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> |
97965478 | 18 | #include <linux/bounds.h> |
1da177e4 | 19 | #include <asm/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 | |
33 | * coelesce naturally under reasonable reclaim pressure and those which | |
34 | * will not. | |
35 | */ | |
36 | #define PAGE_ALLOC_COSTLY_ORDER 3 | |
37 | ||
b2a0ac88 | 38 | #define MIGRATE_UNMOVABLE 0 |
e12ba74d MG |
39 | #define MIGRATE_RECLAIMABLE 1 |
40 | #define MIGRATE_MOVABLE 2 | |
64c5e135 | 41 | #define MIGRATE_RESERVE 3 |
a5d76b54 KH |
42 | #define MIGRATE_ISOLATE 4 /* can't allocate from here */ |
43 | #define MIGRATE_TYPES 5 | |
b2a0ac88 MG |
44 | |
45 | #define for_each_migratetype_order(order, type) \ | |
46 | for (order = 0; order < MAX_ORDER; order++) \ | |
47 | for (type = 0; type < MIGRATE_TYPES; type++) | |
48 | ||
467c996c MG |
49 | extern int page_group_by_mobility_disabled; |
50 | ||
51 | static inline int get_pageblock_migratetype(struct page *page) | |
52 | { | |
467c996c MG |
53 | return get_pageblock_flags_group(page, PB_migrate, PB_migrate_end); |
54 | } | |
55 | ||
1da177e4 | 56 | struct free_area { |
b2a0ac88 | 57 | struct list_head free_list[MIGRATE_TYPES]; |
1da177e4 LT |
58 | unsigned long nr_free; |
59 | }; | |
60 | ||
61 | struct pglist_data; | |
62 | ||
63 | /* | |
64 | * zone->lock and zone->lru_lock are two of the hottest locks in the kernel. | |
65 | * So add a wild amount of padding here to ensure that they fall into separate | |
66 | * cachelines. There are very few zone structures in the machine, so space | |
67 | * consumption is not a concern here. | |
68 | */ | |
69 | #if defined(CONFIG_SMP) | |
70 | struct zone_padding { | |
71 | char x[0]; | |
22fc6ecc | 72 | } ____cacheline_internodealigned_in_smp; |
1da177e4 LT |
73 | #define ZONE_PADDING(name) struct zone_padding name; |
74 | #else | |
75 | #define ZONE_PADDING(name) | |
76 | #endif | |
77 | ||
2244b95a | 78 | enum zone_stat_item { |
51ed4491 | 79 | /* First 128 byte cacheline (assuming 64 bit words) */ |
d23ad423 | 80 | NR_FREE_PAGES, |
b69408e8 | 81 | NR_LRU_BASE, |
4f98a2fe RR |
82 | NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */ |
83 | NR_ACTIVE_ANON, /* " " " " " */ | |
84 | NR_INACTIVE_FILE, /* " " " " " */ | |
85 | NR_ACTIVE_FILE, /* " " " " " */ | |
894bc310 | 86 | NR_UNEVICTABLE, /* " " " " " */ |
5344b7e6 | 87 | NR_MLOCK, /* mlock()ed pages found and moved off LRU */ |
f3dbd344 CL |
88 | NR_ANON_PAGES, /* Mapped anonymous pages */ |
89 | NR_FILE_MAPPED, /* pagecache pages mapped into pagetables. | |
65ba55f5 | 90 | only modified from process context */ |
347ce434 | 91 | NR_FILE_PAGES, |
b1e7a8fd | 92 | NR_FILE_DIRTY, |
ce866b34 | 93 | NR_WRITEBACK, |
51ed4491 CL |
94 | NR_SLAB_RECLAIMABLE, |
95 | NR_SLAB_UNRECLAIMABLE, | |
96 | NR_PAGETABLE, /* used for pagetables */ | |
fd39fc85 | 97 | NR_UNSTABLE_NFS, /* NFS unstable pages */ |
d2c5e30c | 98 | NR_BOUNCE, |
e129b5c2 | 99 | NR_VMSCAN_WRITE, |
4f98a2fe | 100 | /* Second 128 byte cacheline */ |
fc3ba692 | 101 | NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */ |
ca889e6c CL |
102 | #ifdef CONFIG_NUMA |
103 | NUMA_HIT, /* allocated in intended node */ | |
104 | NUMA_MISS, /* allocated in non intended node */ | |
105 | NUMA_FOREIGN, /* was intended here, hit elsewhere */ | |
106 | NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */ | |
107 | NUMA_LOCAL, /* allocation from local node */ | |
108 | NUMA_OTHER, /* allocation from other node */ | |
109 | #endif | |
2244b95a CL |
110 | NR_VM_ZONE_STAT_ITEMS }; |
111 | ||
4f98a2fe RR |
112 | /* |
113 | * We do arithmetic on the LRU lists in various places in the code, | |
114 | * so it is important to keep the active lists LRU_ACTIVE higher in | |
115 | * the array than the corresponding inactive lists, and to keep | |
116 | * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists. | |
117 | * | |
118 | * This has to be kept in sync with the statistics in zone_stat_item | |
119 | * above and the descriptions in vmstat_text in mm/vmstat.c | |
120 | */ | |
121 | #define LRU_BASE 0 | |
122 | #define LRU_ACTIVE 1 | |
123 | #define LRU_FILE 2 | |
124 | ||
b69408e8 | 125 | enum lru_list { |
4f98a2fe RR |
126 | LRU_INACTIVE_ANON = LRU_BASE, |
127 | LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE, | |
128 | LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE, | |
129 | LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE, | |
894bc310 | 130 | LRU_UNEVICTABLE, |
894bc310 LS |
131 | NR_LRU_LISTS |
132 | }; | |
b69408e8 CL |
133 | |
134 | #define for_each_lru(l) for (l = 0; l < NR_LRU_LISTS; l++) | |
135 | ||
894bc310 LS |
136 | #define for_each_evictable_lru(l) for (l = 0; l <= LRU_ACTIVE_FILE; l++) |
137 | ||
4f98a2fe RR |
138 | static inline int is_file_lru(enum lru_list l) |
139 | { | |
140 | return (l == LRU_INACTIVE_FILE || l == LRU_ACTIVE_FILE); | |
141 | } | |
142 | ||
b69408e8 CL |
143 | static inline int is_active_lru(enum lru_list l) |
144 | { | |
4f98a2fe | 145 | return (l == LRU_ACTIVE_ANON || l == LRU_ACTIVE_FILE); |
b69408e8 CL |
146 | } |
147 | ||
894bc310 LS |
148 | static inline int is_unevictable_lru(enum lru_list l) |
149 | { | |
894bc310 | 150 | return (l == LRU_UNEVICTABLE); |
894bc310 LS |
151 | } |
152 | ||
41858966 MG |
153 | enum zone_watermarks { |
154 | WMARK_MIN, | |
155 | WMARK_LOW, | |
156 | WMARK_HIGH, | |
157 | NR_WMARK | |
158 | }; | |
159 | ||
160 | #define min_wmark_pages(z) (z->watermark[WMARK_MIN]) | |
161 | #define low_wmark_pages(z) (z->watermark[WMARK_LOW]) | |
162 | #define high_wmark_pages(z) (z->watermark[WMARK_HIGH]) | |
163 | ||
1da177e4 LT |
164 | struct per_cpu_pages { |
165 | int count; /* number of pages in the list */ | |
1da177e4 LT |
166 | int high; /* high watermark, emptying needed */ |
167 | int batch; /* chunk size for buddy add/remove */ | |
168 | struct list_head list; /* the list of pages */ | |
169 | }; | |
170 | ||
