blob: 428328a05fa123779a832b4b51731bee45984ea6 [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>
Christoph Lameter97965472008-04-28 02:12:54 -070018#include <linux/bounds.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <asm/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 Gorman64c5e132007-10-16 01:25:59 -070041#define MIGRATE_RESERVE 3
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -070042#define MIGRATE_ISOLATE 4 /* can't allocate from here */
43#define MIGRATE_TYPES 5
Mel Gormanb2a0ac82007-10-16 01:25:48 -070044
45#define for_each_migratetype_order(order, type) \
46 for (order = 0; order < MAX_ORDER; order++) \
47 for (type = 0; type < MIGRATE_TYPES; type++)
48
Mel Gorman467c9962007-10-16 01:26:02 -070049extern int page_group_by_mobility_disabled;
50
51static inline int get_pageblock_migratetype(struct page *page)
52{
53 if (unlikely(page_group_by_mobility_disabled))
54 return MIGRATE_UNMOVABLE;
55
56 return get_pageblock_flags_group(page, PB_migrate, PB_migrate_end);
57}
58
Linus Torvalds1da177e2005-04-16 15:20:36 -070059struct free_area {
Mel Gormanb2a0ac82007-10-16 01:25:48 -070060 struct list_head free_list[MIGRATE_TYPES];
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 unsigned long nr_free;
62};
63
64struct pglist_data;
65
66/*
67 * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
68 * So add a wild amount of padding here to ensure that they fall into separate
69 * cachelines. There are very few zone structures in the machine, so space
70 * consumption is not a concern here.
71 */
72#if defined(CONFIG_SMP)
73struct zone_padding {
74 char x[0];
Ravikiran G Thirumalai22fc6ec2006-01-08 01:01:27 -080075} ____cacheline_internodealigned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070076#define ZONE_PADDING(name) struct zone_padding name;
77#else
78#define ZONE_PADDING(name)
79#endif
80
Christoph Lameter2244b952006-06-30 01:55:33 -070081enum zone_stat_item {
Christoph Lameter51ed4492007-02-10 01:43:02 -080082 /* First 128 byte cacheline (assuming 64 bit words) */
Christoph Lameterd23ad422007-02-10 01:43:02 -080083 NR_FREE_PAGES,
Christoph Lameterc8785382007-02-10 01:43:01 -080084 NR_INACTIVE,
85 NR_ACTIVE,
Christoph Lameterf3dbd342006-06-30 01:55:36 -070086 NR_ANON_PAGES, /* Mapped anonymous pages */
87 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
Christoph Lameter65ba55f2006-06-30 01:55:34 -070088 only modified from process context */
Christoph Lameter347ce432006-06-30 01:55:35 -070089 NR_FILE_PAGES,
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -070090 NR_FILE_DIRTY,
Christoph Lameterce866b32006-06-30 01:55:40 -070091 NR_WRITEBACK,
Christoph Lameter51ed4492007-02-10 01:43:02 -080092 /* Second 128 byte cacheline */
93 NR_SLAB_RECLAIMABLE,
94 NR_SLAB_UNRECLAIMABLE,
95 NR_PAGETABLE, /* used for pagetables */
Christoph Lameterfd39fc82006-06-30 01:55:40 -070096 NR_UNSTABLE_NFS, /* NFS unstable pages */
Christoph Lameterd2c5e302006-06-30 01:55:41 -070097 NR_BOUNCE,
Andrew Mortone129b5c2006-09-27 01:50:00 -070098 NR_VMSCAN_WRITE,
Miklos Szeredifc3ba692008-04-30 00:54:38 -070099 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
Christoph Lameterca889e62006-06-30 01:55:44 -0700100#ifdef CONFIG_NUMA
101 NUMA_HIT, /* allocated in intended node */
102 NUMA_MISS, /* allocated in non intended node */
103 NUMA_FOREIGN, /* was intended here, hit elsewhere */
104 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
105 NUMA_LOCAL, /* allocation from local node */
106 NUMA_OTHER, /* allocation from other node */
107#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700108 NR_VM_ZONE_STAT_ITEMS };
109
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110struct per_cpu_pages {
111 int count; /* number of pages in the list */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 int high; /* high watermark, emptying needed */
113 int batch; /* chunk size for buddy add/remove */
114 struct list_head list; /* the list of pages */
115};
116
117struct per_cpu_pageset {
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800118 struct per_cpu_pages pcp;
Christoph Lameter4037d452007-05-09 02:35:14 -0700119#ifdef CONFIG_NUMA
120 s8 expire;
121#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700122#ifdef CONFIG_SMP
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700123 s8 stat_threshold;
Christoph Lameter2244b952006-06-30 01:55:33 -0700124 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
125#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126} ____cacheline_aligned_in_smp;
127
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700128#ifdef CONFIG_NUMA
129#define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)])
130#else
131#define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)])
132#endif
133
Christoph Lameter97965472008-04-28 02:12:54 -0700134#endif /* !__GENERATING_BOUNDS.H */
135
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700136enum zone_type {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800137#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700138 /*
139 * ZONE_DMA is used when there are devices that are not able
140 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
141 * carve out the portion of memory that is needed for these devices.
142 * The range is arch specific.
143 *
144 * Some examples
145 *
146 * Architecture Limit
147 * ---------------------------
148 * parisc, ia64, sparc <4G
149 * s390 <2G
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700150 * arm Various
151 * alpha Unlimited or 0-16MB.
