]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | #ifndef _LINUX_MMZONE_H |
2 | #define _LINUX_MMZONE_H | |
3 | ||
1da177e4 | 4 | #ifndef __ASSEMBLY__ |
97965478 | 5 | #ifndef __GENERATING_BOUNDS_H |
1da177e4 | 6 | |
1da177e4 LT |
7 | #include <linux/spinlock.h> |
8 | #include <linux/list.h> | |
9 | #include <linux/wait.h> | |
e815af95 | 10 | #include <linux/bitops.h> |
1da177e4 LT |
11 | #include <linux/cache.h> |
12 | #include <linux/threads.h> | |
13 | #include <linux/numa.h> | |
14 | #include <linux/init.h> | |
bdc8cb98 | 15 | #include <linux/seqlock.h> |
8357f869 | 16 | #include <linux/nodemask.h> |
835c134e | 17 | #include <linux/pageblock-flags.h> |
bbeae5b0 | 18 | #include <linux/page-flags-layout.h> |
60063497 | 19 | #include <linux/atomic.h> |
93ff66bf | 20 | #include <asm/page.h> |
1da177e4 LT |
21 | |
22 | /* Free memory management - zoned buddy allocator. */ | |
23 | #ifndef CONFIG_FORCE_MAX_ZONEORDER | |
24 | #define MAX_ORDER 11 | |
25 | #else | |
26 | #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER | |
27 | #endif | |
e984bb43 | 28 | #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1)) |
1da177e4 | 29 | |
5ad333eb AW |
30 | /* |
31 | * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed | |
32 | * costly to service. That is between allocation orders which should | |
35fca53e | 33 | * coalesce naturally under reasonable reclaim pressure and those which |
5ad333eb AW |
34 | * will not. |
35 | */ | |
36 | #define PAGE_ALLOC_COSTLY_ORDER 3 | |
37 | ||
47118af0 MN |
38 | enum { |
39 | MIGRATE_UNMOVABLE, | |
47118af0 | 40 | MIGRATE_MOVABLE, |
016c13da | 41 | MIGRATE_RECLAIMABLE, |
0aaa29a5 MG |
42 | MIGRATE_PCPTYPES, /* the number of types on the pcp lists */ |
43 | MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES, | |
47118af0 MN |
44 | #ifdef CONFIG_CMA |
45 | /* | |
46 | * MIGRATE_CMA migration type is designed to mimic the way | |
47 | * ZONE_MOVABLE works. Only movable pages can be allocated | |
48 | * from MIGRATE_CMA pageblocks and page allocator never | |
49 | * implicitly change migration type of MIGRATE_CMA pageblock. | |
50 | * | |
51 | * The way to use it is to change migratetype of a range of | |
52 | * pageblocks to MIGRATE_CMA which can be done by | |
53 | * __free_pageblock_cma() function. What is important though | |
54 | * is that a range of pageblocks must be aligned to | |
55 | * MAX_ORDER_NR_PAGES should biggest page be bigger then | |
56 | * a single pageblock. | |
57 | */ | |
58 | MIGRATE_CMA, | |
59 | #endif | |
194159fb | 60 | #ifdef CONFIG_MEMORY_ISOLATION |
47118af0 | 61 | MIGRATE_ISOLATE, /* can't allocate from here */ |
194159fb | 62 | #endif |
47118af0 MN |
63 | MIGRATE_TYPES |
64 | }; | |
65 | ||
60f30350 VB |
66 | /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */ |
67 | extern char * const migratetype_names[MIGRATE_TYPES]; | |
68 | ||
47118af0 MN |
69 | #ifdef CONFIG_CMA |
70 | # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA) | |
7c15d9bb | 71 | # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA) |
47118af0 MN |
72 | #else |
73 | # define is_migrate_cma(migratetype) false | |
7c15d9bb | 74 | # define is_migrate_cma_page(_page) false |
47118af0 | 75 | #endif |
b2a0ac88 MG |
76 | |
77 | #define for_each_migratetype_order(order, type) \ | |
78 | for (order = 0; order < MAX_ORDER; order++) \ | |
79 | for (type = 0; type < MIGRATE_TYPES; type++) | |
80 | ||
467c996c MG |
81 | extern int page_group_by_mobility_disabled; |
82 | ||
e58469ba MG |
83 | #define NR_MIGRATETYPE_BITS (PB_migrate_end - PB_migrate + 1) |
84 | #define MIGRATETYPE_MASK ((1UL << NR_MIGRATETYPE_BITS) - 1) | |
85 | ||
dc4b0caf MG |
86 | #define get_pageblock_migratetype(page) \ |
87 | get_pfnblock_flags_mask(page, page_to_pfn(page), \ | |
88 | PB_migrate_end, MIGRATETYPE_MASK) | |
89 | ||
1da177e4 | 90 | struct free_area { |
b2a0ac88 | 91 | struct list_head free_list[MIGRATE_TYPES]; |
1da177e4 LT |
92 | unsigned long nr_free; |
93 | }; | |
94 | ||
95 | struct pglist_data; | |
96 | ||
97 | /* | |
a52633d8 | 98 | * zone->lock and the zone lru_lock are two of the hottest locks in the kernel. |
1da177e4 LT |
99 | * So add a wild amount of padding here to ensure that they fall into separate |
100 | * cachelines. There are very few zone structures in the machine, so space | |
101 | * consumption is not a concern here. | |
102 | */ | |
103 | #if defined(CONFIG_SMP) | |
104 | struct zone_padding { | |
105 | char x[0]; | |
22fc6ecc | 106 | } ____cacheline_internodealigned_in_smp; |
1da177e4 LT |
107 | #define ZONE_PADDING(name) struct zone_padding name; |
108 | #else | |
109 | #define ZONE_PADDING(name) | |
110 | #endif | |
111 | ||
2244b95a | 112 | enum zone_stat_item { |
51ed4491 | 113 | /* First 128 byte cacheline (assuming 64 bit words) */ |
d23ad423 | 114 | NR_FREE_PAGES, |
71c799f4 MK |
115 | NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */ |
116 | NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE, | |
117 | NR_ZONE_ACTIVE_ANON, | |
118 | NR_ZONE_INACTIVE_FILE, | |
119 | NR_ZONE_ACTIVE_FILE, | |
120 | NR_ZONE_UNEVICTABLE, | |
5a1c84b4 | 121 | NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */ |
5344b7e6 | 122 | NR_MLOCK, /* mlock()ed pages found and moved off LRU */ |
51ed4491 CL |
123 | NR_SLAB_RECLAIMABLE, |
124 | NR_SLAB_UNRECLAIMABLE, | |
125 | NR_PAGETABLE, /* used for pagetables */ | |
d30dd8be | 126 | NR_KERNEL_STACK_KB, /* measured in KiB */ |
c6a7f572 | 127 | /* Second 128 byte cacheline */ |
d2c5e30c | 128 | NR_BOUNCE, |
91537fee MK |
129 | #if IS_ENABLED(CONFIG_ZSMALLOC) |
130 | NR_ZSPAGES, /* allocated in zsmalloc */ | |
131 | #endif | |
ca889e6c CL |
132 | #ifdef CONFIG_NUMA |
133 | NUMA_HIT, /* allocated in intended node */ | |
134 | NUMA_MISS, /* allocated in non intended node */ | |
135 | NUMA_FOREIGN, /* was intended here, hit elsewhere */ | |
136 | NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */ | |
137 | NUMA_LOCAL, /* allocation from local node */ | |
138 | NUMA_OTHER, /* allocation from other node */ | |
139 | #endif | |
d1ce749a | 140 | NR_FREE_CMA_PAGES, |
2244b95a CL |
141 | NR_VM_ZONE_STAT_ITEMS }; |
142 | ||
75ef7184 | 143 | enum node_stat_item { |
599d0c95 MG |
144 | NR_LRU_BASE, |
145 | NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */ | |
146 | NR_ACTIVE_ANON, /* " " " " " */ | |
147 | NR_INACTIVE_FILE, /* " " " " " */ | |
148 | NR_ACTIVE_FILE, /* " " " " " */ | |
149 | NR_UNEVICTABLE, /* " " " " " */ | |
150 | NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */ | |
151 | NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */ | |
152 | NR_PAGES_SCANNED, /* pages scanned since last reclaim */ | |
1e6b1085 MG |
153 | WORKINGSET_REFAULT, |
154 | WORKINGSET_ACTIVATE, | |
155 | WORKINGSET_NODERECLAIM, | |
4b9d0fab | 156 | NR_ANON_MAPPED, /* Mapped anonymous pages */ |
50658e2e MG |
157 | NR_FILE_MAPPED, /* pagecache pages mapped into pagetables. |
158 | only modified from process context */ | |
