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 */ | |
49 | int low; /* low watermark, refill needed */ | |
50 | int high; /* high watermark, emptying needed */ | |
51 | int batch; /* chunk size for buddy add/remove */ | |
52 | struct list_head list; /* the list of pages */ | |
53 | }; | |
54 | ||
55 | struct per_cpu_pageset { | |
56 | struct per_cpu_pages pcp[2]; /* 0: hot. 1: cold */ | |
57 | #ifdef CONFIG_NUMA | |
58 | unsigned long numa_hit; /* allocated in intended node */ | |
59 | unsigned long numa_miss; /* allocated in non intended node */ | |
60 | unsigned long numa_foreign; /* was intended here, hit elsewhere */ | |
61 | unsigned long interleave_hit; /* interleaver prefered this zone */ | |
62 | unsigned long local_node; /* allocation from local node */ | |
63 | unsigned long other_node; /* allocation from other node */ | |
64 | #endif | |
65 | } ____cacheline_aligned_in_smp; | |
66 | ||
e7c8d5c9 CL |
67 | #ifdef CONFIG_NUMA |
68 | #define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)]) | |
69 | #else | |
70 | #define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)]) | |
71 | #endif | |
72 | ||
1da177e4 | 73 | #define ZONE_DMA 0 |
a2f1b424 AK |
74 | #define ZONE_DMA32 1 |
75 | #define ZONE_NORMAL 2 | |
76 | #define ZONE_HIGHMEM 3 | |
1da177e4 | 77 | |
a2f1b424 | 78 | #define MAX_NR_ZONES 4 /* Sync this with ZONES_SHIFT */ |
1da177e4 LT |
79 | #define ZONES_SHIFT 2 /* ceil(log2(MAX_NR_ZONES)) */ |
80 | ||
81 | ||
82 | /* | |
83 | * When a memory allocation must conform to specific limitations (such | |
84 | * as being suitable for DMA) the caller will pass in hints to the | |
85 | * allocator in the gfp_mask, in the zone modifier bits. These bits | |
86 | * are used to select a priority ordered list of memory zones which | |
87 | * match the requested limits. GFP_ZONEMASK defines which bits within | |
88 | * the gfp_mask should be considered as zone modifiers. Each valid | |
89 | * combination of the zone modifier bits has a corresponding list | |
90 | * of zones (in node_zonelists). Thus for two zone modifiers there | |
91 | * will be a maximum of 4 (2 ** 2) zonelists, for 3 modifiers there will | |
92 | * be 8 (2 ** 3) zonelists. GFP_ZONETYPES defines the number of possible | |
93 | * combinations of zone modifiers in "zone modifier space". | |
94 | */ | |
95 | #define GFP_ZONEMASK 0x03 | |
96 | /* | |
97 | * As an optimisation any zone modifier bits which are only valid when | |
98 | * no other zone modifier bits are set (loners) should be placed in | |
99 | * the highest order bits of this field. This allows us to reduce the | |
100 | * extent of the zonelists thus saving space. For example in the case | |
101 | * of three zone modifier bits, we could require up to eight zonelists. | |
102 | * If the left most zone modifier is a "loner" then the highest valid | |
103 | * zonelist would be four allowing us to allocate only five zonelists. | |
104 | * Use the first form when the left most bit is not a "loner", otherwise | |
105 | * use the second. | |
106 | */ | |
107 | /* #define GFP_ZONETYPES (GFP_ZONEMASK + 1) */ /* Non-loner */ | |
108 | #define GFP_ZONETYPES ((GFP_ZONEMASK + 1) / 2 + 1) /* Loner */ | |
109 | ||
110 | /* | |
111 | * On machines where it is needed (eg PCs) we divide physical memory | |
a2f1b424 | 112 | * into multiple physical zones. On a PC we have 4 zones: |
1da177e4 LT |
113 | * |
114 | * ZONE_DMA < 16 MB ISA DMA capable memory | |
a2f1b424 | 115 | * ZONE_DMA32 0 MB Empty |
1da177e4 LT |