171 | struct per_cpu_pageset { | |
3dfa5721 | 172 | struct per_cpu_pages pcp; |
4037d452 CL |
173 | #ifdef CONFIG_NUMA |
174 | s8 expire; | |
175 | #endif | |
2244b95a | 176 | #ifdef CONFIG_SMP |
df9ecaba | 177 | s8 stat_threshold; |
2244b95a CL |
178 | s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS]; |
179 | #endif | |
1da177e4 LT |
180 | } ____cacheline_aligned_in_smp; |
181 | ||
e7c8d5c9 CL |
182 | #ifdef CONFIG_NUMA |
183 | #define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)]) | |
184 | #else | |
185 | #define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)]) | |
186 | #endif | |
187 | ||
97965478 CL |
188 | #endif /* !__GENERATING_BOUNDS.H */ |
189 | ||
2f1b6248 | 190 | enum zone_type { |
4b51d669 | 191 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 CL |
192 | /* |
193 | * ZONE_DMA is used when there are devices that are not able | |
194 | * to do DMA to all of addressable memory (ZONE_NORMAL). Then we | |
195 | * carve out the portion of memory that is needed for these devices. | |
196 | * The range is arch specific. | |
197 | * | |
198 | * Some examples | |
199 | * | |
200 | * Architecture Limit | |
201 | * --------------------------- | |
202 | * parisc, ia64, sparc <4G | |
203 | * s390 <2G | |
2f1b6248 CL |
204 | * arm Various |
205 | * alpha Unlimited or 0-16MB. | |
206 | * | |
207 | * i386, x86_64 and multiple other arches | |
208 | * <16M. | |
209 | */ | |
210 | ZONE_DMA, | |
4b51d669 | 211 | #endif |
fb0e7942 | 212 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 CL |
213 | /* |
214 | * x86_64 needs two ZONE_DMAs because it supports devices that are | |
215 | * only able to do DMA to the lower 16M but also 32 bit devices that | |
216 | * can only do DMA areas below 4G. | |
217 | */ | |
218 | ZONE_DMA32, | |
fb0e7942 | 219 | #endif |
2f1b6248 CL |
220 | /* |
221 | * Normal addressable memory is in ZONE_NORMAL. DMA operations can be | |
222 | * performed on pages in ZONE_NORMAL if the DMA devices support | |
223 | * transfers to all addressable memory. | |
224 | */ | |
225 | ZONE_NORMAL, | |
e53ef38d | 226 | #ifdef CONFIG_HIGHMEM |
2f1b6248 CL |
227 | /* |
228 | * A memory area that is only addressable by the kernel through | |
229 | * mapping portions into its own address space. This is for example | |
230 | * used by i386 to allow the kernel to address the memory beyond | |
231 | * 900MB. The kernel will set up special mappings (page | |
232 | * table entries on i386) for each page that the kernel needs to | |
233 | * access. | |
234 | */ | |
235 | ZONE_HIGHMEM, | |
e53ef38d | 236 | #endif |
2a1e274a | 237 | ZONE_MOVABLE, |
97965478 | 238 | __MAX_NR_ZONES |
2f1b6248 | 239 | }; |
1da177e4 | 240 | |
97965478 CL |
241 | #ifndef __GENERATING_BOUNDS_H |
242 | ||
1da177e4 LT |
243 | /* |
244 | * When a memory allocation must conform to specific limitations (such | |
245 | * as being suitable for DMA) the caller will pass in hints to the | |
246 | * allocator in the gfp_mask, in the zone modifier bits. These bits | |
247 | * are used to select a priority ordered list of memory zones which | |
19655d34 | 248 | * match the requested limits. See gfp_zone() in include/linux/gfp.h |
1da177e4 | 249 | */ |
fb0e7942 | 250 | |
97965478 | 251 | #if MAX_NR_ZONES < 2 |
4b51d669 | 252 | #define ZONES_SHIFT 0 |
97965478 | 253 | #elif MAX_NR_ZONES <= 2 |
19655d34 | 254 | #define ZONES_SHIFT 1 |
97965478 | 255 | #elif MAX_NR_ZONES <= 4 |
19655d34 | 256 | #define ZONES_SHIFT 2 |
4b51d669 CL |
257 | #else |
258 | #error ZONES_SHIFT -- too many zones configured adjust calculation | |
fb0e7942 | 259 | #endif |
1da177e4 | 260 | |
6e901571 KM |
261 | struct zone_reclaim_stat { |
262 | /* | |
263 | * The pageout code in vmscan.c keeps track of how many of the | |
264 | * mem/swap backed and file backed pages are refeferenced. | |
265 | * The higher the rotated/scanned ratio, the more valuable | |
266 | * that cache is. | |
267 | * | |
268 | * The anon LRU stats live in [0], file LRU stats in [1] | |
269 | */ | |
270 | unsigned long recent_rotated[2]; | |
271 | unsigned long recent_scanned[2]; | |
272 | }; | |
273 | ||
1da177e4 LT |
274 | struct zone { |
275 | /* Fields commonly accessed by the page allocator */ | |
41858966 MG |
276 | |
277 | /* zone watermarks, access with *_wmark_pages(zone) macros */ | |
278 | unsigned long watermark[NR_WMARK]; | |
279 | ||
1da177e4 LT |
280 | /* |
281 | * We don't know if the memory that we're going to allocate will be freeable | |
282 | * or/and it will be released eventually, so to avoid totally wasting several | |
283 | * GB of ram we must reserve some of the lower zone memory (otherwise we risk | |
284 | * to run OOM on the lower zones despite there's tons of freeable ram | |
285 | * on the higher zones). This array is recalculated at runtime if the | |
286 | * sysctl_lowmem_reserve_ratio sysctl changes. | |
287 | */ | |
288 | unsigned long lowmem_reserve[MAX_NR_ZONES]; | |
289 | ||
e7c8d5c9 | 290 | #ifdef CONFIG_NUMA |
d5f541ed | 291 | int node; |
9614634f CL |
292 | /* |
293 | * zone reclaim becomes active if more unmapped pages exist. | |
294 | */ | |
8417bba4 | 295 | unsigned long min_unmapped_pages; |
0ff38490 | 296 | unsigned long min_slab_pages; |
e7c8d5c9 CL |
297 | struct per_cpu_pageset *pageset[NR_CPUS]; |
298 | #else | |
1da177e4 | 299 | struct per_cpu_pageset pageset[NR_CPUS]; |
e7c8d5c9 | 300 | #endif |
1da177e4 LT |
301 | /* |
302 | * free areas of different sizes | |
303 | */ | |
304 | spinlock_t lock; | |
bdc8cb98 DH |
305 | #ifdef CONFIG_MEMORY_HOTPLUG |
306 | /* see spanned/present_pages for more description */ | |
307 | seqlock_t span_seqlock; | |
308 | #endif | |
1da177e4 LT |
309 | struct free_area free_area[MAX_ORDER]; |
310 | ||
835c134e MG |
311 | #ifndef CONFIG_SPARSEMEM |
312 | /* | |
d9c23400 | 313 | * Flags for a pageblock_nr_pages block. See pageblock-flags.h. |
835c134e MG |
314 | * In SPARSEMEM, this map is stored in struct mem_section |
