[ALSA] Merge ad1816a-lib module to ad1816a
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / include / linux / mmzone.h
blob9f22090df7dde381a177707cf745733ececbd320
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 <asm/atomic.h>
18 /* Free memory management - zoned buddy allocator. */
19 #ifndef CONFIG_FORCE_MAX_ZONEORDER
20 #define MAX_ORDER 11
21 #else
22 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
23 #endif
25 struct free_area {
26 struct list_head free_list;
27 unsigned long nr_free;
30 struct pglist_data;
33 * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
34 * So add a wild amount of padding here to ensure that they fall into separate
35 * cachelines. There are very few zone structures in the machine, so space
36 * consumption is not a concern here.
38 #if defined(CONFIG_SMP)
39 struct zone_padding {
40 char x[0];
41 } ____cacheline_maxaligned_in_smp;
42 #define ZONE_PADDING(name) struct zone_padding name;
43 #else
44 #define ZONE_PADDING(name)
45 #endif
47 struct per_cpu_pages {
48 int count; /* number of pages in the list */
49 int low; /* low watermark, refill needed */
50 int high; /* high watermark, emptying needed */
51 int batch; /* chunk size for buddy add/remove */
52 struct list_head list; /* the list of pages */
55 struct per_cpu_pageset {
56 struct per_cpu_pages pcp[2]; /* 0: hot. 1: cold */
57 #ifdef CONFIG_NUMA
58 unsigned long numa_hit; /* allocated in intended node */
59 unsigned long numa_miss; /* allocated in non intended node */
60 unsigned long numa_foreign; /* was intended here, hit elsewhere */
61 unsigned long interleave_hit; /* interleaver prefered this zone */
62 unsigned long local_node; /* allocation from local node */
63 unsigned long other_node; /* allocation from other node */
64 #endif
65 } ____cacheline_aligned_in_smp;
67 #ifdef CONFIG_NUMA
68 #define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)])
69 #else
70 #define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)])
71 #endif
73 #define ZONE_DMA 0
74 #define ZONE_DMA32 1
75 #define ZONE_NORMAL 2
76 #define ZONE_HIGHMEM 3
78 #define MAX_NR_ZONES 4 /* Sync this with ZONES_SHIFT */
79 #define ZONES_SHIFT 2 /* ceil(log2(MAX_NR_ZONES)) */
83 * When a memory allocation must conform to specific limitations (such
84 * as being suitable for DMA) the caller will pass in hints to the
85 * allocator in the gfp_mask, in the zone modifier bits. These bits
86 * are used to select a priority ordered list of memory zones which
87 * match the requested limits. GFP_ZONEMASK defines which bits within
88 * the gfp_mask should be considered as zone modifiers. Each valid
89 * combination of the zone modifier bits has a corresponding list
90 * of zones (in node_zonelists). Thus for two zone modifiers there
91 * will be a maximum of 4 (2 ** 2) zonelists, for 3 modifiers there will
92 * be 8 (2 ** 3) zonelists. GFP_ZONETYPES defines the number of possible
93 * combinations of zone modifiers in "zone modifier space".
95 * NOTE! Make sure this matches the zones in <linux/gfp.h>
97 #define GFP_ZONEMASK 0x07
98 #define GFP_ZONETYPES 5
101 * On machines where it is needed (eg PCs) we divide physical memory
102 * into multiple physical zones. On a PC we have 4 zones:
104 * ZONE_DMA < 16 MB ISA DMA capable memory
105 * ZONE_DMA32 0 MB Empty
106 * ZONE_NORMAL 16-896 MB direct mapped by the kernel
107 * ZONE_HIGHMEM > 896 MB only page cache and user processes
110 struct zone {
111 /* Fields commonly accessed by the page allocator */
112 unsigned long free_pages;
113 unsigned long pages_min, pages_low, pages_high;
115 * We don't know if the memory that we're going to allocate will be freeable
116 * or/and it will be released eventually, so to avoid totally wasting several
117 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
118 * to run OOM on the lower zones despite there's tons of freeable ram
119 * on the higher zones). This array is recalculated at runtime if the
120 * sysctl_lowmem_reserve_ratio sysctl changes.
