blob: 3693f1a5278839c23243691564a5085af0617313 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001#ifndef _LINUX_MMZONE_H
2#define _LINUX_MMZONE_H
3
4#ifdef __KERNEL__
5#ifndef __ASSEMBLY__
6
Linus Torvalds1da177e2005-04-16 15:20:36 -07007#include <linux/spinlock.h>
8#include <linux/list.h>
9#include <linux/wait.h>
10#include <linux/cache.h>
11#include <linux/threads.h>
12#include <linux/numa.h>
13#include <linux/init.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070014#include <linux/seqlock.h>
KAMEZAWA Hiroyuki8357f862006-03-27 01:15:57 -080015#include <linux/nodemask.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <asm/atomic.h>
Ralf Baechle93ff66b2006-06-04 02:51:29 -070017#include <asm/page.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018
19/* Free memory management - zoned buddy allocator. */
20#ifndef CONFIG_FORCE_MAX_ZONEORDER
21#define MAX_ORDER 11
22#else
23#define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
24#endif
Bob Piccoe984bb42006-05-20 15:00:31 -070025#define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -070026
27struct free_area {
28 struct list_head free_list;
29 unsigned long nr_free;
30};
31
32struct pglist_data;
33
34/*
35 * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
36 * So add a wild amount of padding here to ensure that they fall into separate
37 * cachelines. There are very few zone structures in the machine, so space
38 * consumption is not a concern here.
39 */
40#if defined(CONFIG_SMP)
41struct zone_padding {
42 char x[0];
Ravikiran G Thirumalai22fc6ec2006-01-08 01:01:27 -080043} ____cacheline_internodealigned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#define ZONE_PADDING(name) struct zone_padding name;
45#else
46#define ZONE_PADDING(name)
47#endif
48
Christoph Lameter2244b952006-06-30 01:55:33 -070049enum zone_stat_item {
Christoph Lameterf3dbd342006-06-30 01:55:36 -070050 NR_ANON_PAGES, /* Mapped anonymous pages */
51 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
Christoph Lameter65ba55f2006-06-30 01:55:34 -070052 only modified from process context */
Christoph Lameter347ce432006-06-30 01:55:35 -070053 NR_FILE_PAGES,
Christoph Lameter972d1a72006-09-25 23:31:51 -070054 NR_SLAB_RECLAIMABLE,
55 NR_SLAB_UNRECLAIMABLE,
Christoph Lameterdf849a12006-06-30 01:55:38 -070056 NR_PAGETABLE, /* used for pagetables */
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -070057 NR_FILE_DIRTY,
Christoph Lameterce866b32006-06-30 01:55:40 -070058 NR_WRITEBACK,
Christoph Lameterfd39fc82006-06-30 01:55:40 -070059 NR_UNSTABLE_NFS, /* NFS unstable pages */
Christoph Lameterd2c5e302006-06-30 01:55:41 -070060 NR_BOUNCE,
Christoph Lameterca889e62006-06-30 01:55:44 -070061#ifdef CONFIG_NUMA
62 NUMA_HIT, /* allocated in intended node */
63 NUMA_MISS, /* allocated in non intended node */
64 NUMA_FOREIGN, /* was intended here, hit elsewhere */
65 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
66 NUMA_LOCAL, /* allocation from local node */
67 NUMA_OTHER, /* allocation from other node */
68#endif
Christoph Lameter2244b952006-06-30 01:55:33 -070069 NR_VM_ZONE_STAT_ITEMS };
70
Linus Torvalds1da177e2005-04-16 15:20:36 -070071struct per_cpu_pages {
72 int count; /* number of pages in the list */
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 int high; /* high watermark, emptying needed */
74 int batch; /* chunk size for buddy add/remove */
75 struct list_head list; /* the list of pages */
76};
77
78struct per_cpu_pageset {
79 struct per_cpu_pages pcp[2]; /* 0: hot. 1: cold */
Christoph Lameter2244b952006-06-30 01:55:33 -070080#ifdef CONFIG_SMP
Christoph Lameterdf9ecab2006-08-31 21:27:35 -070081 s8 stat_threshold;
Christoph Lameter2244b952006-06-30 01:55:33 -070082 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
83#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070084} ____cacheline_aligned_in_smp;
85
Christoph Lametere7c8d5c2005-06-21 17:14:47 -070086#ifdef CONFIG_NUMA
87#define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)])
88#else
89#define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)])
90#endif
91
Christoph Lameter2f1b6242006-09-25 23:31:13 -070092enum zone_type {
93 /*
94 * ZONE_DMA is used when there are devices that are not able
95 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
96 * carve out the portion of memory that is needed for these devices.