315 | */ | |
316 | unsigned long *pageblock_flags; | |
317 | #endif /* CONFIG_SPARSEMEM */ | |
318 | ||
1da177e4 LT |
319 | |
320 | ZONE_PADDING(_pad1_) | |
321 | ||
322 | /* Fields commonly accessed by the page reclaim scanner */ | |
323 | spinlock_t lru_lock; | |
6e08a369 | 324 | struct zone_lru { |
b69408e8 | 325 | struct list_head list; |
6e08a369 | 326 | unsigned long nr_saved_scan; /* accumulated for batching */ |
b69408e8 | 327 | } lru[NR_LRU_LISTS]; |
4f98a2fe | 328 | |
6e901571 | 329 | struct zone_reclaim_stat reclaim_stat; |
4f98a2fe | 330 | |
1da177e4 | 331 | unsigned long pages_scanned; /* since last reclaim */ |
e815af95 | 332 | unsigned long flags; /* zone flags, see below */ |
753ee728 | 333 | |
2244b95a CL |
334 | /* Zone statistics */ |
335 | atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS]; | |
9eeff239 | 336 | |
1da177e4 LT |
337 | /* |
338 | * prev_priority holds the scanning priority for this zone. It is | |
339 | * defined as the scanning priority at which we achieved our reclaim | |
340 | * target at the previous try_to_free_pages() or balance_pgdat() | |
341 | * invokation. | |
342 | * | |
343 | * We use prev_priority as a measure of how much stress page reclaim is | |
344 | * under - it drives the swappiness decision: whether to unmap mapped | |
345 | * pages. | |
346 | * | |
3bb1a852 | 347 | * Access to both this field is quite racy even on uniprocessor. But |
1da177e4 LT |
348 | * it is expected to average out OK. |
349 | */ | |
1da177e4 LT |
350 | int prev_priority; |
351 | ||
556adecb RR |
352 | /* |
353 | * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on | |
354 | * this zone's LRU. Maintained by the pageout code. | |
355 | */ | |
356 | unsigned int inactive_ratio; | |
357 | ||
1da177e4 LT |
358 | |
359 | ZONE_PADDING(_pad2_) | |
360 | /* Rarely used or read-mostly fields */ | |
361 | ||
362 | /* | |
363 | * wait_table -- the array holding the hash table | |
02b694de | 364 | * wait_table_hash_nr_entries -- the size of the hash table array |
1da177e4 LT |
365 | * wait_table_bits -- wait_table_size == (1 << wait_table_bits) |
366 | * | |
367 | * The purpose of all these is to keep track of the people | |
368 | * waiting for a page to become available and make them | |
369 | * runnable again when possible. The trouble is that this | |
370 | * consumes a lot of space, especially when so few things | |
371 | * wait on pages at a given time. So instead of using | |
372 | * per-page waitqueues, we use a waitqueue hash table. | |
373 | * | |
374 | * The bucket discipline is to sleep on the same queue when | |
375 | * colliding and wake all in that wait queue when removing. | |
376 | * When something wakes, it must check to be sure its page is | |
377 | * truly available, a la thundering herd. The cost of a | |
378 | * collision is great, but given the expected load of the | |
379 | * table, they should be so rare as to be outweighed by the | |
380 | * benefits from the saved space. | |
381 | * | |
382 | * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the | |
383 | * primary users of these fields, and in mm/page_alloc.c | |
384 | * free_area_init_core() performs the initialization of them. | |
385 | */ | |
386 | wait_queue_head_t * wait_table; | |
02b694de | 387 | unsigned long wait_table_hash_nr_entries; |
1da177e4 LT |
388 | unsigned long wait_table_bits; |
389 | ||
390 | /* | |
391 | * Discontig memory support fields. | |
392 | */ | |
393 | struct pglist_data *zone_pgdat; | |
1da177e4 LT |
394 | /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */ |
395 | unsigned long zone_start_pfn; | |
396 | ||
bdc8cb98 DH |
397 | /* |
398 | * zone_start_pfn, spanned_pages and present_pages are all | |
399 | * protected by span_seqlock. It is a seqlock because it has | |
400 | * to be read outside of zone->lock, and it is done in the main | |
401 | * allocator path. But, it is written quite infrequently. | |
402 | * | |
403 | * The lock is declared along with zone->lock because it is | |
404 | * frequently read in proximity to zone->lock. It's good to | |
405 | * give them a chance of being in the same cacheline. | |
406 | */ | |
1da177e4 LT |
407 | unsigned long spanned_pages; /* total size, including holes */ |
408 | unsigned long present_pages; /* amount of memory (excluding holes) */ | |
409 | ||
410 | /* | |
411 | * rarely used fields: | |
412 | */ | |
15ad7cdc | 413 | const char *name; |
22fc6ecc | 414 | } ____cacheline_internodealigned_in_smp; |
1da177e4 | 415 | |
e815af95 DR |
416 | typedef enum { |
417 | ZONE_ALL_UNRECLAIMABLE, /* all pages pinned */ | |
418 | ZONE_RECLAIM_LOCKED, /* prevents concurrent reclaim */ | |
098d7f12 | 419 | ZONE_OOM_LOCKED, /* zone is in OOM killer zonelist */ |
e815af95 DR |
420 | } zone_flags_t; |
421 | ||
422 | static inline void zone_set_flag(struct zone *zone, zone_flags_t flag) | |
423 | { | |
424 | set_bit(flag, &zone->flags); | |
425 | } | |
d773ed6b DR |
426 | |
427 | static inline int zone_test_and_set_flag(struct zone *zone, zone_flags_t flag) | |
428 | { | |
429 | return test_and_set_bit(flag, &zone->flags); | |
430 | } | |
431 | ||
e815af95 DR |
432 | static inline void zone_clear_flag(struct zone *zone, zone_flags_t flag) |
433 | { | |
434 | clear_bit(flag, &zone->flags); | |
435 | } | |
436 | ||
437 | static inline int zone_is_all_unreclaimable(const struct zone *zone) | |
438 | { | |
439 | return test_bit(ZONE_ALL_UNRECLAIMABLE, &zone->flags); | |
440 | } | |
d773ed6b | 441 | |
e815af95 DR |
442 | static inline int zone_is_reclaim_locked(const struct zone *zone) |
443 | { | |
444 | return test_bit(ZONE_RECLAIM_LOCKED, &zone->flags); | |
445 | } | |
d773ed6b | 446 | |
098d7f12 DR |
447 | static inline int zone_is_oom_locked(const struct zone *zone) |
448 | { | |
449 | return test_bit(ZONE_OOM_LOCKED, &zone->flags); | |
450 | } | |
e815af95 | 451 | |
1da177e4 LT |
452 | /* |
