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
];
173 * zone reclaim becomes active if more unmapped pages exist.
175 unsigned long min_unmapped_pages
;
176 unsigned long min_slab_pages
;
177 struct per_cpu_pageset
*pageset
[NR_CPUS
];
179 struct per_cpu_pageset pageset
[NR_CPUS
];
182 * free areas of different sizes
185 #ifdef CONFIG_MEMORY_HOTPLUG
186 /* see spanned/present_pages for more description */
187 seqlock_t span_seqlock
;
189 struct free_area free_area
[MAX_ORDER
];
194 /* Fields commonly accessed by the page reclaim scanner */
196 struct list_head active_list
;
197 struct list_head inactive_list
;
198 unsigned long nr_scan_active
;
199 unsigned long nr_scan_inactive
;
200 unsigned long nr_active
;
201 unsigned long nr_inactive
;
202 unsigned long pages_scanned
; /* since last reclaim */
203 int all_unreclaimable
; /* All pages pinned */
205 /* A count of how many reclaimers are scanning this zone */
206 atomic_t reclaim_in_progress
;
208 /* Zone statistics */
209 atomic_long_t vm_stat
[NR_VM_ZONE_STAT_ITEMS
];
212 * prev_priority holds the scanning priority for this zone. It is
213 * defined as the scanning priority at which we achieved our reclaim
214 * target at the previous try_to_free_pages() or balance_pgdat()
217 * We use prev_priority as a measure of how much stress page reclaim is
218 * under - it drives the swappiness decision: whether to unmap mapped
221 * Access to both this field is quite racy even on uniprocessor. But
222 * it is expected to average out OK.
228 /* Rarely used or read-mostly fields */
231 * wait_table -- the array holding the hash table
232 * wait_table_hash_nr_entries -- the size of the hash table array
233 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
235 * The purpose of all these is to keep track of the people
236 * waiting for a page to become available and make them
237 * runnable again when possible. The trouble is that this
238 * consumes a lot of space, especially when so few things
239 * wait on pages at a given time. So instead of using
240 * per-page waitqueues, we use a waitqueue hash table.
242 * The bucket discipline is to sleep on the same queue when
243 * colliding and wake all in that wait queue when removing.
244 * When something wakes, it must check to be sure its page is
245 * truly available, a la thundering herd. The cost of a
246 * collision is great, but given the expected load of the
247 * table, they should be so rare as to be outweighed by the
248 * benefits from the saved space.
250 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
251 * primary users of these fields, and in mm/page_alloc.c
252 * free_area_init_core() performs the initialization of them.
254 wait_queue_head_t
* wait_table
;
255 unsigned long wait_table_hash_nr_entries
;
256 unsigned long wait_table_bits
;
259 * Discontig memory support fields.
261 struct pglist_data
*zone_pgdat
;
262 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
263 unsigned long zone_start_pfn
;
266 * zone_start_pfn, spanned_pages and present_pages are all
267 * protected by span_seqlock. It is a seqlock because it has
268 * to be read outside of zone->lock, and it is done in the main
269 * allocator path. But, it is written quite infrequently.
271 * The lock is declared along with zone->lock because it is
272 * frequently read in proximity to zone->lock. It's good to
273 * give them a chance of being in the same cacheline.
275 unsigned long spanned_pages
; /* total size, including holes */
276 unsigned long present_pages
; /* amount of memory (excluding holes) */
279 * rarely used fields:
282 } ____cacheline_internodealigned_in_smp
;
285 * The "priority" of VM scanning is how much of the queues we will scan in one
286 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
287 * queues ("queue_length >> 12") during an aging round.
289 #define DEF_PRIORITY 12
292 * One allocation request operates on a zonelist. A zonelist
293 * is a list of zones, the first one is the 'goal' of the
294 * allocation, the other zones are fallback zones, in decreasing
297 * Right now a zonelist takes up less than a cacheline. We never
298 * modify it apart from boot-up, and only a few indices are used,
299 * so despite the zonelist table being relatively big, the cache
300 * footprint of this construct is very small.
303 struct zone
*zones
[MAX_NUMNODES
* MAX_NR_ZONES
+ 1]; // NULL delimited
306 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
307 struct node_active_region
{
308 unsigned long start_pfn
;
309 unsigned long end_pfn
;
312 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
314 #ifndef CONFIG_DISCONTIGMEM
315 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
316 extern struct page
*mem_map
;
320 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
321 * (mostly NUMA machines?) to denote a higher-level memory zone than the
324 * On NUMA machines, each NUMA node would have a pg_data_t to describe
325 * it's memory layout.
