blob: be1ac8d7789be37b21dd5fdf72832aea621926bd [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001#ifndef _LINUX_MMZONE_H
2#define _LINUX_MMZONE_H
3
Linus Torvalds1da177e2005-04-16 15:20:36 -07004#ifndef __ASSEMBLY__
Christoph Lameter97965472008-04-28 02:12:54 -07005#ifndef __GENERATING_BOUNDS_H
Linus Torvalds1da177e2005-04-16 15:20:36 -07006
Linus Torvalds1da177e2005-04-16 15:20:36 -07007#include <linux/spinlock.h>
8#include <linux/list.h>
9#include <linux/wait.h>
David Rientjese815af92007-10-16 23:25:54 -070010#include <linux/bitops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/cache.h>
12#include <linux/threads.h>
13#include <linux/numa.h>
14#include <linux/init.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070015#include <linux/seqlock.h>
KAMEZAWA Hiroyuki8357f862006-03-27 01:15:57 -080016#include <linux/nodemask.h>
Mel Gorman835c1342007-10-16 01:25:47 -070017#include <linux/pageblock-flags.h>
Sam Ravnborg01fc0ac2009-04-19 21:57:19 +020018#include <generated/bounds.h>
Arun Sharma600634972011-07-26 16:09:06 -070019#include <linux/atomic.h>
Ralf Baechle93ff66b2006-06-04 02:51:29 -070020#include <asm/page.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021
22/* Free memory management - zoned buddy allocator. */
23#ifndef CONFIG_FORCE_MAX_ZONEORDER
24#define MAX_ORDER 11
25#else
26#define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
27#endif
Bob Piccoe984bb42006-05-20 15:00:31 -070028#define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -070029
Andy Whitcroft5ad333e2007-07-17 04:03:16 -070030/*
31 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
32 * costly to service. That is between allocation orders which should
33 * coelesce naturally under reasonable reclaim pressure and those which
34 * will not.
35 */
36#define PAGE_ALLOC_COSTLY_ORDER 3
37
Mel Gormanb2a0ac82007-10-16 01:25:48 -070038#define MIGRATE_UNMOVABLE 0
Mel Gormane12ba742007-10-16 01:25:52 -070039#define MIGRATE_RECLAIMABLE 1
40#define MIGRATE_MOVABLE 2
Mel Gorman5f8dcc22009-09-21 17:03:19 -070041#define MIGRATE_PCPTYPES 3 /* the number of types on the pcp lists */
Mel Gorman64c5e132007-10-16 01:25:59 -070042#define MIGRATE_RESERVE 3
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -070043#define MIGRATE_ISOLATE 4 /* can't allocate from here */
44#define MIGRATE_TYPES 5
Mel Gormanb2a0ac82007-10-16 01:25:48 -070045
46#define for_each_migratetype_order(order, type) \
47 for (order = 0; order < MAX_ORDER; order++) \
48 for (type = 0; type < MIGRATE_TYPES; type++)
49
Mel Gorman467c9962007-10-16 01:26:02 -070050extern int page_group_by_mobility_disabled;
51
52static inline int get_pageblock_migratetype(struct page *page)
53{
Mel Gorman467c9962007-10-16 01:26:02 -070054 return get_pageblock_flags_group(page, PB_migrate, PB_migrate_end);
55}
56
Linus Torvalds1da177e2005-04-16 15:20:36 -070057struct free_area {
Mel Gormanb2a0ac82007-10-16 01:25:48 -070058 struct list_head free_list[MIGRATE_TYPES];
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 unsigned long nr_free;
60};
61
62struct pglist_data;
63
64/*
65 * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
66 * So add a wild amount of padding here to ensure that they fall into separate
67 * cachelines. There are very few zone structures in the machine, so space
68 * consumption is not a concern here.
69 */
70#if defined(CONFIG_SMP)
71struct zone_padding {
72 char x[0];
Ravikiran G Thirumalai22fc6ec2006-01-08 01:01:27 -080073} ____cacheline_internodealigned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070074#define ZONE_PADDING(name) struct zone_padding name;
75#else
76#define ZONE_PADDING(name)
77#endif
78
Christoph Lameter2244b952006-06-30 01:55:33 -070079enum zone_stat_item {
Christoph Lameter51ed4492007-02-10 01:43:02 -080080 /* First 128 byte cacheline (assuming 64 bit words) */
Christoph Lameterd23ad422007-02-10 01:43:02 -080081 NR_FREE_PAGES,
Christoph Lameterb69408e2008-10-18 20:26:14 -070082 NR_LRU_BASE,
Rik van Riel4f98a2f2008-10-18 20:26:32 -070083 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
84 NR_ACTIVE_ANON, /* " " " " " */
85 NR_INACTIVE_FILE, /* " " " " " */
86 NR_ACTIVE_FILE, /* " " " " " */
Lee Schermerhorn894bc312008-10-18 20:26:39 -070087 NR_UNEVICTABLE, /* " " " " " */
Nick Piggin5344b7e2008-10-18 20:26:51 -070088 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
Christoph Lameterf3dbd342006-06-30 01:55:36 -070089 NR_ANON_PAGES, /* Mapped anonymous pages */
90 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
Christoph Lameter65ba55f2006-06-30 01:55:34 -070091 only modified from process context */
Christoph Lameter347ce432006-06-30 01:55:35 -070092 NR_FILE_PAGES,
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -070093 NR_FILE_DIRTY,
Christoph Lameterce866b32006-06-30 01:55:40 -070094 NR_WRITEBACK,
Christoph Lameter51ed4492007-02-10 01:43:02 -080095 NR_SLAB_RECLAIMABLE,
96 NR_SLAB_UNRECLAIMABLE,
97 NR_PAGETABLE, /* used for pagetables */
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -070098 NR_KERNEL_STACK,
99 /* Second 128 byte cacheline */
Christoph Lameterfd39fc82006-06-30 01:55:40 -0700100 NR_UNSTABLE_NFS, /* NFS unstable pages */
Christoph Lameterd2c5e302006-06-30 01:55:41 -0700101 NR_BOUNCE,
Andrew Mortone129b5c2006-09-27 01:50:00 -0700102 NR_VMSCAN_WRITE,
Miklos Szeredifc3ba692008-04-30 00:54:38 -0700103 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
KOSAKI Motohiroa7312862009-09-21 17:01:37 -0700104 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
105 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700106 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
Michael Rubinea941f02010-10-26 14:21:35 -0700107 NR_DIRTIED, /* page dirtyings since bootup */
108 NR_WRITTEN, /* page writings since bootup */
Christoph Lameterca889e62006-06-30 01:55:44 -0700109#ifdef CONFIG_NUMA
110 NUMA_HIT, /* allocated in intended node */
111 NUMA_MISS, /* allocated in non intended node */
112 NUMA_FOREIGN, /* was intended here, hit elsewhere */
113 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
114 NUMA_LOCAL, /* allocation from local node */
115 NUMA_OTHER, /* allocation from other node */
116#endif
Andrea Arcangeli79134172011-01-13 15:46:58 -0800117 NR_ANON_TRANSPARENT_HUGEPAGES,
Christoph Lameter2244b952006-06-30 01:55:33 -0700118 NR_VM_ZONE_STAT_ITEMS };
119
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700120/*
121 * We do arithmetic on the LRU lists in various places in the code,
122 * so it is important to keep the active lists LRU_ACTIVE higher in
123 * the array than the corresponding inactive lists, and to keep
124 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
125 *
126 * This has to be kept in sync with the statistics in zone_stat_item
127 * above and the descriptions in vmstat_text in mm/vmstat.c
128 */
129#define LRU_BASE 0
130#define LRU_ACTIVE 1
131#define LRU_FILE 2
132
Christoph Lameterb69408e2008-10-18 20:26:14 -0700133enum lru_list {
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700134 LRU_INACTIVE_ANON = LRU_BASE,