453 | * The "priority" of VM scanning is how much of the queues we will scan in one | |
454 | * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the | |
455 | * queues ("queue_length >> 12") during an aging round. | |
456 | */ | |
457 | #define DEF_PRIORITY 12 | |
458 | ||
9276b1bc PJ |
459 | /* Maximum number of zones on a zonelist */ |
460 | #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES) | |
461 | ||
462 | #ifdef CONFIG_NUMA | |
523b9458 CL |
463 | |
464 | /* | |
465 | * The NUMA zonelists are doubled becausse we need zonelists that restrict the | |
466 | * allocations to a single node for GFP_THISNODE. | |
467 | * | |
54a6eb5c MG |
468 | * [0] : Zonelist with fallback |
469 | * [1] : No fallback (GFP_THISNODE) | |
523b9458 | 470 | */ |
54a6eb5c | 471 | #define MAX_ZONELISTS 2 |
523b9458 CL |
472 | |
473 | ||
9276b1bc PJ |
474 | /* |
475 | * We cache key information from each zonelist for smaller cache | |
476 | * footprint when scanning for free pages in get_page_from_freelist(). | |
477 | * | |
478 | * 1) The BITMAP fullzones tracks which zones in a zonelist have come | |
479 | * up short of free memory since the last time (last_fullzone_zap) | |
480 | * we zero'd fullzones. | |
481 | * 2) The array z_to_n[] maps each zone in the zonelist to its node | |
482 | * id, so that we can efficiently evaluate whether that node is | |
483 | * set in the current tasks mems_allowed. | |
484 | * | |
485 | * Both fullzones and z_to_n[] are one-to-one with the zonelist, | |
486 | * indexed by a zones offset in the zonelist zones[] array. | |
487 | * | |
488 | * The get_page_from_freelist() routine does two scans. During the | |
489 | * first scan, we skip zones whose corresponding bit in 'fullzones' | |
490 | * is set or whose corresponding node in current->mems_allowed (which | |
491 | * comes from cpusets) is not set. During the second scan, we bypass | |
492 | * this zonelist_cache, to ensure we look methodically at each zone. | |
493 | * | |
494 | * Once per second, we zero out (zap) fullzones, forcing us to | |
495 | * reconsider nodes that might have regained more free memory. | |
496 | * The field last_full_zap is the time we last zapped fullzones. | |
497 | * | |
498 | * This mechanism reduces the amount of time we waste repeatedly | |
499 | * reexaming zones for free memory when they just came up low on | |
500 | * memory momentarilly ago. | |
501 | * | |
502 | * The zonelist_cache struct members logically belong in struct | |
503 | * zonelist. However, the mempolicy zonelists constructed for | |
504 | * MPOL_BIND are intentionally variable length (and usually much | |
505 | * shorter). A general purpose mechanism for handling structs with | |
506 | * multiple variable length members is more mechanism than we want | |
507 | * here. We resort to some special case hackery instead. | |
508 | * | |
509 | * The MPOL_BIND zonelists don't need this zonelist_cache (in good | |
510 | * part because they are shorter), so we put the fixed length stuff | |
511 | * at the front of the zonelist struct, ending in a variable length | |
512 | * zones[], as is needed by MPOL_BIND. | |
513 | * | |
514 | * Then we put the optional zonelist cache on the end of the zonelist | |
515 | * struct. This optional stuff is found by a 'zlcache_ptr' pointer in | |
516 | * the fixed length portion at the front of the struct. This pointer | |
517 | * both enables us to find the zonelist cache, and in the case of | |
518 | * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL) | |
519 | * to know that the zonelist cache is not there. | |
520 | * | |
521 | * The end result is that struct zonelists come in two flavors: | |
522 | * 1) The full, fixed length version, shown below, and | |
523 | * 2) The custom zonelists for MPOL_BIND. | |
524 | * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache. | |
525 | * | |
526 | * Even though there may be multiple CPU cores on a node modifying | |
527 | * fullzones or last_full_zap in the same zonelist_cache at the same | |
528 | * time, we don't lock it. This is just hint data - if it is wrong now | |
529 | * and then, the allocator will still function, perhaps a bit slower. | |
530 | */ | |
531 | ||
532 | ||
533 | struct zonelist_cache { | |
9276b1bc | 534 | unsigned short z_to_n[MAX_ZONES_PER_ZONELIST]; /* zone->nid */ |
7253f4ef | 535 | DECLARE_BITMAP(fullzones, MAX_ZONES_PER_ZONELIST); /* zone full? */ |
9276b1bc PJ |
536 | unsigned long last_full_zap; /* when last zap'd (jiffies) */ |
537 | }; | |
538 | #else | |
54a6eb5c | 539 | #define MAX_ZONELISTS 1 |
9276b1bc PJ |
540 | struct zonelist_cache; |
541 | #endif | |
542 | ||
dd1a239f MG |
543 | /* |
544 | * This struct contains information about a zone in a zonelist. It is stored | |
545 | * here to avoid dereferences into large structures and lookups of tables | |
546 | */ | |
547 | struct zoneref { | |
548 | struct zone *zone; /* Pointer to actual zone */ | |
549 | int zone_idx; /* zone_idx(zoneref->zone) */ | |
550 | }; | |
551 | ||
1da177e4 LT |
552 | /* |
553 | * One allocation request operates on a zonelist. A zonelist | |
554 | * is a list of zones, the first one is the 'goal' of the | |
555 | * allocation, the other zones are fallback zones, in decreasing | |
556 | * priority. | |
557 | * | |
9276b1bc PJ |
558 | * If zlcache_ptr is not NULL, then it is just the address of zlcache, |
559 | * as explained above. If zlcache_ptr is NULL, there is no zlcache. | |
dd1a239f MG |
560 | * * |
561 | * To speed the reading of the zonelist, the zonerefs contain the zone index | |
562 | * of the entry being read. Helper functions to access information given | |
563 | * a struct zoneref are | |
564 | * | |
565 | * zonelist_zone() - Return the struct zone * for an entry in _zonerefs | |
566 | * zonelist_zone_idx() - Return the index of the zone for an entry | |
567 | * zonelist_node_idx() - Return the index of the node for an entry | |
1da177e4 LT |
568 | */ |
569 | struct zonelist { | |
