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_VM_ZONE_STAT_ITEMS };
60 struct per_cpu_pages {
61 int count; /* number of pages in the list */
62 int high; /* high watermark, emptying needed */
63 int batch; /* chunk size for buddy add/remove */
64 struct list_head list; /* the list of pages */
67 struct per_cpu_pageset {
68 struct per_cpu_pages pcp[2]; /* 0: hot. 1: cold */
70 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
74 unsigned long numa_hit; /* allocated in intended node */
75 unsigned long numa_miss; /* allocated in non intended node */
76 unsigned long numa_foreign; /* was intended here, hit elsewhere */
77 unsigned long interleave_hit; /* interleaver prefered this zone */
78 unsigned long local_node; /* allocation from local node */
79 unsigned long other_node; /* allocation from other node */
81 } ____cacheline_aligned_in_smp;
84 #define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)])
86 #define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)])
92 #define ZONE_HIGHMEM 3
94 #define MAX_NR_ZONES 4 /* Sync this with ZONES_SHIFT */
95 #define ZONES_SHIFT 2 /* ceil(log2(MAX_NR_ZONES)) */
99 * When a memory allocation must conform to specific limitations (such
100 * as being suitable for DMA) the caller will pass in hints to the
101 * allocator in the gfp_mask, in the zone modifier bits. These bits
102 * are used to select a priority ordered list of memory zones which
103 * match the requested limits. GFP_ZONEMASK defines which bits within
104 * the gfp_mask should be considered as zone modifiers. Each valid
105 * combination of the zone modifier bits has a corresponding list
106 * of zones (in node_zonelists). Thus for two zone modifiers there
107 * will be a maximum of 4 (2 ** 2) zonelists, for 3 modifiers there will
108 * be 8 (2 ** 3) zonelists. GFP_ZONETYPES defines the number of possible
109 * combinations of zone modifiers in "zone modifier space".
111 * As an optimisation any zone modifier bits which are only valid when
112 * no other zone modifier bits are set (loners) should be placed in
113 * the highest order bits of this field. This allows us to reduce the
114 * extent of the zonelists thus saving space. For example in the case
115 * of three zone modifier bits, we could require up to eight zonelists.
116 * If the left most zone modifier is a "loner" then the highest valid
117 * zonelist would be four allowing us to allocate only five zonelists.
118 * Use the first form for GFP_ZONETYPES when the left most bit is not
119 * a "loner", otherwise use the second.
121 * NOTE! Make sure this matches the zones in <linux/gfp.h>
123 #define GFP_ZONEMASK 0x07
124 /* #define GFP_ZONETYPES (GFP_ZONEMASK + 1) */ /* Non-loner */
125 #define GFP_ZONETYPES ((GFP_ZONEMASK + 1) / 2 + 1) /* Loner */
128 * On machines where it is needed (eg PCs) we divide physical memory
129 * into multiple physical zones. On a 32bit PC we have 4 zones:
131 * ZONE_DMA < 16 MB ISA DMA capable memory
132 * ZONE_DMA32 0 MB Empty
133 * ZONE_NORMAL 16-896 MB direct mapped by the kernel
134 * ZONE_HIGHMEM > 896 MB only page cache and user processes
138 /* Fields commonly accessed by the page allocator */
139 unsigned long free_pages;
140 unsigned long pages_min, pages_low, pages_high;
142 * We don't know if the memory that we're going to allocate will be freeable
143 * or/and it will be released eventually, so to avoid totally wasting several
144 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
145 * to run OOM on the lower zones despite there's tons of freeable ram
146 * on the higher zones). This array is recalculated at runtime if the
147 * sysctl_lowmem_reserve_ratio sysctl changes.