11fb9989 MG |
159 | NR_FILE_PAGES, |
160 | NR_FILE_DIRTY, | |
161 | NR_WRITEBACK, | |
162 | NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */ | |
163 | NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */ | |
164 | NR_SHMEM_THPS, | |
165 | NR_SHMEM_PMDMAPPED, | |
166 | NR_ANON_THPS, | |
167 | NR_UNSTABLE_NFS, /* NFS unstable pages */ | |
c4a25635 MG |
168 | NR_VMSCAN_WRITE, |
169 | NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */ | |
170 | NR_DIRTIED, /* page dirtyings since bootup */ | |
171 | NR_WRITTEN, /* page writings since bootup */ | |
75ef7184 MG |
172 | NR_VM_NODE_STAT_ITEMS |
173 | }; | |
174 | ||
4f98a2fe RR |
175 | /* |
176 | * We do arithmetic on the LRU lists in various places in the code, | |
177 | * so it is important to keep the active lists LRU_ACTIVE higher in | |
178 | * the array than the corresponding inactive lists, and to keep | |
179 | * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists. | |
180 | * | |
181 | * This has to be kept in sync with the statistics in zone_stat_item | |
182 | * above and the descriptions in vmstat_text in mm/vmstat.c | |
183 | */ | |
184 | #define LRU_BASE 0 | |
185 | #define LRU_ACTIVE 1 | |
186 | #define LRU_FILE 2 | |
187 | ||
b69408e8 | 188 | enum lru_list { |
4f98a2fe RR |
189 | LRU_INACTIVE_ANON = LRU_BASE, |
190 | LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE, | |
191 | LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE, | |
192 | LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE, | |
894bc310 | 193 | LRU_UNEVICTABLE, |
894bc310 LS |
194 | NR_LRU_LISTS |
195 | }; | |
b69408e8 | 196 | |
4111304d | 197 | #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++) |
b69408e8 | 198 | |
4111304d | 199 | #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++) |
894bc310 | 200 | |
4111304d | 201 | static inline int is_file_lru(enum lru_list lru) |
4f98a2fe | 202 | { |
4111304d | 203 | return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE); |
4f98a2fe RR |
204 | } |
205 | ||
4111304d | 206 | static inline int is_active_lru(enum lru_list lru) |
b69408e8 | 207 | { |
4111304d | 208 | return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE); |
b69408e8 CL |
209 | } |
210 | ||
89abfab1 HD |
211 | struct zone_reclaim_stat { |
212 | /* | |
213 | * The pageout code in vmscan.c keeps track of how many of the | |
59f91e5d | 214 | * mem/swap backed and file backed pages are referenced. |
89abfab1 HD |
215 | * The higher the rotated/scanned ratio, the more valuable |
216 | * that cache is. | |
217 | * | |
218 | * The anon LRU stats live in [0], file LRU stats in [1] | |
219 | */ | |
220 | unsigned long recent_rotated[2]; | |
221 | unsigned long recent_scanned[2]; | |
222 | }; | |
223 | ||
6290df54 | 224 | struct lruvec { |
23047a96 JW |
225 | struct list_head lists[NR_LRU_LISTS]; |
226 | struct zone_reclaim_stat reclaim_stat; | |
227 | /* Evictions & activations on the inactive file list */ | |
228 | atomic_long_t inactive_age; | |
c255a458 | 229 | #ifdef CONFIG_MEMCG |
599d0c95 | 230 | struct pglist_data *pgdat; |
7f5e86c2 | 231 | #endif |
6290df54 JW |
232 | }; |
233 | ||
bb2a0de9 KH |
234 | /* Mask used at gathering information at once (see memcontrol.c) */ |
235 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) | |
236 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) | |
bb2a0de9 KH |
237 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) |
238 | ||
39deaf85 | 239 | /* Isolate clean file */ |
f3fd4a61 | 240 | #define ISOLATE_CLEAN ((__force isolate_mode_t)0x1) |
f80c0673 | 241 | /* Isolate unmapped file */ |
f3fd4a61 | 242 | #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2) |
c8244935 | 243 | /* Isolate for asynchronous migration */ |
f3fd4a61 | 244 | #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4) |
e46a2879 MK |
245 | /* Isolate unevictable pages */ |
246 | #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8) | |
4356f21d MK |
247 | |
248 | /* LRU Isolation modes. */ | |
249 | typedef unsigned __bitwise__ isolate_mode_t; | |
250 | ||
41858966 MG |
251 | enum zone_watermarks { |
252 | WMARK_MIN, | |
253 | WMARK_LOW, | |
254 | WMARK_HIGH, | |
255 | NR_WMARK | |
256 | }; | |
257 | ||
258 | #define min_wmark_pages(z) (z->watermark[WMARK_MIN]) | |
259 | #define low_wmark_pages(z) (z->watermark[WMARK_LOW]) | |
260 | #define high_wmark_pages(z) (z->watermark[WMARK_HIGH]) | |
261 | ||
1da177e4 LT |
262 | struct per_cpu_pages { |
263 | int count; /* number of pages in the list */ | |
1da177e4 LT |
264 | int high; /* high watermark, emptying needed */ |
265 | int batch; /* chunk size for buddy add/remove */ | |
5f8dcc21 MG |
266 | |
267 | /* Lists of pages, one per migrate type stored on the pcp-lists */ | |
268 | struct list_head lists[MIGRATE_PCPTYPES]; | |
1da177e4 LT |
269 | }; |
270 | ||
271 | struct per_cpu_pageset { | |
3dfa5721 | 272 | struct per_cpu_pages pcp; |
4037d452 CL |
273 | #ifdef CONFIG_NUMA |
274 | s8 expire; | |
275 | #endif | |
2244b95a | 276 | #ifdef CONFIG_SMP |
df9ecaba | 277 | s8 stat_threshold; |
2244b95a CL |
278 | s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS]; |
279 | #endif | |
99dcc3e5 | 280 | }; |
e7c8d5c9 | 281 | |
75ef7184 MG |
282 | struct per_cpu_nodestat { |
283 | s8 stat_threshold; | |
284 | s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS]; | |
285 | }; | |
286 | ||
97965478 CL |
287 | #endif /* !__GENERATING_BOUNDS.H */ |
288 | ||
2f1b6248 | 289 | enum zone_type { |
4b51d669 | 290 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 CL |
291 | /* |
292 | * ZONE_DMA is used when there are devices that are not able | |
293 | * to do DMA to all of addressable memory (ZONE_NORMAL). Then we | |
294 | * carve out the portion of memory that is needed for these devices. | |
295 | * The range is arch specific. | |
296 | * | |
297 | * Some examples | |
298 | * | |
299 | * Architecture Limit | |
300 | * --------------------------- | |
301 | * parisc, ia64, sparc <4G | |
302 | * s390 <2G | |
2f1b6248 CL |
303 | * arm Various |
304 | * alpha Unlimited or 0-16MB. | |
305 | * | |
306 | * i386, x86_64 and multiple other arches | |
307 | * <16M. | |
308 | */ | |
309 | ZONE_DMA, | |
4b51d669 | 310 | #endif |
fb0e7942 | 311 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 CL |
312 | /* |
313 | * x86_64 needs two ZONE_DMAs because it supports devices that are | |
314 | * only able to do DMA to the lower 16M but also 32 bit devices that | |
315 | * can only do DMA areas below 4G. | |
316 | */ | |
317 | ZONE_DMA32, | |
fb0e7942 | 318 | #endif |
2f1b6248 CL |
319 | /* |
320 | * Normal addressable memory is in ZONE_NORMAL. DMA operations can be | |
321 | * performed on pages in ZONE_NORMAL if the DMA devices support | |
322 | * transfers to all addressable memory. | |
323 | */ | |
324 | ZONE_NORMAL, | |
e53ef38d | 325 | #ifdef CONFIG_HIGHMEM |
2f1b6248 CL |
326 | /* |
327 | * A memory area that is only addressable by the kernel through | |
