Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/page_alloc.c | |
3 | * | |
4 | * Manages the free list, the system allocates free pages here. | |
5 | * Note that kmalloc() lives in slab.c | |
6 | * | |
7 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
8 | * Swap reorganised 29.12.95, Stephen Tweedie | |
9 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 | |
10 | * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999 | |
11 | * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999 | |
12 | * Zone balancing, Kanoj Sarcar, SGI, Jan 2000 | |
13 | * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002 | |
14 | * (lots of bits borrowed from Ingo Molnar & Andrew Morton) | |
15 | */ | |
16 | ||
1da177e4 LT |
17 | #include <linux/stddef.h> |
18 | #include <linux/mm.h> | |
19 | #include <linux/swap.h> | |
20 | #include <linux/interrupt.h> | |
21 | #include <linux/pagemap.h> | |
22 | #include <linux/bootmem.h> | |
23 | #include <linux/compiler.h> | |
9f158333 | 24 | #include <linux/kernel.h> |
1da177e4 LT |
25 | #include <linux/module.h> |
26 | #include <linux/suspend.h> | |
27 | #include <linux/pagevec.h> | |
28 | #include <linux/blkdev.h> | |
29 | #include <linux/slab.h> | |
30 | #include <linux/notifier.h> | |
31 | #include <linux/topology.h> | |
32 | #include <linux/sysctl.h> | |
33 | #include <linux/cpu.h> | |
34 | #include <linux/cpuset.h> | |
bdc8cb98 | 35 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
36 | #include <linux/nodemask.h> |
37 | #include <linux/vmalloc.h> | |
4be38e35 | 38 | #include <linux/mempolicy.h> |
6811378e | 39 | #include <linux/stop_machine.h> |
c713216d MG |
40 | #include <linux/sort.h> |
41 | #include <linux/pfn.h> | |
3fcfab16 | 42 | #include <linux/backing-dev.h> |
933e312e | 43 | #include <linux/fault-inject.h> |
1da177e4 LT |
44 | |
45 | #include <asm/tlbflush.h> | |
ac924c60 | 46 | #include <asm/div64.h> |
1da177e4 LT |
47 | #include "internal.h" |
48 | ||
49 | /* | |
13808910 | 50 | * Array of node states. |
1da177e4 | 51 | */ |
13808910 CL |
52 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
53 | [N_POSSIBLE] = NODE_MASK_ALL, | |
54 | [N_ONLINE] = { { [0] = 1UL } }, | |
55 | #ifndef CONFIG_NUMA | |
56 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
57 | #ifdef CONFIG_HIGHMEM | |
58 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
59 | #endif | |
60 | [N_CPU] = { { [0] = 1UL } }, | |
61 | #endif /* NUMA */ | |
62 | }; | |
63 | EXPORT_SYMBOL(node_states); | |
64 | ||
6c231b7b | 65 | unsigned long totalram_pages __read_mostly; |
cb45b0e9 | 66 | unsigned long totalreserve_pages __read_mostly; |
1da177e4 | 67 | long nr_swap_pages; |
8ad4b1fb | 68 | int percpu_pagelist_fraction; |
1da177e4 | 69 | |
d98c7a09 | 70 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 71 | |
1da177e4 LT |
72 | /* |
73 | * results with 256, 32 in the lowmem_reserve sysctl: | |
74 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
75 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
76 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
77 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
78 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA | |
a2f1b424 AK |
79 | * |
80 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
81 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 82 | */ |
2f1b6248 | 83 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
4b51d669 | 84 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 85 | 256, |
4b51d669 | 86 | #endif |
fb0e7942 | 87 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 88 | 256, |
fb0e7942 | 89 | #endif |
e53ef38d | 90 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 91 | 32, |
e53ef38d | 92 | #endif |
2a1e274a | 93 | 32, |
2f1b6248 | 94 | }; |
1da177e4 LT |
95 | |
96 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 | 97 | |
15ad7cdc | 98 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 99 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 100 | "DMA", |
4b51d669 | 101 | #endif |
fb0e7942 | 102 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 103 | "DMA32", |
fb0e7942 | 104 | #endif |
2f1b6248 | 105 | "Normal", |
e53ef38d | 106 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 107 | "HighMem", |
e53ef38d | 108 | #endif |
2a1e274a | 109 | "Movable", |
2f1b6248 CL |
110 | }; |
111 | ||
1da177e4 LT |
112 | int min_free_kbytes = 1024; |
113 | ||
86356ab1 YG |
114 | unsigned long __meminitdata nr_kernel_pages; |
115 | unsigned long __meminitdata nr_all_pages; | |
a3142c8e | 116 | static unsigned long __meminitdata dma_reserve; |
1da177e4 | 117 | |
c713216d MG |
118 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
119 | /* | |
120 | * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct | |
121 | * ranges of memory (RAM) that may be registered with add_active_range(). | |
122 | * Ranges passed to add_active_range() will be merged if possible | |
123 | * so the number of times add_active_range() can be called is | |
124 | * related to the number of nodes and the number of holes | |
125 | */ | |
126 | #ifdef CONFIG_MAX_ACTIVE_REGIONS | |
127 | /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */ | |
128 | #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS | |
129 | #else | |
130 | #if MAX_NUMNODES >= 32 | |
131 | /* If there can be many nodes, allow up to 50 holes per node */ | |
132 | #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50) | |
133 | #else | |
134 | /* By default, allow up to 256 distinct regions */ | |
135 | #define MAX_ACTIVE_REGIONS 256 | |
136 | #endif | |
137 | #endif | |
138 | ||
98011f56 JB |
139 | static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS]; |
140 | static int __meminitdata nr_nodemap_entries; | |
141 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; | |
142 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
fb01439c | 143 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE |
98011f56 JB |
144 | static unsigned long __meminitdata node_boundary_start_pfn[MAX_NUMNODES]; |
145 | static unsigned long __meminitdata node_boundary_end_pfn[MAX_NUMNODES]; | |
fb01439c | 146 | #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */ |
2a1e274a | 147 | unsigned long __initdata required_kernelcore; |
7e63efef | 148 | unsigned long __initdata required_movablecore; |
e228929b | 149 | unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES]; |
2a1e274a MG |
150 | |
151 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
152 | int movable_zone; | |
153 | EXPORT_SYMBOL(movable_zone); | |
c713216d MG |
154 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
155 | ||
418508c1 MS |
156 | #if MAX_NUMNODES > 1 |
157 | int nr_node_ids __read_mostly = MAX_NUMNODES; | |
158 | EXPORT_SYMBOL(nr_node_ids); | |
159 | #endif | |
160 | ||
b92a6edd | 161 | #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY |
b2a0ac88 MG |
162 | static inline int get_pageblock_migratetype(struct page *page) |
163 | { | |
164 | return get_pageblock_flags_group(page, PB_migrate, PB_migrate_end); | |
165 | } | |
166 | ||
167 | static void set_pageblock_migratetype(struct page *page, int migratetype) | |
168 | { | |
169 | set_pageblock_flags_group(page, (unsigned long)migratetype, | |
170 | PB_migrate, PB_migrate_end); | |
171 | } | |
172 | ||
173 | static inline int gfpflags_to_migratetype(gfp_t gfp_flags) | |
174 | { | |
e12ba74d MG |
175 | WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK); |
176 | ||
177 | return (((gfp_flags & __GFP_MOVABLE) != 0) << 1) | | |
178 | ((gfp_flags & __GFP_RECLAIMABLE) != 0); | |
b2a0ac88 MG |
179 | } |
180 | ||
b92a6edd MG |
181 | #else |
182 | static inline int get_pageblock_migratetype(struct page *page) | |
183 | { | |
184 | return MIGRATE_UNMOVABLE; | |
185 | } | |
186 | ||
187 | static void set_pageblock_migratetype(struct page *page, int migratetype) | |
188 | { | |
189 | } | |
190 | ||
191 | static inline int gfpflags_to_migratetype(gfp_t gfp_flags) | |
192 | { | |
193 | return MIGRATE_UNMOVABLE; | |
194 | } | |
195 | #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */ | |
196 | ||
13e7444b | 197 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 198 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 199 | { |
bdc8cb98 DH |
200 | int ret = 0; |
201 | unsigned seq; | |
202 | unsigned long pfn = page_to_pfn(page); | |
c6a57e19 | 203 | |
bdc8cb98 DH |
204 | do { |
205 | seq = zone_span_seqbegin(zone); | |
206 | if (pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
207 | ret = 1; | |
208 | else if (pfn < zone->zone_start_pfn) | |
209 | ret = 1; | |
210 | } while (zone_span_seqretry(zone, seq)); | |
211 | ||
212 | return ret; | |
c6a57e19 DH |
213 | } |
214 | ||
215 | static int page_is_consistent(struct zone *zone, struct page *page) | |
216 | { | |
14e07298 | 217 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 218 | return 0; |
1da177e4 | 219 | if (zone != page_zone(page)) |
c6a57e19 DH |
220 | return 0; |
221 | ||
222 | return 1; | |
223 | } | |
224 | /* | |
225 | * Temporary debugging check for pages not lying within a given zone. | |
226 | */ | |
227 | static int bad_range(struct zone *zone, struct page *page) | |
228 | { | |
229 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 230 | return 1; |
c6a57e19 DH |
231 | if (!page_is_consistent(zone, page)) |
232 | return 1; | |
233 | ||
1da177e4 LT |
234 | return 0; |
235 | } | |
13e7444b NP |
236 | #else |
237 | static inline int bad_range(struct zone *zone, struct page *page) | |
238 | { | |
239 | return 0; | |
240 | } | |
241 | #endif | |
242 | ||
224abf92 | 243 | static void bad_page(struct page *page) |
1da177e4 | 244 | { |
224abf92 | 245 | printk(KERN_EMERG "Bad page state in process '%s'\n" |
7365f3d1 HD |
246 | KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n" |
247 | KERN_EMERG "Trying to fix it up, but a reboot is needed\n" | |
248 | KERN_EMERG "Backtrace:\n", | |
224abf92 NP |
249 | current->comm, page, (int)(2*sizeof(unsigned long)), |
250 | (unsigned long)page->flags, page->mapping, | |
251 | page_mapcount(page), page_count(page)); | |
1da177e4 | 252 | dump_stack(); |
334795ec HD |
253 | page->flags &= ~(1 << PG_lru | |
254 | 1 << PG_private | | |
1da177e4 | 255 | 1 << PG_locked | |
1da177e4 LT |
256 | 1 << PG_active | |
257 | 1 << PG_dirty | | |
334795ec HD |
258 | 1 << PG_reclaim | |
259 | 1 << PG_slab | | |
1da177e4 | 260 | 1 << PG_swapcache | |
676165a8 NP |
261 | 1 << PG_writeback | |
262 | 1 << PG_buddy ); | |
1da177e4 LT |
263 | set_page_count(page, 0); |
264 | reset_page_mapcount(page); | |
265 | page->mapping = NULL; | |
9f158333 | 266 | add_taint(TAINT_BAD_PAGE); |
1da177e4 LT |
267 | } |
268 | ||
1da177e4 LT |
269 | /* |
270 | * Higher-order pages are called "compound pages". They are structured thusly: | |
271 | * | |
272 | * The first PAGE_SIZE page is called the "head page". | |
273 | * | |
274 | * The remaining PAGE_SIZE pages are called "tail pages". | |
275 | * | |
276 | * All pages have PG_compound set. All pages have their ->private pointing at | |
277 | * the head page (even the head page has this). | |
278 | * | |
41d78ba5 HD |
279 | * The first tail page's ->lru.next holds the address of the compound page's |
280 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
281 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 282 | */ |
d98c7a09 HD |
283 | |
284 | static void free_compound_page(struct page *page) | |
285 | { | |
d85f3385 | 286 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
287 | } |
288 | ||
1da177e4 LT |
289 | static void prep_compound_page(struct page *page, unsigned long order) |
290 | { | |
291 | int i; | |
292 | int nr_pages = 1 << order; | |
293 | ||
33f2ef89 | 294 | set_compound_page_dtor(page, free_compound_page); |
d85f3385 | 295 | set_compound_order(page, order); |
6d777953 | 296 | __SetPageHead(page); |
d85f3385 | 297 | for (i = 1; i < nr_pages; i++) { |
1da177e4 LT |
298 | struct page *p = page + i; |
299 | ||
d85f3385 | 300 | __SetPageTail(p); |
d85f3385 | 301 | p->first_page = page; |
1da177e4 LT |
302 | } |
303 | } | |
304 | ||
305 | static void destroy_compound_page(struct page *page, unsigned long order) | |
306 | { | |
307 | int i; | |
308 | int nr_pages = 1 << order; | |
309 | ||
d85f3385 | 310 | if (unlikely(compound_order(page) != order)) |
224abf92 | 311 | bad_page(page); |
1da177e4 | 312 | |
6d777953 | 313 | if (unlikely(!PageHead(page))) |
d85f3385 | 314 | bad_page(page); |
6d777953 | 315 | __ClearPageHead(page); |
d85f3385 | 316 | for (i = 1; i < nr_pages; i++) { |
1da177e4 LT |
317 | struct page *p = page + i; |
318 | ||
6d777953 | 319 | if (unlikely(!PageTail(p) | |
d85f3385 | 320 | (p->first_page != page))) |
224abf92 | 321 | bad_page(page); |
d85f3385 | 322 | __ClearPageTail(p); |
1da177e4 LT |
323 | } |
324 | } | |
1da177e4 | 325 | |
17cf4406 NP |
326 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) |
327 | { | |
328 | int i; | |
329 | ||
725d704e | 330 | VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM); |
6626c5d5 AM |
331 | /* |
332 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO | |
333 | * and __GFP_HIGHMEM from hard or soft interrupt context. | |
334 | */ | |
725d704e | 335 | VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); |
17cf4406 NP |
336 | for (i = 0; i < (1 << order); i++) |
337 | clear_highpage(page + i); | |
338 | } | |
339 | ||
1da177e4 LT |
340 | /* |
341 | * function for dealing with page's order in buddy system. | |
342 | * zone->lock is already acquired when we use these. | |
343 | * So, we don't need atomic page->flags operations here. | |
344 | */ | |
6aa3001b AM |
345 | static inline unsigned long page_order(struct page *page) |
346 | { | |
4c21e2f2 | 347 | return page_private(page); |
1da177e4 LT |
348 | } |
349 | ||
6aa3001b AM |
350 | static inline void set_page_order(struct page *page, int order) |
351 | { | |
4c21e2f2 | 352 | set_page_private(page, order); |
676165a8 | 353 | __SetPageBuddy(page); |
1da177e4 LT |
354 | } |
355 | ||
356 | static inline void rmv_page_order(struct page *page) | |
357 | { | |
676165a8 | 358 | __ClearPageBuddy(page); |
4c21e2f2 | 359 | set_page_private(page, 0); |
1da177e4 LT |
360 | } |
361 | ||
362 | /* | |
363 | * Locate the struct page for both the matching buddy in our | |
364 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
365 | * | |
366 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
367 | * the following equation: | |
368 | * B2 = B1 ^ (1 << O) | |
369 | * For example, if the starting buddy (buddy2) is #8 its order | |
370 | * 1 buddy is #10: | |
371 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
372 | * | |
373 | * 2) Any buddy B will have an order O+1 parent P which | |
374 | * satisfies the following equation: | |
375 | * P = B & ~(1 << O) | |
376 | * | |
d6e05edc | 377 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER |
1da177e4 LT |
378 | */ |
379 | static inline struct page * | |
380 | __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order) | |
381 | { | |
382 | unsigned long buddy_idx = page_idx ^ (1 << order); | |
383 | ||
384 | return page + (buddy_idx - page_idx); | |
385 | } | |
386 | ||
387 | static inline unsigned long | |
388 | __find_combined_index(unsigned long page_idx, unsigned int order) | |
389 | { | |
390 | return (page_idx & ~(1 << order)); | |
391 | } | |
392 | ||
393 | /* | |
394 | * This function checks whether a page is free && is the buddy | |
395 | * we can do coalesce a page and its buddy if | |
13e7444b | 396 | * (a) the buddy is not in a hole && |
676165a8 | 397 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
398 | * (c) a page and its buddy have the same order && |
399 | * (d) a page and its buddy are in the same zone. | |
676165a8 NP |
400 | * |
401 | * For recording whether a page is in the buddy system, we use PG_buddy. | |
402 | * Setting, clearing, and testing PG_buddy is serialized by zone->lock. | |
1da177e4 | 403 | * |
676165a8 | 404 | * For recording page's order, we use page_private(page). |
1da177e4 | 405 | */ |
cb2b95e1 AW |
406 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
407 | int order) | |
1da177e4 | 408 | { |
14e07298 | 409 | if (!pfn_valid_within(page_to_pfn(buddy))) |
13e7444b | 410 | return 0; |
13e7444b | 411 | |
cb2b95e1 AW |
412 | if (page_zone_id(page) != page_zone_id(buddy)) |
413 | return 0; | |
414 | ||
415 | if (PageBuddy(buddy) && page_order(buddy) == order) { | |
416 | BUG_ON(page_count(buddy) != 0); | |
6aa3001b | 417 | return 1; |
676165a8 | 418 | } |
6aa3001b | 419 | return 0; |
1da177e4 LT |
420 | } |
421 | ||
422 | /* | |
423 | * Freeing function for a buddy system allocator. | |
424 | * | |
425 | * The concept of a buddy system is to maintain direct-mapped table | |
426 | * (containing bit values) for memory blocks of various "orders". | |
427 | * The bottom level table contains the map for the smallest allocatable | |
428 | * units of memory (here, pages), and each level above it describes | |
429 | * pairs of units from the levels below, hence, "buddies". | |
430 | * At a high level, all that happens here is marking the table entry | |
431 | * at the bottom level available, and propagating the changes upward | |
432 | * as necessary, plus some accounting needed to play nicely with other | |
433 | * parts of the VM system. | |
434 | * At each level, we keep a list of pages, which are heads of continuous | |
676165a8 | 435 | * free pages of length of (1 << order) and marked with PG_buddy. Page's |
4c21e2f2 | 436 | * order is recorded in page_private(page) field. |
1da177e4 LT |
437 | * So when we are allocating or freeing one, we can derive the state of the |
438 | * other. That is, if we allocate a small block, and both were | |
439 | * free, the remainder of the region must be split into blocks. | |
440 | * If a block is freed, and its buddy is also free, then this | |
441 | * triggers coalescing into a block of larger size. | |
442 | * | |
443 | * -- wli | |
444 | */ | |
445 | ||
48db57f8 | 446 | static inline void __free_one_page(struct page *page, |
1da177e4 LT |
447 | struct zone *zone, unsigned int order) |
448 | { | |
449 | unsigned long page_idx; | |
450 | int order_size = 1 << order; | |
b2a0ac88 | 451 | int migratetype = get_pageblock_migratetype(page); |
1da177e4 | 452 | |
224abf92 | 453 | if (unlikely(PageCompound(page))) |
1da177e4 LT |
454 | destroy_compound_page(page, order); |
455 | ||
456 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); | |
457 | ||
725d704e NP |
458 | VM_BUG_ON(page_idx & (order_size - 1)); |
459 | VM_BUG_ON(bad_range(zone, page)); | |
1da177e4 | 460 | |
d23ad423 | 461 | __mod_zone_page_state(zone, NR_FREE_PAGES, order_size); |
1da177e4 LT |
462 | while (order < MAX_ORDER-1) { |
463 | unsigned long combined_idx; | |
1da177e4 LT |
464 | struct page *buddy; |
465 | ||
1da177e4 | 466 | buddy = __page_find_buddy(page, page_idx, order); |
cb2b95e1 | 467 | if (!page_is_buddy(page, buddy, order)) |
1da177e4 | 468 | break; /* Move the buddy up one level. */ |
13e7444b | 469 | |
1da177e4 | 470 | list_del(&buddy->lru); |
b2a0ac88 | 471 | zone->free_area[order].nr_free--; |
1da177e4 | 472 | rmv_page_order(buddy); |
13e7444b | 473 | combined_idx = __find_combined_index(page_idx, order); |
1da177e4 LT |
474 | page = page + (combined_idx - page_idx); |
475 | page_idx = combined_idx; | |
476 | order++; | |
477 | } | |
478 | set_page_order(page, order); | |
b2a0ac88 MG |
479 | list_add(&page->lru, |
480 | &zone->free_area[order].free_list[migratetype]); | |
1da177e4 LT |
481 | zone->free_area[order].nr_free++; |
482 | } | |
483 | ||
224abf92 | 484 | static inline int free_pages_check(struct page *page) |
1da177e4 | 485 | { |
92be2e33 NP |
486 | if (unlikely(page_mapcount(page) | |
487 | (page->mapping != NULL) | | |
488 | (page_count(page) != 0) | | |
1da177e4 LT |
489 | (page->flags & ( |
490 | 1 << PG_lru | | |
491 | 1 << PG_private | | |
492 | 1 << PG_locked | | |
493 | 1 << PG_active | | |
