Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
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/* Defines the order for the number of pages that have a migrate type. */
41#ifndef CONFIG_PAGE_BLOCK_MAX_ORDER
42#define PAGE_BLOCK_MAX_ORDER MAX_PAGE_ORDER
43#else
44#define PAGE_BLOCK_MAX_ORDER CONFIG_PAGE_BLOCK_MAX_ORDER
45#endif /* CONFIG_PAGE_BLOCK_MAX_ORDER */
46
47/*
48 * The MAX_PAGE_ORDER, which defines the max order of pages to be allocated
49 * by the buddy allocator, has to be larger or equal to the PAGE_BLOCK_MAX_ORDER,
50 * which defines the order for the number of pages that can have a migrate type
51 */
52#if (PAGE_BLOCK_MAX_ORDER > MAX_PAGE_ORDER)
53#error MAX_PAGE_ORDER must be >= PAGE_BLOCK_MAX_ORDER
54#endif
55
56/*
57 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
58 * costly to service. That is between allocation orders which should
59 * coalesce naturally under reasonable reclaim pressure and those which
60 * will not.
61 */
62#define PAGE_ALLOC_COSTLY_ORDER 3
63
64enum migratetype {
65 MIGRATE_UNMOVABLE,
66 MIGRATE_MOVABLE,
67 MIGRATE_RECLAIMABLE,
68 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
69 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
70#ifdef CONFIG_CMA
71 /*
72 * MIGRATE_CMA migration type is designed to mimic the way
73 * ZONE_MOVABLE works. Only movable pages can be allocated
74 * from MIGRATE_CMA pageblocks and page allocator never
75 * implicitly change migration type of MIGRATE_CMA pageblock.
76 *
77 * The way to use it is to change migratetype of a range of
78 * pageblocks to MIGRATE_CMA which can be done by
79 * __free_pageblock_cma() function.
80 */
81 MIGRATE_CMA,
82 __MIGRATE_TYPE_END = MIGRATE_CMA,
83#else
84 __MIGRATE_TYPE_END = MIGRATE_HIGHATOMIC,
85#endif
86#ifdef CONFIG_MEMORY_ISOLATION
87 MIGRATE_ISOLATE, /* can't allocate from here */
88#endif
89 MIGRATE_TYPES
90};
91
92/* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
93extern const char * const migratetype_names[MIGRATE_TYPES];
94
95#ifdef CONFIG_CMA
96# define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
97# define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
98/*
99 * __dump_folio() in mm/debug.c passes a folio pointer to on-stack struct folio,
100 * so folio_pfn() cannot be used and pfn is needed.
101 */
102# define is_migrate_cma_folio(folio, pfn) \
103 (get_pfnblock_migratetype(&folio->page, pfn) == MIGRATE_CMA)
104#else
105# define is_migrate_cma(migratetype) false
106# define is_migrate_cma_page(_page) false
107# define is_migrate_cma_folio(folio, pfn) false
108#endif
109
110static inline bool is_migrate_movable(int mt)
111{
112 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
113}
114
115/*
116 * Check whether a migratetype can be merged with another migratetype.
117 *
118 * It is only mergeable when it can fall back to other migratetypes for
119 * allocation. See fallbacks[MIGRATE_TYPES][3] in page_alloc.c.
120 */
121static inline bool migratetype_is_mergeable(int mt)
122{
123 return mt < MIGRATE_PCPTYPES;
124}
125
126#define for_each_migratetype_order(order, type) \
127 for (order = 0; order < NR_PAGE_ORDERS; order++) \
128 for (type = 0; type < MIGRATE_TYPES; type++)
129
130extern int page_group_by_mobility_disabled;
131
132#define get_pageblock_migratetype(page) \
133 get_pfnblock_migratetype(page, page_to_pfn(page))
134
135#define folio_migratetype(folio) \
136 get_pageblock_migratetype(&folio->page)
137
138struct free_area {
139 struct list_head free_list[MIGRATE_TYPES];
140 unsigned long nr_free;
141};
142
143struct pglist_data;
144
145#ifdef CONFIG_NUMA
146enum numa_stat_item {
147 NUMA_HIT, /* allocated in intended node */
148 NUMA_MISS, /* allocated in non intended node */
149 NUMA_FOREIGN, /* was intended here, hit elsewhere */
150 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
151 NUMA_LOCAL, /* allocation from local node */
152 NUMA_OTHER, /* allocation from other node */
153 NR_VM_NUMA_EVENT_ITEMS
154};
155#else
156#define NR_VM_NUMA_EVENT_ITEMS 0
157#endif
158
159enum zone_stat_item {
160 /* First 128 byte cacheline (assuming 64 bit words) */
161 NR_FREE_PAGES,
162 NR_FREE_PAGES_BLOCKS,
163 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
164 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
165 NR_ZONE_ACTIVE_ANON,
166 NR_ZONE_INACTIVE_FILE,
167 NR_ZONE_ACTIVE_FILE,
168 NR_ZONE_UNEVICTABLE,
169 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
170 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
171 /* Second 128 byte cacheline */
172#if IS_ENABLED(CONFIG_ZSMALLOC)
173 NR_ZSPAGES, /* allocated in zsmalloc */
174#endif
175 NR_FREE_CMA_PAGES,
176#ifdef CONFIG_UNACCEPTED_MEMORY
177 NR_UNACCEPTED,
178#endif
179 NR_VM_ZONE_STAT_ITEMS };
180
181enum node_stat_item {
182 NR_LRU_BASE,
183 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
184 NR_ACTIVE_ANON, /* " " " " " */
185 NR_INACTIVE_FILE, /* " " " " " */
186 NR_ACTIVE_FILE, /* " " " " " */
187 NR_UNEVICTABLE, /* " " " " " */
188 NR_SLAB_RECLAIMABLE_B,
189 NR_SLAB_UNRECLAIMABLE_B,
190 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
191 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
192 WORKINGSET_NODES,
193 WORKINGSET_REFAULT_BASE,
194 WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE,
195 WORKINGSET_REFAULT_FILE,
196 WORKINGSET_ACTIVATE_BASE,
197 WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE,
198 WORKINGSET_ACTIVATE_FILE,
199 WORKINGSET_RESTORE_BASE,
200 WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE,
201 WORKINGSET_RESTORE_FILE,
202 WORKINGSET_NODERECLAIM,
203 NR_ANON_MAPPED, /* Mapped anonymous pages */
204 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
205 only modified from process context */
206 NR_FILE_PAGES,
207 NR_FILE_DIRTY,
208 NR_WRITEBACK,
209 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
210 NR_SHMEM_THPS,
211 NR_SHMEM_PMDMAPPED,
212 NR_FILE_THPS,
213 NR_FILE_PMDMAPPED,
214 NR_ANON_THPS,
215 NR_VMSCAN_WRITE,
216 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
217 NR_DIRTIED, /* page dirtyings since bootup */
218 NR_WRITTEN, /* page writings since bootup */
219 NR_THROTTLED_WRITTEN, /* NR_WRITTEN while reclaim throttled */
220 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
221 NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */
222 NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */
223 NR_KERNEL_STACK_KB, /* measured in KiB */
224#if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
225 NR_KERNEL_SCS_KB, /* measured in KiB */
226#endif
227 NR_PAGETABLE, /* used for pagetables */
228 NR_SECONDARY_PAGETABLE, /* secondary pagetables, KVM & IOMMU */
229#ifdef CONFIG_IOMMU_SUPPORT
230 NR_IOMMU_PAGES, /* # of pages allocated by IOMMU */
231#endif
232#ifdef CONFIG_SWAP
233 NR_SWAPCACHE,
234#endif
235#ifdef CONFIG_NUMA_BALANCING
236 PGPROMOTE_SUCCESS, /* promote successfully */
237 /**
238 * Candidate pages for promotion based on hint fault latency. This
239 * counter is used to control the promotion rate and adjust the hot
240 * threshold.
241 */
242 PGPROMOTE_CANDIDATE,
243 /**
244 * Not rate-limited (NRL) candidate pages for those can be promoted
245 * without considering hot threshold because of enough free pages in
246 * fast-tier node. These promotions bypass the regular hotness checks
247 * and do NOT influence the promotion rate-limiter or
248 * threshold-adjustment logic.
249 * This is for statistics/monitoring purposes.
250 */
251 PGPROMOTE_CANDIDATE_NRL,
252#endif
253 /* PGDEMOTE_*: pages demoted */
254 PGDEMOTE_KSWAPD,
255 PGDEMOTE_DIRECT,
256 PGDEMOTE_KHUGEPAGED,
257 PGDEMOTE_PROACTIVE,
258#ifdef CONFIG_HUGETLB_PAGE
259 NR_HUGETLB,
260#endif
261 NR_BALLOON_PAGES,
262 NR_KERNEL_FILE_PAGES,
263 NR_VM_NODE_STAT_ITEMS
264};
265
266/*
267 * Returns true if the item should be printed in THPs (/proc/vmstat
268 * currently prints number of anon, file and shmem THPs. But the item
269 * is charged in pages).
270 */
271static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item)
272{
273 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
274 return false;
275
276 return item == NR_ANON_THPS ||
277 item == NR_FILE_THPS ||
278 item == NR_SHMEM_THPS ||
279 item == NR_SHMEM_PMDMAPPED ||
280 item == NR_FILE_PMDMAPPED;
281}
282
283/*
284 * Returns true if the value is measured in bytes (most vmstat values are
285 * measured in pages). This defines the API part, the internal representation
286 * might be different.
287 */
288static __always_inline bool vmstat_item_in_bytes(int idx)
289{
290 /*
291 * Global and per-node slab counters track slab pages.
292 * It's expected that changes are multiples of PAGE_SIZE.
293 * Internally values are stored in pages.
294 *
295 * Per-memcg and per-lruvec counters track memory, consumed
296 * by individual slab objects. These counters are actually
297 * byte-precise.