9276b1bc | 570 | struct zonelist_cache *zlcache_ptr; // NULL or &zlcache |
dd1a239f | 571 | struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1]; |
9276b1bc PJ |
572 | #ifdef CONFIG_NUMA |
573 | struct zonelist_cache zlcache; // optional ... | |
574 | #endif | |
1da177e4 LT |
575 | }; |
576 | ||
c713216d MG |
577 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
578 | struct node_active_region { | |
579 | unsigned long start_pfn; | |
580 | unsigned long end_pfn; | |
581 | int nid; | |
582 | }; | |
583 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ | |
1da177e4 | 584 | |
5b99cd0e HC |
585 | #ifndef CONFIG_DISCONTIGMEM |
586 | /* The array of struct pages - for discontigmem use pgdat->lmem_map */ | |
587 | extern struct page *mem_map; | |
588 | #endif | |
589 | ||
1da177e4 LT |
590 | /* |
591 | * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM | |
592 | * (mostly NUMA machines?) to denote a higher-level memory zone than the | |
593 | * zone denotes. | |
594 | * | |
595 | * On NUMA machines, each NUMA node would have a pg_data_t to describe | |
596 | * it's memory layout. | |
597 | * | |
598 | * Memory statistics and page replacement data structures are maintained on a | |
599 | * per-zone basis. | |
600 | */ | |
601 | struct bootmem_data; | |
602 | typedef struct pglist_data { | |
603 | struct zone node_zones[MAX_NR_ZONES]; | |
523b9458 | 604 | struct zonelist node_zonelists[MAX_ZONELISTS]; |
1da177e4 | 605 | int nr_zones; |
52d4b9ac | 606 | #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */ |
1da177e4 | 607 | struct page *node_mem_map; |
52d4b9ac KH |
608 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
609 | struct page_cgroup *node_page_cgroup; | |
610 | #endif | |
d41dee36 | 611 | #endif |
1da177e4 | 612 | struct bootmem_data *bdata; |
208d54e5 DH |
613 | #ifdef CONFIG_MEMORY_HOTPLUG |
614 | /* | |
615 | * Must be held any time you expect node_start_pfn, node_present_pages | |
616 | * or node_spanned_pages stay constant. Holding this will also | |
617 | * guarantee that any pfn_valid() stays that way. | |
618 | * | |
619 | * Nests above zone->lock and zone->size_seqlock. | |
620 | */ | |
621 | spinlock_t node_size_lock; | |
622 | #endif | |
1da177e4 LT |
623 | unsigned long node_start_pfn; |
624 | unsigned long node_present_pages; /* total number of physical pages */ | |
625 | unsigned long node_spanned_pages; /* total size of physical page | |
626 | range, including holes */ | |
627 | int node_id; | |
1da177e4 LT |
628 | wait_queue_head_t kswapd_wait; |
629 | struct task_struct *kswapd; | |
630 | int kswapd_max_order; | |
631 | } pg_data_t; | |
632 | ||
633 | #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages) | |
634 | #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages) | |
d41dee36 | 635 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
408fde81 | 636 | #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr)) |
d41dee36 AW |
637 | #else |
638 | #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr)) | |
639 | #endif | |
408fde81 | 640 | #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr)) |
1da177e4 | 641 | |
208d54e5 DH |
642 | #include <linux/memory_hotplug.h> |
643 | ||
1da177e4 LT |
644 | void get_zone_counts(unsigned long *active, unsigned long *inactive, |
645 | unsigned long *free); | |
646 | void build_all_zonelists(void); | |
647 | void wakeup_kswapd(struct zone *zone, int order); | |
648 | int zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
7fb1d9fc | 649 | int classzone_idx, int alloc_flags); |
a2f3aa02 DH |
650 | enum memmap_context { |
651 | MEMMAP_EARLY, | |
652 | MEMMAP_HOTPLUG, | |
653 | }; | |
718127cc | 654 | extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn, |
a2f3aa02 DH |
655 | unsigned long size, |
656 | enum memmap_context context); | |
718127cc | 657 | |
1da177e4 LT |
658 | #ifdef CONFIG_HAVE_MEMORY_PRESENT |
659 | void memory_present(int nid, unsigned long start, unsigned long end); | |
660 | #else | |
661 | static inline void memory_present(int nid, unsigned long start, unsigned long end) {} | |
662 | #endif | |
663 | ||
664 | #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE | |
665 | unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long); | |
666 | #endif | |
667 | ||
668 | /* | |
669 | * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc. | |
670 | */ | |
671 | #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones) | |
672 | ||
f3fe6512 CK |
673 | static inline int populated_zone(struct zone *zone) |
674 | { | |
675 | return (!!zone->present_pages); | |
676 | } | |
677 | ||
2a1e274a MG |
678 | extern int movable_zone; |
679 | ||
680 | static inline int zone_movable_is_highmem(void) | |
681 | { | |
682 | #if defined(CONFIG_HIGHMEM) && defined(CONFIG_ARCH_POPULATES_NODE_MAP) | |
683 | return movable_zone == ZONE_HIGHMEM; | |
684 | #else | |
685 | return 0; | |
686 | #endif | |
687 | } | |
688 | ||
2f1b6248 | 689 | static inline int is_highmem_idx(enum zone_type idx) |
1da177e4 | 690 | { |
e53ef38d | 691 | #ifdef CONFIG_HIGHMEM |
2a1e274a MG |
692 | return (idx == ZONE_HIGHMEM || |
693 | (idx == ZONE_MOVABLE && zone_movable_is_highmem())); | |
e53ef38d CL |
694 | #else |
695 | return 0; | |
696 | #endif | |
1da177e4 LT |
697 | } |
698 | ||
2f1b6248 | 699 | static inline int is_normal_idx(enum zone_type idx) |
1da177e4 LT |
700 | { |
701 | return (idx == ZONE_NORMAL); | |
702 | } | |
9328b8fa | 703 | |
1da177e4 LT |
704 | /** |
705 | * is_highmem - helper function to quickly check if a struct zone is a | |
706 | * highmem zone or not. This is an attempt to keep references | |
707 | * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum. | |
708 | * @zone - pointer to struct zone variable | |
709 | */ | |
710 | static inline int is_highmem(struct zone *zone) | |
711 | { | |
e53ef38d | 712 | #ifdef CONFIG_HIGHMEM |
ddc81ed2 HH |
713 | int zone_off = (char *)zone - (char *)zone->zone_pgdat->node_zones; |
