Commit | Line | Data |
---|---|---|
b2441318 | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
1da177e4 LT |
2 | #ifndef _LINUX_MMZONE_H |
3 | #define _LINUX_MMZONE_H | |
4 | ||
1da177e4 | 5 | #ifndef __ASSEMBLY__ |
97965478 | 6 | #ifndef __GENERATING_BOUNDS_H |
1da177e4 | 7 | |
1da177e4 LT |
8 | #include <linux/spinlock.h> |
9 | #include <linux/list.h> | |
10 | #include <linux/wait.h> | |
e815af95 | 11 | #include <linux/bitops.h> |
1da177e4 LT |
12 | #include <linux/cache.h> |
13 | #include <linux/threads.h> | |
14 | #include <linux/numa.h> | |
15 | #include <linux/init.h> | |
bdc8cb98 | 16 | #include <linux/seqlock.h> |
8357f869 | 17 | #include <linux/nodemask.h> |
835c134e | 18 | #include <linux/pageblock-flags.h> |
bbeae5b0 | 19 | #include <linux/page-flags-layout.h> |
60063497 | 20 | #include <linux/atomic.h> |
b03641af DW |
21 | #include <linux/mm_types.h> |
22 | #include <linux/page-flags.h> | |
dbbee9d5 | 23 | #include <linux/local_lock.h> |
93ff66bf | 24 | #include <asm/page.h> |
1da177e4 LT |
25 | |
26 | /* Free memory management - zoned buddy allocator. */ | |
0192445c | 27 | #ifndef CONFIG_ARCH_FORCE_MAX_ORDER |
1da177e4 LT |
28 | #define MAX_ORDER 11 |
29 | #else | |
0192445c | 30 | #define MAX_ORDER CONFIG_ARCH_FORCE_MAX_ORDER |
1da177e4 | 31 | #endif |
e984bb43 | 32 | #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1)) |
1da177e4 | 33 | |
5ad333eb AW |
34 | /* |
35 | * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed | |
36 | * costly to service. That is between allocation orders which should | |
35fca53e | 37 | * coalesce naturally under reasonable reclaim pressure and those which |
5ad333eb AW |
38 | * will not. |
39 | */ | |
40 | #define PAGE_ALLOC_COSTLY_ORDER 3 | |
41 | ||
a6ffdc07 | 42 | enum migratetype { |
47118af0 | 43 | MIGRATE_UNMOVABLE, |
47118af0 | 44 | MIGRATE_MOVABLE, |
016c13da | 45 | MIGRATE_RECLAIMABLE, |
0aaa29a5 MG |
46 | MIGRATE_PCPTYPES, /* the number of types on the pcp lists */ |
47 | MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES, | |
47118af0 MN |
48 | #ifdef CONFIG_CMA |
49 | /* | |
50 | * MIGRATE_CMA migration type is designed to mimic the way | |
51 | * ZONE_MOVABLE works. Only movable pages can be allocated | |
52 | * from MIGRATE_CMA pageblocks and page allocator never | |
53 | * implicitly change migration type of MIGRATE_CMA pageblock. | |
54 | * | |
55 | * The way to use it is to change migratetype of a range of | |
56 | * pageblocks to MIGRATE_CMA which can be done by | |
11ac3e87 | 57 | * __free_pageblock_cma() function. |
47118af0 MN |
58 | */ |
59 | MIGRATE_CMA, | |
60 | #endif | |
194159fb | 61 | #ifdef CONFIG_MEMORY_ISOLATION |
47118af0 | 62 | MIGRATE_ISOLATE, /* can't allocate from here */ |
194159fb | 63 | #endif |
47118af0 MN |
64 | MIGRATE_TYPES |
65 | }; | |
66 | ||
60f30350 | 67 | /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */ |
c999fbd3 | 68 | extern const char * const migratetype_names[MIGRATE_TYPES]; |
60f30350 | 69 | |
47118af0 MN |
70 | #ifdef CONFIG_CMA |
71 | # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA) | |
7c15d9bb | 72 | # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA) |
47118af0 MN |
73 | #else |
74 | # define is_migrate_cma(migratetype) false | |
7c15d9bb | 75 | # define is_migrate_cma_page(_page) false |
47118af0 | 76 | #endif |
b2a0ac88 | 77 | |
b682debd VB |
78 | static inline bool is_migrate_movable(int mt) |
79 | { | |
80 | return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE; | |
81 | } | |
82 | ||
1dd214b8 ZY |
83 | /* |
84 | * Check whether a migratetype can be merged with another migratetype. | |
85 | * | |
86 | * It is only mergeable when it can fall back to other migratetypes for | |
87 | * allocation. See fallbacks[MIGRATE_TYPES][3] in page_alloc.c. | |
88 | */ | |
89 | static inline bool migratetype_is_mergeable(int mt) | |
90 | { | |
91 | return mt < MIGRATE_PCPTYPES; | |
92 | } | |
93 | ||
b2a0ac88 MG |
94 | #define for_each_migratetype_order(order, type) \ |
95 | for (order = 0; order < MAX_ORDER; order++) \ | |
96 | for (type = 0; type < MIGRATE_TYPES; type++) | |
97 | ||
467c996c MG |
98 | extern int page_group_by_mobility_disabled; |
99 | ||
d38ac97f | 100 | #define MIGRATETYPE_MASK ((1UL << PB_migratetype_bits) - 1) |
e58469ba | 101 | |
dc4b0caf | 102 | #define get_pageblock_migratetype(page) \ |
535b81e2 | 103 | get_pfnblock_flags_mask(page, page_to_pfn(page), MIGRATETYPE_MASK) |
dc4b0caf | 104 | |
1da177e4 | 105 | struct free_area { |
b2a0ac88 | 106 | struct list_head free_list[MIGRATE_TYPES]; |
1da177e4 LT |
107 | unsigned long nr_free; |
108 | }; | |
109 | ||
b03641af DW |
110 | static inline struct page *get_page_from_free_area(struct free_area *area, |
111 | int migratetype) | |
112 | { | |
113 | return list_first_entry_or_null(&area->free_list[migratetype], | |
114 | struct page, lru); | |
115 | } | |
116 | ||
b03641af DW |
117 | static inline bool free_area_empty(struct free_area *area, int migratetype) |
118 | { | |
119 | return list_empty(&area->free_list[migratetype]); | |
120 | } | |
121 | ||
1da177e4 LT |
122 | struct pglist_data; |
123 | ||
124 | /* | |
041711ce | 125 | * Add a wild amount of padding here to ensure data fall into separate |
1da177e4 LT |
126 | * cachelines. There are very few zone structures in the machine, so space |
127 | * consumption is not a concern here. | |
128 | */ | |
129 | #if defined(CONFIG_SMP) | |
130 | struct zone_padding { | |
131 | char x[0]; | |
22fc6ecc | 132 | } ____cacheline_internodealigned_in_smp; |
1da177e4 LT |
133 | #define ZONE_PADDING(name) struct zone_padding name; |
134 | #else | |
135 | #define ZONE_PADDING(name) | |
136 | #endif | |
137 | ||
3a321d2a KW |
138 | #ifdef CONFIG_NUMA |
139 | enum numa_stat_item { | |
140 | NUMA_HIT, /* allocated in intended node */ | |
141 | NUMA_MISS, /* allocated in non intended node */ | |
142 | NUMA_FOREIGN, /* was intended here, hit elsewhere */ | |
143 | NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */ | |
144 | NUMA_LOCAL, /* allocation from local node */ | |
145 | NUMA_OTHER, /* allocation from other node */ | |
f19298b9 | 146 | NR_VM_NUMA_EVENT_ITEMS |
3a321d2a KW |
147 | }; |
148 | #else | |
f19298b9 | 149 | #define NR_VM_NUMA_EVENT_ITEMS 0 |
3a321d2a KW |
150 | #endif |
151 | ||
2244b95a | 152 | enum zone_stat_item { |
51ed4491 | 153 | /* First 128 byte cacheline (assuming 64 bit words) */ |
d23ad423 | 154 | NR_FREE_PAGES, |
71c799f4 MK |
155 | NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */ |
156 | NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE, | |
157 | NR_ZONE_ACTIVE_ANON, | |
158 | NR_ZONE_INACTIVE_FILE, | |
159 | NR_ZONE_ACTIVE_FILE, | |
160 | NR_ZONE_UNEVICTABLE, | |
5a1c84b4 | 161 | NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */ |
5344b7e6 | 162 | NR_MLOCK, /* mlock()ed pages found and moved off LRU */ |
c6a7f572 | 163 | /* Second 128 byte cacheline */ |
d2c5e30c | 164 | NR_BOUNCE, |
91537fee MK |
165 | #if IS_ENABLED(CONFIG_ZSMALLOC) |
166 | NR_ZSPAGES, /* allocated in zsmalloc */ | |
ca889e6c | 167 | #endif |
d1ce749a | 168 | NR_FREE_CMA_PAGES, |
2244b95a CL |
169 | NR_VM_ZONE_STAT_ITEMS }; |
170 | ||
75ef7184 | 171 | enum node_stat_item { |
599d0c95 MG |
172 | NR_LRU_BASE, |
173 | NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */ | |
174 | NR_ACTIVE_ANON, /* " " " " " */ | |
175 | NR_INACTIVE_FILE, /* " " " " " */ | |
176 | NR_ACTIVE_FILE, /* " " " " " */ | |
177 | NR_UNEVICTABLE, /* " " " " " */ | |
d42f3245 RG |
178 | NR_SLAB_RECLAIMABLE_B, |
179 | NR_SLAB_UNRECLAIMABLE_B, | |
599d0c95 MG |
180 | NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */ |
181 | NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */ | |
68d48e6a | 182 | WORKINGSET_NODES, |
170b04b7 JK |
183 | WORKINGSET_REFAULT_BASE, |
184 | WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE, | |
185 | WORKINGSET_REFAULT_FILE, | |
186 | WORKINGSET_ACTIVATE_BASE, | |
187 | WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE, | |
188 | WORKINGSET_ACTIVATE_FILE, | |
189 | WORKINGSET_RESTORE_BASE, | |
190 | WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE, | |
191 | WORKINGSET_RESTORE_FILE, | |
1e6b1085 | 192 | WORKINGSET_NODERECLAIM, |
4b9d0fab | 193 | NR_ANON_MAPPED, /* Mapped anonymous pages */ |
50658e2e MG |
194 | NR_FILE_MAPPED, /* pagecache pages mapped into pagetables. |
195 | only modified from process context */ | |
11fb9989 MG |
196 | NR_FILE_PAGES, |
197 | NR_FILE_DIRTY, | |
198 | NR_WRITEBACK, | |
199 | NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */ | |
200 | NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */ | |
201 | NR_SHMEM_THPS, | |
202 | NR_SHMEM_PMDMAPPED, | |
60fbf0ab SL |
203 | NR_FILE_THPS, |
204 | NR_FILE_PMDMAPPED, | |
11fb9989 | 205 | NR_ANON_THPS, |
c4a25635 MG |
206 | NR_VMSCAN_WRITE, |
207 | NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */ | |
208 | NR_DIRTIED, /* page dirtyings since bootup */ | |
209 | NR_WRITTEN, /* page writings since bootup */ | |
8cd7c588 | 210 | NR_THROTTLED_WRITTEN, /* NR_WRITTEN while reclaim throttled */ |
b29940c1 | 211 | NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */ |
1970dc6f JH |
212 | NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */ |
213 | NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */ | |
991e7673 SB |
214 | NR_KERNEL_STACK_KB, /* measured in KiB */ |
