]>
Commit | Line | Data |
---|---|---|
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 | |
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 | ||
2244b95a | 49 | enum 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 */ |
2244b95a CL |
55 | NR_VM_ZONE_STAT_ITEMS }; |
56 | ||
1da177e4 LT |
57 | struct per_cpu_pages { |
58 | int count; /* number of pages in the list */ | |
1da177e4 LT |
59 | int high; /* high watermark, emptying needed */ |
60 | int batch; /* chunk size for buddy add/remove */ | |
61 | struct list_head list; /* the list of pages */ | |
62 | }; | |
63 | ||
64 | struct per_cpu_pageset { | |
65 | struct per_cpu_pages pcp[2]; /* 0: hot. 1: cold */ | |
2244b95a CL |
66 | #ifdef CONFIG_SMP |
67 | s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS]; | |
68 | #endif | |
69 | ||
1da177e4 LT |
70 | #ifdef CONFIG_NUMA |
71 | unsigned long numa_hit; /* allocated in intended node */ | |
72 | unsigned long numa_miss; /* allocated in non intended node */ | |
73 | unsigned long numa_foreign; /* was intended here, hit elsewhere */ | |
74 | unsigned long interleave_hit; /* interleaver prefered this zone */ | |
75 | unsigned long local_node; /* allocation from local node */ | |
76 | unsigned long other_node; /* allocation from other node */ | |
77 | #endif | |
78 | } ____cacheline_aligned_in_smp; | |
79 | ||
e7c8d5c9 CL |
80 | #ifdef CONFIG_NUMA |
81 | #define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)]) | |
82 | #else | |
83 | #define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)]) | |
84 | #endif | |
85 | ||
1da177e4 | 86 | #define ZONE_DMA 0 |
a2f1b424 AK |
87 | #define ZONE_DMA32 1 |
88 | #define ZONE_NORMAL 2 | |
89 | #define ZONE_HIGHMEM 3 | |
1da177e4 | 90 | |
a2f1b424 | 91 | #define MAX_NR_ZONES 4 /* Sync this with ZONES_SHIFT */ |
1da177e4 LT |
92 | #define ZONES_SHIFT 2 /* ceil(log2(MAX_NR_ZONES)) */ |
93 | ||
94 | ||
95 | /* | |
96 | * When a memory allocation must conform to specific limitations (such | |
97 | * as being suitable for DMA) the caller will pass in hints to the | |
98 | * allocator in the gfp_mask, in the zone modifier bits. These bits | |
99 | * are used to select a priority ordered list of memory zones which | |
100 | * match the requested limits. GFP_ZONEMASK defines which bits within | |
101 | * the gfp_mask should be considered as zone modifiers. Each valid | |
102 | * combination of the zone modifier bits has a corresponding list | |
103 | * of zones (in node_zonelists). Thus for two zone modifiers there | |
104 | * will be a maximum of 4 (2 ** 2) zonelists, for 3 modifiers there will | |
105 | * be 8 (2 ** 3) zonelists. GFP_ZONETYPES defines the number of possible | |
106 | * combinations of zone modifiers in "zone modifier space". | |
ac3461ad | 107 | * |
79046ae0 AW |
108 | * As an optimisation any zone modifier bits which are only valid when |
109 | * no other zone modifier bits are set (loners) should be placed in | |
110 | * the highest order bits of this field. This allows us to reduce the | |
111 | * extent of the zonelists thus saving space. For example in the case | |
112 | * of three zone modifier bits, we could require up to eight zonelists. | |
113 | * If the left most zone modifier is a "loner" then the highest valid | |
114 | * zonelist would be four allowing us to allocate only five zonelists. | |
ce2ea89b AW |
115 | * Use the first form for GFP_ZONETYPES when the left most bit is not |
116 | * a "loner", otherwise use the second. | |
79046ae0 | 117 | * |
ac3461ad | 118 | * NOTE! Make sure this matches the zones in <linux/gfp.h> |
