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 <linux/sizes.h>
27#include <asm/page.h>
28
29/* Free memory management - zoned buddy allocator. */
30#ifndef CONFIG_ARCH_FORCE_MAX_ORDER
31#define MAX_PAGE_ORDER 10
32#else
33#define MAX_PAGE_ORDER CONFIG_ARCH_FORCE_MAX_ORDER
34#endif
35#define MAX_ORDER_NR_PAGES (1 << MAX_PAGE_ORDER)
36
37#define IS_MAX_ORDER_ALIGNED(pfn) IS_ALIGNED(pfn, MAX_ORDER_NR_PAGES)
38
39#define NR_PAGE_ORDERS (MAX_PAGE_ORDER + 1)
40
41/* Defines the order for the number of pages that have a migrate type. */
42#ifndef CONFIG_PAGE_BLOCK_MAX_ORDER
43#define PAGE_BLOCK_MAX_ORDER MAX_PAGE_ORDER
44#else
45#define PAGE_BLOCK_MAX_ORDER CONFIG_PAGE_BLOCK_MAX_ORDER
46#endif /* CONFIG_PAGE_BLOCK_MAX_ORDER */
47
48/*
49 * The MAX_PAGE_ORDER, which defines the max order of pages to be allocated
50 * by the buddy allocator, has to be larger or equal to the PAGE_BLOCK_MAX_ORDER,
51 * which defines the order for the number of pages that can have a migrate type
52 */
53#if (PAGE_BLOCK_MAX_ORDER > MAX_PAGE_ORDER)
54#error MAX_PAGE_ORDER must be >= PAGE_BLOCK_MAX_ORDER
55#endif
56
57/*
58 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
59 * costly to service. That is between allocation orders which should
60 * coalesce naturally under reasonable reclaim pressure and those which
61 * will not.
62 */
63#define PAGE_ALLOC_COSTLY_ORDER 3
64
65#if !defined(CONFIG_HAVE_GIGANTIC_FOLIOS)
66/*
67 * We don't expect any folios that exceed buddy sizes (and consequently
68 * memory sections).
69 */
70#define MAX_FOLIO_ORDER MAX_PAGE_ORDER
71#elif defined(CONFIG_SPARSEMEM) && !defined(CONFIG_SPARSEMEM_VMEMMAP)
72/*
73 * Only pages within a single memory section are guaranteed to be
74 * contiguous. By limiting folios to a single memory section, all folio
75 * pages are guaranteed to be contiguous.
76 */
77#define MAX_FOLIO_ORDER PFN_SECTION_SHIFT
78#elif defined(CONFIG_HUGETLB_PAGE)
79/*
80 * There is no real limit on the folio size. We limit them to the maximum we
81 * currently expect (see CONFIG_HAVE_GIGANTIC_FOLIOS): with hugetlb, we expect
82 * no folios larger than 16 GiB on 64bit and 1 GiB on 32bit.
83 */
84#ifdef CONFIG_64BIT
85#define MAX_FOLIO_ORDER (ilog2(SZ_16G) - PAGE_SHIFT)
86#else
87#define MAX_FOLIO_ORDER (ilog2(SZ_1G) - PAGE_SHIFT)
88#endif
89#else
90/*
91 * Without hugetlb, gigantic folios that are bigger than a single PUD are
92 * currently impossible.
93 */
94#define MAX_FOLIO_ORDER (PUD_SHIFT - PAGE_SHIFT)
95#endif
96
97#define MAX_FOLIO_NR_PAGES (1UL << MAX_FOLIO_ORDER)
98
99/*
100 * HugeTLB Vmemmap Optimization (HVO) requires struct pages of the head page to
101 * be naturally aligned with regard to the folio size.
102 *
103 * HVO which is only active if the size of struct page is a power of 2.
104 */
105#define MAX_FOLIO_VMEMMAP_ALIGN \
106 (IS_ENABLED(CONFIG_HUGETLB_PAGE_OPTIMIZE_VMEMMAP) && \
107 is_power_of_2(sizeof(struct page)) ? \
108 MAX_FOLIO_NR_PAGES * sizeof(struct page) : 0)
109
110/*
111 * vmemmap optimization (like HVO) is only possible for page orders that fill
112 * two or more pages with struct pages.
113 */
114#define VMEMMAP_TAIL_MIN_ORDER (ilog2(2 * PAGE_SIZE / sizeof(struct page)))
115#define __NR_VMEMMAP_TAILS (MAX_FOLIO_ORDER - VMEMMAP_TAIL_MIN_ORDER + 1)
116#define NR_VMEMMAP_TAILS (__NR_VMEMMAP_TAILS > 0 ? __NR_VMEMMAP_TAILS : 0)
117
118enum migratetype {
119 MIGRATE_UNMOVABLE,
120 MIGRATE_MOVABLE,
121 MIGRATE_RECLAIMABLE,
122 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
123 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
124#ifdef CONFIG_CMA
125 /*
126 * MIGRATE_CMA migration type is designed to mimic the way
127 * ZONE_MOVABLE works. Only movable pages can be allocated
128 * from MIGRATE_CMA pageblocks and page allocator never
129 * implicitly change migration type of MIGRATE_CMA pageblock.
130 *
131 * The way to use it is to change migratetype of a range of
132 * pageblocks to MIGRATE_CMA which can be done by
133 * __free_pageblock_cma() function.
134 */
135 MIGRATE_CMA,
136 __MIGRATE_TYPE_END = MIGRATE_CMA,
137#else
138 __MIGRATE_TYPE_END = MIGRATE_HIGHATOMIC,
139#endif
140#ifdef CONFIG_MEMORY_ISOLATION
141 MIGRATE_ISOLATE, /* can't allocate from here */
142#endif
143 MIGRATE_TYPES
144};
145
146/* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
147extern const char * const migratetype_names[MIGRATE_TYPES];
148
149#ifdef CONFIG_CMA
150# define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
151# define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
152/*
153 * __dump_folio() in mm/debug.c passes a folio pointer to on-stack struct folio,
154 * so folio_pfn() cannot be used and pfn is needed.
155 */
156# define is_migrate_cma_folio(folio, pfn) \
157 (get_pfnblock_migratetype(&folio->page, pfn) == MIGRATE_CMA)
158#else
159# define is_migrate_cma(migratetype) false
160# define is_migrate_cma_page(_page) false
161# define is_migrate_cma_folio(folio, pfn) false
162#endif
163
164static inline bool is_migrate_movable(int mt)
165{
166 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
167}
168
169/*
170 * Check whether a migratetype can be merged with another migratetype.
171 *
172 * It is only mergeable when it can fall back to other migratetypes for
173 * allocation. See fallbacks[MIGRATE_TYPES][3] in page_alloc.c.
174 */
175static inline bool migratetype_is_mergeable(int mt)
176{
177 return mt < MIGRATE_PCPTYPES;
178}
179
180#define for_each_migratetype_order(order, type) \
181 for (order = 0; order < NR_PAGE_ORDERS; order++) \
182 for (type = 0; type < MIGRATE_TYPES; type++)
183
184extern int page_group_by_mobility_disabled;
185
186#define get_pageblock_migratetype(page) \
187 get_pfnblock_migratetype(page, page_to_pfn(page))
188
189#define folio_migratetype(folio) \
190 get_pageblock_migratetype(&folio->page)
191
192struct free_area {
193 struct list_head free_list[MIGRATE_TYPES];
194 unsigned long nr_free;
195};
196
197struct pglist_data;
198
199#ifdef CONFIG_NUMA
200enum numa_stat_item {
201 NUMA_HIT, /* allocated in intended node */
202 NUMA_MISS, /* allocated in non intended node */
203 NUMA_FOREIGN, /* was intended here, hit elsewhere */
204 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
205 NUMA_LOCAL, /* allocation from local node */
206 NUMA_OTHER, /* allocation from other node */
207 NR_VM_NUMA_EVENT_ITEMS
208};
209#else
210#define NR_VM_NUMA_EVENT_ITEMS 0
211#endif
212
213enum zone_stat_item {
214 /* First 128 byte cacheline (assuming 64 bit words) */
215 NR_FREE_PAGES,
216 NR_FREE_PAGES_BLOCKS,
217 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
218 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
219 NR_ZONE_ACTIVE_ANON,
220 NR_ZONE_INACTIVE_FILE,
221 NR_ZONE_ACTIVE_FILE,
222 NR_ZONE_UNEVICTABLE,
223 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
224 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
225 /* Second 128 byte cacheline */
226#if IS_ENABLED(CONFIG_ZSMALLOC)
227 NR_ZSPAGES, /* allocated in zsmalloc */
228#endif
229 NR_FREE_CMA_PAGES,
230#ifdef CONFIG_UNACCEPTED_MEMORY
231 NR_UNACCEPTED,
232#endif
233 NR_VM_ZONE_STAT_ITEMS };
234
235enum node_stat_item {
236 NR_LRU_BASE,
237 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
238 NR_ACTIVE_ANON, /* " " " " " */
239 NR_INACTIVE_FILE, /* " " " " " */
240 NR_ACTIVE_FILE, /* " " " " " */
241 NR_UNEVICTABLE, /* " " " " " */
242 NR_SLAB_RECLAIMABLE_B,
243 NR_SLAB_UNRECLAIMABLE_B,
244 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
245 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
246 WORKINGSET_NODES,
247 WORKINGSET_REFAULT_BASE,
248 WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE,
249 WORKINGSET_REFAULT_FILE,
250 WORKINGSET_ACTIVATE_BASE,
251 WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE,
252 WORKINGSET_ACTIVATE_FILE,
253 WORKINGSET_RESTORE_BASE,
254 WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE,
255 WORKINGSET_RESTORE_FILE,
256 WORKINGSET_NODERECLAIM,
257 NR_ANON_MAPPED, /* Mapped anonymous pages */
258 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
259 only modified from process context */
260 NR_FILE_PAGES,
261 NR_FILE_DIRTY,
262 NR_WRITEBACK,
263 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
264 NR_SHMEM_THPS,
265 NR_SHMEM_PMDMAPPED,
266 NR_FILE_THPS,
267 NR_FILE_PMDMAPPED,
268 NR_ANON_THPS,
269 NR_VMSCAN_WRITE,
270 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
271 NR_DIRTIED, /* page dirtyings since bootup */
272 NR_WRITTEN, /* page writings since bootup */
273 NR_THROTTLED_WRITTEN, /* NR_WRITTEN while reclaim throttled */
274 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
275 NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */
276 NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */
277 NR_VMALLOC,
278 NR_KERNEL_STACK_KB, /* measured in KiB */
279#if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
280 NR_KERNEL_SCS_KB, /* measured in KiB */
281#endif
282 NR_PAGETABLE, /* used for pagetables */
283 NR_SECONDARY_PAGETABLE, /* secondary pagetables, KVM & IOMMU */
284#ifdef CONFIG_IOMMU_SUPPORT
285 NR_IOMMU_PAGES, /* # of pages allocated by IOMMU */
286#endif
287#ifdef CONFIG_SWAP
288 NR_SWAPCACHE,
289#endif
290#ifdef CONFIG_NUMA_BALANCING
291 PGPROMOTE_SUCCESS, /* promote successfully */
292 /**
293 * Candidate pages for promotion based on hint fault latency. This
294 * counter is used to control the promotion rate and adjust the hot
295 * threshold.