298 */
299 return (idx == NR_SLAB_RECLAIMABLE_B ||
300 idx == NR_SLAB_UNRECLAIMABLE_B);
301}
302
303/*
304 * We do arithmetic on the LRU lists in various places in the code,
305 * so it is important to keep the active lists LRU_ACTIVE higher in
306 * the array than the corresponding inactive lists, and to keep
307 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
308 *
309 * This has to be kept in sync with the statistics in zone_stat_item
310 * above and the descriptions in vmstat_text in mm/vmstat.c
311 */
312#define LRU_BASE 0
313#define LRU_ACTIVE 1
314#define LRU_FILE 2
315
316enum lru_list {
317 LRU_INACTIVE_ANON = LRU_BASE,
318 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
319 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
320 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
321 LRU_UNEVICTABLE,
322 NR_LRU_LISTS
323};
324
325enum vmscan_throttle_state {
326 VMSCAN_THROTTLE_WRITEBACK,
327 VMSCAN_THROTTLE_ISOLATED,
328 VMSCAN_THROTTLE_NOPROGRESS,
329 VMSCAN_THROTTLE_CONGESTED,
330 NR_VMSCAN_THROTTLE,
331};
332
333#define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
334
335#define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
336
337static inline bool is_file_lru(enum lru_list lru)
338{
339 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
340}
341
342static inline bool is_active_lru(enum lru_list lru)
343{
344 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
345}
346
347#define WORKINGSET_ANON 0
348#define WORKINGSET_FILE 1
349#define ANON_AND_FILE 2
350
351enum lruvec_flags {
352 /*
353 * An lruvec has many dirty pages backed by a congested BDI:
354 * 1. LRUVEC_CGROUP_CONGESTED is set by cgroup-level reclaim.
355 * It can be cleared by cgroup reclaim or kswapd.
356 * 2. LRUVEC_NODE_CONGESTED is set by kswapd node-level reclaim.
357 * It can only be cleared by kswapd.
358 *
359 * Essentially, kswapd can unthrottle an lruvec throttled by cgroup
360 * reclaim, but not vice versa. This only applies to the root cgroup.
361 * The goal is to prevent cgroup reclaim on the root cgroup (e.g.
362 * memory.reclaim) to unthrottle an unbalanced node (that was throttled
363 * by kswapd).
364 */
365 LRUVEC_CGROUP_CONGESTED,
366 LRUVEC_NODE_CONGESTED,
367};
368
369#endif /* !__GENERATING_BOUNDS_H */
370
371/*
372 * Evictable folios are divided into multiple generations. The youngest and the
373 * oldest generation numbers, max_seq and min_seq, are monotonically increasing.
374 * They form a sliding window of a variable size [MIN_NR_GENS, MAX_NR_GENS]. An
375 * offset within MAX_NR_GENS, i.e., gen, indexes the LRU list of the
376 * corresponding generation. The gen counter in folio->flags stores gen+1 while
377 * a folio is on one of lrugen->folios[]. Otherwise it stores 0.
378 *
379 * After a folio is faulted in, the aging needs to check the accessed bit at
380 * least twice before handing this folio over to the eviction. The first check
381 * clears the accessed bit from the initial fault; the second check makes sure
382 * this folio hasn't been used since then. This process, AKA second chance,
383 * requires a minimum of two generations, hence MIN_NR_GENS. And to maintain ABI
384 * compatibility with the active/inactive LRU, e.g., /proc/vmstat, these two
385 * generations are considered active; the rest of generations, if they exist,
386 * are considered inactive. See lru_gen_is_active().
387 *
388 * PG_active is always cleared while a folio is on one of lrugen->folios[] so
389 * that the sliding window needs not to worry about it. And it's set again when
390 * a folio considered active is isolated for non-reclaiming purposes, e.g.,
391 * migration. See lru_gen_add_folio() and lru_gen_del_folio().
392 *
393 * MAX_NR_GENS is set to 4 so that the multi-gen LRU can support twice the
394 * number of categories of the active/inactive LRU when keeping track of
395 * accesses through page tables. This requires order_base_2(MAX_NR_GENS+1) bits
396 * in folio->flags, masked by LRU_GEN_MASK.
397 */
398#define MIN_NR_GENS 2U
399#define MAX_NR_GENS 4U
400
401/*
402 * Each generation is divided into multiple tiers. A folio accessed N times
403 * through file descriptors is in tier order_base_2(N). A folio in the first
404 * tier (N=0,1) is marked by PG_referenced unless it was faulted in through page
405 * tables or read ahead. A folio in the last tier (MAX_NR_TIERS-1) is marked by
406 * PG_workingset. A folio in any other tier (1<N<5) between the first and last
407 * is marked by additional bits of LRU_REFS_WIDTH in folio->flags.
408 *
409 * In contrast to moving across generations which requires the LRU lock, moving
410 * across tiers only involves atomic operations on folio->flags and therefore
411 * has a negligible cost in the buffered access path. In the eviction path,
412 * comparisons of refaulted/(evicted+protected) from the first tier and the rest
413 * infer whether folios accessed multiple times through file descriptors are
414 * statistically hot and thus worth protecting.
415 *
416 * MAX_NR_TIERS is set to 4 so that the multi-gen LRU can support twice the
417 * number of categories of the active/inactive LRU when keeping track of
418 * accesses through file descriptors. This uses MAX_NR_TIERS-2 spare bits in
419 * folio->flags, masked by LRU_REFS_MASK.
420 */
421#define MAX_NR_TIERS 4U
422
423#ifndef __GENERATING_BOUNDS_H
424
425#define LRU_GEN_MASK ((BIT(LRU_GEN_WIDTH) - 1) << LRU_GEN_PGOFF)
426#define LRU_REFS_MASK ((BIT(LRU_REFS_WIDTH) - 1) << LRU_REFS_PGOFF)
427
428/*
429 * For folios accessed multiple times through file descriptors,
430 * lru_gen_inc_refs() sets additional bits of LRU_REFS_WIDTH in folio->flags
431 * after PG_referenced, then PG_workingset after LRU_REFS_WIDTH. After all its
432 * bits are set, i.e., LRU_REFS_FLAGS|BIT(PG_workingset), a folio is lazily
433 * promoted into the second oldest generation in the eviction path. And when
434 * folio_inc_gen() does that, it clears LRU_REFS_FLAGS so that
435 * lru_gen_inc_refs() can start over. Note that for this case, LRU_REFS_MASK is
436 * only valid when PG_referenced is set.
437 *
438 * For folios accessed multiple times through page tables, folio_update_gen()
439 * from a page table walk or lru_gen_set_refs() from a rmap walk sets
440 * PG_referenced after the accessed bit is cleared for the first time.
441 * Thereafter, those two paths set PG_workingset and promote folios to the
442 * youngest generation. Like folio_inc_gen(), folio_update_gen() also clears
443 * PG_referenced. Note that for this case, LRU_REFS_MASK is not used.
444 *
445 * For both cases above, after PG_workingset is set on a folio, it remains until
446 * this folio is either reclaimed, or "deactivated" by lru_gen_clear_refs(). It
447 * can be set again if lru_gen_test_recent() returns true upon a refault.
448 */
449#define LRU_REFS_FLAGS (LRU_REFS_MASK | BIT(PG_referenced))
450
451struct lruvec;
452struct page_vma_mapped_walk;
453
454#ifdef CONFIG_LRU_GEN
455
456enum {
457 LRU_GEN_ANON,
458 LRU_GEN_FILE,
459};
460
461enum {
462 LRU_GEN_CORE,
463 LRU_GEN_MM_WALK,
464 LRU_GEN_NONLEAF_YOUNG,
465 NR_LRU_GEN_CAPS
466};
467
468#define MIN_LRU_BATCH BITS_PER_LONG
469#define MAX_LRU_BATCH (MIN_LRU_BATCH * 64)
470
471/* whether to keep historical stats from evicted generations */
472#ifdef CONFIG_LRU_GEN_STATS
473#define NR_HIST_GENS MAX_NR_GENS
474#else
475#define NR_HIST_GENS 1U
476#endif
477
478/*
479 * The youngest generation number is stored in max_seq for both anon and file
480 * types as they are aged on an equal footing. The oldest generation numbers are
481 * stored in min_seq[] separately for anon and file types so that they can be
482 * incremented independently. Ideally min_seq[] are kept in sync when both anon
483 * and file types are evictable. However, to adapt to situations like extreme
484 * swappiness, they are allowed to be out of sync by at most
485 * MAX_NR_GENS-MIN_NR_GENS-1.
486 *
487 * The number of pages in each generation is eventually consistent and therefore
488 * can be transiently negative when reset_batch_size() is pending.
489 */
490struct lru_gen_folio {
491 /* the aging increments the youngest generation number */
492 unsigned long max_seq;
493 /* the eviction increments the oldest generation numbers */
494 unsigned long min_seq[ANON_AND_FILE];
495 /* the birth time of each generation in jiffies */
496 unsigned long timestamps[MAX_NR_GENS];
497 /* the multi-gen LRU lists, lazily sorted on eviction */
498 struct list_head folios[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
499 /* the multi-gen LRU sizes, eventually consistent */
500 long nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
501 /* the exponential moving average of refaulted */
502 unsigned long avg_refaulted[ANON_AND_FILE][MAX_NR_TIERS];
503 /* the exponential moving average of evicted+protected */
504 unsigned long avg_total[ANON_AND_FILE][MAX_NR_TIERS];
505 /* can only be modified under the LRU lock */
506 unsigned long protected[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
507 /* can be modified without holding the LRU lock */
508 atomic_long_t evicted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
509 atomic_long_t refaulted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
510 /* whether the multi-gen LRU is enabled */
511 bool enabled;
512 /* the memcg generation this lru_gen_folio belongs to */
513 u8 gen;
514 /* the list segment this lru_gen_folio belongs to */
515 u8 seg;
516 /* per-node lru_gen_folio list for global reclaim */
517 struct hlist_nulls_node list;
518};
519
520enum {
521 MM_LEAF_TOTAL, /* total leaf entries */
522 MM_LEAF_YOUNG, /* young leaf entries */
523 MM_NONLEAF_FOUND, /* non-leaf entries found in Bloom filters */
524 MM_NONLEAF_ADDED, /* non-leaf entries added to Bloom filters */
525 NR_MM_STATS
526};
527
528/* double-buffering Bloom filters */
529#define NR_BLOOM_FILTERS 2
530
531struct lru_gen_mm_state {
532 /* synced with max_seq after each iteration */
533 unsigned long seq;
534 /* where the current iteration continues after */
535 struct list_head *head;
536 /* where the last iteration ended before */
537 struct list_head *tail;
538 /* Bloom filters flip after each iteration */
539 unsigned long *filters[NR_BLOOM_FILTERS];
540 /* the mm stats for debugging */
541 unsigned long stats[NR_HIST_GENS][NR_MM_STATS];
542};
543
544struct lru_gen_mm_walk {
545 /* the lruvec under reclaim */
546 struct lruvec *lruvec;
547 /* max_seq from lru_gen_folio: can be out of date */
548 unsigned long seq;
549 /* the next address within an mm to scan */
550 unsigned long next_addr;
551 /* to batch promoted pages */
552 int nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
553 /* to batch the mm stats */
554 int mm_stats[NR_MM_STATS];
555 /* total batched items */
556 int batched;
557 int swappiness;
558 bool force_scan;
559};
560
561/*
562 * For each node, memcgs are divided into two generations: the old and the
563 * young. For each generation, memcgs are randomly sharded into multiple bins
564 * to improve scalability. For each bin, the hlist_nulls is virtually divided
565 * into three segments: the head, the tail and the default.