1da177e4 | 119 | */ |
ac3461ad | 120 | #define GFP_ZONEMASK 0x07 |
ce2ea89b AW |
121 | /* #define GFP_ZONETYPES (GFP_ZONEMASK + 1) */ /* Non-loner */ |
122 | #define GFP_ZONETYPES ((GFP_ZONEMASK + 1) / 2 + 1) /* Loner */ | |
1da177e4 LT |
123 | |
124 | /* | |
125 | * On machines where it is needed (eg PCs) we divide physical memory | |
1f6818b9 | 126 | * into multiple physical zones. On a 32bit PC we have 4 zones: |
1da177e4 LT |
127 | * |
128 | * ZONE_DMA < 16 MB ISA DMA capable memory | |
a2f1b424 | 129 | * ZONE_DMA32 0 MB Empty |
1da177e4 LT |
130 | * ZONE_NORMAL 16-896 MB direct mapped by the kernel |
131 | * ZONE_HIGHMEM > 896 MB only page cache and user processes | |
132 | */ | |
133 | ||
134 | struct zone { | |
135 | /* Fields commonly accessed by the page allocator */ | |
136 | unsigned long free_pages; | |
137 | unsigned long pages_min, pages_low, pages_high; | |
138 | /* | |
139 | * We don't know if the memory that we're going to allocate will be freeable | |
140 | * or/and it will be released eventually, so to avoid totally wasting several | |
141 | * GB of ram we must reserve some of the lower zone memory (otherwise we risk | |
142 | * to run OOM on the lower zones despite there's tons of freeable ram | |
143 | * on the higher zones). This array is recalculated at runtime if the | |
144 | * sysctl_lowmem_reserve_ratio sysctl changes. | |
145 | */ | |
146 | unsigned long lowmem_reserve[MAX_NR_ZONES]; | |
147 | ||
e7c8d5c9 CL |
148 | #ifdef CONFIG_NUMA |
149 | struct per_cpu_pageset *pageset[NR_CPUS]; | |
150 | #else | |
1da177e4 | 151 | struct per_cpu_pageset pageset[NR_CPUS]; |
e7c8d5c9 | 152 | #endif |
1da177e4 LT |
153 | /* |
154 | * free areas of different sizes | |
155 | */ | |
156 | spinlock_t lock; | |
bdc8cb98 DH |
157 | #ifdef CONFIG_MEMORY_HOTPLUG |
158 | /* see spanned/present_pages for more description */ | |
159 | seqlock_t span_seqlock; | |
160 | #endif | |
1da177e4 LT |
161 | struct free_area free_area[MAX_ORDER]; |
162 | ||
163 | ||
164 | ZONE_PADDING(_pad1_) | |
165 | ||
166 | /* Fields commonly accessed by the page reclaim scanner */ | |
167 | spinlock_t lru_lock; | |
168 | struct list_head active_list; | |
169 | struct list_head inactive_list; | |
170 | unsigned long nr_scan_active; | |
171 | unsigned long nr_scan_inactive; | |
172 | unsigned long nr_active; | |
173 | unsigned long nr_inactive; | |
174 | unsigned long pages_scanned; /* since last reclaim */ | |
175 | int all_unreclaimable; /* All pages pinned */ | |
176 | ||
1e7e5a90 MH |
177 | /* A count of how many reclaimers are scanning this zone */ |
178 | atomic_t reclaim_in_progress; | |
753ee728 | 179 | |
2244b95a CL |
180 | /* Zone statistics */ |
181 | atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS]; | |
9eeff239 | 182 | |
1da177e4 LT |
183 | /* |
184 | * prev_priority holds the scanning priority for this zone. It is | |
185 | * defined as the scanning priority at which we achieved our reclaim | |
186 | * target at the previous try_to_free_pages() or balance_pgdat() | |
187 | * invokation. | |
188 | * | |
189 | * We use prev_priority as a measure of how much stress page reclaim is | |
190 | * under - it drives the swappiness decision: whether to unmap mapped | |
191 | * pages. | |
192 | * | |
193 | * temp_priority is used to remember the scanning priority at which | |
194 | * this zone was successfully refilled to free_pages == pages_high. | |
195 | * | |
196 | * Access to both these fields is quite racy even on uniprocessor. But | |
197 | * it is expected to average out OK. | |
