]>
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> |
8357f869 | 16 | #include <linux/nodemask.h> |
1da177e4 LT |
17 | #include <asm/atomic.h> |
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 | |
27 | struct free_area { | |
28 | struct list_head free_list; | |
29 | unsigned long nr_free; | |
30 | }; | |
31 | ||
32 | struct 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) | |
41 | struct 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 | ||
49 | struct per_cpu_pages { | |
50 | int count; /* number of pages in the list */ | |
1da177e4 LT |
51 | int high; /* high watermark, emptying needed */ |
52 | int batch; /* chunk size for buddy add/remove */ | |
53 | struct list_head list; /* the list of pages */ | |
54 | }; | |
55 | ||
56 | struct per_cpu_pageset { | |
57 | struct per_cpu_pages pcp[2]; /* 0: hot. 1: cold */ | |
58 | #ifdef CONFIG_NUMA | |
59 | unsigned long numa_hit; /* allocated in intended node */ | |
60 | unsigned long numa_miss; /* allocated in non intended node */ | |
61 | unsigned long numa_foreign; /* was intended here, hit elsewhere */ | |
62 | unsigned long interleave_hit; /* interleaver prefered this zone */ | |
63 | unsigned long local_node; /* allocation from local node */ | |
64 | unsigned long other_node; /* allocation from other node */ | |
65 | #endif | |
66 | } ____cacheline_aligned_in_smp; | |
67 | ||
e7c8d5c9 CL |
68 | #ifdef CONFIG_NUMA |
69 | #define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)]) | |
70 | #else | |
71 | #define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)]) | |
72 | #endif | |
73 | ||
1da177e4 | 74 | #define ZONE_DMA 0 |
a2f1b424 AK |
75 | #define ZONE_DMA32 1 |
76 | #define ZONE_NORMAL 2 | |
77 | #define ZONE_HIGHMEM 3 | |
1da177e4 | 78 | |
a2f1b424 | 79 | #define MAX_NR_ZONES 4 /* Sync this with ZONES_SHIFT */ |
1da177e4 LT |
80 | #define ZONES_SHIFT 2 /* ceil(log2(MAX_NR_ZONES)) */ |
81 | ||
82 | ||
83 | /* | |
84 | * When a memory allocation must conform to specific limitations (such | |
85 | * as being suitable for DMA) the caller will pass in hints to the | |
86 | * allocator in the gfp_mask, in the zone modifier bits. These bits | |
87 | * are used to select a priority ordered list of memory zones which | |
88 | * match the requested limits. GFP_ZONEMASK defines which bits within | |
89 | * the gfp_mask should be considered as zone modifiers. Each valid | |
90 | * combination of the zone modifier bits has a corresponding list | |
91 | * of zones (in node_zonelists). Thus for two zone modifiers there | |
92 | * will be a maximum of 4 (2 ** 2) zonelists, for 3 modifiers there will | |
93 | * be 8 (2 ** 3) zonelists. GFP_ZONETYPES defines the number of possible | |
94 | * combinations of zone modifiers in "zone modifier space". | |
ac3461ad | 95 | * |
79046ae0 AW |
96 | * As an optimisation any zone modifier bits which are only valid when |
97 | * no other zone modifier bits are set (loners) should be placed in | |
98 | * the highest order bits of this field. This allows us to reduce the | |
99 | * extent of the zonelists thus saving space. For example in the case | |
100 | * of three zone modifier bits, we could require up to eight zonelists. | |
101 | * If the left most zone modifier is a "loner" then the highest valid | |
102 | * zonelist would be four allowing us to allocate only five zonelists. | |
ce2ea89b AW |
103 | * Use the first form for GFP_ZONETYPES when the left most bit is not |
104 | * a "loner", otherwise use the second. | |
79046ae0 | 105 | * |
ac3461ad | 106 | * NOTE! Make sure this matches the zones in <linux/gfp.h> |
1da177e4 | 107 | */ |
ac3461ad | 108 | #define GFP_ZONEMASK 0x07 |
ce2ea89b AW |
109 | /* #define GFP_ZONETYPES (GFP_ZONEMASK + 1) */ /* Non-loner */ |
110 | #define GFP_ZONETYPES ((GFP_ZONEMASK + 1) / 2 + 1) /* Loner */ | |
1da177e4 LT |
111 | |
112 | /* | |
113 | * On machines where it is needed (eg PCs) we divide physical memory | |