328 | * mapping portions into its own address space. This is for example | |
329 | * used by i386 to allow the kernel to address the memory beyond | |
330 | * 900MB. The kernel will set up special mappings (page | |
331 | * table entries on i386) for each page that the kernel needs to | |
332 | * access. | |
333 | */ | |
334 | ZONE_HIGHMEM, | |
e53ef38d | 335 | #endif |
2a1e274a | 336 | ZONE_MOVABLE, |
033fbae9 DW |
337 | #ifdef CONFIG_ZONE_DEVICE |
338 | ZONE_DEVICE, | |
339 | #endif | |
97965478 | 340 | __MAX_NR_ZONES |
033fbae9 | 341 | |
2f1b6248 | 342 | }; |
1da177e4 | 343 | |
97965478 CL |
344 | #ifndef __GENERATING_BOUNDS_H |
345 | ||
1da177e4 | 346 | struct zone { |
3484b2de | 347 | /* Read-mostly fields */ |
41858966 MG |
348 | |
349 | /* zone watermarks, access with *_wmark_pages(zone) macros */ | |
350 | unsigned long watermark[NR_WMARK]; | |
351 | ||
0aaa29a5 MG |
352 | unsigned long nr_reserved_highatomic; |
353 | ||
1da177e4 | 354 | /* |
89903327 AM |
355 | * We don't know if the memory that we're going to allocate will be |
356 | * freeable or/and it will be released eventually, so to avoid totally | |
357 | * wasting several GB of ram we must reserve some of the lower zone | |
358 | * memory (otherwise we risk to run OOM on the lower zones despite | |
359 | * there being tons of freeable ram on the higher zones). This array is | |
360 | * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl | |
361 | * changes. | |
1da177e4 | 362 | */ |
3484b2de | 363 | long lowmem_reserve[MAX_NR_ZONES]; |
ab8fabd4 | 364 | |
e7c8d5c9 | 365 | #ifdef CONFIG_NUMA |
d5f541ed | 366 | int node; |
3484b2de | 367 | #endif |
3484b2de | 368 | struct pglist_data *zone_pgdat; |
43cf38eb | 369 | struct per_cpu_pageset __percpu *pageset; |
3484b2de | 370 | |
835c134e MG |
371 | #ifndef CONFIG_SPARSEMEM |
372 | /* | |
d9c23400 | 373 | * Flags for a pageblock_nr_pages block. See pageblock-flags.h. |
835c134e MG |
374 | * In SPARSEMEM, this map is stored in struct mem_section |
375 | */ | |
376 | unsigned long *pageblock_flags; | |
377 | #endif /* CONFIG_SPARSEMEM */ | |
378 | ||
1da177e4 LT |
379 | /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */ |
380 | unsigned long zone_start_pfn; | |
381 | ||
bdc8cb98 | 382 | /* |
9feedc9d JL |
383 | * spanned_pages is the total pages spanned by the zone, including |
384 | * holes, which is calculated as: | |
385 | * spanned_pages = zone_end_pfn - zone_start_pfn; | |
bdc8cb98 | 386 | * |
9feedc9d JL |
387 | * present_pages is physical pages existing within the zone, which |
388 | * is calculated as: | |
8761e31c | 389 | * present_pages = spanned_pages - absent_pages(pages in holes); |
9feedc9d JL |
390 | * |
391 | * managed_pages is present pages managed by the buddy system, which | |
392 | * is calculated as (reserved_pages includes pages allocated by the | |
393 | * bootmem allocator): | |
394 | * managed_pages = present_pages - reserved_pages; | |
395 | * | |
396 | * So present_pages may be used by memory hotplug or memory power | |
397 | * management logic to figure out unmanaged pages by checking | |
398 | * (present_pages - managed_pages). And managed_pages should be used | |
399 | * by page allocator and vm scanner to calculate all kinds of watermarks | |
400 | * and thresholds. | |
401 | * | |
402 | * Locking rules: | |
403 | * | |
404 | * zone_start_pfn and spanned_pages are protected by span_seqlock. | |
405 | * It is a seqlock because it has to be read outside of zone->lock, | |
406 | * and it is done in the main allocator path. But, it is written | |
407 | * quite infrequently. | |
408 | * | |
409 | * The span_seq lock is declared along with zone->lock because it is | |
bdc8cb98 DH |
410 | * frequently read in proximity to zone->lock. It's good to |
411 | * give them a chance of being in the same cacheline. | |
9feedc9d | 412 | * |
c3d5f5f0 | 413 | * Write access to present_pages at runtime should be protected by |
bfc8c901 VD |
414 | * mem_hotplug_begin/end(). Any reader who can't tolerant drift of |
415 | * present_pages should get_online_mems() to get a stable value. | |
c3d5f5f0 JL |
416 | * |
417 | * Read access to managed_pages should be safe because it's unsigned | |
418 | * long. Write access to zone->managed_pages and totalram_pages are | |
419 | * protected by managed_page_count_lock at runtime. Idealy only | |
420 | * adjust_managed_page_count() should be used instead of directly | |
421 | * touching zone->managed_pages and totalram_pages. | |
bdc8cb98 | 422 | */ |
3484b2de | 423 | unsigned long managed_pages; |
9feedc9d JL |
424 | unsigned long spanned_pages; |
425 | unsigned long present_pages; | |
3484b2de MG |
426 | |
427 | const char *name; | |
1da177e4 | 428 | |
ad53f92e JK |
429 | #ifdef CONFIG_MEMORY_ISOLATION |
430 | /* | |
431 | * Number of isolated pageblock. It is used to solve incorrect | |
432 | * freepage counting problem due to racy retrieving migratetype | |
433 | * of pageblock. Protected by zone->lock. | |
434 | */ | |
435 | unsigned long nr_isolate_pageblock; | |
436 | #endif | |
437 | ||
3484b2de MG |
438 | #ifdef CONFIG_MEMORY_HOTPLUG |
439 | /* see spanned/present_pages for more description */ | |
440 | seqlock_t span_seqlock; | |
441 | #endif | |
442 | ||
1da177e4 | 443 | /* |
3484b2de MG |
444 | * wait_table -- the array holding the hash table |
445 | * wait_table_hash_nr_entries -- the size of the hash table array | |
446 | * wait_table_bits -- wait_table_size == (1 << wait_table_bits) | |
447 | * | |
448 | * The purpose of all these is to keep track of the people | |
449 | * waiting for a page to become available and make them | |
450 | * runnable again when possible. The trouble is that this | |
451 | * consumes a lot of space, especially when so few things | |
452 | * wait on pages at a given time. So instead of using | |
453 | * per-page waitqueues, we use a waitqueue hash table. | |
454 | * | |
455 | * The bucket discipline is to sleep on the same queue when | |
456 | * colliding and wake all in that wait queue when removing. | |
457 | * When something wakes, it must check to be sure its page is | |
458 | * truly available, a la thundering herd. The cost of a | |
459 | * collision is great, but given the expected load of the | |
460 | * table, they should be so rare as to be outweighed by the | |
461 | * benefits from the saved space. | |
462 | * | |
463 | * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the | |
464 | * primary users of these fields, and in mm/page_alloc.c | |
465 | * free_area_init_core() performs the initialization of them. | |
1da177e4 | 466 | */ |
3484b2de MG |
467 | wait_queue_head_t *wait_table; |
468 | unsigned long wait_table_hash_nr_entries; | |
469 | unsigned long wait_table_bits; | |
470 | ||
0f661148 | 471 | /* Write-intensive fields used from the page allocator */ |
3484b2de | 472 | ZONE_PADDING(_pad1_) |
0f661148 | 473 | |
3484b2de MG |
474 | /* free areas of different sizes */ |
475 | struct free_area free_area[MAX_ORDER]; | |