296 */
297 PGPROMOTE_CANDIDATE,
298 /**
299 * Not rate-limited (NRL) candidate pages for those can be promoted
300 * without considering hot threshold because of enough free pages in
301 * fast-tier node. These promotions bypass the regular hotness checks
302 * and do NOT influence the promotion rate-limiter or
303 * threshold-adjustment logic.
304 * This is for statistics/monitoring purposes.
305 */
306 PGPROMOTE_CANDIDATE_NRL,
307#endif
308 /* PGDEMOTE_*: pages demoted */
309 PGDEMOTE_KSWAPD,
310 PGDEMOTE_DIRECT,
311 PGDEMOTE_KHUGEPAGED,
312 PGDEMOTE_PROACTIVE,
313 PGSTEAL_KSWAPD,
314 PGSTEAL_DIRECT,
315 PGSTEAL_KHUGEPAGED,
316 PGSTEAL_PROACTIVE,
317 PGSTEAL_ANON,
318 PGSTEAL_FILE,
319 PGSCAN_KSWAPD,
320 PGSCAN_DIRECT,
321 PGSCAN_KHUGEPAGED,
322 PGSCAN_PROACTIVE,
323 PGSCAN_ANON,
324 PGSCAN_FILE,
325 PGREFILL,
326#ifdef CONFIG_HUGETLB_PAGE
327 NR_HUGETLB,
328#endif
329 NR_BALLOON_PAGES,
330 NR_KERNEL_FILE_PAGES,
331 NR_GPU_ACTIVE, /* Pages assigned to GPU objects */
332 NR_GPU_RECLAIM, /* Pages in shrinkable GPU pools */
333 NR_VM_NODE_STAT_ITEMS
334};
335
336/*
337 * Returns true if the item should be printed in THPs (/proc/vmstat
338 * currently prints number of anon, file and shmem THPs. But the item
339 * is charged in pages).
340 */
341static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item)
342{
343 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
344 return false;
345
346 return item == NR_ANON_THPS ||
347 item == NR_FILE_THPS ||
348 item == NR_SHMEM_THPS ||
349 item == NR_SHMEM_PMDMAPPED ||
350 item == NR_FILE_PMDMAPPED;
351}
352
353/*
354 * Returns true if the value is measured in bytes (most vmstat values are
355 * measured in pages). This defines the API part, the internal representation
356 * might be different.
357 */
358static __always_inline bool vmstat_item_in_bytes(int idx)
359{
360 /*
361 * Global and per-node slab counters track slab pages.
362 * It's expected that changes are multiples of PAGE_SIZE.
363 * Internally values are stored in pages.
364 *
365 * Per-memcg and per-lruvec counters track memory, consumed
366 * by individual slab objects. These counters are actually
367 * byte-precise.
368 */
369 return (idx == NR_SLAB_RECLAIMABLE_B ||
370 idx == NR_SLAB_UNRECLAIMABLE_B);
371}
372
373/*
374 * We do arithmetic on the LRU lists in various places in the code,
375 * so it is important to keep the active lists LRU_ACTIVE higher in
376 * the array than the corresponding inactive lists, and to keep
377 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
378 *
379 * This has to be kept in sync with the statistics in zone_stat_item
380 * above and the descriptions in vmstat_text in mm/vmstat.c
381 */
382#define LRU_BASE 0
383#define LRU_ACTIVE 1
384#define LRU_FILE 2
385
386enum lru_list {
387 LRU_INACTIVE_ANON = LRU_BASE,
388 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
389 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
390 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
391 LRU_UNEVICTABLE,
392 NR_LRU_LISTS
393};
394
395enum vmscan_throttle_state {
396 VMSCAN_THROTTLE_WRITEBACK,
397 VMSCAN_THROTTLE_ISOLATED,
398 VMSCAN_THROTTLE_NOPROGRESS,
399 VMSCAN_THROTTLE_CONGESTED,
400 NR_VMSCAN_THROTTLE,
401};
402
403#define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
404
405#define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
406
407static inline bool is_file_lru(enum lru_list lru)
408{
409 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
410}
411
412static inline bool is_active_lru(enum lru_list lru)
413{
414 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
415}
416
417#define WORKINGSET_ANON 0
418#define WORKINGSET_FILE 1
419#define ANON_AND_FILE 2
420
421enum lruvec_flags {
422 /*
423 * An lruvec has many dirty pages backed by a congested BDI:
424 * 1. LRUVEC_CGROUP_CONGESTED is set by cgroup-level reclaim.
425 * It can be cleared by cgroup reclaim or kswapd.
426 * 2. LRUVEC_NODE_CONGESTED is set by kswapd node-level reclaim.
427 * It can only be cleared by kswapd.
428 *
429 * Essentially, kswapd can unthrottle an lruvec throttled by cgroup
430 * reclaim, but not vice versa. This only applies to the root cgroup.
431 * The goal is to prevent cgroup reclaim on the root cgroup (e.g.
432 * memory.reclaim) to unthrottle an unbalanced node (that was throttled
433 * by kswapd).
434 */
435 LRUVEC_CGROUP_CONGESTED,
436 LRUVEC_NODE_CONGESTED,
437};
438
439#endif /* !__GENERATING_BOUNDS_H */
440
441/*
442 * Evictable folios are divided into multiple generations. The youngest and the
443 * oldest generation numbers, max_seq and min_seq, are monotonically increasing.
444 * They form a sliding window of a variable size [MIN_NR_GENS, MAX_NR_GENS]. An
445 * offset within MAX_NR_GENS, i.e., gen, indexes the LRU list of the
446 * corresponding generation. The gen counter in folio->flags stores gen+1 while
447 * a folio is on one of lrugen->folios[]. Otherwise it stores 0.
448 *
449 * After a folio is faulted in, the aging needs to check the accessed bit at
450 * least twice before handing this folio over to the eviction. The first check
451 * clears the accessed bit from the initial fault; the second check makes sure
452 * this folio hasn't been used since then. This process, AKA second chance,
453 * requires a minimum of two generations, hence MIN_NR_GENS. And to maintain ABI
454 * compatibility with the active/inactive LRU, e.g., /proc/vmstat, these two
455 * generations are considered active; the rest of generations, if they exist,
456 * are considered inactive. See lru_gen_is_active().
457 *
458 * PG_active is always cleared while a folio is on one of lrugen->folios[] so
459 * that the sliding window needs not to worry about it. And it's set again when
460 * a folio considered active is isolated for non-reclaiming purposes, e.g.,
461 * migration. See lru_gen_add_folio() and lru_gen_del_folio().
462 *
463 * MAX_NR_GENS is set to 4 so that the multi-gen LRU can support twice the
464 * number of categories of the active/inactive LRU when keeping track of
465 * accesses through page tables. This requires order_base_2(MAX_NR_GENS+1) bits
466 * in folio->flags, masked by LRU_GEN_MASK.
467 */
468#define MIN_NR_GENS 2U
469#define MAX_NR_GENS 4U
470
471/*
472 * Each generation is divided into multiple tiers. A folio accessed N times
473 * through file descriptors is in tier order_base_2(N). A folio in the first
474 * tier (N=0,1) is marked by PG_referenced unless it was faulted in through page
475 * tables or read ahead. A folio in the last tier (MAX_NR_TIERS-1) is marked by
476 * PG_workingset. A folio in any other tier (1<N<5) between the first and last
477 * is marked by additional bits of LRU_REFS_WIDTH in folio->flags.
478 *
479 * In contrast to moving across generations which requires the LRU lock, moving
480 * across tiers only involves atomic operations on folio->flags and therefore
481 * has a negligible cost in the buffered access path. In the eviction path,
482 * comparisons of refaulted/(evicted+protected) from the first tier and the rest
483 * infer whether folios accessed multiple times through file descriptors are
484 * statistically hot and thus worth protecting.
485 *
486 * MAX_NR_TIERS is set to 4 so that the multi-gen LRU can support twice the
487 * number of categories of the active/inactive LRU when keeping track of
488 * accesses through file descriptors. This uses MAX_NR_TIERS-2 spare bits in
489 * folio->flags, masked by LRU_REFS_MASK.
490 */
491#define MAX_NR_TIERS 4U
492
493#ifndef __GENERATING_BOUNDS_H
494
495#define LRU_GEN_MASK ((BIT(LRU_GEN_WIDTH) - 1) << LRU_GEN_PGOFF)
496#define LRU_REFS_MASK ((BIT(LRU_REFS_WIDTH) - 1) << LRU_REFS_PGOFF)
497
498/*
499 * For folios accessed multiple times through file descriptors,
500 * lru_gen_inc_refs() sets additional bits of LRU_REFS_WIDTH in folio->flags
501 * after PG_referenced, then PG_workingset after LRU_REFS_WIDTH. After all its
502 * bits are set, i.e., LRU_REFS_FLAGS|BIT(PG_workingset), a folio is lazily
503 * promoted into the second oldest generation in the eviction path. And when
504 * folio_inc_gen() does that, it clears LRU_REFS_FLAGS so that
505 * lru_gen_inc_refs() can start over. Note that for this case, LRU_REFS_MASK is
506 * only valid when PG_referenced is set.
507 *
508 * For folios accessed multiple times through page tables, folio_update_gen()
509 * from a page table walk or lru_gen_set_refs() from a rmap walk sets
510 * PG_referenced after the accessed bit is cleared for the first time.
511 * Thereafter, those two paths set PG_workingset and promote folios to the
512 * youngest generation. Like folio_inc_gen(), folio_update_gen() also clears
513 * PG_referenced. Note that for this case, LRU_REFS_MASK is not used.
514 *
515 * For both cases above, after PG_workingset is set on a folio, it remains until
516 * this folio is either reclaimed, or "deactivated" by lru_gen_clear_refs(). It
517 * can be set again if lru_gen_test_recent() returns true upon a refault.
518 */
519#define LRU_REFS_FLAGS (LRU_REFS_MASK | BIT(PG_referenced))
520
521struct lruvec;
522struct page_vma_mapped_walk;
523
524#ifdef CONFIG_LRU_GEN
525
526enum {
527 LRU_GEN_ANON,
528 LRU_GEN_FILE,
529};
530
531enum {
532 LRU_GEN_CORE,
533 LRU_GEN_MM_WALK,
534 LRU_GEN_NONLEAF_YOUNG,
535 NR_LRU_GEN_CAPS
536};
537
538#define MIN_LRU_BATCH BITS_PER_LONG
539#define MAX_LRU_BATCH (MIN_LRU_BATCH * 64)
540
541/* whether to keep historical stats from evicted generations */
542#ifdef CONFIG_LRU_GEN_STATS
543#define NR_HIST_GENS MAX_NR_GENS
544#else
545#define NR_HIST_GENS 1U
546#endif
547
548/*
549 * The youngest generation number is stored in max_seq for both anon and file
550 * types as they are aged on an equal footing. The oldest generation numbers are
551 * stored in min_seq[] separately for anon and file types so that they can be
552 * incremented independently. Ideally min_seq[] are kept in sync when both anon
553 * and file types are evictable. However, to adapt to situations like extreme
554 * swappiness, they are allowed to be out of sync by at most
555 * MAX_NR_GENS-MIN_NR_GENS-1.