1f6818b9 | 114 | * into multiple physical zones. On a 32bit PC we have 4 zones: |
1da177e4 LT |
115 | * |
116 | * ZONE_DMA < 16 MB ISA DMA capable memory | |
a2f1b424 | 117 | * ZONE_DMA32 0 MB Empty |
1da177e4 LT |
118 | * ZONE_NORMAL 16-896 MB direct mapped by the kernel |
119 | * ZONE_HIGHMEM > 896 MB only page cache and user processes | |
120 | */ | |
121 | ||
122 | struct zone { | |
123 | /* Fields commonly accessed by the page allocator */ | |
124 | unsigned long free_pages; | |
125 | unsigned long pages_min, pages_low, pages_high; | |
126 | /* | |
127 | * We don't know if the memory that we're going to allocate will be freeable | |
128 | * or/and it will be released eventually, so to avoid totally wasting several | |
129 | * GB of ram we must reserve some of the lower zone memory (otherwise we risk | |
130 | * to run OOM on the lower zones despite there's tons of freeable ram | |
131 | * on the higher zones). This array is recalculated at runtime if the | |
132 | * sysctl_lowmem_reserve_ratio sysctl changes. | |
133 | */ | |
134 | unsigned long lowmem_reserve[MAX_NR_ZONES]; | |
135 | ||
e7c8d5c9 CL |
136 | #ifdef CONFIG_NUMA |
137 | struct per_cpu_pageset *pageset[NR_CPUS]; | |
138 | #else | |
1da177e4 | 139 | struct per_cpu_pageset pageset[NR_CPUS]; |
e7c8d5c9 | 140 | #endif |
1da177e4 LT |
141 | /* |
142 | * free areas of different sizes | |
143 | */ | |
144 | spinlock_t lock; | |
bdc8cb98 DH |
145 | #ifdef CONFIG_MEMORY_HOTPLUG |
146 | /* see spanned/present_pages for more description */ | |
147 | seqlock_t span_seqlock; | |
148 | #endif | |
1da177e4 LT |
149 | struct free_area free_area[MAX_ORDER]; |
150 | ||
151 | ||
152 | ZONE_PADDING(_pad1_) | |
153 | ||
154 | /* Fields commonly accessed by the page reclaim scanner */ | |
155 | spinlock_t lru_lock; | |
156 | struct list_head active_list; | |
157 | struct list_head inactive_list; | |
158 | unsigned long nr_scan_active; | |
159 | unsigned long nr_scan_inactive; | |
160 | unsigned long nr_active; | |
161 | unsigned long nr_inactive; | |
162 | unsigned long pages_scanned; /* since last reclaim */ | |
163 | int all_unreclaimable; /* All pages pinned */ | |
164 | ||
1e7e5a90 MH |
165 | /* A count of how many reclaimers are scanning this zone */ |
166 | atomic_t reclaim_in_progress; | |
753ee728 | 167 | |
9eeff239 CL |
168 | /* |
169 | * timestamp (in jiffies) of the last zone reclaim that did not | |
170 | * result in freeing of pages. This is used to avoid repeated scans | |
171 | * if all memory in the zone is in use. | |
172 | */ | |
173 | unsigned long last_unsuccessful_zone_reclaim; | |
174 | ||
1da177e4 LT |
175 | /* |
176 | * prev_priority holds the scanning priority for this zone. It is | |
177 | * defined as the scanning priority at which we achieved our reclaim | |
178 | * target at the previous try_to_free_pages() or balance_pgdat() | |
179 | * invokation. | |
180 | * | |
181 | * We use prev_priority as a measure of how much stress page reclaim is | |
182 | * under - it drives the swappiness decision: whether to unmap mapped | |
183 | * pages. | |
184 | * | |
185 | * temp_priority is used to remember the scanning priority at which | |
186 | * this zone was successfully refilled to free_pages == pages_high. | |
187 | * | |
188 | * Access to both these fields is quite racy even on uniprocessor. But | |
189 | * it is expected to average out OK. | |
190 | */ | |
191 | int temp_priority; | |
192 | int prev_priority; | |
193 | ||
194 | ||
195 | ZONE_PADDING(_pad2_) | |
196 | /* Rarely used or read-mostly fields */ | |
197 | ||
198 | /* | |
199 | * wait_table -- the array holding the hash table | |
200 | * wait_table_size -- the size of the hash table array | |
201 | * wait_table_bits -- wait_table_size == (1 << wait_table_bits) | |
202 | * | |
203 | * The purpose of all these is to keep track of the people | |
204 | * waiting for a page to become available and make them | |
205 | * runnable again when possible. The trouble is that this | |
