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_MEMMAP, /* page metadata allocated through buddy allocator */
224 NR_MEMMAP_BOOT, /* page metadata allocated through boot allocator */
225 NR_VM_NODE_STAT_ITEMS
226 };
227
228 /*
229 * Returns true if the item should be printed in THPs (/proc/vmstat
230 * currently prints number of anon, file and shmem THPs. But the item
231 * is charged in pages).
232 */
vmstat_item_print_in_thp(enum node_stat_item item)233 static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item)
234 {
235 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
236 return false;
237
238 return item == NR_ANON_THPS ||
239 item == NR_FILE_THPS ||
240 item == NR_SHMEM_THPS ||
241 item == NR_SHMEM_PMDMAPPED ||
242 item == NR_FILE_PMDMAPPED;
243 }
244
245 /*
246 * Returns true if the value is measured in bytes (most vmstat values are
247 * measured in pages). This defines the API part, the internal representation
248 * might be different.
249 */
vmstat_item_in_bytes(int idx)250 static __always_inline bool vmstat_item_in_bytes(int idx)
251 {
252 /*
253 * Global and per-node slab counters track slab pages.
254 * It's expected that changes are multiples of PAGE_SIZE.
255 * Internally values are stored in pages.
256 *
257 * Per-memcg and per-lruvec counters track memory, consumed
258 * by individual slab objects. These counters are actually
259 * byte-precise.
260 */
261 return (idx == NR_SLAB_RECLAIMABLE_B ||
262 idx == NR_SLAB_UNRECLAIMABLE_B);
263 }
264
265 /*
266 * We do arithmetic on the LRU lists in various places in the code,
267 * so it is important to keep the active lists LRU_ACTIVE higher in
268 * the array than the corresponding inactive lists, and to keep
269 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
270 *
271 * This has to be kept in sync with the statistics in zone_stat_item
272 * above and the descriptions in vmstat_text in mm/vmstat.c
273 */
274 #define LRU_BASE 0
275 #define LRU_ACTIVE 1
276 #define LRU_FILE 2
277
278 enum lru_list {
279 LRU_INACTIVE_ANON = LRU_BASE,
280 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
281 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
282 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
283 LRU_UNEVICTABLE,
284 NR_LRU_LISTS
285 };
286
287 enum vmscan_throttle_state {
288 VMSCAN_THROTTLE_WRITEBACK,
289 VMSCAN_THROTTLE_ISOLATED,
290 VMSCAN_THROTTLE_NOPROGRESS,
291 VMSCAN_THROTTLE_CONGESTED,
292 NR_VMSCAN_THROTTLE,
293 };
294
295 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
296
297 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
298
is_file_lru(enum lru_list lru)299 static inline bool is_file_lru(enum lru_list lru)
300 {
301 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
302 }
303
is_active_lru(enum lru_list lru)304 static inline bool is_active_lru(enum lru_list lru)
305 {
306 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
307 }
308
309 #define WORKINGSET_ANON 0
310 #define WORKINGSET_FILE 1
311 #define ANON_AND_FILE 2
312
313 enum lruvec_flags {
314 /*
315 * An lruvec has many dirty pages backed by a congested BDI:
316 * 1. LRUVEC_CGROUP_CONGESTED is set by cgroup-level reclaim.
317 * It can be cleared by cgroup reclaim or kswapd.
318 * 2. LRUVEC_NODE_CONGESTED is set by kswapd node-level reclaim.
319 * It can only be cleared by kswapd.
320 *
321 * Essentially, kswapd can unthrottle an lruvec throttled by cgroup
322 * reclaim, but not vice versa. This only applies to the root cgroup.
323 * The goal is to prevent cgroup reclaim on the root cgroup (e.g.
324 * memory.reclaim) to unthrottle an unbalanced node (that was throttled
325 * by kswapd).
326 */
327 LRUVEC_CGROUP_CONGESTED,
328 LRUVEC_NODE_CONGESTED,
329 };
330
331 #endif /* !__GENERATING_BOUNDS_H */
332
333 /*
334 * Evictable pages are divided into multiple generations. The youngest and the
335 * oldest generation numbers, max_seq and min_seq, are monotonically increasing.
336 * They form a sliding window of a variable size [MIN_NR_GENS, MAX_NR_GENS]. An
337 * offset within MAX_NR_GENS, i.e., gen, indexes the LRU list of the
338 * corresponding generation. The gen counter in folio->flags stores gen+1 while
339 * a page is on one of lrugen->folios[]. Otherwise it stores 0.
340 *
341 * A page is added to the youngest generation on faulting. The aging needs to
342 * check the accessed bit at least twice before handing this page over to the
343 * eviction. The first check takes care of the accessed bit set on the initial
344 * fault; the second check makes sure this page hasn't been used since then.
345 * This process, AKA second chance, requires a minimum of two generations,
346 * hence MIN_NR_GENS. And to maintain ABI compatibility with the active/inactive
347 * LRU, e.g., /proc/vmstat, these two generations are considered active; the
348 * rest of generations, if they exist, are considered inactive. See
349 * lru_gen_is_active().
350 *
351 * PG_active is always cleared while a page is on one of lrugen->folios[] so
352 * that the aging needs not to worry about it. And it's set again when a page
353 * considered active is isolated for non-reclaiming purposes, e.g., migration.
354 * See lru_gen_add_folio() and lru_gen_del_folio().
355 *
356 * MAX_NR_GENS 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 page tables. This requires order_base_2(MAX_NR_GENS+1) bits
359 * in folio->flags.
360 */
361 #define MIN_NR_GENS 2U
362 #define MAX_NR_GENS 4U
363
364 /*
365 * Each generation is divided into multiple tiers. A page accessed N times
366 * through file descriptors is in tier order_base_2(N). A page in the first tier
367 * (N=0,1) is marked by PG_referenced unless it was faulted in through page
368 * tables or read ahead. A page in any other tier (N>1) is marked by
369 * PG_referenced and PG_workingset. This implies a minimum of two tiers is
370 * supported without using additional bits in folio->flags.
371 *
372 * In contrast to moving across generations which requires the LRU lock, moving
373 * across tiers only involves atomic operations on folio->flags and therefore
374 * has a negligible cost in the buffered access path. In the eviction path,
375 * comparisons of refaulted/(evicted+protected) from the first tier and the
376 * rest infer whether pages accessed multiple times through file descriptors
377 * are statistically hot and thus worth protecting.
378 *
379 * MAX_NR_TIERS is set to 4 so that the multi-gen LRU can support twice the
380 * number of categories of the active/inactive LRU when keeping track of
381 * accesses through file descriptors. This uses MAX_NR_TIERS-2 spare bits in
382 * folio->flags.
383 */
384 #define MAX_NR_TIERS 4U
385
386 #ifndef __GENERATING_BOUNDS_H
387
388 struct lruvec;
389 struct page_vma_mapped_walk;
390
391 #define LRU_GEN_MASK ((BIT(LRU_GEN_WIDTH) - 1) << LRU_GEN_PGOFF)
392 #define LRU_REFS_MASK ((BIT(LRU_REFS_WIDTH) - 1) << LRU_REFS_PGOFF)
393
394 #ifdef CONFIG_LRU_GEN
395
396 enum {
397 LRU_GEN_ANON,
398 LRU_GEN_FILE,
399 };
400
401 enum {
402 LRU_GEN_CORE,
403 LRU_GEN_MM_WALK,
404 LRU_GEN_NONLEAF_YOUNG,
405 NR_LRU_GEN_CAPS
406 };
407
408 #define MIN_LRU_BATCH BITS_PER_LONG
409 #define MAX_LRU_BATCH (MIN_LRU_BATCH * 64)
410
411 /* whether to keep historical stats from evicted generations */
412 #ifdef CONFIG_LRU_GEN_STATS
413 #define NR_HIST_GENS MAX_NR_GENS
414 #else
415 #define NR_HIST_GENS 1U
416 #endif
417
418 /*
419 * The youngest generation number is stored in max_seq for both anon and file
420 * types as they are aged on an equal footing. The oldest generation numbers are
421 * stored in min_seq[] separately for anon and file types as clean file pages
422 * can be evicted regardless of swap constraints.
423 *
424 * Normally anon and file min_seq are in sync. But if swapping is constrained,
425 * e.g., out of swap space, file min_seq is allowed to advance and leave anon
426 * min_seq behind.
427 *
428 * The number of pages in each generation is eventually consistent and therefore
429 * can be transiently negative when reset_batch_size() is pending.