566 *
567 * An onlining memcg is added to the tail of a random bin in the old generation.
568 * The eviction starts at the head of a random bin in the old generation. The
569 * per-node memcg generation counter, whose reminder (mod MEMCG_NR_GENS) indexes
570 * the old generation, is incremented when all its bins become empty.
571 *
572 * There are four operations:
573 * 1. MEMCG_LRU_HEAD, which moves a memcg to the head of a random bin in its
574 * current generation (old or young) and updates its "seg" to "head";
575 * 2. MEMCG_LRU_TAIL, which moves a memcg to the tail of a random bin in its
576 * current generation (old or young) and updates its "seg" to "tail";
577 * 3. MEMCG_LRU_OLD, which moves a memcg to the head of a random bin in the old
578 * generation, updates its "gen" to "old" and resets its "seg" to "default";
579 * 4. MEMCG_LRU_YOUNG, which moves a memcg to the tail of a random bin in the
580 * young generation, updates its "gen" to "young" and resets its "seg" to
581 * "default".
582 *
583 * The events that trigger the above operations are:
584 * 1. Exceeding the soft limit, which triggers MEMCG_LRU_HEAD;
585 * 2. The first attempt to reclaim a memcg below low, which triggers
586 * MEMCG_LRU_TAIL;
587 * 3. The first attempt to reclaim a memcg offlined or below reclaimable size
588 * threshold, which triggers MEMCG_LRU_TAIL;
589 * 4. The second attempt to reclaim a memcg offlined or below reclaimable size
590 * threshold, which triggers MEMCG_LRU_YOUNG;
591 * 5. Attempting to reclaim a memcg below min, which triggers MEMCG_LRU_YOUNG;
592 * 6. Finishing the aging on the eviction path, which triggers MEMCG_LRU_YOUNG;
593 * 7. Offlining a memcg, which triggers MEMCG_LRU_OLD.
594 *
595 * Notes:
596 * 1. Memcg LRU only applies to global reclaim, and the round-robin incrementing
597 * of their max_seq counters ensures the eventual fairness to all eligible
598 * memcgs. For memcg reclaim, it still relies on mem_cgroup_iter().
599 * 2. There are only two valid generations: old (seq) and young (seq+1).
600 * MEMCG_NR_GENS is set to three so that when reading the generation counter
601 * locklessly, a stale value (seq-1) does not wraparound to young.
602 */
603#define MEMCG_NR_GENS 3
604#define MEMCG_NR_BINS 8
605
606struct lru_gen_memcg {
607 /* the per-node memcg generation counter */
608 unsigned long seq;
609 /* each memcg has one lru_gen_folio per node */
610 unsigned long nr_memcgs[MEMCG_NR_GENS];
611 /* per-node lru_gen_folio list for global reclaim */
612 struct hlist_nulls_head fifo[MEMCG_NR_GENS][MEMCG_NR_BINS];
613 /* protects the above */
614 spinlock_t lock;
615};
616
617void lru_gen_init_pgdat(struct pglist_data *pgdat);
618void lru_gen_init_lruvec(struct lruvec *lruvec);
619bool lru_gen_look_around(struct page_vma_mapped_walk *pvmw);
620
621void lru_gen_init_memcg(struct mem_cgroup *memcg);
622void lru_gen_exit_memcg(struct mem_cgroup *memcg);
623void lru_gen_online_memcg(struct mem_cgroup *memcg);
624void lru_gen_offline_memcg(struct mem_cgroup *memcg);
625void lru_gen_release_memcg(struct mem_cgroup *memcg);
626void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid);
627
628#else /* !CONFIG_LRU_GEN */
629
630static inline void lru_gen_init_pgdat(struct pglist_data *pgdat)
631{
632}
633
634static inline void lru_gen_init_lruvec(struct lruvec *lruvec)
635{
636}
637
638static inline bool lru_gen_look_around(struct page_vma_mapped_walk *pvmw)
639{
640 return false;
641}
642
643static inline void lru_gen_init_memcg(struct mem_cgroup *memcg)
644{
645}
646
647static inline void lru_gen_exit_memcg(struct mem_cgroup *memcg)
648{
649}
650
651static inline void lru_gen_online_memcg(struct mem_cgroup *memcg)
652{
653}
654
655static inline void lru_gen_offline_memcg(struct mem_cgroup *memcg)
656{
657}
658
659static inline void lru_gen_release_memcg(struct mem_cgroup *memcg)
660{
661}
662
663static inline void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid)
664{
665}
666
667#endif /* CONFIG_LRU_GEN */
668
669struct lruvec {
670 struct list_head lists[NR_LRU_LISTS];
671 /* per lruvec lru_lock for memcg */
672 spinlock_t lru_lock;
673 /*
674 * These track the cost of reclaiming one LRU - file or anon -
675 * over the other. As the observed cost of reclaiming one LRU
676 * increases, the reclaim scan balance tips toward the other.
677 */
678 unsigned long anon_cost;
679 unsigned long file_cost;
680 /* Non-resident age, driven by LRU movement */
681 atomic_long_t nonresident_age;
682 /* Refaults at the time of last reclaim cycle */
683 unsigned long refaults[ANON_AND_FILE];
684 /* Various lruvec state flags (enum lruvec_flags) */
685 unsigned long flags;
686#ifdef CONFIG_LRU_GEN
687 /* evictable pages divided into generations */
688 struct lru_gen_folio lrugen;
689#ifdef CONFIG_LRU_GEN_WALKS_MMU
690 /* to concurrently iterate lru_gen_mm_list */
691 struct lru_gen_mm_state mm_state;
692#endif
693#endif /* CONFIG_LRU_GEN */
694#ifdef CONFIG_MEMCG
695 struct pglist_data *pgdat;
696#endif
697 struct zswap_lruvec_state zswap_lruvec_state;
698};
699
700/* Isolate for asynchronous migration */
701#define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
702/* Isolate unevictable pages */
703#define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
704
705/* LRU Isolation modes. */
706typedef unsigned __bitwise isolate_mode_t;
707
708enum zone_watermarks {
709 WMARK_MIN,
710 WMARK_LOW,
711 WMARK_HIGH,
712 WMARK_PROMO,
713 NR_WMARK
714};
715
716/*
717 * One per migratetype for each PAGE_ALLOC_COSTLY_ORDER. Two additional lists
718 * are added for THP. One PCP list is used by GPF_MOVABLE, and the other PCP list
719 * is used by GFP_UNMOVABLE and GFP_RECLAIMABLE.
720 */
721#ifdef CONFIG_TRANSPARENT_HUGEPAGE
722#define NR_PCP_THP 2
723#else
724#define NR_PCP_THP 0
725#endif
726#define NR_LOWORDER_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1))
727#define NR_PCP_LISTS (NR_LOWORDER_PCP_LISTS + NR_PCP_THP)
728
729/*
730 * Flags used in pcp->flags field.
731 *
732 * PCPF_PREV_FREE_HIGH_ORDER: a high-order page is freed in the
733 * previous page freeing. To avoid to drain PCP for an accident
734 * high-order page freeing.
735 *
736 * PCPF_FREE_HIGH_BATCH: preserve "pcp->batch" pages in PCP before
737 * draining PCP for consecutive high-order pages freeing without
738 * allocation if data cache slice of CPU is large enough. To reduce
739 * zone lock contention and keep cache-hot pages reusing.
740 */
741#define PCPF_PREV_FREE_HIGH_ORDER BIT(0)
742#define PCPF_FREE_HIGH_BATCH BIT(1)
743
744struct per_cpu_pages {
745 spinlock_t lock; /* Protects lists field */
746 int count; /* number of pages in the list */
747 int high; /* high watermark, emptying needed */
748 int high_min; /* min high watermark */
749 int high_max; /* max high watermark */
750 int batch; /* chunk size for buddy add/remove */
751 u8 flags; /* protected by pcp->lock */
752 u8 alloc_factor; /* batch scaling factor during allocate */
753#ifdef CONFIG_NUMA
754 u8 expire; /* When 0, remote pagesets are drained */
755#endif
756 short free_count; /* consecutive free count */
757
758 /* Lists of pages, one per migrate type stored on the pcp-lists */
759 struct list_head lists[NR_PCP_LISTS];
760} ____cacheline_aligned_in_smp;
761
762struct per_cpu_zonestat {
763#ifdef CONFIG_SMP
764 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
765 s8 stat_threshold;
766#endif
767#ifdef CONFIG_NUMA
768 /*
769 * Low priority inaccurate counters that are only folded
770 * on demand. Use a large type to avoid the overhead of
771 * folding during refresh_cpu_vm_stats.
772 */
773 unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
774#endif
775};
776
777struct per_cpu_nodestat {
778 s8 stat_threshold;
779 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
780};
781
782#endif /* !__GENERATING_BOUNDS.H */
783
784enum zone_type {
785 /*
786 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
787 * to DMA to all of the addressable memory (ZONE_NORMAL).
788 * On architectures where this area covers the whole 32 bit address
789 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
790 * DMA addressing constraints. This distinction is important as a 32bit
791 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
792 * platforms may need both zones as they support peripherals with
793 * different DMA addressing limitations.
794 */
795#ifdef CONFIG_ZONE_DMA
796 ZONE_DMA,
797#endif
798#ifdef CONFIG_ZONE_DMA32
799 ZONE_DMA32,
800#endif
801 /*
802 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
803 * performed on pages in ZONE_NORMAL if the DMA devices support
804 * transfers to all addressable memory.
805 */
806 ZONE_NORMAL,
807#ifdef CONFIG_HIGHMEM
808 /*
809 * A memory area that is only addressable by the kernel through
810 * mapping portions into its own address space. This is for example
811 * used by i386 to allow the kernel to address the memory beyond
812 * 900MB. The kernel will set up special mappings (page
813 * table entries on i386) for each page that the kernel needs to
814 * access.
