Line data Source code
1 : /* SPDX-License-Identifier: GPL-2.0 */
2 : #ifndef _LINUX_MMZONE_H
3 : #define _LINUX_MMZONE_H
4 :
5 : #ifndef __ASSEMBLY__
6 : #ifndef __GENERATING_BOUNDS_H
7 :
8 : #include <linux/spinlock.h>
9 : #include <linux/list.h>
10 : #include <linux/list_nulls.h>
11 : #include <linux/wait.h>
12 : #include <linux/bitops.h>
13 : #include <linux/cache.h>
14 : #include <linux/threads.h>
15 : #include <linux/numa.h>
16 : #include <linux/init.h>
17 : #include <linux/seqlock.h>
18 : #include <linux/nodemask.h>
19 : #include <linux/pageblock-flags.h>
20 : #include <linux/page-flags-layout.h>
21 : #include <linux/atomic.h>
22 : #include <linux/mm_types.h>
23 : #include <linux/page-flags.h>
24 : #include <linux/local_lock.h>
25 : #include <asm/page.h>
26 :
27 : /* Free memory management - zoned buddy allocator. */
28 : #ifndef CONFIG_ARCH_FORCE_MAX_ORDER
29 : #define MAX_ORDER 11
30 : #else
31 : #define MAX_ORDER CONFIG_ARCH_FORCE_MAX_ORDER
32 : #endif
33 : #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
34 :
35 : /*
36 : * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
37 : * costly to service. That is between allocation orders which should
38 : * coalesce naturally under reasonable reclaim pressure and those which
39 : * will not.
40 : */
41 : #define PAGE_ALLOC_COSTLY_ORDER 3
42 :
43 : enum migratetype {
44 : MIGRATE_UNMOVABLE,
45 : MIGRATE_MOVABLE,
46 : MIGRATE_RECLAIMABLE,
47 : MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
48 : MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
49 : #ifdef CONFIG_CMA
50 : /*
51 : * MIGRATE_CMA migration type is designed to mimic the way
52 : * ZONE_MOVABLE works. Only movable pages can be allocated
53 : * from MIGRATE_CMA pageblocks and page allocator never
54 : * implicitly change migration type of MIGRATE_CMA pageblock.
55 : *
56 : * The way to use it is to change migratetype of a range of
57 : * pageblocks to MIGRATE_CMA which can be done by
58 : * __free_pageblock_cma() function.
59 : */
60 : MIGRATE_CMA,
61 : #endif
62 : #ifdef CONFIG_MEMORY_ISOLATION
63 : MIGRATE_ISOLATE, /* can't allocate from here */
64 : #endif
65 : MIGRATE_TYPES
66 : };
67 :
68 : /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
69 : extern const char * const migratetype_names[MIGRATE_TYPES];
70 :
71 : #ifdef CONFIG_CMA
72 : # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
73 : # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
74 : #else
75 : # define is_migrate_cma(migratetype) false
76 : # define is_migrate_cma_page(_page) false
77 : #endif
78 :
79 : static inline bool is_migrate_movable(int mt)
80 : {
81 0 : return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
82 : }
83 :
84 : /*
85 : * Check whether a migratetype can be merged with another migratetype.
86 : *
87 : * It is only mergeable when it can fall back to other migratetypes for
88 : * allocation. See fallbacks[MIGRATE_TYPES][3] in page_alloc.c.
89 : */
90 : static inline bool migratetype_is_mergeable(int mt)
91 : {
92 : return mt < MIGRATE_PCPTYPES;
93 : }
94 :
95 : #define for_each_migratetype_order(order, type) \
96 : for (order = 0; order < MAX_ORDER; order++) \
97 : for (type = 0; type < MIGRATE_TYPES; type++)
98 :
99 : extern int page_group_by_mobility_disabled;
100 :
101 : #define MIGRATETYPE_MASK ((1UL << PB_migratetype_bits) - 1)
102 :
103 : #define get_pageblock_migratetype(page) \
104 : get_pfnblock_flags_mask(page, page_to_pfn(page), MIGRATETYPE_MASK)
105 :
106 : struct free_area {
107 : struct list_head free_list[MIGRATE_TYPES];
108 : unsigned long nr_free;
109 : };
110 :
111 : static inline struct page *get_page_from_free_area(struct free_area *area,
112 : int migratetype)
113 : {
114 12451 : return list_first_entry_or_null(&area->free_list[migratetype],
115 : struct page, lru);
116 : }
117 :
118 : static inline bool free_area_empty(struct free_area *area, int migratetype)
119 : {
120 356 : return list_empty(&area->free_list[migratetype]);
121 : }
122 :
123 : struct pglist_data;
124 :
125 : #ifdef CONFIG_NUMA
126 : enum numa_stat_item {
127 : NUMA_HIT, /* allocated in intended node */
128 : NUMA_MISS, /* allocated in non intended node */
129 : NUMA_FOREIGN, /* was intended here, hit elsewhere */
130 : NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
131 : NUMA_LOCAL, /* allocation from local node */
132 : NUMA_OTHER, /* allocation from other node */
133 : NR_VM_NUMA_EVENT_ITEMS
134 : };
135 : #else
136 : #define NR_VM_NUMA_EVENT_ITEMS 0
137 : #endif
138 :
139 : enum zone_stat_item {
140 : /* First 128 byte cacheline (assuming 64 bit words) */
141 : NR_FREE_PAGES,
142 : NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
143 : NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
144 : NR_ZONE_ACTIVE_ANON,
145 : NR_ZONE_INACTIVE_FILE,
146 : NR_ZONE_ACTIVE_FILE,
147 : NR_ZONE_UNEVICTABLE,
148 : NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
149 : NR_MLOCK, /* mlock()ed pages found and moved off LRU */
150 : /* Second 128 byte cacheline */
151 : NR_BOUNCE,
152 : #if IS_ENABLED(CONFIG_ZSMALLOC)
153 : NR_ZSPAGES, /* allocated in zsmalloc */
154 : #endif
155 : NR_FREE_CMA_PAGES,
156 : NR_VM_ZONE_STAT_ITEMS };
157 :
158 : enum node_stat_item {
159 : NR_LRU_BASE,
160 : NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
161 : NR_ACTIVE_ANON, /* " " " " " */
162 : NR_INACTIVE_FILE, /* " " " " " */
163 : NR_ACTIVE_FILE, /* " " " " " */
164 : NR_UNEVICTABLE, /* " " " " " */
165 : NR_SLAB_RECLAIMABLE_B,
166 : NR_SLAB_UNRECLAIMABLE_B,
167 : NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
168 : NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
169 : WORKINGSET_NODES,
170 : WORKINGSET_REFAULT_BASE,
171 : WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE,
172 : WORKINGSET_REFAULT_FILE,
173 : WORKINGSET_ACTIVATE_BASE,
174 : WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE,
175 : WORKINGSET_ACTIVATE_FILE,
176 : WORKINGSET_RESTORE_BASE,
177 : WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE,
178 : WORKINGSET_RESTORE_FILE,
179 : WORKINGSET_NODERECLAIM,
180 : NR_ANON_MAPPED, /* Mapped anonymous pages */
181 : NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
182 : only modified from process context */
183 : NR_FILE_PAGES,
184 : NR_FILE_DIRTY,
185 : NR_WRITEBACK,
186 : NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
187 : NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
188 : NR_SHMEM_THPS,
189 : NR_SHMEM_PMDMAPPED,
190 : NR_FILE_THPS,
191 : NR_FILE_PMDMAPPED,
192 : NR_ANON_THPS,
193 : NR_VMSCAN_WRITE,
194 : NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
195 : NR_DIRTIED, /* page dirtyings since bootup */
196 : NR_WRITTEN, /* page writings since bootup */
197 : NR_THROTTLED_WRITTEN, /* NR_WRITTEN while reclaim throttled */
198 : NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
199 : NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */
200 : NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */
201 : NR_KERNEL_STACK_KB, /* measured in KiB */
202 : #if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
203 : NR_KERNEL_SCS_KB, /* measured in KiB */
204 : #endif
205 : NR_PAGETABLE, /* used for pagetables */
206 : NR_SECONDARY_PAGETABLE, /* secondary pagetables, e.g. KVM pagetables */
207 : #ifdef CONFIG_SWAP
208 : NR_SWAPCACHE,
209 : #endif
210 : #ifdef CONFIG_NUMA_BALANCING
211 : PGPROMOTE_SUCCESS, /* promote successfully */
212 : PGPROMOTE_CANDIDATE, /* candidate pages to promote */
213 : #endif
214 : NR_VM_NODE_STAT_ITEMS
215 : };
216 :
217 : /*
218 : * Returns true if the item should be printed in THPs (/proc/vmstat
219 : * currently prints number of anon, file and shmem THPs. But the item
220 : * is charged in pages).
221 : */
222 : static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item)
223 : {
224 : if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
225 : return false;
226 :
227 : return item == NR_ANON_THPS ||
228 : item == NR_FILE_THPS ||
229 : item == NR_SHMEM_THPS ||
230 : item == NR_SHMEM_PMDMAPPED ||
231 : item == NR_FILE_PMDMAPPED;
232 : }
233 :
234 : /*
235 : * Returns true if the value is measured in bytes (most vmstat values are
236 : * measured in pages). This defines the API part, the internal representation
237 : * might be different.
238 : */
239 : static __always_inline bool vmstat_item_in_bytes(int idx)
240 : {
241 : /*
242 : * Global and per-node slab counters track slab pages.
243 : * It's expected that changes are multiples of PAGE_SIZE.
244 : * Internally values are stored in pages.
245 : *
246 : * Per-memcg and per-lruvec counters track memory, consumed
247 : * by individual slab objects. These counters are actually
248 : * byte-precise.
