Line data Source code
1 : #ifndef _LINUX_MMZONE_H
2 : #define _LINUX_MMZONE_H
3 :
4 : #ifndef __ASSEMBLY__
5 : #ifndef __GENERATING_BOUNDS_H
6 :
7 : #include <linux/spinlock.h>
8 : #include <linux/list.h>
9 : #include <linux/wait.h>
10 : #include <linux/bitops.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 <linux/nodemask.h>
17 : #include <linux/pageblock-flags.h>
18 : #include <generated/bounds.h>
19 : #include <asm/atomic.h>
20 : #include <asm/page.h>
21 :
22 : /* Free memory management - zoned buddy allocator. */
23 : #ifndef CONFIG_FORCE_MAX_ZONEORDER
24 : #define MAX_ORDER 11
25 : #else
26 : #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
27 : #endif
28 : #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
29 :
30 : /*
31 : * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
32 : * costly to service. That is between allocation orders which should
33 : * coelesce naturally under reasonable reclaim pressure and those which
34 : * will not.
35 : */
36 : #define PAGE_ALLOC_COSTLY_ORDER 3
37 :
38 : #define MIGRATE_UNMOVABLE 0
39 : #define MIGRATE_RECLAIMABLE 1
40 : #define MIGRATE_MOVABLE 2
41 : #define MIGRATE_PCPTYPES 3 /* the number of types on the pcp lists */
42 : #define MIGRATE_RESERVE 3
43 : #define MIGRATE_ISOLATE 4 /* can't allocate from here */
44 : #define MIGRATE_TYPES 5
45 :
46 : #define for_each_migratetype_order(order, type) \
47 : for (order = 0; order < MAX_ORDER; order++) \
48 : for (type = 0; type < MIGRATE_TYPES; type++)
49 :
50 : extern int page_group_by_mobility_disabled;
51 :
52 : static inline int get_pageblock_migratetype(struct page *page)
53 : {
54 : return get_pageblock_flags_group(page, PB_migrate, PB_migrate_end);
55 : }
56 :
57 : struct free_area {
58 : struct list_head free_list[MIGRATE_TYPES];
59 : unsigned long nr_free;
60 : };
61 :
62 : struct pglist_data;
63 :
64 : /*
65 : * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
66 : * So add a wild amount of padding here to ensure that they fall into separate
67 : * cachelines. There are very few zone structures in the machine, so space
68 : * consumption is not a concern here.
69 : */
70 : #if defined(CONFIG_SMP)
71 : struct zone_padding {
72 : char x[0];
73 : } ____cacheline_internodealigned_in_smp;
74 : #define ZONE_PADDING(name) struct zone_padding name;
75 : #else
76 : #define ZONE_PADDING(name)
77 : #endif
78 :
79 : enum zone_stat_item {
80 : /* First 128 byte cacheline (assuming 64 bit words) */
81 : NR_FREE_PAGES,
82 : NR_LRU_BASE,
83 : NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
84 : NR_ACTIVE_ANON, /* " " " " " */
85 : NR_INACTIVE_FILE, /* " " " " " */
86 : NR_ACTIVE_FILE, /* " " " " " */
87 : NR_UNEVICTABLE, /* " " " " " */
88 : NR_MLOCK, /* mlock()ed pages found and moved off LRU */
89 : NR_ANON_PAGES, /* Mapped anonymous pages */
90 : NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
91 : only modified from process context */
92 : NR_FILE_PAGES,
93 : NR_FILE_DIRTY,
94 : NR_WRITEBACK,
95 : NR_SLAB_RECLAIMABLE,
96 : NR_SLAB_UNRECLAIMABLE,
97 : NR_PAGETABLE, /* used for pagetables */
98 : NR_KERNEL_STACK,
99 : /* Second 128 byte cacheline */
100 : NR_UNSTABLE_NFS, /* NFS unstable pages */
101 : NR_BOUNCE,
102 : NR_VMSCAN_WRITE,
103 : NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
104 : NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
105 : NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
106 : NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
107 : #ifdef CONFIG_NUMA
108 : NUMA_HIT, /* allocated in intended node */
109 : NUMA_MISS, /* allocated in non intended node */
110 : NUMA_FOREIGN, /* was intended here, hit elsewhere */
111 : NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
112 : NUMA_LOCAL, /* allocation from local node */
113 : NUMA_OTHER, /* allocation from other node */
114 : #endif
115 : NR_VM_ZONE_STAT_ITEMS };
116 :
117 : /*
118 : * We do arithmetic on the LRU lists in various places in the code,
119 : * so it is important to keep the active lists LRU_ACTIVE higher in
120 : * the array than the corresponding inactive lists, and to keep
121 : * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
122 : *
123 : * This has to be kept in sync with the statistics in zone_stat_item
124 : * above and the descriptions in vmstat_text in mm/vmstat.c
125 : */
126 : #define LRU_BASE 0
127 : #define LRU_ACTIVE 1
128 : #define LRU_FILE 2
129 :
130 : enum lru_list {
131 : LRU_INACTIVE_ANON = LRU_BASE,
132 : LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