430 */
431 struct lru_gen_folio {
432 /* the aging increments the youngest generation number */
433 unsigned long max_seq;
434 /* the eviction increments the oldest generation numbers */
435 unsigned long min_seq[ANON_AND_FILE];
436 /* the birth time of each generation in jiffies */
437 unsigned long timestamps[MAX_NR_GENS];
438 /* the multi-gen LRU lists, lazily sorted on eviction */
439 struct list_head folios[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
440 /* the multi-gen LRU sizes, eventually consistent */
441 long nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
442 /* the exponential moving average of refaulted */
443 unsigned long avg_refaulted[ANON_AND_FILE][MAX_NR_TIERS];
444 /* the exponential moving average of evicted+protected */
445 unsigned long avg_total[ANON_AND_FILE][MAX_NR_TIERS];
446 /* the first tier doesn't need protection, hence the minus one */
447 unsigned long protected[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS - 1];
448 /* can be modified without holding the LRU lock */
449 atomic_long_t evicted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
450 atomic_long_t refaulted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
451 /* whether the multi-gen LRU is enabled */
452 bool enabled;
453 /* the memcg generation this lru_gen_folio belongs to */
454 u8 gen;
455 /* the list segment this lru_gen_folio belongs to */
456 u8 seg;
457 /* per-node lru_gen_folio list for global reclaim */
458 struct hlist_nulls_node list;
459 };
460
461 enum {
462 MM_LEAF_TOTAL, /* total leaf entries */
463 MM_LEAF_OLD, /* old leaf entries */
464 MM_LEAF_YOUNG, /* young leaf entries */
465 MM_NONLEAF_TOTAL, /* total non-leaf entries */
466 MM_NONLEAF_FOUND, /* non-leaf entries found in Bloom filters */
467 MM_NONLEAF_ADDED, /* non-leaf entries added to Bloom filters */
468 NR_MM_STATS
469 };
470
471 /* double-buffering Bloom filters */
472 #define NR_BLOOM_FILTERS 2
473
474 struct lru_gen_mm_state {
475 /* synced with max_seq after each iteration */
476 unsigned long seq;
477 /* where the current iteration continues after */
478 struct list_head *head;
479 /* where the last iteration ended before */
480 struct list_head *tail;
481 /* Bloom filters flip after each iteration */
482 unsigned long *filters[NR_BLOOM_FILTERS];
483 /* the mm stats for debugging */
484 unsigned long stats[NR_HIST_GENS][NR_MM_STATS];
485 };
486
487 struct lru_gen_mm_walk {
488 /* the lruvec under reclaim */
489 struct lruvec *lruvec;
490 /* max_seq from lru_gen_folio: can be out of date */
491 unsigned long seq;
492 /* the next address within an mm to scan */
493 unsigned long next_addr;
494 /* to batch promoted pages */
495 int nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
496 /* to batch the mm stats */
497 int mm_stats[NR_MM_STATS];
498 /* total batched items */
499 int batched;
500 bool can_swap;
501 bool force_scan;
502 };
503
504 /*
505 * For each node, memcgs are divided into two generations: the old and the
506 * young. For each generation, memcgs are randomly sharded into multiple bins
507 * to improve scalability. For each bin, the hlist_nulls is virtually divided
508 * into three segments: the head, the tail and the default.
509 *
510 * An onlining memcg is added to the tail of a random bin in the old generation.
511 * The eviction starts at the head of a random bin in the old generation. The
512 * per-node memcg generation counter, whose reminder (mod MEMCG_NR_GENS) indexes
513 * the old generation, is incremented when all its bins become empty.
514 *
515 * There are four operations:
516 * 1. MEMCG_LRU_HEAD, which moves a memcg to the head of a random bin in its
517 * current generation (old or young) and updates its "seg" to "head";
518 * 2. MEMCG_LRU_TAIL, which moves a memcg to the tail of a random bin in its
519 * current generation (old or young) and updates its "seg" to "tail";
520 * 3. MEMCG_LRU_OLD, which moves a memcg to the head of a random bin in the old
521 * generation, updates its "gen" to "old" and resets its "seg" to "default";
522 * 4. MEMCG_LRU_YOUNG, which moves a memcg to the tail of a random bin in the
523 * young generation, updates its "gen" to "young" and resets its "seg" to
524 * "default".
525 *
526 * The events that trigger the above operations are:
527 * 1. Exceeding the soft limit, which triggers MEMCG_LRU_HEAD;
528 * 2. The first attempt to reclaim a memcg below low, which triggers
529 * MEMCG_LRU_TAIL;
530 * 3. The first attempt to reclaim a memcg offlined or below reclaimable size
531 * threshold, which triggers MEMCG_LRU_TAIL;
532 * 4. The second attempt to reclaim a memcg offlined or below reclaimable size
533 * threshold, which triggers MEMCG_LRU_YOUNG;
534 * 5. Attempting to reclaim a memcg below min, which triggers MEMCG_LRU_YOUNG;
535 * 6. Finishing the aging on the eviction path, which triggers MEMCG_LRU_YOUNG;
536 * 7. Offlining a memcg, which triggers MEMCG_LRU_OLD.
537 *
538 * Notes:
539 * 1. Memcg LRU only applies to global reclaim, and the round-robin incrementing
540 * of their max_seq counters ensures the eventual fairness to all eligible
541 * memcgs. For memcg reclaim, it still relies on mem_cgroup_iter().
542 * 2. There are only two valid generations: old (seq) and young (seq+1).
543 * MEMCG_NR_GENS is set to three so that when reading the generation counter
544 * locklessly, a stale value (seq-1) does not wraparound to young.
545 */
546 #define MEMCG_NR_GENS 3
547 #define MEMCG_NR_BINS 8
548
549 struct lru_gen_memcg {
550 /* the per-node memcg generation counter */
551 unsigned long seq;
552 /* each memcg has one lru_gen_folio per node */
553 unsigned long nr_memcgs[MEMCG_NR_GENS];
554 /* per-node lru_gen_folio list for global reclaim */
555 struct hlist_nulls_head fifo[MEMCG_NR_GENS][MEMCG_NR_BINS];
556 /* protects the above */
557 spinlock_t lock;
558 };
559
560 void lru_gen_init_pgdat(struct pglist_data *pgdat);
561 void lru_gen_init_lruvec(struct lruvec *lruvec);
562 void lru_gen_look_around(struct page_vma_mapped_walk *pvmw);
563
564 void lru_gen_init_memcg(struct mem_cgroup *memcg);
565 void lru_gen_exit_memcg(struct mem_cgroup *memcg);
566 void lru_gen_online_memcg(struct mem_cgroup *memcg);
567 void lru_gen_offline_memcg(struct mem_cgroup *memcg);
568 void lru_gen_release_memcg(struct mem_cgroup *memcg);
569 void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid);
570
571 #else /* !CONFIG_LRU_GEN */
572
lru_gen_init_pgdat(struct pglist_data * pgdat)573 static inline void lru_gen_init_pgdat(struct pglist_data *pgdat)
574 {
575 }
576
lru_gen_init_lruvec(struct lruvec * lruvec)577 static inline void lru_gen_init_lruvec(struct lruvec *lruvec)
578 {
579 }
580
lru_gen_look_around(struct page_vma_mapped_walk * pvmw)581 static inline void lru_gen_look_around(struct page_vma_mapped_walk *pvmw)
582 {
583 }
584
lru_gen_init_memcg(struct mem_cgroup * memcg)585 static inline void lru_gen_init_memcg(struct mem_cgroup *memcg)
586 {
587 }
588
lru_gen_exit_memcg(struct mem_cgroup * memcg)589 static inline void lru_gen_exit_memcg(struct mem_cgroup *memcg)
590 {
591 }
592
lru_gen_online_memcg(struct mem_cgroup * memcg)593 static inline void lru_gen_online_memcg(struct mem_cgroup *memcg)
594 {
595 }
596
lru_gen_offline_memcg(struct mem_cgroup * memcg)597 static inline void lru_gen_offline_memcg(struct mem_cgroup *memcg)
598 {
599 }
600
lru_gen_release_memcg(struct mem_cgroup * memcg)601 static inline void lru_gen_release_memcg(struct mem_cgroup *memcg)
602 {
603 }
604
lru_gen_soft_reclaim(struct mem_cgroup * memcg,int nid)605 static inline void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid)
606 {
607 }
608
609 #endif /* CONFIG_LRU_GEN */
610
611 struct lruvec {
612 struct list_head lists[NR_LRU_LISTS];
613 /* per lruvec lru_lock for memcg */
614 spinlock_t lru_lock;
615 /*
616 * These track the cost of reclaiming one LRU - file or anon -
617 * over the other. As the observed cost of reclaiming one LRU
618 * increases, the reclaim scan balance tips toward the other.