198 | */ | |
199 | int temp_priority; | |
200 | int prev_priority; | |
201 | ||
202 | ||
203 | ZONE_PADDING(_pad2_) | |
204 | /* Rarely used or read-mostly fields */ | |
205 | ||
206 | /* | |
207 | * wait_table -- the array holding the hash table | |
02b694de | 208 | * wait_table_hash_nr_entries -- the size of the hash table array |
1da177e4 LT |
209 | * wait_table_bits -- wait_table_size == (1 << wait_table_bits) |
210 | * | |
211 | * The purpose of all these is to keep track of the people | |
212 | * waiting for a page to become available and make them | |
213 | * runnable again when possible. The trouble is that this | |
214 | * consumes a lot of space, especially when so few things | |
215 | * wait on pages at a given time. So instead of using | |
216 | * per-page waitqueues, we use a waitqueue hash table. | |
217 | * | |
218 | * The bucket discipline is to sleep on the same queue when | |
219 | * colliding and wake all in that wait queue when removing. | |
220 | * When something wakes, it must check to be sure its page is | |
221 | * truly available, a la thundering herd. The cost of a | |
222 | * collision is great, but given the expected load of the | |
223 | * table, they should be so rare as to be outweighed by the | |
224 | * benefits from the saved space. | |
225 | * | |
226 | * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the | |
227 | * primary users of these fields, and in mm/page_alloc.c | |
228 | * free_area_init_core() performs the initialization of them. | |
229 | */ | |
230 | wait_queue_head_t * wait_table; | |
02b694de | 231 | unsigned long wait_table_hash_nr_entries; |
1da177e4 LT |
232 | unsigned long wait_table_bits; |
233 | ||
234 | /* | |
235 | * Discontig memory support fields. | |
236 | */ | |
237 | struct pglist_data *zone_pgdat; | |
1da177e4 LT |
238 | /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */ |
239 | unsigned long zone_start_pfn; | |
240 | ||
bdc8cb98 DH |
241 | /* |
242 | * zone_start_pfn, spanned_pages and present_pages are all | |
243 | * protected by span_seqlock. It is a seqlock because it has | |
244 | * to be read outside of zone->lock, and it is done in the main | |
245 | * allocator path. But, it is written quite infrequently. | |
246 | * | |
247 | * The lock is declared along with zone->lock because it is | |
248 | * frequently read in proximity to zone->lock. It's good to | |
249 | * give them a chance of being in the same cacheline. | |
250 | */ | |
1da177e4 LT |
251 | unsigned long spanned_pages; /* total size, including holes */ |
252 | unsigned long present_pages; /* amount of memory (excluding holes) */ | |
253 | ||
254 | /* | |
255 | * rarely used fields: | |
256 | */ | |
257 | char *name; | |
22fc6ecc | 258 | } ____cacheline_internodealigned_in_smp; |
1da177e4 LT |
259 | |
260 | ||
261 | /* | |
262 | * The "priority" of VM scanning is how much of the queues we will scan in one | |
263 | * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the | |
264 | * queues ("queue_length >> 12") during an aging round. | |
265 | */ | |
266 | #define DEF_PRIORITY 12 | |
267 | ||
268 | /* | |
269 | * One allocation request operates on a zonelist. A zonelist | |
270 | * is a list of zones, the first one is the 'goal' of the | |
271 | * allocation, the other zones are fallback zones, in decreasing | |
272 | * priority. | |
273 | * | |
274 | * Right now a zonelist takes up less than a cacheline. We never | |
275 | * modify it apart from boot-up, and only a few indices are used, | |