152 *
153 * i386, x86_64 and multiple other arches
154 * <16M.
155 */
156 ZONE_DMA,
Christoph Lameter4b51d662007-02-10 01:43:10 -0800157#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700158#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700159 /*
160 * x86_64 needs two ZONE_DMAs because it supports devices that are
161 * only able to do DMA to the lower 16M but also 32 bit devices that
162 * can only do DMA areas below 4G.
163 */
164 ZONE_DMA32,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700165#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700166 /*
167 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
168 * performed on pages in ZONE_NORMAL if the DMA devices support
169 * transfers to all addressable memory.
170 */
171 ZONE_NORMAL,
Christoph Lametere53ef382006-09-25 23:31:14 -0700172#ifdef CONFIG_HIGHMEM
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700173 /*
174 * A memory area that is only addressable by the kernel through
175 * mapping portions into its own address space. This is for example
176 * used by i386 to allow the kernel to address the memory beyond
177 * 900MB. The kernel will set up special mappings (page
178 * table entries on i386) for each page that the kernel needs to
179 * access.
180 */
181 ZONE_HIGHMEM,
Christoph Lametere53ef382006-09-25 23:31:14 -0700182#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700183 ZONE_MOVABLE,
Christoph Lameter97965472008-04-28 02:12:54 -0700184 __MAX_NR_ZONES
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700185};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186
Christoph Lameter97965472008-04-28 02:12:54 -0700187#ifndef __GENERATING_BOUNDS_H
188
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189/*
190 * When a memory allocation must conform to specific limitations (such
191 * as being suitable for DMA) the caller will pass in hints to the
192 * allocator in the gfp_mask, in the zone modifier bits. These bits
193 * are used to select a priority ordered list of memory zones which
Christoph Lameter19655d32006-09-25 23:31:19 -0700194 * match the requested limits. See gfp_zone() in include/linux/gfp.h
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 */
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700196
Christoph Lameter97965472008-04-28 02:12:54 -0700197#if MAX_NR_ZONES < 2
Christoph Lameter4b51d662007-02-10 01:43:10 -0800198#define ZONES_SHIFT 0
Christoph Lameter97965472008-04-28 02:12:54 -0700199#elif MAX_NR_ZONES <= 2
Christoph Lameter19655d32006-09-25 23:31:19 -0700200#define ZONES_SHIFT 1
Christoph Lameter97965472008-04-28 02:12:54 -0700201#elif MAX_NR_ZONES <= 4
Christoph Lameter19655d32006-09-25 23:31:19 -0700202#define ZONES_SHIFT 2
Christoph Lameter4b51d662007-02-10 01:43:10 -0800203#else
204#error ZONES_SHIFT -- too many zones configured adjust calculation
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700205#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207struct zone {
208 /* Fields commonly accessed by the page allocator */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 unsigned long pages_min, pages_low, pages_high;
210 /*
211 * We don't know if the memory that we're going to allocate will be freeable
212 * or/and it will be released eventually, so to avoid totally wasting several
213 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
214 * to run OOM on the lower zones despite there's tons of freeable ram
215 * on the higher zones). This array is recalculated at runtime if the
216 * sysctl_lowmem_reserve_ratio sysctl changes.
217 */
218 unsigned long lowmem_reserve[MAX_NR_ZONES];
219
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700220#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -0700221 int node;
Christoph Lameter96146342006-07-03 00:24:13 -0700222 /*
223 * zone reclaim becomes active if more unmapped pages exist.
224 */
Christoph Lameter8417bba2006-09-25 23:31:51 -0700225 unsigned long min_unmapped_pages;
Christoph Lameter0ff38492006-09-25 23:31:52 -0700226 unsigned long min_slab_pages;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700227 struct per_cpu_pageset *pageset[NR_CPUS];
228#else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 struct per_cpu_pageset pageset[NR_CPUS];
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700230#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 /*
232 * free areas of different sizes
233 */
234 spinlock_t lock;
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700235#ifdef CONFIG_MEMORY_HOTPLUG
236 /* see spanned/present_pages for more description */
237 seqlock_t span_seqlock;
238#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 struct free_area free_area[MAX_ORDER];
240
Mel Gorman835c1342007-10-16 01:25:47 -0700241#ifndef CONFIG_SPARSEMEM
242 /*
Mel Gormand9c23402007-10-16 01:26:01 -0700243 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
Mel Gorman835c1342007-10-16 01:25:47 -0700244 * In SPARSEMEM, this map is stored in struct mem_section
245 */
246 unsigned long *pageblock_flags;
247#endif /* CONFIG_SPARSEMEM */
248
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249
250 ZONE_PADDING(_pad1_)
251
252 /* Fields commonly accessed by the page reclaim scanner */
253 spinlock_t lru_lock;
254 struct list_head active_list;
255 struct list_head inactive_list;
256 unsigned long nr_scan_active;
257 unsigned long nr_scan_inactive;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 unsigned long pages_scanned; /* since last reclaim */
David Rientjese815af92007-10-16 23:25:54 -0700259 unsigned long flags; /* zone flags, see below */
Martin Hicks753ee722005-06-21 17:14:41 -0700260
Christoph Lameter2244b952006-06-30 01:55:33 -0700261 /* Zone statistics */
262 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
Christoph Lameter9eeff232006-01-18 17:42:31 -0800263
264 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 * prev_priority holds the scanning priority for this zone. It is
266 * defined as the scanning priority at which we achieved our reclaim
267 * target at the previous try_to_free_pages() or balance_pgdat()
268 * invokation.