206 | * consumes a lot of space, especially when so few things | |
207 | * wait on pages at a given time. So instead of using | |
208 | * per-page waitqueues, we use a waitqueue hash table. | |
209 | * | |
210 | * The bucket discipline is to sleep on the same queue when | |
211 | * colliding and wake all in that wait queue when removing. | |
212 | * When something wakes, it must check to be sure its page is | |
213 | * truly available, a la thundering herd. The cost of a | |
214 | * collision is great, but given the expected load of the | |
215 | * table, they should be so rare as to be outweighed by the | |
216 | * benefits from the saved space. | |
217 | * | |
218 | * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the | |
219 | * primary users of these fields, and in mm/page_alloc.c | |
220 | * free_area_init_core() performs the initialization of them. | |
221 | */ | |
222 | wait_queue_head_t * wait_table; | |
223 | unsigned long wait_table_size; | |
224 | unsigned long wait_table_bits; | |
225 | ||
226 | /* | |
227 | * Discontig memory support fields. | |
228 | */ | |
229 | struct pglist_data *zone_pgdat; | |
1da177e4 LT |
230 | /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */ |
231 | unsigned long zone_start_pfn; | |
232 | ||
bdc8cb98 DH |
233 | /* |
234 | * zone_start_pfn, spanned_pages and present_pages are all | |
235 | * protected by span_seqlock. It is a seqlock because it has | |
236 | * to be read outside of zone->lock, and it is done in the main | |
237 | * allocator path. But, it is written quite infrequently. | |
238 | * | |
239 | * The lock is declared along with zone->lock because it is | |
240 | * frequently read in proximity to zone->lock. It's good to | |
241 | * give them a chance of being in the same cacheline. | |
242 | */ | |
1da177e4 LT |
243 | unsigned long spanned_pages; /* total size, including holes */ |
244 | unsigned long present_pages; /* amount of memory (excluding holes) */ | |
245 | ||
246 | /* | |
247 | * rarely used fields: | |
248 | */ | |
249 | char *name; | |
22fc6ecc | 250 | } ____cacheline_internodealigned_in_smp; |
1da177e4 LT |
251 | |
252 | ||
253 | /* | |
254 | * The "priority" of VM scanning is how much of the queues we will scan in one | |
255 | * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the | |
256 | * queues ("queue_length >> 12") during an aging round. | |
257 | */ | |
258 | #define DEF_PRIORITY 12 | |
259 | ||
260 | /* | |
261 | * One allocation request operates on a zonelist. A zonelist | |
262 | * is a list of zones, the first one is the 'goal' of the | |
263 | * allocation, the other zones are fallback zones, in decreasing | |
264 | * priority. | |
265 | * | |
266 | * Right now a zonelist takes up less than a cacheline. We never | |
267 | * modify it apart from boot-up, and only a few indices are used, | |
268 | * so despite the zonelist table being relatively big, the cache | |
269 | * footprint of this construct is very small. | |
270 | */ | |
271 | struct zonelist { | |
272 | struct zone *zones[MAX_NUMNODES * MAX_NR_ZONES + 1]; // NULL delimited | |
273 | }; | |
274 | ||
275 | ||
276 | /* | |
277 | * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM | |
278 | * (mostly NUMA machines?) to denote a higher-level memory zone than the | |
279 | * zone denotes. | |
280 | * | |
281 | * On NUMA machines, each NUMA node would have a pg_data_t to describe | |
282 | * it's memory layout. | |
283 | * | |
284 | * Memory statistics and page replacement data structures are maintained on a | |
285 | * per-zone basis. | |
286 | */ | |
287 | struct bootmem_data; | |
288 | typedef struct pglist_data { | |
289 | struct zone node_zones[MAX_NR_ZONES]; | |
290 | struct zonelist node_zonelists[GFP_ZONETYPES]; | |
291 | int nr_zones; | |
d41dee36 | 292 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 | 293 | struct page *node_mem_map; |
d41dee36 | 294 | #endif |