133 : LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
134 : LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
135 : LRU_UNEVICTABLE,
136 : NR_LRU_LISTS
137 : };
138 :
139 : #define for_each_lru(l) for (l = 0; l < NR_LRU_LISTS; l++)
140 :
141 : #define for_each_evictable_lru(l) for (l = 0; l <= LRU_ACTIVE_FILE; l++)
142 :
143 : static inline int is_file_lru(enum lru_list l)
144 : {
145 : return (l == LRU_INACTIVE_FILE || l == LRU_ACTIVE_FILE);
146 : }
147 :
148 : static inline int is_active_lru(enum lru_list l)
149 : {
150 : return (l == LRU_ACTIVE_ANON || l == LRU_ACTIVE_FILE);
151 : }
152 :
153 : static inline int is_unevictable_lru(enum lru_list l)
154 : {
155 : return (l == LRU_UNEVICTABLE);
156 : }
157 :
158 : enum zone_watermarks {
159 : WMARK_MIN,
160 : WMARK_LOW,
161 : WMARK_HIGH,
162 : NR_WMARK
163 : };
164 :
165 : #define min_wmark_pages(z) (z->watermark[WMARK_MIN])
166 : #define low_wmark_pages(z) (z->watermark[WMARK_LOW])
167 : #define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
168 :
169 : struct per_cpu_pages {
170 : int count; /* number of pages in the list */
171 : int high; /* high watermark, emptying needed */
172 : int batch; /* chunk size for buddy add/remove */
173 :
174 : /* Lists of pages, one per migrate type stored on the pcp-lists */
175 : struct list_head lists[MIGRATE_PCPTYPES];
176 : };
177 :
178 : struct per_cpu_pageset {
179 : struct per_cpu_pages pcp;
180 : #ifdef CONFIG_NUMA
181 : s8 expire;
182 : #endif
183 : #ifdef CONFIG_SMP
184 : s8 stat_threshold;
185 : s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
186 : #endif
187 : } ____cacheline_aligned_in_smp;
188 :
189 : #ifdef CONFIG_NUMA
190 : #define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)])
191 : #else
192 : #define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)])
193 : #endif
194 :
195 : #endif /* !__GENERATING_BOUNDS.H */
196 :
197 : enum zone_type {
198 : #ifdef CONFIG_ZONE_DMA
199 : /*
200 : * ZONE_DMA is used when there are devices that are not able
201 : * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
202 : * carve out the portion of memory that is needed for these devices.
203 : * The range is arch specific.
204 : *
205 : * Some examples
206 : *
207 : * Architecture Limit
208 : * ---------------------------
209 : * parisc, ia64, sparc <4G
210 : * s390 <2G
211 : * arm Various
212 : * alpha Unlimited or 0-16MB.
213 : *
214 : * i386, x86_64 and multiple other arches
215 : * <16M.
216 : */
217 : ZONE_DMA,
218 : #endif
219 : #ifdef CONFIG_ZONE_DMA32
220 : /*
221 : * x86_64 needs two ZONE_DMAs because it supports devices that are
222 : * only able to do DMA to the lower 16M but also 32 bit devices that
223 : * can only do DMA areas below 4G.
224 : */
225 : ZONE_DMA32,
226 : #endif
227 : /*
228 : * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
229 : * performed on pages in ZONE_NORMAL if the DMA devices support
230 : * transfers to all addressable memory.
231 : */
232 : ZONE_NORMAL,
233 : #ifdef CONFIG_HIGHMEM
234 : /*
235 : * A memory area that is only addressable by the kernel through
236 : * mapping portions into its own address space. This is for example
237 : * used by i386 to allow the kernel to address the memory beyond
238 : * 900MB. The kernel will set up special mappings (page
239 : * table entries on i386) for each page that the kernel needs to
240 : * access.
241 : */
242 : ZONE_HIGHMEM,
243 : #endif
244 : ZONE_MOVABLE,
245 : __MAX_NR_ZONES
246 : };
247 :
248 : #ifndef __GENERATING_BOUNDS_H
249 :
250 : /*
251 : * When a memory allocation must conform to specific limitations (such
252 : * as being suitable for DMA) the caller will pass in hints to the
253 : * allocator in the gfp_mask, in the zone modifier bits. These bits
254 : * are used to select a priority ordered list of memory zones which
255 : * match the requested limits. See gfp_zone() in include/linux/gfp.h
256 : */
257 :
258 : #if MAX_NR_ZONES < 2
259 : #define ZONES_SHIFT 0
260 : #elif MAX_NR_ZONES <= 2
261 : #define ZONES_SHIFT 1
262 : #elif MAX_NR_ZONES <= 4
263 : #define ZONES_SHIFT 2
264 : #else
265 : #error ZONES_SHIFT -- too many zones configured adjust calculation
266 : #endif
267 :
268 : struct zone_reclaim_stat {
269 : /*
270 : * The pageout code in vmscan.c keeps track of how many of the
271 : * mem/swap backed and file backed pages are refeferenced.
272 : * The higher the rotated/scanned ratio, the more valuable
273 : * that cache is.