97 * The range is arch specific.
98 *
99 * Some examples
100 *
101 * Architecture Limit
102 * ---------------------------
103 * parisc, ia64, sparc <4G
104 * s390 <2G
105 * arm26 <48M
106 * arm Various
107 * alpha Unlimited or 0-16MB.
108 *
109 * i386, x86_64 and multiple other arches
110 * <16M.
111 */
112 ZONE_DMA,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700113#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700114 /*
115 * x86_64 needs two ZONE_DMAs because it supports devices that are
116 * only able to do DMA to the lower 16M but also 32 bit devices that
117 * can only do DMA areas below 4G.
118 */
119 ZONE_DMA32,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700120#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700121 /*
122 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
123 * performed on pages in ZONE_NORMAL if the DMA devices support
124 * transfers to all addressable memory.
125 */
126 ZONE_NORMAL,
Christoph Lametere53ef382006-09-25 23:31:14 -0700127#ifdef CONFIG_HIGHMEM
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700128 /*
129 * A memory area that is only addressable by the kernel through
130 * mapping portions into its own address space. This is for example
131 * used by i386 to allow the kernel to address the memory beyond
132 * 900MB. The kernel will set up special mappings (page
133 * table entries on i386) for each page that the kernel needs to
134 * access.
135 */
136 ZONE_HIGHMEM,
Christoph Lametere53ef382006-09-25 23:31:14 -0700137#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700138 MAX_NR_ZONES
139};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141/*
142 * When a memory allocation must conform to specific limitations (such
143 * as being suitable for DMA) the caller will pass in hints to the
144 * allocator in the gfp_mask, in the zone modifier bits. These bits
145 * are used to select a priority ordered list of memory zones which
Christoph Lameter19655d32006-09-25 23:31:19 -0700146 * match the requested limits. See gfp_zone() in include/linux/gfp.h
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 */
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700148
Christoph Lameter19655d32006-09-25 23:31:19 -0700149#if !defined(CONFIG_ZONE_DMA32) && !defined(CONFIG_HIGHMEM)
150#define ZONES_SHIFT 1
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700151#else
Christoph Lameter19655d32006-09-25 23:31:19 -0700152#define ZONES_SHIFT 2
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700153#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155struct zone {
156 /* Fields commonly accessed by the page allocator */
157 unsigned long free_pages;
158 unsigned long pages_min, pages_low, pages_high;
159 /*
160 * We don't know if the memory that we're going to allocate will be freeable
161 * or/and it will be released eventually, so to avoid totally wasting several
162 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
163 * to run OOM on the lower zones despite there's tons of freeable ram
164 * on the higher zones). This array is recalculated at runtime if the
165 * sysctl_lowmem_reserve_ratio sysctl changes.
166 */
167 unsigned long lowmem_reserve[MAX_NR_ZONES];
168
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700169#ifdef CONFIG_NUMA
Christoph Lameter96146342006-07-03 00:24:13 -0700170 /*
171 * zone reclaim becomes active if more unmapped pages exist.