249 : */
250 16760 : return (idx == NR_SLAB_RECLAIMABLE_B ||
251 : idx == NR_SLAB_UNRECLAIMABLE_B);
252 : }
253 :
254 : /*
255 : * We do arithmetic on the LRU lists in various places in the code,
256 : * so it is important to keep the active lists LRU_ACTIVE higher in
257 : * the array than the corresponding inactive lists, and to keep
258 : * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
259 : *
260 : * This has to be kept in sync with the statistics in zone_stat_item
261 : * above and the descriptions in vmstat_text in mm/vmstat.c
262 : */
263 : #define LRU_BASE 0
264 : #define LRU_ACTIVE 1
265 : #define LRU_FILE 2
266 :
267 : enum lru_list {
268 : LRU_INACTIVE_ANON = LRU_BASE,
269 : LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
270 : LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
271 : LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
272 : LRU_UNEVICTABLE,
273 : NR_LRU_LISTS
274 : };
275 :
276 : enum vmscan_throttle_state {
277 : VMSCAN_THROTTLE_WRITEBACK,
278 : VMSCAN_THROTTLE_ISOLATED,
279 : VMSCAN_THROTTLE_NOPROGRESS,
280 : VMSCAN_THROTTLE_CONGESTED,
281 : NR_VMSCAN_THROTTLE,
282 : };
283 :
284 : #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
285 :
286 : #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
287 :
288 : static inline bool is_file_lru(enum lru_list lru)
289 : {
290 0 : return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
291 : }
292 :
293 : static inline bool is_active_lru(enum lru_list lru)
294 : {
295 0 : return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
296 : }
297 :
298 : #define WORKINGSET_ANON 0
299 : #define WORKINGSET_FILE 1
300 : #define ANON_AND_FILE 2
301 :
302 : enum lruvec_flags {
303 : LRUVEC_CONGESTED, /* lruvec has many dirty pages
304 : * backed by a congested BDI
305 : */
306 : };
307 :
308 : #endif /* !__GENERATING_BOUNDS_H */
309 :
310 : /*
311 : * Evictable pages are divided into multiple generations. The youngest and the
312 : * oldest generation numbers, max_seq and min_seq, are monotonically increasing.
313 : * They form a sliding window of a variable size [MIN_NR_GENS, MAX_NR_GENS]. An
314 : * offset within MAX_NR_GENS, i.e., gen, indexes the LRU list of the
315 : * corresponding generation. The gen counter in folio->flags stores gen+1 while
316 : * a page is on one of lrugen->folios[]. Otherwise it stores 0.
317 : *
318 : * A page is added to the youngest generation on faulting. The aging needs to
319 : * check the accessed bit at least twice before handing this page over to the
320 : * eviction. The first check takes care of the accessed bit set on the initial
321 : * fault; the second check makes sure this page hasn't been used since then.
322 : * This process, AKA second chance, requires a minimum of two generations,
323 : * hence MIN_NR_GENS. And to maintain ABI compatibility with the active/inactive
324 : * LRU, e.g., /proc/vmstat, these two generations are considered active; the
325 : * rest of generations, if they exist, are considered inactive. See
326 : * lru_gen_is_active().
327 : *
328 : * PG_active is always cleared while a page is on one of lrugen->folios[] so
329 : * that the aging needs not to worry about it. And it's set again when a page
330 : * considered active is isolated for non-reclaiming purposes, e.g., migration.
331 : * See lru_gen_add_folio() and lru_gen_del_folio().
332 : *
333 : * MAX_NR_GENS is set to 4 so that the multi-gen LRU can support twice the
334 : * number of categories of the active/inactive LRU when keeping track of
335 : * accesses through page tables. This requires order_base_2(MAX_NR_GENS+1) bits
336 : * in folio->flags.
337 : */
338 : #define MIN_NR_GENS 2U
339 : #define MAX_NR_GENS 4U
340 :
341 : /*
342 : * Each generation is divided into multiple tiers. A page accessed N times
343 : * through file descriptors is in tier order_base_2(N). A page in the first tier
344 : * (N=0,1) is marked by PG_referenced unless it was faulted in through page
345 : * tables or read ahead. A page in any other tier (N>1) is marked by
346 : * PG_referenced and PG_workingset. This implies a minimum of two tiers is
347 : * supported without using additional bits in folio->flags.
348 : *
349 : * In contrast to moving across generations which requires the LRU lock, moving
350 : * across tiers only involves atomic operations on folio->flags and therefore
351 : * has a negligible cost in the buffered access path. In the eviction path,
352 : * comparisons of refaulted/(evicted+protected) from the first tier and the
353 : * rest infer whether pages accessed multiple times through file descriptors
354 : * are statistically hot and thus worth protecting.
355 : *
356 : * MAX_NR_TIERS is set to 4 so that the multi-gen LRU can support twice the
357 : * number of categories of the active/inactive LRU when keeping track of
358 : * accesses through file descriptors. This uses MAX_NR_TIERS-2 spare bits in
359 : * folio->flags.
360 : */
361 : #define MAX_NR_TIERS 4U
362 :
363 : #ifndef __GENERATING_BOUNDS_H
364 :
365 : struct lruvec;
366 : struct page_vma_mapped_walk;
367 :
368 : #define LRU_GEN_MASK ((BIT(LRU_GEN_WIDTH) - 1) << LRU_GEN_PGOFF)
369 : #define LRU_REFS_MASK ((BIT(LRU_REFS_WIDTH) - 1) << LRU_REFS_PGOFF)
370 :
371 : #ifdef CONFIG_LRU_GEN
372 :
373 : enum {
374 : LRU_GEN_ANON,
375 : LRU_GEN_FILE,
376 : };
377 :
378 : enum {
379 : LRU_GEN_CORE,
380 : LRU_GEN_MM_WALK,
381 : LRU_GEN_NONLEAF_YOUNG,
382 : NR_LRU_GEN_CAPS
383 : };
384 :
385 : #define MIN_LRU_BATCH BITS_PER_LONG
386 : #define MAX_LRU_BATCH (MIN_LRU_BATCH * 64)
387 :
388 : /* whether to keep historical stats from evicted generations */
389 : #ifdef CONFIG_LRU_GEN_STATS
390 : #define NR_HIST_GENS MAX_NR_GENS
391 : #else
392 : #define NR_HIST_GENS 1U
393 : #endif
394 :
395 : /*
396 : * The youngest generation number is stored in max_seq for both anon and file
397 : * types as they are aged on an equal footing. The oldest generation numbers are
398 : * stored in min_seq[] separately for anon and file types as clean file pages
399 : * can be evicted regardless of swap constraints.
400 : *
401 : * Normally anon and file min_seq are in sync. But if swapping is constrained,
402 : * e.g., out of swap space, file min_seq is allowed to advance and leave anon
403 : * min_seq behind.
404 : *
405 : * The number of pages in each generation is eventually consistent and therefore
406 : * can be transiently negative when reset_batch_size() is pending.
407 : */
408 : struct lru_gen_folio {
409 : /* the aging increments the youngest generation number */
410 : unsigned long max_seq;
411 : /* the eviction increments the oldest generation numbers */
412 : unsigned long min_seq[ANON_AND_FILE];
413 : /* the birth time of each generation in jiffies */
414 : unsigned long timestamps[MAX_NR_GENS];
415 : /* the multi-gen LRU lists, lazily sorted on eviction */
416 : struct list_head folios[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
417 : /* the multi-gen LRU sizes, eventually consistent */
418 : long nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
419 : /* the exponential moving average of refaulted */
420 : unsigned long avg_refaulted[ANON_AND_FILE][MAX_NR_TIERS];
421 : /* the exponential moving average of evicted+protected */
422 : unsigned long avg_total[ANON_AND_FILE][MAX_NR_TIERS];
423 : /* the first tier doesn't need protection, hence the minus one */
424 : unsigned long protected[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS - 1];
425 : /* can be modified without holding the LRU lock */
426 : atomic_long_t evicted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
427 : atomic_long_t refaulted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
428 : /* whether the multi-gen LRU is enabled */
429 : bool enabled;
430 : #ifdef CONFIG_MEMCG
431 : /* the memcg generation this lru_gen_folio belongs to */
432 : u8 gen;
433 : /* the list segment this lru_gen_folio belongs to */
434 : u8 seg;
435 : /* per-node lru_gen_folio list for global reclaim */
436 : struct hlist_nulls_node list;
437 : #endif
438 : };
439 :
440 : enum {
441 : MM_LEAF_TOTAL, /* total leaf entries */
442 : MM_LEAF_OLD, /* old leaf entries */
443 : MM_LEAF_YOUNG, /* young leaf entries */
444 : MM_NONLEAF_TOTAL, /* total non-leaf entries */
445 : MM_NONLEAF_FOUND, /* non-leaf entries found in Bloom filters */
446 : MM_NONLEAF_ADDED, /* non-leaf entries added to Bloom filters */
447 : NR_MM_STATS
448 : };
449 :
450 : /* double-buffering Bloom filters */
451 : #define NR_BLOOM_FILTERS 2
452 :
453 : struct lru_gen_mm_state {
454 : /* set to max_seq after each iteration */
455 : unsigned long seq;
456 : /* where the current iteration continues (inclusive) */
457 : struct list_head *head;
458 : /* where the last iteration ended (exclusive) */
459 : struct list_head *tail;
460 : /* to wait for the last page table walker to finish */
461 : struct wait_queue_head wait;
462 : /* Bloom filters flip after each iteration */
463 : unsigned long *filters[NR_BLOOM_FILTERS];
464 : /* the mm stats for debugging */
465 : unsigned long stats[NR_HIST_GENS][NR_MM_STATS];
466 : /* the number of concurrent page table walkers */
467 : int nr_walkers;
468 : };
469 :
470 : struct lru_gen_mm_walk {
471 : /* the lruvec under reclaim */
472 : struct lruvec *lruvec;
473 : /* unstable max_seq from lru_gen_folio */
474 : unsigned long max_seq;
475 : /* the next address within an mm to scan */
476 : unsigned long next_addr;
477 : /* to batch promoted pages */
478 : int nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
479 : /* to batch the mm stats */
480 : int mm_stats[NR_MM_STATS];
481 : /* total batched items */
482 : int batched;
483 : bool can_swap;
484 : bool force_scan;
485 : };
486 :
487 : void lru_gen_init_lruvec(struct lruvec *lruvec);
488 : void lru_gen_look_around(struct page_vma_mapped_walk *pvmw);
489 :
490 : #ifdef CONFIG_MEMCG
491 :
492 : /*
493 : * For each node, memcgs are divided into two generations: the old and the
494 : * young. For each generation, memcgs are randomly sharded into multiple bins
495 : * to improve scalability. For each bin, the hlist_nulls is virtually divided
496 : * into three segments: the head, the tail and the default.