135 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
136 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
137 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700138 LRU_UNEVICTABLE,
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700139 NR_LRU_LISTS
140};
Christoph Lameterb69408e2008-10-18 20:26:14 -0700141
142#define for_each_lru(l) for (l = 0; l < NR_LRU_LISTS; l++)
143
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700144#define for_each_evictable_lru(l) for (l = 0; l <= LRU_ACTIVE_FILE; l++)
145
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700146static inline int is_file_lru(enum lru_list l)
147{
148 return (l == LRU_INACTIVE_FILE || l == LRU_ACTIVE_FILE);
149}
150
Christoph Lameterb69408e2008-10-18 20:26:14 -0700151static inline int is_active_lru(enum lru_list l)
152{
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700153 return (l == LRU_ACTIVE_ANON || l == LRU_ACTIVE_FILE);
Christoph Lameterb69408e2008-10-18 20:26:14 -0700154}
155
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700156static inline int is_unevictable_lru(enum lru_list l)
157{
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700158 return (l == LRU_UNEVICTABLE);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700159}
160
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700161/* Mask used at gathering information at once (see memcontrol.c) */
162#define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
163#define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
164#define LRU_ALL_EVICTABLE (LRU_ALL_FILE | LRU_ALL_ANON)
165#define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
166
Mel Gorman41858962009-06-16 15:32:12 -0700167enum zone_watermarks {
168 WMARK_MIN,
169 WMARK_LOW,
170 WMARK_HIGH,
171 NR_WMARK
172};
173
174#define min_wmark_pages(z) (z->watermark[WMARK_MIN])
175#define low_wmark_pages(z) (z->watermark[WMARK_LOW])
176#define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
177
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178struct per_cpu_pages {
179 int count; /* number of pages in the list */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 int high; /* high watermark, emptying needed */
181 int batch; /* chunk size for buddy add/remove */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700182
183 /* Lists of pages, one per migrate type stored on the pcp-lists */
184 struct list_head lists[MIGRATE_PCPTYPES];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185};
186
187struct per_cpu_pageset {
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800188 struct per_cpu_pages pcp;
Christoph Lameter4037d452007-05-09 02:35:14 -0700189#ifdef CONFIG_NUMA
190 s8 expire;
191#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700192#ifdef CONFIG_SMP
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700193 s8 stat_threshold;
Christoph Lameter2244b952006-06-30 01:55:33 -0700194 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
195#endif
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900196};
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700197
Christoph Lameter97965472008-04-28 02:12:54 -0700198#endif /* !__GENERATING_BOUNDS.H */
199
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700200enum zone_type {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800201#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700202 /*
203 * ZONE_DMA is used when there are devices that are not able
204 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
205 * carve out the portion of memory that is needed for these devices.
206 * The range is arch specific.
207 *
208 * Some examples
209 *
210 * Architecture Limit
211 * ---------------------------
212 * parisc, ia64, sparc <4G
213 * s390 <2G
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700214 * arm Various
215 * alpha Unlimited or 0-16MB.
216 *
217 * i386, x86_64 and multiple other arches
218 * <16M.
219 */
220 ZONE_DMA,
Christoph Lameter4b51d662007-02-10 01:43:10 -0800221#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700222#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700223 /*
224 * x86_64 needs two ZONE_DMAs because it supports devices that are
225 * only able to do DMA to the lower 16M but also 32 bit devices that
226 * can only do DMA areas below 4G.
227 */
228 ZONE_DMA32,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700229#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700230 /*
231 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
232 * performed on pages in ZONE_NORMAL if the DMA devices support
233 * transfers to all addressable memory.
234 */
235 ZONE_NORMAL,
Christoph Lametere53ef382006-09-25 23:31:14 -0700236#ifdef CONFIG_HIGHMEM
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700237 /*
238 * A memory area that is only addressable by the kernel through
239 * mapping portions into its own address space. This is for example
240 * used by i386 to allow the kernel to address the memory beyond
241 * 900MB. The kernel will set up special mappings (page
242 * table entries on i386) for each page that the kernel needs to
243 * access.
244 */
245 ZONE_HIGHMEM,
Christoph Lametere53ef382006-09-25 23:31:14 -0700246#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700247 ZONE_MOVABLE,
Christoph Lameter97965472008-04-28 02:12:54 -0700248 __MAX_NR_ZONES
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700249};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250
Christoph Lameter97965472008-04-28 02:12:54 -0700251#ifndef __GENERATING_BOUNDS_H
252
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253/*
254 * When a memory allocation must conform to specific limitations (such
255 * as being suitable for DMA) the caller will pass in hints to the
256 * allocator in the gfp_mask, in the zone modifier bits. These bits
257 * are used to select a priority ordered list of memory zones which
Christoph Lameter19655d32006-09-25 23:31:19 -0700258 * match the requested limits. See gfp_zone() in include/linux/gfp.h
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 */
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700260
Christoph Lameter97965472008-04-28 02:12:54 -0700261#if MAX_NR_ZONES < 2
Christoph Lameter4b51d662007-02-10 01:43:10 -0800262#define ZONES_SHIFT 0
Christoph Lameter97965472008-04-28 02:12:54 -0700263#elif MAX_NR_ZONES <= 2
Christoph Lameter19655d32006-09-25 23:31:19 -0700264#define ZONES_SHIFT 1
Christoph Lameter97965472008-04-28 02:12:54 -0700265#elif MAX_NR_ZONES <= 4
Christoph Lameter19655d32006-09-25 23:31:19 -0700266#define ZONES_SHIFT 2
Christoph Lameter4b51d662007-02-10 01:43:10 -0800267#else
268#error ZONES_SHIFT -- too many zones configured adjust calculation
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700269#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270
KOSAKI Motohiro6e901572009-01-07 18:08:15 -0800271struct zone_reclaim_stat {
272 /*
273 * The pageout code in vmscan.c keeps track of how many of the
274 * mem/swap backed and file backed pages are refeferenced.
