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 */
81 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
83 } ____cacheline_aligned_in_smp;
86 #define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)])
88 #define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)])
93 * ZONE_DMA is used when there are devices that are not able
94 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
95 * carve out the portion of memory that is needed for these devices.
96 * The range is arch specific.
101 * ---------------------------
102 * parisc, ia64, sparc <4G
106 * alpha Unlimited or 0-16MB.
108 * i386, x86_64 and multiple other arches
112 #ifdef CONFIG_ZONE_DMA32
114 * x86_64 needs two ZONE_DMAs because it supports devices that are
115 * only able to do DMA to the lower 16M but also 32 bit devices that
116 * can only do DMA areas below 4G.
121 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
122 * performed on pages in ZONE_NORMAL if the DMA devices support
123 * transfers to all addressable memory.
126 #ifdef CONFIG_HIGHMEM
128 * A memory area that is only addressable by the kernel through
129 * mapping portions into its own address space. This is for example
130 * used by i386 to allow the kernel to address the memory beyond
131 * 900MB. The kernel will set up special mappings (page
132 * table entries on i386) for each page that the kernel needs to
142 * When a memory allocation must conform to specific limitations (such
143 * as being suitable for DMA) the caller will pass in hints to the
144 * allocator in the gfp_mask, in the zone modifier bits. These bits
145 * are used to select a priority ordered list of memory zones which
146 * match the requested limits. GFP_ZONEMASK defines which bits within
147 * the gfp_mask should be considered as zone modifiers. Each valid
148 * combination of the zone modifier bits has a corresponding list
149 * of zones (in node_zonelists). Thus for two zone modifiers there
150 * will be a maximum of 4 (2 ** 2) zonelists, for 3 modifiers there will
151 * be 8 (2 ** 3) zonelists. GFP_ZONETYPES defines the number of possible
152 * combinations of zone modifiers in "zone modifier space".
154 * As an optimisation any zone modifier bits which are only valid when
155 * no other zone modifier bits are set (loners) should be placed in
156 * the highest order bits of this field. This allows us to reduce the
157 * extent of the zonelists thus saving space. For example in the case
158 * of three zone modifier bits, we could require up to eight zonelists.
159 * If the left most zone modifier is a "loner" then the highest valid
160 * zonelist would be four allowing us to allocate only five zonelists.
161 * Use the first form for GFP_ZONETYPES when the left most bit is not
162 * a "loner", otherwise use the second.
164 * NOTE! Make sure this matches the zones in <linux/gfp.h>
167 #ifdef CONFIG_ZONE_DMA32
169 #ifdef CONFIG_HIGHMEM
170 #define GFP_ZONETYPES ((GFP_ZONEMASK + 1) / 2 + 1) /* Loner */
171 #define GFP_ZONEMASK 0x07
172 #define ZONES_SHIFT 2 /* ceil(log2(MAX_NR_ZONES)) */
174 #define GFP_ZONETYPES ((0x07 + 1) / 2 + 1) /* Loner */
175 /* Mask __GFP_HIGHMEM */
176 #define GFP_ZONEMASK 0x05
177 #define ZONES_SHIFT 2
181 #ifdef CONFIG_HIGHMEM
183 #define GFP_ZONEMASK 0x03
184 #define ZONES_SHIFT 2
185 #define GFP_ZONETYPES 3
189 #define GFP_ZONEMASK 0x01
190 #define ZONES_SHIFT 1
191 #define GFP_ZONETYPES 2
197 /* Fields commonly accessed by the page allocator */
198 unsigned long free_pages;
199 unsigned long pages_min, pages_low, pages_high;
201 * We don't know if the memory that we're going to allocate will be freeable
202 * or/and it will be released eventually, so to avoid totally wasting several
203 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
204 * to run OOM on the lower zones despite there's tons of freeable ram
205 * on the higher zones). This array is recalculated at runtime if the
206 * sysctl_lowmem_reserve_ratio sysctl changes.
208 unsigned long lowmem_reserve[MAX_NR_ZONES];
212 * zone reclaim becomes active if more unmapped pages exist.