815 */
816 ZONE_HIGHMEM,
817#endif
818 /*
819 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
820 * movable pages with few exceptional cases described below. Main use
821 * cases for ZONE_MOVABLE are to make memory offlining/unplug more
822 * likely to succeed, and to locally limit unmovable allocations - e.g.,
823 * to increase the number of THP/huge pages. Notable special cases are:
824 *
825 * 1. Pinned pages: (long-term) pinning of movable pages might
826 * essentially turn such pages unmovable. Therefore, we do not allow
827 * pinning long-term pages in ZONE_MOVABLE. When pages are pinned and
828 * faulted, they come from the right zone right away. However, it is
829 * still possible that address space already has pages in
830 * ZONE_MOVABLE at the time when pages are pinned (i.e. user has
831 * touches that memory before pinning). In such case we migrate them
832 * to a different zone. When migration fails - pinning fails.
833 * 2. memblock allocations: kernelcore/movablecore setups might create
834 * situations where ZONE_MOVABLE contains unmovable allocations
835 * after boot. Memory offlining and allocations fail early.
836 * 3. Memory holes: kernelcore/movablecore setups might create very rare
837 * situations where ZONE_MOVABLE contains memory holes after boot,
838 * for example, if we have sections that are only partially
839 * populated. Memory offlining and allocations fail early.
840 * 4. PG_hwpoison pages: while poisoned pages can be skipped during
841 * memory offlining, such pages cannot be allocated.
842 * 5. Unmovable PG_offline pages: in paravirtualized environments,
843 * hotplugged memory blocks might only partially be managed by the
844 * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
845 * parts not manged by the buddy are unmovable PG_offline pages. In
846 * some cases (virtio-mem), such pages can be skipped during
847 * memory offlining, however, cannot be moved/allocated. These
848 * techniques might use alloc_contig_range() to hide previously
849 * exposed pages from the buddy again (e.g., to implement some sort
850 * of memory unplug in virtio-mem).
851 * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create
852 * situations where ZERO_PAGE(0) which is allocated differently
853 * on different platforms may end up in a movable zone. ZERO_PAGE(0)
854 * cannot be migrated.
855 * 7. Memory-hotplug: when using memmap_on_memory and onlining the
856 * memory to the MOVABLE zone, the vmemmap pages are also placed in
857 * such zone. Such pages cannot be really moved around as they are
858 * self-stored in the range, but they are treated as movable when
859 * the range they describe is about to be offlined.
860 *
861 * In general, no unmovable allocations that degrade memory offlining
862 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
863 * have to expect that migrating pages in ZONE_MOVABLE can fail (even
864 * if has_unmovable_pages() states that there are no unmovable pages,
865 * there can be false negatives).
866 */
867 ZONE_MOVABLE,
868#ifdef CONFIG_ZONE_DEVICE
869 ZONE_DEVICE,
870#endif
871 __MAX_NR_ZONES
872
873};
874
875#ifndef __GENERATING_BOUNDS_H
876
877#define ASYNC_AND_SYNC 2
878
879struct zone {
880 /* Read-mostly fields */
881
882 /* zone watermarks, access with *_wmark_pages(zone) macros */
883 unsigned long _watermark[NR_WMARK];
884 unsigned long watermark_boost;
885
886 unsigned long nr_reserved_highatomic;
887 unsigned long nr_free_highatomic;
888
889 /*
890 * We don't know if the memory that we're going to allocate will be
891 * freeable or/and it will be released eventually, so to avoid totally
892 * wasting several GB of ram we must reserve some of the lower zone
893 * memory (otherwise we risk to run OOM on the lower zones despite
894 * there being tons of freeable ram on the higher zones). This array is
895 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
896 * changes.
897 */
898 long lowmem_reserve[MAX_NR_ZONES];
899
900#ifdef CONFIG_NUMA
901 int node;
902#endif
903 struct pglist_data *zone_pgdat;
904 struct per_cpu_pages __percpu *per_cpu_pageset;
905 struct per_cpu_zonestat __percpu *per_cpu_zonestats;
906 /*
907 * the high and batch values are copied to individual pagesets for
908 * faster access
909 */
910 int pageset_high_min;
911 int pageset_high_max;
912 int pageset_batch;
913
914#ifndef CONFIG_SPARSEMEM
915 /*
916 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
917 * In SPARSEMEM, this map is stored in struct mem_section
918 */
919 unsigned long *pageblock_flags;
920#endif /* CONFIG_SPARSEMEM */
921
922 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
923 unsigned long zone_start_pfn;
924
925 /*
926 * spanned_pages is the total pages spanned by the zone, including
927 * holes, which is calculated as:
928 * spanned_pages = zone_end_pfn - zone_start_pfn;
929 *
930 * present_pages is physical pages existing within the zone, which
931 * is calculated as:
932 * present_pages = spanned_pages - absent_pages(pages in holes);
933 *
934 * present_early_pages is present pages existing within the zone
935 * located on memory available since early boot, excluding hotplugged
936 * memory.
937 *
938 * managed_pages is present pages managed by the buddy system, which
939 * is calculated as (reserved_pages includes pages allocated by the
940 * bootmem allocator):
941 * managed_pages = present_pages - reserved_pages;
942 *
943 * cma pages is present pages that are assigned for CMA use
944 * (MIGRATE_CMA).
945 *
946 * So present_pages may be used by memory hotplug or memory power
947 * management logic to figure out unmanaged pages by checking
948 * (present_pages - managed_pages). And managed_pages should be used
949 * by page allocator and vm scanner to calculate all kinds of watermarks
950 * and thresholds.
951 *
952 * Locking rules:
953 *
954 * zone_start_pfn and spanned_pages are protected by span_seqlock.
955 * It is a seqlock because it has to be read outside of zone->lock,
956 * and it is done in the main allocator path. But, it is written
957 * quite infrequently.
958 *
959 * The span_seq lock is declared along with zone->lock because it is
960 * frequently read in proximity to zone->lock. It's good to
961 * give them a chance of being in the same cacheline.
962 *
963 * Write access to present_pages at runtime should be protected by
964 * mem_hotplug_begin/done(). Any reader who can't tolerant drift of
965 * present_pages should use get_online_mems() to get a stable value.
966 */
967 atomic_long_t managed_pages;
968 unsigned long spanned_pages;
969 unsigned long present_pages;
970#if defined(CONFIG_MEMORY_HOTPLUG)
971 unsigned long present_early_pages;
972#endif
973#ifdef CONFIG_CMA
974 unsigned long cma_pages;
975#endif
976
977 const char *name;
978
979#ifdef CONFIG_MEMORY_ISOLATION
980 /*
981 * Number of isolated pageblock. It is used to solve incorrect
982 * freepage counting problem due to racy retrieving migratetype
983 * of pageblock. Protected by zone->lock.
984 */
985 unsigned long nr_isolate_pageblock;
986#endif
987
988#ifdef CONFIG_MEMORY_HOTPLUG
989 /* see spanned/present_pages for more description */
990 seqlock_t span_seqlock;
991#endif
992
993 int initialized;
994
995 /* Write-intensive fields used from the page allocator */
996 CACHELINE_PADDING(_pad1_);
997
998 /* free areas of different sizes */
999 struct free_area free_area[NR_PAGE_ORDERS];
1000
1001#ifdef CONFIG_UNACCEPTED_MEMORY
1002 /* Pages to be accepted. All pages on the list are MAX_PAGE_ORDER */
1003 struct list_head unaccepted_pages;
1004
1005 /* To be called once the last page in the zone is accepted */
1006 struct work_struct unaccepted_cleanup;
1007#endif
1008
1009 /* zone flags, see below */
1010 unsigned long flags;
1011
1012 /* Primarily protects free_area */
1013 spinlock_t lock;
1014
1015 /* Pages to be freed when next trylock succeeds */
1016 struct llist_head trylock_free_pages;
1017
1018 /* Write-intensive fields used by compaction and vmstats. */
1019 CACHELINE_PADDING(_pad2_);
1020
1021 /*
1022 * When free pages are below this point, additional steps are taken
1023 * when reading the number of free pages to avoid per-cpu counter
1024 * drift allowing watermarks to be breached
1025 */
1026 unsigned long percpu_drift_mark;
1027
1028#if defined CONFIG_COMPACTION || defined CONFIG_CMA
1029 /* pfn where compaction free scanner should start */
1030 unsigned long compact_cached_free_pfn;
1031 /* pfn where compaction migration scanner should start */
1032 unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC];
1033 unsigned long compact_init_migrate_pfn;
1034 unsigned long compact_init_free_pfn;
1035#endif
1036
1037#ifdef CONFIG_COMPACTION
1038 /*
1039 * On compaction failure, 1<<compact_defer_shift compactions
1040 * are skipped before trying again. The number attempted since
1041 * last failure is tracked with compact_considered.
1042 * compact_order_failed is the minimum compaction failed order.
1043 */
1044 unsigned int compact_considered;
1045 unsigned int compact_defer_shift;
1046 int compact_order_failed;
1047#endif
1048
1049#if defined CONFIG_COMPACTION || defined CONFIG_CMA
1050 /* Set to true when the PG_migrate_skip bits should be cleared */
1051 bool compact_blockskip_flush;
1052#endif
1053
1054 bool contiguous;
1055
1056 CACHELINE_PADDING(_pad3_);
1057 /* Zone statistics */
1058 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
1059 atomic_long_t vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
1060} ____cacheline_internodealigned_in_smp;
1061
1062enum pgdat_flags {
1063 PGDAT_WRITEBACK, /* reclaim scanning has recently found
1064 * many pages under writeback
1065 */
1066 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
1067};
1068
1069enum zone_flags {
1070 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
1071 * Cleared when kswapd is woken.