274 : *
275 : * The anon LRU stats live in [0], file LRU stats in [1]
276 : */
277 : unsigned long recent_rotated[2];
278 : unsigned long recent_scanned[2];
279 :
280 : /*
281 : * accumulated for batching
282 : */
283 : unsigned long nr_saved_scan[NR_LRU_LISTS];
284 : };
285 :
286 : struct zone {
287 : /* Fields commonly accessed by the page allocator */
288 :
289 : /* zone watermarks, access with *_wmark_pages(zone) macros */
290 : unsigned long watermark[NR_WMARK];
291 :
292 : /*
293 : * When free pages are below this point, additional steps are taken
294 : * when reading the number of free pages to avoid per-cpu counter
295 : * drift allowing watermarks to be breached
296 : */
297 : unsigned long percpu_drift_mark;
298 :
299 : /*
300 : * We don't know if the memory that we're going to allocate will be freeable
301 : * or/and it will be released eventually, so to avoid totally wasting several
302 : * GB of ram we must reserve some of the lower zone memory (otherwise we risk
303 : * to run OOM on the lower zones despite there's tons of freeable ram
304 : * on the higher zones). This array is recalculated at runtime if the
305 : * sysctl_lowmem_reserve_ratio sysctl changes.
306 : */
307 : unsigned long lowmem_reserve[MAX_NR_ZONES];
308 :
309 : #ifdef CONFIG_NUMA
310 : int node;
311 : /*
312 : * zone reclaim becomes active if more unmapped pages exist.
313 : */
314 : unsigned long min_unmapped_pages;
315 : unsigned long min_slab_pages;
316 : struct per_cpu_pageset *pageset[NR_CPUS];
317 : #else
318 : struct per_cpu_pageset pageset[NR_CPUS];
319 : #endif
320 : /*
321 : * free areas of different sizes
322 : */
323 : spinlock_t lock;
324 : #ifdef CONFIG_MEMORY_HOTPLUG
325 : /* see spanned/present_pages for more description */
326 : seqlock_t span_seqlock;
327 : #endif
328 : struct free_area free_area[MAX_ORDER];
329 :
330 : #ifndef CONFIG_SPARSEMEM
331 : /*
332 : * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
333 : * In SPARSEMEM, this map is stored in struct mem_section
334 : */
335 : unsigned long *pageblock_flags;
336 : #endif /* CONFIG_SPARSEMEM */
337 :
338 :
339 : ZONE_PADDING(_pad1_)
340 :
341 : /* Fields commonly accessed by the page reclaim scanner */
342 : spinlock_t lru_lock;
343 : struct zone_lru {
344 : struct list_head list;
345 : } lru[NR_LRU_LISTS];
346 :
347 : struct zone_reclaim_stat reclaim_stat;
348 :
349 : unsigned long pages_scanned; /* since last reclaim */
350 : unsigned long flags; /* zone flags, see below */
351 :
352 : /* Zone statistics */
353 : atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
354 :
355 : /*
356 : * prev_priority holds the scanning priority for this zone. It is
357 : * defined as the scanning priority at which we achieved our reclaim
358 : * target at the previous try_to_free_pages() or balance_pgdat()
359 : * invokation.
360 : *
361 : * We use prev_priority as a measure of how much stress page reclaim is
362 : * under - it drives the swappiness decision: whether to unmap mapped
363 : * pages.
364 : *
365 : * Access to both this field is quite racy even on uniprocessor. But
366 : * it is expected to average out OK.
367 : */
368 : int prev_priority;
369 :
370 : /*
371 : * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on
372 : * this zone's LRU. Maintained by the pageout code.
373 : */
374 : unsigned int inactive_ratio;
375 :
376 :
377 : ZONE_PADDING(_pad2_)
378 : /* Rarely used or read-mostly fields */
379 :
380 : /*
381 : * wait_table -- the array holding the hash table
382 : * wait_table_hash_nr_entries -- the size of the hash table array
383 : * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
384 : *
385 : * The purpose of all these is to keep track of the people
386 : * waiting for a page to become available and make them
387 : * runnable again when possible. The trouble is that this
388 : * consumes a lot of space, especially when so few things
389 : * wait on pages at a given time. So instead of using
390 : * per-page waitqueues, we use a waitqueue hash table.
391 : *
392 : * The bucket discipline is to sleep on the same queue when
393 : * colliding and wake all in that wait queue when removing.
394 : * When something wakes, it must check to be sure its page is
395 : * truly available, a la thundering herd. The cost of a
396 : * collision is great, but given the expected load of the
397 : * table, they should be so rare as to be outweighed by the
398 : * benefits from the saved space.
399 : *
400 : * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
401 : * primary users of these fields, and in mm/page_alloc.c
402 : * free_area_init_core() performs the initialization of them.
403 : */
404 : wait_queue_head_t * wait_table;
405 : unsigned long wait_table_hash_nr_entries;
406 : unsigned long wait_table_bits;
407 :
408 : /*
409 : * Discontig memory support fields.
410 : */
411 : struct pglist_data *zone_pgdat;
412 : /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
413 : unsigned long zone_start_pfn;
414 :
415 : /*
416 : * zone_start_pfn, spanned_pages and present_pages are all
417 : * protected by span_seqlock. It is a seqlock because it has
418 : * to be read outside of zone->lock, and it is done in the main
419 : * allocator path. But, it is written quite infrequently.
420 : *
421 : * The lock is declared along with zone->lock because it is
422 : * frequently read in proximity to zone->lock. It's good to
423 : * give them a chance of being in the same cacheline.