556 *
557 * The number of pages in each generation is eventually consistent and therefore
558 * can be transiently negative when reset_batch_size() is pending.
559 */
560struct lru_gen_folio {
561 /* the aging increments the youngest generation number */
562 unsigned long max_seq;
563 /* the eviction increments the oldest generation numbers */
564 unsigned long min_seq[ANON_AND_FILE];
565 /* the birth time of each generation in jiffies */
566 unsigned long timestamps[MAX_NR_GENS];
567 /* the multi-gen LRU lists, lazily sorted on eviction */
568 struct list_head folios[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
569 /* the multi-gen LRU sizes, eventually consistent */
570 long nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
571 /* the exponential moving average of refaulted */
572 unsigned long avg_refaulted[ANON_AND_FILE][MAX_NR_TIERS];
573 /* the exponential moving average of evicted+protected */
574 unsigned long avg_total[ANON_AND_FILE][MAX_NR_TIERS];
575 /* can only be modified under the LRU lock */
576 unsigned long protected[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
577 /* can be modified without holding the LRU lock */
578 atomic_long_t evicted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
579 atomic_long_t refaulted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS];
580 /* whether the multi-gen LRU is enabled */
581 bool enabled;
582 /* the memcg generation this lru_gen_folio belongs to */
583 u8 gen;
584 /* the list segment this lru_gen_folio belongs to */
585 u8 seg;
586 /* per-node lru_gen_folio list for global reclaim */
587 struct hlist_nulls_node list;
588};
589
590enum {
591 MM_LEAF_TOTAL, /* total leaf entries */
592 MM_LEAF_YOUNG, /* young leaf entries */
593 MM_NONLEAF_FOUND, /* non-leaf entries found in Bloom filters */
594 MM_NONLEAF_ADDED, /* non-leaf entries added to Bloom filters */
595 NR_MM_STATS
596};
597
598/* double-buffering Bloom filters */
599#define NR_BLOOM_FILTERS 2
600
601struct lru_gen_mm_state {
602 /* synced with max_seq after each iteration */
603 unsigned long seq;
604 /* where the current iteration continues after */
605 struct list_head *head;
606 /* where the last iteration ended before */
607 struct list_head *tail;
608 /* Bloom filters flip after each iteration */
609 unsigned long *filters[NR_BLOOM_FILTERS];
610 /* the mm stats for debugging */
611 unsigned long stats[NR_HIST_GENS][NR_MM_STATS];
612};
613
614struct lru_gen_mm_walk {
615 /* the lruvec under reclaim */
616 struct lruvec *lruvec;
617 /* max_seq from lru_gen_folio: can be out of date */
618 unsigned long seq;
619 /* the next address within an mm to scan */
620 unsigned long next_addr;
621 /* to batch promoted pages */
622 int nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
623 /* to batch the mm stats */
624 int mm_stats[NR_MM_STATS];
625 /* total batched items */
626 int batched;
627 int swappiness;
628 bool force_scan;
629};
630
631/*
632 * For each node, memcgs are divided into two generations: the old and the
633 * young. For each generation, memcgs are randomly sharded into multiple bins
634 * to improve scalability. For each bin, the hlist_nulls is virtually divided
635 * into three segments: the head, the tail and the default.
636 *
637 * An onlining memcg is added to the tail of a random bin in the old generation.
638 * The eviction starts at the head of a random bin in the old generation. The
639 * per-node memcg generation counter, whose reminder (mod MEMCG_NR_GENS) indexes
640 * the old generation, is incremented when all its bins become empty.
641 *
642 * There are four operations:
643 * 1. MEMCG_LRU_HEAD, which moves a memcg to the head of a random bin in its
644 * current generation (old or young) and updates its "seg" to "head";
645 * 2. MEMCG_LRU_TAIL, which moves a memcg to the tail of a random bin in its
646 * current generation (old or young) and updates its "seg" to "tail";
647 * 3. MEMCG_LRU_OLD, which moves a memcg to the head of a random bin in the old
648 * generation, updates its "gen" to "old" and resets its "seg" to "default";
649 * 4. MEMCG_LRU_YOUNG, which moves a memcg to the tail of a random bin in the
650 * young generation, updates its "gen" to "young" and resets its "seg" to
651 * "default".
652 *
653 * The events that trigger the above operations are:
654 * 1. Exceeding the soft limit, which triggers MEMCG_LRU_HEAD;
655 * 2. The first attempt to reclaim a memcg below low, which triggers
656 * MEMCG_LRU_TAIL;
657 * 3. The first attempt to reclaim a memcg offlined or below reclaimable size
658 * threshold, which triggers MEMCG_LRU_TAIL;
659 * 4. The second attempt to reclaim a memcg offlined or below reclaimable size
660 * threshold, which triggers MEMCG_LRU_YOUNG;
661 * 5. Attempting to reclaim a memcg below min, which triggers MEMCG_LRU_YOUNG;
662 * 6. Finishing the aging on the eviction path, which triggers MEMCG_LRU_YOUNG;
663 * 7. Offlining a memcg, which triggers MEMCG_LRU_OLD.
664 *
665 * Notes:
666 * 1. Memcg LRU only applies to global reclaim, and the round-robin incrementing
667 * of their max_seq counters ensures the eventual fairness to all eligible
668 * memcgs. For memcg reclaim, it still relies on mem_cgroup_iter().
669 * 2. There are only two valid generations: old (seq) and young (seq+1).
670 * MEMCG_NR_GENS is set to three so that when reading the generation counter
671 * locklessly, a stale value (seq-1) does not wraparound to young.
672 */
673#define MEMCG_NR_GENS 3
674#define MEMCG_NR_BINS 8
675
676struct lru_gen_memcg {
677 /* the per-node memcg generation counter */
678 unsigned long seq;
679 /* each memcg has one lru_gen_folio per node */
680 unsigned long nr_memcgs[MEMCG_NR_GENS];
681 /* per-node lru_gen_folio list for global reclaim */
682 struct hlist_nulls_head fifo[MEMCG_NR_GENS][MEMCG_NR_BINS];
683 /* protects the above */
684 spinlock_t lock;
685};
686
687void lru_gen_init_pgdat(struct pglist_data *pgdat);
688void lru_gen_init_lruvec(struct lruvec *lruvec);
689bool lru_gen_look_around(struct page_vma_mapped_walk *pvmw, unsigned int nr);
690
691void lru_gen_init_memcg(struct mem_cgroup *memcg);
692void lru_gen_exit_memcg(struct mem_cgroup *memcg);
693void lru_gen_online_memcg(struct mem_cgroup *memcg);
694void lru_gen_offline_memcg(struct mem_cgroup *memcg);
695void lru_gen_release_memcg(struct mem_cgroup *memcg);
696void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid);
697void max_lru_gen_memcg(struct mem_cgroup *memcg, int nid);
698bool recheck_lru_gen_max_memcg(struct mem_cgroup *memcg, int nid);
699void lru_gen_reparent_memcg(struct mem_cgroup *memcg, struct mem_cgroup *parent, int nid);
700
701#else /* !CONFIG_LRU_GEN */
702
703static inline void lru_gen_init_pgdat(struct pglist_data *pgdat)
704{
705}
706
707static inline void lru_gen_init_lruvec(struct lruvec *lruvec)
708{
709}
710
711static inline bool lru_gen_look_around(struct page_vma_mapped_walk *pvmw,
712 unsigned int nr)
713{
714 return false;
715}
716
717static inline void lru_gen_init_memcg(struct mem_cgroup *memcg)
718{
719}
720
721static inline void lru_gen_exit_memcg(struct mem_cgroup *memcg)
722{
723}
724
725static inline void lru_gen_online_memcg(struct mem_cgroup *memcg)
726{
727}
728
729static inline void lru_gen_offline_memcg(struct mem_cgroup *memcg)
730{
731}
732
733static inline void lru_gen_release_memcg(struct mem_cgroup *memcg)
734{
735}
736
737static inline void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid)
738{
739}
740
741static inline void max_lru_gen_memcg(struct mem_cgroup *memcg, int nid)
742{
743}
744
745static inline bool recheck_lru_gen_max_memcg(struct mem_cgroup *memcg, int nid)
746{
747 return true;
748}
749
750static inline
751void lru_gen_reparent_memcg(struct mem_cgroup *memcg, struct mem_cgroup *parent, int nid)
752{
753}
754
755#endif /* CONFIG_LRU_GEN */
756
757struct lruvec {
758 struct list_head lists[NR_LRU_LISTS];
759 /* per lruvec lru_lock for memcg */
760 spinlock_t lru_lock;
761 /*
762 * These track the cost of reclaiming one LRU - file or anon -
763 * over the other. As the observed cost of reclaiming one LRU
764 * increases, the reclaim scan balance tips toward the other.
765 */
766 unsigned long anon_cost;
767 unsigned long file_cost;
768 /* Non-resident age, driven by LRU movement */
769 atomic_long_t nonresident_age;
770 /* Refaults at the time of last reclaim cycle */
771 unsigned long refaults[ANON_AND_FILE];
772 /* Various lruvec state flags (enum lruvec_flags) */
773 unsigned long flags;
774#ifdef CONFIG_LRU_GEN
775 /* evictable pages divided into generations */
776 struct lru_gen_folio lrugen;
777#ifdef CONFIG_LRU_GEN_WALKS_MMU
778 /* to concurrently iterate lru_gen_mm_list */
779 struct lru_gen_mm_state mm_state;
780#endif
781#endif /* CONFIG_LRU_GEN */
782#ifdef CONFIG_MEMCG
783 struct pglist_data *pgdat;
784#endif
785 struct zswap_lruvec_state zswap_lruvec_state;
786};
787
788/* Isolate for asynchronous migration */
789#define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
790/* Isolate unevictable pages */
791#define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
792
793/* LRU Isolation modes. */
794typedef unsigned __bitwise isolate_mode_t;
795
796enum zone_watermarks {
797 WMARK_MIN,
798 WMARK_LOW,
799 WMARK_HIGH,
800 WMARK_PROMO,
801 NR_WMARK
802};
803
804/*
805 * One per migratetype for each PAGE_ALLOC_COSTLY_ORDER. Two additional lists
806 * are added for THP. One PCP list is used by GPF_MOVABLE, and the other PCP list
807 * is used by GFP_UNMOVABLE and GFP_RECLAIMABLE.
808 */
809#ifdef CONFIG_TRANSPARENT_HUGEPAGE
810#define NR_PCP_THP 2
811#else
812#define NR_PCP_THP 0
813#endif
814#define NR_LOWORDER_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1))
815#define NR_PCP_LISTS (NR_LOWORDER_PCP_LISTS + NR_PCP_THP)
816
817/*
818 * Flags used in pcp->flags field.
819 *
820 * PCPF_PREV_FREE_HIGH_ORDER: a high-order page is freed in the
821 * previous page freeing. To avoid to drain PCP for an accident
822 * high-order page freeing.