276 | * so despite the zonelist table being relatively big, the cache | |
277 | * footprint of this construct is very small. | |
278 | */ | |
279 | struct zonelist { | |
280 | struct zone *zones[MAX_NUMNODES * MAX_NR_ZONES + 1]; // NULL delimited | |
281 | }; | |
282 | ||
283 | ||
284 | /* | |
285 | * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM | |
286 | * (mostly NUMA machines?) to denote a higher-level memory zone than the | |
287 | * zone denotes. | |
288 | * | |
289 | * On NUMA machines, each NUMA node would have a pg_data_t to describe | |
290 | * it's memory layout. | |
291 | * | |
292 | * Memory statistics and page replacement data structures are maintained on a | |
293 | * per-zone basis. | |
294 | */ | |
295 | struct bootmem_data; | |
296 | typedef struct pglist_data { | |
297 | struct zone node_zones[MAX_NR_ZONES]; | |
298 | struct zonelist node_zonelists[GFP_ZONETYPES]; | |
299 | int nr_zones; | |
d41dee36 | 300 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 | 301 | struct page *node_mem_map; |
d41dee36 | 302 | #endif |
1da177e4 | 303 | struct bootmem_data *bdata; |
208d54e5 DH |
304 | #ifdef CONFIG_MEMORY_HOTPLUG |
305 | /* | |
306 | * Must be held any time you expect node_start_pfn, node_present_pages | |
307 | * or node_spanned_pages stay constant. Holding this will also | |
308 | * guarantee that any pfn_valid() stays that way. | |
309 | * | |
310 | * Nests above zone->lock and zone->size_seqlock. | |
311 | */ | |
312 | spinlock_t node_size_lock; | |
313 | #endif | |
1da177e4 LT |
314 | unsigned long node_start_pfn; |
315 | unsigned long node_present_pages; /* total number of physical pages */ | |
316 | unsigned long node_spanned_pages; /* total size of physical page | |
317 | range, including holes */ | |
318 | int node_id; | |
1da177e4 LT |
319 | wait_queue_head_t kswapd_wait; |
320 | struct task_struct *kswapd; | |
321 | int kswapd_max_order; | |
322 | } pg_data_t; | |
323 | ||
324 | #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages) | |
325 | #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages) | |
d41dee36 | 326 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
408fde81 | 327 | #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr)) |
d41dee36 AW |
328 | #else |
329 | #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr)) | |
330 | #endif | |
408fde81 | 331 | #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr)) |
1da177e4 | 332 | |
208d54e5 DH |
333 | #include <linux/memory_hotplug.h> |
334 | ||
1da177e4 LT |
335 | void __get_zone_counts(unsigned long *active, unsigned long *inactive, |
336 | unsigned long *free, struct pglist_data *pgdat); | |
337 | void get_zone_counts(unsigned long *active, unsigned long *inactive, | |
338 | unsigned long *free); | |
339 | void build_all_zonelists(void); | |
340 | void wakeup_kswapd(struct zone *zone, int order); | |
341 | int zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
7fb1d9fc | 342 | int classzone_idx, int alloc_flags); |
1da177e4 | 343 | |
718127cc YG |
344 | extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn, |
345 | unsigned long size); | |
346 | ||
1da177e4 LT |
347 | #ifdef CONFIG_HAVE_MEMORY_PRESENT |
348 | void memory_present(int nid, unsigned long start, unsigned long end); | |
349 | #else | |
350 | static inline void memory_present(int nid, unsigned long start, unsigned long end) {} | |
351 | #endif | |
352 | ||
353 | #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE | |
354 | unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long); | |
355 | #endif | |
356 | ||
357 | /* | |
