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 */ | |
c6a7f572 KM |
97 | NR_KERNEL_STACK, |
98 | /* Second 128 byte cacheline */ | |
fd39fc85 | 99 | NR_UNSTABLE_NFS, /* NFS unstable pages */ |
d2c5e30c | 100 | NR_BOUNCE, |
e129b5c2 | 101 | NR_VMSCAN_WRITE, |
fc3ba692 | 102 | NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */ |
4b02108a | 103 | NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */ |
ca889e6c CL |
104 | #ifdef CONFIG_NUMA |
105 | NUMA_HIT, /* allocated in intended node */ | |
106 | NUMA_MISS, /* allocated in non intended node */ | |
107 | NUMA_FOREIGN, /* was intended here, hit elsewhere */ | |
108 | NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */ | |
109 | NUMA_LOCAL, /* allocation from local node */ | |
110 | NUMA_OTHER, /* allocation from other node */ | |
111 | #endif | |
2244b95a CL |
112 | NR_VM_ZONE_STAT_ITEMS }; |
113 | ||
4f98a2fe RR |
114 | /* |
115 | * We do arithmetic on the LRU lists in various places in the code, | |
116 | * so it is important to keep the active lists LRU_ACTIVE higher in | |
117 | * the array than the corresponding inactive lists, and to keep | |
118 | * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists. | |
119 | * | |
120 | * This has to be kept in sync with the statistics in zone_stat_item | |
121 | * above and the descriptions in vmstat_text in mm/vmstat.c | |
122 | */ | |
123 | #define LRU_BASE 0 | |
124 | #define LRU_ACTIVE 1 | |
125 | #define LRU_FILE 2 | |
126 | ||
b69408e8 | 127 | enum lru_list { |
4f98a2fe RR |
128 | LRU_INACTIVE_ANON = LRU_BASE, |
129 | LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE, | |
130 | LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE, | |
131 | LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE, | |
894bc310 | 132 | LRU_UNEVICTABLE, |
894bc310 LS |
133 | NR_LRU_LISTS |
134 | }; | |
b69408e8 CL |
135 | |
136 | #define for_each_lru(l) for (l = 0; l < NR_LRU_LISTS; l++) | |
137 | ||
894bc310 LS |
138 | #define for_each_evictable_lru(l) for (l = 0; l <= LRU_ACTIVE_FILE; l++) |
139 | ||
4f98a2fe RR |
140 | static inline int is_file_lru(enum lru_list l) |
141 | { | |
142 | return (l == LRU_INACTIVE_FILE || l == LRU_ACTIVE_FILE); | |
143 | } | |
144 | ||
b69408e8 CL |
145 | static inline int is_active_lru(enum lru_list l) |
146 | { | |
4f98a2fe | 147 | return (l == LRU_ACTIVE_ANON || l == LRU_ACTIVE_FILE); |
b69408e8 CL |
148 | } |
149 | ||
894bc310 LS |
150 | static inline int is_unevictable_lru(enum lru_list l) |
151 | { | |
894bc310 | 152 | return (l == LRU_UNEVICTABLE); |
894bc310 LS |
153 | } |
154 | ||
41858966 MG |
155 | enum zone_watermarks { |
156 | WMARK_MIN, | |
157 | WMARK_LOW, | |
158 | WMARK_HIGH, | |
159 | NR_WMARK | |
160 | }; | |
161 | ||
162 | #define min_wmark_pages(z) (z->watermark[WMARK_MIN]) | |
163 | #define low_wmark_pages(z) (z->watermark[WMARK_LOW]) | |
164 | #define high_wmark_pages(z) (z->watermark[WMARK_HIGH]) | |
165 | ||
1da177e4 LT |
166 | struct per_cpu_pages { |
167 | int count; /* number of pages in the list */ | |
1da177e4 LT |
168 | int high; /* high watermark, emptying needed */ |
169 | int batch; /* chunk size for buddy add/remove */ | |
170 | struct list_head list; /* the list of pages */ | |
171 | }; | |
172 | ||
173 | struct per_cpu_pageset { | |
3dfa5721 | 174 | struct per_cpu_pages pcp; |
4037d452 CL |
175 | #ifdef CONFIG_NUMA |
176 | s8 expire; | |
177 | #endif | |
2244b95a | 178 | #ifdef CONFIG_SMP |
df9ecaba | 179 | s8 stat_threshold; |
2244b95a CL |
180 | s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS]; |
181 | #endif | |
1da177e4 LT |
182 | } ____cacheline_aligned_in_smp; |
183 | ||
e7c8d5c9 CL |
184 | #ifdef CONFIG_NUMA |
185 | #define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)]) | |
186 | #else | |
187 | #define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)]) | |
188 | #endif | |
189 | ||
97965478 CL |
190 | #endif /* !__GENERATING_BOUNDS.H */ |
191 | ||
2f1b6248 | 192 | enum zone_type { |
4b51d669 | 193 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 CL |
194 | /* |
195 | * ZONE_DMA is used when there are devices that are not able | |
196 | * to do DMA to all of addressable memory (ZONE_NORMAL). Then we | |
197 | * carve out the portion of memory that is needed for these devices. | |
198 | * The range is arch specific. | |
199 | * | |
200 | * Some examples | |
201 | * | |
202 | * Architecture Limit | |
203 | * --------------------------- | |
204 | * parisc, ia64, sparc <4G | |
205 | * s390 <2G | |
2f1b6248 CL |
206 | * arm Various |
207 | * alpha Unlimited or 0-16MB. | |
208 | * | |
209 | * i386, x86_64 and multiple other arches | |
210 | * <16M. | |
211 | */ | |
212 | ZONE_DMA, | |
4b51d669 | 213 | #endif |
fb0e7942 | 214 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 CL |
215 | /* |
216 | * x86_64 needs two ZONE_DMAs because it supports devices that are | |
217 | * only able to do DMA to the lower 16M but also 32 bit devices that | |
218 | * can only do DMA areas below 4G. | |
219 | */ | |
220 | ZONE_DMA32, | |
fb0e7942 | 221 | #endif |
2f1b6248 CL |
222 | /* |
223 | * Normal addressable memory is in ZONE_NORMAL. DMA operations can be | |
224 | * performed on pages in ZONE_NORMAL if the DMA devices support | |
225 | * transfers to all addressable memory. | |
226 | */ | |
227 | ZONE_NORMAL, | |
e53ef38d | 228 | #ifdef CONFIG_HIGHMEM |
2f1b6248 CL |
229 | /* |
230 | * A memory area that is only addressable by the kernel through | |
231 | * mapping portions into its own address space. This is for example | |
232 | * used by i386 to allow the kernel to address the memory beyond | |
