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 <linux/zswap.h>
26 #include <asm/page.h>
27
28 /* Free memory management - zoned buddy allocator. */
29 #ifndef CONFIG_ARCH_FORCE_MAX_ORDER
30 #define MAX_PAGE_ORDER 10
31 #else
32 #define MAX_PAGE_ORDER CONFIG_ARCH_FORCE_MAX_ORDER
33 #endif
34 #define MAX_ORDER_NR_PAGES (1 << MAX_PAGE_ORDER)
35
36 #define IS_MAX_ORDER_ALIGNED(pfn) IS_ALIGNED(pfn, MAX_ORDER_NR_PAGES)
37
38 #define NR_PAGE_ORDERS (MAX_PAGE_ORDER + 1)
39
40 /*
41 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
42 * costly to service. That is between allocation orders which should
43 * coalesce naturally under reasonable reclaim pressure and those which
44 * will not.
45 */
46 #define PAGE_ALLOC_COSTLY_ORDER 3
47
48 enum migratetype {
49 MIGRATE_UNMOVABLE,
50 MIGRATE_MOVABLE,
51 MIGRATE_RECLAIMABLE,
52 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
53 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
54 #ifdef CONFIG_CMA
55 /*
56 * MIGRATE_CMA migration type is designed to mimic the way
57 * ZONE_MOVABLE works. Only movable pages can be allocated
58 * from MIGRATE_CMA pageblocks and page allocator never
59 * implicitly change migration type of MIGRATE_CMA pageblock.
60 *
61 * The way to use it is to change migratetype of a range of
62 * pageblocks to MIGRATE_CMA which can be done by
63 * __free_pageblock_cma() function.
64 */
65 MIGRATE_CMA,
66 #endif
67 #ifdef CONFIG_MEMORY_ISOLATION
68 MIGRATE_ISOLATE, /* can't allocate from here */
69 #endif
70 MIGRATE_TYPES
71 };
72
73 /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
74 extern const char * const migratetype_names[MIGRATE_TYPES];
75
76 #ifdef CONFIG_CMA
77 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
78 # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
79 # define is_migrate_cma_folio(folio, pfn) (MIGRATE_CMA == \
80 get_pfnblock_flags_mask(&folio->page, pfn, MIGRATETYPE_MASK))
81 #else
82 # define is_migrate_cma(migratetype) false
83 # define is_migrate_cma_page(_page) false
84 # define is_migrate_cma_folio(folio, pfn) false
85 #endif
86
is_migrate_movable(int mt)87 static inline bool is_migrate_movable(int mt)
88 {
89 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
90 }
91
92 /*
93 * Check whether a migratetype can be merged with another migratetype.
94 *
95 * It is only mergeable when it can fall back to other migratetypes for
96 * allocation. See fallbacks[MIGRATE_TYPES][3] in page_alloc.c.
97 */
migratetype_is_mergeable(int mt)98 static inline bool migratetype_is_mergeable(int mt)
99 {
100 return mt < MIGRATE_PCPTYPES;
101 }
102
103 #define for_each_migratetype_order(order, type) \
104 for (order = 0; order < NR_PAGE_ORDERS; order++) \
105 for (type = 0; type < MIGRATE_TYPES; type++)
106
107 extern int page_group_by_mobility_disabled;
108
109 #define MIGRATETYPE_MASK ((1UL << PB_migratetype_bits) - 1)
110
111 #define get_pageblock_migratetype(page) \
112 get_pfnblock_flags_mask(page, page_to_pfn(page), MIGRATETYPE_MASK)
113
114 #define folio_migratetype(folio) \
115 get_pfnblock_flags_mask(&folio->page, folio_pfn(folio), \
116 MIGRATETYPE_MASK)
117 struct free_area {
118 struct list_head free_list[MIGRATE_TYPES];
119 unsigned long nr_free;
120 };
121
122 struct pglist_data;
123
124 #ifdef CONFIG_NUMA
125 enum numa_stat_item {
126 NUMA_HIT, /* allocated in intended node */
127 NUMA_MISS, /* allocated in non intended node */
128 NUMA_FOREIGN, /* was intended here, hit elsewhere */
129 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
130 NUMA_LOCAL, /* allocation from local node */
131 NUMA_OTHER, /* allocation from other node */
132 NR_VM_NUMA_EVENT_ITEMS
133 };
134 #else
135 #define NR_VM_NUMA_EVENT_ITEMS 0
136 #endif
137
138 enum zone_stat_item {
139 /* First 128 byte cacheline (assuming 64 bit words) */
140 NR_FREE_PAGES,
141 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
142 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
143 NR_ZONE_ACTIVE_ANON,
144 NR_ZONE_INACTIVE_FILE,
145 NR_ZONE_ACTIVE_FILE,
146 NR_ZONE_UNEVICTABLE,
147 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
148 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
149 /* Second 128 byte cacheline */
150 NR_BOUNCE,
151 #if IS_ENABLED(CONFIG_ZSMALLOC)
152 NR_ZSPAGES, /* allocated in zsmalloc */
153 #endif
154 NR_FREE_CMA_PAGES,
155 #ifdef CONFIG_UNACCEPTED_MEMORY
156 NR_UNACCEPTED,
157 #endif
158 NR_VM_ZONE_STAT_ITEMS };
159
160 enum node_stat_item {
161 NR_LRU_BASE,
162 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
163 NR_ACTIVE_ANON, /* " " " " " */
164 NR_INACTIVE_FILE, /* " " " " " */
165 NR_ACTIVE_FILE, /* " " " " " */
166 NR_UNEVICTABLE, /* " " " " " */
167 NR_SLAB_RECLAIMABLE_B,
168 NR_SLAB_UNRECLAIMABLE_B,
169 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
170 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
171 WORKINGSET_NODES,
172 WORKINGSET_REFAULT_BASE,
173 WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE,
174 WORKINGSET_REFAULT_FILE,
175 WORKINGSET_ACTIVATE_BASE,
176 WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE,
177 WORKINGSET_ACTIVATE_FILE,
178 WORKINGSET_RESTORE_BASE,
179 WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE,
180 WORKINGSET_RESTORE_FILE,
181 WORKINGSET_NODERECLAIM,
182 NR_ANON_MAPPED, /* Mapped anonymous pages */
183 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
184 only modified from process context */
185 NR_FILE_PAGES,
186 NR_FILE_DIRTY,
187 NR_WRITEBACK,
188 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
189 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
190 NR_SHMEM_THPS,
191 NR_SHMEM_PMDMAPPED,
192 NR_FILE_THPS,
193 NR_FILE_PMDMAPPED,
194 NR_ANON_THPS,
195 NR_VMSCAN_WRITE,
196 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
197 NR_DIRTIED, /* page dirtyings since bootup */
198 NR_WRITTEN, /* page writings since bootup */
199 NR_THROTTLED_WRITTEN, /* NR_WRITTEN while reclaim throttled */
200 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
201 NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */
202 NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */
203 NR_KERNEL_STACK_KB, /* measured in KiB */
204 #if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
205 NR_KERNEL_SCS_KB, /* measured in KiB */
206 #endif
207 NR_PAGETABLE, /* used for pagetables */
208 NR_SECONDARY_PAGETABLE, /* secondary pagetables, KVM & IOMMU */
209 #ifdef CONFIG_IOMMU_SUPPORT
210 NR_IOMMU_PAGES, /* # of pages allocated by IOMMU */
211 #endif
212 #ifdef CONFIG_SWAP
213 NR_SWAPCACHE,
214 #endif
215 #ifdef CONFIG_NUMA_BALANCING
216 PGPROMOTE_SUCCESS, /* promote successfully */
217 PGPROMOTE_CANDIDATE, /* candidate pages to promote */
218 #endif
219 /* PGDEMOTE_*: pages demoted */
220 PGDEMOTE_KSWAPD,
221 PGDEMOTE_DIRECT,
222 PGDEMOTE_KHUGEPAGED,
223 NR_VM_NODE_STAT_ITEMS
224 };
225
226 /*
227 * Returns true if the item should be printed in THPs (/proc/vmstat
228 * currently prints number of anon, file and shmem THPs. But the item
229 * is charged in pages).
230 */
vmstat_item_print_in_thp(enum node_stat_item item)231 static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item)
232 {
233 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
234 return false;
235
236 return item == NR_ANON_THPS ||
237 item == NR_FILE_THPS ||
238 item == NR_SHMEM_THPS ||
239 item == NR_SHMEM_PMDMAPPED ||
240 item == NR_FILE_PMDMAPPED;
241 }
242
243 /*
244 * Returns true if the value is measured in bytes (most vmstat values are
245 * measured in pages). This defines the API part, the internal representation
246 * might be different.
247 */
vmstat_item_in_bytes(int idx)248 static __always_inline bool vmstat_item_in_bytes(int idx)
249 {
250 /*
251 * Global and per-node slab counters track slab pages.
252 * It's expected that changes are multiples of PAGE_SIZE.
253 * Internally values are stored in pages.
254 *
255 * Per-memcg and per-lruvec counters track memory, consumed
256 * by individual slab objects. These counters are actually
257 * byte-precise.