269 *
270 * We use prev_priority as a measure of how much stress page reclaim is
271 * under - it drives the swappiness decision: whether to unmap mapped
272 * pages.
273 *
Martin Bligh3bb1a8522006-10-28 10:38:24 -0700274 * Access to both this field is quite racy even on uniprocessor. But
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 * it is expected to average out OK.
276 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 int prev_priority;
278
279
280 ZONE_PADDING(_pad2_)
281 /* Rarely used or read-mostly fields */
282
283 /*
284 * wait_table -- the array holding the hash table
Yasunori Goto02b694d2006-06-23 02:03:08 -0700285 * wait_table_hash_nr_entries -- the size of the hash table array
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
287 *
288 * The purpose of all these is to keep track of the people
289 * waiting for a page to become available and make them
290 * runnable again when possible. The trouble is that this
291 * consumes a lot of space, especially when so few things
292 * wait on pages at a given time. So instead of using
293 * per-page waitqueues, we use a waitqueue hash table.
294 *
295 * The bucket discipline is to sleep on the same queue when
296 * colliding and wake all in that wait queue when removing.
297 * When something wakes, it must check to be sure its page is
298 * truly available, a la thundering herd. The cost of a
299 * collision is great, but given the expected load of the
300 * table, they should be so rare as to be outweighed by the
301 * benefits from the saved space.
302 *
303 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
304 * primary users of these fields, and in mm/page_alloc.c
305 * free_area_init_core() performs the initialization of them.
306 */
307 wait_queue_head_t * wait_table;
Yasunori Goto02b694d2006-06-23 02:03:08 -0700308 unsigned long wait_table_hash_nr_entries;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 unsigned long wait_table_bits;
310
311 /*
312 * Discontig memory support fields.
313 */
314 struct pglist_data *zone_pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
316 unsigned long zone_start_pfn;
317
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700318 /*
319 * zone_start_pfn, spanned_pages and present_pages are all
320 * protected by span_seqlock. It is a seqlock because it has
321 * to be read outside of zone->lock, and it is done in the main
322 * allocator path. But, it is written quite infrequently.
323 *
324 * The lock is declared along with zone->lock because it is
325 * frequently read in proximity to zone->lock. It's good to
326 * give them a chance of being in the same cacheline.
327 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 unsigned long spanned_pages; /* total size, including holes */
329 unsigned long present_pages; /* amount of memory (excluding holes) */
330
331 /*
332 * rarely used fields:
333 */
Helge Deller15ad7cd2006-12-06 20:40:36 -0800334 const char *name;
Ravikiran G Thirumalai22fc6ec2006-01-08 01:01:27 -0800335} ____cacheline_internodealigned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336
David Rientjese815af92007-10-16 23:25:54 -0700337typedef enum {
338 ZONE_ALL_UNRECLAIMABLE, /* all pages pinned */
339 ZONE_RECLAIM_LOCKED, /* prevents concurrent reclaim */
David Rientjes098d7f12007-10-16 23:25:55 -0700340 ZONE_OOM_LOCKED, /* zone is in OOM killer zonelist */
David Rientjese815af92007-10-16 23:25:54 -0700341} zone_flags_t;
342
343static inline void zone_set_flag(struct zone *zone, zone_flags_t flag)
344{
345 set_bit(flag, &zone->flags);
346}
David Rientjesd773ed62007-10-16 23:26:01 -0700347
348static inline int zone_test_and_set_flag(struct zone *zone, zone_flags_t flag)
349{
350 return test_and_set_bit(flag, &zone->flags);
351}
352
David Rientjese815af92007-10-16 23:25:54 -0700353static inline void zone_clear_flag(struct zone *zone, zone_flags_t flag)
354{
355 clear_bit(flag, &zone->flags);
356}
357
358static inline int zone_is_all_unreclaimable(const struct zone *zone)
359{
360 return test_bit(ZONE_ALL_UNRECLAIMABLE, &zone->flags);
361}
David Rientjesd773ed62007-10-16 23:26:01 -0700362
David Rientjese815af92007-10-16 23:25:54 -0700363static inline int zone_is_reclaim_locked(const struct zone *zone)
364{
365 return test_bit(ZONE_RECLAIM_LOCKED, &zone->flags);
366}
David Rientjesd773ed62007-10-16 23:26:01 -0700367
David Rientjes098d7f12007-10-16 23:25:55 -0700368static inline int zone_is_oom_locked(const struct zone *zone)
369{
370 return test_bit(ZONE_OOM_LOCKED, &zone->flags);
371}
David Rientjese815af92007-10-16 23:25:54 -0700372
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373/*
374 * The "priority" of VM scanning is how much of the queues we will scan in one
375 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
376 * queues ("queue_length >> 12") during an aging round.
377 */
378#define DEF_PRIORITY 12
379
Paul Jackson9276b1bc2006-12-06 20:31:48 -0800380/* Maximum number of zones on a zonelist */
381#define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
382
383#ifdef CONFIG_NUMA
Christoph Lameter523b9452007-10-16 01:25:37 -0700384
385/*
386 * The NUMA zonelists are doubled becausse we need zonelists that restrict the
387 * allocations to a single node for GFP_THISNODE.
