blob: acd6f9cb01bd59c3e16e8930e6d965a6f0e6a7b5 [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001/* SPDX-License-Identifier: GPL-2.0 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002#ifndef _LINUX_MMZONE_H
3#define _LINUX_MMZONE_H
4
Linus Torvalds1da177e2005-04-16 15:20:36 -07005#ifndef __ASSEMBLY__
Christoph Lameter97965472008-04-28 02:12:54 -07006#ifndef __GENERATING_BOUNDS_H
Linus Torvalds1da177e2005-04-16 15:20:36 -07007
Linus Torvalds1da177e2005-04-16 15:20:36 -07008#include <linux/spinlock.h>
9#include <linux/list.h>
10#include <linux/wait.h>
David Rientjese815af92007-10-16 23:25:54 -070011#include <linux/bitops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/cache.h>
13#include <linux/threads.h>
14#include <linux/numa.h>
15#include <linux/init.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070016#include <linux/seqlock.h>
KAMEZAWA Hiroyuki8357f862006-03-27 01:15:57 -080017#include <linux/nodemask.h>
Mel Gorman835c1342007-10-16 01:25:47 -070018#include <linux/pageblock-flags.h>
Peter Zijlstrabbeae5b2013-02-22 16:34:30 -080019#include <linux/page-flags-layout.h>
Arun Sharma600634972011-07-26 16:09:06 -070020#include <linux/atomic.h>
Dan Williamsb03641a2019-05-14 15:41:32 -070021#include <linux/mm_types.h>
22#include <linux/page-flags.h>
Ralf Baechle93ff66b2006-06-04 02:51:29 -070023#include <asm/page.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024
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
Bob Piccoe984bb42006-05-20 15:00:31 -070031#define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
Andy Whitcroft5ad333e2007-07-17 04:03:16 -070033/*
34 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
35 * costly to service. That is between allocation orders which should
Wang YanQing35fca532012-04-15 20:42:28 +080036 * coalesce naturally under reasonable reclaim pressure and those which
Andy Whitcroft5ad333e2007-07-17 04:03:16 -070037 * will not.
38 */
39#define PAGE_ALLOC_COSTLY_ORDER 3
40
Xishi Qiua6ffdc02017-05-03 14:52:52 -070041enum migratetype {
Michal Nazarewicz47118af2011-12-29 13:09:50 +010042 MIGRATE_UNMOVABLE,
Michal Nazarewicz47118af2011-12-29 13:09:50 +010043 MIGRATE_MOVABLE,
Mel Gorman016c13d2015-11-06 16:28:18 -080044 MIGRATE_RECLAIMABLE,
Mel Gorman0aaa29a2015-11-06 16:28:37 -080045 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
46 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
Michal Nazarewicz47118af2011-12-29 13:09:50 +010047#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
Minchan Kim194159f2013-02-22 16:33:58 -080063#ifdef CONFIG_MEMORY_ISOLATION
Michal Nazarewicz47118af2011-12-29 13:09:50 +010064 MIGRATE_ISOLATE, /* can't allocate from here */
Minchan Kim194159f2013-02-22 16:33:58 -080065#endif
Michal Nazarewicz47118af2011-12-29 13:09:50 +010066 MIGRATE_TYPES
67};
68
Vlastimil Babka60f30352016-03-15 14:56:08 -070069/* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
Alexey Dobriyanc999fbd2018-12-28 00:35:55 -080070extern const char * const migratetype_names[MIGRATE_TYPES];
Vlastimil Babka60f30352016-03-15 14:56:08 -070071
Michal Nazarewicz47118af2011-12-29 13:09:50 +010072#ifdef CONFIG_CMA
73# define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
Laura Abbott7c15d9b2016-07-19 15:00:04 -070074# define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
Michal Nazarewicz47118af2011-12-29 13:09:50 +010075#else
76# define is_migrate_cma(migratetype) false
Laura Abbott7c15d9b2016-07-19 15:00:04 -070077# define is_migrate_cma_page(_page) false
Michal Nazarewicz47118af2011-12-29 13:09:50 +010078#endif
Mel Gormanb2a0ac82007-10-16 01:25:48 -070079
Vlastimil Babkab682deb2017-05-08 15:54:43 -070080static inline bool is_migrate_movable(int mt)
81{
82 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
83}
84
Mel Gormanb2a0ac82007-10-16 01:25:48 -070085#define for_each_migratetype_order(order, type) \
86 for (order = 0; order < MAX_ORDER; order++) \
87 for (type = 0; type < MIGRATE_TYPES; type++)
88
Mel Gorman467c9962007-10-16 01:26:02 -070089extern int page_group_by_mobility_disabled;
90
Mel Gormane58469b2014-06-04 16:10:16 -070091#define NR_MIGRATETYPE_BITS (PB_migrate_end - PB_migrate + 1)
92#define MIGRATETYPE_MASK ((1UL << NR_MIGRATETYPE_BITS) - 1)
93
Mel Gormandc4b0ca2014-06-04 16:10:17 -070094#define get_pageblock_migratetype(page) \
95 get_pfnblock_flags_mask(page, page_to_pfn(page), \
96 PB_migrate_end, MIGRATETYPE_MASK)
97
Linus Torvalds1da177e2005-04-16 15:20:36 -070098struct free_area {
Mel Gormanb2a0ac82007-10-16 01:25:48 -070099 struct list_head free_list[MIGRATE_TYPES];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 unsigned long nr_free;
101};
102
Dan Williamsb03641a2019-05-14 15:41:32 -0700103/* Used for pages not on another list */
104static inline void add_to_free_area(struct page *page, struct free_area *area,
105 int migratetype)
106{
107 list_add(&page->lru, &area->free_list[migratetype]);
108 area->nr_free++;
109}
110
111/* Used for pages not on another list */
112static inline void add_to_free_area_tail(struct page *page, struct free_area *area,
113 int migratetype)
114{
115 list_add_tail(&page->lru, &area->free_list[migratetype]);
116 area->nr_free++;
117}
118
119/* Used for pages which are on another list */
120static inline void move_to_free_area(struct page *page, struct free_area *area,
121 int migratetype)
122{
123 list_move(&page->lru, &area->free_list[migratetype]);
124}
125
126static inline struct page *get_page_from_free_area(struct free_area *area,
127 int migratetype)
128{
129 return list_first_entry_or_null(&area->free_list[migratetype],
130 struct page, lru);
131}
132
133static inline void del_page_from_free_area(struct page *page,
134 struct free_area *area)
135{
136 list_del(&page->lru);
137 __ClearPageBuddy(page);
138 set_page_private(page, 0);
139 area->nr_free--;
140}
141
142static inline bool free_area_empty(struct free_area *area, int migratetype)
143{
144 return list_empty(&area->free_list[migratetype]);
145}
146
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147struct pglist_data;
148
149/*
Mel Gormana52633d2016-07-28 15:45:28 -0700150 * zone->lock and the zone lru_lock are two of the hottest locks in the kernel.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 * So add a wild amount of padding here to ensure that they fall into separate
152 * cachelines. There are very few zone structures in the machine, so space
153 * consumption is not a concern here.
