[PATCH] Apply type enum zone_type
[deliverable/linux.git] / include / linux / mmzone.h
CommitLineData
1da177e4
LT
1#ifndef _LINUX_MMZONE_H
2#define _LINUX_MMZONE_H
3
4#ifdef __KERNEL__
5#ifndef __ASSEMBLY__
6
1da177e4
LT
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>
bdc8cb98 14#include <linux/seqlock.h>
8357f869 15#include <linux/nodemask.h>
1da177e4 16#include <asm/atomic.h>
93ff66bf 17#include <asm/page.h>
1da177e4
LT
18
19/* Free memory management - zoned buddy allocator. */
20#ifndef CONFIG_FORCE_MAX_ZONEORDER
21#define MAX_ORDER 11
22#else
23#define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
24#endif
e984bb43 25#define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
1da177e4
LT
26
27struct free_area {
28 struct list_head free_list;
29 unsigned long nr_free;
30};
31
32struct pglist_data;
33
34/*
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.
39 */
40#if defined(CONFIG_SMP)
41struct zone_padding {
42 char x[0];
22fc6ecc 43} ____cacheline_internodealigned_in_smp;
1da177e4
LT
44#define ZONE_PADDING(name) struct zone_padding name;
45#else
46#define ZONE_PADDING(name)
47#endif
48
2244b95a 49enum zone_stat_item {
f3dbd344
CL
50 NR_ANON_PAGES, /* Mapped anonymous pages */
51 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
65ba55f5 52 only modified from process context */
347ce434 53 NR_FILE_PAGES,
9a865ffa 54 NR_SLAB, /* Pages used by slab allocator */
df849a15 55 NR_PAGETABLE, /* used for pagetables */
b1e7a8fd 56 NR_FILE_DIRTY,
ce866b34 57 NR_WRITEBACK,
fd39fc85 58 NR_UNSTABLE_NFS, /* NFS unstable pages */
d2c5e30c 59 NR_BOUNCE,
ca889e6c
CL
60#ifdef CONFIG_NUMA
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 */
67#endif
2244b95a
CL
68 NR_VM_ZONE_STAT_ITEMS };
69
1da177e4
LT
70struct per_cpu_pages {
71 int count; /* number of pages in the list */
1da177e4
LT
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 */
75};
76
77struct per_cpu_pageset {
78 struct per_cpu_pages pcp[2]; /* 0: hot. 1: cold */
2244b95a 79#ifdef CONFIG_SMP
df9ecaba 80 s8 stat_threshold;
2244b95a
CL
81 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
82#endif
1da177e4
LT
83} ____cacheline_aligned_in_smp;
84
e7c8d5c9
CL
85#ifdef CONFIG_NUMA
86#define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)])
87#else
88#define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)])
89#endif
90
2f1b6248
CL
91enum zone_type {
92 /*
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.
97 *
98 * Some examples
99 *
100 * Architecture Limit
101 * ---------------------------
102 * parisc, ia64, sparc <4G
103 * s390 <2G
104 * arm26 <48M
105 * arm Various
106 * alpha Unlimited or 0-16MB.
107 *
108 * i386, x86_64 and multiple other arches
109 * <16M.
110 */
111 ZONE_DMA,
fb0e7942 112#ifdef CONFIG_ZONE_DMA32
2f1b6248
CL
113 /*
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.
117 */
118 ZONE_DMA32,
fb0e7942 119#endif
2f1b6248
CL
120 /*
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.
124 */
125 ZONE_NORMAL,
e53ef38d 126#ifdef CONFIG_HIGHMEM
2f1b6248
CL
127 /*
128 * A memory area that is only addressable by the kernel through
129 * mapping portions into its own address space. This is for example
130 * used by i386 to allow the kernel to address the memory beyond
131 * 900MB. The kernel will set up special mappings (page
132 * table entries on i386) for each page that the kernel needs to
133 * access.