388 *
Mel Gorman54a6eb52008-04-28 02:12:16 -0700389 * [0] : Zonelist with fallback
390 * [1] : No fallback (GFP_THISNODE)
Christoph Lameter523b9452007-10-16 01:25:37 -0700391 */
Mel Gorman54a6eb52008-04-28 02:12:16 -0700392#define MAX_ZONELISTS 2
Christoph Lameter523b9452007-10-16 01:25:37 -0700393
394
Paul Jackson9276b1bc2006-12-06 20:31:48 -0800395/*
396 * We cache key information from each zonelist for smaller cache
397 * footprint when scanning for free pages in get_page_from_freelist().
398 *
399 * 1) The BITMAP fullzones tracks which zones in a zonelist have come
400 * up short of free memory since the last time (last_fullzone_zap)
401 * we zero'd fullzones.
402 * 2) The array z_to_n[] maps each zone in the zonelist to its node
403 * id, so that we can efficiently evaluate whether that node is
404 * set in the current tasks mems_allowed.
405 *
406 * Both fullzones and z_to_n[] are one-to-one with the zonelist,
407 * indexed by a zones offset in the zonelist zones[] array.
408 *
409 * The get_page_from_freelist() routine does two scans. During the
410 * first scan, we skip zones whose corresponding bit in 'fullzones'
411 * is set or whose corresponding node in current->mems_allowed (which
412 * comes from cpusets) is not set. During the second scan, we bypass
413 * this zonelist_cache, to ensure we look methodically at each zone.
414 *
415 * Once per second, we zero out (zap) fullzones, forcing us to
416 * reconsider nodes that might have regained more free memory.
417 * The field last_full_zap is the time we last zapped fullzones.
418 *
419 * This mechanism reduces the amount of time we waste repeatedly
420 * reexaming zones for free memory when they just came up low on
421 * memory momentarilly ago.
422 *
423 * The zonelist_cache struct members logically belong in struct
424 * zonelist. However, the mempolicy zonelists constructed for
425 * MPOL_BIND are intentionally variable length (and usually much
426 * shorter). A general purpose mechanism for handling structs with
427 * multiple variable length members is more mechanism than we want
428 * here. We resort to some special case hackery instead.
429 *
430 * The MPOL_BIND zonelists don't need this zonelist_cache (in good
431 * part because they are shorter), so we put the fixed length stuff
432 * at the front of the zonelist struct, ending in a variable length
433 * zones[], as is needed by MPOL_BIND.
434 *
435 * Then we put the optional zonelist cache on the end of the zonelist
436 * struct. This optional stuff is found by a 'zlcache_ptr' pointer in
437 * the fixed length portion at the front of the struct. This pointer
438 * both enables us to find the zonelist cache, and in the case of
439 * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL)
440 * to know that the zonelist cache is not there.
441 *
442 * The end result is that struct zonelists come in two flavors:
443 * 1) The full, fixed length version, shown below, and
444 * 2) The custom zonelists for MPOL_BIND.
445 * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache.
446 *
447 * Even though there may be multiple CPU cores on a node modifying
448 * fullzones or last_full_zap in the same zonelist_cache at the same
449 * time, we don't lock it. This is just hint data - if it is wrong now
450 * and then, the allocator will still function, perhaps a bit slower.
451 */
452
453
454struct zonelist_cache {
Paul Jackson9276b1bc2006-12-06 20:31:48 -0800455 unsigned short z_to_n[MAX_ZONES_PER_ZONELIST]; /* zone->nid */
Paul Jackson7253f4e2006-12-06 20:31:49 -0800456 DECLARE_BITMAP(fullzones, MAX_ZONES_PER_ZONELIST); /* zone full? */
Paul Jackson9276b1bc2006-12-06 20:31:48 -0800457 unsigned long last_full_zap; /* when last zap'd (jiffies) */
458};
459#else
Mel Gorman54a6eb52008-04-28 02:12:16 -0700460#define MAX_ZONELISTS 1
Paul Jackson9276b1bc2006-12-06 20:31:48 -0800461struct zonelist_cache;
462#endif
463
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464/*
Mel Gormandd1a2392008-04-28 02:12:17 -0700465 * This struct contains information about a zone in a zonelist. It is stored
466 * here to avoid dereferences into large structures and lookups of tables
467 */
468struct zoneref {
469 struct zone *zone; /* Pointer to actual zone */
470 int zone_idx; /* zone_idx(zoneref->zone) */
471};
472
473/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 * One allocation request operates on a zonelist. A zonelist
475 * is a list of zones, the first one is the 'goal' of the
476 * allocation, the other zones are fallback zones, in decreasing
477 * priority.
478 *
Paul Jackson9276b1bc2006-12-06 20:31:48 -0800479 * If zlcache_ptr is not NULL, then it is just the address of zlcache,
480 * as explained above. If zlcache_ptr is NULL, there is no zlcache.
Mel Gormandd1a2392008-04-28 02:12:17 -0700481 * *
482 * To speed the reading of the zonelist, the zonerefs contain the zone index
483 * of the entry being read. Helper functions to access information given
484 * a struct zoneref are
485 *
486 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
487 * zonelist_zone_idx() - Return the index of the zone for an entry
488 * zonelist_node_idx() - Return the index of the node for an entry
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 */
490struct zonelist {
Paul Jackson9276b1bc2006-12-06 20:31:48 -0800491 struct zonelist_cache *zlcache_ptr; // NULL or &zlcache
Mel Gormandd1a2392008-04-28 02:12:17 -0700492 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
Paul Jackson9276b1bc2006-12-06 20:31:48 -0800493#ifdef CONFIG_NUMA
494 struct zonelist_cache zlcache; // optional ...