714 | return zone_off == ZONE_HIGHMEM * sizeof(*zone) || | |
715 | (zone_off == ZONE_MOVABLE * sizeof(*zone) && | |
716 | zone_movable_is_highmem()); | |
e53ef38d CL |
717 | #else |
718 | return 0; | |
719 | #endif | |
1da177e4 LT |
720 | } |
721 | ||
722 | static inline int is_normal(struct zone *zone) | |
723 | { | |
724 | return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL; | |
725 | } | |
726 | ||
9328b8fa NP |
727 | static inline int is_dma32(struct zone *zone) |
728 | { | |
fb0e7942 | 729 | #ifdef CONFIG_ZONE_DMA32 |
9328b8fa | 730 | return zone == zone->zone_pgdat->node_zones + ZONE_DMA32; |
fb0e7942 CL |
731 | #else |
732 | return 0; | |
733 | #endif | |
9328b8fa NP |
734 | } |
735 | ||
736 | static inline int is_dma(struct zone *zone) | |
737 | { | |
4b51d669 | 738 | #ifdef CONFIG_ZONE_DMA |
9328b8fa | 739 | return zone == zone->zone_pgdat->node_zones + ZONE_DMA; |
4b51d669 CL |
740 | #else |
741 | return 0; | |
742 | #endif | |
9328b8fa NP |
743 | } |
744 | ||
1da177e4 LT |
745 | /* These two functions are used to setup the per zone pages min values */ |
746 | struct ctl_table; | |
747 | struct file; | |
748 | int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *, | |
749 | void __user *, size_t *, loff_t *); | |
750 | extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1]; | |
751 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *, | |
752 | void __user *, size_t *, loff_t *); | |
8ad4b1fb RS |
753 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, struct file *, |
754 | void __user *, size_t *, loff_t *); | |
9614634f CL |
755 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int, |
756 | struct file *, void __user *, size_t *, loff_t *); | |
0ff38490 CL |
757 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int, |
758 | struct file *, void __user *, size_t *, loff_t *); | |
1da177e4 | 759 | |
f0c0b2b8 KH |
760 | extern int numa_zonelist_order_handler(struct ctl_table *, int, |
761 | struct file *, void __user *, size_t *, loff_t *); | |
762 | extern char numa_zonelist_order[]; | |
763 | #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */ | |
764 | ||
93b7504e | 765 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
766 | |
767 | extern struct pglist_data contig_page_data; | |
768 | #define NODE_DATA(nid) (&contig_page_data) | |
769 | #define NODE_MEM_MAP(nid) mem_map | |
1da177e4 | 770 | |
93b7504e | 771 | #else /* CONFIG_NEED_MULTIPLE_NODES */ |
1da177e4 LT |
772 | |
773 | #include <asm/mmzone.h> | |
774 | ||
93b7504e | 775 | #endif /* !CONFIG_NEED_MULTIPLE_NODES */ |
348f8b6c | 776 | |
95144c78 KH |
777 | extern struct pglist_data *first_online_pgdat(void); |
778 | extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat); | |
779 | extern struct zone *next_zone(struct zone *zone); | |
8357f869 KH |
780 | |
781 | /** | |
12d15f0d | 782 | * for_each_online_pgdat - helper macro to iterate over all online nodes |
8357f869 KH |
783 | * @pgdat - pointer to a pg_data_t variable |
784 | */ | |
785 | #define for_each_online_pgdat(pgdat) \ | |
786 | for (pgdat = first_online_pgdat(); \ | |
787 | pgdat; \ | |
788 | pgdat = next_online_pgdat(pgdat)) | |
8357f869 KH |
789 | /** |
790 | * for_each_zone - helper macro to iterate over all memory zones | |
791 | * @zone - pointer to struct zone variable | |
792 | * | |
793 | * The user only needs to declare the zone variable, for_each_zone | |
794 | * fills it in. | |
795 | */ | |
796 | #define for_each_zone(zone) \ | |
797 | for (zone = (first_online_pgdat())->node_zones; \ | |
798 | zone; \ | |
799 | zone = next_zone(zone)) | |
800 | ||
ee99c71c KM |
801 | #define for_each_populated_zone(zone) \ |
802 | for (zone = (first_online_pgdat())->node_zones; \ | |
803 | zone; \ | |
804 | zone = next_zone(zone)) \ | |
805 | if (!populated_zone(zone)) \ | |
806 | ; /* do nothing */ \ | |
807 | else | |
808 | ||
dd1a239f MG |
809 | static inline struct zone *zonelist_zone(struct zoneref *zoneref) |
810 | { | |
811 | return zoneref->zone; | |
812 | } | |
813 | ||
814 | static inline int zonelist_zone_idx(struct zoneref *zoneref) | |
815 | { | |
816 | return zoneref->zone_idx; | |
817 | } | |
818 | ||
819 | static inline int zonelist_node_idx(struct zoneref *zoneref) | |
820 | { | |
821 | #ifdef CONFIG_NUMA | |
822 | /* zone_to_nid not available in this context */ | |
823 | return zoneref->zone->node; | |
824 | #else | |
825 | return 0; | |
826 | #endif /* CONFIG_NUMA */ | |
827 | } | |
828 | ||
19770b32 MG |
829 | /** |
830 | * 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 | |
831 | * @z - The cursor used as a starting point for the search | |
832 | * @highest_zoneidx - The zone index of the highest zone to return | |
833 | * @nodes - An optional nodemask to filter the zonelist with | |
834 | * @zone - The first suitable zone found is returned via this parameter | |
835 | * | |
836 | * This function returns the next zone at or below a given zone index that is | |
837 | * within the allowed nodemask using a cursor as the starting point for the | |
5bead2a0 MG |
838 | * search. The zoneref returned is a cursor that represents the current zone |
839 | * being examined. It should be advanced by one before calling | |
840 | * next_zones_zonelist again. | |
19770b32 MG |
841 | */ |
842 | struct zoneref *next_zones_zonelist(struct zoneref *z, | |
843 | enum zone_type highest_zoneidx, | |
844 | nodemask_t *nodes, | |
845 | struct zone **zone); | |
dd1a239f | 846 | |
19770b32 MG |
847 | /** |
848 | * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist | |
849 | * @zonelist - The zonelist to search for a suitable zone | |
850 | * @highest_zoneidx - The zone index of the highest zone to return | |
851 | * @nodes - An optional nodemask to filter the zonelist with | |
852 | * @zone - The first suitable zone found is returned via this parameter | |
853 | * | |