233 | * 900MB. The kernel will set up special mappings (page | |
234 | * table entries on i386) for each page that the kernel needs to | |
235 | * access. | |
236 | */ | |
237 | ZONE_HIGHMEM, | |
e53ef38d | 238 | #endif |
2a1e274a | 239 | ZONE_MOVABLE, |
97965478 | 240 | __MAX_NR_ZONES |
2f1b6248 | 241 | }; |
1da177e4 | 242 | |
97965478 CL |
243 | #ifndef __GENERATING_BOUNDS_H |
244 | ||
1da177e4 LT |
245 | /* |
246 | * When a memory allocation must conform to specific limitations (such | |
247 | * as being suitable for DMA) the caller will pass in hints to the | |
248 | * allocator in the gfp_mask, in the zone modifier bits. These bits | |
249 | * are used to select a priority ordered list of memory zones which | |
19655d34 | 250 | * match the requested limits. See gfp_zone() in include/linux/gfp.h |
1da177e4 | 251 | */ |
fb0e7942 | 252 | |
97965478 | 253 | #if MAX_NR_ZONES < 2 |
4b51d669 | 254 | #define ZONES_SHIFT 0 |
97965478 | 255 | #elif MAX_NR_ZONES <= 2 |
19655d34 | 256 | #define ZONES_SHIFT 1 |
97965478 | 257 | #elif MAX_NR_ZONES <= 4 |
19655d34 | 258 | #define ZONES_SHIFT 2 |
4b51d669 CL |
259 | #else |
260 | #error ZONES_SHIFT -- too many zones configured adjust calculation | |
fb0e7942 | 261 | #endif |
1da177e4 | 262 | |
6e901571 KM |
263 | struct zone_reclaim_stat { |
264 | /* | |
265 | * The pageout code in vmscan.c keeps track of how many of the | |
266 | * mem/swap backed and file backed pages are refeferenced. | |
267 | * The higher the rotated/scanned ratio, the more valuable | |
268 | * that cache is. | |
269 | * | |
270 | * The anon LRU stats live in [0], file LRU stats in [1] | |
271 | */ | |
272 | unsigned long recent_rotated[2]; | |
273 | unsigned long recent_scanned[2]; | |
274 | }; | |
275 | ||
1da177e4 LT |
276 | struct zone { |
277 | /* Fields commonly accessed by the page allocator */ | |
41858966 MG |
278 | |
279 | /* zone watermarks, access with *_wmark_pages(zone) macros */ | |
280 | unsigned long watermark[NR_WMARK]; | |
281 | ||
1da177e4 LT |
282 | /* |
283 | * We don't know if the memory that we're going to allocate will be freeable | |
284 | * or/and it will be released eventually, so to avoid totally wasting several | |
285 | * GB of ram we must reserve some of the lower zone memory (otherwise we risk | |
286 | * to run OOM on the lower zones despite there's tons of freeable ram | |
287 | * on the higher zones). This array is recalculated at runtime if the | |
288 | * sysctl_lowmem_reserve_ratio sysctl changes. | |
289 | */ | |
290 | unsigned long lowmem_reserve[MAX_NR_ZONES]; | |
291 | ||
e7c8d5c9 | 292 | #ifdef CONFIG_NUMA |
d5f541ed | 293 | int node; |
9614634f CL |
294 | /* |
295 | * zone reclaim becomes active if more unmapped pages exist. | |
296 | */ | |
8417bba4 | 297 | unsigned long min_unmapped_pages; |
0ff38490 | 298 | unsigned long min_slab_pages; |
e7c8d5c9 CL |
299 | struct per_cpu_pageset *pageset[NR_CPUS]; |
300 | #else | |
1da177e4 | 301 | struct per_cpu_pageset pageset[NR_CPUS]; |
e7c8d5c9 | 302 | #endif |
1da177e4 LT |
303 | /* |
304 | * free areas of different sizes | |
305 | */ | |
306 | spinlock_t lock; | |
bdc8cb98 DH |
307 | #ifdef CONFIG_MEMORY_HOTPLUG |
308 | /* see spanned/present_pages for more description */ | |
309 | seqlock_t span_seqlock; | |
310 | #endif | |
1da177e4 LT |
311 | struct free_area free_area[MAX_ORDER]; |
312 | ||
835c134e MG |
313 | #ifndef CONFIG_SPARSEMEM |
314 | /* | |
d9c23400 | 315 | * Flags for a pageblock_nr_pages block. See pageblock-flags.h. |
835c134e MG |
316 | * In SPARSEMEM, this map is stored in struct mem_section |
317 | */ | |
318 | unsigned long *pageblock_flags; | |
319 | #endif /* CONFIG_SPARSEMEM */ | |
320 | ||
1da177e4 LT |
321 | |
322 | ZONE_PADDING(_pad1_) | |
323 | ||
324 | /* Fields commonly accessed by the page reclaim scanner */ | |
325 | spinlock_t lru_lock; | |
6e08a369 | 326 | struct zone_lru { |
b69408e8 | 327 | struct list_head list; |
6e08a369 | 328 | unsigned long nr_saved_scan; /* accumulated for batching */ |
b69408e8 | 329 | } lru[NR_LRU_LISTS]; |
4f98a2fe | 330 | |
6e901571 | 331 | struct zone_reclaim_stat reclaim_stat; |
4f98a2fe | 332 | |
1da177e4 | 333 | unsigned long pages_scanned; /* since last reclaim */ |
e815af95 | 334 | unsigned long flags; /* zone flags, see below */ |
753ee728 | 335 | |
2244b95a CL |
336 | /* Zone statistics */ |
337 | atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS]; | |
9eeff239 | 338 | |
1da177e4 LT |
339 | /* |
340 | * prev_priority holds the scanning priority for this zone. It is | |
341 | * defined as the scanning priority at which we achieved our reclaim | |
342 | * target at the previous try_to_free_pages() or balance_pgdat() | |
343 | * invokation. | |
344 | * | |
345 | * We use prev_priority as a measure of how much stress page reclaim is | |
346 | * under - it drives the swappiness decision: whether to unmap mapped | |
347 | * pages. | |
348 | * | |
3bb1a852 | 349 | * Access to both this field is quite racy even on uniprocessor. But |
1da177e4 LT |
350 | * it is expected to average out OK. |
351 | */ | |
1da177e4 LT |
352 | int prev_priority; |
353 | ||
556adecb RR |
354 | /* |
355 | * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on | |
356 | * this zone's LRU. Maintained by the pageout code. | |
357 | */ | |
358 | unsigned int inactive_ratio; | |
359 | ||
1da177e4 LT |
360 | |
361 | ZONE_PADDING(_pad2_) | |
362 | /* Rarely used or read-mostly fields */ | |
363 | ||
364 | /* | |
365 | * wait_table -- the array holding the hash table | |
02b694de | 366 | * wait_table_hash_nr_entries -- the size of the hash table array |
