[PATCH] zoned vm counters: basic ZVC (zoned vm counter) implementation
[linux-2.6/openmoko-kernel/knife-kernel.git] / include / linux / mmzone.h
blob543f9e411563acd8bfc87b3747ba6c97e68a63fb
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_VM_ZONE_STAT_ITEMS };
52 struct per_cpu_pages {
53 int count; /* number of pages in the list */
54 int high; /* high watermark, emptying needed */
55 int batch; /* chunk size for buddy add/remove */
56 struct list_head list; /* the list of pages */
59 struct per_cpu_pageset {
60 struct per_cpu_pages pcp[2]; /* 0: hot. 1: cold */
61 #ifdef CONFIG_SMP
62 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
63 #endif
65 #ifdef CONFIG_NUMA
66 unsigned long numa_hit; /* allocated in intended node */
67 unsigned long numa_miss; /* allocated in non intended node */
68 unsigned long numa_foreign; /* was intended here, hit elsewhere */
69 unsigned long interleave_hit; /* interleaver prefered this zone */
70 unsigned long local_node; /* allocation from local node */
71 unsigned long other_node; /* allocation from other node */
72 #endif
73 } ____cacheline_aligned_in_smp;
75 #ifdef CONFIG_NUMA
76 #define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)])
77 #else
78 #define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)])
79 #endif
81 #define ZONE_DMA 0
82 #define ZONE_DMA32 1
83 #define ZONE_NORMAL 2
84 #define ZONE_HIGHMEM 3
86 #define MAX_NR_ZONES 4 /* Sync this with ZONES_SHIFT */
87 #define ZONES_SHIFT 2 /* ceil(log2(MAX_NR_ZONES)) */
91 * When a memory allocation must conform to specific limitations (such
92 * as being suitable for DMA) the caller will pass in hints to the
93 * allocator in the gfp_mask, in the zone modifier bits. These bits
94 * are used to select a priority ordered list of memory zones which
95 * match the requested limits. GFP_ZONEMASK defines which bits within
96 * the gfp_mask should be considered as zone modifiers. Each valid
97 * combination of the zone modifier bits has a corresponding list
98 * of zones (in node_zonelists). Thus for two zone modifiers there
99 * will be a maximum of 4 (2 ** 2) zonelists, for 3 modifiers there will
100 * be 8 (2 ** 3) zonelists. GFP_ZONETYPES defines the number of possible
101 * combinations of zone modifiers in "zone modifier space".
103 * As an optimisation any zone modifier bits which are only valid when
104 * no other zone modifier bits are set (loners) should be placed in
105 * the highest order bits of this field. This allows us to reduce the
106 * extent of the zonelists thus saving space. For example in the case
107 * of three zone modifier bits, we could require up to eight zonelists.
108 * If the left most zone modifier is a "loner" then the highest valid
109 * zonelist would be four allowing us to allocate only five zonelists.
110 * Use the first form for GFP_ZONETYPES when the left most bit is not
111 * a "loner", otherwise use the second.
113 * NOTE! Make sure this matches the zones in <linux/gfp.h>
115 #define GFP_ZONEMASK 0x07
116 /* #define GFP_ZONETYPES (GFP_ZONEMASK + 1) */ /* Non-loner */
117 #define GFP_ZONETYPES ((GFP_ZONEMASK + 1) / 2 + 1) /* Loner */
120 * On machines where it is needed (eg PCs) we divide physical memory
121 * into multiple physical zones. On a 32bit PC we have 4 zones:
123 * ZONE_DMA < 16 MB ISA DMA capable memory
124 * ZONE_DMA32 0 MB Empty
125 * ZONE_NORMAL 16-896 MB direct mapped by the kernel
126 * ZONE_HIGHMEM > 896 MB only page cache and user processes
129 struct zone {
130 /* Fields commonly accessed by the page allocator */
131 unsigned long free_pages;
132 unsigned long pages_min, pages_low, pages_high;
134 * We don't know if the memory that we're going to allocate will be freeable
135 * or/and it will be released eventually, so to avoid totally wasting several
136 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
137 * to run OOM on the lower zones despite there's tons of freeable ram
138 * on the higher zones). This array is recalculated at runtime if the
139 * sysctl_lowmem_reserve_ratio sysctl changes.