495#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496};
497
Mel Gormanc7132162006-09-27 01:49:43 -0700498#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
499struct node_active_region {
500 unsigned long start_pfn;
501 unsigned long end_pfn;
502 int nid;
503};
504#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505
Heiko Carstens5b99cd02006-09-27 01:50:01 -0700506#ifndef CONFIG_DISCONTIGMEM
507/* The array of struct pages - for discontigmem use pgdat->lmem_map */
508extern struct page *mem_map;
509#endif
510
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511/*
512 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
513 * (mostly NUMA machines?) to denote a higher-level memory zone than the
514 * zone denotes.
515 *
516 * On NUMA machines, each NUMA node would have a pg_data_t to describe
517 * it's memory layout.
518 *
519 * Memory statistics and page replacement data structures are maintained on a
520 * per-zone basis.
521 */
522struct bootmem_data;
523typedef struct pglist_data {
524 struct zone node_zones[MAX_NR_ZONES];
Christoph Lameter523b9452007-10-16 01:25:37 -0700525 struct zonelist node_zonelists[MAX_ZONELISTS];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 int nr_zones;
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700527#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 struct page *node_mem_map;
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700529#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 struct bootmem_data *bdata;
Dave Hansen208d54e2005-10-29 18:16:52 -0700531#ifdef CONFIG_MEMORY_HOTPLUG
532 /*
533 * Must be held any time you expect node_start_pfn, node_present_pages
534 * or node_spanned_pages stay constant. Holding this will also
535 * guarantee that any pfn_valid() stays that way.
536 *
537 * Nests above zone->lock and zone->size_seqlock.
538 */
539 spinlock_t node_size_lock;
540#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 unsigned long node_start_pfn;
542 unsigned long node_present_pages; /* total number of physical pages */
543 unsigned long node_spanned_pages; /* total size of physical page
544 range, including holes */
545 int node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 wait_queue_head_t kswapd_wait;
547 struct task_struct *kswapd;
548 int kswapd_max_order;
549} pg_data_t;
550
551#define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
552#define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700553#ifdef CONFIG_FLAT_NODE_MEM_MAP
Dave Hansen408fde82005-06-23 00:07:37 -0700554#define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700555#else
556#define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
557#endif
Dave Hansen408fde82005-06-23 00:07:37 -0700558#define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559
Dave Hansen208d54e2005-10-29 18:16:52 -0700560#include <linux/memory_hotplug.h>
561
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562void get_zone_counts(unsigned long *active, unsigned long *inactive,
563 unsigned long *free);
564void build_all_zonelists(void);
565void wakeup_kswapd(struct zone *zone, int order);
566int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800567 int classzone_idx, int alloc_flags);
Dave Hansena2f3aa022007-01-10 23:15:30 -0800568enum memmap_context {
569 MEMMAP_EARLY,
570 MEMMAP_HOTPLUG,
571};
Yasunori Goto718127c2006-06-23 02:03:10 -0700572extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
Dave Hansena2f3aa022007-01-10 23:15:30 -0800573 unsigned long size,
574 enum memmap_context context);
Yasunori Goto718127c2006-06-23 02:03:10 -0700575
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576#ifdef CONFIG_HAVE_MEMORY_PRESENT
577void memory_present(int nid, unsigned long start, unsigned long end);
578#else
579static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
580#endif
581
582#ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
583unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
584#endif
585
586/*
587 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
588 */
589#define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
590
Con Kolivasf3fe6512006-01-06 00:11:15 -0800591static inline int populated_zone(struct zone *zone)
592{
593 return (!!zone->present_pages);
594}
595
Mel Gorman2a1e2742007-07-17 04:03:12 -0700596extern int movable_zone;
597
598static inline int zone_movable_is_highmem(void)
599{
600#if defined(CONFIG_HIGHMEM) && defined(CONFIG_ARCH_POPULATES_NODE_MAP)
601 return movable_zone == ZONE_HIGHMEM;
602#else
603 return 0;
604#endif
605}
606
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700607static inline int is_highmem_idx(enum zone_type idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608{
Christoph Lametere53ef382006-09-25 23:31:14 -0700609#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700610 return (idx == ZONE_HIGHMEM ||
611 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
Christoph Lametere53ef382006-09-25 23:31:14 -0700612#else
613 return 0;
614#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615}
616
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700617static inline int is_normal_idx(enum zone_type idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618{
619 return (idx == ZONE_NORMAL);
620}
Nick Piggin9328b8f2006-01-06 00:11:10 -0800621
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622/**
623 * is_highmem - helper function to quickly check if a struct zone is a
624 * highmem zone or not. This is an attempt to keep references
625 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
626 * @zone - pointer to struct zone variable
627 */
628static inline int is_highmem(struct zone *zone)
629{
Christoph Lametere53ef382006-09-25 23:31:14 -0700630#ifdef CONFIG_HIGHMEM