1da177e4 | 295 | struct bootmem_data *bdata; |
208d54e5 DH |
296 | #ifdef CONFIG_MEMORY_HOTPLUG |
297 | /* | |
298 | * Must be held any time you expect node_start_pfn, node_present_pages | |
299 | * or node_spanned_pages stay constant. Holding this will also | |
300 | * guarantee that any pfn_valid() stays that way. | |
301 | * | |
302 | * Nests above zone->lock and zone->size_seqlock. | |
303 | */ | |
304 | spinlock_t node_size_lock; | |
305 | #endif | |
1da177e4 LT |
306 | unsigned long node_start_pfn; |
307 | unsigned long node_present_pages; /* total number of physical pages */ | |
308 | unsigned long node_spanned_pages; /* total size of physical page | |
309 | range, including holes */ | |
310 | int node_id; | |
1da177e4 LT |
311 | wait_queue_head_t kswapd_wait; |
312 | struct task_struct *kswapd; | |
313 | int kswapd_max_order; | |
314 | } pg_data_t; | |
315 | ||
316 | #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages) | |
317 | #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages) | |
d41dee36 | 318 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
408fde81 | 319 | #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr)) |
d41dee36 AW |
320 | #else |
321 | #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr)) | |
322 | #endif | |
408fde81 | 323 | #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr)) |
1da177e4 | 324 | |
208d54e5 DH |
325 | #include <linux/memory_hotplug.h> |
326 | ||
1da177e4 LT |
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, | |
7fb1d9fc | 334 | int classzone_idx, int alloc_flags); |
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 | ||
f3fe6512 CK |
351 | static inline int populated_zone(struct zone *zone) |
352 | { | |
353 | return (!!zone->present_pages); | |
354 | } | |
355 | ||
1da177e4 LT |
356 | static inline int is_highmem_idx(int idx) |
357 | { | |
358 | return (idx == ZONE_HIGHMEM); | |
359 | } | |
360 | ||
361 | static inline int is_normal_idx(int idx) | |
362 | { | |
363 | return (idx == ZONE_NORMAL); | |
364 | } | |
9328b8fa | 365 | |
1da177e4 LT |
366 | /** |
367 | * is_highmem - helper function to quickly check if a struct zone is a | |
368 | * highmem zone or not. This is an attempt to keep references | |
369 | * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum. | |
370 | * @zone - pointer to struct zone variable | |
371 | */ | |
372 | static inline int is_highmem(struct zone *zone) | |
373 | { | |
374 | return zone == zone->zone_pgdat->node_zones + ZONE_HIGHMEM; | |
375 | } | |
376 | ||
377 | static inline int is_normal(struct zone *zone) | |
378 | { | |
379 | return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL; | |
380 | } | |
381 | ||
9328b8fa NP |
382 | static inline int is_dma32(struct zone *zone) |
383 | { | |
384 | return zone == zone->zone_pgdat->node_zones + ZONE_DMA32; | |
385 | } | |
386 | ||
387 | static inline int is_dma(struct zone *zone) | |
388 | { | |
389 | return zone == zone->zone_pgdat->node_zones + ZONE_DMA; | |
390 | } | |
391 | ||
1da177e4 LT |
392 | /* These two functions are used to setup the per zone pages min values */ |
393 | struct ctl_table; | |
394 | struct file; | |
395 | int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *, | |
396 | void __user *, size_t *, loff_t *); | |
397 | extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1]; | |
398 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *, | |
399 | void __user *, size_t *, loff_t *); | |
8ad4b1fb RS |
400 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, struct file *, |
401 | void __user *, size_t *, loff_t *); | |
1da177e4 LT |
402 | |
403 | #include <linux/topology.h> | |
404 | /* Returns the number of the current Node. */ | |
69d81fcd | 405 | #ifndef numa_node_id |
39c715b7 | 406 | #define numa_node_id() (cpu_to_node(raw_smp_processor_id())) |
69d81fcd | 407 | #endif |
1da177e4 | 408 | |
93b7504e | 409 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
410 | |