214 unsigned long min_unmapped_ratio;
215 struct per_cpu_pageset *pageset[NR_CPUS];
217 struct per_cpu_pageset pageset[NR_CPUS];
220 * free areas of different sizes
223 #ifdef CONFIG_MEMORY_HOTPLUG
224 /* see spanned/present_pages for more description */
225 seqlock_t span_seqlock;
227 struct free_area free_area[MAX_ORDER];
232 /* Fields commonly accessed by the page reclaim scanner */
234 struct list_head active_list;
235 struct list_head inactive_list;
236 unsigned long nr_scan_active;
237 unsigned long nr_scan_inactive;
238 unsigned long nr_active;
239 unsigned long nr_inactive;
240 unsigned long pages_scanned; /* since last reclaim */
241 int all_unreclaimable; /* All pages pinned */
243 /* A count of how many reclaimers are scanning this zone */
244 atomic_t reclaim_in_progress;
246 /* Zone statistics */
247 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
250 * prev_priority holds the scanning priority for this zone. It is
251 * defined as the scanning priority at which we achieved our reclaim
252 * target at the previous try_to_free_pages() or balance_pgdat()
255 * We use prev_priority as a measure of how much stress page reclaim is
256 * under - it drives the swappiness decision: whether to unmap mapped
259 * temp_priority is used to remember the scanning priority at which
260 * this zone was successfully refilled to free_pages == pages_high.
262 * Access to both these fields is quite racy even on uniprocessor. But
263 * it is expected to average out OK.
270 /* Rarely used or read-mostly fields */
273 * wait_table -- the array holding the hash table
274 * wait_table_hash_nr_entries -- the size of the hash table array
275 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
277 * The purpose of all these is to keep track of the people
278 * waiting for a page to become available and make them
279 * runnable again when possible. The trouble is that this
280 * consumes a lot of space, especially when so few things
281 * wait on pages at a given time. So instead of using
282 * per-page waitqueues, we use a waitqueue hash table.
284 * The bucket discipline is to sleep on the same queue when
285 * colliding and wake all in that wait queue when removing.
286 * When something wakes, it must check to be sure its page is
287 * truly available, a la thundering herd. The cost of a
288 * collision is great, but given the expected load of the
289 * table, they should be so rare as to be outweighed by the
290 * benefits from the saved space.
292 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
293 * primary users of these fields, and in mm/page_alloc.c
294 * free_area_init_core() performs the initialization of them.
296 wait_queue_head_t * wait_table;
297 unsigned long wait_table_hash_nr_entries;
298 unsigned long wait_table_bits;
301 * Discontig memory support fields.
303 struct pglist_data *zone_pgdat;
304 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
305 unsigned long zone_start_pfn;
308 * zone_start_pfn, spanned_pages and present_pages are all
309 * protected by span_seqlock. It is a seqlock because it has
310 * to be read outside of zone->lock, and it is done in the main
311 * allocator path. But, it is written quite infrequently.
313 * The lock is declared along with zone->lock because it is
314 * frequently read in proximity to zone->lock. It's good to
315 * give them a chance of being in the same cacheline.
317 unsigned long spanned_pages; /* total size, including holes */
318 unsigned long present_pages; /* amount of memory (excluding holes) */
321 * rarely used fields:
324 } ____cacheline_internodealigned_in_smp;
327 * The "priority" of VM scanning is how much of the queues we will scan in one
328 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
329 * queues ("queue_length >> 12") during an aging round.
331 #define DEF_PRIORITY 12
334 * One allocation request operates on a zonelist. A zonelist
335 * is a list of zones, the first one is the 'goal' of the
336 * allocation, the other zones are fallback zones, in decreasing
339 * Right now a zonelist takes up less than a cacheline. We never
340 * modify it apart from boot-up, and only a few indices are used,
341 * so despite the zonelist table being relatively big, the cache
342 * footprint of this construct is very small.
345 struct zone *zones[MAX_NUMNODES * MAX_NR_ZONES + 1]; // NULL delimited
350 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
351 * (mostly NUMA machines?) to denote a higher-level memory zone than the
354 * On NUMA machines, each NUMA node would have a pg_data_t to describe
355 * it's memory layout.