358 | * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc. | |
359 | */ | |
360 | #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones) | |
361 | ||
f3fe6512 CK |
362 | static inline int populated_zone(struct zone *zone) |
363 | { | |
364 | return (!!zone->present_pages); | |
365 | } | |
366 | ||
1da177e4 LT |
367 | static inline int is_highmem_idx(int idx) |
368 | { | |
369 | return (idx == ZONE_HIGHMEM); | |
370 | } | |
371 | ||
372 | static inline int is_normal_idx(int idx) | |
373 | { | |
374 | return (idx == ZONE_NORMAL); | |
375 | } | |
9328b8fa | 376 | |
1da177e4 LT |
377 | /** |
378 | * is_highmem - helper function to quickly check if a struct zone is a | |
379 | * highmem zone or not. This is an attempt to keep references | |
380 | * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum. | |
381 | * @zone - pointer to struct zone variable | |
382 | */ | |
383 | static inline int is_highmem(struct zone *zone) | |
384 | { | |
385 | return zone == zone->zone_pgdat->node_zones + ZONE_HIGHMEM; | |
386 | } | |
387 | ||
388 | static inline int is_normal(struct zone *zone) | |
389 | { | |
390 | return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL; | |
391 | } | |
392 | ||
9328b8fa NP |
393 | static inline int is_dma32(struct zone *zone) |
394 | { | |
395 | return zone == zone->zone_pgdat->node_zones + ZONE_DMA32; | |
396 | } | |
397 | ||
398 | static inline int is_dma(struct zone *zone) | |
399 | { | |
400 | return zone == zone->zone_pgdat->node_zones + ZONE_DMA; | |
401 | } | |
402 | ||
1da177e4 LT |
403 | /* These two functions are used to setup the per zone pages min values */ |
404 | struct ctl_table; | |
405 | struct file; | |
406 | int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *, | |
407 | void __user *, size_t *, loff_t *); | |
408 | extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1]; | |
409 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *, | |
410 | void __user *, size_t *, loff_t *); | |
8ad4b1fb RS |
411 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, struct file *, |
412 | void __user *, size_t *, loff_t *); | |
1da177e4 LT |
413 | |
414 | #include <linux/topology.h> | |
415 | /* Returns the number of the current Node. */ | |
69d81fcd | 416 | #ifndef numa_node_id |
39c715b7 | 417 | #define numa_node_id() (cpu_to_node(raw_smp_processor_id())) |
69d81fcd | 418 | #endif |
1da177e4 | 419 | |
93b7504e | 420 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
421 | |
422 | extern struct pglist_data contig_page_data; | |
423 | #define NODE_DATA(nid) (&contig_page_data) | |
424 | #define NODE_MEM_MAP(nid) mem_map | |
425 | #define MAX_NODES_SHIFT 1 | |
1da177e4 | 426 | |
93b7504e | 427 | #else /* CONFIG_NEED_MULTIPLE_NODES */ |
1da177e4 LT |
428 | |
429 | #include <asm/mmzone.h> | |
430 | ||
93b7504e | 431 | #endif /* !CONFIG_NEED_MULTIPLE_NODES */ |
348f8b6c | 432 | |
95144c78 KH |
433 | extern struct pglist_data *first_online_pgdat(void); |
434 | extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat); | |
435 | extern struct zone *next_zone(struct zone *zone); | |
8357f869 KH |
436 | |
437 | /** | |
438 | * for_each_pgdat - helper macro to iterate over all nodes | |
439 | * @pgdat - pointer to a pg_data_t variable | |
440 | */ | |
441 | #define for_each_online_pgdat(pgdat) \ | |
442 | for (pgdat = first_online_pgdat(); \ | |
443 | pgdat; \ | |
444 | pgdat = next_online_pgdat(pgdat)) | |
8357f869 KH |
445 | /** |
446 | * for_each_zone - helper macro to iterate over all memory zones | |
447 | * @zone - pointer to struct zone variable | |
