blob: c142a152d6a41cfaf7ddfc0dd3162e8e96c459ff [file] [log] [blame]
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001/* 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/wait.h>
11#include <linux/bitops.h>
12#include <linux/cache.h>
13#include <linux/threads.h>
14#include <linux/numa.h>
15#include <linux/init.h>
16#include <linux/seqlock.h>
17#include <linux/nodemask.h>
18#include <linux/pageblock-flags.h>
19#include <linux/page-flags-layout.h>
20#include <linux/atomic.h>
David Brazdil0f672f62019-12-10 10:32:29 +000021#include <linux/mm_types.h>
22#include <linux/page-flags.h>
Andrew Scullb4b6d4a2019-01-02 15:54:55 +000023#include <asm/page.h>
24
25/* Free memory management - zoned buddy allocator. */
26#ifndef CONFIG_FORCE_MAX_ZONEORDER
27#define MAX_ORDER 11
28#else
29#define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
30#endif
31#define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
32
33/*
34 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
35 * costly to service. That is between allocation orders which should
36 * coalesce naturally under reasonable reclaim pressure and those which
37 * will not.
38 */
39#define PAGE_ALLOC_COSTLY_ORDER 3
40
41enum migratetype {
42 MIGRATE_UNMOVABLE,
43 MIGRATE_MOVABLE,
44 MIGRATE_RECLAIMABLE,
45 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
46 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
47#ifdef CONFIG_CMA
48 /*
49 * MIGRATE_CMA migration type is designed to mimic the way
50 * ZONE_MOVABLE works. Only movable pages can be allocated
51 * from MIGRATE_CMA pageblocks and page allocator never
52 * implicitly change migration type of MIGRATE_CMA pageblock.
53 *
54 * The way to use it is to change migratetype of a range of
55 * pageblocks to MIGRATE_CMA which can be done by
56 * __free_pageblock_cma() function. What is important though
57 * is that a range of pageblocks must be aligned to
58 * MAX_ORDER_NR_PAGES should biggest page be bigger then
59 * a single pageblock.
60 */
61 MIGRATE_CMA,
62#endif
63#ifdef CONFIG_MEMORY_ISOLATION
64 MIGRATE_ISOLATE, /* can't allocate from here */
65#endif
66 MIGRATE_TYPES
67};
68
69/* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
David Brazdil0f672f62019-12-10 10:32:29 +000070extern const char * const migratetype_names[MIGRATE_TYPES];
Andrew Scullb4b6d4a2019-01-02 15:54:55 +000071
72#ifdef CONFIG_CMA
73# define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
74# define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
75#else
76# define is_migrate_cma(migratetype) false
77# define is_migrate_cma_page(_page) false
78#endif
79
80static inline bool is_migrate_movable(int mt)
81{
82 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
83}
84
85#define for_each_migratetype_order(order, type) \
86 for (order = 0; order < MAX_ORDER; order++) \
87 for (type = 0; type < MIGRATE_TYPES; type++)
88
89extern int page_group_by_mobility_disabled;
90
Olivier Deprez157378f2022-04-04 15:47:50 +020091#define MIGRATETYPE_MASK ((1UL << PB_migratetype_bits) - 1)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +000092
93#define get_pageblock_migratetype(page) \
Olivier Deprez157378f2022-04-04 15:47:50 +020094 get_pfnblock_flags_mask(page, page_to_pfn(page), MIGRATETYPE_MASK)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +000095
96struct free_area {
97 struct list_head free_list[MIGRATE_TYPES];
98 unsigned long nr_free;
99};
100
David Brazdil0f672f62019-12-10 10:32:29 +0000101static inline struct page *get_page_from_free_area(struct free_area *area,
102 int migratetype)
103{
104 return list_first_entry_or_null(&area->free_list[migratetype],
105 struct page, lru);
106}
107
David Brazdil0f672f62019-12-10 10:32:29 +0000108static inline bool free_area_empty(struct free_area *area, int migratetype)
109{
110 return list_empty(&area->free_list[migratetype]);
111}
112
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000113struct pglist_data;
114
115/*
116 * zone->lock and the zone lru_lock are two of the hottest locks in the kernel.
117 * So add a wild amount of padding here to ensure that they fall into separate
118 * cachelines. There are very few zone structures in the machine, so space
119 * consumption is not a concern here.
120 */
121#if defined(CONFIG_SMP)
122struct zone_padding {
123 char x[0];
124} ____cacheline_internodealigned_in_smp;
125#define ZONE_PADDING(name) struct zone_padding name;
126#else
127#define ZONE_PADDING(name)
128#endif
129
130#ifdef CONFIG_NUMA
131enum numa_stat_item {
132 NUMA_HIT, /* allocated in intended node */
133 NUMA_MISS, /* allocated in non intended node */
134 NUMA_FOREIGN, /* was intended here, hit elsewhere */
135 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
136 NUMA_LOCAL, /* allocation from local node */
137 NUMA_OTHER, /* allocation from other node */
138 NR_VM_NUMA_STAT_ITEMS
139};
140#else
141#define NR_VM_NUMA_STAT_ITEMS 0
142#endif
143
144enum zone_stat_item {
145 /* First 128 byte cacheline (assuming 64 bit words) */
146 NR_FREE_PAGES,
147 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
148 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
149 NR_ZONE_ACTIVE_ANON,
150 NR_ZONE_INACTIVE_FILE,
151 NR_ZONE_ACTIVE_FILE,
152 NR_ZONE_UNEVICTABLE,
153 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
154 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
155 NR_PAGETABLE, /* used for pagetables */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000156 /* Second 128 byte cacheline */
157 NR_BOUNCE,
158#if IS_ENABLED(CONFIG_ZSMALLOC)
159 NR_ZSPAGES, /* allocated in zsmalloc */
160#endif
161 NR_FREE_CMA_PAGES,
162 NR_VM_ZONE_STAT_ITEMS };
163
164enum node_stat_item {
165 NR_LRU_BASE,
166 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
167 NR_ACTIVE_ANON, /* " " " " " */
168 NR_INACTIVE_FILE, /* " " " " " */
169 NR_ACTIVE_FILE, /* " " " " " */
170 NR_UNEVICTABLE, /* " " " " " */
Olivier Deprez157378f2022-04-04 15:47:50 +0200171 NR_SLAB_RECLAIMABLE_B,
172 NR_SLAB_UNRECLAIMABLE_B,
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000173 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
174 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
David Brazdil0f672f62019-12-10 10:32:29 +0000175 WORKINGSET_NODES,
Olivier Deprez157378f2022-04-04 15:47:50 +0200176 WORKINGSET_REFAULT_BASE,
177 WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE,
178 WORKINGSET_REFAULT_FILE,
179 WORKINGSET_ACTIVATE_BASE,
180 WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE,
181 WORKINGSET_ACTIVATE_FILE,
182 WORKINGSET_RESTORE_BASE,
183 WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE,
184 WORKINGSET_RESTORE_FILE,
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000185 WORKINGSET_NODERECLAIM,
186 NR_ANON_MAPPED, /* Mapped anonymous pages */
187 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
188 only modified from process context */
189 NR_FILE_PAGES,
190 NR_FILE_DIRTY,
191 NR_WRITEBACK,
192 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
193 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
194 NR_SHMEM_THPS,
195 NR_SHMEM_PMDMAPPED,
David Brazdil0f672f62019-12-10 10:32:29 +0000196 NR_FILE_THPS,
197 NR_FILE_PMDMAPPED,
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000198 NR_ANON_THPS,
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000199 NR_VMSCAN_WRITE,
200 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
201 NR_DIRTIED, /* page dirtyings since bootup */
202 NR_WRITTEN, /* page writings since bootup */
David Brazdil0f672f62019-12-10 10:32:29 +0000203 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
Olivier Deprez157378f2022-04-04 15:47:50 +0200204 NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */
205 NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */
206 NR_KERNEL_STACK_KB, /* measured in KiB */
207#if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
208 NR_KERNEL_SCS_KB, /* measured in KiB */
209#endif
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000210 NR_VM_NODE_STAT_ITEMS
211};
212
213/*
Olivier Deprez157378f2022-04-04 15:47:50 +0200214 * Returns true if the value is measured in bytes (most vmstat values are
215 * measured in pages). This defines the API part, the internal representation
216 * might be different.