258 */
259 return (idx == NR_SLAB_RECLAIMABLE_B ||
260 idx == NR_SLAB_UNRECLAIMABLE_B);
261 }
262
263 /*
264 * We do arithmetic on the LRU lists in various places in the code,
265 * so it is important to keep the active lists LRU_ACTIVE higher in
266 * the array than the corresponding inactive lists, and to keep
267 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
268 *
269 * This has to be kept in sync with the statistics in zone_stat_item
270 * above and the descriptions in vmstat_text in mm/vmstat.c
271 */
272 #define LRU_BASE 0
273 #define LRU_ACTIVE 1
274 #define LRU_FILE 2
275
276 enum lru_list {
277 LRU_INACTIVE_ANON = LRU_BASE,
278 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
279 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
280 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
281 LRU_UNEVICTABLE,
282 NR_LRU_LISTS
283 };
284
285 enum vmscan_throttle_state {
286 VMSCAN_THROTTLE_WRITEBACK,
287 VMSCAN_THROTTLE_ISOLATED,
288 VMSCAN_THROTTLE_NOPROGRESS,
289 VMSCAN_THROTTLE_CONGESTED,
290 NR_VMSCAN_THROTTLE,
291 };
292
293 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
294
295 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
296
is_file_lru(enum lru_list lru)297 static inline bool is_file_lru(enum lru_list lru)
298 {
299 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
300 }
301
is_active_lru(enum lru_list lru)302 static inline bool is_active_lru(enum lru_list lru)
303 {
304 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
305 }
306
307 #define WORKINGSET_ANON 0
308 #define WORKINGSET_FILE 1
309 #define ANON_AND_FILE 2
310
311 enum lruvec_flags {
312 /*
313 * An lruvec has many dirty pages backed by a congested BDI:
314 * 1. LRUVEC_CGROUP_CONGESTED is set by cgroup-level reclaim.
315 * It can be cleared by cgroup reclaim or kswapd.
316 * 2. LRUVEC_NODE_CONGESTED is set by kswapd node-level reclaim.
317 * It can only be cleared by kswapd.
318 *
319 * Essentially, kswapd can unthrottle an lruvec throttled by cgroup
320 * reclaim, but not vice versa. This only applies to the root cgroup.
321 * The goal is to prevent cgroup reclaim on the root cgroup (e.g.
322 * memory.reclaim) to unthrottle an unbalanced node (that was throttled
323 * by kswapd).
324 */
325 LRUVEC_CGROUP_CONGESTED,
326 LRUVEC_NODE_CONGESTED,
327 };
328
329 #endif /* !__GENERATING_BOUNDS_H */
330
331 /*
332 * Evictable pages are divided into multiple generations. The youngest and the
333 * oldest generation numbers, max_seq and min_seq, are monotonically increasing.
334 * They form a sliding window of a variable size [MIN_NR_GENS, MAX_NR_GENS]. An
335 * offset within MAX_NR_GENS, i.e., gen, indexes the LRU list of the
336 * corresponding generation. The gen counter in folio->flags stores gen+1 while
337 * a page is on one of lrugen->folios[]. Otherwise it stores 0.
338 *
339 * A page is added to the youngest generation on faulting. The aging needs to
340 * check the accessed bit at least twice before handing this page over to the
341 * eviction. The first check takes care of the accessed bit set on the initial
342 * fault; the second check makes sure this page hasn't been used since then.
343 * This process, AKA second chance, requires a minimum of two generations,
344 * hence MIN_NR_GENS. And to maintain ABI compatibility with the active/inactive
345 * LRU, e.g., /proc/vmstat, these two generations are considered active; the
346 * rest of generations, if they exist, are considered inactive. See
347 * lru_gen_is_active().
348 *
349 * PG_active is always cleared while a page is on one of lrugen->folios[] so
350 * that the aging needs not to worry about it. And it's set again when a page
351 * considered active is isolated for non-reclaiming purposes, e.g., migration.
352 * See lru_gen_add_folio() and lru_gen_del_folio().
353 *
354 * MAX_NR_GENS is set to 4 so that the multi-gen LRU can support twice the
355 * number of categories of the active/inactive LRU when keeping track of
356 * accesses through page tables. This requires order_base_2(MAX_NR_GENS+1) bits
357 * in folio->flags.
358 */
359 #define MIN_NR_GENS 2U
360 #define MAX_NR_GENS 4U
361
362 /*
363 * Each generation is divided into multiple tiers. A page accessed N times
364 * through file descriptors is in tier order_base_2(N). A page in the first tier
365 * (N=0,1) is marked by PG_referenced unless it was faulted in through page
366 * tables or read ahead. A page in any other tier (N>1) is marked by
367 * PG_referenced and PG_workingset. This implies a minimum of two tiers is
368 * supported without using additional bits in folio->flags.
369 *
370 * In contrast to moving across generations which requires the LRU lock, moving
371 * across tiers only involves atomic operations on folio->flags and therefore
372 * has a negligible cost in the buffered access path. In the eviction path,
373 * comparisons of refaulted/(evicted+protected) from the first tier and the
374 * rest infer whether pages accessed multiple times through file descriptors
375 * are statistically hot and thus worth protecting.
376 *
377 * MAX_NR_TIERS is set to 4 so that the multi-gen LRU can support twice the
378 * number of categories of the active/inactive LRU when keeping track of
379 * accesses through file descriptors. This uses MAX_NR_TIERS-2 spare bits in
380 * folio->flags.
381 */
382 #define MAX_NR_TIERS 4U
383
384 #ifndef __GENERATING_BOUNDS_H
385
386 struct lruvec;
387 struct page_vma_mapped_walk;
388
389 #define LRU_GEN_MASK ((BIT(LRU_GEN_WIDTH) - 1) << LRU_GEN_PGOFF)
390 #define LRU_REFS_MASK ((BIT(LRU_REFS_WIDTH) - 1) << LRU_REFS_PGOFF)
391
392 #ifdef CONFIG_LRU_GEN
393
394 enum {
395 LRU_GEN_ANON,
396 LRU_GEN_FILE,
397 };
398
399 enum {
400 LRU_GEN_CORE,
401 LRU_GEN_MM_WALK,
402 LRU_GEN_NONLEAF_YOUNG,
403 NR_LRU_GEN_CAPS
404 };
405
406 #define MIN_LRU_BATCH BITS_PER_LONG
407 #define MAX_LRU_BATCH (MIN_LRU_BATCH * 64)
408
409 /* whether to keep historical stats from evicted generations */
410 #ifdef CONFIG_LRU_GEN_STATS
411 #define NR_HIST_GENS MAX_NR_GENS
412 #else
413 #define NR_HIST_GENS 1U
414 #endif
415
416 /*
417 * The youngest generation number is stored in max_seq for both anon and file
418 * types as they are aged on an equal footing. The oldest generation numbers are
419 * stored in min_seq[] separately for anon and file types as clean file pages
420 * can be evicted regardless of swap constraints.
421 *
422 * Normally anon and file min_seq are in sync. But if swapping is constrained,
423 * e.g., out of swap space, file min_seq is allowed to advance and leave anon
424 * min_seq behind.
425 *
426 * The number of pages in each generation is eventually consistent and therefore
427 * can be transiently negative when reset_batch_size() is pending.
428 */
429 struct lru_gen_folio {
430 /* the aging increments the youngest generation number */
431 unsigned long max_seq;
432 /* the eviction increments the oldest generation numbers */
433 unsigned long min_seq[ANON_AND_FILE];
434 /* the birth time of each generation in jiffies */
435 unsigned long timestamps[MAX_NR_GENS];
436 /* the multi-gen LRU lists, lazily sorted on eviction */
437 struct list_head folios[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
438 /* the multi-gen LRU sizes, eventually consistent */
439 long nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
440 /* the exponential moving average of refaulted */
441 unsigned long avg_refaulted[ANON_AND_FILE][MAX_NR_TIERS];
442 /* the exponential moving average of evicted+protected */
443 unsigned long avg_total[ANON_AND_FILE][MAX_NR_TIERS];
444 /* the first tier doesn't need protection, hence the minus one */
445 unsigned long protected[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS - 1];
446 /* can be modified without holding the LRU lock */
447 atomic_long_t evicted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
448 atomic_long_t refaulted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
449 /* whether the multi-gen LRU is enabled */
450 bool enabled;
451 /* the memcg generation this lru_gen_folio belongs to */
452 u8 gen;
453 /* the list segment this lru_gen_folio belongs to */
454 u8 seg;
455 /* per-node lru_gen_folio list for global reclaim */
456 struct hlist_nulls_node list;
457 };
458
459 enum {
460 MM_LEAF_TOTAL, /* total leaf entries */
461 MM_LEAF_OLD, /* old leaf entries */
462 MM_LEAF_YOUNG, /* young leaf entries */
463 MM_NONLEAF_TOTAL, /* total non-leaf entries */
464 MM_NONLEAF_FOUND, /* non-leaf entries found in Bloom filters */
465 MM_NONLEAF_ADDED, /* non-leaf entries added to Bloom filters */
466 NR_MM_STATS
467 };
468
469 /* double-buffering Bloom filters */
470 #define NR_BLOOM_FILTERS 2
471
472 struct lru_gen_mm_state {
473 /* synced with max_seq after each iteration */
474 unsigned long seq;
475 /* where the current iteration continues after */
476 struct list_head *head;
477 /* where the last iteration ended before */
478 struct list_head *tail;
479 /* Bloom filters flip after each iteration */
480 unsigned long *filters[NR_BLOOM_FILTERS];
481 /* the mm stats for debugging */
482 unsigned long stats[NR_HIST_GENS][NR_MM_STATS];
483 };
484
485 struct lru_gen_mm_walk {
486 /* the lruvec under reclaim */
487 struct lruvec *lruvec;
488 /* max_seq from lru_gen_folio: can be out of date */
489 unsigned long seq;
490 /* the next address within an mm to scan */
491 unsigned long next_addr;
492 /* to batch promoted pages */
493 int nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
494 /* to batch the mm stats */
495 int mm_stats[NR_MM_STATS];
496 /* total batched items */
497 int batched;
498 bool can_swap;
499 bool force_scan;
500 };
501
502 /*
503 * For each node, memcgs are divided into two generations: the old and the
504 * young. For each generation, memcgs are randomly sharded into multiple bins
505 * to improve scalability. For each bin, the hlist_nulls is virtually divided
506 * into three segments: the head, the tail and the default.
