1 #ifndef _LINUX_MMZONE_H
2 #define _LINUX_MMZONE_H
7 #include <linux/config.h>
8 #include <linux/spinlock.h>
9 #include <linux/list.h>
10 #include <linux/wait.h>
11 #include <linux/cache.h>
12 #include <linux/threads.h>
13 #include <linux/numa.h>
14 #include <linux/init.h>
15 #include <linux/seqlock.h>
16 #include <asm/atomic.h>
18 /* Free memory management - zoned buddy allocator. */
19 #ifndef CONFIG_FORCE_MAX_ZONEORDER
22 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
26 struct list_head free_list
;
27 unsigned long nr_free
;
33 * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
34 * So add a wild amount of padding here to ensure that they fall into separate
35 * cachelines. There are very few zone structures in the machine, so space
36 * consumption is not a concern here.
38 #if defined(CONFIG_SMP)
41 } ____cacheline_internodealigned_in_smp
;
42 #define ZONE_PADDING(name) struct zone_padding name;
44 #define ZONE_PADDING(name)
47 struct per_cpu_pages
{
48 int count
; /* number of pages in the list */
49 int high
; /* high watermark, emptying needed */
50 int batch
; /* chunk size for buddy add/remove */
51 struct list_head list
; /* the list of pages */
54 struct per_cpu_pageset
{
55 struct per_cpu_pages pcp
[2]; /* 0: hot. 1: cold */
57 unsigned long numa_hit
; /* allocated in intended node */
58 unsigned long numa_miss
; /* allocated in non intended node */
59 unsigned long numa_foreign
; /* was intended here, hit elsewhere */
60 unsigned long interleave_hit
; /* interleaver prefered this zone */
61 unsigned long local_node
; /* allocation from local node */
62 unsigned long other_node
; /* allocation from other node */
64 } ____cacheline_aligned_in_smp
;
67 #define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)])
69 #define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)])
75 #define ZONE_HIGHMEM 3
77 #define MAX_NR_ZONES 4 /* Sync this with ZONES_SHIFT */
78 #define ZONES_SHIFT 2 /* ceil(log2(MAX_NR_ZONES)) */
82 * When a memory allocation must conform to specific limitations (such
83 * as being suitable for DMA) the caller will pass in hints to the
84 * allocator in the gfp_mask, in the zone modifier bits. These bits
85 * are used to select a priority ordered list of memory zones which
86 * match the requested limits. GFP_ZONEMASK defines which bits within
87 * the gfp_mask should be considered as zone modifiers. Each valid
88 * combination of the zone modifier bits has a corresponding list
89 * of zones (in node_zonelists). Thus for two zone modifiers there
90 * will be a maximum of 4 (2 ** 2) zonelists, for 3 modifiers there will
91 * be 8 (2 ** 3) zonelists. GFP_ZONETYPES defines the number of possible
92 * combinations of zone modifiers in "zone modifier space".
94 * NOTE! Make sure this matches the zones in <linux/gfp.h>
96 #define GFP_ZONEMASK 0x07
97 #define GFP_ZONETYPES 5
100 * On machines where it is needed (eg PCs) we divide physical memory
101 * into multiple physical zones. On a 32bit PC we have 4 zones:
103 * ZONE_DMA < 16 MB ISA DMA capable memory
104 * ZONE_DMA32 0 MB Empty
105 * ZONE_NORMAL 16-896 MB direct mapped by the kernel
106 * ZONE_HIGHMEM > 896 MB only page cache and user processes
110 /* Fields commonly accessed by the page allocator */
111 unsigned long free_pages
;
112 unsigned long pages_min
, pages_low
, pages_high
;
114 * We don't know if the memory that we're going to allocate will be freeable
115 * or/and it will be released eventually, so to avoid totally wasting several
116 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
117 * to run OOM on the lower zones despite there's tons of freeable ram
118 * on the higher zones). This array is recalculated at runtime if the
119 * sysctl_lowmem_reserve_ratio sysctl changes.