275 * 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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284struct zone {
285 /* Fields commonly accessed by the page allocator */
Mel Gorman41858962009-06-16 15:32:12 -0700286
287 /* zone watermarks, access with *_wmark_pages(zone) macros */
288 unsigned long watermark[NR_WMARK];
289
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 /*
Christoph Lameteraa454842010-09-09 16:38:17 -0700291 * When free pages are below this point, additional steps are taken
292 * when reading the number of free pages to avoid per-cpu counter
293 * drift allowing watermarks to be breached
294 */
295 unsigned long percpu_drift_mark;
296
297 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 * We don't know if the memory that we're going to allocate will be freeable
299 * or/and it will be released eventually, so to avoid totally wasting several
300 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
301 * to run OOM on the lower zones despite there's tons of freeable ram
302 * on the higher zones). This array is recalculated at runtime if the
303 * sysctl_lowmem_reserve_ratio sysctl changes.
304 */
305 unsigned long lowmem_reserve[MAX_NR_ZONES];
306
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700307#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -0700308 int node;
Christoph Lameter96146342006-07-03 00:24:13 -0700309 /*
310 * zone reclaim becomes active if more unmapped pages exist.
311 */
Christoph Lameter8417bba2006-09-25 23:31:51 -0700312 unsigned long min_unmapped_pages;
Christoph Lameter0ff38492006-09-25 23:31:52 -0700313 unsigned long min_slab_pages;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700314#endif
Tejun Heo43cf38e2010-02-02 14:38:57 +0900315 struct per_cpu_pageset __percpu *pageset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 /*
317 * free areas of different sizes
318 */
319 spinlock_t lock;
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -0800320 int all_unreclaimable; /* All pages pinned */
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700321#ifdef CONFIG_MEMORY_HOTPLUG
322 /* see spanned/present_pages for more description */
323 seqlock_t span_seqlock;
324#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 struct free_area free_area[MAX_ORDER];
326
Mel Gorman835c1342007-10-16 01:25:47 -0700327#ifndef CONFIG_SPARSEMEM
328 /*
Mel Gormand9c23402007-10-16 01:26:01 -0700329 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
Mel Gorman835c1342007-10-16 01:25:47 -0700330 * In SPARSEMEM, this map is stored in struct mem_section
331 */
332 unsigned long *pageblock_flags;
333#endif /* CONFIG_SPARSEMEM */
334
Mel Gorman4f92e252010-05-24 14:32:32 -0700335#ifdef CONFIG_COMPACTION
336 /*
337 * On compaction failure, 1<<compact_defer_shift compactions
338 * are skipped before trying again. The number attempted since
339 * last failure is tracked with compact_considered.
340 */
341 unsigned int compact_considered;
342 unsigned int compact_defer_shift;
343#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344
345 ZONE_PADDING(_pad1_)
346
347 /* Fields commonly accessed by the page reclaim scanner */
348 spinlock_t lru_lock;
Wu Fengguang6e08a362009-06-16 15:32:29 -0700349 struct zone_lru {
Christoph Lameterb69408e2008-10-18 20:26:14 -0700350 struct list_head list;
Christoph Lameterb69408e2008-10-18 20:26:14 -0700351 } lru[NR_LRU_LISTS];
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700352
KOSAKI Motohiro6e901572009-01-07 18:08:15 -0800353 struct zone_reclaim_stat reclaim_stat;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700354
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 unsigned long pages_scanned; /* since last reclaim */
David Rientjese815af92007-10-16 23:25:54 -0700356 unsigned long flags; /* zone flags, see below */
Martin Hicks753ee722005-06-21 17:14:41 -0700357
Christoph Lameter2244b952006-06-30 01:55:33 -0700358 /* Zone statistics */
359 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
Christoph Lameter9eeff232006-01-18 17:42:31 -0800360
361 /*
Rik van Riel556adec2008-10-18 20:26:34 -0700362 * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on
363 * this zone's LRU. Maintained by the pageout code.
364 */
365 unsigned int inactive_ratio;
366
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367
368 ZONE_PADDING(_pad2_)
369 /* Rarely used or read-mostly fields */
370
371 /*
372 * wait_table -- the array holding the hash table
Yasunori Goto02b694d2006-06-23 02:03:08 -0700373 * wait_table_hash_nr_entries -- the size of the hash table array
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
375 *
376 * The purpose of all these is to keep track of the people
377 * waiting for a page to become available and make them
378 * runnable again when possible. The trouble is that this
379 * consumes a lot of space, especially when so few things
380 * wait on pages at a given time. So instead of using
381 * per-page waitqueues, we use a waitqueue hash table.
382 *
383 * The bucket discipline is to sleep on the same queue when
384 * colliding and wake all in that wait queue when removing.
385 * When something wakes, it must check to be sure its page is
386 * truly available, a la thundering herd. The cost of a
387 * collision is great, but given the expected load of the
388 * table, they should be so rare as to be outweighed by the
389 * benefits from the saved space.
390 *
391 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
392 * primary users of these fields, and in mm/page_alloc.c
393 * free_area_init_core() performs the initialization of them.
394 */
395 wait_queue_head_t * wait_table;
Yasunori Goto02b694d2006-06-23 02:03:08 -0700396 unsigned long wait_table_hash_nr_entries;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 unsigned long wait_table_bits;
398
399 /*
400 * Discontig memory support fields.
401 */
402 struct pglist_data *zone_pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
404 unsigned long zone_start_pfn;
405
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700406 /*
407 * zone_start_pfn, spanned_pages and present_pages are all
408 * protected by span_seqlock. It is a seqlock because it has
409 * to be read outside of zone->lock, and it is done in the main
410 * allocator path. But, it is written quite infrequently.
411 *
412 * The lock is declared along with zone->lock because it is
413 * frequently read in proximity to zone->lock. It's good to
414 * give them a chance of being in the same cacheline.