1072 */
1073 ZONE_RECLAIM_ACTIVE, /* kswapd may be scanning the zone. */
1074 ZONE_BELOW_HIGH, /* zone is below high watermark. */
1075};
1076
1077static inline unsigned long wmark_pages(const struct zone *z,
1078 enum zone_watermarks w)
1079{
1080 return z->_watermark[w] + z->watermark_boost;
1081}
1082
1083static inline unsigned long min_wmark_pages(const struct zone *z)
1084{
1085 return wmark_pages(z, WMARK_MIN);
1086}
1087
1088static inline unsigned long low_wmark_pages(const struct zone *z)
1089{
1090 return wmark_pages(z, WMARK_LOW);
1091}
1092
1093static inline unsigned long high_wmark_pages(const struct zone *z)
1094{
1095 return wmark_pages(z, WMARK_HIGH);
1096}
1097
1098static inline unsigned long promo_wmark_pages(const struct zone *z)
1099{
1100 return wmark_pages(z, WMARK_PROMO);
1101}
1102
1103static inline unsigned long zone_managed_pages(const struct zone *zone)
1104{
1105 return (unsigned long)atomic_long_read(&zone->managed_pages);
1106}
1107
1108static inline unsigned long zone_cma_pages(struct zone *zone)
1109{
1110#ifdef CONFIG_CMA
1111 return zone->cma_pages;
1112#else
1113 return 0;
1114#endif
1115}
1116
1117static inline unsigned long zone_end_pfn(const struct zone *zone)
1118{
1119 return zone->zone_start_pfn + zone->spanned_pages;
1120}
1121
1122static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
1123{
1124 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
1125}
1126
1127static inline bool zone_is_initialized(const struct zone *zone)
1128{
1129 return zone->initialized;
1130}
1131
1132static inline bool zone_is_empty(const struct zone *zone)
1133{
1134 return zone->spanned_pages == 0;
1135}
1136
1137#ifndef BUILD_VDSO32_64
1138/*
1139 * The zone field is never updated after free_area_init_core()
1140 * sets it, so none of the operations on it need to be atomic.
1141 */
1142
1143/* Page flags: | [SECTION] | [NODE] | ZONE | [LAST_CPUPID] | ... | FLAGS | */
1144#define SECTIONS_PGOFF ((sizeof(unsigned long)*8) - SECTIONS_WIDTH)
1145#define NODES_PGOFF (SECTIONS_PGOFF - NODES_WIDTH)
1146#define ZONES_PGOFF (NODES_PGOFF - ZONES_WIDTH)
1147#define LAST_CPUPID_PGOFF (ZONES_PGOFF - LAST_CPUPID_WIDTH)
1148#define KASAN_TAG_PGOFF (LAST_CPUPID_PGOFF - KASAN_TAG_WIDTH)
1149#define LRU_GEN_PGOFF (KASAN_TAG_PGOFF - LRU_GEN_WIDTH)
1150#define LRU_REFS_PGOFF (LRU_GEN_PGOFF - LRU_REFS_WIDTH)
1151
1152/*
1153 * Define the bit shifts to access each section. For non-existent
1154 * sections we define the shift as 0; that plus a 0 mask ensures
1155 * the compiler will optimise away reference to them.
1156 */
1157#define SECTIONS_PGSHIFT (SECTIONS_PGOFF * (SECTIONS_WIDTH != 0))
1158#define NODES_PGSHIFT (NODES_PGOFF * (NODES_WIDTH != 0))
1159#define ZONES_PGSHIFT (ZONES_PGOFF * (ZONES_WIDTH != 0))
1160#define LAST_CPUPID_PGSHIFT (LAST_CPUPID_PGOFF * (LAST_CPUPID_WIDTH != 0))
1161#define KASAN_TAG_PGSHIFT (KASAN_TAG_PGOFF * (KASAN_TAG_WIDTH != 0))
1162
1163/* NODE:ZONE or SECTION:ZONE is used to ID a zone for the buddy allocator */
1164#ifdef NODE_NOT_IN_PAGE_FLAGS
1165#define ZONEID_SHIFT (SECTIONS_SHIFT + ZONES_SHIFT)
1166#define ZONEID_PGOFF ((SECTIONS_PGOFF < ZONES_PGOFF) ? \
1167 SECTIONS_PGOFF : ZONES_PGOFF)
1168#else
1169#define ZONEID_SHIFT (NODES_SHIFT + ZONES_SHIFT)
1170#define ZONEID_PGOFF ((NODES_PGOFF < ZONES_PGOFF) ? \
1171 NODES_PGOFF : ZONES_PGOFF)
1172#endif
1173
1174#define ZONEID_PGSHIFT (ZONEID_PGOFF * (ZONEID_SHIFT != 0))
1175
1176#define ZONES_MASK ((1UL << ZONES_WIDTH) - 1)
1177#define NODES_MASK ((1UL << NODES_WIDTH) - 1)
1178#define SECTIONS_MASK ((1UL << SECTIONS_WIDTH) - 1)
1179#define LAST_CPUPID_MASK ((1UL << LAST_CPUPID_SHIFT) - 1)
1180#define KASAN_TAG_MASK ((1UL << KASAN_TAG_WIDTH) - 1)
1181#define ZONEID_MASK ((1UL << ZONEID_SHIFT) - 1)
1182
1183static inline enum zone_type memdesc_zonenum(memdesc_flags_t flags)
1184{
1185 ASSERT_EXCLUSIVE_BITS(flags.f, ZONES_MASK << ZONES_PGSHIFT);
1186 return (flags.f >> ZONES_PGSHIFT) & ZONES_MASK;
1187}
1188
1189static inline enum zone_type page_zonenum(const struct page *page)
1190{
1191 return memdesc_zonenum(page->flags);
1192}
1193
1194static inline enum zone_type folio_zonenum(const struct folio *folio)
1195{
1196 return memdesc_zonenum(folio->flags);
1197}
1198
1199#ifdef CONFIG_ZONE_DEVICE
1200static inline bool memdesc_is_zone_device(memdesc_flags_t mdf)
1201{
1202 return memdesc_zonenum(mdf) == ZONE_DEVICE;
1203}
1204
1205static inline struct dev_pagemap *page_pgmap(const struct page *page)
1206{
1207 VM_WARN_ON_ONCE_PAGE(!memdesc_is_zone_device(page->flags), page);
1208 return page_folio(page)->pgmap;
1209}
1210
1211/*
1212 * Consecutive zone device pages should not be merged into the same sgl
1213 * or bvec segment with other types of pages or if they belong to different
1214 * pgmaps. Otherwise getting the pgmap of a given segment is not possible
1215 * without scanning the entire segment. This helper returns true either if
1216 * both pages are not zone device pages or both pages are zone device pages
1217 * with the same pgmap.
1218 */
1219static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1220 const struct page *b)
1221{
1222 if (memdesc_is_zone_device(a->flags) != memdesc_is_zone_device(b->flags))
1223 return false;
1224 if (!memdesc_is_zone_device(a->flags))
1225 return true;
1226 return page_pgmap(a) == page_pgmap(b);
1227}
1228
1229extern void memmap_init_zone_device(struct zone *, unsigned long,
1230 unsigned long, struct dev_pagemap *);
1231#else
1232static inline bool memdesc_is_zone_device(memdesc_flags_t mdf)
1233{
1234 return false;
1235}
1236static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1237 const struct page *b)
1238{
1239 return true;
1240}
1241static inline struct dev_pagemap *page_pgmap(const struct page *page)
1242{
1243 return NULL;
1244}
1245#endif
1246
1247static inline bool is_zone_device_page(const struct page *page)
1248{
1249 return memdesc_is_zone_device(page->flags);
1250}
1251
1252static inline bool folio_is_zone_device(const struct folio *folio)
1253{
1254 return memdesc_is_zone_device(folio->flags);
1255}
1256
1257static inline bool is_zone_movable_page(const struct page *page)
1258{
1259 return page_zonenum(page) == ZONE_MOVABLE;
1260}
1261
1262static inline bool folio_is_zone_movable(const struct folio *folio)
1263{
1264 return folio_zonenum(folio) == ZONE_MOVABLE;
1265}
1266#endif
1267
1268/*
1269 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
1270 * intersection with the given zone
1271 */
1272static inline bool zone_intersects(const struct zone *zone,
1273 unsigned long start_pfn, unsigned long nr_pages)
1274{
1275 if (zone_is_empty(zone))
1276 return false;
1277 if (start_pfn >= zone_end_pfn(zone) ||
1278 start_pfn + nr_pages <= zone->zone_start_pfn)
1279 return false;
1280
1281 return true;
1282}
1283
1284/*
1285 * The "priority" of VM scanning is how much of the queues we will scan in one
1286 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
1287 * queues ("queue_length >> 12") during an aging round.
1288 */
1289#define DEF_PRIORITY 12
1290
1291/* Maximum number of zones on a zonelist */
1292#define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
1293
1294enum {
1295 ZONELIST_FALLBACK, /* zonelist with fallback */
1296#ifdef CONFIG_NUMA
1297 /*
1298 * The NUMA zonelists are doubled because we need zonelists that
1299 * restrict the allocations to a single node for __GFP_THISNODE.
1300 */
1301 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
1302#endif
1303 MAX_ZONELISTS
1304};
1305
1306/*
1307 * This struct contains information about a zone in a zonelist. It is stored
1308 * here to avoid dereferences into large structures and lookups of tables
1309 */
1310struct zoneref {
1311 struct zone *zone; /* Pointer to actual zone */
1312 int zone_idx; /* zone_idx(zoneref->zone) */
1313};
1314
1315/*
1316 * One allocation request operates on a zonelist. A zonelist
1317 * is a list of zones, the first one is the 'goal' of the
1318 * allocation, the other zones are fallback zones, in decreasing
1319 * priority.
1320 *
1321 * To speed the reading of the zonelist, the zonerefs contain the zone index
1322 * of the entry being read. Helper functions to access information given
1323 * a struct zoneref are
1324 *
1325 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
1326 * zonelist_zone_idx() - Return the index of the zone for an entry
1327 * zonelist_node_idx() - Return the index of the node for an entry
1328 */
1329struct zonelist {
1330 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
1331};
1332
1333/*
1334 * The array of struct pages for flatmem.
1335 * It must be declared for SPARSEMEM as well because there are configurations
1336 * that rely on that.
1337 */
1338extern struct page *mem_map;
1339
1340#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1341struct deferred_split {
1342 spinlock_t split_queue_lock;
1343 struct list_head split_queue;
1344 unsigned long split_queue_len;
1345};
1346#endif
1347
1348#ifdef CONFIG_MEMORY_FAILURE
1349/*
1350 * Per NUMA node memory failure handling statistics.
1351 */
1352struct memory_failure_stats {
1353 /*
1354 * Number of raw pages poisoned.
1355 * Cases not accounted: memory outside kernel control, offline page,
1356 * arch-specific memory_failure (SGX), hwpoison_filter() filtered
1357 * error events, and unpoison actions from hwpoison_unpoison.
1358 */
1359 unsigned long total;
1360 /*
1361 * Recovery results of poisoned raw pages handled by memory_failure,
1362 * in sync with mf_result.
1363 * total = ignored + failed + delayed + recovered.
1364 * total * PAGE_SIZE * #nodes = /proc/meminfo/HardwareCorrupted.