424 : */
425 : unsigned long spanned_pages; /* total size, including holes */
426 : unsigned long present_pages; /* amount of memory (excluding holes) */
427 :
428 : /*
429 : * rarely used fields:
430 : */
431 : const char *name;
432 : } ____cacheline_internodealigned_in_smp;
433 :
434 : typedef enum {
435 : ZONE_ALL_UNRECLAIMABLE, /* all pages pinned */
436 : ZONE_RECLAIM_LOCKED, /* prevents concurrent reclaim */
437 : ZONE_OOM_LOCKED, /* zone is in OOM killer zonelist */
438 : } zone_flags_t;
439 :
440 : static inline void zone_set_flag(struct zone *zone, zone_flags_t flag)
441 : {
442 : set_bit(flag, &zone->flags);
443 : }
444 :
445 : static inline int zone_test_and_set_flag(struct zone *zone, zone_flags_t flag)
446 : {
447 : return test_and_set_bit(flag, &zone->flags);
448 : }
449 :
450 : static inline void zone_clear_flag(struct zone *zone, zone_flags_t flag)
451 : {
452 : clear_bit(flag, &zone->flags);
453 : }
454 :
455 : static inline int zone_is_all_unreclaimable(const struct zone *zone)
456 : {
457 : return test_bit(ZONE_ALL_UNRECLAIMABLE, &zone->flags);
458 : }
459 :
460 : static inline int zone_is_reclaim_locked(const struct zone *zone)
461 : {
462 : return test_bit(ZONE_RECLAIM_LOCKED, &zone->flags);
463 : }
464 :
465 : static inline int zone_is_oom_locked(const struct zone *zone)
466 : {
467 : return test_bit(ZONE_OOM_LOCKED, &zone->flags);
468 : }
469 :
470 : #ifdef CONFIG_SMP
471 : unsigned long zone_nr_free_pages(struct zone *zone);
472 : #else
473 : #define zone_nr_free_pages(zone) zone_page_state(zone, NR_FREE_PAGES)
474 : #endif /* CONFIG_SMP */
475 :
476 : /*
477 : * The "priority" of VM scanning is how much of the queues we will scan in one
478 : * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
479 : * queues ("queue_length >> 12") during an aging round.
480 : */
481 : #define DEF_PRIORITY 12
482 :
483 : /* Maximum number of zones on a zonelist */
484 : #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
485 :
486 : #ifdef CONFIG_NUMA
487 :
488 : /*
489 : * The NUMA zonelists are doubled becausse we need zonelists that restrict the
490 : * allocations to a single node for GFP_THISNODE.
491 : *
492 : * [0] : Zonelist with fallback
493 : * [1] : No fallback (GFP_THISNODE)
494 : */
495 : #define MAX_ZONELISTS 2
496 :
497 :
498 : /*
499 : * We cache key information from each zonelist for smaller cache
500 : * footprint when scanning for free pages in get_page_from_freelist().
501 : *
502 : * 1) The BITMAP fullzones tracks which zones in a zonelist have come
503 : * up short of free memory since the last time (last_fullzone_zap)
504 : * we zero'd fullzones.
505 : * 2) The array z_to_n[] maps each zone in the zonelist to its node
506 : * id, so that we can efficiently evaluate whether that node is
507 : * set in the current tasks mems_allowed.
508 : *
509 : * Both fullzones and z_to_n[] are one-to-one with the zonelist,
510 : * indexed by a zones offset in the zonelist zones[] array.
511 : *
512 : * The get_page_from_freelist() routine does two scans. During the
513 : * first scan, we skip zones whose corresponding bit in 'fullzones'
514 : * is set or whose corresponding node in current->mems_allowed (which
515 : * comes from cpusets) is not set. During the second scan, we bypass
516 : * this zonelist_cache, to ensure we look methodically at each zone.
517 : *
518 : * Once per second, we zero out (zap) fullzones, forcing us to
519 : * reconsider nodes that might have regained more free memory.
520 : * The field last_full_zap is the time we last zapped fullzones.
521 : *
522 : * This mechanism reduces the amount of time we waste repeatedly
523 : * reexaming zones for free memory when they just came up low on
524 : * memory momentarilly ago.
525 : *
526 : * The zonelist_cache struct members logically belong in struct
527 : * zonelist. However, the mempolicy zonelists constructed for
528 : * MPOL_BIND are intentionally variable length (and usually much
529 : * shorter). A general purpose mechanism for handling structs with
530 : * multiple variable length members is more mechanism than we want
531 : * here. We resort to some special case hackery instead.
532 : *
533 : * The MPOL_BIND zonelists don't need this zonelist_cache (in good
534 : * part because they are shorter), so we put the fixed length stuff
535 : * at the front of the zonelist struct, ending in a variable length
536 : * zones[], as is needed by MPOL_BIND.
537 : *
538 : * Then we put the optional zonelist cache on the end of the zonelist
539 : * struct. This optional stuff is found by a 'zlcache_ptr' pointer in
540 : * the fixed length portion at the front of the struct. This pointer
541 : * both enables us to find the zonelist cache, and in the case of
542 : * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL)
543 : * to know that the zonelist cache is not there.
544 : *
545 : * The end result is that struct zonelists come in two flavors:
546 : * 1) The full, fixed length version, shown below, and
547 : * 2) The custom zonelists for MPOL_BIND.
548 : * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache.
549 : *
550 : * Even though there may be multiple CPU cores on a node modifying
551 : * fullzones or last_full_zap in the same zonelist_cache at the same
552 : * time, we don't lock it. This is just hint data - if it is wrong now
553 : * and then, the allocator will still function, perhaps a bit slower.
