[PATCH] zoned vm counters: conversion of nr_dirty to per zone counter
[linux-2.6/cjktty.git] / include / linux / mmzone.h
blob1cc8412ac26406d1d6b4e2c552a045b15813ac6c
1 #ifndef _LINUX_MMZONE_H
2 #define _LINUX_MMZONE_H
4 #ifdef __KERNEL__
5 #ifndef __ASSEMBLY__
7 #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>
14 #include <linux/seqlock.h>
15 #include <linux/nodemask.h>
16 #include <asm/atomic.h>
17 #include <asm/page.h>
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
25 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
27 struct free_area {
28 struct list_head free_list;
29 unsigned long nr_free;
32 struct pglist_data;
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.
40 #if defined(CONFIG_SMP)
41 struct zone_padding {
42 char x[0];
43 } ____cacheline_internodealigned_in_smp;
44 #define ZONE_PADDING(name) struct zone_padding name;
45 #else
46 #define ZONE_PADDING(name)
47 #endif
49 enum zone_stat_item {
50 NR_ANON_PAGES, /* Mapped anonymous pages */
51 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
52 only modified from process context */
53 NR_FILE_PAGES,
54 NR_SLAB, /* Pages used by slab allocator */
55 NR_PAGETABLE, /* used for pagetables */
56 NR_FILE_DIRTY,
57 NR_VM_ZONE_STAT_ITEMS };
59 struct per_cpu_pages {
60 int count; /* number of pages in the list */
61 int high; /* high watermark, emptying needed */
62 int batch; /* chunk size for buddy add/remove */
63 struct list_head list; /* the list of pages */
66 struct per_cpu_pageset {
67 struct per_cpu_pages pcp[2]; /* 0: hot. 1: cold */
68 #ifdef CONFIG_SMP
69 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
70 #endif
72 #ifdef CONFIG_NUMA
73 unsigned long numa_hit; /* allocated in intended node */
74 unsigned long numa_miss; /* allocated in non intended node */
75 unsigned long numa_foreign; /* was intended here, hit elsewhere */
76 unsigned long interleave_hit; /* interleaver prefered this zone */
77 unsigned long local_node; /* allocation from local node */
78 unsigned long other_node; /* allocation from other node */
79 #endif
80 } ____cacheline_aligned_in_smp;
82 #ifdef CONFIG_NUMA
83 #define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)])
84 #else
85 #define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)])
86 #endif
88 #define ZONE_DMA 0
89 #define ZONE_DMA32 1
90 #define ZONE_NORMAL 2
91 #define ZONE_HIGHMEM 3
93 #define MAX_NR_ZONES 4 /* Sync this with ZONES_SHIFT */
94 #define ZONES_SHIFT 2 /* ceil(log2(MAX_NR_ZONES)) */
98 * When a memory allocation must conform to specific limitations (such
99 * as being suitable for DMA) the caller will pass in hints to the
100 * allocator in the gfp_mask, in the zone modifier bits. These bits
101 * are used to select a priority ordered list of memory zones which
102 * match the requested limits. GFP_ZONEMASK defines which bits within
103 * the gfp_mask should be considered as zone modifiers. Each valid
104 * combination of the zone modifier bits has a corresponding list
105 * of zones (in node_zonelists). Thus for two zone modifiers there
106 * will be a maximum of 4 (2 ** 2) zonelists, for 3 modifiers there will
107 * be 8 (2 ** 3) zonelists. GFP_ZONETYPES defines the number of possible
108 * combinations of zone modifiers in "zone modifier space".
110 * As an optimisation any zone modifier bits which are only valid when
111 * no other zone modifier bits are set (loners) should be placed in
112 * the highest order bits of this field. This allows us to reduce the
113 * extent of the zonelists thus saving space. For example in the case
114 * of three zone modifier bits, we could require up to eight zonelists.
115 * If the left most zone modifier is a "loner" then the highest valid
116 * zonelist would be four allowing us to allocate only five zonelists.
117 * Use the first form for GFP_ZONETYPES when the left most bit is not
118 * a "loner", otherwise use the second.
120 * NOTE! Make sure this matches the zones in <linux/gfp.h>
122 #define GFP_ZONEMASK 0x07
123 /* #define GFP_ZONETYPES (GFP_ZONEMASK + 1) */ /* Non-loner */
124 #define GFP_ZONETYPES ((GFP_ZONEMASK + 1) / 2 + 1) /* Loner */
127 * On machines where it is needed (eg PCs) we divide physical memory
128 * into multiple physical zones. On a 32bit PC we have 4 zones:
130 * ZONE_DMA < 16 MB ISA DMA capable memory
131 * ZONE_DMA32 0 MB Empty
132 * ZONE_NORMAL 16-896 MB direct mapped by the kernel
133 * ZONE_HIGHMEM > 896 MB only page cache and user processes
136 struct zone {
137 /* Fields commonly accessed by the page allocator */
138 unsigned long free_pages;
139 unsigned long pages_min, pages_low, pages_high;
141 * We don't know if the memory that we're going to allocate will be freeable
142 * or/and it will be released eventually, so to avoid totally wasting several
143 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
144 * to run OOM on the lower zones despite there's tons of freeable ram
145 * on the higher zones). This array is recalculated at runtime if the
146 * sysctl_lowmem_reserve_ratio sysctl changes.