823 *
824 * PCPF_FREE_HIGH_BATCH: preserve "pcp->batch" pages in PCP before
825 * draining PCP for consecutive high-order pages freeing without
826 * allocation if data cache slice of CPU is large enough. To reduce
827 * zone lock contention and keep cache-hot pages reusing.
828 */
829#define PCPF_PREV_FREE_HIGH_ORDER BIT(0)
830#define PCPF_FREE_HIGH_BATCH BIT(1)
831
832struct per_cpu_pages {
833 spinlock_t lock; /* Protects lists field */
834 int count; /* number of pages in the list */
835 int high; /* high watermark, emptying needed */
836 int high_min; /* min high watermark */
837 int high_max; /* max high watermark */
838 int batch; /* chunk size for buddy add/remove */
839 u8 flags; /* protected by pcp->lock */
840 u8 alloc_factor; /* batch scaling factor during allocate */
841#ifdef CONFIG_NUMA
842 u8 expire; /* When 0, remote pagesets are drained */
843#endif
844 short free_count; /* consecutive free count */
845
846 /* Lists of pages, one per migrate type stored on the pcp-lists */
847 struct list_head lists[NR_PCP_LISTS];
848} ____cacheline_aligned_in_smp;
849
850struct per_cpu_zonestat {
851#ifdef CONFIG_SMP
852 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
853 s8 stat_threshold;
854#endif
855#ifdef CONFIG_NUMA
856 /*
857 * Low priority inaccurate counters that are only folded
858 * on demand. Use a large type to avoid the overhead of
859 * folding during refresh_cpu_vm_stats.
860 */
861 unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
862#endif
863};
864
865struct per_cpu_nodestat {
866 s8 stat_threshold;
867 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
868};
869
870#endif /* !__GENERATING_BOUNDS.H */
871
872enum zone_type {
873 /*
874 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
875 * to DMA to all of the addressable memory (ZONE_NORMAL).
876 * On architectures where this area covers the whole 32 bit address
877 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
878 * DMA addressing constraints. This distinction is important as a 32bit
879 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
880 * platforms may need both zones as they support peripherals with
881 * different DMA addressing limitations.
882 */
883#ifdef CONFIG_ZONE_DMA
884 ZONE_DMA,
885#endif
886#ifdef CONFIG_ZONE_DMA32
887 ZONE_DMA32,
888#endif
889 /*
890 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
891 * performed on pages in ZONE_NORMAL if the DMA devices support
892 * transfers to all addressable memory.
893 */
894 ZONE_NORMAL,
895#ifdef CONFIG_HIGHMEM
896 /*
897 * A memory area that is only addressable by the kernel through
898 * mapping portions into its own address space. This is for example
899 * used by i386 to allow the kernel to address the memory beyond
900 * 900MB. The kernel will set up special mappings (page
901 * table entries on i386) for each page that the kernel needs to
902 * access.
903 */
904 ZONE_HIGHMEM,
905#endif
906 /*
907 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
908 * movable pages with few exceptional cases described below. Main use
909 * cases for ZONE_MOVABLE are to make memory offlining/unplug more
910 * likely to succeed, and to locally limit unmovable allocations - e.g.,
911 * to increase the number of THP/huge pages. Notable special cases are:
912 *
913 * 1. Pinned pages: (long-term) pinning of movable pages might
914 * essentially turn such pages unmovable. Therefore, we do not allow
915 * pinning long-term pages in ZONE_MOVABLE. When pages are pinned and
916 * faulted, they come from the right zone right away. However, it is
917 * still possible that address space already has pages in
918 * ZONE_MOVABLE at the time when pages are pinned (i.e. user has
919 * touches that memory before pinning). In such case we migrate them
920 * to a different zone. When migration fails - pinning fails.
921 * 2. memblock allocations: kernelcore/movablecore setups might create
922 * situations where ZONE_MOVABLE contains unmovable allocations
923 * after boot. Memory offlining and allocations fail early.
924 * 3. Memory holes: kernelcore/movablecore setups might create very rare
925 * situations where ZONE_MOVABLE contains memory holes after boot,
926 * for example, if we have sections that are only partially
927 * populated. Memory offlining and allocations fail early.
928 * 4. PG_hwpoison pages: while poisoned pages can be skipped during
929 * memory offlining, such pages cannot be allocated.
930 * 5. Unmovable PG_offline pages: in paravirtualized environments,
931 * hotplugged memory blocks might only partially be managed by the
932 * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
933 * parts not manged by the buddy are unmovable PG_offline pages. In
934 * some cases (virtio-mem), such pages can be skipped during
935 * memory offlining, however, cannot be moved/allocated. These
936 * techniques might use alloc_contig_range() to hide previously
937 * exposed pages from the buddy again (e.g., to implement some sort
938 * of memory unplug in virtio-mem).
939 * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create
940 * situations where ZERO_PAGE(0) which is allocated differently
941 * on different platforms may end up in a movable zone. ZERO_PAGE(0)
942 * cannot be migrated.
943 * 7. Memory-hotplug: when using memmap_on_memory and onlining the
944 * memory to the MOVABLE zone, the vmemmap pages are also placed in
945 * such zone. Such pages cannot be really moved around as they are
946 * self-stored in the range, but they are treated as movable when
947 * the range they describe is about to be offlined.
948 *
949 * In general, no unmovable allocations that degrade memory offlining
950 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
951 * have to expect that migrating pages in ZONE_MOVABLE can fail (even
952 * if has_unmovable_pages() states that there are no unmovable pages,
953 * there can be false negatives).
954 */
955 ZONE_MOVABLE,
956#ifdef CONFIG_ZONE_DEVICE
957 ZONE_DEVICE,
958#endif
959 __MAX_NR_ZONES
960
961};
962
963#ifndef __GENERATING_BOUNDS_H
964
965#define ASYNC_AND_SYNC 2
966
967struct zone {
968 /* Read-mostly fields */
969
970 /* zone watermarks, access with *_wmark_pages(zone) macros */
971 unsigned long _watermark[NR_WMARK];
972 unsigned long watermark_boost;
973
974 unsigned long nr_reserved_highatomic;
975 unsigned long nr_free_highatomic;
976
977 /*
978 * We don't know if the memory that we're going to allocate will be
979 * freeable or/and it will be released eventually, so to avoid totally
980 * wasting several GB of ram we must reserve some of the lower zone
981 * memory (otherwise we risk to run OOM on the lower zones despite
982 * there being tons of freeable ram on the higher zones). This array is
983 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
984 * changes.
985 */
986 long lowmem_reserve[MAX_NR_ZONES];
987
988#ifdef CONFIG_NUMA
989 int node;
990#endif
991 struct pglist_data *zone_pgdat;
992 struct per_cpu_pages __percpu *per_cpu_pageset;
993 struct per_cpu_zonestat __percpu *per_cpu_zonestats;
994 /*
995 * the high and batch values are copied to individual pagesets for
996 * faster access
997 */
998 int pageset_high_min;
999 int pageset_high_max;
1000 int pageset_batch;
1001
1002#ifndef CONFIG_SPARSEMEM
1003 /*
1004 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
1005 * In SPARSEMEM, this map is stored in struct mem_section
1006 */
1007 unsigned long *pageblock_flags;
1008#endif /* CONFIG_SPARSEMEM */
1009
1010 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
1011 unsigned long zone_start_pfn;
1012
1013 /*
1014 * spanned_pages is the total pages spanned by the zone, including
1015 * holes, which is calculated as:
1016 * spanned_pages = zone_end_pfn - zone_start_pfn;
1017 *
1018 * present_pages is physical pages existing within the zone, which
1019 * is calculated as:
1020 * present_pages = spanned_pages - absent_pages(pages in holes);
1021 *
1022 * present_early_pages is present pages existing within the zone
1023 * located on memory available since early boot, excluding hotplugged
1024 * memory.
1025 *
1026 * managed_pages is present pages managed by the buddy system, which
1027 * is calculated as (reserved_pages includes pages allocated by the
1028 * bootmem allocator):
1029 * managed_pages = present_pages - reserved_pages;
1030 *
1031 * cma pages is present pages that are assigned for CMA use
1032 * (MIGRATE_CMA).
1033 *
1034 * So present_pages may be used by memory hotplug or memory power
1035 * management logic to figure out unmanaged pages by checking
1036 * (present_pages - managed_pages). And managed_pages should be used
1037 * by page allocator and vm scanner to calculate all kinds of watermarks
1038 * and thresholds.
1039 *
1040 * Locking rules:
1041 *
1042 * zone_start_pfn and spanned_pages are protected by span_seqlock.
1043 * It is a seqlock because it has to be read outside of zone->lock,
1044 * and it is done in the main allocator path. But, it is written
1045 * quite infrequently.
1046 *
1047 * The span_seq lock is declared along with zone->lock because it is
1048 * frequently read in proximity to zone->lock. It's good to
1049 * give them a chance of being in the same cacheline.
1050 *
1051 * Write access to present_pages at runtime should be protected by
1052 * mem_hotplug_begin/done(). Any reader who can't tolerant drift of
1053 * present_pages should use get_online_mems() to get a stable value.
1054 */
1055 atomic_long_t managed_pages;
1056 unsigned long spanned_pages;
1057 unsigned long present_pages;
1058#if defined(CONFIG_MEMORY_HOTPLUG)
1059 unsigned long present_early_pages;
1060#endif
1061#ifdef CONFIG_CMA
1062 unsigned long cma_pages;
1063#endif
1064
1065 const char *name;
1066
1067#ifdef CONFIG_MEMORY_ISOLATION
1068 /*
1069 * Number of isolated pageblock. It is used to solve incorrect
1070 * freepage counting problem due to racy retrieving migratetype
1071 * of pageblock. Protected by zone->lock.
1072 */
1073 unsigned long nr_isolate_pageblock;
1074#endif
1075
1076#ifdef CONFIG_MEMORY_HOTPLUG
1077 /* see spanned/present_pages for more description */
1078 seqlock_t span_seqlock;
1079#endif
1080
1081 int initialized;
1082
1083 /* Write-intensive fields used from the page allocator */
1084 CACHELINE_PADDING(_pad1_);
1085
1086 /* free areas of different sizes */
1087 struct free_area free_area[NR_PAGE_ORDERS];
1088
1089#ifdef CONFIG_UNACCEPTED_MEMORY
1090 /* Pages to be accepted. All pages on the list are MAX_PAGE_ORDER */
1091 struct list_head unaccepted_pages;
1092
1093 /* To be called once the last page in the zone is accepted */
1094 struct work_struct unaccepted_cleanup;
1095#endif
1096
1097 /* zone flags, see below */
1098 unsigned long flags;
1099
1100 /* Primarily protects free_area */
1101 spinlock_t lock;
1102
1103 /* Pages to be freed when next trylock succeeds */
1104 struct llist_head trylock_free_pages;
1105
1106 /* Write-intensive fields used by compaction and vmstats. */
1107 CACHELINE_PADDING(_pad2_);
1108
1109 /*
1110 * When free pages are below this point, additional steps are taken
1111 * when reading the number of free pages to avoid per-cpu counter
1112 * drift allowing watermarks to be breached
1113 */
1114 unsigned long percpu_drift_mark;
1115
1116#if defined CONFIG_COMPACTION || defined CONFIG_CMA
1117 /* pfn where compaction free scanner should start */
1118 unsigned long compact_cached_free_pfn;
1119 /* pfn where compaction migration scanner should start */
1120 unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC];
1121 unsigned long compact_init_migrate_pfn;
1122 unsigned long compact_init_free_pfn;
1123#endif
1124
1125#ifdef CONFIG_COMPACTION
1126 /*
1127 * On compaction failure, 1<<compact_defer_shift compactions
1128 * are skipped before trying again. The number attempted since
1129 * last failure is tracked with compact_considered.