122 unsigned long lowmem_reserve[MAX_NR_ZONES];
124 #ifdef CONFIG_NUMA
125 struct per_cpu_pageset *pageset[NR_CPUS];
126 #else
127 struct per_cpu_pageset pageset[NR_CPUS];
128 #endif
130 * free areas of different sizes
132 spinlock_t lock;
133 #ifdef CONFIG_MEMORY_HOTPLUG
134 /* see spanned/present_pages for more description */
135 seqlock_t span_seqlock;
136 #endif
137 struct free_area free_area[MAX_ORDER];
140 ZONE_PADDING(_pad1_)
142 /* Fields commonly accessed by the page reclaim scanner */
143 spinlock_t lru_lock;
144 struct list_head active_list;
145 struct list_head inactive_list;
146 unsigned long nr_scan_active;
147 unsigned long nr_scan_inactive;
148 unsigned long nr_active;
149 unsigned long nr_inactive;
150 unsigned long pages_scanned; /* since last reclaim */
151 int all_unreclaimable; /* All pages pinned */
154 * Does the allocator try to reclaim pages from the zone as soon
155 * as it fails a watermark_ok() in __alloc_pages?
157 int reclaim_pages;
158 /* A count of how many reclaimers are scanning this zone */
159 atomic_t reclaim_in_progress;
162 * prev_priority holds the scanning priority for this zone. It is
163 * defined as the scanning priority at which we achieved our reclaim
164 * target at the previous try_to_free_pages() or balance_pgdat()
165 * invokation.
167 * We use prev_priority as a measure of how much stress page reclaim is
168 * under - it drives the swappiness decision: whether to unmap mapped
169 * pages.
171 * temp_priority is used to remember the scanning priority at which
172 * this zone was successfully refilled to free_pages == pages_high.
174 * Access to both these fields is quite racy even on uniprocessor. But
175 * it is expected to average out OK.
177 int temp_priority;
178 int prev_priority;
181 ZONE_PADDING(_pad2_)
182 /* Rarely used or read-mostly fields */
185 * wait_table -- the array holding the hash table
186 * wait_table_size -- the size of the hash table array
187 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
189 * The purpose of all these is to keep track of the people
190 * waiting for a page to become available and make them
191 * runnable again when possible. The trouble is that this
192 * consumes a lot of space, especially when so few things
193 * wait on pages at a given time. So instead of using
194 * per-page waitqueues, we use a waitqueue hash table.
196 * The bucket discipline is to sleep on the same queue when
197 * colliding and wake all in that wait queue when removing.
198 * When something wakes, it must check to be sure its page is
199 * truly available, a la thundering herd. The cost of a
200 * collision is great, but given the expected load of the
201 * table, they should be so rare as to be outweighed by the
202 * benefits from the saved space.
204 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
205 * primary users of these fields, and in mm/page_alloc.c
206 * free_area_init_core() performs the initialization of them.
208 wait_queue_head_t * wait_table;
209 unsigned long wait_table_size;
210 unsigned long wait_table_bits;
213 * Discontig memory support fields.
215 struct pglist_data *zone_pgdat;
216 struct page *zone_mem_map;
217 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
218 unsigned long zone_start_pfn;
221 * zone_start_pfn, spanned_pages and present_pages are all
222 * protected by span_seqlock. It is a seqlock because it has
223 * to be read outside of zone->lock, and it is done in the main
224 * allocator path. But, it is written quite infrequently.
226 * The lock is declared along with zone->lock because it is
227 * frequently read in proximity to zone->lock. It's good to
228 * give them a chance of being in the same cacheline.
230 unsigned long spanned_pages; /* total size, including holes */
231 unsigned long present_pages; /* amount of memory (excluding holes) */
234 * rarely used fields:
236 char *name;
237 } ____cacheline_maxaligned_in_smp;
241 * The "priority" of VM scanning is how much of the queues we will scan in one
242 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
243 * queues ("queue_length >> 12") during an aging round.
245 #define DEF_PRIORITY 12
248 * One allocation request operates on a zonelist. A zonelist
249 * is a list of zones, the first one is the 'goal' of the
250 * allocation, the other zones are fallback zones, in decreasing
251 * priority.
253 * Right now a zonelist takes up less than a cacheline. We never
254 * modify it apart from boot-up, and only a few indices are used,
255 * so despite the zonelist table being relatively big, the cache
256 * footprint of this construct is very small.
258 struct zonelist {
259 struct zone *zones[MAX_NUMNODES * MAX_NR_ZONES + 1]; // NULL delimited
264 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
265 * (mostly NUMA machines?) to denote a higher-level memory zone than the
266 * zone denotes.
268 * On NUMA machines, each NUMA node would have a pg_data_t to describe
269 * it's memory layout.
271 * Memory statistics and page replacement data structures are maintained on a
272 * per-zone basis.
274 struct bootmem_data;
275 typedef struct pglist_data {
276 struct zone node_zones[MAX_NR_ZONES];
277 struct zonelist node_zonelists[GFP_ZONETYPES];
278 int nr_zones;
279 #ifdef CONFIG_FLAT_NODE_MEM_MAP
280 struct page *node_mem_map;
281 #endif
282 struct bootmem_data *bdata;
283 #ifdef CONFIG_MEMORY_HOTPLUG
285 * Must be held any time you expect node_start_pfn, node_present_pages
286 * or node_spanned_pages stay constant. Holding this will also
287 * guarantee that any pfn_valid() stays that way.