215 | #if IS_ENABLED(CONFIG_SHADOW_CALL_STACK) | |
216 | NR_KERNEL_SCS_KB, /* measured in KiB */ | |
217 | #endif | |
f0c0c115 | 218 | NR_PAGETABLE, /* used for pagetables */ |
b6038942 SB |
219 | #ifdef CONFIG_SWAP |
220 | NR_SWAPCACHE, | |
e39bb6be HY |
221 | #endif |
222 | #ifdef CONFIG_NUMA_BALANCING | |
223 | PGPROMOTE_SUCCESS, /* promote successfully */ | |
c6833e10 | 224 | PGPROMOTE_CANDIDATE, /* candidate pages to promote */ |
b6038942 | 225 | #endif |
75ef7184 MG |
226 | NR_VM_NODE_STAT_ITEMS |
227 | }; | |
228 | ||
69473e5d MS |
229 | /* |
230 | * Returns true if the item should be printed in THPs (/proc/vmstat | |
231 | * currently prints number of anon, file and shmem THPs. But the item | |
232 | * is charged in pages). | |
233 | */ | |
234 | static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item) | |
235 | { | |
236 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) | |
237 | return false; | |
238 | ||
bf9ecead | 239 | return item == NR_ANON_THPS || |
57b2847d | 240 | item == NR_FILE_THPS || |
a1528e21 | 241 | item == NR_SHMEM_THPS || |
380780e7 MS |
242 | item == NR_SHMEM_PMDMAPPED || |
243 | item == NR_FILE_PMDMAPPED; | |
69473e5d MS |
244 | } |
245 | ||
ea426c2a RG |
246 | /* |
247 | * Returns true if the value is measured in bytes (most vmstat values are | |
248 | * measured in pages). This defines the API part, the internal representation | |
249 | * might be different. | |
250 | */ | |
251 | static __always_inline bool vmstat_item_in_bytes(int idx) | |
252 | { | |
d42f3245 RG |
253 | /* |
254 | * Global and per-node slab counters track slab pages. | |
255 | * It's expected that changes are multiples of PAGE_SIZE. | |
256 | * Internally values are stored in pages. | |
257 | * | |
258 | * Per-memcg and per-lruvec counters track memory, consumed | |
259 | * by individual slab objects. These counters are actually | |
260 | * byte-precise. | |
261 | */ | |
262 | return (idx == NR_SLAB_RECLAIMABLE_B || | |
263 | idx == NR_SLAB_UNRECLAIMABLE_B); | |
ea426c2a RG |
264 | } |
265 | ||
4f98a2fe RR |
266 | /* |
267 | * We do arithmetic on the LRU lists in various places in the code, | |
268 | * so it is important to keep the active lists LRU_ACTIVE higher in | |
269 | * the array than the corresponding inactive lists, and to keep | |
270 | * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists. | |
271 | * | |
272 | * This has to be kept in sync with the statistics in zone_stat_item | |
273 | * above and the descriptions in vmstat_text in mm/vmstat.c | |
274 | */ | |
275 | #define LRU_BASE 0 | |
276 | #define LRU_ACTIVE 1 | |
277 | #define LRU_FILE 2 | |
278 | ||
b69408e8 | 279 | enum lru_list { |
4f98a2fe RR |
280 | LRU_INACTIVE_ANON = LRU_BASE, |
281 | LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE, | |
282 | LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE, | |
283 | LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE, | |
894bc310 | 284 | LRU_UNEVICTABLE, |
894bc310 LS |
285 | NR_LRU_LISTS |
286 | }; | |
b69408e8 | 287 | |
8cd7c588 MG |
288 | enum vmscan_throttle_state { |
289 | VMSCAN_THROTTLE_WRITEBACK, | |
d818fca1 | 290 | VMSCAN_THROTTLE_ISOLATED, |
69392a40 | 291 | VMSCAN_THROTTLE_NOPROGRESS, |
1b4e3f26 | 292 | VMSCAN_THROTTLE_CONGESTED, |
8cd7c588 MG |
293 | NR_VMSCAN_THROTTLE, |
294 | }; | |
295 | ||
4111304d | 296 | #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++) |
b69408e8 | 297 | |
4111304d | 298 | #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++) |
894bc310 | 299 | |
b91ac374 | 300 | static inline bool is_file_lru(enum lru_list lru) |
4f98a2fe | 301 | { |
4111304d | 302 | return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE); |
4f98a2fe RR |
303 | } |
304 | ||
b91ac374 | 305 | static inline bool is_active_lru(enum lru_list lru) |
b69408e8 | 306 | { |
4111304d | 307 | return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE); |
b69408e8 CL |
308 | } |
309 | ||
e9c2dbc8 YY |
310 | #define WORKINGSET_ANON 0 |
311 | #define WORKINGSET_FILE 1 | |
ed017373 YZ |
312 | #define ANON_AND_FILE 2 |
313 | ||
1b05117d JW |
314 | enum lruvec_flags { |
315 | LRUVEC_CONGESTED, /* lruvec has many dirty pages | |
316 | * backed by a congested BDI | |
317 | */ | |
318 | }; | |
319 | ||
6290df54 | 320 | struct lruvec { |
23047a96 | 321 | struct list_head lists[NR_LRU_LISTS]; |
6168d0da AS |
322 | /* per lruvec lru_lock for memcg */ |
323 | spinlock_t lru_lock; | |
1431d4d1 JW |
324 | /* |
325 | * These track the cost of reclaiming one LRU - file or anon - | |
326 | * over the other. As the observed cost of reclaiming one LRU | |
327 | * increases, the reclaim scan balance tips toward the other. | |
328 | */ | |
329 | unsigned long anon_cost; | |
330 | unsigned long file_cost; | |
31d8fcac JW |
331 | /* Non-resident age, driven by LRU movement */ |
332 | atomic_long_t nonresident_age; | |
ed017373 YZ |
333 | /* Refaults at the time of last reclaim cycle */ |
334 | unsigned long refaults[ANON_AND_FILE]; | |
1b05117d JW |
335 | /* Various lruvec state flags (enum lruvec_flags) */ |
336 | unsigned long flags; | |
c255a458 | 337 | #ifdef CONFIG_MEMCG |
599d0c95 | 338 | struct pglist_data *pgdat; |
7f5e86c2 | 339 | #endif |
6290df54 JW |
340 | }; |
341 | ||
653e003d | 342 | /* Isolate unmapped pages */ |
f3fd4a61 | 343 | #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2) |
c8244935 | 344 | /* Isolate for asynchronous migration */ |
f3fd4a61 | 345 | #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4) |
e46a2879 MK |
346 | /* Isolate unevictable pages */ |
347 | #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8) | |
4356f21d MK |
348 | |
349 | /* LRU Isolation modes. */ | |
9efeccac | 350 | typedef unsigned __bitwise isolate_mode_t; |
4356f21d | 351 | |
41858966 MG |
352 | enum zone_watermarks { |
353 | WMARK_MIN, | |
354 | WMARK_LOW, | |
355 | WMARK_HIGH, | |
c574bbe9 | 356 | WMARK_PROMO, |
41858966 MG |
357 | NR_WMARK |
358 | }; | |
359 | ||
44042b44 | 360 | /* |
5d0a661d MG |
361 | * One per migratetype for each PAGE_ALLOC_COSTLY_ORDER. One additional list |
362 | * for THP which will usually be GFP_MOVABLE. Even if it is another type, | |
363 | * it should not contribute to serious fragmentation causing THP allocation | |
364 | * failures. | |
44042b44 MG |
365 | */ |
366 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
367 | #define NR_PCP_THP 1 | |
368 | #else | |
369 | #define NR_PCP_THP 0 | |
370 | #endif | |
5d0a661d MG |
371 | #define NR_LOWORDER_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1)) |
372 | #define NR_PCP_LISTS (NR_LOWORDER_PCP_LISTS + NR_PCP_THP) | |
44042b44 MG |
373 | |
374 | /* | |
375 | * Shift to encode migratetype and order in the same integer, with order | |
376 | * in the least significant bits. | |
377 | */ | |
378 | #define NR_PCP_ORDER_WIDTH 8 | |
379 | #define NR_PCP_ORDER_MASK ((1<<NR_PCP_ORDER_WIDTH) - 1) | |
380 | ||
1c30844d MG |
381 | #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost) |
382 | #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost) | |
383 | #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost) | |
384 | #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost) | |
41858966 | 385 | |
dbbee9d5 | 386 | /* Fields and list protected by pagesets local_lock in page_alloc.c */ |
1da177e4 | 387 | struct per_cpu_pages { |
4b23a68f | 388 | spinlock_t lock; /* Protects lists field */ |
1da177e4 | 389 | int count; /* number of pages in the list */ |
1da177e4 LT |
390 | int high; /* high watermark, emptying needed */ |
391 | int batch; /* chunk size for buddy add/remove */ | |
3b12e7e9 | 392 | short free_factor; /* batch scaling factor during free */ |
28f836b6 | 393 | #ifdef CONFIG_NUMA |
3b12e7e9 | 394 | short expire; /* When 0, remote pagesets are drained */ |
28f836b6 | 395 | #endif |
5f8dcc21 MG |
396 | |
397 | /* Lists of pages, one per migrate type stored on the pcp-lists */ | |
44042b44 | 398 | struct list_head lists[NR_PCP_LISTS]; |
5d0a661d | 399 | } ____cacheline_aligned_in_smp; |
1da177e4 | 400 | |
28f836b6 | 401 | struct per_cpu_zonestat { |
2244b95a CL |
402 | #ifdef CONFIG_SMP |
403 | s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS]; | |
28f836b6 MG |
404 | s8 stat_threshold; |
405 | #endif | |
406 | #ifdef CONFIG_NUMA | |
f19298b9 MG |
407 | /* |
408 | * Low priority inaccurate counters that are only folded | |
409 | * on demand. Use a large type to avoid the overhead of | |
410 | * folding during refresh_cpu_vm_stats. | |
411 | */ | |
412 | unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS]; | |
2244b95a | 413 | #endif |
99dcc3e5 | 414 | }; |
e7c8d5c9 | 415 | |
75ef7184 MG |
416 | struct per_cpu_nodestat { |
417 | s8 stat_threshold; | |
418 | s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS]; | |
419 | }; | |
420 | ||
97965478 CL |
421 | #endif /* !__GENERATING_BOUNDS.H */ |
422 | ||
2f1b6248 CL |
423 | enum zone_type { |
424 | /* | |
734f9246 NSJ |
425 | * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able |
426 | * to DMA to all of the addressable memory (ZONE_NORMAL). | |
427 | * On architectures where this area covers the whole 32 bit address | |
428 | * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller | |
429 | * DMA addressing constraints. This distinction is important as a 32bit | |