411 | extern struct pglist_data contig_page_data; | |
412 | #define NODE_DATA(nid) (&contig_page_data) | |
413 | #define NODE_MEM_MAP(nid) mem_map | |
414 | #define MAX_NODES_SHIFT 1 | |
1da177e4 | 415 | |
93b7504e | 416 | #else /* CONFIG_NEED_MULTIPLE_NODES */ |
1da177e4 LT |
417 | |
418 | #include <asm/mmzone.h> | |
419 | ||
93b7504e | 420 | #endif /* !CONFIG_NEED_MULTIPLE_NODES */ |
348f8b6c | 421 | |
95144c78 KH |
422 | extern struct pglist_data *first_online_pgdat(void); |
423 | extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat); | |
424 | extern struct zone *next_zone(struct zone *zone); | |
8357f869 KH |
425 | |
426 | /** | |
427 | * for_each_pgdat - helper macro to iterate over all nodes | |
428 | * @pgdat - pointer to a pg_data_t variable | |
429 | */ | |
430 | #define for_each_online_pgdat(pgdat) \ | |
431 | for (pgdat = first_online_pgdat(); \ | |
432 | pgdat; \ | |
433 | pgdat = next_online_pgdat(pgdat)) | |
8357f869 KH |
434 | /** |
435 | * for_each_zone - helper macro to iterate over all memory zones | |
436 | * @zone - pointer to struct zone variable | |
437 | * | |
438 | * The user only needs to declare the zone variable, for_each_zone | |
439 | * fills it in. | |
440 | */ | |
441 | #define for_each_zone(zone) \ | |
442 | for (zone = (first_online_pgdat())->node_zones; \ | |
443 | zone; \ | |
444 | zone = next_zone(zone)) | |
445 | ||
d41dee36 AW |
446 | #ifdef CONFIG_SPARSEMEM |
447 | #include <asm/sparsemem.h> | |
448 | #endif | |
449 | ||
07808b74 | 450 | #if BITS_PER_LONG == 32 |
1da177e4 | 451 | /* |
a2f1b424 AK |
452 | * with 32 bit page->flags field, we reserve 9 bits for node/zone info. |
453 | * there are 4 zones (3 bits) and this leaves 9-3=6 bits for nodes. | |
1da177e4 | 454 | */ |
a2f1b424 | 455 | #define FLAGS_RESERVED 9 |
348f8b6c | 456 | |
1da177e4 LT |
457 | #elif BITS_PER_LONG == 64 |
458 | /* | |
459 | * with 64 bit flags field, there's plenty of room. | |
460 | */ | |
348f8b6c | 461 | #define FLAGS_RESERVED 32 |
1da177e4 | 462 | |
348f8b6c | 463 | #else |
1da177e4 | 464 | |
348f8b6c | 465 | #error BITS_PER_LONG not defined |
1da177e4 | 466 | |
1da177e4 LT |
467 | #endif |
468 | ||
b159d43f AW |
469 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
470 | #define early_pfn_to_nid(nid) (0UL) | |
471 | #endif | |
472 | ||
2bdaf115 AW |
473 | #ifdef CONFIG_FLATMEM |
474 | #define pfn_to_nid(pfn) (0) | |
475 | #endif | |
476 | ||
d41dee36 AW |
477 | #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT) |
478 | #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT) | |
479 | ||
480 | #ifdef CONFIG_SPARSEMEM | |
481 | ||
482 | /* | |
483 | * SECTION_SHIFT #bits space required to store a section # | |
484 | * | |
485 | * PA_SECTION_SHIFT physical address to/from section number | |
486 | * PFN_SECTION_SHIFT pfn to/from section number | |
487 | */ | |
488 | #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS) | |
489 | ||
490 | #define PA_SECTION_SHIFT (SECTION_SIZE_BITS) | |
491 | #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT) | |
492 | ||
493 | #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT) | |
494 | ||
495 | #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT) | |
496 | #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1)) | |
497 | ||
498 | #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS | |
499 | #error Allocator MAX_ORDER exceeds SECTION_SIZE | |
500 | #endif | |
501 | ||
502 | struct page; | |
503 | struct mem_section { | |
29751f69 AW |
504 | /* |
505 | * This is, logically, a pointer to an array of struct | |
506 | * pages. However, it is stored with some other magic. | |
507 | * (see sparse.c::sparse_init_one_section()) | |
508 | * | |
509 | * Making it a UL at least makes someone do a cast | |
510 | * before using it wrong. | |
511 | */ | |
512 | unsigned long section_mem_map; | |
d41dee36 AW |
513 | }; |
514 | ||
3e347261 BP |
515 | #ifdef CONFIG_SPARSEMEM_EXTREME |