217 */
218static __always_inline bool vmstat_item_in_bytes(int idx)
219{
220 /*
221 * Global and per-node slab counters track slab pages.
222 * It's expected that changes are multiples of PAGE_SIZE.
223 * Internally values are stored in pages.
224 *
225 * Per-memcg and per-lruvec counters track memory, consumed
226 * by individual slab objects. These counters are actually
227 * byte-precise.
228 */
229 return (idx == NR_SLAB_RECLAIMABLE_B ||
230 idx == NR_SLAB_UNRECLAIMABLE_B);
231}
232
233/*
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000234 * We do arithmetic on the LRU lists in various places in the code,
235 * so it is important to keep the active lists LRU_ACTIVE higher in
236 * the array than the corresponding inactive lists, and to keep
237 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
238 *
239 * This has to be kept in sync with the statistics in zone_stat_item
240 * above and the descriptions in vmstat_text in mm/vmstat.c
241 */
242#define LRU_BASE 0
243#define LRU_ACTIVE 1
244#define LRU_FILE 2
245
246enum lru_list {
247 LRU_INACTIVE_ANON = LRU_BASE,
248 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
249 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
250 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
251 LRU_UNEVICTABLE,
252 NR_LRU_LISTS
253};
254
255#define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
256
257#define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
258
Olivier Deprez157378f2022-04-04 15:47:50 +0200259static inline bool is_file_lru(enum lru_list lru)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000260{
261 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
262}
263
Olivier Deprez157378f2022-04-04 15:47:50 +0200264static inline bool is_active_lru(enum lru_list lru)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000265{
266 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
267}
268
Olivier Deprez157378f2022-04-04 15:47:50 +0200269#define ANON_AND_FILE 2
270
271enum lruvec_flags {
272 LRUVEC_CONGESTED, /* lruvec has many dirty pages
273 * backed by a congested BDI
274 */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000275};
276
277struct lruvec {
278 struct list_head lists[NR_LRU_LISTS];
Olivier Deprez157378f2022-04-04 15:47:50 +0200279 /*
280 * These track the cost of reclaiming one LRU - file or anon -
281 * over the other. As the observed cost of reclaiming one LRU
282 * increases, the reclaim scan balance tips toward the other.
283 */
284 unsigned long anon_cost;
285 unsigned long file_cost;
286 /* Non-resident age, driven by LRU movement */
287 atomic_long_t nonresident_age;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000288 /* Refaults at the time of last reclaim cycle */
Olivier Deprez157378f2022-04-04 15:47:50 +0200289 unsigned long refaults[ANON_AND_FILE];
290 /* Various lruvec state flags (enum lruvec_flags) */
291 unsigned long flags;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000292#ifdef CONFIG_MEMCG
293 struct pglist_data *pgdat;
294#endif
295};
296
Olivier Deprez157378f2022-04-04 15:47:50 +0200297/* Isolate unmapped pages */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000298#define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
299/* Isolate for asynchronous migration */
300#define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
301/* Isolate unevictable pages */
302#define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
303
304/* LRU Isolation modes. */
305typedef unsigned __bitwise isolate_mode_t;
306
307enum zone_watermarks {
308 WMARK_MIN,
309 WMARK_LOW,
310 WMARK_HIGH,
311 NR_WMARK
312};
313
David Brazdil0f672f62019-12-10 10:32:29 +0000314#define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
315#define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
316#define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
317#define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000318
319struct per_cpu_pages {
320 int count; /* number of pages in the list */
321 int high; /* high watermark, emptying needed */
322 int batch; /* chunk size for buddy add/remove */
323
324 /* Lists of pages, one per migrate type stored on the pcp-lists */
325 struct list_head lists[MIGRATE_PCPTYPES];
326};
327
328struct per_cpu_pageset {
329 struct per_cpu_pages pcp;
330#ifdef CONFIG_NUMA
331 s8 expire;
332 u16 vm_numa_stat_diff[NR_VM_NUMA_STAT_ITEMS];
333#endif
334#ifdef CONFIG_SMP
335 s8 stat_threshold;
336 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
337#endif
338};
339
340struct per_cpu_nodestat {
341 s8 stat_threshold;
342 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
343};
344
345#endif /* !__GENERATING_BOUNDS.H */
346
347enum zone_type {
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000348 /*
Olivier Deprez157378f2022-04-04 15:47:50 +0200349 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
350 * to DMA to all of the addressable memory (ZONE_NORMAL).