116 | * ZONE_NORMAL 16-896 MB direct mapped by the kernel |
117 | * ZONE_HIGHMEM > 896 MB only page cache and user processes | |
118 | */ | |
119 | ||
120 | struct zone { | |
121 | /* Fields commonly accessed by the page allocator */ | |
122 | unsigned long free_pages; | |
123 | unsigned long pages_min, pages_low, pages_high; | |
124 | /* | |
125 | * We don't know if the memory that we're going to allocate will be freeable | |
126 | * or/and it will be released eventually, so to avoid totally wasting several | |
127 | * GB of ram we must reserve some of the lower zone memory (otherwise we risk | |
128 | * to run OOM on the lower zones despite there's tons of freeable ram | |
129 | * on the higher zones). This array is recalculated at runtime if the | |
130 | * sysctl_lowmem_reserve_ratio sysctl changes. | |
131 | */ | |
132 | unsigned long lowmem_reserve[MAX_NR_ZONES]; | |
133 | ||
e7c8d5c9 CL |
134 | #ifdef CONFIG_NUMA |
135 | struct per_cpu_pageset *pageset[NR_CPUS]; | |
136 | #else | |
1da177e4 | 137 | struct per_cpu_pageset pageset[NR_CPUS]; |
e7c8d5c9 | 138 | #endif |
1da177e4 LT |
139 | /* |
140 | * free areas of different sizes | |
141 | */ | |
142 | spinlock_t lock; | |
bdc8cb98 DH |
143 | #ifdef CONFIG_MEMORY_HOTPLUG |
144 | /* see spanned/present_pages for more description */ | |
145 | seqlock_t span_seqlock; | |
146 | #endif | |
1da177e4 LT |
147 | struct free_area free_area[MAX_ORDER]; |
148 | ||
149 | ||
150 | ZONE_PADDING(_pad1_) | |
151 | ||
152 | /* Fields commonly accessed by the page reclaim scanner */ | |
153 | spinlock_t lru_lock; | |
154 | struct list_head active_list; | |
155 | struct list_head inactive_list; | |
156 | unsigned long nr_scan_active; | |
157 | unsigned long nr_scan_inactive; | |
158 | unsigned long nr_active; | |
159 | unsigned long nr_inactive; | |
160 | unsigned long pages_scanned; /* since last reclaim */ | |
161 | int all_unreclaimable; /* All pages pinned */ | |
162 | ||
753ee728 MH |
163 | /* |
164 | * Does the allocator try to reclaim pages from the zone as soon | |
165 | * as it fails a watermark_ok() in __alloc_pages? | |
166 | */ | |
167 | int reclaim_pages; | |
1e7e5a90 MH |
168 | /* A count of how many reclaimers are scanning this zone */ |
169 | atomic_t reclaim_in_progress; | |
753ee728 | 170 | |
1da177e4 LT |
171 | /* |
172 | * prev_priority holds the scanning priority for this zone. It is | |
173 | * defined as the scanning priority at which we achieved our reclaim | |
174 | * target at the previous try_to_free_pages() or balance_pgdat() | |
175 | * invokation. | |
176 | * | |
177 | * We use prev_priority as a measure of how much stress page reclaim is | |
178 | * under - it drives the swappiness decision: whether to unmap mapped | |
179 | * pages. | |
180 | * | |
181 | * temp_priority is used to remember the scanning priority at which | |
182 | * this zone was successfully refilled to free_pages == pages_high. | |
183 | * | |
184 | * Access to both these fields is quite racy even on uniprocessor. But | |
185 | * it is expected to average out OK. | |
186 | */ | |
187 | int temp_priority; | |
188 | int prev_priority; | |
189 | ||
190 | ||
191 | ZONE_PADDING(_pad2_) | |
192 | /* Rarely used or read-mostly fields */ | |
193 | ||
194 | /* | |
195 | * wait_table -- the array holding the hash table | |
196 | * wait_table_size -- the size of the hash table array | |
197 | * wait_table_bits -- wait_table_size == (1 << wait_table_bits) | |
198 | * | |
199 | * The purpose of all these is to keep track of the people | |