289 * Nests above zone->lock and zone->size_seqlock.
291 spinlock_t node_size_lock;
292 #endif
293 unsigned long node_start_pfn;
294 unsigned long node_present_pages; /* total number of physical pages */
295 unsigned long node_spanned_pages; /* total size of physical page
296 range, including holes */
297 int node_id;
298 struct pglist_data *pgdat_next;
299 wait_queue_head_t kswapd_wait;
300 struct task_struct *kswapd;
301 int kswapd_max_order;
302 } pg_data_t;
304 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
305 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
306 #ifdef CONFIG_FLAT_NODE_MEM_MAP
307 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
308 #else
309 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
310 #endif
311 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
313 #include <linux/memory_hotplug.h>
315 extern struct pglist_data *pgdat_list;
317 void __get_zone_counts(unsigned long *active, unsigned long *inactive,
318 unsigned long *free, struct pglist_data *pgdat);
319 void get_zone_counts(unsigned long *active, unsigned long *inactive,
320 unsigned long *free);
321 void build_all_zonelists(void);
322 void wakeup_kswapd(struct zone *zone, int order);
323 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
324 int classzone_idx, int alloc_flags);
326 #ifdef CONFIG_HAVE_MEMORY_PRESENT
327 void memory_present(int nid, unsigned long start, unsigned long end);
328 #else
329 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
330 #endif
332 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
333 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
334 #endif
337 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
339 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
342 * for_each_pgdat - helper macro to iterate over all nodes
343 * @pgdat - pointer to a pg_data_t variable
345 * Meant to help with common loops of the form
346 * pgdat = pgdat_list;
347 * while(pgdat) {
348 * ...
349 * pgdat = pgdat->pgdat_next;
352 #define for_each_pgdat(pgdat) \
353 for (pgdat = pgdat_list; pgdat; pgdat = pgdat->pgdat_next)
356 * next_zone - helper magic for for_each_zone()
357 * Thanks to William Lee Irwin III for this piece of ingenuity.
359 static inline struct zone *next_zone(struct zone *zone)
361 pg_data_t *pgdat = zone->zone_pgdat;
363 if (zone < pgdat->node_zones + MAX_NR_ZONES - 1)
364 zone++;
365 else if (pgdat->pgdat_next) {
366 pgdat = pgdat->pgdat_next;
367 zone = pgdat->node_zones;
368 } else
369 zone = NULL;
371 return zone;
375 * for_each_zone - helper macro to iterate over all memory zones
376 * @zone - pointer to struct zone variable
378 * The user only needs to declare the zone variable, for_each_zone
379 * fills it in. This basically means for_each_zone() is an
380 * easier to read version of this piece of code:
382 * for (pgdat = pgdat_list; pgdat; pgdat = pgdat->node_next)
383 * for (i = 0; i < MAX_NR_ZONES; ++i) {
384 * struct zone * z = pgdat->node_zones + i;
385 * ...
389 #define for_each_zone(zone) \
390 for (zone = pgdat_list->node_zones; zone; zone = next_zone(zone))
392 static inline int is_highmem_idx(int idx)
394 return (idx == ZONE_HIGHMEM);
397 static inline int is_normal_idx(int idx)
399 return (idx == ZONE_NORMAL);
402 * is_highmem - helper function to quickly check if a struct zone is a
403 * highmem zone or not. This is an attempt to keep references
404 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
405 * @zone - pointer to struct zone variable
407 static inline int is_highmem(struct zone *zone)
409 return zone == zone->zone_pgdat->node_zones + ZONE_HIGHMEM;
412 static inline int is_normal(struct zone *zone)
414 return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
417 /* These two functions are used to setup the per zone pages min values */
418 struct ctl_table;
419 struct file;
420 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *,
421 void __user *, size_t *, loff_t *);
422 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
423 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *,
424 void __user *, size_t *, loff_t *);
426 #include <linux/topology.h>
427 /* Returns the number of the current Node. */
428 #ifndef numa_node_id
429 #define numa_node_id() (cpu_to_node(raw_smp_processor_id()))
430 #endif
432 #ifndef CONFIG_NEED_MULTIPLE_NODES
434 extern struct pglist_data contig_page_data;
435 #define NODE_DATA(nid) (&contig_page_data)
436 #define NODE_MEM_MAP(nid) mem_map
437 #define MAX_NODES_SHIFT 1
438 #define pfn_to_nid(pfn) (0)
440 #else /* CONFIG_NEED_MULTIPLE_NODES */
442 #include <asm/mmzone.h>
444 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
446 #ifdef CONFIG_SPARSEMEM
447 #include <asm/sparsemem.h>
448 #endif
450 #if BITS_PER_LONG == 32
452 * with 32 bit page->flags field, we reserve 9 bits for node/zone info.