149 unsigned long lowmem_reserve[MAX_NR_ZONES];
152 struct per_cpu_pageset *pageset[NR_CPUS];
154 struct per_cpu_pageset pageset[NR_CPUS];
157 * free areas of different sizes
160 #ifdef CONFIG_MEMORY_HOTPLUG
161 /* see spanned/present_pages for more description */
162 seqlock_t span_seqlock;
164 struct free_area free_area[MAX_ORDER];
169 /* Fields commonly accessed by the page reclaim scanner */
171 struct list_head active_list;
172 struct list_head inactive_list;
173 unsigned long nr_scan_active;
174 unsigned long nr_scan_inactive;
175 unsigned long nr_active;
176 unsigned long nr_inactive;
177 unsigned long pages_scanned; /* since last reclaim */
178 int all_unreclaimable; /* All pages pinned */
180 /* A count of how many reclaimers are scanning this zone */
181 atomic_t reclaim_in_progress;
183 /* Zone statistics */
184 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
187 * prev_priority holds the scanning priority for this zone. It is
188 * defined as the scanning priority at which we achieved our reclaim
189 * target at the previous try_to_free_pages() or balance_pgdat()
192 * We use prev_priority as a measure of how much stress page reclaim is
193 * under - it drives the swappiness decision: whether to unmap mapped
196 * temp_priority is used to remember the scanning priority at which
197 * this zone was successfully refilled to free_pages == pages_high.
199 * Access to both these fields is quite racy even on uniprocessor. But
200 * it is expected to average out OK.
207 /* Rarely used or read-mostly fields */
210 * wait_table -- the array holding the hash table
211 * wait_table_hash_nr_entries -- the size of the hash table array
212 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
214 * The purpose of all these is to keep track of the people
215 * waiting for a page to become available and make them
216 * runnable again when possible. The trouble is that this
217 * consumes a lot of space, especially when so few things
218 * wait on pages at a given time. So instead of using
219 * per-page waitqueues, we use a waitqueue hash table.
221 * The bucket discipline is to sleep on the same queue when
222 * colliding and wake all in that wait queue when removing.
223 * When something wakes, it must check to be sure its page is
224 * truly available, a la thundering herd. The cost of a
225 * collision is great, but given the expected load of the
226 * table, they should be so rare as to be outweighed by the
227 * benefits from the saved space.
229 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
230 * primary users of these fields, and in mm/page_alloc.c
231 * free_area_init_core() performs the initialization of them.
233 wait_queue_head_t * wait_table;
234 unsigned long wait_table_hash_nr_entries;
235 unsigned long wait_table_bits;
238 * Discontig memory support fields.
240 struct pglist_data *zone_pgdat;
241 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
242 unsigned long zone_start_pfn;
245 * zone_start_pfn, spanned_pages and present_pages are all
246 * protected by span_seqlock. It is a seqlock because it has
247 * to be read outside of zone->lock, and it is done in the main
248 * allocator path. But, it is written quite infrequently.
250 * The lock is declared along with zone->lock because it is
251 * frequently read in proximity to zone->lock. It's good to
252 * give them a chance of being in the same cacheline.
254 unsigned long spanned_pages; /* total size, including holes */
255 unsigned long present_pages; /* amount of memory (excluding holes) */
258 * rarely used fields:
261 } ____cacheline_internodealigned_in_smp;
265 * The "priority" of VM scanning is how much of the queues we will scan in one
266 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
267 * queues ("queue_length >> 12") during an aging round.
269 #define DEF_PRIORITY 12
272 * One allocation request operates on a zonelist. A zonelist
273 * is a list of zones, the first one is the 'goal' of the
274 * allocation, the other zones are fallback zones, in decreasing
277 * Right now a zonelist takes up less than a cacheline. We never
278 * modify it apart from boot-up, and only a few indices are used,
279 * so despite the zonelist table being relatively big, the cache
280 * footprint of this construct is very small.
283 struct zone *zones[MAX_NUMNODES * MAX_NR_ZONES + 1]; // NULL delimited
288 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
289 * (mostly NUMA machines?) to denote a higher-level memory zone than the
292 * On NUMA machines, each NUMA node would have a pg_data_t to describe
293 * it's memory layout.