619 */
620 unsigned long anon_cost;
621 unsigned long file_cost;
622 /* Non-resident age, driven by LRU movement */
623 atomic_long_t nonresident_age;
624 /* Refaults at the time of last reclaim cycle */
625 unsigned long refaults[ANON_AND_FILE];
626 /* Various lruvec state flags (enum lruvec_flags) */
627 unsigned long flags;
628 #ifdef CONFIG_LRU_GEN
629 /* evictable pages divided into generations */
630 struct lru_gen_folio lrugen;
631 #ifdef CONFIG_LRU_GEN_WALKS_MMU
632 /* to concurrently iterate lru_gen_mm_list */
633 struct lru_gen_mm_state mm_state;
634 #endif
635 #endif /* CONFIG_LRU_GEN */
636 #ifdef CONFIG_MEMCG
637 struct pglist_data *pgdat;
638 #endif
639 struct zswap_lruvec_state zswap_lruvec_state;
640 };
641
642 /* Isolate for asynchronous migration */
643 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
644 /* Isolate unevictable pages */
645 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
646
647 /* LRU Isolation modes. */
648 typedef unsigned __bitwise isolate_mode_t;
649
650 enum zone_watermarks {
651 WMARK_MIN,
652 WMARK_LOW,
653 WMARK_HIGH,
654 WMARK_PROMO,
655 NR_WMARK
656 };
657
658 /*
659 * One per migratetype for each PAGE_ALLOC_COSTLY_ORDER. Two additional lists
660 * are added for THP. One PCP list is used by GPF_MOVABLE, and the other PCP list
661 * is used by GFP_UNMOVABLE and GFP_RECLAIMABLE.
662 */
663 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
664 #define NR_PCP_THP 2
665 #else
666 #define NR_PCP_THP 0
667 #endif
668 #define NR_LOWORDER_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1))
669 #define NR_PCP_LISTS (NR_LOWORDER_PCP_LISTS + NR_PCP_THP)
670
671 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
672 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
673 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
674 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
675
676 /*
677 * Flags used in pcp->flags field.
678 *
679 * PCPF_PREV_FREE_HIGH_ORDER: a high-order page is freed in the
680 * previous page freeing. To avoid to drain PCP for an accident
681 * high-order page freeing.
682 *
683 * PCPF_FREE_HIGH_BATCH: preserve "pcp->batch" pages in PCP before
684 * draining PCP for consecutive high-order pages freeing without
685 * allocation if data cache slice of CPU is large enough. To reduce
686 * zone lock contention and keep cache-hot pages reusing.
687 */
688 #define PCPF_PREV_FREE_HIGH_ORDER BIT(0)
689 #define PCPF_FREE_HIGH_BATCH BIT(1)
690
691 struct per_cpu_pages {
692 spinlock_t lock; /* Protects lists field */
693 int count; /* number of pages in the list */
694 int high; /* high watermark, emptying needed */
695 int high_min; /* min high watermark */
696 int high_max; /* max high watermark */
697 int batch; /* chunk size for buddy add/remove */
698 u8 flags; /* protected by pcp->lock */
699 u8 alloc_factor; /* batch scaling factor during allocate */
700 #ifdef CONFIG_NUMA
701 u8 expire; /* When 0, remote pagesets are drained */
702 #endif
703 short free_count; /* consecutive free count */
704
705 /* Lists of pages, one per migrate type stored on the pcp-lists */
706 struct list_head lists[NR_PCP_LISTS];
707 } ____cacheline_aligned_in_smp;
708
709 struct per_cpu_zonestat {
710 #ifdef CONFIG_SMP
711 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
712 s8 stat_threshold;
713 #endif
714 #ifdef CONFIG_NUMA
715 /*
716 * Low priority inaccurate counters that are only folded
717 * on demand. Use a large type to avoid the overhead of
718 * folding during refresh_cpu_vm_stats.
719 */
720 unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
721 #endif
722 };
723
724 struct per_cpu_nodestat {
725 s8 stat_threshold;
726 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
727 };
728
729 #endif /* !__GENERATING_BOUNDS.H */
730
731 enum zone_type {
732 /*
733 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
734 * to DMA to all of the addressable memory (ZONE_NORMAL).
735 * On architectures where this area covers the whole 32 bit address
736 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
737 * DMA addressing constraints. This distinction is important as a 32bit
738 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
739 * platforms may need both zones as they support peripherals with
740 * different DMA addressing limitations.
741 */
742 #ifdef CONFIG_ZONE_DMA
743 ZONE_DMA,
744 #endif
745 #ifdef CONFIG_ZONE_DMA32
746 ZONE_DMA32,
747 #endif
748 /*
749 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
750 * performed on pages in ZONE_NORMAL if the DMA devices support
751 * transfers to all addressable memory.
752 */
753 ZONE_NORMAL,
754 #ifdef CONFIG_HIGHMEM
755 /*
756 * A memory area that is only addressable by the kernel through
757 * mapping portions into its own address space. This is for example
758 * used by i386 to allow the kernel to address the memory beyond
759 * 900MB. The kernel will set up special mappings (page
760 * table entries on i386) for each page that the kernel needs to
761 * access.
762 */
763 ZONE_HIGHMEM,
764 #endif
765 /*
766 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
767 * movable pages with few exceptional cases described below. Main use
768 * cases for ZONE_MOVABLE are to make memory offlining/unplug more
769 * likely to succeed, and to locally limit unmovable allocations - e.g.,
770 * to increase the number of THP/huge pages. Notable special cases are:
771 *
772 * 1. Pinned pages: (long-term) pinning of movable pages might
773 * essentially turn such pages unmovable. Therefore, we do not allow
774 * pinning long-term pages in ZONE_MOVABLE. When pages are pinned and
775 * faulted, they come from the right zone right away. However, it is
776 * still possible that address space already has pages in
777 * ZONE_MOVABLE at the time when pages are pinned (i.e. user has
778 * touches that memory before pinning). In such case we migrate them
779 * to a different zone. When migration fails - pinning fails.
780 * 2. memblock allocations: kernelcore/movablecore setups might create
781 * situations where ZONE_MOVABLE contains unmovable allocations
782 * after boot. Memory offlining and allocations fail early.
783 * 3. Memory holes: kernelcore/movablecore setups might create very rare
784 * situations where ZONE_MOVABLE contains memory holes after boot,
785 * for example, if we have sections that are only partially
786 * populated. Memory offlining and allocations fail early.
787 * 4. PG_hwpoison pages: while poisoned pages can be skipped during
788 * memory offlining, such pages cannot be allocated.
789 * 5. Unmovable PG_offline pages: in paravirtualized environments,
790 * hotplugged memory blocks might only partially be managed by the
791 * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
792 * parts not manged by the buddy are unmovable PG_offline pages. In
793 * some cases (virtio-mem), such pages can be skipped during
794 * memory offlining, however, cannot be moved/allocated. These
795 * techniques might use alloc_contig_range() to hide previously
796 * exposed pages from the buddy again (e.g., to implement some sort
797 * of memory unplug in virtio-mem).
798 * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create
799 * situations where ZERO_PAGE(0) which is allocated differently
800 * on different platforms may end up in a movable zone. ZERO_PAGE(0)
801 * cannot be migrated.
802 * 7. Memory-hotplug: when using memmap_on_memory and onlining the
803 * memory to the MOVABLE zone, the vmemmap pages are also placed in
804 * such zone. Such pages cannot be really moved around as they are
805 * self-stored in the range, but they are treated as movable when
806 * the range they describe is about to be offlined.
807 *
808 * In general, no unmovable allocations that degrade memory offlining
809 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
810 * have to expect that migrating pages in ZONE_MOVABLE can fail (even
811 * if has_unmovable_pages() states that there are no unmovable pages,
812 * there can be false negatives).
813 */
814 ZONE_MOVABLE,
815 #ifdef CONFIG_ZONE_DEVICE
816 ZONE_DEVICE,
817 #endif
818 __MAX_NR_ZONES
819
820 };
821
822 #ifndef __GENERATING_BOUNDS_H
823
824 #define ASYNC_AND_SYNC 2
825
826 struct zone {
827 /* Read-mostly fields */
828
829 /* zone watermarks, access with *_wmark_pages(zone) macros */
830 unsigned long _watermark[NR_WMARK];
831 unsigned long watermark_boost;
832
833 unsigned long nr_reserved_highatomic;
834
835 /*
836 * We don't know if the memory that we're going to allocate will be
837 * freeable or/and it will be released eventually, so to avoid totally
838 * wasting several GB of ram we must reserve some of the lower zone
839 * memory (otherwise we risk to run OOM on the lower zones despite
840 * there being tons of freeable ram on the higher zones). This array is
841 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
842 * changes.