448 | * | |
449 | * The user only needs to declare the zone variable, for_each_zone | |
450 | * fills it in. | |
451 | */ | |
452 | #define for_each_zone(zone) \ | |
453 | for (zone = (first_online_pgdat())->node_zones; \ | |
454 | zone; \ | |
455 | zone = next_zone(zone)) | |
456 | ||
d41dee36 AW |
457 | #ifdef CONFIG_SPARSEMEM |
458 | #include <asm/sparsemem.h> | |
459 | #endif | |
460 | ||
07808b74 | 461 | #if BITS_PER_LONG == 32 |
1da177e4 | 462 | /* |
a2f1b424 AK |
463 | * with 32 bit page->flags field, we reserve 9 bits for node/zone info. |
464 | * there are 4 zones (3 bits) and this leaves 9-3=6 bits for nodes. | |
1da177e4 | 465 | */ |
a2f1b424 | 466 | #define FLAGS_RESERVED 9 |
348f8b6c | 467 | |
1da177e4 LT |
468 | #elif BITS_PER_LONG == 64 |
469 | /* | |
470 | * with 64 bit flags field, there's plenty of room. | |
471 | */ | |
348f8b6c | 472 | #define FLAGS_RESERVED 32 |
1da177e4 | 473 | |
348f8b6c | 474 | #else |
1da177e4 | 475 | |
348f8b6c | 476 | #error BITS_PER_LONG not defined |
1da177e4 | 477 | |
1da177e4 LT |
478 | #endif |
479 | ||
b159d43f AW |
480 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
481 | #define early_pfn_to_nid(nid) (0UL) | |
482 | #endif | |
483 | ||
2bdaf115 AW |
484 | #ifdef CONFIG_FLATMEM |
485 | #define pfn_to_nid(pfn) (0) | |
486 | #endif | |
487 | ||
d41dee36 AW |
488 | #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT) |
489 | #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT) | |
490 | ||
491 | #ifdef CONFIG_SPARSEMEM | |
492 | ||
493 | /* | |
494 | * SECTION_SHIFT #bits space required to store a section # | |
495 | * | |
496 | * PA_SECTION_SHIFT physical address to/from section number | |
497 | * PFN_SECTION_SHIFT pfn to/from section number | |
498 | */ | |
499 | #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS) | |
500 | ||
501 | #define PA_SECTION_SHIFT (SECTION_SIZE_BITS) | |
502 | #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT) | |
503 | ||
504 | #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT) | |
505 | ||
506 | #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT) | |
507 | #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1)) | |
508 | ||
509 | #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS | |
510 | #error Allocator MAX_ORDER exceeds SECTION_SIZE | |
511 | #endif | |
512 | ||
513 | struct page; | |
514 | struct mem_section { | |
29751f69 AW |
515 | /* |
516 | * This is, logically, a pointer to an array of struct | |
517 | * pages. However, it is stored with some other magic. | |
518 | * (see sparse.c::sparse_init_one_section()) | |
519 | * | |
30c253e6 AW |
520 | * Additionally during early boot we encode node id of |
521 | * the location of the section here to guide allocation. | |
522 | * (see sparse.c::memory_present()) | |
523 | * | |
29751f69 AW |
524 | * Making it a UL at least makes someone do a cast |
525 | * before using it wrong. | |
526 | */ | |
527 | unsigned long section_mem_map; | |
d41dee36 AW |
528 | }; |
529 | ||
3e347261 BP |
530 | #ifdef CONFIG_SPARSEMEM_EXTREME |
531 | #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section)) | |
532 | #else | |
533 | #define SECTIONS_PER_ROOT 1 | |
534 | #endif | |
802f192e | 535 | |
3e347261 BP |
536 | #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT) |
537 | #define NR_SECTION_ROOTS (NR_MEM_SECTIONS / SECTIONS_PER_ROOT) | |
538 | #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1) | |
802f192e | 539 | |
3e347261 BP |
540 | #ifdef CONFIG_SPARSEMEM_EXTREME |
