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.
127 * A memory area that is only addressable by the kernel through
128 * mapping portions into its own address space. This is for example
129 * used by i386 to allow the kernel to address the memory beyond
130 * 900MB. The kernel will set up special mappings (page
131 * table entries on i386) for each page that the kernel needs to
139 #define ZONES_SHIFT 2 /* ceil(log2(MAX_NR_ZONES)) */
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>
166 #define GFP_ZONETYPES ((GFP_ZONEMASK + 1) / 2 + 1) /* Loner */
168 #ifdef CONFIG_ZONE_DMA32
169 #define GFP_ZONEMASK 0x07
171 #define GFP_ZONEMASK 0x03
175 /* Fields commonly accessed by the page allocator */
176 unsigned long free_pages;
177 unsigned long pages_min, pages_low, pages_high;
179 * We don't know if the memory that we're going to allocate will be freeable
180 * or/and it will be released eventually, so to avoid totally wasting several
181 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
182 * to run OOM on the lower zones despite there's tons of freeable ram
183 * on the higher zones). This array is recalculated at runtime if the
184 * sysctl_lowmem_reserve_ratio sysctl changes.
186 unsigned long lowmem_reserve[MAX_NR_ZONES];
190 * zone reclaim becomes active if more unmapped pages exist.
192 unsigned long min_unmapped_ratio;
193 struct per_cpu_pageset *pageset[NR_CPUS];
195 struct per_cpu_pageset pageset[NR_CPUS];
198 * free areas of different sizes
201 #ifdef CONFIG_MEMORY_HOTPLUG
202 /* see spanned/present_pages for more description */
203 seqlock_t span_seqlock;
205 struct free_area free_area[MAX_ORDER];
210 /* Fields commonly accessed by the page reclaim scanner */
212 struct list_head active_list;
213 struct list_head inactive_list;
214 unsigned long nr_scan_active;
215 unsigned long nr_scan_inactive;
216 unsigned long nr_active;
217 unsigned long nr_inactive;
218 unsigned long pages_scanned; /* since last reclaim */
219 int all_unreclaimable; /* All pages pinned */
221 /* A count of how many reclaimers are scanning this zone */
222 atomic_t reclaim_in_progress;
224 /* Zone statistics */
225 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
228 * prev_priority holds the scanning priority for this zone. It is
229 * defined as the scanning priority at which we achieved our reclaim
230 * target at the previous try_to_free_pages() or balance_pgdat()
233 * We use prev_priority as a measure of how much stress page reclaim is
234 * under - it drives the swappiness decision: whether to unmap mapped
237 * temp_priority is used to remember the scanning priority at which
238 * this zone was successfully refilled to free_pages == pages_high.
240 * Access to both these fields is quite racy even on uniprocessor. But
241 * it is expected to average out OK.
248 /* Rarely used or read-mostly fields */
251 * wait_table -- the array holding the hash table
252 * wait_table_hash_nr_entries -- the size of the hash table array
253 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
255 * The purpose of all these is to keep track of the people
256 * waiting for a page to become available and make them
257 * runnable again when possible. The trouble is that this
258 * consumes a lot of space, especially when so few things
259 * wait on pages at a given time. So instead of using
260 * per-page waitqueues, we use a waitqueue hash table.
262 * The bucket discipline is to sleep on the same queue when
263 * colliding and wake all in that wait queue when removing.
264 * When something wakes, it must check to be sure its page is
265 * truly available, a la thundering herd. The cost of a
266 * collision is great, but given the expected load of the
267 * table, they should be so rare as to be outweighed by the
268 * benefits from the saved space.
270 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
271 * primary users of these fields, and in mm/page_alloc.c
272 * free_area_init_core() performs the initialization of them.
274 wait_queue_head_t * wait_table;
275 unsigned long wait_table_hash_nr_entries;
276 unsigned long wait_table_bits;
279 * Discontig memory support fields.