415 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 unsigned long spanned_pages; /* total size, including holes */
417 unsigned long present_pages; /* amount of memory (excluding holes) */
418
419 /*
420 * rarely used fields:
421 */
Helge Deller15ad7cd2006-12-06 20:40:36 -0800422 const char *name;
Ravikiran G Thirumalai22fc6ec2006-01-08 01:01:27 -0800423} ____cacheline_internodealigned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424
David Rientjese815af92007-10-16 23:25:54 -0700425typedef enum {
David Rientjese815af92007-10-16 23:25:54 -0700426 ZONE_RECLAIM_LOCKED, /* prevents concurrent reclaim */
David Rientjes098d7f12007-10-16 23:25:55 -0700427 ZONE_OOM_LOCKED, /* zone is in OOM killer zonelist */
Mel Gorman0e093d992010-10-26 14:21:45 -0700428 ZONE_CONGESTED, /* zone has many dirty pages backed by
429 * a congested BDI
430 */
David Rientjese815af92007-10-16 23:25:54 -0700431} zone_flags_t;
432
433static inline void zone_set_flag(struct zone *zone, zone_flags_t flag)
434{
435 set_bit(flag, &zone->flags);
436}
David Rientjesd773ed62007-10-16 23:26:01 -0700437
438static inline int zone_test_and_set_flag(struct zone *zone, zone_flags_t flag)
439{
440 return test_and_set_bit(flag, &zone->flags);
441}
442
David Rientjese815af92007-10-16 23:25:54 -0700443static inline void zone_clear_flag(struct zone *zone, zone_flags_t flag)
444{
445 clear_bit(flag, &zone->flags);
446}
447
Mel Gorman0e093d992010-10-26 14:21:45 -0700448static inline int zone_is_reclaim_congested(const struct zone *zone)
449{
450 return test_bit(ZONE_CONGESTED, &zone->flags);
451}
452
David Rientjese815af92007-10-16 23:25:54 -0700453static inline int zone_is_reclaim_locked(const struct zone *zone)
454{
455 return test_bit(ZONE_RECLAIM_LOCKED, &zone->flags);
456}
David Rientjesd773ed62007-10-16 23:26:01 -0700457
David Rientjes098d7f12007-10-16 23:25:55 -0700458static inline int zone_is_oom_locked(const struct zone *zone)
459{
460 return test_bit(ZONE_OOM_LOCKED, &zone->flags);
461}
David Rientjese815af92007-10-16 23:25:54 -0700462
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463/*
464 * The "priority" of VM scanning is how much of the queues we will scan in one
465 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
466 * queues ("queue_length >> 12") during an aging round.
467 */
468#define DEF_PRIORITY 12
469
Paul Jackson9276b1bc2006-12-06 20:31:48 -0800470/* Maximum number of zones on a zonelist */
471#define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
472
473#ifdef CONFIG_NUMA
Christoph Lameter523b9452007-10-16 01:25:37 -0700474
475/*
Pete Zaitcev25a64ec2011-02-03 22:43:48 -0700476 * The NUMA zonelists are doubled because we need zonelists that restrict the
Christoph Lameter523b9452007-10-16 01:25:37 -0700477 * allocations to a single node for GFP_THISNODE.
478 *
Mel Gorman54a6eb52008-04-28 02:12:16 -0700479 * [0] : Zonelist with fallback
480 * [1] : No fallback (GFP_THISNODE)
Christoph Lameter523b9452007-10-16 01:25:37 -0700481 */
Mel Gorman54a6eb52008-04-28 02:12:16 -0700482#define MAX_ZONELISTS 2
Christoph Lameter523b9452007-10-16 01:25:37 -0700483
484
Paul Jackson9276b1bc2006-12-06 20:31:48 -0800485/*
486 * We cache key information from each zonelist for smaller cache
487 * footprint when scanning for free pages in get_page_from_freelist().
488 *
489 * 1) The BITMAP fullzones tracks which zones in a zonelist have come
490 * up short of free memory since the last time (last_fullzone_zap)
491 * we zero'd fullzones.
492 * 2) The array z_to_n[] maps each zone in the zonelist to its node
493 * id, so that we can efficiently evaluate whether that node is
494 * set in the current tasks mems_allowed.
495 *
496 * Both fullzones and z_to_n[] are one-to-one with the zonelist,
497 * indexed by a zones offset in the zonelist zones[] array.
498 *
499 * The get_page_from_freelist() routine does two scans. During the
500 * first scan, we skip zones whose corresponding bit in 'fullzones'
501 * is set or whose corresponding node in current->mems_allowed (which
502 * comes from cpusets) is not set. During the second scan, we bypass
503 * this zonelist_cache, to ensure we look methodically at each zone.
504 *
505 * Once per second, we zero out (zap) fullzones, forcing us to
506 * reconsider nodes that might have regained more free memory.
507 * The field last_full_zap is the time we last zapped fullzones.
508 *
509 * This mechanism reduces the amount of time we waste repeatedly
510 * reexaming zones for free memory when they just came up low on
511 * memory momentarilly ago.
512 *
513 * The zonelist_cache struct members logically belong in struct
514 * zonelist. However, the mempolicy zonelists constructed for
515 * MPOL_BIND are intentionally variable length (and usually much
516 * shorter). A general purpose mechanism for handling structs with
517 * multiple variable length members is more mechanism than we want
518 * here. We resort to some special case hackery instead.
519 *
520 * The MPOL_BIND zonelists don't need this zonelist_cache (in good
521 * part because they are shorter), so we put the fixed length stuff
522 * at the front of the zonelist struct, ending in a variable length
523 * zones[], as is needed by MPOL_BIND.
524 *
525 * Then we put the optional zonelist cache on the end of the zonelist
526 * struct. This optional stuff is found by a 'zlcache_ptr' pointer in
527 * the fixed length portion at the front of the struct. This pointer
528 * both enables us to find the zonelist cache, and in the case of
529 * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL)
530 * to know that the zonelist cache is not there.
531 *
532 * The end result is that struct zonelists come in two flavors:
533 * 1) The full, fixed length version, shown below, and
534 * 2) The custom zonelists for MPOL_BIND.
535 * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache.
536 *
537 * Even though there may be multiple CPU cores on a node modifying
538 * fullzones or last_full_zap in the same zonelist_cache at the same
539 * time, we don't lock it. This is just hint data - if it is wrong now
540 * and then, the allocator will still function, perhaps a bit slower.