554 : */
555 :
556 :
557 : struct zonelist_cache {
558 : unsigned short z_to_n[MAX_ZONES_PER_ZONELIST]; /* zone->nid */
559 : DECLARE_BITMAP(fullzones, MAX_ZONES_PER_ZONELIST); /* zone full? */
560 : unsigned long last_full_zap; /* when last zap'd (jiffies) */
561 : };
562 : #else
563 : #define MAX_ZONELISTS 1
564 : struct zonelist_cache;
565 : #endif
566 :
567 : /*
568 : * This struct contains information about a zone in a zonelist. It is stored
569 : * here to avoid dereferences into large structures and lookups of tables
570 : */
571 : struct zoneref {
572 : struct zone *zone; /* Pointer to actual zone */
573 : int zone_idx; /* zone_idx(zoneref->zone) */
574 : };
575 :
576 : /*
577 : * One allocation request operates on a zonelist. A zonelist
578 : * is a list of zones, the first one is the 'goal' of the
579 : * allocation, the other zones are fallback zones, in decreasing
580 : * priority.
581 : *
582 : * If zlcache_ptr is not NULL, then it is just the address of zlcache,
583 : * as explained above. If zlcache_ptr is NULL, there is no zlcache.
584 : * *
585 : * To speed the reading of the zonelist, the zonerefs contain the zone index
586 : * of the entry being read. Helper functions to access information given
587 : * a struct zoneref are
588 : *
589 : * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
590 : * zonelist_zone_idx() - Return the index of the zone for an entry
591 : * zonelist_node_idx() - Return the index of the node for an entry
592 : */
593 : struct zonelist {
594 : struct zonelist_cache *zlcache_ptr; // NULL or &zlcache
595 : struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
596 : #ifdef CONFIG_NUMA
597 : struct zonelist_cache zlcache; // optional ...
598 : #endif
599 : };
600 :
601 : #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
602 : struct node_active_region {
603 : unsigned long start_pfn;
604 : unsigned long end_pfn;
605 : int nid;
606 : };
607 : #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
608 :
609 : #ifndef CONFIG_DISCONTIGMEM
610 : /* The array of struct pages - for discontigmem use pgdat->lmem_map */
611 : extern struct page *mem_map;
612 : #endif
613 :
614 : /*
615 : * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
616 : * (mostly NUMA machines?) to denote a higher-level memory zone than the
617 : * zone denotes.
618 : *
619 : * On NUMA machines, each NUMA node would have a pg_data_t to describe
620 : * it's memory layout.
621 : *
622 : * Memory statistics and page replacement data structures are maintained on a
623 : * per-zone basis.
624 : */
625 : struct bootmem_data;
626 : typedef struct pglist_data {
627 : struct zone node_zones[MAX_NR_ZONES];
628 : struct zonelist node_zonelists[MAX_ZONELISTS];
629 : int nr_zones;
630 : #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
631 : struct page *node_mem_map;
632 : #ifdef CONFIG_CGROUP_MEM_RES_CTLR
633 : struct page_cgroup *node_page_cgroup;
634 : #endif
635 : #endif
636 : struct bootmem_data *bdata;
637 : #ifdef CONFIG_MEMORY_HOTPLUG
638 : /*
639 : * Must be held any time you expect node_start_pfn, node_present_pages
640 : * or node_spanned_pages stay constant. Holding this will also
641 : * guarantee that any pfn_valid() stays that way.
642 : *
643 : * Nests above zone->lock and zone->size_seqlock.
644 : */
645 : spinlock_t node_size_lock;
646 : #endif
647 : unsigned long node_start_pfn;
648 : unsigned long node_present_pages; /* total number of physical pages */
649 : unsigned long node_spanned_pages; /* total size of physical page
650 : range, including holes */
651 : int node_id;
652 : wait_queue_head_t kswapd_wait;
653 : struct task_struct *kswapd;
654 : int kswapd_max_order;
655 : } pg_data_t;
656 1 :
657 : #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
658 : #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
659 : #ifdef CONFIG_FLAT_NODE_MEM_MAP
660 : #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
661 : #else
662 : #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
663 : #endif
664 : #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
665 :
666 : #include <linux/memory_hotplug.h>
667 :
668 : void get_zone_counts(unsigned long *active, unsigned long *inactive,
669 : unsigned long *free);
670 : void build_all_zonelists(void);
671 : void wakeup_kswapd(struct zone *zone, int order);
672 : int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
673 : int classzone_idx, int alloc_flags);
674 : enum memmap_context {
675 : MEMMAP_EARLY,
676 : MEMMAP_HOTPLUG,
677 : };
678 : extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
679 : unsigned long size,
680 : enum memmap_context context);
681 :
682 : #ifdef CONFIG_HAVE_MEMORY_PRESENT
683 : void memory_present(int nid, unsigned long start, unsigned long end);
684 : #else
685 : static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
686 : #endif
687 :
688 : #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
689 : unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
690 : #endif
691 :
692 : /*
693 : * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
694 : */
695 : #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
696 :
697 : static inline int populated_zone(struct zone *zone)
698 : {
699 : return (!!zone->present_pages);
700 : }
701 :
702 : extern int movable_zone;
703 :
704 : static inline int zone_movable_is_highmem(void)
705 : {