1da177e4 LT |
494 | 1 << PG_slab | |
495 | 1 << PG_swapcache | | |
b5810039 | 496 | 1 << PG_writeback | |
676165a8 NP |
497 | 1 << PG_reserved | |
498 | 1 << PG_buddy )))) | |
224abf92 | 499 | bad_page(page); |
1da177e4 | 500 | if (PageDirty(page)) |
242e5468 | 501 | __ClearPageDirty(page); |
689bcebf HD |
502 | /* |
503 | * For now, we report if PG_reserved was found set, but do not | |
504 | * clear it, and do not free the page. But we shall soon need | |
505 | * to do more, for when the ZERO_PAGE count wraps negative. | |
506 | */ | |
507 | return PageReserved(page); | |
1da177e4 LT |
508 | } |
509 | ||
510 | /* | |
511 | * Frees a list of pages. | |
512 | * Assumes all pages on list are in same zone, and of same order. | |
207f36ee | 513 | * count is the number of pages to free. |
1da177e4 LT |
514 | * |
515 | * If the zone was previously in an "all pages pinned" state then look to | |
516 | * see if this freeing clears that state. | |
517 | * | |
518 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
519 | * pinned" detection logic. | |
520 | */ | |
48db57f8 NP |
521 | static void free_pages_bulk(struct zone *zone, int count, |
522 | struct list_head *list, int order) | |
1da177e4 | 523 | { |
c54ad30c | 524 | spin_lock(&zone->lock); |
1da177e4 LT |
525 | zone->all_unreclaimable = 0; |
526 | zone->pages_scanned = 0; | |
48db57f8 NP |
527 | while (count--) { |
528 | struct page *page; | |
529 | ||
725d704e | 530 | VM_BUG_ON(list_empty(list)); |
1da177e4 | 531 | page = list_entry(list->prev, struct page, lru); |
48db57f8 | 532 | /* have to delete it as __free_one_page list manipulates */ |
1da177e4 | 533 | list_del(&page->lru); |
48db57f8 | 534 | __free_one_page(page, zone, order); |
1da177e4 | 535 | } |
c54ad30c | 536 | spin_unlock(&zone->lock); |
1da177e4 LT |
537 | } |
538 | ||
48db57f8 | 539 | static void free_one_page(struct zone *zone, struct page *page, int order) |
1da177e4 | 540 | { |
006d22d9 CL |
541 | spin_lock(&zone->lock); |
542 | zone->all_unreclaimable = 0; | |
543 | zone->pages_scanned = 0; | |
0798e519 | 544 | __free_one_page(page, zone, order); |
006d22d9 | 545 | spin_unlock(&zone->lock); |
48db57f8 NP |
546 | } |
547 | ||
548 | static void __free_pages_ok(struct page *page, unsigned int order) | |
549 | { | |
550 | unsigned long flags; | |
1da177e4 | 551 | int i; |
689bcebf | 552 | int reserved = 0; |
1da177e4 | 553 | |
1da177e4 | 554 | for (i = 0 ; i < (1 << order) ; ++i) |
224abf92 | 555 | reserved += free_pages_check(page + i); |
689bcebf HD |
556 | if (reserved) |
557 | return; | |
558 | ||
9858db50 NP |
559 | if (!PageHighMem(page)) |
560 | debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order); | |
dafb1367 | 561 | arch_free_page(page, order); |
48db57f8 | 562 | kernel_map_pages(page, 1 << order, 0); |
dafb1367 | 563 | |
c54ad30c | 564 | local_irq_save(flags); |
f8891e5e | 565 | __count_vm_events(PGFREE, 1 << order); |
48db57f8 | 566 | free_one_page(page_zone(page), page, order); |
c54ad30c | 567 | local_irq_restore(flags); |
1da177e4 LT |
568 | } |
569 | ||
a226f6c8 DH |
570 | /* |
571 | * permit the bootmem allocator to evade page validation on high-order frees | |
572 | */ | |
573 | void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order) | |
574 | { | |
575 | if (order == 0) { | |
576 | __ClearPageReserved(page); | |
577 | set_page_count(page, 0); | |
7835e98b | 578 | set_page_refcounted(page); |
545b1ea9 | 579 | __free_page(page); |
a226f6c8 | 580 | } else { |
a226f6c8 DH |
581 | int loop; |
582 | ||
545b1ea9 | 583 | prefetchw(page); |
a226f6c8 DH |
584 | for (loop = 0; loop < BITS_PER_LONG; loop++) { |
585 | struct page *p = &page[loop]; | |
586 | ||
545b1ea9 NP |
587 | if (loop + 1 < BITS_PER_LONG) |
588 | prefetchw(p + 1); | |
a226f6c8 DH |
589 | __ClearPageReserved(p); |
590 | set_page_count(p, 0); | |
591 | } | |
592 | ||
7835e98b | 593 | set_page_refcounted(page); |
545b1ea9 | 594 | __free_pages(page, order); |
a226f6c8 DH |
595 | } |
596 | } | |
597 | ||
1da177e4 LT |
598 | |
599 | /* | |
600 | * The order of subdivision here is critical for the IO subsystem. | |
601 | * Please do not alter this order without good reasons and regression | |
602 | * testing. Specifically, as large blocks of memory are subdivided, | |
603 | * the order in which smaller blocks are delivered depends on the order | |
604 | * they're subdivided in this function. This is the primary factor | |
605 | * influencing the order in which pages are delivered to the IO | |
606 | * subsystem according to empirical testing, and this is also justified | |
607 | * by considering the behavior of a buddy system containing a single | |
608 | * large block of memory acted on by a series of small allocations. | |
609 | * This behavior is a critical factor in sglist merging's success. | |
610 | * | |
611 | * -- wli | |
612 | */ | |
085cc7d5 | 613 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
614 | int low, int high, struct free_area *area, |
615 | int migratetype) | |
1da177e4 LT |
616 | { |
617 | unsigned long size = 1 << high; | |
618 | ||
619 | while (high > low) { | |
620 | area--; | |
621 | high--; | |
622 | size >>= 1; | |
725d704e | 623 | VM_BUG_ON(bad_range(zone, &page[size])); |
b2a0ac88 | 624 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
625 | area->nr_free++; |
626 | set_page_order(&page[size], high); | |
627 | } | |
1da177e4 LT |
628 | } |
629 | ||
1da177e4 LT |
630 | /* |
631 | * This page is about to be returned from the page allocator | |
632 | */ | |
17cf4406 | 633 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) |
1da177e4 | 634 | { |
92be2e33 NP |
635 | if (unlikely(page_mapcount(page) | |
636 | (page->mapping != NULL) | | |
637 | (page_count(page) != 0) | | |
334795ec HD |
638 | (page->flags & ( |
639 | 1 << PG_lru | | |
1da177e4 LT |
640 | 1 << PG_private | |
641 | 1 << PG_locked | | |
1da177e4 LT |
642 | 1 << PG_active | |
643 | 1 << PG_dirty | | |
334795ec | 644 | 1 << PG_slab | |
1da177e4 | 645 | 1 << PG_swapcache | |
b5810039 | 646 | 1 << PG_writeback | |
676165a8 NP |
647 | 1 << PG_reserved | |
648 | 1 << PG_buddy )))) | |
224abf92 | 649 | bad_page(page); |
1da177e4 | 650 | |
689bcebf HD |
651 | /* |
652 | * For now, we report if PG_reserved was found set, but do not | |
653 | * clear it, and do not allocate the page: as a safety net. | |
654 | */ | |
655 | if (PageReserved(page)) | |
656 | return 1; | |
657 | ||
d77c2d7c | 658 | page->flags &= ~(1 << PG_uptodate | 1 << PG_error | 1 << PG_readahead | |
1da177e4 | 659 | 1 << PG_referenced | 1 << PG_arch_1 | |
5409bae0 | 660 | 1 << PG_owner_priv_1 | 1 << PG_mappedtodisk); |
4c21e2f2 | 661 | set_page_private(page, 0); |
7835e98b | 662 | set_page_refcounted(page); |
cc102509 NP |
663 | |
664 | arch_alloc_page(page, order); | |
1da177e4 | 665 | kernel_map_pages(page, 1 << order, 1); |
17cf4406 NP |
666 | |
667 | if (gfp_flags & __GFP_ZERO) | |
668 | prep_zero_page(page, order, gfp_flags); | |
669 | ||
670 | if (order && (gfp_flags & __GFP_COMP)) | |
671 | prep_compound_page(page, order); | |
672 | ||
689bcebf | 673 | return 0; |
1da177e4 LT |
674 | } |
675 | ||
b92a6edd | 676 | #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY |
b2a0ac88 MG |
677 | /* |
678 | * This array describes the order lists are fallen back to when | |
679 | * the free lists for the desirable migrate type are depleted | |
680 | */ | |
681 | static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = { | |
e12ba74d MG |
682 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE }, |
683 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE }, | |
684 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE }, | |
b2a0ac88 MG |
685 | }; |
686 | ||
c361be55 MG |
687 | /* |
688 | * Move the free pages in a range to the free lists of the requested type. | |
689 | * Note that start_page and end_pages are not aligned in a MAX_ORDER_NR_PAGES | |
690 | * boundary. If alignment is required, use move_freepages_block() | |
691 | */ | |
692 | int move_freepages(struct zone *zone, | |
693 | struct page *start_page, struct page *end_page, | |
694 | int migratetype) | |
695 | { | |
696 | struct page *page; | |
697 | unsigned long order; | |
698 | int blocks_moved = 0; | |
699 | ||
700 | #ifndef CONFIG_HOLES_IN_ZONE | |
701 | /* | |
702 | * page_zone is not safe to call in this context when | |
703 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
704 | * anyway as we check zone boundaries in move_freepages_block(). | |
705 | * Remove at a later date when no bug reports exist related to | |
706 | * CONFIG_PAGE_GROUP_BY_MOBILITY | |
707 | */ | |
708 | BUG_ON(page_zone(start_page) != page_zone(end_page)); | |
709 | #endif | |
710 | ||
711 | for (page = start_page; page <= end_page;) { | |
712 | if (!pfn_valid_within(page_to_pfn(page))) { | |
713 | page++; | |
714 | continue; | |
715 | } | |
716 | ||
717 | if (!PageBuddy(page)) { | |
718 | page++; | |
719 | continue; | |
720 | } | |
721 | ||
722 | order = page_order(page); | |
723 | list_del(&page->lru); | |
724 | list_add(&page->lru, | |
725 | &zone->free_area[order].free_list[migratetype]); | |
726 | page += 1 << order; | |
727 | blocks_moved++; | |
728 | } | |
729 | ||
730 | return blocks_moved; | |
731 | } | |
732 | ||
733 | int move_freepages_block(struct zone *zone, struct page *page, int migratetype) | |
734 | { | |
735 | unsigned long start_pfn, end_pfn; | |
736 | struct page *start_page, *end_page; | |
737 | ||
738 | start_pfn = page_to_pfn(page); | |
739 | start_pfn = start_pfn & ~(MAX_ORDER_NR_PAGES-1); | |
740 | start_page = pfn_to_page(start_pfn); | |
741 | end_page = start_page + MAX_ORDER_NR_PAGES - 1; | |
742 | end_pfn = start_pfn + MAX_ORDER_NR_PAGES - 1; | |
743 | ||
744 | /* Do not cross zone boundaries */ | |
745 | if (start_pfn < zone->zone_start_pfn) | |
746 | start_page = page; | |
747 | if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
748 | return 0; | |
749 | ||
750 | return move_freepages(zone, start_page, end_page, migratetype); | |
751 | } | |
752 | ||
b2a0ac88 MG |
753 | /* Remove an element from the buddy allocator from the fallback list */ |
754 | static struct page *__rmqueue_fallback(struct zone *zone, int order, | |
755 | int start_migratetype) | |
756 | { | |
757 | struct free_area * area; | |
758 | int current_order; | |
759 | struct page *page; | |
760 | int migratetype, i; | |
761 | ||
762 | /* Find the largest possible block of pages in the other list */ | |
763 | for (current_order = MAX_ORDER-1; current_order >= order; | |
764 | --current_order) { | |
765 | for (i = 0; i < MIGRATE_TYPES - 1; i++) { | |
766 | migratetype = fallbacks[start_migratetype][i]; | |
767 | ||
768 | area = &(zone->free_area[current_order]); | |
769 | if (list_empty(&area->free_list[migratetype])) | |
770 | continue; | |
771 | ||
772 | page = list_entry(area->free_list[migratetype].next, | |
773 | struct page, lru); | |
774 | area->nr_free--; | |
775 | ||
776 | /* | |
c361be55 MG |
777 | * If breaking a large block of pages, move all free |
778 | * pages to the preferred allocation list | |
b2a0ac88 | 779 | */ |
c361be55 | 780 | if (unlikely(current_order >= MAX_ORDER / 2)) { |
b2a0ac88 | 781 | migratetype = start_migratetype; |
c361be55 MG |
782 | move_freepages_block(zone, page, migratetype); |
783 | } | |
b2a0ac88 MG |
784 | |
785 | /* Remove the page from the freelists */ | |
786 | list_del(&page->lru); | |
787 | rmv_page_order(page); | |
788 | __mod_zone_page_state(zone, NR_FREE_PAGES, | |
789 | -(1UL << order)); | |
790 | ||
791 | if (current_order == MAX_ORDER - 1) | |
792 | set_pageblock_migratetype(page, | |
793 | start_migratetype); | |
794 | ||
795 | expand(zone, page, order, current_order, area, migratetype); | |
796 | return page; | |
797 | } | |
798 | } | |
799 | ||
800 | return NULL; | |
801 | } | |
b92a6edd MG |
802 | #else |
803 | static struct page *__rmqueue_fallback(struct zone *zone, int order, | |
804 | int start_migratetype) | |
805 | { | |
806 | return NULL; | |
807 | } | |
808 | #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */ | |
b2a0ac88 | 809 | |
1da177e4 LT |
810 | /* |
811 | * Do the hard work of removing an element from the buddy allocator. | |
812 | * Call me with the zone->lock already held. | |
813 | */ | |
b2a0ac88 MG |
814 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
815 | int migratetype) | |
1da177e4 LT |
816 | { |
817 | struct free_area * area; | |
818 | unsigned int current_order; | |
819 | struct page *page; | |
820 | ||
b2a0ac88 | 821 | /* Find a page of the appropriate size in the preferred list */ |
1da177e4 | 822 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { |
b2a0ac88 MG |
823 | area = &(zone->free_area[current_order]); |
824 | if (list_empty(&area->free_list[migratetype])) | |
1da177e4 LT |
825 | continue; |
826 | ||
b2a0ac88 MG |
827 | page = list_entry(area->free_list[migratetype].next, |
828 | struct page, lru); | |
1da177e4 LT |
829 | list_del(&page->lru); |
830 | rmv_page_order(page); | |
831 | area->nr_free--; | |
d23ad423 | 832 | __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order)); |
b2a0ac88 MG |
833 | expand(zone, page, order, current_order, area, migratetype); |
834 | goto got_page; | |
1da177e4 LT |
835 | } |
836 | ||
b2a0ac88 MG |
837 | page = __rmqueue_fallback(zone, order, migratetype); |
838 | ||
839 | got_page: | |
840 | ||
841 | return page; | |
1da177e4 LT |
842 | } |
843 | ||
844 | /* | |
845 | * Obtain a specified number of elements from the buddy allocator, all under | |
846 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
847 | * Returns the number of new pages which were placed at *list. | |
848 | */ | |
849 | static int rmqueue_bulk(struct zone *zone, unsigned int order, | |
b2a0ac88 MG |
850 | unsigned long count, struct list_head *list, |
851 | int migratetype) | |
1da177e4 | 852 | { |
1da177e4 | 853 | int i; |
1da177e4 | 854 | |
c54ad30c | 855 | spin_lock(&zone->lock); |
1da177e4 | 856 | for (i = 0; i < count; ++i) { |
b2a0ac88 | 857 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 858 | if (unlikely(page == NULL)) |
1da177e4 | 859 | break; |
535131e6 MG |
860 | list_add(&page->lru, list); |
861 | set_page_private(page, migratetype); | |
1da177e4 | 862 | } |
c54ad30c | 863 | spin_unlock(&zone->lock); |
085cc7d5 | 864 | return i; |
1da177e4 LT |
865 | } |
866 | ||
4ae7c039 | 867 | #ifdef CONFIG_NUMA |
8fce4d8e | 868 | /* |
4037d452 CL |
869 | * Called from the vmstat counter updater to drain pagesets of this |
870 | * currently executing processor on remote nodes after they have | |
871 | * expired. | |
872 | * | |
879336c3 CL |
873 | * Note that this function must be called with the thread pinned to |
874 | * a single processor. | |
8fce4d8e | 875 | */ |
4037d452 | 876 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 877 | { |
4ae7c039 | 878 | unsigned long flags; |
4037d452 | 879 | int to_drain; |
4ae7c039 | 880 | |
4037d452 CL |
881 | local_irq_save(flags); |
882 | if (pcp->count >= pcp->batch) | |
883 | to_drain = pcp->batch; | |
884 | else | |
885 | to_drain = pcp->count; | |
886 | free_pages_bulk(zone, to_drain, &pcp->list, 0); | |
887 | pcp->count -= to_drain; | |
888 | local_irq_restore(flags); | |
4ae7c039 CL |
889 | } |
890 | #endif | |
891 | ||
1da177e4 LT |
892 | static void __drain_pages(unsigned int cpu) |
893 | { | |
c54ad30c | 894 | unsigned long flags; |
1da177e4 LT |
895 | struct zone *zone; |
896 | int i; | |
897 | ||
898 | for_each_zone(zone) { | |
899 | struct per_cpu_pageset *pset; | |
900 | ||
f2e12bb2 CL |
901 | if (!populated_zone(zone)) |
902 | continue; | |
903 | ||
e7c8d5c9 | 904 | pset = zone_pcp(zone, cpu); |
1da177e4 LT |
905 | for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) { |
906 | struct per_cpu_pages *pcp; | |
907 | ||
908 | pcp = &pset->pcp[i]; | |
c54ad30c | 909 | local_irq_save(flags); |
48db57f8 NP |
910 | free_pages_bulk(zone, pcp->count, &pcp->list, 0); |
911 | pcp->count = 0; | |
c54ad30c | 912 | local_irq_restore(flags); |
1da177e4 LT |
913 | } |
914 | } | |
915 | } | |
1da177e4 | 916 | |
296699de | 917 | #ifdef CONFIG_HIBERNATION |
1da177e4 LT |
918 | |
919 | void mark_free_pages(struct zone *zone) | |
920 | { | |
f623f0db RW |
921 | unsigned long pfn, max_zone_pfn; |
922 | unsigned long flags; | |
b2a0ac88 | 923 | int order, t; |
1da177e4 LT |
924 | struct list_head *curr; |
925 | ||
926 | if (!zone->spanned_pages) | |
927 | return; | |
928 | ||
929 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db RW |
930 | |
931 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
932 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
933 | if (pfn_valid(pfn)) { | |
934 | struct page *page = pfn_to_page(pfn); | |
935 | ||
7be98234 RW |
936 | if (!swsusp_page_is_forbidden(page)) |
937 | swsusp_unset_page_free(page); | |
f623f0db | 938 | } |
1da177e4 | 939 | |
b2a0ac88 MG |
940 | for_each_migratetype_order(order, t) { |
941 | list_for_each(curr, &zone->free_area[order].free_list[t]) { | |
f623f0db | 942 | unsigned long i; |
1da177e4 | 943 | |
f623f0db RW |
944 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); |
945 | for (i = 0; i < (1UL << order); i++) | |
7be98234 | 946 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 947 | } |
b2a0ac88 | 948 | } |
1da177e4 LT |
949 | spin_unlock_irqrestore(&zone->lock, flags); |
950 | } | |
e2c55dc8 | 951 | #endif /* CONFIG_PM */ |
1da177e4 | 952 | |
e2c55dc8 | 953 | #if defined(CONFIG_HIBERNATION) || defined(CONFIG_PAGE_GROUP_BY_MOBILITY) |
1da177e4 LT |
954 | /* |
955 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
956 | */ | |
957 | void drain_local_pages(void) | |
958 | { | |
959 | unsigned long flags; | |
960 | ||
961 | local_irq_save(flags); | |
962 | __drain_pages(smp_processor_id()); | |
963 | local_irq_restore(flags); | |
964 | } | |
e2c55dc8 MG |
965 | |
966 | void smp_drain_local_pages(void *arg) | |
967 | { | |
968 | drain_local_pages(); | |
969 | } | |
970 | ||
971 | /* | |
972 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator | |
973 | */ | |
974 | void drain_all_local_pages(void) | |
975 | { | |
976 | unsigned long flags; | |
977 | ||
978 | local_irq_save(flags); | |
979 | __drain_pages(smp_processor_id()); | |
980 | local_irq_restore(flags); | |
981 | ||
982 | smp_call_function(smp_drain_local_pages, NULL, 0, 1); | |
983 | } | |
984 | #else | |
985 | void drain_all_local_pages(void) {} | |
986 | #endif /* CONFIG_HIBERNATION || CONFIG_PAGE_GROUP_BY_MOBILITY */ | |
1da177e4 | 987 | |
1da177e4 LT |
988 | /* |
989 | * Free a 0-order page | |
990 | */ | |
1da177e4 LT |
991 | static void fastcall free_hot_cold_page(struct page *page, int cold) |
992 | { | |
993 | struct zone *zone = page_zone(page); | |
994 | struct per_cpu_pages *pcp; | |
995 | unsigned long flags; | |
996 | ||
1da177e4 LT |
997 | if (PageAnon(page)) |
998 | page->mapping = NULL; | |
224abf92 | 999 | if (free_pages_check(page)) |
689bcebf HD |
1000 | return; |
1001 | ||
9858db50 NP |
1002 | if (!PageHighMem(page)) |
1003 | debug_check_no_locks_freed(page_address(page), PAGE_SIZE); | |
dafb1367 | 1004 | arch_free_page(page, 0); |
689bcebf HD |
1005 | kernel_map_pages(page, 1, 0); |
1006 | ||
e7c8d5c9 | 1007 | pcp = &zone_pcp(zone, get_cpu())->pcp[cold]; |
1da177e4 | 1008 | local_irq_save(flags); |
f8891e5e | 1009 | __count_vm_event(PGFREE); |
1da177e4 | 1010 | list_add(&page->lru, &pcp->list); |
535131e6 | 1011 | set_page_private(page, get_pageblock_migratetype(page)); |
1da177e4 | 1012 | pcp->count++; |
48db57f8 NP |
1013 | if (pcp->count >= pcp->high) { |
1014 | free_pages_bulk(zone, pcp->batch, &pcp->list, 0); | |
1015 | pcp->count -= pcp->batch; | |
1016 | } | |
1da177e4 LT |
1017 | local_irq_restore(flags); |
1018 | put_cpu(); | |
1019 | } | |
1020 | ||
1021 | void fastcall free_hot_page(struct page *page) | |
1022 | { | |
1023 | free_hot_cold_page(page, 0); | |
1024 | } | |
1025 | ||
1026 | void fastcall free_cold_page(struct page *page) | |
