[PATCH] Sparsemem build fix
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / include / linux / mmzone.h
blob2d8337150493a9ffe7eb9060b6ee84c5a3c45fb1
1 #ifndef _LINUX_MMZONE_H
2 #define _LINUX_MMZONE_H
4 #ifdef __KERNEL__
5 #ifndef __ASSEMBLY__
7 #include <linux/config.h>
8 #include <linux/spinlock.h>
9 #include <linux/list.h>
10 #include <linux/wait.h>
11 #include <linux/cache.h>
12 #include <linux/threads.h>
13 #include <linux/numa.h>
14 #include <linux/init.h>
15 #include <linux/seqlock.h>
16 #include <linux/nodemask.h>
17 #include <asm/atomic.h>
18 #include <asm/page.h>
20 /* Free memory management - zoned buddy allocator. */
21 #ifndef CONFIG_FORCE_MAX_ZONEORDER
22 #define MAX_ORDER 11
23 #else
24 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
25 #endif
26 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
28 struct free_area {
29 struct list_head free_list;
30 unsigned long nr_free;
33 struct pglist_data;
36 * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
37 * So add a wild amount of padding here to ensure that they fall into separate
38 * cachelines. There are very few zone structures in the machine, so space
39 * consumption is not a concern here.
41 #if defined(CONFIG_SMP)
42 struct zone_padding {
43 char x[0];
44 } ____cacheline_internodealigned_in_smp;
45 #define ZONE_PADDING(name) struct zone_padding name;
46 #else
47 #define ZONE_PADDING(name)
48 #endif
50 struct per_cpu_pages {
51 int count; /* number of pages in the list */
52 int high; /* high watermark, emptying needed */
53 int batch; /* chunk size for buddy add/remove */
54 struct list_head list; /* the list of pages */
57 struct per_cpu_pageset {
58 struct per_cpu_pages pcp[2]; /* 0: hot. 1: cold */
59 #ifdef CONFIG_NUMA
60 unsigned long numa_hit; /* allocated in intended node */
61 unsigned long numa_miss; /* allocated in non intended node */
62 unsigned long numa_foreign; /* was intended here, hit elsewhere */
63 unsigned long interleave_hit; /* interleaver prefered this zone */
64 unsigned long local_node; /* allocation from local node */
65 unsigned long other_node; /* allocation from other node */
66 #endif
67 } ____cacheline_aligned_in_smp;
69 #ifdef CONFIG_NUMA
70 #define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)])
71 #else
72 #define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)])
73 #endif
75 #define ZONE_DMA 0
76 #define ZONE_DMA32 1
77 #define ZONE_NORMAL 2
78 #define ZONE_HIGHMEM 3
80 #define MAX_NR_ZONES 4 /* Sync this with ZONES_SHIFT */
81 #define ZONES_SHIFT 2 /* ceil(log2(MAX_NR_ZONES)) */
85 * When a memory allocation must conform to specific limitations (such
86 * as being suitable for DMA) the caller will pass in hints to the
87 * allocator in the gfp_mask, in the zone modifier bits. These bits
88 * are used to select a priority ordered list of memory zones which
89 * match the requested limits. GFP_ZONEMASK defines which bits within
90 * the gfp_mask should be considered as zone modifiers. Each valid
91 * combination of the zone modifier bits has a corresponding list
92 * of zones (in node_zonelists). Thus for two zone modifiers there
93 * will be a maximum of 4 (2 ** 2) zonelists, for 3 modifiers there will
94 * be 8 (2 ** 3) zonelists. GFP_ZONETYPES defines the number of possible
95 * combinations of zone modifiers in "zone modifier space".
97 * As an optimisation any zone modifier bits which are only valid when
98 * no other zone modifier bits are set (loners) should be placed in
99 * the highest order bits of this field. This allows us to reduce the
100 * extent of the zonelists thus saving space. For example in the case
101 * of three zone modifier bits, we could require up to eight zonelists.
102 * If the left most zone modifier is a "loner" then the highest valid
103 * zonelist would be four allowing us to allocate only five zonelists.
104 * Use the first form for GFP_ZONETYPES when the left most bit is not
105 * a "loner", otherwise use the second.