1da177e4 LT |
367 | * wait_table_bits -- wait_table_size == (1 << wait_table_bits) |
368 | * | |
369 | * The purpose of all these is to keep track of the people | |
370 | * waiting for a page to become available and make them | |
371 | * runnable again when possible. The trouble is that this | |
372 | * consumes a lot of space, especially when so few things | |
373 | * wait on pages at a given time. So instead of using | |
374 | * per-page waitqueues, we use a waitqueue hash table. | |
375 | * | |
376 | * The bucket discipline is to sleep on the same queue when | |
377 | * colliding and wake all in that wait queue when removing. | |
378 | * When something wakes, it must check to be sure its page is | |
379 | * truly available, a la thundering herd. The cost of a | |
380 | * collision is great, but given the expected load of the | |
381 | * table, they should be so rare as to be outweighed by the | |
382 | * benefits from the saved space. | |
383 | * | |
384 | * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the | |
385 | * primary users of these fields, and in mm/page_alloc.c | |
386 | * free_area_init_core() performs the initialization of them. | |
387 | */ | |
388 | wait_queue_head_t * wait_table; | |
02b694de | 389 | unsigned long wait_table_hash_nr_entries; |
1da177e4 LT |
390 | unsigned long wait_table_bits; |
391 | ||
392 | /* | |
393 | * Discontig memory support fields. | |
394 | */ | |
395 | struct pglist_data *zone_pgdat; | |
1da177e4 LT |
396 | /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */ |
397 | unsigned long zone_start_pfn; | |
398 | ||
bdc8cb98 DH |
399 | /* |
400 | * zone_start_pfn, spanned_pages and present_pages are all | |
401 | * protected by span_seqlock. It is a seqlock because it has | |
402 | * to be read outside of zone->lock, and it is done in the main | |
403 | * allocator path. But, it is written quite infrequently. | |
404 | * | |
405 | * The lock is declared along with zone->lock because it is | |
406 | * frequently read in proximity to zone->lock. It's good to | |
407 | * give them a chance of being in the same cacheline. | |
408 | */ | |
1da177e4 LT |
409 | unsigned long spanned_pages; /* total size, including holes */ |
410 | unsigned long present_pages; /* amount of memory (excluding holes) */ | |
411 | ||
412 | /* | |
413 | * rarely used fields: | |
414 | */ | |
15ad7cdc | 415 | const char *name; |
22fc6ecc | 416 | } ____cacheline_internodealigned_in_smp; |
1da177e4 | 417 | |
e815af95 DR |
418 | typedef enum { |
419 | ZONE_ALL_UNRECLAIMABLE, /* all pages pinned */ | |
420 | ZONE_RECLAIM_LOCKED, /* prevents concurrent reclaim */ | |
098d7f12 | 421 | ZONE_OOM_LOCKED, /* zone is in OOM killer zonelist */ |
e815af95 DR |
422 | } zone_flags_t; |
423 | ||
424 | static inline void zone_set_flag(struct zone *zone, zone_flags_t flag) | |
425 | { | |
426 | set_bit(flag, &zone->flags); | |
427 | } | |
d773ed6b DR |
428 | |
429 | static inline int zone_test_and_set_flag(struct zone *zone, zone_flags_t flag) | |
430 | { | |
431 | return test_and_set_bit(flag, &zone->flags); | |
432 | } | |
433 | ||
e815af95 DR |
434 | static inline void zone_clear_flag(struct zone *zone, zone_flags_t flag) |
435 | { | |
436 | clear_bit(flag, &zone->flags); | |
437 | } | |
438 | ||
439 | static inline int zone_is_all_unreclaimable(const struct zone *zone) | |
440 | { | |
441 | return test_bit(ZONE_ALL_UNRECLAIMABLE, &zone->flags); | |
442 | } | |
d773ed6b | 443 | |
e815af95 DR |
444 | static inline int zone_is_reclaim_locked(const struct zone *zone) |
445 | { | |
446 | return test_bit(ZONE_RECLAIM_LOCKED, &zone->flags); | |
447 | } | |
d773ed6b | 448 | |
098d7f12 DR |
449 | static inline int zone_is_oom_locked(const struct zone *zone) |
450 | { | |
451 | return test_bit(ZONE_OOM_LOCKED, &zone->flags); | |
452 | } | |
e815af95 | 453 | |
1da177e4 LT |
454 | /* |
455 | * The "priority" of VM scanning is how much of the queues we will scan in one | |
456 | * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the | |
457 | * queues ("queue_length >> 12") during an aging round. | |
458 | */ | |
459 | #define DEF_PRIORITY 12 | |
460 | ||
9276b1bc PJ |
461 | /* Maximum number of zones on a zonelist */ |
462 | #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES) | |
463 | ||
464 | #ifdef CONFIG_NUMA | |
523b9458 CL |
465 | |
466 | /* | |
467 | * The NUMA zonelists are doubled becausse we need zonelists that restrict the | |
468 | * allocations to a single node for GFP_THISNODE. | |
469 | * | |
54a6eb5c MG |
470 | * [0] : Zonelist with fallback |
471 | * [1] : No fallback (GFP_THISNODE) | |
523b9458 | 472 | */ |
54a6eb5c | 473 | #define MAX_ZONELISTS 2 |
523b9458 CL |
474 | |
475 | ||
9276b1bc PJ |
476 | /* |
477 | * We cache key information from each zonelist for smaller cache | |
478 | * footprint when scanning for free pages in get_page_from_freelist(). | |
479 | * | |
480 | * 1) The BITMAP fullzones tracks which zones in a zonelist have come | |
481 | * up short of free memory since the last time (last_fullzone_zap) | |
482 | * we zero'd fullzones. | |
483 | * 2) The array z_to_n[] maps each zone in the zonelist to its node | |
484 | * id, so that we can efficiently evaluate whether that node is | |
485 | * set in the current tasks mems_allowed. | |
486 | * | |
487 | * Both fullzones and z_to_n[] are one-to-one with the zonelist, | |
488 | * indexed by a zones offset in the zonelist zones[] array. | |
489 | * | |
490 | * The get_page_from_freelist() routine does two scans. During the | |
491 | * first scan, we skip zones whose corresponding bit in 'fullzones' | |