351 * On architectures where this area covers the whole 32 bit address
352 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
353 * DMA addressing constraints. This distinction is important as a 32bit
354 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
355 * platforms may need both zones as they support peripherals with
356 * different DMA addressing limitations.
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000357 */
Olivier Deprez157378f2022-04-04 15:47:50 +0200358#ifdef CONFIG_ZONE_DMA
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000359 ZONE_DMA,
360#endif
361#ifdef CONFIG_ZONE_DMA32
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000362 ZONE_DMA32,
363#endif
364 /*
365 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
366 * performed on pages in ZONE_NORMAL if the DMA devices support
367 * transfers to all addressable memory.
368 */
369 ZONE_NORMAL,
370#ifdef CONFIG_HIGHMEM
371 /*
372 * A memory area that is only addressable by the kernel through
373 * mapping portions into its own address space. This is for example
374 * used by i386 to allow the kernel to address the memory beyond
375 * 900MB. The kernel will set up special mappings (page
376 * table entries on i386) for each page that the kernel needs to
377 * access.
378 */
379 ZONE_HIGHMEM,
380#endif
Olivier Deprez157378f2022-04-04 15:47:50 +0200381 /*
382 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
383 * movable pages with few exceptional cases described below. Main use
384 * cases for ZONE_MOVABLE are to make memory offlining/unplug more
385 * likely to succeed, and to locally limit unmovable allocations - e.g.,
386 * to increase the number of THP/huge pages. Notable special cases are:
387 *
388 * 1. Pinned pages: (long-term) pinning of movable pages might
389 * essentially turn such pages unmovable. Memory offlining might
390 * retry a long time.
391 * 2. memblock allocations: kernelcore/movablecore setups might create
392 * situations where ZONE_MOVABLE contains unmovable allocations
393 * after boot. Memory offlining and allocations fail early.
394 * 3. Memory holes: kernelcore/movablecore setups might create very rare
395 * situations where ZONE_MOVABLE contains memory holes after boot,
396 * for example, if we have sections that are only partially
397 * populated. Memory offlining and allocations fail early.
398 * 4. PG_hwpoison pages: while poisoned pages can be skipped during
399 * memory offlining, such pages cannot be allocated.
400 * 5. Unmovable PG_offline pages: in paravirtualized environments,
401 * hotplugged memory blocks might only partially be managed by the
402 * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
403 * parts not manged by the buddy are unmovable PG_offline pages. In
404 * some cases (virtio-mem), such pages can be skipped during
405 * memory offlining, however, cannot be moved/allocated. These
406 * techniques might use alloc_contig_range() to hide previously
407 * exposed pages from the buddy again (e.g., to implement some sort
408 * of memory unplug in virtio-mem).
409 *
410 * In general, no unmovable allocations that degrade memory offlining
411 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
412 * have to expect that migrating pages in ZONE_MOVABLE can fail (even
413 * if has_unmovable_pages() states that there are no unmovable pages,
414 * there can be false negatives).
415 */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000416 ZONE_MOVABLE,
417#ifdef CONFIG_ZONE_DEVICE
418 ZONE_DEVICE,
419#endif
420 __MAX_NR_ZONES
421
422};
423
424#ifndef __GENERATING_BOUNDS_H
425
Olivier Deprez157378f2022-04-04 15:47:50 +0200426#define ASYNC_AND_SYNC 2
427
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000428struct zone {
429 /* Read-mostly fields */
430
431 /* zone watermarks, access with *_wmark_pages(zone) macros */
David Brazdil0f672f62019-12-10 10:32:29 +0000432 unsigned long _watermark[NR_WMARK];
433 unsigned long watermark_boost;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000434
435 unsigned long nr_reserved_highatomic;
436
437 /*
438 * We don't know if the memory that we're going to allocate will be
439 * freeable or/and it will be released eventually, so to avoid totally
440 * wasting several GB of ram we must reserve some of the lower zone
441 * memory (otherwise we risk to run OOM on the lower zones despite
442 * there being tons of freeable ram on the higher zones). This array is
443 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
444 * changes.
445 */
446 long lowmem_reserve[MAX_NR_ZONES];
447
Olivier Deprez157378f2022-04-04 15:47:50 +0200448#ifdef CONFIG_NEED_MULTIPLE_NODES
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000449 int node;
450#endif
451 struct pglist_data *zone_pgdat;
452 struct per_cpu_pageset __percpu *pageset;
453
454#ifndef CONFIG_SPARSEMEM
455 /*
456 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
457 * In SPARSEMEM, this map is stored in struct mem_section
458 */
459 unsigned long *pageblock_flags;
460#endif /* CONFIG_SPARSEMEM */
461
462 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
463 unsigned long zone_start_pfn;
464
465 /*
466 * spanned_pages is the total pages spanned by the zone, including
467 * holes, which is calculated as:
468 * spanned_pages = zone_end_pfn - zone_start_pfn;
469 *
470 * present_pages is physical pages existing within the zone, which
471 * is calculated as:
472 * present_pages = spanned_pages - absent_pages(pages in holes);
473 *
474 * managed_pages is present pages managed by the buddy system, which
475 * is calculated as (reserved_pages includes pages allocated by the
476 * bootmem allocator):
477 * managed_pages = present_pages - reserved_pages;
478 *
479 * So present_pages may be used by memory hotplug or memory power
480 * management logic to figure out unmanaged pages by checking
481 * (present_pages - managed_pages). And managed_pages should be used
482 * by page allocator and vm scanner to calculate all kinds of watermarks
483 * and thresholds.
484 *
485 * Locking rules:
486 *
487 * zone_start_pfn and spanned_pages are protected by span_seqlock.
488 * It is a seqlock because it has to be read outside of zone->lock,
489 * and it is done in the main allocator path. But, it is written
490 * quite infrequently.
491 *
492 * The span_seq lock is declared along with zone->lock because it is
493 * frequently read in proximity to zone->lock. It's good to
494 * give them a chance of being in the same cacheline.
495 *
496 * Write access to present_pages at runtime should be protected by
497 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
498 * present_pages should get_online_mems() to get a stable value.
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000499 */
David Brazdil0f672f62019-12-10 10:32:29 +0000500 atomic_long_t managed_pages;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000501 unsigned long spanned_pages;
502 unsigned long present_pages;
503
504 const char *name;
505
506#ifdef CONFIG_MEMORY_ISOLATION
507 /*
508 * Number of isolated pageblock. It is used to solve incorrect
509 * freepage counting problem due to racy retrieving migratetype
510 * of pageblock. Protected by zone->lock.