107 * NOTE! Make sure this matches the zones in <linux/gfp.h>
109 #define GFP_ZONEMASK 0x07
110 /* #define GFP_ZONETYPES (GFP_ZONEMASK + 1) */ /* Non-loner */
111 #define GFP_ZONETYPES ((GFP_ZONEMASK + 1) / 2 + 1) /* Loner */
114 * On machines where it is needed (eg PCs) we divide physical memory
115 * into multiple physical zones. On a 32bit PC we have 4 zones:
117 * ZONE_DMA < 16 MB ISA DMA capable memory
118 * ZONE_DMA32 0 MB Empty
119 * ZONE_NORMAL 16-896 MB direct mapped by the kernel
120 * ZONE_HIGHMEM > 896 MB only page cache and user processes
123 struct zone {
124 /* Fields commonly accessed by the page allocator */
125 unsigned long free_pages;
126 unsigned long pages_min, pages_low, pages_high;
128 * We don't know if the memory that we're going to allocate will be freeable
129 * or/and it will be released eventually, so to avoid totally wasting several
130 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
131 * to run OOM on the lower zones despite there's tons of freeable ram
132 * on the higher zones). This array is recalculated at runtime if the
133 * sysctl_lowmem_reserve_ratio sysctl changes.
135 unsigned long lowmem_reserve[MAX_NR_ZONES];
137 #ifdef CONFIG_NUMA
138 struct per_cpu_pageset *pageset[NR_CPUS];
139 #else
140 struct per_cpu_pageset pageset[NR_CPUS];
141 #endif
143 * free areas of different sizes
145 spinlock_t lock;
146 #ifdef CONFIG_MEMORY_HOTPLUG
147 /* see spanned/present_pages for more description */
148 seqlock_t span_seqlock;
149 #endif
150 struct free_area free_area[MAX_ORDER];
153 ZONE_PADDING(_pad1_)
155 /* Fields commonly accessed by the page reclaim scanner */
156 spinlock_t lru_lock;
157 struct list_head active_list;
158 struct list_head inactive_list;
159 unsigned long nr_scan_active;
160 unsigned long nr_scan_inactive;
161 unsigned long nr_active;
162 unsigned long nr_inactive;
163 unsigned long pages_scanned; /* since last reclaim */
164 int all_unreclaimable; /* All pages pinned */
166 /* A count of how many reclaimers are scanning this zone */
167 atomic_t reclaim_in_progress;
170 * timestamp (in jiffies) of the last zone reclaim that did not
171 * result in freeing of pages. This is used to avoid repeated scans
172 * if all memory in the zone is in use.
174 unsigned long last_unsuccessful_zone_reclaim;
177 * prev_priority holds the scanning priority for this zone. It is
178 * defined as the scanning priority at which we achieved our reclaim
179 * target at the previous try_to_free_pages() or balance_pgdat()
180 * invokation.
182 * We use prev_priority as a measure of how much stress page reclaim is
183 * under - it drives the swappiness decision: whether to unmap mapped
184 * pages.
186 * temp_priority is used to remember the scanning priority at which
187 * this zone was successfully refilled to free_pages == pages_high.
189 * Access to both these fields is quite racy even on uniprocessor. But
190 * it is expected to average out OK.
192 int temp_priority;
193 int prev_priority;
196 ZONE_PADDING(_pad2_)
197 /* Rarely used or read-mostly fields */
200 * wait_table -- the array holding the hash table
201 * wait_table_size -- the size of the hash table array
202 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
204 * The purpose of all these is to keep track of the people
205 * waiting for a page to become available and make them
206 * runnable again when possible. The trouble is that this
207 * consumes a lot of space, especially when so few things
208 * wait on pages at a given time. So instead of using
209 * per-page waitqueues, we use a waitqueue hash table.
211 * The bucket discipline is to sleep on the same queue when
212 * colliding and wake all in that wait queue when removing.
213 * When something wakes, it must check to be sure its page is
214 * truly available, a la thundering herd. The cost of a
215 * collision is great, but given the expected load of the
216 * table, they should be so rare as to be outweighed by the
217 * benefits from the saved space.
219 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
220 * primary users of these fields, and in mm/page_alloc.c
221 * free_area_init_core() performs the initialization of them.
223 wait_queue_head_t * wait_table;
224 unsigned long wait_table_size;
225 unsigned long wait_table_bits;
228 * Discontig memory support fields.
230 struct pglist_data *zone_pgdat;
231 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
232 unsigned long zone_start_pfn;
235 * zone_start_pfn, spanned_pages and present_pages are all
236 * protected by span_seqlock. It is a seqlock because it has
237 * to be read outside of zone->lock, and it is done in the main
238 * allocator path. But, it is written quite infrequently.
240 * The lock is declared along with zone->lock because it is
241 * frequently read in proximity to zone->lock. It's good to
242 * give them a chance of being in the same cacheline.
244 unsigned long spanned_pages; /* total size, including holes */
245 unsigned long present_pages; /* amount of memory (excluding holes) */
248 * rarely used fields:
250 char *name;
251 } ____cacheline_internodealigned_in_smp;
255 * The "priority" of VM scanning is how much of the queues we will scan in one
256 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
257 * queues ("queue_length >> 12") during an aging round.