121 unsigned long lowmem_reserve
[MAX_NR_ZONES
];
124 struct per_cpu_pageset
*pageset
[NR_CPUS
];
126 struct per_cpu_pageset pageset
[NR_CPUS
];
129 * free areas of different sizes
132 #ifdef CONFIG_MEMORY_HOTPLUG
133 /* see spanned/present_pages for more description */
134 seqlock_t span_seqlock
;
136 struct free_area free_area
[MAX_ORDER
];
141 /* Fields commonly accessed by the page reclaim scanner */
143 struct list_head active_list
;
144 struct list_head inactive_list
;
145 unsigned long nr_scan_active
;
146 unsigned long nr_scan_inactive
;
147 unsigned long nr_active
;
148 unsigned long nr_inactive
;
149 unsigned long pages_scanned
; /* since last reclaim */
150 int all_unreclaimable
; /* All pages pinned */
152 /* A count of how many reclaimers are scanning this zone */
153 atomic_t reclaim_in_progress
;
156 * timestamp (in jiffies) of the last zone reclaim that did not
157 * result in freeing of pages. This is used to avoid repeated scans
158 * if all memory in the zone is in use.
160 unsigned long last_unsuccessful_zone_reclaim
;
163 * prev_priority holds the scanning priority for this zone. It is
164 * defined as the scanning priority at which we achieved our reclaim
165 * target at the previous try_to_free_pages() or balance_pgdat()
168 * We use prev_priority as a measure of how much stress page reclaim is
169 * under - it drives the swappiness decision: whether to unmap mapped
172 * temp_priority is used to remember the scanning priority at which
173 * this zone was successfully refilled to free_pages == pages_high.
175 * Access to both these fields is quite racy even on uniprocessor. But
176 * it is expected to average out OK.
183 /* Rarely used or read-mostly fields */
186 * wait_table -- the array holding the hash table
187 * wait_table_size -- the size of the hash table array
188 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
190 * The purpose of all these is to keep track of the people
191 * waiting for a page to become available and make them
192 * runnable again when possible. The trouble is that this
193 * consumes a lot of space, especially when so few things
194 * wait on pages at a given time. So instead of using
195 * per-page waitqueues, we use a waitqueue hash table.
197 * The bucket discipline is to sleep on the same queue when
198 * colliding and wake all in that wait queue when removing.
199 * When something wakes, it must check to be sure its page is
200 * truly available, a la thundering herd. The cost of a
201 * collision is great, but given the expected load of the
202 * table, they should be so rare as to be outweighed by the
203 * benefits from the saved space.
205 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
206 * primary users of these fields, and in mm/page_alloc.c
207 * free_area_init_core() performs the initialization of them.
209 wait_queue_head_t
* wait_table
;
210 unsigned long wait_table_size
;
211 unsigned long wait_table_bits
;
214 * Discontig memory support fields.
216 struct pglist_data
*zone_pgdat
;
217 struct page
*zone_mem_map
;
218 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
219 unsigned long zone_start_pfn
;
222 * zone_start_pfn, spanned_pages and present_pages are all
223 * protected by span_seqlock. It is a seqlock because it has
224 * to be read outside of zone->lock, and it is done in the main
225 * allocator path. But, it is written quite infrequently.
227 * The lock is declared along with zone->lock because it is
228 * frequently read in proximity to zone->lock. It's good to
229 * give them a chance of being in the same cacheline.
231 unsigned long spanned_pages
; /* total size, including holes */
232 unsigned long present_pages
; /* amount of memory (excluding holes) */
235 * rarely used fields:
238 } ____cacheline_internodealigned_in_smp
;
242 * The "priority" of VM scanning is how much of the queues we will scan in one
243 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
244 * queues ("queue_length >> 12") during an aging round.
246 #define DEF_PRIORITY 12
249 * One allocation request operates on a zonelist. A zonelist
250 * is a list of zones, the first one is the 'goal' of the
251 * allocation, the other zones are fallback zones, in decreasing
254 * Right now a zonelist takes up less than a cacheline. We never
255 * modify it apart from boot-up, and only a few indices are used,
256 * so despite the zonelist table being relatively big, the cache
257 * footprint of this construct is very small.
260 struct zone
*zones
[MAX_NUMNODES
* MAX_NR_ZONES
+ 1]; // NULL delimited
265 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
266 * (mostly NUMA machines?) to denote a higher-level memory zone than the
269 * On NUMA machines, each NUMA node would have a pg_data_t to describe
270 * it's memory layout.
272 * Memory statistics and page replacement data structures are maintained on a
276 typedef struct pglist_data
{
277 struct zone node_zones
[MAX_NR_ZONES
];
278 struct zonelist node_zonelists
[GFP_ZONETYPES
];
280 #ifdef CONFIG_FLAT_NODE_MEM_MAP
281 struct page
*node_mem_map
;
283 struct bootmem_data
*bdata
;
284 #ifdef CONFIG_MEMORY_HOTPLUG
286 * Must be held any time you expect node_start_pfn, node_present_pages
287 * or node_spanned_pages stay constant. Holding this will also
288 * guarantee that any pfn_valid() stays that way.