134 */
135 ZONE_HIGHMEM,
e53ef38d 136#endif
2f1b6248
CL
137 MAX_NR_ZONES
138};
1da177e4
LT
139
140
141/*
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".
ac3461ad 153 *
79046ae0
AW
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.
ce2ea89b
AW
161 * Use the first form for GFP_ZONETYPES when the left most bit is not
162 * a "loner", otherwise use the second.
79046ae0 163 *
ac3461ad 164 * NOTE! Make sure this matches the zones in <linux/gfp.h>
1da177e4 165 */
fb0e7942
CL
166
167#ifdef CONFIG_ZONE_DMA32
e53ef38d
CL
168
169#ifdef CONFIG_HIGHMEM
170#define GFP_ZONETYPES ((GFP_ZONEMASK + 1) / 2 + 1) /* Loner */
fb0e7942 171#define GFP_ZONEMASK 0x07
e53ef38d 172#define ZONES_SHIFT 2 /* ceil(log2(MAX_NR_ZONES)) */
fb0e7942 173#else
e53ef38d
CL
174#define GFP_ZONETYPES ((0x07 + 1) / 2 + 1) /* Loner */
175/* Mask __GFP_HIGHMEM */
176#define GFP_ZONEMASK 0x05
177#define ZONES_SHIFT 2
178#endif
179
180#else
181#ifdef CONFIG_HIGHMEM
182
fb0e7942 183#define GFP_ZONEMASK 0x03
e53ef38d
CL
184#define ZONES_SHIFT 2
185#define GFP_ZONETYPES 3
186
187#else
188
189#define GFP_ZONEMASK 0x01
190#define ZONES_SHIFT 1
191#define GFP_ZONETYPES 2
192
193#endif
fb0e7942 194#endif
1da177e4 195
1da177e4
LT
196struct zone {
197 /* Fields commonly accessed by the page allocator */
198 unsigned long free_pages;
199 unsigned long pages_min, pages_low, pages_high;
200 /*
201 * We don't know if the memory that we're going to allocate will be freeable
202 * or/and it will be released eventually, so to avoid totally wasting several
203 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
204 * to run OOM on the lower zones despite there's tons of freeable ram
205 * on the higher zones). This array is recalculated at runtime if the
206 * sysctl_lowmem_reserve_ratio sysctl changes.
207 */
208 unsigned long lowmem_reserve[MAX_NR_ZONES];
209
e7c8d5c9 210#ifdef CONFIG_NUMA
9614634f
CL
211 /*
212 * zone reclaim becomes active if more unmapped pages exist.
213 */
214 unsigned long min_unmapped_ratio;
e7c8d5c9
CL
215 struct per_cpu_pageset *pageset[NR_CPUS];
216#else
1da177e4 217 struct per_cpu_pageset pageset[NR_CPUS];
e7c8d5c9 218#endif
1da177e4
LT
219 /*
220 * free areas of different sizes
221 */
222 spinlock_t lock;
bdc8cb98
DH
223#ifdef CONFIG_MEMORY_HOTPLUG
224 /* see spanned/present_pages for more description */
225 seqlock_t span_seqlock;
226#endif
1da177e4
LT
227 struct free_area free_area[MAX_ORDER];
228
229
230 ZONE_PADDING(_pad1_)
231
232 /* Fields commonly accessed by the page reclaim scanner */
233 spinlock_t lru_lock;
234 struct list_head active_list;
235 struct list_head inactive_list;
236 unsigned long nr_scan_active;
237 unsigned long nr_scan_inactive;
238 unsigned long nr_active;
239 unsigned long nr_inactive;
240 unsigned long pages_scanned; /* since last reclaim */
241 int all_unreclaimable; /* All pages pinned */
242
1e7e5a90
MH
243 /* A count of how many reclaimers are scanning this zone */
244 atomic_t reclaim_in_progress;
753ee728 245
2244b95a
CL
246 /* Zone statistics */
247 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
9eeff239 248
1da177e4
LT
249 /*
250 * prev_priority holds the scanning priority for this zone. It is
251 * defined as the scanning priority at which we achieved our reclaim
252 * target at the previous try_to_free_pages() or balance_pgdat()
253 * invokation.
