1 #ifndef _LINUX_MMZONE_H
2 #define _LINUX_MMZONE_H
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>
19 /* Free memory management - zoned buddy allocator. */
20 #ifndef CONFIG_FORCE_MAX_ZONEORDER
23 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
25 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
28 struct list_head free_list;
29 unsigned long nr_free;
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)
43 } ____cacheline_internodealigned_in_smp;
44 #define ZONE_PADDING(name) struct zone_padding name;
46 #define ZONE_PADDING(name)
50 NR_ANON_PAGES, /* Mapped anonymous pages */
51 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
52 only modified from process context */
54 NR_SLAB, /* Pages used by slab allocator */
55 NR_PAGETABLE, /* used for pagetables */
58 NR_UNSTABLE_NFS, /* NFS unstable pages */
61 NUMA_HIT, /* allocated in intended node */
62 NUMA_MISS, /* allocated in non intended node */
63 NUMA_FOREIGN, /* was intended here, hit elsewhere */
64 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
65 NUMA_LOCAL, /* allocation from local node */
66 NUMA_OTHER, /* allocation from other node */
68 NR_VM_ZONE_STAT_ITEMS };
70 struct per_cpu_pages {
71 int count; /* number of pages in the list */
72 int high; /* high watermark, emptying needed */
73 int batch; /* chunk size for buddy add/remove */
74 struct list_head list; /* the list of pages */
77 struct per_cpu_pageset {
78 struct per_cpu_pages pcp[2]; /* 0: hot. 1: cold */
80 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
82 } ____cacheline_aligned_in_smp;
85 #define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)])
87 #define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)])
93 #define ZONE_HIGHMEM 3
95 #define MAX_NR_ZONES 4 /* Sync this with ZONES_SHIFT */
96 #define ZONES_SHIFT 2 /* ceil(log2(MAX_NR_ZONES)) */
100 * When a memory allocation must conform to specific limitations (such
101 * as being suitable for DMA) the caller will pass in hints to the
102 * allocator in the gfp_mask, in the zone modifier bits. These bits
103 * are used to select a priority ordered list of memory zones which
104 * match the requested limits. GFP_ZONEMASK defines which bits within
105 * the gfp_mask should be considered as zone modifiers. Each valid
106 * combination of the zone modifier bits has a corresponding list
107 * of zones (in node_zonelists). Thus for two zone modifiers there
108 * will be a maximum of 4 (2 ** 2) zonelists, for 3 modifiers there will
109 * be 8 (2 ** 3) zonelists. GFP_ZONETYPES defines the number of possible
110 * combinations of zone modifiers in "zone modifier space".
112 * As an optimisation any zone modifier bits which are only valid when
113 * no other zone modifier bits are set (loners) should be placed in
114 * the highest order bits of this field. This allows us to reduce the
115 * extent of the zonelists thus saving space. For example in the case
116 * of three zone modifier bits, we could require up to eight zonelists.
117 * If the left most zone modifier is a "loner" then the highest valid
118 * zonelist would be four allowing us to allocate only five zonelists.
119 * Use the first form for GFP_ZONETYPES when the left most bit is not
120 * a "loner", otherwise use the second.
122 * NOTE! Make sure this matches the zones in <linux/gfp.h>
124 #define GFP_ZONEMASK 0x07
125 /* #define GFP_ZONETYPES (GFP_ZONEMASK + 1) */ /* Non-loner */
126 #define GFP_ZONETYPES ((GFP_ZONEMASK + 1) / 2 + 1) /* Loner */
129 * On machines where it is needed (eg PCs) we divide physical memory
130 * into multiple physical zones. On a 32bit PC we have 4 zones:
132 * ZONE_DMA < 16 MB ISA DMA capable memory
133 * ZONE_DMA32 0 MB Empty
134 * ZONE_NORMAL 16-896 MB direct mapped by the kernel
135 * ZONE_HIGHMEM > 896 MB only page cache and user processes
139 /* Fields commonly accessed by the page allocator */
140 unsigned long free_pages;
141 unsigned long pages_min, pages_low, pages_high;
143 * We don't know if the memory that we're going to allocate will be freeable
144 * or/and it will be released eventually, so to avoid totally wasting several
145 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
146 * to run OOM on the lower zones despite there's tons of freeable ram
147 * on the higher zones). This array is recalculated at runtime if the
148 * sysctl_lowmem_reserve_ratio sysctl changes.