1027 | { | |
1028 | free_hot_cold_page(page, 1); | |
1029 | } | |
1030 | ||
8dfcc9ba NP |
1031 | /* |
1032 | * split_page takes a non-compound higher-order page, and splits it into | |
1033 | * n (1<<order) sub-pages: page[0..n] | |
1034 | * Each sub-page must be freed individually. | |
1035 | * | |
1036 | * Note: this is probably too low level an operation for use in drivers. | |
1037 | * Please consult with lkml before using this in your driver. | |
1038 | */ | |
1039 | void split_page(struct page *page, unsigned int order) | |
1040 | { | |
1041 | int i; | |
1042 | ||
725d704e NP |
1043 | VM_BUG_ON(PageCompound(page)); |
1044 | VM_BUG_ON(!page_count(page)); | |
7835e98b NP |
1045 | for (i = 1; i < (1 << order); i++) |
1046 | set_page_refcounted(page + i); | |
8dfcc9ba | 1047 | } |
8dfcc9ba | 1048 | |
1da177e4 LT |
1049 | /* |
1050 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
1051 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
1052 | * or two. | |
1053 | */ | |
a74609fa NP |
1054 | static struct page *buffered_rmqueue(struct zonelist *zonelist, |
1055 | struct zone *zone, int order, gfp_t gfp_flags) | |
1da177e4 LT |
1056 | { |
1057 | unsigned long flags; | |
689bcebf | 1058 | struct page *page; |
1da177e4 | 1059 | int cold = !!(gfp_flags & __GFP_COLD); |
a74609fa | 1060 | int cpu; |
b2a0ac88 | 1061 | int migratetype = gfpflags_to_migratetype(gfp_flags); |
1da177e4 | 1062 | |
689bcebf | 1063 | again: |
a74609fa | 1064 | cpu = get_cpu(); |
48db57f8 | 1065 | if (likely(order == 0)) { |
1da177e4 LT |
1066 | struct per_cpu_pages *pcp; |
1067 | ||
a74609fa | 1068 | pcp = &zone_pcp(zone, cpu)->pcp[cold]; |
1da177e4 | 1069 | local_irq_save(flags); |
a74609fa | 1070 | if (!pcp->count) { |
941c7105 | 1071 | pcp->count = rmqueue_bulk(zone, 0, |
b2a0ac88 | 1072 | pcp->batch, &pcp->list, migratetype); |
a74609fa NP |
1073 | if (unlikely(!pcp->count)) |
1074 | goto failed; | |
1da177e4 | 1075 | } |
b92a6edd MG |
1076 | |
1077 | #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY | |
535131e6 | 1078 | /* Find a page of the appropriate migrate type */ |
b92a6edd MG |
1079 | list_for_each_entry(page, &pcp->list, lru) |
1080 | if (page_private(page) == migratetype) | |
535131e6 | 1081 | break; |
535131e6 | 1082 | |
b92a6edd MG |
1083 | /* Allocate more to the pcp list if necessary */ |
1084 | if (unlikely(&page->lru == &pcp->list)) { | |
535131e6 MG |
1085 | pcp->count += rmqueue_bulk(zone, 0, |
1086 | pcp->batch, &pcp->list, migratetype); | |
1087 | page = list_entry(pcp->list.next, struct page, lru); | |
535131e6 | 1088 | } |
b92a6edd MG |
1089 | #else |
1090 | page = list_entry(pcp->list.next, struct page, lru); | |
1091 | #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */ | |
1092 | ||
1093 | list_del(&page->lru); | |
1094 | pcp->count--; | |
7fb1d9fc | 1095 | } else { |
1da177e4 | 1096 | spin_lock_irqsave(&zone->lock, flags); |
b2a0ac88 | 1097 | page = __rmqueue(zone, order, migratetype); |
a74609fa NP |
1098 | spin_unlock(&zone->lock); |
1099 | if (!page) | |
1100 | goto failed; | |
1da177e4 LT |
1101 | } |
1102 | ||
f8891e5e | 1103 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
ca889e6c | 1104 | zone_statistics(zonelist, zone); |
a74609fa NP |
1105 | local_irq_restore(flags); |
1106 | put_cpu(); | |
1da177e4 | 1107 | |
725d704e | 1108 | VM_BUG_ON(bad_range(zone, page)); |
17cf4406 | 1109 | if (prep_new_page(page, order, gfp_flags)) |
a74609fa | 1110 | goto again; |
1da177e4 | 1111 | return page; |
a74609fa NP |
1112 | |
1113 | failed: | |
1114 | local_irq_restore(flags); | |
1115 | put_cpu(); | |
1116 | return NULL; | |
1da177e4 LT |
1117 | } |
1118 | ||
7fb1d9fc | 1119 | #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */ |
3148890b NP |
1120 | #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */ |
1121 | #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */ | |
1122 | #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */ | |
1123 | #define ALLOC_HARDER 0x10 /* try to alloc harder */ | |
1124 | #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */ | |
1125 | #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ | |
7fb1d9fc | 1126 | |
933e312e AM |
1127 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
1128 | ||
1129 | static struct fail_page_alloc_attr { | |
1130 | struct fault_attr attr; | |
1131 | ||
1132 | u32 ignore_gfp_highmem; | |
1133 | u32 ignore_gfp_wait; | |
54114994 | 1134 | u32 min_order; |
933e312e AM |
1135 | |
1136 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1137 | ||
1138 | struct dentry *ignore_gfp_highmem_file; | |
1139 | struct dentry *ignore_gfp_wait_file; | |
54114994 | 1140 | struct dentry *min_order_file; |
933e312e AM |
1141 | |
1142 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1143 | ||
1144 | } fail_page_alloc = { | |
1145 | .attr = FAULT_ATTR_INITIALIZER, | |
6b1b60f4 DM |
1146 | .ignore_gfp_wait = 1, |
1147 | .ignore_gfp_highmem = 1, | |
54114994 | 1148 | .min_order = 1, |
933e312e AM |
1149 | }; |
1150 | ||
1151 | static int __init setup_fail_page_alloc(char *str) | |
1152 | { | |
1153 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
1154 | } | |
1155 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
1156 | ||
1157 | static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
1158 | { | |
54114994 AM |
1159 | if (order < fail_page_alloc.min_order) |
1160 | return 0; | |
933e312e AM |
1161 | if (gfp_mask & __GFP_NOFAIL) |
1162 | return 0; | |
1163 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) | |
1164 | return 0; | |
1165 | if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT)) | |
1166 | return 0; | |
1167 | ||
1168 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
1169 | } | |
1170 | ||
1171 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1172 | ||
1173 | static int __init fail_page_alloc_debugfs(void) | |
1174 | { | |
1175 | mode_t mode = S_IFREG | S_IRUSR | S_IWUSR; | |
1176 | struct dentry *dir; | |
1177 | int err; | |
1178 | ||
1179 | err = init_fault_attr_dentries(&fail_page_alloc.attr, | |
1180 | "fail_page_alloc"); | |
1181 | if (err) | |
1182 | return err; | |
1183 | dir = fail_page_alloc.attr.dentries.dir; | |
1184 | ||
1185 | fail_page_alloc.ignore_gfp_wait_file = | |
1186 | debugfs_create_bool("ignore-gfp-wait", mode, dir, | |
1187 | &fail_page_alloc.ignore_gfp_wait); | |
1188 | ||
1189 | fail_page_alloc.ignore_gfp_highmem_file = | |
1190 | debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
1191 | &fail_page_alloc.ignore_gfp_highmem); | |
54114994 AM |
1192 | fail_page_alloc.min_order_file = |
1193 | debugfs_create_u32("min-order", mode, dir, | |
1194 | &fail_page_alloc.min_order); | |
933e312e AM |
1195 | |
1196 | if (!fail_page_alloc.ignore_gfp_wait_file || | |
54114994 AM |
1197 | !fail_page_alloc.ignore_gfp_highmem_file || |
1198 | !fail_page_alloc.min_order_file) { | |
933e312e AM |
1199 | err = -ENOMEM; |
1200 | debugfs_remove(fail_page_alloc.ignore_gfp_wait_file); | |
1201 | debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file); | |
54114994 | 1202 | debugfs_remove(fail_page_alloc.min_order_file); |
933e312e AM |
1203 | cleanup_fault_attr_dentries(&fail_page_alloc.attr); |
1204 | } | |
1205 | ||
1206 | return err; | |
1207 | } | |
1208 | ||
1209 | late_initcall(fail_page_alloc_debugfs); | |
1210 | ||
1211 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1212 | ||
1213 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
1214 | ||
1215 | static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
1216 | { | |
1217 | return 0; | |
1218 | } | |
1219 | ||
1220 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
1221 | ||
1da177e4 LT |
1222 | /* |
1223 | * Return 1 if free pages are above 'mark'. This takes into account the order | |
1224 | * of the allocation. | |
1225 | */ | |
1226 | int zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
7fb1d9fc | 1227 | int classzone_idx, int alloc_flags) |
1da177e4 LT |
1228 | { |
1229 | /* free_pages my go negative - that's OK */ | |
d23ad423 CL |
1230 | long min = mark; |
1231 | long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1; | |
1da177e4 LT |
1232 | int o; |
1233 | ||
7fb1d9fc | 1234 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 1235 | min -= min / 2; |
7fb1d9fc | 1236 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 LT |
1237 | min -= min / 4; |
1238 | ||
1239 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
1240 | return 0; | |
1241 | for (o = 0; o < order; o++) { | |
1242 | /* At the next order, this order's pages become unavailable */ | |
1243 | free_pages -= z->free_area[o].nr_free << o; | |
1244 | ||
1245 | /* Require fewer higher order pages to be free */ | |
1246 | min >>= 1; | |
1247 | ||
1248 | if (free_pages <= min) | |
1249 | return 0; | |
1250 | } | |
1251 | return 1; | |
1252 | } | |
1253 | ||
9276b1bc PJ |
1254 | #ifdef CONFIG_NUMA |
1255 | /* | |
1256 | * zlc_setup - Setup for "zonelist cache". Uses cached zone data to | |
1257 | * skip over zones that are not allowed by the cpuset, or that have | |
1258 | * been recently (in last second) found to be nearly full. See further | |
1259 | * comments in mmzone.h. Reduces cache footprint of zonelist scans | |
1260 | * that have to skip over alot of full or unallowed zones. | |
1261 | * | |
1262 | * If the zonelist cache is present in the passed in zonelist, then | |
1263 | * returns a pointer to the allowed node mask (either the current | |
37b07e41 | 1264 | * tasks mems_allowed, or node_states[N_HIGH_MEMORY].) |
9276b1bc PJ |
1265 | * |
1266 | * If the zonelist cache is not available for this zonelist, does | |
1267 | * nothing and returns NULL. | |
1268 | * | |
1269 | * If the fullzones BITMAP in the zonelist cache is stale (more than | |
1270 | * a second since last zap'd) then we zap it out (clear its bits.) | |
1271 | * | |
1272 | * We hold off even calling zlc_setup, until after we've checked the | |
1273 | * first zone in the zonelist, on the theory that most allocations will | |
1274 | * be satisfied from that first zone, so best to examine that zone as | |
1275 | * quickly as we can. | |
1276 | */ | |
1277 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1278 | { | |
1279 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1280 | nodemask_t *allowednodes; /* zonelist_cache approximation */ | |
1281 | ||
1282 | zlc = zonelist->zlcache_ptr; | |
1283 | if (!zlc) | |
1284 | return NULL; | |
1285 | ||
1286 | if (jiffies - zlc->last_full_zap > 1 * HZ) { | |
1287 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
1288 | zlc->last_full_zap = jiffies; | |
1289 | } | |
1290 | ||
1291 | allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ? | |
1292 | &cpuset_current_mems_allowed : | |
37b07e41 | 1293 | &node_states[N_HIGH_MEMORY]; |
9276b1bc PJ |
1294 | return allowednodes; |
1295 | } | |
1296 | ||
1297 | /* | |
1298 | * Given 'z' scanning a zonelist, run a couple of quick checks to see | |
1299 | * if it is worth looking at further for free memory: | |
1300 | * 1) Check that the zone isn't thought to be full (doesn't have its | |
1301 | * bit set in the zonelist_cache fullzones BITMAP). | |
1302 | * 2) Check that the zones node (obtained from the zonelist_cache | |
1303 | * z_to_n[] mapping) is allowed in the passed in allowednodes mask. | |
1304 | * Return true (non-zero) if zone is worth looking at further, or | |
1305 | * else return false (zero) if it is not. | |
1306 | * | |
1307 | * This check -ignores- the distinction between various watermarks, | |
1308 | * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is | |
1309 | * found to be full for any variation of these watermarks, it will | |
1310 | * be considered full for up to one second by all requests, unless | |
1311 | * we are so low on memory on all allowed nodes that we are forced | |
1312 | * into the second scan of the zonelist. | |
1313 | * | |
1314 | * In the second scan we ignore this zonelist cache and exactly | |
1315 | * apply the watermarks to all zones, even it is slower to do so. | |
1316 | * We are low on memory in the second scan, and should leave no stone | |
1317 | * unturned looking for a free page. | |
1318 | */ | |
1319 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z, | |
1320 | nodemask_t *allowednodes) | |
1321 | { | |
1322 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1323 | int i; /* index of *z in zonelist zones */ | |
1324 | int n; /* node that zone *z is on */ | |
1325 | ||
1326 | zlc = zonelist->zlcache_ptr; | |
1327 | if (!zlc) | |
1328 | return 1; | |
1329 | ||
1330 | i = z - zonelist->zones; | |
1331 | n = zlc->z_to_n[i]; | |
1332 | ||
1333 | /* This zone is worth trying if it is allowed but not full */ | |
1334 | return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones); | |
1335 | } | |
1336 | ||
1337 | /* | |
1338 | * Given 'z' scanning a zonelist, set the corresponding bit in | |
1339 | * zlc->fullzones, so that subsequent attempts to allocate a page | |
1340 | * from that zone don't waste time re-examining it. | |
1341 | */ | |
1342 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z) | |
1343 | { | |
1344 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1345 | int i; /* index of *z in zonelist zones */ | |
1346 | ||
1347 | zlc = zonelist->zlcache_ptr; | |
1348 | if (!zlc) | |
1349 | return; | |
1350 | ||
1351 | i = z - zonelist->zones; | |
1352 | ||
1353 | set_bit(i, zlc->fullzones); | |
1354 | } | |
1355 | ||
1356 | #else /* CONFIG_NUMA */ | |
1357 | ||
1358 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1359 | { | |
1360 | return NULL; | |
1361 | } | |
1362 | ||
1363 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z, | |
1364 | nodemask_t *allowednodes) | |
1365 | { | |
1366 | return 1; | |
1367 | } | |
1368 | ||
1369 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z) | |
1370 | { | |
1371 | } | |
1372 | #endif /* CONFIG_NUMA */ | |
1373 | ||
7fb1d9fc | 1374 | /* |
0798e519 | 1375 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
1376 | * a page. |
1377 | */ | |
1378 | static struct page * | |
1379 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, | |
1380 | struct zonelist *zonelist, int alloc_flags) | |
753ee728 | 1381 | { |
9276b1bc | 1382 | struct zone **z; |
7fb1d9fc | 1383 | struct page *page = NULL; |
9276b1bc | 1384 | int classzone_idx = zone_idx(zonelist->zones[0]); |
1192d526 | 1385 | struct zone *zone; |
9276b1bc PJ |
1386 | nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */ |
1387 | int zlc_active = 0; /* set if using zonelist_cache */ | |
1388 | int did_zlc_setup = 0; /* just call zlc_setup() one time */ | |
b377fd39 | 1389 | enum zone_type highest_zoneidx = -1; /* Gets set for policy zonelists */ |
7fb1d9fc | 1390 | |
9276b1bc | 1391 | zonelist_scan: |
7fb1d9fc | 1392 | /* |
9276b1bc | 1393 | * Scan zonelist, looking for a zone with enough free. |
7fb1d9fc RS |
1394 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1395 | */ | |
9276b1bc PJ |
1396 | z = zonelist->zones; |
1397 | ||
7fb1d9fc | 1398 | do { |
b377fd39 MG |
1399 | /* |
1400 | * In NUMA, this could be a policy zonelist which contains | |
1401 | * zones that may not be allowed by the current gfp_mask. | |
1402 | * Check the zone is allowed by the current flags | |
1403 | */ | |
1404 | if (unlikely(alloc_should_filter_zonelist(zonelist))) { | |
1405 | if (highest_zoneidx == -1) | |
1406 | highest_zoneidx = gfp_zone(gfp_mask); | |
1407 | if (zone_idx(*z) > highest_zoneidx) | |
1408 | continue; | |
1409 | } | |
1410 | ||
9276b1bc PJ |
1411 | if (NUMA_BUILD && zlc_active && |
1412 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) | |
1413 | continue; | |
1192d526 | 1414 | zone = *z; |
7fb1d9fc | 1415 | if ((alloc_flags & ALLOC_CPUSET) && |
02a0e53d | 1416 | !cpuset_zone_allowed_softwall(zone, gfp_mask)) |
9276b1bc | 1417 | goto try_next_zone; |
7fb1d9fc RS |
1418 | |
1419 | if (!(alloc_flags & ALLOC_NO_WATERMARKS)) { | |
3148890b NP |
1420 | unsigned long mark; |
1421 | if (alloc_flags & ALLOC_WMARK_MIN) | |
1192d526 | 1422 | mark = zone->pages_min; |
3148890b | 1423 | else if (alloc_flags & ALLOC_WMARK_LOW) |
1192d526 | 1424 | mark = zone->pages_low; |
3148890b | 1425 | else |
1192d526 | 1426 | mark = zone->pages_high; |
0798e519 PJ |
1427 | if (!zone_watermark_ok(zone, order, mark, |
1428 | classzone_idx, alloc_flags)) { | |
9eeff239 | 1429 | if (!zone_reclaim_mode || |
1192d526 | 1430 | !zone_reclaim(zone, gfp_mask, order)) |
9276b1bc | 1431 | goto this_zone_full; |
0798e519 | 1432 | } |
7fb1d9fc RS |
1433 | } |
1434 | ||
1192d526 | 1435 | page = buffered_rmqueue(zonelist, zone, order, gfp_mask); |
0798e519 | 1436 | if (page) |
7fb1d9fc | 1437 | break; |
9276b1bc PJ |
1438 | this_zone_full: |
1439 | if (NUMA_BUILD) | |
1440 | zlc_mark_zone_full(zonelist, z); | |
1441 | try_next_zone: | |
1442 | if (NUMA_BUILD && !did_zlc_setup) { | |
1443 | /* we do zlc_setup after the first zone is tried */ | |
1444 | allowednodes = zlc_setup(zonelist, alloc_flags); | |
1445 | zlc_active = 1; | |
1446 | did_zlc_setup = 1; | |
1447 | } | |
7fb1d9fc | 1448 | } while (*(++z) != NULL); |
9276b1bc PJ |
1449 | |
1450 | if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) { | |
1451 | /* Disable zlc cache for second zonelist scan */ | |
1452 | zlc_active = 0; | |
1453 | goto zonelist_scan; | |
1454 | } | |
7fb1d9fc | 1455 | return page; |
753ee728 MH |
1456 | } |
1457 | ||
1da177e4 LT |
1458 | /* |
1459 | * This is the 'heart' of the zoned buddy allocator. | |
1460 | */ | |
1461 | struct page * fastcall | |
dd0fc66f | 1462 | __alloc_pages(gfp_t gfp_mask, unsigned int order, |
1da177e4 LT |
1463 | struct zonelist *zonelist) |
1464 | { | |
260b2367 | 1465 | const gfp_t wait = gfp_mask & __GFP_WAIT; |
7fb1d9fc | 1466 | struct zone **z; |
1da177e4 LT |
1467 | struct page *page; |
1468 | struct reclaim_state reclaim_state; | |
1469 | struct task_struct *p = current; | |
1da177e4 | 1470 | int do_retry; |
7fb1d9fc | 1471 | int alloc_flags; |
1da177e4 LT |
1472 | int did_some_progress; |
1473 | ||
1474 | might_sleep_if(wait); | |
1475 | ||
933e312e AM |
1476 | if (should_fail_alloc_page(gfp_mask, order)) |
1477 | return NULL; | |
1478 | ||
6b1de916 | 1479 | restart: |
7fb1d9fc | 1480 | z = zonelist->zones; /* the list of zones suitable for gfp_mask */ |
1da177e4 | 1481 | |
7fb1d9fc | 1482 | if (unlikely(*z == NULL)) { |
523b9458 CL |
1483 | /* |
1484 | * Happens if we have an empty zonelist as a result of | |
1485 | * GFP_THISNODE being used on a memoryless node | |
1486 | */ | |
1da177e4 LT |
1487 | return NULL; |
1488 | } | |
6b1de916 | 1489 | |
7fb1d9fc | 1490 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, |
3148890b | 1491 | zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET); |
7fb1d9fc RS |
1492 | if (page) |
1493 | goto got_pg; | |
1da177e4 | 1494 | |
952f3b51 CL |
1495 | /* |
1496 | * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and | |
1497 | * __GFP_NOWARN set) should not cause reclaim since the subsystem | |
1498 | * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim | |
1499 | * using a larger set of nodes after it has established that the | |
1500 | * allowed per node queues are empty and that nodes are | |
1501 | * over allocated. | |
1502 | */ | |
1503 | if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE) | |
1504 | goto nopage; | |
1505 | ||
0798e519 | 1506 | for (z = zonelist->zones; *z; z++) |
43b0bc00 | 1507 | wakeup_kswapd(*z, order); |
1da177e4 | 1508 | |
9bf2229f | 1509 | /* |
7fb1d9fc RS |
1510 | * OK, we're below the kswapd watermark and have kicked background |
1511 | * reclaim. Now things get more complex, so set up alloc_flags according | |
1512 | * to how we want to proceed. | |
1513 | * | |
1514 | * The caller may dip into page reserves a bit more if the caller | |
1515 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
4eac915d PJ |
1516 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will |