476 | ||
477 | /* zone flags, see below */ | |
478 | unsigned long flags; | |
479 | ||
0f661148 | 480 | /* Primarily protects free_area */ |
a368ab67 MG |
481 | spinlock_t lock; |
482 | ||
0f661148 | 483 | /* Write-intensive fields used by compaction and vmstats. */ |
3484b2de MG |
484 | ZONE_PADDING(_pad2_) |
485 | ||
3484b2de MG |
486 | /* |
487 | * When free pages are below this point, additional steps are taken | |
488 | * when reading the number of free pages to avoid per-cpu counter | |
489 | * drift allowing watermarks to be breached | |
490 | */ | |
491 | unsigned long percpu_drift_mark; | |
492 | ||
493 | #if defined CONFIG_COMPACTION || defined CONFIG_CMA | |
494 | /* pfn where compaction free scanner should start */ | |
495 | unsigned long compact_cached_free_pfn; | |
496 | /* pfn where async and sync compaction migration scanner should start */ | |
497 | unsigned long compact_cached_migrate_pfn[2]; | |
498 | #endif | |
499 | ||
500 | #ifdef CONFIG_COMPACTION | |
501 | /* | |
502 | * On compaction failure, 1<<compact_defer_shift compactions | |
503 | * are skipped before trying again. The number attempted since | |
504 | * last failure is tracked with compact_considered. | |
505 | */ | |
506 | unsigned int compact_considered; | |
507 | unsigned int compact_defer_shift; | |
508 | int compact_order_failed; | |
509 | #endif | |
510 | ||
511 | #if defined CONFIG_COMPACTION || defined CONFIG_CMA | |
512 | /* Set to true when the PG_migrate_skip bits should be cleared */ | |
513 | bool compact_blockskip_flush; | |
514 | #endif | |
515 | ||
7cf91a98 JK |
516 | bool contiguous; |
517 | ||
3484b2de MG |
518 | ZONE_PADDING(_pad3_) |
519 | /* Zone statistics */ | |
520 | atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS]; | |
22fc6ecc | 521 | } ____cacheline_internodealigned_in_smp; |
1da177e4 | 522 | |
599d0c95 MG |
523 | enum pgdat_flags { |
524 | PGDAT_CONGESTED, /* pgdat has many dirty pages backed by | |
0e093d99 MG |
525 | * a congested BDI |
526 | */ | |
599d0c95 | 527 | PGDAT_DIRTY, /* reclaim scanning has recently found |
d43006d5 MG |
528 | * many dirty file pages at the tail |
529 | * of the LRU. | |
530 | */ | |
599d0c95 | 531 | PGDAT_WRITEBACK, /* reclaim scanning has recently found |
283aba9f MG |
532 | * many pages under writeback |
533 | */ | |
a5f5f91d | 534 | PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */ |
57054651 | 535 | }; |
e815af95 | 536 | |
f9228b20 | 537 | static inline unsigned long zone_end_pfn(const struct zone *zone) |
108bcc96 CS |
538 | { |
539 | return zone->zone_start_pfn + zone->spanned_pages; | |
540 | } | |
541 | ||
542 | static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn) | |
543 | { | |
544 | return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone); | |
545 | } | |
546 | ||
2a6e3ebe CS |
547 | static inline bool zone_is_initialized(struct zone *zone) |
548 | { | |
549 | return !!zone->wait_table; | |
550 | } | |
551 | ||
552 | static inline bool zone_is_empty(struct zone *zone) | |
553 | { | |
554 | return zone->spanned_pages == 0; | |
555 | } | |
556 | ||
1da177e4 LT |
557 | /* |
558 | * The "priority" of VM scanning is how much of the queues we will scan in one | |
559 | * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the | |
560 | * queues ("queue_length >> 12") during an aging round. | |
561 | */ | |
562 | #define DEF_PRIORITY 12 | |
563 | ||
9276b1bc PJ |
564 | /* Maximum number of zones on a zonelist */ |
565 | #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES) | |
566 | ||
c00eb15a YB |
567 | enum { |
568 | ZONELIST_FALLBACK, /* zonelist with fallback */ | |
9276b1bc | 569 | #ifdef CONFIG_NUMA |
c00eb15a YB |
570 | /* |
571 | * The NUMA zonelists are doubled because we need zonelists that | |
572 | * restrict the allocations to a single node for __GFP_THISNODE. | |
573 | */ | |
574 | ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */ | |
9276b1bc | 575 | #endif |
c00eb15a YB |
576 | MAX_ZONELISTS |
577 | }; | |
9276b1bc | 578 | |
dd1a239f MG |
579 | /* |
580 | * This struct contains information about a zone in a zonelist. It is stored | |
581 | * here to avoid dereferences into large structures and lookups of tables | |
582 | */ | |
583 | struct zoneref { | |
584 | struct zone *zone; /* Pointer to actual zone */ | |
585 | int zone_idx; /* zone_idx(zoneref->zone) */ | |
586 | }; | |
587 | ||
1da177e4 LT |
588 | /* |
589 | * One allocation request operates on a zonelist. A zonelist | |
590 | * is a list of zones, the first one is the 'goal' of the | |
591 | * allocation, the other zones are fallback zones, in decreasing | |
592 | * priority. | |
593 | * | |
dd1a239f MG |
594 | * To speed the reading of the zonelist, the zonerefs contain the zone index |
595 | * of the entry being read. Helper functions to access information given | |
596 | * a struct zoneref are | |
597 | * | |
598 | * zonelist_zone() - Return the struct zone * for an entry in _zonerefs | |
599 | * zonelist_zone_idx() - Return the index of the zone for an entry | |
600 | * zonelist_node_idx() - Return the index of the node for an entry | |
1da177e4 LT |
601 | */ |
602 | struct zonelist { | |
dd1a239f | 603 | struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1]; |
1da177e4 LT |
604 | }; |
605 | ||
5b99cd0e HC |
606 | #ifndef CONFIG_DISCONTIGMEM |
607 | /* The array of struct pages - for discontigmem use pgdat->lmem_map */ | |
608 | extern struct page *mem_map; | |
609 | #endif | |
610 | ||
1da177e4 LT |
611 | /* |
612 | * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM | |
613 | * (mostly NUMA machines?) to denote a higher-level memory zone than the | |
614 | * zone denotes. | |
615 | * | |
616 | * On NUMA machines, each NUMA node would have a pg_data_t to describe | |
617 | * it's memory layout. | |
618 | * | |
619 | * Memory statistics and page replacement data structures are maintained on a | |
620 | * per-zone basis. | |
621 | */ | |
622 | struct bootmem_data; | |
623 | typedef struct pglist_data { | |
624 | struct zone node_zones[MAX_NR_ZONES]; | |
523b9458 | 625 | struct zonelist node_zonelists[MAX_ZONELISTS]; |
1da177e4 | 626 | int nr_zones; |
52d4b9ac | 627 | #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */ |
1da177e4 | 628 | struct page *node_mem_map; |
eefa864b JK |
629 | #ifdef CONFIG_PAGE_EXTENSION |
630 | struct page_ext *node_page_ext; | |
631 | #endif | |
d41dee36 | 632 | #endif |
08677214 | 633 | #ifndef CONFIG_NO_BOOTMEM |
1da177e4 | 634 | struct bootmem_data *bdata; |
08677214 | 635 | #endif |
208d54e5 DH |
636 | #ifdef CONFIG_MEMORY_HOTPLUG |
637 | /* | |
638 | * Must be held any time you expect node_start_pfn, node_present_pages | |
639 | * or node_spanned_pages stay constant. Holding this will also | |
640 | * guarantee that any pfn_valid() stays that way. | |
641 | * | |
114d4b79 CS |
642 | * pgdat_resize_lock() and pgdat_resize_unlock() are provided to |
643 | * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG. | |