Harvey Harrisonddc81ed2008-04-28 02:12:07 -0700631 int zone_off = (char *)zone - (char *)zone->zone_pgdat->node_zones;
632 return zone_off == ZONE_HIGHMEM * sizeof(*zone) ||
633 (zone_off == ZONE_MOVABLE * sizeof(*zone) &&
634 zone_movable_is_highmem());
Christoph Lametere53ef382006-09-25 23:31:14 -0700635#else
636 return 0;
637#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638}
639
640static inline int is_normal(struct zone *zone)
641{
642 return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
643}
644
Nick Piggin9328b8f2006-01-06 00:11:10 -0800645static inline int is_dma32(struct zone *zone)
646{
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700647#ifdef CONFIG_ZONE_DMA32
Nick Piggin9328b8f2006-01-06 00:11:10 -0800648 return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700649#else
650 return 0;
651#endif
Nick Piggin9328b8f2006-01-06 00:11:10 -0800652}
653
654static inline int is_dma(struct zone *zone)
655{
Christoph Lameter4b51d662007-02-10 01:43:10 -0800656#ifdef CONFIG_ZONE_DMA
Nick Piggin9328b8f2006-01-06 00:11:10 -0800657 return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
Christoph Lameter4b51d662007-02-10 01:43:10 -0800658#else
659 return 0;
660#endif
Nick Piggin9328b8f2006-01-06 00:11:10 -0800661}
662
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663/* These two functions are used to setup the per zone pages min values */
664struct ctl_table;
665struct file;
666int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *,
667 void __user *, size_t *, loff_t *);
668extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
669int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *,
670 void __user *, size_t *, loff_t *);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -0800671int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, struct file *,
672 void __user *, size_t *, loff_t *);
Christoph Lameter96146342006-07-03 00:24:13 -0700673int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
674 struct file *, void __user *, size_t *, loff_t *);
Christoph Lameter0ff38492006-09-25 23:31:52 -0700675int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
676 struct file *, void __user *, size_t *, loff_t *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -0700678extern int numa_zonelist_order_handler(struct ctl_table *, int,
679 struct file *, void __user *, size_t *, loff_t *);
680extern char numa_zonelist_order[];
681#define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */
682
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683#include <linux/topology.h>
684/* Returns the number of the current Node. */
Andi Kleen69d81fc2005-11-05 17:25:53 +0100685#ifndef numa_node_id
Ingo Molnar39c715b2005-06-21 17:14:34 -0700686#define numa_node_id() (cpu_to_node(raw_smp_processor_id()))
Andi Kleen69d81fc2005-11-05 17:25:53 +0100687#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688
Dave Hansen93b75042005-06-23 00:07:47 -0700689#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690
691extern struct pglist_data contig_page_data;
692#define NODE_DATA(nid) (&contig_page_data)
693#define NODE_MEM_MAP(nid) mem_map
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694
Dave Hansen93b75042005-06-23 00:07:47 -0700695#else /* CONFIG_NEED_MULTIPLE_NODES */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696
697#include <asm/mmzone.h>
698
Dave Hansen93b75042005-06-23 00:07:47 -0700699#endif /* !CONFIG_NEED_MULTIPLE_NODES */
Dave Hansen348f8b62005-06-23 00:07:40 -0700700
KAMEZAWA Hiroyuki95144c72006-03-27 01:16:02 -0800701extern struct pglist_data *first_online_pgdat(void);
702extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
703extern struct zone *next_zone(struct zone *zone);
KAMEZAWA Hiroyuki8357f862006-03-27 01:15:57 -0800704
705/**
Fernando Luis Vazquez Cao12d15f02008-05-23 13:05:01 -0700706 * for_each_online_pgdat - helper macro to iterate over all online nodes
KAMEZAWA Hiroyuki8357f862006-03-27 01:15:57 -0800707 * @pgdat - pointer to a pg_data_t variable
708 */
709#define for_each_online_pgdat(pgdat) \
710 for (pgdat = first_online_pgdat(); \
711 pgdat; \
712 pgdat = next_online_pgdat(pgdat))
KAMEZAWA Hiroyuki8357f862006-03-27 01:15:57 -0800713/**
714 * for_each_zone - helper macro to iterate over all memory zones
715 * @zone - pointer to struct zone variable
716 *
717 * The user only needs to declare the zone variable, for_each_zone
718 * fills it in.
719 */
720#define for_each_zone(zone) \
721 for (zone = (first_online_pgdat())->node_zones; \
722 zone; \
723 zone = next_zone(zone))
724
Mel Gormandd1a2392008-04-28 02:12:17 -0700725static inline struct zone *zonelist_zone(struct zoneref *zoneref)
726{
727 return zoneref->zone;
728}
729
730static inline int zonelist_zone_idx(struct zoneref *zoneref)
731{
732 return zoneref->zone_idx;
733}
734
735static inline int zonelist_node_idx(struct zoneref *zoneref)
736{
737#ifdef CONFIG_NUMA
738 /* zone_to_nid not available in this context */
739 return zoneref->zone->node;
740#else
741 return 0;
742#endif /* CONFIG_NUMA */
743}
744
Mel Gorman19770b32008-04-28 02:12:18 -0700745/**
746 * 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
747 * @z - The cursor used as a starting point for the search
748 * @highest_zoneidx - The zone index of the highest zone to return
749 * @nodes - An optional nodemask to filter the zonelist with
750 * @zone - The first suitable zone found is returned via this parameter
751 *
752 * This function returns the next zone at or below a given zone index that is
753 * within the allowed nodemask using a cursor as the starting point for the
Mel Gorman5bead2a2008-09-13 02:33:19 -0700754 * search. The zoneref returned is a cursor that represents the current zone
755 * being examined. It should be advanced by one before calling
756 * next_zones_zonelist again.
