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 */
55 NR_SLAB_UNRECLAIMABLE,
56 NR_PAGETABLE, /* used for pagetables */
59 NR_UNSTABLE_NFS, /* NFS unstable pages */
63 NUMA_HIT, /* allocated in intended node */
64 NUMA_MISS, /* allocated in non intended node */
65 NUMA_FOREIGN, /* was intended here, hit elsewhere */
66 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
67 NUMA_LOCAL, /* allocation from local node */
68 NUMA_OTHER, /* allocation from other node */
70 NR_VM_ZONE_STAT_ITEMS };
72 struct per_cpu_pages {
73 int count; /* number of pages in the list */
74 int high; /* high watermark, emptying needed */
75 int batch; /* chunk size for buddy add/remove */
76 struct list_head list; /* the list of pages */
79 struct per_cpu_pageset {
80 struct per_cpu_pages pcp[2]; /* 0: hot. 1: cold */
83 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
85 } ____cacheline_aligned_in_smp;
88 #define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)])
90 #define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)])
95 * ZONE_DMA is used when there are devices that are not able
96 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
97 * carve out the portion of memory that is needed for these devices.
98 * The range is arch specific.
103 * ---------------------------
104 * parisc, ia64, sparc <4G
108 * alpha Unlimited or 0-16MB.
110 * i386, x86_64 and multiple other arches
114 #ifdef CONFIG_ZONE_DMA32
116 * x86_64 needs two ZONE_DMAs because it supports devices that are
117 * only able to do DMA to the lower 16M but also 32 bit devices that
118 * can only do DMA areas below 4G.
123 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
124 * performed on pages in ZONE_NORMAL if the DMA devices support
125 * transfers to all addressable memory.
128 #ifdef CONFIG_HIGHMEM
130 * A memory area that is only addressable by the kernel through
131 * mapping portions into its own address space. This is for example
132 * used by i386 to allow the kernel to address the memory beyond
133 * 900MB. The kernel will set up special mappings (page
134 * table entries on i386) for each page that the kernel needs to
143 * When a memory allocation must conform to specific limitations (such
144 * as being suitable for DMA) the caller will pass in hints to the
145 * allocator in the gfp_mask, in the zone modifier bits. These bits
146 * are used to select a priority ordered list of memory zones which
147 * match the requested limits. See gfp_zone() in include/linux/gfp.h
150 #if !defined(CONFIG_ZONE_DMA32) && !defined(CONFIG_HIGHMEM)
151 #define ZONES_SHIFT 1
153 #define ZONES_SHIFT 2
157 /* Fields commonly accessed by the page allocator */
158 unsigned long free_pages;
159 unsigned long pages_min, pages_low, pages_high;
161 * We don't know if the memory that we're going to allocate will be freeable
162 * or/and it will be released eventually, so to avoid totally wasting several
163 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
164 * to run OOM on the lower zones despite there's tons of freeable ram
165 * on the higher zones). This array is recalculated at runtime if the
166 * sysctl_lowmem_reserve_ratio sysctl changes.
168 unsigned long lowmem_reserve[MAX_NR_ZONES];
172 * zone reclaim becomes active if more unmapped pages exist.
174 unsigned long min_unmapped_pages;
175 unsigned long min_slab_pages;
176 struct per_cpu_pageset *pageset[NR_CPUS];
178 struct per_cpu_pageset pageset[NR_CPUS];
181 * free areas of different sizes
184 #ifdef CONFIG_MEMORY_HOTPLUG
185 /* see spanned/present_pages for more description */
186 seqlock_t span_seqlock;
188 struct free_area free_area[MAX_ORDER];
193 /* Fields commonly accessed by the page reclaim scanner */
195 struct list_head active_list;
196 struct list_head inactive_list;
197 unsigned long nr_scan_active;
198 unsigned long nr_scan_inactive;
199 unsigned long nr_active;
200 unsigned long nr_inactive;
201 unsigned long pages_scanned; /* since last reclaim */
202 int all_unreclaimable; /* All pages pinned */
204 /* A count of how many reclaimers are scanning this zone */
205 atomic_t reclaim_in_progress;
207 /* Zone statistics */
208 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
211 * prev_priority holds the scanning priority for this zone. It is
212 * defined as the scanning priority at which we achieved our reclaim
213 * target at the previous try_to_free_pages() or balance_pgdat()
216 * We use prev_priority as a measure of how much stress page reclaim is
217 * under - it drives the swappiness decision: whether to unmap mapped
220 * temp_priority is used to remember the scanning priority at which
221 * this zone was successfully refilled to free_pages == pages_high.