497 : *
498 : * An onlining memcg is added to the tail of a random bin in the old generation.
499 : * The eviction starts at the head of a random bin in the old generation. The
500 : * per-node memcg generation counter, whose reminder (mod MEMCG_NR_GENS) indexes
501 : * the old generation, is incremented when all its bins become empty.
502 : *
503 : * There are four operations:
504 : * 1. MEMCG_LRU_HEAD, which moves an memcg to the head of a random bin in its
505 : * current generation (old or young) and updates its "seg" to "head";
506 : * 2. MEMCG_LRU_TAIL, which moves an memcg to the tail of a random bin in its
507 : * current generation (old or young) and updates its "seg" to "tail";
508 : * 3. MEMCG_LRU_OLD, which moves an memcg to the head of a random bin in the old
509 : * generation, updates its "gen" to "old" and resets its "seg" to "default";
510 : * 4. MEMCG_LRU_YOUNG, which moves an memcg to the tail of a random bin in the
511 : * young generation, updates its "gen" to "young" and resets its "seg" to
512 : * "default".
513 : *
514 : * The events that trigger the above operations are:
515 : * 1. Exceeding the soft limit, which triggers MEMCG_LRU_HEAD;
516 : * 2. The first attempt to reclaim an memcg below low, which triggers
517 : * MEMCG_LRU_TAIL;
518 : * 3. The first attempt to reclaim an memcg below reclaimable size threshold,
519 : * which triggers MEMCG_LRU_TAIL;
520 : * 4. The second attempt to reclaim an memcg below reclaimable size threshold,
521 : * which triggers MEMCG_LRU_YOUNG;
522 : * 5. Attempting to reclaim an memcg below min, which triggers MEMCG_LRU_YOUNG;
523 : * 6. Finishing the aging on the eviction path, which triggers MEMCG_LRU_YOUNG;
524 : * 7. Offlining an memcg, which triggers MEMCG_LRU_OLD.
525 : *
526 : * Note that memcg LRU only applies to global reclaim, and the round-robin
527 : * incrementing of their max_seq counters ensures the eventual fairness to all
528 : * eligible memcgs. For memcg reclaim, it still relies on mem_cgroup_iter().
529 : */
530 : #define MEMCG_NR_GENS 2
531 : #define MEMCG_NR_BINS 8
532 :
533 : struct lru_gen_memcg {
534 : /* the per-node memcg generation counter */
535 : unsigned long seq;
536 : /* each memcg has one lru_gen_folio per node */
537 : unsigned long nr_memcgs[MEMCG_NR_GENS];
538 : /* per-node lru_gen_folio list for global reclaim */
539 : struct hlist_nulls_head fifo[MEMCG_NR_GENS][MEMCG_NR_BINS];
540 : /* protects the above */
541 : spinlock_t lock;
542 : };
543 :
544 : void lru_gen_init_pgdat(struct pglist_data *pgdat);
545 :
546 : void lru_gen_init_memcg(struct mem_cgroup *memcg);
547 : void lru_gen_exit_memcg(struct mem_cgroup *memcg);
548 : void lru_gen_online_memcg(struct mem_cgroup *memcg);
549 : void lru_gen_offline_memcg(struct mem_cgroup *memcg);
550 : void lru_gen_release_memcg(struct mem_cgroup *memcg);
551 : void lru_gen_soft_reclaim(struct lruvec *lruvec);
552 :
553 : #else /* !CONFIG_MEMCG */
554 :
555 : #define MEMCG_NR_GENS 1
556 :
557 : struct lru_gen_memcg {
558 : };
559 :
560 : static inline void lru_gen_init_pgdat(struct pglist_data *pgdat)
561 : {
562 : }
563 :
564 : #endif /* CONFIG_MEMCG */
565 :
566 : #else /* !CONFIG_LRU_GEN */
567 :
568 : static inline void lru_gen_init_pgdat(struct pglist_data *pgdat)
569 : {
570 : }
571 :
572 : static inline void lru_gen_init_lruvec(struct lruvec *lruvec)
573 : {
574 : }
575 :
576 : static inline void lru_gen_look_around(struct page_vma_mapped_walk *pvmw)
577 : {
578 : }
579 :
580 : #ifdef CONFIG_MEMCG
581 :
582 : static inline void lru_gen_init_memcg(struct mem_cgroup *memcg)
583 : {
584 : }
585 :
586 : static inline void lru_gen_exit_memcg(struct mem_cgroup *memcg)
587 : {
588 : }
589 :
590 : static inline void lru_gen_online_memcg(struct mem_cgroup *memcg)
591 : {
592 : }
593 :
594 : static inline void lru_gen_offline_memcg(struct mem_cgroup *memcg)
595 : {
596 : }
597 :
598 : static inline void lru_gen_release_memcg(struct mem_cgroup *memcg)
599 : {
600 : }
601 :
602 : static inline void lru_gen_soft_reclaim(struct lruvec *lruvec)
603 : {
604 : }
605 :
606 : #endif /* CONFIG_MEMCG */
607 :
608 : #endif /* CONFIG_LRU_GEN */
609 :
610 : struct lruvec {
611 : struct list_head lists[NR_LRU_LISTS];
612 : /* per lruvec lru_lock for memcg */
613 : spinlock_t lru_lock;
614 : /*
615 : * These track the cost of reclaiming one LRU - file or anon -
616 : * over the other. As the observed cost of reclaiming one LRU
617 : * increases, the reclaim scan balance tips toward the other.
618 : */
619 : unsigned long anon_cost;
620 : unsigned long file_cost;
621 : /* Non-resident age, driven by LRU movement */
622 : atomic_long_t nonresident_age;
623 : /* Refaults at the time of last reclaim cycle */
624 : unsigned long refaults[ANON_AND_FILE];
625 : /* Various lruvec state flags (enum lruvec_flags) */
626 : unsigned long flags;
627 : #ifdef CONFIG_LRU_GEN
628 : /* evictable pages divided into generations */
629 : struct lru_gen_folio lrugen;
630 : /* to concurrently iterate lru_gen_mm_list */
631 : struct lru_gen_mm_state mm_state;
632 : #endif
633 : #ifdef CONFIG_MEMCG
634 : struct pglist_data *pgdat;
635 : #endif
636 : };
637 :
638 : /* Isolate unmapped pages */
639 : #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
640 : /* Isolate for asynchronous migration */
641 : #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
642 : /* Isolate unevictable pages */
643 : #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
644 :
645 : /* LRU Isolation modes. */
646 : typedef unsigned __bitwise isolate_mode_t;
647 :
648 : enum zone_watermarks {
649 : WMARK_MIN,
650 : WMARK_LOW,
651 : WMARK_HIGH,
652 : WMARK_PROMO,
653 : NR_WMARK
654 : };
655 :
656 : /*
657 : * One per migratetype for each PAGE_ALLOC_COSTLY_ORDER. One additional list
658 : * for THP which will usually be GFP_MOVABLE. Even if it is another type,
659 : * it should not contribute to serious fragmentation causing THP allocation
660 : * failures.
661 : */
662 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
663 : #define NR_PCP_THP 1
664 : #else
665 : #define NR_PCP_THP 0
666 : #endif
667 : #define NR_LOWORDER_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1))
668 : #define NR_PCP_LISTS (NR_LOWORDER_PCP_LISTS + NR_PCP_THP)
669 :
670 : #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
671 : #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
672 : #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
673 : #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
674 :
675 : /* Fields and list protected by pagesets local_lock in page_alloc.c */
676 : struct per_cpu_pages {
677 : spinlock_t lock; /* Protects lists field */
678 : int count; /* number of pages in the list */
679 : int high; /* high watermark, emptying needed */
680 : int batch; /* chunk size for buddy add/remove */
681 : short free_factor; /* batch scaling factor during free */
682 : #ifdef CONFIG_NUMA
683 : short expire; /* When 0, remote pagesets are drained */
684 : #endif
685 :
686 : /* Lists of pages, one per migrate type stored on the pcp-lists */
687 : struct list_head lists[NR_PCP_LISTS];
688 : } ____cacheline_aligned_in_smp;
689 :
690 : struct per_cpu_zonestat {
691 : #ifdef CONFIG_SMP
692 : s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
693 : s8 stat_threshold;
694 : #endif
695 : #ifdef CONFIG_NUMA
696 : /*
697 : * Low priority inaccurate counters that are only folded
698 : * on demand. Use a large type to avoid the overhead of
699 : * folding during refresh_cpu_vm_stats.
700 : */
701 : unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
702 : #endif
703 : };
704 :
705 : struct per_cpu_nodestat {
706 : s8 stat_threshold;
707 : s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
708 : };
709 :
710 : #endif /* !__GENERATING_BOUNDS.H */
711 :
712 : enum zone_type {
713 : /*
714 : * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
715 : * to DMA to all of the addressable memory (ZONE_NORMAL).
716 : * On architectures where this area covers the whole 32 bit address
717 : * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
718 : * DMA addressing constraints. This distinction is important as a 32bit
719 : * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
720 : * platforms may need both zones as they support peripherals with
721 : * different DMA addressing limitations.
722 : */
723 : #ifdef CONFIG_ZONE_DMA
724 : ZONE_DMA,
725 : #endif
726 : #ifdef CONFIG_ZONE_DMA32
727 : ZONE_DMA32,
728 : #endif
729 : /*
730 : * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
731 : * performed on pages in ZONE_NORMAL if the DMA devices support
732 : * transfers to all addressable memory.