843 */
844 long lowmem_reserve[MAX_NR_ZONES];
845
846 #ifdef CONFIG_NUMA
847 int node;
848 #endif
849 struct pglist_data *zone_pgdat;
850 struct per_cpu_pages __percpu *per_cpu_pageset;
851 struct per_cpu_zonestat __percpu *per_cpu_zonestats;
852 /*
853 * the high and batch values are copied to individual pagesets for
854 * faster access
855 */
856 int pageset_high_min;
857 int pageset_high_max;
858 int pageset_batch;
859
860 #ifndef CONFIG_SPARSEMEM
861 /*
862 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
863 * In SPARSEMEM, this map is stored in struct mem_section
864 */
865 unsigned long *pageblock_flags;
866 #endif /* CONFIG_SPARSEMEM */
867
868 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
869 unsigned long zone_start_pfn;
870
871 /*
872 * spanned_pages is the total pages spanned by the zone, including
873 * holes, which is calculated as:
874 * spanned_pages = zone_end_pfn - zone_start_pfn;
875 *
876 * present_pages is physical pages existing within the zone, which
877 * is calculated as:
878 * present_pages = spanned_pages - absent_pages(pages in holes);
879 *
880 * present_early_pages is present pages existing within the zone
881 * located on memory available since early boot, excluding hotplugged
882 * memory.
883 *
884 * managed_pages is present pages managed by the buddy system, which
885 * is calculated as (reserved_pages includes pages allocated by the
886 * bootmem allocator):
887 * managed_pages = present_pages - reserved_pages;
888 *
889 * cma pages is present pages that are assigned for CMA use
890 * (MIGRATE_CMA).
891 *
892 * So present_pages may be used by memory hotplug or memory power
893 * management logic to figure out unmanaged pages by checking
894 * (present_pages - managed_pages). And managed_pages should be used
895 * by page allocator and vm scanner to calculate all kinds of watermarks
896 * and thresholds.
897 *
898 * Locking rules:
899 *
900 * zone_start_pfn and spanned_pages are protected by span_seqlock.
901 * It is a seqlock because it has to be read outside of zone->lock,
902 * and it is done in the main allocator path. But, it is written
903 * quite infrequently.
904 *
905 * The span_seq lock is declared along with zone->lock because it is
906 * frequently read in proximity to zone->lock. It's good to
907 * give them a chance of being in the same cacheline.
908 *
909 * Write access to present_pages at runtime should be protected by
910 * mem_hotplug_begin/done(). Any reader who can't tolerant drift of
911 * present_pages should use get_online_mems() to get a stable value.
912 */
913 atomic_long_t managed_pages;
914 unsigned long spanned_pages;
915 unsigned long present_pages;
916 #if defined(CONFIG_MEMORY_HOTPLUG)
917 unsigned long present_early_pages;
918 #endif
919 #ifdef CONFIG_CMA
920 unsigned long cma_pages;
921 #endif
922
923 const char *name;
924
925 #ifdef CONFIG_MEMORY_ISOLATION
926 /*
927 * Number of isolated pageblock. It is used to solve incorrect
928 * freepage counting problem due to racy retrieving migratetype
929 * of pageblock. Protected by zone->lock.
930 */
931 unsigned long nr_isolate_pageblock;
932 #endif
933
934 #ifdef CONFIG_MEMORY_HOTPLUG
935 /* see spanned/present_pages for more description */
936 seqlock_t span_seqlock;
937 #endif
938
939 int initialized;
940
941 /* Write-intensive fields used from the page allocator */
942 CACHELINE_PADDING(_pad1_);
943
944 /* free areas of different sizes */
945 struct free_area free_area[NR_PAGE_ORDERS];
946
947 #ifdef CONFIG_UNACCEPTED_MEMORY
948 /* Pages to be accepted. All pages on the list are MAX_PAGE_ORDER */
949 struct list_head unaccepted_pages;
950 #endif
951
952 /* zone flags, see below */
953 unsigned long flags;
954
955 /* Primarily protects free_area */
956 spinlock_t lock;
957
958 /* Write-intensive fields used by compaction and vmstats. */
959 CACHELINE_PADDING(_pad2_);
960
961 /*
962 * When free pages are below this point, additional steps are taken
963 * when reading the number of free pages to avoid per-cpu counter
964 * drift allowing watermarks to be breached
965 */
966 unsigned long percpu_drift_mark;
967
968 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
969 /* pfn where compaction free scanner should start */
970 unsigned long compact_cached_free_pfn;
971 /* pfn where compaction migration scanner should start */
972 unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC];
973 unsigned long compact_init_migrate_pfn;
974 unsigned long compact_init_free_pfn;
975 #endif
976
977 #ifdef CONFIG_COMPACTION
978 /*
979 * On compaction failure, 1<<compact_defer_shift compactions
980 * are skipped before trying again. The number attempted since
981 * last failure is tracked with compact_considered.
982 * compact_order_failed is the minimum compaction failed order.
983 */
984 unsigned int compact_considered;
985 unsigned int compact_defer_shift;
986 int compact_order_failed;
987 #endif
988
989 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
990 /* Set to true when the PG_migrate_skip bits should be cleared */
991 bool compact_blockskip_flush;
992 #endif
993
994 bool contiguous;
995
996 CACHELINE_PADDING(_pad3_);
997 /* Zone statistics */
998 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
999 atomic_long_t vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
1000 } ____cacheline_internodealigned_in_smp;
1001
1002 enum pgdat_flags {
1003 PGDAT_DIRTY, /* reclaim scanning has recently found
1004 * many dirty file pages at the tail
1005 * of the LRU.
1006 */
1007 PGDAT_WRITEBACK, /* reclaim scanning has recently found
1008 * many pages under writeback
1009 */
1010 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
1011 };
1012
1013 enum zone_flags {
1014 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
1015 * Cleared when kswapd is woken.
1016 */
1017 ZONE_RECLAIM_ACTIVE, /* kswapd may be scanning the zone. */
1018 ZONE_BELOW_HIGH, /* zone is below high watermark. */
1019 };
1020
zone_managed_pages(struct zone * zone)1021 static inline unsigned long zone_managed_pages(struct zone *zone)
1022 {
1023 return (unsigned long)atomic_long_read(&zone->managed_pages);
1024 }
1025
zone_cma_pages(struct zone * zone)1026 static inline unsigned long zone_cma_pages(struct zone *zone)
1027 {
1028 #ifdef CONFIG_CMA
1029 return zone->cma_pages;
1030 #else
1031 return 0;
1032 #endif
1033 }
1034
zone_end_pfn(const struct zone * zone)1035 static inline unsigned long zone_end_pfn(const struct zone *zone)
1036 {
1037 return zone->zone_start_pfn + zone->spanned_pages;
1038 }
1039
zone_spans_pfn(const struct zone * zone,unsigned long pfn)1040 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
1041 {
1042 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
1043 }
1044
zone_is_initialized(struct zone * zone)1045 static inline bool zone_is_initialized(struct zone *zone)
1046 {
1047 return zone->initialized;
1048 }
1049
zone_is_empty(struct zone * zone)1050 static inline bool zone_is_empty(struct zone *zone)
1051 {
1052 return zone->spanned_pages == 0;
1053 }
1054
1055 #ifndef BUILD_VDSO32_64
1056 /*
1057 * The zone field is never updated after free_area_init_core()
1058 * sets it, so none of the operations on it need to be atomic.
1059 */
1060
1061 /* Page flags: | [SECTION] | [NODE] | ZONE | [LAST_CPUPID] | ... | FLAGS | */
1062 #define SECTIONS_PGOFF ((sizeof(unsigned long)*8) - SECTIONS_WIDTH)
1063 #define NODES_PGOFF (SECTIONS_PGOFF - NODES_WIDTH)
1064 #define ZONES_PGOFF (NODES_PGOFF - ZONES_WIDTH)
1065 #define LAST_CPUPID_PGOFF (ZONES_PGOFF - LAST_CPUPID_WIDTH)
1066 #define KASAN_TAG_PGOFF (LAST_CPUPID_PGOFF - KASAN_TAG_WIDTH)
1067 #define LRU_GEN_PGOFF (KASAN_TAG_PGOFF - LRU_GEN_WIDTH)
1068 #define LRU_REFS_PGOFF (LRU_GEN_PGOFF - LRU_REFS_WIDTH)
1069
1070 /*
1071 * Define the bit shifts to access each section. For non-existent
1072 * sections we define the shift as 0; that plus a 0 mask ensures
1073 * the compiler will optimise away reference to them.