1130 * compact_order_failed is the minimum compaction failed order.
1131 */
1132 unsigned int compact_considered;
1133 unsigned int compact_defer_shift;
1134 int compact_order_failed;
1135#endif
1136
1137#if defined CONFIG_COMPACTION || defined CONFIG_CMA
1138 /* Set to true when the PG_migrate_skip bits should be cleared */
1139 bool compact_blockskip_flush;
1140#endif
1141
1142 bool contiguous;
1143
1144 CACHELINE_PADDING(_pad3_);
1145 /* Zone statistics */
1146 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
1147 atomic_long_t vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
1148#ifdef CONFIG_HUGETLB_PAGE_OPTIMIZE_VMEMMAP
1149 struct page *vmemmap_tails[NR_VMEMMAP_TAILS];
1150#endif
1151} ____cacheline_internodealigned_in_smp;
1152
1153enum pgdat_flags {
1154 PGDAT_WRITEBACK, /* reclaim scanning has recently found
1155 * many pages under writeback
1156 */
1157 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
1158};
1159
1160enum zone_flags {
1161 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
1162 * Cleared when kswapd is woken.
1163 */
1164 ZONE_RECLAIM_ACTIVE, /* kswapd may be scanning the zone. */
1165 ZONE_BELOW_HIGH, /* zone is below high watermark. */
1166};
1167
1168static inline unsigned long wmark_pages(const struct zone *z,
1169 enum zone_watermarks w)
1170{
1171 return z->_watermark[w] + z->watermark_boost;
1172}
1173
1174static inline unsigned long min_wmark_pages(const struct zone *z)
1175{
1176 return wmark_pages(z, WMARK_MIN);
1177}
1178
1179static inline unsigned long low_wmark_pages(const struct zone *z)
1180{
1181 return wmark_pages(z, WMARK_LOW);
1182}
1183
1184static inline unsigned long high_wmark_pages(const struct zone *z)
1185{
1186 return wmark_pages(z, WMARK_HIGH);
1187}
1188
1189static inline unsigned long promo_wmark_pages(const struct zone *z)
1190{
1191 return wmark_pages(z, WMARK_PROMO);
1192}
1193
1194static inline unsigned long zone_managed_pages(const struct zone *zone)
1195{
1196 return (unsigned long)atomic_long_read(&zone->managed_pages);
1197}
1198
1199static inline unsigned long zone_cma_pages(struct zone *zone)
1200{
1201#ifdef CONFIG_CMA
1202 return zone->cma_pages;
1203#else
1204 return 0;
1205#endif
1206}
1207
1208static inline unsigned long zone_end_pfn(const struct zone *zone)
1209{
1210 return zone->zone_start_pfn + zone->spanned_pages;
1211}
1212
1213static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
1214{
1215 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
1216}
1217
1218static inline bool zone_is_initialized(const struct zone *zone)
1219{
1220 return zone->initialized;
1221}
1222
1223static inline bool zone_is_empty(const struct zone *zone)
1224{
1225 return zone->spanned_pages == 0;
1226}
1227
1228#ifndef BUILD_VDSO32_64
1229/*
1230 * The zone field is never updated after free_area_init_core()
1231 * sets it, so none of the operations on it need to be atomic.
1232 */
1233
1234/* Page flags: | [SECTION] | [NODE] | ZONE | [LAST_CPUPID] | ... | FLAGS | */
1235#define SECTIONS_PGOFF ((sizeof(unsigned long)*8) - SECTIONS_WIDTH)
1236#define NODES_PGOFF (SECTIONS_PGOFF - NODES_WIDTH)
1237#define ZONES_PGOFF (NODES_PGOFF - ZONES_WIDTH)
1238#define LAST_CPUPID_PGOFF (ZONES_PGOFF - LAST_CPUPID_WIDTH)
1239#define KASAN_TAG_PGOFF (LAST_CPUPID_PGOFF - KASAN_TAG_WIDTH)
1240#define LRU_GEN_PGOFF (KASAN_TAG_PGOFF - LRU_GEN_WIDTH)
1241#define LRU_REFS_PGOFF (LRU_GEN_PGOFF - LRU_REFS_WIDTH)
1242
1243/*
1244 * Define the bit shifts to access each section. For non-existent
1245 * sections we define the shift as 0; that plus a 0 mask ensures
1246 * the compiler will optimise away reference to them.
1247 */
1248#define SECTIONS_PGSHIFT (SECTIONS_PGOFF * (SECTIONS_WIDTH != 0))
1249#define NODES_PGSHIFT (NODES_PGOFF * (NODES_WIDTH != 0))
1250#define ZONES_PGSHIFT (ZONES_PGOFF * (ZONES_WIDTH != 0))
1251#define LAST_CPUPID_PGSHIFT (LAST_CPUPID_PGOFF * (LAST_CPUPID_WIDTH != 0))
1252#define KASAN_TAG_PGSHIFT (KASAN_TAG_PGOFF * (KASAN_TAG_WIDTH != 0))
1253
1254/* NODE:ZONE or SECTION:ZONE is used to ID a zone for the buddy allocator */
1255#ifdef NODE_NOT_IN_PAGE_FLAGS
1256#define ZONEID_SHIFT (SECTIONS_SHIFT + ZONES_SHIFT)
1257#define ZONEID_PGOFF ((SECTIONS_PGOFF < ZONES_PGOFF) ? \
1258 SECTIONS_PGOFF : ZONES_PGOFF)
1259#else
1260#define ZONEID_SHIFT (NODES_SHIFT + ZONES_SHIFT)
1261#define ZONEID_PGOFF ((NODES_PGOFF < ZONES_PGOFF) ? \
1262 NODES_PGOFF : ZONES_PGOFF)
1263#endif
1264
1265#define ZONEID_PGSHIFT (ZONEID_PGOFF * (ZONEID_SHIFT != 0))
1266
1267#define ZONES_MASK ((1UL << ZONES_WIDTH) - 1)
1268#define NODES_MASK ((1UL << NODES_WIDTH) - 1)
1269#define SECTIONS_MASK ((1UL << SECTIONS_WIDTH) - 1)
1270#define LAST_CPUPID_MASK ((1UL << LAST_CPUPID_SHIFT) - 1)
1271#define KASAN_TAG_MASK ((1UL << KASAN_TAG_WIDTH) - 1)
1272#define ZONEID_MASK ((1UL << ZONEID_SHIFT) - 1)
1273
1274static inline enum zone_type memdesc_zonenum(memdesc_flags_t flags)
1275{
1276 ASSERT_EXCLUSIVE_BITS(flags.f, ZONES_MASK << ZONES_PGSHIFT);
1277 return (flags.f >> ZONES_PGSHIFT) & ZONES_MASK;
1278}
1279
1280static inline enum zone_type page_zonenum(const struct page *page)
1281{
1282 return memdesc_zonenum(page->flags);
1283}
1284
1285static inline enum zone_type folio_zonenum(const struct folio *folio)
1286{
1287 return memdesc_zonenum(folio->flags);
1288}
1289
1290#ifdef CONFIG_ZONE_DEVICE
1291static inline bool memdesc_is_zone_device(memdesc_flags_t mdf)
1292{
1293 return memdesc_zonenum(mdf) == ZONE_DEVICE;
1294}
1295
1296static inline struct dev_pagemap *page_pgmap(const struct page *page)
1297{
1298 VM_WARN_ON_ONCE_PAGE(!memdesc_is_zone_device(page->flags), page);
1299 return page_folio(page)->pgmap;
1300}
1301
1302/*
1303 * Consecutive zone device pages should not be merged into the same sgl
1304 * or bvec segment with other types of pages or if they belong to different
1305 * pgmaps. Otherwise getting the pgmap of a given segment is not possible
1306 * without scanning the entire segment. This helper returns true either if
1307 * both pages are not zone device pages or both pages are zone device pages
1308 * with the same pgmap.
1309 */
1310static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1311 const struct page *b)
1312{
1313 if (memdesc_is_zone_device(a->flags) != memdesc_is_zone_device(b->flags))
1314 return false;
1315 if (!memdesc_is_zone_device(a->flags))
1316 return true;
1317 return page_pgmap(a) == page_pgmap(b);
1318}
1319
1320extern void memmap_init_zone_device(struct zone *, unsigned long,
1321 unsigned long, struct dev_pagemap *);
1322#else
1323static inline bool memdesc_is_zone_device(memdesc_flags_t mdf)
1324{
1325 return false;
1326}
1327static inline bool zone_device_pages_have_same_pgmap(const struct page *a,
1328 const struct page *b)
1329{
1330 return true;
1331}
1332static inline struct dev_pagemap *page_pgmap(const struct page *page)
1333{
1334 return NULL;
1335}
1336#endif
1337
1338static inline bool is_zone_device_page(const struct page *page)
1339{
1340 return memdesc_is_zone_device(page->flags);
1341}
1342
1343static inline bool folio_is_zone_device(const struct folio *folio)
1344{
1345 return memdesc_is_zone_device(folio->flags);
1346}
1347
1348static inline bool is_zone_movable_page(const struct page *page)
1349{
1350 return page_zonenum(page) == ZONE_MOVABLE;
1351}
1352
1353static inline bool folio_is_zone_movable(const struct folio *folio)
1354{
1355 return folio_zonenum(folio) == ZONE_MOVABLE;
1356}
1357#endif
1358
1359/*
1360 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
1361 * intersection with the given zone
1362 */
1363static inline bool zone_intersects(const struct zone *zone,
1364 unsigned long start_pfn, unsigned long nr_pages)
1365{
1366 if (zone_is_empty(zone))
1367 return false;
1368 if (start_pfn >= zone_end_pfn(zone) ||
1369 start_pfn + nr_pages <= zone->zone_start_pfn)
1370 return false;
1371
1372 return true;
1373}
1374
1375/*
1376 * The "priority" of VM scanning is how much of the queues we will scan in one
1377 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
1378 * queues ("queue_length >> 12") during an aging round.
1379 */
1380#define DEF_PRIORITY 12
1381
1382/* Maximum number of zones on a zonelist */
1383#define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
1384
1385enum {
1386 ZONELIST_FALLBACK, /* zonelist with fallback */
1387#ifdef CONFIG_NUMA
1388 /*
1389 * The NUMA zonelists are doubled because we need zonelists that
1390 * restrict the allocations to a single node for __GFP_THISNODE.