172 */
Christoph Lameter8417bba2006-09-25 23:31:51 -0700173 unsigned long min_unmapped_pages;
Christoph Lameter0ff38492006-09-25 23:31:52 -0700174 unsigned long min_slab_pages;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700175 struct per_cpu_pageset *pageset[NR_CPUS];
176#else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 struct per_cpu_pageset pageset[NR_CPUS];
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700178#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 /*
180 * free areas of different sizes
181 */
182 spinlock_t lock;
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700183#ifdef CONFIG_MEMORY_HOTPLUG
184 /* see spanned/present_pages for more description */
185 seqlock_t span_seqlock;
186#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 struct free_area free_area[MAX_ORDER];
188
189
190 ZONE_PADDING(_pad1_)
191
192 /* Fields commonly accessed by the page reclaim scanner */
193 spinlock_t lru_lock;
194 struct list_head active_list;
195 struct list_head inactive_list;
196 unsigned long nr_scan_active;
197 unsigned long nr_scan_inactive;
198 unsigned long nr_active;
199 unsigned long nr_inactive;
200 unsigned long pages_scanned; /* since last reclaim */
201 int all_unreclaimable; /* All pages pinned */
202
Martin Hicks1e7e5a92005-06-21 17:14:43 -0700203 /* A count of how many reclaimers are scanning this zone */
204 atomic_t reclaim_in_progress;
Martin Hicks753ee722005-06-21 17:14:41 -0700205
Christoph Lameter2244b952006-06-30 01:55:33 -0700206 /* Zone statistics */
207 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
Christoph Lameter9eeff232006-01-18 17:42:31 -0800208
209 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 * prev_priority holds the scanning priority for this zone. It is
211 * defined as the scanning priority at which we achieved our reclaim
212 * target at the previous try_to_free_pages() or balance_pgdat()
213 * invokation.
214 *
215 * We use prev_priority as a measure of how much stress page reclaim is
216 * under - it drives the swappiness decision: whether to unmap mapped
217 * pages.
218 *
219 * temp_priority is used to remember the scanning priority at which
220 * this zone was successfully refilled to free_pages == pages_high.
221 *
222 * Access to both these fields is quite racy even on uniprocessor. But
223 * it is expected to average out OK.
224 */
225 int temp_priority;
226 int prev_priority;
227
228
229 ZONE_PADDING(_pad2_)
230 /* Rarely used or read-mostly fields */
231
232 /*
233 * wait_table -- the array holding the hash table
Yasunori Goto02b694d2006-06-23 02:03:08 -0700234 * wait_table_hash_nr_entries -- the size of the hash table array
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
236 *
237 * The purpose of all these is to keep track of the people
238 * waiting for a page to become available and make them
239 * runnable again when possible. The trouble is that this
240 * consumes a lot of space, especially when so few things
241 * wait on pages at a given time. So instead of using
242 * per-page waitqueues, we use a waitqueue hash table.
243 *
244 * The bucket discipline is to sleep on the same queue when
245 * colliding and wake all in that wait queue when removing.
246 * When something wakes, it must check to be sure its page is
247 * truly available, a la thundering herd. The cost of a
248 * collision is great, but given the expected load of the
249 * table, they should be so rare as to be outweighed by the
250 * benefits from the saved space.
251 *
252 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
253 * primary users of these fields, and in mm/page_alloc.c
254 * free_area_init_core() performs the initialization of them.
255 */
256 wait_queue_head_t * wait_table;
Yasunori Goto02b694d2006-06-23 02:03:08 -0700257 unsigned long wait_table_hash_nr_entries;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 unsigned long wait_table_bits;
259
260 /*
261 * Discontig memory support fields.
262 */
263 struct pglist_data *zone_pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
265 unsigned long zone_start_pfn;
266
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700267 /*
268 * zone_start_pfn, spanned_pages and present_pages are all
269 * protected by span_seqlock. It is a seqlock because it has
270 * to be read outside of zone->lock, and it is done in the main
271 * allocator path. But, it is written quite infrequently.