644 | * | |
72c3b51b | 645 | * Nests above zone->lock and zone->span_seqlock |
208d54e5 DH |
646 | */ |
647 | spinlock_t node_size_lock; | |
648 | #endif | |
1da177e4 LT |
649 | unsigned long node_start_pfn; |
650 | unsigned long node_present_pages; /* total number of physical pages */ | |
651 | unsigned long node_spanned_pages; /* total size of physical page | |
652 | range, including holes */ | |
653 | int node_id; | |
1da177e4 | 654 | wait_queue_head_t kswapd_wait; |
5515061d | 655 | wait_queue_head_t pfmemalloc_wait; |
bfc8c901 VD |
656 | struct task_struct *kswapd; /* Protected by |
657 | mem_hotplug_begin/end() */ | |
38087d9b MG |
658 | int kswapd_order; |
659 | enum zone_type kswapd_classzone_idx; | |
660 | ||
698b1b30 VB |
661 | #ifdef CONFIG_COMPACTION |
662 | int kcompactd_max_order; | |
663 | enum zone_type kcompactd_classzone_idx; | |
664 | wait_queue_head_t kcompactd_wait; | |
665 | struct task_struct *kcompactd; | |
666 | #endif | |
8177a420 | 667 | #ifdef CONFIG_NUMA_BALANCING |
1c5e9c27 | 668 | /* Lock serializing the migrate rate limiting window */ |
8177a420 AA |
669 | spinlock_t numabalancing_migrate_lock; |
670 | ||
671 | /* Rate limiting time interval */ | |
672 | unsigned long numabalancing_migrate_next_window; | |
673 | ||
674 | /* Number of pages migrated during the rate limiting time interval */ | |
675 | unsigned long numabalancing_migrate_nr_pages; | |
676 | #endif | |
281e3726 MG |
677 | /* |
678 | * This is a per-node reserve of pages that are not available | |
679 | * to userspace allocations. | |
680 | */ | |
681 | unsigned long totalreserve_pages; | |
682 | ||
a5f5f91d MG |
683 | #ifdef CONFIG_NUMA |
684 | /* | |
685 | * zone reclaim becomes active if more unmapped pages exist. | |
686 | */ | |
687 | unsigned long min_unmapped_pages; | |
688 | unsigned long min_slab_pages; | |
689 | #endif /* CONFIG_NUMA */ | |
690 | ||
a52633d8 MG |
691 | /* Write-intensive fields used by page reclaim */ |
692 | ZONE_PADDING(_pad1_) | |
693 | spinlock_t lru_lock; | |
3a80a7fa MG |
694 | |
695 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
696 | /* | |
697 | * If memory initialisation on large machines is deferred then this | |
698 | * is the first PFN that needs to be initialised. | |
699 | */ | |
700 | unsigned long first_deferred_pfn; | |
701 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ | |
a3d0a918 KS |
702 | |
703 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
704 | spinlock_t split_queue_lock; | |
705 | struct list_head split_queue; | |
706 | unsigned long split_queue_len; | |
707 | #endif | |
75ef7184 | 708 | |
599d0c95 MG |
709 | /* Fields commonly accessed by the page reclaim scanner */ |
710 | struct lruvec lruvec; | |
711 | ||
712 | /* | |
713 | * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on | |
714 | * this node's LRU. Maintained by the pageout code. | |
715 | */ | |
716 | unsigned int inactive_ratio; | |
717 | ||
718 | unsigned long flags; | |
719 | ||
720 | ZONE_PADDING(_pad2_) | |
721 | ||
75ef7184 MG |
722 | /* Per-node vmstats */ |
723 | struct per_cpu_nodestat __percpu *per_cpu_nodestats; | |
724 | atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS]; | |
1da177e4 LT |
725 | } pg_data_t; |
726 | ||
727 | #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages) | |
728 | #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages) | |
d41dee36 | 729 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
408fde81 | 730 | #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr)) |
d41dee36 AW |
731 | #else |
732 | #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr)) | |
733 | #endif | |
408fde81 | 734 | #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr)) |
1da177e4 | 735 | |
c6830c22 | 736 | #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn) |
da3649e1 | 737 | #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid)) |
a52633d8 MG |
738 | static inline spinlock_t *zone_lru_lock(struct zone *zone) |
739 | { | |
740 | return &zone->zone_pgdat->lru_lock; | |
741 | } | |
c6830c22 | 742 | |
a9dd0a83 | 743 | static inline struct lruvec *node_lruvec(struct pglist_data *pgdat) |
599d0c95 | 744 | { |
a9dd0a83 | 745 | return &pgdat->lruvec; |
599d0c95 MG |
746 | } |
747 | ||
da3649e1 CS |
748 | static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat) |
749 | { | |
750 | return pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
751 | } | |
752 | ||
753 | static inline bool pgdat_is_empty(pg_data_t *pgdat) | |
754 | { | |
755 | return !pgdat->node_start_pfn && !pgdat->node_spanned_pages; | |
756 | } | |
c6830c22 | 757 | |
033fbae9 DW |
758 | static inline int zone_id(const struct zone *zone) |
759 | { | |
760 | struct pglist_data *pgdat = zone->zone_pgdat; | |
761 | ||
762 | return zone - pgdat->node_zones; | |
763 | } | |
764 | ||
765 | #ifdef CONFIG_ZONE_DEVICE | |
766 | static inline bool is_dev_zone(const struct zone *zone) | |
767 | { | |
768 | return zone_id(zone) == ZONE_DEVICE; | |
769 | } | |
770 | #else | |
771 | static inline bool is_dev_zone(const struct zone *zone) | |
772 | { | |
773 | return false; | |
774 | } | |
775 | #endif | |
776 | ||
208d54e5 DH |
777 | #include <linux/memory_hotplug.h> |
778 | ||
4eaf3f64 | 779 | extern struct mutex zonelists_mutex; |
9adb62a5 | 780 | void build_all_zonelists(pg_data_t *pgdat, struct zone *zone); |
99504748 | 781 | void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx); |
86a294a8 MH |
782 | bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
783 | int classzone_idx, unsigned int alloc_flags, | |
784 | long free_pages); | |
7aeb09f9 | 785 | bool zone_watermark_ok(struct zone *z, unsigned int order, |
c603844b MG |
786 | unsigned long mark, int classzone_idx, |
787 | unsigned int alloc_flags); | |
7aeb09f9 | 788 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
e2b19197 | 789 | unsigned long mark, int classzone_idx); |
a2f3aa02 DH |
790 | enum memmap_context { |
791 | MEMMAP_EARLY, | |
792 | MEMMAP_HOTPLUG, | |
793 | }; | |
718127cc | 794 | extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn, |
b171e409 | 795 | unsigned long size); |
718127cc | 796 | |
bea8c150 | 797 | extern void lruvec_init(struct lruvec *lruvec); |
7f5e86c2 | 798 | |
599d0c95 | 799 | static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec) |
7f5e86c2 | 800 | { |
c255a458 | 801 | #ifdef CONFIG_MEMCG |
599d0c95 | 802 | return lruvec->pgdat; |
7f5e86c2 | 803 | #else |
599d0c95 | 804 | return container_of(lruvec, struct pglist_data, lruvec); |
7f5e86c2 KK |
805 | #endif |
806 | } | |
807 | ||
23047a96 JW |
808 | extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru); |
809 | ||
1da177e4 LT |
810 | #ifdef CONFIG_HAVE_MEMORY_PRESENT |
811 | void memory_present(int nid, unsigned long start, unsigned long end); | |
812 | #else | |
813 | static inline void memory_present(int nid, unsigned long start, unsigned long end) {} | |
814 | #endif | |
815 | ||
7aac7898 LS |