150 unsigned long lowmem_reserve[MAX_NR_ZONES];
153 struct per_cpu_pageset *pageset[NR_CPUS];
155 struct per_cpu_pageset pageset[NR_CPUS];
158 * free areas of different sizes
161 #ifdef CONFIG_MEMORY_HOTPLUG
162 /* see spanned/present_pages for more description */
163 seqlock_t span_seqlock;
165 struct free_area free_area[MAX_ORDER];
170 /* Fields commonly accessed by the page reclaim scanner */
172 struct list_head active_list;
173 struct list_head inactive_list;
174 unsigned long nr_scan_active;
175 unsigned long nr_scan_inactive;
176 unsigned long nr_active;
177 unsigned long nr_inactive;
178 unsigned long pages_scanned; /* since last reclaim */
179 int all_unreclaimable; /* All pages pinned */
181 /* A count of how many reclaimers are scanning this zone */
182 atomic_t reclaim_in_progress;
184 /* Zone statistics */
185 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
188 * prev_priority holds the scanning priority for this zone. It is
189 * defined as the scanning priority at which we achieved our reclaim
190 * target at the previous try_to_free_pages() or balance_pgdat()
193 * We use prev_priority as a measure of how much stress page reclaim is
194 * under - it drives the swappiness decision: whether to unmap mapped
197 * temp_priority is used to remember the scanning priority at which
198 * this zone was successfully refilled to free_pages == pages_high.
200 * Access to both these fields is quite racy even on uniprocessor. But
201 * it is expected to average out OK.
208 /* Rarely used or read-mostly fields */
211 * wait_table -- the array holding the hash table
212 * wait_table_hash_nr_entries -- the size of the hash table array
213 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
215 * The purpose of all these is to keep track of the people
216 * waiting for a page to become available and make them
217 * runnable again when possible. The trouble is that this
218 * consumes a lot of space, especially when so few things
219 * wait on pages at a given time. So instead of using
220 * per-page waitqueues, we use a waitqueue hash table.
222 * The bucket discipline is to sleep on the same queue when
223 * colliding and wake all in that wait queue when removing.
224 * When something wakes, it must check to be sure its page is
225 * truly available, a la thundering herd. The cost of a
226 * collision is great, but given the expected load of the
227 * table, they should be so rare as to be outweighed by the
228 * benefits from the saved space.
230 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
231 * primary users of these fields, and in mm/page_alloc.c
232 * free_area_init_core() performs the initialization of them.
234 wait_queue_head_t * wait_table;
235 unsigned long wait_table_hash_nr_entries;
236 unsigned long wait_table_bits;
239 * Discontig memory support fields.
241 struct pglist_data *zone_pgdat;
242 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
243 unsigned long zone_start_pfn;
246 * zone_start_pfn, spanned_pages and present_pages are all
247 * protected by span_seqlock. It is a seqlock because it has
248 * to be read outside of zone->lock, and it is done in the main
249 * allocator path. But, it is written quite infrequently.
251 * The lock is declared along with zone->lock because it is
252 * frequently read in proximity to zone->lock. It's good to
253 * give them a chance of being in the same cacheline.
255 unsigned long spanned_pages; /* total size, including holes */
256 unsigned long present_pages; /* amount of memory (excluding holes) */
259 * rarely used fields:
262 } ____cacheline_internodealigned_in_smp;
266 * The "priority" of VM scanning is how much of the queues we will scan in one
267 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
268 * queues ("queue_length >> 12") during an aging round.
270 #define DEF_PRIORITY 12
273 * One allocation request operates on a zonelist. A zonelist
274 * is a list of zones, the first one is the 'goal' of the
275 * allocation, the other zones are fallback zones, in decreasing
278 * Right now a zonelist takes up less than a cacheline. We never
279 * modify it apart from boot-up, and only a few indices are used,
280 * so despite the zonelist table being relatively big, the cache
281 * footprint of this construct is very small.