430 | * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit | |
431 | * platforms may need both zones as they support peripherals with | |
432 | * different DMA addressing limitations. | |
2f1b6248 | 433 | */ |
734f9246 | 434 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 435 | ZONE_DMA, |
4b51d669 | 436 | #endif |
fb0e7942 | 437 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 438 | ZONE_DMA32, |
fb0e7942 | 439 | #endif |
2f1b6248 CL |
440 | /* |
441 | * Normal addressable memory is in ZONE_NORMAL. DMA operations can be | |
442 | * performed on pages in ZONE_NORMAL if the DMA devices support | |
443 | * transfers to all addressable memory. | |
444 | */ | |
445 | ZONE_NORMAL, | |
e53ef38d | 446 | #ifdef CONFIG_HIGHMEM |
2f1b6248 CL |
447 | /* |
448 | * A memory area that is only addressable by the kernel through | |
449 | * mapping portions into its own address space. This is for example | |
450 | * used by i386 to allow the kernel to address the memory beyond | |
451 | * 900MB. The kernel will set up special mappings (page | |
452 | * table entries on i386) for each page that the kernel needs to | |
453 | * access. | |
454 | */ | |
455 | ZONE_HIGHMEM, | |
e53ef38d | 456 | #endif |
9181a980 DH |
457 | /* |
458 | * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains | |
459 | * movable pages with few exceptional cases described below. Main use | |
460 | * cases for ZONE_MOVABLE are to make memory offlining/unplug more | |
461 | * likely to succeed, and to locally limit unmovable allocations - e.g., | |
462 | * to increase the number of THP/huge pages. Notable special cases are: | |
463 | * | |
464 | * 1. Pinned pages: (long-term) pinning of movable pages might | |
d1e153fe PT |
465 | * essentially turn such pages unmovable. Therefore, we do not allow |
466 | * pinning long-term pages in ZONE_MOVABLE. When pages are pinned and | |
467 | * faulted, they come from the right zone right away. However, it is | |
468 | * still possible that address space already has pages in | |
469 | * ZONE_MOVABLE at the time when pages are pinned (i.e. user has | |
470 | * touches that memory before pinning). In such case we migrate them | |
471 | * to a different zone. When migration fails - pinning fails. | |
9181a980 DH |
472 | * 2. memblock allocations: kernelcore/movablecore setups might create |
473 | * situations where ZONE_MOVABLE contains unmovable allocations | |
474 | * after boot. Memory offlining and allocations fail early. | |
475 | * 3. Memory holes: kernelcore/movablecore setups might create very rare | |
476 | * situations where ZONE_MOVABLE contains memory holes after boot, | |
477 | * for example, if we have sections that are only partially | |
478 | * populated. Memory offlining and allocations fail early. | |
479 | * 4. PG_hwpoison pages: while poisoned pages can be skipped during | |
480 | * memory offlining, such pages cannot be allocated. | |
481 | * 5. Unmovable PG_offline pages: in paravirtualized environments, | |
482 | * hotplugged memory blocks might only partially be managed by the | |
483 | * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The | |
484 | * parts not manged by the buddy are unmovable PG_offline pages. In | |
485 | * some cases (virtio-mem), such pages can be skipped during | |
486 | * memory offlining, however, cannot be moved/allocated. These | |
487 | * techniques might use alloc_contig_range() to hide previously | |
488 | * exposed pages from the buddy again (e.g., to implement some sort | |
489 | * of memory unplug in virtio-mem). | |
9afaf30f PT |
490 | * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create |
491 | * situations where ZERO_PAGE(0) which is allocated differently | |
492 | * on different platforms may end up in a movable zone. ZERO_PAGE(0) | |
493 | * cannot be migrated. | |
a08a2ae3 OS |
494 | * 7. Memory-hotplug: when using memmap_on_memory and onlining the |
495 | * memory to the MOVABLE zone, the vmemmap pages are also placed in | |
496 | * such zone. Such pages cannot be really moved around as they are | |
497 | * self-stored in the range, but they are treated as movable when | |
498 | * the range they describe is about to be offlined. | |
9181a980 DH |
499 | * |
500 | * In general, no unmovable allocations that degrade memory offlining | |
501 | * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range()) | |
502 | * have to expect that migrating pages in ZONE_MOVABLE can fail (even | |
503 | * if has_unmovable_pages() states that there are no unmovable pages, | |
504 | * there can be false negatives). | |
505 | */ | |
2a1e274a | 506 | ZONE_MOVABLE, |
033fbae9 DW |
507 | #ifdef CONFIG_ZONE_DEVICE |
508 | ZONE_DEVICE, | |
509 | #endif | |
97965478 | 510 | __MAX_NR_ZONES |
033fbae9 | 511 | |
2f1b6248 | 512 | }; |
1da177e4 | 513 | |
97965478 CL |
514 | #ifndef __GENERATING_BOUNDS_H |
515 | ||
ed017373 YZ |
516 | #define ASYNC_AND_SYNC 2 |
517 | ||
1da177e4 | 518 | struct zone { |
3484b2de | 519 | /* Read-mostly fields */ |
41858966 MG |
520 | |
521 | /* zone watermarks, access with *_wmark_pages(zone) macros */ | |
a9214443 | 522 | unsigned long _watermark[NR_WMARK]; |
1c30844d | 523 | unsigned long watermark_boost; |
41858966 | 524 | |
0aaa29a5 MG |
525 | unsigned long nr_reserved_highatomic; |
526 | ||
1da177e4 | 527 | /* |
89903327 AM |
528 | * We don't know if the memory that we're going to allocate will be |
529 | * freeable or/and it will be released eventually, so to avoid totally | |
530 | * wasting several GB of ram we must reserve some of the lower zone | |
531 | * memory (otherwise we risk to run OOM on the lower zones despite | |
532 | * there being tons of freeable ram on the higher zones). This array is | |
533 | * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl | |
534 | * changes. | |
1da177e4 | 535 | */ |
3484b2de | 536 | long lowmem_reserve[MAX_NR_ZONES]; |
ab8fabd4 | 537 | |
e7c8d5c9 | 538 | #ifdef CONFIG_NUMA |
d5f541ed | 539 | int node; |
3484b2de | 540 | #endif |
3484b2de | 541 | struct pglist_data *zone_pgdat; |
28f836b6 MG |
542 | struct per_cpu_pages __percpu *per_cpu_pageset; |
543 | struct per_cpu_zonestat __percpu *per_cpu_zonestats; | |
952eaf81 VB |
544 | /* |
545 | * the high and batch values are copied to individual pagesets for | |
546 | * faster access | |
547 | */ | |
548 | int pageset_high; | |
549 | int pageset_batch; | |
3484b2de | 550 | |
835c134e MG |
551 | #ifndef CONFIG_SPARSEMEM |
552 | /* | |
d9c23400 | 553 | * Flags for a pageblock_nr_pages block. See pageblock-flags.h. |
835c134e MG |
554 | * In SPARSEMEM, this map is stored in struct mem_section |
555 | */ | |
556 | unsigned long *pageblock_flags; | |
557 | #endif /* CONFIG_SPARSEMEM */ | |
558 | ||
1da177e4 LT |
559 | /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */ |
560 | unsigned long zone_start_pfn; | |
561 | ||
bdc8cb98 | 562 | /* |
9feedc9d JL |
563 | * spanned_pages is the total pages spanned by the zone, including |
564 | * holes, which is calculated as: | |
565 | * spanned_pages = zone_end_pfn - zone_start_pfn; | |
bdc8cb98 | 566 | * |
9feedc9d JL |
567 | * present_pages is physical pages existing within the zone, which |
568 | * is calculated as: | |
8761e31c | 569 | * present_pages = spanned_pages - absent_pages(pages in holes); |
9feedc9d | 570 | * |
4b097002 DH |
571 | * present_early_pages is present pages existing within the zone |
572 | * located on memory available since early boot, excluding hotplugged | |
573 | * memory. | |
574 | * | |
9feedc9d JL |
575 | * managed_pages is present pages managed by the buddy system, which |
576 | * is calculated as (reserved_pages includes pages allocated by the | |
577 | * bootmem allocator): | |
578 | * managed_pages = present_pages - reserved_pages; | |
579 | * | |
3c381db1 DH |
580 | * cma pages is present pages that are assigned for CMA use |
581 | * (MIGRATE_CMA). | |
582 | * | |
9feedc9d JL |
583 | * So present_pages may be used by memory hotplug or memory power |
584 | * management logic to figure out unmanaged pages by checking | |
585 | * (present_pages - managed_pages). And managed_pages should be used | |
586 | * by page allocator and vm scanner to calculate all kinds of watermarks | |
587 | * and thresholds. | |
588 | * | |
589 | * Locking rules: | |
590 | * | |
591 | * zone_start_pfn and spanned_pages are protected by span_seqlock. | |
592 | * It is a seqlock because it has to be read outside of zone->lock, | |
593 | * and it is done in the main allocator path. But, it is written | |
594 | * quite infrequently. | |
595 | * | |
596 | * The span_seq lock is declared along with zone->lock because it is | |
bdc8cb98 DH |
597 | * frequently read in proximity to zone->lock. It's good to |
598 | * give them a chance of being in the same cacheline. | |
9feedc9d | 599 | * |
c3d5f5f0 | 600 | * Write access to present_pages at runtime should be protected by |
e8da368a YZL |
601 | * mem_hotplug_begin/done(). Any reader who can't tolerant drift of |
602 | * present_pages should use get_online_mems() to get a stable value. | |
bdc8cb98 | 603 | */ |
9705bea5 | 604 | atomic_long_t managed_pages; |
9feedc9d JL |
605 | unsigned long spanned_pages; |
606 | unsigned long present_pages; | |
4b097002 DH |
607 | #if defined(CONFIG_MEMORY_HOTPLUG) |
608 | unsigned long present_early_pages; | |
609 | #endif | |
3c381db1 DH |