154 */
155#if defined(CONFIG_SMP)
156struct zone_padding {
157 char x[0];
Ravikiran G Thirumalai22fc6ec2006-01-08 01:01:27 -0800158} ____cacheline_internodealigned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159#define ZONE_PADDING(name) struct zone_padding name;
160#else
161#define ZONE_PADDING(name)
162#endif
163
Kemi Wang3a321d22017-09-08 16:12:48 -0700164#ifdef CONFIG_NUMA
165enum numa_stat_item {
166 NUMA_HIT, /* allocated in intended node */
167 NUMA_MISS, /* allocated in non intended node */
168 NUMA_FOREIGN, /* was intended here, hit elsewhere */
169 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
170 NUMA_LOCAL, /* allocation from local node */
171 NUMA_OTHER, /* allocation from other node */
172 NR_VM_NUMA_STAT_ITEMS
173};
174#else
175#define NR_VM_NUMA_STAT_ITEMS 0
176#endif
177
Christoph Lameter2244b952006-06-30 01:55:33 -0700178enum zone_stat_item {
Christoph Lameter51ed4492007-02-10 01:43:02 -0800179 /* First 128 byte cacheline (assuming 64 bit words) */
Christoph Lameterd23ad422007-02-10 01:43:02 -0800180 NR_FREE_PAGES,
Minchan Kim71c799f2016-07-28 15:47:26 -0700181 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
182 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
183 NR_ZONE_ACTIVE_ANON,
184 NR_ZONE_INACTIVE_FILE,
185 NR_ZONE_ACTIVE_FILE,
186 NR_ZONE_UNEVICTABLE,
Mel Gorman5a1c84b2016-07-28 15:47:31 -0700187 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
Nick Piggin5344b7e2008-10-18 20:26:51 -0700188 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
Christoph Lameter51ed4492007-02-10 01:43:02 -0800189 NR_PAGETABLE, /* used for pagetables */
Andy Lutomirskid30dd8b2016-07-28 15:48:14 -0700190 NR_KERNEL_STACK_KB, /* measured in KiB */
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -0700191 /* Second 128 byte cacheline */
Christoph Lameterd2c5e302006-06-30 01:55:41 -0700192 NR_BOUNCE,
Minchan Kim91537fe2016-07-26 15:24:45 -0700193#if IS_ENABLED(CONFIG_ZSMALLOC)
194 NR_ZSPAGES, /* allocated in zsmalloc */
195#endif
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -0700196 NR_FREE_CMA_PAGES,
Christoph Lameter2244b952006-06-30 01:55:33 -0700197 NR_VM_ZONE_STAT_ITEMS };
198
Mel Gorman75ef7182016-07-28 15:45:24 -0700199enum node_stat_item {
Mel Gorman599d0c92016-07-28 15:45:31 -0700200 NR_LRU_BASE,
201 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
202 NR_ACTIVE_ANON, /* " " " " " */
203 NR_INACTIVE_FILE, /* " " " " " */
204 NR_ACTIVE_FILE, /* " " " " " */
205 NR_UNEVICTABLE, /* " " " " " */
Johannes Weiner385386c2017-07-06 15:40:43 -0700206 NR_SLAB_RECLAIMABLE,
207 NR_SLAB_UNRECLAIMABLE,
Mel Gorman599d0c92016-07-28 15:45:31 -0700208 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
209 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
Johannes Weiner68d48e62018-10-26 15:06:39 -0700210 WORKINGSET_NODES,
Mel Gorman1e6b10852016-07-28 15:46:08 -0700211 WORKINGSET_REFAULT,
212 WORKINGSET_ACTIVATE,
Johannes Weiner1899ad12018-10-26 15:06:04 -0700213 WORKINGSET_RESTORE,
Mel Gorman1e6b10852016-07-28 15:46:08 -0700214 WORKINGSET_NODERECLAIM,
Mel Gorman4b9d0fa2016-07-28 15:46:17 -0700215 NR_ANON_MAPPED, /* Mapped anonymous pages */
Mel Gorman50658e22016-07-28 15:46:14 -0700216 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
217 only modified from process context */
Mel Gorman11fb9982016-07-28 15:46:20 -0700218 NR_FILE_PAGES,
219 NR_FILE_DIRTY,
220 NR_WRITEBACK,
221 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
222 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
223 NR_SHMEM_THPS,
224 NR_SHMEM_PMDMAPPED,
225 NR_ANON_THPS,
226 NR_UNSTABLE_NFS, /* NFS unstable pages */
Mel Gormanc4a25632016-07-28 15:46:23 -0700227 NR_VMSCAN_WRITE,
228 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
229 NR_DIRTIED, /* page dirtyings since bootup */
230 NR_WRITTEN, /* page writings since bootup */
Vlastimil Babkab29940c2018-10-26 15:05:46 -0700231 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
Mel Gorman75ef7182016-07-28 15:45:24 -0700232 NR_VM_NODE_STAT_ITEMS
233};
234
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700235/*
236 * We do arithmetic on the LRU lists in various places in the code,
237 * so it is important to keep the active lists LRU_ACTIVE higher in
238 * the array than the corresponding inactive lists, and to keep
239 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
240 *
241 * This has to be kept in sync with the statistics in zone_stat_item
242 * above and the descriptions in vmstat_text in mm/vmstat.c
243 */
244#define LRU_BASE 0
245#define LRU_ACTIVE 1
246#define LRU_FILE 2
247
Christoph Lameterb69408e2008-10-18 20:26:14 -0700248enum lru_list {
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700249 LRU_INACTIVE_ANON = LRU_BASE,
250 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
251 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
252 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700253 LRU_UNEVICTABLE,
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700254 NR_LRU_LISTS
255};
Christoph Lameterb69408e2008-10-18 20:26:14 -0700256
Hugh Dickins41113042012-01-12 17:20:01 -0800257#define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
Christoph Lameterb69408e2008-10-18 20:26:14 -0700258
Hugh Dickins41113042012-01-12 17:20:01 -0800259#define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700260
Hugh Dickins41113042012-01-12 17:20:01 -0800261static inline int is_file_lru(enum lru_list lru)
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700262{
Hugh Dickins41113042012-01-12 17:20:01 -0800263 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700264}
265
Hugh Dickins41113042012-01-12 17:20:01 -0800266static inline int is_active_lru(enum lru_list lru)
Christoph Lameterb69408e2008-10-18 20:26:14 -0700267{
Hugh Dickins41113042012-01-12 17:20:01 -0800268 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
Christoph Lameterb69408e2008-10-18 20:26:14 -0700269}
270
Hugh Dickins89abfab2012-05-29 15:06:53 -0700271struct zone_reclaim_stat {
272 /*
273 * The pageout code in vmscan.c keeps track of how many of the
Jiri Kosina59f91e52012-06-29 14:45:58 +0200274 * mem/swap backed and file backed pages are referenced.
Hugh Dickins89abfab2012-05-29 15:06:53 -0700275 * The higher the rotated/scanned ratio, the more valuable
276 * that cache is.
277 *
278 * The anon LRU stats live in [0], file LRU stats in [1]
279 */
280 unsigned long recent_rotated[2];
281 unsigned long recent_scanned[2];
282};
283
Johannes Weiner6290df52012-01-12 17:18:10 -0800284struct lruvec {
Johannes Weiner23047a92016-03-15 14:57:16 -0700285 struct list_head lists[NR_LRU_LISTS];
286 struct zone_reclaim_stat reclaim_stat;
287 /* Evictions & activations on the inactive file list */
288 atomic_long_t inactive_age;
Johannes Weiner2a2e4882017-05-03 14:55:03 -0700289 /* Refaults at the time of last reclaim cycle */
290 unsigned long refaults;
Andrew Mortonc255a452012-07-31 16:43:02 -0700291#ifdef CONFIG_MEMCG
Mel Gorman599d0c92016-07-28 15:45:31 -0700292 struct pglist_data *pgdat;
Konstantin Khlebnikov7f5e86c2012-05-29 15:06:58 -0700293#endif
Johannes Weiner6290df52012-01-12 17:18:10 -0800294};
295
Minchan Kimf80c0672011-10-31 17:06:55 -0700296/* Isolate unmapped file */
Konstantin Khlebnikovf3fd4a62012-05-29 15:06:54 -0700297#define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
Mel Gormanc8244932012-01-12 17:19:38 -0800298/* Isolate for asynchronous migration */
Konstantin Khlebnikovf3fd4a62012-05-29 15:06:54 -0700299#define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
Minchan Kime46a2872012-10-08 16:33:48 -0700300/* Isolate unevictable pages */
301#define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
Minchan Kim4356f212011-10-31 17:06:47 -0700302
303/* LRU Isolation modes. */
Michael S. Tsirkin9efecca2016-12-11 06:34:53 +0200304typedef unsigned __bitwise isolate_mode_t;
Minchan Kim4356f212011-10-31 17:06:47 -0700305
Mel Gorman41858962009-06-16 15:32:12 -0700306enum zone_watermarks {
307 WMARK_MIN,
308 WMARK_LOW,
309 WMARK_HIGH,
310 NR_WMARK
311};
312
Mel Gorman1c308442018-12-28 00:35:52 -0800313#define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
314#define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
315#define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
316#define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
Mel Gorman41858962009-06-16 15:32:12 -0700317
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318struct per_cpu_pages {
319 int count; /* number of pages in the list */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 int high; /* high watermark, emptying needed */
321 int batch; /* chunk size for buddy add/remove */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700322
323 /* Lists of pages, one per migrate type stored on the pcp-lists */
324 struct list_head lists[MIGRATE_PCPTYPES];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325};
326
327struct per_cpu_pageset {
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800328 struct per_cpu_pages pcp;
Christoph Lameter4037d452007-05-09 02:35:14 -0700329#ifdef CONFIG_NUMA
330 s8 expire;
Kemi Wang1d90ca82017-09-08 16:12:52 -0700331 u16 vm_numa_stat_diff[NR_VM_NUMA_STAT_ITEMS];
Christoph Lameter4037d452007-05-09 02:35:14 -0700332#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700333#ifdef CONFIG_SMP
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700334 s8 stat_threshold;
Christoph Lameter2244b952006-06-30 01:55:33 -0700335 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
336#endif
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900337};
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700338
Mel Gorman75ef7182016-07-28 15:45:24 -0700339struct per_cpu_nodestat {
340 s8 stat_threshold;
341 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
342};
343
Christoph Lameter97965472008-04-28 02:12:54 -0700344#endif /* !__GENERATING_BOUNDS.H */
345
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700346enum zone_type {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800347#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700348 /*
349 * ZONE_DMA is used when there are devices that are not able
350 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
351 * carve out the portion of memory that is needed for these devices.