1365 */
1366 unsigned long ignored;
1367 unsigned long failed;
1368 unsigned long delayed;
1369 unsigned long recovered;
1370};
1371#endif
1372
1373/*
1374 * On NUMA machines, each NUMA node would have a pg_data_t to describe
1375 * it's memory layout. On UMA machines there is a single pglist_data which
1376 * describes the whole memory.
1377 *
1378 * Memory statistics and page replacement data structures are maintained on a
1379 * per-zone basis.
1380 */
1381typedef struct pglist_data {
1382 /*
1383 * node_zones contains just the zones for THIS node. Not all of the
1384 * zones may be populated, but it is the full list. It is referenced by
1385 * this node's node_zonelists as well as other node's node_zonelists.
1386 */
1387 struct zone node_zones[MAX_NR_ZONES];
1388
1389 /*
1390 * node_zonelists contains references to all zones in all nodes.
1391 * Generally the first zones will be references to this node's
1392 * node_zones.
1393 */
1394 struct zonelist node_zonelists[MAX_ZONELISTS];
1395
1396 int nr_zones; /* number of populated zones in this node */
1397#ifdef CONFIG_FLATMEM /* means !SPARSEMEM */
1398 struct page *node_mem_map;
1399#ifdef CONFIG_PAGE_EXTENSION
1400 struct page_ext *node_page_ext;
1401#endif
1402#endif
1403#if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
1404 /*
1405 * Must be held any time you expect node_start_pfn,
1406 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
1407 * Also synchronizes pgdat->first_deferred_pfn during deferred page
1408 * init.
1409 *
1410 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
1411 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
1412 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
1413 *
1414 * Nests above zone->lock and zone->span_seqlock
1415 */
1416 spinlock_t node_size_lock;
1417#endif
1418 unsigned long node_start_pfn;
1419 unsigned long node_present_pages; /* total number of physical pages */
1420 unsigned long node_spanned_pages; /* total size of physical page
1421 range, including holes */
1422 int node_id;
1423 wait_queue_head_t kswapd_wait;
1424 wait_queue_head_t pfmemalloc_wait;
1425
1426 /* workqueues for throttling reclaim for different reasons. */
1427 wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE];
1428
1429 atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */
1430 unsigned long nr_reclaim_start; /* nr pages written while throttled
1431 * when throttling started. */
1432#ifdef CONFIG_MEMORY_HOTPLUG
1433 struct mutex kswapd_lock;
1434#endif
1435 struct task_struct *kswapd; /* Protected by kswapd_lock */
1436 int kswapd_order;
1437 enum zone_type kswapd_highest_zoneidx;
1438
1439 atomic_t kswapd_failures; /* Number of 'reclaimed == 0' runs */
1440
1441#ifdef CONFIG_COMPACTION
1442 int kcompactd_max_order;
1443 enum zone_type kcompactd_highest_zoneidx;
1444 wait_queue_head_t kcompactd_wait;
1445 struct task_struct *kcompactd;
1446 bool proactive_compact_trigger;
1447#endif
1448 /*
1449 * This is a per-node reserve of pages that are not available
1450 * to userspace allocations.
1451 */
1452 unsigned long totalreserve_pages;
1453
1454#ifdef CONFIG_NUMA
1455 /*
1456 * node reclaim becomes active if more unmapped pages exist.
1457 */
1458 unsigned long min_unmapped_pages;
1459 unsigned long min_slab_pages;
1460#endif /* CONFIG_NUMA */
1461
1462 /* Write-intensive fields used by page reclaim */
1463 CACHELINE_PADDING(_pad1_);
1464
1465#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1466 /*
1467 * If memory initialisation on large machines is deferred then this
1468 * is the first PFN that needs to be initialised.
1469 */
1470 unsigned long first_deferred_pfn;
1471#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1472
1473#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1474 struct deferred_split deferred_split_queue;
1475#endif
1476
1477#ifdef CONFIG_NUMA_BALANCING
1478 /* start time in ms of current promote rate limit period */
1479 unsigned int nbp_rl_start;
1480 /* number of promote candidate pages at start time of current rate limit period */
1481 unsigned long nbp_rl_nr_cand;
1482 /* promote threshold in ms */
1483 unsigned int nbp_threshold;
1484 /* start time in ms of current promote threshold adjustment period */
1485 unsigned int nbp_th_start;
1486 /*
1487 * number of promote candidate pages at start time of current promote
1488 * threshold adjustment period
1489 */
1490 unsigned long nbp_th_nr_cand;
1491#endif
1492 /* Fields commonly accessed by the page reclaim scanner */
1493
1494 /*
1495 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
1496 *
1497 * Use mem_cgroup_lruvec() to look up lruvecs.
1498 */
1499 struct lruvec __lruvec;
1500
1501 unsigned long flags;
1502
1503#ifdef CONFIG_LRU_GEN
1504 /* kswap mm walk data */
1505 struct lru_gen_mm_walk mm_walk;
1506 /* lru_gen_folio list */
1507 struct lru_gen_memcg memcg_lru;
1508#endif
1509
1510 CACHELINE_PADDING(_pad2_);
1511
1512 /* Per-node vmstats */
1513 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
1514 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
1515#ifdef CONFIG_NUMA
1516 struct memory_tier __rcu *memtier;
1517#endif
1518#ifdef CONFIG_MEMORY_FAILURE
1519 struct memory_failure_stats mf_stats;
1520#endif
1521} pg_data_t;
1522
1523#define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
1524#define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
1525
1526#define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
1527#define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
1528
1529static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
1530{
1531 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
1532}
1533
1534#include <linux/memory_hotplug.h>
1535
1536void build_all_zonelists(pg_data_t *pgdat);
1537bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
1538 int highest_zoneidx, unsigned int alloc_flags,
1539 long free_pages);
1540bool zone_watermark_ok(struct zone *z, unsigned int order,
1541 unsigned long mark, int highest_zoneidx,
1542 unsigned int alloc_flags);
1543
1544enum kswapd_clear_hopeless_reason {
1545 KSWAPD_CLEAR_HOPELESS_OTHER = 0,
1546 KSWAPD_CLEAR_HOPELESS_KSWAPD,
1547 KSWAPD_CLEAR_HOPELESS_DIRECT,
1548 KSWAPD_CLEAR_HOPELESS_PCP,
1549};
1550
1551void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
1552 enum zone_type highest_zoneidx);
1553void kswapd_try_clear_hopeless(struct pglist_data *pgdat,
1554 unsigned int order, int highest_zoneidx);
1555void kswapd_clear_hopeless(pg_data_t *pgdat, enum kswapd_clear_hopeless_reason reason);
1556bool kswapd_test_hopeless(pg_data_t *pgdat);
1557
1558/*
1559 * Memory initialization context, use to differentiate memory added by
1560 * the platform statically or via memory hotplug interface.
1561 */
1562enum meminit_context {
1563 MEMINIT_EARLY,
1564 MEMINIT_HOTPLUG,
1565};
1566
1567extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
1568 unsigned long size);
1569
1570extern void lruvec_init(struct lruvec *lruvec);
1571
1572static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
1573{
1574#ifdef CONFIG_MEMCG
1575 return lruvec->pgdat;
1576#else
1577 return container_of(lruvec, struct pglist_data, __lruvec);
1578#endif
1579}
1580
1581#ifdef CONFIG_HAVE_MEMORYLESS_NODES
1582int local_memory_node(int node_id);
1583#else
1584static inline int local_memory_node(int node_id) { return node_id; };
1585#endif
1586
1587/*
1588 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
1589 */
1590#define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
1591
1592#ifdef CONFIG_ZONE_DEVICE
1593static inline bool zone_is_zone_device(const struct zone *zone)
1594{
1595 return zone_idx(zone) == ZONE_DEVICE;
1596}
1597#else
1598static inline bool zone_is_zone_device(const struct zone *zone)
1599{
1600 return false;
1601}
1602#endif
1603
1604/*
1605 * Returns true if a zone has pages managed by the buddy allocator.
1606 * All the reclaim decisions have to use this function rather than
1607 * populated_zone(). If the whole zone is reserved then we can easily
1608 * end up with populated_zone() && !managed_zone().
1609 */
1610static inline bool managed_zone(const struct zone *zone)
1611{
1612 return zone_managed_pages(zone);
1613}
1614
1615/* Returns true if a zone has memory */
1616static inline bool populated_zone(const struct zone *zone)
1617{
1618 return zone->present_pages;
1619}
1620
1621#ifdef CONFIG_NUMA
1622static inline int zone_to_nid(const struct zone *zone)
1623{
1624 return zone->node;
1625}
1626
1627static inline void zone_set_nid(struct zone *zone, int nid)
1628{
1629 zone->node = nid;
1630}
1631#else
1632static inline int zone_to_nid(const struct zone *zone)
1633{
1634 return 0;
1635}
1636
1637static inline void zone_set_nid(struct zone *zone, int nid) {}
1638#endif
1639
1640extern int movable_zone;
1641
1642static inline int is_highmem_idx(enum zone_type idx)
1643{
1644#ifdef CONFIG_HIGHMEM
1645 return (idx == ZONE_HIGHMEM ||
1646 (idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM));
1647#else
1648 return 0;
1649#endif
1650}
1651
1652/**
1653 * is_highmem - helper function to quickly check if a struct zone is a
1654 * highmem zone or not. This is an attempt to keep references
1655 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
1656 * @zone: pointer to struct zone variable
1657 * Return: 1 for a highmem zone, 0 otherwise
1658 */
1659static inline int is_highmem(const struct zone *zone)
1660{
1661 return is_highmem_idx(zone_idx(zone));
1662}
1663
1664bool has_managed_zone(enum zone_type zone);
1665static inline bool has_managed_dma(void)
1666{
1667#ifdef CONFIG_ZONE_DMA
1668 return has_managed_zone(ZONE_DMA);
1669#else
1670 return false;
1671#endif
1672}
1673
1674
1675#ifndef CONFIG_NUMA
1676
1677extern struct pglist_data contig_page_data;
1678static inline struct pglist_data *NODE_DATA(int nid)
1679{
1680 return &contig_page_data;
1681}
1682
1683#else /* CONFIG_NUMA */
1684
1685#include <asm/mmzone.h>
1686
1687#endif /* !CONFIG_NUMA */
1688
1689extern struct pglist_data *first_online_pgdat(void);
1690extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
1691extern struct zone *next_zone(struct zone *zone);
1692
1693/**
1694 * for_each_online_pgdat - helper macro to iterate over all online nodes
1695 * @pgdat: pointer to a pg_data_t variable
1696 */
1697#define for_each_online_pgdat(pgdat) \
1698 for (pgdat = first_online_pgdat(); \
1699 pgdat; \
1700 pgdat = next_online_pgdat(pgdat))
1701/**
1702 * for_each_zone - helper macro to iterate over all memory zones
1703 * @zone: pointer to struct zone variable
1704 *
1705 * The user only needs to declare the zone variable, for_each_zone
1706 * fills it in.