507 *
508 * An onlining memcg is added to the tail of a random bin in the old generation.
509 * The eviction starts at the head of a random bin in the old generation. The
510 * per-node memcg generation counter, whose reminder (mod MEMCG_NR_GENS) indexes
511 * the old generation, is incremented when all its bins become empty.
512 *
513 * There are four operations:
514 * 1. MEMCG_LRU_HEAD, which moves a memcg to the head of a random bin in its
515 * current generation (old or young) and updates its "seg" to "head";
516 * 2. MEMCG_LRU_TAIL, which moves a memcg to the tail of a random bin in its
517 * current generation (old or young) and updates its "seg" to "tail";
518 * 3. MEMCG_LRU_OLD, which moves a memcg to the head of a random bin in the old
519 * generation, updates its "gen" to "old" and resets its "seg" to "default";
520 * 4. MEMCG_LRU_YOUNG, which moves a memcg to the tail of a random bin in the
521 * young generation, updates its "gen" to "young" and resets its "seg" to
522 * "default".
523 *
524 * The events that trigger the above operations are:
525 * 1. Exceeding the soft limit, which triggers MEMCG_LRU_HEAD;
526 * 2. The first attempt to reclaim a memcg below low, which triggers
527 * MEMCG_LRU_TAIL;
528 * 3. The first attempt to reclaim a memcg offlined or below reclaimable size
529 * threshold, which triggers MEMCG_LRU_TAIL;
530 * 4. The second attempt to reclaim a memcg offlined or below reclaimable size
531 * threshold, which triggers MEMCG_LRU_YOUNG;
532 * 5. Attempting to reclaim a memcg below min, which triggers MEMCG_LRU_YOUNG;
533 * 6. Finishing the aging on the eviction path, which triggers MEMCG_LRU_YOUNG;
534 * 7. Offlining a memcg, which triggers MEMCG_LRU_OLD.
535 *
536 * Notes:
537 * 1. Memcg LRU only applies to global reclaim, and the round-robin incrementing
538 * of their max_seq counters ensures the eventual fairness to all eligible
539 * memcgs. For memcg reclaim, it still relies on mem_cgroup_iter().
540 * 2. There are only two valid generations: old (seq) and young (seq+1).
541 * MEMCG_NR_GENS is set to three so that when reading the generation counter
542 * locklessly, a stale value (seq-1) does not wraparound to young.
543 */
544 #define MEMCG_NR_GENS 3
545 #define MEMCG_NR_BINS 8
546
547 struct lru_gen_memcg {
548 /* the per-node memcg generation counter */
549 unsigned long seq;
550 /* each memcg has one lru_gen_folio per node */
551 unsigned long nr_memcgs[MEMCG_NR_GENS];
552 /* per-node lru_gen_folio list for global reclaim */
553 struct hlist_nulls_head fifo[MEMCG_NR_GENS][MEMCG_NR_BINS];
554 /* protects the above */
555 spinlock_t lock;
556 };
557
558 void lru_gen_init_pgdat(struct pglist_data *pgdat);
559 void lru_gen_init_lruvec(struct lruvec *lruvec);
560 void lru_gen_look_around(struct page_vma_mapped_walk *pvmw);
561
562 void lru_gen_init_memcg(struct mem_cgroup *memcg);
563 void lru_gen_exit_memcg(struct mem_cgroup *memcg);
564 void lru_gen_online_memcg(struct mem_cgroup *memcg);
565 void lru_gen_offline_memcg(struct mem_cgroup *memcg);
566 void lru_gen_release_memcg(struct mem_cgroup *memcg);
567 void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid);
568
569 #else /* !CONFIG_LRU_GEN */
570
lru_gen_init_pgdat(struct pglist_data * pgdat)571 static inline void lru_gen_init_pgdat(struct pglist_data *pgdat)
572 {
573 }
574
lru_gen_init_lruvec(struct lruvec * lruvec)575 static inline void lru_gen_init_lruvec(struct lruvec *lruvec)
576 {
577 }
578
lru_gen_look_around(struct page_vma_mapped_walk * pvmw)579 static inline void lru_gen_look_around(struct page_vma_mapped_walk *pvmw)
580 {
581 }
582
lru_gen_init_memcg(struct mem_cgroup * memcg)583 static inline void lru_gen_init_memcg(struct mem_cgroup *memcg)
584 {
585 }
586
lru_gen_exit_memcg(struct mem_cgroup * memcg)587 static inline void lru_gen_exit_memcg(struct mem_cgroup *memcg)
588 {
589 }
590
lru_gen_online_memcg(struct mem_cgroup * memcg)591 static inline void lru_gen_online_memcg(struct mem_cgroup *memcg)
592 {
593 }
594
lru_gen_offline_memcg(struct mem_cgroup * memcg)595 static inline void lru_gen_offline_memcg(struct mem_cgroup *memcg)
596 {
597 }
598
lru_gen_release_memcg(struct mem_cgroup * memcg)599 static inline void lru_gen_release_memcg(struct mem_cgroup *memcg)
600 {
601 }
602
lru_gen_soft_reclaim(struct mem_cgroup * memcg,int nid)603 static inline void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid)
604 {
605 }
606
607 #endif /* CONFIG_LRU_GEN */
608
609 struct lruvec {
610 struct list_head lists[NR_LRU_LISTS];
611 /* per lruvec lru_lock for memcg */
612 spinlock_t lru_lock;
613 /*
614 * These track the cost of reclaiming one LRU - file or anon -
615 * over the other. As the observed cost of reclaiming one LRU
616 * increases, the reclaim scan balance tips toward the other.
617 */
618 unsigned long anon_cost;
619 unsigned long file_cost;
620 /* Non-resident age, driven by LRU movement */
621 atomic_long_t nonresident_age;
622 /* Refaults at the time of last reclaim cycle */
623 unsigned long refaults[ANON_AND_FILE];
624 /* Various lruvec state flags (enum lruvec_flags) */
625 unsigned long flags;
626 #ifdef CONFIG_LRU_GEN
627 /* evictable pages divided into generations */
628 struct lru_gen_folio lrugen;
629 #ifdef CONFIG_LRU_GEN_WALKS_MMU
630 /* to concurrently iterate lru_gen_mm_list */
631 struct lru_gen_mm_state mm_state;
632 #endif
633 #endif /* CONFIG_LRU_GEN */
634 #ifdef CONFIG_MEMCG
635 struct pglist_data *pgdat;
636 #endif
637 struct zswap_lruvec_state zswap_lruvec_state;
638 };
639
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. Two additional lists
658 * are added for THP. One PCP list is used by GPF_MOVABLE, and the other PCP list
659 * is used by GFP_UNMOVABLE and GFP_RECLAIMABLE.
660 */
661 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
662 #define NR_PCP_THP 2
663 #else
664 #define NR_PCP_THP 0
665 #endif
666 #define NR_LOWORDER_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1))
667 #define NR_PCP_LISTS (NR_LOWORDER_PCP_LISTS + NR_PCP_THP)
668
669 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
670 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
671 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
672 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
673
674 /*
675 * Flags used in pcp->flags field.
676 *
677 * PCPF_PREV_FREE_HIGH_ORDER: a high-order page is freed in the
678 * previous page freeing. To avoid to drain PCP for an accident
679 * high-order page freeing.
680 *
681 * PCPF_FREE_HIGH_BATCH: preserve "pcp->batch" pages in PCP before
682 * draining PCP for consecutive high-order pages freeing without
683 * allocation if data cache slice of CPU is large enough. To reduce
684 * zone lock contention and keep cache-hot pages reusing.
685 */
686 #define PCPF_PREV_FREE_HIGH_ORDER BIT(0)
687 #define PCPF_FREE_HIGH_BATCH BIT(1)
688
689 struct per_cpu_pages {
690 spinlock_t lock; /* Protects lists field */
691 int count; /* number of pages in the list */
692 int high; /* high watermark, emptying needed */
693 int high_min; /* min high watermark */
694 int high_max; /* max high watermark */
695 int batch; /* chunk size for buddy add/remove */
696 u8 flags; /* protected by pcp->lock */
697 u8 alloc_factor; /* batch scaling factor during allocate */
698 #ifdef CONFIG_NUMA
699 u8 expire; /* When 0, remote pagesets are drained */
700 #endif
701 short free_count; /* consecutive free count */
702
703 /* Lists of pages, one per migrate type stored on the pcp-lists */
704 struct list_head lists[NR_PCP_LISTS];
705 } ____cacheline_aligned_in_smp;
706
707 struct per_cpu_zonestat {
708 #ifdef CONFIG_SMP
709 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
710 s8 stat_threshold;
711 #endif
712 #ifdef CONFIG_NUMA
713 /*
714 * Low priority inaccurate counters that are only folded
715 * on demand. Use a large type to avoid the overhead of
716 * folding during refresh_cpu_vm_stats.
717 */
718 unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
719 #endif
720 };
721
722 struct per_cpu_nodestat {
723 s8 stat_threshold;
724 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
725 };
726
727 #endif /* !__GENERATING_BOUNDS.H */
728
729 enum zone_type {
730 /*
731 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
732 * to DMA to all of the addressable memory (ZONE_NORMAL).
733 * On architectures where this area covers the whole 32 bit address
734 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
735 * DMA addressing constraints. This distinction is important as a 32bit
736 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
737 * platforms may need both zones as they support peripherals with
738 * different DMA addressing limitations.