295 * Memory statistics and page replacement data structures are maintained on a
299 typedef struct pglist_data {
300 struct zone node_zones[MAX_NR_ZONES];
301 struct zonelist node_zonelists[GFP_ZONETYPES];
303 #ifdef CONFIG_FLAT_NODE_MEM_MAP
304 struct page *node_mem_map;
306 struct bootmem_data *bdata;
307 #ifdef CONFIG_MEMORY_HOTPLUG
309 * Must be held any time you expect node_start_pfn, node_present_pages
310 * or node_spanned_pages stay constant. Holding this will also
311 * guarantee that any pfn_valid() stays that way.
313 * Nests above zone->lock and zone->size_seqlock.
315 spinlock_t node_size_lock;
317 unsigned long node_start_pfn;
318 unsigned long node_present_pages; /* total number of physical pages */
319 unsigned long node_spanned_pages; /* total size of physical page
320 range, including holes */
322 wait_queue_head_t kswapd_wait;
323 struct task_struct *kswapd;
324 int kswapd_max_order;
327 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
328 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
329 #ifdef CONFIG_FLAT_NODE_MEM_MAP
330 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
332 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
334 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
336 #include <linux/memory_hotplug.h>
338 void __get_zone_counts(unsigned long *active, unsigned long *inactive,
339 unsigned long *free, struct pglist_data *pgdat);
340 void get_zone_counts(unsigned long *active, unsigned long *inactive,
341 unsigned long *free);
342 void build_all_zonelists(void);
343 void wakeup_kswapd(struct zone *zone, int order);
344 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
345 int classzone_idx, int alloc_flags);
347 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
350 #ifdef CONFIG_HAVE_MEMORY_PRESENT
351 void memory_present(int nid, unsigned long start, unsigned long end);
353 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
356 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
357 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
361 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
363 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
365 static inline int populated_zone(struct zone *zone)
367 return (!!zone->present_pages);
370 static inline int is_highmem_idx(int idx)
372 return (idx == ZONE_HIGHMEM);
375 static inline int is_normal_idx(int idx)
377 return (idx == ZONE_NORMAL);
381 * is_highmem - helper function to quickly check if a struct zone is a
382 * highmem zone or not. This is an attempt to keep references
383 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
384 * @zone - pointer to struct zone variable
386 static inline int is_highmem(struct zone *zone)
388 return zone == zone->zone_pgdat->node_zones + ZONE_HIGHMEM;
391 static inline int is_normal(struct zone *zone)
393 return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
396 static inline int is_dma32(struct zone *zone)
398 return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
401 static inline int is_dma(struct zone *zone)
403 return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
406 /* These two functions are used to setup the per zone pages min values */
409 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *,
410 void __user *, size_t *, loff_t *);
411 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
412 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *,
413 void __user *, size_t *, loff_t *);
414 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, struct file *,
415 void __user *, size_t *, loff_t *);
417 #include <linux/topology.h>
418 /* Returns the number of the current Node. */
420 #define numa_node_id() (cpu_to_node(raw_smp_processor_id()))
423 #ifndef CONFIG_NEED_MULTIPLE_NODES
425 extern struct pglist_data contig_page_data;
426 #define NODE_DATA(nid) (&contig_page_data)
427 #define NODE_MEM_MAP(nid) mem_map
428 #define MAX_NODES_SHIFT 1
430 #else /* CONFIG_NEED_MULTIPLE_NODES */
432 #include <asm/mmzone.h>
434 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
436 extern struct pglist_data *first_online_pgdat(void);
437 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
438 extern struct zone *next_zone(struct zone *zone);
441 * for_each_pgdat - helper macro to iterate over all nodes
442 * @pgdat - pointer to a pg_data_t variable
444 #define for_each_online_pgdat(pgdat) \
445 for (pgdat = first_online_pgdat(); \
447 pgdat = next_online_pgdat(pgdat))
449 * for_each_zone - helper macro to iterate over all memory zones
450 * @zone - pointer to struct zone variable
452 * The user only needs to declare the zone variable, for_each_zone
455 #define for_each_zone(zone) \
456 for (zone = (first_online_pgdat())->node_zones; \
458 zone = next_zone(zone))
460 #ifdef CONFIG_SPARSEMEM
461 #include <asm/sparsemem.h>
464 #if BITS_PER_LONG == 32
466 * with 32 bit page->flags field, we reserve 9 bits for node/zone info.