511 */
512 unsigned long nr_isolate_pageblock;
513#endif
514
515#ifdef CONFIG_MEMORY_HOTPLUG
516 /* see spanned/present_pages for more description */
517 seqlock_t span_seqlock;
518#endif
519
520 int initialized;
521
522 /* Write-intensive fields used from the page allocator */
523 ZONE_PADDING(_pad1_)
524
525 /* free areas of different sizes */
526 struct free_area free_area[MAX_ORDER];
527
528 /* zone flags, see below */
529 unsigned long flags;
530
531 /* Primarily protects free_area */
532 spinlock_t lock;
533
534 /* Write-intensive fields used by compaction and vmstats. */
535 ZONE_PADDING(_pad2_)
536
537 /*
538 * When free pages are below this point, additional steps are taken
539 * when reading the number of free pages to avoid per-cpu counter
540 * drift allowing watermarks to be breached
541 */
542 unsigned long percpu_drift_mark;
543
544#if defined CONFIG_COMPACTION || defined CONFIG_CMA
545 /* pfn where compaction free scanner should start */
546 unsigned long compact_cached_free_pfn;
Olivier Deprez157378f2022-04-04 15:47:50 +0200547 /* pfn where compaction migration scanner should start */
548 unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC];
David Brazdil0f672f62019-12-10 10:32:29 +0000549 unsigned long compact_init_migrate_pfn;
550 unsigned long compact_init_free_pfn;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000551#endif
552
553#ifdef CONFIG_COMPACTION
554 /*
555 * On compaction failure, 1<<compact_defer_shift compactions
556 * are skipped before trying again. The number attempted since
557 * last failure is tracked with compact_considered.
Olivier Deprez157378f2022-04-04 15:47:50 +0200558 * compact_order_failed is the minimum compaction failed order.
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000559 */
560 unsigned int compact_considered;
561 unsigned int compact_defer_shift;
562 int compact_order_failed;
563#endif
564
565#if defined CONFIG_COMPACTION || defined CONFIG_CMA
566 /* Set to true when the PG_migrate_skip bits should be cleared */
567 bool compact_blockskip_flush;
568#endif
569
570 bool contiguous;
571
572 ZONE_PADDING(_pad3_)
573 /* Zone statistics */
574 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
575 atomic_long_t vm_numa_stat[NR_VM_NUMA_STAT_ITEMS];
576} ____cacheline_internodealigned_in_smp;
577
578enum pgdat_flags {
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000579 PGDAT_DIRTY, /* reclaim scanning has recently found
580 * many dirty file pages at the tail
581 * of the LRU.
582 */
583 PGDAT_WRITEBACK, /* reclaim scanning has recently found
584 * many pages under writeback
585 */
586 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
587};
588
David Brazdil0f672f62019-12-10 10:32:29 +0000589enum zone_flags {
590 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
591 * Cleared when kswapd is woken.
592 */
593};
594
595static inline unsigned long zone_managed_pages(struct zone *zone)
596{
597 return (unsigned long)atomic_long_read(&zone->managed_pages);
598}
599
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000600static inline unsigned long zone_end_pfn(const struct zone *zone)
601{
602 return zone->zone_start_pfn + zone->spanned_pages;
603}
604
605static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
606{
607 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
608}
609
610static inline bool zone_is_initialized(struct zone *zone)
611{
612 return zone->initialized;
613}
614
615static inline bool zone_is_empty(struct zone *zone)
616{
617 return zone->spanned_pages == 0;
618}
619
620/*
621 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
622 * intersection with the given zone
623 */
624static inline bool zone_intersects(struct zone *zone,
625 unsigned long start_pfn, unsigned long nr_pages)
626{
627 if (zone_is_empty(zone))
628 return false;
629 if (start_pfn >= zone_end_pfn(zone) ||
630 start_pfn + nr_pages <= zone->zone_start_pfn)
631 return false;
632
633 return true;
634}
635
636/*
637 * The "priority" of VM scanning is how much of the queues we will scan in one
638 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
639 * queues ("queue_length >> 12") during an aging round.
640 */
641#define DEF_PRIORITY 12
642
643/* Maximum number of zones on a zonelist */
644#define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
645
646enum {
647 ZONELIST_FALLBACK, /* zonelist with fallback */
648#ifdef CONFIG_NUMA
649 /*
650 * The NUMA zonelists are doubled because we need zonelists that
651 * restrict the allocations to a single node for __GFP_THISNODE.
652 */
653 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
654#endif
655 MAX_ZONELISTS
656};
657
658/*
659 * This struct contains information about a zone in a zonelist. It is stored
660 * here to avoid dereferences into large structures and lookups of tables
661 */
662struct zoneref {
663 struct zone *zone; /* Pointer to actual zone */
664 int zone_idx; /* zone_idx(zoneref->zone) */
665};
666
667/*
668 * One allocation request operates on a zonelist. A zonelist
669 * is a list of zones, the first one is the 'goal' of the
670 * allocation, the other zones are fallback zones, in decreasing
671 * priority.
672 *
673 * To speed the reading of the zonelist, the zonerefs contain the zone index
674 * of the entry being read. Helper functions to access information given
675 * a struct zoneref are
676 *
677 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
678 * zonelist_zone_idx() - Return the index of the zone for an entry
679 * zonelist_node_idx() - Return the index of the node for an entry
680 */
681struct zonelist {
682 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
683};
684
685#ifndef CONFIG_DISCONTIGMEM
686/* The array of struct pages - for discontigmem use pgdat->lmem_map */
687extern struct page *mem_map;
688#endif
689
David Brazdil0f672f62019-12-10 10:32:29 +0000690#ifdef CONFIG_TRANSPARENT_HUGEPAGE
691struct deferred_split {
692 spinlock_t split_queue_lock;
693 struct list_head split_queue;
694 unsigned long split_queue_len;
695};
696#endif
697
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000698/*
699 * On NUMA machines, each NUMA node would have a pg_data_t to describe
700 * it's memory layout. On UMA machines there is a single pglist_data which
701 * describes the whole memory.
702 *
703 * Memory statistics and page replacement data structures are maintained on a
704 * per-zone basis.
705 */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000706typedef struct pglist_data {
Olivier Deprez157378f2022-04-04 15:47:50 +0200707 /*
708 * node_zones contains just the zones for THIS node. Not all of the
709 * zones may be populated, but it is the full list. It is referenced by
710 * this node's node_zonelists as well as other node's node_zonelists.