706 : #if defined(CONFIG_HIGHMEM) && defined(CONFIG_ARCH_POPULATES_NODE_MAP)
707 : return movable_zone == ZONE_HIGHMEM;
708 : #else
709 : return 0;
710 : #endif
711 : }
712 :
713 : static inline int is_highmem_idx(enum zone_type idx)
714 : {
715 : #ifdef CONFIG_HIGHMEM
716 : return (idx == ZONE_HIGHMEM ||
717 : (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
718 : #else
719 : return 0;
720 : #endif
721 : }
722 :
723 : static inline int is_normal_idx(enum zone_type idx)
724 : {
725 : return (idx == ZONE_NORMAL);
726 : }
727 :
728 : /**
729 : * is_highmem - helper function to quickly check if a struct zone is a
730 : * highmem zone or not. This is an attempt to keep references
731 : * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
732 : * @zone - pointer to struct zone variable
733 : */
734 : static inline int is_highmem(struct zone *zone)
735 : {
736 : #ifdef CONFIG_HIGHMEM
737 : int zone_off = (char *)zone - (char *)zone->zone_pgdat->node_zones;
738 : return zone_off == ZONE_HIGHMEM * sizeof(*zone) ||
739 : (zone_off == ZONE_MOVABLE * sizeof(*zone) &&
740 : zone_movable_is_highmem());
741 : #else
742 : return 0;
743 : #endif
744 : }
745 :
746 : static inline int is_normal(struct zone *zone)
747 : {
748 : return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
749 : }
750 :
751 : static inline int is_dma32(struct zone *zone)
752 : {
753 : #ifdef CONFIG_ZONE_DMA32
754 : return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
755 : #else
756 : return 0;
757 : #endif
758 : }
759 :
760 : static inline int is_dma(struct zone *zone)
761 : {
762 : #ifdef CONFIG_ZONE_DMA
763 : return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
764 : #else
765 : return 0;
766 : #endif
767 : }
768 :
769 : /* These two functions are used to setup the per zone pages min values */
770 : struct ctl_table;
771 : int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
772 : void __user *, size_t *, loff_t *);
773 : extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
774 : int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
775 : void __user *, size_t *, loff_t *);
776 : int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
777 : void __user *, size_t *, loff_t *);
778 : int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
779 : void __user *, size_t *, loff_t *);
780 : int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
781 : void __user *, size_t *, loff_t *);
782 :
783 : extern int numa_zonelist_order_handler(struct ctl_table *, int,
784 : void __user *, size_t *, loff_t *);
785 : extern char numa_zonelist_order[];
786 : #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */
787 :
788 : #ifndef CONFIG_NEED_MULTIPLE_NODES
789 :
790 : extern struct pglist_data contig_page_data;
791 : #define NODE_DATA(nid) (&contig_page_data)
792 : #define NODE_MEM_MAP(nid) mem_map
793 :
794 : #else /* CONFIG_NEED_MULTIPLE_NODES */
795 :
796 : #include <asm/mmzone.h>
797 :
798 : #endif /* !CONFIG_NEED_MULTIPLE_NODES */
799 :
800 : extern struct pglist_data *first_online_pgdat(void);
801 : extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
802 : extern struct zone *next_zone(struct zone *zone);
803 :
804 : /**
805 : * for_each_online_pgdat - helper macro to iterate over all online nodes
806 : * @pgdat - pointer to a pg_data_t variable
807 : */
808 : #define for_each_online_pgdat(pgdat) \
809 : for (pgdat = first_online_pgdat(); \
810 : pgdat; \
811 : pgdat = next_online_pgdat(pgdat))
812 : /**
813 : * for_each_zone - helper macro to iterate over all memory zones
814 : * @zone - pointer to struct zone variable
815 : *
816 : * The user only needs to declare the zone variable, for_each_zone
817 : * fills it in.
818 : */
819 : #define for_each_zone(zone) \
820 : for (zone = (first_online_pgdat())->node_zones; \
821 : zone; \
822 : zone = next_zone(zone))
823 :
824 : #define for_each_populated_zone(zone) \
825 : for (zone = (first_online_pgdat())->node_zones; \
826 : zone; \
827 : zone = next_zone(zone)) \
828 : if (!populated_zone(zone)) \
829 : ; /* do nothing */ \
830 : else
831 :
832 : static inline struct zone *zonelist_zone(struct zoneref *zoneref)
833 : {
834 : return zoneref->zone;
835 : }
836 :
837 : static inline int zonelist_zone_idx(struct zoneref *zoneref)
838 : {
839 : return zoneref->zone_idx;
840 : }
841 :
842 : static inline int zonelist_node_idx(struct zoneref *zoneref)
843 : {
844 : #ifdef CONFIG_NUMA
845 : /* zone_to_nid not available in this context */
846 : return zoneref->zone->node;
847 : #else
848 : return 0;
849 : #endif /* CONFIG_NUMA */
850 : }
851 :
852 : /**
853 : * next_zones_zonelist - Returns the next zone at or below highest_zoneidx within the allowed nodemask using a cursor within a zonelist as a starting point
854 : * @z - The cursor used as a starting point for the search
855 : * @highest_zoneidx - The zone index of the highest zone to return
856 : * @nodes - An optional nodemask to filter the zonelist with
857 : * @zone - The first suitable zone found is returned via this parameter
858 : *
859 : * This function returns the next zone at or below a given zone index that is
860 : * within the allowed nodemask using a cursor as the starting point for the
861 : * search. The zoneref returned is a cursor that represents the current zone
862 : * being examined. It should be advanced by one before calling
863 : * next_zones_zonelist again.