357 * Memory statistics and page replacement data structures are maintained on a
361 typedef struct pglist_data {
362 struct zone node_zones[MAX_NR_ZONES];
363 struct zonelist node_zonelists[GFP_ZONETYPES];
365 #ifdef CONFIG_FLAT_NODE_MEM_MAP
366 struct page *node_mem_map;
368 struct bootmem_data *bdata;
369 #ifdef CONFIG_MEMORY_HOTPLUG
371 * Must be held any time you expect node_start_pfn, node_present_pages
372 * or node_spanned_pages stay constant. Holding this will also
373 * guarantee that any pfn_valid() stays that way.
375 * Nests above zone->lock and zone->size_seqlock.
377 spinlock_t node_size_lock;
379 unsigned long node_start_pfn;
380 unsigned long node_present_pages; /* total number of physical pages */
381 unsigned long node_spanned_pages; /* total size of physical page
382 range, including holes */
384 wait_queue_head_t kswapd_wait;
385 struct task_struct *kswapd;
386 int kswapd_max_order;
389 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
390 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
391 #ifdef CONFIG_FLAT_NODE_MEM_MAP
392 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
394 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
396 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
398 #include <linux/memory_hotplug.h>
400 void __get_zone_counts(unsigned long *active, unsigned long *inactive,
401 unsigned long *free, struct pglist_data *pgdat);
402 void get_zone_counts(unsigned long *active, unsigned long *inactive,
403 unsigned long *free);
404 void build_all_zonelists(void);
405 void wakeup_kswapd(struct zone *zone, int order);
406 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
407 int classzone_idx, int alloc_flags);
409 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
412 #ifdef CONFIG_HAVE_MEMORY_PRESENT
413 void memory_present(int nid, unsigned long start, unsigned long end);
415 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
418 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
419 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
423 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
425 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
427 static inline int populated_zone(struct zone *zone)
429 return (!!zone->present_pages);
432 static inline int is_highmem_idx(enum zone_type idx)
434 #ifdef CONFIG_HIGHMEM
435 return (idx == ZONE_HIGHMEM);
441 static inline int is_normal_idx(enum zone_type idx)
443 return (idx == ZONE_NORMAL);
447 * is_highmem - helper function to quickly check if a struct zone is a
448 * highmem zone or not. This is an attempt to keep references
449 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
450 * @zone - pointer to struct zone variable
452 static inline int is_highmem(struct zone *zone)
454 #ifdef CONFIG_HIGHMEM
455 return zone == zone->zone_pgdat->node_zones + ZONE_HIGHMEM;
461 static inline int is_normal(struct zone *zone)
463 return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
466 static inline int is_dma32(struct zone *zone)
468 #ifdef CONFIG_ZONE_DMA32
469 return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
475 static inline int is_dma(struct zone *zone)
477 return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
480 /* These two functions are used to setup the per zone pages min values */
483 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *,
484 void __user *, size_t *, loff_t *);
485 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
486 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *,
487 void __user *, size_t *, loff_t *);
488 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, struct file *,
489 void __user *, size_t *, loff_t *);
490 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
491 struct file *, void __user *, size_t *, loff_t *);
493 #include <linux/topology.h>
494 /* Returns the number of the current Node. */
496 #define numa_node_id() (cpu_to_node(raw_smp_processor_id()))
499 #ifndef CONFIG_NEED_MULTIPLE_NODES
501 extern struct pglist_data contig_page_data;
502 #define NODE_DATA(nid) (&contig_page_data)
503 #define NODE_MEM_MAP(nid) mem_map
504 #define MAX_NODES_SHIFT 1
506 #else /* CONFIG_NEED_MULTIPLE_NODES */
508 #include <asm/mmzone.h>
510 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
512 extern struct pglist_data *first_online_pgdat(void);
513 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
514 extern struct zone *next_zone(struct zone *zone);
517 * for_each_pgdat - helper macro to iterate over all nodes
518 * @pgdat - pointer to a pg_data_t variable
520 #define for_each_online_pgdat(pgdat) \
521 for (pgdat = first_online_pgdat(); \
523 pgdat = next_online_pgdat(pgdat))
525 * for_each_zone - helper macro to iterate over all memory zones
526 * @zone - pointer to struct zone variable
528 * The user only needs to declare the zone variable, for_each_zone
531 #define for_each_zone(zone) \
532 for (zone = (first_online_pgdat())->node_zones; \
534 zone = next_zone(zone))
536 #ifdef CONFIG_SPARSEMEM
537 #include <asm/sparsemem.h>
540 #if BITS_PER_LONG == 32
542 * with 32 bit page->flags field, we reserve 9 bits for node/zone info.