733 : */
734 : ZONE_NORMAL,
735 : #ifdef CONFIG_HIGHMEM
736 : /*
737 : * A memory area that is only addressable by the kernel through
738 : * mapping portions into its own address space. This is for example
739 : * used by i386 to allow the kernel to address the memory beyond
740 : * 900MB. The kernel will set up special mappings (page
741 : * table entries on i386) for each page that the kernel needs to
742 : * access.
743 : */
744 : ZONE_HIGHMEM,
745 : #endif
746 : /*
747 : * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
748 : * movable pages with few exceptional cases described below. Main use
749 : * cases for ZONE_MOVABLE are to make memory offlining/unplug more
750 : * likely to succeed, and to locally limit unmovable allocations - e.g.,
751 : * to increase the number of THP/huge pages. Notable special cases are:
752 : *
753 : * 1. Pinned pages: (long-term) pinning of movable pages might
754 : * essentially turn such pages unmovable. Therefore, we do not allow
755 : * pinning long-term pages in ZONE_MOVABLE. When pages are pinned and
756 : * faulted, they come from the right zone right away. However, it is
757 : * still possible that address space already has pages in
758 : * ZONE_MOVABLE at the time when pages are pinned (i.e. user has
759 : * touches that memory before pinning). In such case we migrate them
760 : * to a different zone. When migration fails - pinning fails.
761 : * 2. memblock allocations: kernelcore/movablecore setups might create
762 : * situations where ZONE_MOVABLE contains unmovable allocations
763 : * after boot. Memory offlining and allocations fail early.
764 : * 3. Memory holes: kernelcore/movablecore setups might create very rare
765 : * situations where ZONE_MOVABLE contains memory holes after boot,
766 : * for example, if we have sections that are only partially
767 : * populated. Memory offlining and allocations fail early.
768 : * 4. PG_hwpoison pages: while poisoned pages can be skipped during
769 : * memory offlining, such pages cannot be allocated.
770 : * 5. Unmovable PG_offline pages: in paravirtualized environments,
771 : * hotplugged memory blocks might only partially be managed by the
772 : * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
773 : * parts not manged by the buddy are unmovable PG_offline pages. In
774 : * some cases (virtio-mem), such pages can be skipped during
775 : * memory offlining, however, cannot be moved/allocated. These
776 : * techniques might use alloc_contig_range() to hide previously
777 : * exposed pages from the buddy again (e.g., to implement some sort
778 : * of memory unplug in virtio-mem).
779 : * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create
780 : * situations where ZERO_PAGE(0) which is allocated differently
781 : * on different platforms may end up in a movable zone. ZERO_PAGE(0)
782 : * cannot be migrated.
783 : * 7. Memory-hotplug: when using memmap_on_memory and onlining the
784 : * memory to the MOVABLE zone, the vmemmap pages are also placed in
785 : * such zone. Such pages cannot be really moved around as they are
786 : * self-stored in the range, but they are treated as movable when
787 : * the range they describe is about to be offlined.
788 : *
789 : * In general, no unmovable allocations that degrade memory offlining
790 : * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
791 : * have to expect that migrating pages in ZONE_MOVABLE can fail (even
792 : * if has_unmovable_pages() states that there are no unmovable pages,
793 : * there can be false negatives).
794 : */
795 : ZONE_MOVABLE,
796 : #ifdef CONFIG_ZONE_DEVICE
797 : ZONE_DEVICE,
798 : #endif
799 : __MAX_NR_ZONES
800 :
801 : };
802 :
803 : #ifndef __GENERATING_BOUNDS_H
804 :
805 : #define ASYNC_AND_SYNC 2
806 :
807 : struct zone {
808 : /* Read-mostly fields */
809 :
810 : /* zone watermarks, access with *_wmark_pages(zone) macros */
811 : unsigned long _watermark[NR_WMARK];
812 : unsigned long watermark_boost;
813 :
814 : unsigned long nr_reserved_highatomic;
815 :
816 : /*
817 : * We don't know if the memory that we're going to allocate will be
818 : * freeable or/and it will be released eventually, so to avoid totally
819 : * wasting several GB of ram we must reserve some of the lower zone
820 : * memory (otherwise we risk to run OOM on the lower zones despite
821 : * there being tons of freeable ram on the higher zones). This array is
822 : * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
823 : * changes.
824 : */
825 : long lowmem_reserve[MAX_NR_ZONES];
826 :
827 : #ifdef CONFIG_NUMA
828 : int node;
829 : #endif
830 : struct pglist_data *zone_pgdat;
831 : struct per_cpu_pages __percpu *per_cpu_pageset;
832 : struct per_cpu_zonestat __percpu *per_cpu_zonestats;
833 : /*
834 : * the high and batch values are copied to individual pagesets for
835 : * faster access
836 : */
837 : int pageset_high;
838 : int pageset_batch;
839 :
840 : #ifndef CONFIG_SPARSEMEM
841 : /*
842 : * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
843 : * In SPARSEMEM, this map is stored in struct mem_section
844 : */
845 : unsigned long *pageblock_flags;
846 : #endif /* CONFIG_SPARSEMEM */
847 :
848 : /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
849 : unsigned long zone_start_pfn;
850 :
851 : /*
852 : * spanned_pages is the total pages spanned by the zone, including
853 : * holes, which is calculated as:
854 : * spanned_pages = zone_end_pfn - zone_start_pfn;
855 : *
856 : * present_pages is physical pages existing within the zone, which
857 : * is calculated as:
858 : * present_pages = spanned_pages - absent_pages(pages in holes);
859 : *
860 : * present_early_pages is present pages existing within the zone
861 : * located on memory available since early boot, excluding hotplugged
862 : * memory.
863 : *
864 : * managed_pages is present pages managed by the buddy system, which
865 : * is calculated as (reserved_pages includes pages allocated by the
866 : * bootmem allocator):
867 : * managed_pages = present_pages - reserved_pages;
868 : *
869 : * cma pages is present pages that are assigned for CMA use
870 : * (MIGRATE_CMA).
871 : *
872 : * So present_pages may be used by memory hotplug or memory power
873 : * management logic to figure out unmanaged pages by checking
874 : * (present_pages - managed_pages). And managed_pages should be used
875 : * by page allocator and vm scanner to calculate all kinds of watermarks
876 : * and thresholds.
877 : *
878 : * Locking rules:
879 : *
880 : * zone_start_pfn and spanned_pages are protected by span_seqlock.
881 : * It is a seqlock because it has to be read outside of zone->lock,
882 : * and it is done in the main allocator path. But, it is written
883 : * quite infrequently.
884 : *
885 : * The span_seq lock is declared along with zone->lock because it is
886 : * frequently read in proximity to zone->lock. It's good to
887 : * give them a chance of being in the same cacheline.
888 : *
889 : * Write access to present_pages at runtime should be protected by
890 : * mem_hotplug_begin/done(). Any reader who can't tolerant drift of
891 : * present_pages should use get_online_mems() to get a stable value.
892 : */
893 : atomic_long_t managed_pages;
894 : unsigned long spanned_pages;
895 : unsigned long present_pages;
896 : #if defined(CONFIG_MEMORY_HOTPLUG)
897 : unsigned long present_early_pages;
898 : #endif
899 : #ifdef CONFIG_CMA
900 : unsigned long cma_pages;
901 : #endif
902 :
903 : const char *name;
904 :
905 : #ifdef CONFIG_MEMORY_ISOLATION
906 : /*
907 : * Number of isolated pageblock. It is used to solve incorrect
908 : * freepage counting problem due to racy retrieving migratetype
909 : * of pageblock. Protected by zone->lock.
910 : */
911 : unsigned long nr_isolate_pageblock;
912 : #endif
913 :
914 : #ifdef CONFIG_MEMORY_HOTPLUG
915 : /* see spanned/present_pages for more description */
916 : seqlock_t span_seqlock;
917 : #endif
918 :
919 : int initialized;
920 :
921 : /* Write-intensive fields used from the page allocator */
922 : CACHELINE_PADDING(_pad1_);
923 :
924 : /* free areas of different sizes */
925 : struct free_area free_area[MAX_ORDER];
926 :
927 : /* zone flags, see below */
928 : unsigned long flags;
929 :
930 : /* Primarily protects free_area */
931 : spinlock_t lock;
932 :
933 : /* Write-intensive fields used by compaction and vmstats. */
934 : CACHELINE_PADDING(_pad2_);
935 :
936 : /*
937 : * When free pages are below this point, additional steps are taken
938 : * when reading the number of free pages to avoid per-cpu counter
939 : * drift allowing watermarks to be breached
940 : */
941 : unsigned long percpu_drift_mark;
942 :
943 : #if defined CONFIG_COMPACTION || defined CONFIG_CMA
944 : /* pfn where compaction free scanner should start */
945 : unsigned long compact_cached_free_pfn;
946 : /* pfn where compaction migration scanner should start */
947 : unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC];
948 : unsigned long compact_init_migrate_pfn;
949 : unsigned long compact_init_free_pfn;
950 : #endif
951 :
952 : #ifdef CONFIG_COMPACTION
953 : /*
954 : * On compaction failure, 1<<compact_defer_shift compactions
955 : * are skipped before trying again. The number attempted since
956 : * last failure is tracked with compact_considered.
957 : * compact_order_failed is the minimum compaction failed order.
958 : */
959 : unsigned int compact_considered;
960 : unsigned int compact_defer_shift;
961 : int compact_order_failed;
962 : #endif
963 :
964 : #if defined CONFIG_COMPACTION || defined CONFIG_CMA
965 : /* Set to true when the PG_migrate_skip bits should be cleared */
966 : bool compact_blockskip_flush;
967 : #endif
968 :
969 : bool contiguous;
970 :
971 : CACHELINE_PADDING(_pad3_);
972 : /* Zone statistics */
973 : atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
974 : atomic_long_t vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
975 : } ____cacheline_internodealigned_in_smp;
976 :
977 : enum pgdat_flags {
978 : PGDAT_DIRTY, /* reclaim scanning has recently found
979 : * many dirty file pages at the tail
980 : * of the LRU.
981 : */
982 : PGDAT_WRITEBACK, /* reclaim scanning has recently found
983 : * many pages under writeback
984 : */
985 : PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
986 : };
987 :
988 : enum zone_flags {
989 : ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
990 : * Cleared when kswapd is woken.