1074 */
1075 #define SECTIONS_PGSHIFT (SECTIONS_PGOFF * (SECTIONS_WIDTH != 0))
1076 #define NODES_PGSHIFT (NODES_PGOFF * (NODES_WIDTH != 0))
1077 #define ZONES_PGSHIFT (ZONES_PGOFF * (ZONES_WIDTH != 0))
1078 #define LAST_CPUPID_PGSHIFT (LAST_CPUPID_PGOFF * (LAST_CPUPID_WIDTH != 0))
1079 #define KASAN_TAG_PGSHIFT (KASAN_TAG_PGOFF * (KASAN_TAG_WIDTH != 0))
1080
1081 /* NODE:ZONE or SECTION:ZONE is used to ID a zone for the buddy allocator */
1082 #ifdef NODE_NOT_IN_PAGE_FLAGS
1083 #define ZONEID_SHIFT (SECTIONS_SHIFT + ZONES_SHIFT)
1084 #define ZONEID_PGOFF ((SECTIONS_PGOFF < ZONES_PGOFF) ? \
1085 SECTIONS_PGOFF : ZONES_PGOFF)
1086 #else
1087 #define ZONEID_SHIFT (NODES_SHIFT + ZONES_SHIFT)
1088 #define ZONEID_PGOFF ((NODES_PGOFF < ZONES_PGOFF) ? \
1089 NODES_PGOFF : ZONES_PGOFF)
1090 #endif
1091
1092 #define ZONEID_PGSHIFT (ZONEID_PGOFF * (ZONEID_SHIFT != 0))
1093
1094 #define ZONES_MASK ((1UL << ZONES_WIDTH) - 1)
1095 #define NODES_MASK ((1UL << NODES_WIDTH) - 1)
1096 #define SECTIONS_MASK ((1UL << SECTIONS_WIDTH) - 1)
1097 #define LAST_CPUPID_MASK ((1UL << LAST_CPUPID_SHIFT) - 1)
1098 #define KASAN_TAG_MASK ((1UL << KASAN_TAG_WIDTH) - 1)
1099 #define ZONEID_MASK ((1UL << ZONEID_SHIFT) - 1)
1100
page_zonenum(const struct page * page)1101 static inline enum zone_type page_zonenum(const struct page *page)
1102 {
1103 ASSERT_EXCLUSIVE_BITS(page->flags, ZONES_MASK << ZONES_PGSHIFT);
1104 return (page->flags >> ZONES_PGSHIFT) & ZONES_MASK;
1105 }
1106
folio_zonenum(const struct folio * folio)1107 static inline enum zone_type folio_zonenum(const struct folio *folio)
1108 {
1109 return page_zonenum(&folio->page);
1110 }
1111
1112 #ifdef CONFIG_ZONE_DEVICE
is_zone_device_page(const struct page * page)1113 static inline bool is_zone_device_page(const struct page *page)
1114 {
1115 return page_zonenum(page) == ZONE_DEVICE;
1116 }
1117
1118 /*
1119 * Consecutive zone device pages should not be merged into the same sgl
1120 * or bvec segment with other types of pages or if they belong to different
1121 * pgmaps. Otherwise getting the pgmap of a given segment is not possible
1122 * without scanning the entire segment. This helper returns true either if
1123 * both pages are not zone device pages or both pages are zone device pages
1124 * with the same pgmap.
1125 */
zone_device_pages_have_same_pgmap(const struct page * a,const struct page * b)1126 static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1127 const struct page *b)
1128 {
1129 if (is_zone_device_page(a) != is_zone_device_page(b))
1130 return false;
1131 if (!is_zone_device_page(a))
1132 return true;
1133 return a->pgmap == b->pgmap;
1134 }
1135
1136 extern void memmap_init_zone_device(struct zone *, unsigned long,
1137 unsigned long, struct dev_pagemap *);
1138 #else
is_zone_device_page(const struct page * page)1139 static inline bool is_zone_device_page(const struct page *page)
1140 {
1141 return false;
1142 }
zone_device_pages_have_same_pgmap(const struct page * a,const struct page * b)1143 static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1144 const struct page *b)
1145 {
1146 return true;
1147 }
1148 #endif
1149
folio_is_zone_device(const struct folio * folio)1150 static inline bool folio_is_zone_device(const struct folio *folio)
1151 {
1152 return is_zone_device_page(&folio->page);
1153 }
1154
is_zone_movable_page(const struct page * page)1155 static inline bool is_zone_movable_page(const struct page *page)
1156 {
1157 return page_zonenum(page) == ZONE_MOVABLE;
1158 }
1159
folio_is_zone_movable(const struct folio * folio)1160 static inline bool folio_is_zone_movable(const struct folio *folio)
1161 {
1162 return folio_zonenum(folio) == ZONE_MOVABLE;
1163 }
1164 #endif
1165
1166 /*
1167 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
1168 * intersection with the given zone
1169 */
zone_intersects(struct zone * zone,unsigned long start_pfn,unsigned long nr_pages)1170 static inline bool zone_intersects(struct zone *zone,
1171 unsigned long start_pfn, unsigned long nr_pages)
1172 {
1173 if (zone_is_empty(zone))
1174 return false;
1175 if (start_pfn >= zone_end_pfn(zone) ||
1176 start_pfn + nr_pages <= zone->zone_start_pfn)
1177 return false;
1178
1179 return true;
1180 }
1181
1182 /*
1183 * The "priority" of VM scanning is how much of the queues we will scan in one
1184 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
1185 * queues ("queue_length >> 12") during an aging round.
1186 */
1187 #define DEF_PRIORITY 12
1188
1189 /* Maximum number of zones on a zonelist */
1190 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
1191
1192 enum {
1193 ZONELIST_FALLBACK, /* zonelist with fallback */
1194 #ifdef CONFIG_NUMA
1195 /*
1196 * The NUMA zonelists are doubled because we need zonelists that
1197 * restrict the allocations to a single node for __GFP_THISNODE.
1198 */
1199 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
1200 #endif
1201 MAX_ZONELISTS
1202 };
1203
1204 /*
1205 * This struct contains information about a zone in a zonelist. It is stored
1206 * here to avoid dereferences into large structures and lookups of tables
1207 */
1208 struct zoneref {
1209 struct zone *zone; /* Pointer to actual zone */
1210 int zone_idx; /* zone_idx(zoneref->zone) */
1211 };
1212
1213 /*
1214 * One allocation request operates on a zonelist. A zonelist
1215 * is a list of zones, the first one is the 'goal' of the
1216 * allocation, the other zones are fallback zones, in decreasing
1217 * priority.
1218 *
1219 * To speed the reading of the zonelist, the zonerefs contain the zone index
1220 * of the entry being read. Helper functions to access information given
1221 * a struct zoneref are
1222 *
1223 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
1224 * zonelist_zone_idx() - Return the index of the zone for an entry
1225 * zonelist_node_idx() - Return the index of the node for an entry
1226 */
1227 struct zonelist {
1228 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
1229 };
1230
1231 /*
1232 * The array of struct pages for flatmem.
1233 * It must be declared for SPARSEMEM as well because there are configurations
1234 * that rely on that.
1235 */
1236 extern struct page *mem_map;
1237
1238 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1239 struct deferred_split {
1240 spinlock_t split_queue_lock;
1241 struct list_head split_queue;
1242 unsigned long split_queue_len;
1243 };
1244 #endif
1245
1246 #ifdef CONFIG_MEMORY_FAILURE
1247 /*
1248 * Per NUMA node memory failure handling statistics.
1249 */
1250 struct memory_failure_stats {
1251 /*
1252 * Number of raw pages poisoned.
1253 * Cases not accounted: memory outside kernel control, offline page,
1254 * arch-specific memory_failure (SGX), hwpoison_filter() filtered
1255 * error events, and unpoison actions from hwpoison_unpoison.
1256 */
1257 unsigned long total;
1258 /*
1259 * Recovery results of poisoned raw pages handled by memory_failure,
1260 * in sync with mf_result.
1261 * total = ignored + failed + delayed + recovered.
1262 * total * PAGE_SIZE * #nodes = /proc/meminfo/HardwareCorrupted.
1263 */
1264 unsigned long ignored;
1265 unsigned long failed;
1266 unsigned long delayed;
1267 unsigned long recovered;
1268 };
1269 #endif
1270
1271 /*
1272 * On NUMA machines, each NUMA node would have a pg_data_t to describe
1273 * it's memory layout. On UMA machines there is a single pglist_data which
1274 * describes the whole memory.
1275 *
1276 * Memory statistics and page replacement data structures are maintained on a
1277 * per-zone basis.
1278 */
1279 typedef struct pglist_data {
1280 /*
1281 * node_zones contains just the zones for THIS node. Not all of the
1282 * zones may be populated, but it is the full list. It is referenced by
1283 * this node's node_zonelists as well as other node's node_zonelists.
1284 */
1285 struct zone node_zones[MAX_NR_ZONES];
1286
1287 /*
1288 * node_zonelists contains references to all zones in all nodes.
1289 * Generally the first zones will be references to this node's
1290 * node_zones.
1291 */
1292 struct zonelist node_zonelists[MAX_ZONELISTS];
1293
1294 int nr_zones; /* number of populated zones in this node */
1295 #ifdef CONFIG_FLATMEM /* means !SPARSEMEM */
1296 struct page *node_mem_map;
1297 #ifdef CONFIG_PAGE_EXTENSION
1298 struct page_ext *node_page_ext;
1299 #endif
1300 #endif
1301 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
1302 /*
1303 * Must be held any time you expect node_start_pfn,
1304 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
1305 * Also synchronizes pgdat->first_deferred_pfn during deferred page
1306 * init.