284 struct zone *zones[MAX_NUMNODES * MAX_NR_ZONES + 1]; // NULL delimited
289 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
290 * (mostly NUMA machines?) to denote a higher-level memory zone than the
293 * On NUMA machines, each NUMA node would have a pg_data_t to describe
294 * it's memory layout.
296 * Memory statistics and page replacement data structures are maintained on a
300 typedef struct pglist_data {
301 struct zone node_zones[MAX_NR_ZONES];
302 struct zonelist node_zonelists[GFP_ZONETYPES];
304 #ifdef CONFIG_FLAT_NODE_MEM_MAP
305 struct page *node_mem_map;
307 struct bootmem_data *bdata;
308 #ifdef CONFIG_MEMORY_HOTPLUG
310 * Must be held any time you expect node_start_pfn, node_present_pages
311 * or node_spanned_pages stay constant. Holding this will also
312 * guarantee that any pfn_valid() stays that way.
314 * Nests above zone->lock and zone->size_seqlock.
316 spinlock_t node_size_lock;
318 unsigned long node_start_pfn;
319 unsigned long node_present_pages; /* total number of physical pages */
320 unsigned long node_spanned_pages; /* total size of physical page
321 range, including holes */
323 wait_queue_head_t kswapd_wait;
324 struct task_struct *kswapd;
325 int kswapd_max_order;
328 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
329 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
330 #ifdef CONFIG_FLAT_NODE_MEM_MAP
331 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
333 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
335 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
337 #include <linux/memory_hotplug.h>
339 void __get_zone_counts(unsigned long *active, unsigned long *inactive,
340 unsigned long *free, struct pglist_data *pgdat);
341 void get_zone_counts(unsigned long *active, unsigned long *inactive,
342 unsigned long *free);
343 void build_all_zonelists(void);
344 void wakeup_kswapd(struct zone *zone, int order);
345 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
346 int classzone_idx, int alloc_flags);
348 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
351 #ifdef CONFIG_HAVE_MEMORY_PRESENT
352 void memory_present(int nid, unsigned long start, unsigned long end);
354 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
357 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
358 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
362 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
364 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
366 static inline int populated_zone(struct zone *zone)
368 return (!!zone->present_pages);
371 static inline int is_highmem_idx(int idx)
373 return (idx == ZONE_HIGHMEM);
376 static inline int is_normal_idx(int idx)
378 return (idx == ZONE_NORMAL);
382 * is_highmem - helper function to quickly check if a struct zone is a
383 * highmem zone or not. This is an attempt to keep references
384 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
385 * @zone - pointer to struct zone variable
387 static inline int is_highmem(struct zone *zone)
389 return zone == zone->zone_pgdat->node_zones + ZONE_HIGHMEM;
392 static inline int is_normal(struct zone *zone)
394 return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
397 static inline int is_dma32(struct zone *zone)
399 return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
402 static inline int is_dma(struct zone *zone)
404 return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
407 /* These two functions are used to setup the per zone pages min values */
410 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *,
411 void __user *, size_t *, loff_t *);
412 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
413 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *,
414 void __user *, size_t *, loff_t *);
415 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, struct file *,
416 void __user *, size_t *, loff_t *);
418 #include <linux/topology.h>
419 /* Returns the number of the current Node. */
421 #define numa_node_id() (cpu_to_node(raw_smp_processor_id()))
424 #ifndef CONFIG_NEED_MULTIPLE_NODES
426 extern struct pglist_data contig_page_data;
427 #define NODE_DATA(nid) (&contig_page_data)
428 #define NODE_MEM_MAP(nid) mem_map
429 #define MAX_NODES_SHIFT 1
431 #else /* CONFIG_NEED_MULTIPLE_NODES */
433 #include <asm/mmzone.h>
435 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
437 extern struct pglist_data *first_online_pgdat(void);
438 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
439 extern struct zone *next_zone(struct zone *zone);
442 * for_each_pgdat - helper macro to iterate over all nodes
443 * @pgdat - pointer to a pg_data_t variable
445 #define for_each_online_pgdat(pgdat) \
446 for (pgdat = first_online_pgdat(); \
448 pgdat = next_online_pgdat(pgdat))
450 * for_each_zone - helper macro to iterate over all memory zones
451 * @zone - pointer to struct zone variable
453 * The user only needs to declare the zone variable, for_each_zone
456 #define for_each_zone(zone) \
457 for (zone = (first_online_pgdat())->node_zones; \
459 zone = next_zone(zone))
461 #ifdef CONFIG_SPARSEMEM
462 #include <asm/sparsemem.h>
465 #if BITS_PER_LONG == 32
467 * with 32 bit page->flags field, we reserve 9 bits for node/zone info.