610 | #ifdef CONFIG_CMA |
611 | unsigned long cma_pages; | |
612 | #endif | |
3484b2de MG |
613 | |
614 | const char *name; | |
1da177e4 | 615 | |
ad53f92e JK |
616 | #ifdef CONFIG_MEMORY_ISOLATION |
617 | /* | |
618 | * Number of isolated pageblock. It is used to solve incorrect | |
619 | * freepage counting problem due to racy retrieving migratetype | |
620 | * of pageblock. Protected by zone->lock. | |
621 | */ | |
622 | unsigned long nr_isolate_pageblock; | |
623 | #endif | |
624 | ||
3484b2de MG |
625 | #ifdef CONFIG_MEMORY_HOTPLUG |
626 | /* see spanned/present_pages for more description */ | |
627 | seqlock_t span_seqlock; | |
628 | #endif | |
629 | ||
9dcb8b68 | 630 | int initialized; |
3484b2de | 631 | |
0f661148 | 632 | /* Write-intensive fields used from the page allocator */ |
3484b2de | 633 | ZONE_PADDING(_pad1_) |
0f661148 | 634 | |
3484b2de MG |
635 | /* free areas of different sizes */ |
636 | struct free_area free_area[MAX_ORDER]; | |
637 | ||
638 | /* zone flags, see below */ | |
639 | unsigned long flags; | |
640 | ||
0f661148 | 641 | /* Primarily protects free_area */ |
a368ab67 MG |
642 | spinlock_t lock; |
643 | ||
0f661148 | 644 | /* Write-intensive fields used by compaction and vmstats. */ |
3484b2de MG |
645 | ZONE_PADDING(_pad2_) |
646 | ||
3484b2de MG |
647 | /* |
648 | * When free pages are below this point, additional steps are taken | |
649 | * when reading the number of free pages to avoid per-cpu counter | |
650 | * drift allowing watermarks to be breached | |
651 | */ | |
652 | unsigned long percpu_drift_mark; | |
653 | ||
654 | #if defined CONFIG_COMPACTION || defined CONFIG_CMA | |
655 | /* pfn where compaction free scanner should start */ | |
656 | unsigned long compact_cached_free_pfn; | |
ed017373 YZ |
657 | /* pfn where compaction migration scanner should start */ |
658 | unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC]; | |
e332f741 MG |
659 | unsigned long compact_init_migrate_pfn; |
660 | unsigned long compact_init_free_pfn; | |
3484b2de MG |
661 | #endif |
662 | ||
663 | #ifdef CONFIG_COMPACTION | |
664 | /* | |
665 | * On compaction failure, 1<<compact_defer_shift compactions | |
666 | * are skipped before trying again. The number attempted since | |
667 | * last failure is tracked with compact_considered. | |
860b3272 | 668 | * compact_order_failed is the minimum compaction failed order. |
3484b2de MG |
669 | */ |
670 | unsigned int compact_considered; | |
671 | unsigned int compact_defer_shift; | |
672 | int compact_order_failed; | |
673 | #endif | |
674 | ||
675 | #if defined CONFIG_COMPACTION || defined CONFIG_CMA | |
676 | /* Set to true when the PG_migrate_skip bits should be cleared */ | |
677 | bool compact_blockskip_flush; | |
678 | #endif | |
679 | ||
7cf91a98 JK |
680 | bool contiguous; |
681 | ||
3484b2de MG |
682 | ZONE_PADDING(_pad3_) |
683 | /* Zone statistics */ | |
684 | atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS]; | |
f19298b9 | 685 | atomic_long_t vm_numa_event[NR_VM_NUMA_EVENT_ITEMS]; |
22fc6ecc | 686 | } ____cacheline_internodealigned_in_smp; |
1da177e4 | 687 | |
599d0c95 | 688 | enum pgdat_flags { |
599d0c95 | 689 | PGDAT_DIRTY, /* reclaim scanning has recently found |
d43006d5 MG |
690 | * many dirty file pages at the tail |
691 | * of the LRU. | |
692 | */ | |
599d0c95 | 693 | PGDAT_WRITEBACK, /* reclaim scanning has recently found |
283aba9f MG |
694 | * many pages under writeback |
695 | */ | |
a5f5f91d | 696 | PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */ |
57054651 | 697 | }; |
e815af95 | 698 | |
73444bc4 MG |
699 | enum zone_flags { |
700 | ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks. | |
701 | * Cleared when kswapd is woken. | |
702 | */ | |
c49c2c47 | 703 | ZONE_RECLAIM_ACTIVE, /* kswapd may be scanning the zone. */ |
73444bc4 MG |
704 | }; |
705 | ||
9705bea5 AK |
706 | static inline unsigned long zone_managed_pages(struct zone *zone) |
707 | { | |
708 | return (unsigned long)atomic_long_read(&zone->managed_pages); | |
709 | } | |
710 | ||
3c381db1 DH |
711 | static inline unsigned long zone_cma_pages(struct zone *zone) |
712 | { | |
713 | #ifdef CONFIG_CMA | |
714 | return zone->cma_pages; | |
715 | #else | |
716 | return 0; | |
717 | #endif | |
718 | } | |
719 | ||
f9228b20 | 720 | static inline unsigned long zone_end_pfn(const struct zone *zone) |
108bcc96 CS |
721 | { |
722 | return zone->zone_start_pfn + zone->spanned_pages; | |
723 | } | |
724 | ||
725 | static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn) | |
726 | { | |
727 | return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone); | |
728 | } | |
729 | ||
2a6e3ebe CS |
730 | static inline bool zone_is_initialized(struct zone *zone) |
731 | { | |
9dcb8b68 | 732 | return zone->initialized; |
2a6e3ebe CS |
733 | } |
734 | ||
735 | static inline bool zone_is_empty(struct zone *zone) | |
736 | { | |
737 | return zone->spanned_pages == 0; | |
738 | } | |
739 | ||
5bb88dc5 AS |
740 | #ifndef BUILD_VDSO32_64 |
741 | /* | |
742 | * The zone field is never updated after free_area_init_core() | |
743 | * sets it, so none of the operations on it need to be atomic. | |
744 | */ | |
745 | ||
746 | /* Page flags: | [SECTION] | [NODE] | ZONE | [LAST_CPUPID] | ... | FLAGS | */ | |
747 | #define SECTIONS_PGOFF ((sizeof(unsigned long)*8) - SECTIONS_WIDTH) | |
748 | #define NODES_PGOFF (SECTIONS_PGOFF - NODES_WIDTH) | |
749 | #define ZONES_PGOFF (NODES_PGOFF - ZONES_WIDTH) | |
750 | #define LAST_CPUPID_PGOFF (ZONES_PGOFF - LAST_CPUPID_WIDTH) | |
751 | #define KASAN_TAG_PGOFF (LAST_CPUPID_PGOFF - KASAN_TAG_WIDTH) | |
752 | ||
753 | /* | |
754 | * Define the bit shifts to access each section. For non-existent | |
755 | * sections we define the shift as 0; that plus a 0 mask ensures | |
756 | * the compiler will optimise away reference to them. | |
757 | */ | |
758 | #define SECTIONS_PGSHIFT (SECTIONS_PGOFF * (SECTIONS_WIDTH != 0)) | |
759 | #define NODES_PGSHIFT (NODES_PGOFF * (NODES_WIDTH != 0)) | |
760 | #define ZONES_PGSHIFT (ZONES_PGOFF * (ZONES_WIDTH != 0)) | |
761 | #define LAST_CPUPID_PGSHIFT (LAST_CPUPID_PGOFF * (LAST_CPUPID_WIDTH != 0)) | |
762 | #define KASAN_TAG_PGSHIFT (KASAN_TAG_PGOFF * (KASAN_TAG_WIDTH != 0)) | |
763 | ||
764 | /* NODE:ZONE or SECTION:ZONE is used to ID a zone for the buddy allocator */ | |
765 | #ifdef NODE_NOT_IN_PAGE_FLAGS | |
766 | #define ZONEID_SHIFT (SECTIONS_SHIFT + ZONES_SHIFT) | |
767 | #define ZONEID_PGOFF ((SECTIONS_PGOFF < ZONES_PGOFF) ? \ | |
768 | SECTIONS_PGOFF : ZONES_PGOFF) | |
769 | #else | |
770 | #define ZONEID_SHIFT (NODES_SHIFT + ZONES_SHIFT) | |
771 | #define ZONEID_PGOFF ((NODES_PGOFF < ZONES_PGOFF) ? \ | |
772 | NODES_PGOFF : ZONES_PGOFF) | |
773 | #endif | |
774 | ||
775 | #define ZONEID_PGSHIFT (ZONEID_PGOFF * (ZONEID_SHIFT != 0)) | |
776 | ||
777 | #define ZONES_MASK ((1UL << ZONES_WIDTH) - 1) | |
778 | #define NODES_MASK ((1UL << NODES_WIDTH) - 1) | |
779 | #define SECTIONS_MASK ((1UL << SECTIONS_WIDTH) - 1) | |
780 | #define LAST_CPUPID_MASK ((1UL << LAST_CPUPID_SHIFT) - 1) | |
781 | #define KASAN_TAG_MASK ((1UL << KASAN_TAG_WIDTH) - 1) | |
782 | #define ZONEID_MASK ((1UL << ZONEID_SHIFT) - 1) | |
783 | ||
784 | static inline enum zone_type page_zonenum(const struct page *page) | |
785 | { | |
786 | ASSERT_EXCLUSIVE_BITS(page->flags, ZONES_MASK << ZONES_PGSHIFT); | |
787 | return (page->flags >> ZONES_PGSHIFT) & ZONES_MASK; | |
788 | } | |
789 | ||
790 | static inline enum zone_type folio_zonenum(const struct folio *folio) | |
791 | { | |
792 | return page_zonenum(&folio->page); | |
793 | } | |
794 | ||
795 | #ifdef CONFIG_ZONE_DEVICE | |
796 | static inline bool is_zone_device_page(const struct page *page) | |
797 | { | |
798 | return page_zonenum(page) == ZONE_DEVICE; | |
799 | } | |
800 | extern void memmap_init_zone_device(struct zone *, unsigned long, | |
801 | unsigned long, struct dev_pagemap *); | |
802 | #else | |
803 | static inline bool is_zone_device_page(const struct page *page) | |
804 | { | |
805 | return false; | |
806 | } | |
807 | #endif | |
808 | ||
809 | static inline bool folio_is_zone_device(const struct folio *folio) | |
810 | { | |
811 | return is_zone_device_page(&folio->page); | |
812 | } | |
813 | ||
814 | static inline bool is_zone_movable_page(const struct page *page) | |
815 | { | |
816 | return page_zonenum(page) == ZONE_MOVABLE; | |
817 | } | |
818 | #endif | |
819 | ||
f1dd2cd1 MH |
820 | /* |
821 | * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty | |
822 | * intersection with the given zone | |
823 | */ | |
824 | static inline bool zone_intersects(struct zone *zone, | |
825 | unsigned long start_pfn, unsigned long nr_pages) | |
826 | { | |
827 | if (zone_is_empty(zone)) | |
828 | return false; | |
829 | if (start_pfn >= zone_end_pfn(zone) || | |
830 | start_pfn + nr_pages <= zone->zone_start_pfn) | |
831 | return false; | |
832 | ||
833 | return true; | |
834 | } | |
835 | ||
1da177e4 LT |
836 | /* |
837 | * The "priority" of VM scanning is how much of the queues we will scan in one | |
838 | * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the | |
839 | * queues ("queue_length >> 12") during an aging round. | |
840 | */ | |
841 | #define DEF_PRIORITY 12 | |
842 | ||
9276b1bc PJ |
843 | /* Maximum number of zones on a zonelist */ |