1307 *
1308 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
1309 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
1310 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
1311 *
1312 * Nests above zone->lock and zone->span_seqlock
1313 */
1314 spinlock_t node_size_lock;
1315 #endif
1316 unsigned long node_start_pfn;
1317 unsigned long node_present_pages; /* total number of physical pages */
1318 unsigned long node_spanned_pages; /* total size of physical page
1319 range, including holes */
1320 int node_id;
1321 wait_queue_head_t kswapd_wait;
1322 wait_queue_head_t pfmemalloc_wait;
1323
1324 /* workqueues for throttling reclaim for different reasons. */
1325 wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE];
1326
1327 atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */
1328 unsigned long nr_reclaim_start; /* nr pages written while throttled
1329 * when throttling started. */
1330 #ifdef CONFIG_MEMORY_HOTPLUG
1331 struct mutex kswapd_lock;
1332 #endif
1333 struct task_struct *kswapd; /* Protected by kswapd_lock */
1334 int kswapd_order;
1335 enum zone_type kswapd_highest_zoneidx;
1336
1337 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
1338
1339 #ifdef CONFIG_COMPACTION
1340 int kcompactd_max_order;
1341 enum zone_type kcompactd_highest_zoneidx;
1342 wait_queue_head_t kcompactd_wait;
1343 struct task_struct *kcompactd;
1344 bool proactive_compact_trigger;
1345 #endif
1346 /*
1347 * This is a per-node reserve of pages that are not available
1348 * to userspace allocations.
1349 */
1350 unsigned long totalreserve_pages;
1351
1352 #ifdef CONFIG_NUMA
1353 /*
1354 * node reclaim becomes active if more unmapped pages exist.
1355 */
1356 unsigned long min_unmapped_pages;
1357 unsigned long min_slab_pages;
1358 #endif /* CONFIG_NUMA */
1359
1360 /* Write-intensive fields used by page reclaim */
1361 CACHELINE_PADDING(_pad1_);
1362
1363 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1364 /*
1365 * If memory initialisation on large machines is deferred then this
1366 * is the first PFN that needs to be initialised.
1367 */
1368 unsigned long first_deferred_pfn;
1369 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1370
1371 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1372 struct deferred_split deferred_split_queue;
1373 #endif
1374
1375 #ifdef CONFIG_NUMA_BALANCING
1376 /* start time in ms of current promote rate limit period */
1377 unsigned int nbp_rl_start;
1378 /* number of promote candidate pages at start time of current rate limit period */
1379 unsigned long nbp_rl_nr_cand;
1380 /* promote threshold in ms */
1381 unsigned int nbp_threshold;
1382 /* start time in ms of current promote threshold adjustment period */
1383 unsigned int nbp_th_start;
1384 /*
1385 * number of promote candidate pages at start time of current promote
1386 * threshold adjustment period
1387 */
1388 unsigned long nbp_th_nr_cand;
1389 #endif
1390 /* Fields commonly accessed by the page reclaim scanner */
1391
1392 /*
1393 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
1394 *
1395 * Use mem_cgroup_lruvec() to look up lruvecs.
1396 */
1397 struct lruvec __lruvec;
1398
1399 unsigned long flags;
1400
1401 #ifdef CONFIG_LRU_GEN
1402 /* kswap mm walk data */
1403 struct lru_gen_mm_walk mm_walk;
1404 /* lru_gen_folio list */
1405 struct lru_gen_memcg memcg_lru;
1406 #endif
1407
1408 CACHELINE_PADDING(_pad2_);
1409
1410 /* Per-node vmstats */
1411 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
1412 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
1413 #ifdef CONFIG_NUMA
1414 struct memory_tier __rcu *memtier;
1415 #endif
1416 #ifdef CONFIG_MEMORY_FAILURE
1417 struct memory_failure_stats mf_stats;
1418 #endif
1419 } pg_data_t;
1420
1421 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
1422 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
1423
1424 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
1425 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
1426
pgdat_end_pfn(pg_data_t * pgdat)1427 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
1428 {
1429 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
1430 }
1431
1432 #include <linux/memory_hotplug.h>
1433
1434 void build_all_zonelists(pg_data_t *pgdat);
1435 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
1436 enum zone_type highest_zoneidx);
1437 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
1438 int highest_zoneidx, unsigned int alloc_flags,
1439 long free_pages);
1440 bool zone_watermark_ok(struct zone *z, unsigned int order,
1441 unsigned long mark, int highest_zoneidx,
1442 unsigned int alloc_flags);
1443 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
1444 unsigned long mark, int highest_zoneidx);
1445 /*
1446 * Memory initialization context, use to differentiate memory added by
1447 * the platform statically or via memory hotplug interface.
1448 */
1449 enum meminit_context {
1450 MEMINIT_EARLY,
1451 MEMINIT_HOTPLUG,
1452 };
1453
1454 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
1455 unsigned long size);
1456
1457 extern void lruvec_init(struct lruvec *lruvec);
1458
lruvec_pgdat(struct lruvec * lruvec)1459 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
1460 {
1461 #ifdef CONFIG_MEMCG
1462 return lruvec->pgdat;
1463 #else
1464 return container_of(lruvec, struct pglist_data, __lruvec);
1465 #endif
1466 }
1467
1468 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
1469 int local_memory_node(int node_id);
1470 #else
local_memory_node(int node_id)1471 static inline int local_memory_node(int node_id) { return node_id; };
1472 #endif
1473
1474 /*
1475 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
1476 */
1477 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
1478
1479 #ifdef CONFIG_ZONE_DEVICE
zone_is_zone_device(struct zone * zone)1480 static inline bool zone_is_zone_device(struct zone *zone)
1481 {
1482 return zone_idx(zone) == ZONE_DEVICE;
1483 }
1484 #else
zone_is_zone_device(struct zone * zone)1485 static inline bool zone_is_zone_device(struct zone *zone)
1486 {
1487 return false;
1488 }
1489 #endif
1490
1491 /*
1492 * Returns true if a zone has pages managed by the buddy allocator.
1493 * All the reclaim decisions have to use this function rather than
1494 * populated_zone(). If the whole zone is reserved then we can easily
1495 * end up with populated_zone() && !managed_zone().
1496 */
managed_zone(struct zone * zone)1497 static inline bool managed_zone(struct zone *zone)
1498 {
1499 return zone_managed_pages(zone);
1500 }
1501
1502 /* Returns true if a zone has memory */
populated_zone(struct zone * zone)1503 static inline bool populated_zone(struct zone *zone)
1504 {
1505 return zone->present_pages;
1506 }
1507
1508 #ifdef CONFIG_NUMA
zone_to_nid(struct zone * zone)1509 static inline int zone_to_nid(struct zone *zone)
1510 {
1511 return zone->node;
1512 }
1513
zone_set_nid(struct zone * zone,int nid)1514 static inline void zone_set_nid(struct zone *zone, int nid)
1515 {
1516 zone->node = nid;
1517 }
1518 #else
zone_to_nid(struct zone * zone)1519 static inline int zone_to_nid(struct zone *zone)
1520 {
1521 return 0;
1522 }
1523
zone_set_nid(struct zone * zone,int nid)1524 static inline void zone_set_nid(struct zone *zone, int nid) {}
1525 #endif
1526
1527 extern int movable_zone;
1528
is_highmem_idx(enum zone_type idx)1529 static inline int is_highmem_idx(enum zone_type idx)
1530 {
1531 #ifdef CONFIG_HIGHMEM
1532 return (idx == ZONE_HIGHMEM ||
1533 (idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM));
1534 #else
1535 return 0;
1536 #endif
1537 }
1538
1539 /**
1540 * is_highmem - helper function to quickly check if a struct zone is a
1541 * highmem zone or not. This is an attempt to keep references
1542 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
1543 * @zone: pointer to struct zone variable
1544 * Return: 1 for a highmem zone, 0 otherwise
1545 */
is_highmem(struct zone * zone)1546 static inline int is_highmem(struct zone *zone)
1547 {
1548 return is_highmem_idx(zone_idx(zone));
1549 }
1550
1551 #ifdef CONFIG_ZONE_DMA
1552 bool has_managed_dma(void);
1553 #else
has_managed_dma(void)1554 static inline bool has_managed_dma(void)
1555 {
1556 return false;
1557 }
1558 #endif
1559
1560
1561 #ifndef CONFIG_NUMA
1562
1563 extern struct pglist_data contig_page_data;
NODE_DATA(int nid)1564 static inline struct pglist_data *NODE_DATA(int nid)
1565 {
1566 return &contig_page_data;
1567 }
1568
1569 #else /* CONFIG_NUMA */
1570
1571 #include <asm/mmzone.h>
1572
1573 #endif /* !CONFIG_NUMA */
1574
1575 extern struct pglist_data *first_online_pgdat(void);
1576 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
1577 extern struct zone *next_zone(struct zone *zone);
1578
1579 /**
1580 * for_each_online_pgdat - helper macro to iterate over all online nodes
1581 * @pgdat: pointer to a pg_data_t variable
1582 */
1583 #define for_each_online_pgdat(pgdat) \
1584 for (pgdat = first_online_pgdat(); \
1585 pgdat; \
1586 pgdat = next_online_pgdat(pgdat))
1587 /**
1588 * for_each_zone - helper macro to iterate over all memory zones
1589 * @zone: pointer to struct zone variable
1590 *
1591 * The user only needs to declare the zone variable, for_each_zone
1592 * fills it in.