991 : */
992 : ZONE_RECLAIM_ACTIVE, /* kswapd may be scanning the zone. */
993 : };
994 :
995 : static inline unsigned long zone_managed_pages(struct zone *zone)
996 : {
997 44 : return (unsigned long)atomic_long_read(&zone->managed_pages);
998 : }
999 :
1000 : static inline unsigned long zone_cma_pages(struct zone *zone)
1001 : {
1002 : #ifdef CONFIG_CMA
1003 : return zone->cma_pages;
1004 : #else
1005 : return 0;
1006 : #endif
1007 : }
1008 :
1009 : static inline unsigned long zone_end_pfn(const struct zone *zone)
1010 : {
1011 258 : return zone->zone_start_pfn + zone->spanned_pages;
1012 : }
1013 :
1014 : static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
1015 : {
1016 0 : return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
1017 : }
1018 :
1019 : static inline bool zone_is_initialized(struct zone *zone)
1020 : {
1021 : return zone->initialized;
1022 : }
1023 :
1024 : static inline bool zone_is_empty(struct zone *zone)
1025 : {
1026 : return zone->spanned_pages == 0;
1027 : }
1028 :
1029 : #ifndef BUILD_VDSO32_64
1030 : /*
1031 : * The zone field is never updated after free_area_init_core()
1032 : * sets it, so none of the operations on it need to be atomic.
1033 : */
1034 :
1035 : /* Page flags: | [SECTION] | [NODE] | ZONE | [LAST_CPUPID] | ... | FLAGS | */
1036 : #define SECTIONS_PGOFF ((sizeof(unsigned long)*8) - SECTIONS_WIDTH)
1037 : #define NODES_PGOFF (SECTIONS_PGOFF - NODES_WIDTH)
1038 : #define ZONES_PGOFF (NODES_PGOFF - ZONES_WIDTH)
1039 : #define LAST_CPUPID_PGOFF (ZONES_PGOFF - LAST_CPUPID_WIDTH)
1040 : #define KASAN_TAG_PGOFF (LAST_CPUPID_PGOFF - KASAN_TAG_WIDTH)
1041 : #define LRU_GEN_PGOFF (KASAN_TAG_PGOFF - LRU_GEN_WIDTH)
1042 : #define LRU_REFS_PGOFF (LRU_GEN_PGOFF - LRU_REFS_WIDTH)
1043 :
1044 : /*
1045 : * Define the bit shifts to access each section. For non-existent
1046 : * sections we define the shift as 0; that plus a 0 mask ensures
1047 : * the compiler will optimise away reference to them.
1048 : */
1049 : #define SECTIONS_PGSHIFT (SECTIONS_PGOFF * (SECTIONS_WIDTH != 0))
1050 : #define NODES_PGSHIFT (NODES_PGOFF * (NODES_WIDTH != 0))
1051 : #define ZONES_PGSHIFT (ZONES_PGOFF * (ZONES_WIDTH != 0))
1052 : #define LAST_CPUPID_PGSHIFT (LAST_CPUPID_PGOFF * (LAST_CPUPID_WIDTH != 0))
1053 : #define KASAN_TAG_PGSHIFT (KASAN_TAG_PGOFF * (KASAN_TAG_WIDTH != 0))
1054 :
1055 : /* NODE:ZONE or SECTION:ZONE is used to ID a zone for the buddy allocator */
1056 : #ifdef NODE_NOT_IN_PAGE_FLAGS
1057 : #define ZONEID_SHIFT (SECTIONS_SHIFT + ZONES_SHIFT)
1058 : #define ZONEID_PGOFF ((SECTIONS_PGOFF < ZONES_PGOFF) ? \
1059 : SECTIONS_PGOFF : ZONES_PGOFF)
1060 : #else
1061 : #define ZONEID_SHIFT (NODES_SHIFT + ZONES_SHIFT)
1062 : #define ZONEID_PGOFF ((NODES_PGOFF < ZONES_PGOFF) ? \
1063 : NODES_PGOFF : ZONES_PGOFF)
1064 : #endif
1065 :
1066 : #define ZONEID_PGSHIFT (ZONEID_PGOFF * (ZONEID_SHIFT != 0))
1067 :
1068 : #define ZONES_MASK ((1UL << ZONES_WIDTH) - 1)
1069 : #define NODES_MASK ((1UL << NODES_WIDTH) - 1)
1070 : #define SECTIONS_MASK ((1UL << SECTIONS_WIDTH) - 1)
1071 : #define LAST_CPUPID_MASK ((1UL << LAST_CPUPID_SHIFT) - 1)
1072 : #define KASAN_TAG_MASK ((1UL << KASAN_TAG_WIDTH) - 1)
1073 : #define ZONEID_MASK ((1UL << ZONEID_SHIFT) - 1)
1074 :
1075 : static inline enum zone_type page_zonenum(const struct page *page)
1076 : {
1077 162380 : ASSERT_EXCLUSIVE_BITS(page->flags, ZONES_MASK << ZONES_PGSHIFT);
1078 111212 : return (page->flags >> ZONES_PGSHIFT) & ZONES_MASK;
1079 : }
1080 :
1081 : static inline enum zone_type folio_zonenum(const struct folio *folio)
1082 : {
1083 0 : return page_zonenum(&folio->page);
1084 : }
1085 :
1086 : #ifdef CONFIG_ZONE_DEVICE
1087 : static inline bool is_zone_device_page(const struct page *page)
1088 : {
1089 : return page_zonenum(page) == ZONE_DEVICE;
1090 : }
1091 :
1092 : /*
1093 : * Consecutive zone device pages should not be merged into the same sgl
1094 : * or bvec segment with other types of pages or if they belong to different
1095 : * pgmaps. Otherwise getting the pgmap of a given segment is not possible
1096 : * without scanning the entire segment. This helper returns true either if
1097 : * both pages are not zone device pages or both pages are zone device pages
1098 : * with the same pgmap.
1099 : */
1100 : static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1101 : const struct page *b)
1102 : {
1103 : if (is_zone_device_page(a) != is_zone_device_page(b))
1104 : return false;
1105 : if (!is_zone_device_page(a))
1106 : return true;
1107 : return a->pgmap == b->pgmap;
1108 : }
1109 :
1110 : extern void memmap_init_zone_device(struct zone *, unsigned long,
1111 : unsigned long, struct dev_pagemap *);
1112 : #else
1113 : static inline bool is_zone_device_page(const struct page *page)
1114 : {
1115 : return false;
1116 : }
1117 : static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1118 : const struct page *b)
1119 : {
1120 : return true;
1121 : }
1122 : #endif
1123 :
1124 : static inline bool folio_is_zone_device(const struct folio *folio)
1125 : {
1126 0 : return is_zone_device_page(&folio->page);
1127 : }
1128 :
1129 : static inline bool is_zone_movable_page(const struct page *page)
1130 : {
1131 0 : return page_zonenum(page) == ZONE_MOVABLE;
1132 : }
1133 : #endif
1134 :
1135 : /*
1136 : * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
1137 : * intersection with the given zone
1138 : */
1139 : static inline bool zone_intersects(struct zone *zone,
1140 : unsigned long start_pfn, unsigned long nr_pages)
1141 : {
1142 : if (zone_is_empty(zone))
1143 : return false;
1144 : if (start_pfn >= zone_end_pfn(zone) ||
1145 : start_pfn + nr_pages <= zone->zone_start_pfn)
1146 : return false;
1147 :
1148 : return true;
1149 : }
1150 :
1151 : /*
1152 : * The "priority" of VM scanning is how much of the queues we will scan in one
1153 : * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
1154 : * queues ("queue_length >> 12") during an aging round.
1155 : */
1156 : #define DEF_PRIORITY 12
1157 :
1158 : /* Maximum number of zones on a zonelist */
1159 : #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
1160 :
1161 : enum {
1162 : ZONELIST_FALLBACK, /* zonelist with fallback */
1163 : #ifdef CONFIG_NUMA
1164 : /*
1165 : * The NUMA zonelists are doubled because we need zonelists that
1166 : * restrict the allocations to a single node for __GFP_THISNODE.
1167 : */
1168 : ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
1169 : #endif
1170 : MAX_ZONELISTS
1171 : };
1172 :
1173 : /*
1174 : * This struct contains information about a zone in a zonelist. It is stored
1175 : * here to avoid dereferences into large structures and lookups of tables
1176 : */
1177 : struct zoneref {
1178 : struct zone *zone; /* Pointer to actual zone */
1179 : int zone_idx; /* zone_idx(zoneref->zone) */
1180 : };
1181 :
1182 : /*
1183 : * One allocation request operates on a zonelist. A zonelist
1184 : * is a list of zones, the first one is the 'goal' of the
1185 : * allocation, the other zones are fallback zones, in decreasing
1186 : * priority.
1187 : *
1188 : * To speed the reading of the zonelist, the zonerefs contain the zone index
1189 : * of the entry being read. Helper functions to access information given
1190 : * a struct zoneref are
1191 : *
1192 : * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
1193 : * zonelist_zone_idx() - Return the index of the zone for an entry
1194 : * zonelist_node_idx() - Return the index of the node for an entry
1195 : */
1196 : struct zonelist {
1197 : struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
1198 : };
1199 :
1200 : /*
1201 : * The array of struct pages for flatmem.
1202 : * It must be declared for SPARSEMEM as well because there are configurations
1203 : * that rely on that.
1204 : */
1205 : extern struct page *mem_map;
1206 :
1207 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1208 : struct deferred_split {
1209 : spinlock_t split_queue_lock;
1210 : struct list_head split_queue;
1211 : unsigned long split_queue_len;
1212 : };
1213 : #endif
1214 :
1215 : #ifdef CONFIG_MEMORY_FAILURE
1216 : /*
1217 : * Per NUMA node memory failure handling statistics.
1218 : */
1219 : struct memory_failure_stats {
1220 : /*
1221 : * Number of raw pages poisoned.