711 */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000712 struct zone node_zones[MAX_NR_ZONES];
Olivier Deprez157378f2022-04-04 15:47:50 +0200713
714 /*
715 * node_zonelists contains references to all zones in all nodes.
716 * Generally the first zones will be references to this node's
717 * node_zones.
718 */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000719 struct zonelist node_zonelists[MAX_ZONELISTS];
Olivier Deprez157378f2022-04-04 15:47:50 +0200720
721 int nr_zones; /* number of populated zones in this node */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000722#ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
723 struct page *node_mem_map;
724#ifdef CONFIG_PAGE_EXTENSION
725 struct page_ext *node_page_ext;
726#endif
727#endif
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000728#if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
729 /*
David Brazdil0f672f62019-12-10 10:32:29 +0000730 * Must be held any time you expect node_start_pfn,
731 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
Olivier Deprez0e641232021-09-23 10:07:05 +0200732 * Also synchronizes pgdat->first_deferred_pfn during deferred page
733 * init.
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000734 *
735 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
736 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
737 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
738 *
739 * Nests above zone->lock and zone->span_seqlock
740 */
741 spinlock_t node_size_lock;
742#endif
743 unsigned long node_start_pfn;
744 unsigned long node_present_pages; /* total number of physical pages */
745 unsigned long node_spanned_pages; /* total size of physical page
746 range, including holes */
747 int node_id;
748 wait_queue_head_t kswapd_wait;
749 wait_queue_head_t pfmemalloc_wait;
750 struct task_struct *kswapd; /* Protected by
751 mem_hotplug_begin/end() */
752 int kswapd_order;
Olivier Deprez157378f2022-04-04 15:47:50 +0200753 enum zone_type kswapd_highest_zoneidx;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000754
755 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
756
757#ifdef CONFIG_COMPACTION
758 int kcompactd_max_order;
Olivier Deprez157378f2022-04-04 15:47:50 +0200759 enum zone_type kcompactd_highest_zoneidx;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000760 wait_queue_head_t kcompactd_wait;
761 struct task_struct *kcompactd;
762#endif
763 /*
764 * This is a per-node reserve of pages that are not available
765 * to userspace allocations.
766 */
767 unsigned long totalreserve_pages;
768
769#ifdef CONFIG_NUMA
770 /*
Olivier Deprez157378f2022-04-04 15:47:50 +0200771 * node reclaim becomes active if more unmapped pages exist.
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000772 */
773 unsigned long min_unmapped_pages;
774 unsigned long min_slab_pages;
775#endif /* CONFIG_NUMA */
776
777 /* Write-intensive fields used by page reclaim */
778 ZONE_PADDING(_pad1_)
779 spinlock_t lru_lock;
780
781#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
782 /*
783 * If memory initialisation on large machines is deferred then this
784 * is the first PFN that needs to be initialised.
785 */
786 unsigned long first_deferred_pfn;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000787#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
788
789#ifdef CONFIG_TRANSPARENT_HUGEPAGE
David Brazdil0f672f62019-12-10 10:32:29 +0000790 struct deferred_split deferred_split_queue;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000791#endif
792
793 /* Fields commonly accessed by the page reclaim scanner */
Olivier Deprez157378f2022-04-04 15:47:50 +0200794
795 /*
796 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
797 *
798 * Use mem_cgroup_lruvec() to look up lruvecs.
799 */
800 struct lruvec __lruvec;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000801
802 unsigned long flags;
803
804 ZONE_PADDING(_pad2_)
805
806 /* Per-node vmstats */
807 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
808 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
809} pg_data_t;
810
811#define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
812#define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
813#ifdef CONFIG_FLAT_NODE_MEM_MAP
814#define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
815#else
816#define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
817#endif
818#define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
819
820#define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
821#define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000822
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000823static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
824{
825 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
826}
827
828static inline bool pgdat_is_empty(pg_data_t *pgdat)
829{
830 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
831}
832
833#include <linux/memory_hotplug.h>
834
835void build_all_zonelists(pg_data_t *pgdat);
836void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
Olivier Deprez157378f2022-04-04 15:47:50 +0200837 enum zone_type highest_zoneidx);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000838bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
Olivier Deprez157378f2022-04-04 15:47:50 +0200839 int highest_zoneidx, unsigned int alloc_flags,
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000840 long free_pages);
841bool zone_watermark_ok(struct zone *z, unsigned int order,
Olivier Deprez157378f2022-04-04 15:47:50 +0200842 unsigned long mark, int highest_zoneidx,
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000843 unsigned int alloc_flags);
844bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
Olivier Deprez157378f2022-04-04 15:47:50 +0200845 unsigned long mark, int highest_zoneidx);
Olivier Deprez0e641232021-09-23 10:07:05 +0200846/*
847 * Memory initialization context, use to differentiate memory added by
848 * the platform statically or via memory hotplug interface.
849 */
850enum meminit_context {
851 MEMINIT_EARLY,
852 MEMINIT_HOTPLUG,
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000853};
Olivier Deprez0e641232021-09-23 10:07:05 +0200854
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000855extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
856 unsigned long size);
857
858extern void lruvec_init(struct lruvec *lruvec);
859
860static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
861{
862#ifdef CONFIG_MEMCG
863 return lruvec->pgdat;
864#else
Olivier Deprez157378f2022-04-04 15:47:50 +0200865 return container_of(lruvec, struct pglist_data, __lruvec);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000866#endif
867}
868
869extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru, int zone_idx);
870
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000871#ifdef CONFIG_HAVE_MEMORYLESS_NODES
872int local_memory_node(int node_id);
873#else
874static inline int local_memory_node(int node_id) { return node_id; };
875#endif
876
877/*
878 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
879 */
880#define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
881
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000882/*
883 * Returns true if a zone has pages managed by the buddy allocator.
884 * All the reclaim decisions have to use this function rather than
885 * populated_zone(). If the whole zone is reserved then we can easily
886 * end up with populated_zone() && !managed_zone().