864 : */
865 : struct zoneref *next_zones_zonelist(struct zoneref *z,
866 : enum zone_type highest_zoneidx,
867 : nodemask_t *nodes,
868 : struct zone **zone);
869 :
870 : /**
871 : * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
872 : * @zonelist - The zonelist to search for a suitable zone
873 : * @highest_zoneidx - The zone index of the highest zone to return
874 : * @nodes - An optional nodemask to filter the zonelist with
875 : * @zone - The first suitable zone found is returned via this parameter
876 : *
877 : * This function returns the first zone at or below a given zone index that is
878 : * within the allowed nodemask. The zoneref returned is a cursor that can be
879 : * used to iterate the zonelist with next_zones_zonelist by advancing it by
880 : * one before calling.
881 : */
882 : static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
883 : enum zone_type highest_zoneidx,
884 : nodemask_t *nodes,
885 : struct zone **zone)
886 : {
887 : return next_zones_zonelist(zonelist->_zonerefs, highest_zoneidx, nodes,
888 : zone);
889 : }
890 1 :
891 : /**
892 : * for_each_zone_zonelist_nodemask - helper macro to iterate over valid zones in a zonelist at or below a given zone index and within a nodemask
893 : * @zone - The current zone in the iterator
894 : * @z - The current pointer within zonelist->zones being iterated
895 : * @zlist - The zonelist being iterated
896 : * @highidx - The zone index of the highest zone to return
897 : * @nodemask - Nodemask allowed by the allocator
898 : *
899 : * This iterator iterates though all zones at or below a given zone index and
900 : * within a given nodemask
901 : */
902 : #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
903 : for (z = first_zones_zonelist(zlist, highidx, nodemask, &zone); \
904 : zone; \
905 : z = next_zones_zonelist(++z, highidx, nodemask, &zone)) \
906 :
907 : /**
908 : * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
909 : * @zone - The current zone in the iterator
910 : * @z - The current pointer within zonelist->zones being iterated
911 : * @zlist - The zonelist being iterated
912 : * @highidx - The zone index of the highest zone to return
913 : *
914 : * This iterator iterates though all zones at or below a given zone index.
915 : */
916 : #define for_each_zone_zonelist(zone, z, zlist, highidx) \
917 : for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
918 :
919 : #ifdef CONFIG_SPARSEMEM
920 : #include <asm/sparsemem.h>
921 : #endif
922 :
923 : #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
924 : !defined(CONFIG_ARCH_POPULATES_NODE_MAP)
925 : static inline unsigned long early_pfn_to_nid(unsigned long pfn)
926 : {
927 : return 0;
928 : }
929 : #endif
930 :
931 : #ifdef CONFIG_FLATMEM
932 : #define pfn_to_nid(pfn) (0)
933 : #endif
934 :
935 : #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
936 : #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
937 :
938 : #ifdef CONFIG_SPARSEMEM
939 :
940 : /*
941 : * SECTION_SHIFT #bits space required to store a section #
942 : *
943 : * PA_SECTION_SHIFT physical address to/from section number
944 : * PFN_SECTION_SHIFT pfn to/from section number
945 : */
946 : #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
947 :
948 : #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
949 : #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
950 :
951 : #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
952 :
953 : #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
954 : #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
955 :
956 : #define SECTION_BLOCKFLAGS_BITS \
957 : ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
958 :
959 : #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
960 : #error Allocator MAX_ORDER exceeds SECTION_SIZE
961 : #endif
962 :
963 : struct page;
964 : struct page_cgroup;
965 : struct mem_section {
966 : /*
967 : * This is, logically, a pointer to an array of struct
968 : * pages. However, it is stored with some other magic.
969 : * (see sparse.c::sparse_init_one_section())
970 : *
971 : * Additionally during early boot we encode node id of
972 : * the location of the section here to guide allocation.
973 : * (see sparse.c::memory_present())
974 : *
975 : * Making it a UL at least makes someone do a cast
976 : * before using it wrong.
977 : */
978 : unsigned long section_mem_map;
979 :
980 : /* See declaration of similar field in struct zone */
981 : unsigned long *pageblock_flags;
982 : #ifdef CONFIG_CGROUP_MEM_RES_CTLR
983 : /*
984 : * If !SPARSEMEM, pgdat doesn't have page_cgroup pointer. We use
985 : * section. (see memcontrol.h/page_cgroup.h about this.)