352 * The range is arch specific.
353 *
354 * Some examples
355 *
356 * Architecture Limit
357 * ---------------------------
358 * parisc, ia64, sparc <4G
Christoph Hellwig25078dc2018-12-16 17:53:49 +0100359 * s390, powerpc <2G
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700360 * arm Various
361 * alpha Unlimited or 0-16MB.
362 *
363 * i386, x86_64 and multiple other arches
364 * <16M.
365 */
366 ZONE_DMA,
Christoph Lameter4b51d662007-02-10 01:43:10 -0800367#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700368#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700369 /*
370 * x86_64 needs two ZONE_DMAs because it supports devices that are
371 * only able to do DMA to the lower 16M but also 32 bit devices that
372 * can only do DMA areas below 4G.
373 */
374 ZONE_DMA32,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700375#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700376 /*
377 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
378 * performed on pages in ZONE_NORMAL if the DMA devices support
379 * transfers to all addressable memory.
380 */
381 ZONE_NORMAL,
Christoph Lametere53ef382006-09-25 23:31:14 -0700382#ifdef CONFIG_HIGHMEM
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700383 /*
384 * A memory area that is only addressable by the kernel through
385 * mapping portions into its own address space. This is for example
386 * used by i386 to allow the kernel to address the memory beyond
387 * 900MB. The kernel will set up special mappings (page
388 * table entries on i386) for each page that the kernel needs to
389 * access.
390 */
391 ZONE_HIGHMEM,
Christoph Lametere53ef382006-09-25 23:31:14 -0700392#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700393 ZONE_MOVABLE,
Dan Williams033fbae2015-08-09 15:29:06 -0400394#ifdef CONFIG_ZONE_DEVICE
395 ZONE_DEVICE,
396#endif
Christoph Lameter97965472008-04-28 02:12:54 -0700397 __MAX_NR_ZONES
Dan Williams033fbae2015-08-09 15:29:06 -0400398
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700399};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400
Christoph Lameter97965472008-04-28 02:12:54 -0700401#ifndef __GENERATING_BOUNDS_H
402
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403struct zone {
Mel Gorman3484b2d2014-08-06 16:07:14 -0700404 /* Read-mostly fields */
Mel Gorman41858962009-06-16 15:32:12 -0700405
406 /* zone watermarks, access with *_wmark_pages(zone) macros */
Mel Gormana9214442018-12-28 00:35:44 -0800407 unsigned long _watermark[NR_WMARK];
Mel Gorman1c308442018-12-28 00:35:52 -0800408 unsigned long watermark_boost;
Mel Gorman41858962009-06-16 15:32:12 -0700409
Mel Gorman0aaa29a2015-11-06 16:28:37 -0800410 unsigned long nr_reserved_highatomic;
411
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 /*
Andrew Morton89903322015-11-06 16:28:46 -0800413 * We don't know if the memory that we're going to allocate will be
414 * freeable or/and it will be released eventually, so to avoid totally
415 * wasting several GB of ram we must reserve some of the lower zone
416 * memory (otherwise we risk to run OOM on the lower zones despite
417 * there being tons of freeable ram on the higher zones). This array is
418 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
419 * changes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 */
Mel Gorman3484b2d2014-08-06 16:07:14 -0700421 long lowmem_reserve[MAX_NR_ZONES];
422
423#ifdef CONFIG_NUMA
424 int node;
425#endif
Mel Gorman3484b2d2014-08-06 16:07:14 -0700426 struct pglist_data *zone_pgdat;
427 struct per_cpu_pageset __percpu *pageset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428
Mel Gorman835c1342007-10-16 01:25:47 -0700429#ifndef CONFIG_SPARSEMEM
430 /*
Mel Gormand9c23402007-10-16 01:26:01 -0700431 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
Mel Gorman835c1342007-10-16 01:25:47 -0700432 * In SPARSEMEM, this map is stored in struct mem_section
433 */
434 unsigned long *pageblock_flags;
435#endif /* CONFIG_SPARSEMEM */
436
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
438 unsigned long zone_start_pfn;
439
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700440 /*
Jiang Liu9feedc92012-12-12 13:52:12 -0800441 * spanned_pages is the total pages spanned by the zone, including
442 * holes, which is calculated as:
443 * spanned_pages = zone_end_pfn - zone_start_pfn;
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700444 *
Jiang Liu9feedc92012-12-12 13:52:12 -0800445 * present_pages is physical pages existing within the zone, which
446 * is calculated as:
Cody P Schafer8761e312013-03-26 10:30:44 -0700447 * present_pages = spanned_pages - absent_pages(pages in holes);
Jiang Liu9feedc92012-12-12 13:52:12 -0800448 *
449 * managed_pages is present pages managed by the buddy system, which
450 * is calculated as (reserved_pages includes pages allocated by the
451 * bootmem allocator):
452 * managed_pages = present_pages - reserved_pages;
453 *
454 * So present_pages may be used by memory hotplug or memory power
455 * management logic to figure out unmanaged pages by checking
456 * (present_pages - managed_pages). And managed_pages should be used
457 * by page allocator and vm scanner to calculate all kinds of watermarks
458 * and thresholds.
459 *
460 * Locking rules:
461 *
462 * zone_start_pfn and spanned_pages are protected by span_seqlock.
463 * It is a seqlock because it has to be read outside of zone->lock,
464 * and it is done in the main allocator path. But, it is written
465 * quite infrequently.
466 *
467 * The span_seq lock is declared along with zone->lock because it is
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700468 * frequently read in proximity to zone->lock. It's good to
469 * give them a chance of being in the same cacheline.
Jiang Liu9feedc92012-12-12 13:52:12 -0800470 *
Jiang Liuc3d5f5f2013-07-03 15:03:14 -0700471 * Write access to present_pages at runtime should be protected by
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700472 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
473 * present_pages should get_online_mems() to get a stable value.
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700474 */
Arun KS9705bea2018-12-28 00:34:24 -0800475 atomic_long_t managed_pages;
Jiang Liu9feedc92012-12-12 13:52:12 -0800476 unsigned long spanned_pages;
477 unsigned long present_pages;
Mel Gorman3484b2d2014-08-06 16:07:14 -0700478
479 const char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480
Joonsoo Kimad53f922014-11-13 15:19:11 -0800481#ifdef CONFIG_MEMORY_ISOLATION
482 /*
483 * Number of isolated pageblock. It is used to solve incorrect
484 * freepage counting problem due to racy retrieving migratetype
485 * of pageblock. Protected by zone->lock.