259 #define DEF_PRIORITY 12
262 * One allocation request operates on a zonelist. A zonelist
263 * is a list of zones, the first one is the 'goal' of the
264 * allocation, the other zones are fallback zones, in decreasing
265 * priority.
267 * Right now a zonelist takes up less than a cacheline. We never
268 * modify it apart from boot-up, and only a few indices are used,
269 * so despite the zonelist table being relatively big, the cache
270 * footprint of this construct is very small.
272 struct zonelist {
273 struct zone *zones[MAX_NUMNODES * MAX_NR_ZONES + 1]; // NULL delimited
278 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
279 * (mostly NUMA machines?) to denote a higher-level memory zone than the
280 * zone denotes.
282 * On NUMA machines, each NUMA node would have a pg_data_t to describe
283 * it's memory layout.
285 * Memory statistics and page replacement data structures are maintained on a
286 * per-zone basis.
288 struct bootmem_data;
289 typedef struct pglist_data {
290 struct zone node_zones[MAX_NR_ZONES];
291 struct zonelist node_zonelists[GFP_ZONETYPES];
292 int nr_zones;
293 #ifdef CONFIG_FLAT_NODE_MEM_MAP
294 struct page *node_mem_map;
295 #endif
296 struct bootmem_data *bdata;
297 #ifdef CONFIG_MEMORY_HOTPLUG
299 * Must be held any time you expect node_start_pfn, node_present_pages
300 * or node_spanned_pages stay constant. Holding this will also
301 * guarantee that any pfn_valid() stays that way.
303 * Nests above zone->lock and zone->size_seqlock.
305 spinlock_t node_size_lock;
306 #endif
307 unsigned long node_start_pfn;
308 unsigned long node_present_pages; /* total number of physical pages */
309 unsigned long node_spanned_pages; /* total size of physical page
310 range, including holes */
311 int node_id;
312 wait_queue_head_t kswapd_wait;
313 struct task_struct *kswapd;
314 int kswapd_max_order;
315 } pg_data_t;
317 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
318 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
319 #ifdef CONFIG_FLAT_NODE_MEM_MAP
320 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
321 #else
322 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
323 #endif
324 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
326 #include <linux/memory_hotplug.h>
328 void __get_zone_counts(unsigned long *active, unsigned long *inactive,
329 unsigned long *free, struct pglist_data *pgdat);
330 void get_zone_counts(unsigned long *active, unsigned long *inactive,
331 unsigned long *free);
332 void build_all_zonelists(void);
333 void wakeup_kswapd(struct zone *zone, int order);
334 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
335 int classzone_idx, int alloc_flags);
337 #ifdef CONFIG_HAVE_MEMORY_PRESENT
338 void memory_present(int nid, unsigned long start, unsigned long end);
339 #else
340 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
341 #endif
343 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
344 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
345 #endif
348 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
350 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
352 static inline int populated_zone(struct zone *zone)
354 return (!!zone->present_pages);
357 static inline int is_highmem_idx(int idx)
359 return (idx == ZONE_HIGHMEM);
362 static inline int is_normal_idx(int idx)
364 return (idx == ZONE_NORMAL);
368 * is_highmem - helper function to quickly check if a struct zone is a
369 * highmem zone or not. This is an attempt to keep references
370 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
371 * @zone - pointer to struct zone variable
373 static inline int is_highmem(struct zone *zone)
375 return zone == zone->zone_pgdat->node_zones + ZONE_HIGHMEM;
378 static inline int is_normal(struct zone *zone)
380 return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
383 static inline int is_dma32(struct zone *zone)
385 return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
388 static inline int is_dma(struct zone *zone)
390 return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
393 /* These two functions are used to setup the per zone pages min values */
394 struct ctl_table;
395 struct file;
396 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *,
397 void __user *, size_t *, loff_t *);
398 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
399 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *,
400 void __user *, size_t *, loff_t *);
401 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, struct file *,
402 void __user *, size_t *, loff_t *);
404 #include <linux/topology.h>
405 /* Returns the number of the current Node. */
406 #ifndef numa_node_id
407 #define numa_node_id() (cpu_to_node(raw_smp_processor_id()))
408 #endif
410 #ifndef CONFIG_NEED_MULTIPLE_NODES
412 extern struct pglist_data contig_page_data;
413 #define NODE_DATA(nid) (&contig_page_data)
414 #define NODE_MEM_MAP(nid) mem_map
415 #define MAX_NODES_SHIFT 1
417 #else /* CONFIG_NEED_MULTIPLE_NODES */
419 #include <asm/mmzone.h>
421 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
423 extern struct pglist_data *first_online_pgdat(void);
424 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
425 extern struct zone *next_zone(struct zone *zone);
428 * for_each_pgdat - helper macro to iterate over all nodes
429 * @pgdat - pointer to a pg_data_t variable
431 #define for_each_online_pgdat(pgdat) \
432 for (pgdat = first_online_pgdat(); \
433 pgdat; \
434 pgdat = next_online_pgdat(pgdat))
436 * for_each_zone - helper macro to iterate over all memory zones
437 * @zone - pointer to struct zone variable
439 * The user only needs to declare the zone variable, for_each_zone
440 * fills it in.