148 unsigned long lowmem_reserve[MAX_NR_ZONES];
150 #ifdef CONFIG_NUMA
151 struct per_cpu_pageset *pageset[NR_CPUS];
152 #else
153 struct per_cpu_pageset pageset[NR_CPUS];
154 #endif
156 * free areas of different sizes
158 spinlock_t lock;
159 #ifdef CONFIG_MEMORY_HOTPLUG
160 /* see spanned/present_pages for more description */
161 seqlock_t span_seqlock;
162 #endif
163 struct free_area free_area[MAX_ORDER];
166 ZONE_PADDING(_pad1_)
168 /* Fields commonly accessed by the page reclaim scanner */
169 spinlock_t lru_lock;
170 struct list_head active_list;
171 struct list_head inactive_list;
172 unsigned long nr_scan_active;
173 unsigned long nr_scan_inactive;
174 unsigned long nr_active;
175 unsigned long nr_inactive;
176 unsigned long pages_scanned; /* since last reclaim */
177 int all_unreclaimable; /* All pages pinned */
179 /* A count of how many reclaimers are scanning this zone */
180 atomic_t reclaim_in_progress;
182 /* Zone statistics */
183 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
186 * prev_priority holds the scanning priority for this zone. It is
187 * defined as the scanning priority at which we achieved our reclaim
188 * target at the previous try_to_free_pages() or balance_pgdat()
189 * invokation.
191 * We use prev_priority as a measure of how much stress page reclaim is
192 * under - it drives the swappiness decision: whether to unmap mapped
193 * pages.
195 * temp_priority is used to remember the scanning priority at which
196 * this zone was successfully refilled to free_pages == pages_high.
198 * Access to both these fields is quite racy even on uniprocessor. But
199 * it is expected to average out OK.
201 int temp_priority;
202 int prev_priority;
205 ZONE_PADDING(_pad2_)
206 /* Rarely used or read-mostly fields */
209 * wait_table -- the array holding the hash table
210 * wait_table_hash_nr_entries -- the size of the hash table array
211 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
213 * The purpose of all these is to keep track of the people
214 * waiting for a page to become available and make them
215 * runnable again when possible. The trouble is that this
216 * consumes a lot of space, especially when so few things
217 * wait on pages at a given time. So instead of using
218 * per-page waitqueues, we use a waitqueue hash table.
220 * The bucket discipline is to sleep on the same queue when
221 * colliding and wake all in that wait queue when removing.
222 * When something wakes, it must check to be sure its page is
223 * truly available, a la thundering herd. The cost of a
224 * collision is great, but given the expected load of the
225 * table, they should be so rare as to be outweighed by the
226 * benefits from the saved space.
228 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
229 * primary users of these fields, and in mm/page_alloc.c
230 * free_area_init_core() performs the initialization of them.
232 wait_queue_head_t * wait_table;
233 unsigned long wait_table_hash_nr_entries;
234 unsigned long wait_table_bits;
237 * Discontig memory support fields.
239 struct pglist_data *zone_pgdat;
240 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
241 unsigned long zone_start_pfn;
244 * zone_start_pfn, spanned_pages and present_pages are all
245 * protected by span_seqlock. It is a seqlock because it has
246 * to be read outside of zone->lock, and it is done in the main
247 * allocator path. But, it is written quite infrequently.
249 * The lock is declared along with zone->lock because it is
250 * frequently read in proximity to zone->lock. It's good to
251 * give them a chance of being in the same cacheline.
253 unsigned long spanned_pages; /* total size, including holes */
254 unsigned long present_pages; /* amount of memory (excluding holes) */
257 * rarely used fields:
259 char *name;
260 } ____cacheline_internodealigned_in_smp;
264 * The "priority" of VM scanning is how much of the queues we will scan in one
265 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
266 * queues ("queue_length >> 12") during an aging round.
268 #define DEF_PRIORITY 12
271 * One allocation request operates on a zonelist. A zonelist
272 * is a list of zones, the first one is the 'goal' of the
273 * allocation, the other zones are fallback zones, in decreasing
274 * priority.