281 struct pglist_data *zone_pgdat;
282 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
283 unsigned long zone_start_pfn;
286 * zone_start_pfn, spanned_pages and present_pages are all
287 * protected by span_seqlock. It is a seqlock because it has
288 * to be read outside of zone->lock, and it is done in the main
289 * allocator path. But, it is written quite infrequently.
291 * The lock is declared along with zone->lock because it is
292 * frequently read in proximity to zone->lock. It's good to
293 * give them a chance of being in the same cacheline.
295 unsigned long spanned_pages; /* total size, including holes */
296 unsigned long present_pages; /* amount of memory (excluding holes) */
299 * rarely used fields:
302 } ____cacheline_internodealigned_in_smp;
305 * The "priority" of VM scanning is how much of the queues we will scan in one
306 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
307 * queues ("queue_length >> 12") during an aging round.
309 #define DEF_PRIORITY 12
312 * One allocation request operates on a zonelist. A zonelist
313 * is a list of zones, the first one is the 'goal' of the
314 * allocation, the other zones are fallback zones, in decreasing
317 * Right now a zonelist takes up less than a cacheline. We never
318 * modify it apart from boot-up, and only a few indices are used,
319 * so despite the zonelist table being relatively big, the cache
320 * footprint of this construct is very small.
323 struct zone *zones[MAX_NUMNODES * MAX_NR_ZONES + 1]; // NULL delimited
328 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
329 * (mostly NUMA machines?) to denote a higher-level memory zone than the
332 * On NUMA machines, each NUMA node would have a pg_data_t to describe
333 * it's memory layout.
335 * Memory statistics and page replacement data structures are maintained on a
339 typedef struct pglist_data {
340 struct zone node_zones[MAX_NR_ZONES];
341 struct zonelist node_zonelists[GFP_ZONETYPES];
343 #ifdef CONFIG_FLAT_NODE_MEM_MAP
344 struct page *node_mem_map;
346 struct bootmem_data *bdata;
347 #ifdef CONFIG_MEMORY_HOTPLUG
349 * Must be held any time you expect node_start_pfn, node_present_pages
350 * or node_spanned_pages stay constant. Holding this will also
351 * guarantee that any pfn_valid() stays that way.
353 * Nests above zone->lock and zone->size_seqlock.
355 spinlock_t node_size_lock;
357 unsigned long node_start_pfn;
358 unsigned long node_present_pages; /* total number of physical pages */
359 unsigned long node_spanned_pages; /* total size of physical page
360 range, including holes */
362 wait_queue_head_t kswapd_wait;
363 struct task_struct *kswapd;
364 int kswapd_max_order;
367 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
368 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
369 #ifdef CONFIG_FLAT_NODE_MEM_MAP
370 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
372 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
374 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
376 #include <linux/memory_hotplug.h>
378 void __get_zone_counts(unsigned long *active, unsigned long *inactive,
379 unsigned long *free, struct pglist_data *pgdat);
380 void get_zone_counts(unsigned long *active, unsigned long *inactive,
381 unsigned long *free);
382 void build_all_zonelists(void);
383 void wakeup_kswapd(struct zone *zone, int order);
384 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
385 int classzone_idx, int alloc_flags);
387 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
390 #ifdef CONFIG_HAVE_MEMORY_PRESENT
391 void memory_present(int nid, unsigned long start, unsigned long end);
393 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
396 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
397 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
401 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
403 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
405 static inline int populated_zone(struct zone *zone)
407 return (!!zone->present_pages);
410 static inline int is_highmem_idx(enum zone_type idx)
412 return (idx == ZONE_HIGHMEM);
415 static inline int is_normal_idx(enum zone_type idx)
417 return (idx == ZONE_NORMAL);
421 * is_highmem - helper function to quickly check if a struct zone is a
422 * highmem zone or not. This is an attempt to keep references
423 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
424 * @zone - pointer to struct zone variable
426 static inline int is_highmem(struct zone *zone)
428 return zone == zone->zone_pgdat->node_zones + ZONE_HIGHMEM;
431 static inline int is_normal(struct zone *zone)
433 return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
436 static inline int is_dma32(struct zone *zone)
438 #ifdef CONFIG_ZONE_DMA32