1222 : * Cases not accounted: memory outside kernel control, offline page,
1223 : * arch-specific memory_failure (SGX), hwpoison_filter() filtered
1224 : * error events, and unpoison actions from hwpoison_unpoison.
1225 : */
1226 : unsigned long total;
1227 : /*
1228 : * Recovery results of poisoned raw pages handled by memory_failure,
1229 : * in sync with mf_result.
1230 : * total = ignored + failed + delayed + recovered.
1231 : * total * PAGE_SIZE * #nodes = /proc/meminfo/HardwareCorrupted.
1232 : */
1233 : unsigned long ignored;
1234 : unsigned long failed;
1235 : unsigned long delayed;
1236 : unsigned long recovered;
1237 : };
1238 : #endif
1239 :
1240 : /*
1241 : * On NUMA machines, each NUMA node would have a pg_data_t to describe
1242 : * it's memory layout. On UMA machines there is a single pglist_data which
1243 : * describes the whole memory.
1244 : *
1245 : * Memory statistics and page replacement data structures are maintained on a
1246 : * per-zone basis.
1247 : */
1248 : typedef struct pglist_data {
1249 : /*
1250 : * node_zones contains just the zones for THIS node. Not all of the
1251 : * zones may be populated, but it is the full list. It is referenced by
1252 : * this node's node_zonelists as well as other node's node_zonelists.
1253 : */
1254 : struct zone node_zones[MAX_NR_ZONES];
1255 :
1256 : /*
1257 : * node_zonelists contains references to all zones in all nodes.
1258 : * Generally the first zones will be references to this node's
1259 : * node_zones.
1260 : */
1261 : struct zonelist node_zonelists[MAX_ZONELISTS];
1262 :
1263 : int nr_zones; /* number of populated zones in this node */
1264 : #ifdef CONFIG_FLATMEM /* means !SPARSEMEM */
1265 : struct page *node_mem_map;
1266 : #ifdef CONFIG_PAGE_EXTENSION
1267 : struct page_ext *node_page_ext;
1268 : #endif
1269 : #endif
1270 : #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
1271 : /*
1272 : * Must be held any time you expect node_start_pfn,
1273 : * node_present_pages, node_spanned_pages or nr_zones to stay constant.
1274 : * Also synchronizes pgdat->first_deferred_pfn during deferred page
1275 : * init.
1276 : *
1277 : * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
1278 : * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
1279 : * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
1280 : *
1281 : * Nests above zone->lock and zone->span_seqlock
1282 : */
1283 : spinlock_t node_size_lock;
1284 : #endif
1285 : unsigned long node_start_pfn;
1286 : unsigned long node_present_pages; /* total number of physical pages */
1287 : unsigned long node_spanned_pages; /* total size of physical page
1288 : range, including holes */
1289 : int node_id;
1290 : wait_queue_head_t kswapd_wait;
1291 : wait_queue_head_t pfmemalloc_wait;
1292 :
1293 : /* workqueues for throttling reclaim for different reasons. */
1294 : wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE];
1295 :
1296 : atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */
1297 : unsigned long nr_reclaim_start; /* nr pages written while throttled
1298 : * when throttling started. */
1299 : #ifdef CONFIG_MEMORY_HOTPLUG
1300 : struct mutex kswapd_lock;
1301 : #endif
1302 : struct task_struct *kswapd; /* Protected by kswapd_lock */
1303 : int kswapd_order;
1304 : enum zone_type kswapd_highest_zoneidx;
1305 :
1306 : int kswapd_failures; /* Number of 'reclaimed == 0' runs */
1307 :
1308 : #ifdef CONFIG_COMPACTION
1309 : int kcompactd_max_order;
1310 : enum zone_type kcompactd_highest_zoneidx;
1311 : wait_queue_head_t kcompactd_wait;
1312 : struct task_struct *kcompactd;
1313 : bool proactive_compact_trigger;
1314 : #endif
1315 : /*
1316 : * This is a per-node reserve of pages that are not available
1317 : * to userspace allocations.
1318 : */
1319 : unsigned long totalreserve_pages;
1320 :
1321 : #ifdef CONFIG_NUMA
1322 : /*
1323 : * node reclaim becomes active if more unmapped pages exist.
1324 : */
1325 : unsigned long min_unmapped_pages;
1326 : unsigned long min_slab_pages;
1327 : #endif /* CONFIG_NUMA */
1328 :
1329 : /* Write-intensive fields used by page reclaim */
1330 : CACHELINE_PADDING(_pad1_);
1331 :
1332 : #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1333 : /*
1334 : * If memory initialisation on large machines is deferred then this
1335 : * is the first PFN that needs to be initialised.
1336 : */
1337 : unsigned long first_deferred_pfn;
1338 : #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1339 :
1340 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1341 : struct deferred_split deferred_split_queue;
1342 : #endif
1343 :
1344 : #ifdef CONFIG_NUMA_BALANCING
1345 : /* start time in ms of current promote rate limit period */
1346 : unsigned int nbp_rl_start;
1347 : /* number of promote candidate pages at start time of current rate limit period */
1348 : unsigned long nbp_rl_nr_cand;
1349 : /* promote threshold in ms */
1350 : unsigned int nbp_threshold;
1351 : /* start time in ms of current promote threshold adjustment period */
1352 : unsigned int nbp_th_start;
1353 : /*
1354 : * number of promote candidate pages at start time of current promote
1355 : * threshold adjustment period
1356 : */
1357 : unsigned long nbp_th_nr_cand;
1358 : #endif
1359 : /* Fields commonly accessed by the page reclaim scanner */
1360 :
1361 : /*
1362 : * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
1363 : *
1364 : * Use mem_cgroup_lruvec() to look up lruvecs.
1365 : */
1366 : struct lruvec __lruvec;
1367 :
1368 : unsigned long flags;
1369 :
1370 : #ifdef CONFIG_LRU_GEN
1371 : /* kswap mm walk data */
1372 : struct lru_gen_mm_walk mm_walk;
1373 : /* lru_gen_folio list */
1374 : struct lru_gen_memcg memcg_lru;
1375 : #endif
1376 :
1377 : CACHELINE_PADDING(_pad2_);
1378 :
1379 : /* Per-node vmstats */
1380 : struct per_cpu_nodestat __percpu *per_cpu_nodestats;
1381 : atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
1382 : #ifdef CONFIG_NUMA
1383 : struct memory_tier __rcu *memtier;
1384 : #endif
1385 : #ifdef CONFIG_MEMORY_FAILURE
1386 : struct memory_failure_stats mf_stats;
1387 : #endif
1388 : } pg_data_t;
1389 :
1390 : #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
1391 : #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
1392 :
1393 : #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
1394 : #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
1395 :
1396 : static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
1397 : {
1398 1 : return pgdat->node_start_pfn + pgdat->node_spanned_pages;
1399 : }
1400 :
1401 : #include <linux/memory_hotplug.h>
1402 :
1403 : void build_all_zonelists(pg_data_t *pgdat);
1404 : void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
1405 : enum zone_type highest_zoneidx);
1406 : bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
1407 : int highest_zoneidx, unsigned int alloc_flags,
1408 : long free_pages);
1409 : bool zone_watermark_ok(struct zone *z, unsigned int order,
1410 : unsigned long mark, int highest_zoneidx,
1411 : unsigned int alloc_flags);
1412 : bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
1413 : unsigned long mark, int highest_zoneidx);
1414 : /*
1415 : * Memory initialization context, use to differentiate memory added by
1416 : * the platform statically or via memory hotplug interface.
1417 : */
1418 : enum meminit_context {
1419 : MEMINIT_EARLY,
1420 : MEMINIT_HOTPLUG,
1421 : };
1422 :
1423 : extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
1424 : unsigned long size);
1425 :
1426 : extern void lruvec_init(struct lruvec *lruvec);
1427 :
1428 : static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
1429 : {
1430 : #ifdef CONFIG_MEMCG
1431 : return lruvec->pgdat;
1432 : #else
1433 0 : return container_of(lruvec, struct pglist_data, __lruvec);
1434 : #endif
1435 : }
1436 :
1437 : #ifdef CONFIG_HAVE_MEMORYLESS_NODES
1438 : int local_memory_node(int node_id);
1439 : #else
1440 : static inline int local_memory_node(int node_id) { return node_id; };
1441 : #endif
1442 :
1443 : /*
1444 : * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
1445 : */
1446 : #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
1447 :
1448 : #ifdef CONFIG_ZONE_DEVICE
1449 : static inline bool zone_is_zone_device(struct zone *zone)
1450 : {
1451 : return zone_idx(zone) == ZONE_DEVICE;
1452 : }
1453 : #else
1454 : static inline bool zone_is_zone_device(struct zone *zone)
1455 : {
1456 : return false;
1457 : }
1458 : #endif
1459 :
1460 : /*
1461 : * Returns true if a zone has pages managed by the buddy allocator.
1462 : * All the reclaim decisions have to use this function rather than
1463 : * populated_zone(). If the whole zone is reserved then we can easily
1464 : * end up with populated_zone() && !managed_zone().