739 */
740 #ifdef CONFIG_ZONE_DMA
741 ZONE_DMA,
742 #endif
743 #ifdef CONFIG_ZONE_DMA32
744 ZONE_DMA32,
745 #endif
746 /*
747 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
748 * performed on pages in ZONE_NORMAL if the DMA devices support
749 * transfers to all addressable memory.
750 */
751 ZONE_NORMAL,
752 #ifdef CONFIG_HIGHMEM
753 /*
754 * A memory area that is only addressable by the kernel through
755 * mapping portions into its own address space. This is for example
756 * used by i386 to allow the kernel to address the memory beyond
757 * 900MB. The kernel will set up special mappings (page
758 * table entries on i386) for each page that the kernel needs to
759 * access.
760 */
761 ZONE_HIGHMEM,
762 #endif
763 /*
764 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
765 * movable pages with few exceptional cases described below. Main use
766 * cases for ZONE_MOVABLE are to make memory offlining/unplug more
767 * likely to succeed, and to locally limit unmovable allocations - e.g.,
768 * to increase the number of THP/huge pages. Notable special cases are:
769 *
770 * 1. Pinned pages: (long-term) pinning of movable pages might
771 * essentially turn such pages unmovable. Therefore, we do not allow
772 * pinning long-term pages in ZONE_MOVABLE. When pages are pinned and
773 * faulted, they come from the right zone right away. However, it is
774 * still possible that address space already has pages in
775 * ZONE_MOVABLE at the time when pages are pinned (i.e. user has
776 * touches that memory before pinning). In such case we migrate them
777 * to a different zone. When migration fails - pinning fails.
778 * 2. memblock allocations: kernelcore/movablecore setups might create
779 * situations where ZONE_MOVABLE contains unmovable allocations
780 * after boot. Memory offlining and allocations fail early.
781 * 3. Memory holes: kernelcore/movablecore setups might create very rare
782 * situations where ZONE_MOVABLE contains memory holes after boot,
783 * for example, if we have sections that are only partially
784 * populated. Memory offlining and allocations fail early.
785 * 4. PG_hwpoison pages: while poisoned pages can be skipped during
786 * memory offlining, such pages cannot be allocated.
787 * 5. Unmovable PG_offline pages: in paravirtualized environments,
788 * hotplugged memory blocks might only partially be managed by the
789 * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
790 * parts not manged by the buddy are unmovable PG_offline pages. In
791 * some cases (virtio-mem), such pages can be skipped during
792 * memory offlining, however, cannot be moved/allocated. These
793 * techniques might use alloc_contig_range() to hide previously
794 * exposed pages from the buddy again (e.g., to implement some sort
795 * of memory unplug in virtio-mem).
796 * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create
797 * situations where ZERO_PAGE(0) which is allocated differently
798 * on different platforms may end up in a movable zone. ZERO_PAGE(0)
799 * cannot be migrated.
800 * 7. Memory-hotplug: when using memmap_on_memory and onlining the
801 * memory to the MOVABLE zone, the vmemmap pages are also placed in
802 * such zone. Such pages cannot be really moved around as they are
803 * self-stored in the range, but they are treated as movable when
804 * the range they describe is about to be offlined.
805 *
806 * In general, no unmovable allocations that degrade memory offlining
807 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
808 * have to expect that migrating pages in ZONE_MOVABLE can fail (even
809 * if has_unmovable_pages() states that there are no unmovable pages,
810 * there can be false negatives).
811 */
812 ZONE_MOVABLE,
813 #ifdef CONFIG_ZONE_DEVICE
814 ZONE_DEVICE,
815 #endif
816 __MAX_NR_ZONES
817
818 };
819
820 #ifndef __GENERATING_BOUNDS_H
821
822 #define ASYNC_AND_SYNC 2
823
824 struct zone {
825 /* Read-mostly fields */
826
827 /* zone watermarks, access with *_wmark_pages(zone) macros */
828 unsigned long _watermark[NR_WMARK];
829 unsigned long watermark_boost;
830
831 unsigned long nr_reserved_highatomic;
832
833 /*
834 * We don't know if the memory that we're going to allocate will be
835 * freeable or/and it will be released eventually, so to avoid totally
836 * wasting several GB of ram we must reserve some of the lower zone
837 * memory (otherwise we risk to run OOM on the lower zones despite
838 * there being tons of freeable ram on the higher zones). This array is
839 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
840 * changes.
841 */
842 long lowmem_reserve[MAX_NR_ZONES];
843
844 #ifdef CONFIG_NUMA
845 int node;
846 #endif
847 struct pglist_data *zone_pgdat;
848 struct per_cpu_pages __percpu *per_cpu_pageset;
849 struct per_cpu_zonestat __percpu *per_cpu_zonestats;
850 /*
851 * the high and batch values are copied to individual pagesets for
852 * faster access
853 */
854 int pageset_high_min;
855 int pageset_high_max;
856 int pageset_batch;
857
858 #ifndef CONFIG_SPARSEMEM
859 /*
860 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
861 * In SPARSEMEM, this map is stored in struct mem_section
862 */
863 unsigned long *pageblock_flags;
864 #endif /* CONFIG_SPARSEMEM */
865
866 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
867 unsigned long zone_start_pfn;
868
869 /*
870 * spanned_pages is the total pages spanned by the zone, including
871 * holes, which is calculated as:
872 * spanned_pages = zone_end_pfn - zone_start_pfn;
873 *
874 * present_pages is physical pages existing within the zone, which
875 * is calculated as:
876 * present_pages = spanned_pages - absent_pages(pages in holes);
877 *
878 * present_early_pages is present pages existing within the zone
879 * located on memory available since early boot, excluding hotplugged
880 * memory.
881 *
882 * managed_pages is present pages managed by the buddy system, which
883 * is calculated as (reserved_pages includes pages allocated by the
884 * bootmem allocator):
885 * managed_pages = present_pages - reserved_pages;
886 *
887 * cma pages is present pages that are assigned for CMA use
888 * (MIGRATE_CMA).
889 *
890 * So present_pages may be used by memory hotplug or memory power
891 * management logic to figure out unmanaged pages by checking
892 * (present_pages - managed_pages). And managed_pages should be used
893 * by page allocator and vm scanner to calculate all kinds of watermarks
894 * and thresholds.
895 *
896 * Locking rules:
897 *
898 * zone_start_pfn and spanned_pages are protected by span_seqlock.
899 * It is a seqlock because it has to be read outside of zone->lock,
900 * and it is done in the main allocator path. But, it is written
901 * quite infrequently.
902 *
903 * The span_seq lock is declared along with zone->lock because it is
904 * frequently read in proximity to zone->lock. It's good to
905 * give them a chance of being in the same cacheline.
906 *
907 * Write access to present_pages at runtime should be protected by
908 * mem_hotplug_begin/done(). Any reader who can't tolerant drift of
909 * present_pages should use get_online_mems() to get a stable value.
910 */
911 atomic_long_t managed_pages;
912 unsigned long spanned_pages;
913 unsigned long present_pages;
914 #if defined(CONFIG_MEMORY_HOTPLUG)
915 unsigned long present_early_pages;
916 #endif
917 #ifdef CONFIG_CMA
918 unsigned long cma_pages;
919 #endif
920
921 const char *name;
922
923 #ifdef CONFIG_MEMORY_ISOLATION
924 /*
925 * Number of isolated pageblock. It is used to solve incorrect
926 * freepage counting problem due to racy retrieving migratetype
927 * of pageblock. Protected by zone->lock.
928 */
929 unsigned long nr_isolate_pageblock;
930 #endif
931
932 #ifdef CONFIG_MEMORY_HOTPLUG
933 /* see spanned/present_pages for more description */
934 seqlock_t span_seqlock;
935 #endif
936
937 int initialized;
938
939 /* Write-intensive fields used from the page allocator */
940 CACHELINE_PADDING(_pad1_);
941
942 /* free areas of different sizes */
943 struct free_area free_area[NR_PAGE_ORDERS];
944
945 #ifdef CONFIG_UNACCEPTED_MEMORY
946 /* Pages to be accepted. All pages on the list are MAX_PAGE_ORDER */
947 struct list_head unaccepted_pages;
948 #endif
949
950 /* zone flags, see below */
951 unsigned long flags;
952
953 /* Primarily protects free_area */
954 spinlock_t lock;
955
956 /* Write-intensive fields used by compaction and vmstats. */
957 CACHELINE_PADDING(_pad2_);
958
959 /*
960 * When free pages are below this point, additional steps are taken
961 * when reading the number of free pages to avoid per-cpu counter
962 * drift allowing watermarks to be breached
963 */
964 unsigned long percpu_drift_mark;
965
966 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
967 /* pfn where compaction free scanner should start */
968 unsigned long compact_cached_free_pfn;
969 /* pfn where compaction migration scanner should start */
970 unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC];
971 unsigned long compact_init_migrate_pfn;
972 unsigned long compact_init_free_pfn;
973 #endif
974
975 #ifdef CONFIG_COMPACTION
976 /*
977 * On compaction failure, 1<<compact_defer_shift compactions
978 * are skipped before trying again. The number attempted since
979 * last failure is tracked with compact_considered.
980 * compact_order_failed is the minimum compaction failed order.
981 */
982 unsigned int compact_considered;
983 unsigned int compact_defer_shift;
984 int compact_order_failed;
985 #endif
986
987 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
988 /* Set to true when the PG_migrate_skip bits should be cleared */
989 bool compact_blockskip_flush;
990 #endif
991
992 bool contiguous;
993
994 CACHELINE_PADDING(_pad3_);
995 /* Zone statistics */
996 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
997 atomic_long_t vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
998 } ____cacheline_internodealigned_in_smp;
999
1000 enum pgdat_flags {
1001 PGDAT_DIRTY, /* reclaim scanning has recently found
1002 * many dirty file pages at the tail
1003 * of the LRU.