816 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
817 | int local_memory_node(int node_id); | |
818 | #else | |
819 | static inline int local_memory_node(int node_id) { return node_id; }; | |
820 | #endif | |
821 | ||
1da177e4 LT |
822 | #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE |
823 | unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long); | |
824 | #endif | |
825 | ||
826 | /* | |
827 | * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc. | |
828 | */ | |
829 | #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones) | |
830 | ||
6aa303de MG |
831 | /* |
832 | * Returns true if a zone has pages managed by the buddy allocator. | |
833 | * All the reclaim decisions have to use this function rather than | |
834 | * populated_zone(). If the whole zone is reserved then we can easily | |
835 | * end up with populated_zone() && !managed_zone(). | |
836 | */ | |
837 | static inline bool managed_zone(struct zone *zone) | |
838 | { | |
839 | return zone->managed_pages; | |
840 | } | |
841 | ||
842 | /* Returns true if a zone has memory */ | |
843 | static inline bool populated_zone(struct zone *zone) | |
f3fe6512 | 844 | { |
6aa303de | 845 | return zone->present_pages; |
f3fe6512 CK |
846 | } |
847 | ||
2a1e274a MG |
848 | extern int movable_zone; |
849 | ||
d7e4a2ea | 850 | #ifdef CONFIG_HIGHMEM |
2a1e274a MG |
851 | static inline int zone_movable_is_highmem(void) |
852 | { | |
d7e4a2ea | 853 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
2a1e274a MG |
854 | return movable_zone == ZONE_HIGHMEM; |
855 | #else | |
d7e4a2ea | 856 | return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM; |
2a1e274a MG |
857 | #endif |
858 | } | |
d7e4a2ea | 859 | #endif |
2a1e274a | 860 | |
2f1b6248 | 861 | static inline int is_highmem_idx(enum zone_type idx) |
1da177e4 | 862 | { |
e53ef38d | 863 | #ifdef CONFIG_HIGHMEM |
2a1e274a MG |
864 | return (idx == ZONE_HIGHMEM || |
865 | (idx == ZONE_MOVABLE && zone_movable_is_highmem())); | |
e53ef38d CL |
866 | #else |
867 | return 0; | |
868 | #endif | |
1da177e4 LT |
869 | } |
870 | ||
1da177e4 LT |
871 | /** |
872 | * is_highmem - helper function to quickly check if a struct zone is a | |
873 | * highmem zone or not. This is an attempt to keep references | |
874 | * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum. | |
875 | * @zone - pointer to struct zone variable | |
876 | */ | |
877 | static inline int is_highmem(struct zone *zone) | |
878 | { | |
e53ef38d | 879 | #ifdef CONFIG_HIGHMEM |
29f9cb53 | 880 | return is_highmem_idx(zone_idx(zone)); |
e53ef38d CL |
881 | #else |
882 | return 0; | |
883 | #endif | |
1da177e4 LT |
884 | } |
885 | ||
1da177e4 LT |
886 | /* These two functions are used to setup the per zone pages min values */ |
887 | struct ctl_table; | |
8d65af78 | 888 | int min_free_kbytes_sysctl_handler(struct ctl_table *, int, |
1da177e4 | 889 | void __user *, size_t *, loff_t *); |
795ae7a0 JW |
890 | int watermark_scale_factor_sysctl_handler(struct ctl_table *, int, |
891 | void __user *, size_t *, loff_t *); | |
1da177e4 | 892 | extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1]; |
8d65af78 | 893 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, |
1da177e4 | 894 | void __user *, size_t *, loff_t *); |
8d65af78 | 895 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, |
8ad4b1fb | 896 | void __user *, size_t *, loff_t *); |
9614634f | 897 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int, |
8d65af78 | 898 | void __user *, size_t *, loff_t *); |
0ff38490 | 899 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int, |
8d65af78 | 900 | void __user *, size_t *, loff_t *); |
1da177e4 | 901 | |
f0c0b2b8 | 902 | extern int numa_zonelist_order_handler(struct ctl_table *, int, |
8d65af78 | 903 | void __user *, size_t *, loff_t *); |
f0c0b2b8 KH |
904 | extern char numa_zonelist_order[]; |
905 | #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */ | |
906 | ||
93b7504e | 907 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
908 | |
909 | extern struct pglist_data contig_page_data; | |
910 | #define NODE_DATA(nid) (&contig_page_data) | |
911 | #define NODE_MEM_MAP(nid) mem_map | |
1da177e4 | 912 | |
93b7504e | 913 | #else /* CONFIG_NEED_MULTIPLE_NODES */ |
1da177e4 LT |
914 | |
915 | #include <asm/mmzone.h> | |
916 | ||
93b7504e | 917 | #endif /* !CONFIG_NEED_MULTIPLE_NODES */ |
348f8b6c | 918 | |
95144c78 KH |
919 | extern struct pglist_data *first_online_pgdat(void); |
920 | extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat); | |
921 | extern struct zone *next_zone(struct zone *zone); | |
8357f869 KH |
922 | |
923 | /** | |
12d15f0d | 924 | * for_each_online_pgdat - helper macro to iterate over all online nodes |
8357f869 KH |
925 | * @pgdat - pointer to a pg_data_t variable |
926 | */ | |
927 | #define for_each_online_pgdat(pgdat) \ | |
928 | for (pgdat = first_online_pgdat(); \ | |
929 | pgdat; \ | |
930 | pgdat = next_online_pgdat(pgdat)) | |
8357f869 KH |
931 | /** |
932 | * for_each_zone - helper macro to iterate over all memory zones | |
933 | * @zone - pointer to struct zone variable | |
934 | * | |
935 | * The user only needs to declare the zone variable, for_each_zone | |
936 | * fills it in. | |
937 | */ | |
938 | #define for_each_zone(zone) \ | |
939 | for (zone = (first_online_pgdat())->node_zones; \ | |
940 | zone; \ | |
941 | zone = next_zone(zone)) | |
942 | ||
ee99c71c KM |
943 | #define for_each_populated_zone(zone) \ |
944 | for (zone = (first_online_pgdat())->node_zones; \ | |
945 | zone; \ | |
946 | zone = next_zone(zone)) \ | |
947 | if (!populated_zone(zone)) \ | |
948 | ; /* do nothing */ \ | |
949 | else | |
950 | ||
dd1a239f MG |
951 | static inline struct zone *zonelist_zone(struct zoneref *zoneref) |
952 | { | |
953 | return zoneref->zone; | |
954 | } | |
955 | ||
956 | static inline int zonelist_zone_idx(struct zoneref *zoneref) | |
957 | { | |
958 | return zoneref->zone_idx; | |
959 | } | |
960 | ||
961 | static inline int zonelist_node_idx(struct zoneref *zoneref) | |
962 | { | |
963 | #ifdef CONFIG_NUMA | |
964 | /* zone_to_nid not available in this context */ | |
965 | return zoneref->zone->node; | |
966 | #else | |
967 | return 0; | |
968 | #endif /* CONFIG_NUMA */ | |
969 | } | |
970 | ||
682a3385 MG |
971 | struct zoneref *__next_zones_zonelist(struct zoneref *z, |
972 | enum zone_type highest_zoneidx, | |
973 | nodemask_t *nodes); | |
974 | ||
19770b32 MG |
975 | /** |
976 | * 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 | |
977 | * @z - The cursor used as a starting point for the search | |
978 | * @highest_zoneidx - The zone index of the highest zone to return | |
979 | * @nodes - An optional nodemask to filter the zonelist with | |
19770b32 MG |
980 | * |
981 | * This function returns the next zone at or below a given zone index that is | |
982 | * within the allowed nodemask using a cursor as the starting point for the | |
5bead2a0 MG |