1517 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). | |
9bf2229f | 1518 | */ |
3148890b | 1519 | alloc_flags = ALLOC_WMARK_MIN; |
7fb1d9fc RS |
1520 | if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait) |
1521 | alloc_flags |= ALLOC_HARDER; | |
1522 | if (gfp_mask & __GFP_HIGH) | |
1523 | alloc_flags |= ALLOC_HIGH; | |
bdd804f4 PJ |
1524 | if (wait) |
1525 | alloc_flags |= ALLOC_CPUSET; | |
1da177e4 LT |
1526 | |
1527 | /* | |
1528 | * Go through the zonelist again. Let __GFP_HIGH and allocations | |
7fb1d9fc | 1529 | * coming from realtime tasks go deeper into reserves. |
1da177e4 LT |
1530 | * |
1531 | * This is the last chance, in general, before the goto nopage. | |
1532 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. | |
9bf2229f | 1533 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1da177e4 | 1534 | */ |
7fb1d9fc RS |
1535 | page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags); |
1536 | if (page) | |
1537 | goto got_pg; | |
1da177e4 LT |
1538 | |
1539 | /* This allocation should allow future memory freeing. */ | |
b84a35be | 1540 | |
b43a57bb | 1541 | rebalance: |
b84a35be NP |
1542 | if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE))) |
1543 | && !in_interrupt()) { | |
1544 | if (!(gfp_mask & __GFP_NOMEMALLOC)) { | |
885036d3 | 1545 | nofail_alloc: |
b84a35be | 1546 | /* go through the zonelist yet again, ignoring mins */ |
7fb1d9fc | 1547 | page = get_page_from_freelist(gfp_mask, order, |
47f3a867 | 1548 | zonelist, ALLOC_NO_WATERMARKS); |
7fb1d9fc RS |
1549 | if (page) |
1550 | goto got_pg; | |
885036d3 | 1551 | if (gfp_mask & __GFP_NOFAIL) { |
3fcfab16 | 1552 | congestion_wait(WRITE, HZ/50); |
885036d3 KK |
1553 | goto nofail_alloc; |
1554 | } | |
1da177e4 LT |
1555 | } |
1556 | goto nopage; | |
1557 | } | |
1558 | ||
1559 | /* Atomic allocations - we can't balance anything */ | |
1560 | if (!wait) | |
1561 | goto nopage; | |
1562 | ||
1da177e4 LT |
1563 | cond_resched(); |
1564 | ||
1565 | /* We now go into synchronous reclaim */ | |
3e0d98b9 | 1566 | cpuset_memory_pressure_bump(); |
1da177e4 LT |
1567 | p->flags |= PF_MEMALLOC; |
1568 | reclaim_state.reclaimed_slab = 0; | |
1569 | p->reclaim_state = &reclaim_state; | |
1570 | ||
5ad333eb | 1571 | did_some_progress = try_to_free_pages(zonelist->zones, order, gfp_mask); |
1da177e4 LT |
1572 | |
1573 | p->reclaim_state = NULL; | |
1574 | p->flags &= ~PF_MEMALLOC; | |
1575 | ||
1576 | cond_resched(); | |
1577 | ||
e2c55dc8 MG |
1578 | if (order != 0) |
1579 | drain_all_local_pages(); | |
1580 | ||
1da177e4 | 1581 | if (likely(did_some_progress)) { |
7fb1d9fc RS |
1582 | page = get_page_from_freelist(gfp_mask, order, |
1583 | zonelist, alloc_flags); | |
1584 | if (page) | |
1585 | goto got_pg; | |
1da177e4 LT |
1586 | } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) { |
1587 | /* | |
1588 | * Go through the zonelist yet one more time, keep | |
1589 | * very high watermark here, this is only to catch | |
1590 | * a parallel oom killing, we must fail if we're still | |
1591 | * under heavy pressure. | |
1592 | */ | |
7fb1d9fc | 1593 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, |
3148890b | 1594 | zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET); |
7fb1d9fc RS |
1595 | if (page) |
1596 | goto got_pg; | |
1da177e4 | 1597 | |
a8bbf72a MG |
1598 | /* The OOM killer will not help higher order allocs so fail */ |
1599 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
1600 | goto nopage; | |
1601 | ||
9b0f8b04 | 1602 | out_of_memory(zonelist, gfp_mask, order); |
1da177e4 LT |
1603 | goto restart; |
1604 | } | |
1605 | ||
1606 | /* | |
1607 | * Don't let big-order allocations loop unless the caller explicitly | |
1608 | * requests that. Wait for some write requests to complete then retry. | |
1609 | * | |
1610 | * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order | |
1611 | * <= 3, but that may not be true in other implementations. | |
1612 | */ | |
1613 | do_retry = 0; | |
1614 | if (!(gfp_mask & __GFP_NORETRY)) { | |
5ad333eb AW |
1615 | if ((order <= PAGE_ALLOC_COSTLY_ORDER) || |
1616 | (gfp_mask & __GFP_REPEAT)) | |
1da177e4 LT |
1617 | do_retry = 1; |
1618 | if (gfp_mask & __GFP_NOFAIL) | |
1619 | do_retry = 1; | |
1620 | } | |
1621 | if (do_retry) { | |
3fcfab16 | 1622 | congestion_wait(WRITE, HZ/50); |
1da177e4 LT |
1623 | goto rebalance; |
1624 | } | |
1625 | ||
1626 | nopage: | |
1627 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) { | |
1628 | printk(KERN_WARNING "%s: page allocation failure." | |
1629 | " order:%d, mode:0x%x\n", | |
1630 | p->comm, order, gfp_mask); | |
1631 | dump_stack(); | |
578c2fd6 | 1632 | show_mem(); |
1da177e4 | 1633 | } |
1da177e4 | 1634 | got_pg: |
1da177e4 LT |
1635 | return page; |
1636 | } | |
1637 | ||
1638 | EXPORT_SYMBOL(__alloc_pages); | |
1639 | ||
1640 | /* | |
1641 | * Common helper functions. | |
1642 | */ | |
dd0fc66f | 1643 | fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 LT |
1644 | { |
1645 | struct page * page; | |
1646 | page = alloc_pages(gfp_mask, order); | |
1647 | if (!page) | |
1648 | return 0; | |
1649 | return (unsigned long) page_address(page); | |
1650 | } | |
1651 | ||
1652 | EXPORT_SYMBOL(__get_free_pages); | |
1653 | ||
dd0fc66f | 1654 | fastcall unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 LT |
1655 | { |
1656 | struct page * page; | |
1657 | ||
1658 | /* | |
1659 | * get_zeroed_page() returns a 32-bit address, which cannot represent | |
1660 | * a highmem page | |
1661 | */ | |
725d704e | 1662 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); |
1da177e4 LT |
1663 | |
1664 | page = alloc_pages(gfp_mask | __GFP_ZERO, 0); | |
1665 | if (page) | |
1666 | return (unsigned long) page_address(page); | |
1667 | return 0; | |
1668 | } | |
1669 | ||
1670 | EXPORT_SYMBOL(get_zeroed_page); | |
1671 | ||
1672 | void __pagevec_free(struct pagevec *pvec) | |
1673 | { | |
1674 | int i = pagevec_count(pvec); | |
1675 | ||
1676 | while (--i >= 0) | |
1677 | free_hot_cold_page(pvec->pages[i], pvec->cold); | |
1678 | } | |
1679 | ||
1680 | fastcall void __free_pages(struct page *page, unsigned int order) | |
1681 | { | |
b5810039 | 1682 | if (put_page_testzero(page)) { |
1da177e4 LT |
1683 | if (order == 0) |
1684 | free_hot_page(page); | |
1685 | else | |
1686 | __free_pages_ok(page, order); | |
1687 | } | |
1688 | } | |
1689 | ||
1690 | EXPORT_SYMBOL(__free_pages); | |
1691 | ||
1692 | fastcall void free_pages(unsigned long addr, unsigned int order) | |
1693 | { | |
1694 | if (addr != 0) { | |
725d704e | 1695 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
1696 | __free_pages(virt_to_page((void *)addr), order); |
1697 | } | |
1698 | } | |
1699 | ||
1700 | EXPORT_SYMBOL(free_pages); | |
1701 | ||
1da177e4 LT |
1702 | static unsigned int nr_free_zone_pages(int offset) |
1703 | { | |
e310fd43 MB |
1704 | /* Just pick one node, since fallback list is circular */ |
1705 | pg_data_t *pgdat = NODE_DATA(numa_node_id()); | |
1da177e4 LT |
1706 | unsigned int sum = 0; |
1707 | ||
e310fd43 MB |
1708 | struct zonelist *zonelist = pgdat->node_zonelists + offset; |
1709 | struct zone **zonep = zonelist->zones; | |
1710 | struct zone *zone; | |
1da177e4 | 1711 | |
e310fd43 MB |
1712 | for (zone = *zonep++; zone; zone = *zonep++) { |
1713 | unsigned long size = zone->present_pages; | |
1714 | unsigned long high = zone->pages_high; | |
1715 | if (size > high) | |
1716 | sum += size - high; | |
1da177e4 LT |
1717 | } |
1718 | ||
1719 | return sum; | |
1720 | } | |
1721 | ||
1722 | /* | |
1723 | * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL | |
1724 | */ | |
1725 | unsigned int nr_free_buffer_pages(void) | |
1726 | { | |
af4ca457 | 1727 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 1728 | } |
c2f1a551 | 1729 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 LT |
1730 | |
1731 | /* | |
1732 | * Amount of free RAM allocatable within all zones | |
1733 | */ | |
1734 | unsigned int nr_free_pagecache_pages(void) | |
1735 | { | |
2a1e274a | 1736 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 1737 | } |
08e0f6a9 CL |
1738 | |
1739 | static inline void show_node(struct zone *zone) | |
1da177e4 | 1740 | { |
08e0f6a9 | 1741 | if (NUMA_BUILD) |
25ba77c1 | 1742 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 1743 | } |
1da177e4 | 1744 | |
1da177e4 LT |
1745 | void si_meminfo(struct sysinfo *val) |
1746 | { | |
1747 | val->totalram = totalram_pages; | |
1748 | val->sharedram = 0; | |
d23ad423 | 1749 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 1750 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
1751 | val->totalhigh = totalhigh_pages; |
1752 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
1753 | val->mem_unit = PAGE_SIZE; |
1754 | } | |
1755 | ||
1756 | EXPORT_SYMBOL(si_meminfo); | |
1757 | ||
1758 | #ifdef CONFIG_NUMA | |
1759 | void si_meminfo_node(struct sysinfo *val, int nid) | |
1760 | { | |
1761 | pg_data_t *pgdat = NODE_DATA(nid); | |
1762 | ||
1763 | val->totalram = pgdat->node_present_pages; | |
d23ad423 | 1764 | val->freeram = node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 1765 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 1766 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
d23ad423 CL |
1767 | val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM], |
1768 | NR_FREE_PAGES); | |
98d2b0eb CL |
1769 | #else |
1770 | val->totalhigh = 0; | |
1771 | val->freehigh = 0; | |
1772 | #endif | |
1da177e4 LT |
1773 | val->mem_unit = PAGE_SIZE; |
1774 | } | |
1775 | #endif | |
1776 | ||
1777 | #define K(x) ((x) << (PAGE_SHIFT-10)) | |
1778 | ||
1779 | /* | |
1780 | * Show free area list (used inside shift_scroll-lock stuff) | |
1781 | * We also calculate the percentage fragmentation. We do this by counting the | |
1782 | * memory on each free list with the exception of the first item on the list. | |
1783 | */ | |
1784 | void show_free_areas(void) | |
1785 | { | |
c7241913 | 1786 | int cpu; |
1da177e4 LT |
1787 | struct zone *zone; |
1788 | ||
1789 | for_each_zone(zone) { | |
c7241913 | 1790 | if (!populated_zone(zone)) |
1da177e4 | 1791 | continue; |
c7241913 JS |
1792 | |
1793 | show_node(zone); | |
1794 | printk("%s per-cpu:\n", zone->name); | |
1da177e4 | 1795 | |
6b482c67 | 1796 | for_each_online_cpu(cpu) { |
1da177e4 LT |
1797 | struct per_cpu_pageset *pageset; |
1798 | ||
e7c8d5c9 | 1799 | pageset = zone_pcp(zone, cpu); |
1da177e4 | 1800 | |
c7241913 JS |
1801 | printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d " |
1802 | "Cold: hi:%5d, btch:%4d usd:%4d\n", | |
1803 | cpu, pageset->pcp[0].high, | |
1804 | pageset->pcp[0].batch, pageset->pcp[0].count, | |
1805 | pageset->pcp[1].high, pageset->pcp[1].batch, | |
1806 | pageset->pcp[1].count); | |
1da177e4 LT |
1807 | } |
1808 | } | |
1809 | ||
a25700a5 | 1810 | printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n" |
d23ad423 | 1811 | " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n", |
65e458d4 CL |
1812 | global_page_state(NR_ACTIVE), |
1813 | global_page_state(NR_INACTIVE), | |
b1e7a8fd | 1814 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 1815 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 1816 | global_page_state(NR_UNSTABLE_NFS), |
d23ad423 | 1817 | global_page_state(NR_FREE_PAGES), |
972d1a7b CL |
1818 | global_page_state(NR_SLAB_RECLAIMABLE) + |
1819 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 1820 | global_page_state(NR_FILE_MAPPED), |
a25700a5 AM |
1821 | global_page_state(NR_PAGETABLE), |
1822 | global_page_state(NR_BOUNCE)); | |
1da177e4 LT |
1823 | |
1824 | for_each_zone(zone) { | |
1825 | int i; | |
1826 | ||
c7241913 JS |
1827 | if (!populated_zone(zone)) |
1828 | continue; | |
1829 | ||
1da177e4 LT |
1830 | show_node(zone); |
1831 | printk("%s" | |
1832 | " free:%lukB" | |
1833 | " min:%lukB" | |
1834 | " low:%lukB" | |
1835 | " high:%lukB" | |
1836 | " active:%lukB" | |
1837 | " inactive:%lukB" | |
1838 | " present:%lukB" | |
1839 | " pages_scanned:%lu" | |
1840 | " all_unreclaimable? %s" | |
1841 | "\n", | |
1842 | zone->name, | |
d23ad423 | 1843 | K(zone_page_state(zone, NR_FREE_PAGES)), |
1da177e4 LT |
1844 | K(zone->pages_min), |
1845 | K(zone->pages_low), | |
1846 | K(zone->pages_high), | |
c8785385 CL |
1847 | K(zone_page_state(zone, NR_ACTIVE)), |
1848 | K(zone_page_state(zone, NR_INACTIVE)), | |
1da177e4 LT |
1849 | K(zone->present_pages), |
1850 | zone->pages_scanned, | |
1851 | (zone->all_unreclaimable ? "yes" : "no") | |
1852 | ); | |
1853 | printk("lowmem_reserve[]:"); | |
1854 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1855 | printk(" %lu", zone->lowmem_reserve[i]); | |
1856 | printk("\n"); | |
1857 | } | |
1858 | ||
1859 | for_each_zone(zone) { | |
8f9de51a | 1860 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
1da177e4 | 1861 | |
c7241913 JS |
1862 | if (!populated_zone(zone)) |
1863 | continue; | |
1864 | ||
1da177e4 LT |
1865 | show_node(zone); |
1866 | printk("%s: ", zone->name); | |
1da177e4 LT |
1867 | |
1868 | spin_lock_irqsave(&zone->lock, flags); | |
1869 | for (order = 0; order < MAX_ORDER; order++) { | |
8f9de51a KK |
1870 | nr[order] = zone->free_area[order].nr_free; |
1871 | total += nr[order] << order; | |
1da177e4 LT |
1872 | } |
1873 | spin_unlock_irqrestore(&zone->lock, flags); | |
8f9de51a KK |
1874 | for (order = 0; order < MAX_ORDER; order++) |
1875 | printk("%lu*%lukB ", nr[order], K(1UL) << order); | |
1da177e4 LT |
1876 | printk("= %lukB\n", K(total)); |
1877 | } | |
1878 | ||
1879 | show_swap_cache_info(); | |
1880 | } | |
1881 | ||
1882 | /* | |
1883 | * Builds allocation fallback zone lists. | |
1a93205b CL |
1884 | * |
1885 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 1886 | */ |
f0c0b2b8 KH |
1887 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
1888 | int nr_zones, enum zone_type zone_type) | |
1da177e4 | 1889 | { |
1a93205b CL |
1890 | struct zone *zone; |
1891 | ||
98d2b0eb | 1892 | BUG_ON(zone_type >= MAX_NR_ZONES); |
2f6726e5 | 1893 | zone_type++; |
02a68a5e CL |
1894 | |
1895 | do { | |
2f6726e5 | 1896 | zone_type--; |
070f8032 | 1897 | zone = pgdat->node_zones + zone_type; |
1a93205b | 1898 | if (populated_zone(zone)) { |
070f8032 CL |
1899 | zonelist->zones[nr_zones++] = zone; |
1900 | check_highest_zone(zone_type); | |
1da177e4 | 1901 | } |
02a68a5e | 1902 | |
2f6726e5 | 1903 | } while (zone_type); |
070f8032 | 1904 | return nr_zones; |
1da177e4 LT |
1905 | } |
1906 | ||
f0c0b2b8 KH |
1907 | |
1908 | /* | |
1909 | * zonelist_order: | |
1910 | * 0 = automatic detection of better ordering. | |
1911 | * 1 = order by ([node] distance, -zonetype) | |
1912 | * 2 = order by (-zonetype, [node] distance) | |
1913 | * | |
1914 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
1915 | * the same zonelist. So only NUMA can configure this param. | |
1916 | */ | |
1917 | #define ZONELIST_ORDER_DEFAULT 0 | |
1918 | #define ZONELIST_ORDER_NODE 1 | |
1919 | #define ZONELIST_ORDER_ZONE 2 | |
1920 | ||
1921 | /* zonelist order in the kernel. | |
1922 | * set_zonelist_order() will set this to NODE or ZONE. | |
1923 | */ | |
1924 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
1925 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
1926 | ||
1927 | ||
1da177e4 | 1928 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
1929 | /* The value user specified ....changed by config */ |
1930 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
1931 | /* string for sysctl */ | |
1932 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
1933 | char numa_zonelist_order[16] = "default"; | |
1934 | ||
1935 | /* | |
1936 | * interface for configure zonelist ordering. | |
1937 | * command line option "numa_zonelist_order" | |
1938 | * = "[dD]efault - default, automatic configuration. | |
1939 | * = "[nN]ode - order by node locality, then by zone within node | |
1940 | * = "[zZ]one - order by zone, then by locality within zone | |
1941 | */ | |
1942 | ||
1943 | static int __parse_numa_zonelist_order(char *s) | |
1944 | { | |
1945 | if (*s == 'd' || *s == 'D') { | |
1946 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
1947 | } else if (*s == 'n' || *s == 'N') { | |
1948 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
1949 | } else if (*s == 'z' || *s == 'Z') { | |
1950 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
1951 | } else { | |
1952 | printk(KERN_WARNING | |
1953 | "Ignoring invalid numa_zonelist_order value: " | |
1954 | "%s\n", s); | |
1955 | return -EINVAL; | |
1956 | } | |
1957 | return 0; | |
1958 | } | |
1959 | ||
1960 | static __init int setup_numa_zonelist_order(char *s) | |
1961 | { | |
1962 | if (s) | |
1963 | return __parse_numa_zonelist_order(s); | |
1964 | return 0; | |
1965 | } | |
1966 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
1967 | ||
1968 | /* | |
1969 | * sysctl handler for numa_zonelist_order | |
1970 | */ | |
1971 | int numa_zonelist_order_handler(ctl_table *table, int write, | |
1972 | struct file *file, void __user *buffer, size_t *length, | |
1973 | loff_t *ppos) | |
1974 | { | |
1975 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
1976 | int ret; | |
1977 | ||
1978 | if (write) | |
1979 | strncpy(saved_string, (char*)table->data, | |
1980 | NUMA_ZONELIST_ORDER_LEN); | |
1981 | ret = proc_dostring(table, write, file, buffer, length, ppos); | |
1982 | if (ret) | |
1983 | return ret; | |
1984 | if (write) { | |
1985 | int oldval = user_zonelist_order; | |
1986 | if (__parse_numa_zonelist_order((char*)table->data)) { | |
1987 | /* | |
1988 | * bogus value. restore saved string | |
1989 | */ | |
1990 | strncpy((char*)table->data, saved_string, | |
1991 | NUMA_ZONELIST_ORDER_LEN); | |
1992 | user_zonelist_order = oldval; | |
1993 | } else if (oldval != user_zonelist_order) | |
1994 | build_all_zonelists(); | |
1995 | } | |
1996 | return 0; | |
1997 | } | |
1998 | ||
1999 | ||
1da177e4 | 2000 | #define MAX_NODE_LOAD (num_online_nodes()) |
f0c0b2b8 KH |
2001 | static int node_load[MAX_NUMNODES]; |
2002 | ||
1da177e4 | 2003 | /** |
4dc3b16b | 2004 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
2005 | * @node: node whose fallback list we're appending |
2006 | * @used_node_mask: nodemask_t of already used nodes | |
2007 | * | |
2008 | * We use a number of factors to determine which is the next node that should | |
2009 | * appear on a given node's fallback list. The node should not have appeared | |
2010 | * already in @node's fallback list, and it should be the next closest node | |
2011 | * according to the distance array (which contains arbitrary distance values | |
2012 | * from each node to each node in the system), and should also prefer nodes | |
2013 | * with no CPUs, since presumably they'll have very little allocation pressure | |
2014 | * on them otherwise. | |
2015 | * It returns -1 if no node is found. | |
2016 | */ | |
f0c0b2b8 | 2017 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 2018 | { |
4cf808eb | 2019 | int n, val; |
1da177e4 LT |
2020 | int min_val = INT_MAX; |
2021 | int best_node = -1; | |
2022 | ||
4cf808eb LT |
2023 | /* Use the local node if we haven't already */ |
2024 | if (!node_isset(node, *used_node_mask)) { | |
2025 | node_set(node, *used_node_mask); | |
2026 | return node; | |
2027 | } | |
1da177e4 | 2028 | |
37b07e41 | 2029 | for_each_node_state(n, N_HIGH_MEMORY) { |
4cf808eb | 2030 | cpumask_t tmp; |
1da177e4 LT |
2031 | |
2032 | /* Don't want a node to appear more than once */ | |
2033 | if (node_isset(n, *used_node_mask)) | |
2034 | continue; | |
2035 | ||
1da177e4 LT |
2036 | /* Use the distance array to find the distance */ |
2037 | val = node_distance(node, n); | |
2038 | ||
4cf808eb LT |
2039 | /* Penalize nodes under us ("prefer the next node") */ |
2040 | val += (n < node); | |
2041 | ||
1da177e4 LT |
2042 | /* Give preference to headless and unused nodes */ |
2043 | tmp = node_to_cpumask(n); | |
2044 | if (!cpus_empty(tmp)) | |
2045 | val += PENALTY_FOR_NODE_WITH_CPUS; | |