492 | * is set or whose corresponding node in current->mems_allowed (which | |
493 | * comes from cpusets) is not set. During the second scan, we bypass | |
494 | * this zonelist_cache, to ensure we look methodically at each zone. | |
495 | * | |
496 | * Once per second, we zero out (zap) fullzones, forcing us to | |
497 | * reconsider nodes that might have regained more free memory. | |
498 | * The field last_full_zap is the time we last zapped fullzones. | |
499 | * | |
500 | * This mechanism reduces the amount of time we waste repeatedly | |
501 | * reexaming zones for free memory when they just came up low on | |
502 | * memory momentarilly ago. | |
503 | * | |
504 | * The zonelist_cache struct members logically belong in struct | |
505 | * zonelist. However, the mempolicy zonelists constructed for | |
506 | * MPOL_BIND are intentionally variable length (and usually much | |
507 | * shorter). A general purpose mechanism for handling structs with | |
508 | * multiple variable length members is more mechanism than we want | |
509 | * here. We resort to some special case hackery instead. | |
510 | * | |
511 | * The MPOL_BIND zonelists don't need this zonelist_cache (in good | |
512 | * part because they are shorter), so we put the fixed length stuff | |
513 | * at the front of the zonelist struct, ending in a variable length | |
514 | * zones[], as is needed by MPOL_BIND. | |
515 | * | |
516 | * Then we put the optional zonelist cache on the end of the zonelist | |
517 | * struct. This optional stuff is found by a 'zlcache_ptr' pointer in | |
518 | * the fixed length portion at the front of the struct. This pointer | |
519 | * both enables us to find the zonelist cache, and in the case of | |
520 | * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL) | |
521 | * to know that the zonelist cache is not there. | |
522 | * | |
523 | * The end result is that struct zonelists come in two flavors: | |
524 | * 1) The full, fixed length version, shown below, and | |
525 | * 2) The custom zonelists for MPOL_BIND. | |
526 | * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache. | |
527 | * | |
528 | * Even though there may be multiple CPU cores on a node modifying | |
529 | * fullzones or last_full_zap in the same zonelist_cache at the same | |
530 | * time, we don't lock it. This is just hint data - if it is wrong now | |
531 | * and then, the allocator will still function, perhaps a bit slower. | |
532 | */ | |
533 | ||
534 | ||
535 | struct zonelist_cache { | |
9276b1bc | 536 | unsigned short z_to_n[MAX_ZONES_PER_ZONELIST]; /* zone->nid */ |
7253f4ef | 537 | DECLARE_BITMAP(fullzones, MAX_ZONES_PER_ZONELIST); /* zone full? */ |
9276b1bc PJ |
538 | unsigned long last_full_zap; /* when last zap'd (jiffies) */ |
539 | }; | |
540 | #else | |
54a6eb5c | 541 | #define MAX_ZONELISTS 1 |
9276b1bc PJ |
542 | struct zonelist_cache; |
543 | #endif | |
544 | ||
dd1a239f MG |
545 | /* |
546 | * This struct contains information about a zone in a zonelist. It is stored | |
547 | * here to avoid dereferences into large structures and lookups of tables | |
548 | */ | |
549 | struct zoneref { | |
550 | struct zone *zone; /* Pointer to actual zone */ | |
551 | int zone_idx; /* zone_idx(zoneref->zone) */ | |
552 | }; | |
553 | ||
1da177e4 LT |
554 | /* |
555 | * One allocation request operates on a zonelist. A zonelist | |
556 | * is a list of zones, the first one is the 'goal' of the | |
557 | * allocation, the other zones are fallback zones, in decreasing | |
558 | * priority. | |
559 | * | |
9276b1bc PJ |
560 | * If zlcache_ptr is not NULL, then it is just the address of zlcache, |
561 | * as explained above. If zlcache_ptr is NULL, there is no zlcache. | |
dd1a239f MG |
562 | * * |
563 | * To speed the reading of the zonelist, the zonerefs contain the zone index | |
564 | * of the entry being read. Helper functions to access information given | |
565 | * a struct zoneref are | |
566 | * | |
567 | * zonelist_zone() - Return the struct zone * for an entry in _zonerefs | |
568 | * zonelist_zone_idx() - Return the index of the zone for an entry | |
569 | * zonelist_node_idx() - Return the index of the node for an entry | |
1da177e4 LT |
570 | */ |
571 | struct zonelist { | |
9276b1bc | 572 | struct zonelist_cache *zlcache_ptr; // NULL or &zlcache |
dd1a239f | 573 | struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1]; |
9276b1bc PJ |
574 | #ifdef CONFIG_NUMA |
575 | struct zonelist_cache zlcache; // optional ... | |
576 | #endif | |
1da177e4 LT |
577 | }; |
578 | ||
c713216d MG |
579 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
580 | struct node_active_region { | |
581 | unsigned long start_pfn; | |
582 | unsigned long end_pfn; | |
583 | int nid; | |
584 | }; | |
585 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ | |
1da177e4 | 586 | |
5b99cd0e HC |
587 | #ifndef CONFIG_DISCONTIGMEM |
588 | /* The array of struct pages - for discontigmem use pgdat->lmem_map */ | |
589 | extern struct page *mem_map; | |
590 | #endif | |
591 | ||
1da177e4 LT |
592 | /* |
593 | * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM | |
594 | * (mostly NUMA machines?) to denote a higher-level memory zone than the | |
595 | * zone denotes. | |
596 | * | |
597 | * On NUMA machines, each NUMA node would have a pg_data_t to describe | |
598 | * it's memory layout. | |
599 | * | |
600 | * Memory statistics and page replacement data structures are maintained on a | |
601 | * per-zone basis. | |
602 | */ | |
603 | struct bootmem_data; | |
604 | typedef struct pglist_data { | |