1465 : */
1466 : static inline bool managed_zone(struct zone *zone)
1467 : {
1468 2 : return zone_managed_pages(zone);
1469 : }
1470 :
1471 : /* Returns true if a zone has memory */
1472 : static inline bool populated_zone(struct zone *zone)
1473 : {
1474 : return zone->present_pages;
1475 : }
1476 :
1477 : #ifdef CONFIG_NUMA
1478 : static inline int zone_to_nid(struct zone *zone)
1479 : {
1480 : return zone->node;
1481 : }
1482 :
1483 : static inline void zone_set_nid(struct zone *zone, int nid)
1484 : {
1485 : zone->node = nid;
1486 : }
1487 : #else
1488 : static inline int zone_to_nid(struct zone *zone)
1489 : {
1490 : return 0;
1491 : }
1492 :
1493 : static inline void zone_set_nid(struct zone *zone, int nid) {}
1494 : #endif
1495 :
1496 : extern int movable_zone;
1497 :
1498 : static inline int is_highmem_idx(enum zone_type idx)
1499 : {
1500 : #ifdef CONFIG_HIGHMEM
1501 : return (idx == ZONE_HIGHMEM ||
1502 : (idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM));
1503 : #else
1504 : return 0;
1505 : #endif
1506 : }
1507 :
1508 : /**
1509 : * is_highmem - helper function to quickly check if a struct zone is a
1510 : * highmem zone or not. This is an attempt to keep references
1511 : * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
1512 : * @zone: pointer to struct zone variable
1513 : * Return: 1 for a highmem zone, 0 otherwise
1514 : */
1515 : static inline int is_highmem(struct zone *zone)
1516 : {
1517 4 : return is_highmem_idx(zone_idx(zone));
1518 : }
1519 :
1520 : #ifdef CONFIG_ZONE_DMA
1521 : bool has_managed_dma(void);
1522 : #else
1523 : static inline bool has_managed_dma(void)
1524 : {
1525 : return false;
1526 : }
1527 : #endif
1528 :
1529 : /* These two functions are used to setup the per zone pages min values */
1530 : struct ctl_table;
1531 :
1532 : int min_free_kbytes_sysctl_handler(struct ctl_table *, int, void *, size_t *,
1533 : loff_t *);
1534 : int watermark_scale_factor_sysctl_handler(struct ctl_table *, int, void *,
1535 : size_t *, loff_t *);
1536 : extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
1537 : int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, void *,
1538 : size_t *, loff_t *);
1539 : int percpu_pagelist_high_fraction_sysctl_handler(struct ctl_table *, int,
1540 : void *, size_t *, loff_t *);
1541 : int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
1542 : void *, size_t *, loff_t *);
1543 : int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
1544 : void *, size_t *, loff_t *);
1545 : int numa_zonelist_order_handler(struct ctl_table *, int,
1546 : void *, size_t *, loff_t *);
1547 : extern int percpu_pagelist_high_fraction;
1548 : extern char numa_zonelist_order[];
1549 : #define NUMA_ZONELIST_ORDER_LEN 16
1550 :
1551 : #ifndef CONFIG_NUMA
1552 :
1553 : extern struct pglist_data contig_page_data;
1554 : static inline struct pglist_data *NODE_DATA(int nid)
1555 : {
1556 : return &contig_page_data;
1557 : }
1558 :
1559 : #else /* CONFIG_NUMA */
1560 :
1561 : #include <asm/mmzone.h>
1562 :
1563 : #endif /* !CONFIG_NUMA */
1564 :
1565 : extern struct pglist_data *first_online_pgdat(void);
1566 : extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
1567 : extern struct zone *next_zone(struct zone *zone);
1568 :
1569 : /**
1570 : * for_each_online_pgdat - helper macro to iterate over all online nodes
1571 : * @pgdat: pointer to a pg_data_t variable
1572 : */
1573 : #define for_each_online_pgdat(pgdat) \
1574 : for (pgdat = first_online_pgdat(); \
1575 : pgdat; \
1576 : pgdat = next_online_pgdat(pgdat))
1577 : /**
1578 : * for_each_zone - helper macro to iterate over all memory zones
1579 : * @zone: pointer to struct zone variable
1580 : *
1581 : * The user only needs to declare the zone variable, for_each_zone
1582 : * fills it in.
1583 : */
1584 : #define for_each_zone(zone) \
1585 : for (zone = (first_online_pgdat())->node_zones; \
1586 : zone; \
1587 : zone = next_zone(zone))
1588 :
1589 : #define for_each_populated_zone(zone) \
1590 : for (zone = (first_online_pgdat())->node_zones; \
1591 : zone; \
1592 : zone = next_zone(zone)) \
1593 : if (!populated_zone(zone)) \
1594 : ; /* do nothing */ \
1595 : else
1596 :
1597 : static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1598 : {
1599 : return zoneref->zone;
1600 : }
1601 :
1602 : static inline int zonelist_zone_idx(struct zoneref *zoneref)
1603 : {
1604 : return zoneref->zone_idx;
1605 : }
1606 :
1607 : static inline int zonelist_node_idx(struct zoneref *zoneref)
1608 : {
1609 : return zone_to_nid(zoneref->zone);
1610 : }
1611 :
1612 : struct zoneref *__next_zones_zonelist(struct zoneref *z,
1613 : enum zone_type highest_zoneidx,
1614 : nodemask_t *nodes);
1615 :
1616 : /**
1617 : * 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
1618 : * @z: The cursor used as a starting point for the search
1619 : * @highest_zoneidx: The zone index of the highest zone to return
1620 : * @nodes: An optional nodemask to filter the zonelist with
1621 : *
1622 : * This function returns the next zone at or below a given zone index that is
1623 : * within the allowed nodemask using a cursor as the starting point for the
1624 : * search. The zoneref returned is a cursor that represents the current zone
1625 : * being examined. It should be advanced by one before calling
1626 : * next_zones_zonelist again.
1627 : *
1628 : * Return: the next zone at or below highest_zoneidx within the allowed
1629 : * nodemask using a cursor within a zonelist as a starting point
1630 : */
1631 : static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1632 : enum zone_type highest_zoneidx,
1633 : nodemask_t *nodes)
1634 : {
1635 9826 : if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1636 : return z;
1637 0 : return __next_zones_zonelist(z, highest_zoneidx, nodes);
1638 : }
1639 :
1640 : /**
1641 : * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1642 : * @zonelist: The zonelist to search for a suitable zone
1643 : * @highest_zoneidx: The zone index of the highest zone to return
1644 : * @nodes: An optional nodemask to filter the zonelist with
1645 : *
1646 : * This function returns the first zone at or below a given zone index that is
1647 : * within the allowed nodemask. The zoneref returned is a cursor that can be
1648 : * used to iterate the zonelist with next_zones_zonelist by advancing it by
1649 : * one before calling.
1650 : *
1651 : * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1652 : * never NULL). This may happen either genuinely, or due to concurrent nodemask
1653 : * update due to cpuset modification.
1654 : *
1655 : * Return: Zoneref pointer for the first suitable zone found
1656 : */
1657 : static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1658 : enum zone_type highest_zoneidx,
1659 : nodemask_t *nodes)
1660 : {
1661 19646 : return next_zones_zonelist(zonelist->_zonerefs,
1662 : highest_zoneidx, nodes);
1663 : }
1664 :
1665 : /**
1666 : * 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
1667 : * @zone: The current zone in the iterator
1668 : * @z: The current pointer within zonelist->_zonerefs being iterated
1669 : * @zlist: The zonelist being iterated
1670 : * @highidx: The zone index of the highest zone to return
1671 : * @nodemask: Nodemask allowed by the allocator
1672 : *
1673 : * This iterator iterates though all zones at or below a given zone index and
1674 : * within a given nodemask
1675 : */
1676 : #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1677 : for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1678 : zone; \
1679 : z = next_zones_zonelist(++z, highidx, nodemask), \
1680 : zone = zonelist_zone(z))
1681 :
1682 : #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
1683 : for (zone = z->zone; \
1684 : zone; \
1685 : z = next_zones_zonelist(++z, highidx, nodemask), \
1686 : zone = zonelist_zone(z))
1687 :
1688 :
1689 : /**
1690 : * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1691 : * @zone: The current zone in the iterator
1692 : * @z: The current pointer within zonelist->zones being iterated
1693 : * @zlist: The zonelist being iterated
1694 : * @highidx: The zone index of the highest zone to return
1695 : *
1696 : * This iterator iterates though all zones at or below a given zone index.
1697 : */
1698 : #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1699 : for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1700 :
1701 : /* Whether the 'nodes' are all movable nodes */
1702 : static inline bool movable_only_nodes(nodemask_t *nodes)
1703 : {
1704 : struct zonelist *zonelist;
1705 : struct zoneref *z;
1706 : int nid;
1707 :
1708 : if (nodes_empty(*nodes))
1709 : return false;
1710 :
1711 : /*
1712 : * We can chose arbitrary node from the nodemask to get a
1713 : * zonelist as they are interlinked. We just need to find
1714 : * at least one zone that can satisfy kernel allocations.
1715 : */
1716 : nid = first_node(*nodes);
1717 : zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK];
1718 : z = first_zones_zonelist(zonelist, ZONE_NORMAL, nodes);
1719 : return (!z->zone) ? true : false;
1720 : }
1721 :
1722 :
1723 : #ifdef CONFIG_SPARSEMEM
1724 : #include <asm/sparsemem.h>
1725 : #endif
1726 :
1727 : #ifdef CONFIG_FLATMEM
1728 : #define pfn_to_nid(pfn) (0)
1729 : #endif
1730 :
1731 : #ifdef CONFIG_SPARSEMEM
1732 :
1733 : /*
1734 : * PA_SECTION_SHIFT physical address to/from section number
1735 : * PFN_SECTION_SHIFT pfn to/from section number
1736 : */
1737 : #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1738 : #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1739 :
1740 : #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1741 :
1742 : #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1743 : #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1744 :
1745 : #define SECTION_BLOCKFLAGS_BITS \
1746 : ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1747 :
1748 : #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1749 : #error Allocator MAX_ORDER exceeds SECTION_SIZE
1750 : #endif
1751 :
1752 : static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1753 : {
1754 : return pfn >> PFN_SECTION_SHIFT;
1755 : }
1756 : static inline unsigned long section_nr_to_pfn(unsigned long sec)
1757 : {
1758 : return sec << PFN_SECTION_SHIFT;
1759 : }
1760 :
1761 : #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1762 : #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1763 :
1764 : #define SUBSECTION_SHIFT 21
1765 : #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1766 :
1767 : #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1768 : #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1769 : #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1770 :
1771 : #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1772 : #error Subsection size exceeds section size
1773 : #else
1774 : #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1775 : #endif
1776 :
1777 : #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1778 : #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1779 :
1780 : struct mem_section_usage {
1781 : #ifdef CONFIG_SPARSEMEM_VMEMMAP
1782 : DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1783 : #endif
1784 : /* See declaration of similar field in struct zone */
1785 : unsigned long pageblock_flags[0];
1786 : };
1787 :
1788 : void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1789 :
1790 : struct page;
1791 : struct page_ext;
1792 : struct mem_section {
1793 : /*
1794 : * This is, logically, a pointer to an array of struct
1795 : * pages. However, it is stored with some other magic.