1593 */
1594 #define for_each_zone(zone) \
1595 for (zone = (first_online_pgdat())->node_zones; \
1596 zone; \
1597 zone = next_zone(zone))
1598
1599 #define for_each_populated_zone(zone) \
1600 for (zone = (first_online_pgdat())->node_zones; \
1601 zone; \
1602 zone = next_zone(zone)) \
1603 if (!populated_zone(zone)) \
1604 ; /* do nothing */ \
1605 else
1606
zonelist_zone(struct zoneref * zoneref)1607 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1608 {
1609 return zoneref->zone;
1610 }
1611
zonelist_zone_idx(struct zoneref * zoneref)1612 static inline int zonelist_zone_idx(struct zoneref *zoneref)
1613 {
1614 return zoneref->zone_idx;
1615 }
1616
zonelist_node_idx(struct zoneref * zoneref)1617 static inline int zonelist_node_idx(struct zoneref *zoneref)
1618 {
1619 return zone_to_nid(zoneref->zone);
1620 }
1621
1622 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1623 enum zone_type highest_zoneidx,
1624 nodemask_t *nodes);
1625
1626 /**
1627 * 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
1628 * @z: The cursor used as a starting point for the search
1629 * @highest_zoneidx: The zone index of the highest zone to return
1630 * @nodes: An optional nodemask to filter the zonelist with
1631 *
1632 * This function returns the next zone at or below a given zone index that is
1633 * within the allowed nodemask using a cursor as the starting point for the
1634 * search. The zoneref returned is a cursor that represents the current zone
1635 * being examined. It should be advanced by one before calling
1636 * next_zones_zonelist again.
1637 *
1638 * Return: the next zone at or below highest_zoneidx within the allowed
1639 * nodemask using a cursor within a zonelist as a starting point
1640 */
next_zones_zonelist(struct zoneref * z,enum zone_type highest_zoneidx,nodemask_t * nodes)1641 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1642 enum zone_type highest_zoneidx,
1643 nodemask_t *nodes)
1644 {
1645 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1646 return z;
1647 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1648 }
1649
1650 /**
1651 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1652 * @zonelist: The zonelist to search for a suitable zone
1653 * @highest_zoneidx: The zone index of the highest zone to return
1654 * @nodes: An optional nodemask to filter the zonelist with
1655 *
1656 * This function returns the first zone at or below a given zone index that is
1657 * within the allowed nodemask. The zoneref returned is a cursor that can be
1658 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1659 * one before calling.
1660 *
1661 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1662 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1663 * update due to cpuset modification.
1664 *
1665 * Return: Zoneref pointer for the first suitable zone found
1666 */
first_zones_zonelist(struct zonelist * zonelist,enum zone_type highest_zoneidx,nodemask_t * nodes)1667 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1668 enum zone_type highest_zoneidx,
1669 nodemask_t *nodes)
1670 {
1671 return next_zones_zonelist(zonelist->_zonerefs,
1672 highest_zoneidx, nodes);
1673 }
1674
1675 /**
1676 * 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
1677 * @zone: The current zone in the iterator
1678 * @z: The current pointer within zonelist->_zonerefs being iterated
1679 * @zlist: The zonelist being iterated
1680 * @highidx: The zone index of the highest zone to return
1681 * @nodemask: Nodemask allowed by the allocator
1682 *
1683 * This iterator iterates though all zones at or below a given zone index and
1684 * within a given nodemask
1685 */
1686 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1687 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1688 zone; \
1689 z = next_zones_zonelist(++z, highidx, nodemask), \
1690 zone = zonelist_zone(z))
1691
1692 #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
1693 for (zone = z->zone; \
1694 zone; \
1695 z = next_zones_zonelist(++z, highidx, nodemask), \
1696 zone = zonelist_zone(z))
1697
1698
1699 /**
1700 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1701 * @zone: The current zone in the iterator
1702 * @z: The current pointer within zonelist->zones being iterated
1703 * @zlist: The zonelist being iterated
1704 * @highidx: The zone index of the highest zone to return
1705 *
1706 * This iterator iterates though all zones at or below a given zone index.
1707 */
1708 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1709 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1710
1711 /* Whether the 'nodes' are all movable nodes */
movable_only_nodes(nodemask_t * nodes)1712 static inline bool movable_only_nodes(nodemask_t *nodes)
1713 {
1714 struct zonelist *zonelist;
1715 struct zoneref *z;
1716 int nid;
1717
1718 if (nodes_empty(*nodes))
1719 return false;
1720
1721 /*
1722 * We can chose arbitrary node from the nodemask to get a
1723 * zonelist as they are interlinked. We just need to find
1724 * at least one zone that can satisfy kernel allocations.
1725 */
1726 nid = first_node(*nodes);
1727 zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK];
1728 z = first_zones_zonelist(zonelist, ZONE_NORMAL, nodes);
1729 return (!z->zone) ? true : false;
1730 }
1731
1732
1733 #ifdef CONFIG_SPARSEMEM
1734 #include <asm/sparsemem.h>
1735 #endif
1736
1737 #ifdef CONFIG_FLATMEM
1738 #define pfn_to_nid(pfn) (0)
1739 #endif
1740
1741 #ifdef CONFIG_SPARSEMEM
1742
1743 /*
1744 * PA_SECTION_SHIFT physical address to/from section number
1745 * PFN_SECTION_SHIFT pfn to/from section number
1746 */
1747 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1748 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1749
1750 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1751
1752 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1753 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1754
1755 #define SECTION_BLOCKFLAGS_BITS \
1756 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1757
1758 #if (MAX_PAGE_ORDER + PAGE_SHIFT) > SECTION_SIZE_BITS
1759 #error Allocator MAX_PAGE_ORDER exceeds SECTION_SIZE
1760 #endif
1761
pfn_to_section_nr(unsigned long pfn)1762 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1763 {
1764 return pfn >> PFN_SECTION_SHIFT;
1765 }
section_nr_to_pfn(unsigned long sec)1766 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1767 {
1768 return sec << PFN_SECTION_SHIFT;
1769 }
1770
1771 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1772 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1773
1774 #define SUBSECTION_SHIFT 21
1775 #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1776
1777 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1778 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1779 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1780
1781 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1782 #error Subsection size exceeds section size
1783 #else
1784 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1785 #endif
1786
1787 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1788 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1789
1790 struct mem_section_usage {
1791 struct rcu_head rcu;
1792 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1793 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1794 #endif
1795 /* See declaration of similar field in struct zone */
1796 unsigned long pageblock_flags[0];
1797 };
1798
1799 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1800
1801 struct page;
1802 struct page_ext;
1803 struct mem_section {
1804 /*
1805 * This is, logically, a pointer to an array of struct
1806 * pages. However, it is stored with some other magic.
1807 * (see sparse.c::sparse_init_one_section())
1808 *
1809 * Additionally during early boot we encode node id of
1810 * the location of the section here to guide allocation.
1811 * (see sparse.c::memory_present())
1812 *
1813 * Making it a UL at least makes someone do a cast
1814 * before using it wrong.
1815 */
1816 unsigned long section_mem_map;
1817
1818 struct mem_section_usage *usage;
1819 #ifdef CONFIG_PAGE_EXTENSION
1820 /*
1821 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1822 * section. (see page_ext.h about this.)
1823 */
1824 struct page_ext *page_ext;
1825 unsigned long pad;
1826 #endif
1827 /*
1828 * WARNING: mem_section must be a power-of-2 in size for the
1829 * calculation and use of SECTION_ROOT_MASK to make sense.