605 | struct zone node_zones[MAX_NR_ZONES]; | |
523b9458 | 606 | struct zonelist node_zonelists[MAX_ZONELISTS]; |
1da177e4 | 607 | int nr_zones; |
52d4b9ac | 608 | #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */ |
1da177e4 | 609 | struct page *node_mem_map; |
52d4b9ac KH |
610 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
611 | struct page_cgroup *node_page_cgroup; | |
612 | #endif | |
d41dee36 | 613 | #endif |
1da177e4 | 614 | struct bootmem_data *bdata; |
208d54e5 DH |
615 | #ifdef CONFIG_MEMORY_HOTPLUG |
616 | /* | |
617 | * Must be held any time you expect node_start_pfn, node_present_pages | |
618 | * or node_spanned_pages stay constant. Holding this will also | |
619 | * guarantee that any pfn_valid() stays that way. | |
620 | * | |
621 | * Nests above zone->lock and zone->size_seqlock. | |
622 | */ | |
623 | spinlock_t node_size_lock; | |
624 | #endif | |
1da177e4 LT |
625 | unsigned long node_start_pfn; |
626 | unsigned long node_present_pages; /* total number of physical pages */ | |
627 | unsigned long node_spanned_pages; /* total size of physical page | |
628 | range, including holes */ | |
629 | int node_id; | |
1da177e4 LT |
630 | wait_queue_head_t kswapd_wait; |
631 | struct task_struct *kswapd; | |
632 | int kswapd_max_order; | |
633 | } pg_data_t; | |
634 | ||
635 | #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages) | |
636 | #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages) | |
d41dee36 | 637 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
408fde81 | 638 | #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr)) |
d41dee36 AW |
639 | #else |
640 | #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr)) | |
641 | #endif | |
408fde81 | 642 | #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr)) |
1da177e4 | 643 | |
208d54e5 DH |
644 | #include <linux/memory_hotplug.h> |
645 | ||
1da177e4 LT |
646 | void get_zone_counts(unsigned long *active, unsigned long *inactive, |
647 | unsigned long *free); | |
648 | void build_all_zonelists(void); | |
649 | void wakeup_kswapd(struct zone *zone, int order); | |
650 | int zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
7fb1d9fc | 651 | int classzone_idx, int alloc_flags); |
a2f3aa02 DH |
652 | enum memmap_context { |
653 | MEMMAP_EARLY, | |
654 | MEMMAP_HOTPLUG, | |
655 | }; | |
718127cc | 656 | extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn, |
a2f3aa02 DH |
657 | unsigned long size, |
658 | enum memmap_context context); | |
718127cc | 659 | |
1da177e4 LT |
660 | #ifdef CONFIG_HAVE_MEMORY_PRESENT |
661 | void memory_present(int nid, unsigned long start, unsigned long end); | |
662 | #else | |
663 | static inline void memory_present(int nid, unsigned long start, unsigned long end) {} | |
664 | #endif | |
665 | ||
666 | #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE | |
667 | unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long); | |
668 | #endif | |
669 | ||
670 | /* | |
671 | * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc. | |
672 | */ | |
673 | #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones) | |
674 | ||
f3fe6512 CK |
675 | static inline int populated_zone(struct zone *zone) |
676 | { | |
677 | return (!!zone->present_pages); | |
678 | } | |
679 | ||
2a1e274a MG |
680 | extern int movable_zone; |
681 | ||
682 | static inline int zone_movable_is_highmem(void) | |
683 | { | |
684 | #if defined(CONFIG_HIGHMEM) && defined(CONFIG_ARCH_POPULATES_NODE_MAP) | |
685 | return movable_zone == ZONE_HIGHMEM; | |
686 | #else | |
687 | return 0; | |
688 | #endif | |
689 | } | |
690 | ||
2f1b6248 | 691 | static inline int is_highmem_idx(enum zone_type idx) |
1da177e4 | 692 | { |
e53ef38d | 693 | #ifdef CONFIG_HIGHMEM |
2a1e274a MG |
694 | return (idx == ZONE_HIGHMEM || |
695 | (idx == ZONE_MOVABLE && zone_movable_is_highmem())); | |
e53ef38d CL |
696 | #else |
697 | return 0; | |
698 | #endif | |
1da177e4 LT |
699 | } |
700 | ||
2f1b6248 | 701 | static inline int is_normal_idx(enum zone_type idx) |
1da177e4 LT |
702 | { |
703 | return (idx == ZONE_NORMAL); | |
704 | } | |
9328b8fa | 705 | |
1da177e4 LT |
706 | /** |
707 | * is_highmem - helper function to quickly check if a struct zone is a | |
708 | * highmem zone or not. This is an attempt to keep references | |
709 | * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum. | |
710 | * @zone - pointer to struct zone variable | |
711 | */ | |
712 | static inline int is_highmem(struct zone *zone) | |
713 | { | |
e53ef38d | 714 | #ifdef CONFIG_HIGHMEM |
ddc81ed2 HH |
715 | int zone_off = (char *)zone - (char *)zone->zone_pgdat->node_zones; |
716 | return zone_off == ZONE_HIGHMEM * sizeof(*zone) || | |
717 | (zone_off == ZONE_MOVABLE * sizeof(*zone) && | |
718 | zone_movable_is_highmem()); | |
e53ef38d CL |
719 | #else |
720 | return 0; | |
721 | #endif | |
1da177e4 LT |
722 | } |
723 | ||
724 | static inline int is_normal(struct zone *zone) | |
725 | { | |
726 | return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL; | |
727 | } | |
728 | ||
9328b8fa NP |
729 | static inline int is_dma32(struct zone *zone) |
730 | { | |
fb0e7942 | 731 | #ifdef CONFIG_ZONE_DMA32 |
9328b8fa | 732 | return zone == zone->zone_pgdat->node_zones + ZONE_DMA32; |
fb0e7942 CL |
733 | #else |
734 | return 0; | |
735 | #endif | |
9328b8fa NP |
736 | } |
737 | ||
738 | static inline int is_dma(struct zone *zone) | |
739 | { | |
4b51d669 | 740 | #ifdef CONFIG_ZONE_DMA |
9328b8fa | 741 | return zone == zone->zone_pgdat->node_zones + ZONE_DMA; |
4b51d669 CL |
742 | #else |
743 | return 0; | |
744 | #endif | |