442 #define for_each_zone(zone) \
443 for (zone = (first_online_pgdat())->node_zones; \
444 zone; \
445 zone = next_zone(zone))
447 #ifdef CONFIG_SPARSEMEM
448 #include <asm/sparsemem.h>
449 #endif
451 #if BITS_PER_LONG == 32
453 * with 32 bit page->flags field, we reserve 9 bits for node/zone info.
454 * there are 4 zones (3 bits) and this leaves 9-3=6 bits for nodes.
456 #define FLAGS_RESERVED 9
458 #elif BITS_PER_LONG == 64
460 * with 64 bit flags field, there's plenty of room.
462 #define FLAGS_RESERVED 32
464 #else
466 #error BITS_PER_LONG not defined
468 #endif
470 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
471 #define early_pfn_to_nid(nid) (0UL)
472 #endif
474 #ifdef CONFIG_FLATMEM
475 #define pfn_to_nid(pfn) (0)
476 #endif
478 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
479 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
481 #ifdef CONFIG_SPARSEMEM
484 * SECTION_SHIFT #bits space required to store a section #
486 * PA_SECTION_SHIFT physical address to/from section number
487 * PFN_SECTION_SHIFT pfn to/from section number
489 #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
491 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
492 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
494 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
496 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
497 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
499 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
500 #error Allocator MAX_ORDER exceeds SECTION_SIZE
501 #endif
503 struct page;
504 struct mem_section {
506 * This is, logically, a pointer to an array of struct
507 * pages. However, it is stored with some other magic.
508 * (see sparse.c::sparse_init_one_section())
510 * Making it a UL at least makes someone do a cast
511 * before using it wrong.
513 unsigned long section_mem_map;
516 #ifdef CONFIG_SPARSEMEM_EXTREME
517 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
518 #else
519 #define SECTIONS_PER_ROOT 1
520 #endif
522 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
523 #define NR_SECTION_ROOTS (NR_MEM_SECTIONS / SECTIONS_PER_ROOT)
524 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
526 #ifdef CONFIG_SPARSEMEM_EXTREME
527 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
528 #else
529 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
530 #endif
532 static inline struct mem_section *__nr_to_section(unsigned long nr)
534 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
535 return NULL;
536 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
538 extern int __section_nr(struct mem_section* ms);
541 * We use the lower bits of the mem_map pointer to store
542 * a little bit of information. There should be at least
543 * 3 bits here due to 32-bit alignment.
545 #define SECTION_MARKED_PRESENT (1UL<<0)
546 #define SECTION_HAS_MEM_MAP (1UL<<1)
547 #define SECTION_MAP_LAST_BIT (1UL<<2)
548 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
550 static inline struct page *__section_mem_map_addr(struct mem_section *section)
552 unsigned long map = section->section_mem_map;
553 map &= SECTION_MAP_MASK;
554 return (struct page *)map;
557 static inline int valid_section(struct mem_section *section)
559 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
562 static inline int section_has_mem_map(struct mem_section *section)
564 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
567 static inline int valid_section_nr(unsigned long nr)
569 return valid_section(__nr_to_section(nr));
572 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
574 return __nr_to_section(pfn_to_section_nr(pfn));
577 static inline int pfn_valid(unsigned long pfn)
579 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
580 return 0;
581 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
585 * These are _only_ used during initialisation, therefore they
586 * can use __initdata ... They could have names to indicate
587 * this restriction.
589 #ifdef CONFIG_NUMA
590 #define pfn_to_nid(pfn) \
591 ({ \
592 unsigned long __pfn_to_nid_pfn = (pfn); \
593 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
595 #else
596 #define pfn_to_nid(pfn) (0)
597 #endif
599 #define early_pfn_valid(pfn) pfn_valid(pfn)
600 void sparse_init(void);
601 #else
602 #define sparse_init() do {} while (0)
603 #define sparse_index_init(_sec, _nid) do {} while (0)
604 #endif /* CONFIG_SPARSEMEM */
606 #ifndef early_pfn_valid
607 #define early_pfn_valid(pfn) (1)
608 #endif
610 void memory_present(int nid, unsigned long start, unsigned long end);
611 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
613 #endif /* !__ASSEMBLY__ */
614 #endif /* __KERNEL__ */
615 #endif /* _LINUX_MMZONE_H */