486 */
487 unsigned long nr_isolate_pageblock;
488#endif
489
Mel Gorman3484b2d2014-08-06 16:07:14 -0700490#ifdef CONFIG_MEMORY_HOTPLUG
491 /* see spanned/present_pages for more description */
492 seqlock_t span_seqlock;
493#endif
494
Linus Torvalds9dcb8b62016-10-26 10:15:30 -0700495 int initialized;
Mel Gorman3484b2d2014-08-06 16:07:14 -0700496
Mel Gorman0f661142016-07-28 15:45:34 -0700497 /* Write-intensive fields used from the page allocator */
Mel Gorman3484b2d2014-08-06 16:07:14 -0700498 ZONE_PADDING(_pad1_)
Mel Gorman0f661142016-07-28 15:45:34 -0700499
Mel Gorman3484b2d2014-08-06 16:07:14 -0700500 /* free areas of different sizes */
501 struct free_area free_area[MAX_ORDER];
502
503 /* zone flags, see below */
504 unsigned long flags;
505
Mel Gorman0f661142016-07-28 15:45:34 -0700506 /* Primarily protects free_area */
Mel Gormana368ab62015-04-07 14:26:41 -0700507 spinlock_t lock;
508
Mel Gorman0f661142016-07-28 15:45:34 -0700509 /* Write-intensive fields used by compaction and vmstats. */
Mel Gorman3484b2d2014-08-06 16:07:14 -0700510 ZONE_PADDING(_pad2_)
511
Mel Gorman3484b2d2014-08-06 16:07:14 -0700512 /*
513 * When free pages are below this point, additional steps are taken
514 * when reading the number of free pages to avoid per-cpu counter
515 * drift allowing watermarks to be breached
516 */
517 unsigned long percpu_drift_mark;
518
519#if defined CONFIG_COMPACTION || defined CONFIG_CMA
520 /* pfn where compaction free scanner should start */
521 unsigned long compact_cached_free_pfn;
522 /* pfn where async and sync compaction migration scanner should start */
523 unsigned long compact_cached_migrate_pfn[2];
Mel Gormane332f742019-03-05 15:45:38 -0800524 unsigned long compact_init_migrate_pfn;
525 unsigned long compact_init_free_pfn;
Mel Gorman3484b2d2014-08-06 16:07:14 -0700526#endif
527
528#ifdef CONFIG_COMPACTION
529 /*
530 * On compaction failure, 1<<compact_defer_shift compactions
531 * are skipped before trying again. The number attempted since
532 * last failure is tracked with compact_considered.
533 */
534 unsigned int compact_considered;
535 unsigned int compact_defer_shift;
536 int compact_order_failed;
537#endif
538
539#if defined CONFIG_COMPACTION || defined CONFIG_CMA
540 /* Set to true when the PG_migrate_skip bits should be cleared */
541 bool compact_blockskip_flush;
542#endif
543
Joonsoo Kim7cf91a92016-03-15 14:57:51 -0700544 bool contiguous;
545
Mel Gorman3484b2d2014-08-06 16:07:14 -0700546 ZONE_PADDING(_pad3_)
547 /* Zone statistics */
548 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
Kemi Wang3a321d22017-09-08 16:12:48 -0700549 atomic_long_t vm_numa_stat[NR_VM_NUMA_STAT_ITEMS];
Ravikiran G Thirumalai22fc6ec2006-01-08 01:01:27 -0800550} ____cacheline_internodealigned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
Mel Gorman599d0c92016-07-28 15:45:31 -0700552enum pgdat_flags {
553 PGDAT_CONGESTED, /* pgdat has many dirty pages backed by
Mel Gorman0e093d992010-10-26 14:21:45 -0700554 * a congested BDI
555 */
Mel Gorman599d0c92016-07-28 15:45:31 -0700556 PGDAT_DIRTY, /* reclaim scanning has recently found
Mel Gormand43006d2013-07-03 15:01:50 -0700557 * many dirty file pages at the tail
558 * of the LRU.
559 */
Mel Gorman599d0c92016-07-28 15:45:31 -0700560 PGDAT_WRITEBACK, /* reclaim scanning has recently found
Mel Gorman283aba92013-07-03 15:01:51 -0700561 * many pages under writeback
562 */
Mel Gormana5f5f912016-07-28 15:46:32 -0700563 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
Johannes Weiner57054652014-10-09 15:28:17 -0700564};
David Rientjese815af92007-10-16 23:25:54 -0700565
Mel Gorman73444bc2019-01-08 15:23:39 -0800566enum zone_flags {
567 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
568 * Cleared when kswapd is woken.
569 */
570};
571
Arun KS9705bea2018-12-28 00:34:24 -0800572static inline unsigned long zone_managed_pages(struct zone *zone)
573{
574 return (unsigned long)atomic_long_read(&zone->managed_pages);
575}
576
Russ Andersonf9228b22013-03-22 15:04:43 -0700577static inline unsigned long zone_end_pfn(const struct zone *zone)
Cody P Schafer108bcc92013-02-22 16:35:23 -0800578{
579 return zone->zone_start_pfn + zone->spanned_pages;
580}
581
582static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
583{
584 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
585}
586
Cody P Schafer2a6e3eb2013-02-22 16:35:24 -0800587static inline bool zone_is_initialized(struct zone *zone)
588{
Linus Torvalds9dcb8b62016-10-26 10:15:30 -0700589 return zone->initialized;
Cody P Schafer2a6e3eb2013-02-22 16:35:24 -0800590}
591
592static inline bool zone_is_empty(struct zone *zone)
593{
594 return zone->spanned_pages == 0;
595}
596
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597/*
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700598 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
599 * intersection with the given zone
600 */
601static inline bool zone_intersects(struct zone *zone,
602 unsigned long start_pfn, unsigned long nr_pages)
603{
604 if (zone_is_empty(zone))
605 return false;
606 if (start_pfn >= zone_end_pfn(zone) ||
607 start_pfn + nr_pages <= zone->zone_start_pfn)
608 return false;
609
610 return true;
611}
612
613/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 * The "priority" of VM scanning is how much of the queues we will scan in one
615 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
616 * queues ("queue_length >> 12") during an aging round.
617 */
618#define DEF_PRIORITY 12
619
Paul Jackson9276b1bc2006-12-06 20:31:48 -0800620/* Maximum number of zones on a zonelist */
621#define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
622
Yaowei Baic00eb152016-01-14 15:19:00 -0800623enum {
624 ZONELIST_FALLBACK, /* zonelist with fallback */
Paul Jackson9276b1bc2006-12-06 20:31:48 -0800625#ifdef CONFIG_NUMA
Yaowei Baic00eb152016-01-14 15:19:00 -0800626 /*
627 * The NUMA zonelists are doubled because we need zonelists that
628 * restrict the allocations to a single node for __GFP_THISNODE.
629 */
630 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
Paul Jackson9276b1bc2006-12-06 20:31:48 -0800631#endif
Yaowei Baic00eb152016-01-14 15:19:00 -0800632 MAX_ZONELISTS
633};
Paul Jackson9276b1bc2006-12-06 20:31:48 -0800634
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635/*
Mel Gormandd1a2392008-04-28 02:12:17 -0700636 * This struct contains information about a zone in a zonelist. It is stored
637 * here to avoid dereferences into large structures and lookups of tables
638 */
639struct zoneref {
640 struct zone *zone; /* Pointer to actual zone */
641 int zone_idx; /* zone_idx(zoneref->zone) */
642};
643
644/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 * One allocation request operates on a zonelist. A zonelist
646 * is a list of zones, the first one is the 'goal' of the
647 * allocation, the other zones are fallback zones, in decreasing
648 * priority.
649 *
Mel Gormandd1a2392008-04-28 02:12:17 -0700650 * To speed the reading of the zonelist, the zonerefs contain the zone index
651 * of the entry being read. Helper functions to access information given
652 * a struct zoneref are
653 *
654 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
655 * zonelist_zone_idx() - Return the index of the zone for an entry
656 * zonelist_node_idx() - Return the index of the node for an entry
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 */
658struct zonelist {
Mel Gormandd1a2392008-04-28 02:12:17 -0700659 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660};
661
Heiko Carstens5b99cd02006-09-27 01:50:01 -0700662#ifndef CONFIG_DISCONTIGMEM
663/* The array of struct pages - for discontigmem use pgdat->lmem_map */
664extern struct page *mem_map;
665#endif
666
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 * On NUMA machines, each NUMA node would have a pg_data_t to describe
Nikolay Borisov618b8c22017-07-10 15:49:32 -0700669 * it's memory layout. On UMA machines there is a single pglist_data which
670 * describes the whole memory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 *
672 * Memory statistics and page replacement data structures are maintained on a
673 * per-zone basis.
