Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | #ifndef _LINUX_MMZONE_H |
2 | #define _LINUX_MMZONE_H | |
3 | ||
4 | #ifdef __KERNEL__ | |
5 | #ifndef __ASSEMBLY__ | |
6 | ||
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> | |
bdc8cb98 | 15 | #include <linux/seqlock.h> |
1da177e4 LT |
16 | #include <asm/atomic.h> |
17 | ||
18 | /* Free memory management - zoned buddy allocator. */ | |
19 | #ifndef CONFIG_FORCE_MAX_ZONEORDER | |
20 | #define MAX_ORDER 11 | |
21 | #else | |
22 | #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER | |
23 | #endif | |
24 | ||
25 | struct free_area { | |
26 | struct list_head free_list; | |
27 | unsigned long nr_free; | |
28 | }; | |
29 | ||
30 | struct pglist_data; | |
31 | ||
32 | /* | |
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. | |
37 | */ | |
38 | #if defined(CONFIG_SMP) | |
39 | struct zone_padding { | |
40 | char x[0]; | |
41 | } ____cacheline_maxaligned_in_smp; | |
42 | #define ZONE_PADDING(name) struct zone_padding name; | |
43 | #else | |
44 | #define ZONE_PADDING(name) | |
45 | #endif | |
46 | ||
47 | struct per_cpu_pages { | |
48 | int count; /* number of pages in the list */ | |
1da177e4 LT |
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 */ | |
52 | }; | |
53 | ||
54 | struct per_cpu_pageset { | |
55 | struct per_cpu_pages pcp[2]; /* 0: hot. 1: cold */ | |
56 | #ifdef CONFIG_NUMA | |
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 */ | |
63 | #endif | |
64 | } ____cacheline_aligned_in_smp; | |
65 | ||
e7c8d5c9 CL |
66 | #ifdef CONFIG_NUMA |
67 | #define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)]) | |
68 | #else | |
69 | #define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)]) | |
70 | #endif | |
71 | ||
1da177e4 | 72 | #define ZONE_DMA 0 |
a2f1b424 AK |
73 | #define ZONE_DMA32 1 |
74 | #define ZONE_NORMAL 2 | |
75 | #define ZONE_HIGHMEM 3 | |
1da177e4 | 76 | |
a2f1b424 | 77 | #define MAX_NR_ZONES 4 /* Sync this with ZONES_SHIFT */ |
1da177e4 LT |
78 | #define ZONES_SHIFT 2 /* ceil(log2(MAX_NR_ZONES)) */ |
79 | ||
80 | ||
81 | /* | |
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". | |
ac3461ad LT |
93 | * |
94 | * NOTE! Make sure this matches the zones in <linux/gfp.h> | |
1da177e4 | 95 | */ |
ac3461ad LT |
96 | #define GFP_ZONEMASK 0x07 |
97 | #define GFP_ZONETYPES 5 | |
1da177e4 LT |
98 | |
99 | /* | |
100 | * On machines where it is needed (eg PCs) we divide physical memory | |
a2f1b424 | 101 | * into multiple physical zones. On a PC we have 4 zones: |
1da177e4 LT |
102 | * |
103 | * ZONE_DMA < 16 MB ISA DMA capable memory | |
a2f1b424 | 104 | * ZONE_DMA32 0 MB Empty |
1da177e4 LT |
105 | * ZONE_NORMAL 16-896 MB direct mapped by the kernel |
106 | * ZONE_HIGHMEM > 896 MB only page cache and user processes | |
107 | */ | |
108 | ||
109 | struct zone { | |
110 | /* Fields commonly accessed by the page allocator */ | |
111 | unsigned long free_pages; | |
112 | unsigned long pages_min, pages_low, pages_high; | |
113 | /* | |
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. | |
120 | */ | |
121 | unsigned long lowmem_reserve[MAX_NR_ZONES]; | |
122 | ||
e7c8d5c9 CL |
123 | #ifdef CONFIG_NUMA |
124 | struct per_cpu_pageset *pageset[NR_CPUS]; | |
125 | #else | |
1da177e4 | 126 | struct per_cpu_pageset pageset[NR_CPUS]; |
e7c8d5c9 | 127 | #endif |
1da177e4 LT |
128 | /* |
129 | * free areas of different sizes | |
130 | */ | |
131 | spinlock_t lock; | |
bdc8cb98 DH |
132 | #ifdef CONFIG_MEMORY_HOTPLUG |
133 | /* see spanned/present_pages for more description */ | |
134 | seqlock_t span_seqlock; | |
135 | #endif | |
1da177e4 LT |
136 | struct free_area free_area[MAX_ORDER]; |
137 | ||
138 | ||
139 | ZONE_PADDING(_pad1_) | |
140 | ||
141 | /* Fields commonly accessed by the page reclaim scanner */ | |
142 | spinlock_t lru_lock; | |