327 * Memory statistics and page replacement data structures are maintained on a
331 typedef struct pglist_data
{
332 struct zone node_zones
[MAX_NR_ZONES
];
333 struct zonelist node_zonelists
[MAX_NR_ZONES
];
335 #ifdef CONFIG_FLAT_NODE_MEM_MAP
336 struct page
*node_mem_map
;
338 struct bootmem_data
*bdata
;
339 #ifdef CONFIG_MEMORY_HOTPLUG
341 * Must be held any time you expect node_start_pfn, node_present_pages
342 * or node_spanned_pages stay constant. Holding this will also
343 * guarantee that any pfn_valid() stays that way.
345 * Nests above zone->lock and zone->size_seqlock.
347 spinlock_t node_size_lock
;
349 unsigned long node_start_pfn
;
350 unsigned long node_present_pages
; /* total number of physical pages */
351 unsigned long node_spanned_pages
; /* total size of physical page
352 range, including holes */
354 wait_queue_head_t kswapd_wait
;
355 struct task_struct
*kswapd
;
356 int kswapd_max_order
;
359 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
360 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
361 #ifdef CONFIG_FLAT_NODE_MEM_MAP
362 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
364 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
366 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
368 #include <linux/memory_hotplug.h>
370 void __get_zone_counts(unsigned long *active
, unsigned long *inactive
,
371 unsigned long *free
, struct pglist_data
*pgdat
);
372 void get_zone_counts(unsigned long *active
, unsigned long *inactive
,
373 unsigned long *free
);
374 void build_all_zonelists(void);
375 void wakeup_kswapd(struct zone
*zone
, int order
);
376 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
377 int classzone_idx
, int alloc_flags
);
379 extern int init_currently_empty_zone(struct zone
*zone
, unsigned long start_pfn
,
382 #ifdef CONFIG_HAVE_MEMORY_PRESENT
383 void memory_present(int nid
, unsigned long start
, unsigned long end
);
385 static inline void memory_present(int nid
, unsigned long start
, unsigned long end
) {}
388 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
389 unsigned long __init
node_memmap_size_bytes(int, unsigned long, unsigned long);
393 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
395 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
397 static inline int populated_zone(struct zone
*zone
)
399 return (!!zone
->present_pages
);
402 static inline int is_highmem_idx(enum zone_type idx
)
404 #ifdef CONFIG_HIGHMEM
405 return (idx
== ZONE_HIGHMEM
);
411 static inline int is_normal_idx(enum zone_type idx
)
413 return (idx
== ZONE_NORMAL
);
417 * is_highmem - helper function to quickly check if a struct zone is a
418 * highmem zone or not. This is an attempt to keep references
419 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
420 * @zone - pointer to struct zone variable
422 static inline int is_highmem(struct zone
*zone
)
424 #ifdef CONFIG_HIGHMEM
425 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
*);
462 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table
*, int,
463 struct file
*, void __user
*, size_t *, loff_t
*);
465 #include <linux/topology.h>
466 /* Returns the number of the current Node. */
468 #define numa_node_id() (cpu_to_node(raw_smp_processor_id()))
471 #ifndef CONFIG_NEED_MULTIPLE_NODES
473 extern struct pglist_data contig_page_data
;
474 #define NODE_DATA(nid) (&contig_page_data)
475 #define NODE_MEM_MAP(nid) mem_map
476 #define MAX_NODES_SHIFT 1
478 #else /* CONFIG_NEED_MULTIPLE_NODES */
480 #include <asm/mmzone.h>
482 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
484 extern struct pglist_data
*first_online_pgdat(void);
485 extern struct pglist_data
*next_online_pgdat(struct pglist_data
*pgdat
);
486 extern struct zone
*next_zone(struct zone
*zone
);
489 * for_each_pgdat - helper macro to iterate over all nodes
490 * @pgdat - pointer to a pg_data_t variable
492 #define for_each_online_pgdat(pgdat) \
493 for (pgdat = first_online_pgdat(); \
495 pgdat = next_online_pgdat(pgdat))
497 * for_each_zone - helper macro to iterate over all memory zones
498 * @zone - pointer to struct zone variable
500 * The user only needs to declare the zone variable, for_each_zone
503 #define for_each_zone(zone) \
504 for (zone = (first_online_pgdat())->node_zones; \
506 zone = next_zone(zone))
508 #ifdef CONFIG_SPARSEMEM
509 #include <asm/sparsemem.h>
512 #if BITS_PER_LONG == 32
514 * with 32 bit page->flags field, we reserve 9 bits for node/zone info.
515 * there are 4 zones (3 bits) and this leaves 9-3=6 bits for nodes.
517 #define FLAGS_RESERVED 9
519 #elif BITS_PER_LONG == 64
521 * with 64 bit flags field, there's plenty of room.