674 */
675struct bootmem_data;
676typedef struct pglist_data {
677 struct zone node_zones[MAX_NR_ZONES];
Christoph Lameter523b9452007-10-16 01:25:37 -0700678 struct zonelist node_zonelists[MAX_ZONELISTS];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 int nr_zones;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700680#ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 struct page *node_mem_map;
Joonsoo Kimeefa864b2014-12-12 16:55:46 -0800682#ifdef CONFIG_PAGE_EXTENSION
683 struct page_ext *node_page_ext;
684#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700685#endif
Pavel Tatashin3a2d7fa2018-04-05 16:22:27 -0700686#if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
Dave Hansen208d54e2005-10-29 18:16:52 -0700687 /*
Wei Yangfa004ab2018-12-28 00:37:10 -0800688 * Must be held any time you expect node_start_pfn,
689 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
Dave Hansen208d54e2005-10-29 18:16:52 -0700690 *
Cody P Schafer114d4b72013-07-03 15:02:09 -0700691 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
Pavel Tatashin3a2d7fa2018-04-05 16:22:27 -0700692 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
693 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
Cody P Schafer114d4b72013-07-03 15:02:09 -0700694 *
Cody P Schafer72c3b512013-07-03 15:02:08 -0700695 * Nests above zone->lock and zone->span_seqlock
Dave Hansen208d54e2005-10-29 18:16:52 -0700696 */
697 spinlock_t node_size_lock;
698#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 unsigned long node_start_pfn;
700 unsigned long node_present_pages; /* total number of physical pages */
701 unsigned long node_spanned_pages; /* total size of physical page
702 range, including holes */
703 int node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 wait_queue_head_t kswapd_wait;
Mel Gorman55150612012-07-31 16:44:35 -0700705 wait_queue_head_t pfmemalloc_wait;
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700706 struct task_struct *kswapd; /* Protected by
707 mem_hotplug_begin/end() */
Mel Gorman38087d92016-07-28 15:45:49 -0700708 int kswapd_order;
709 enum zone_type kswapd_classzone_idx;
710
Johannes Weinerc73322d2017-05-03 14:51:51 -0700711 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
712
Vlastimil Babka698b1b32016-03-17 14:18:08 -0700713#ifdef CONFIG_COMPACTION
714 int kcompactd_max_order;
715 enum zone_type kcompactd_classzone_idx;
716 wait_queue_head_t kcompactd_wait;
717 struct task_struct *kcompactd;
718#endif
Mel Gorman281e3722016-07-28 15:46:11 -0700719 /*
720 * This is a per-node reserve of pages that are not available
721 * to userspace allocations.
722 */
723 unsigned long totalreserve_pages;
724
Mel Gormana5f5f912016-07-28 15:46:32 -0700725#ifdef CONFIG_NUMA
726 /*
727 * zone reclaim becomes active if more unmapped pages exist.
728 */
729 unsigned long min_unmapped_pages;
730 unsigned long min_slab_pages;
731#endif /* CONFIG_NUMA */
732
Mel Gormana52633d2016-07-28 15:45:28 -0700733 /* Write-intensive fields used by page reclaim */
734 ZONE_PADDING(_pad1_)
735 spinlock_t lru_lock;
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700736
737#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
738 /*
739 * If memory initialisation on large machines is deferred then this
740 * is the first PFN that needs to be initialised.
741 */
742 unsigned long first_deferred_pfn;
743#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
Kirill A. Shutemova3d0a9182016-02-02 16:57:08 -0800744
745#ifdef CONFIG_TRANSPARENT_HUGEPAGE
746 spinlock_t split_queue_lock;
747 struct list_head split_queue;
748 unsigned long split_queue_len;
749#endif
Mel Gorman75ef7182016-07-28 15:45:24 -0700750
Mel Gorman599d0c92016-07-28 15:45:31 -0700751 /* Fields commonly accessed by the page reclaim scanner */
752 struct lruvec lruvec;
753
Mel Gorman599d0c92016-07-28 15:45:31 -0700754 unsigned long flags;
755
756 ZONE_PADDING(_pad2_)
757
Mel Gorman75ef7182016-07-28 15:45:24 -0700758 /* Per-node vmstats */
759 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
760 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761} pg_data_t;
762
763#define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
764#define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700765#ifdef CONFIG_FLAT_NODE_MEM_MAP
Dave Hansen408fde82005-06-23 00:07:37 -0700766#define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700767#else
768#define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
769#endif
Dave Hansen408fde82005-06-23 00:07:37 -0700770#define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771
KAMEZAWA Hiroyukic6830c22011-06-16 17:28:07 +0900772#define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
Cody P Schaferda3649e2013-02-22 16:35:27 -0800773#define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
KAMEZAWA Hiroyukic6830c22011-06-16 17:28:07 +0900774
Mel Gormana9dd0a82016-07-28 15:46:02 -0700775static inline struct lruvec *node_lruvec(struct pglist_data *pgdat)
Mel Gorman599d0c92016-07-28 15:45:31 -0700776{
Mel Gormana9dd0a82016-07-28 15:46:02 -0700777 return &pgdat->lruvec;
Mel Gorman599d0c92016-07-28 15:45:31 -0700778}
779
Cody P Schaferda3649e2013-02-22 16:35:27 -0800780static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
781{
782 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
783}
784
785static inline bool pgdat_is_empty(pg_data_t *pgdat)
786{
787 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
788}
KAMEZAWA Hiroyukic6830c22011-06-16 17:28:07 +0900789
Dave Hansen208d54e2005-10-29 18:16:52 -0700790#include <linux/memory_hotplug.h>
791
Michal Hocko72675e12017-09-06 16:20:24 -0700792void build_all_zonelists(pg_data_t *pgdat);
David Rientjes5ecd9d42018-04-05 16:25:16 -0700793void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
794 enum zone_type classzone_idx);
Michal Hocko86a294a2016-05-20 16:57:12 -0700795bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
796 int classzone_idx, unsigned int alloc_flags,
797 long free_pages);
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700798bool zone_watermark_ok(struct zone *z, unsigned int order,
Mel Gormanc6038442016-05-19 17:13:38 -0700799 unsigned long mark, int classzone_idx,
800 unsigned int alloc_flags);
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700801bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
Mel Gormane2b19192015-11-06 16:28:09 -0800802 unsigned long mark, int classzone_idx);
Dave Hansena2f3aa022007-01-10 23:15:30 -0800803enum memmap_context {
804 MEMMAP_EARLY,
805 MEMMAP_HOTPLUG,
806};
Michal Hockodc0bbf32017-07-06 15:37:35 -0700807extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
Yaowei Baib171e402015-11-05 18:47:06 -0800808 unsigned long size);
Yasunori Goto718127c2006-06-23 02:03:10 -0700809
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800810extern void lruvec_init(struct lruvec *lruvec);
Konstantin Khlebnikov7f5e86c2012-05-29 15:06:58 -0700811
Mel Gorman599d0c92016-07-28 15:45:31 -0700812static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
Konstantin Khlebnikov7f5e86c2012-05-29 15:06:58 -0700813{
Andrew Mortonc255a452012-07-31 16:43:02 -0700814#ifdef CONFIG_MEMCG
Mel Gorman599d0c92016-07-28 15:45:31 -0700815 return lruvec->pgdat;
Konstantin Khlebnikov7f5e86c2012-05-29 15:06:58 -0700816#else
Mel Gorman599d0c92016-07-28 15:45:31 -0700817 return container_of(lruvec, struct pglist_data, lruvec);
Konstantin Khlebnikov7f5e86c2012-05-29 15:06:58 -0700818#endif
819}
820
Michal Hockofd538802017-02-22 15:45:58 -0800821extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru, int zone_idx);
Johannes Weiner23047a92016-03-15 14:57:16 -0700822
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823#ifdef CONFIG_HAVE_MEMORY_PRESENT
824void memory_present(int nid, unsigned long start, unsigned long end);
825#else
826static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
827#endif
828
Logan Gunthorpe9def36e2018-12-14 14:16:57 -0800829#if defined(CONFIG_SPARSEMEM)
830void memblocks_present(void);
831#else
832static inline void memblocks_present(void) {}
833#endif
834
Lee Schermerhorn7aac7892010-05-26 14:45:00 -0700835#ifdef CONFIG_HAVE_MEMORYLESS_NODES
836int local_memory_node(int node_id);
837#else
838static inline int local_memory_node(int node_id) { return node_id; };
839#endif
840
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841/*
842 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
843 */
844#define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
845
Oscar Salvador89696702018-08-21 21:53:24 -0700846#ifdef CONFIG_ZONE_DEVICE
847static inline bool is_dev_zone(const struct zone *zone)
848{
849 return zone_idx(zone) == ZONE_DEVICE;
850}
851#else
852static inline bool is_dev_zone(const struct zone *zone)
853{
854 return false;
855}
856#endif
857
Mel Gorman6aa303d2016-09-01 16:14:55 -0700858/*
859 * Returns true if a zone has pages managed by the buddy allocator.
860 * All the reclaim decisions have to use this function rather than
861 * populated_zone(). If the whole zone is reserved then we can easily
862 * end up with populated_zone() && !managed_zone().