200 | * waiting for a page to become available and make them | |
201 | * runnable again when possible. The trouble is that this | |
202 | * consumes a lot of space, especially when so few things | |
203 | * wait on pages at a given time. So instead of using | |
204 | * per-page waitqueues, we use a waitqueue hash table. | |
205 | * | |
206 | * The bucket discipline is to sleep on the same queue when | |
207 | * colliding and wake all in that wait queue when removing. | |
208 | * When something wakes, it must check to be sure its page is | |
209 | * truly available, a la thundering herd. The cost of a | |
210 | * collision is great, but given the expected load of the | |
211 | * table, they should be so rare as to be outweighed by the | |
212 | * benefits from the saved space. | |
213 | * | |
214 | * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the | |
215 | * primary users of these fields, and in mm/page_alloc.c | |
216 | * free_area_init_core() performs the initialization of them. | |
217 | */ | |
218 | wait_queue_head_t * wait_table; | |
219 | unsigned long wait_table_size; | |
220 | unsigned long wait_table_bits; | |
221 | ||
222 | /* | |
223 | * Discontig memory support fields. | |
224 | */ | |
225 | struct pglist_data *zone_pgdat; | |
226 | struct page *zone_mem_map; | |
227 | /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */ | |
228 | unsigned long zone_start_pfn; | |
229 | ||
bdc8cb98 DH |
230 | /* |
231 | * zone_start_pfn, spanned_pages and present_pages are all | |
232 | * protected by span_seqlock. It is a seqlock because it has | |
233 | * to be read outside of zone->lock, and it is done in the main | |
234 | * allocator path. But, it is written quite infrequently. | |
235 | * | |
236 | * The lock is declared along with zone->lock because it is | |
237 | * frequently read in proximity to zone->lock. It's good to | |
238 | * give them a chance of being in the same cacheline. | |
239 | */ | |
1da177e4 LT |
240 | unsigned long spanned_pages; /* total size, including holes */ |
241 | unsigned long present_pages; /* amount of memory (excluding holes) */ | |
242 | ||
243 | /* | |
244 | * rarely used fields: | |
245 | */ | |
246 | char *name; | |
247 | } ____cacheline_maxaligned_in_smp; | |
248 | ||
249 | ||
250 | /* | |
251 | * The "priority" of VM scanning is how much of the queues we will scan in one | |
252 | * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the | |
253 | * queues ("queue_length >> 12") during an aging round. | |
254 | */ | |
255 | #define DEF_PRIORITY 12 | |
256 | ||
257 | /* | |
258 | * One allocation request operates on a zonelist. A zonelist | |
259 | * is a list of zones, the first one is the 'goal' of the | |
260 | * allocation, the other zones are fallback zones, in decreasing | |
261 | * priority. | |
262 | * | |
263 | * Right now a zonelist takes up less than a cacheline. We never | |
264 | * modify it apart from boot-up, and only a few indices are used, | |
265 | * so despite the zonelist table being relatively big, the cache | |
266 | * footprint of this construct is very small. | |
267 | */ | |
268 | struct zonelist { | |
269 | struct zone *zones[MAX_NUMNODES * MAX_NR_ZONES + 1]; // NULL delimited | |
270 | }; | |
271 | ||
272 | ||
273 | /* | |
274 | * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM | |
275 | * (mostly NUMA machines?) to denote a higher-level memory zone than the | |
276 | * zone denotes. | |
277 | * | |
278 | * On NUMA machines, each NUMA node would have a pg_data_t to describe | |