290 * Nests above zone->lock and zone->size_seqlock.
292 spinlock_t node_size_lock
;
294 unsigned long node_start_pfn
;
295 unsigned long node_present_pages
; /* total number of physical pages */
296 unsigned long node_spanned_pages
; /* total size of physical page
297 range, including holes */
299 struct pglist_data
*pgdat_next
;
300 wait_queue_head_t kswapd_wait
;
301 struct task_struct
*kswapd
;
302 int kswapd_max_order
;
305 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
306 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
307 #ifdef CONFIG_FLAT_NODE_MEM_MAP
308 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
310 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
312 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
314 #include <linux/memory_hotplug.h>
316 extern struct pglist_data
*pgdat_list
;
318 void __get_zone_counts(unsigned long *active
, unsigned long *inactive
,
319 unsigned long *free
, struct pglist_data
*pgdat
);
320 void get_zone_counts(unsigned long *active
, unsigned long *inactive
,
321 unsigned long *free
);
322 void build_all_zonelists(void);
323 void wakeup_kswapd(struct zone
*zone
, int order
);
324 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
325 int classzone_idx
, int alloc_flags
);
327 #ifdef CONFIG_HAVE_MEMORY_PRESENT
328 void memory_present(int nid
, unsigned long start
, unsigned long end
);
330 static inline void memory_present(int nid
, unsigned long start
, unsigned long end
) {}
333 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
334 unsigned long __init
node_memmap_size_bytes(int, unsigned long, unsigned long);
338 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
340 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
343 * for_each_pgdat - helper macro to iterate over all nodes
344 * @pgdat - pointer to a pg_data_t variable
346 * Meant to help with common loops of the form
347 * pgdat = pgdat_list;
350 * pgdat = pgdat->pgdat_next;
353 #define for_each_pgdat(pgdat) \
354 for (pgdat = pgdat_list; pgdat; pgdat = pgdat->pgdat_next)
357 * next_zone - helper magic for for_each_zone()
358 * Thanks to William Lee Irwin III for this piece of ingenuity.
360 static inline struct zone
*next_zone(struct zone
*zone
)
362 pg_data_t
*pgdat
= zone
->zone_pgdat
;
364 if (zone
< pgdat
->node_zones
+ MAX_NR_ZONES
- 1)
366 else if (pgdat
->pgdat_next
) {
367 pgdat
= pgdat
->pgdat_next
;
368 zone
= pgdat
->node_zones
;
376 * for_each_zone - helper macro to iterate over all memory zones
377 * @zone - pointer to struct zone variable
379 * The user only needs to declare the zone variable, for_each_zone
380 * fills it in. This basically means for_each_zone() is an
381 * easier to read version of this piece of code:
383 * for (pgdat = pgdat_list; pgdat; pgdat = pgdat->node_next)
384 * for (i = 0; i < MAX_NR_ZONES; ++i) {
385 * struct zone * z = pgdat->node_zones + i;
390 #define for_each_zone(zone) \
391 for (zone = pgdat_list->node_zones; zone; zone = next_zone(zone))
393 static inline int populated_zone(struct zone
*zone
)
395 return (!!zone
->present_pages
);
398 static inline int is_highmem_idx(int idx
)
400 return (idx
== ZONE_HIGHMEM
);
403 static inline int is_normal_idx(int idx
)
405 return (idx
== ZONE_NORMAL
);
409 * is_highmem - helper function to quickly check if a struct zone is a
410 * highmem zone or not. This is an attempt to keep references
411 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
412 * @zone - pointer to struct zone variable
414 static inline int is_highmem(struct zone
*zone
)
416 return zone
== zone
->zone_pgdat
->node_zones
+ ZONE_HIGHMEM
;
419 static inline int is_normal(struct zone
*zone
)
421 return zone
== zone
->zone_pgdat
->node_zones
+ ZONE_NORMAL
;
424 static inline int is_dma32(struct zone
*zone
)
426 return zone
== zone
->zone_pgdat
->node_zones
+ ZONE_DMA32
;
429 static inline int is_dma(struct zone
*zone
)
431 return zone
== zone
->zone_pgdat
->node_zones
+ ZONE_DMA
;
434 /* These two functions are used to setup the per zone pages min values */
437 int min_free_kbytes_sysctl_handler(struct ctl_table
*, int, struct file
*,
438 void __user
*, size_t *, loff_t
*);
439 extern int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1];
440 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table
*, int, struct file
*,
441 void __user
*, size_t *, loff_t
*);
442 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table
*, int, struct file
*,
443 void __user
*, size_t *, loff_t
*);
445 #include <linux/topology.h>
446 /* Returns the number of the current Node. */
448 #define numa_node_id() (cpu_to_node(raw_smp_processor_id()))
451 #ifndef CONFIG_NEED_MULTIPLE_NODES
453 extern struct pglist_data contig_page_data
;
454 #define NODE_DATA(nid) (&contig_page_data)
455 #define NODE_MEM_MAP(nid) mem_map
456 #define MAX_NODES_SHIFT 1
458 #else /* CONFIG_NEED_MULTIPLE_NODES */
460 #include <asm/mmzone.h>
462 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
464 #ifdef CONFIG_SPARSEMEM
465 #include <asm/sparsemem.h>
468 #if BITS_PER_LONG == 32
470 * with 32 bit page->flags field, we reserve 9 bits for node/zone info.