1004 */
1005 PGDAT_WRITEBACK, /* reclaim scanning has recently found
1006 * many pages under writeback
1007 */
1008 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
1009 };
1010
1011 enum zone_flags {
1012 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
1013 * Cleared when kswapd is woken.
1014 */
1015 ZONE_RECLAIM_ACTIVE, /* kswapd may be scanning the zone. */
1016 ZONE_BELOW_HIGH, /* zone is below high watermark. */
1017 };
1018
zone_managed_pages(struct zone * zone)1019 static inline unsigned long zone_managed_pages(struct zone *zone)
1020 {
1021 return (unsigned long)atomic_long_read(&zone->managed_pages);
1022 }
1023
zone_cma_pages(struct zone * zone)1024 static inline unsigned long zone_cma_pages(struct zone *zone)
1025 {
1026 #ifdef CONFIG_CMA
1027 return zone->cma_pages;
1028 #else
1029 return 0;
1030 #endif
1031 }
1032
zone_end_pfn(const struct zone * zone)1033 static inline unsigned long zone_end_pfn(const struct zone *zone)
1034 {
1035 return zone->zone_start_pfn + zone->spanned_pages;
1036 }
1037
zone_spans_pfn(const struct zone * zone,unsigned long pfn)1038 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
1039 {
1040 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
1041 }
1042
zone_is_initialized(struct zone * zone)1043 static inline bool zone_is_initialized(struct zone *zone)
1044 {
1045 return zone->initialized;
1046 }
1047
zone_is_empty(struct zone * zone)1048 static inline bool zone_is_empty(struct zone *zone)
1049 {
1050 return zone->spanned_pages == 0;
1051 }
1052
1053 #ifndef BUILD_VDSO32_64
1054 /*
1055 * The zone field is never updated after free_area_init_core()
1056 * sets it, so none of the operations on it need to be atomic.
1057 */
1058
1059 /* Page flags: | [SECTION] | [NODE] | ZONE | [LAST_CPUPID] | ... | FLAGS | */
1060 #define SECTIONS_PGOFF ((sizeof(unsigned long)*8) - SECTIONS_WIDTH)
1061 #define NODES_PGOFF (SECTIONS_PGOFF - NODES_WIDTH)
1062 #define ZONES_PGOFF (NODES_PGOFF - ZONES_WIDTH)
1063 #define LAST_CPUPID_PGOFF (ZONES_PGOFF - LAST_CPUPID_WIDTH)
1064 #define KASAN_TAG_PGOFF (LAST_CPUPID_PGOFF - KASAN_TAG_WIDTH)
1065 #define LRU_GEN_PGOFF (KASAN_TAG_PGOFF - LRU_GEN_WIDTH)
1066 #define LRU_REFS_PGOFF (LRU_GEN_PGOFF - LRU_REFS_WIDTH)
1067
1068 /*
1069 * Define the bit shifts to access each section. For non-existent
1070 * sections we define the shift as 0; that plus a 0 mask ensures
1071 * the compiler will optimise away reference to them.
1072 */
1073 #define SECTIONS_PGSHIFT (SECTIONS_PGOFF * (SECTIONS_WIDTH != 0))
1074 #define NODES_PGSHIFT (NODES_PGOFF * (NODES_WIDTH != 0))
1075 #define ZONES_PGSHIFT (ZONES_PGOFF * (ZONES_WIDTH != 0))
1076 #define LAST_CPUPID_PGSHIFT (LAST_CPUPID_PGOFF * (LAST_CPUPID_WIDTH != 0))
1077 #define KASAN_TAG_PGSHIFT (KASAN_TAG_PGOFF * (KASAN_TAG_WIDTH != 0))
1078
1079 /* NODE:ZONE or SECTION:ZONE is used to ID a zone for the buddy allocator */
1080 #ifdef NODE_NOT_IN_PAGE_FLAGS
1081 #define ZONEID_SHIFT (SECTIONS_SHIFT + ZONES_SHIFT)
1082 #define ZONEID_PGOFF ((SECTIONS_PGOFF < ZONES_PGOFF) ? \
1083 SECTIONS_PGOFF : ZONES_PGOFF)
1084 #else
1085 #define ZONEID_SHIFT (NODES_SHIFT + ZONES_SHIFT)
1086 #define ZONEID_PGOFF ((NODES_PGOFF < ZONES_PGOFF) ? \
1087 NODES_PGOFF : ZONES_PGOFF)
1088 #endif
1089
1090 #define ZONEID_PGSHIFT (ZONEID_PGOFF * (ZONEID_SHIFT != 0))
1091
1092 #define ZONES_MASK ((1UL << ZONES_WIDTH) - 1)
1093 #define NODES_MASK ((1UL << NODES_WIDTH) - 1)
1094 #define SECTIONS_MASK ((1UL << SECTIONS_WIDTH) - 1)
1095 #define LAST_CPUPID_MASK ((1UL << LAST_CPUPID_SHIFT) - 1)
1096 #define KASAN_TAG_MASK ((1UL << KASAN_TAG_WIDTH) - 1)
1097 #define ZONEID_MASK ((1UL << ZONEID_SHIFT) - 1)
1098
page_zonenum(const struct page * page)1099 static inline enum zone_type page_zonenum(const struct page *page)
1100 {
1101 ASSERT_EXCLUSIVE_BITS(page->flags, ZONES_MASK << ZONES_PGSHIFT);
1102 return (page->flags >> ZONES_PGSHIFT) & ZONES_MASK;
1103 }
1104
folio_zonenum(const struct folio * folio)1105 static inline enum zone_type folio_zonenum(const struct folio *folio)
1106 {
1107 return page_zonenum(&folio->page);
1108 }
1109
1110 #ifdef CONFIG_ZONE_DEVICE
is_zone_device_page(const struct page * page)1111 static inline bool is_zone_device_page(const struct page *page)
1112 {
1113 return page_zonenum(page) == ZONE_DEVICE;
1114 }
1115
1116 /*
1117 * Consecutive zone device pages should not be merged into the same sgl
1118 * or bvec segment with other types of pages or if they belong to different
1119 * pgmaps. Otherwise getting the pgmap of a given segment is not possible
1120 * without scanning the entire segment. This helper returns true either if
1121 * both pages are not zone device pages or both pages are zone device pages
1122 * with the same pgmap.
1123 */
zone_device_pages_have_same_pgmap(const struct page * a,const struct page * b)1124 static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1125 const struct page *b)
1126 {
1127 if (is_zone_device_page(a) != is_zone_device_page(b))
1128 return false;
1129 if (!is_zone_device_page(a))
1130 return true;
1131 return a->pgmap == b->pgmap;
1132 }
1133
1134 extern void memmap_init_zone_device(struct zone *, unsigned long,
1135 unsigned long, struct dev_pagemap *);
1136 #else
is_zone_device_page(const struct page * page)1137 static inline bool is_zone_device_page(const struct page *page)
1138 {
1139 return false;
1140 }
zone_device_pages_have_same_pgmap(const struct page * a,const struct page * b)1141 static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1142 const struct page *b)
1143 {
1144 return true;
1145 }
1146 #endif
1147
folio_is_zone_device(const struct folio * folio)1148 static inline bool folio_is_zone_device(const struct folio *folio)
1149 {
1150 return is_zone_device_page(&folio->page);
1151 }
1152
is_zone_movable_page(const struct page * page)1153 static inline bool is_zone_movable_page(const struct page *page)
1154 {
1155 return page_zonenum(page) == ZONE_MOVABLE;
1156 }
1157
folio_is_zone_movable(const struct folio * folio)1158 static inline bool folio_is_zone_movable(const struct folio *folio)
1159 {
1160 return folio_zonenum(folio) == ZONE_MOVABLE;
1161 }
1162 #endif
1163
1164 /*
1165 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
1166 * intersection with the given zone
1167 */
zone_intersects(struct zone * zone,unsigned long start_pfn,unsigned long nr_pages)1168 static inline bool zone_intersects(struct zone *zone,
1169 unsigned long start_pfn, unsigned long nr_pages)
1170 {
1171 if (zone_is_empty(zone))
1172 return false;
1173 if (start_pfn >= zone_end_pfn(zone) ||
1174 start_pfn + nr_pages <= zone->zone_start_pfn)
1175 return false;
1176
1177 return true;
1178 }
1179
1180 /*
1181 * The "priority" of VM scanning is how much of the queues we will scan in one
1182 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
1183 * queues ("queue_length >> 12") during an aging round.
1184 */
1185 #define DEF_PRIORITY 12
1186
1187 /* Maximum number of zones on a zonelist */
1188 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
1189
1190 enum {
1191 ZONELIST_FALLBACK, /* zonelist with fallback */
1192 #ifdef CONFIG_NUMA
1193 /*
1194 * The NUMA zonelists are doubled because we need zonelists that
1195 * restrict the allocations to a single node for __GFP_THISNODE.
1196 */
1197 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
1198 #endif
1199 MAX_ZONELISTS
1200 };
1201
1202 /*
1203 * This struct contains information about a zone in a zonelist. It is stored
1204 * here to avoid dereferences into large structures and lookups of tables
1205 */
1206 struct zoneref {
1207 struct zone *zone; /* Pointer to actual zone */
1208 int zone_idx; /* zone_idx(zoneref->zone) */
1209 };
1210
1211 /*
1212 * One allocation request operates on a zonelist. A zonelist
1213 * is a list of zones, the first one is the 'goal' of the
1214 * allocation, the other zones are fallback zones, in decreasing
1215 * priority.