983 | * search. The zoneref returned is a cursor that represents the current zone |
984 | * being examined. It should be advanced by one before calling | |
985 | * next_zones_zonelist again. | |
19770b32 | 986 | */ |
682a3385 | 987 | static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z, |
19770b32 | 988 | enum zone_type highest_zoneidx, |
682a3385 MG |
989 | nodemask_t *nodes) |
990 | { | |
991 | if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx)) | |
992 | return z; | |
993 | return __next_zones_zonelist(z, highest_zoneidx, nodes); | |
994 | } | |
dd1a239f | 995 | |
19770b32 MG |
996 | /** |
997 | * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist | |
998 | * @zonelist - The zonelist to search for a suitable zone | |
999 | * @highest_zoneidx - The zone index of the highest zone to return | |
1000 | * @nodes - An optional nodemask to filter the zonelist with | |
1001 | * @zone - The first suitable zone found is returned via this parameter | |
1002 | * | |
1003 | * This function returns the first zone at or below a given zone index that is | |
1004 | * within the allowed nodemask. The zoneref returned is a cursor that can be | |
5bead2a0 MG |
1005 | * used to iterate the zonelist with next_zones_zonelist by advancing it by |
1006 | * one before calling. | |
19770b32 | 1007 | */ |
dd1a239f | 1008 | static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist, |
19770b32 | 1009 | enum zone_type highest_zoneidx, |
c33d6c06 | 1010 | nodemask_t *nodes) |
54a6eb5c | 1011 | { |
c33d6c06 | 1012 | return next_zones_zonelist(zonelist->_zonerefs, |
05891fb0 | 1013 | highest_zoneidx, nodes); |
54a6eb5c MG |
1014 | } |
1015 | ||
19770b32 MG |
1016 | /** |
1017 | * 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 | |
1018 | * @zone - The current zone in the iterator | |
1019 | * @z - The current pointer within zonelist->zones being iterated | |
1020 | * @zlist - The zonelist being iterated | |
1021 | * @highidx - The zone index of the highest zone to return | |
1022 | * @nodemask - Nodemask allowed by the allocator | |
1023 | * | |
1024 | * This iterator iterates though all zones at or below a given zone index and | |
1025 | * within a given nodemask | |
1026 | */ | |
1027 | #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \ | |
c33d6c06 | 1028 | for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \ |
19770b32 | 1029 | zone; \ |
05891fb0 | 1030 | z = next_zones_zonelist(++z, highidx, nodemask), \ |
c33d6c06 MG |
1031 | zone = zonelist_zone(z)) |
1032 | ||
1033 | #define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \ | |
1034 | for (zone = z->zone; \ | |
1035 | zone; \ | |
1036 | z = next_zones_zonelist(++z, highidx, nodemask), \ | |
1037 | zone = zonelist_zone(z)) | |
1038 | ||
54a6eb5c MG |
1039 | |
1040 | /** | |
1041 | * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index | |
1042 | * @zone - The current zone in the iterator | |
1043 | * @z - The current pointer within zonelist->zones being iterated | |
1044 | * @zlist - The zonelist being iterated | |
1045 | * @highidx - The zone index of the highest zone to return | |
1046 | * | |
1047 | * This iterator iterates though all zones at or below a given zone index. | |
1048 | */ | |
1049 | #define for_each_zone_zonelist(zone, z, zlist, highidx) \ | |
19770b32 | 1050 | for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL) |
54a6eb5c | 1051 | |
d41dee36 AW |
1052 | #ifdef CONFIG_SPARSEMEM |
1053 | #include <asm/sparsemem.h> | |
1054 | #endif | |
1055 | ||
c713216d | 1056 | #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \ |
0ee332c1 | 1057 | !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) |
b4544568 AM |
1058 | static inline unsigned long early_pfn_to_nid(unsigned long pfn) |
1059 | { | |
1060 | return 0; | |
1061 | } | |
b159d43f AW |
1062 | #endif |
1063 | ||
2bdaf115 AW |
1064 | #ifdef CONFIG_FLATMEM |
1065 | #define pfn_to_nid(pfn) (0) | |
1066 | #endif | |
1067 | ||
d41dee36 AW |
1068 | #ifdef CONFIG_SPARSEMEM |
1069 | ||
1070 | /* | |
1071 | * SECTION_SHIFT #bits space required to store a section # | |
1072 | * | |
1073 | * PA_SECTION_SHIFT physical address to/from section number | |
1074 | * PFN_SECTION_SHIFT pfn to/from section number | |
1075 | */ | |
d41dee36 AW |
1076 | #define PA_SECTION_SHIFT (SECTION_SIZE_BITS) |
1077 | #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT) | |
1078 | ||
1079 | #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT) | |
1080 | ||
1081 | #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT) | |
1082 | #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1)) | |
1083 | ||
835c134e | 1084 | #define SECTION_BLOCKFLAGS_BITS \ |
d9c23400 | 1085 | ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS) |
835c134e | 1086 | |
d41dee36 AW |
1087 | #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS |
1088 | #error Allocator MAX_ORDER exceeds SECTION_SIZE | |
1089 | #endif | |
1090 | ||
e3c40f37 DK |
1091 | #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT) |
1092 | #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT) | |
1093 | ||
a539f353 DK |
1094 | #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK) |
1095 | #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK) | |
1096 | ||
d41dee36 | 1097 | struct page; |
eefa864b | 1098 | struct page_ext; |
d41dee36 | 1099 | struct mem_section { |
29751f69 AW |
1100 | /* |
1101 | * This is, logically, a pointer to an array of struct | |
1102 | * pages. However, it is stored with some other magic. | |
1103 | * (see sparse.c::sparse_init_one_section()) | |
1104 | * | |
30c253e6 AW |
1105 | * Additionally during early boot we encode node id of |
1106 | * the location of the section here to guide allocation. | |
1107 | * (see sparse.c::memory_present()) | |
1108 | * | |
29751f69 AW |
1109 | * Making it a UL at least makes someone do a cast |
1110 | * before using it wrong. | |
1111 | */ | |
1112 | unsigned long section_mem_map; | |
5c0e3066 MG |
1113 | |
1114 | /* See declaration of similar field in struct zone */ | |
1115 | unsigned long *pageblock_flags; | |
eefa864b JK |
1116 | #ifdef CONFIG_PAGE_EXTENSION |
1117 | /* | |
0c9ad804 | 1118 | * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use |
eefa864b JK |
1119 | * section. (see page_ext.h about this.) |
1120 | */ | |
1121 | struct page_ext *page_ext; | |
1122 | unsigned long pad; | |
1123 | #endif | |
55878e88 CS |
1124 | /* |
1125 | * WARNING: mem_section must be a power-of-2 in size for the | |
1126 | * calculation and use of SECTION_ROOT_MASK to make sense. | |
1127 | */ | |
d41dee36 AW |
1128 | }; |
1129 | ||
3e347261 BP |
1130 | #ifdef CONFIG_SPARSEMEM_EXTREME |
1131 | #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section)) | |
1132 | #else | |
1133 | #define SECTIONS_PER_ROOT 1 | |
1134 | #endif | |
802f192e | 1135 | |
3e347261 | 1136 | #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT) |
0faa5638 | 1137 | #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT) |