541 */
542
543
544struct zonelist_cache {
Paul Jackson9276b1bc2006-12-06 20:31:48 -0800545 unsigned short z_to_n[MAX_ZONES_PER_ZONELIST]; /* zone->nid */
Paul Jackson7253f4e2006-12-06 20:31:49 -0800546 DECLARE_BITMAP(fullzones, MAX_ZONES_PER_ZONELIST); /* zone full? */
Paul Jackson9276b1bc2006-12-06 20:31:48 -0800547 unsigned long last_full_zap; /* when last zap'd (jiffies) */
548};
549#else
Mel Gorman54a6eb52008-04-28 02:12:16 -0700550#define MAX_ZONELISTS 1
Paul Jackson9276b1bc2006-12-06 20:31:48 -0800551struct zonelist_cache;
552#endif
553
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554/*
Mel Gormandd1a2392008-04-28 02:12:17 -0700555 * This struct contains information about a zone in a zonelist. It is stored
556 * here to avoid dereferences into large structures and lookups of tables
557 */
558struct zoneref {
559 struct zone *zone; /* Pointer to actual zone */
560 int zone_idx; /* zone_idx(zoneref->zone) */
561};
562
563/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 * One allocation request operates on a zonelist. A zonelist
565 * is a list of zones, the first one is the 'goal' of the
566 * allocation, the other zones are fallback zones, in decreasing
567 * priority.
568 *
Paul Jackson9276b1bc2006-12-06 20:31:48 -0800569 * If zlcache_ptr is not NULL, then it is just the address of zlcache,
570 * as explained above. If zlcache_ptr is NULL, there is no zlcache.
Mel Gormandd1a2392008-04-28 02:12:17 -0700571 * *
572 * To speed the reading of the zonelist, the zonerefs contain the zone index
573 * of the entry being read. Helper functions to access information given
574 * a struct zoneref are
575 *
576 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
577 * zonelist_zone_idx() - Return the index of the zone for an entry
578 * zonelist_node_idx() - Return the index of the node for an entry
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 */
580struct zonelist {
Paul Jackson9276b1bc2006-12-06 20:31:48 -0800581 struct zonelist_cache *zlcache_ptr; // NULL or &zlcache
Mel Gormandd1a2392008-04-28 02:12:17 -0700582 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
Paul Jackson9276b1bc2006-12-06 20:31:48 -0800583#ifdef CONFIG_NUMA
584 struct zonelist_cache zlcache; // optional ...
585#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586};
587
Mel Gormanc7132162006-09-27 01:49:43 -0700588#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
589struct node_active_region {
590 unsigned long start_pfn;
591 unsigned long end_pfn;
592 int nid;
593};
594#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595
Heiko Carstens5b99cd02006-09-27 01:50:01 -0700596#ifndef CONFIG_DISCONTIGMEM
597/* The array of struct pages - for discontigmem use pgdat->lmem_map */
598extern struct page *mem_map;
599#endif
600
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601/*
602 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
603 * (mostly NUMA machines?) to denote a higher-level memory zone than the
604 * zone denotes.
605 *
606 * On NUMA machines, each NUMA node would have a pg_data_t to describe
607 * it's memory layout.
608 *
609 * Memory statistics and page replacement data structures are maintained on a
610 * per-zone basis.
611 */
612struct bootmem_data;
613typedef struct pglist_data {
614 struct zone node_zones[MAX_NR_ZONES];
Christoph Lameter523b9452007-10-16 01:25:37 -0700615 struct zonelist node_zonelists[MAX_ZONELISTS];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 int nr_zones;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700617#ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 struct page *node_mem_map;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700619#ifdef CONFIG_CGROUP_MEM_RES_CTLR
620 struct page_cgroup *node_page_cgroup;
621#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700622#endif
Yinghai Lu08677212010-02-10 01:20:20 -0800623#ifndef CONFIG_NO_BOOTMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 struct bootmem_data *bdata;
Yinghai Lu08677212010-02-10 01:20:20 -0800625#endif
Dave Hansen208d54e2005-10-29 18:16:52 -0700626#ifdef CONFIG_MEMORY_HOTPLUG
627 /*
628 * Must be held any time you expect node_start_pfn, node_present_pages
629 * or node_spanned_pages stay constant. Holding this will also
630 * guarantee that any pfn_valid() stays that way.
631 *
632 * Nests above zone->lock and zone->size_seqlock.
633 */
634 spinlock_t node_size_lock;
635#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 unsigned long node_start_pfn;
637 unsigned long node_present_pages; /* total number of physical pages */
638 unsigned long node_spanned_pages; /* total size of physical page
639 range, including holes */
640 int node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 wait_queue_head_t kswapd_wait;
642 struct task_struct *kswapd;
643 int kswapd_max_order;
Mel Gorman99504742011-01-13 15:46:20 -0800644 enum zone_type classzone_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645} pg_data_t;
646
647#define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
648#define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700649#ifdef CONFIG_FLAT_NODE_MEM_MAP
Dave Hansen408fde82005-06-23 00:07:37 -0700650#define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700651#else
652#define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
653#endif
Dave Hansen408fde82005-06-23 00:07:37 -0700654#define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655
KAMEZAWA Hiroyukic6830c22011-06-16 17:28:07 +0900656#define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
657
658#define node_end_pfn(nid) ({\
659 pg_data_t *__pgdat = NODE_DATA(nid);\
660 __pgdat->node_start_pfn + __pgdat->node_spanned_pages;\
661})
662
Dave Hansen208d54e2005-10-29 18:16:52 -0700663#include <linux/memory_hotplug.h>
664
Haicheng Li4eaf3f62010-05-24 14:32:52 -0700665extern struct mutex zonelists_mutex;
Haicheng Li1f522502010-05-24 14:32:51 -0700666void build_all_zonelists(void *data);
Mel Gorman99504742011-01-13 15:46:20 -0800667void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx);
Mel Gorman88f5acf2011-01-13 15:45:41 -0800668bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
669 int classzone_idx, int alloc_flags);
670bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800671 int classzone_idx, int alloc_flags);
Dave Hansena2f3aa022007-01-10 23:15:30 -0800672enum memmap_context {
673 MEMMAP_EARLY,
674 MEMMAP_HOTPLUG,
675};
Yasunori Goto718127c2006-06-23 02:03:10 -0700676extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
Dave Hansena2f3aa022007-01-10 23:15:30 -0800677 unsigned long size,
678 enum memmap_context context);
Yasunori Goto718127c2006-06-23 02:03:10 -0700679
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680#ifdef CONFIG_HAVE_MEMORY_PRESENT
681void memory_present(int nid, unsigned long start, unsigned long end);
682#else
683static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
684#endif
685
Lee Schermerhorn7aac7892010-05-26 14:45:00 -0700686#ifdef CONFIG_HAVE_MEMORYLESS_NODES
687int local_memory_node(int node_id);
688#else
689static inline int local_memory_node(int node_id) { return node_id; };
690#endif
691
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692#ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
693unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
694#endif
695
696/*
697 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
698 */
699#define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