2046 | ||
2047 | /* Slight preference for less loaded node */ | |
2048 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
2049 | val += node_load[n]; | |
2050 | ||
2051 | if (val < min_val) { | |
2052 | min_val = val; | |
2053 | best_node = n; | |
2054 | } | |
2055 | } | |
2056 | ||
2057 | if (best_node >= 0) | |
2058 | node_set(best_node, *used_node_mask); | |
2059 | ||
2060 | return best_node; | |
2061 | } | |
2062 | ||
f0c0b2b8 KH |
2063 | |
2064 | /* | |
2065 | * Build zonelists ordered by node and zones within node. | |
2066 | * This results in maximum locality--normal zone overflows into local | |
2067 | * DMA zone, if any--but risks exhausting DMA zone. | |
2068 | */ | |
2069 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 2070 | { |
19655d34 | 2071 | enum zone_type i; |
f0c0b2b8 | 2072 | int j; |
1da177e4 | 2073 | struct zonelist *zonelist; |
f0c0b2b8 KH |
2074 | |
2075 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2076 | zonelist = pgdat->node_zonelists + i; | |
2077 | for (j = 0; zonelist->zones[j] != NULL; j++) | |
2078 | ; | |
2079 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); | |
2080 | zonelist->zones[j] = NULL; | |
2081 | } | |
2082 | } | |
2083 | ||
523b9458 CL |
2084 | /* |
2085 | * Build gfp_thisnode zonelists | |
2086 | */ | |
2087 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
2088 | { | |
2089 | enum zone_type i; | |
2090 | int j; | |
2091 | struct zonelist *zonelist; | |
2092 | ||
2093 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2094 | zonelist = pgdat->node_zonelists + MAX_NR_ZONES + i; | |
2095 | j = build_zonelists_node(pgdat, zonelist, 0, i); | |
2096 | zonelist->zones[j] = NULL; | |
2097 | } | |
2098 | } | |
2099 | ||
f0c0b2b8 KH |
2100 | /* |
2101 | * Build zonelists ordered by zone and nodes within zones. | |
2102 | * This results in conserving DMA zone[s] until all Normal memory is | |
2103 | * exhausted, but results in overflowing to remote node while memory | |
2104 | * may still exist in local DMA zone. | |
2105 | */ | |
2106 | static int node_order[MAX_NUMNODES]; | |
2107 | ||
2108 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
2109 | { | |
2110 | enum zone_type i; | |
2111 | int pos, j, node; | |
2112 | int zone_type; /* needs to be signed */ | |
2113 | struct zone *z; | |
2114 | struct zonelist *zonelist; | |
2115 | ||
2116 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2117 | zonelist = pgdat->node_zonelists + i; | |
2118 | pos = 0; | |
2119 | for (zone_type = i; zone_type >= 0; zone_type--) { | |
2120 | for (j = 0; j < nr_nodes; j++) { | |
2121 | node = node_order[j]; | |
2122 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
2123 | if (populated_zone(z)) { | |
2124 | zonelist->zones[pos++] = z; | |
2125 | check_highest_zone(zone_type); | |
2126 | } | |
2127 | } | |
2128 | } | |
2129 | zonelist->zones[pos] = NULL; | |
2130 | } | |
2131 | } | |
2132 | ||
2133 | static int default_zonelist_order(void) | |
2134 | { | |
2135 | int nid, zone_type; | |
2136 | unsigned long low_kmem_size,total_size; | |
2137 | struct zone *z; | |
2138 | int average_size; | |
2139 | /* | |
2140 | * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem. | |
2141 | * If they are really small and used heavily, the system can fall | |
2142 | * into OOM very easily. | |
2143 | * This function detect ZONE_DMA/DMA32 size and confgigures zone order. | |
2144 | */ | |
2145 | /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */ | |
2146 | low_kmem_size = 0; | |
2147 | total_size = 0; | |
2148 | for_each_online_node(nid) { | |
2149 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
2150 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
2151 | if (populated_zone(z)) { | |
2152 | if (zone_type < ZONE_NORMAL) | |
2153 | low_kmem_size += z->present_pages; | |
2154 | total_size += z->present_pages; | |
2155 | } | |
2156 | } | |
2157 | } | |
2158 | if (!low_kmem_size || /* there are no DMA area. */ | |
2159 | low_kmem_size > total_size/2) /* DMA/DMA32 is big. */ | |
2160 | return ZONELIST_ORDER_NODE; | |
2161 | /* | |
2162 | * look into each node's config. | |
2163 | * If there is a node whose DMA/DMA32 memory is very big area on | |
2164 | * local memory, NODE_ORDER may be suitable. | |
2165 | */ | |
37b07e41 LS |
2166 | average_size = total_size / |
2167 | (nodes_weight(node_states[N_HIGH_MEMORY]) + 1); | |
f0c0b2b8 KH |
2168 | for_each_online_node(nid) { |
2169 | low_kmem_size = 0; | |
2170 | total_size = 0; | |
2171 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
2172 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
2173 | if (populated_zone(z)) { | |
2174 | if (zone_type < ZONE_NORMAL) | |
2175 | low_kmem_size += z->present_pages; | |
2176 | total_size += z->present_pages; | |
2177 | } | |
2178 | } | |
2179 | if (low_kmem_size && | |
2180 | total_size > average_size && /* ignore small node */ | |
2181 | low_kmem_size > total_size * 70/100) | |
2182 | return ZONELIST_ORDER_NODE; | |
2183 | } | |
2184 | return ZONELIST_ORDER_ZONE; | |
2185 | } | |
2186 | ||
2187 | static void set_zonelist_order(void) | |
2188 | { | |
2189 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
2190 | current_zonelist_order = default_zonelist_order(); | |
2191 | else | |
2192 | current_zonelist_order = user_zonelist_order; | |
2193 | } | |
2194 | ||
2195 | static void build_zonelists(pg_data_t *pgdat) | |
2196 | { | |
2197 | int j, node, load; | |
2198 | enum zone_type i; | |
1da177e4 | 2199 | nodemask_t used_mask; |
f0c0b2b8 KH |
2200 | int local_node, prev_node; |
2201 | struct zonelist *zonelist; | |
2202 | int order = current_zonelist_order; | |
1da177e4 LT |
2203 | |
2204 | /* initialize zonelists */ | |
523b9458 | 2205 | for (i = 0; i < MAX_ZONELISTS; i++) { |
1da177e4 LT |
2206 | zonelist = pgdat->node_zonelists + i; |
2207 | zonelist->zones[0] = NULL; | |
2208 | } | |
2209 | ||
2210 | /* NUMA-aware ordering of nodes */ | |
2211 | local_node = pgdat->node_id; | |
2212 | load = num_online_nodes(); | |
2213 | prev_node = local_node; | |
2214 | nodes_clear(used_mask); | |
f0c0b2b8 KH |
2215 | |
2216 | memset(node_load, 0, sizeof(node_load)); | |
2217 | memset(node_order, 0, sizeof(node_order)); | |
2218 | j = 0; | |
2219 | ||
1da177e4 | 2220 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
9eeff239 CL |
2221 | int distance = node_distance(local_node, node); |
2222 | ||
2223 | /* | |
2224 | * If another node is sufficiently far away then it is better | |
2225 | * to reclaim pages in a zone before going off node. | |
2226 | */ | |
2227 | if (distance > RECLAIM_DISTANCE) | |
2228 | zone_reclaim_mode = 1; | |
2229 | ||
1da177e4 LT |
2230 | /* |
2231 | * We don't want to pressure a particular node. | |
2232 | * So adding penalty to the first node in same | |
2233 | * distance group to make it round-robin. | |
2234 | */ | |
9eeff239 | 2235 | if (distance != node_distance(local_node, prev_node)) |
f0c0b2b8 KH |
2236 | node_load[node] = load; |
2237 | ||
1da177e4 LT |
2238 | prev_node = node; |
2239 | load--; | |
f0c0b2b8 KH |
2240 | if (order == ZONELIST_ORDER_NODE) |
2241 | build_zonelists_in_node_order(pgdat, node); | |
2242 | else | |
2243 | node_order[j++] = node; /* remember order */ | |
2244 | } | |
1da177e4 | 2245 | |
f0c0b2b8 KH |
2246 | if (order == ZONELIST_ORDER_ZONE) { |
2247 | /* calculate node order -- i.e., DMA last! */ | |
2248 | build_zonelists_in_zone_order(pgdat, j); | |
1da177e4 | 2249 | } |
523b9458 CL |
2250 | |
2251 | build_thisnode_zonelists(pgdat); | |
1da177e4 LT |
2252 | } |
2253 | ||
9276b1bc | 2254 | /* Construct the zonelist performance cache - see further mmzone.h */ |
f0c0b2b8 | 2255 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc PJ |
2256 | { |
2257 | int i; | |
2258 | ||
2259 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2260 | struct zonelist *zonelist; | |
2261 | struct zonelist_cache *zlc; | |
2262 | struct zone **z; | |
2263 | ||
2264 | zonelist = pgdat->node_zonelists + i; | |
2265 | zonelist->zlcache_ptr = zlc = &zonelist->zlcache; | |
2266 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
2267 | for (z = zonelist->zones; *z; z++) | |
2268 | zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z); | |
2269 | } | |
2270 | } | |
2271 | ||
f0c0b2b8 | 2272 | |
1da177e4 LT |
2273 | #else /* CONFIG_NUMA */ |
2274 | ||
f0c0b2b8 KH |
2275 | static void set_zonelist_order(void) |
2276 | { | |
2277 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
2278 | } | |
2279 | ||
2280 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 2281 | { |
19655d34 CL |
2282 | int node, local_node; |
2283 | enum zone_type i,j; | |
1da177e4 LT |
2284 | |
2285 | local_node = pgdat->node_id; | |
19655d34 | 2286 | for (i = 0; i < MAX_NR_ZONES; i++) { |
1da177e4 LT |
2287 | struct zonelist *zonelist; |
2288 | ||
2289 | zonelist = pgdat->node_zonelists + i; | |
2290 | ||
19655d34 | 2291 | j = build_zonelists_node(pgdat, zonelist, 0, i); |
1da177e4 LT |
2292 | /* |
2293 | * Now we build the zonelist so that it contains the zones | |
2294 | * of all the other nodes. | |
2295 | * We don't want to pressure a particular node, so when | |
2296 | * building the zones for node N, we make sure that the | |
2297 | * zones coming right after the local ones are those from | |
2298 | * node N+1 (modulo N) | |
2299 | */ | |
2300 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
2301 | if (!node_online(node)) | |
2302 | continue; | |
19655d34 | 2303 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); |
1da177e4 LT |
2304 | } |
2305 | for (node = 0; node < local_node; node++) { | |
2306 | if (!node_online(node)) | |
2307 | continue; | |
19655d34 | 2308 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); |
1da177e4 LT |
2309 | } |
2310 | ||
2311 | zonelist->zones[j] = NULL; | |
2312 | } | |
2313 | } | |
2314 | ||
9276b1bc | 2315 | /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */ |
f0c0b2b8 | 2316 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc PJ |
2317 | { |
2318 | int i; | |
2319 | ||
2320 | for (i = 0; i < MAX_NR_ZONES; i++) | |
2321 | pgdat->node_zonelists[i].zlcache_ptr = NULL; | |
2322 | } | |
2323 | ||
1da177e4 LT |
2324 | #endif /* CONFIG_NUMA */ |
2325 | ||
6811378e | 2326 | /* return values int ....just for stop_machine_run() */ |
f0c0b2b8 | 2327 | static int __build_all_zonelists(void *dummy) |
1da177e4 | 2328 | { |
6811378e | 2329 | int nid; |
9276b1bc PJ |
2330 | |
2331 | for_each_online_node(nid) { | |
7ea1530a CL |
2332 | pg_data_t *pgdat = NODE_DATA(nid); |
2333 | ||
2334 | build_zonelists(pgdat); | |
2335 | build_zonelist_cache(pgdat); | |
9276b1bc | 2336 | } |
6811378e YG |
2337 | return 0; |
2338 | } | |
2339 | ||
f0c0b2b8 | 2340 | void build_all_zonelists(void) |
6811378e | 2341 | { |
f0c0b2b8 KH |
2342 | set_zonelist_order(); |
2343 | ||
6811378e | 2344 | if (system_state == SYSTEM_BOOTING) { |
423b41d7 | 2345 | __build_all_zonelists(NULL); |
6811378e YG |
2346 | cpuset_init_current_mems_allowed(); |
2347 | } else { | |
2348 | /* we have to stop all cpus to guaranntee there is no user | |
2349 | of zonelist */ | |
2350 | stop_machine_run(__build_all_zonelists, NULL, NR_CPUS); | |
2351 | /* cpuset refresh routine should be here */ | |
2352 | } | |
bd1e22b8 | 2353 | vm_total_pages = nr_free_pagecache_pages(); |
f0c0b2b8 KH |
2354 | printk("Built %i zonelists in %s order. Total pages: %ld\n", |
2355 | num_online_nodes(), | |
2356 | zonelist_order_name[current_zonelist_order], | |
2357 | vm_total_pages); | |
2358 | #ifdef CONFIG_NUMA | |
2359 | printk("Policy zone: %s\n", zone_names[policy_zone]); | |
2360 | #endif | |
1da177e4 LT |
2361 | } |
2362 | ||
2363 | /* | |
2364 | * Helper functions to size the waitqueue hash table. | |
2365 | * Essentially these want to choose hash table sizes sufficiently | |
2366 | * large so that collisions trying to wait on pages are rare. | |
2367 | * But in fact, the number of active page waitqueues on typical | |
2368 | * systems is ridiculously low, less than 200. So this is even | |
2369 | * conservative, even though it seems large. | |
2370 | * | |
2371 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
2372 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
2373 | */ | |
2374 | #define PAGES_PER_WAITQUEUE 256 | |
2375 | ||
cca448fe | 2376 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 2377 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
2378 | { |
2379 | unsigned long size = 1; | |
2380 | ||
2381 | pages /= PAGES_PER_WAITQUEUE; | |
2382 | ||
2383 | while (size < pages) | |
2384 | size <<= 1; | |
2385 | ||
2386 | /* | |
2387 | * Once we have dozens or even hundreds of threads sleeping | |
2388 | * on IO we've got bigger problems than wait queue collision. | |
2389 | * Limit the size of the wait table to a reasonable size. | |
2390 | */ | |
2391 | size = min(size, 4096UL); | |
2392 | ||
2393 | return max(size, 4UL); | |
2394 | } | |
cca448fe YG |
2395 | #else |
2396 | /* | |
2397 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
2398 | * a suitable size for its wait_table. So we use the maximum size now. | |
2399 | * | |
2400 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
2401 | * | |
2402 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
2403 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
2404 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
2405 | * | |
2406 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
2407 | * or more by the traditional way. (See above). It equals: | |
2408 | * | |
2409 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
2410 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
2411 | * powerpc (64K page size) : = (32G +16M)byte. | |
2412 | */ | |
2413 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
2414 | { | |
2415 | return 4096UL; | |
2416 | } | |
2417 | #endif | |
1da177e4 LT |
2418 | |
2419 | /* | |
2420 | * This is an integer logarithm so that shifts can be used later | |
2421 | * to extract the more random high bits from the multiplicative | |
2422 | * hash function before the remainder is taken. | |
2423 | */ | |
2424 | static inline unsigned long wait_table_bits(unsigned long size) | |
2425 | { | |
2426 | return ffz(~size); | |
2427 | } | |
2428 | ||
2429 | #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1)) | |
2430 | ||
1da177e4 LT |
2431 | /* |
2432 | * Initially all pages are reserved - free ones are freed | |
2433 | * up by free_all_bootmem() once the early boot process is | |
2434 | * done. Non-atomic initialization, single-pass. | |
2435 | */ | |
c09b4240 | 2436 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 2437 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 2438 | { |
1da177e4 | 2439 | struct page *page; |
29751f69 AW |
2440 | unsigned long end_pfn = start_pfn + size; |
2441 | unsigned long pfn; | |
1da177e4 | 2442 | |
cbe8dd4a | 2443 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 DH |
2444 | /* |
2445 | * There can be holes in boot-time mem_map[]s | |
2446 | * handed to this function. They do not | |
2447 | * exist on hotplugged memory. | |
2448 | */ | |
2449 | if (context == MEMMAP_EARLY) { | |
2450 | if (!early_pfn_valid(pfn)) | |
2451 | continue; | |
2452 | if (!early_pfn_in_nid(pfn, nid)) | |
2453 | continue; | |
2454 | } | |
d41dee36 AW |
2455 | page = pfn_to_page(pfn); |
2456 | set_page_links(page, zone, nid, pfn); | |
7835e98b | 2457 | init_page_count(page); |
1da177e4 LT |
2458 | reset_page_mapcount(page); |
2459 | SetPageReserved(page); | |
b2a0ac88 MG |
2460 | |
2461 | /* | |
2462 | * Mark the block movable so that blocks are reserved for | |
2463 | * movable at startup. This will force kernel allocations | |
2464 | * to reserve their blocks rather than leaking throughout | |
2465 | * the address space during boot when many long-lived | |
2466 | * kernel allocations are made | |
2467 | */ | |
2468 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
2469 | ||
1da177e4 LT |
2470 | INIT_LIST_HEAD(&page->lru); |
2471 | #ifdef WANT_PAGE_VIRTUAL | |
2472 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
2473 | if (!is_highmem_idx(zone)) | |
3212c6be | 2474 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1da177e4 | 2475 | #endif |
1da177e4 LT |
2476 | } |
2477 | } | |
2478 | ||
6ea6e688 PM |
2479 | static void __meminit zone_init_free_lists(struct pglist_data *pgdat, |
2480 | struct zone *zone, unsigned long size) | |
1da177e4 | 2481 | { |
b2a0ac88 MG |
2482 | int order, t; |
2483 | for_each_migratetype_order(order, t) { | |
2484 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
2485 | zone->free_area[order].nr_free = 0; |
2486 | } | |
2487 | } | |
2488 | ||
2489 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
2490 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 2491 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
2492 | #endif |
2493 | ||
d09c6b80 | 2494 | static int __devinit zone_batchsize(struct zone *zone) |
e7c8d5c9 CL |
2495 | { |
2496 | int batch; | |
2497 | ||
2498 | /* | |
2499 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 2500 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
2501 | * |
2502 | * OK, so we don't know how big the cache is. So guess. | |
2503 | */ | |
2504 | batch = zone->present_pages / 1024; | |
ba56e91c SR |
2505 | if (batch * PAGE_SIZE > 512 * 1024) |
2506 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
2507 | batch /= 4; /* We effectively *= 4 below */ |
2508 | if (batch < 1) | |
2509 | batch = 1; | |
2510 | ||
2511 | /* | |
0ceaacc9 NP |
2512 | * Clamp the batch to a 2^n - 1 value. Having a power |
2513 | * of 2 value was found to be more likely to have | |
2514 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 2515 | * |
0ceaacc9 NP |
2516 | * For example if 2 tasks are alternately allocating |
2517 | * batches of pages, one task can end up with a lot | |
2518 | * of pages of one half of the possible page colors | |
2519 | * and the other with pages of the other colors. | |
e7c8d5c9 | 2520 | */ |
0ceaacc9 | 2521 | batch = (1 << (fls(batch + batch/2)-1)) - 1; |
ba56e91c | 2522 | |
e7c8d5c9 CL |
2523 | return batch; |
2524 | } | |
2525 | ||
2caaad41 CL |
2526 | inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
2527 | { | |
2528 | struct per_cpu_pages *pcp; | |
2529 | ||
1c6fe946 MD |
2530 | memset(p, 0, sizeof(*p)); |
2531 | ||
2caaad41 CL |
2532 | pcp = &p->pcp[0]; /* hot */ |
2533 | pcp->count = 0; | |
2caaad41 CL |
2534 | pcp->high = 6 * batch; |
2535 | pcp->batch = max(1UL, 1 * batch); | |
2536 | INIT_LIST_HEAD(&pcp->list); | |
2537 | ||
2538 | pcp = &p->pcp[1]; /* cold*/ | |
2539 | pcp->count = 0; | |
2caaad41 | 2540 | pcp->high = 2 * batch; |
e46a5e28 | 2541 | pcp->batch = max(1UL, batch/2); |
2caaad41 CL |
2542 | INIT_LIST_HEAD(&pcp->list); |
2543 | } | |
2544 | ||
8ad4b1fb RS |
2545 | /* |
2546 | * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist | |
2547 | * to the value high for the pageset p. | |
2548 | */ | |
2549 | ||
2550 | static void setup_pagelist_highmark(struct per_cpu_pageset *p, | |
2551 | unsigned long high) | |
2552 | { | |
2553 | struct per_cpu_pages *pcp; | |
2554 | ||
2555 | pcp = &p->pcp[0]; /* hot list */ | |
2556 | pcp->high = high; | |
2557 | pcp->batch = max(1UL, high/4); | |
2558 | if ((high/4) > (PAGE_SHIFT * 8)) | |
2559 | pcp->batch = PAGE_SHIFT * 8; | |
2560 | } | |
2561 | ||
2562 | ||
e7c8d5c9 CL |
2563 | #ifdef CONFIG_NUMA |
2564 | /* | |
2caaad41 CL |
2565 | * Boot pageset table. One per cpu which is going to be used for all |
2566 | * zones and all nodes. The parameters will be set in such a way | |
2567 | * that an item put on a list will immediately be handed over to | |
2568 | * the buddy list. This is safe since pageset manipulation is done | |
2569 | * with interrupts disabled. | |
2570 | * | |
2571 | * Some NUMA counter updates may also be caught by the boot pagesets. | |
b7c84c6a CL |
2572 | * |
2573 | * The boot_pagesets must be kept even after bootup is complete for | |