223 * Access to both these fields is quite racy even on uniprocessor. But
224 * it is expected to average out OK.
231 /* Rarely used or read-mostly fields */
234 * wait_table -- the array holding the hash table
235 * wait_table_hash_nr_entries -- the size of the hash table array
236 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
238 * The purpose of all these is to keep track of the people
239 * waiting for a page to become available and make them
240 * runnable again when possible. The trouble is that this
241 * consumes a lot of space, especially when so few things
242 * wait on pages at a given time. So instead of using
243 * per-page waitqueues, we use a waitqueue hash table.
245 * The bucket discipline is to sleep on the same queue when
246 * colliding and wake all in that wait queue when removing.
247 * When something wakes, it must check to be sure its page is
248 * truly available, a la thundering herd. The cost of a
249 * collision is great, but given the expected load of the
250 * table, they should be so rare as to be outweighed by the
251 * benefits from the saved space.
253 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
254 * primary users of these fields, and in mm/page_alloc.c
255 * free_area_init_core() performs the initialization of them.
257 wait_queue_head_t * wait_table;
258 unsigned long wait_table_hash_nr_entries;
259 unsigned long wait_table_bits;
262 * Discontig memory support fields.
264 struct pglist_data *zone_pgdat;
265 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
266 unsigned long zone_start_pfn;
269 * zone_start_pfn, spanned_pages and present_pages are all
270 * protected by span_seqlock. It is a seqlock because it has
271 * to be read outside of zone->lock, and it is done in the main
272 * allocator path. But, it is written quite infrequently.
274 * The lock is declared along with zone->lock because it is
275 * frequently read in proximity to zone->lock. It's good to
276 * give them a chance of being in the same cacheline.
278 unsigned long spanned_pages; /* total size, including holes */
279 unsigned long present_pages; /* amount of memory (excluding holes) */
282 * rarely used fields:
285 } ____cacheline_internodealigned_in_smp;
288 * The "priority" of VM scanning is how much of the queues we will scan in one
289 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
290 * queues ("queue_length >> 12") during an aging round.
292 #define DEF_PRIORITY 12
295 * One allocation request operates on a zonelist. A zonelist
296 * is a list of zones, the first one is the 'goal' of the
297 * allocation, the other zones are fallback zones, in decreasing
300 * Right now a zonelist takes up less than a cacheline. We never
301 * modify it apart from boot-up, and only a few indices are used,
302 * so despite the zonelist table being relatively big, the cache
303 * footprint of this construct is very small.
306 struct zone *zones[MAX_NUMNODES * MAX_NR_ZONES + 1]; // NULL delimited
309 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
310 struct node_active_region {
311 unsigned long start_pfn;
312 unsigned long end_pfn;
315 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
318 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
319 * (mostly NUMA machines?) to denote a higher-level memory zone than the
322 * On NUMA machines, each NUMA node would have a pg_data_t to describe
323 * it's memory layout.
325 * Memory statistics and page replacement data structures are maintained on a
329 typedef struct pglist_data {
330 struct zone node_zones[MAX_NR_ZONES];
331 struct zonelist node_zonelists[MAX_NR_ZONES];
333 #ifdef CONFIG_FLAT_NODE_MEM_MAP
334 struct page *node_mem_map;
336 struct bootmem_data *bdata;
337 #ifdef CONFIG_MEMORY_HOTPLUG
339 * Must be held any time you expect node_start_pfn, node_present_pages
340 * or node_spanned_pages stay constant. Holding this will also
341 * guarantee that any pfn_valid() stays that way.