254 *
255 * We use prev_priority as a measure of how much stress page reclaim is
256 * under - it drives the swappiness decision: whether to unmap mapped
257 * pages.
258 *
259 * temp_priority is used to remember the scanning priority at which
260 * this zone was successfully refilled to free_pages == pages_high.
261 *
262 * Access to both these fields is quite racy even on uniprocessor. But
263 * it is expected to average out OK.
264 */
265 int temp_priority;
266 int prev_priority;
267
268
269 ZONE_PADDING(_pad2_)
270 /* Rarely used or read-mostly fields */
271
272 /*
273 * wait_table -- the array holding the hash table
02b694de 274 * wait_table_hash_nr_entries -- the size of the hash table array
1da177e4
LT
275 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
276 *
277 * The purpose of all these is to keep track of the people
278 * waiting for a page to become available and make them
279 * runnable again when possible. The trouble is that this
280 * consumes a lot of space, especially when so few things
281 * wait on pages at a given time. So instead of using
282 * per-page waitqueues, we use a waitqueue hash table.
283 *
284 * The bucket discipline is to sleep on the same queue when
285 * colliding and wake all in that wait queue when removing.
286 * When something wakes, it must check to be sure its page is
287 * truly available, a la thundering herd. The cost of a
288 * collision is great, but given the expected load of the
289 * table, they should be so rare as to be outweighed by the
290 * benefits from the saved space.
291 *
292 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
293 * primary users of these fields, and in mm/page_alloc.c
294 * free_area_init_core() performs the initialization of them.
295 */
296 wait_queue_head_t * wait_table;
02b694de 297 unsigned long wait_table_hash_nr_entries;
1da177e4
LT
298 unsigned long wait_table_bits;
299
300 /*
301 * Discontig memory support fields.
302 */
303 struct pglist_data *zone_pgdat;
1da177e4
LT
304 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
305 unsigned long zone_start_pfn;
306
bdc8cb98
DH
307 /*
308 * zone_start_pfn, spanned_pages and present_pages are all
309 * protected by span_seqlock. It is a seqlock because it has
310 * to be read outside of zone->lock, and it is done in the main
311 * allocator path. But, it is written quite infrequently.
312 *
313 * The lock is declared along with zone->lock because it is
314 * frequently read in proximity to zone->lock. It's good to
315 * give them a chance of being in the same cacheline.
316 */
1da177e4
LT
317 unsigned long spanned_pages; /* total size, including holes */
318 unsigned long present_pages; /* amount of memory (excluding holes) */
319
320 /*
321 * rarely used fields:
322 */
323 char *name;
22fc6ecc 324} ____cacheline_internodealigned_in_smp;
1da177e4 325
1da177e4
LT
326/*
327 * The "priority" of VM scanning is how much of the queues we will scan in one
328 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
329 * queues ("queue_length >> 12") during an aging round.
330 */
331#define DEF_PRIORITY 12
332
333/*
334 * One allocation request operates on a zonelist. A zonelist
335 * is a list of zones, the first one is the 'goal' of the
336 * allocation, the other zones are fallback zones, in decreasing
337 * priority.
338 *
339 * Right now a zonelist takes up less than a cacheline. We never
340 * modify it apart from boot-up, and only a few indices are used,
341 * so despite the zonelist table being relatively big, the cache
342 * footprint of this construct is very small.
343 */
344struct zonelist {
345 struct zone *zones[MAX_NUMNODES * MAX_NR_ZONES + 1]; // NULL delimited
346};
347
348
349/*
350 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
351 * (mostly NUMA machines?) to denote a higher-level memory zone than the
352 * zone denotes.
353 *
354 * On NUMA machines, each NUMA node would have a pg_data_t to describe
355 * it's memory layout.
356 *
357 * Memory statistics and page replacement data structures are maintained on a
358 * per-zone basis.