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 */ | |
151 | ||
753ee728 MH |
152 | /* |
153 | * Does the allocator try to reclaim pages from the zone as soon | |
154 | * as it fails a watermark_ok() in __alloc_pages? | |
155 | */ | |
156 | int reclaim_pages; | |
1e7e5a90 MH |
157 | /* A count of how many reclaimers are scanning this zone */ |
158 | atomic_t reclaim_in_progress; | |
753ee728 | 159 | |
1da177e4 LT |
160 | /* |
161 | * prev_priority holds the scanning priority for this zone. It is | |
162 | * defined as the scanning priority at which we achieved our reclaim | |
163 | * target at the previous try_to_free_pages() or balance_pgdat() | |
164 | * invokation. | |
165 | * | |
166 | * We use prev_priority as a measure of how much stress page reclaim is | |
167 | * under - it drives the swappiness decision: whether to unmap mapped | |
168 | * pages. | |
169 | * | |
170 | * temp_priority is used to remember the scanning priority at which | |
171 | * this zone was successfully refilled to free_pages == pages_high. | |
172 | * | |
173 | * Access to both these fields is quite racy even on uniprocessor. But | |
174 | * it is expected to average out OK. | |
175 | */ | |
176 | int temp_priority; | |
177 | int prev_priority; | |
178 | ||
179 | ||
180 | ZONE_PADDING(_pad2_) | |
181 | /* Rarely used or read-mostly fields */ | |
182 | ||
183 | /* | |
184 | * wait_table -- the array holding the hash table | |
185 | * wait_table_size -- the size of the hash table array | |
186 | * wait_table_bits -- wait_table_size == (1 << wait_table_bits) | |
187 | * | |
188 | * The purpose of all these is to keep track of the people | |
189 | * waiting for a page to become available and make them | |
190 | * runnable again when possible. The trouble is that this | |
191 | * consumes a lot of space, especially when so few things | |
192 | * wait on pages at a given time. So instead of using | |
193 | * per-page waitqueues, we use a waitqueue hash table. | |
194 | * | |
195 | * The bucket discipline is to sleep on the same queue when | |
196 | * colliding and wake all in that wait queue when removing. | |
197 | * When something wakes, it must check to be sure its page is | |
198 | * truly available, a la thundering herd. The cost of a | |
199 | * collision is great, but given the expected load of the | |
200 | * table, they should be so rare as to be outweighed by the | |
201 | * benefits from the saved space. | |
202 | * | |
203 | * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the | |
204 | * primary users of these fields, and in mm/page_alloc.c | |
205 | * free_area_init_core() performs the initialization of them. | |
206 | */ | |
207 | wait_queue_head_t * wait_table; | |
208 | unsigned long wait_table_size; | |
209 | unsigned long wait_table_bits; | |
210 | ||
211 | /* | |
212 | * Discontig memory support fields. | |
213 | */ | |
214 | struct pglist_data *zone_pgdat; | |
215 | struct page *zone_mem_map; | |
216 | /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */ | |
217 | unsigned long zone_start_pfn; | |
218 | ||
bdc8cb98 DH |
219 | /* |
220 | * zone_start_pfn, spanned_pages and present_pages are all | |
221 | * protected by span_seqlock. It is a seqlock because it has | |
222 | * to be read outside of zone->lock, and it is done in the main | |
223 | * allocator path. But, it is written quite infrequently. | |
224 | * | |
225 | * The lock is declared along with zone->lock because it is | |
226 | * frequently read in proximity to zone->lock. It's good to | |
227 | * give them a chance of being in the same cacheline. | |
228 | */ | |
1da177e4 LT |
229 | unsigned long spanned_pages; /* total size, including holes */ |
230 | unsigned long present_pages; /* amount of memory (excluding holes) */ | |
231 | ||
232 | /* | |