854 | * This function returns the first zone at or below a given zone index that is | |
855 | * within the allowed nodemask. The zoneref returned is a cursor that can be | |
5bead2a0 MG |
856 | * used to iterate the zonelist with next_zones_zonelist by advancing it by |
857 | * one before calling. | |
19770b32 | 858 | */ |
dd1a239f | 859 | static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist, |
19770b32 MG |
860 | enum zone_type highest_zoneidx, |
861 | nodemask_t *nodes, | |
862 | struct zone **zone) | |
54a6eb5c | 863 | { |
19770b32 MG |
864 | return next_zones_zonelist(zonelist->_zonerefs, highest_zoneidx, nodes, |
865 | zone); | |
54a6eb5c MG |
866 | } |
867 | ||
19770b32 MG |
868 | /** |
869 | * 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 | |
870 | * @zone - The current zone in the iterator | |
871 | * @z - The current pointer within zonelist->zones being iterated | |
872 | * @zlist - The zonelist being iterated | |
873 | * @highidx - The zone index of the highest zone to return | |
874 | * @nodemask - Nodemask allowed by the allocator | |
875 | * | |
876 | * This iterator iterates though all zones at or below a given zone index and | |
877 | * within a given nodemask | |
878 | */ | |
879 | #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \ | |
880 | for (z = first_zones_zonelist(zlist, highidx, nodemask, &zone); \ | |
881 | zone; \ | |
5bead2a0 | 882 | z = next_zones_zonelist(++z, highidx, nodemask, &zone)) \ |
54a6eb5c MG |
883 | |
884 | /** | |
885 | * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index | |
886 | * @zone - The current zone in the iterator | |
887 | * @z - The current pointer within zonelist->zones being iterated | |
888 | * @zlist - The zonelist being iterated | |
889 | * @highidx - The zone index of the highest zone to return | |
890 | * | |
891 | * This iterator iterates though all zones at or below a given zone index. | |
892 | */ | |
893 | #define for_each_zone_zonelist(zone, z, zlist, highidx) \ | |
19770b32 | 894 | for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL) |
54a6eb5c | 895 | |
d41dee36 AW |
896 | #ifdef CONFIG_SPARSEMEM |
897 | #include <asm/sparsemem.h> | |
898 | #endif | |
899 | ||
c713216d MG |
900 | #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \ |
901 | !defined(CONFIG_ARCH_POPULATES_NODE_MAP) | |
b4544568 AM |
902 | static inline unsigned long early_pfn_to_nid(unsigned long pfn) |
903 | { | |
904 | return 0; | |
905 | } | |
b159d43f AW |
906 | #endif |
907 | ||
2bdaf115 AW |
908 | #ifdef CONFIG_FLATMEM |
909 | #define pfn_to_nid(pfn) (0) | |
910 | #endif | |
911 | ||
d41dee36 AW |
912 | #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT) |
913 | #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT) | |
914 | ||
915 | #ifdef CONFIG_SPARSEMEM | |
916 | ||
917 | /* | |
918 | * SECTION_SHIFT #bits space required to store a section # | |
919 | * | |
920 | * PA_SECTION_SHIFT physical address to/from section number | |
921 | * PFN_SECTION_SHIFT pfn to/from section number | |
922 | */ | |
923 | #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS) | |
924 | ||
925 | #define PA_SECTION_SHIFT (SECTION_SIZE_BITS) | |
926 | #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT) | |
927 | ||
928 | #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT) | |
929 | ||
930 | #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT) | |
931 | #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1)) | |
932 | ||
835c134e | 933 | #define SECTION_BLOCKFLAGS_BITS \ |
d9c23400 | 934 | ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS) |
835c134e | 935 | |
d41dee36 AW |
936 | #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS |
937 | #error Allocator MAX_ORDER exceeds SECTION_SIZE | |
938 | #endif | |
939 | ||
940 | struct page; | |
52d4b9ac | 941 | struct page_cgroup; |
d41dee36 | 942 | struct mem_section { |
29751f69 AW |
943 | /* |
944 | * This is, logically, a pointer to an array of struct | |
945 | * pages. However, it is stored with some other magic. | |
946 | * (see sparse.c::sparse_init_one_section()) | |
947 | * | |
30c253e6 AW |
948 | * Additionally during early boot we encode node id of |
949 | * the location of the section here to guide allocation. | |
950 | * (see sparse.c::memory_present()) | |
951 | * | |
29751f69 AW |
952 | * Making it a UL at least makes someone do a cast |
953 | * before using it wrong. | |
954 | */ | |
955 | unsigned long section_mem_map; | |
5c0e3066 MG |
956 | |
957 | /* See declaration of similar field in struct zone */ | |
958 | unsigned long *pageblock_flags; | |
52d4b9ac KH |
959 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
960 | /* | |
961 | * If !SPARSEMEM, pgdat doesn't have page_cgroup pointer. We use | |
962 | * section. (see memcontrol.h/page_cgroup.h about this.) | |
963 | */ | |
964 | struct page_cgroup *page_cgroup; | |
965 | unsigned long pad; | |
966 | #endif | |
d41dee36 AW |
967 | }; |
968 | ||
3e347261 BP |
969 | #ifdef CONFIG_SPARSEMEM_EXTREME |
970 | #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section)) | |
971 | #else | |
972 | #define SECTIONS_PER_ROOT 1 | |
973 | #endif | |
802f192e | 974 | |
3e347261 BP |
975 | #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT) |
976 | #define NR_SECTION_ROOTS (NR_MEM_SECTIONS / SECTIONS_PER_ROOT) | |
977 | #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1) | |
802f192e | 978 | |
3e347261 BP |
979 | #ifdef CONFIG_SPARSEMEM_EXTREME |
980 | extern struct mem_section *mem_section[NR_SECTION_ROOTS]; | |
802f192e | 981 | #else |
3e347261 BP |
982 | extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]; |
983 | #endif | |
d41dee36 | 984 | |
29751f69 AW |
985 | static inline struct mem_section *__nr_to_section(unsigned long nr) |
986 | { | |
3e347261 BP |
987 | if (!mem_section[SECTION_NR_TO_ROOT(nr)]) |
988 | return NULL; | |
989 | return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK]; | |
29751f69 | 990 | } |