467 * there are 4 zones (3 bits) and this leaves 9-3=6 bits for nodes.
469 #define FLAGS_RESERVED 9
471 #elif BITS_PER_LONG == 64
473 * with 64 bit flags field, there's plenty of room.
475 #define FLAGS_RESERVED 32
479 #error BITS_PER_LONG not defined
483 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
484 #define early_pfn_to_nid(nid) (0UL)
487 #ifdef CONFIG_FLATMEM
488 #define pfn_to_nid(pfn) (0)
491 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
492 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
494 #ifdef CONFIG_SPARSEMEM
497 * SECTION_SHIFT #bits space required to store a section #
499 * PA_SECTION_SHIFT physical address to/from section number
500 * PFN_SECTION_SHIFT pfn to/from section number
502 #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
504 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
505 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
507 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
509 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
510 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
512 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
513 #error Allocator MAX_ORDER exceeds SECTION_SIZE
519 * This is, logically, a pointer to an array of struct
520 * pages. However, it is stored with some other magic.
521 * (see sparse.c::sparse_init_one_section())
523 * Additionally during early boot we encode node id of
524 * the location of the section here to guide allocation.
525 * (see sparse.c::memory_present())
527 * Making it a UL at least makes someone do a cast
528 * before using it wrong.
530 unsigned long section_mem_map;
533 #ifdef CONFIG_SPARSEMEM_EXTREME
534 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
536 #define SECTIONS_PER_ROOT 1
539 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
540 #define NR_SECTION_ROOTS (NR_MEM_SECTIONS / SECTIONS_PER_ROOT)
541 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
543 #ifdef CONFIG_SPARSEMEM_EXTREME
544 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
546 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
549 static inline struct mem_section *__nr_to_section(unsigned long nr)
551 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
553 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
555 extern int __section_nr(struct mem_section* ms);
558 * We use the lower bits of the mem_map pointer to store
559 * a little bit of information. There should be at least
560 * 3 bits here due to 32-bit alignment.
562 #define SECTION_MARKED_PRESENT (1UL<<0)
563 #define SECTION_HAS_MEM_MAP (1UL<<1)
564 #define SECTION_MAP_LAST_BIT (1UL<<2)
565 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
566 #define SECTION_NID_SHIFT 2
568 static inline struct page *__section_mem_map_addr(struct mem_section *section)
570 unsigned long map = section->section_mem_map;
571 map &= SECTION_MAP_MASK;
572 return (struct page *)map;
575 static inline int valid_section(struct mem_section *section)
577 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
580 static inline int section_has_mem_map(struct mem_section *section)
582 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
585 static inline int valid_section_nr(unsigned long nr)
587 return valid_section(__nr_to_section(nr));
590 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
592 return __nr_to_section(pfn_to_section_nr(pfn));
595 static inline int pfn_valid(unsigned long pfn)
597 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
599 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
603 * These are _only_ used during initialisation, therefore they
604 * can use __initdata ... They could have names to indicate
608 #define pfn_to_nid(pfn) \
610 unsigned long __pfn_to_nid_pfn = (pfn); \
611 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
614 #define pfn_to_nid(pfn) (0)
617 #define early_pfn_valid(pfn) pfn_valid(pfn)
618 void sparse_init(void);
620 #define sparse_init() do {} while (0)
621 #define sparse_index_init(_sec, _nid) do {} while (0)
622 #endif /* CONFIG_SPARSEMEM */
624 #ifndef early_pfn_valid
625 #define early_pfn_valid(pfn) (1)
628 void memory_present(int nid, unsigned long start, unsigned long end);
629 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
631 #endif /* !__ASSEMBLY__ */
632 #endif /* __KERNEL__ */
633 #endif /* _LINUX_MMZONE_H */