439 return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
445 static inline int is_dma(struct zone *zone)
447 return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
450 /* These two functions are used to setup the per zone pages min values */
453 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *,
454 void __user *, size_t *, loff_t *);
455 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
456 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *,
457 void __user *, size_t *, loff_t *);
458 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, struct file *,
459 void __user *, size_t *, loff_t *);
460 int sysctl_min_unmapped_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 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
530 #define early_pfn_to_nid(nid) (0UL)
533 #ifdef CONFIG_FLATMEM
534 #define pfn_to_nid(pfn) (0)
537 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
538 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
540 #ifdef CONFIG_SPARSEMEM
543 * SECTION_SHIFT #bits space required to store a section #
545 * PA_SECTION_SHIFT physical address to/from section number
546 * PFN_SECTION_SHIFT pfn to/from section number
548 #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
550 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
551 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
553 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
555 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
556 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
558 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
559 #error Allocator MAX_ORDER exceeds SECTION_SIZE
565 * This is, logically, a pointer to an array of struct
566 * pages. However, it is stored with some other magic.
567 * (see sparse.c::sparse_init_one_section())
569 * Additionally during early boot we encode node id of
570 * the location of the section here to guide allocation.
571 * (see sparse.c::memory_present())
573 * Making it a UL at least makes someone do a cast
574 * before using it wrong.
576 unsigned long section_mem_map;
579 #ifdef CONFIG_SPARSEMEM_EXTREME
580 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
582 #define SECTIONS_PER_ROOT 1
585 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
586 #define NR_SECTION_ROOTS (NR_MEM_SECTIONS / SECTIONS_PER_ROOT)
587 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
589 #ifdef CONFIG_SPARSEMEM_EXTREME
590 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
592 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
595 static inline struct mem_section *__nr_to_section(unsigned long nr)
597 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
599 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
601 extern int __section_nr(struct mem_section* ms);
604 * We use the lower bits of the mem_map pointer to store
605 * a little bit of information. There should be at least
606 * 3 bits here due to 32-bit alignment.
608 #define SECTION_MARKED_PRESENT (1UL<<0)
609 #define SECTION_HAS_MEM_MAP (1UL<<1)
610 #define SECTION_MAP_LAST_BIT (1UL<<2)
611 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
612 #define SECTION_NID_SHIFT 2
614 static inline struct page *__section_mem_map_addr(struct mem_section *section)
616 unsigned long map = section->section_mem_map;
617 map &= SECTION_MAP_MASK;
618 return (struct page *)map;
621 static inline int valid_section(struct mem_section *section)
623 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
626 static inline int section_has_mem_map(struct mem_section *section)
628 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
631 static inline int valid_section_nr(unsigned long nr)
633 return valid_section(__nr_to_section(nr));
636 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
638 return __nr_to_section(pfn_to_section_nr(pfn));
641 static inline int pfn_valid(unsigned long pfn)
643 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
645 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
649 * These are _only_ used during initialisation, therefore they
650 * can use __initdata ... They could have names to indicate
654 #define pfn_to_nid(pfn) \
656 unsigned long __pfn_to_nid_pfn = (pfn); \
657 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
660 #define pfn_to_nid(pfn) (0)
663 #define early_pfn_valid(pfn) pfn_valid(pfn)
664 void sparse_init(void);
666 #define sparse_init() do {} while (0)
667 #define sparse_index_init(_sec, _nid) do {} while (0)
668 #endif /* CONFIG_SPARSEMEM */
670 #ifndef early_pfn_valid
671 #define early_pfn_valid(pfn) (1)
674 void memory_present(int nid, unsigned long start, unsigned long end);
675 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
677 #endif /* !__ASSEMBLY__ */
678 #endif /* __KERNEL__ */
679 #endif /* _LINUX_MMZONE_H */