516 | #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section)) | |
517 | #else | |
518 | #define SECTIONS_PER_ROOT 1 | |
519 | #endif | |
802f192e | 520 | |
3e347261 BP |
521 | #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT) |
522 | #define NR_SECTION_ROOTS (NR_MEM_SECTIONS / SECTIONS_PER_ROOT) | |
523 | #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1) | |
802f192e | 524 | |
3e347261 BP |
525 | #ifdef CONFIG_SPARSEMEM_EXTREME |
526 | extern struct mem_section *mem_section[NR_SECTION_ROOTS]; | |
802f192e | 527 | #else |
3e347261 BP |
528 | extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]; |
529 | #endif | |
d41dee36 | 530 | |
29751f69 AW |
531 | static inline struct mem_section *__nr_to_section(unsigned long nr) |
532 | { | |
3e347261 BP |
533 | if (!mem_section[SECTION_NR_TO_ROOT(nr)]) |
534 | return NULL; | |
535 | return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK]; | |
29751f69 | 536 | } |
4ca644d9 | 537 | extern int __section_nr(struct mem_section* ms); |
29751f69 AW |
538 | |
539 | /* | |
540 | * We use the lower bits of the mem_map pointer to store | |
541 | * a little bit of information. There should be at least | |
542 | * 3 bits here due to 32-bit alignment. | |
543 | */ | |
544 | #define SECTION_MARKED_PRESENT (1UL<<0) | |
545 | #define SECTION_HAS_MEM_MAP (1UL<<1) | |
546 | #define SECTION_MAP_LAST_BIT (1UL<<2) | |
547 | #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1)) | |
548 | ||
549 | static inline struct page *__section_mem_map_addr(struct mem_section *section) | |
550 | { | |
551 | unsigned long map = section->section_mem_map; | |
552 | map &= SECTION_MAP_MASK; | |
553 | return (struct page *)map; | |
554 | } | |
555 | ||
556 | static inline int valid_section(struct mem_section *section) | |
557 | { | |
802f192e | 558 | return (section && (section->section_mem_map & SECTION_MARKED_PRESENT)); |
29751f69 AW |
559 | } |
560 | ||
561 | static inline int section_has_mem_map(struct mem_section *section) | |
562 | { | |
802f192e | 563 | return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP)); |
29751f69 AW |
564 | } |
565 | ||
566 | static inline int valid_section_nr(unsigned long nr) | |
567 | { | |
568 | return valid_section(__nr_to_section(nr)); | |
569 | } | |
570 | ||
d41dee36 AW |
571 | static inline struct mem_section *__pfn_to_section(unsigned long pfn) |
572 | { | |
29751f69 | 573 | return __nr_to_section(pfn_to_section_nr(pfn)); |
d41dee36 AW |
574 | } |
575 | ||
d41dee36 AW |
576 | static inline int pfn_valid(unsigned long pfn) |
577 | { | |
578 | if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) | |
579 | return 0; | |
29751f69 | 580 | return valid_section(__nr_to_section(pfn_to_section_nr(pfn))); |
d41dee36 AW |
581 | } |
582 | ||
583 | /* | |
584 | * These are _only_ used during initialisation, therefore they | |
585 | * can use __initdata ... They could have names to indicate | |
586 | * this restriction. | |
587 | */ | |
588 | #ifdef CONFIG_NUMA | |
161599ff AW |
589 | #define pfn_to_nid(pfn) \ |
590 | ({ \ | |
591 | unsigned long __pfn_to_nid_pfn = (pfn); \ | |
592 | page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \ | |
593 | }) | |
2bdaf115 AW |
594 | #else |
595 | #define pfn_to_nid(pfn) (0) | |
d41dee36 AW |
596 | #endif |
597 | ||
d41dee36 AW |
598 | #define early_pfn_valid(pfn) pfn_valid(pfn) |
599 | void sparse_init(void); | |
600 | #else | |
601 | #define sparse_init() do {} while (0) | |
28ae55c9 | 602 | #define sparse_index_init(_sec, _nid) do {} while (0) |
d41dee36 AW |
603 | #endif /* CONFIG_SPARSEMEM */ |
604 | ||
605 | #ifndef early_pfn_valid | |
606 | #define early_pfn_valid(pfn) (1) | |
607 | #endif | |
608 | ||
609 | void memory_present(int nid, unsigned long start, unsigned long end); | |
610 | unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long); | |
611 | ||
1da177e4 LT |
612 | #endif /* !__ASSEMBLY__ */ |
613 | #endif /* __KERNEL__ */ | |
614 | #endif /* _LINUX_MMZONE_H */ |