1707 */
1708#define for_each_zone(zone) \
1709 for (zone = (first_online_pgdat())->node_zones; \
1710 zone; \
1711 zone = next_zone(zone))
1712
1713#define for_each_populated_zone(zone) \
1714 for (zone = (first_online_pgdat())->node_zones; \
1715 zone; \
1716 zone = next_zone(zone)) \
1717 if (!populated_zone(zone)) \
1718 ; /* do nothing */ \
1719 else
1720
1721static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1722{
1723 return zoneref->zone;
1724}
1725
1726static inline int zonelist_zone_idx(const struct zoneref *zoneref)
1727{
1728 return zoneref->zone_idx;
1729}
1730
1731static inline int zonelist_node_idx(const struct zoneref *zoneref)
1732{
1733 return zone_to_nid(zoneref->zone);
1734}
1735
1736struct zoneref *__next_zones_zonelist(struct zoneref *z,
1737 enum zone_type highest_zoneidx,
1738 nodemask_t *nodes);
1739
1740/**
1741 * 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
1742 * @z: The cursor used as a starting point for the search
1743 * @highest_zoneidx: The zone index of the highest zone to return
1744 * @nodes: An optional nodemask to filter the zonelist with
1745 *
1746 * This function returns the next zone at or below a given zone index that is
1747 * within the allowed nodemask using a cursor as the starting point for the
1748 * search. The zoneref returned is a cursor that represents the current zone
1749 * being examined. It should be advanced by one before calling
1750 * next_zones_zonelist again.
1751 *
1752 * Return: the next zone at or below highest_zoneidx within the allowed
1753 * nodemask using a cursor within a zonelist as a starting point
1754 */
1755static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1756 enum zone_type highest_zoneidx,
1757 nodemask_t *nodes)
1758{
1759 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1760 return z;
1761 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1762}
1763
1764/**
1765 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1766 * @zonelist: The zonelist to search for a suitable zone
1767 * @highest_zoneidx: The zone index of the highest zone to return
1768 * @nodes: An optional nodemask to filter the zonelist with
1769 *
1770 * This function returns the first zone at or below a given zone index that is
1771 * within the allowed nodemask. The zoneref returned is a cursor that can be
1772 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1773 * one before calling.
1774 *
1775 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1776 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1777 * update due to cpuset modification.
1778 *
1779 * Return: Zoneref pointer for the first suitable zone found
1780 */
1781static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1782 enum zone_type highest_zoneidx,
1783 nodemask_t *nodes)
1784{
1785 return next_zones_zonelist(zonelist->_zonerefs,
1786 highest_zoneidx, nodes);
1787}
1788
1789/**
1790 * 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
1791 * @zone: The current zone in the iterator
1792 * @z: The current pointer within zonelist->_zonerefs being iterated
1793 * @zlist: The zonelist being iterated
1794 * @highidx: The zone index of the highest zone to return
1795 * @nodemask: Nodemask allowed by the allocator
1796 *
1797 * This iterator iterates though all zones at or below a given zone index and
1798 * within a given nodemask
1799 */
1800#define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1801 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1802 zone; \
1803 z = next_zones_zonelist(++z, highidx, nodemask), \
1804 zone = zonelist_zone(z))
1805
1806#define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
1807 for (zone = zonelist_zone(z); \
1808 zone; \
1809 z = next_zones_zonelist(++z, highidx, nodemask), \
1810 zone = zonelist_zone(z))
1811
1812
1813/**
1814 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1815 * @zone: The current zone in the iterator
1816 * @z: The current pointer within zonelist->zones being iterated
1817 * @zlist: The zonelist being iterated
1818 * @highidx: The zone index of the highest zone to return
1819 *
1820 * This iterator iterates though all zones at or below a given zone index.
1821 */
1822#define for_each_zone_zonelist(zone, z, zlist, highidx) \
1823 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1824
1825/* Whether the 'nodes' are all movable nodes */
1826static inline bool movable_only_nodes(nodemask_t *nodes)
1827{
1828 struct zonelist *zonelist;
1829 struct zoneref *z;
1830 int nid;
1831
1832 if (nodes_empty(*nodes))
1833 return false;
1834
1835 /*
1836 * We can chose arbitrary node from the nodemask to get a
1837 * zonelist as they are interlinked. We just need to find
1838 * at least one zone that can satisfy kernel allocations.
1839 */
1840 nid = first_node(*nodes);
1841 zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK];
1842 z = first_zones_zonelist(zonelist, ZONE_NORMAL, nodes);
1843 return (!zonelist_zone(z)) ? true : false;
1844}
1845
1846
1847#ifdef CONFIG_SPARSEMEM
1848#include <asm/sparsemem.h>
1849#endif
1850
1851#ifdef CONFIG_FLATMEM
1852#define pfn_to_nid(pfn) (0)
1853#endif
1854
1855#ifdef CONFIG_SPARSEMEM
1856
1857/*
1858 * PA_SECTION_SHIFT physical address to/from section number
1859 * PFN_SECTION_SHIFT pfn to/from section number
1860 */
1861#define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1862#define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1863
1864#define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1865
1866#define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1867#define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1868
1869#define SECTION_BLOCKFLAGS_BITS \
1870 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1871
1872#if (MAX_PAGE_ORDER + PAGE_SHIFT) > SECTION_SIZE_BITS
1873#error Allocator MAX_PAGE_ORDER exceeds SECTION_SIZE
1874#endif
1875
1876static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1877{
1878 return pfn >> PFN_SECTION_SHIFT;
1879}
1880static inline unsigned long section_nr_to_pfn(unsigned long sec)
1881{
1882 return sec << PFN_SECTION_SHIFT;
1883}
1884
1885#define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1886#define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1887
1888#define SUBSECTION_SHIFT 21
1889#define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1890
1891#define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1892#define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1893#define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1894
1895#if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1896#error Subsection size exceeds section size
1897#else
1898#define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1899#endif
1900
1901#define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1902#define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1903
1904struct mem_section_usage {
1905 struct rcu_head rcu;
1906#ifdef CONFIG_SPARSEMEM_VMEMMAP
1907 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1908#endif
1909 /* See declaration of similar field in struct zone */
1910 unsigned long pageblock_flags[0];
1911};
1912
1913void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1914
1915struct page;
1916struct page_ext;
1917struct mem_section {
1918 /*
1919 * This is, logically, a pointer to an array of struct
1920 * pages. However, it is stored with some other magic.
1921 * (see sparse.c::sparse_init_one_section())
1922 *
1923 * Additionally during early boot we encode node id of
1924 * the location of the section here to guide allocation.
1925 * (see sparse.c::memory_present())
1926 *
1927 * Making it a UL at least makes someone do a cast
1928 * before using it wrong.
1929 */
1930 unsigned long section_mem_map;
1931
1932 struct mem_section_usage *usage;
1933#ifdef CONFIG_PAGE_EXTENSION
1934 /*
1935 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1936 * section. (see page_ext.h about this.)
1937 */
1938 struct page_ext *page_ext;
1939 unsigned long pad;
1940#endif
1941 /*
1942 * WARNING: mem_section must be a power-of-2 in size for the
1943 * calculation and use of SECTION_ROOT_MASK to make sense.
1944 */
1945};
1946
1947#ifdef CONFIG_SPARSEMEM_EXTREME
1948#define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1949#else
1950#define SECTIONS_PER_ROOT 1
1951#endif
1952
1953#define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1954#define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1955#define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1956
1957#ifdef CONFIG_SPARSEMEM_EXTREME
1958extern struct mem_section **mem_section;
1959#else
1960extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1961#endif
1962
1963static inline unsigned long *section_to_usemap(struct mem_section *ms)
1964{
1965 return ms->usage->pageblock_flags;
1966}
1967
1968static inline struct mem_section *__nr_to_section(unsigned long nr)
1969{
1970 unsigned long root = SECTION_NR_TO_ROOT(nr);
1971
1972 if (unlikely(root >= NR_SECTION_ROOTS))
1973 return NULL;
1974
1975#ifdef CONFIG_SPARSEMEM_EXTREME
1976 if (!mem_section || !mem_section[root])
1977 return NULL;
1978#endif
1979 return &mem_section[root][nr & SECTION_ROOT_MASK];
1980}
1981extern size_t mem_section_usage_size(void);
1982
1983/*
1984 * We use the lower bits of the mem_map pointer to store
1985 * a little bit of information. The pointer is calculated
1986 * as mem_map - section_nr_to_pfn(pnum). The result is
1987 * aligned to the minimum alignment of the two values:
1988 * 1. All mem_map arrays are page-aligned.
1989 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1990 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1991 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1992 * worst combination is powerpc with 256k pages,
1993 * which results in PFN_SECTION_SHIFT equal 6.
1994 * To sum it up, at least 6 bits are available on all architectures.
1995 * However, we can exceed 6 bits on some other architectures except
1996 * powerpc (e.g. 15 bits are available on x86_64, 13 bits are available
1997 * with the worst case of 64K pages on arm64) if we make sure the
1998 * exceeded bit is not applicable to powerpc.