541 | extern struct mem_section *mem_section[NR_SECTION_ROOTS]; | |
802f192e | 542 | #else |
3e347261 BP |
543 | extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]; |
544 | #endif | |
d41dee36 | 545 | |
29751f69 AW |
546 | static inline struct mem_section *__nr_to_section(unsigned long nr) |
547 | { | |
3e347261 BP |
548 | if (!mem_section[SECTION_NR_TO_ROOT(nr)]) |
549 | return NULL; | |
550 | return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK]; | |
29751f69 | 551 | } |
4ca644d9 | 552 | extern int __section_nr(struct mem_section* ms); |
29751f69 AW |
553 | |
554 | /* | |
555 | * We use the lower bits of the mem_map pointer to store | |
556 | * a little bit of information. There should be at least | |
557 | * 3 bits here due to 32-bit alignment. | |
558 | */ | |
559 | #define SECTION_MARKED_PRESENT (1UL<<0) | |
560 | #define SECTION_HAS_MEM_MAP (1UL<<1) | |
561 | #define SECTION_MAP_LAST_BIT (1UL<<2) | |
562 | #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1)) | |
30c253e6 | 563 | #define SECTION_NID_SHIFT 2 |
29751f69 AW |
564 | |
565 | static inline struct page *__section_mem_map_addr(struct mem_section *section) | |
566 | { | |
567 | unsigned long map = section->section_mem_map; | |
568 | map &= SECTION_MAP_MASK; | |
569 | return (struct page *)map; | |
570 | } | |
571 | ||
572 | static inline int valid_section(struct mem_section *section) | |
573 | { | |
802f192e | 574 | return (section && (section->section_mem_map & SECTION_MARKED_PRESENT)); |
29751f69 AW |
575 | } |
576 | ||
577 | static inline int section_has_mem_map(struct mem_section *section) | |
578 | { | |
802f192e | 579 | return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP)); |
29751f69 AW |
580 | } |
581 | ||
582 | static inline int valid_section_nr(unsigned long nr) | |
583 | { | |
584 | return valid_section(__nr_to_section(nr)); | |
585 | } | |
586 | ||
d41dee36 AW |
587 | static inline struct mem_section *__pfn_to_section(unsigned long pfn) |
588 | { | |
29751f69 | 589 | return __nr_to_section(pfn_to_section_nr(pfn)); |
d41dee36 AW |
590 | } |
591 | ||
d41dee36 AW |
592 | static inline int pfn_valid(unsigned long pfn) |
593 | { | |
594 | if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) | |
595 | return 0; | |
29751f69 | 596 | return valid_section(__nr_to_section(pfn_to_section_nr(pfn))); |
d41dee36 AW |
597 | } |
598 | ||
599 | /* | |
600 | * These are _only_ used during initialisation, therefore they | |
601 | * can use __initdata ... They could have names to indicate | |
602 | * this restriction. | |
603 | */ | |
604 | #ifdef CONFIG_NUMA | |
161599ff AW |
605 | #define pfn_to_nid(pfn) \ |
606 | ({ \ | |
607 | unsigned long __pfn_to_nid_pfn = (pfn); \ | |
608 | page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \ | |
609 | }) | |
2bdaf115 AW |
610 | #else |
611 | #define pfn_to_nid(pfn) (0) | |
d41dee36 AW |
612 | #endif |
613 | ||
d41dee36 AW |
614 | #define early_pfn_valid(pfn) pfn_valid(pfn) |
615 | void sparse_init(void); | |
616 | #else | |
617 | #define sparse_init() do {} while (0) | |
28ae55c9 | 618 | #define sparse_index_init(_sec, _nid) do {} while (0) |
d41dee36 AW |
619 | #endif /* CONFIG_SPARSEMEM */ |
620 | ||
621 | #ifndef early_pfn_valid | |
622 | #define early_pfn_valid(pfn) (1) | |
623 | #endif | |
624 | ||
625 | void memory_present(int nid, unsigned long start, unsigned long end); | |
626 | unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long); | |
627 | ||
1da177e4 LT |
628 | #endif /* !__ASSEMBLY__ */ |
629 | #endif /* __KERNEL__ */ | |
630 | #endif /* _LINUX_MMZONE_H */ |