141 unsigned long lowmem_reserve[MAX_NR_ZONES];
143 #ifdef CONFIG_NUMA
144 struct per_cpu_pageset *pageset[NR_CPUS];
145 #else
146 struct per_cpu_pageset pageset[NR_CPUS];
147 #endif
149 * free areas of different sizes
151 spinlock_t lock;
152 #ifdef CONFIG_MEMORY_HOTPLUG
153 /* see spanned/present_pages for more description */
154 seqlock_t span_seqlock;
155 #endif
156 struct free_area free_area[MAX_ORDER];
159 ZONE_PADDING(_pad1_)
161 /* Fields commonly accessed by the page reclaim scanner */
162 spinlock_t lru_lock;
163 struct list_head active_list;
164 struct list_head inactive_list;
165 unsigned long nr_scan_active;
166 unsigned long nr_scan_inactive;
167 unsigned long nr_active;
168 unsigned long nr_inactive;
169 unsigned long pages_scanned; /* since last reclaim */
170 int all_unreclaimable; /* All pages pinned */
172 /* A count of how many reclaimers are scanning this zone */
173 atomic_t reclaim_in_progress;
175 /* Zone statistics */
176 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
178 * timestamp (in jiffies) of the last zone reclaim that did not
179 * result in freeing of pages. This is used to avoid repeated scans
180 * if all memory in the zone is in use.
182 unsigned long last_unsuccessful_zone_reclaim;
185 * prev_priority holds the scanning priority for this zone. It is
186 * defined as the scanning priority at which we achieved our reclaim
187 * target at the previous try_to_free_pages() or balance_pgdat()
188 * invokation.
190 * We use prev_priority as a measure of how much stress page reclaim is
191 * under - it drives the swappiness decision: whether to unmap mapped
192 * pages.
194 * temp_priority is used to remember the scanning priority at which
195 * this zone was successfully refilled to free_pages == pages_high.
197 * Access to both these fields is quite racy even on uniprocessor. But
198 * it is expected to average out OK.
200 int temp_priority;
201 int prev_priority;
204 ZONE_PADDING(_pad2_)
205 /* Rarely used or read-mostly fields */
208 * wait_table -- the array holding the hash table
209 * wait_table_hash_nr_entries -- the size of the hash table array
210 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
212 * The purpose of all these is to keep track of the people
213 * waiting for a page to become available and make them
214 * runnable again when possible. The trouble is that this
215 * consumes a lot of space, especially when so few things
216 * wait on pages at a given time. So instead of using
217 * per-page waitqueues, we use a waitqueue hash table.
219 * The bucket discipline is to sleep on the same queue when
220 * colliding and wake all in that wait queue when removing.
221 * When something wakes, it must check to be sure its page is
222 * truly available, a la thundering herd. The cost of a
223 * collision is great, but given the expected load of the
224 * table, they should be so rare as to be outweighed by the
225 * benefits from the saved space.
227 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
228 * primary users of these fields, and in mm/page_alloc.c
229 * free_area_init_core() performs the initialization of them.
231 wait_queue_head_t * wait_table;
232 unsigned long wait_table_hash_nr_entries;
233 unsigned long wait_table_bits;
236 * Discontig memory support fields.
238 struct pglist_data *zone_pgdat;
239 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
240 unsigned long zone_start_pfn;
243 * zone_start_pfn, spanned_pages and present_pages are all
244 * protected by span_seqlock. It is a seqlock because it has
245 * to be read outside of zone->lock, and it is done in the main
246 * allocator path. But, it is written quite infrequently.
248 * The lock is declared along with zone->lock because it is
249 * frequently read in proximity to zone->lock. It's good to
250 * give them a chance of being in the same cacheline.
252 unsigned long spanned_pages; /* total size, including holes */
253 unsigned long present_pages; /* amount of memory (excluding holes) */
256 * rarely used fields:
258 char *name;
259 } ____cacheline_internodealigned_in_smp;
263 * The "priority" of VM scanning is how much of the queues we will scan in one
264 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
265 * queues ("queue_length >> 12") during an aging round.
267 #define DEF_PRIORITY 12
270 * One allocation request operates on a zonelist. A zonelist
271 * is a list of zones, the first one is the 'goal' of the
272 * allocation, the other zones are fallback zones, in decreasing
273 * priority.
275 * Right now a zonelist takes up less than a cacheline. We never
276 * modify it apart from boot-up, and only a few indices are used,
277 * so despite the zonelist table being relatively big, the cache
278 * footprint of this construct is very small.