2574 | * unused processors and/or zones. They do play a role for bootstrapping | |
2575 | * hotplugged processors. | |
2576 | * | |
2577 | * zoneinfo_show() and maybe other functions do | |
2578 | * not check if the processor is online before following the pageset pointer. | |
2579 | * Other parts of the kernel may not check if the zone is available. | |
2caaad41 | 2580 | */ |
88a2a4ac | 2581 | static struct per_cpu_pageset boot_pageset[NR_CPUS]; |
2caaad41 CL |
2582 | |
2583 | /* | |
2584 | * Dynamically allocate memory for the | |
e7c8d5c9 CL |
2585 | * per cpu pageset array in struct zone. |
2586 | */ | |
6292d9aa | 2587 | static int __cpuinit process_zones(int cpu) |
e7c8d5c9 CL |
2588 | { |
2589 | struct zone *zone, *dzone; | |
37c0708d CL |
2590 | int node = cpu_to_node(cpu); |
2591 | ||
2592 | node_set_state(node, N_CPU); /* this node has a cpu */ | |
e7c8d5c9 CL |
2593 | |
2594 | for_each_zone(zone) { | |
e7c8d5c9 | 2595 | |
66a55030 CL |
2596 | if (!populated_zone(zone)) |
2597 | continue; | |
2598 | ||
23316bc8 | 2599 | zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset), |
37c0708d | 2600 | GFP_KERNEL, node); |
23316bc8 | 2601 | if (!zone_pcp(zone, cpu)) |
e7c8d5c9 | 2602 | goto bad; |
e7c8d5c9 | 2603 | |
23316bc8 | 2604 | setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone)); |
8ad4b1fb RS |
2605 | |
2606 | if (percpu_pagelist_fraction) | |
2607 | setup_pagelist_highmark(zone_pcp(zone, cpu), | |
2608 | (zone->present_pages / percpu_pagelist_fraction)); | |
e7c8d5c9 CL |
2609 | } |
2610 | ||
2611 | return 0; | |
2612 | bad: | |
2613 | for_each_zone(dzone) { | |
64191688 AM |
2614 | if (!populated_zone(dzone)) |
2615 | continue; | |
e7c8d5c9 CL |
2616 | if (dzone == zone) |
2617 | break; | |
23316bc8 NP |
2618 | kfree(zone_pcp(dzone, cpu)); |
2619 | zone_pcp(dzone, cpu) = NULL; | |
e7c8d5c9 CL |
2620 | } |
2621 | return -ENOMEM; | |
2622 | } | |
2623 | ||
2624 | static inline void free_zone_pagesets(int cpu) | |
2625 | { | |
e7c8d5c9 CL |
2626 | struct zone *zone; |
2627 | ||
2628 | for_each_zone(zone) { | |
2629 | struct per_cpu_pageset *pset = zone_pcp(zone, cpu); | |
2630 | ||
f3ef9ead DR |
2631 | /* Free per_cpu_pageset if it is slab allocated */ |
2632 | if (pset != &boot_pageset[cpu]) | |
2633 | kfree(pset); | |
e7c8d5c9 | 2634 | zone_pcp(zone, cpu) = NULL; |
e7c8d5c9 | 2635 | } |
e7c8d5c9 CL |
2636 | } |
2637 | ||
9c7b216d | 2638 | static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb, |
e7c8d5c9 CL |
2639 | unsigned long action, |
2640 | void *hcpu) | |
2641 | { | |
2642 | int cpu = (long)hcpu; | |
2643 | int ret = NOTIFY_OK; | |
2644 | ||
2645 | switch (action) { | |
ce421c79 | 2646 | case CPU_UP_PREPARE: |
8bb78442 | 2647 | case CPU_UP_PREPARE_FROZEN: |
ce421c79 AW |
2648 | if (process_zones(cpu)) |
2649 | ret = NOTIFY_BAD; | |
2650 | break; | |
2651 | case CPU_UP_CANCELED: | |
8bb78442 | 2652 | case CPU_UP_CANCELED_FROZEN: |
ce421c79 | 2653 | case CPU_DEAD: |
8bb78442 | 2654 | case CPU_DEAD_FROZEN: |
ce421c79 AW |
2655 | free_zone_pagesets(cpu); |
2656 | break; | |
2657 | default: | |
2658 | break; | |
e7c8d5c9 CL |
2659 | } |
2660 | return ret; | |
2661 | } | |
2662 | ||
74b85f37 | 2663 | static struct notifier_block __cpuinitdata pageset_notifier = |
e7c8d5c9 CL |
2664 | { &pageset_cpuup_callback, NULL, 0 }; |
2665 | ||
78d9955b | 2666 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 CL |
2667 | { |
2668 | int err; | |
2669 | ||
2670 | /* Initialize per_cpu_pageset for cpu 0. | |
2671 | * A cpuup callback will do this for every cpu | |
2672 | * as it comes online | |
2673 | */ | |
2674 | err = process_zones(smp_processor_id()); | |
2675 | BUG_ON(err); | |
2676 | register_cpu_notifier(&pageset_notifier); | |
2677 | } | |
2678 | ||
2679 | #endif | |
2680 | ||
577a32f6 | 2681 | static noinline __init_refok |
cca448fe | 2682 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
2683 | { |
2684 | int i; | |
2685 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe | 2686 | size_t alloc_size; |
ed8ece2e DH |
2687 | |
2688 | /* | |
2689 | * The per-page waitqueue mechanism uses hashed waitqueues | |
2690 | * per zone. | |
2691 | */ | |
02b694de YG |
2692 | zone->wait_table_hash_nr_entries = |
2693 | wait_table_hash_nr_entries(zone_size_pages); | |
2694 | zone->wait_table_bits = | |
2695 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
2696 | alloc_size = zone->wait_table_hash_nr_entries |
2697 | * sizeof(wait_queue_head_t); | |
2698 | ||
2699 | if (system_state == SYSTEM_BOOTING) { | |
2700 | zone->wait_table = (wait_queue_head_t *) | |
2701 | alloc_bootmem_node(pgdat, alloc_size); | |
2702 | } else { | |
2703 | /* | |
2704 | * This case means that a zone whose size was 0 gets new memory | |
2705 | * via memory hot-add. | |
2706 | * But it may be the case that a new node was hot-added. In | |
2707 | * this case vmalloc() will not be able to use this new node's | |
2708 | * memory - this wait_table must be initialized to use this new | |
2709 | * node itself as well. | |
2710 | * To use this new node's memory, further consideration will be | |
2711 | * necessary. | |
2712 | */ | |
8691f3a7 | 2713 | zone->wait_table = vmalloc(alloc_size); |
cca448fe YG |
2714 | } |
2715 | if (!zone->wait_table) | |
2716 | return -ENOMEM; | |
ed8ece2e | 2717 | |
02b694de | 2718 | for(i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 2719 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
2720 | |
2721 | return 0; | |
ed8ece2e DH |
2722 | } |
2723 | ||
c09b4240 | 2724 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e DH |
2725 | { |
2726 | int cpu; | |
2727 | unsigned long batch = zone_batchsize(zone); | |
2728 | ||
2729 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
2730 | #ifdef CONFIG_NUMA | |
2731 | /* Early boot. Slab allocator not functional yet */ | |
23316bc8 | 2732 | zone_pcp(zone, cpu) = &boot_pageset[cpu]; |
ed8ece2e DH |
2733 | setup_pageset(&boot_pageset[cpu],0); |
2734 | #else | |
2735 | setup_pageset(zone_pcp(zone,cpu), batch); | |
2736 | #endif | |
2737 | } | |
f5335c0f AB |
2738 | if (zone->present_pages) |
2739 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n", | |
2740 | zone->name, zone->present_pages, batch); | |
ed8ece2e DH |
2741 | } |
2742 | ||
718127cc YG |
2743 | __meminit int init_currently_empty_zone(struct zone *zone, |
2744 | unsigned long zone_start_pfn, | |
a2f3aa02 DH |
2745 | unsigned long size, |
2746 | enum memmap_context context) | |
ed8ece2e DH |
2747 | { |
2748 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
2749 | int ret; |
2750 | ret = zone_wait_table_init(zone, size); | |
2751 | if (ret) | |
2752 | return ret; | |
ed8ece2e DH |
2753 | pgdat->nr_zones = zone_idx(zone) + 1; |
2754 | ||
ed8ece2e DH |
2755 | zone->zone_start_pfn = zone_start_pfn; |
2756 | ||
2757 | memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn); | |
2758 | ||
2759 | zone_init_free_lists(pgdat, zone, zone->spanned_pages); | |
718127cc YG |
2760 | |
2761 | return 0; | |
ed8ece2e DH |
2762 | } |
2763 | ||
c713216d MG |
2764 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
2765 | /* | |
2766 | * Basic iterator support. Return the first range of PFNs for a node | |
2767 | * Note: nid == MAX_NUMNODES returns first region regardless of node | |
2768 | */ | |
a3142c8e | 2769 | static int __meminit first_active_region_index_in_nid(int nid) |
c713216d MG |
2770 | { |
2771 | int i; | |
2772 | ||
2773 | for (i = 0; i < nr_nodemap_entries; i++) | |
2774 | if (nid == MAX_NUMNODES || early_node_map[i].nid == nid) | |
2775 | return i; | |
2776 | ||
2777 | return -1; | |
2778 | } | |
2779 | ||
2780 | /* | |
2781 | * Basic iterator support. Return the next active range of PFNs for a node | |
2782 | * Note: nid == MAX_NUMNODES returns next region regardles of node | |
2783 | */ | |
a3142c8e | 2784 | static int __meminit next_active_region_index_in_nid(int index, int nid) |
c713216d MG |
2785 | { |
2786 | for (index = index + 1; index < nr_nodemap_entries; index++) | |
2787 | if (nid == MAX_NUMNODES || early_node_map[index].nid == nid) | |
2788 | return index; | |
2789 | ||
2790 | return -1; | |
2791 | } | |
2792 | ||
2793 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID | |
2794 | /* | |
2795 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
2796 | * Architectures may implement their own version but if add_active_range() | |
2797 | * was used and there are no special requirements, this is a convenient | |
2798 | * alternative | |
2799 | */ | |
6f076f5d | 2800 | int __meminit early_pfn_to_nid(unsigned long pfn) |
c713216d MG |
2801 | { |
2802 | int i; | |
2803 | ||
2804 | for (i = 0; i < nr_nodemap_entries; i++) { | |
2805 | unsigned long start_pfn = early_node_map[i].start_pfn; | |
2806 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
2807 | ||
2808 | if (start_pfn <= pfn && pfn < end_pfn) | |
2809 | return early_node_map[i].nid; | |
2810 | } | |
2811 | ||
2812 | return 0; | |
2813 | } | |
2814 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
2815 | ||
2816 | /* Basic iterator support to walk early_node_map[] */ | |
2817 | #define for_each_active_range_index_in_nid(i, nid) \ | |
2818 | for (i = first_active_region_index_in_nid(nid); i != -1; \ | |
2819 | i = next_active_region_index_in_nid(i, nid)) | |
2820 | ||
2821 | /** | |
2822 | * free_bootmem_with_active_regions - Call free_bootmem_node for each active range | |
88ca3b94 RD |
2823 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
2824 | * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node | |
c713216d MG |
2825 | * |
2826 | * If an architecture guarantees that all ranges registered with | |
2827 | * add_active_ranges() contain no holes and may be freed, this | |
2828 | * this function may be used instead of calling free_bootmem() manually. | |
2829 | */ | |
2830 | void __init free_bootmem_with_active_regions(int nid, | |
2831 | unsigned long max_low_pfn) | |
2832 | { | |
2833 | int i; | |
2834 | ||
2835 | for_each_active_range_index_in_nid(i, nid) { | |
2836 | unsigned long size_pages = 0; | |
2837 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
2838 | ||
2839 | if (early_node_map[i].start_pfn >= max_low_pfn) | |
2840 | continue; | |
2841 | ||
2842 | if (end_pfn > max_low_pfn) | |
2843 | end_pfn = max_low_pfn; | |
2844 | ||
2845 | size_pages = end_pfn - early_node_map[i].start_pfn; | |
2846 | free_bootmem_node(NODE_DATA(early_node_map[i].nid), | |
2847 | PFN_PHYS(early_node_map[i].start_pfn), | |
2848 | size_pages << PAGE_SHIFT); | |
2849 | } | |
2850 | } | |
2851 | ||
2852 | /** | |
2853 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 2854 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d MG |
2855 | * |
2856 | * If an architecture guarantees that all ranges registered with | |
2857 | * add_active_ranges() contain no holes and may be freed, this | |
88ca3b94 | 2858 | * function may be used instead of calling memory_present() manually. |
c713216d MG |
2859 | */ |
2860 | void __init sparse_memory_present_with_active_regions(int nid) | |
2861 | { | |
2862 | int i; | |
2863 | ||
2864 | for_each_active_range_index_in_nid(i, nid) | |
2865 | memory_present(early_node_map[i].nid, | |
2866 | early_node_map[i].start_pfn, | |
2867 | early_node_map[i].end_pfn); | |
2868 | } | |
2869 | ||
fb01439c MG |
2870 | /** |
2871 | * push_node_boundaries - Push node boundaries to at least the requested boundary | |
2872 | * @nid: The nid of the node to push the boundary for | |
2873 | * @start_pfn: The start pfn of the node | |
2874 | * @end_pfn: The end pfn of the node | |
2875 | * | |
2876 | * In reserve-based hot-add, mem_map is allocated that is unused until hotadd | |
2877 | * time. Specifically, on x86_64, SRAT will report ranges that can potentially | |
2878 | * be hotplugged even though no physical memory exists. This function allows | |
2879 | * an arch to push out the node boundaries so mem_map is allocated that can | |
2880 | * be used later. | |
2881 | */ | |
2882 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE | |
2883 | void __init push_node_boundaries(unsigned int nid, | |
2884 | unsigned long start_pfn, unsigned long end_pfn) | |
2885 | { | |
2886 | printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n", | |
2887 | nid, start_pfn, end_pfn); | |
2888 | ||
2889 | /* Initialise the boundary for this node if necessary */ | |
2890 | if (node_boundary_end_pfn[nid] == 0) | |
2891 | node_boundary_start_pfn[nid] = -1UL; | |
2892 | ||
2893 | /* Update the boundaries */ | |
2894 | if (node_boundary_start_pfn[nid] > start_pfn) | |
2895 | node_boundary_start_pfn[nid] = start_pfn; | |
2896 | if (node_boundary_end_pfn[nid] < end_pfn) | |
2897 | node_boundary_end_pfn[nid] = end_pfn; | |
2898 | } | |
2899 | ||
2900 | /* If necessary, push the node boundary out for reserve hotadd */ | |
98011f56 | 2901 | static void __meminit account_node_boundary(unsigned int nid, |
fb01439c MG |
2902 | unsigned long *start_pfn, unsigned long *end_pfn) |
2903 | { | |
2904 | printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n", | |
2905 | nid, *start_pfn, *end_pfn); | |
2906 | ||
2907 | /* Return if boundary information has not been provided */ | |
2908 | if (node_boundary_end_pfn[nid] == 0) | |
2909 | return; | |
2910 | ||
2911 | /* Check the boundaries and update if necessary */ | |
2912 | if (node_boundary_start_pfn[nid] < *start_pfn) | |
2913 | *start_pfn = node_boundary_start_pfn[nid]; | |
2914 | if (node_boundary_end_pfn[nid] > *end_pfn) | |
2915 | *end_pfn = node_boundary_end_pfn[nid]; | |
2916 | } | |
2917 | #else | |
2918 | void __init push_node_boundaries(unsigned int nid, | |
2919 | unsigned long start_pfn, unsigned long end_pfn) {} | |
2920 | ||
98011f56 | 2921 | static void __meminit account_node_boundary(unsigned int nid, |
fb01439c MG |
2922 | unsigned long *start_pfn, unsigned long *end_pfn) {} |
2923 | #endif | |
2924 | ||
2925 | ||
c713216d MG |
2926 | /** |
2927 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
2928 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
2929 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
2930 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
2931 | * |
2932 | * It returns the start and end page frame of a node based on information | |
2933 | * provided by an arch calling add_active_range(). If called for a node | |
2934 | * with no available memory, a warning is printed and the start and end | |
88ca3b94 | 2935 | * PFNs will be 0. |
c713216d | 2936 | */ |
a3142c8e | 2937 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
2938 | unsigned long *start_pfn, unsigned long *end_pfn) |
2939 | { | |
2940 | int i; | |
2941 | *start_pfn = -1UL; | |
2942 | *end_pfn = 0; | |
2943 | ||
2944 | for_each_active_range_index_in_nid(i, nid) { | |
2945 | *start_pfn = min(*start_pfn, early_node_map[i].start_pfn); | |
2946 | *end_pfn = max(*end_pfn, early_node_map[i].end_pfn); | |
2947 | } | |
2948 | ||
633c0666 | 2949 | if (*start_pfn == -1UL) |
c713216d | 2950 | *start_pfn = 0; |
fb01439c MG |
2951 | |
2952 | /* Push the node boundaries out if requested */ | |
2953 | account_node_boundary(nid, start_pfn, end_pfn); | |
c713216d MG |
2954 | } |
2955 | ||
2a1e274a MG |
2956 | /* |
2957 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
2958 | * assumption is made that zones within a node are ordered in monotonic | |
2959 | * increasing memory addresses so that the "highest" populated zone is used | |
2960 | */ | |
2961 | void __init find_usable_zone_for_movable(void) | |
2962 | { | |
2963 | int zone_index; | |
2964 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
2965 | if (zone_index == ZONE_MOVABLE) | |
2966 | continue; | |
2967 | ||
2968 | if (arch_zone_highest_possible_pfn[zone_index] > | |
2969 | arch_zone_lowest_possible_pfn[zone_index]) | |
2970 | break; | |
2971 | } | |
2972 | ||
2973 | VM_BUG_ON(zone_index == -1); | |
2974 | movable_zone = zone_index; | |
2975 | } | |
2976 | ||
2977 | /* | |
2978 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
2979 | * because it is sized independant of architecture. Unlike the other zones, | |
2980 | * the starting point for ZONE_MOVABLE is not fixed. It may be different | |
2981 | * in each node depending on the size of each node and how evenly kernelcore | |
2982 | * is distributed. This helper function adjusts the zone ranges | |
2983 | * provided by the architecture for a given node by using the end of the | |
2984 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
2985 | * zones within a node are in order of monotonic increases memory addresses | |
2986 | */ | |
2987 | void __meminit adjust_zone_range_for_zone_movable(int nid, | |
2988 | unsigned long zone_type, | |
2989 | unsigned long node_start_pfn, | |
2990 | unsigned long node_end_pfn, | |
2991 | unsigned long *zone_start_pfn, | |
2992 | unsigned long *zone_end_pfn) | |
2993 | { | |
2994 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
2995 | if (zone_movable_pfn[nid]) { | |
2996 | /* Size ZONE_MOVABLE */ | |
2997 | if (zone_type == ZONE_MOVABLE) { | |
2998 | *zone_start_pfn = zone_movable_pfn[nid]; | |
2999 | *zone_end_pfn = min(node_end_pfn, | |
3000 | arch_zone_highest_possible_pfn[movable_zone]); | |
3001 | ||
3002 | /* Adjust for ZONE_MOVABLE starting within this range */ | |
3003 | } else if (*zone_start_pfn < zone_movable_pfn[nid] && | |
3004 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
3005 | *zone_end_pfn = zone_movable_pfn[nid]; | |
3006 | ||
3007 | /* Check if this whole range is within ZONE_MOVABLE */ | |
3008 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
3009 | *zone_start_pfn = *zone_end_pfn; | |
3010 | } | |
3011 | } | |
3012 | ||
c713216d MG |
3013 | /* |
3014 | * Return the number of pages a zone spans in a node, including holes | |
3015 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
3016 | */ | |
6ea6e688 | 3017 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
3018 | unsigned long zone_type, |
3019 | unsigned long *ignored) | |
3020 | { | |
3021 | unsigned long node_start_pfn, node_end_pfn; | |
3022 | unsigned long zone_start_pfn, zone_end_pfn; | |
3023 | ||
3024 | /* Get the start and end of the node and zone */ | |
3025 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
3026 | zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; | |
3027 | zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
3028 | adjust_zone_range_for_zone_movable(nid, zone_type, |
3029 | node_start_pfn, node_end_pfn, | |
3030 | &zone_start_pfn, &zone_end_pfn); | |
c713216d MG |
3031 | |
3032 | /* Check that this node has pages within the zone's required range */ | |
3033 | if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn) | |
3034 | return 0; | |
3035 | ||
3036 | /* Move the zone boundaries inside the node if necessary */ | |
3037 | zone_end_pfn = min(zone_end_pfn, node_end_pfn); | |
3038 | zone_start_pfn = max(zone_start_pfn, node_start_pfn); | |
3039 | ||
3040 | /* Return the spanned pages */ | |
3041 | return zone_end_pfn - zone_start_pfn; | |
3042 | } | |
3043 | ||
3044 | /* | |
3045 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 3046 | * then all holes in the requested range will be accounted for. |
c713216d | 3047 | */ |
a3142c8e | 3048 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
3049 | unsigned long range_start_pfn, |
3050 | unsigned long range_end_pfn) | |
3051 | { | |
3052 | int i = 0; | |
3053 | unsigned long prev_end_pfn = 0, hole_pages = 0; | |
3054 | unsigned long start_pfn; | |
3055 | ||
3056 | /* Find the end_pfn of the first active range of pfns in the node */ | |
3057 | i = first_active_region_index_in_nid(nid); | |
3058 | if (i == -1) | |
3059 | return 0; | |
3060 | ||
b5445f95 MG |
3061 | prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn); |
3062 | ||
9c7cd687 MG |
3063 | /* Account for ranges before physical memory on this node */ |
3064 | if (early_node_map[i].start_pfn > range_start_pfn) | |