3e347261 | 1138 | #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1) |
802f192e | 1139 | |
3e347261 BP |
1140 | #ifdef CONFIG_SPARSEMEM_EXTREME |
1141 | extern struct mem_section *mem_section[NR_SECTION_ROOTS]; | |
802f192e | 1142 | #else |
3e347261 BP |
1143 | extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]; |
1144 | #endif | |
d41dee36 | 1145 | |
29751f69 AW |
1146 | static inline struct mem_section *__nr_to_section(unsigned long nr) |
1147 | { | |
3e347261 BP |
1148 | if (!mem_section[SECTION_NR_TO_ROOT(nr)]) |
1149 | return NULL; | |
1150 | return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK]; | |
29751f69 | 1151 | } |
4ca644d9 | 1152 | extern int __section_nr(struct mem_section* ms); |
04753278 | 1153 | extern unsigned long usemap_size(void); |
29751f69 AW |
1154 | |
1155 | /* | |
1156 | * We use the lower bits of the mem_map pointer to store | |
1157 | * a little bit of information. There should be at least | |
1158 | * 3 bits here due to 32-bit alignment. | |
1159 | */ | |
1160 | #define SECTION_MARKED_PRESENT (1UL<<0) | |
1161 | #define SECTION_HAS_MEM_MAP (1UL<<1) | |
1162 | #define SECTION_MAP_LAST_BIT (1UL<<2) | |
1163 | #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1)) | |
30c253e6 | 1164 | #define SECTION_NID_SHIFT 2 |
29751f69 AW |
1165 | |
1166 | static inline struct page *__section_mem_map_addr(struct mem_section *section) | |
1167 | { | |
1168 | unsigned long map = section->section_mem_map; | |
1169 | map &= SECTION_MAP_MASK; | |
1170 | return (struct page *)map; | |
1171 | } | |
1172 | ||
540557b9 | 1173 | static inline int present_section(struct mem_section *section) |
29751f69 | 1174 | { |
802f192e | 1175 | return (section && (section->section_mem_map & SECTION_MARKED_PRESENT)); |
29751f69 AW |
1176 | } |
1177 | ||
540557b9 AW |
1178 | static inline int present_section_nr(unsigned long nr) |
1179 | { | |
1180 | return present_section(__nr_to_section(nr)); | |
1181 | } | |
1182 | ||
1183 | static inline int valid_section(struct mem_section *section) | |
29751f69 | 1184 | { |
802f192e | 1185 | return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP)); |
29751f69 AW |
1186 | } |
1187 | ||
1188 | static inline int valid_section_nr(unsigned long nr) | |
1189 | { | |
1190 | return valid_section(__nr_to_section(nr)); | |
1191 | } | |
1192 | ||
d41dee36 AW |
1193 | static inline struct mem_section *__pfn_to_section(unsigned long pfn) |
1194 | { | |
29751f69 | 1195 | return __nr_to_section(pfn_to_section_nr(pfn)); |
d41dee36 AW |
1196 | } |
1197 | ||
7b7bf499 | 1198 | #ifndef CONFIG_HAVE_ARCH_PFN_VALID |
d41dee36 AW |
1199 | static inline int pfn_valid(unsigned long pfn) |
1200 | { | |
1201 | if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) | |
1202 | return 0; | |
29751f69 | 1203 | return valid_section(__nr_to_section(pfn_to_section_nr(pfn))); |
d41dee36 | 1204 | } |
7b7bf499 | 1205 | #endif |
d41dee36 | 1206 | |
540557b9 AW |
1207 | static inline int pfn_present(unsigned long pfn) |
1208 | { | |
1209 | if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) | |
1210 | return 0; | |
1211 | return present_section(__nr_to_section(pfn_to_section_nr(pfn))); | |
1212 | } | |
1213 | ||
d41dee36 AW |
1214 | /* |
1215 | * These are _only_ used during initialisation, therefore they | |
1216 | * can use __initdata ... They could have names to indicate | |
1217 | * this restriction. | |
1218 | */ | |
1219 | #ifdef CONFIG_NUMA | |
161599ff AW |
1220 | #define pfn_to_nid(pfn) \ |
1221 | ({ \ | |
1222 | unsigned long __pfn_to_nid_pfn = (pfn); \ | |
1223 | page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \ | |
1224 | }) | |
2bdaf115 AW |
1225 | #else |
1226 | #define pfn_to_nid(pfn) (0) | |
d41dee36 AW |
1227 | #endif |
1228 | ||
d41dee36 AW |
1229 | #define early_pfn_valid(pfn) pfn_valid(pfn) |
1230 | void sparse_init(void); | |
1231 | #else | |
1232 | #define sparse_init() do {} while (0) | |
28ae55c9 | 1233 | #define sparse_index_init(_sec, _nid) do {} while (0) |
d41dee36 AW |
1234 | #endif /* CONFIG_SPARSEMEM */ |
1235 | ||
8a942fde MG |
1236 | /* |
1237 | * During memory init memblocks map pfns to nids. The search is expensive and | |
1238 | * this caches recent lookups. The implementation of __early_pfn_to_nid | |
1239 | * may treat start/end as pfns or sections. | |
1240 | */ | |
1241 | struct mminit_pfnnid_cache { | |
1242 | unsigned long last_start; | |
1243 | unsigned long last_end; | |
1244 | int last_nid; | |
1245 | }; | |
1246 | ||
d41dee36 AW |
1247 | #ifndef early_pfn_valid |
1248 | #define early_pfn_valid(pfn) (1) | |
1249 | #endif | |
1250 | ||
1251 | void memory_present(int nid, unsigned long start, unsigned long end); | |
1252 | unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long); | |
1253 | ||
14e07298 AW |
1254 | /* |
1255 | * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we | |
1256 | * need to check pfn validility within that MAX_ORDER_NR_PAGES block. | |
1257 | * pfn_valid_within() should be used in this case; we optimise this away | |
1258 | * when we have no holes within a MAX_ORDER_NR_PAGES block. | |
1259 | */ | |
1260 | #ifdef CONFIG_HOLES_IN_ZONE | |
1261 | #define pfn_valid_within(pfn) pfn_valid(pfn) | |
1262 | #else | |
1263 | #define pfn_valid_within(pfn) (1) | |
1264 | #endif | |
1265 | ||
eb33575c MG |
1266 | #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL |
1267 | /* | |
1268 | * pfn_valid() is meant to be able to tell if a given PFN has valid memmap | |
1269 | * associated with it or not. In FLATMEM, it is expected that holes always | |
1270 | * have valid memmap as long as there is valid PFNs either side of the hole. | |
1271 | * In SPARSEMEM, it is assumed that a valid section has a memmap for the | |
1272 | * entire section. | |
1273 | * | |
1274 | * However, an ARM, and maybe other embedded architectures in the future | |
1275 | * free memmap backing holes to save memory on the assumption the memmap is | |
1276 | * never used. The page_zone linkages are then broken even though pfn_valid() | |
1277 | * returns true. A walker of the full memmap must then do this additional | |
1278 | * check to ensure the memmap they are looking at is sane by making sure | |
1279 | * the zone and PFN linkages are still valid. This is expensive, but walkers | |
1280 | * of the full memmap are extremely rare. | |
1281 | */ | |
5b80287a | 1282 | bool memmap_valid_within(unsigned long pfn, |
eb33575c MG |
1283 | struct page *page, struct zone *zone); |
1284 | #else | |
5b80287a | 1285 | static inline bool memmap_valid_within(unsigned long pfn, |
eb33575c MG |
1286 | struct page *page, struct zone *zone) |
1287 | { | |
5b80287a | 1288 | return true; |
eb33575c MG |
1289 | } |
1290 | #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */ | |
1291 | ||
97965478 | 1292 | #endif /* !__GENERATING_BOUNDS.H */ |
1da177e4 | 1293 | #endif /* !__ASSEMBLY__ */ |
1da177e4 | 1294 | #endif /* _LINUX_MMZONE_H */ |