359 */
360struct bootmem_data;
361typedef struct pglist_data {
362 struct zone node_zones[MAX_NR_ZONES];
363 struct zonelist node_zonelists[GFP_ZONETYPES];
364 int nr_zones;
d41dee36 365#ifdef CONFIG_FLAT_NODE_MEM_MAP
1da177e4 366 struct page *node_mem_map;
d41dee36 367#endif
1da177e4 368 struct bootmem_data *bdata;
208d54e5
DH
369#ifdef CONFIG_MEMORY_HOTPLUG
370 /*
371 * Must be held any time you expect node_start_pfn, node_present_pages
372 * or node_spanned_pages stay constant. Holding this will also
373 * guarantee that any pfn_valid() stays that way.
374 *
375 * Nests above zone->lock and zone->size_seqlock.
376 */
377 spinlock_t node_size_lock;
378#endif
1da177e4
LT
379 unsigned long node_start_pfn;
380 unsigned long node_present_pages; /* total number of physical pages */
381 unsigned long node_spanned_pages; /* total size of physical page
382 range, including holes */
383 int node_id;
1da177e4
LT
384 wait_queue_head_t kswapd_wait;
385 struct task_struct *kswapd;
386 int kswapd_max_order;
387} pg_data_t;
388
389#define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
390#define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
d41dee36 391#ifdef CONFIG_FLAT_NODE_MEM_MAP
408fde81 392#define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
d41dee36
AW
393#else
394#define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
395#endif
408fde81 396#define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
1da177e4 397
208d54e5
DH
398#include <linux/memory_hotplug.h>
399
1da177e4
LT
400void __get_zone_counts(unsigned long *active, unsigned long *inactive,
401 unsigned long *free, struct pglist_data *pgdat);
402void get_zone_counts(unsigned long *active, unsigned long *inactive,
403 unsigned long *free);
404void build_all_zonelists(void);
405void wakeup_kswapd(struct zone *zone, int order);
406int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
7fb1d9fc 407 int classzone_idx, int alloc_flags);
1da177e4 408
718127cc
YG
409extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
410 unsigned long size);
411
1da177e4
LT
412#ifdef CONFIG_HAVE_MEMORY_PRESENT
413void memory_present(int nid, unsigned long start, unsigned long end);
414#else
415static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
416#endif
417
418#ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
419unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
420#endif
421
422/*
423 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
424 */
425#define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
426
f3fe6512
CK
427static inline int populated_zone(struct zone *zone)
428{
429 return (!!zone->present_pages);
430}
431
2f1b6248 432static inline int is_highmem_idx(enum zone_type idx)
1da177e4 433{
e53ef38d 434#ifdef CONFIG_HIGHMEM
1da177e4 435 return (idx == ZONE_HIGHMEM);
e53ef38d
CL
436#else
437 return 0;
438#endif
1da177e4
LT
439}
440
2f1b6248 441static inline int is_normal_idx(enum zone_type idx)
1da177e4
LT
442{
443 return (idx == ZONE_NORMAL);
444}
9328b8fa 445
1da177e4
LT
446/**
447 * is_highmem - helper function to quickly check if a struct zone is a
448 * highmem zone or not. This is an attempt to keep references
449 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
450 * @zone - pointer to struct zone variable
451 */
452static inline int is_highmem(struct zone *zone)
453{
e53ef38d 454#ifdef CONFIG_HIGHMEM
1da177e4 455 return zone == zone->zone_pgdat->node_zones + ZONE_HIGHMEM;
e53ef38d
CL
456#else
457 return 0;
458#endif
1da177e4
LT
459}
460
461static inline int is_normal(struct zone *zone)
462{
463 return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
464}
465
9328b8fa
NP
466static inline int is_dma32(struct zone *zone)
467{
fb0e7942 468#ifdef CONFIG_ZONE_DMA32
9328b8fa 469 return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
fb0e7942
CL
470#else
471 return 0;
472#endif
9328b8fa
NP
473}
474
475static inline int is_dma(struct zone *zone)
476{
477 return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
478}
479