b5445f95 | 3065 | hole_pages = prev_end_pfn - range_start_pfn; |
c713216d MG |
3066 | |
3067 | /* Find all holes for the zone within the node */ | |
3068 | for (; i != -1; i = next_active_region_index_in_nid(i, nid)) { | |
3069 | ||
3070 | /* No need to continue if prev_end_pfn is outside the zone */ | |
3071 | if (prev_end_pfn >= range_end_pfn) | |
3072 | break; | |
3073 | ||
3074 | /* Make sure the end of the zone is not within the hole */ | |
3075 | start_pfn = min(early_node_map[i].start_pfn, range_end_pfn); | |
3076 | prev_end_pfn = max(prev_end_pfn, range_start_pfn); | |
3077 | ||
3078 | /* Update the hole size cound and move on */ | |
3079 | if (start_pfn > range_start_pfn) { | |
3080 | BUG_ON(prev_end_pfn > start_pfn); | |
3081 | hole_pages += start_pfn - prev_end_pfn; | |
3082 | } | |
3083 | prev_end_pfn = early_node_map[i].end_pfn; | |
3084 | } | |
3085 | ||
9c7cd687 MG |
3086 | /* Account for ranges past physical memory on this node */ |
3087 | if (range_end_pfn > prev_end_pfn) | |
0c6cb974 | 3088 | hole_pages += range_end_pfn - |
9c7cd687 MG |
3089 | max(range_start_pfn, prev_end_pfn); |
3090 | ||
c713216d MG |
3091 | return hole_pages; |
3092 | } | |
3093 | ||
3094 | /** | |
3095 | * absent_pages_in_range - Return number of page frames in holes within a range | |
3096 | * @start_pfn: The start PFN to start searching for holes | |
3097 | * @end_pfn: The end PFN to stop searching for holes | |
3098 | * | |
88ca3b94 | 3099 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
3100 | */ |
3101 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
3102 | unsigned long end_pfn) | |
3103 | { | |
3104 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
3105 | } | |
3106 | ||
3107 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 3108 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
3109 | unsigned long zone_type, |
3110 | unsigned long *ignored) | |
3111 | { | |
9c7cd687 MG |
3112 | unsigned long node_start_pfn, node_end_pfn; |
3113 | unsigned long zone_start_pfn, zone_end_pfn; | |
3114 | ||
3115 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
3116 | zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type], | |
3117 | node_start_pfn); | |
3118 | zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type], | |
3119 | node_end_pfn); | |
3120 | ||
2a1e274a MG |
3121 | adjust_zone_range_for_zone_movable(nid, zone_type, |
3122 | node_start_pfn, node_end_pfn, | |
3123 | &zone_start_pfn, &zone_end_pfn); | |
9c7cd687 | 3124 | return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
c713216d | 3125 | } |
0e0b864e | 3126 | |
c713216d | 3127 | #else |
6ea6e688 | 3128 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
3129 | unsigned long zone_type, |
3130 | unsigned long *zones_size) | |
3131 | { | |
3132 | return zones_size[zone_type]; | |
3133 | } | |
3134 | ||
6ea6e688 | 3135 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
3136 | unsigned long zone_type, |
3137 | unsigned long *zholes_size) | |
3138 | { | |
3139 | if (!zholes_size) | |
3140 | return 0; | |
3141 | ||
3142 | return zholes_size[zone_type]; | |
3143 | } | |
0e0b864e | 3144 | |
c713216d MG |
3145 | #endif |
3146 | ||
a3142c8e | 3147 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
c713216d MG |
3148 | unsigned long *zones_size, unsigned long *zholes_size) |
3149 | { | |
3150 | unsigned long realtotalpages, totalpages = 0; | |
3151 | enum zone_type i; | |
3152 | ||
3153 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3154 | totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, | |
3155 | zones_size); | |
3156 | pgdat->node_spanned_pages = totalpages; | |
3157 | ||
3158 | realtotalpages = totalpages; | |
3159 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3160 | realtotalpages -= | |
3161 | zone_absent_pages_in_node(pgdat->node_id, i, | |
3162 | zholes_size); | |
3163 | pgdat->node_present_pages = realtotalpages; | |
3164 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
3165 | realtotalpages); | |
3166 | } | |
3167 | ||
835c134e MG |
3168 | #ifndef CONFIG_SPARSEMEM |
3169 | /* | |
3170 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
3171 | * Start by making sure zonesize is a multiple of MAX_ORDER-1 by rounding up | |
3172 | * Then figure 1 NR_PAGEBLOCK_BITS worth of bits per MAX_ORDER-1, finally | |
3173 | * round what is now in bits to nearest long in bits, then return it in | |
3174 | * bytes. | |
3175 | */ | |
3176 | static unsigned long __init usemap_size(unsigned long zonesize) | |
3177 | { | |
3178 | unsigned long usemapsize; | |
3179 | ||
3180 | usemapsize = roundup(zonesize, MAX_ORDER_NR_PAGES); | |
3181 | usemapsize = usemapsize >> (MAX_ORDER-1); | |
3182 | usemapsize *= NR_PAGEBLOCK_BITS; | |
3183 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
3184 | ||
3185 | return usemapsize / 8; | |
3186 | } | |
3187 | ||
3188 | static void __init setup_usemap(struct pglist_data *pgdat, | |
3189 | struct zone *zone, unsigned long zonesize) | |
3190 | { | |
3191 | unsigned long usemapsize = usemap_size(zonesize); | |
3192 | zone->pageblock_flags = NULL; | |
3193 | if (usemapsize) { | |
3194 | zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize); | |
3195 | memset(zone->pageblock_flags, 0, usemapsize); | |
3196 | } | |
3197 | } | |
3198 | #else | |
3199 | static void inline setup_usemap(struct pglist_data *pgdat, | |
3200 | struct zone *zone, unsigned long zonesize) {} | |
3201 | #endif /* CONFIG_SPARSEMEM */ | |
3202 | ||
1da177e4 LT |
3203 | /* |
3204 | * Set up the zone data structures: | |
3205 | * - mark all pages reserved | |
3206 | * - mark all memory queues empty | |
3207 | * - clear the memory bitmaps | |
3208 | */ | |
86356ab1 | 3209 | static void __meminit free_area_init_core(struct pglist_data *pgdat, |
1da177e4 LT |
3210 | unsigned long *zones_size, unsigned long *zholes_size) |
3211 | { | |
2f1b6248 | 3212 | enum zone_type j; |
ed8ece2e | 3213 | int nid = pgdat->node_id; |
1da177e4 | 3214 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
718127cc | 3215 | int ret; |
1da177e4 | 3216 | |
208d54e5 | 3217 | pgdat_resize_init(pgdat); |
1da177e4 LT |
3218 | pgdat->nr_zones = 0; |
3219 | init_waitqueue_head(&pgdat->kswapd_wait); | |
3220 | pgdat->kswapd_max_order = 0; | |
3221 | ||
3222 | for (j = 0; j < MAX_NR_ZONES; j++) { | |
3223 | struct zone *zone = pgdat->node_zones + j; | |
0e0b864e | 3224 | unsigned long size, realsize, memmap_pages; |
1da177e4 | 3225 | |
c713216d MG |
3226 | size = zone_spanned_pages_in_node(nid, j, zones_size); |
3227 | realsize = size - zone_absent_pages_in_node(nid, j, | |
3228 | zholes_size); | |
1da177e4 | 3229 | |
0e0b864e MG |
3230 | /* |
3231 | * Adjust realsize so that it accounts for how much memory | |
3232 | * is used by this zone for memmap. This affects the watermark | |
3233 | * and per-cpu initialisations | |
3234 | */ | |
3235 | memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT; | |
3236 | if (realsize >= memmap_pages) { | |
3237 | realsize -= memmap_pages; | |
3238 | printk(KERN_DEBUG | |
3239 | " %s zone: %lu pages used for memmap\n", | |
3240 | zone_names[j], memmap_pages); | |
3241 | } else | |
3242 | printk(KERN_WARNING | |
3243 | " %s zone: %lu pages exceeds realsize %lu\n", | |
3244 | zone_names[j], memmap_pages, realsize); | |
3245 | ||
6267276f CL |
3246 | /* Account for reserved pages */ |
3247 | if (j == 0 && realsize > dma_reserve) { | |
0e0b864e | 3248 | realsize -= dma_reserve; |
6267276f CL |
3249 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
3250 | zone_names[0], dma_reserve); | |
0e0b864e MG |
3251 | } |
3252 | ||
98d2b0eb | 3253 | if (!is_highmem_idx(j)) |
1da177e4 LT |
3254 | nr_kernel_pages += realsize; |
3255 | nr_all_pages += realsize; | |
3256 | ||
3257 | zone->spanned_pages = size; | |
3258 | zone->present_pages = realsize; | |
9614634f | 3259 | #ifdef CONFIG_NUMA |
d5f541ed | 3260 | zone->node = nid; |
8417bba4 | 3261 | zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio) |
9614634f | 3262 | / 100; |
0ff38490 | 3263 | zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100; |
9614634f | 3264 | #endif |
1da177e4 LT |
3265 | zone->name = zone_names[j]; |
3266 | spin_lock_init(&zone->lock); | |
3267 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 3268 | zone_seqlock_init(zone); |
1da177e4 | 3269 | zone->zone_pgdat = pgdat; |
1da177e4 | 3270 | |
3bb1a852 | 3271 | zone->prev_priority = DEF_PRIORITY; |
1da177e4 | 3272 | |
ed8ece2e | 3273 | zone_pcp_init(zone); |
1da177e4 LT |
3274 | INIT_LIST_HEAD(&zone->active_list); |
3275 | INIT_LIST_HEAD(&zone->inactive_list); | |
3276 | zone->nr_scan_active = 0; | |
3277 | zone->nr_scan_inactive = 0; | |
2244b95a | 3278 | zap_zone_vm_stats(zone); |
53e9a615 | 3279 | atomic_set(&zone->reclaim_in_progress, 0); |
1da177e4 LT |
3280 | if (!size) |
3281 | continue; | |
3282 | ||
835c134e | 3283 | setup_usemap(pgdat, zone, size); |
a2f3aa02 DH |
3284 | ret = init_currently_empty_zone(zone, zone_start_pfn, |
3285 | size, MEMMAP_EARLY); | |
718127cc | 3286 | BUG_ON(ret); |
1da177e4 | 3287 | zone_start_pfn += size; |
1da177e4 LT |
3288 | } |
3289 | } | |
3290 | ||
577a32f6 | 3291 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 3292 | { |
1da177e4 LT |
3293 | /* Skip empty nodes */ |
3294 | if (!pgdat->node_spanned_pages) | |
3295 | return; | |
3296 | ||
d41dee36 | 3297 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
3298 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
3299 | if (!pgdat->node_mem_map) { | |
e984bb43 | 3300 | unsigned long size, start, end; |
d41dee36 AW |
3301 | struct page *map; |
3302 | ||
e984bb43 BP |
3303 | /* |
3304 | * The zone's endpoints aren't required to be MAX_ORDER | |
3305 | * aligned but the node_mem_map endpoints must be in order | |
3306 | * for the buddy allocator to function correctly. | |
3307 | */ | |
3308 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
3309 | end = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
3310 | end = ALIGN(end, MAX_ORDER_NR_PAGES); | |
3311 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
3312 | map = alloc_remap(pgdat->node_id, size); |
3313 | if (!map) | |
3314 | map = alloc_bootmem_node(pgdat, size); | |
e984bb43 | 3315 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
1da177e4 | 3316 | } |
12d810c1 | 3317 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
3318 | /* |
3319 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
3320 | */ | |
c713216d | 3321 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 3322 | mem_map = NODE_DATA(0)->node_mem_map; |
c713216d MG |
3323 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
3324 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) | |
3325 | mem_map -= pgdat->node_start_pfn; | |
3326 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ | |
3327 | } | |
1da177e4 | 3328 | #endif |
d41dee36 | 3329 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
3330 | } |
3331 | ||
86356ab1 | 3332 | void __meminit free_area_init_node(int nid, struct pglist_data *pgdat, |
1da177e4 LT |
3333 | unsigned long *zones_size, unsigned long node_start_pfn, |
3334 | unsigned long *zholes_size) | |
3335 | { | |
3336 | pgdat->node_id = nid; | |
3337 | pgdat->node_start_pfn = node_start_pfn; | |
c713216d | 3338 | calculate_node_totalpages(pgdat, zones_size, zholes_size); |
1da177e4 LT |
3339 | |
3340 | alloc_node_mem_map(pgdat); | |
3341 | ||
3342 | free_area_init_core(pgdat, zones_size, zholes_size); | |
3343 | } | |
3344 | ||
c713216d | 3345 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
418508c1 MS |
3346 | |
3347 | #if MAX_NUMNODES > 1 | |
3348 | /* | |
3349 | * Figure out the number of possible node ids. | |
3350 | */ | |
3351 | static void __init setup_nr_node_ids(void) | |
3352 | { | |
3353 | unsigned int node; | |
3354 | unsigned int highest = 0; | |
3355 | ||
3356 | for_each_node_mask(node, node_possible_map) | |
3357 | highest = node; | |
3358 | nr_node_ids = highest + 1; | |
3359 | } | |
3360 | #else | |
3361 | static inline void setup_nr_node_ids(void) | |
3362 | { | |
3363 | } | |
3364 | #endif | |
3365 | ||
c713216d MG |
3366 | /** |
3367 | * add_active_range - Register a range of PFNs backed by physical memory | |
3368 | * @nid: The node ID the range resides on | |
3369 | * @start_pfn: The start PFN of the available physical memory | |
3370 | * @end_pfn: The end PFN of the available physical memory | |
3371 | * | |
3372 | * These ranges are stored in an early_node_map[] and later used by | |
3373 | * free_area_init_nodes() to calculate zone sizes and holes. If the | |
3374 | * range spans a memory hole, it is up to the architecture to ensure | |
3375 | * the memory is not freed by the bootmem allocator. If possible | |
3376 | * the range being registered will be merged with existing ranges. | |
3377 | */ | |
3378 | void __init add_active_range(unsigned int nid, unsigned long start_pfn, | |
3379 | unsigned long end_pfn) | |
3380 | { | |
3381 | int i; | |
3382 | ||
3383 | printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) " | |
3384 | "%d entries of %d used\n", | |
3385 | nid, start_pfn, end_pfn, | |
3386 | nr_nodemap_entries, MAX_ACTIVE_REGIONS); | |
3387 | ||
3388 | /* Merge with existing active regions if possible */ | |
3389 | for (i = 0; i < nr_nodemap_entries; i++) { | |
3390 | if (early_node_map[i].nid != nid) | |
3391 | continue; | |
3392 | ||
3393 | /* Skip if an existing region covers this new one */ | |
3394 | if (start_pfn >= early_node_map[i].start_pfn && | |
3395 | end_pfn <= early_node_map[i].end_pfn) | |
3396 | return; | |
3397 | ||
3398 | /* Merge forward if suitable */ | |
3399 | if (start_pfn <= early_node_map[i].end_pfn && | |
3400 | end_pfn > early_node_map[i].end_pfn) { | |
3401 | early_node_map[i].end_pfn = end_pfn; | |
3402 | return; | |
3403 | } | |
3404 | ||
3405 | /* Merge backward if suitable */ | |
3406 | if (start_pfn < early_node_map[i].end_pfn && | |
3407 | end_pfn >= early_node_map[i].start_pfn) { | |
3408 | early_node_map[i].start_pfn = start_pfn; | |
3409 | return; | |
3410 | } | |
3411 | } | |
3412 | ||
3413 | /* Check that early_node_map is large enough */ | |
3414 | if (i >= MAX_ACTIVE_REGIONS) { | |
3415 | printk(KERN_CRIT "More than %d memory regions, truncating\n", | |
3416 | MAX_ACTIVE_REGIONS); | |
3417 | return; | |
3418 | } | |
3419 | ||
3420 | early_node_map[i].nid = nid; | |
3421 | early_node_map[i].start_pfn = start_pfn; | |
3422 | early_node_map[i].end_pfn = end_pfn; | |
3423 | nr_nodemap_entries = i + 1; | |
3424 | } | |
3425 | ||
3426 | /** | |
3427 | * shrink_active_range - Shrink an existing registered range of PFNs | |
3428 | * @nid: The node id the range is on that should be shrunk | |
3429 | * @old_end_pfn: The old end PFN of the range | |
3430 | * @new_end_pfn: The new PFN of the range | |
3431 | * | |
3432 | * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node. | |
3433 | * The map is kept at the end physical page range that has already been | |
3434 | * registered with add_active_range(). This function allows an arch to shrink | |
3435 | * an existing registered range. | |
3436 | */ | |
3437 | void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn, | |
3438 | unsigned long new_end_pfn) | |
3439 | { | |
3440 | int i; | |
3441 | ||
3442 | /* Find the old active region end and shrink */ | |
3443 | for_each_active_range_index_in_nid(i, nid) | |
3444 | if (early_node_map[i].end_pfn == old_end_pfn) { | |
3445 | early_node_map[i].end_pfn = new_end_pfn; | |
3446 | break; | |
3447 | } | |
3448 | } | |
3449 | ||
3450 | /** | |
3451 | * remove_all_active_ranges - Remove all currently registered regions | |
88ca3b94 | 3452 | * |
c713216d MG |
3453 | * During discovery, it may be found that a table like SRAT is invalid |
3454 | * and an alternative discovery method must be used. This function removes | |
3455 | * all currently registered regions. | |
3456 | */ | |
88ca3b94 | 3457 | void __init remove_all_active_ranges(void) |
c713216d MG |
3458 | { |
3459 | memset(early_node_map, 0, sizeof(early_node_map)); | |
3460 | nr_nodemap_entries = 0; | |
fb01439c MG |
3461 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE |
3462 | memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn)); | |
3463 | memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn)); | |
3464 | #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */ | |
c713216d MG |
3465 | } |
3466 | ||
3467 | /* Compare two active node_active_regions */ | |
3468 | static int __init cmp_node_active_region(const void *a, const void *b) | |
3469 | { | |
3470 | struct node_active_region *arange = (struct node_active_region *)a; | |
3471 | struct node_active_region *brange = (struct node_active_region *)b; | |
3472 | ||
3473 | /* Done this way to avoid overflows */ | |
3474 | if (arange->start_pfn > brange->start_pfn) | |
3475 | return 1; | |
3476 | if (arange->start_pfn < brange->start_pfn) | |
3477 | return -1; | |
3478 | ||
3479 | return 0; | |
3480 | } | |
3481 | ||
3482 | /* sort the node_map by start_pfn */ | |
3483 | static void __init sort_node_map(void) | |
3484 | { | |
3485 | sort(early_node_map, (size_t)nr_nodemap_entries, | |
3486 | sizeof(struct node_active_region), | |
3487 | cmp_node_active_region, NULL); | |
3488 | } | |
3489 | ||
a6af2bc3 | 3490 | /* Find the lowest pfn for a node */ |
c713216d MG |
3491 | unsigned long __init find_min_pfn_for_node(unsigned long nid) |
3492 | { | |
3493 | int i; | |
a6af2bc3 | 3494 | unsigned long min_pfn = ULONG_MAX; |
1abbfb41 | 3495 | |
c713216d MG |
3496 | /* Assuming a sorted map, the first range found has the starting pfn */ |
3497 | for_each_active_range_index_in_nid(i, nid) | |
a6af2bc3 | 3498 | min_pfn = min(min_pfn, early_node_map[i].start_pfn); |
c713216d | 3499 | |
a6af2bc3 MG |
3500 | if (min_pfn == ULONG_MAX) { |
3501 | printk(KERN_WARNING | |
3502 | "Could not find start_pfn for node %lu\n", nid); | |
3503 | return 0; | |
3504 | } | |
3505 | ||
3506 | return min_pfn; | |
c713216d MG |
3507 | } |
3508 | ||
3509 | /** | |
3510 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
3511 | * | |
3512 | * It returns the minimum PFN based on information provided via | |
88ca3b94 | 3513 | * add_active_range(). |
c713216d MG |
3514 | */ |
3515 | unsigned long __init find_min_pfn_with_active_regions(void) | |
3516 | { | |
3517 | return find_min_pfn_for_node(MAX_NUMNODES); | |
3518 | } | |
3519 | ||
3520 | /** | |
3521 | * find_max_pfn_with_active_regions - Find the maximum PFN registered | |
3522 | * | |
3523 | * It returns the maximum PFN based on information provided via | |
88ca3b94 | 3524 | * add_active_range(). |
c713216d MG |
3525 | */ |
3526 | unsigned long __init find_max_pfn_with_active_regions(void) | |
3527 | { | |
3528 | int i; | |
3529 | unsigned long max_pfn = 0; | |
3530 | ||
3531 | for (i = 0; i < nr_nodemap_entries; i++) | |
3532 | max_pfn = max(max_pfn, early_node_map[i].end_pfn); | |
3533 | ||
3534 | return max_pfn; | |
3535 | } | |
3536 | ||
37b07e41 LS |
3537 | /* |
3538 | * early_calculate_totalpages() | |
3539 | * Sum pages in active regions for movable zone. | |
3540 | * Populate N_HIGH_MEMORY for calculating usable_nodes. | |
3541 | */ | |
7e63efef MG |
3542 | unsigned long __init early_calculate_totalpages(void) |
3543 | { | |
3544 | int i; | |
3545 | unsigned long totalpages = 0; | |
3546 | ||
37b07e41 LS |
3547 | for (i = 0; i < nr_nodemap_entries; i++) { |
3548 | unsigned long pages = early_node_map[i].end_pfn - | |
7e63efef | 3549 | early_node_map[i].start_pfn; |
37b07e41 LS |
3550 | totalpages += pages; |
3551 | if (pages) | |
3552 | node_set_state(early_node_map[i].nid, N_HIGH_MEMORY); | |
3553 | } | |
3554 | return totalpages; | |
7e63efef MG |
3555 | } |
3556 | ||
2a1e274a MG |
3557 | /* |
3558 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
3559 | * is spread evenly between nodes as long as the nodes have enough | |
3560 | * memory. When they don't, some nodes will have more kernelcore than | |