844 | #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES) | |
845 | ||
c00eb15a YB |
846 | enum { |
847 | ZONELIST_FALLBACK, /* zonelist with fallback */ | |
9276b1bc | 848 | #ifdef CONFIG_NUMA |
c00eb15a YB |
849 | /* |
850 | * The NUMA zonelists are doubled because we need zonelists that | |
851 | * restrict the allocations to a single node for __GFP_THISNODE. | |
852 | */ | |
853 | ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */ | |
9276b1bc | 854 | #endif |
c00eb15a YB |
855 | MAX_ZONELISTS |
856 | }; | |
9276b1bc | 857 | |
dd1a239f MG |
858 | /* |
859 | * This struct contains information about a zone in a zonelist. It is stored | |
860 | * here to avoid dereferences into large structures and lookups of tables | |
861 | */ | |
862 | struct zoneref { | |
863 | struct zone *zone; /* Pointer to actual zone */ | |
864 | int zone_idx; /* zone_idx(zoneref->zone) */ | |
865 | }; | |
866 | ||
1da177e4 LT |
867 | /* |
868 | * One allocation request operates on a zonelist. A zonelist | |
869 | * is a list of zones, the first one is the 'goal' of the | |
870 | * allocation, the other zones are fallback zones, in decreasing | |
871 | * priority. | |
872 | * | |
dd1a239f MG |
873 | * To speed the reading of the zonelist, the zonerefs contain the zone index |
874 | * of the entry being read. Helper functions to access information given | |
875 | * a struct zoneref are | |
876 | * | |
877 | * zonelist_zone() - Return the struct zone * for an entry in _zonerefs | |
878 | * zonelist_zone_idx() - Return the index of the zone for an entry | |
879 | * zonelist_node_idx() - Return the index of the node for an entry | |
1da177e4 LT |
880 | */ |
881 | struct zonelist { | |
dd1a239f | 882 | struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1]; |
1da177e4 LT |
883 | }; |
884 | ||
bb1c50d3 MR |
885 | /* |
886 | * The array of struct pages for flatmem. | |
887 | * It must be declared for SPARSEMEM as well because there are configurations | |
888 | * that rely on that. | |
889 | */ | |
5b99cd0e | 890 | extern struct page *mem_map; |
5b99cd0e | 891 | |
364c1eeb YS |
892 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
893 | struct deferred_split { | |
894 | spinlock_t split_queue_lock; | |
895 | struct list_head split_queue; | |
896 | unsigned long split_queue_len; | |
897 | }; | |
898 | #endif | |
899 | ||
1da177e4 | 900 | /* |
1da177e4 | 901 | * On NUMA machines, each NUMA node would have a pg_data_t to describe |
618b8c20 NB |
902 | * it's memory layout. On UMA machines there is a single pglist_data which |
903 | * describes the whole memory. | |
1da177e4 LT |
904 | * |
905 | * Memory statistics and page replacement data structures are maintained on a | |
906 | * per-zone basis. | |
907 | */ | |
1da177e4 | 908 | typedef struct pglist_data { |
496df3d3 BW |
909 | /* |
910 | * node_zones contains just the zones for THIS node. Not all of the | |
911 | * zones may be populated, but it is the full list. It is referenced by | |
912 | * this node's node_zonelists as well as other node's node_zonelists. | |
913 | */ | |
1da177e4 | 914 | struct zone node_zones[MAX_NR_ZONES]; |
496df3d3 BW |
915 | |
916 | /* | |
917 | * node_zonelists contains references to all zones in all nodes. | |
918 | * Generally the first zones will be references to this node's | |
919 | * node_zones. | |
920 | */ | |
523b9458 | 921 | struct zonelist node_zonelists[MAX_ZONELISTS]; |
496df3d3 BW |
922 | |
923 | int nr_zones; /* number of populated zones in this node */ | |
43b02ba9 | 924 | #ifdef CONFIG_FLATMEM /* means !SPARSEMEM */ |
1da177e4 | 925 | struct page *node_mem_map; |
eefa864b JK |
926 | #ifdef CONFIG_PAGE_EXTENSION |
927 | struct page_ext *node_page_ext; | |
928 | #endif | |
d41dee36 | 929 | #endif |
3a2d7fa8 | 930 | #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT) |
208d54e5 | 931 | /* |
fa004ab7 WY |
932 | * Must be held any time you expect node_start_pfn, |
933 | * node_present_pages, node_spanned_pages or nr_zones to stay constant. | |
3d060856 PT |
934 | * Also synchronizes pgdat->first_deferred_pfn during deferred page |
935 | * init. | |
208d54e5 | 936 | * |
114d4b79 | 937 | * pgdat_resize_lock() and pgdat_resize_unlock() are provided to |
3a2d7fa8 PT |
938 | * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG |
939 | * or CONFIG_DEFERRED_STRUCT_PAGE_INIT. | |
114d4b79 | 940 | * |
72c3b51b | 941 | * Nests above zone->lock and zone->span_seqlock |
208d54e5 DH |
942 | */ |
943 | spinlock_t node_size_lock; | |
944 | #endif | |
1da177e4 LT |
945 | unsigned long node_start_pfn; |
946 | unsigned long node_present_pages; /* total number of physical pages */ | |
947 | unsigned long node_spanned_pages; /* total size of physical page | |
948 | range, including holes */ | |
949 | int node_id; | |
1da177e4 | 950 | wait_queue_head_t kswapd_wait; |
5515061d | 951 | wait_queue_head_t pfmemalloc_wait; |
8cd7c588 MG |
952 | |
953 | /* workqueues for throttling reclaim for different reasons. */ | |
954 | wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE]; | |
955 | ||
956 | atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */ | |
957 | unsigned long nr_reclaim_start; /* nr pages written while throttled | |
958 | * when throttling started. */ | |
b4a0215e KW |
959 | #ifdef CONFIG_MEMORY_HOTPLUG |
960 | struct mutex kswapd_lock; | |
961 | #endif | |
962 | struct task_struct *kswapd; /* Protected by kswapd_lock */ | |
38087d9b | 963 | int kswapd_order; |
97a225e6 | 964 | enum zone_type kswapd_highest_zoneidx; |
38087d9b | 965 | |
c73322d0 JW |
966 | int kswapd_failures; /* Number of 'reclaimed == 0' runs */ |
967 | ||
698b1b30 VB |
968 | #ifdef CONFIG_COMPACTION |
969 | int kcompactd_max_order; | |
97a225e6 | 970 | enum zone_type kcompactd_highest_zoneidx; |
698b1b30 VB |
971 | wait_queue_head_t kcompactd_wait; |
972 | struct task_struct *kcompactd; | |
65d759c8 | 973 | bool proactive_compact_trigger; |
8177a420 | 974 | #endif |
281e3726 MG |
975 | /* |
976 | * This is a per-node reserve of pages that are not available | |
977 | * to userspace allocations. | |
978 | */ | |
979 | unsigned long totalreserve_pages; | |
980 | ||
a5f5f91d MG |
981 | #ifdef CONFIG_NUMA |
982 | /* | |
0a3c5772 | 983 | * node reclaim becomes active if more unmapped pages exist. |
a5f5f91d MG |
984 | */ |
985 | unsigned long min_unmapped_pages; | |
986 | unsigned long min_slab_pages; | |
987 | #endif /* CONFIG_NUMA */ | |
988 | ||
a52633d8 MG |
989 | /* Write-intensive fields used by page reclaim */ |
990 | ZONE_PADDING(_pad1_) | |
3a80a7fa MG |
991 | |
992 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
993 | /* | |
994 | * If memory initialisation on large machines is deferred then this | |
995 | * is the first PFN that needs to be initialised. | |
996 | */ | |
997 | unsigned long first_deferred_pfn; | |
998 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ | |
a3d0a918 KS |
999 | |
1000 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
364c1eeb | 1001 | struct deferred_split deferred_split_queue; |
a3d0a918 | 1002 | #endif |
75ef7184 | 1003 | |
c6833e10 HY |
1004 | #ifdef CONFIG_NUMA_BALANCING |
1005 | /* start time in ms of current promote rate limit period */ | |
1006 | unsigned int nbp_rl_start; | |
1007 | /* number of promote candidate pages at start time of current rate limit period */ | |
1008 | unsigned long nbp_rl_nr_cand; | |
c959924b HY |
1009 | /* promote threshold in ms */ |
1010 | unsigned int nbp_threshold; | |
1011 | /* start time in ms of current promote threshold adjustment period */ | |
1012 | unsigned int nbp_th_start; | |
1013 | /* | |
1014 | * number of promote candidate pages at stat time of current promote | |
1015 | * threshold adjustment period | |
1016 | */ | |
1017 | unsigned long nbp_th_nr_cand; | |
c6833e10 | 1018 | #endif |
599d0c95 | 1019 | /* Fields commonly accessed by the page reclaim scanner */ |
867e5e1d JW |
1020 | |
1021 | /* | |
1022 | * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED. | |
1023 | * | |
1024 | * Use mem_cgroup_lruvec() to look up lruvecs. | |
1025 | */ | |
1026 | struct lruvec __lruvec; | |
599d0c95 | 1027 | |
599d0c95 MG |
1028 | unsigned long flags; |
1029 | ||
1030 | ZONE_PADDING(_pad2_) | |
1031 | ||
75ef7184 MG |
1032 | /* Per-node vmstats */ |
1033 | struct per_cpu_nodestat __percpu *per_cpu_nodestats; | |
1034 | atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS]; | |
1da177e4 LT |
1035 | } pg_data_t; |
1036 | ||
1037 | #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages) | |
1038 | #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages) | |
1039 | ||
c6830c22 | 1040 | #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn) |
da3649e1 | 1041 | #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid)) |
c6830c22 | 1042 | |
da3649e1 CS |
1043 | static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat) |
1044 | { | |
1045 | return pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
1046 | } | |
1047 | ||
1048 | static inline bool pgdat_is_empty(pg_data_t *pgdat) | |
1049 | { | |
1050 | return !pgdat->node_start_pfn && !pgdat->node_spanned_pages; | |
1051 | } | |
c6830c22 | 1052 | |
208d54e5 DH |
1053 | #include <linux/memory_hotplug.h> |
1054 | ||
72675e13 | 1055 | void build_all_zonelists(pg_data_t *pgdat); |
5ecd9d40 | 1056 | void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order, |