700
Con Kolivasf3fe6512006-01-06 00:11:15 -0800701static inline int populated_zone(struct zone *zone)
702{
703 return (!!zone->present_pages);
704}
705
Mel Gorman2a1e2742007-07-17 04:03:12 -0700706extern int movable_zone;
707
708static inline int zone_movable_is_highmem(void)
709{
710#if defined(CONFIG_HIGHMEM) && defined(CONFIG_ARCH_POPULATES_NODE_MAP)
711 return movable_zone == ZONE_HIGHMEM;
712#else
713 return 0;
714#endif
715}
716
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700717static inline int is_highmem_idx(enum zone_type idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718{
Christoph Lametere53ef382006-09-25 23:31:14 -0700719#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700720 return (idx == ZONE_HIGHMEM ||
721 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
Christoph Lametere53ef382006-09-25 23:31:14 -0700722#else
723 return 0;
724#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725}
726
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700727static inline int is_normal_idx(enum zone_type idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728{
729 return (idx == ZONE_NORMAL);
730}
Nick Piggin9328b8f2006-01-06 00:11:10 -0800731
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732/**
733 * is_highmem - helper function to quickly check if a struct zone is a
734 * highmem zone or not. This is an attempt to keep references
735 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
736 * @zone - pointer to struct zone variable
737 */
738static inline int is_highmem(struct zone *zone)
739{
Christoph Lametere53ef382006-09-25 23:31:14 -0700740#ifdef CONFIG_HIGHMEM
Harvey Harrisonddc81ed2008-04-28 02:12:07 -0700741 int zone_off = (char *)zone - (char *)zone->zone_pgdat->node_zones;
742 return zone_off == ZONE_HIGHMEM * sizeof(*zone) ||
743 (zone_off == ZONE_MOVABLE * sizeof(*zone) &&
744 zone_movable_is_highmem());
Christoph Lametere53ef382006-09-25 23:31:14 -0700745#else
746 return 0;
747#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748}
749
750static inline int is_normal(struct zone *zone)
751{
752 return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
753}
754
Nick Piggin9328b8f2006-01-06 00:11:10 -0800755static inline int is_dma32(struct zone *zone)
756{
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700757#ifdef CONFIG_ZONE_DMA32
Nick Piggin9328b8f2006-01-06 00:11:10 -0800758 return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700759#else
760 return 0;
761#endif
Nick Piggin9328b8f2006-01-06 00:11:10 -0800762}
763
764static inline int is_dma(struct zone *zone)
765{
Christoph Lameter4b51d662007-02-10 01:43:10 -0800766#ifdef CONFIG_ZONE_DMA
Nick Piggin9328b8f2006-01-06 00:11:10 -0800767 return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
Christoph Lameter4b51d662007-02-10 01:43:10 -0800768#else
769 return 0;
770#endif
Nick Piggin9328b8f2006-01-06 00:11:10 -0800771}
772
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773/* These two functions are used to setup the per zone pages min values */
774struct ctl_table;
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700775int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 void __user *, size_t *, loff_t *);
777extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700778int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 void __user *, size_t *, loff_t *);
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700780int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
Rohit Seth8ad4b1f2006-01-08 01:00:40 -0800781 void __user *, size_t *, loff_t *);
Christoph Lameter96146342006-07-03 00:24:13 -0700782int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700783 void __user *, size_t *, loff_t *);
Christoph Lameter0ff38492006-09-25 23:31:52 -0700784int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700785 void __user *, size_t *, loff_t *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -0700787extern int numa_zonelist_order_handler(struct ctl_table *, int,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700788 void __user *, size_t *, loff_t *);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -0700789extern char numa_zonelist_order[];
790#define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */
791
Dave Hansen93b75042005-06-23 00:07:47 -0700792#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793
794extern struct pglist_data contig_page_data;
795#define NODE_DATA(nid) (&contig_page_data)
796#define NODE_MEM_MAP(nid) mem_map
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797
Dave Hansen93b75042005-06-23 00:07:47 -0700798#else /* CONFIG_NEED_MULTIPLE_NODES */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799
800#include <asm/mmzone.h>
801
Dave Hansen93b75042005-06-23 00:07:47 -0700802#endif /* !CONFIG_NEED_MULTIPLE_NODES */
Dave Hansen348f8b62005-06-23 00:07:40 -0700803
KAMEZAWA Hiroyuki95144c72006-03-27 01:16:02 -0800804extern struct pglist_data *first_online_pgdat(void);
805extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
806extern struct zone *next_zone(struct zone *zone);
KAMEZAWA Hiroyuki8357f862006-03-27 01:15:57 -0800807
808/**
Fernando Luis Vazquez Cao12d15f02008-05-23 13:05:01 -0700809 * for_each_online_pgdat - helper macro to iterate over all online nodes
KAMEZAWA Hiroyuki8357f862006-03-27 01:15:57 -0800810 * @pgdat - pointer to a pg_data_t variable
811 */
812#define for_each_online_pgdat(pgdat) \
813 for (pgdat = first_online_pgdat(); \
814 pgdat; \
815 pgdat = next_online_pgdat(pgdat))
KAMEZAWA Hiroyuki8357f862006-03-27 01:15:57 -0800816/**
817 * for_each_zone - helper macro to iterate over all memory zones
818 * @zone - pointer to struct zone variable
819 *
820 * The user only needs to declare the zone variable, for_each_zone
821 * fills it in.
822 */
823#define for_each_zone(zone) \
824 for (zone = (first_online_pgdat())->node_zones; \
825 zone; \
826 zone = next_zone(zone))
827
KOSAKI Motohiroee99c712009-03-31 15:19:31 -0700828#define for_each_populated_zone(zone) \
829 for (zone = (first_online_pgdat())->node_zones; \
830 zone; \
831 zone = next_zone(zone)) \
832 if (!populated_zone(zone)) \
833 ; /* do nothing */ \
834 else
835
Mel Gormandd1a2392008-04-28 02:12:17 -0700836static inline struct zone *zonelist_zone(struct zoneref *zoneref)
837{
838 return zoneref->zone;
839}
840
841static inline int zonelist_zone_idx(struct zoneref *zoneref)
842{
843 return zoneref->zone_idx;
844}
845
846static inline int zonelist_node_idx(struct zoneref *zoneref)
847{
848#ifdef CONFIG_NUMA
849 /* zone_to_nid not available in this context */
850 return zoneref->zone->node;
851#else
852 return 0;
853#endif /* CONFIG_NUMA */
854}
855
Mel Gorman19770b32008-04-28 02:12:18 -0700856/**
857 * 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
858 * @z - The cursor used as a starting point for the search
859 * @highest_zoneidx - The zone index of the highest zone to return
860 * @nodes - An optional nodemask to filter the zonelist with
861 * @zone - The first suitable zone found is returned via this parameter
862 *
863 * This function returns the next zone at or below a given zone index that is
864 * within the allowed nodemask using a cursor as the starting point for the
Mel Gorman5bead2a2008-09-13 02:33:19 -0700865 * search. The zoneref returned is a cursor that represents the current zone
866 * being examined. It should be advanced by one before calling
867 * next_zones_zonelist again.