272 *
273 * The lock is declared along with zone->lock because it is
274 * frequently read in proximity to zone->lock. It's good to
275 * give them a chance of being in the same cacheline.
276 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 unsigned long spanned_pages; /* total size, including holes */
278 unsigned long present_pages; /* amount of memory (excluding holes) */
279
280 /*
281 * rarely used fields:
282 */
283 char *name;
Ravikiran G Thirumalai22fc6ec2006-01-08 01:01:27 -0800284} ____cacheline_internodealigned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286/*
287 * The "priority" of VM scanning is how much of the queues we will scan in one
288 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
289 * queues ("queue_length >> 12") during an aging round.
290 */
291#define DEF_PRIORITY 12
292
293/*
294 * One allocation request operates on a zonelist. A zonelist
295 * is a list of zones, the first one is the 'goal' of the
296 * allocation, the other zones are fallback zones, in decreasing
297 * priority.
298 *
299 * Right now a zonelist takes up less than a cacheline. We never
300 * modify it apart from boot-up, and only a few indices are used,
301 * so despite the zonelist table being relatively big, the cache
302 * footprint of this construct is very small.
303 */
304struct zonelist {
305 struct zone *zones[MAX_NUMNODES * MAX_NR_ZONES + 1]; // NULL delimited
306};
307
308
309/*
310 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
311 * (mostly NUMA machines?) to denote a higher-level memory zone than the
312 * zone denotes.
313 *
314 * On NUMA machines, each NUMA node would have a pg_data_t to describe
315 * it's memory layout.
316 *
317 * Memory statistics and page replacement data structures are maintained on a
318 * per-zone basis.
319 */
320struct bootmem_data;
321typedef struct pglist_data {
322 struct zone node_zones[MAX_NR_ZONES];
Christoph Lameter19655d32006-09-25 23:31:19 -0700323 struct zonelist node_zonelists[MAX_NR_ZONES];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 int nr_zones;
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700325#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 struct page *node_mem_map;
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700327#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 struct bootmem_data *bdata;
Dave Hansen208d54e2005-10-29 18:16:52 -0700329#ifdef CONFIG_MEMORY_HOTPLUG
330 /*
331 * Must be held any time you expect node_start_pfn, node_present_pages
332 * or node_spanned_pages stay constant. Holding this will also
333 * guarantee that any pfn_valid() stays that way.
334 *
335 * Nests above zone->lock and zone->size_seqlock.
336 */
337 spinlock_t node_size_lock;
338#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 unsigned long node_start_pfn;
340 unsigned long node_present_pages; /* total number of physical pages */
341 unsigned long node_spanned_pages; /* total size of physical page
342 range, including holes */
343 int node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 wait_queue_head_t kswapd_wait;
345 struct task_struct *kswapd;
346 int kswapd_max_order;
347} pg_data_t;
348
349#define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
350#define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700351#ifdef CONFIG_FLAT_NODE_MEM_MAP
Dave Hansen408fde82005-06-23 00:07:37 -0700352#define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700353#else
354#define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
355#endif
Dave Hansen408fde82005-06-23 00:07:37 -0700356#define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357
Dave Hansen208d54e2005-10-29 18:16:52 -0700358#include <linux/memory_hotplug.h>
359
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360void __get_zone_counts(unsigned long *active, unsigned long *inactive,
361 unsigned long *free, struct pglist_data *pgdat);
362void get_zone_counts(unsigned long *active, unsigned long *inactive,
363 unsigned long *free);
364void build_all_zonelists(void);
365void wakeup_kswapd(struct zone *zone, int order);
366int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800367 int classzone_idx, int alloc_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368
Yasunori Goto718127c2006-06-23 02:03:10 -0700369extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
370 unsigned long size);
371
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372#ifdef CONFIG_HAVE_MEMORY_PRESENT
373void memory_present(int nid, unsigned long start, unsigned long end);
374#else
375static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
376#endif
377
378#ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
379unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
380#endif
381
382/*
383 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
384 */
385#define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
386