279 | * it's memory layout. | |
280 | * | |
281 | * Memory statistics and page replacement data structures are maintained on a | |
282 | * per-zone basis. | |
283 | */ | |
284 | struct bootmem_data; | |
285 | typedef struct pglist_data { | |
286 | struct zone node_zones[MAX_NR_ZONES]; | |
287 | struct zonelist node_zonelists[GFP_ZONETYPES]; | |
288 | int nr_zones; | |
d41dee36 | 289 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 | 290 | struct page *node_mem_map; |
d41dee36 | 291 | #endif |
1da177e4 | 292 | struct bootmem_data *bdata; |
208d54e5 DH |
293 | #ifdef CONFIG_MEMORY_HOTPLUG |
294 | /* | |
295 | * Must be held any time you expect node_start_pfn, node_present_pages | |
296 | * or node_spanned_pages stay constant. Holding this will also | |
297 | * guarantee that any pfn_valid() stays that way. | |
298 | * | |
299 | * Nests above zone->lock and zone->size_seqlock. | |
300 | */ | |
301 | spinlock_t node_size_lock; | |
302 | #endif | |
1da177e4 LT |
303 | unsigned long node_start_pfn; |
304 | unsigned long node_present_pages; /* total number of physical pages */ | |
305 | unsigned long node_spanned_pages; /* total size of physical page | |
306 | range, including holes */ | |
307 | int node_id; | |
308 | struct pglist_data *pgdat_next; | |
309 | wait_queue_head_t kswapd_wait; | |
310 | struct task_struct *kswapd; | |
311 | int kswapd_max_order; | |
312 | } pg_data_t; | |
313 | ||
314 | #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages) | |
315 | #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages) | |
d41dee36 | 316 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
408fde81 | 317 | #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr)) |
d41dee36 AW |
318 | #else |
319 | #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr)) | |
320 | #endif | |
408fde81 | 321 | #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr)) |
1da177e4 | 322 | |
208d54e5 DH |
323 | #include <linux/memory_hotplug.h> |
324 | ||
1da177e4 LT |
325 | extern struct pglist_data *pgdat_list; |
326 | ||
327 | void __get_zone_counts(unsigned long *active, unsigned long *inactive, | |
328 | unsigned long *free, struct pglist_data *pgdat); | |
329 | void get_zone_counts(unsigned long *active, unsigned long *inactive, | |
330 | unsigned long *free); | |
331 | void build_all_zonelists(void); | |
332 | void wakeup_kswapd(struct zone *zone, int order); | |
333 | int zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
260b2367 | 334 | int alloc_type, int can_try_harder, gfp_t gfp_high); |
1da177e4 LT |
335 | |
336 | #ifdef CONFIG_HAVE_MEMORY_PRESENT | |
337 | void memory_present(int nid, unsigned long start, unsigned long end); | |
338 | #else | |
339 | static inline void memory_present(int nid, unsigned long start, unsigned long end) {} | |
340 | #endif | |
341 | ||
342 | #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE | |
343 | unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long); | |
344 | #endif | |
345 | ||
346 | /* | |
347 | * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc. | |
348 | */ | |
349 | #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones) | |
350 | ||
351 | /** | |
352 | * for_each_pgdat - helper macro to iterate over all nodes | |
353 | * @pgdat - pointer to a pg_data_t variable | |
354 | * | |
355 | * Meant to help with common loops of the form | |
356 | * pgdat = pgdat_list; | |
357 | * while(pgdat) { | |
358 | * ... | |
359 | * pgdat = pgdat->pgdat_next; | |
360 | * } | |
361 | */ | |