863 */
864static inline bool managed_zone(struct zone *zone)
Con Kolivasf3fe6512006-01-06 00:11:15 -0800865{
Arun KS9705bea2018-12-28 00:34:24 -0800866 return zone_managed_pages(zone);
Mel Gorman6aa303d2016-09-01 16:14:55 -0700867}
868
869/* Returns true if a zone has memory */
870static inline bool populated_zone(struct zone *zone)
871{
872 return zone->present_pages;
Con Kolivasf3fe6512006-01-06 00:11:15 -0800873}
874
Pavel Tatashinc1093b72018-08-21 21:53:32 -0700875#ifdef CONFIG_NUMA
876static inline int zone_to_nid(struct zone *zone)
877{
878 return zone->node;
879}
880
881static inline void zone_set_nid(struct zone *zone, int nid)
882{
883 zone->node = nid;
884}
885#else
886static inline int zone_to_nid(struct zone *zone)
887{
888 return 0;
889}
890
891static inline void zone_set_nid(struct zone *zone, int nid) {}
892#endif
893
Mel Gorman2a1e2742007-07-17 04:03:12 -0700894extern int movable_zone;
895
Zhang Zhend7e4a2e2015-04-15 16:12:57 -0700896#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700897static inline int zone_movable_is_highmem(void)
898{
Zhang Zhend7e4a2e2015-04-15 16:12:57 -0700899#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gorman2a1e2742007-07-17 04:03:12 -0700900 return movable_zone == ZONE_HIGHMEM;
901#else
Zhang Zhend7e4a2e2015-04-15 16:12:57 -0700902 return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
Mel Gorman2a1e2742007-07-17 04:03:12 -0700903#endif
904}
Zhang Zhend7e4a2e2015-04-15 16:12:57 -0700905#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700906
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700907static inline int is_highmem_idx(enum zone_type idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908{
Christoph Lametere53ef382006-09-25 23:31:14 -0700909#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700910 return (idx == ZONE_HIGHMEM ||
911 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
Christoph Lametere53ef382006-09-25 23:31:14 -0700912#else
913 return 0;
914#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915}
916
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917/**
Mike Rapoportb4a991e2018-10-30 15:07:40 -0700918 * is_highmem - helper function to quickly check if a struct zone is a
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 * highmem zone or not. This is an attempt to keep references
920 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
921 * @zone - pointer to struct zone variable
922 */
923static inline int is_highmem(struct zone *zone)
924{
Christoph Lametere53ef382006-09-25 23:31:14 -0700925#ifdef CONFIG_HIGHMEM
Chanho Min29f9cb52016-05-19 17:11:57 -0700926 return is_highmem_idx(zone_idx(zone));
Christoph Lametere53ef382006-09-25 23:31:14 -0700927#else
928 return 0;
929#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930}
931
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932/* These two functions are used to setup the per zone pages min values */
933struct ctl_table;
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700934int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 void __user *, size_t *, loff_t *);
Mel Gorman1c308442018-12-28 00:35:52 -0800936int watermark_boost_factor_sysctl_handler(struct ctl_table *, int,
937 void __user *, size_t *, loff_t *);
Johannes Weiner795ae7a2016-03-17 14:19:14 -0700938int watermark_scale_factor_sysctl_handler(struct ctl_table *, int,
939 void __user *, size_t *, loff_t *);
Joonsoo Kimd3cda232018-04-10 16:30:11 -0700940extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700941int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 void __user *, size_t *, loff_t *);
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700943int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
Rohit Seth8ad4b1f2006-01-08 01:00:40 -0800944 void __user *, size_t *, loff_t *);
Christoph Lameter96146342006-07-03 00:24:13 -0700945int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700946 void __user *, size_t *, loff_t *);
Christoph Lameter0ff38492006-09-25 23:31:52 -0700947int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700948 void __user *, size_t *, loff_t *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -0700950extern int numa_zonelist_order_handler(struct ctl_table *, int,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700951 void __user *, size_t *, loff_t *);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -0700952extern char numa_zonelist_order[];
Michal Hockoc9bff3e2017-09-06 16:20:13 -0700953#define NUMA_ZONELIST_ORDER_LEN 16
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -0700954
Dave Hansen93b75042005-06-23 00:07:47 -0700955#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956
957extern struct pglist_data contig_page_data;
958#define NODE_DATA(nid) (&contig_page_data)
959#define NODE_MEM_MAP(nid) mem_map
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960
Dave Hansen93b75042005-06-23 00:07:47 -0700961#else /* CONFIG_NEED_MULTIPLE_NODES */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962
963#include <asm/mmzone.h>
964
Dave Hansen93b75042005-06-23 00:07:47 -0700965#endif /* !CONFIG_NEED_MULTIPLE_NODES */
Dave Hansen348f8b62005-06-23 00:07:40 -0700966
KAMEZAWA Hiroyuki95144c72006-03-27 01:16:02 -0800967extern struct pglist_data *first_online_pgdat(void);
968extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
969extern struct zone *next_zone(struct zone *zone);
KAMEZAWA Hiroyuki8357f862006-03-27 01:15:57 -0800970
971/**
Fernando Luis Vazquez Cao12d15f02008-05-23 13:05:01 -0700972 * for_each_online_pgdat - helper macro to iterate over all online nodes
KAMEZAWA Hiroyuki8357f862006-03-27 01:15:57 -0800973 * @pgdat - pointer to a pg_data_t variable
974 */
975#define for_each_online_pgdat(pgdat) \
976 for (pgdat = first_online_pgdat(); \
977 pgdat; \
978 pgdat = next_online_pgdat(pgdat))
KAMEZAWA Hiroyuki8357f862006-03-27 01:15:57 -0800979/**
980 * for_each_zone - helper macro to iterate over all memory zones
981 * @zone - pointer to struct zone variable
982 *
983 * The user only needs to declare the zone variable, for_each_zone
984 * fills it in.
985 */
986#define for_each_zone(zone) \
987 for (zone = (first_online_pgdat())->node_zones; \
988 zone; \
989 zone = next_zone(zone))
990
KOSAKI Motohiroee99c712009-03-31 15:19:31 -0700991#define for_each_populated_zone(zone) \
992 for (zone = (first_online_pgdat())->node_zones; \
993 zone; \
994 zone = next_zone(zone)) \
995 if (!populated_zone(zone)) \
996 ; /* do nothing */ \
997 else
998
Mel Gormandd1a2392008-04-28 02:12:17 -0700999static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1000{
1001 return zoneref->zone;
1002}
1003
1004static inline int zonelist_zone_idx(struct zoneref *zoneref)
1005{
1006 return zoneref->zone_idx;
1007}
1008
1009static inline int zonelist_node_idx(struct zoneref *zoneref)
1010{
Pavel Tatashinc1093b72018-08-21 21:53:32 -07001011 return zone_to_nid(zoneref->zone);
Mel Gormandd1a2392008-04-28 02:12:17 -07001012}
1013
Mel Gorman682a3382016-05-19 17:13:30 -07001014struct zoneref *__next_zones_zonelist(struct zoneref *z,
1015 enum zone_type highest_zoneidx,
1016 nodemask_t *nodes);
1017
Mel Gorman19770b32008-04-28 02:12:18 -07001018/**
1019 * 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
1020 * @z - The cursor used as a starting point for the search
1021 * @highest_zoneidx - The zone index of the highest zone to return
1022 * @nodes - An optional nodemask to filter the zonelist with
Mel Gorman19770b32008-04-28 02:12:18 -07001023 *
1024 * This function returns the next zone at or below a given zone index that is
1025 * within the allowed nodemask using a cursor as the starting point for the
Mel Gorman5bead2a2008-09-13 02:33:19 -07001026 * search. The zoneref returned is a cursor that represents the current zone
1027 * being examined. It should be advanced by one before calling
1028 * next_zones_zonelist again.
Mel Gorman19770b32008-04-28 02:12:18 -07001029 */
Mel Gorman682a3382016-05-19 17:13:30 -07001030static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
Mel Gorman19770b32008-04-28 02:12:18 -07001031 enum zone_type highest_zoneidx,
Mel Gorman682a3382016-05-19 17:13:30 -07001032 nodemask_t *nodes)
1033{
1034 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1035 return z;
1036 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1037}
Mel Gormandd1a2392008-04-28 02:12:17 -07001038
Mel Gorman19770b32008-04-28 02:12:18 -07001039/**
1040 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1041 * @zonelist - The zonelist to search for a suitable zone
1042 * @highest_zoneidx - The zone index of the highest zone to return
1043 * @nodes - An optional nodemask to filter the zonelist with
Vlastimil Babkaea574852017-01-24 15:18:32 -08001044 * @return - Zoneref pointer for the first suitable zone found (see below)
Mel Gorman19770b32008-04-28 02:12:18 -07001045 *
1046 * This function returns the first zone at or below a given zone index that is
1047 * within the allowed nodemask. The zoneref returned is a cursor that can be
Mel Gorman5bead2a2008-09-13 02:33:19 -07001048 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1049 * one before calling.