543 * there are 4 zones (3 bits) and this leaves 9-3=6 bits for nodes.
545 #define FLAGS_RESERVED 9
547 #elif BITS_PER_LONG == 64
549 * with 64 bit flags field, there's plenty of room.
551 #define FLAGS_RESERVED 32
555 #error BITS_PER_LONG not defined
559 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
560 #define early_pfn_to_nid(nid) (0UL)
563 #ifdef CONFIG_FLATMEM
564 #define pfn_to_nid(pfn) (0)
567 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
568 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
570 #ifdef CONFIG_SPARSEMEM
573 * SECTION_SHIFT #bits space required to store a section #
575 * PA_SECTION_SHIFT physical address to/from section number
576 * PFN_SECTION_SHIFT pfn to/from section number
578 #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
580 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
581 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
583 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
585 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
586 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
588 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
589 #error Allocator MAX_ORDER exceeds SECTION_SIZE
595 * This is, logically, a pointer to an array of struct
596 * pages. However, it is stored with some other magic.
597 * (see sparse.c::sparse_init_one_section())
599 * Additionally during early boot we encode node id of
600 * the location of the section here to guide allocation.
601 * (see sparse.c::memory_present())
603 * Making it a UL at least makes someone do a cast
604 * before using it wrong.
606 unsigned long section_mem_map;
609 #ifdef CONFIG_SPARSEMEM_EXTREME
610 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
612 #define SECTIONS_PER_ROOT 1
615 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
616 #define NR_SECTION_ROOTS (NR_MEM_SECTIONS / SECTIONS_PER_ROOT)
617 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
619 #ifdef CONFIG_SPARSEMEM_EXTREME
620 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
622 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
625 static inline struct mem_section *__nr_to_section(unsigned long nr)
627 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
629 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
631 extern int __section_nr(struct mem_section* ms);
634 * We use the lower bits of the mem_map pointer to store
635 * a little bit of information. There should be at least
636 * 3 bits here due to 32-bit alignment.
638 #define SECTION_MARKED_PRESENT (1UL<<0)
639 #define SECTION_HAS_MEM_MAP (1UL<<1)
640 #define SECTION_MAP_LAST_BIT (1UL<<2)
641 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
642 #define SECTION_NID_SHIFT 2
644 static inline struct page *__section_mem_map_addr(struct mem_section *section)
646 unsigned long map = section->section_mem_map;
647 map &= SECTION_MAP_MASK;
648 return (struct page *)map;
651 static inline int valid_section(struct mem_section *section)
653 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
656 static inline int section_has_mem_map(struct mem_section *section)
658 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
661 static inline int valid_section_nr(unsigned long nr)
663 return valid_section(__nr_to_section(nr));
666 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
668 return __nr_to_section(pfn_to_section_nr(pfn));
671 static inline int pfn_valid(unsigned long pfn)
673 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
675 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
679 * These are _only_ used during initialisation, therefore they
680 * can use __initdata ... They could have names to indicate
684 #define pfn_to_nid(pfn) \
686 unsigned long __pfn_to_nid_pfn = (pfn); \
687 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
690 #define pfn_to_nid(pfn) (0)
693 #define early_pfn_valid(pfn) pfn_valid(pfn)
694 void sparse_init(void);
696 #define sparse_init() do {} while (0)
697 #define sparse_index_init(_sec, _nid) do {} while (0)
698 #endif /* CONFIG_SPARSEMEM */
700 #ifndef early_pfn_valid
701 #define early_pfn_valid(pfn) (1)
704 void memory_present(int nid, unsigned long start, unsigned long end);
705 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
707 #endif /* !__ASSEMBLY__ */
708 #endif /* __KERNEL__ */
709 #endif /* _LINUX_MMZONE_H */