233 | * rarely used fields: | |
234 | */ | |
235 | char *name; | |
236 | } ____cacheline_maxaligned_in_smp; | |
237 | ||
238 | ||
239 | /* | |
240 | * The "priority" of VM scanning is how much of the queues we will scan in one | |
241 | * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the | |
242 | * queues ("queue_length >> 12") during an aging round. | |
243 | */ | |
244 | #define DEF_PRIORITY 12 | |
245 | ||
246 | /* | |
247 | * One allocation request operates on a zonelist. A zonelist | |
248 | * is a list of zones, the first one is the 'goal' of the | |
249 | * allocation, the other zones are fallback zones, in decreasing | |
250 | * priority. | |
251 | * | |
252 | * Right now a zonelist takes up less than a cacheline. We never | |
253 | * modify it apart from boot-up, and only a few indices are used, | |
254 | * so despite the zonelist table being relatively big, the cache | |
255 | * footprint of this construct is very small. | |
256 | */ | |
257 | struct zonelist { | |
258 | struct zone *zones[MAX_NUMNODES * MAX_NR_ZONES + 1]; // NULL delimited | |
259 | }; | |
260 | ||
261 | ||
262 | /* | |
263 | * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM | |
264 | * (mostly NUMA machines?) to denote a higher-level memory zone than the | |
265 | * zone denotes. | |
266 | * | |
267 | * On NUMA machines, each NUMA node would have a pg_data_t to describe | |
268 | * it's memory layout. | |
269 | * | |
270 | * Memory statistics and page replacement data structures are maintained on a | |
271 | * per-zone basis. | |
272 | */ | |
273 | struct bootmem_data; | |
274 | typedef struct pglist_data { | |
275 | struct zone node_zones[MAX_NR_ZONES]; | |
276 | struct zonelist node_zonelists[GFP_ZONETYPES]; | |
277 | int nr_zones; | |
d41dee36 | 278 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 | 279 | struct page *node_mem_map; |
d41dee36 | 280 | #endif |
1da177e4 | 281 | struct bootmem_data *bdata; |
208d54e5 DH |
282 | #ifdef CONFIG_MEMORY_HOTPLUG |
283 | /* | |
284 | * Must be held any time you expect node_start_pfn, node_present_pages | |
285 | * or node_spanned_pages stay constant. Holding this will also | |
286 | * guarantee that any pfn_valid() stays that way. | |
287 | * | |
288 | * Nests above zone->lock and zone->size_seqlock. | |
289 | */ | |
290 | spinlock_t node_size_lock; | |
291 | #endif | |
1da177e4 LT |
292 | unsigned long node_start_pfn; |
293 | unsigned long node_present_pages; /* total number of physical pages */ | |
294 | unsigned long node_spanned_pages; /* total size of physical page | |
295 | range, including holes */ | |
296 | int node_id; | |
297 | struct pglist_data *pgdat_next; | |
298 | wait_queue_head_t kswapd_wait; | |
299 | struct task_struct *kswapd; | |
300 | int kswapd_max_order; | |
301 | } pg_data_t; | |
302 | ||
303 | #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages) | |
304 | #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages) | |
d41dee36 | 305 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
408fde81 | 306 | #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr)) |
d41dee36 AW |
307 | #else |
308 | #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr)) | |
309 | #endif | |
408fde81 | 310 | #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr)) |
1da177e4 | 311 | |
208d54e5 DH |
312 | #include <linux/memory_hotplug.h> |
313 | ||
1da177e4 LT |
314 | extern struct pglist_data *pgdat_list; |
315 | ||
316 | void __get_zone_counts(unsigned long *active, unsigned long *inactive, | |
317 | unsigned long *free, struct pglist_data *pgdat); | |
318 | void get_zone_counts(unsigned long *active, unsigned long *inactive, | |
319 | unsigned long *free); | |
320 | void build_all_zonelists(void); | |
321 | void wakeup_kswapd(struct zone *zone, int order); | |