986 : */
987 : struct page_cgroup *page_cgroup;
988 : unsigned long pad;
989 : #endif
990 : };
991 :
992 : #ifdef CONFIG_SPARSEMEM_EXTREME
993 : #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
994 : #else
995 : #define SECTIONS_PER_ROOT 1
996 : #endif
997 :
998 : #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
999 : #define NR_SECTION_ROOTS (NR_MEM_SECTIONS / SECTIONS_PER_ROOT)
1000 : #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1001 :
1002 : #ifdef CONFIG_SPARSEMEM_EXTREME
1003 1 : extern struct mem_section *mem_section[NR_SECTION_ROOTS];
1004 : #else
1005 : extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1006 : #endif
1007 :
1008 : static inline struct mem_section *__nr_to_section(unsigned long nr)
1009 : {
1010 8 : if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1011 4 : return NULL;
1012 4 : return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1013 : }
1014 : extern int __section_nr(struct mem_section* ms);
1015 : extern unsigned long usemap_size(void);
1016 :
1017 : /*
1018 : * We use the lower bits of the mem_map pointer to store
1019 : * a little bit of information. There should be at least
1020 : * 3 bits here due to 32-bit alignment.
1021 : */
1022 : #define SECTION_MARKED_PRESENT (1UL<<0)
1023 : #define SECTION_HAS_MEM_MAP (1UL<<1)
1024 : #define SECTION_MAP_LAST_BIT (1UL<<2)
1025 : #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1026 : #define SECTION_NID_SHIFT 2
1027 :
1028 : static inline struct page *__section_mem_map_addr(struct mem_section *section)
1029 : {
1030 8 : unsigned long map = section->section_mem_map;
1031 4 : map &= SECTION_MAP_MASK;
1032 4 : return (struct page *)map;
1033 : }
1034 :
1035 : static inline int present_section(struct mem_section *section)
1036 : {
1037 : return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1038 : }
1039 :
1040 : static inline int present_section_nr(unsigned long nr)
1041 : {
1042 : return present_section(__nr_to_section(nr));
1043 : }
1044 :
1045 : static inline int valid_section(struct mem_section *section)
1046 : {
1047 : return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1048 : }
1049 :
1050 : static inline int valid_section_nr(unsigned long nr)
1051 : {
1052 : return valid_section(__nr_to_section(nr));
1053 : }
1054 :
1055 : static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1056 : {
1057 4 : return __nr_to_section(pfn_to_section_nr(pfn));
1058 : }
1059 :
1060 : static inline int pfn_valid(unsigned long pfn)
1061 : {
1062 : if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1063 : return 0;
1064 : return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
1065 : }
1066 :
1067 : static inline int pfn_present(unsigned long pfn)
1068 : {
1069 : if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1070 : return 0;
1071 : return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1072 : }
1073 :
1074 : /*
1075 : * These are _only_ used during initialisation, therefore they
1076 : * can use __initdata ... They could have names to indicate
1077 : * this restriction.
1078 : */
1079 : #ifdef CONFIG_NUMA
1080 : #define pfn_to_nid(pfn) \
1081 : ({ \
1082 : unsigned long __pfn_to_nid_pfn = (pfn); \
1083 : page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1084 : })
1085 : #else
1086 : #define pfn_to_nid(pfn) (0)
1087 : #endif
1088 :
1089 : #define early_pfn_valid(pfn) pfn_valid(pfn)
1090 : void sparse_init(void);
1091 : #else
1092 : #define sparse_init() do {} while (0)
1093 : #define sparse_index_init(_sec, _nid) do {} while (0)
1094 : #endif /* CONFIG_SPARSEMEM */
1095 :
1096 : #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1097 : bool early_pfn_in_nid(unsigned long pfn, int nid);
1098 : #else
1099 : #define early_pfn_in_nid(pfn, nid) (1)
1100 : #endif
1101 :
1102 : #ifndef early_pfn_valid
1103 : #define early_pfn_valid(pfn) (1)
1104 : #endif
1105 :
1106 : void memory_present(int nid, unsigned long start, unsigned long end);
1107 : unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
1108 :
1109 : /*
1110 : * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1111 : * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1112 : * pfn_valid_within() should be used in this case; we optimise this away
1113 : * when we have no holes within a MAX_ORDER_NR_PAGES block.
1114 : */
1115 : #ifdef CONFIG_HOLES_IN_ZONE
1116 : #define pfn_valid_within(pfn) pfn_valid(pfn)
1117 : #else
1118 : #define pfn_valid_within(pfn) (1)
1119 : #endif
1120 :
1121 : #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1122 : /*
1123 : * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1124 : * associated with it or not. In FLATMEM, it is expected that holes always
1125 : * have valid memmap as long as there is valid PFNs either side of the hole.
1126 : * In SPARSEMEM, it is assumed that a valid section has a memmap for the
1127 : * entire section.
1128 : *
1129 : * However, an ARM, and maybe other embedded architectures in the future
1130 : * free memmap backing holes to save memory on the assumption the memmap is
1131 : * never used. The page_zone linkages are then broken even though pfn_valid()
1132 : * returns true. A walker of the full memmap must then do this additional
1133 : * check to ensure the memmap they are looking at is sane by making sure
1134 : * the zone and PFN linkages are still valid. This is expensive, but walkers
1135 : * of the full memmap are extremely rare.
1136 : */
1137 : int memmap_valid_within(unsigned long pfn,
1138 : struct page *page, struct zone *zone);
1139 : #else
1140 : static inline int memmap_valid_within(unsigned long pfn,
1141 : struct page *page, struct zone *zone)
1142 : {
1143 : return 1;
1144 : }
1145 : #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1146 :
1147 : #endif /* !__GENERATING_BOUNDS.H */
1148 : #endif /* !__ASSEMBLY__ */
1149 : #endif /* _LINUX_MMZONE_H */
|