468 * there are 4 zones (3 bits) and this leaves 9-3=6 bits for nodes.
470 #define FLAGS_RESERVED 9
472 #elif BITS_PER_LONG == 64
474 * with 64 bit flags field, there's plenty of room.
476 #define FLAGS_RESERVED 32
480 #error BITS_PER_LONG not defined
484 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
485 #define early_pfn_to_nid(nid) (0UL)
488 #ifdef CONFIG_FLATMEM
489 #define pfn_to_nid(pfn) (0)
492 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
493 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
495 #ifdef CONFIG_SPARSEMEM
498 * SECTION_SHIFT #bits space required to store a section #
500 * PA_SECTION_SHIFT physical address to/from section number
501 * PFN_SECTION_SHIFT pfn to/from section number
503 #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
505 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
506 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
508 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
510 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
511 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
513 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
514 #error Allocator MAX_ORDER exceeds SECTION_SIZE
520 * This is, logically, a pointer to an array of struct
521 * pages. However, it is stored with some other magic.
522 * (see sparse.c::sparse_init_one_section())
524 * Additionally during early boot we encode node id of
525 * the location of the section here to guide allocation.
526 * (see sparse.c::memory_present())
528 * Making it a UL at least makes someone do a cast
529 * before using it wrong.
531 unsigned long section_mem_map;
534 #ifdef CONFIG_SPARSEMEM_EXTREME
535 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
537 #define SECTIONS_PER_ROOT 1
540 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
541 #define NR_SECTION_ROOTS (NR_MEM_SECTIONS / SECTIONS_PER_ROOT)
542 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
544 #ifdef CONFIG_SPARSEMEM_EXTREME
545 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
547 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
550 static inline struct mem_section *__nr_to_section(unsigned long nr)
552 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
554 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
556 extern int __section_nr(struct mem_section* ms);
559 * We use the lower bits of the mem_map pointer to store
560 * a little bit of information. There should be at least
561 * 3 bits here due to 32-bit alignment.
563 #define SECTION_MARKED_PRESENT (1UL<<0)
564 #define SECTION_HAS_MEM_MAP (1UL<<1)
565 #define SECTION_MAP_LAST_BIT (1UL<<2)
566 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
567 #define SECTION_NID_SHIFT 2
569 static inline struct page *__section_mem_map_addr(struct mem_section *section)
571 unsigned long map = section->section_mem_map;
572 map &= SECTION_MAP_MASK;
573 return (struct page *)map;
576 static inline int valid_section(struct mem_section *section)
578 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
581 static inline int section_has_mem_map(struct mem_section *section)
583 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
586 static inline int valid_section_nr(unsigned long nr)
588 return valid_section(__nr_to_section(nr));
591 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
593 return __nr_to_section(pfn_to_section_nr(pfn));
596 static inline int pfn_valid(unsigned long pfn)
598 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
600 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
604 * These are _only_ used during initialisation, therefore they
605 * can use __initdata ... They could have names to indicate
609 #define pfn_to_nid(pfn) \
611 unsigned long __pfn_to_nid_pfn = (pfn); \
612 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
615 #define pfn_to_nid(pfn) (0)
618 #define early_pfn_valid(pfn) pfn_valid(pfn)
619 void sparse_init(void);
621 #define sparse_init() do {} while (0)
622 #define sparse_index_init(_sec, _nid) do {} while (0)
623 #endif /* CONFIG_SPARSEMEM */
625 #ifndef early_pfn_valid
626 #define early_pfn_valid(pfn) (1)
629 void memory_present(int nid, unsigned long start, unsigned long end);
630 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
632 #endif /* !__ASSEMBLY__ */
633 #endif /* __KERNEL__ */
634 #endif /* _LINUX_MMZONE_H */