322 | int zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
7fb1d9fc | 323 | int classzone_idx, int alloc_flags); |
1da177e4 LT |
324 | |
325 | #ifdef CONFIG_HAVE_MEMORY_PRESENT | |
326 | void memory_present(int nid, unsigned long start, unsigned long end); | |
327 | #else | |
328 | static inline void memory_present(int nid, unsigned long start, unsigned long end) {} | |
329 | #endif | |
330 | ||
331 | #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE | |
332 | unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long); | |
333 | #endif | |
334 | ||
335 | /* | |
336 | * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc. | |
337 | */ | |
338 | #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones) | |
339 | ||
340 | /** | |
341 | * for_each_pgdat - helper macro to iterate over all nodes | |
342 | * @pgdat - pointer to a pg_data_t variable | |
343 | * | |
344 | * Meant to help with common loops of the form | |
345 | * pgdat = pgdat_list; | |
346 | * while(pgdat) { | |
347 | * ... | |
348 | * pgdat = pgdat->pgdat_next; | |
349 | * } | |
350 | */ | |
351 | #define for_each_pgdat(pgdat) \ | |
352 | for (pgdat = pgdat_list; pgdat; pgdat = pgdat->pgdat_next) | |
353 | ||
354 | /* | |
355 | * next_zone - helper magic for for_each_zone() | |
356 | * Thanks to William Lee Irwin III for this piece of ingenuity. | |
357 | */ | |
358 | static inline struct zone *next_zone(struct zone *zone) | |
359 | { | |
360 | pg_data_t *pgdat = zone->zone_pgdat; | |
361 | ||
362 | if (zone < pgdat->node_zones + MAX_NR_ZONES - 1) | |
363 | zone++; | |
364 | else if (pgdat->pgdat_next) { | |
365 | pgdat = pgdat->pgdat_next; | |
366 | zone = pgdat->node_zones; | |
367 | } else | |
368 | zone = NULL; | |
369 | ||
370 | return zone; | |
371 | } | |
372 | ||
373 | /** | |
374 | * for_each_zone - helper macro to iterate over all memory zones | |
375 | * @zone - pointer to struct zone variable | |
376 | * | |
377 | * The user only needs to declare the zone variable, for_each_zone | |
378 | * fills it in. This basically means for_each_zone() is an | |
379 | * easier to read version of this piece of code: | |
380 | * | |
381 | * for (pgdat = pgdat_list; pgdat; pgdat = pgdat->node_next) | |
382 | * for (i = 0; i < MAX_NR_ZONES; ++i) { | |
383 | * struct zone * z = pgdat->node_zones + i; | |
384 | * ... | |
385 | * } | |
386 | * } | |
387 | */ | |
388 | #define for_each_zone(zone) \ | |
389 | for (zone = pgdat_list->node_zones; zone; zone = next_zone(zone)) | |
390 | ||
391 | static inline int is_highmem_idx(int idx) | |
392 | { | |
393 | return (idx == ZONE_HIGHMEM); | |
394 | } | |
395 | ||
396 | static inline int is_normal_idx(int idx) | |
397 | { | |
398 | return (idx == ZONE_NORMAL); | |
399 | } | |
400 | /** | |
401 | * is_highmem - helper function to quickly check if a struct zone is a | |
402 | * highmem zone or not. This is an attempt to keep references | |
403 | * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum. | |
404 | * @zone - pointer to struct zone variable | |
405 | */ | |
406 | static inline int is_highmem(struct zone *zone) | |
407 | { | |
408 | return zone == zone->zone_pgdat->node_zones + ZONE_HIGHMEM; | |
409 | } | |
410 | ||
411 | static inline int is_normal(struct zone *zone) | |
412 | { | |
413 | return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL; | |
414 | } | |
415 | ||
416 | /* These two functions are used to setup the per zone pages min values */ | |
417 | struct ctl_table; | |
418 | struct file; | |
419 | int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *, | |
420 | void __user *, size_t *, loff_t *); | |
421 | extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1]; | |
422 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *, | |
423 | void __user *, size_t *, loff_t *); | |
424 | ||
425 | #include <linux/topology.h> | |