1999 */
2000enum {
2001 SECTION_MARKED_PRESENT_BIT,
2002 SECTION_HAS_MEM_MAP_BIT,
2003 SECTION_IS_ONLINE_BIT,
2004 SECTION_IS_EARLY_BIT,
2005#ifdef CONFIG_ZONE_DEVICE
2006 SECTION_TAINT_ZONE_DEVICE_BIT,
2007#endif
2008#ifdef CONFIG_SPARSEMEM_VMEMMAP_PREINIT
2009 SECTION_IS_VMEMMAP_PREINIT_BIT,
2010#endif
2011 SECTION_MAP_LAST_BIT,
2012};
2013
2014#define SECTION_MARKED_PRESENT BIT(SECTION_MARKED_PRESENT_BIT)
2015#define SECTION_HAS_MEM_MAP BIT(SECTION_HAS_MEM_MAP_BIT)
2016#define SECTION_IS_ONLINE BIT(SECTION_IS_ONLINE_BIT)
2017#define SECTION_IS_EARLY BIT(SECTION_IS_EARLY_BIT)
2018#ifdef CONFIG_ZONE_DEVICE
2019#define SECTION_TAINT_ZONE_DEVICE BIT(SECTION_TAINT_ZONE_DEVICE_BIT)
2020#endif
2021#ifdef CONFIG_SPARSEMEM_VMEMMAP_PREINIT
2022#define SECTION_IS_VMEMMAP_PREINIT BIT(SECTION_IS_VMEMMAP_PREINIT_BIT)
2023#endif
2024#define SECTION_MAP_MASK (~(BIT(SECTION_MAP_LAST_BIT) - 1))
2025#define SECTION_NID_SHIFT SECTION_MAP_LAST_BIT
2026
2027static inline struct page *__section_mem_map_addr(struct mem_section *section)
2028{
2029 unsigned long map = section->section_mem_map;
2030 map &= SECTION_MAP_MASK;
2031 return (struct page *)map;
2032}
2033
2034static inline int present_section(const struct mem_section *section)
2035{
2036 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
2037}
2038
2039static inline int present_section_nr(unsigned long nr)
2040{
2041 return present_section(__nr_to_section(nr));
2042}
2043
2044static inline int valid_section(const struct mem_section *section)
2045{
2046 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
2047}
2048
2049static inline int early_section(const struct mem_section *section)
2050{
2051 return (section && (section->section_mem_map & SECTION_IS_EARLY));
2052}
2053
2054static inline int valid_section_nr(unsigned long nr)
2055{
2056 return valid_section(__nr_to_section(nr));
2057}
2058
2059static inline int online_section(const struct mem_section *section)
2060{
2061 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
2062}
2063
2064#ifdef CONFIG_ZONE_DEVICE
2065static inline int online_device_section(const struct mem_section *section)
2066{
2067 unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE;
2068
2069 return section && ((section->section_mem_map & flags) == flags);
2070}
2071#else
2072static inline int online_device_section(const struct mem_section *section)
2073{
2074 return 0;
2075}
2076#endif
2077
2078#ifdef CONFIG_SPARSEMEM_VMEMMAP_PREINIT
2079static inline int preinited_vmemmap_section(const struct mem_section *section)
2080{
2081 return (section &&
2082 (section->section_mem_map & SECTION_IS_VMEMMAP_PREINIT));
2083}
2084
2085void sparse_vmemmap_init_nid_early(int nid);
2086void sparse_vmemmap_init_nid_late(int nid);
2087
2088#else
2089static inline int preinited_vmemmap_section(const struct mem_section *section)
2090{
2091 return 0;
2092}
2093static inline void sparse_vmemmap_init_nid_early(int nid)
2094{
2095}
2096
2097static inline void sparse_vmemmap_init_nid_late(int nid)
2098{
2099}
2100#endif
2101
2102static inline int online_section_nr(unsigned long nr)
2103{
2104 return online_section(__nr_to_section(nr));
2105}
2106
2107#ifdef CONFIG_MEMORY_HOTPLUG
2108void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
2109void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
2110#endif
2111
2112static inline struct mem_section *__pfn_to_section(unsigned long pfn)
2113{
2114 return __nr_to_section(pfn_to_section_nr(pfn));
2115}
2116
2117extern unsigned long __highest_present_section_nr;
2118
2119static inline int subsection_map_index(unsigned long pfn)
2120{
2121 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
2122}
2123
2124#ifdef CONFIG_SPARSEMEM_VMEMMAP
2125static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
2126{
2127 int idx = subsection_map_index(pfn);
2128 struct mem_section_usage *usage = READ_ONCE(ms->usage);
2129
2130 return usage ? test_bit(idx, usage->subsection_map) : 0;
2131}
2132
2133static inline bool pfn_section_first_valid(struct mem_section *ms, unsigned long *pfn)
2134{
2135 struct mem_section_usage *usage = READ_ONCE(ms->usage);
2136 int idx = subsection_map_index(*pfn);
2137 unsigned long bit;
2138
2139 if (!usage)
2140 return false;
2141
2142 if (test_bit(idx, usage->subsection_map))
2143 return true;
2144
2145 /* Find the next subsection that exists */
2146 bit = find_next_bit(usage->subsection_map, SUBSECTIONS_PER_SECTION, idx);
2147 if (bit == SUBSECTIONS_PER_SECTION)
2148 return false;
2149
2150 *pfn = (*pfn & PAGE_SECTION_MASK) + (bit * PAGES_PER_SUBSECTION);
2151 return true;
2152}
2153#else
2154static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
2155{
2156 return 1;
2157}
2158
2159static inline bool pfn_section_first_valid(struct mem_section *ms, unsigned long *pfn)
2160{
2161 return true;
2162}
2163#endif
2164
2165void sparse_init_early_section(int nid, struct page *map, unsigned long pnum,
2166 unsigned long flags);
2167
2168#ifndef CONFIG_HAVE_ARCH_PFN_VALID
2169/**
2170 * pfn_valid - check if there is a valid memory map entry for a PFN
2171 * @pfn: the page frame number to check
2172 *
2173 * Check if there is a valid memory map entry aka struct page for the @pfn.
2174 * Note, that availability of the memory map entry does not imply that
2175 * there is actual usable memory at that @pfn. The struct page may
2176 * represent a hole or an unusable page frame.
2177 *
2178 * Return: 1 for PFNs that have memory map entries and 0 otherwise
2179 */
2180static inline int pfn_valid(unsigned long pfn)
2181{
2182 struct mem_section *ms;
2183 int ret;
2184
2185 /*
2186 * Ensure the upper PAGE_SHIFT bits are clear in the
2187 * pfn. Else it might lead to false positives when
2188 * some of the upper bits are set, but the lower bits
2189 * match a valid pfn.
2190 */
2191 if (PHYS_PFN(PFN_PHYS(pfn)) != pfn)
2192 return 0;
2193
2194 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2195 return 0;
2196 ms = __pfn_to_section(pfn);
2197 rcu_read_lock_sched();
2198 if (!valid_section(ms)) {
2199 rcu_read_unlock_sched();
2200 return 0;
2201 }
2202 /*
2203 * Traditionally early sections always returned pfn_valid() for
2204 * the entire section-sized span.
2205 */
2206 ret = early_section(ms) || pfn_section_valid(ms, pfn);
2207 rcu_read_unlock_sched();
2208
2209 return ret;
2210}
2211
2212/* Returns end_pfn or higher if no valid PFN remaining in range */
2213static inline unsigned long first_valid_pfn(unsigned long pfn, unsigned long end_pfn)
2214{
2215 unsigned long nr = pfn_to_section_nr(pfn);
2216
2217 rcu_read_lock_sched();
2218
2219 while (nr <= __highest_present_section_nr && pfn < end_pfn) {
2220 struct mem_section *ms = __pfn_to_section(pfn);
2221
2222 if (valid_section(ms) &&
2223 (early_section(ms) || pfn_section_first_valid(ms, &pfn))) {
2224 rcu_read_unlock_sched();
2225 return pfn;
2226 }
2227
2228 /* Nothing left in this section? Skip to next section */
2229 nr++;
2230 pfn = section_nr_to_pfn(nr);
2231 }
2232
2233 rcu_read_unlock_sched();
2234 return end_pfn;
2235}
2236
2237static inline unsigned long next_valid_pfn(unsigned long pfn, unsigned long end_pfn)
2238{
2239 pfn++;
2240
2241 if (pfn >= end_pfn)
2242 return end_pfn;
2243
2244 /*
2245 * Either every PFN within the section (or subsection for VMEMMAP) is
2246 * valid, or none of them are. So there's no point repeating the check
2247 * for every PFN; only call first_valid_pfn() again when crossing a
2248 * (sub)section boundary (i.e. !(pfn & ~PAGE_{SUB,}SECTION_MASK)).
2249 */
2250 if (pfn & ~(IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP) ?
2251 PAGE_SUBSECTION_MASK : PAGE_SECTION_MASK))
2252 return pfn;
2253
2254 return first_valid_pfn(pfn, end_pfn);
2255}
2256
2257
2258#define for_each_valid_pfn(_pfn, _start_pfn, _end_pfn) \
2259 for ((_pfn) = first_valid_pfn((_start_pfn), (_end_pfn)); \
2260 (_pfn) < (_end_pfn); \
2261 (_pfn) = next_valid_pfn((_pfn), (_end_pfn)))
2262
2263#endif
2264
2265static inline int pfn_in_present_section(unsigned long pfn)
2266{
2267 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2268 return 0;
2269 return present_section(__pfn_to_section(pfn));
2270}
2271
2272static inline unsigned long next_present_section_nr(unsigned long section_nr)
2273{
2274 while (++section_nr <= __highest_present_section_nr) {
2275 if (present_section_nr(section_nr))
2276 return section_nr;
2277 }
2278
2279 return -1;
2280}
2281
2282#define for_each_present_section_nr(start, section_nr) \
2283 for (section_nr = next_present_section_nr(start - 1); \
2284 section_nr != -1; \
2285 section_nr = next_present_section_nr(section_nr))
2286
2287/*
2288 * These are _only_ used during initialisation, therefore they
2289 * can use __initdata ... They could have names to indicate
2290 * this restriction.
2291 */
2292#ifdef CONFIG_NUMA
2293#define pfn_to_nid(pfn) \
2294({ \
2295 unsigned long __pfn_to_nid_pfn = (pfn); \
2296 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
2297})
2298#else
2299#define pfn_to_nid(pfn) (0)
2300#endif
2301
2302#else
2303#define sparse_index_init(_sec, _nid) do {} while (0)
2304#define sparse_vmemmap_init_nid_early(_nid) do {} while (0)
2305#define sparse_vmemmap_init_nid_late(_nid) do {} while (0)
2306#define pfn_in_present_section pfn_valid
2307#define subsection_map_init(_pfn, _nr_pages) do {} while (0)
2308#endif /* CONFIG_SPARSEMEM */
2309
2310/*
2311 * Fallback case for when the architecture provides its own pfn_valid() but
2312 * not a corresponding for_each_valid_pfn().
2313 */
2314#ifndef for_each_valid_pfn
2315#define for_each_valid_pfn(_pfn, _start_pfn, _end_pfn) \
2316 for ((_pfn) = (_start_pfn); (_pfn) < (_end_pfn); (_pfn)++) \
2317 if (pfn_valid(_pfn))
2318#endif
2319
2320#endif /* !__GENERATING_BOUNDS.H */
2321#endif /* !__ASSEMBLY__ */
2322#endif /* _LINUX_MMZONE_H */