280 struct zonelist {
281 struct zone *zones[MAX_NUMNODES * MAX_NR_ZONES + 1]; // NULL delimited
286 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
287 * (mostly NUMA machines?) to denote a higher-level memory zone than the
288 * zone denotes.
290 * On NUMA machines, each NUMA node would have a pg_data_t to describe
291 * it's memory layout.
293 * Memory statistics and page replacement data structures are maintained on a
294 * per-zone basis.
296 struct bootmem_data;
297 typedef struct pglist_data {
298 struct zone node_zones[MAX_NR_ZONES];
299 struct zonelist node_zonelists[GFP_ZONETYPES];
300 int nr_zones;
301 #ifdef CONFIG_FLAT_NODE_MEM_MAP
302 struct page *node_mem_map;
303 #endif
304 struct bootmem_data *bdata;
305 #ifdef CONFIG_MEMORY_HOTPLUG
307 * Must be held any time you expect node_start_pfn, node_present_pages
308 * or node_spanned_pages stay constant. Holding this will also
309 * guarantee that any pfn_valid() stays that way.
311 * Nests above zone->lock and zone->size_seqlock.
313 spinlock_t node_size_lock;
314 #endif
315 unsigned long node_start_pfn;
316 unsigned long node_present_pages; /* total number of physical pages */
317 unsigned long node_spanned_pages; /* total size of physical page
318 range, including holes */
319 int node_id;
320 wait_queue_head_t kswapd_wait;
321 struct task_struct *kswapd;
322 int kswapd_max_order;
323 } pg_data_t;
325 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
326 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
327 #ifdef CONFIG_FLAT_NODE_MEM_MAP
328 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
329 #else
330 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
331 #endif
332 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
334 #include <linux/memory_hotplug.h>
336 void __get_zone_counts(unsigned long *active, unsigned long *inactive,
337 unsigned long *free, struct pglist_data *pgdat);
338 void get_zone_counts(unsigned long *active, unsigned long *inactive,
339 unsigned long *free);
340 void build_all_zonelists(void);
341 void wakeup_kswapd(struct zone *zone, int order);
342 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
343 int classzone_idx, int alloc_flags);
345 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
346 unsigned long size);
348 #ifdef CONFIG_HAVE_MEMORY_PRESENT
349 void memory_present(int nid, unsigned long start, unsigned long end);
350 #else
351 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
352 #endif
354 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
355 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
356 #endif
359 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
361 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
363 static inline int populated_zone(struct zone *zone)
365 return (!!zone->present_pages);
368 static inline int is_highmem_idx(int idx)
370 return (idx == ZONE_HIGHMEM);
373 static inline int is_normal_idx(int idx)
375 return (idx == ZONE_NORMAL);
379 * is_highmem - helper function to quickly check if a struct zone is a
380 * highmem zone or not. This is an attempt to keep references
381 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
382 * @zone - pointer to struct zone variable
384 static inline int is_highmem(struct zone *zone)
386 return zone == zone->zone_pgdat->node_zones + ZONE_HIGHMEM;
389 static inline int is_normal(struct zone *zone)
391 return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
394 static inline int is_dma32(struct zone *zone)
396 return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
399 static inline int is_dma(struct zone *zone)
401 return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
404 /* These two functions are used to setup the per zone pages min values */
405 struct ctl_table;
406 struct file;
407 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *,
408 void __user *, size_t *, loff_t *);
409 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
410 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *,
411 void __user *, size_t *, loff_t *);
412 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, struct file *,
413 void __user *, size_t *, loff_t *);
415 #include <linux/topology.h>
416 /* Returns the number of the current Node. */
417 #ifndef numa_node_id
418 #define numa_node_id() (cpu_to_node(raw_smp_processor_id()))
419 #endif
421 #ifndef CONFIG_NEED_MULTIPLE_NODES
423 extern struct pglist_data contig_page_data;
424 #define NODE_DATA(nid) (&contig_page_data)
425 #define NODE_MEM_MAP(nid) mem_map
426 #define MAX_NODES_SHIFT 1
428 #else /* CONFIG_NEED_MULTIPLE_NODES */
430 #include <asm/mmzone.h>
432 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
434 extern struct pglist_data *first_online_pgdat(void);
435 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
436 extern struct zone *next_zone(struct zone *zone);
439 * for_each_pgdat - helper macro to iterate over all nodes
440 * @pgdat - pointer to a pg_data_t variable
442 #define for_each_online_pgdat(pgdat) \
443 for (pgdat = first_online_pgdat(); \
444 pgdat; \
445 pgdat = next_online_pgdat(pgdat))
447 * for_each_zone - helper macro to iterate over all memory zones
448 * @zone - pointer to struct zone variable
450 * The user only needs to declare the zone variable, for_each_zone
451 * fills it in.