887 */
888static inline bool managed_zone(struct zone *zone)
889{
David Brazdil0f672f62019-12-10 10:32:29 +0000890 return zone_managed_pages(zone);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000891}
892
893/* Returns true if a zone has memory */
894static inline bool populated_zone(struct zone *zone)
895{
896 return zone->present_pages;
897}
898
Olivier Deprez157378f2022-04-04 15:47:50 +0200899#ifdef CONFIG_NEED_MULTIPLE_NODES
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000900static inline int zone_to_nid(struct zone *zone)
901{
902 return zone->node;
903}
904
905static inline void zone_set_nid(struct zone *zone, int nid)
906{
907 zone->node = nid;
908}
909#else
910static inline int zone_to_nid(struct zone *zone)
911{
912 return 0;
913}
914
915static inline void zone_set_nid(struct zone *zone, int nid) {}
916#endif
917
918extern int movable_zone;
919
920#ifdef CONFIG_HIGHMEM
921static inline int zone_movable_is_highmem(void)
922{
Olivier Deprez157378f2022-04-04 15:47:50 +0200923#ifdef CONFIG_NEED_MULTIPLE_NODES
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000924 return movable_zone == ZONE_HIGHMEM;
925#else
926 return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
927#endif
928}
929#endif
930
931static inline int is_highmem_idx(enum zone_type idx)
932{
933#ifdef CONFIG_HIGHMEM
934 return (idx == ZONE_HIGHMEM ||
935 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
936#else
937 return 0;
938#endif
939}
940
Olivier Deprez157378f2022-04-04 15:47:50 +0200941#ifdef CONFIG_ZONE_DMA
942bool has_managed_dma(void);
943#else
944static inline bool has_managed_dma(void)
945{
946 return false;
947}
948#endif
949
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000950/**
David Brazdil0f672f62019-12-10 10:32:29 +0000951 * is_highmem - helper function to quickly check if a struct zone is a
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000952 * highmem zone or not. This is an attempt to keep references
953 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
954 * @zone - pointer to struct zone variable
955 */
956static inline int is_highmem(struct zone *zone)
957{
958#ifdef CONFIG_HIGHMEM
959 return is_highmem_idx(zone_idx(zone));
960#else
961 return 0;
962#endif
963}
964
965/* These two functions are used to setup the per zone pages min values */
966struct ctl_table;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000967
Olivier Deprez157378f2022-04-04 15:47:50 +0200968int min_free_kbytes_sysctl_handler(struct ctl_table *, int, void *, size_t *,
969 loff_t *);
970int watermark_scale_factor_sysctl_handler(struct ctl_table *, int, void *,
971 size_t *, loff_t *);
972extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
973int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, void *,
974 size_t *, loff_t *);
975int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
976 void *, size_t *, loff_t *);
977int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
978 void *, size_t *, loff_t *);
979int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
980 void *, size_t *, loff_t *);
981int numa_zonelist_order_handler(struct ctl_table *, int,
982 void *, size_t *, loff_t *);
983extern int percpu_pagelist_fraction;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000984extern char numa_zonelist_order[];
985#define NUMA_ZONELIST_ORDER_LEN 16
986
987#ifndef CONFIG_NEED_MULTIPLE_NODES
988
989extern struct pglist_data contig_page_data;
990#define NODE_DATA(nid) (&contig_page_data)
991#define NODE_MEM_MAP(nid) mem_map
992
993#else /* CONFIG_NEED_MULTIPLE_NODES */
994
995#include <asm/mmzone.h>
996
997#endif /* !CONFIG_NEED_MULTIPLE_NODES */
998
999extern struct pglist_data *first_online_pgdat(void);
1000extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
1001extern struct zone *next_zone(struct zone *zone);
1002
1003/**
1004 * for_each_online_pgdat - helper macro to iterate over all online nodes
1005 * @pgdat - pointer to a pg_data_t variable
1006 */
1007#define for_each_online_pgdat(pgdat) \
1008 for (pgdat = first_online_pgdat(); \
1009 pgdat; \
1010 pgdat = next_online_pgdat(pgdat))
1011/**
1012 * for_each_zone - helper macro to iterate over all memory zones
1013 * @zone - pointer to struct zone variable
1014 *
1015 * The user only needs to declare the zone variable, for_each_zone
1016 * fills it in.
1017 */
1018#define for_each_zone(zone) \
1019 for (zone = (first_online_pgdat())->node_zones; \
1020 zone; \
1021 zone = next_zone(zone))
1022
1023#define for_each_populated_zone(zone) \
1024 for (zone = (first_online_pgdat())->node_zones; \
1025 zone; \
1026 zone = next_zone(zone)) \
1027 if (!populated_zone(zone)) \
1028 ; /* do nothing */ \
1029 else
1030
1031static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1032{
1033 return zoneref->zone;
1034}
1035
1036static inline int zonelist_zone_idx(struct zoneref *zoneref)
1037{
1038 return zoneref->zone_idx;
1039}
1040
1041static inline int zonelist_node_idx(struct zoneref *zoneref)
1042{
1043 return zone_to_nid(zoneref->zone);
1044}
1045
1046struct zoneref *__next_zones_zonelist(struct zoneref *z,
1047 enum zone_type highest_zoneidx,
1048 nodemask_t *nodes);
1049
1050/**
1051 * 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
1052 * @z - The cursor used as a starting point for the search
1053 * @highest_zoneidx - The zone index of the highest zone to return
1054 * @nodes - An optional nodemask to filter the zonelist with
1055 *
1056 * This function returns the next zone at or below a given zone index that is
1057 * within the allowed nodemask using a cursor as the starting point for the
1058 * search. The zoneref returned is a cursor that represents the current zone
1059 * being examined. It should be advanced by one before calling
1060 * next_zones_zonelist again.
1061 */
1062static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1063 enum zone_type highest_zoneidx,
1064 nodemask_t *nodes)
1065{
1066 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1067 return z;
1068 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1069}
1070
1071/**
1072 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1073 * @zonelist - The zonelist to search for a suitable zone
1074 * @highest_zoneidx - The zone index of the highest zone to return
1075 * @nodes - An optional nodemask to filter the zonelist with
1076 * @return - Zoneref pointer for the first suitable zone found (see below)
1077 *
1078 * This function returns the first zone at or below a given zone index that is
1079 * within the allowed nodemask. The zoneref returned is a cursor that can be
1080 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1081 * one before calling.
1082 *
1083 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1084 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1085 * update due to cpuset modification.