97a225e6 | 1057 | enum zone_type highest_zoneidx); |
86a294a8 | 1058 | bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
97a225e6 | 1059 | int highest_zoneidx, unsigned int alloc_flags, |
86a294a8 | 1060 | long free_pages); |
7aeb09f9 | 1061 | bool zone_watermark_ok(struct zone *z, unsigned int order, |
97a225e6 | 1062 | unsigned long mark, int highest_zoneidx, |
c603844b | 1063 | unsigned int alloc_flags); |
7aeb09f9 | 1064 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
97a225e6 | 1065 | unsigned long mark, int highest_zoneidx); |
c1d0da83 LD |
1066 | /* |
1067 | * Memory initialization context, use to differentiate memory added by | |
1068 | * the platform statically or via memory hotplug interface. | |
1069 | */ | |
1070 | enum meminit_context { | |
1071 | MEMINIT_EARLY, | |
1072 | MEMINIT_HOTPLUG, | |
a2f3aa02 | 1073 | }; |
c1d0da83 | 1074 | |
dc0bbf3b | 1075 | extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn, |
b171e409 | 1076 | unsigned long size); |
718127cc | 1077 | |
bea8c150 | 1078 | extern void lruvec_init(struct lruvec *lruvec); |
7f5e86c2 | 1079 | |
599d0c95 | 1080 | static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec) |
7f5e86c2 | 1081 | { |
c255a458 | 1082 | #ifdef CONFIG_MEMCG |
599d0c95 | 1083 | return lruvec->pgdat; |
7f5e86c2 | 1084 | #else |
867e5e1d | 1085 | return container_of(lruvec, struct pglist_data, __lruvec); |
7f5e86c2 KK |
1086 | #endif |
1087 | } | |
1088 | ||
7aac7898 LS |
1089 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
1090 | int local_memory_node(int node_id); | |
1091 | #else | |
1092 | static inline int local_memory_node(int node_id) { return node_id; }; | |
1093 | #endif | |
1094 | ||
1da177e4 LT |
1095 | /* |
1096 | * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc. | |
1097 | */ | |
1098 | #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones) | |
1099 | ||
1f90a347 DW |
1100 | #ifdef CONFIG_ZONE_DEVICE |
1101 | static inline bool zone_is_zone_device(struct zone *zone) | |
1102 | { | |
1103 | return zone_idx(zone) == ZONE_DEVICE; | |
1104 | } | |
1105 | #else | |
1106 | static inline bool zone_is_zone_device(struct zone *zone) | |
1107 | { | |
1108 | return false; | |
1109 | } | |
1110 | #endif | |
1111 | ||
6aa303de MG |
1112 | /* |
1113 | * Returns true if a zone has pages managed by the buddy allocator. | |
1114 | * All the reclaim decisions have to use this function rather than | |
1115 | * populated_zone(). If the whole zone is reserved then we can easily | |
1116 | * end up with populated_zone() && !managed_zone(). | |
1117 | */ | |
1118 | static inline bool managed_zone(struct zone *zone) | |
1119 | { | |
9705bea5 | 1120 | return zone_managed_pages(zone); |
6aa303de MG |
1121 | } |
1122 | ||
1123 | /* Returns true if a zone has memory */ | |
1124 | static inline bool populated_zone(struct zone *zone) | |
f3fe6512 | 1125 | { |
6aa303de | 1126 | return zone->present_pages; |
f3fe6512 CK |
1127 | } |
1128 | ||
c1093b74 PT |
1129 | #ifdef CONFIG_NUMA |
1130 | static inline int zone_to_nid(struct zone *zone) | |
1131 | { | |
1132 | return zone->node; | |
1133 | } | |
1134 | ||
1135 | static inline void zone_set_nid(struct zone *zone, int nid) | |
1136 | { | |
1137 | zone->node = nid; | |
1138 | } | |
1139 | #else | |
1140 | static inline int zone_to_nid(struct zone *zone) | |
1141 | { | |
1142 | return 0; | |
1143 | } | |
1144 | ||
1145 | static inline void zone_set_nid(struct zone *zone, int nid) {} | |
1146 | #endif | |
1147 | ||
2a1e274a MG |
1148 | extern int movable_zone; |
1149 | ||
2f1b6248 | 1150 | static inline int is_highmem_idx(enum zone_type idx) |
1da177e4 | 1151 | { |
e53ef38d | 1152 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 1153 | return (idx == ZONE_HIGHMEM || |
b19bd1c9 | 1154 | (idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM)); |
e53ef38d CL |
1155 | #else |
1156 | return 0; | |
1157 | #endif | |
1da177e4 LT |
1158 | } |
1159 | ||
1da177e4 | 1160 | /** |
b4a991ec | 1161 | * is_highmem - helper function to quickly check if a struct zone is a |
1da177e4 LT |
1162 | * highmem zone or not. This is an attempt to keep references |
1163 | * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum. | |
198fba41 MR |
1164 | * @zone: pointer to struct zone variable |
1165 | * Return: 1 for a highmem zone, 0 otherwise | |
1da177e4 LT |
1166 | */ |
1167 | static inline int is_highmem(struct zone *zone) | |
1168 | { | |
29f9cb53 | 1169 | return is_highmem_idx(zone_idx(zone)); |
bb077c3f KW |
1170 | } |
1171 | ||
1172 | #ifdef CONFIG_ZONE_DMA | |
1173 | bool has_managed_dma(void); | |
e53ef38d | 1174 | #else |
bb077c3f KW |
1175 | static inline bool has_managed_dma(void) |
1176 | { | |
1177 | return false; | |
1da177e4 | 1178 | } |
bb077c3f | 1179 | #endif |
1da177e4 | 1180 | |
1da177e4 LT |
1181 | /* These two functions are used to setup the per zone pages min values */ |
1182 | struct ctl_table; | |
2374c09b | 1183 | |
32927393 CH |
1184 | int min_free_kbytes_sysctl_handler(struct ctl_table *, int, void *, size_t *, |
1185 | loff_t *); | |
1186 | int watermark_scale_factor_sysctl_handler(struct ctl_table *, int, void *, | |
1187 | size_t *, loff_t *); | |
d3cda233 | 1188 | extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES]; |
32927393 CH |
1189 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, void *, |
1190 | size_t *, loff_t *); | |
74f44822 MG |
1191 | int percpu_pagelist_high_fraction_sysctl_handler(struct ctl_table *, int, |
1192 | void *, size_t *, loff_t *); | |
9614634f | 1193 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int, |
32927393 | 1194 | void *, size_t *, loff_t *); |
0ff38490 | 1195 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int, |
32927393 CH |
1196 | void *, size_t *, loff_t *); |
1197 | int numa_zonelist_order_handler(struct ctl_table *, int, | |
1198 | void *, size_t *, loff_t *); | |
74f44822 | 1199 | extern int percpu_pagelist_high_fraction; |
f0c0b2b8 | 1200 | extern char numa_zonelist_order[]; |
c9bff3ee | 1201 | #define NUMA_ZONELIST_ORDER_LEN 16 |
f0c0b2b8 | 1202 | |
a9ee6cf5 | 1203 | #ifndef CONFIG_NUMA |
1da177e4 LT |
1204 | |
1205 | extern struct pglist_data contig_page_data; | |
351de44f MG |
1206 | static inline struct pglist_data *NODE_DATA(int nid) |
1207 | { | |
1208 | return &contig_page_data; | |
1209 | } | |
1da177e4 | 1210 | |
a9ee6cf5 | 1211 | #else /* CONFIG_NUMA */ |
1da177e4 LT |
1212 | |
1213 | #include <asm/mmzone.h> | |
1214 | ||
a9ee6cf5 | 1215 | #endif /* !CONFIG_NUMA */ |
348f8b6c | 1216 | |
95144c78 KH |
1217 | extern struct pglist_data *first_online_pgdat(void); |
1218 | extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat); | |
1219 | extern struct zone *next_zone(struct zone *zone); | |
8357f869 KH |
1220 | |
1221 | /** | |
12d15f0d | 1222 | * for_each_online_pgdat - helper macro to iterate over all online nodes |
198fba41 | 1223 | * @pgdat: pointer to a pg_data_t variable |
8357f869 KH |
1224 | */ |
1225 | #define for_each_online_pgdat(pgdat) \ | |
1226 | for (pgdat = first_online_pgdat(); \ | |
1227 | pgdat; \ | |
1228 | pgdat = next_online_pgdat(pgdat)) | |
8357f869 KH |
1229 | /** |
1230 | * for_each_zone - helper macro to iterate over all memory zones | |
198fba41 | 1231 | * @zone: pointer to struct zone variable |
8357f869 KH |
1232 | * |
1233 | * The user only needs to declare the zone variable, for_each_zone | |
1234 | * fills it in. | |
1235 | */ | |
1236 | #define for_each_zone(zone) \ | |
1237 | for (zone = (first_online_pgdat())->node_zones; \ | |
1238 | zone; \ | |
1239 | zone = next_zone(zone)) | |
1240 | ||
ee99c71c KM |
1241 | #define for_each_populated_zone(zone) \ |
1242 | for (zone = (first_online_pgdat())->node_zones; \ | |
1243 | zone; \ | |
1244 | zone = next_zone(zone)) \ | |
1245 | if (!populated_zone(zone)) \ | |
1246 | ; /* do nothing */ \ | |
1247 | else | |
1248 | ||
dd1a239f MG |
1249 | static inline struct zone *zonelist_zone(struct zoneref *zoneref) |
1250 | { | |
1251 | return zoneref->zone; | |
1252 | } | |
1253 | ||
1254 | static inline int zonelist_zone_idx(struct zoneref *zoneref) | |
1255 | { | |
1256 | return zoneref->zone_idx; | |
1257 | } | |
1258 | ||
1259 | static inline int zonelist_node_idx(struct zoneref *zoneref) | |
1260 | { | |
c1093b74 | 1261 | return zone_to_nid(zoneref->zone); |
dd1a239f MG |
1262 | } |
1263 | ||
682a3385 MG |
1264 | struct zoneref *__next_zones_zonelist(struct zoneref *z, |
1265 | enum zone_type highest_zoneidx, | |
1266 | nodemask_t *nodes); | |
1267 | ||
19770b32 MG |
1268 | /** |
1269 | * 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 | |
198fba41 MR |
1270 | * @z: The cursor used as a starting point for the search |
1271 | * @highest_zoneidx: The zone index of the highest zone to return | |
1272 | * @nodes: An optional nodemask to filter the zonelist with | |
19770b32 MG |
1273 | * |
1274 | * This function returns the next zone at or below a given zone index that is | |
1275 | * within the allowed nodemask using a cursor as the starting point for the | |
5bead2a0 MG |
1276 | * search. The zoneref returned is a cursor that represents the current zone |
1277 | * being examined. It should be advanced by one before calling | |
1278 | * next_zones_zonelist again. | |
198fba41 MR |
1279 | * |
1280 | * Return: the next zone at or below highest_zoneidx within the allowed | |