453 * there are 4 zones (3 bits) and this leaves 9-3=6 bits for nodes.
455 #define FLAGS_RESERVED 9
457 #elif BITS_PER_LONG == 64
459 * with 64 bit flags field, there's plenty of room.
461 #define FLAGS_RESERVED 32
463 #else
465 #error BITS_PER_LONG not defined
467 #endif
469 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
470 #define early_pfn_to_nid(nid) (0UL)
471 #endif
473 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
474 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
476 #ifdef CONFIG_SPARSEMEM
479 * SECTION_SHIFT #bits space required to store a section #
481 * PA_SECTION_SHIFT physical address to/from section number
482 * PFN_SECTION_SHIFT pfn to/from section number
484 #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
486 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
487 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
489 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
491 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
492 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
494 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
495 #error Allocator MAX_ORDER exceeds SECTION_SIZE
496 #endif
498 struct page;
499 struct mem_section {
501 * This is, logically, a pointer to an array of struct
502 * pages. However, it is stored with some other magic.
503 * (see sparse.c::sparse_init_one_section())
505 * Making it a UL at least makes someone do a cast
506 * before using it wrong.
508 unsigned long section_mem_map;
511 #ifdef CONFIG_SPARSEMEM_EXTREME
512 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
513 #else
514 #define SECTIONS_PER_ROOT 1
515 #endif
517 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
518 #define NR_SECTION_ROOTS (NR_MEM_SECTIONS / SECTIONS_PER_ROOT)
519 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
521 #ifdef CONFIG_SPARSEMEM_EXTREME
522 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
523 #else
524 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
525 #endif
527 static inline struct mem_section *__nr_to_section(unsigned long nr)
529 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
530 return NULL;
531 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
533 extern int __section_nr(struct mem_section* ms);
536 * We use the lower bits of the mem_map pointer to store
537 * a little bit of information. There should be at least
538 * 3 bits here due to 32-bit alignment.
540 #define SECTION_MARKED_PRESENT (1UL<<0)
541 #define SECTION_HAS_MEM_MAP (1UL<<1)
542 #define SECTION_MAP_LAST_BIT (1UL<<2)
543 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
545 static inline struct page *__section_mem_map_addr(struct mem_section *section)
547 unsigned long map = section->section_mem_map;
548 map &= SECTION_MAP_MASK;
549 return (struct page *)map;
552 static inline int valid_section(struct mem_section *section)
554 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
557 static inline int section_has_mem_map(struct mem_section *section)
559 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
562 static inline int valid_section_nr(unsigned long nr)
564 return valid_section(__nr_to_section(nr));
568 * Given a kernel address, find the home node of the underlying memory.
570 #define kvaddr_to_nid(kaddr) pfn_to_nid(__pa(kaddr) >> PAGE_SHIFT)
572 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
574 return __nr_to_section(pfn_to_section_nr(pfn));
577 #define pfn_to_page(pfn) \
578 ({ \
579 unsigned long __pfn = (pfn); \
580 __section_mem_map_addr(__pfn_to_section(__pfn)) + __pfn; \
582 #define page_to_pfn(page) \
583 ({ \
584 page - __section_mem_map_addr(__nr_to_section( \
585 page_to_section(page))); \
588 static inline int pfn_valid(unsigned long pfn)
590 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
591 return 0;
592 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
596 * These are _only_ used during initialisation, therefore they
597 * can use __initdata ... They could have names to indicate
598 * this restriction.
600 #ifdef CONFIG_NUMA
601 #define pfn_to_nid early_pfn_to_nid
602 #endif
604 #define pfn_to_pgdat(pfn) \
605 ({ \
606 NODE_DATA(pfn_to_nid(pfn)); \
609 #define early_pfn_valid(pfn) pfn_valid(pfn)
610 void sparse_init(void);
611 #else
612 #define sparse_init() do {} while (0)
613 #define sparse_index_init(_sec, _nid) do {} while (0)
614 #endif /* CONFIG_SPARSEMEM */
616 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
617 #define early_pfn_in_nid(pfn, nid) (early_pfn_to_nid(pfn) == (nid))
618 #else
619 #define early_pfn_in_nid(pfn, nid) (1)
620 #endif
622 #ifndef early_pfn_valid
623 #define early_pfn_valid(pfn) (1)
624 #endif
626 void memory_present(int nid, unsigned long start, unsigned long end);
627 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
629 #endif /* !__ASSEMBLY__ */
630 #endif /* __KERNEL__ */
631 #endif /* _LINUX_MMZONE_H */