1216 *
1217 * To speed the reading of the zonelist, the zonerefs contain the zone index
1218 * of the entry being read. Helper functions to access information given
1219 * a struct zoneref are
1220 *
1221 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
1222 * zonelist_zone_idx() - Return the index of the zone for an entry
1223 * zonelist_node_idx() - Return the index of the node for an entry
1224 */
1225 struct zonelist {
1226 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
1227 };
1228
1229 /*
1230 * The array of struct pages for flatmem.
1231 * It must be declared for SPARSEMEM as well because there are configurations
1232 * that rely on that.
1233 */
1234 extern struct page *mem_map;
1235
1236 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1237 struct deferred_split {
1238 spinlock_t split_queue_lock;
1239 struct list_head split_queue;
1240 unsigned long split_queue_len;
1241 };
1242 #endif
1243
1244 #ifdef CONFIG_MEMORY_FAILURE
1245 /*
1246 * Per NUMA node memory failure handling statistics.
1247 */
1248 struct memory_failure_stats {
1249 /*
1250 * Number of raw pages poisoned.
1251 * Cases not accounted: memory outside kernel control, offline page,
1252 * arch-specific memory_failure (SGX), hwpoison_filter() filtered
1253 * error events, and unpoison actions from hwpoison_unpoison.
1254 */
1255 unsigned long total;
1256 /*
1257 * Recovery results of poisoned raw pages handled by memory_failure,
1258 * in sync with mf_result.
1259 * total = ignored + failed + delayed + recovered.
1260 * total * PAGE_SIZE * #nodes = /proc/meminfo/HardwareCorrupted.
1261 */
1262 unsigned long ignored;
1263 unsigned long failed;
1264 unsigned long delayed;
1265 unsigned long recovered;
1266 };
1267 #endif
1268
1269 /*
1270 * On NUMA machines, each NUMA node would have a pg_data_t to describe
1271 * it's memory layout. On UMA machines there is a single pglist_data which
1272 * describes the whole memory.
1273 *
1274 * Memory statistics and page replacement data structures are maintained on a
1275 * per-zone basis.
1276 */
1277 typedef struct pglist_data {
1278 /*
1279 * node_zones contains just the zones for THIS node. Not all of the
1280 * zones may be populated, but it is the full list. It is referenced by
1281 * this node's node_zonelists as well as other node's node_zonelists.
1282 */
1283 struct zone node_zones[MAX_NR_ZONES];
1284
1285 /*
1286 * node_zonelists contains references to all zones in all nodes.
1287 * Generally the first zones will be references to this node's
1288 * node_zones.
1289 */
1290 struct zonelist node_zonelists[MAX_ZONELISTS];
1291
1292 int nr_zones; /* number of populated zones in this node */
1293 #ifdef CONFIG_FLATMEM /* means !SPARSEMEM */
1294 struct page *node_mem_map;
1295 #ifdef CONFIG_PAGE_EXTENSION
1296 struct page_ext *node_page_ext;
1297 #endif
1298 #endif
1299 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
1300 /*
1301 * Must be held any time you expect node_start_pfn,
1302 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
1303 * Also synchronizes pgdat->first_deferred_pfn during deferred page
1304 * init.
1305 *
1306 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
1307 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
1308 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
1309 *
1310 * Nests above zone->lock and zone->span_seqlock
1311 */
1312 spinlock_t node_size_lock;
1313 #endif
1314 unsigned long node_start_pfn;
1315 unsigned long node_present_pages; /* total number of physical pages */
1316 unsigned long node_spanned_pages; /* total size of physical page
1317 range, including holes */
1318 int node_id;
1319 wait_queue_head_t kswapd_wait;
1320 wait_queue_head_t pfmemalloc_wait;
1321
1322 /* workqueues for throttling reclaim for different reasons. */
1323 wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE];
1324
1325 atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */
1326 unsigned long nr_reclaim_start; /* nr pages written while throttled
1327 * when throttling started. */
1328 #ifdef CONFIG_MEMORY_HOTPLUG
1329 struct mutex kswapd_lock;
1330 #endif
1331 struct task_struct *kswapd; /* Protected by kswapd_lock */
1332 int kswapd_order;
1333 enum zone_type kswapd_highest_zoneidx;
1334
1335 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
1336
1337 #ifdef CONFIG_COMPACTION
1338 int kcompactd_max_order;
1339 enum zone_type kcompactd_highest_zoneidx;
1340 wait_queue_head_t kcompactd_wait;
1341 struct task_struct *kcompactd;
1342 bool proactive_compact_trigger;
1343 #endif
1344 /*
1345 * This is a per-node reserve of pages that are not available
1346 * to userspace allocations.
1347 */
1348 unsigned long totalreserve_pages;
1349
1350 #ifdef CONFIG_NUMA
1351 /*
1352 * node reclaim becomes active if more unmapped pages exist.
1353 */
1354 unsigned long min_unmapped_pages;
1355 unsigned long min_slab_pages;
1356 #endif /* CONFIG_NUMA */
1357
1358 /* Write-intensive fields used by page reclaim */
1359 CACHELINE_PADDING(_pad1_);
1360
1361 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1362 /*
1363 * If memory initialisation on large machines is deferred then this
1364 * is the first PFN that needs to be initialised.
1365 */
1366 unsigned long first_deferred_pfn;
1367 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1368
1369 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1370 struct deferred_split deferred_split_queue;
1371 #endif
1372
1373 #ifdef CONFIG_NUMA_BALANCING
1374 /* start time in ms of current promote rate limit period */
1375 unsigned int nbp_rl_start;
1376 /* number of promote candidate pages at start time of current rate limit period */
1377 unsigned long nbp_rl_nr_cand;
1378 /* promote threshold in ms */
1379 unsigned int nbp_threshold;
1380 /* start time in ms of current promote threshold adjustment period */
1381 unsigned int nbp_th_start;
1382 /*
1383 * number of promote candidate pages at start time of current promote
1384 * threshold adjustment period
1385 */
1386 unsigned long nbp_th_nr_cand;
1387 #endif
1388 /* Fields commonly accessed by the page reclaim scanner */
1389
1390 /*
1391 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
1392 *
1393 * Use mem_cgroup_lruvec() to look up lruvecs.
1394 */
1395 struct lruvec __lruvec;
1396
1397 unsigned long flags;
1398
1399 #ifdef CONFIG_LRU_GEN
1400 /* kswap mm walk data */
1401 struct lru_gen_mm_walk mm_walk;
1402 /* lru_gen_folio list */
1403 struct lru_gen_memcg memcg_lru;
1404 #endif
1405
1406 CACHELINE_PADDING(_pad2_);
1407
1408 /* Per-node vmstats */
1409 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
1410 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
1411 #ifdef CONFIG_NUMA
1412 struct memory_tier __rcu *memtier;
1413 #endif
1414 #ifdef CONFIG_MEMORY_FAILURE
1415 struct memory_failure_stats mf_stats;
1416 #endif
1417 } pg_data_t;
1418
1419 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
1420 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
1421
1422 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
1423 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
1424
pgdat_end_pfn(pg_data_t * pgdat)1425 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
1426 {
1427 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
1428 }
1429
1430 #include <linux/memory_hotplug.h>
1431
1432 void build_all_zonelists(pg_data_t *pgdat);
1433 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
1434 enum zone_type highest_zoneidx);
1435 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
1436 int highest_zoneidx, unsigned int alloc_flags,
1437 long free_pages);
1438 bool zone_watermark_ok(struct zone *z, unsigned int order,
1439 unsigned long mark, int highest_zoneidx,
1440 unsigned int alloc_flags);
1441 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
1442 unsigned long mark, int highest_zoneidx);
1443 /*
1444 * Memory initialization context, use to differentiate memory added by
1445 * the platform statically or via memory hotplug interface.
1446 */
1447 enum meminit_context {
1448 MEMINIT_EARLY,
1449 MEMINIT_HOTPLUG,
1450 };
1451
1452 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
1453 unsigned long size);
1454
1455 extern void lruvec_init(struct lruvec *lruvec);
1456
lruvec_pgdat(struct lruvec * lruvec)1457 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
1458 {
1459 #ifdef CONFIG_MEMCG
1460 return lruvec->pgdat;
1461 #else
1462 return container_of(lruvec, struct pglist_data, __lruvec);
1463 #endif
1464 }
1465
1466 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
1467 int local_memory_node(int node_id);
1468 #else
local_memory_node(int node_id)1469 static inline int local_memory_node(int node_id) { return node_id; };
1470 #endif
1471
1472 /*
1473 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
1474 */
1475 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
1476
1477 #ifdef CONFIG_ZONE_DEVICE
zone_is_zone_device(struct zone * zone)1478 static inline bool zone_is_zone_device(struct zone *zone)
1479 {
1480 return zone_idx(zone) == ZONE_DEVICE;
1481 }
1482 #else
zone_is_zone_device(struct zone * zone)1483 static inline bool zone_is_zone_device(struct zone *zone)
1484 {
1485 return false;
1486 }
1487 #endif
1488
1489 /*
1490 * Returns true if a zone has pages managed by the buddy allocator.
1491 * All the reclaim decisions have to use this function rather than
1492 * populated_zone(). If the whole zone is reserved then we can easily
1493 * end up with populated_zone() && !managed_zone().