362 | #define for_each_pgdat(pgdat) \ | |
363 | for (pgdat = pgdat_list; pgdat; pgdat = pgdat->pgdat_next) | |
364 | ||
365 | /* | |
366 | * next_zone - helper magic for for_each_zone() | |
367 | * Thanks to William Lee Irwin III for this piece of ingenuity. | |
368 | */ | |
369 | static inline struct zone *next_zone(struct zone *zone) | |
370 | { | |
371 | pg_data_t *pgdat = zone->zone_pgdat; | |
372 | ||
373 | if (zone < pgdat->node_zones + MAX_NR_ZONES - 1) | |
374 | zone++; | |
375 | else if (pgdat->pgdat_next) { | |
376 | pgdat = pgdat->pgdat_next; | |
377 | zone = pgdat->node_zones; | |
378 | } else | |
379 | zone = NULL; | |
380 | ||
381 | return zone; | |
382 | } | |
383 | ||
384 | /** | |
385 | * for_each_zone - helper macro to iterate over all memory zones | |
386 | * @zone - pointer to struct zone variable | |
387 | * | |
388 | * The user only needs to declare the zone variable, for_each_zone | |
389 | * fills it in. This basically means for_each_zone() is an | |
390 | * easier to read version of this piece of code: | |
391 | * | |
392 | * for (pgdat = pgdat_list; pgdat; pgdat = pgdat->node_next) | |
393 | * for (i = 0; i < MAX_NR_ZONES; ++i) { | |
394 | * struct zone * z = pgdat->node_zones + i; | |
395 | * ... | |
396 | * } | |
397 | * } | |
398 | */ | |
399 | #define for_each_zone(zone) \ | |
400 | for (zone = pgdat_list->node_zones; zone; zone = next_zone(zone)) | |
401 | ||
402 | static inline int is_highmem_idx(int idx) | |
403 | { | |
404 | return (idx == ZONE_HIGHMEM); | |
405 | } | |
406 | ||
407 | static inline int is_normal_idx(int idx) | |
408 | { | |
409 | return (idx == ZONE_NORMAL); | |
410 | } | |
411 | /** | |
412 | * is_highmem - helper function to quickly check if a struct zone is a | |
413 | * highmem zone or not. This is an attempt to keep references | |
414 | * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum. | |
415 | * @zone - pointer to struct zone variable | |
416 | */ | |
417 | static inline int is_highmem(struct zone *zone) | |
418 | { | |
419 | return zone == zone->zone_pgdat->node_zones + ZONE_HIGHMEM; | |
420 | } | |
421 | ||
422 | static inline int is_normal(struct zone *zone) | |
423 | { | |
424 | return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL; | |
425 | } | |
426 | ||
427 | /* These two functions are used to setup the per zone pages min values */ | |
428 | struct ctl_table; | |
429 | struct file; | |
430 | int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *, | |
431 | void __user *, size_t *, loff_t *); | |
432 | extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1]; | |
433 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *, | |
434 | void __user *, size_t *, loff_t *); | |
435 | ||
436 | #include <linux/topology.h> | |
437 | /* Returns the number of the current Node. */ | |
39c715b7 | 438 | #define numa_node_id() (cpu_to_node(raw_smp_processor_id())) |
1da177e4 | 439 | |
93b7504e | 440 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
441 | |
442 | extern struct pglist_data contig_page_data; | |
443 | #define NODE_DATA(nid) (&contig_page_data) | |
444 | #define NODE_MEM_MAP(nid) mem_map | |
445 | #define MAX_NODES_SHIFT 1 | |
446 | #define pfn_to_nid(pfn) (0) | |
447 | ||
93b7504e | 448 | #else /* CONFIG_NEED_MULTIPLE_NODES */ |
1da177e4 LT |
449 | |
450 | #include <asm/mmzone.h> | |
451 | ||
93b7504e | 452 | #endif /* !CONFIG_NEED_MULTIPLE_NODES */ |
348f8b6c | 453 | |
d41dee36 AW |
454 | #ifdef CONFIG_SPARSEMEM |
455 | #include <asm/sparsemem.h> | |
456 | #endif | |