276 * Right now a zonelist takes up less than a cacheline. We never
277 * modify it apart from boot-up, and only a few indices are used,
278 * so despite the zonelist table being relatively big, the cache
279 * footprint of this construct is very small.
281 struct zonelist {
282 struct zone *zones[MAX_NUMNODES * MAX_NR_ZONES + 1]; // NULL delimited
287 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
288 * (mostly NUMA machines?) to denote a higher-level memory zone than the
289 * zone denotes.
291 * On NUMA machines, each NUMA node would have a pg_data_t to describe
292 * it's memory layout.
294 * Memory statistics and page replacement data structures are maintained on a
295 * per-zone basis.
297 struct bootmem_data;
298 typedef struct pglist_data {
299 struct zone node_zones[MAX_NR_ZONES];
300 struct zonelist node_zonelists[GFP_ZONETYPES];
301 int nr_zones;
302 #ifdef CONFIG_FLAT_NODE_MEM_MAP
303 struct page *node_mem_map;
304 #endif
305 struct bootmem_data *bdata;
306 #ifdef CONFIG_MEMORY_HOTPLUG
308 * Must be held any time you expect node_start_pfn, node_present_pages
309 * or node_spanned_pages stay constant. Holding this will also
310 * guarantee that any pfn_valid() stays that way.
312 * Nests above zone->lock and zone->size_seqlock.
314 spinlock_t node_size_lock;
315 #endif
316 unsigned long node_start_pfn;
317 unsigned long node_present_pages; /* total number of physical pages */
318 unsigned long node_spanned_pages; /* total size of physical page
319 range, including holes */
320 int node_id;
321 wait_queue_head_t kswapd_wait;
322 struct task_struct *kswapd;
323 int kswapd_max_order;
324 } pg_data_t;
326 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
327 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
328 #ifdef CONFIG_FLAT_NODE_MEM_MAP
329 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
330 #else
331 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
332 #endif
333 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
335 #include <linux/memory_hotplug.h>
337 void __get_zone_counts(unsigned long *active, unsigned long *inactive,
338 unsigned long *free, struct pglist_data *pgdat);
339 void get_zone_counts(unsigned long *active, unsigned long *inactive,
340 unsigned long *free);
341 void build_all_zonelists(void);
342 void wakeup_kswapd(struct zone *zone, int order);
343 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
344 int classzone_idx, int alloc_flags);
346 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
347 unsigned long size);
349 #ifdef CONFIG_HAVE_MEMORY_PRESENT
350 void memory_present(int nid, unsigned long start, unsigned long end);
351 #else
352 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
353 #endif
355 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
356 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
357 #endif
360 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
362 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
364 static inline int populated_zone(struct zone *zone)
366 return (!!zone->present_pages);
369 static inline int is_highmem_idx(int idx)
371 return (idx == ZONE_HIGHMEM);
374 static inline int is_normal_idx(int idx)
376 return (idx == ZONE_NORMAL);
380 * is_highmem - helper function to quickly check if a struct zone is a
381 * highmem zone or not. This is an attempt to keep references
382 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
383 * @zone - pointer to struct zone variable
385 static inline int is_highmem(struct zone *zone)
387 return zone == zone->zone_pgdat->node_zones + ZONE_HIGHMEM;
390 static inline int is_normal(struct zone *zone)
392 return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
395 static inline int is_dma32(struct zone *zone)
397 return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
400 static inline int is_dma(struct zone *zone)
402 return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
405 /* These two functions are used to setup the per zone pages min values */
406 struct ctl_table;
407 struct file;
408 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *,
409 void __user *, size_t *, loff_t *);
410 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
411 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *,
412 void __user *, size_t *, loff_t *);
413 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, struct file *,
414 void __user *, size_t *, loff_t *);
416 #include <linux/topology.h>
417 /* Returns the number of the current Node. */
418 #ifndef numa_node_id
419 #define numa_node_id() (cpu_to_node(raw_smp_processor_id()))
420 #endif
422 #ifndef CONFIG_NEED_MULTIPLE_NODES
424 extern struct pglist_data contig_page_data;
425 #define NODE_DATA(nid) (&contig_page_data)
426 #define NODE_MEM_MAP(nid) mem_map
427 #define MAX_NODES_SHIFT 1
429 #else /* CONFIG_NEED_MULTIPLE_NODES */
431 #include <asm/mmzone.h>
433 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
435 extern struct pglist_data *first_online_pgdat(void);
436 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
437 extern struct zone *next_zone(struct zone *zone);
440 * for_each_pgdat - helper macro to iterate over all nodes
441 * @pgdat - pointer to a pg_data_t variable
443 #define for_each_online_pgdat(pgdat) \
444 for (pgdat = first_online_pgdat(); \
445 pgdat; \
446 pgdat = next_online_pgdat(pgdat))
448 * for_each_zone - helper macro to iterate over all memory zones
449 * @zone - pointer to struct zone variable
451 * The user only needs to declare the zone variable, for_each_zone
452 * fills it in.