1391 */
1392 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
1393#endif
1394 MAX_ZONELISTS
1395};
1396
1397/*
1398 * This struct contains information about a zone in a zonelist. It is stored
1399 * here to avoid dereferences into large structures and lookups of tables
1400 */
1401struct zoneref {
1402 struct zone *zone; /* Pointer to actual zone */
1403 int zone_idx; /* zone_idx(zoneref->zone) */
1404};
1405
1406/*
1407 * One allocation request operates on a zonelist. A zonelist
1408 * is a list of zones, the first one is the 'goal' of the
1409 * allocation, the other zones are fallback zones, in decreasing
1410 * priority.
1411 *
1412 * To speed the reading of the zonelist, the zonerefs contain the zone index
1413 * of the entry being read. Helper functions to access information given
1414 * a struct zoneref are
1415 *
1416 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
1417 * zonelist_zone_idx() - Return the index of the zone for an entry
1418 * zonelist_node_idx() - Return the index of the node for an entry
1419 */
1420struct zonelist {
1421 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
1422};
1423
1424/*
1425 * The array of struct pages for flatmem.
1426 * It must be declared for SPARSEMEM as well because there are configurations
1427 * that rely on that.
1428 */
1429extern struct page *mem_map;
1430
1431#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1432struct deferred_split {
1433 spinlock_t split_queue_lock;
1434 struct list_head split_queue;
1435 unsigned long split_queue_len;
1436};
1437#endif
1438
1439#ifdef CONFIG_MEMORY_FAILURE
1440/*
1441 * Per NUMA node memory failure handling statistics.
1442 */
1443struct memory_failure_stats {
1444 /*
1445 * Number of raw pages poisoned.
1446 * Cases not accounted: memory outside kernel control, offline page,
1447 * arch-specific memory_failure (SGX), hwpoison_filter() filtered
1448 * error events, and unpoison actions from hwpoison_unpoison.
1449 */
1450 unsigned long total;
1451 /*
1452 * Recovery results of poisoned raw pages handled by memory_failure,
1453 * in sync with mf_result.
1454 * total = ignored + failed + delayed + recovered.
1455 * total * PAGE_SIZE * #nodes = /proc/meminfo/HardwareCorrupted.
1456 */
1457 unsigned long ignored;
1458 unsigned long failed;
1459 unsigned long delayed;
1460 unsigned long recovered;
1461};
1462#endif
1463
1464/*
1465 * On NUMA machines, each NUMA node would have a pg_data_t to describe
1466 * it's memory layout. On UMA machines there is a single pglist_data which
1467 * describes the whole memory.
1468 *
1469 * Memory statistics and page replacement data structures are maintained on a
1470 * per-zone basis.
1471 */
1472typedef struct pglist_data {
1473 /*
1474 * node_zones contains just the zones for THIS node. Not all of the
1475 * zones may be populated, but it is the full list. It is referenced by
1476 * this node's node_zonelists as well as other node's node_zonelists.
1477 */
1478 struct zone node_zones[MAX_NR_ZONES];
1479
1480 /*
1481 * node_zonelists contains references to all zones in all nodes.
1482 * Generally the first zones will be references to this node's
1483 * node_zones.
1484 */
1485 struct zonelist node_zonelists[MAX_ZONELISTS];
1486
1487 int nr_zones; /* number of populated zones in this node */
1488#ifdef CONFIG_FLATMEM /* means !SPARSEMEM */
1489 struct page *node_mem_map;
1490#ifdef CONFIG_PAGE_EXTENSION
1491 struct page_ext *node_page_ext;
1492#endif
1493#endif
1494#if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
1495 /*
1496 * Must be held any time you expect node_start_pfn,
1497 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
1498 * Also synchronizes pgdat->first_deferred_pfn during deferred page
1499 * init.
1500 *
1501 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
1502 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
1503 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
1504 *
1505 * Nests above zone->lock and zone->span_seqlock
1506 */
1507 spinlock_t node_size_lock;
1508#endif
1509 unsigned long node_start_pfn;
1510 unsigned long node_present_pages; /* total number of physical pages */
1511 unsigned long node_spanned_pages; /* total size of physical page
1512 range, including holes */
1513 int node_id;
1514 wait_queue_head_t kswapd_wait;
1515 wait_queue_head_t pfmemalloc_wait;
1516
1517 /* workqueues for throttling reclaim for different reasons. */
1518 wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE];
1519
1520 atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */
1521 unsigned long nr_reclaim_start; /* nr pages written while throttled
1522 * when throttling started. */
1523#ifdef CONFIG_MEMORY_HOTPLUG
1524 struct mutex kswapd_lock;
1525#endif
1526 struct task_struct *kswapd; /* Protected by kswapd_lock */
1527 int kswapd_order;
1528 enum zone_type kswapd_highest_zoneidx;
1529
1530 atomic_t kswapd_failures; /* Number of 'reclaimed == 0' runs */
1531
1532#ifdef CONFIG_COMPACTION
1533 int kcompactd_max_order;
1534 enum zone_type kcompactd_highest_zoneidx;
1535 wait_queue_head_t kcompactd_wait;
1536 struct task_struct *kcompactd;
1537 bool proactive_compact_trigger;
1538#endif
1539 /*
1540 * This is a per-node reserve of pages that are not available
1541 * to userspace allocations.
1542 */
1543 unsigned long totalreserve_pages;
1544
1545#ifdef CONFIG_NUMA
1546 /*
1547 * node reclaim becomes active if more unmapped pages exist.
1548 */
1549 unsigned long min_unmapped_pages;
1550 unsigned long min_slab_pages;
1551#endif /* CONFIG_NUMA */
1552
1553 /* Write-intensive fields used by page reclaim */
1554 CACHELINE_PADDING(_pad1_);
1555
1556#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1557 /*
1558 * If memory initialisation on large machines is deferred then this
1559 * is the first PFN that needs to be initialised.
1560 */
1561 unsigned long first_deferred_pfn;
1562#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1563
1564#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1565 struct deferred_split deferred_split_queue;
1566#endif
1567
1568#ifdef CONFIG_NUMA_BALANCING
1569 /* start time in ms of current promote rate limit period */
1570 unsigned int nbp_rl_start;
1571 /* number of promote candidate pages at start time of current rate limit period */
1572 unsigned long nbp_rl_nr_cand;
1573 /* promote threshold in ms */
1574 unsigned int nbp_threshold;
1575 /* start time in ms of current promote threshold adjustment period */
1576 unsigned int nbp_th_start;
1577 /*
1578 * number of promote candidate pages at start time of current promote
1579 * threshold adjustment period
1580 */
1581 unsigned long nbp_th_nr_cand;
1582#endif
1583 /* Fields commonly accessed by the page reclaim scanner */
1584
1585 /*
1586 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
1587 *
1588 * Use mem_cgroup_lruvec() to look up lruvecs.
1589 */
1590 struct lruvec __lruvec;
1591
1592 unsigned long flags;
1593
1594#ifdef CONFIG_LRU_GEN
1595 /* kswap mm walk data */
1596 struct lru_gen_mm_walk mm_walk;
1597 /* lru_gen_folio list */
1598 struct lru_gen_memcg memcg_lru;
1599#endif
1600
1601 CACHELINE_PADDING(_pad2_);
1602
1603 /* Per-node vmstats */
1604 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
1605 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
1606#ifdef CONFIG_NUMA
1607 struct memory_tier __rcu *memtier;
1608#endif
1609#ifdef CONFIG_MEMORY_FAILURE
1610 struct memory_failure_stats mf_stats;
1611#endif
1612} pg_data_t;
1613
1614#define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
1615#define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
1616
1617#define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
1618#define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
1619
1620static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
1621{
1622 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
1623}
1624
1625#include <linux/memory_hotplug.h>
1626
1627void build_all_zonelists(pg_data_t *pgdat);
1628bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
1629 int highest_zoneidx, unsigned int alloc_flags,
1630 long free_pages);
1631bool zone_watermark_ok(struct zone *z, unsigned int order,
1632 unsigned long mark, int highest_zoneidx,
1633 unsigned int alloc_flags);
1634
1635enum kswapd_clear_hopeless_reason {
1636 KSWAPD_CLEAR_HOPELESS_OTHER = 0,
1637 KSWAPD_CLEAR_HOPELESS_KSWAPD,
1638 KSWAPD_CLEAR_HOPELESS_DIRECT,
1639 KSWAPD_CLEAR_HOPELESS_PCP,
1640};
1641
1642void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
1643 enum zone_type highest_zoneidx);
1644void kswapd_try_clear_hopeless(struct pglist_data *pgdat,
1645 unsigned int order, int highest_zoneidx);
1646void kswapd_clear_hopeless(pg_data_t *pgdat, enum kswapd_clear_hopeless_reason reason);
1647bool kswapd_test_hopeless(pg_data_t *pgdat);
1648
1649/*
1650 * Memory initialization context, use to differentiate memory added by
1651 * the platform statically or via memory hotplug interface.
1652 */
1653enum meminit_context {
1654 MEMINIT_EARLY,
1655 MEMINIT_HOTPLUG,
1656};
1657
1658extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
1659 unsigned long size);
1660
1661extern void lruvec_init(struct lruvec *lruvec);
1662
1663static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
1664{
1665#ifdef CONFIG_MEMCG
1666 return lruvec->pgdat;
1667#else
1668 return container_of(lruvec, struct pglist_data, __lruvec);
1669#endif
1670}
1671
1672#ifdef CONFIG_HAVE_MEMORYLESS_NODES
1673int local_memory_node(int node_id);
1674#else
1675static inline int local_memory_node(int node_id) { return node_id; };
1676#endif
1677
1678/*
1679 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
1680 */
1681#define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
1682
1683#ifdef CONFIG_ZONE_DEVICE
1684static inline bool zone_is_zone_device(const struct zone *zone)
1685{
1686 return zone_idx(zone) == ZONE_DEVICE;
1687}
1688#else
1689static inline bool zone_is_zone_device(const struct zone *zone)
1690{
1691 return false;
1692}
1693#endif
1694
1695/*
1696 * Returns true if a zone has pages managed by the buddy allocator.
1697 * All the reclaim decisions have to use this function rather than
1698 * populated_zone(). If the whole zone is reserved then we can easily
1699 * end up with populated_zone() && !managed_zone().