1796 : * (see sparse.c::sparse_init_one_section())
1797 : *
1798 : * Additionally during early boot we encode node id of
1799 : * the location of the section here to guide allocation.
1800 : * (see sparse.c::memory_present())
1801 : *
1802 : * Making it a UL at least makes someone do a cast
1803 : * before using it wrong.
1804 : */
1805 : unsigned long section_mem_map;
1806 :
1807 : struct mem_section_usage *usage;
1808 : #ifdef CONFIG_PAGE_EXTENSION
1809 : /*
1810 : * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1811 : * section. (see page_ext.h about this.)
1812 : */
1813 : struct page_ext *page_ext;
1814 : unsigned long pad;
1815 : #endif
1816 : /*
1817 : * WARNING: mem_section must be a power-of-2 in size for the
1818 : * calculation and use of SECTION_ROOT_MASK to make sense.
1819 : */
1820 : };
1821 :
1822 : #ifdef CONFIG_SPARSEMEM_EXTREME
1823 : #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1824 : #else
1825 : #define SECTIONS_PER_ROOT 1
1826 : #endif
1827 :
1828 : #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1829 : #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1830 : #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1831 :
1832 : #ifdef CONFIG_SPARSEMEM_EXTREME
1833 : extern struct mem_section **mem_section;
1834 : #else
1835 : extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1836 : #endif
1837 :
1838 : static inline unsigned long *section_to_usemap(struct mem_section *ms)
1839 : {
1840 : return ms->usage->pageblock_flags;
1841 : }
1842 :
1843 : static inline struct mem_section *__nr_to_section(unsigned long nr)
1844 : {
1845 : unsigned long root = SECTION_NR_TO_ROOT(nr);
1846 :
1847 : if (unlikely(root >= NR_SECTION_ROOTS))
1848 : return NULL;
1849 :
1850 : #ifdef CONFIG_SPARSEMEM_EXTREME
1851 : if (!mem_section || !mem_section[root])
1852 : return NULL;
1853 : #endif
1854 : return &mem_section[root][nr & SECTION_ROOT_MASK];
1855 : }
1856 : extern size_t mem_section_usage_size(void);
1857 :
1858 : /*
1859 : * We use the lower bits of the mem_map pointer to store
1860 : * a little bit of information. The pointer is calculated
1861 : * as mem_map - section_nr_to_pfn(pnum). The result is
1862 : * aligned to the minimum alignment of the two values:
1863 : * 1. All mem_map arrays are page-aligned.
1864 : * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1865 : * lowest bits. PFN_SECTION_SHIFT is arch-specific
1866 : * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1867 : * worst combination is powerpc with 256k pages,
1868 : * which results in PFN_SECTION_SHIFT equal 6.
1869 : * To sum it up, at least 6 bits are available on all architectures.
1870 : * However, we can exceed 6 bits on some other architectures except
1871 : * powerpc (e.g. 15 bits are available on x86_64, 13 bits are available
1872 : * with the worst case of 64K pages on arm64) if we make sure the
1873 : * exceeded bit is not applicable to powerpc.
1874 : */
1875 : enum {
1876 : SECTION_MARKED_PRESENT_BIT,
1877 : SECTION_HAS_MEM_MAP_BIT,
1878 : SECTION_IS_ONLINE_BIT,
1879 : SECTION_IS_EARLY_BIT,
1880 : #ifdef CONFIG_ZONE_DEVICE
1881 : SECTION_TAINT_ZONE_DEVICE_BIT,
1882 : #endif
1883 : SECTION_MAP_LAST_BIT,
1884 : };
1885 :
1886 : #define SECTION_MARKED_PRESENT BIT(SECTION_MARKED_PRESENT_BIT)
1887 : #define SECTION_HAS_MEM_MAP BIT(SECTION_HAS_MEM_MAP_BIT)
1888 : #define SECTION_IS_ONLINE BIT(SECTION_IS_ONLINE_BIT)
1889 : #define SECTION_IS_EARLY BIT(SECTION_IS_EARLY_BIT)
1890 : #ifdef CONFIG_ZONE_DEVICE
1891 : #define SECTION_TAINT_ZONE_DEVICE BIT(SECTION_TAINT_ZONE_DEVICE_BIT)
1892 : #endif
1893 : #define SECTION_MAP_MASK (~(BIT(SECTION_MAP_LAST_BIT) - 1))
1894 : #define SECTION_NID_SHIFT SECTION_MAP_LAST_BIT
1895 :
1896 : static inline struct page *__section_mem_map_addr(struct mem_section *section)
1897 : {
1898 : unsigned long map = section->section_mem_map;
1899 : map &= SECTION_MAP_MASK;
1900 : return (struct page *)map;
1901 : }
1902 :
1903 : static inline int present_section(struct mem_section *section)
1904 : {
1905 : return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1906 : }
1907 :
1908 : static inline int present_section_nr(unsigned long nr)
1909 : {
1910 : return present_section(__nr_to_section(nr));
1911 : }
1912 :
1913 : static inline int valid_section(struct mem_section *section)
1914 : {
1915 : return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1916 : }
1917 :
1918 : static inline int early_section(struct mem_section *section)
1919 : {
1920 : return (section && (section->section_mem_map & SECTION_IS_EARLY));
1921 : }
1922 :
1923 : static inline int valid_section_nr(unsigned long nr)
1924 : {
1925 : return valid_section(__nr_to_section(nr));
1926 : }
1927 :
1928 : static inline int online_section(struct mem_section *section)
1929 : {
1930 : return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1931 : }
1932 :
1933 : #ifdef CONFIG_ZONE_DEVICE
1934 : static inline int online_device_section(struct mem_section *section)
1935 : {
1936 : unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE;
1937 :
1938 : return section && ((section->section_mem_map & flags) == flags);
1939 : }
1940 : #else
1941 : static inline int online_device_section(struct mem_section *section)
1942 : {
1943 : return 0;
1944 : }
1945 : #endif
1946 :
1947 : static inline int online_section_nr(unsigned long nr)
1948 : {
1949 : return online_section(__nr_to_section(nr));
1950 : }
1951 :
1952 : #ifdef CONFIG_MEMORY_HOTPLUG
1953 : void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1954 : void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1955 : #endif
1956 :
1957 : static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1958 : {
1959 : return __nr_to_section(pfn_to_section_nr(pfn));
1960 : }
1961 :
1962 : extern unsigned long __highest_present_section_nr;
1963 :
1964 : static inline int subsection_map_index(unsigned long pfn)
1965 : {
1966 : return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1967 : }
1968 :
1969 : #ifdef CONFIG_SPARSEMEM_VMEMMAP
1970 : static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1971 : {
1972 : int idx = subsection_map_index(pfn);
1973 :
1974 : return test_bit(idx, ms->usage->subsection_map);
1975 : }
1976 : #else
1977 : static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1978 : {
1979 : return 1;
1980 : }
1981 : #endif
1982 :
1983 : #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1984 : /**
1985 : * pfn_valid - check if there is a valid memory map entry for a PFN
1986 : * @pfn: the page frame number to check
1987 : *
1988 : * Check if there is a valid memory map entry aka struct page for the @pfn.
1989 : * Note, that availability of the memory map entry does not imply that
1990 : * there is actual usable memory at that @pfn. The struct page may
1991 : * represent a hole or an unusable page frame.
1992 : *
1993 : * Return: 1 for PFNs that have memory map entries and 0 otherwise
1994 : */
1995 : static inline int pfn_valid(unsigned long pfn)
1996 : {
1997 : struct mem_section *ms;
1998 :
1999 : /*
2000 : * Ensure the upper PAGE_SHIFT bits are clear in the
2001 : * pfn. Else it might lead to false positives when
2002 : * some of the upper bits are set, but the lower bits
2003 : * match a valid pfn.
2004 : */
2005 : if (PHYS_PFN(PFN_PHYS(pfn)) != pfn)
2006 : return 0;
2007 :
2008 : if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2009 : return 0;
2010 : ms = __pfn_to_section(pfn);
2011 : if (!valid_section(ms))
2012 : return 0;
2013 : /*
2014 : * Traditionally early sections always returned pfn_valid() for
2015 : * the entire section-sized span.
2016 : */
2017 : return early_section(ms) || pfn_section_valid(ms, pfn);
2018 : }
2019 : #endif
2020 :
2021 : static inline int pfn_in_present_section(unsigned long pfn)
2022 : {
2023 : if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2024 : return 0;
2025 : return present_section(__pfn_to_section(pfn));
2026 : }
2027 :
2028 : static inline unsigned long next_present_section_nr(unsigned long section_nr)
2029 : {
2030 : while (++section_nr <= __highest_present_section_nr) {
2031 : if (present_section_nr(section_nr))
2032 : return section_nr;
2033 : }
2034 :
2035 : return -1;
2036 : }
2037 :
2038 : /*
2039 : * These are _only_ used during initialisation, therefore they
2040 : * can use __initdata ... They could have names to indicate
2041 : * this restriction.
2042 : */
2043 : #ifdef CONFIG_NUMA
2044 : #define pfn_to_nid(pfn) \
2045 : ({ \
2046 : unsigned long __pfn_to_nid_pfn = (pfn); \
2047 : page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
2048 : })
2049 : #else
2050 : #define pfn_to_nid(pfn) (0)
2051 : #endif
2052 :
2053 : void sparse_init(void);
2054 : #else
2055 : #define sparse_init() do {} while (0)
2056 : #define sparse_index_init(_sec, _nid) do {} while (0)
2057 : #define pfn_in_present_section pfn_valid
2058 : #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
2059 : #endif /* CONFIG_SPARSEMEM */
2060 :
2061 : #endif /* !__GENERATING_BOUNDS.H */
2062 : #endif /* !__ASSEMBLY__ */
2063 : #endif /* _LINUX_MMZONE_H */
|