9328b8fa NP |
745 | } |
746 | ||
1da177e4 LT |
747 | /* These two functions are used to setup the per zone pages min values */ |
748 | struct ctl_table; | |
749 | struct file; | |
750 | int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *, | |
751 | void __user *, size_t *, loff_t *); | |
752 | extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1]; | |
753 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *, | |
754 | void __user *, size_t *, loff_t *); | |
8ad4b1fb RS |
755 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, struct file *, |
756 | void __user *, size_t *, loff_t *); | |
9614634f CL |
757 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int, |
758 | struct file *, void __user *, size_t *, loff_t *); | |
0ff38490 CL |
759 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int, |
760 | struct file *, void __user *, size_t *, loff_t *); | |
1da177e4 | 761 | |
f0c0b2b8 KH |
762 | extern int numa_zonelist_order_handler(struct ctl_table *, int, |
763 | struct file *, void __user *, size_t *, loff_t *); | |
764 | extern char numa_zonelist_order[]; | |
765 | #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */ | |
766 | ||
93b7504e | 767 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
768 | |
769 | extern struct pglist_data contig_page_data; | |
770 | #define NODE_DATA(nid) (&contig_page_data) | |
771 | #define NODE_MEM_MAP(nid) mem_map | |
1da177e4 | 772 | |
93b7504e | 773 | #else /* CONFIG_NEED_MULTIPLE_NODES */ |
1da177e4 LT |
774 | |
775 | #include <asm/mmzone.h> | |
776 | ||
93b7504e | 777 | #endif /* !CONFIG_NEED_MULTIPLE_NODES */ |
348f8b6c | 778 | |
95144c78 KH |
779 | extern struct pglist_data *first_online_pgdat(void); |
780 | extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat); | |
781 | extern struct zone *next_zone(struct zone *zone); | |
8357f869 KH |
782 | |
783 | /** | |
12d15f0d | 784 | * for_each_online_pgdat - helper macro to iterate over all online nodes |
8357f869 KH |
785 | * @pgdat - pointer to a pg_data_t variable |
786 | */ | |
787 | #define for_each_online_pgdat(pgdat) \ | |
788 | for (pgdat = first_online_pgdat(); \ | |
789 | pgdat; \ | |
790 | pgdat = next_online_pgdat(pgdat)) | |
8357f869 KH |
791 | /** |
792 | * for_each_zone - helper macro to iterate over all memory zones | |
793 | * @zone - pointer to struct zone variable | |
794 | * | |
795 | * The user only needs to declare the zone variable, for_each_zone | |
796 | * fills it in. | |
797 | */ | |
798 | #define for_each_zone(zone) \ | |
799 | for (zone = (first_online_pgdat())->node_zones; \ | |
800 | zone; \ | |
801 | zone = next_zone(zone)) | |
802 | ||
ee99c71c KM |
803 | #define for_each_populated_zone(zone) \ |
804 | for (zone = (first_online_pgdat())->node_zones; \ | |
805 | zone; \ | |
806 | zone = next_zone(zone)) \ | |
807 | if (!populated_zone(zone)) \ | |
808 | ; /* do nothing */ \ | |
809 | else | |
810 | ||
dd1a239f MG |
811 | static inline struct zone *zonelist_zone(struct zoneref *zoneref) |
812 | { | |
813 | return zoneref->zone; | |
814 | } | |
815 | ||
816 | static inline int zonelist_zone_idx(struct zoneref *zoneref) | |
817 | { | |
818 | return zoneref->zone_idx; | |
819 | } | |
820 | ||
821 | static inline int zonelist_node_idx(struct zoneref *zoneref) | |
822 | { | |
823 | #ifdef CONFIG_NUMA | |
824 | /* zone_to_nid not available in this context */ | |
825 | return zoneref->zone->node; | |
826 | #else | |
827 | return 0; | |
828 | #endif /* CONFIG_NUMA */ | |
829 | } | |
830 | ||
19770b32 MG |
831 | /** |
832 | * 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 | |
833 | * @z - The cursor used as a starting point for the search | |
834 | * @highest_zoneidx - The zone index of the highest zone to return | |
835 | * @nodes - An optional nodemask to filter the zonelist with | |
836 | * @zone - The first suitable zone found is returned via this parameter | |
837 | * | |
838 | * This function returns the next zone at or below a given zone index that is | |
839 | * within the allowed nodemask using a cursor as the starting point for the | |
5bead2a0 MG |
840 | * search. The zoneref returned is a cursor that represents the current zone |
841 | * being examined. It should be advanced by one before calling | |
842 | * next_zones_zonelist again. | |
19770b32 MG |
843 | */ |
844 | struct zoneref *next_zones_zonelist(struct zoneref *z, | |
845 | enum zone_type highest_zoneidx, | |
846 | nodemask_t *nodes, | |
847 | struct zone **zone); | |
dd1a239f | 848 | |
19770b32 MG |
849 | /** |
850 | * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist | |
851 | * @zonelist - The zonelist to search for a suitable zone | |
852 | * @highest_zoneidx - The zone index of the highest zone to return | |
853 | * @nodes - An optional nodemask to filter the zonelist with | |
854 | * @zone - The first suitable zone found is returned via this parameter | |
855 | * | |
856 | * This function returns the first zone at or below a given zone index that is | |
857 | * within the allowed nodemask. The zoneref returned is a cursor that can be | |
5bead2a0 MG |
858 | * used to iterate the zonelist with next_zones_zonelist by advancing it by |
859 | * one before calling. | |
19770b32 | 860 | */ |
dd1a239f | 861 | static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist, |
19770b32 MG |
862 | enum zone_type highest_zoneidx, |
863 | nodemask_t *nodes, | |
864 | struct zone **zone) | |
54a6eb5c | 865 | { |
19770b32 MG |
866 | return next_zones_zonelist(zonelist->_zonerefs, highest_zoneidx, nodes, |
867 | zone); | |
54a6eb5c MG |
868 | } |
869 | ||
19770b32 MG |