1700 */
1701static inline bool managed_zone(const struct zone *zone)
1702{
1703 return zone_managed_pages(zone);
1704}
1705
1706/* Returns true if a zone has memory */
1707static inline bool populated_zone(const struct zone *zone)
1708{
1709 return zone->present_pages;
1710}
1711
1712#ifdef CONFIG_NUMA
1713static inline int zone_to_nid(const struct zone *zone)
1714{
1715 return zone->node;
1716}
1717
1718static inline void zone_set_nid(struct zone *zone, int nid)
1719{
1720 zone->node = nid;
1721}
1722#else
1723static inline int zone_to_nid(const struct zone *zone)
1724{
1725 return 0;
1726}
1727
1728static inline void zone_set_nid(struct zone *zone, int nid) {}
1729#endif
1730
1731extern int movable_zone;
1732
1733static inline int is_highmem_idx(enum zone_type idx)
1734{
1735#ifdef CONFIG_HIGHMEM
1736 return (idx == ZONE_HIGHMEM ||
1737 (idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM));
1738#else
1739 return 0;
1740#endif
1741}
1742
1743/**
1744 * is_highmem - helper function to quickly check if a struct zone is a
1745 * highmem zone or not. This is an attempt to keep references
1746 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
1747 * @zone: pointer to struct zone variable
1748 * Return: 1 for a highmem zone, 0 otherwise
1749 */
1750static inline int is_highmem(const struct zone *zone)
1751{
1752 return is_highmem_idx(zone_idx(zone));
1753}
1754
1755bool has_managed_zone(enum zone_type zone);
1756static inline bool has_managed_dma(void)
1757{
1758#ifdef CONFIG_ZONE_DMA
1759 return has_managed_zone(ZONE_DMA);
1760#else
1761 return false;
1762#endif
1763}
1764
1765
1766#ifndef CONFIG_NUMA
1767
1768extern struct pglist_data contig_page_data;
1769static inline struct pglist_data *NODE_DATA(int nid)
1770{
1771 return &contig_page_data;
1772}
1773
1774#else /* CONFIG_NUMA */
1775
1776#include <asm/mmzone.h>
1777
1778#endif /* !CONFIG_NUMA */
1779
1780extern struct pglist_data *first_online_pgdat(void);
1781extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
1782extern struct zone *next_zone(struct zone *zone);
1783
1784/**
1785 * for_each_online_pgdat - helper macro to iterate over all online nodes
1786 * @pgdat: pointer to a pg_data_t variable
1787 */
1788#define for_each_online_pgdat(pgdat) \
1789 for (pgdat = first_online_pgdat(); \
1790 pgdat; \
1791 pgdat = next_online_pgdat(pgdat))
1792/**
1793 * for_each_zone - helper macro to iterate over all memory zones
1794 * @zone: pointer to struct zone variable
1795 *
1796 * The user only needs to declare the zone variable, for_each_zone
1797 * fills it in.
1798 */
1799#define for_each_zone(zone) \
1800 for (zone = (first_online_pgdat())->node_zones; \
1801 zone; \
1802 zone = next_zone(zone))
1803
1804#define for_each_populated_zone(zone) \
1805 for (zone = (first_online_pgdat())->node_zones; \
1806 zone; \
1807 zone = next_zone(zone)) \
1808 if (!populated_zone(zone)) \
1809 ; /* do nothing */ \
1810 else
1811
1812static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1813{
1814 return zoneref->zone;
1815}
1816
1817static inline int zonelist_zone_idx(const struct zoneref *zoneref)
1818{
1819 return zoneref->zone_idx;
1820}
1821
1822static inline int zonelist_node_idx(const struct zoneref *zoneref)
1823{
1824 return zone_to_nid(zoneref->zone);
1825}
1826
1827struct zoneref *__next_zones_zonelist(struct zoneref *z,
1828 enum zone_type highest_zoneidx,
1829 nodemask_t *nodes);
1830
1831/**
1832 * 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
1833 * @z: The cursor used as a starting point for the search
1834 * @highest_zoneidx: The zone index of the highest zone to return
1835 * @nodes: An optional nodemask to filter the zonelist with
1836 *
1837 * This function returns the next zone at or below a given zone index that is
1838 * within the allowed nodemask using a cursor as the starting point for the
1839 * search. The zoneref returned is a cursor that represents the current zone
1840 * being examined. It should be advanced by one before calling
1841 * next_zones_zonelist again.
1842 *
1843 * Return: the next zone at or below highest_zoneidx within the allowed
1844 * nodemask using a cursor within a zonelist as a starting point
1845 */
1846static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1847 enum zone_type highest_zoneidx,
1848 nodemask_t *nodes)
1849{
1850 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1851 return z;
1852 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1853}
1854
1855/**
1856 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1857 * @zonelist: The zonelist to search for a suitable zone
1858 * @highest_zoneidx: The zone index of the highest zone to return
1859 * @nodes: An optional nodemask to filter the zonelist with
1860 *
1861 * This function returns the first zone at or below a given zone index that is
1862 * within the allowed nodemask. The zoneref returned is a cursor that can be
1863 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1864 * one before calling.
1865 *
1866 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1867 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1868 * update due to cpuset modification.
1869 *
1870 * Return: Zoneref pointer for the first suitable zone found
1871 */
1872static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1873 enum zone_type highest_zoneidx,
1874 nodemask_t *nodes)
1875{
1876 return next_zones_zonelist(zonelist->_zonerefs,
1877 highest_zoneidx, nodes);
1878}
1879
1880/**
1881 * 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
1882 * @zone: The current zone in the iterator
1883 * @z: The current pointer within zonelist->_zonerefs being iterated
1884 * @zlist: The zonelist being iterated
1885 * @highidx: The zone index of the highest zone to return
1886 * @nodemask: Nodemask allowed by the allocator
1887 *
1888 * This iterator iterates though all zones at or below a given zone index and
1889 * within a given nodemask
1890 */
1891#define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1892 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1893 zone; \
1894 z = next_zones_zonelist(++z, highidx, nodemask), \
1895 zone = zonelist_zone(z))
1896
1897#define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
1898 for (zone = zonelist_zone(z); \
1899 zone; \
1900 z = next_zones_zonelist(++z, highidx, nodemask), \
1901 zone = zonelist_zone(z))
1902
1903
1904/**
1905 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1906 * @zone: The current zone in the iterator
1907 * @z: The current pointer within zonelist->zones being iterated
1908 * @zlist: The zonelist being iterated
1909 * @highidx: The zone index of the highest zone to return
1910 *
1911 * This iterator iterates though all zones at or below a given zone index.
1912 */
1913#define for_each_zone_zonelist(zone, z, zlist, highidx) \
1914 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1915
1916/* Whether the 'nodes' are all movable nodes */
1917static inline bool movable_only_nodes(nodemask_t *nodes)
1918{
1919 struct zonelist *zonelist;
1920 struct zoneref *z;
1921 int nid;
1922
1923 if (nodes_empty(*nodes))
1924 return false;
1925
1926 /*
1927 * We can chose arbitrary node from the nodemask to get a
1928 * zonelist as they are interlinked. We just need to find
1929 * at least one zone that can satisfy kernel allocations.
1930 */
1931 nid = first_node(*nodes);
1932 zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK];
1933 z = first_zones_zonelist(zonelist, ZONE_NORMAL, nodes);
1934 return (!zonelist_zone(z)) ? true : false;
1935}
1936
1937
1938#ifdef CONFIG_SPARSEMEM
1939#include <asm/sparsemem.h>
1940#endif
1941
1942#ifdef CONFIG_FLATMEM
1943#define pfn_to_nid(pfn) (0)
1944#endif
1945
1946#ifdef CONFIG_SPARSEMEM
1947
1948/*
1949 * PA_SECTION_SHIFT physical address to/from section number
1950 * PFN_SECTION_SHIFT pfn to/from section number
1951 */
1952#define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1953#define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1954
1955#define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1956
1957#define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1958#define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1959
1960#define SECTION_BLOCKFLAGS_BITS \
1961 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1962
1963#if (MAX_PAGE_ORDER + PAGE_SHIFT) > SECTION_SIZE_BITS
1964#error Allocator MAX_PAGE_ORDER exceeds SECTION_SIZE
1965#endif
1966
1967static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1968{
1969 return pfn >> PFN_SECTION_SHIFT;
1970}
1971static inline unsigned long section_nr_to_pfn(unsigned long sec)
1972{
1973 return sec << PFN_SECTION_SHIFT;
1974}
1975
1976#define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1977#define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1978
1979#define SUBSECTION_SHIFT 21
1980#define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1981
1982#define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1983#define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1984#define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1985
1986#if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1987#error Subsection size exceeds section size
1988#else
1989#define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1990#endif
1991
1992#define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1993#define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1994
1995struct mem_section_usage {
1996 struct rcu_head rcu;
1997#ifdef CONFIG_SPARSEMEM_VMEMMAP
1998 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1999#endif
2000 /* See declaration of similar field in struct zone */
2001 unsigned long pageblock_flags[0];
2002};
2003
2004struct page;
2005struct page_ext;
2006struct mem_section {
2007 /*
2008 * This is, logically, a pointer to an array of struct
2009 * pages. However, it is stored with some other magic.
2010 * (see sparse_init_one_section())
2011 *
2012 * Additionally during early boot we encode node id of
2013 * the location of the section here to guide allocation.
2014 * (see sparse.c::memory_present())
2015 *
2016 * Making it a UL at least makes someone do a cast
2017 * before using it wrong.
2018 */
2019 unsigned long section_mem_map;
2020
2021 struct mem_section_usage *usage;
2022#ifdef CONFIG_PAGE_EXTENSION
2023 /*
2024 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
2025 * section. (see page_ext.h about this.)
2026 */
2027 struct page_ext *page_ext;
2028 unsigned long pad;
2029#endif
2030 /*
2031 * WARNING: mem_section must be a power-of-2 in size for the
2032 * calculation and use of SECTION_ROOT_MASK to make sense.
2033 */
2034};
2035
2036#ifdef CONFIG_SPARSEMEM_EXTREME
2037#define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
2038#else
2039#define SECTIONS_PER_ROOT 1
2040#endif
2041
2042#define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
2043#define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
2044#define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
2045
2046#ifdef CONFIG_SPARSEMEM_EXTREME
2047extern struct mem_section **mem_section;
2048#else
2049extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
2050#endif
2051
2052static inline unsigned long *section_to_usemap(struct mem_section *ms)
2053{
2054 return ms->usage->pageblock_flags;
2055}
2056
2057static inline struct mem_section *__nr_to_section(unsigned long nr)
2058{
2059 unsigned long root = SECTION_NR_TO_ROOT(nr);
2060
2061 if (unlikely(root >= NR_SECTION_ROOTS))
2062 return NULL;
2063
2064#ifdef CONFIG_SPARSEMEM_EXTREME
2065 if (!mem_section || !mem_section[root])
2066 return NULL;
2067#endif
2068 return &mem_section[root][nr & SECTION_ROOT_MASK];
2069}
2070extern size_t mem_section_usage_size(void);
2071
2072/*
2073 * We use the lower bits of the mem_map pointer to store a little bit of
2074 * information. The pointer is calculated as mem_map - section_nr_to_pfn().
2075 * The result is aligned to the minimum alignment of the two values:
2076 *
2077 * 1. All mem_map arrays are page-aligned.
2078 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT lowest bits.
2079 *
2080 * We always expect a single section to cover full pages. Therefore,
2081 * we can safely assume that PFN_SECTION_SHIFT is large enough to
2082 * accommodate SECTION_MAP_LAST_BIT. We use BUILD_BUG_ON() to ensure this.