1494 */
managed_zone(struct zone * zone)1495 static inline bool managed_zone(struct zone *zone)
1496 {
1497 return zone_managed_pages(zone);
1498 }
1499
1500 /* Returns true if a zone has memory */
populated_zone(struct zone * zone)1501 static inline bool populated_zone(struct zone *zone)
1502 {
1503 return zone->present_pages;
1504 }
1505
1506 #ifdef CONFIG_NUMA
zone_to_nid(struct zone * zone)1507 static inline int zone_to_nid(struct zone *zone)
1508 {
1509 return zone->node;
1510 }
1511
zone_set_nid(struct zone * zone,int nid)1512 static inline void zone_set_nid(struct zone *zone, int nid)
1513 {
1514 zone->node = nid;
1515 }
1516 #else
zone_to_nid(struct zone * zone)1517 static inline int zone_to_nid(struct zone *zone)
1518 {
1519 return 0;
1520 }
1521
zone_set_nid(struct zone * zone,int nid)1522 static inline void zone_set_nid(struct zone *zone, int nid) {}
1523 #endif
1524
1525 extern int movable_zone;
1526
is_highmem_idx(enum zone_type idx)1527 static inline int is_highmem_idx(enum zone_type idx)
1528 {
1529 #ifdef CONFIG_HIGHMEM
1530 return (idx == ZONE_HIGHMEM ||
1531 (idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM));
1532 #else
1533 return 0;
1534 #endif
1535 }
1536
1537 /**
1538 * is_highmem - helper function to quickly check if a struct zone is a
1539 * highmem zone or not. This is an attempt to keep references
1540 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
1541 * @zone: pointer to struct zone variable
1542 * Return: 1 for a highmem zone, 0 otherwise
1543 */
is_highmem(struct zone * zone)1544 static inline int is_highmem(struct zone *zone)
1545 {
1546 return is_highmem_idx(zone_idx(zone));
1547 }
1548
1549 #ifdef CONFIG_ZONE_DMA
1550 bool has_managed_dma(void);
1551 #else
has_managed_dma(void)1552 static inline bool has_managed_dma(void)
1553 {
1554 return false;
1555 }
1556 #endif
1557
1558
1559 #ifndef CONFIG_NUMA
1560
1561 extern struct pglist_data contig_page_data;
NODE_DATA(int nid)1562 static inline struct pglist_data *NODE_DATA(int nid)
1563 {
1564 return &contig_page_data;
1565 }
1566
1567 #else /* CONFIG_NUMA */
1568
1569 #include <asm/mmzone.h>
1570
1571 #endif /* !CONFIG_NUMA */
1572
1573 extern struct pglist_data *first_online_pgdat(void);
1574 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
1575 extern struct zone *next_zone(struct zone *zone);
1576
1577 /**
1578 * for_each_online_pgdat - helper macro to iterate over all online nodes
1579 * @pgdat: pointer to a pg_data_t variable
1580 */
1581 #define for_each_online_pgdat(pgdat) \
1582 for (pgdat = first_online_pgdat(); \
1583 pgdat; \
1584 pgdat = next_online_pgdat(pgdat))
1585 /**
1586 * for_each_zone - helper macro to iterate over all memory zones
1587 * @zone: pointer to struct zone variable
1588 *
1589 * The user only needs to declare the zone variable, for_each_zone
1590 * fills it in.
1591 */
1592 #define for_each_zone(zone) \
1593 for (zone = (first_online_pgdat())->node_zones; \
1594 zone; \
1595 zone = next_zone(zone))
1596
1597 #define for_each_populated_zone(zone) \
1598 for (zone = (first_online_pgdat())->node_zones; \
1599 zone; \
1600 zone = next_zone(zone)) \
1601 if (!populated_zone(zone)) \
1602 ; /* do nothing */ \
1603 else
1604
zonelist_zone(struct zoneref * zoneref)1605 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1606 {
1607 return zoneref->zone;
1608 }
1609
zonelist_zone_idx(struct zoneref * zoneref)1610 static inline int zonelist_zone_idx(struct zoneref *zoneref)
1611 {
1612 return zoneref->zone_idx;
1613 }
1614
zonelist_node_idx(struct zoneref * zoneref)1615 static inline int zonelist_node_idx(struct zoneref *zoneref)
1616 {
1617 return zone_to_nid(zoneref->zone);
1618 }
1619
1620 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1621 enum zone_type highest_zoneidx,
1622 nodemask_t *nodes);
1623
1624 /**
1625 * 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
1626 * @z: The cursor used as a starting point for the search
1627 * @highest_zoneidx: The zone index of the highest zone to return
1628 * @nodes: An optional nodemask to filter the zonelist with
1629 *
1630 * This function returns the next zone at or below a given zone index that is
1631 * within the allowed nodemask using a cursor as the starting point for the
1632 * search. The zoneref returned is a cursor that represents the current zone
1633 * being examined. It should be advanced by one before calling
1634 * next_zones_zonelist again.
1635 *
1636 * Return: the next zone at or below highest_zoneidx within the allowed
1637 * nodemask using a cursor within a zonelist as a starting point
1638 */
next_zones_zonelist(struct zoneref * z,enum zone_type highest_zoneidx,nodemask_t * nodes)1639 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1640 enum zone_type highest_zoneidx,
1641 nodemask_t *nodes)
1642 {
1643 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1644 return z;
1645 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1646 }
1647
1648 /**
1649 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1650 * @zonelist: The zonelist to search for a suitable zone
1651 * @highest_zoneidx: The zone index of the highest zone to return
1652 * @nodes: An optional nodemask to filter the zonelist with
1653 *
1654 * This function returns the first zone at or below a given zone index that is
1655 * within the allowed nodemask. The zoneref returned is a cursor that can be
1656 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1657 * one before calling.
1658 *
1659 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1660 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1661 * update due to cpuset modification.
1662 *
1663 * Return: Zoneref pointer for the first suitable zone found
1664 */
first_zones_zonelist(struct zonelist * zonelist,enum zone_type highest_zoneidx,nodemask_t * nodes)1665 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1666 enum zone_type highest_zoneidx,
1667 nodemask_t *nodes)
1668 {
1669 return next_zones_zonelist(zonelist->_zonerefs,
1670 highest_zoneidx, nodes);
1671 }
1672
1673 /**
1674 * 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
1675 * @zone: The current zone in the iterator
1676 * @z: The current pointer within zonelist->_zonerefs being iterated
1677 * @zlist: The zonelist being iterated
1678 * @highidx: The zone index of the highest zone to return
1679 * @nodemask: Nodemask allowed by the allocator
1680 *
1681 * This iterator iterates though all zones at or below a given zone index and
1682 * within a given nodemask
1683 */
1684 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1685 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1686 zone; \
1687 z = next_zones_zonelist(++z, highidx, nodemask), \
1688 zone = zonelist_zone(z))
1689
1690 #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
1691 for (zone = z->zone; \
1692 zone; \
1693 z = next_zones_zonelist(++z, highidx, nodemask), \
1694 zone = zonelist_zone(z))
1695
1696
1697 /**
1698 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1699 * @zone: The current zone in the iterator
1700 * @z: The current pointer within zonelist->zones being iterated
1701 * @zlist: The zonelist being iterated
1702 * @highidx: The zone index of the highest zone to return
1703 *
1704 * This iterator iterates though all zones at or below a given zone index.
1705 */
1706 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1707 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1708
1709 /* Whether the 'nodes' are all movable nodes */
movable_only_nodes(nodemask_t * nodes)1710 static inline bool movable_only_nodes(nodemask_t *nodes)
1711 {
1712 struct zonelist *zonelist;
1713 struct zoneref *z;
1714 int nid;
1715
1716 if (nodes_empty(*nodes))
1717 return false;
1718
1719 /*
1720 * We can chose arbitrary node from the nodemask to get a
1721 * zonelist as they are interlinked. We just need to find
1722 * at least one zone that can satisfy kernel allocations.
1723 */
1724 nid = first_node(*nodes);
1725 zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK];
1726 z = first_zones_zonelist(zonelist, ZONE_NORMAL, nodes);
1727 return (!z->zone) ? true : false;
1728 }
1729
1730
1731 #ifdef CONFIG_SPARSEMEM
1732 #include <asm/sparsemem.h>
1733 #endif
1734
1735 #ifdef CONFIG_FLATMEM
1736 #define pfn_to_nid(pfn) (0)
1737 #endif
1738
1739 #ifdef CONFIG_SPARSEMEM
1740
1741 /*
1742 * PA_SECTION_SHIFT physical address to/from section number
1743 * PFN_SECTION_SHIFT pfn to/from section number
1744 */
1745 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1746 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1747
1748 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1749
1750 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1751 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1752
1753 #define SECTION_BLOCKFLAGS_BITS \
1754 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1755
1756 #if (MAX_PAGE_ORDER + PAGE_SHIFT) > SECTION_SIZE_BITS
1757 #error Allocator MAX_PAGE_ORDER exceeds SECTION_SIZE
1758 #endif
1759
pfn_to_section_nr(unsigned long pfn)1760 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1761 {
1762 return pfn >> PFN_SECTION_SHIFT;
1763 }
section_nr_to_pfn(unsigned long sec)1764 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1765 {
1766 return sec << PFN_SECTION_SHIFT;
1767 }
1768
1769 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1770 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1771
1772 #define SUBSECTION_SHIFT 21
1773 #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1774
1775 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1776 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1777 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1778
1779 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1780 #error Subsection size exceeds section size
1781 #else
1782 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1783 #endif
1784
1785 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1786 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1787
1788 struct mem_section_usage {
1789 struct rcu_head rcu;
1790 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1791 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1792 #endif
1793 /* See declaration of similar field in struct zone */
1794 unsigned long pageblock_flags[0];
1795 };
1796
1797 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1798
1799 struct page;
1800 struct page_ext;
1801 struct mem_section {
1802 /*
1803 * This is, logically, a pointer to an array of struct
1804 * pages. However, it is stored with some other magic.
1805 * (see sparse.c::sparse_init_one_section())
1806 *
1807 * Additionally during early boot we encode node id of
1808 * the location of the section here to guide allocation.