1086 */
1087static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1088 enum zone_type highest_zoneidx,
1089 nodemask_t *nodes)
1090{
1091 return next_zones_zonelist(zonelist->_zonerefs,
1092 highest_zoneidx, nodes);
1093}
1094
1095/**
1096 * 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
1097 * @zone - The current zone in the iterator
Olivier Deprez157378f2022-04-04 15:47:50 +02001098 * @z - The current pointer within zonelist->_zonerefs being iterated
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001099 * @zlist - The zonelist being iterated
1100 * @highidx - The zone index of the highest zone to return
1101 * @nodemask - Nodemask allowed by the allocator
1102 *
1103 * This iterator iterates though all zones at or below a given zone index and
1104 * within a given nodemask
1105 */
1106#define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1107 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1108 zone; \
1109 z = next_zones_zonelist(++z, highidx, nodemask), \
1110 zone = zonelist_zone(z))
1111
Olivier Deprez157378f2022-04-04 15:47:50 +02001112#define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001113 for (zone = z->zone; \
1114 zone; \
1115 z = next_zones_zonelist(++z, highidx, nodemask), \
1116 zone = zonelist_zone(z))
1117
1118
1119/**
1120 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1121 * @zone - The current zone in the iterator
1122 * @z - The current pointer within zonelist->zones being iterated
1123 * @zlist - The zonelist being iterated
1124 * @highidx - The zone index of the highest zone to return
1125 *
1126 * This iterator iterates though all zones at or below a given zone index.
1127 */
1128#define for_each_zone_zonelist(zone, z, zlist, highidx) \
1129 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1130
1131#ifdef CONFIG_SPARSEMEM
1132#include <asm/sparsemem.h>
1133#endif
1134
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001135#ifdef CONFIG_FLATMEM
1136#define pfn_to_nid(pfn) (0)
1137#endif
1138
1139#ifdef CONFIG_SPARSEMEM
1140
1141/*
1142 * SECTION_SHIFT #bits space required to store a section #
1143 *
1144 * PA_SECTION_SHIFT physical address to/from section number
1145 * PFN_SECTION_SHIFT pfn to/from section number
1146 */
1147#define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1148#define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1149
1150#define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1151
1152#define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1153#define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1154
1155#define SECTION_BLOCKFLAGS_BITS \
1156 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1157
1158#if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1159#error Allocator MAX_ORDER exceeds SECTION_SIZE
1160#endif
1161
1162static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1163{
1164 return pfn >> PFN_SECTION_SHIFT;
1165}
1166static inline unsigned long section_nr_to_pfn(unsigned long sec)
1167{
1168 return sec << PFN_SECTION_SHIFT;
1169}
1170
1171#define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1172#define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1173
David Brazdil0f672f62019-12-10 10:32:29 +00001174#define SUBSECTION_SHIFT 21
Olivier Deprez157378f2022-04-04 15:47:50 +02001175#define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
David Brazdil0f672f62019-12-10 10:32:29 +00001176
1177#define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1178#define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1179#define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1180
1181#if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1182#error Subsection size exceeds section size
1183#else
1184#define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1185#endif
1186
1187#define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1188#define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1189
1190struct mem_section_usage {
Olivier Deprez157378f2022-04-04 15:47:50 +02001191#ifdef CONFIG_SPARSEMEM_VMEMMAP
David Brazdil0f672f62019-12-10 10:32:29 +00001192 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
Olivier Deprez157378f2022-04-04 15:47:50 +02001193#endif
David Brazdil0f672f62019-12-10 10:32:29 +00001194 /* See declaration of similar field in struct zone */
1195 unsigned long pageblock_flags[0];
1196};
1197
1198void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1199
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001200struct page;
1201struct page_ext;
1202struct mem_section {
1203 /*
1204 * This is, logically, a pointer to an array of struct
1205 * pages. However, it is stored with some other magic.
1206 * (see sparse.c::sparse_init_one_section())
1207 *
1208 * Additionally during early boot we encode node id of
1209 * the location of the section here to guide allocation.
1210 * (see sparse.c::memory_present())
1211 *
1212 * Making it a UL at least makes someone do a cast
1213 * before using it wrong.
1214 */
1215 unsigned long section_mem_map;
1216
David Brazdil0f672f62019-12-10 10:32:29 +00001217 struct mem_section_usage *usage;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001218#ifdef CONFIG_PAGE_EXTENSION
1219 /*
1220 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1221 * section. (see page_ext.h about this.)
1222 */
1223 struct page_ext *page_ext;
1224 unsigned long pad;
1225#endif
1226 /*
1227 * WARNING: mem_section must be a power-of-2 in size for the
1228 * calculation and use of SECTION_ROOT_MASK to make sense.
1229 */
1230};
1231
1232#ifdef CONFIG_SPARSEMEM_EXTREME
1233#define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1234#else
1235#define SECTIONS_PER_ROOT 1
1236#endif
1237
1238#define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1239#define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1240#define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1241
1242#ifdef CONFIG_SPARSEMEM_EXTREME
1243extern struct mem_section **mem_section;
1244#else
1245extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1246#endif
1247
David Brazdil0f672f62019-12-10 10:32:29 +00001248static inline unsigned long *section_to_usemap(struct mem_section *ms)
1249{
1250 return ms->usage->pageblock_flags;
1251}
1252
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001253static inline struct mem_section *__nr_to_section(unsigned long nr)
1254{
1255#ifdef CONFIG_SPARSEMEM_EXTREME
1256 if (!mem_section)
1257 return NULL;
1258#endif
1259 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1260 return NULL;
1261 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1262}
David Brazdil0f672f62019-12-10 10:32:29 +00001263extern unsigned long __section_nr(struct mem_section *ms);
1264extern size_t mem_section_usage_size(void);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001265
1266/*
1267 * We use the lower bits of the mem_map pointer to store
1268 * a little bit of information. The pointer is calculated
1269 * as mem_map - section_nr_to_pfn(pnum). The result is
1270 * aligned to the minimum alignment of the two values:
1271 * 1. All mem_map arrays are page-aligned.
1272 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1273 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1274 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1275 * worst combination is powerpc with 256k pages,
1276 * which results in PFN_SECTION_SHIFT equal 6.
1277 * To sum it up, at least 6 bits are available.