Mel Gorman19770b32008-04-28 02:12:18 -0700868 */
869struct zoneref *next_zones_zonelist(struct zoneref *z,
870 enum zone_type highest_zoneidx,
871 nodemask_t *nodes,
872 struct zone **zone);
Mel Gormandd1a2392008-04-28 02:12:17 -0700873
Mel Gorman19770b32008-04-28 02:12:18 -0700874/**
875 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
876 * @zonelist - The zonelist to search for a suitable zone
877 * @highest_zoneidx - The zone index of the highest zone to return
878 * @nodes - An optional nodemask to filter the zonelist with
879 * @zone - The first suitable zone found is returned via this parameter
880 *
881 * This function returns the first zone at or below a given zone index that is
882 * within the allowed nodemask. The zoneref returned is a cursor that can be
Mel Gorman5bead2a2008-09-13 02:33:19 -0700883 * used to iterate the zonelist with next_zones_zonelist by advancing it by
884 * one before calling.
Mel Gorman19770b32008-04-28 02:12:18 -0700885 */
Mel Gormandd1a2392008-04-28 02:12:17 -0700886static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
Mel Gorman19770b32008-04-28 02:12:18 -0700887 enum zone_type highest_zoneidx,
888 nodemask_t *nodes,
889 struct zone **zone)
Mel Gorman54a6eb52008-04-28 02:12:16 -0700890{
Mel Gorman19770b32008-04-28 02:12:18 -0700891 return next_zones_zonelist(zonelist->_zonerefs, highest_zoneidx, nodes,
892 zone);
Mel Gorman54a6eb52008-04-28 02:12:16 -0700893}
894
Mel Gorman19770b32008-04-28 02:12:18 -0700895/**
896 * 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
897 * @zone - The current zone in the iterator
898 * @z - The current pointer within zonelist->zones being iterated
899 * @zlist - The zonelist being iterated
900 * @highidx - The zone index of the highest zone to return
901 * @nodemask - Nodemask allowed by the allocator
902 *
903 * This iterator iterates though all zones at or below a given zone index and
904 * within a given nodemask
905 */
906#define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
907 for (z = first_zones_zonelist(zlist, highidx, nodemask, &zone); \
908 zone; \
Mel Gorman5bead2a2008-09-13 02:33:19 -0700909 z = next_zones_zonelist(++z, highidx, nodemask, &zone)) \
Mel Gorman54a6eb52008-04-28 02:12:16 -0700910
911/**
912 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
913 * @zone - The current zone in the iterator
914 * @z - The current pointer within zonelist->zones being iterated
915 * @zlist - The zonelist being iterated
916 * @highidx - The zone index of the highest zone to return
917 *
918 * This iterator iterates though all zones at or below a given zone index.
919 */
920#define for_each_zone_zonelist(zone, z, zlist, highidx) \
Mel Gorman19770b32008-04-28 02:12:18 -0700921 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
Mel Gorman54a6eb52008-04-28 02:12:16 -0700922
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700923#ifdef CONFIG_SPARSEMEM
924#include <asm/sparsemem.h>
925#endif
926
Mel Gormanc7132162006-09-27 01:49:43 -0700927#if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
928 !defined(CONFIG_ARCH_POPULATES_NODE_MAP)
Andrew Mortonb4544562008-04-28 02:12:39 -0700929static inline unsigned long early_pfn_to_nid(unsigned long pfn)
930{
931 return 0;
932}
Andy Whitcroftb159d432005-06-23 00:07:52 -0700933#endif
934
Andy Whitcroft2bdaf112006-01-06 00:10:53 -0800935#ifdef CONFIG_FLATMEM
936#define pfn_to_nid(pfn) (0)
937#endif
938
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700939#ifdef CONFIG_SPARSEMEM
940
941/*
942 * SECTION_SHIFT #bits space required to store a section #
943 *
944 * PA_SECTION_SHIFT physical address to/from section number
945 * PFN_SECTION_SHIFT pfn to/from section number
946 */
947#define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
948
949#define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
950#define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
951
952#define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
953
954#define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
955#define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
956
Mel Gorman835c1342007-10-16 01:25:47 -0700957#define SECTION_BLOCKFLAGS_BITS \
Mel Gormand9c23402007-10-16 01:26:01 -0700958 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
Mel Gorman835c1342007-10-16 01:25:47 -0700959
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700960#if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
961#error Allocator MAX_ORDER exceeds SECTION_SIZE
962#endif
963
Daniel Kipere3c40f32011-05-24 17:12:33 -0700964#define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
965#define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
966
Daniel Kipera539f352011-05-24 17:12:51 -0700967#define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
968#define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
969
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700970struct page;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700971struct page_cgroup;
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700972struct mem_section {
Andy Whitcroft29751f62005-06-23 00:08:00 -0700973 /*
974 * This is, logically, a pointer to an array of struct
975 * pages. However, it is stored with some other magic.
976 * (see sparse.c::sparse_init_one_section())
977 *
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700978 * Additionally during early boot we encode node id of
979 * the location of the section here to guide allocation.
980 * (see sparse.c::memory_present())
981 *
Andy Whitcroft29751f62005-06-23 00:08:00 -0700982 * Making it a UL at least makes someone do a cast
983 * before using it wrong.
984 */
985 unsigned long section_mem_map;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700986
987 /* See declaration of similar field in struct zone */
988 unsigned long *pageblock_flags;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700989#ifdef CONFIG_CGROUP_MEM_RES_CTLR
990 /*
991 * If !SPARSEMEM, pgdat doesn't have page_cgroup pointer. We use
992 * section. (see memcontrol.h/page_cgroup.h about this.)