Con Kolivasf3fe6512006-01-06 00:11:15 -0800387static inline int populated_zone(struct zone *zone)
388{
389 return (!!zone->present_pages);
390}
391
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700392static inline int is_highmem_idx(enum zone_type idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393{
Christoph Lametere53ef382006-09-25 23:31:14 -0700394#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 return (idx == ZONE_HIGHMEM);
Christoph Lametere53ef382006-09-25 23:31:14 -0700396#else
397 return 0;
398#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399}
400
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700401static inline int is_normal_idx(enum zone_type idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402{
403 return (idx == ZONE_NORMAL);
404}
Nick Piggin9328b8f2006-01-06 00:11:10 -0800405
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406/**
407 * is_highmem - helper function to quickly check if a struct zone is a
408 * highmem zone or not. This is an attempt to keep references
409 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
410 * @zone - pointer to struct zone variable
411 */
412static inline int is_highmem(struct zone *zone)
413{
Christoph Lametere53ef382006-09-25 23:31:14 -0700414#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 return zone == zone->zone_pgdat->node_zones + ZONE_HIGHMEM;
Christoph Lametere53ef382006-09-25 23:31:14 -0700416#else
417 return 0;
418#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419}
420
421static inline int is_normal(struct zone *zone)
422{
423 return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
424}
425
Nick Piggin9328b8f2006-01-06 00:11:10 -0800426static inline int is_dma32(struct zone *zone)
427{
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700428#ifdef CONFIG_ZONE_DMA32
Nick Piggin9328b8f2006-01-06 00:11:10 -0800429 return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700430#else
431 return 0;
432#endif
Nick Piggin9328b8f2006-01-06 00:11:10 -0800433}
434
435static inline int is_dma(struct zone *zone)
436{
437 return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
438}
439
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440/* These two functions are used to setup the per zone pages min values */
441struct ctl_table;
442struct file;
443int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *,
444 void __user *, size_t *, loff_t *);
445extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
446int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *,
447 void __user *, size_t *, loff_t *);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -0800448int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, struct file *,
449 void __user *, size_t *, loff_t *);
Christoph Lameter96146342006-07-03 00:24:13 -0700450int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
451 struct file *, void __user *, size_t *, loff_t *);
Christoph Lameter0ff38492006-09-25 23:31:52 -0700452int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
453 struct file *, void __user *, size_t *, loff_t *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454
455#include <linux/topology.h>
456/* Returns the number of the current Node. */
Andi Kleen69d81fc2005-11-05 17:25:53 +0100457#ifndef numa_node_id
Ingo Molnar39c715b2005-06-21 17:14:34 -0700458#define numa_node_id() (cpu_to_node(raw_smp_processor_id()))
Andi Kleen69d81fc2005-11-05 17:25:53 +0100459#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460
Dave Hansen93b75042005-06-23 00:07:47 -0700461#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462
463extern struct pglist_data contig_page_data;
464#define NODE_DATA(nid) (&contig_page_data)
465#define NODE_MEM_MAP(nid) mem_map
466#define MAX_NODES_SHIFT 1
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467
Dave Hansen93b75042005-06-23 00:07:47 -0700468#else /* CONFIG_NEED_MULTIPLE_NODES */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469
470#include <asm/mmzone.h>
471
Dave Hansen93b75042005-06-23 00:07:47 -0700472#endif /* !CONFIG_NEED_MULTIPLE_NODES */
Dave Hansen348f8b62005-06-23 00:07:40 -0700473
KAMEZAWA Hiroyuki95144c72006-03-27 01:16:02 -0800474extern struct pglist_data *first_online_pgdat(void);
475extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
476extern struct zone *next_zone(struct zone *zone);
KAMEZAWA Hiroyuki8357f862006-03-27 01:15:57 -0800477
478/**
479 * for_each_pgdat - helper macro to iterate over all nodes
480 * @pgdat - pointer to a pg_data_t variable
481 */
482#define for_each_online_pgdat(pgdat) \
483 for (pgdat = first_online_pgdat(); \
484 pgdat; \
485 pgdat = next_online_pgdat(pgdat))
KAMEZAWA Hiroyuki8357f862006-03-27 01:15:57 -0800486/**
487 * for_each_zone - helper macro to iterate over all memory zones
488 * @zone - pointer to struct zone variable
489 *
490 * The user only needs to declare the zone variable, for_each_zone
491 * fills it in.