1278 */
1279#define SECTION_MARKED_PRESENT (1UL<<0)
1280#define SECTION_HAS_MEM_MAP (1UL<<1)
1281#define SECTION_IS_ONLINE (1UL<<2)
David Brazdil0f672f62019-12-10 10:32:29 +00001282#define SECTION_IS_EARLY (1UL<<3)
1283#define SECTION_MAP_LAST_BIT (1UL<<4)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001284#define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1285#define SECTION_NID_SHIFT 3
1286
1287static inline struct page *__section_mem_map_addr(struct mem_section *section)
1288{
1289 unsigned long map = section->section_mem_map;
1290 map &= SECTION_MAP_MASK;
1291 return (struct page *)map;
1292}
1293
1294static inline int present_section(struct mem_section *section)
1295{
1296 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1297}
1298
1299static inline int present_section_nr(unsigned long nr)
1300{
1301 return present_section(__nr_to_section(nr));
1302}
1303
1304static inline int valid_section(struct mem_section *section)
1305{
1306 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1307}
1308
David Brazdil0f672f62019-12-10 10:32:29 +00001309static inline int early_section(struct mem_section *section)
1310{
1311 return (section && (section->section_mem_map & SECTION_IS_EARLY));
1312}
1313
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001314static inline int valid_section_nr(unsigned long nr)
1315{
1316 return valid_section(__nr_to_section(nr));
1317}
1318
1319static inline int online_section(struct mem_section *section)
1320{
1321 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1322}
1323
1324static inline int online_section_nr(unsigned long nr)
1325{
1326 return online_section(__nr_to_section(nr));
1327}
1328
1329#ifdef CONFIG_MEMORY_HOTPLUG
1330void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1331#ifdef CONFIG_MEMORY_HOTREMOVE
1332void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1333#endif
1334#endif
1335
1336static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1337{
1338 return __nr_to_section(pfn_to_section_nr(pfn));
1339}
1340
David Brazdil0f672f62019-12-10 10:32:29 +00001341extern unsigned long __highest_present_section_nr;
1342
1343static inline int subsection_map_index(unsigned long pfn)
1344{
1345 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1346}
1347
1348#ifdef CONFIG_SPARSEMEM_VMEMMAP
1349static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1350{
1351 int idx = subsection_map_index(pfn);
1352
1353 return test_bit(idx, ms->usage->subsection_map);
1354}
1355#else
1356static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1357{
1358 return 1;
1359}
1360#endif
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001361
1362#ifndef CONFIG_HAVE_ARCH_PFN_VALID
1363static inline int pfn_valid(unsigned long pfn)
1364{
David Brazdil0f672f62019-12-10 10:32:29 +00001365 struct mem_section *ms;
1366
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001367 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1368 return 0;
David Brazdil0f672f62019-12-10 10:32:29 +00001369 ms = __nr_to_section(pfn_to_section_nr(pfn));
1370 if (!valid_section(ms))
1371 return 0;
1372 /*
1373 * Traditionally early sections always returned pfn_valid() for
1374 * the entire section-sized span.
1375 */
1376 return early_section(ms) || pfn_section_valid(ms, pfn);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001377}
1378#endif
1379
Olivier Deprez157378f2022-04-04 15:47:50 +02001380static inline int pfn_in_present_section(unsigned long pfn)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001381{
1382 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1383 return 0;
1384 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1385}
1386
Olivier Deprez157378f2022-04-04 15:47:50 +02001387static inline unsigned long next_present_section_nr(unsigned long section_nr)
1388{
1389 while (++section_nr <= __highest_present_section_nr) {
1390 if (present_section_nr(section_nr))
1391 return section_nr;
1392 }
1393
1394 return -1;
1395}
1396
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001397/*
1398 * These are _only_ used during initialisation, therefore they
1399 * can use __initdata ... They could have names to indicate
1400 * this restriction.
1401 */
1402#ifdef CONFIG_NUMA
1403#define pfn_to_nid(pfn) \
1404({ \
1405 unsigned long __pfn_to_nid_pfn = (pfn); \
1406 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1407})
1408#else
1409#define pfn_to_nid(pfn) (0)
1410#endif
1411
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001412void sparse_init(void);
1413#else
1414#define sparse_init() do {} while (0)
1415#define sparse_index_init(_sec, _nid) do {} while (0)
Olivier Deprez157378f2022-04-04 15:47:50 +02001416#define pfn_in_present_section pfn_valid
David Brazdil0f672f62019-12-10 10:32:29 +00001417#define subsection_map_init(_pfn, _nr_pages) do {} while (0)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001418#endif /* CONFIG_SPARSEMEM */
1419
1420/*
1421 * During memory init memblocks map pfns to nids. The search is expensive and
1422 * this caches recent lookups. The implementation of __early_pfn_to_nid
1423 * may treat start/end as pfns or sections.
1424 */
1425struct mminit_pfnnid_cache {
1426 unsigned long last_start;
1427 unsigned long last_end;
1428 int last_nid;
1429};
1430
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001431/*
1432 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
David Brazdil0f672f62019-12-10 10:32:29 +00001433 * need to check pfn validity within that MAX_ORDER_NR_PAGES block.
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001434 * pfn_valid_within() should be used in this case; we optimise this away
1435 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1436 */
1437#ifdef CONFIG_HOLES_IN_ZONE
1438#define pfn_valid_within(pfn) pfn_valid(pfn)
1439#else
1440#define pfn_valid_within(pfn) (1)
1441#endif
1442
1443#ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1444/*
1445 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1446 * associated with it or not. This means that a struct page exists for this
1447 * pfn. The caller cannot assume the page is fully initialized in general.
1448 * Hotplugable pages might not have been onlined yet. pfn_to_online_page()
1449 * will ensure the struct page is fully online and initialized. Special pages
1450 * (e.g. ZONE_DEVICE) are never onlined and should be treated accordingly.
1451 *
1452 * In FLATMEM, it is expected that holes always have valid memmap as long as
1453 * there is valid PFNs either side of the hole. In SPARSEMEM, it is assumed
1454 * that a valid section has a memmap for the entire section.
1455 *
1456 * However, an ARM, and maybe other embedded architectures in the future
1457 * free memmap backing holes to save memory on the assumption the memmap is
1458 * never used. The page_zone linkages are then broken even though pfn_valid()
1459 * returns true. A walker of the full memmap must then do this additional
1460 * check to ensure the memmap they are looking at is sane by making sure
1461 * the zone and PFN linkages are still valid. This is expensive, but walkers
1462 * of the full memmap are extremely rare.
1463 */
1464bool memmap_valid_within(unsigned long pfn,
1465 struct page *page, struct zone *zone);
1466#else
1467static inline bool memmap_valid_within(unsigned long pfn,
1468 struct page *page, struct zone *zone)
1469{
1470 return true;
1471}
1472#endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1473
1474#endif /* !__GENERATING_BOUNDS.H */
1475#endif /* !__ASSEMBLY__ */
1476#endif /* _LINUX_MMZONE_H */