457 | ||
07808b74 | 458 | #if BITS_PER_LONG == 32 |
1da177e4 | 459 | /* |
a2f1b424 AK |
460 | * with 32 bit page->flags field, we reserve 9 bits for node/zone info. |
461 | * there are 4 zones (3 bits) and this leaves 9-3=6 bits for nodes. | |
1da177e4 | 462 | */ |
a2f1b424 | 463 | #define FLAGS_RESERVED 9 |
348f8b6c | 464 | |
1da177e4 LT |
465 | #elif BITS_PER_LONG == 64 |
466 | /* | |
467 | * with 64 bit flags field, there's plenty of room. | |
468 | */ | |
348f8b6c | 469 | #define FLAGS_RESERVED 32 |
1da177e4 | 470 | |
348f8b6c | 471 | #else |
1da177e4 | 472 | |
348f8b6c | 473 | #error BITS_PER_LONG not defined |
1da177e4 | 474 | |
1da177e4 LT |
475 | #endif |
476 | ||
b159d43f AW |
477 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
478 | #define early_pfn_to_nid(nid) (0UL) | |
479 | #endif | |
480 | ||
d41dee36 AW |
481 | #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT) |
482 | #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT) | |
483 | ||
484 | #ifdef CONFIG_SPARSEMEM | |
485 | ||
486 | /* | |
487 | * SECTION_SHIFT #bits space required to store a section # | |
488 | * | |
489 | * PA_SECTION_SHIFT physical address to/from section number | |
490 | * PFN_SECTION_SHIFT pfn to/from section number | |
491 | */ | |
492 | #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS) | |
493 | ||
494 | #define PA_SECTION_SHIFT (SECTION_SIZE_BITS) | |
495 | #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT) | |
496 | ||
497 | #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT) | |
498 | ||
499 | #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT) | |
500 | #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1)) | |
501 | ||
502 | #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS | |
503 | #error Allocator MAX_ORDER exceeds SECTION_SIZE | |
504 | #endif | |
505 | ||
506 | struct page; | |
507 | struct mem_section { | |
29751f69 AW |
508 | /* |
509 | * This is, logically, a pointer to an array of struct | |
510 | * pages. However, it is stored with some other magic. | |
511 | * (see sparse.c::sparse_init_one_section()) | |
512 | * | |
513 | * Making it a UL at least makes someone do a cast | |
514 | * before using it wrong. | |
515 | */ | |
516 | unsigned long section_mem_map; | |
d41dee36 AW |
517 | }; |
518 | ||
3e347261 BP |
519 | #ifdef CONFIG_SPARSEMEM_EXTREME |
520 | #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section)) | |
521 | #else | |
522 | #define SECTIONS_PER_ROOT 1 | |
523 | #endif | |
802f192e | 524 | |
3e347261 BP |
525 | #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT) |
526 | #define NR_SECTION_ROOTS (NR_MEM_SECTIONS / SECTIONS_PER_ROOT) | |
527 | #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1) | |
802f192e | 528 | |
3e347261 BP |
529 | #ifdef CONFIG_SPARSEMEM_EXTREME |
530 | extern struct mem_section *mem_section[NR_SECTION_ROOTS]; | |
802f192e | 531 | #else |
3e347261 BP |
532 | extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]; |
533 | #endif | |
d41dee36 | 534 | |
29751f69 AW |
535 | static inline struct mem_section *__nr_to_section(unsigned long nr) |
536 | { | |
3e347261 BP |
537 | if (!mem_section[SECTION_NR_TO_ROOT(nr)]) |
538 | return NULL; | |
539 | return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK]; | |
29751f69 | 540 | } |
4ca644d9 | 541 | extern int __section_nr(struct mem_section* ms); |
29751f69 AW |
542 | |
543 | /* | |
544 | * We use the lower bits of the mem_map pointer to store | |
545 | * a little bit of information. There should be at least | |