471 * there are 4 zones (3 bits) and this leaves 9-3=6 bits for nodes.
473 #define FLAGS_RESERVED 9
475 #elif BITS_PER_LONG == 64
477 * with 64 bit flags field, there's plenty of room.
479 #define FLAGS_RESERVED 32
483 #error BITS_PER_LONG not defined
487 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
488 #define early_pfn_to_nid(nid) (0UL)
491 #ifdef CONFIG_FLATMEM
492 #define pfn_to_nid(pfn) (0)
495 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
496 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
498 #ifdef CONFIG_SPARSEMEM
501 * SECTION_SHIFT #bits space required to store a section #
503 * PA_SECTION_SHIFT physical address to/from section number
504 * PFN_SECTION_SHIFT pfn to/from section number
506 #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
508 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
509 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
511 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
513 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
514 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
516 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
517 #error Allocator MAX_ORDER exceeds SECTION_SIZE
523 * This is, logically, a pointer to an array of struct
524 * pages. However, it is stored with some other magic.
525 * (see sparse.c::sparse_init_one_section())
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))
567 static inline struct page
*__section_mem_map_addr(struct mem_section
*section
)
569 unsigned long map
= section
->section_mem_map
;
570 map
&= SECTION_MAP_MASK
;
571 return (struct page
*)map
;
574 static inline int valid_section(struct mem_section
*section
)
576 return (section
&& (section
->section_mem_map
& SECTION_MARKED_PRESENT
));
579 static inline int section_has_mem_map(struct mem_section
*section
)
581 return (section
&& (section
->section_mem_map
& SECTION_HAS_MEM_MAP
));
584 static inline int valid_section_nr(unsigned long nr
)
586 return valid_section(__nr_to_section(nr
));
589 static inline struct mem_section
*__pfn_to_section(unsigned long pfn
)
591 return __nr_to_section(pfn_to_section_nr(pfn
));
594 #define pfn_to_page(pfn) \
596 unsigned long __pfn = (pfn); \
597 __section_mem_map_addr(__pfn_to_section(__pfn)) + __pfn; \
599 #define page_to_pfn(page) \
601 page - __section_mem_map_addr(__nr_to_section( \
602 page_to_section(page))); \
605 static inline int pfn_valid(unsigned long pfn
)
607 if (pfn_to_section_nr(pfn
) >= NR_MEM_SECTIONS
)
609 return valid_section(__nr_to_section(pfn_to_section_nr(pfn
)));
613 * These are _only_ used during initialisation, therefore they
614 * can use __initdata ... They could have names to indicate
618 #define pfn_to_nid(pfn) \
620 unsigned long __pfn_to_nid_pfn = (pfn); \
621 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
624 #define pfn_to_nid(pfn) (0)
627 #define early_pfn_valid(pfn) pfn_valid(pfn)
628 void sparse_init(void);
630 #define sparse_init() do {} while (0)
631 #define sparse_index_init(_sec, _nid) do {} while (0)
632 #endif /* CONFIG_SPARSEMEM */
634 #ifndef early_pfn_valid
635 #define early_pfn_valid(pfn) (1)
638 void memory_present(int nid
, unsigned long start
, unsigned long end
);
639 unsigned long __init
node_memmap_size_bytes(int, unsigned long, unsigned long);
641 #endif /* !__ASSEMBLY__ */
642 #endif /* __KERNEL__ */
643 #endif /* _LINUX_MMZONE_H */