2083 */
2084enum {
2085 SECTION_MARKED_PRESENT_BIT,
2086 SECTION_HAS_MEM_MAP_BIT,
2087 SECTION_IS_ONLINE_BIT,
2088 SECTION_IS_EARLY_BIT,
2089#ifdef CONFIG_ZONE_DEVICE
2090 SECTION_TAINT_ZONE_DEVICE_BIT,
2091#endif
2092#ifdef CONFIG_SPARSEMEM_VMEMMAP_PREINIT
2093 SECTION_IS_VMEMMAP_PREINIT_BIT,
2094#endif
2095 SECTION_MAP_LAST_BIT,
2096};
2097
2098#define SECTION_MARKED_PRESENT BIT(SECTION_MARKED_PRESENT_BIT)
2099#define SECTION_HAS_MEM_MAP BIT(SECTION_HAS_MEM_MAP_BIT)
2100#define SECTION_IS_ONLINE BIT(SECTION_IS_ONLINE_BIT)
2101#define SECTION_IS_EARLY BIT(SECTION_IS_EARLY_BIT)
2102#ifdef CONFIG_ZONE_DEVICE
2103#define SECTION_TAINT_ZONE_DEVICE BIT(SECTION_TAINT_ZONE_DEVICE_BIT)
2104#endif
2105#ifdef CONFIG_SPARSEMEM_VMEMMAP_PREINIT
2106#define SECTION_IS_VMEMMAP_PREINIT BIT(SECTION_IS_VMEMMAP_PREINIT_BIT)
2107#endif
2108#define SECTION_MAP_MASK (~(BIT(SECTION_MAP_LAST_BIT) - 1))
2109#define SECTION_NID_SHIFT SECTION_MAP_LAST_BIT
2110
2111static inline struct page *__section_mem_map_addr(struct mem_section *section)
2112{
2113 unsigned long map = section->section_mem_map;
2114 map &= SECTION_MAP_MASK;
2115 return (struct page *)map;
2116}
2117
2118static inline int present_section(const struct mem_section *section)
2119{
2120 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
2121}
2122
2123static inline int present_section_nr(unsigned long nr)
2124{
2125 return present_section(__nr_to_section(nr));
2126}
2127
2128static inline int valid_section(const struct mem_section *section)
2129{
2130 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
2131}
2132
2133static inline int early_section(const struct mem_section *section)
2134{
2135 return (section && (section->section_mem_map & SECTION_IS_EARLY));
2136}
2137
2138static inline int valid_section_nr(unsigned long nr)
2139{
2140 return valid_section(__nr_to_section(nr));
2141}
2142
2143static inline int online_section(const struct mem_section *section)
2144{
2145 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
2146}
2147
2148#ifdef CONFIG_ZONE_DEVICE
2149static inline int online_device_section(const struct mem_section *section)
2150{
2151 unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE;
2152
2153 return section && ((section->section_mem_map & flags) == flags);
2154}
2155#else
2156static inline int online_device_section(const struct mem_section *section)
2157{
2158 return 0;
2159}
2160#endif
2161
2162#ifdef CONFIG_SPARSEMEM_VMEMMAP_PREINIT
2163static inline int preinited_vmemmap_section(const struct mem_section *section)
2164{
2165 return (section &&
2166 (section->section_mem_map & SECTION_IS_VMEMMAP_PREINIT));
2167}
2168
2169void sparse_vmemmap_init_nid_early(int nid);
2170void sparse_vmemmap_init_nid_late(int nid);
2171
2172#else
2173static inline int preinited_vmemmap_section(const struct mem_section *section)
2174{
2175 return 0;
2176}
2177static inline void sparse_vmemmap_init_nid_early(int nid)
2178{
2179}
2180
2181static inline void sparse_vmemmap_init_nid_late(int nid)
2182{
2183}
2184#endif
2185
2186static inline int online_section_nr(unsigned long nr)
2187{
2188 return online_section(__nr_to_section(nr));
2189}
2190
2191#ifdef CONFIG_MEMORY_HOTPLUG
2192void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
2193void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
2194#endif
2195
2196static inline struct mem_section *__pfn_to_section(unsigned long pfn)
2197{
2198 return __nr_to_section(pfn_to_section_nr(pfn));
2199}
2200
2201extern unsigned long __highest_present_section_nr;
2202
2203static inline int subsection_map_index(unsigned long pfn)
2204{
2205 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
2206}
2207
2208#ifdef CONFIG_SPARSEMEM_VMEMMAP
2209static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
2210{
2211 int idx = subsection_map_index(pfn);
2212 struct mem_section_usage *usage = READ_ONCE(ms->usage);
2213
2214 return usage ? test_bit(idx, usage->subsection_map) : 0;
2215}
2216
2217static inline bool pfn_section_first_valid(struct mem_section *ms, unsigned long *pfn)
2218{
2219 struct mem_section_usage *usage = READ_ONCE(ms->usage);
2220 int idx = subsection_map_index(*pfn);
2221 unsigned long bit;
2222
2223 if (!usage)
2224 return false;
2225
2226 if (test_bit(idx, usage->subsection_map))
2227 return true;
2228
2229 /* Find the next subsection that exists */
2230 bit = find_next_bit(usage->subsection_map, SUBSECTIONS_PER_SECTION, idx);
2231 if (bit == SUBSECTIONS_PER_SECTION)
2232 return false;
2233
2234 *pfn = (*pfn & PAGE_SECTION_MASK) + (bit * PAGES_PER_SUBSECTION);
2235 return true;
2236}
2237#else
2238static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
2239{
2240 return 1;
2241}
2242
2243static inline bool pfn_section_first_valid(struct mem_section *ms, unsigned long *pfn)
2244{
2245 return true;
2246}
2247#endif
2248
2249void sparse_init_early_section(int nid, struct page *map, unsigned long pnum,
2250 unsigned long flags);
2251
2252#ifndef CONFIG_HAVE_ARCH_PFN_VALID
2253/**
2254 * pfn_valid - check if there is a valid memory map entry for a PFN
2255 * @pfn: the page frame number to check
2256 *
2257 * Check if there is a valid memory map entry aka struct page for the @pfn.
2258 * Note, that availability of the memory map entry does not imply that
2259 * there is actual usable memory at that @pfn. The struct page may
2260 * represent a hole or an unusable page frame.
2261 *
2262 * Return: 1 for PFNs that have memory map entries and 0 otherwise
2263 */
2264static inline int pfn_valid(unsigned long pfn)
2265{
2266 struct mem_section *ms;
2267 int ret;
2268
2269 /*
2270 * Ensure the upper PAGE_SHIFT bits are clear in the
2271 * pfn. Else it might lead to false positives when
2272 * some of the upper bits are set, but the lower bits
2273 * match a valid pfn.
2274 */
2275 if (PHYS_PFN(PFN_PHYS(pfn)) != pfn)
2276 return 0;
2277
2278 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2279 return 0;
2280 ms = __pfn_to_section(pfn);
2281 rcu_read_lock_sched();
2282 if (!valid_section(ms)) {
2283 rcu_read_unlock_sched();
2284 return 0;
2285 }
2286 /*
2287 * Traditionally early sections always returned pfn_valid() for
2288 * the entire section-sized span.
2289 */
2290 ret = early_section(ms) || pfn_section_valid(ms, pfn);
2291 rcu_read_unlock_sched();
2292
2293 return ret;
2294}
2295
2296/* Returns end_pfn or higher if no valid PFN remaining in range */
2297static inline unsigned long first_valid_pfn(unsigned long pfn, unsigned long end_pfn)
2298{
2299 unsigned long nr = pfn_to_section_nr(pfn);
2300
2301 rcu_read_lock_sched();
2302
2303 while (nr <= __highest_present_section_nr && pfn < end_pfn) {
2304 struct mem_section *ms = __pfn_to_section(pfn);
2305
2306 if (valid_section(ms) &&
2307 (early_section(ms) || pfn_section_first_valid(ms, &pfn))) {
2308 rcu_read_unlock_sched();
2309 return pfn;
2310 }
2311
2312 /* Nothing left in this section? Skip to next section */
2313 nr++;
2314 pfn = section_nr_to_pfn(nr);
2315 }
2316
2317 rcu_read_unlock_sched();
2318 return end_pfn;
2319}
2320
2321static inline unsigned long next_valid_pfn(unsigned long pfn, unsigned long end_pfn)
2322{
2323 pfn++;
2324
2325 if (pfn >= end_pfn)
2326 return end_pfn;
2327
2328 /*
2329 * Either every PFN within the section (or subsection for VMEMMAP) is
2330 * valid, or none of them are. So there's no point repeating the check
2331 * for every PFN; only call first_valid_pfn() again when crossing a
2332 * (sub)section boundary (i.e. !(pfn & ~PAGE_{SUB,}SECTION_MASK)).
2333 */
2334 if (pfn & ~(IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP) ?
2335 PAGE_SUBSECTION_MASK : PAGE_SECTION_MASK))
2336 return pfn;
2337
2338 return first_valid_pfn(pfn, end_pfn);
2339}
2340
2341
2342#define for_each_valid_pfn(_pfn, _start_pfn, _end_pfn) \
2343 for ((_pfn) = first_valid_pfn((_start_pfn), (_end_pfn)); \
2344 (_pfn) < (_end_pfn); \
2345 (_pfn) = next_valid_pfn((_pfn), (_end_pfn)))
2346
2347#endif
2348
2349static inline int pfn_in_present_section(unsigned long pfn)
2350{
2351 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
2352 return 0;
2353 return present_section(__pfn_to_section(pfn));
2354}
2355
2356static inline unsigned long next_present_section_nr(unsigned long section_nr)
2357{
2358 while (++section_nr <= __highest_present_section_nr) {
2359 if (present_section_nr(section_nr))
2360 return section_nr;
2361 }
2362
2363 return -1;
2364}
2365
2366#define for_each_present_section_nr(start, section_nr) \
2367 for (section_nr = next_present_section_nr(start - 1); \
2368 section_nr != -1; \
2369 section_nr = next_present_section_nr(section_nr))
2370
2371/*
2372 * These are _only_ used during initialisation, therefore they
2373 * can use __initdata ... They could have names to indicate
2374 * this restriction.
2375 */
2376#ifdef CONFIG_NUMA
2377#define pfn_to_nid(pfn) \
2378({ \
2379 unsigned long __pfn_to_nid_pfn = (pfn); \
2380 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
2381})
2382#else
2383#define pfn_to_nid(pfn) (0)
2384#endif
2385
2386#else
2387#define sparse_vmemmap_init_nid_early(_nid) do {} while (0)
2388#define sparse_vmemmap_init_nid_late(_nid) do {} while (0)
2389#define pfn_in_present_section pfn_valid
2390#endif /* CONFIG_SPARSEMEM */
2391
2392/*
2393 * Fallback case for when the architecture provides its own pfn_valid() but
2394 * not a corresponding for_each_valid_pfn().
2395 */
2396#ifndef for_each_valid_pfn
2397#define for_each_valid_pfn(_pfn, _start_pfn, _end_pfn) \
2398 for ((_pfn) = (_start_pfn); (_pfn) < (_end_pfn); (_pfn)++) \
2399 if (pfn_valid(_pfn))
2400#endif
2401
2402#endif /* !__GENERATING_BOUNDS.H */
2403#endif /* !__ASSEMBLY__ */
2404#endif /* _LINUX_MMZONE_H */