546 | * 3 bits here due to 32-bit alignment. | |
547 | */ | |
548 | #define SECTION_MARKED_PRESENT (1UL<<0) | |
549 | #define SECTION_HAS_MEM_MAP (1UL<<1) | |
550 | #define SECTION_MAP_LAST_BIT (1UL<<2) | |
551 | #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1)) | |
552 | ||
553 | static inline struct page *__section_mem_map_addr(struct mem_section *section) | |
554 | { | |
555 | unsigned long map = section->section_mem_map; | |
556 | map &= SECTION_MAP_MASK; | |
557 | return (struct page *)map; | |
558 | } | |
559 | ||
560 | static inline int valid_section(struct mem_section *section) | |
561 | { | |
802f192e | 562 | return (section && (section->section_mem_map & SECTION_MARKED_PRESENT)); |
29751f69 AW |
563 | } |
564 | ||
565 | static inline int section_has_mem_map(struct mem_section *section) | |
566 | { | |
802f192e | 567 | return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP)); |
29751f69 AW |
568 | } |
569 | ||
570 | static inline int valid_section_nr(unsigned long nr) | |
571 | { | |
572 | return valid_section(__nr_to_section(nr)); | |
573 | } | |
574 | ||
d41dee36 AW |
575 | /* |
576 | * Given a kernel address, find the home node of the underlying memory. | |
577 | */ | |
578 | #define kvaddr_to_nid(kaddr) pfn_to_nid(__pa(kaddr) >> PAGE_SHIFT) | |
579 | ||
580 | static inline struct mem_section *__pfn_to_section(unsigned long pfn) | |
581 | { | |
29751f69 | 582 | return __nr_to_section(pfn_to_section_nr(pfn)); |
d41dee36 AW |
583 | } |
584 | ||
585 | #define pfn_to_page(pfn) \ | |
586 | ({ \ | |
587 | unsigned long __pfn = (pfn); \ | |
29751f69 | 588 | __section_mem_map_addr(__pfn_to_section(__pfn)) + __pfn; \ |
d41dee36 AW |
589 | }) |
590 | #define page_to_pfn(page) \ | |
591 | ({ \ | |
29751f69 AW |
592 | page - __section_mem_map_addr(__nr_to_section( \ |
593 | page_to_section(page))); \ | |
d41dee36 AW |
594 | }) |
595 | ||
596 | static inline int pfn_valid(unsigned long pfn) | |
597 | { | |
598 | if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) | |
599 | return 0; | |
29751f69 | 600 | return valid_section(__nr_to_section(pfn_to_section_nr(pfn))); |
d41dee36 AW |
601 | } |
602 | ||
603 | /* | |
604 | * These are _only_ used during initialisation, therefore they | |
605 | * can use __initdata ... They could have names to indicate | |
606 | * this restriction. | |
607 | */ | |
608 | #ifdef CONFIG_NUMA | |
609 | #define pfn_to_nid early_pfn_to_nid | |
610 | #endif | |
611 | ||
612 | #define pfn_to_pgdat(pfn) \ | |
613 | ({ \ | |
614 | NODE_DATA(pfn_to_nid(pfn)); \ | |
615 | }) | |
616 | ||
617 | #define early_pfn_valid(pfn) pfn_valid(pfn) | |
618 | void sparse_init(void); | |
619 | #else | |
620 | #define sparse_init() do {} while (0) | |
28ae55c9 | 621 | #define sparse_index_init(_sec, _nid) do {} while (0) |
d41dee36 AW |
622 | #endif /* CONFIG_SPARSEMEM */ |
623 | ||
641c7673 AW |
624 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES |
625 | #define early_pfn_in_nid(pfn, nid) (early_pfn_to_nid(pfn) == (nid)) | |
626 | #else | |
627 | #define early_pfn_in_nid(pfn, nid) (1) | |
628 | #endif | |
629 | ||
d41dee36 AW |
630 | #ifndef early_pfn_valid |
631 | #define early_pfn_valid(pfn) (1) | |
632 | #endif | |
633 | ||
634 | void memory_present(int nid, unsigned long start, unsigned long end); | |
635 | unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long); | |
636 | ||
1da177e4 LT |
637 | #endif /* !__ASSEMBLY__ */ |
638 | #endif /* __KERNEL__ */ | |
639 | #endif /* _LINUX_MMZONE_H */ |