Mel Gorman19770b32008-04-28 02:12:18 -0700757 */
758struct zoneref *next_zones_zonelist(struct zoneref *z,
759 enum zone_type highest_zoneidx,
760 nodemask_t *nodes,
761 struct zone **zone);
Mel Gormandd1a2392008-04-28 02:12:17 -0700762
Mel Gorman19770b32008-04-28 02:12:18 -0700763/**
764 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
765 * @zonelist - The zonelist to search for a suitable zone
766 * @highest_zoneidx - The zone index of the highest zone to return
767 * @nodes - An optional nodemask to filter the zonelist with
768 * @zone - The first suitable zone found is returned via this parameter
769 *
770 * This function returns the first zone at or below a given zone index that is
771 * within the allowed nodemask. The zoneref returned is a cursor that can be
Mel Gorman5bead2a2008-09-13 02:33:19 -0700772 * used to iterate the zonelist with next_zones_zonelist by advancing it by
773 * one before calling.
Mel Gorman19770b32008-04-28 02:12:18 -0700774 */
Mel Gormandd1a2392008-04-28 02:12:17 -0700775static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
Mel Gorman19770b32008-04-28 02:12:18 -0700776 enum zone_type highest_zoneidx,
777 nodemask_t *nodes,
778 struct zone **zone)
Mel Gorman54a6eb52008-04-28 02:12:16 -0700779{
Mel Gorman19770b32008-04-28 02:12:18 -0700780 return next_zones_zonelist(zonelist->_zonerefs, highest_zoneidx, nodes,
781 zone);
Mel Gorman54a6eb52008-04-28 02:12:16 -0700782}
783
Mel Gorman19770b32008-04-28 02:12:18 -0700784/**
785 * 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
786 * @zone - The current zone in the iterator
787 * @z - The current pointer within zonelist->zones being iterated
788 * @zlist - The zonelist being iterated
789 * @highidx - The zone index of the highest zone to return
790 * @nodemask - Nodemask allowed by the allocator
791 *
792 * This iterator iterates though all zones at or below a given zone index and
793 * within a given nodemask
794 */
795#define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
796 for (z = first_zones_zonelist(zlist, highidx, nodemask, &zone); \
797 zone; \
Mel Gorman5bead2a2008-09-13 02:33:19 -0700798 z = next_zones_zonelist(++z, highidx, nodemask, &zone)) \
Mel Gorman54a6eb52008-04-28 02:12:16 -0700799
800/**
801 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
802 * @zone - The current zone in the iterator
803 * @z - The current pointer within zonelist->zones being iterated
804 * @zlist - The zonelist being iterated
805 * @highidx - The zone index of the highest zone to return
806 *
807 * This iterator iterates though all zones at or below a given zone index.
808 */
809#define for_each_zone_zonelist(zone, z, zlist, highidx) \
Mel Gorman19770b32008-04-28 02:12:18 -0700810 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
Mel Gorman54a6eb52008-04-28 02:12:16 -0700811
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700812#ifdef CONFIG_SPARSEMEM
813#include <asm/sparsemem.h>
814#endif
815
Mel Gormanc7132162006-09-27 01:49:43 -0700816#if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
817 !defined(CONFIG_ARCH_POPULATES_NODE_MAP)
Andrew Mortonb4544562008-04-28 02:12:39 -0700818static inline unsigned long early_pfn_to_nid(unsigned long pfn)
819{
820 return 0;
821}
Andy Whitcroftb159d432005-06-23 00:07:52 -0700822#endif
823
Andy Whitcroft2bdaf112006-01-06 00:10:53 -0800824#ifdef CONFIG_FLATMEM
825#define pfn_to_nid(pfn) (0)
826#endif
827
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700828#define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
829#define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
830
831#ifdef CONFIG_SPARSEMEM
832
833/*
834 * SECTION_SHIFT #bits space required to store a section #
835 *
836 * PA_SECTION_SHIFT physical address to/from section number
837 * PFN_SECTION_SHIFT pfn to/from section number
838 */
839#define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
840
841#define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
842#define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
843
844#define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
845
846#define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
847#define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
848
Mel Gorman835c1342007-10-16 01:25:47 -0700849#define SECTION_BLOCKFLAGS_BITS \
Mel Gormand9c23402007-10-16 01:26:01 -0700850 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
Mel Gorman835c1342007-10-16 01:25:47 -0700851
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700852#if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
853#error Allocator MAX_ORDER exceeds SECTION_SIZE
854#endif
855
856struct page;
857struct mem_section {
Andy Whitcroft29751f62005-06-23 00:08:00 -0700858 /*
859 * This is, logically, a pointer to an array of struct
860 * pages. However, it is stored with some other magic.
861 * (see sparse.c::sparse_init_one_section())
862 *
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700863 * Additionally during early boot we encode node id of
864 * the location of the section here to guide allocation.
865 * (see sparse.c::memory_present())
866 *
Andy Whitcroft29751f62005-06-23 00:08:00 -0700867 * Making it a UL at least makes someone do a cast
868 * before using it wrong.