870 | /** |
871 | * 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 | |
872 | * @zone - The current zone in the iterator | |
873 | * @z - The current pointer within zonelist->zones being iterated | |
874 | * @zlist - The zonelist being iterated | |
875 | * @highidx - The zone index of the highest zone to return | |
876 | * @nodemask - Nodemask allowed by the allocator | |
877 | * | |
878 | * This iterator iterates though all zones at or below a given zone index and | |
879 | * within a given nodemask | |
880 | */ | |
881 | #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \ | |
882 | for (z = first_zones_zonelist(zlist, highidx, nodemask, &zone); \ | |
883 | zone; \ | |
5bead2a0 | 884 | z = next_zones_zonelist(++z, highidx, nodemask, &zone)) \ |
54a6eb5c MG |
885 | |
886 | /** | |
887 | * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index | |
888 | * @zone - The current zone in the iterator | |
889 | * @z - The current pointer within zonelist->zones being iterated | |
890 | * @zlist - The zonelist being iterated | |
891 | * @highidx - The zone index of the highest zone to return | |
892 | * | |
893 | * This iterator iterates though all zones at or below a given zone index. | |
894 | */ | |
895 | #define for_each_zone_zonelist(zone, z, zlist, highidx) \ | |
19770b32 | 896 | for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL) |
54a6eb5c | 897 | |
d41dee36 AW |
898 | #ifdef CONFIG_SPARSEMEM |
899 | #include <asm/sparsemem.h> | |
900 | #endif | |
901 | ||
c713216d MG |
902 | #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \ |
903 | !defined(CONFIG_ARCH_POPULATES_NODE_MAP) | |
b4544568 AM |
904 | static inline unsigned long early_pfn_to_nid(unsigned long pfn) |
905 | { | |
906 | return 0; | |
907 | } | |
b159d43f AW |
908 | #endif |
909 | ||
2bdaf115 AW |
910 | #ifdef CONFIG_FLATMEM |
911 | #define pfn_to_nid(pfn) (0) | |
912 | #endif | |
913 | ||
d41dee36 AW |
914 | #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT) |
915 | #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT) | |
916 | ||
917 | #ifdef CONFIG_SPARSEMEM | |
918 | ||
919 | /* | |
920 | * SECTION_SHIFT #bits space required to store a section # | |
921 | * | |
922 | * PA_SECTION_SHIFT physical address to/from section number | |
923 | * PFN_SECTION_SHIFT pfn to/from section number | |
924 | */ | |
925 | #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS) | |
926 | ||
927 | #define PA_SECTION_SHIFT (SECTION_SIZE_BITS) | |
928 | #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT) | |
929 | ||
930 | #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT) | |
931 | ||
932 | #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT) | |
933 | #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1)) | |
934 | ||
835c134e | 935 | #define SECTION_BLOCKFLAGS_BITS \ |
d9c23400 | 936 | ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS) |
835c134e | 937 | |
d41dee36 AW |
938 | #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS |
939 | #error Allocator MAX_ORDER exceeds SECTION_SIZE | |
940 | #endif | |
941 | ||
942 | struct page; | |
52d4b9ac | 943 | struct page_cgroup; |
d41dee36 | 944 | struct mem_section { |
29751f69 AW |
945 | /* |
946 | * This is, logically, a pointer to an array of struct | |
947 | * pages. However, it is stored with some other magic. | |
948 | * (see sparse.c::sparse_init_one_section()) | |
949 | * | |
30c253e6 AW |
950 | * Additionally during early boot we encode node id of |
951 | * the location of the section here to guide allocation. | |
952 | * (see sparse.c::memory_present()) | |
953 | * | |
29751f69 AW |
954 | * Making it a UL at least makes someone do a cast |
955 | * before using it wrong. | |
956 | */ | |
957 | unsigned long section_mem_map; | |
5c0e3066 MG |
958 | |
959 | /* See declaration of similar field in struct zone */ | |
960 | unsigned long *pageblock_flags; | |
52d4b9ac KH |
961 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
962 | /* | |
963 | * If !SPARSEMEM, pgdat doesn't have page_cgroup pointer. We use | |
964 | * section. (see memcontrol.h/page_cgroup.h about this.) | |
965 | */ | |
966 | struct page_cgroup *page_cgroup; | |
967 | unsigned long pad; | |
968 | #endif | |
d41dee36 AW |
969 | }; |
970 | ||
3e347261 BP |
971 | #ifdef CONFIG_SPARSEMEM_EXTREME |
972 | #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section)) | |
973 | #else | |
974 | #define SECTIONS_PER_ROOT 1 | |
975 | #endif | |
802f192e | 976 | |
3e347261 BP |
977 | #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT) |
978 | #define NR_SECTION_ROOTS (NR_MEM_SECTIONS / SECTIONS_PER_ROOT) | |
979 | #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1) | |
802f192e | 980 | |
3e347261 BP |
981 | #ifdef CONFIG_SPARSEMEM_EXTREME |
982 | extern struct mem_section *mem_section[NR_SECTION_ROOTS]; | |
802f192e | 983 | #else |
3e347261 BP |
984 | extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]; |
985 | #endif | |
d41dee36 | 986 | |
29751f69 AW |
987 | static inline struct mem_section *__nr_to_section(unsigned long nr) |
988 | { | |
3e347261 BP |
989 | if (!mem_section[SECTION_NR_TO_ROOT(nr)]) |
990 | return NULL; | |
991 | return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK]; | |
29751f69 | 992 | } |
4ca644d9 | 993 | extern int __section_nr(struct mem_section* ms); |
04753278 | 994 | extern unsigned long usemap_size(void); |
29751f69 AW |
995 | |
996 | /* | |
997 | * We use the lower bits of the mem_map pointer to store | |
998 | * a little bit of information. There should be at least | |
999 | * 3 bits here due to 32-bit alignment. | |