523 #define FLAGS_RESERVED 32
527 #error BITS_PER_LONG not defined
531 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
532 !defined(CONFIG_ARCH_POPULATES_NODE_MAP)
533 #define early_pfn_to_nid(nid) (0UL)
536 #ifdef CONFIG_FLATMEM
537 #define pfn_to_nid(pfn) (0)
540 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
541 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
543 #ifdef CONFIG_SPARSEMEM
546 * SECTION_SHIFT #bits space required to store a section #
548 * PA_SECTION_SHIFT physical address to/from section number
549 * PFN_SECTION_SHIFT pfn to/from section number
551 #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
553 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
554 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
556 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
558 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
559 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
561 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
562 #error Allocator MAX_ORDER exceeds SECTION_SIZE
568 * This is, logically, a pointer to an array of struct
569 * pages. However, it is stored with some other magic.
570 * (see sparse.c::sparse_init_one_section())
572 * Additionally during early boot we encode node id of
573 * the location of the section here to guide allocation.
574 * (see sparse.c::memory_present())
576 * Making it a UL at least makes someone do a cast
577 * before using it wrong.
579 unsigned long section_mem_map
;
582 #ifdef CONFIG_SPARSEMEM_EXTREME
583 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
585 #define SECTIONS_PER_ROOT 1
588 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
589 #define NR_SECTION_ROOTS (NR_MEM_SECTIONS / SECTIONS_PER_ROOT)
590 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
592 #ifdef CONFIG_SPARSEMEM_EXTREME
593 extern struct mem_section
*mem_section
[NR_SECTION_ROOTS
];
595 extern struct mem_section mem_section
[NR_SECTION_ROOTS
][SECTIONS_PER_ROOT
];
598 static inline struct mem_section
*__nr_to_section(unsigned long nr
)
600 if (!mem_section
[SECTION_NR_TO_ROOT(nr
)])
602 return &mem_section
[SECTION_NR_TO_ROOT(nr
)][nr
& SECTION_ROOT_MASK
];
604 extern int __section_nr(struct mem_section
* ms
);
607 * We use the lower bits of the mem_map pointer to store
608 * a little bit of information. There should be at least
609 * 3 bits here due to 32-bit alignment.
611 #define SECTION_MARKED_PRESENT (1UL<<0)
612 #define SECTION_HAS_MEM_MAP (1UL<<1)
613 #define SECTION_MAP_LAST_BIT (1UL<<2)
614 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
615 #define SECTION_NID_SHIFT 2
617 static inline struct page
*__section_mem_map_addr(struct mem_section
*section
)
619 unsigned long map
= section
->section_mem_map
;
620 map
&= SECTION_MAP_MASK
;
621 return (struct page
*)map
;
624 static inline int valid_section(struct mem_section
*section
)
626 return (section
&& (section
->section_mem_map
& SECTION_MARKED_PRESENT
));
629 static inline int section_has_mem_map(struct mem_section
*section
)
631 return (section
&& (section
->section_mem_map
& SECTION_HAS_MEM_MAP
));
634 static inline int valid_section_nr(unsigned long nr
)
636 return valid_section(__nr_to_section(nr
));
639 static inline struct mem_section
*__pfn_to_section(unsigned long pfn
)
641 return __nr_to_section(pfn_to_section_nr(pfn
));
644 static inline int pfn_valid(unsigned long pfn
)
646 if (pfn_to_section_nr(pfn
) >= NR_MEM_SECTIONS
)
648 return valid_section(__nr_to_section(pfn_to_section_nr(pfn
)));
652 * These are _only_ used during initialisation, therefore they
653 * can use __initdata ... They could have names to indicate
657 #define pfn_to_nid(pfn) \
659 unsigned long __pfn_to_nid_pfn = (pfn); \
660 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
663 #define pfn_to_nid(pfn) (0)
666 #define early_pfn_valid(pfn) pfn_valid(pfn)
667 void sparse_init(void);
669 #define sparse_init() do {} while (0)
670 #define sparse_index_init(_sec, _nid) do {} while (0)
671 #endif /* CONFIG_SPARSEMEM */
673 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
674 #define early_pfn_in_nid(pfn, nid) (early_pfn_to_nid(pfn) == (nid))
676 #define early_pfn_in_nid(pfn, nid) (1)
679 #ifndef early_pfn_valid
680 #define early_pfn_valid(pfn) (1)
683 void memory_present(int nid
, unsigned long start
, unsigned long end
);
684 unsigned long __init
node_memmap_size_bytes(int, unsigned long, unsigned long);
686 #endif /* !__ASSEMBLY__ */
687 #endif /* __KERNEL__ */
688 #endif /* _LINUX_MMZONE_H */