492 */
493#define for_each_zone(zone) \
494 for (zone = (first_online_pgdat())->node_zones; \
495 zone; \
496 zone = next_zone(zone))
497
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700498#ifdef CONFIG_SPARSEMEM
499#include <asm/sparsemem.h>
500#endif
501
Andi Kleen07808b72005-11-05 17:25:53 +0100502#if BITS_PER_LONG == 32
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503/*
Andi Kleena2f1b422005-11-05 17:25:53 +0100504 * with 32 bit page->flags field, we reserve 9 bits for node/zone info.
505 * there are 4 zones (3 bits) and this leaves 9-3=6 bits for nodes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 */
Andi Kleena2f1b422005-11-05 17:25:53 +0100507#define FLAGS_RESERVED 9
Dave Hansen348f8b62005-06-23 00:07:40 -0700508
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509#elif BITS_PER_LONG == 64
510/*
511 * with 64 bit flags field, there's plenty of room.
512 */
Dave Hansen348f8b62005-06-23 00:07:40 -0700513#define FLAGS_RESERVED 32
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514
Dave Hansen348f8b62005-06-23 00:07:40 -0700515#else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516
Dave Hansen348f8b62005-06-23 00:07:40 -0700517#error BITS_PER_LONG not defined
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519#endif
520
Andy Whitcroftb159d432005-06-23 00:07:52 -0700521#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
522#define early_pfn_to_nid(nid) (0UL)
523#endif
524
Andy Whitcroft2bdaf112006-01-06 00:10:53 -0800525#ifdef CONFIG_FLATMEM
526#define pfn_to_nid(pfn) (0)
527#endif
528
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700529#define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
530#define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
531
532#ifdef CONFIG_SPARSEMEM
533
534/*
535 * SECTION_SHIFT #bits space required to store a section #
536 *
537 * PA_SECTION_SHIFT physical address to/from section number
538 * PFN_SECTION_SHIFT pfn to/from section number
539 */
540#define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
541
542#define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
543#define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
544
545#define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
546
547#define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
548#define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
549
550#if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
551#error Allocator MAX_ORDER exceeds SECTION_SIZE
552#endif
553
554struct page;
555struct mem_section {
Andy Whitcroft29751f62005-06-23 00:08:00 -0700556 /*
557 * This is, logically, a pointer to an array of struct
558 * pages. However, it is stored with some other magic.
559 * (see sparse.c::sparse_init_one_section())
560 *
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700561 * Additionally during early boot we encode node id of
562 * the location of the section here to guide allocation.
563 * (see sparse.c::memory_present())
564 *
Andy Whitcroft29751f62005-06-23 00:08:00 -0700565 * Making it a UL at least makes someone do a cast
566 * before using it wrong.