869 */
870 unsigned long section_mem_map;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700871
872 /* See declaration of similar field in struct zone */
873 unsigned long *pageblock_flags;
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700874};
875
Bob Picco3e347262005-09-03 15:54:28 -0700876#ifdef CONFIG_SPARSEMEM_EXTREME
877#define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
Bob Picco802f1922005-09-03 15:54:26 -0700878#else
Bob Picco3e347262005-09-03 15:54:28 -0700879#define SECTIONS_PER_ROOT 1
880#endif
Bob Picco802f1922005-09-03 15:54:26 -0700881
Bob Picco3e347262005-09-03 15:54:28 -0700882#define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
883#define NR_SECTION_ROOTS (NR_MEM_SECTIONS / SECTIONS_PER_ROOT)
884#define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
885
886#ifdef CONFIG_SPARSEMEM_EXTREME
887extern struct mem_section *mem_section[NR_SECTION_ROOTS];
888#else
889extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
890#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700891
Andy Whitcroft29751f62005-06-23 00:08:00 -0700892static inline struct mem_section *__nr_to_section(unsigned long nr)
893{
Bob Picco3e347262005-09-03 15:54:28 -0700894 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
895 return NULL;
896 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
Andy Whitcroft29751f62005-06-23 00:08:00 -0700897}
Dave Hansen4ca644d2005-10-29 18:16:51 -0700898extern int __section_nr(struct mem_section* ms);
Yasunori Goto04753272008-04-28 02:13:31 -0700899extern unsigned long usemap_size(void);
Andy Whitcroft29751f62005-06-23 00:08:00 -0700900
901/*
902 * We use the lower bits of the mem_map pointer to store
903 * a little bit of information. There should be at least
904 * 3 bits here due to 32-bit alignment.
905 */
906#define SECTION_MARKED_PRESENT (1UL<<0)
907#define SECTION_HAS_MEM_MAP (1UL<<1)
908#define SECTION_MAP_LAST_BIT (1UL<<2)
909#define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700910#define SECTION_NID_SHIFT 2
Andy Whitcroft29751f62005-06-23 00:08:00 -0700911
912static inline struct page *__section_mem_map_addr(struct mem_section *section)
913{
914 unsigned long map = section->section_mem_map;
915 map &= SECTION_MAP_MASK;
916 return (struct page *)map;
917}
918
Andy Whitcroft540557b2007-10-16 01:24:11 -0700919static inline int present_section(struct mem_section *section)
Andy Whitcroft29751f62005-06-23 00:08:00 -0700920{
Bob Picco802f1922005-09-03 15:54:26 -0700921 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
Andy Whitcroft29751f62005-06-23 00:08:00 -0700922}
923
Andy Whitcroft540557b2007-10-16 01:24:11 -0700924static inline int present_section_nr(unsigned long nr)
925{
926 return present_section(__nr_to_section(nr));
927}
928
929static inline int valid_section(struct mem_section *section)
Andy Whitcroft29751f62005-06-23 00:08:00 -0700930{
Bob Picco802f1922005-09-03 15:54:26 -0700931 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
Andy Whitcroft29751f62005-06-23 00:08:00 -0700932}
933
934static inline int valid_section_nr(unsigned long nr)
935{
936 return valid_section(__nr_to_section(nr));
937}
938
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700939static inline struct mem_section *__pfn_to_section(unsigned long pfn)
940{
Andy Whitcroft29751f62005-06-23 00:08:00 -0700941 return __nr_to_section(pfn_to_section_nr(pfn));
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700942}
943
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700944static inline int pfn_valid(unsigned long pfn)
945{
946 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
947 return 0;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700948 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700949}
950
Andy Whitcroft540557b2007-10-16 01:24:11 -0700951static inline int pfn_present(unsigned long pfn)
952{
953 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
954 return 0;
955 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
956}
957
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700958/*
959 * These are _only_ used during initialisation, therefore they
960 * can use __initdata ... They could have names to indicate
961 * this restriction.
962 */
963#ifdef CONFIG_NUMA
Andy Whitcroft161599f2006-01-06 00:10:54 -0800964#define pfn_to_nid(pfn) \
965({ \
966 unsigned long __pfn_to_nid_pfn = (pfn); \
967 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
968})
Andy Whitcroft2bdaf112006-01-06 00:10:53 -0800969#else
970#define pfn_to_nid(pfn) (0)
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700971#endif
972
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700973#define early_pfn_valid(pfn) pfn_valid(pfn)
974void sparse_init(void);
975#else
976#define sparse_init() do {} while (0)
Dave Hansen28ae55c2005-09-03 15:54:29 -0700977#define sparse_index_init(_sec, _nid) do {} while (0)
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700978#endif /* CONFIG_SPARSEMEM */
979
Andy Whitcroft75167952006-10-21 10:24:14 -0700980#ifdef CONFIG_NODES_SPAN_OTHER_NODES
981#define early_pfn_in_nid(pfn, nid) (early_pfn_to_nid(pfn) == (nid))
982#else
983#define early_pfn_in_nid(pfn, nid) (1)
984#endif
985
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700986#ifndef early_pfn_valid
987#define early_pfn_valid(pfn) (1)
988#endif
989
990void memory_present(int nid, unsigned long start, unsigned long end);
991unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
992
Andy Whitcroft14e07292007-05-06 14:49:14 -0700993/*
994 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
995 * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
996 * pfn_valid_within() should be used in this case; we optimise this away
997 * when we have no holes within a MAX_ORDER_NR_PAGES block.
998 */
999#ifdef CONFIG_HOLES_IN_ZONE
1000#define pfn_valid_within(pfn) pfn_valid(pfn)
1001#else
1002#define pfn_valid_within(pfn) (1)
1003#endif
1004
Christoph Lameter97965472008-04-28 02:12:54 -07001005#endif /* !__GENERATING_BOUNDS.H */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006#endif /* !__ASSEMBLY__ */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007#endif /* _LINUX_MMZONE_H */