Vlastimil Babkaea574852017-01-24 15:18:32 -08001050 *
1051 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1052 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1053 * update due to cpuset modification.
Mel Gorman19770b32008-04-28 02:12:18 -07001054 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001055static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
Mel Gorman19770b32008-04-28 02:12:18 -07001056 enum zone_type highest_zoneidx,
Mel Gormanc33d6c02016-05-19 17:14:10 -07001057 nodemask_t *nodes)
Mel Gorman54a6eb52008-04-28 02:12:16 -07001058{
Mel Gormanc33d6c02016-05-19 17:14:10 -07001059 return next_zones_zonelist(zonelist->_zonerefs,
Vlastimil Babka05891fb2015-02-11 15:25:47 -08001060 highest_zoneidx, nodes);
Mel Gorman54a6eb52008-04-28 02:12:16 -07001061}
1062
Mel Gorman19770b32008-04-28 02:12:18 -07001063/**
1064 * 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
1065 * @zone - The current zone in the iterator
1066 * @z - The current pointer within zonelist->zones being iterated
1067 * @zlist - The zonelist being iterated
1068 * @highidx - The zone index of the highest zone to return
1069 * @nodemask - Nodemask allowed by the allocator
1070 *
1071 * This iterator iterates though all zones at or below a given zone index and
1072 * within a given nodemask
1073 */
1074#define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
Mel Gormanc33d6c02016-05-19 17:14:10 -07001075 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
Mel Gorman19770b32008-04-28 02:12:18 -07001076 zone; \
Vlastimil Babka05891fb2015-02-11 15:25:47 -08001077 z = next_zones_zonelist(++z, highidx, nodemask), \
Mel Gormanc33d6c02016-05-19 17:14:10 -07001078 zone = zonelist_zone(z))
1079
1080#define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1081 for (zone = z->zone; \
1082 zone; \
1083 z = next_zones_zonelist(++z, highidx, nodemask), \
1084 zone = zonelist_zone(z))
1085
Mel Gorman54a6eb52008-04-28 02:12:16 -07001086
1087/**
1088 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1089 * @zone - The current zone in the iterator
1090 * @z - The current pointer within zonelist->zones being iterated
1091 * @zlist - The zonelist being iterated
1092 * @highidx - The zone index of the highest zone to return
1093 *
1094 * This iterator iterates though all zones at or below a given zone index.
1095 */
1096#define for_each_zone_zonelist(zone, z, zlist, highidx) \
Mel Gorman19770b32008-04-28 02:12:18 -07001097 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
Mel Gorman54a6eb52008-04-28 02:12:16 -07001098
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001099#ifdef CONFIG_SPARSEMEM
1100#include <asm/sparsemem.h>
1101#endif
1102
Mel Gormanc7132162006-09-27 01:49:43 -07001103#if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
Tejun Heo0ee332c2011-12-08 10:22:09 -08001104 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
Andrew Mortonb4544562008-04-28 02:12:39 -07001105static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1106{
Michal Hocko9d1f4b32017-07-10 15:50:12 -07001107 BUILD_BUG_ON(IS_ENABLED(CONFIG_NUMA));
Andrew Mortonb4544562008-04-28 02:12:39 -07001108 return 0;
1109}
Andy Whitcroftb159d432005-06-23 00:07:52 -07001110#endif
1111
Andy Whitcroft2bdaf112006-01-06 00:10:53 -08001112#ifdef CONFIG_FLATMEM
1113#define pfn_to_nid(pfn) (0)
1114#endif
1115
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001116#ifdef CONFIG_SPARSEMEM
1117
1118/*
1119 * SECTION_SHIFT #bits space required to store a section #
1120 *
1121 * PA_SECTION_SHIFT physical address to/from section number
1122 * PFN_SECTION_SHIFT pfn to/from section number
1123 */
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001124#define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1125#define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1126
1127#define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1128
1129#define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1130#define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1131
Mel Gorman835c1342007-10-16 01:25:47 -07001132#define SECTION_BLOCKFLAGS_BITS \
Mel Gormand9c23402007-10-16 01:26:01 -07001133 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
Mel Gorman835c1342007-10-16 01:25:47 -07001134
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001135#if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1136#error Allocator MAX_ORDER exceeds SECTION_SIZE
1137#endif
1138
YASUAKI ISHIMATSU1dd2bfc2017-10-03 16:16:29 -07001139static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1140{
1141 return pfn >> PFN_SECTION_SHIFT;
1142}
1143static inline unsigned long section_nr_to_pfn(unsigned long sec)
1144{
1145 return sec << PFN_SECTION_SHIFT;
1146}
Daniel Kipere3c40f32011-05-24 17:12:33 -07001147
Daniel Kipera539f352011-05-24 17:12:51 -07001148#define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1149#define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1150
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001151struct page;
Joonsoo Kimeefa864b2014-12-12 16:55:46 -08001152struct page_ext;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001153struct mem_section {
Andy Whitcroft29751f62005-06-23 00:08:00 -07001154 /*
1155 * This is, logically, a pointer to an array of struct
1156 * pages. However, it is stored with some other magic.
1157 * (see sparse.c::sparse_init_one_section())
1158 *
Andy Whitcroft30c253e2006-06-23 02:03:41 -07001159 * Additionally during early boot we encode node id of
1160 * the location of the section here to guide allocation.
1161 * (see sparse.c::memory_present())
1162 *
Andy Whitcroft29751f62005-06-23 00:08:00 -07001163 * Making it a UL at least makes someone do a cast
1164 * before using it wrong.
1165 */
1166 unsigned long section_mem_map;
Mel Gorman5c0e3062007-10-16 01:25:56 -07001167
1168 /* See declaration of similar field in struct zone */
1169 unsigned long *pageblock_flags;
Joonsoo Kimeefa864b2014-12-12 16:55:46 -08001170#ifdef CONFIG_PAGE_EXTENSION
1171 /*
Weijie Yang0c9ad802016-05-20 16:58:04 -07001172 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
Joonsoo Kimeefa864b2014-12-12 16:55:46 -08001173 * section. (see page_ext.h about this.)
1174 */
1175 struct page_ext *page_ext;
1176 unsigned long pad;
1177#endif
Cody P Schafer55878e82013-07-03 15:04:44 -07001178 /*
1179 * WARNING: mem_section must be a power-of-2 in size for the
1180 * calculation and use of SECTION_ROOT_MASK to make sense.
1181 */
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001182};
1183
Bob Picco3e347262005-09-03 15:54:28 -07001184#ifdef CONFIG_SPARSEMEM_EXTREME
1185#define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
Bob Picco802f1922005-09-03 15:54:26 -07001186#else
Bob Picco3e347262005-09-03 15:54:28 -07001187#define SECTIONS_PER_ROOT 1
1188#endif
Bob Picco802f1922005-09-03 15:54:26 -07001189
Bob Picco3e347262005-09-03 15:54:28 -07001190#define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
Marcelo Roberto Jimenez0faa5632010-05-24 14:32:47 -07001191#define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
Bob Picco3e347262005-09-03 15:54:28 -07001192#define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1193
1194#ifdef CONFIG_SPARSEMEM_EXTREME
Kirill A. Shutemov83e3c482017-09-29 17:08:16 +03001195extern struct mem_section **mem_section;
Bob Picco3e347262005-09-03 15:54:28 -07001196#else
1197extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1198#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001199
Andy Whitcroft29751f62005-06-23 00:08:00 -07001200static inline struct mem_section *__nr_to_section(unsigned long nr)
1201{
Kirill A. Shutemov83e3c482017-09-29 17:08:16 +03001202#ifdef CONFIG_SPARSEMEM_EXTREME
1203 if (!mem_section)
1204 return NULL;
1205#endif
Bob Picco3e347262005-09-03 15:54:28 -07001206 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1207 return NULL;
1208 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
Andy Whitcroft29751f62005-06-23 00:08:00 -07001209}
Dave Hansen4ca644d2005-10-29 18:16:51 -07001210extern int __section_nr(struct mem_section* ms);
Yasunori Goto04753272008-04-28 02:13:31 -07001211extern unsigned long usemap_size(void);
Andy Whitcroft29751f62005-06-23 00:08:00 -07001212
1213/*
1214 * We use the lower bits of the mem_map pointer to store
Petr Tesarikdef9b712018-01-31 16:20:26 -08001215 * a little bit of information. The pointer is calculated
1216 * as mem_map - section_nr_to_pfn(pnum). The result is
1217 * aligned to the minimum alignment of the two values:
1218 * 1. All mem_map arrays are page-aligned.