1281 | * nodemask using a cursor within a zonelist as a starting point | |
19770b32 | 1282 | */ |
682a3385 | 1283 | static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z, |
19770b32 | 1284 | enum zone_type highest_zoneidx, |
682a3385 MG |
1285 | nodemask_t *nodes) |
1286 | { | |
1287 | if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx)) | |
1288 | return z; | |
1289 | return __next_zones_zonelist(z, highest_zoneidx, nodes); | |
1290 | } | |
dd1a239f | 1291 | |
19770b32 MG |
1292 | /** |
1293 | * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist | |
198fba41 MR |
1294 | * @zonelist: The zonelist to search for a suitable zone |
1295 | * @highest_zoneidx: The zone index of the highest zone to return | |
1296 | * @nodes: An optional nodemask to filter the zonelist with | |
19770b32 MG |
1297 | * |
1298 | * This function returns the first zone at or below a given zone index that is | |
1299 | * within the allowed nodemask. The zoneref returned is a cursor that can be | |
5bead2a0 MG |
1300 | * used to iterate the zonelist with next_zones_zonelist by advancing it by |
1301 | * one before calling. | |
ea57485a VB |
1302 | * |
1303 | * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is | |
1304 | * never NULL). This may happen either genuinely, or due to concurrent nodemask | |
1305 | * update due to cpuset modification. | |
198fba41 MR |
1306 | * |
1307 | * Return: Zoneref pointer for the first suitable zone found | |
19770b32 | 1308 | */ |
dd1a239f | 1309 | static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist, |
19770b32 | 1310 | enum zone_type highest_zoneidx, |
c33d6c06 | 1311 | nodemask_t *nodes) |
54a6eb5c | 1312 | { |
c33d6c06 | 1313 | return next_zones_zonelist(zonelist->_zonerefs, |
05891fb0 | 1314 | highest_zoneidx, nodes); |
54a6eb5c MG |
1315 | } |
1316 | ||
19770b32 MG |
1317 | /** |
1318 | * 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 | |
198fba41 MR |
1319 | * @zone: The current zone in the iterator |
1320 | * @z: The current pointer within zonelist->_zonerefs being iterated | |
1321 | * @zlist: The zonelist being iterated | |
1322 | * @highidx: The zone index of the highest zone to return | |
1323 | * @nodemask: Nodemask allowed by the allocator | |
19770b32 MG |
1324 | * |
1325 | * This iterator iterates though all zones at or below a given zone index and | |
1326 | * within a given nodemask | |
1327 | */ | |
1328 | #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \ | |
c33d6c06 | 1329 | for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \ |
19770b32 | 1330 | zone; \ |
05891fb0 | 1331 | z = next_zones_zonelist(++z, highidx, nodemask), \ |
c33d6c06 MG |
1332 | zone = zonelist_zone(z)) |
1333 | ||
30d8ec73 | 1334 | #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \ |
c33d6c06 MG |
1335 | for (zone = z->zone; \ |
1336 | zone; \ | |
1337 | z = next_zones_zonelist(++z, highidx, nodemask), \ | |
1338 | zone = zonelist_zone(z)) | |
1339 | ||
54a6eb5c MG |
1340 | |
1341 | /** | |
1342 | * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index | |
198fba41 MR |
1343 | * @zone: The current zone in the iterator |
1344 | * @z: The current pointer within zonelist->zones being iterated | |
1345 | * @zlist: The zonelist being iterated | |
1346 | * @highidx: The zone index of the highest zone to return | |
54a6eb5c MG |
1347 | * |
1348 | * This iterator iterates though all zones at or below a given zone index. | |
1349 | */ | |
1350 | #define for_each_zone_zonelist(zone, z, zlist, highidx) \ | |
19770b32 | 1351 | for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL) |
54a6eb5c | 1352 | |
8ca1b5a4 FT |
1353 | /* Whether the 'nodes' are all movable nodes */ |
1354 | static inline bool movable_only_nodes(nodemask_t *nodes) | |
1355 | { | |
1356 | struct zonelist *zonelist; | |
1357 | struct zoneref *z; | |
1358 | int nid; | |
1359 | ||
1360 | if (nodes_empty(*nodes)) | |
1361 | return false; | |
1362 | ||
1363 | /* | |
1364 | * We can chose arbitrary node from the nodemask to get a | |
1365 | * zonelist as they are interlinked. We just need to find | |
1366 | * at least one zone that can satisfy kernel allocations. | |
1367 | */ | |
1368 | nid = first_node(*nodes); | |
1369 | zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK]; | |
1370 | z = first_zones_zonelist(zonelist, ZONE_NORMAL, nodes); | |
1371 | return (!z->zone) ? true : false; | |
1372 | } | |
1373 | ||
1374 | ||
d41dee36 AW |
1375 | #ifdef CONFIG_SPARSEMEM |
1376 | #include <asm/sparsemem.h> | |
1377 | #endif | |
1378 | ||
2bdaf115 AW |
1379 | #ifdef CONFIG_FLATMEM |
1380 | #define pfn_to_nid(pfn) (0) | |
1381 | #endif | |
1382 | ||
d41dee36 AW |
1383 | #ifdef CONFIG_SPARSEMEM |
1384 | ||
1385 | /* | |
d41dee36 AW |
1386 | * PA_SECTION_SHIFT physical address to/from section number |
1387 | * PFN_SECTION_SHIFT pfn to/from section number | |
1388 | */ | |
d41dee36 AW |
1389 | #define PA_SECTION_SHIFT (SECTION_SIZE_BITS) |
1390 | #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT) | |
1391 | ||
1392 | #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT) | |
1393 | ||
1394 | #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT) | |
1395 | #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1)) | |
1396 | ||
835c134e | 1397 | #define SECTION_BLOCKFLAGS_BITS \ |
d9c23400 | 1398 | ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS) |
835c134e | 1399 | |
d41dee36 AW |
1400 | #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS |
1401 | #error Allocator MAX_ORDER exceeds SECTION_SIZE | |
1402 | #endif | |
1403 | ||
1dd2bfc8 YI |
1404 | static inline unsigned long pfn_to_section_nr(unsigned long pfn) |
1405 | { | |
1406 | return pfn >> PFN_SECTION_SHIFT; | |
1407 | } | |
1408 | static inline unsigned long section_nr_to_pfn(unsigned long sec) | |
1409 | { | |
1410 | return sec << PFN_SECTION_SHIFT; | |
1411 | } | |
e3c40f37 | 1412 | |
a539f353 DK |
1413 | #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK) |
1414 | #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK) | |
1415 | ||
f1eca35a | 1416 | #define SUBSECTION_SHIFT 21 |
9ffc1d19 | 1417 | #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT) |
f1eca35a DW |
1418 | |
1419 | #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT) | |
1420 | #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT) | |
1421 | #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1)) | |
1422 | ||
1423 | #if SUBSECTION_SHIFT > SECTION_SIZE_BITS | |
1424 | #error Subsection size exceeds section size | |
1425 | #else | |
1426 | #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT)) | |
1427 | #endif | |
1428 | ||
a3619190 DW |
1429 | #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION) |
1430 | #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK) | |
1431 | ||
f1eca35a | 1432 | struct mem_section_usage { |
0a9f9f62 | 1433 | #ifdef CONFIG_SPARSEMEM_VMEMMAP |
f1eca35a | 1434 | DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION); |
0a9f9f62 | 1435 | #endif |
f1eca35a DW |
1436 | /* See declaration of similar field in struct zone */ |
1437 | unsigned long pageblock_flags[0]; | |
1438 | }; | |
1439 | ||
f46edbd1 DW |
1440 | void subsection_map_init(unsigned long pfn, unsigned long nr_pages); |
1441 | ||
d41dee36 | 1442 | struct page; |
eefa864b | 1443 | struct page_ext; |
d41dee36 | 1444 | struct mem_section { |
29751f69 AW |
1445 | /* |
1446 | * This is, logically, a pointer to an array of struct | |
1447 | * pages. However, it is stored with some other magic. | |
1448 | * (see sparse.c::sparse_init_one_section()) | |
1449 | * | |
30c253e6 AW |
1450 | * Additionally during early boot we encode node id of |
1451 | * the location of the section here to guide allocation. | |
1452 | * (see sparse.c::memory_present()) | |
1453 | * | |
29751f69 AW |
1454 | * Making it a UL at least makes someone do a cast |
1455 | * before using it wrong. | |
1456 | */ | |
1457 | unsigned long section_mem_map; | |
5c0e3066 | 1458 | |
f1eca35a | 1459 | struct mem_section_usage *usage; |
eefa864b JK |
1460 | #ifdef CONFIG_PAGE_EXTENSION |
1461 | /* | |
0c9ad804 | 1462 | * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use |
eefa864b JK |
1463 | * section. (see page_ext.h about this.) |
1464 | */ | |
1465 | struct page_ext *page_ext; | |
1466 | unsigned long pad; | |
1467 | #endif | |
55878e88 CS |
1468 | /* |
1469 | * WARNING: mem_section must be a power-of-2 in size for the | |
1470 | * calculation and use of SECTION_ROOT_MASK to make sense. | |
1471 | */ | |
d41dee36 AW |
1472 | }; |
1473 | ||
3e347261 BP |
1474 | #ifdef CONFIG_SPARSEMEM_EXTREME |
1475 | #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section)) | |
1476 | #else | |
1477 | #define SECTIONS_PER_ROOT 1 | |
1478 | #endif | |
802f192e | 1479 | |
3e347261 | 1480 | #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT) |
0faa5638 | 1481 | #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT) |
3e347261 | 1482 | #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1) |
802f192e | 1483 | |
3e347261 | 1484 | #ifdef CONFIG_SPARSEMEM_EXTREME |
83e3c487 | 1485 | extern struct mem_section **mem_section; |
802f192e | 1486 | #else |
3e347261 BP |
1487 | extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]; |
1488 | #endif | |
d41dee36 | 1489 | |
f1eca35a DW |