453 #define for_each_zone(zone) \
454 for (zone = (first_online_pgdat())->node_zones; \
455 zone; \
456 zone = next_zone(zone))
458 #ifdef CONFIG_SPARSEMEM
459 #include <asm/sparsemem.h>
460 #endif
462 #if BITS_PER_LONG == 32
464 * with 32 bit page->flags field, we reserve 9 bits for node/zone info.
465 * there are 4 zones (3 bits) and this leaves 9-3=6 bits for nodes.
467 #define FLAGS_RESERVED 9
469 #elif BITS_PER_LONG == 64
471 * with 64 bit flags field, there's plenty of room.
473 #define FLAGS_RESERVED 32
475 #else
477 #error BITS_PER_LONG not defined
479 #endif
481 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
482 #define early_pfn_to_nid(nid) (0UL)
483 #endif
485 #ifdef CONFIG_FLATMEM
486 #define pfn_to_nid(pfn) (0)
487 #endif
489 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
490 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
492 #ifdef CONFIG_SPARSEMEM
495 * SECTION_SHIFT #bits space required to store a section #
497 * PA_SECTION_SHIFT physical address to/from section number
498 * PFN_SECTION_SHIFT pfn to/from section number
500 #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
502 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
503 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
505 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
507 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
508 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
510 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
511 #error Allocator MAX_ORDER exceeds SECTION_SIZE
512 #endif
514 struct page;
515 struct mem_section {
517 * This is, logically, a pointer to an array of struct
518 * pages. However, it is stored with some other magic.
519 * (see sparse.c::sparse_init_one_section())
521 * Additionally during early boot we encode node id of
522 * the location of the section here to guide allocation.
523 * (see sparse.c::memory_present())
525 * Making it a UL at least makes someone do a cast
526 * before using it wrong.
528 unsigned long section_mem_map;
531 #ifdef CONFIG_SPARSEMEM_EXTREME
532 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
533 #else
534 #define SECTIONS_PER_ROOT 1
535 #endif
537 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
538 #define NR_SECTION_ROOTS (NR_MEM_SECTIONS / SECTIONS_PER_ROOT)
539 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
541 #ifdef CONFIG_SPARSEMEM_EXTREME
542 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
543 #else
544 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
545 #endif
547 static inline struct mem_section *__nr_to_section(unsigned long nr)
549 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
550 return NULL;
551 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
553 extern int __section_nr(struct mem_section* ms);
556 * We use the lower bits of the mem_map pointer to store
557 * a little bit of information. There should be at least
558 * 3 bits here due to 32-bit alignment.
560 #define SECTION_MARKED_PRESENT (1UL<<0)
561 #define SECTION_HAS_MEM_MAP (1UL<<1)
562 #define SECTION_MAP_LAST_BIT (1UL<<2)
563 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
564 #define SECTION_NID_SHIFT 2
566 static inline struct page *__section_mem_map_addr(struct mem_section *section)
568 unsigned long map = section->section_mem_map;
569 map &= SECTION_MAP_MASK;
570 return (struct page *)map;
573 static inline int valid_section(struct mem_section *section)
575 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
578 static inline int section_has_mem_map(struct mem_section *section)
580 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
583 static inline int valid_section_nr(unsigned long nr)
585 return valid_section(__nr_to_section(nr));
588 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
590 return __nr_to_section(pfn_to_section_nr(pfn));
593 static inline int pfn_valid(unsigned long pfn)
595 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
596 return 0;
597 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
601 * These are _only_ used during initialisation, therefore they
602 * can use __initdata ... They could have names to indicate
603 * this restriction.
605 #ifdef CONFIG_NUMA
606 #define pfn_to_nid(pfn) \
607 ({ \
608 unsigned long __pfn_to_nid_pfn = (pfn); \
609 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
611 #else
612 #define pfn_to_nid(pfn) (0)
613 #endif
615 #define early_pfn_valid(pfn) pfn_valid(pfn)
616 void sparse_init(void);
617 #else
618 #define sparse_init() do {} while (0)
619 #define sparse_index_init(_sec, _nid) do {} while (0)
620 #endif /* CONFIG_SPARSEMEM */
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 */