1219 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1220 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1221 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1222 * worst combination is powerpc with 256k pages,
1223 * which results in PFN_SECTION_SHIFT equal 6.
1224 * To sum it up, at least 6 bits are available.
Andy Whitcroft29751f62005-06-23 00:08:00 -07001225 */
1226#define SECTION_MARKED_PRESENT (1UL<<0)
1227#define SECTION_HAS_MEM_MAP (1UL<<1)
Michal Hocko2d070ea2017-07-06 15:37:56 -07001228#define SECTION_IS_ONLINE (1UL<<2)
1229#define SECTION_MAP_LAST_BIT (1UL<<3)
Andy Whitcroft29751f62005-06-23 00:08:00 -07001230#define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
Michal Hocko2d070ea2017-07-06 15:37:56 -07001231#define SECTION_NID_SHIFT 3
Andy Whitcroft29751f62005-06-23 00:08:00 -07001232
1233static inline struct page *__section_mem_map_addr(struct mem_section *section)
1234{
1235 unsigned long map = section->section_mem_map;
1236 map &= SECTION_MAP_MASK;
1237 return (struct page *)map;
1238}
1239
Andy Whitcroft540557b2007-10-16 01:24:11 -07001240static inline int present_section(struct mem_section *section)
Andy Whitcroft29751f62005-06-23 00:08:00 -07001241{
Bob Picco802f1922005-09-03 15:54:26 -07001242 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
Andy Whitcroft29751f62005-06-23 00:08:00 -07001243}
1244
Andy Whitcroft540557b2007-10-16 01:24:11 -07001245static inline int present_section_nr(unsigned long nr)
1246{
1247 return present_section(__nr_to_section(nr));
1248}
1249
1250static inline int valid_section(struct mem_section *section)
Andy Whitcroft29751f62005-06-23 00:08:00 -07001251{
Bob Picco802f1922005-09-03 15:54:26 -07001252 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
Andy Whitcroft29751f62005-06-23 00:08:00 -07001253}
1254
1255static inline int valid_section_nr(unsigned long nr)
1256{
1257 return valid_section(__nr_to_section(nr));
1258}
1259
Michal Hocko2d070ea2017-07-06 15:37:56 -07001260static inline int online_section(struct mem_section *section)
1261{
1262 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1263}
1264
1265static inline int online_section_nr(unsigned long nr)
1266{
1267 return online_section(__nr_to_section(nr));
1268}
1269
1270#ifdef CONFIG_MEMORY_HOTPLUG
1271void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1272#ifdef CONFIG_MEMORY_HOTREMOVE
1273void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1274#endif
1275#endif
1276
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001277static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1278{
Andy Whitcroft29751f62005-06-23 00:08:00 -07001279 return __nr_to_section(pfn_to_section_nr(pfn));
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001280}
1281
Dave Hansenc4e1be92017-07-06 15:36:44 -07001282extern int __highest_present_section_nr;
1283
Will Deacon7b7bf492011-05-19 13:21:14 +01001284#ifndef CONFIG_HAVE_ARCH_PFN_VALID
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001285static inline int pfn_valid(unsigned long pfn)
1286{
1287 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1288 return 0;
Andy Whitcroft29751f62005-06-23 00:08:00 -07001289 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001290}
Will Deacon7b7bf492011-05-19 13:21:14 +01001291#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001292
Andy Whitcroft540557b2007-10-16 01:24:11 -07001293static inline int pfn_present(unsigned long pfn)
1294{
1295 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1296 return 0;
1297 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1298}
1299
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001300/*
1301 * These are _only_ used during initialisation, therefore they
1302 * can use __initdata ... They could have names to indicate
1303 * this restriction.
1304 */
1305#ifdef CONFIG_NUMA
Andy Whitcroft161599f2006-01-06 00:10:54 -08001306#define pfn_to_nid(pfn) \
1307({ \
1308 unsigned long __pfn_to_nid_pfn = (pfn); \
1309 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1310})
Andy Whitcroft2bdaf112006-01-06 00:10:53 -08001311#else
1312#define pfn_to_nid(pfn) (0)
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001313#endif
1314
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001315#define early_pfn_valid(pfn) pfn_valid(pfn)
1316void sparse_init(void);
1317#else
1318#define sparse_init() do {} while (0)
Dave Hansen28ae55c2005-09-03 15:54:29 -07001319#define sparse_index_init(_sec, _nid) do {} while (0)
Dan Williamse900a912019-05-14 15:41:28 -07001320#define pfn_present pfn_valid
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001321#endif /* CONFIG_SPARSEMEM */
1322
Mel Gorman8a942fd2015-06-30 14:56:55 -07001323/*
1324 * During memory init memblocks map pfns to nids. The search is expensive and
1325 * this caches recent lookups. The implementation of __early_pfn_to_nid
1326 * may treat start/end as pfns or sections.
1327 */
1328struct mminit_pfnnid_cache {
1329 unsigned long last_start;
1330 unsigned long last_end;
1331 int last_nid;
1332};
1333
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001334#ifndef early_pfn_valid
1335#define early_pfn_valid(pfn) (1)
1336#endif
1337
1338void memory_present(int nid, unsigned long start, unsigned long end);
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001339
Andy Whitcroft14e07292007-05-06 14:49:14 -07001340/*
1341 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
Wei Yang8bb4e7a2019-03-05 15:46:22 -08001342 * need to check pfn validity within that MAX_ORDER_NR_PAGES block.
Andy Whitcroft14e07292007-05-06 14:49:14 -07001343 * pfn_valid_within() should be used in this case; we optimise this away
1344 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1345 */
1346#ifdef CONFIG_HOLES_IN_ZONE
1347#define pfn_valid_within(pfn) pfn_valid(pfn)
1348#else
1349#define pfn_valid_within(pfn) (1)
1350#endif
1351
Mel Gormaneb335752009-05-13 17:34:48 +01001352#ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1353/*
1354 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
Michal Hocko2d070ea2017-07-06 15:37:56 -07001355 * associated with it or not. This means that a struct page exists for this
1356 * pfn. The caller cannot assume the page is fully initialized in general.
1357 * Hotplugable pages might not have been onlined yet. pfn_to_online_page()
1358 * will ensure the struct page is fully online and initialized. Special pages
1359 * (e.g. ZONE_DEVICE) are never onlined and should be treated accordingly.
1360 *
1361 * In FLATMEM, it is expected that holes always have valid memmap as long as
1362 * there is valid PFNs either side of the hole. In SPARSEMEM, it is assumed
1363 * that a valid section has a memmap for the entire section.
Mel Gormaneb335752009-05-13 17:34:48 +01001364 *
1365 * However, an ARM, and maybe other embedded architectures in the future
1366 * free memmap backing holes to save memory on the assumption the memmap is
1367 * never used. The page_zone linkages are then broken even though pfn_valid()
1368 * returns true. A walker of the full memmap must then do this additional
1369 * check to ensure the memmap they are looking at is sane by making sure
1370 * the zone and PFN linkages are still valid. This is expensive, but walkers
1371 * of the full memmap are extremely rare.
1372 */
Yaowei Bai5b802872016-01-14 15:19:11 -08001373bool memmap_valid_within(unsigned long pfn,
Mel Gormaneb335752009-05-13 17:34:48 +01001374 struct page *page, struct zone *zone);
1375#else
Yaowei Bai5b802872016-01-14 15:19:11 -08001376static inline bool memmap_valid_within(unsigned long pfn,
Mel Gormaneb335752009-05-13 17:34:48 +01001377 struct page *page, struct zone *zone)
1378{
Yaowei Bai5b802872016-01-14 15:19:11 -08001379 return true;
Mel Gormaneb335752009-05-13 17:34:48 +01001380}
1381#endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1382
Christoph Lameter97965472008-04-28 02:12:54 -07001383#endif /* !__GENERATING_BOUNDS.H */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384#endif /* !__ASSEMBLY__ */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385#endif /* _LINUX_MMZONE_H */