454 #define for_each_zone(zone) \
455 for (zone = (first_online_pgdat())->node_zones; \
456 zone; \
457 zone = next_zone(zone))
459 #ifdef CONFIG_SPARSEMEM
460 #include <asm/sparsemem.h>
461 #endif
463 #if BITS_PER_LONG == 32
465 * with 32 bit page->flags field, we reserve 9 bits for node/zone info.
466 * there are 4 zones (3 bits) and this leaves 9-3=6 bits for nodes.
468 #define FLAGS_RESERVED 9
470 #elif BITS_PER_LONG == 64
472 * with 64 bit flags field, there's plenty of room.
474 #define FLAGS_RESERVED 32
476 #else
478 #error BITS_PER_LONG not defined
480 #endif
482 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
483 #define early_pfn_to_nid(nid) (0UL)
484 #endif
486 #ifdef CONFIG_FLATMEM
487 #define pfn_to_nid(pfn) (0)
488 #endif
490 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
491 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
493 #ifdef CONFIG_SPARSEMEM
496 * SECTION_SHIFT #bits space required to store a section #
498 * PA_SECTION_SHIFT physical address to/from section number
499 * PFN_SECTION_SHIFT pfn to/from section number
501 #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
503 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
504 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
506 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
508 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
509 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
511 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
512 #error Allocator MAX_ORDER exceeds SECTION_SIZE
513 #endif
515 struct page;
516 struct mem_section {
518 * This is, logically, a pointer to an array of struct
519 * pages. However, it is stored with some other magic.
520 * (see sparse.c::sparse_init_one_section())
522 * Additionally during early boot we encode node id of
523 * the location of the section here to guide allocation.
524 * (see sparse.c::memory_present())
526 * Making it a UL at least makes someone do a cast
527 * before using it wrong.
529 unsigned long section_mem_map;
532 #ifdef CONFIG_SPARSEMEM_EXTREME
533 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
534 #else
535 #define SECTIONS_PER_ROOT 1
536 #endif
538 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
539 #define NR_SECTION_ROOTS (NR_MEM_SECTIONS / SECTIONS_PER_ROOT)
540 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
542 #ifdef CONFIG_SPARSEMEM_EXTREME
543 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
544 #else
545 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
546 #endif
548 static inline struct mem_section *__nr_to_section(unsigned long nr)
550 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
551 return NULL;
552 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
554 extern int __section_nr(struct mem_section* ms);
557 * We use the lower bits of the mem_map pointer to store
558 * a little bit of information. There should be at least
559 * 3 bits here due to 32-bit alignment.
561 #define SECTION_MARKED_PRESENT (1UL<<0)
562 #define SECTION_HAS_MEM_MAP (1UL<<1)
563 #define SECTION_MAP_LAST_BIT (1UL<<2)
564 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
565 #define SECTION_NID_SHIFT 2
567 static inline struct page *__section_mem_map_addr(struct mem_section *section)
569 unsigned long map = section->section_mem_map;
570 map &= SECTION_MAP_MASK;
571 return (struct page *)map;
574 static inline int valid_section(struct mem_section *section)
576 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
579 static inline int section_has_mem_map(struct mem_section *section)
581 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
584 static inline int valid_section_nr(unsigned long nr)
586 return valid_section(__nr_to_section(nr));
589 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
591 return __nr_to_section(pfn_to_section_nr(pfn));
594 static inline int pfn_valid(unsigned long pfn)
596 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
597 return 0;
598 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
602 * These are _only_ used during initialisation, therefore they
603 * can use __initdata ... They could have names to indicate
604 * this restriction.
606 #ifdef CONFIG_NUMA
607 #define pfn_to_nid(pfn) \
608 ({ \
609 unsigned long __pfn_to_nid_pfn = (pfn); \
610 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
612 #else
613 #define pfn_to_nid(pfn) (0)
614 #endif
616 #define early_pfn_valid(pfn) pfn_valid(pfn)
617 void sparse_init(void);
618 #else
619 #define sparse_init() do {} while (0)
620 #define sparse_index_init(_sec, _nid) do {} while (0)
621 #endif /* CONFIG_SPARSEMEM */
623 #ifndef early_pfn_valid
624 #define early_pfn_valid(pfn) (1)
625 #endif
627 void memory_present(int nid, unsigned long start, unsigned long end);
628 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
630 #endif /* !__ASSEMBLY__ */
631 #endif /* __KERNEL__ */
632 #endif /* _LINUX_MMZONE_H */