1000 | */ | |
1001 | #define SECTION_MARKED_PRESENT (1UL<<0) | |
1002 | #define SECTION_HAS_MEM_MAP (1UL<<1) | |
1003 | #define SECTION_MAP_LAST_BIT (1UL<<2) | |
1004 | #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1)) | |
30c253e6 | 1005 | #define SECTION_NID_SHIFT 2 |
29751f69 AW |
1006 | |
1007 | static inline struct page *__section_mem_map_addr(struct mem_section *section) | |
1008 | { | |
1009 | unsigned long map = section->section_mem_map; | |
1010 | map &= SECTION_MAP_MASK; | |
1011 | return (struct page *)map; | |
1012 | } | |
1013 | ||
540557b9 | 1014 | static inline int present_section(struct mem_section *section) |
29751f69 | 1015 | { |
802f192e | 1016 | return (section && (section->section_mem_map & SECTION_MARKED_PRESENT)); |
29751f69 AW |
1017 | } |
1018 | ||
540557b9 AW |
1019 | static inline int present_section_nr(unsigned long nr) |
1020 | { | |
1021 | return present_section(__nr_to_section(nr)); | |
1022 | } | |
1023 | ||
1024 | static inline int valid_section(struct mem_section *section) | |
29751f69 | 1025 | { |
802f192e | 1026 | return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP)); |
29751f69 AW |
1027 | } |
1028 | ||
1029 | static inline int valid_section_nr(unsigned long nr) | |
1030 | { | |
1031 | return valid_section(__nr_to_section(nr)); | |
1032 | } | |
1033 | ||
d41dee36 AW |
1034 | static inline struct mem_section *__pfn_to_section(unsigned long pfn) |
1035 | { | |
29751f69 | 1036 | return __nr_to_section(pfn_to_section_nr(pfn)); |
d41dee36 AW |
1037 | } |
1038 | ||
d41dee36 AW |
1039 | static inline int pfn_valid(unsigned long pfn) |
1040 | { | |
1041 | if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) | |
1042 | return 0; | |
29751f69 | 1043 | return valid_section(__nr_to_section(pfn_to_section_nr(pfn))); |
d41dee36 AW |
1044 | } |
1045 | ||
540557b9 AW |
1046 | static inline int pfn_present(unsigned long pfn) |
1047 | { | |
1048 | if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) | |
1049 | return 0; | |
1050 | return present_section(__nr_to_section(pfn_to_section_nr(pfn))); | |
1051 | } | |
1052 | ||
d41dee36 AW |
1053 | /* |
1054 | * These are _only_ used during initialisation, therefore they | |
1055 | * can use __initdata ... They could have names to indicate | |
1056 | * this restriction. | |
1057 | */ | |
1058 | #ifdef CONFIG_NUMA | |
161599ff AW |
1059 | #define pfn_to_nid(pfn) \ |
1060 | ({ \ | |
1061 | unsigned long __pfn_to_nid_pfn = (pfn); \ | |
1062 | page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \ | |
1063 | }) | |
2bdaf115 AW |
1064 | #else |
1065 | #define pfn_to_nid(pfn) (0) | |
d41dee36 AW |
1066 | #endif |
1067 | ||
d41dee36 AW |
1068 | #define early_pfn_valid(pfn) pfn_valid(pfn) |
1069 | void sparse_init(void); | |
1070 | #else | |
1071 | #define sparse_init() do {} while (0) | |
28ae55c9 | 1072 | #define sparse_index_init(_sec, _nid) do {} while (0) |
d41dee36 AW |
1073 | #endif /* CONFIG_SPARSEMEM */ |
1074 | ||
75167957 | 1075 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES |
cc2559bc | 1076 | bool early_pfn_in_nid(unsigned long pfn, int nid); |
75167957 AW |
1077 | #else |
1078 | #define early_pfn_in_nid(pfn, nid) (1) | |
1079 | #endif | |
1080 | ||
d41dee36 AW |
1081 | #ifndef early_pfn_valid |
1082 | #define early_pfn_valid(pfn) (1) | |
1083 | #endif | |
1084 | ||
1085 | void memory_present(int nid, unsigned long start, unsigned long end); | |
1086 | unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long); | |
1087 | ||
14e07298 AW |
1088 | /* |
1089 | * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we | |
1090 | * need to check pfn validility within that MAX_ORDER_NR_PAGES block. | |
1091 | * pfn_valid_within() should be used in this case; we optimise this away | |
1092 | * when we have no holes within a MAX_ORDER_NR_PAGES block. | |
1093 | */ | |
1094 | #ifdef CONFIG_HOLES_IN_ZONE | |
1095 | #define pfn_valid_within(pfn) pfn_valid(pfn) | |
1096 | #else | |
1097 | #define pfn_valid_within(pfn) (1) | |
1098 | #endif | |
1099 | ||
eb33575c MG |
1100 | #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL |
1101 | /* | |
1102 | * pfn_valid() is meant to be able to tell if a given PFN has valid memmap | |
1103 | * associated with it or not. In FLATMEM, it is expected that holes always | |
1104 | * have valid memmap as long as there is valid PFNs either side of the hole. | |
1105 | * In SPARSEMEM, it is assumed that a valid section has a memmap for the | |
1106 | * entire section. | |
1107 | * | |
1108 | * However, an ARM, and maybe other embedded architectures in the future | |
1109 | * free memmap backing holes to save memory on the assumption the memmap is | |
1110 | * never used. The page_zone linkages are then broken even though pfn_valid() | |
1111 | * returns true. A walker of the full memmap must then do this additional | |
1112 | * check to ensure the memmap they are looking at is sane by making sure | |
1113 | * the zone and PFN linkages are still valid. This is expensive, but walkers | |
1114 | * of the full memmap are extremely rare. | |
1115 | */ | |
1116 | int memmap_valid_within(unsigned long pfn, | |
1117 | struct page *page, struct zone *zone); | |
1118 | #else | |
1119 | static inline int memmap_valid_within(unsigned long pfn, | |
1120 | struct page *page, struct zone *zone) | |
1121 | { | |
1122 | return 1; | |
1123 | } | |
1124 | #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */ | |
1125 | ||
97965478 | 1126 | #endif /* !__GENERATING_BOUNDS.H */ |
1da177e4 | 1127 | #endif /* !__ASSEMBLY__ */ |
1da177e4 | 1128 | #endif /* _LINUX_MMZONE_H */ |