1da177e4
LT
480/* These two functions are used to setup the per zone pages min values */
481struct ctl_table;
482struct file;
483int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *,
484 void __user *, size_t *, loff_t *);
485extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
486int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *,
487 void __user *, size_t *, loff_t *);
8ad4b1fb
RS
488int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, struct file *,
489 void __user *, size_t *, loff_t *);
9614634f
CL
490int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
491 struct file *, void __user *, size_t *, loff_t *);
1da177e4
LT
492
493#include <linux/topology.h>
494/* Returns the number of the current Node. */
69d81fcd 495#ifndef numa_node_id
39c715b7 496#define numa_node_id() (cpu_to_node(raw_smp_processor_id()))
69d81fcd 497#endif
1da177e4 498
93b7504e 499#ifndef CONFIG_NEED_MULTIPLE_NODES
1da177e4
LT
500
501extern struct pglist_data contig_page_data;
502#define NODE_DATA(nid) (&contig_page_data)
503#define NODE_MEM_MAP(nid) mem_map
504#define MAX_NODES_SHIFT 1
1da177e4 505
93b7504e 506#else /* CONFIG_NEED_MULTIPLE_NODES */
1da177e4
LT
507
508#include <asm/mmzone.h>
509
93b7504e 510#endif /* !CONFIG_NEED_MULTIPLE_NODES */
348f8b6c 511
95144c78
KH
512extern struct pglist_data *first_online_pgdat(void);
513extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
514extern struct zone *next_zone(struct zone *zone);
8357f869
KH
515
516/**
517 * for_each_pgdat - helper macro to iterate over all nodes
518 * @pgdat - pointer to a pg_data_t variable
519 */
520#define for_each_online_pgdat(pgdat) \
521 for (pgdat = first_online_pgdat(); \
522 pgdat; \
523 pgdat = next_online_pgdat(pgdat))
8357f869
KH
524/**
525 * for_each_zone - helper macro to iterate over all memory zones
526 * @zone - pointer to struct zone variable
527 *
528 * The user only needs to declare the zone variable, for_each_zone
529 * fills it in.
530 */
531#define for_each_zone(zone) \
532 for (zone = (first_online_pgdat())->node_zones; \
533 zone; \
534 zone = next_zone(zone))
535
d41dee36
AW
536#ifdef CONFIG_SPARSEMEM
537#include <asm/sparsemem.h>
538#endif
539
07808b74 540#if BITS_PER_LONG == 32
1da177e4 541/*
a2f1b424
AK
542 * with 32 bit page->flags field, we reserve 9 bits for node/zone info.
543 * there are 4 zones (3 bits) and this leaves 9-3=6 bits for nodes.
1da177e4 544 */
a2f1b424 545#define FLAGS_RESERVED 9
348f8b6c 546
1da177e4
LT
547#elif BITS_PER_LONG == 64
548/*
549 * with 64 bit flags field, there's plenty of room.
550 */
348f8b6c 551#define FLAGS_RESERVED 32
1da177e4 552
348f8b6c 553#else
1da177e4 554
348f8b6c 555#error BITS_PER_LONG not defined
1da177e4 556
1da177e4
LT
557#endif
558
b159d43f
AW
559#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
560#define early_pfn_to_nid(nid) (0UL)
561#endif
562
2bdaf115
AW
563#ifdef CONFIG_FLATMEM
564#define pfn_to_nid(pfn) (0)
565#endif
566
d41dee36
AW
567#define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
568#define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
569
570#ifdef CONFIG_SPARSEMEM
571
572/*
573 * SECTION_SHIFT #bits space required to store a section #
574 *
575 * PA_SECTION_SHIFT physical address to/from section number
576 * PFN_SECTION_SHIFT pfn to/from section number
577 */
578#define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
579
580#define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
581#define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
582
583#define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
584
585#define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
586#define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
587
588#if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
589#error Allocator MAX_ORDER exceeds SECTION_SIZE
590#endif
591
592struct page;
593struct mem_section {
29751f69
AW
594 /*
595 * This is, logically, a pointer to an array of struct
596 * pages. However, it is stored with some other magic.