426 | /* Returns the number of the current Node. */ | |
69d81fcd | 427 | #ifndef numa_node_id |
39c715b7 | 428 | #define numa_node_id() (cpu_to_node(raw_smp_processor_id())) |
69d81fcd | 429 | #endif |
1da177e4 | 430 | |
93b7504e | 431 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
432 | |
433 | extern struct pglist_data contig_page_data; | |
434 | #define NODE_DATA(nid) (&contig_page_data) | |
435 | #define NODE_MEM_MAP(nid) mem_map | |
436 | #define MAX_NODES_SHIFT 1 | |
1da177e4 | 437 | |
93b7504e | 438 | #else /* CONFIG_NEED_MULTIPLE_NODES */ |
1da177e4 LT |
439 | |
440 | #include <asm/mmzone.h> | |
441 | ||
93b7504e | 442 | #endif /* !CONFIG_NEED_MULTIPLE_NODES */ |
348f8b6c | 443 | |
d41dee36 AW |
444 | #ifdef CONFIG_SPARSEMEM |
445 | #include <asm/sparsemem.h> | |
446 | #endif | |
447 | ||
07808b74 | 448 | #if BITS_PER_LONG == 32 |
1da177e4 | 449 | /* |
a2f1b424 AK |
450 | * with 32 bit page->flags field, we reserve 9 bits for node/zone info. |
451 | * there are 4 zones (3 bits) and this leaves 9-3=6 bits for nodes. | |
1da177e4 | 452 | */ |
a2f1b424 | 453 | #define FLAGS_RESERVED 9 |
348f8b6c | 454 | |
1da177e4 LT |
455 | #elif BITS_PER_LONG == 64 |
456 | /* | |
457 | * with 64 bit flags field, there's plenty of room. | |
458 | */ | |
348f8b6c | 459 | #define FLAGS_RESERVED 32 |
1da177e4 | 460 | |
348f8b6c | 461 | #else |
1da177e4 | 462 | |
348f8b6c | 463 | #error BITS_PER_LONG not defined |
1da177e4 | 464 | |
1da177e4 LT |
465 | #endif |
466 | ||
b159d43f AW |
467 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
468 | #define early_pfn_to_nid(nid) (0UL) | |
469 | #endif | |
470 | ||
2bdaf115 AW |
471 | #ifdef CONFIG_FLATMEM |
472 | #define pfn_to_nid(pfn) (0) | |
473 | #endif | |
474 | ||
d41dee36 AW |
475 | #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT) |
476 | #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT) | |
477 | ||
478 | #ifdef CONFIG_SPARSEMEM | |
479 | ||
480 | /* | |
481 | * SECTION_SHIFT #bits space required to store a section # | |
482 | * | |
483 | * PA_SECTION_SHIFT physical address to/from section number | |
484 | * PFN_SECTION_SHIFT pfn to/from section number | |
485 | */ | |
486 | #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS) | |
487 | ||
488 | #define PA_SECTION_SHIFT (SECTION_SIZE_BITS) | |
489 | #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT) | |
490 | ||
491 | #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT) | |
492 | ||
493 | #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT) | |
494 | #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1)) | |
495 | ||
496 | #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS | |
497 | #error Allocator MAX_ORDER exceeds SECTION_SIZE | |
498 | #endif | |
499 | ||
500 | struct page; | |
501 | struct mem_section { | |
29751f69 AW |
502 | /* |
503 | * This is, logically, a pointer to an array of struct | |
504 | * pages. However, it is stored with some other magic. | |
505 | * (see sparse.c::sparse_init_one_section()) | |
506 | * | |
507 | * Making it a UL at least makes someone do a cast | |
508 | * before using it wrong. | |
509 | */ | |
510 | unsigned long section_mem_map; | |
d41dee36 AW |
511 | }; |
512 | ||
3e347261 BP |
513 | #ifdef CONFIG_SPARSEMEM_EXTREME |
514 | #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section)) | |
515 | #else | |
516 | #define SECTIONS_PER_ROOT 1 | |
517 | #endif | |
802f192e | 518 | |
3e347261 BP |
519 | #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT) |
520 | #define NR_SECTION_ROOTS (NR_MEM_SECTIONS / SECTIONS_PER_ROOT) | |
521 | #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1) | |
802f192e | 522 | |
3e347261 BP |
523 | #ifdef CONFIG_SPARSEMEM_EXTREME |
524 | extern struct mem_section *mem_section[NR_SECTION_ROOTS]; | |
802f192e | 525 | #else |