567 */
568 unsigned long section_mem_map;
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700569};
570
Bob Picco3e347262005-09-03 15:54:28 -0700571#ifdef CONFIG_SPARSEMEM_EXTREME
572#define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
Bob Picco802f1922005-09-03 15:54:26 -0700573#else
Bob Picco3e347262005-09-03 15:54:28 -0700574#define SECTIONS_PER_ROOT 1
575#endif
Bob Picco802f1922005-09-03 15:54:26 -0700576
Bob Picco3e347262005-09-03 15:54:28 -0700577#define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
578#define NR_SECTION_ROOTS (NR_MEM_SECTIONS / SECTIONS_PER_ROOT)
579#define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
580
581#ifdef CONFIG_SPARSEMEM_EXTREME
582extern struct mem_section *mem_section[NR_SECTION_ROOTS];
583#else
584extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
585#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700586
Andy Whitcroft29751f62005-06-23 00:08:00 -0700587static inline struct mem_section *__nr_to_section(unsigned long nr)
588{
Bob Picco3e347262005-09-03 15:54:28 -0700589 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
590 return NULL;
591 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
Andy Whitcroft29751f62005-06-23 00:08:00 -0700592}
Dave Hansen4ca644d2005-10-29 18:16:51 -0700593extern int __section_nr(struct mem_section* ms);
Andy Whitcroft29751f62005-06-23 00:08:00 -0700594
595/*
596 * We use the lower bits of the mem_map pointer to store
597 * a little bit of information. There should be at least
598 * 3 bits here due to 32-bit alignment.
599 */
600#define SECTION_MARKED_PRESENT (1UL<<0)
601#define SECTION_HAS_MEM_MAP (1UL<<1)
602#define SECTION_MAP_LAST_BIT (1UL<<2)
603#define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700604#define SECTION_NID_SHIFT 2
Andy Whitcroft29751f62005-06-23 00:08:00 -0700605
606static inline struct page *__section_mem_map_addr(struct mem_section *section)
607{
608 unsigned long map = section->section_mem_map;
609 map &= SECTION_MAP_MASK;
610 return (struct page *)map;
611}
612
613static inline int valid_section(struct mem_section *section)
614{
Bob Picco802f1922005-09-03 15:54:26 -0700615 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
Andy Whitcroft29751f62005-06-23 00:08:00 -0700616}
617
618static inline int section_has_mem_map(struct mem_section *section)
619{
Bob Picco802f1922005-09-03 15:54:26 -0700620 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
Andy Whitcroft29751f62005-06-23 00:08:00 -0700621}
622
623static inline int valid_section_nr(unsigned long nr)
624{
625 return valid_section(__nr_to_section(nr));
626}
627
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700628static inline struct mem_section *__pfn_to_section(unsigned long pfn)
629{
Andy Whitcroft29751f62005-06-23 00:08:00 -0700630 return __nr_to_section(pfn_to_section_nr(pfn));
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700631}
632
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700633static inline int pfn_valid(unsigned long pfn)
634{
635 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
636 return 0;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700637 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700638}
639
640/*
641 * These are _only_ used during initialisation, therefore they
642 * can use __initdata ... They could have names to indicate
643 * this restriction.
644 */
645#ifdef CONFIG_NUMA
Andy Whitcroft161599f2006-01-06 00:10:54 -0800646#define pfn_to_nid(pfn) \
647({ \
648 unsigned long __pfn_to_nid_pfn = (pfn); \
649 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
650})
Andy Whitcroft2bdaf112006-01-06 00:10:53 -0800651#else
652#define pfn_to_nid(pfn) (0)
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700653#endif
654
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700655#define early_pfn_valid(pfn) pfn_valid(pfn)
656void sparse_init(void);
657#else
658#define sparse_init() do {} while (0)
Dave Hansen28ae55c2005-09-03 15:54:29 -0700659#define sparse_index_init(_sec, _nid) do {} while (0)
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700660#endif /* CONFIG_SPARSEMEM */
661
662#ifndef early_pfn_valid
663#define early_pfn_valid(pfn) (1)
664#endif
665
666void memory_present(int nid, unsigned long start, unsigned long end);
667unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
668
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669#endif /* !__ASSEMBLY__ */
670#endif /* __KERNEL__ */
671#endif /* _LINUX_MMZONE_H */