597 * (see sparse.c::sparse_init_one_section())
598 *
30c253e6
AW
599 * Additionally during early boot we encode node id of
600 * the location of the section here to guide allocation.
601 * (see sparse.c::memory_present())
602 *
29751f69
AW
603 * Making it a UL at least makes someone do a cast
604 * before using it wrong.
605 */
606 unsigned long section_mem_map;
d41dee36
AW
607};
608
3e347261
BP
609#ifdef CONFIG_SPARSEMEM_EXTREME
610#define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
611#else
612#define SECTIONS_PER_ROOT 1
613#endif
802f192e 614
3e347261
BP
615#define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
616#define NR_SECTION_ROOTS (NR_MEM_SECTIONS / SECTIONS_PER_ROOT)
617#define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
802f192e 618
3e347261
BP
619#ifdef CONFIG_SPARSEMEM_EXTREME
620extern struct mem_section *mem_section[NR_SECTION_ROOTS];
802f192e 621#else
3e347261
BP
622extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
623#endif
d41dee36 624
29751f69
AW
625static inline struct mem_section *__nr_to_section(unsigned long nr)
626{
3e347261
BP
627 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
628 return NULL;
629 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
29751f69 630}
4ca644d9 631extern int __section_nr(struct mem_section* ms);
29751f69
AW
632
633/*
634 * We use the lower bits of the mem_map pointer to store
635 * a little bit of information. There should be at least
636 * 3 bits here due to 32-bit alignment.
637 */
638#define SECTION_MARKED_PRESENT (1UL<<0)
639#define SECTION_HAS_MEM_MAP (1UL<<1)
640#define SECTION_MAP_LAST_BIT (1UL<<2)
641#define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
30c253e6 642#define SECTION_NID_SHIFT 2
29751f69
AW
643
644static inline struct page *__section_mem_map_addr(struct mem_section *section)
645{
646 unsigned long map = section->section_mem_map;
647 map &= SECTION_MAP_MASK;
648 return (struct page *)map;
649}
650
651static inline int valid_section(struct mem_section *section)
652{
802f192e 653 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
29751f69
AW
654}
655
656static inline int section_has_mem_map(struct mem_section *section)
657{
802f192e 658 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
29751f69
AW
659}
660
661static inline int valid_section_nr(unsigned long nr)
662{
663 return valid_section(__nr_to_section(nr));
664}
665
d41dee36
AW
666static inline struct mem_section *__pfn_to_section(unsigned long pfn)
667{
29751f69 668 return __nr_to_section(pfn_to_section_nr(pfn));
d41dee36
AW
669}
670
d41dee36
AW
671static inline int pfn_valid(unsigned long pfn)
672{
673 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
674 return 0;
29751f69 675 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
d41dee36
AW
676}
677
678/*
679 * These are _only_ used during initialisation, therefore they
680 * can use __initdata ... They could have names to indicate
681 * this restriction.
682 */
683#ifdef CONFIG_NUMA
161599ff
AW
684#define pfn_to_nid(pfn) \
685({ \
686 unsigned long __pfn_to_nid_pfn = (pfn); \
687 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
688})
2bdaf115
AW
689#else
690#define pfn_to_nid(pfn) (0)
d41dee36
AW
691#endif
692
d41dee36
AW
693#define early_pfn_valid(pfn) pfn_valid(pfn)
694void sparse_init(void);
695#else
696#define sparse_init() do {} while (0)
28ae55c9 697#define sparse_index_init(_sec, _nid) do {} while (0)
d41dee36
AW
698#endif /* CONFIG_SPARSEMEM */
699
700#ifndef early_pfn_valid
701#define early_pfn_valid(pfn) (1)
702#endif
703
704void memory_present(int nid, unsigned long start, unsigned long end);
705unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
706
1da177e4
LT
707#endif /* !__ASSEMBLY__ */
708#endif /* __KERNEL__ */
709#endif /* _LINUX_MMZONE_H */
This page took 0.294379 seconds and 5 git commands to generate.