343 * Nests above zone->lock and zone->size_seqlock.
345 spinlock_t node_size_lock;
347 unsigned long node_start_pfn;
348 unsigned long node_present_pages; /* total number of physical pages */
349 unsigned long node_spanned_pages; /* total size of physical page
350 range, including holes */
352 wait_queue_head_t kswapd_wait;
353 struct task_struct *kswapd;
354 int kswapd_max_order;
357 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
358 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
359 #ifdef CONFIG_FLAT_NODE_MEM_MAP
360 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
362 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
364 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
366 #include <linux/memory_hotplug.h>
368 void __get_zone_counts(unsigned long *active, unsigned long *inactive,
369 unsigned long *free, struct pglist_data *pgdat);
370 void get_zone_counts(unsigned long *active, unsigned long *inactive,
371 unsigned long *free);
372 void build_all_zonelists(void);
373 void wakeup_kswapd(struct zone *zone, int order);
374 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
375 int classzone_idx, int alloc_flags);
377 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
380 #ifdef CONFIG_HAVE_MEMORY_PRESENT
381 void memory_present(int nid, unsigned long start, unsigned long end);
383 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
386 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
387 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
391 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
393 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
395 static inline int populated_zone(struct zone *zone)
397 return (!!zone->present_pages);
400 static inline int is_highmem_idx(enum zone_type idx)
402 #ifdef CONFIG_HIGHMEM
403 return (idx == ZONE_HIGHMEM);
409 static inline int is_normal_idx(enum zone_type idx)
411 return (idx == ZONE_NORMAL);
415 * is_highmem - helper function to quickly check if a struct zone is a
416 * highmem zone or not. This is an attempt to keep references
417 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
418 * @zone - pointer to struct zone variable
420 static inline int is_highmem(struct zone *zone)
422 #ifdef CONFIG_HIGHMEM
423 return zone == zone->zone_pgdat->node_zones + ZONE_HIGHMEM;
429 static inline int is_normal(struct zone *zone)
431 return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
434 static inline int is_dma32(struct zone *zone)
436 #ifdef CONFIG_ZONE_DMA32
437 return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
443 static inline int is_dma(struct zone *zone)
445 return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
448 /* These two functions are used to setup the per zone pages min values */
451 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *,
452 void __user *, size_t *, loff_t *);
453 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
454 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *,
455 void __user *, size_t *, loff_t *);
456 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, struct file *,
457 void __user *, size_t *, loff_t *);
458 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
459 struct file *, void __user *, size_t *, loff_t *);
460 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
461 struct file *, void __user *, size_t *, loff_t *);
463 #include <linux/topology.h>
464 /* Returns the number of the current Node. */
466 #define numa_node_id() (cpu_to_node(raw_smp_processor_id()))
469 #ifndef CONFIG_NEED_MULTIPLE_NODES
471 extern struct pglist_data contig_page_data;
472 #define NODE_DATA(nid) (&contig_page_data)
473 #define NODE_MEM_MAP(nid) mem_map
474 #define MAX_NODES_SHIFT 1
476 #else /* CONFIG_NEED_MULTIPLE_NODES */
478 #include <asm/mmzone.h>
480 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
482 extern struct pglist_data *first_online_pgdat(void);
483 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
484 extern struct zone *next_zone(struct zone *zone);
487 * for_each_pgdat - helper macro to iterate over all nodes
488 * @pgdat - pointer to a pg_data_t variable
490 #define for_each_online_pgdat(pgdat) \
491 for (pgdat = first_online_pgdat(); \
493 pgdat = next_online_pgdat(pgdat))
495 * for_each_zone - helper macro to iterate over all memory zones
496 * @zone - pointer to struct zone variable
498 * The user only needs to declare the zone variable, for_each_zone
501 #define for_each_zone(zone) \
502 for (zone = (first_online_pgdat())->node_zones; \
504 zone = next_zone(zone))
506 #ifdef CONFIG_SPARSEMEM
507 #include <asm/sparsemem.h>
510 #if BITS_PER_LONG == 32
512 * with 32 bit page->flags field, we reserve 9 bits for node/zone info.