1809 * (see sparse.c::memory_present())
1810 *
1811 * Making it a UL at least makes someone do a cast
1812 * before using it wrong.
1813 */
1814 unsigned long section_mem_map;
1815
1816 struct mem_section_usage *usage;
1817 #ifdef CONFIG_PAGE_EXTENSION
1818 /*
1819 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1820 * section. (see page_ext.h about this.)
1821 */
1822 struct page_ext *page_ext;
1823 unsigned long pad;
1824 #endif
1825 /*
1826 * WARNING: mem_section must be a power-of-2 in size for the
1827 * calculation and use of SECTION_ROOT_MASK to make sense.
1828 */
1829 };
1830
1831 #ifdef CONFIG_SPARSEMEM_EXTREME
1832 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1833 #else
1834 #define SECTIONS_PER_ROOT 1
1835 #endif
1836
1837 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1838 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1839 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1840
1841 #ifdef CONFIG_SPARSEMEM_EXTREME
1842 extern struct mem_section **mem_section;
1843 #else
1844 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1845 #endif
1846
section_to_usemap(struct mem_section * ms)1847 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1848 {
1849 return ms->usage->pageblock_flags;
1850 }
1851
__nr_to_section(unsigned long nr)1852 static inline struct mem_section *__nr_to_section(unsigned long nr)
1853 {
1854 unsigned long root = SECTION_NR_TO_ROOT(nr);
1855
1856 if (unlikely(root >= NR_SECTION_ROOTS))
1857 return NULL;
1858
1859 #ifdef CONFIG_SPARSEMEM_EXTREME
1860 if (!mem_section || !mem_section[root])
1861 return NULL;
1862 #endif
1863 return &mem_section[root][nr & SECTION_ROOT_MASK];
1864 }
1865 extern size_t mem_section_usage_size(void);
1866
1867 /*
1868 * We use the lower bits of the mem_map pointer to store
1869 * a little bit of information. The pointer is calculated
1870 * as mem_map - section_nr_to_pfn(pnum). The result is
1871 * aligned to the minimum alignment of the two values:
1872 * 1. All mem_map arrays are page-aligned.
1873 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1874 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1875 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1876 * worst combination is powerpc with 256k pages,
1877 * which results in PFN_SECTION_SHIFT equal 6.
1878 * To sum it up, at least 6 bits are available on all architectures.
1879 * However, we can exceed 6 bits on some other architectures except
1880 * powerpc (e.g. 15 bits are available on x86_64, 13 bits are available
1881 * with the worst case of 64K pages on arm64) if we make sure the
1882 * exceeded bit is not applicable to powerpc.
1883 */
1884 enum {
1885 SECTION_MARKED_PRESENT_BIT,
1886 SECTION_HAS_MEM_MAP_BIT,
1887 SECTION_IS_ONLINE_BIT,
1888 SECTION_IS_EARLY_BIT,
1889 #ifdef CONFIG_ZONE_DEVICE
1890 SECTION_TAINT_ZONE_DEVICE_BIT,
1891 #endif
1892 SECTION_MAP_LAST_BIT,
1893 };
1894
1895 #define SECTION_MARKED_PRESENT BIT(SECTION_MARKED_PRESENT_BIT)
1896 #define SECTION_HAS_MEM_MAP BIT(SECTION_HAS_MEM_MAP_BIT)
1897 #define SECTION_IS_ONLINE BIT(SECTION_IS_ONLINE_BIT)
1898 #define SECTION_IS_EARLY BIT(SECTION_IS_EARLY_BIT)
1899 #ifdef CONFIG_ZONE_DEVICE
1900 #define SECTION_TAINT_ZONE_DEVICE BIT(SECTION_TAINT_ZONE_DEVICE_BIT)
1901 #endif
1902 #define SECTION_MAP_MASK (~(BIT(SECTION_MAP_LAST_BIT) - 1))
1903 #define SECTION_NID_SHIFT SECTION_MAP_LAST_BIT
1904
__section_mem_map_addr(struct mem_section * section)1905 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1906 {
1907 unsigned long map = section->section_mem_map;
1908 map &= SECTION_MAP_MASK;
1909 return (struct page *)map;
1910 }
1911
present_section(struct mem_section * section)1912 static inline int present_section(struct mem_section *section)
1913 {
1914 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1915 }
1916
present_section_nr(unsigned long nr)1917 static inline int present_section_nr(unsigned long nr)
1918 {
1919 return present_section(__nr_to_section(nr));
1920 }
1921
valid_section(struct mem_section * section)1922 static inline int valid_section(struct mem_section *section)
1923 {
1924 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1925 }
1926
early_section(struct mem_section * section)1927 static inline int early_section(struct mem_section *section)
1928 {
1929 return (section && (section->section_mem_map & SECTION_IS_EARLY));
1930 }
1931
valid_section_nr(unsigned long nr)1932 static inline int valid_section_nr(unsigned long nr)
1933 {
1934 return valid_section(__nr_to_section(nr));
1935 }
1936
online_section(struct mem_section * section)1937 static inline int online_section(struct mem_section *section)
1938 {
1939 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1940 }
1941
1942 #ifdef CONFIG_ZONE_DEVICE
online_device_section(struct mem_section * section)1943 static inline int online_device_section(struct mem_section *section)
1944 {
1945 unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE;
1946
1947 return section && ((section->section_mem_map & flags) == flags);
1948 }
1949 #else
online_device_section(struct mem_section * section)1950 static inline int online_device_section(struct mem_section *section)
1951 {
1952 return 0;
1953 }
1954 #endif
1955
online_section_nr(unsigned long nr)1956 static inline int online_section_nr(unsigned long nr)
1957 {
1958 return online_section(__nr_to_section(nr));
1959 }
1960
1961 #ifdef CONFIG_MEMORY_HOTPLUG
1962 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1963 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1964 #endif
1965
__pfn_to_section(unsigned long pfn)1966 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1967 {
1968 return __nr_to_section(pfn_to_section_nr(pfn));
1969 }
1970
1971 extern unsigned long __highest_present_section_nr;
1972
subsection_map_index(unsigned long pfn)1973 static inline int subsection_map_index(unsigned long pfn)
1974 {
1975 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1976 }
1977
1978 #ifdef CONFIG_SPARSEMEM_VMEMMAP
pfn_section_valid(struct mem_section * ms,unsigned long pfn)1979 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1980 {
1981 int idx = subsection_map_index(pfn);
1982 struct mem_section_usage *usage = READ_ONCE(ms->usage);
1983
1984 return usage ? test_bit(idx, usage->subsection_map) : 0;
1985 }
1986 #else
pfn_section_valid(struct mem_section * ms,unsigned long pfn)1987 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1988 {
1989 return 1;
1990 }
1991 #endif
1992
1993 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1994 /**
1995 * pfn_valid - check if there is a valid memory map entry for a PFN
1996 * @pfn: the page frame number to check
1997 *
1998 * Check if there is a valid memory map entry aka struct page for the @pfn.
1999 * Note, that availability of the memory map entry does not imply that
2000 * there is actual usable memory at that @pfn. The struct page may
2001 * represent a hole or an unusable page frame.
2002 *
2003 * Return: 1 for PFNs that have memory map entries and 0 otherwise
2004 */
pfn_valid(unsigned long pfn)2005 static inline int pfn_valid(unsigned long pfn)
2006 {
2007 struct mem_section *ms;
2008 int ret;
2009
2010 /*
2011 * Ensure the upper PAGE_SHIFT bits are clear in the
2012 * pfn. Else it might lead to false positives when
2013 * some of the upper bits are set, but the lower bits
2014 * match a valid pfn.
2015 */
2016 if (PHYS_PFN(PFN_PHYS(pfn)) != pfn)
2017 return 0;
2018
2019 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2020 return 0;
2021 ms = __pfn_to_section(pfn);
2022 rcu_read_lock_sched();
2023 if (!valid_section(ms)) {
2024 rcu_read_unlock_sched();
2025 return 0;
2026 }
2027 /*
2028 * Traditionally early sections always returned pfn_valid() for
2029 * the entire section-sized span.
2030 */
2031 ret = early_section(ms) || pfn_section_valid(ms, pfn);
2032 rcu_read_unlock_sched();
2033
2034 return ret;
2035 }
2036 #endif
2037
pfn_in_present_section(unsigned long pfn)2038 static inline int pfn_in_present_section(unsigned long pfn)
2039 {
2040 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2041 return 0;
2042 return present_section(__pfn_to_section(pfn));
2043 }
2044
next_present_section_nr(unsigned long section_nr)2045 static inline unsigned long next_present_section_nr(unsigned long section_nr)
2046 {
2047 while (++section_nr <= __highest_present_section_nr) {
2048 if (present_section_nr(section_nr))
2049 return section_nr;
2050 }
2051
2052 return -1;
2053 }
2054
2055 /*
2056 * These are _only_ used during initialisation, therefore they
2057 * can use __initdata ... They could have names to indicate
2058 * this restriction.
2059 */
2060 #ifdef CONFIG_NUMA
2061 #define pfn_to_nid(pfn) \
2062 ({ \
2063 unsigned long __pfn_to_nid_pfn = (pfn); \
2064 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
2065 })
2066 #else
2067 #define pfn_to_nid(pfn) (0)
2068 #endif
2069
2070 void sparse_init(void);
2071 #else
2072 #define sparse_init() do {} while (0)
2073 #define sparse_index_init(_sec, _nid) do {} while (0)
2074 #define pfn_in_present_section pfn_valid
2075 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
2076 #endif /* CONFIG_SPARSEMEM */
2077
2078 #endif /* !__GENERATING_BOUNDS.H */
2079 #endif /* !__ASSEMBLY__ */
2080 #endif /* _LINUX_MMZONE_H */
2081