1830 */
1831 };
1832
1833 #ifdef CONFIG_SPARSEMEM_EXTREME
1834 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1835 #else
1836 #define SECTIONS_PER_ROOT 1
1837 #endif
1838
1839 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1840 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1841 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1842
1843 #ifdef CONFIG_SPARSEMEM_EXTREME
1844 extern struct mem_section **mem_section;
1845 #else
1846 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1847 #endif
1848
section_to_usemap(struct mem_section * ms)1849 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1850 {
1851 return ms->usage->pageblock_flags;
1852 }
1853
__nr_to_section(unsigned long nr)1854 static inline struct mem_section *__nr_to_section(unsigned long nr)
1855 {
1856 unsigned long root = SECTION_NR_TO_ROOT(nr);
1857
1858 if (unlikely(root >= NR_SECTION_ROOTS))
1859 return NULL;
1860
1861 #ifdef CONFIG_SPARSEMEM_EXTREME
1862 if (!mem_section || !mem_section[root])
1863 return NULL;
1864 #endif
1865 return &mem_section[root][nr & SECTION_ROOT_MASK];
1866 }
1867 extern size_t mem_section_usage_size(void);
1868
1869 /*
1870 * We use the lower bits of the mem_map pointer to store
1871 * a little bit of information. The pointer is calculated
1872 * as mem_map - section_nr_to_pfn(pnum). The result is
1873 * aligned to the minimum alignment of the two values:
1874 * 1. All mem_map arrays are page-aligned.
1875 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1876 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1877 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1878 * worst combination is powerpc with 256k pages,
1879 * which results in PFN_SECTION_SHIFT equal 6.
1880 * To sum it up, at least 6 bits are available on all architectures.
1881 * However, we can exceed 6 bits on some other architectures except
1882 * powerpc (e.g. 15 bits are available on x86_64, 13 bits are available
1883 * with the worst case of 64K pages on arm64) if we make sure the
1884 * exceeded bit is not applicable to powerpc.
1885 */
1886 enum {
1887 SECTION_MARKED_PRESENT_BIT,
1888 SECTION_HAS_MEM_MAP_BIT,
1889 SECTION_IS_ONLINE_BIT,
1890 SECTION_IS_EARLY_BIT,
1891 #ifdef CONFIG_ZONE_DEVICE
1892 SECTION_TAINT_ZONE_DEVICE_BIT,
1893 #endif
1894 SECTION_MAP_LAST_BIT,
1895 };
1896
1897 #define SECTION_MARKED_PRESENT BIT(SECTION_MARKED_PRESENT_BIT)
1898 #define SECTION_HAS_MEM_MAP BIT(SECTION_HAS_MEM_MAP_BIT)
1899 #define SECTION_IS_ONLINE BIT(SECTION_IS_ONLINE_BIT)
1900 #define SECTION_IS_EARLY BIT(SECTION_IS_EARLY_BIT)
1901 #ifdef CONFIG_ZONE_DEVICE
1902 #define SECTION_TAINT_ZONE_DEVICE BIT(SECTION_TAINT_ZONE_DEVICE_BIT)
1903 #endif
1904 #define SECTION_MAP_MASK (~(BIT(SECTION_MAP_LAST_BIT) - 1))
1905 #define SECTION_NID_SHIFT SECTION_MAP_LAST_BIT
1906
__section_mem_map_addr(struct mem_section * section)1907 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1908 {
1909 unsigned long map = section->section_mem_map;
1910 map &= SECTION_MAP_MASK;
1911 return (struct page *)map;
1912 }
1913
present_section(struct mem_section * section)1914 static inline int present_section(struct mem_section *section)
1915 {
1916 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1917 }
1918
present_section_nr(unsigned long nr)1919 static inline int present_section_nr(unsigned long nr)
1920 {
1921 return present_section(__nr_to_section(nr));
1922 }
1923
valid_section(struct mem_section * section)1924 static inline int valid_section(struct mem_section *section)
1925 {
1926 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1927 }
1928
early_section(struct mem_section * section)1929 static inline int early_section(struct mem_section *section)
1930 {
1931 return (section && (section->section_mem_map & SECTION_IS_EARLY));
1932 }
1933
valid_section_nr(unsigned long nr)1934 static inline int valid_section_nr(unsigned long nr)
1935 {
1936 return valid_section(__nr_to_section(nr));
1937 }
1938
online_section(struct mem_section * section)1939 static inline int online_section(struct mem_section *section)
1940 {
1941 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1942 }
1943
1944 #ifdef CONFIG_ZONE_DEVICE
online_device_section(struct mem_section * section)1945 static inline int online_device_section(struct mem_section *section)
1946 {
1947 unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE;
1948
1949 return section && ((section->section_mem_map & flags) == flags);
1950 }
1951 #else
online_device_section(struct mem_section * section)1952 static inline int online_device_section(struct mem_section *section)
1953 {
1954 return 0;
1955 }
1956 #endif
1957
online_section_nr(unsigned long nr)1958 static inline int online_section_nr(unsigned long nr)
1959 {
1960 return online_section(__nr_to_section(nr));
1961 }
1962
1963 #ifdef CONFIG_MEMORY_HOTPLUG
1964 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1965 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1966 #endif
1967
__pfn_to_section(unsigned long pfn)1968 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1969 {
1970 return __nr_to_section(pfn_to_section_nr(pfn));
1971 }
1972
1973 extern unsigned long __highest_present_section_nr;
1974
subsection_map_index(unsigned long pfn)1975 static inline int subsection_map_index(unsigned long pfn)
1976 {
1977 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1978 }
1979
1980 #ifdef CONFIG_SPARSEMEM_VMEMMAP
pfn_section_valid(struct mem_section * ms,unsigned long pfn)1981 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1982 {
1983 int idx = subsection_map_index(pfn);
1984 struct mem_section_usage *usage = READ_ONCE(ms->usage);
1985
1986 return usage ? test_bit(idx, usage->subsection_map) : 0;
1987 }
1988 #else
pfn_section_valid(struct mem_section * ms,unsigned long pfn)1989 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1990 {
1991 return 1;
1992 }
1993 #endif
1994
1995 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1996 /**
1997 * pfn_valid - check if there is a valid memory map entry for a PFN
1998 * @pfn: the page frame number to check
1999 *
2000 * Check if there is a valid memory map entry aka struct page for the @pfn.
2001 * Note, that availability of the memory map entry does not imply that
2002 * there is actual usable memory at that @pfn. The struct page may
2003 * represent a hole or an unusable page frame.
2004 *
2005 * Return: 1 for PFNs that have memory map entries and 0 otherwise
2006 */
pfn_valid(unsigned long pfn)2007 static inline int pfn_valid(unsigned long pfn)
2008 {
2009 struct mem_section *ms;
2010 int ret;
2011
2012 /*
2013 * Ensure the upper PAGE_SHIFT bits are clear in the
2014 * pfn. Else it might lead to false positives when
2015 * some of the upper bits are set, but the lower bits
2016 * match a valid pfn.
2017 */
2018 if (PHYS_PFN(PFN_PHYS(pfn)) != pfn)
2019 return 0;
2020
2021 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2022 return 0;
2023 ms = __pfn_to_section(pfn);
2024 rcu_read_lock_sched();
2025 if (!valid_section(ms)) {
2026 rcu_read_unlock_sched();
2027 return 0;
2028 }
2029 /*
2030 * Traditionally early sections always returned pfn_valid() for
2031 * the entire section-sized span.
2032 */
2033 ret = early_section(ms) || pfn_section_valid(ms, pfn);
2034 rcu_read_unlock_sched();
2035
2036 return ret;
2037 }
2038 #endif
2039
pfn_in_present_section(unsigned long pfn)2040 static inline int pfn_in_present_section(unsigned long pfn)
2041 {
2042 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2043 return 0;
2044 return present_section(__pfn_to_section(pfn));
2045 }
2046
next_present_section_nr(unsigned long section_nr)2047 static inline unsigned long next_present_section_nr(unsigned long section_nr)
2048 {
2049 while (++section_nr <= __highest_present_section_nr) {
2050 if (present_section_nr(section_nr))
2051 return section_nr;
2052 }
2053
2054 return -1;
2055 }
2056
2057 /*
2058 * These are _only_ used during initialisation, therefore they
2059 * can use __initdata ... They could have names to indicate
2060 * this restriction.
2061 */
2062 #ifdef CONFIG_NUMA
2063 #define pfn_to_nid(pfn) \
2064 ({ \
2065 unsigned long __pfn_to_nid_pfn = (pfn); \
2066 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
2067 })
2068 #else
2069 #define pfn_to_nid(pfn) (0)
2070 #endif
2071
2072 void sparse_init(void);
2073 #else
2074 #define sparse_init() do {} while (0)
2075 #define sparse_index_init(_sec, _nid) do {} while (0)
2076 #define pfn_in_present_section pfn_valid
2077 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
2078 #endif /* CONFIG_SPARSEMEM */
2079
2080 #endif /* !__GENERATING_BOUNDS.H */
2081 #endif /* !__ASSEMBLY__ */
2082 #endif /* _LINUX_MMZONE_H */
2083