4ca644d9 | 991 | extern int __section_nr(struct mem_section* ms); |
04753278 | 992 | extern unsigned long usemap_size(void); |
29751f69 AW |
993 | |
994 | /* | |
995 | * We use the lower bits of the mem_map pointer to store | |
996 | * a little bit of information. There should be at least | |
997 | * 3 bits here due to 32-bit alignment. | |
998 | */ | |
999 | #define SECTION_MARKED_PRESENT (1UL<<0) | |
1000 | #define SECTION_HAS_MEM_MAP (1UL<<1) | |
1001 | #define SECTION_MAP_LAST_BIT (1UL<<2) | |
1002 | #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1)) | |
30c253e6 | 1003 | #define SECTION_NID_SHIFT 2 |
29751f69 AW |
1004 | |
1005 | static inline struct page *__section_mem_map_addr(struct mem_section *section) | |
1006 | { | |
1007 | unsigned long map = section->section_mem_map; | |
1008 | map &= SECTION_MAP_MASK; | |
1009 | return (struct page *)map; | |
1010 | } | |
1011 | ||
540557b9 | 1012 | static inline int present_section(struct mem_section *section) |
29751f69 | 1013 | { |
802f192e | 1014 | return (section && (section->section_mem_map & SECTION_MARKED_PRESENT)); |
29751f69 AW |
1015 | } |
1016 | ||
540557b9 AW |
1017 | static inline int present_section_nr(unsigned long nr) |
1018 | { | |
1019 | return present_section(__nr_to_section(nr)); | |
1020 | } | |
1021 | ||
1022 | static inline int valid_section(struct mem_section *section) | |
29751f69 | 1023 | { |
802f192e | 1024 | return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP)); |
29751f69 AW |
1025 | } |
1026 | ||
1027 | static inline int valid_section_nr(unsigned long nr) | |
1028 | { | |
1029 | return valid_section(__nr_to_section(nr)); | |
1030 | } | |
1031 | ||
d41dee36 AW |
1032 | static inline struct mem_section *__pfn_to_section(unsigned long pfn) |
1033 | { | |
29751f69 | 1034 | return __nr_to_section(pfn_to_section_nr(pfn)); |
d41dee36 AW |
1035 | } |
1036 | ||
d41dee36 AW |
1037 | static inline int pfn_valid(unsigned long pfn) |
1038 | { | |
1039 | if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) | |
1040 | return 0; | |
29751f69 | 1041 | return valid_section(__nr_to_section(pfn_to_section_nr(pfn))); |
d41dee36 AW |
1042 | } |
1043 | ||
540557b9 AW |
1044 | static inline int pfn_present(unsigned long pfn) |
1045 | { | |
1046 | if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) | |
1047 | return 0; | |
1048 | return present_section(__nr_to_section(pfn_to_section_nr(pfn))); | |
1049 | } | |
1050 | ||
d41dee36 AW |
1051 | /* |
1052 | * These are _only_ used during initialisation, therefore they | |
1053 | * can use __initdata ... They could have names to indicate | |
1054 | * this restriction. | |
1055 | */ | |
1056 | #ifdef CONFIG_NUMA | |
161599ff AW |
1057 | #define pfn_to_nid(pfn) \ |
1058 | ({ \ | |
1059 | unsigned long __pfn_to_nid_pfn = (pfn); \ | |
1060 | page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \ | |
1061 | }) | |
2bdaf115 AW |
1062 | #else |
1063 | #define pfn_to_nid(pfn) (0) | |
d41dee36 AW |
1064 | #endif |
1065 | ||
d41dee36 AW |
1066 | #define early_pfn_valid(pfn) pfn_valid(pfn) |
1067 | void sparse_init(void); | |
1068 | #else | |
1069 | #define sparse_init() do {} while (0) | |
28ae55c9 | 1070 | #define sparse_index_init(_sec, _nid) do {} while (0) |
d41dee36 AW |
1071 | #endif /* CONFIG_SPARSEMEM */ |
1072 | ||
75167957 | 1073 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES |
cc2559bc | 1074 | bool early_pfn_in_nid(unsigned long pfn, int nid); |
75167957 AW |
1075 | #else |
1076 | #define early_pfn_in_nid(pfn, nid) (1) | |
1077 | #endif | |
1078 | ||
d41dee36 AW |
1079 | #ifndef early_pfn_valid |
1080 | #define early_pfn_valid(pfn) (1) | |
1081 | #endif | |
1082 | ||
1083 | void memory_present(int nid, unsigned long start, unsigned long end); | |
1084 | unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long); | |
1085 | ||
14e07298 AW |
1086 | /* |
1087 | * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we | |
1088 | * need to check pfn validility within that MAX_ORDER_NR_PAGES block. | |
1089 | * pfn_valid_within() should be used in this case; we optimise this away | |
1090 | * when we have no holes within a MAX_ORDER_NR_PAGES block. | |
1091 | */ | |
1092 | #ifdef CONFIG_HOLES_IN_ZONE | |
1093 | #define pfn_valid_within(pfn) pfn_valid(pfn) | |
1094 | #else | |
1095 | #define pfn_valid_within(pfn) (1) | |
1096 | #endif | |
1097 | ||
eb33575c MG |
1098 | #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL |
1099 | /* | |
1100 | * pfn_valid() is meant to be able to tell if a given PFN has valid memmap | |
1101 | * associated with it or not. In FLATMEM, it is expected that holes always | |
1102 | * have valid memmap as long as there is valid PFNs either side of the hole. | |
1103 | * In SPARSEMEM, it is assumed that a valid section has a memmap for the | |
1104 | * entire section. | |
1105 | * | |
1106 | * However, an ARM, and maybe other embedded architectures in the future | |
1107 | * free memmap backing holes to save memory on the assumption the memmap is | |
1108 | * never used. The page_zone linkages are then broken even though pfn_valid() | |
1109 | * returns true. A walker of the full memmap must then do this additional | |
1110 | * check to ensure the memmap they are looking at is sane by making sure | |
1111 | * the zone and PFN linkages are still valid. This is expensive, but walkers | |
1112 | * of the full memmap are extremely rare. | |
1113 | */ | |
1114 | int memmap_valid_within(unsigned long pfn, | |
1115 | struct page *page, struct zone *zone); | |
1116 | #else | |
1117 | static inline int memmap_valid_within(unsigned long pfn, | |
1118 | struct page *page, struct zone *zone) | |
1119 | { | |
1120 | return 1; | |
1121 | } | |
1122 | #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */ | |
1123 | ||
97965478 | 1124 | #endif /* !__GENERATING_BOUNDS.H */ |
1da177e4 | 1125 | #endif /* !__ASSEMBLY__ */ |
1da177e4 | 1126 | #endif /* _LINUX_MMZONE_H */ |