1490 | static inline unsigned long *section_to_usemap(struct mem_section *ms) |
1491 | { | |
1492 | return ms->usage->pageblock_flags; | |
1493 | } | |
1494 | ||
29751f69 AW |
1495 | static inline struct mem_section *__nr_to_section(unsigned long nr) |
1496 | { | |
a431dbbc WL |
1497 | unsigned long root = SECTION_NR_TO_ROOT(nr); |
1498 | ||
1499 | if (unlikely(root >= NR_SECTION_ROOTS)) | |
1500 | return NULL; | |
1501 | ||
83e3c487 | 1502 | #ifdef CONFIG_SPARSEMEM_EXTREME |
a431dbbc | 1503 | if (!mem_section || !mem_section[root]) |
83e3c487 KS |
1504 | return NULL; |
1505 | #endif | |
a431dbbc | 1506 | return &mem_section[root][nr & SECTION_ROOT_MASK]; |
29751f69 | 1507 | } |
f1eca35a | 1508 | extern size_t mem_section_usage_size(void); |
29751f69 AW |
1509 | |
1510 | /* | |
1511 | * We use the lower bits of the mem_map pointer to store | |
def9b71e PT |
1512 | * a little bit of information. The pointer is calculated |
1513 | * as mem_map - section_nr_to_pfn(pnum). The result is | |
1514 | * aligned to the minimum alignment of the two values: | |
1515 | * 1. All mem_map arrays are page-aligned. | |
1516 | * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT | |
1517 | * lowest bits. PFN_SECTION_SHIFT is arch-specific | |
1518 | * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the | |
1519 | * worst combination is powerpc with 256k pages, | |
1520 | * which results in PFN_SECTION_SHIFT equal 6. | |
ed7802dd MS |
1521 | * To sum it up, at least 6 bits are available on all architectures. |
1522 | * However, we can exceed 6 bits on some other architectures except | |
1523 | * powerpc (e.g. 15 bits are available on x86_64, 13 bits are available | |
1524 | * with the worst case of 64K pages on arm64) if we make sure the | |
1525 | * exceeded bit is not applicable to powerpc. | |
29751f69 | 1526 | */ |
ed7802dd MS |
1527 | enum { |
1528 | SECTION_MARKED_PRESENT_BIT, | |
1529 | SECTION_HAS_MEM_MAP_BIT, | |
1530 | SECTION_IS_ONLINE_BIT, | |
1531 | SECTION_IS_EARLY_BIT, | |
1532 | #ifdef CONFIG_ZONE_DEVICE | |
1533 | SECTION_TAINT_ZONE_DEVICE_BIT, | |
1534 | #endif | |
1535 | SECTION_MAP_LAST_BIT, | |
1536 | }; | |
1537 | ||
1538 | #define SECTION_MARKED_PRESENT BIT(SECTION_MARKED_PRESENT_BIT) | |
1539 | #define SECTION_HAS_MEM_MAP BIT(SECTION_HAS_MEM_MAP_BIT) | |
1540 | #define SECTION_IS_ONLINE BIT(SECTION_IS_ONLINE_BIT) | |
1541 | #define SECTION_IS_EARLY BIT(SECTION_IS_EARLY_BIT) | |
1542 | #ifdef CONFIG_ZONE_DEVICE | |
1543 | #define SECTION_TAINT_ZONE_DEVICE BIT(SECTION_TAINT_ZONE_DEVICE_BIT) | |
1544 | #endif | |
1545 | #define SECTION_MAP_MASK (~(BIT(SECTION_MAP_LAST_BIT) - 1)) | |
1546 | #define SECTION_NID_SHIFT SECTION_MAP_LAST_BIT | |
29751f69 AW |
1547 | |
1548 | static inline struct page *__section_mem_map_addr(struct mem_section *section) | |
1549 | { | |
1550 | unsigned long map = section->section_mem_map; | |
1551 | map &= SECTION_MAP_MASK; | |
1552 | return (struct page *)map; | |
1553 | } | |
1554 | ||
540557b9 | 1555 | static inline int present_section(struct mem_section *section) |
29751f69 | 1556 | { |
802f192e | 1557 | return (section && (section->section_mem_map & SECTION_MARKED_PRESENT)); |
29751f69 AW |
1558 | } |
1559 | ||
540557b9 AW |
1560 | static inline int present_section_nr(unsigned long nr) |
1561 | { | |
1562 | return present_section(__nr_to_section(nr)); | |
1563 | } | |
1564 | ||
1565 | static inline int valid_section(struct mem_section *section) | |
29751f69 | 1566 | { |
802f192e | 1567 | return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP)); |
29751f69 AW |
1568 | } |
1569 | ||
326e1b8f DW |
1570 | static inline int early_section(struct mem_section *section) |
1571 | { | |
1572 | return (section && (section->section_mem_map & SECTION_IS_EARLY)); | |
1573 | } | |
1574 | ||
29751f69 AW |
1575 | static inline int valid_section_nr(unsigned long nr) |
1576 | { | |
1577 | return valid_section(__nr_to_section(nr)); | |
1578 | } | |
1579 | ||
2d070eab MH |
1580 | static inline int online_section(struct mem_section *section) |
1581 | { | |
1582 | return (section && (section->section_mem_map & SECTION_IS_ONLINE)); | |
1583 | } | |
1584 | ||
ed7802dd | 1585 | #ifdef CONFIG_ZONE_DEVICE |
1f90a347 DW |
1586 | static inline int online_device_section(struct mem_section *section) |
1587 | { | |
1588 | unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE; | |
1589 | ||
1590 | return section && ((section->section_mem_map & flags) == flags); | |
1591 | } | |
ed7802dd MS |
1592 | #else |
1593 | static inline int online_device_section(struct mem_section *section) | |
1594 | { | |
1595 | return 0; | |
1596 | } | |
1597 | #endif | |
1f90a347 | 1598 | |
2d070eab MH |
1599 | static inline int online_section_nr(unsigned long nr) |
1600 | { | |
1601 | return online_section(__nr_to_section(nr)); | |
1602 | } | |
1603 | ||
1604 | #ifdef CONFIG_MEMORY_HOTPLUG | |
1605 | void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn); | |
2d070eab MH |
1606 | void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn); |
1607 | #endif | |
2d070eab | 1608 | |
d41dee36 AW |
1609 | static inline struct mem_section *__pfn_to_section(unsigned long pfn) |
1610 | { | |
29751f69 | 1611 | return __nr_to_section(pfn_to_section_nr(pfn)); |
d41dee36 AW |
1612 | } |
1613 | ||
2491f0a2 | 1614 | extern unsigned long __highest_present_section_nr; |
c4e1be9e | 1615 | |
f46edbd1 DW |
1616 | static inline int subsection_map_index(unsigned long pfn) |
1617 | { | |
1618 | return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION; | |
1619 | } | |
1620 | ||
1621 | #ifdef CONFIG_SPARSEMEM_VMEMMAP | |
1622 | static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn) | |
1623 | { | |
1624 | int idx = subsection_map_index(pfn); | |
1625 | ||
1626 | return test_bit(idx, ms->usage->subsection_map); | |
1627 | } | |
1628 | #else | |
1629 | static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn) | |
1630 | { | |
1631 | return 1; | |
1632 | } | |
1633 | #endif | |
1634 | ||
7b7bf499 | 1635 | #ifndef CONFIG_HAVE_ARCH_PFN_VALID |
51c656ae MR |
1636 | /** |
1637 | * pfn_valid - check if there is a valid memory map entry for a PFN | |
1638 | * @pfn: the page frame number to check | |
1639 | * | |
1640 | * Check if there is a valid memory map entry aka struct page for the @pfn. | |
1641 | * Note, that availability of the memory map entry does not imply that | |
1642 | * there is actual usable memory at that @pfn. The struct page may | |
1643 | * represent a hole or an unusable page frame. | |
1644 | * | |
1645 | * Return: 1 for PFNs that have memory map entries and 0 otherwise | |
1646 | */ | |
d41dee36 AW |
1647 | static inline int pfn_valid(unsigned long pfn) |
1648 | { | |
f46edbd1 DW |
1649 | struct mem_section *ms; |
1650 | ||
16c9afc7 AK |
1651 | /* |
1652 | * Ensure the upper PAGE_SHIFT bits are clear in the | |
1653 | * pfn. Else it might lead to false positives when | |
1654 | * some of the upper bits are set, but the lower bits | |
1655 | * match a valid pfn. | |
1656 | */ | |
1657 | if (PHYS_PFN(PFN_PHYS(pfn)) != pfn) | |
1658 | return 0; | |
1659 | ||
d41dee36 AW |
1660 | if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) |
1661 | return 0; | |
f1dc0db2 | 1662 | ms = __pfn_to_section(pfn); |
f46edbd1 DW |
1663 | if (!valid_section(ms)) |
1664 | return 0; | |
1665 | /* | |
1666 | * Traditionally early sections always returned pfn_valid() for | |
1667 | * the entire section-sized span. | |
1668 | */ | |
1669 | return early_section(ms) || pfn_section_valid(ms, pfn); | |
d41dee36 | 1670 | } |
7b7bf499 | 1671 | #endif |
d41dee36 | 1672 | |
e03d1f78 | 1673 | static inline int pfn_in_present_section(unsigned long pfn) |
540557b9 AW |
1674 | { |
1675 | if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) | |
1676 | return 0; | |
f1dc0db2 | 1677 | return present_section(__pfn_to_section(pfn)); |
540557b9 AW |
1678 | } |
1679 | ||
4c605881 DH |
1680 | static inline unsigned long next_present_section_nr(unsigned long section_nr) |
1681 | { | |
1682 | while (++section_nr <= __highest_present_section_nr) { | |
1683 | if (present_section_nr(section_nr)) | |
1684 | return section_nr; | |
1685 | } | |
1686 | ||
1687 | return -1; | |
1688 | } | |
1689 | ||
d41dee36 AW |
1690 | /* |
1691 | * These are _only_ used during initialisation, therefore they | |
1692 | * can use __initdata ... They could have names to indicate | |
1693 | * this restriction. | |
1694 | */ | |
1695 | #ifdef CONFIG_NUMA | |
161599ff AW |
1696 | #define pfn_to_nid(pfn) \ |
1697 | ({ \ | |
1698 | unsigned long __pfn_to_nid_pfn = (pfn); \ | |
1699 | page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \ | |
1700 | }) | |
2bdaf115 AW |
1701 | #else |
1702 | #define pfn_to_nid(pfn) (0) | |
d41dee36 AW |
1703 | #endif |
1704 | ||
d41dee36 AW |
1705 | void sparse_init(void); |
1706 | #else | |
1707 | #define sparse_init() do {} while (0) | |
28ae55c9 | 1708 | #define sparse_index_init(_sec, _nid) do {} while (0) |
e03d1f78 | 1709 | #define pfn_in_present_section pfn_valid |
f46edbd1 | 1710 | #define subsection_map_init(_pfn, _nr_pages) do {} while (0) |
d41dee36 AW |
1711 | #endif /* CONFIG_SPARSEMEM */ |
1712 | ||
97965478 | 1713 | #endif /* !__GENERATING_BOUNDS.H */ |
1da177e4 | 1714 | #endif /* !__ASSEMBLY__ */ |
1da177e4 | 1715 | #endif /* _LINUX_MMZONE_H */ |