993 */
994 struct page_cgroup *page_cgroup;
995 unsigned long pad;
996#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700997};
998
Bob Picco3e347262005-09-03 15:54:28 -0700999#ifdef CONFIG_SPARSEMEM_EXTREME
1000#define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
Bob Picco802f1922005-09-03 15:54:26 -07001001#else
Bob Picco3e347262005-09-03 15:54:28 -07001002#define SECTIONS_PER_ROOT 1
1003#endif
Bob Picco802f1922005-09-03 15:54:26 -07001004
Bob Picco3e347262005-09-03 15:54:28 -07001005#define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
Marcelo Roberto Jimenez0faa5632010-05-24 14:32:47 -07001006#define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
Bob Picco3e347262005-09-03 15:54:28 -07001007#define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1008
1009#ifdef CONFIG_SPARSEMEM_EXTREME
1010extern struct mem_section *mem_section[NR_SECTION_ROOTS];
1011#else
1012extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1013#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001014
Andy Whitcroft29751f62005-06-23 00:08:00 -07001015static inline struct mem_section *__nr_to_section(unsigned long nr)
1016{
Bob Picco3e347262005-09-03 15:54:28 -07001017 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1018 return NULL;
1019 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
Andy Whitcroft29751f62005-06-23 00:08:00 -07001020}
Dave Hansen4ca644d2005-10-29 18:16:51 -07001021extern int __section_nr(struct mem_section* ms);
Yasunori Goto04753272008-04-28 02:13:31 -07001022extern unsigned long usemap_size(void);
Andy Whitcroft29751f62005-06-23 00:08:00 -07001023
1024/*
1025 * We use the lower bits of the mem_map pointer to store
1026 * a little bit of information. There should be at least
1027 * 3 bits here due to 32-bit alignment.
1028 */
1029#define SECTION_MARKED_PRESENT (1UL<<0)
1030#define SECTION_HAS_MEM_MAP (1UL<<1)
1031#define SECTION_MAP_LAST_BIT (1UL<<2)
1032#define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
Andy Whitcroft30c253e2006-06-23 02:03:41 -07001033#define SECTION_NID_SHIFT 2
Andy Whitcroft29751f62005-06-23 00:08:00 -07001034
1035static inline struct page *__section_mem_map_addr(struct mem_section *section)
1036{
1037 unsigned long map = section->section_mem_map;
1038 map &= SECTION_MAP_MASK;
1039 return (struct page *)map;
1040}
1041
Andy Whitcroft540557b2007-10-16 01:24:11 -07001042static inline int present_section(struct mem_section *section)
Andy Whitcroft29751f62005-06-23 00:08:00 -07001043{
Bob Picco802f1922005-09-03 15:54:26 -07001044 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
Andy Whitcroft29751f62005-06-23 00:08:00 -07001045}
1046
Andy Whitcroft540557b2007-10-16 01:24:11 -07001047static inline int present_section_nr(unsigned long nr)
1048{
1049 return present_section(__nr_to_section(nr));
1050}
1051
1052static inline int valid_section(struct mem_section *section)
Andy Whitcroft29751f62005-06-23 00:08:00 -07001053{
Bob Picco802f1922005-09-03 15:54:26 -07001054 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
Andy Whitcroft29751f62005-06-23 00:08:00 -07001055}
1056
1057static inline int valid_section_nr(unsigned long nr)
1058{
1059 return valid_section(__nr_to_section(nr));
1060}
1061
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001062static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1063{
Andy Whitcroft29751f62005-06-23 00:08:00 -07001064 return __nr_to_section(pfn_to_section_nr(pfn));
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001065}
1066
Will Deacon7b7bf492011-05-19 13:21:14 +01001067#ifndef CONFIG_HAVE_ARCH_PFN_VALID
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001068static inline int pfn_valid(unsigned long pfn)
1069{
1070 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1071 return 0;
Andy Whitcroft29751f62005-06-23 00:08:00 -07001072 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001073}
Will Deacon7b7bf492011-05-19 13:21:14 +01001074#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001075
Andy Whitcroft540557b2007-10-16 01:24:11 -07001076static inline int pfn_present(unsigned long pfn)
1077{
1078 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1079 return 0;
1080 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1081}
1082
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001083/*
1084 * These are _only_ used during initialisation, therefore they
1085 * can use __initdata ... They could have names to indicate
1086 * this restriction.
1087 */
1088#ifdef CONFIG_NUMA
Andy Whitcroft161599f2006-01-06 00:10:54 -08001089#define pfn_to_nid(pfn) \
1090({ \
1091 unsigned long __pfn_to_nid_pfn = (pfn); \
1092 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1093})
Andy Whitcroft2bdaf112006-01-06 00:10:53 -08001094#else
1095#define pfn_to_nid(pfn) (0)
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001096#endif
1097
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001098#define early_pfn_valid(pfn) pfn_valid(pfn)
1099void sparse_init(void);
1100#else
1101#define sparse_init() do {} while (0)
Dave Hansen28ae55c2005-09-03 15:54:29 -07001102#define sparse_index_init(_sec, _nid) do {} while (0)
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001103#endif /* CONFIG_SPARSEMEM */
1104
Andy Whitcroft75167952006-10-21 10:24:14 -07001105#ifdef CONFIG_NODES_SPAN_OTHER_NODES
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08001106bool early_pfn_in_nid(unsigned long pfn, int nid);
Andy Whitcroft75167952006-10-21 10:24:14 -07001107#else
1108#define early_pfn_in_nid(pfn, nid) (1)
1109#endif
1110
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001111#ifndef early_pfn_valid
1112#define early_pfn_valid(pfn) (1)
1113#endif
1114
1115void memory_present(int nid, unsigned long start, unsigned long end);
1116unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
1117
Andy Whitcroft14e07292007-05-06 14:49:14 -07001118/*
1119 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1120 * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1121 * pfn_valid_within() should be used in this case; we optimise this away
1122 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1123 */
1124#ifdef CONFIG_HOLES_IN_ZONE
1125#define pfn_valid_within(pfn) pfn_valid(pfn)
1126#else
1127#define pfn_valid_within(pfn) (1)
1128#endif
1129
Mel Gormaneb335752009-05-13 17:34:48 +01001130#ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1131/*
1132 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1133 * associated with it or not. In FLATMEM, it is expected that holes always
1134 * have valid memmap as long as there is valid PFNs either side of the hole.
1135 * In SPARSEMEM, it is assumed that a valid section has a memmap for the
1136 * entire section.
1137 *
1138 * However, an ARM, and maybe other embedded architectures in the future
1139 * free memmap backing holes to save memory on the assumption the memmap is
1140 * never used. The page_zone linkages are then broken even though pfn_valid()
1141 * returns true. A walker of the full memmap must then do this additional
1142 * check to ensure the memmap they are looking at is sane by making sure
1143 * the zone and PFN linkages are still valid. This is expensive, but walkers
1144 * of the full memmap are extremely rare.
1145 */
1146int memmap_valid_within(unsigned long pfn,
1147 struct page *page, struct zone *zone);
1148#else
1149static inline int memmap_valid_within(unsigned long pfn,
1150 struct page *page, struct zone *zone)
1151{
1152 return 1;
1153}
1154#endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1155
Christoph Lameter97965472008-04-28 02:12:54 -07001156#endif /* !__GENERATING_BOUNDS.H */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157#endif /* !__ASSEMBLY__ */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158#endif /* _LINUX_MMZONE_H */