3e347261 BP |
526 | extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]; |
527 | #endif | |
d41dee36 | 528 | |
29751f69 AW |
529 | static inline struct mem_section *__nr_to_section(unsigned long nr) |
530 | { | |
3e347261 BP |
531 | if (!mem_section[SECTION_NR_TO_ROOT(nr)]) |
532 | return NULL; | |
533 | return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK]; | |
29751f69 | 534 | } |
4ca644d9 | 535 | extern int __section_nr(struct mem_section* ms); |
29751f69 AW |
536 | |
537 | /* | |
538 | * We use the lower bits of the mem_map pointer to store | |
539 | * a little bit of information. There should be at least | |
540 | * 3 bits here due to 32-bit alignment. | |
541 | */ | |
542 | #define SECTION_MARKED_PRESENT (1UL<<0) | |
543 | #define SECTION_HAS_MEM_MAP (1UL<<1) | |
544 | #define SECTION_MAP_LAST_BIT (1UL<<2) | |
545 | #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1)) | |
546 | ||
547 | static inline struct page *__section_mem_map_addr(struct mem_section *section) | |
548 | { | |
549 | unsigned long map = section->section_mem_map; | |
550 | map &= SECTION_MAP_MASK; | |
551 | return (struct page *)map; | |
552 | } | |
553 | ||
554 | static inline int valid_section(struct mem_section *section) | |
555 | { | |
802f192e | 556 | return (section && (section->section_mem_map & SECTION_MARKED_PRESENT)); |
29751f69 AW |
557 | } |
558 | ||
559 | static inline int section_has_mem_map(struct mem_section *section) | |
560 | { | |
802f192e | 561 | return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP)); |
29751f69 AW |
562 | } |
563 | ||
564 | static inline int valid_section_nr(unsigned long nr) | |
565 | { | |
566 | return valid_section(__nr_to_section(nr)); | |
567 | } | |
568 | ||
d41dee36 AW |
569 | static inline struct mem_section *__pfn_to_section(unsigned long pfn) |
570 | { | |
29751f69 | 571 | return __nr_to_section(pfn_to_section_nr(pfn)); |
d41dee36 AW |
572 | } |
573 | ||
574 | #define pfn_to_page(pfn) \ | |
575 | ({ \ | |
576 | unsigned long __pfn = (pfn); \ | |
29751f69 | 577 | __section_mem_map_addr(__pfn_to_section(__pfn)) + __pfn; \ |
d41dee36 AW |
578 | }) |
579 | #define page_to_pfn(page) \ | |
580 | ({ \ | |
29751f69 AW |
581 | page - __section_mem_map_addr(__nr_to_section( \ |
582 | page_to_section(page))); \ | |
d41dee36 AW |
583 | }) |
584 | ||
585 | static inline int pfn_valid(unsigned long pfn) | |
586 | { | |
587 | if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) | |
588 | return 0; | |
29751f69 | 589 | return valid_section(__nr_to_section(pfn_to_section_nr(pfn))); |
d41dee36 AW |
590 | } |
591 | ||
592 | /* | |
593 | * These are _only_ used during initialisation, therefore they | |
594 | * can use __initdata ... They could have names to indicate | |
595 | * this restriction. | |
596 | */ | |
597 | #ifdef CONFIG_NUMA | |
161599ff AW |
598 | #define pfn_to_nid(pfn) \ |
599 | ({ \ | |
600 | unsigned long __pfn_to_nid_pfn = (pfn); \ | |
601 | page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \ | |
602 | }) | |
2bdaf115 AW |
603 | #else |
604 | #define pfn_to_nid(pfn) (0) | |
d41dee36 AW |
605 | #endif |
606 | ||
d41dee36 AW |
607 | #define early_pfn_valid(pfn) pfn_valid(pfn) |
608 | void sparse_init(void); | |
609 | #else | |
610 | #define sparse_init() do {} while (0) | |
28ae55c9 | 611 | #define sparse_index_init(_sec, _nid) do {} while (0) |
d41dee36 AW |
612 | #endif /* CONFIG_SPARSEMEM */ |
613 | ||
614 | #ifndef early_pfn_valid | |
615 | #define early_pfn_valid(pfn) (1) | |
616 | #endif | |
617 | ||
618 | void memory_present(int nid, unsigned long start, unsigned long end); | |
619 | unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long); | |
620 | ||
1da177e4 LT |
621 | #endif /* !__ASSEMBLY__ */ |
622 | #endif /* __KERNEL__ */ | |
623 | #endif /* _LINUX_MMZONE_H */ |