513 * there are 4 zones (3 bits) and this leaves 9-3=6 bits for nodes.
515 #define FLAGS_RESERVED 9
517 #elif BITS_PER_LONG == 64
519 * with 64 bit flags field, there's plenty of room.
521 #define FLAGS_RESERVED 32
525 #error BITS_PER_LONG not defined
529 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
530 !defined(CONFIG_ARCH_POPULATES_NODE_MAP)
531 #define early_pfn_to_nid(nid) (0UL)
534 #ifdef CONFIG_FLATMEM
535 #define pfn_to_nid(pfn) (0)
538 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
539 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
541 #ifdef CONFIG_SPARSEMEM
544 * SECTION_SHIFT #bits space required to store a section #
546 * PA_SECTION_SHIFT physical address to/from section number
547 * PFN_SECTION_SHIFT pfn to/from section number
549 #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
551 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
552 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
554 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
556 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
557 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
559 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
560 #error Allocator MAX_ORDER exceeds SECTION_SIZE
566 * This is, logically, a pointer to an array of struct
567 * pages. However, it is stored with some other magic.
568 * (see sparse.c::sparse_init_one_section())
570 * Additionally during early boot we encode node id of
571 * the location of the section here to guide allocation.
572 * (see sparse.c::memory_present())
574 * Making it a UL at least makes someone do a cast
575 * before using it wrong.
577 unsigned long section_mem_map;
580 #ifdef CONFIG_SPARSEMEM_EXTREME
581 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
583 #define SECTIONS_PER_ROOT 1
586 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
587 #define NR_SECTION_ROOTS (NR_MEM_SECTIONS / SECTIONS_PER_ROOT)
588 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
590 #ifdef CONFIG_SPARSEMEM_EXTREME
591 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
593 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
596 static inline struct mem_section *__nr_to_section(unsigned long nr)
598 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
600 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
602 extern int __section_nr(struct mem_section* ms);
605 * We use the lower bits of the mem_map pointer to store
606 * a little bit of information. There should be at least
607 * 3 bits here due to 32-bit alignment.
609 #define SECTION_MARKED_PRESENT (1UL<<0)
610 #define SECTION_HAS_MEM_MAP (1UL<<1)
611 #define SECTION_MAP_LAST_BIT (1UL<<2)
612 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
613 #define SECTION_NID_SHIFT 2
615 static inline struct page *__section_mem_map_addr(struct mem_section *section)
617 unsigned long map = section->section_mem_map;
618 map &= SECTION_MAP_MASK;
619 return (struct page *)map;
622 static inline int valid_section(struct mem_section *section)
624 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
627 static inline int section_has_mem_map(struct mem_section *section)
629 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
632 static inline int valid_section_nr(unsigned long nr)
634 return valid_section(__nr_to_section(nr));
637 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
639 return __nr_to_section(pfn_to_section_nr(pfn));
642 static inline int pfn_valid(unsigned long pfn)
644 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
646 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
650 * These are _only_ used during initialisation, therefore they
651 * can use __initdata ... They could have names to indicate
655 #define pfn_to_nid(pfn) \
657 unsigned long __pfn_to_nid_pfn = (pfn); \
658 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
661 #define pfn_to_nid(pfn) (0)
664 #define early_pfn_valid(pfn) pfn_valid(pfn)
665 void sparse_init(void);
667 #define sparse_init() do {} while (0)
668 #define sparse_index_init(_sec, _nid) do {} while (0)
669 #endif /* CONFIG_SPARSEMEM */
671 #ifndef early_pfn_valid
672 #define early_pfn_valid(pfn) (1)
675 void memory_present(int nid, unsigned long start, unsigned long end);
676 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
678 #endif /* !__ASSEMBLY__ */
679 #endif /* __KERNEL__ */
680 #endif /* _LINUX_MMZONE_H */