3561 | * others | |
3562 | */ | |
3563 | void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn) | |
3564 | { | |
3565 | int i, nid; | |
3566 | unsigned long usable_startpfn; | |
3567 | unsigned long kernelcore_node, kernelcore_remaining; | |
37b07e41 LS |
3568 | unsigned long totalpages = early_calculate_totalpages(); |
3569 | int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]); | |
2a1e274a | 3570 | |
7e63efef MG |
3571 | /* |
3572 | * If movablecore was specified, calculate what size of | |
3573 | * kernelcore that corresponds so that memory usable for | |
3574 | * any allocation type is evenly spread. If both kernelcore | |
3575 | * and movablecore are specified, then the value of kernelcore | |
3576 | * will be used for required_kernelcore if it's greater than | |
3577 | * what movablecore would have allowed. | |
3578 | */ | |
3579 | if (required_movablecore) { | |
7e63efef MG |
3580 | unsigned long corepages; |
3581 | ||
3582 | /* | |
3583 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
3584 | * was requested by the user | |
3585 | */ | |
3586 | required_movablecore = | |
3587 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
3588 | corepages = totalpages - required_movablecore; | |
3589 | ||
3590 | required_kernelcore = max(required_kernelcore, corepages); | |
3591 | } | |
3592 | ||
2a1e274a MG |
3593 | /* If kernelcore was not specified, there is no ZONE_MOVABLE */ |
3594 | if (!required_kernelcore) | |
3595 | return; | |
3596 | ||
3597 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
3598 | find_usable_zone_for_movable(); | |
3599 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; | |
3600 | ||
3601 | restart: | |
3602 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
3603 | kernelcore_node = required_kernelcore / usable_nodes; | |
37b07e41 | 3604 | for_each_node_state(nid, N_HIGH_MEMORY) { |
2a1e274a MG |
3605 | /* |
3606 | * Recalculate kernelcore_node if the division per node | |
3607 | * now exceeds what is necessary to satisfy the requested | |
3608 | * amount of memory for the kernel | |
3609 | */ | |
3610 | if (required_kernelcore < kernelcore_node) | |
3611 | kernelcore_node = required_kernelcore / usable_nodes; | |
3612 | ||
3613 | /* | |
3614 | * As the map is walked, we track how much memory is usable | |
3615 | * by the kernel using kernelcore_remaining. When it is | |
3616 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
3617 | */ | |
3618 | kernelcore_remaining = kernelcore_node; | |
3619 | ||
3620 | /* Go through each range of PFNs within this node */ | |
3621 | for_each_active_range_index_in_nid(i, nid) { | |
3622 | unsigned long start_pfn, end_pfn; | |
3623 | unsigned long size_pages; | |
3624 | ||
3625 | start_pfn = max(early_node_map[i].start_pfn, | |
3626 | zone_movable_pfn[nid]); | |
3627 | end_pfn = early_node_map[i].end_pfn; | |
3628 | if (start_pfn >= end_pfn) | |
3629 | continue; | |
3630 | ||
3631 | /* Account for what is only usable for kernelcore */ | |
3632 | if (start_pfn < usable_startpfn) { | |
3633 | unsigned long kernel_pages; | |
3634 | kernel_pages = min(end_pfn, usable_startpfn) | |
3635 | - start_pfn; | |
3636 | ||
3637 | kernelcore_remaining -= min(kernel_pages, | |
3638 | kernelcore_remaining); | |
3639 | required_kernelcore -= min(kernel_pages, | |
3640 | required_kernelcore); | |
3641 | ||
3642 | /* Continue if range is now fully accounted */ | |
3643 | if (end_pfn <= usable_startpfn) { | |
3644 | ||
3645 | /* | |
3646 | * Push zone_movable_pfn to the end so | |
3647 | * that if we have to rebalance | |
3648 | * kernelcore across nodes, we will | |
3649 | * not double account here | |
3650 | */ | |
3651 | zone_movable_pfn[nid] = end_pfn; | |
3652 | continue; | |
3653 | } | |
3654 | start_pfn = usable_startpfn; | |
3655 | } | |
3656 | ||
3657 | /* | |
3658 | * The usable PFN range for ZONE_MOVABLE is from | |
3659 | * start_pfn->end_pfn. Calculate size_pages as the | |
3660 | * number of pages used as kernelcore | |
3661 | */ | |
3662 | size_pages = end_pfn - start_pfn; | |
3663 | if (size_pages > kernelcore_remaining) | |
3664 | size_pages = kernelcore_remaining; | |
3665 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
3666 | ||
3667 | /* | |
3668 | * Some kernelcore has been met, update counts and | |
3669 | * break if the kernelcore for this node has been | |
3670 | * satisified | |
3671 | */ | |
3672 | required_kernelcore -= min(required_kernelcore, | |
3673 | size_pages); | |
3674 | kernelcore_remaining -= size_pages; | |
3675 | if (!kernelcore_remaining) | |
3676 | break; | |
3677 | } | |
3678 | } | |
3679 | ||
3680 | /* | |
3681 | * If there is still required_kernelcore, we do another pass with one | |
3682 | * less node in the count. This will push zone_movable_pfn[nid] further | |
3683 | * along on the nodes that still have memory until kernelcore is | |
3684 | * satisified | |
3685 | */ | |
3686 | usable_nodes--; | |
3687 | if (usable_nodes && required_kernelcore > usable_nodes) | |
3688 | goto restart; | |
3689 | ||
3690 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ | |
3691 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
3692 | zone_movable_pfn[nid] = | |
3693 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
3694 | } | |
3695 | ||
37b07e41 LS |
3696 | /* Any regular memory on that node ? */ |
3697 | static void check_for_regular_memory(pg_data_t *pgdat) | |
3698 | { | |
3699 | #ifdef CONFIG_HIGHMEM | |
3700 | enum zone_type zone_type; | |
3701 | ||
3702 | for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) { | |
3703 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
3704 | if (zone->present_pages) | |
3705 | node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY); | |
3706 | } | |
3707 | #endif | |
3708 | } | |
3709 | ||
c713216d MG |
3710 | /** |
3711 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 3712 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
3713 | * |
3714 | * This will call free_area_init_node() for each active node in the system. | |
3715 | * Using the page ranges provided by add_active_range(), the size of each | |
3716 | * zone in each node and their holes is calculated. If the maximum PFN | |
3717 | * between two adjacent zones match, it is assumed that the zone is empty. | |
3718 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
3719 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
3720 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
3721 | * at arch_max_dma_pfn. | |
3722 | */ | |
3723 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
3724 | { | |
3725 | unsigned long nid; | |
3726 | enum zone_type i; | |
3727 | ||
a6af2bc3 MG |
3728 | /* Sort early_node_map as initialisation assumes it is sorted */ |
3729 | sort_node_map(); | |
3730 | ||
c713216d MG |
3731 | /* Record where the zone boundaries are */ |
3732 | memset(arch_zone_lowest_possible_pfn, 0, | |
3733 | sizeof(arch_zone_lowest_possible_pfn)); | |
3734 | memset(arch_zone_highest_possible_pfn, 0, | |
3735 | sizeof(arch_zone_highest_possible_pfn)); | |
3736 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
3737 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
3738 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
3739 | if (i == ZONE_MOVABLE) |
3740 | continue; | |
c713216d MG |
3741 | arch_zone_lowest_possible_pfn[i] = |
3742 | arch_zone_highest_possible_pfn[i-1]; | |
3743 | arch_zone_highest_possible_pfn[i] = | |
3744 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
3745 | } | |
2a1e274a MG |
3746 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
3747 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; | |
3748 | ||
3749 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
3750 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
3751 | find_zone_movable_pfns_for_nodes(zone_movable_pfn); | |
c713216d | 3752 | |
c713216d MG |
3753 | /* Print out the zone ranges */ |
3754 | printk("Zone PFN ranges:\n"); | |
2a1e274a MG |
3755 | for (i = 0; i < MAX_NR_ZONES; i++) { |
3756 | if (i == ZONE_MOVABLE) | |
3757 | continue; | |
c713216d MG |
3758 | printk(" %-8s %8lu -> %8lu\n", |
3759 | zone_names[i], | |
3760 | arch_zone_lowest_possible_pfn[i], | |
3761 | arch_zone_highest_possible_pfn[i]); | |
2a1e274a MG |
3762 | } |
3763 | ||
3764 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
3765 | printk("Movable zone start PFN for each node\n"); | |
3766 | for (i = 0; i < MAX_NUMNODES; i++) { | |
3767 | if (zone_movable_pfn[i]) | |
3768 | printk(" Node %d: %lu\n", i, zone_movable_pfn[i]); | |
3769 | } | |
c713216d MG |
3770 | |
3771 | /* Print out the early_node_map[] */ | |
3772 | printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries); | |
3773 | for (i = 0; i < nr_nodemap_entries; i++) | |
3774 | printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid, | |
3775 | early_node_map[i].start_pfn, | |
3776 | early_node_map[i].end_pfn); | |
3777 | ||
3778 | /* Initialise every node */ | |
8ef82866 | 3779 | setup_nr_node_ids(); |
c713216d MG |
3780 | for_each_online_node(nid) { |
3781 | pg_data_t *pgdat = NODE_DATA(nid); | |
3782 | free_area_init_node(nid, pgdat, NULL, | |
3783 | find_min_pfn_for_node(nid), NULL); | |
37b07e41 LS |
3784 | |
3785 | /* Any memory on that node */ | |
3786 | if (pgdat->node_present_pages) | |
3787 | node_set_state(nid, N_HIGH_MEMORY); | |
3788 | check_for_regular_memory(pgdat); | |
c713216d MG |
3789 | } |
3790 | } | |
2a1e274a | 3791 | |
7e63efef | 3792 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
3793 | { |
3794 | unsigned long long coremem; | |
3795 | if (!p) | |
3796 | return -EINVAL; | |
3797 | ||
3798 | coremem = memparse(p, &p); | |
7e63efef | 3799 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 3800 | |
7e63efef | 3801 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
3802 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
3803 | ||
3804 | return 0; | |
3805 | } | |
ed7ed365 | 3806 | |
7e63efef MG |
3807 | /* |
3808 | * kernelcore=size sets the amount of memory for use for allocations that | |
3809 | * cannot be reclaimed or migrated. | |
3810 | */ | |
3811 | static int __init cmdline_parse_kernelcore(char *p) | |
3812 | { | |
3813 | return cmdline_parse_core(p, &required_kernelcore); | |
3814 | } | |
3815 | ||
3816 | /* | |
3817 | * movablecore=size sets the amount of memory for use for allocations that | |
3818 | * can be reclaimed or migrated. | |
3819 | */ | |
3820 | static int __init cmdline_parse_movablecore(char *p) | |
3821 | { | |
3822 | return cmdline_parse_core(p, &required_movablecore); | |
3823 | } | |
3824 | ||
ed7ed365 | 3825 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 3826 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 3827 | |
c713216d MG |
3828 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
3829 | ||
0e0b864e | 3830 | /** |
88ca3b94 RD |
3831 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
3832 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e MG |
3833 | * |
3834 | * The per-cpu batchsize and zone watermarks are determined by present_pages. | |
3835 | * In the DMA zone, a significant percentage may be consumed by kernel image | |
3836 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
3837 | * function may optionally be used to account for unfreeable pages in the |
3838 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
3839 | * smaller per-cpu batchsize. | |
0e0b864e MG |
3840 | */ |
3841 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
3842 | { | |
3843 | dma_reserve = new_dma_reserve; | |
3844 | } | |
3845 | ||
93b7504e | 3846 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
3847 | static bootmem_data_t contig_bootmem_data; |
3848 | struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data }; | |
3849 | ||
3850 | EXPORT_SYMBOL(contig_page_data); | |
93b7504e | 3851 | #endif |
1da177e4 LT |
3852 | |
3853 | void __init free_area_init(unsigned long *zones_size) | |
3854 | { | |
93b7504e | 3855 | free_area_init_node(0, NODE_DATA(0), zones_size, |
1da177e4 LT |
3856 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
3857 | } | |
1da177e4 | 3858 | |
1da177e4 LT |
3859 | static int page_alloc_cpu_notify(struct notifier_block *self, |
3860 | unsigned long action, void *hcpu) | |
3861 | { | |
3862 | int cpu = (unsigned long)hcpu; | |
1da177e4 | 3863 | |
8bb78442 | 3864 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
1da177e4 LT |
3865 | local_irq_disable(); |
3866 | __drain_pages(cpu); | |
f8891e5e | 3867 | vm_events_fold_cpu(cpu); |
1da177e4 | 3868 | local_irq_enable(); |
2244b95a | 3869 | refresh_cpu_vm_stats(cpu); |
1da177e4 LT |
3870 | } |
3871 | return NOTIFY_OK; | |
3872 | } | |
1da177e4 LT |
3873 | |
3874 | void __init page_alloc_init(void) | |
3875 | { | |
3876 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
3877 | } | |
3878 | ||
cb45b0e9 HA |
3879 | /* |
3880 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio | |
3881 | * or min_free_kbytes changes. | |
3882 | */ | |
3883 | static void calculate_totalreserve_pages(void) | |
3884 | { | |
3885 | struct pglist_data *pgdat; | |
3886 | unsigned long reserve_pages = 0; | |
2f6726e5 | 3887 | enum zone_type i, j; |
cb45b0e9 HA |
3888 | |
3889 | for_each_online_pgdat(pgdat) { | |
3890 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
3891 | struct zone *zone = pgdat->node_zones + i; | |
3892 | unsigned long max = 0; | |
3893 | ||
3894 | /* Find valid and maximum lowmem_reserve in the zone */ | |
3895 | for (j = i; j < MAX_NR_ZONES; j++) { | |
3896 | if (zone->lowmem_reserve[j] > max) | |
3897 | max = zone->lowmem_reserve[j]; | |
3898 | } | |
3899 | ||
3900 | /* we treat pages_high as reserved pages. */ | |
3901 | max += zone->pages_high; | |
3902 | ||
3903 | if (max > zone->present_pages) | |
3904 | max = zone->present_pages; | |
3905 | reserve_pages += max; | |
3906 | } | |
3907 | } | |
3908 | totalreserve_pages = reserve_pages; | |
3909 | } | |
3910 | ||
1da177e4 LT |
3911 | /* |
3912 | * setup_per_zone_lowmem_reserve - called whenever | |
3913 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
3914 | * has a correct pages reserved value, so an adequate number of | |
3915 | * pages are left in the zone after a successful __alloc_pages(). | |
3916 | */ | |
3917 | static void setup_per_zone_lowmem_reserve(void) | |
3918 | { | |
3919 | struct pglist_data *pgdat; | |
2f6726e5 | 3920 | enum zone_type j, idx; |
1da177e4 | 3921 | |
ec936fc5 | 3922 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
3923 | for (j = 0; j < MAX_NR_ZONES; j++) { |
3924 | struct zone *zone = pgdat->node_zones + j; | |
3925 | unsigned long present_pages = zone->present_pages; | |
3926 | ||
3927 | zone->lowmem_reserve[j] = 0; | |
3928 | ||
2f6726e5 CL |
3929 | idx = j; |
3930 | while (idx) { | |
1da177e4 LT |
3931 | struct zone *lower_zone; |
3932 | ||
2f6726e5 CL |
3933 | idx--; |
3934 | ||
1da177e4 LT |
3935 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
3936 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
3937 | ||
3938 | lower_zone = pgdat->node_zones + idx; | |
3939 | lower_zone->lowmem_reserve[j] = present_pages / | |
3940 | sysctl_lowmem_reserve_ratio[idx]; | |
3941 | present_pages += lower_zone->present_pages; | |
3942 | } | |
3943 | } | |
3944 | } | |
cb45b0e9 HA |
3945 | |
3946 | /* update totalreserve_pages */ | |
3947 | calculate_totalreserve_pages(); | |
1da177e4 LT |
3948 | } |
3949 | ||
88ca3b94 RD |
3950 | /** |
3951 | * setup_per_zone_pages_min - called when min_free_kbytes changes. | |
3952 | * | |
3953 | * Ensures that the pages_{min,low,high} values for each zone are set correctly | |
3954 | * with respect to min_free_kbytes. | |
1da177e4 | 3955 | */ |
3947be19 | 3956 | void setup_per_zone_pages_min(void) |
1da177e4 LT |
3957 | { |
3958 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
3959 | unsigned long lowmem_pages = 0; | |
3960 | struct zone *zone; | |
3961 | unsigned long flags; | |
3962 | ||
3963 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
3964 | for_each_zone(zone) { | |
3965 | if (!is_highmem(zone)) | |
3966 | lowmem_pages += zone->present_pages; | |
3967 | } | |
3968 | ||
3969 | for_each_zone(zone) { | |
ac924c60 AM |
3970 | u64 tmp; |
3971 | ||
1da177e4 | 3972 | spin_lock_irqsave(&zone->lru_lock, flags); |
ac924c60 AM |
3973 | tmp = (u64)pages_min * zone->present_pages; |
3974 | do_div(tmp, lowmem_pages); | |
1da177e4 LT |
3975 | if (is_highmem(zone)) { |
3976 | /* | |
669ed175 NP |
3977 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
3978 | * need highmem pages, so cap pages_min to a small | |
3979 | * value here. | |
3980 | * | |
3981 | * The (pages_high-pages_low) and (pages_low-pages_min) | |
3982 | * deltas controls asynch page reclaim, and so should | |
3983 | * not be capped for highmem. | |
1da177e4 LT |
3984 | */ |
3985 | int min_pages; | |
3986 | ||
3987 | min_pages = zone->present_pages / 1024; | |
3988 | if (min_pages < SWAP_CLUSTER_MAX) | |
3989 | min_pages = SWAP_CLUSTER_MAX; | |
3990 | if (min_pages > 128) | |
3991 | min_pages = 128; | |
3992 | zone->pages_min = min_pages; | |
3993 | } else { | |
669ed175 NP |
3994 | /* |
3995 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
3996 | * proportionate to the zone's size. |
3997 | */ | |
669ed175 | 3998 | zone->pages_min = tmp; |
1da177e4 LT |
3999 | } |
4000 | ||
ac924c60 AM |
4001 | zone->pages_low = zone->pages_min + (tmp >> 2); |
4002 | zone->pages_high = zone->pages_min + (tmp >> 1); | |
1da177e4 LT |
4003 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
4004 | } | |
cb45b0e9 HA |
4005 | |
4006 | /* update totalreserve_pages */ | |
4007 | calculate_totalreserve_pages(); | |
1da177e4 LT |
4008 | } |
4009 | ||
4010 | /* | |
4011 | * Initialise min_free_kbytes. | |
4012 | * | |
4013 | * For small machines we want it small (128k min). For large machines | |
4014 | * we want it large (64MB max). But it is not linear, because network | |
4015 | * bandwidth does not increase linearly with machine size. We use | |
4016 | * | |
4017 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: | |
4018 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) | |
4019 | * | |
4020 | * which yields | |
4021 | * | |
4022 | * 16MB: 512k | |
4023 | * 32MB: 724k | |
4024 | * 64MB: 1024k | |
4025 | * 128MB: 1448k | |
4026 | * 256MB: 2048k | |
4027 | * 512MB: 2896k | |
4028 | * 1024MB: 4096k | |
4029 | * 2048MB: 5792k | |
4030 | * 4096MB: 8192k | |
4031 | * 8192MB: 11584k | |
4032 | * 16384MB: 16384k | |
4033 | */ | |
4034 | static int __init init_per_zone_pages_min(void) | |
4035 | { | |
4036 | unsigned long lowmem_kbytes; | |
4037 | ||
4038 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
4039 | ||
4040 | min_free_kbytes = int_sqrt(lowmem_kbytes * 16); | |
4041 | if (min_free_kbytes < 128) | |
4042 | min_free_kbytes = 128; | |
4043 | if (min_free_kbytes > 65536) | |
4044 | min_free_kbytes = 65536; | |
4045 | setup_per_zone_pages_min(); | |
4046 | setup_per_zone_lowmem_reserve(); | |
4047 | return 0; | |
4048 | } | |
4049 | module_init(init_per_zone_pages_min) | |
4050 | ||
4051 | /* | |
4052 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so | |
4053 | * that we can call two helper functions whenever min_free_kbytes | |
4054 | * changes. | |
4055 | */ | |
4056 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, | |
4057 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4058 | { | |
4059 | proc_dointvec(table, write, file, buffer, length, ppos); | |
3b1d92c5 MG |
4060 | if (write) |
4061 | setup_per_zone_pages_min(); | |
1da177e4 LT |
4062 | return 0; |
4063 | } | |
4064 | ||
9614634f CL |
4065 | #ifdef CONFIG_NUMA |
4066 | int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write, | |
4067 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4068 | { | |
4069 | struct zone *zone; | |
4070 | int rc; | |
4071 | ||
4072 | rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
4073 | if (rc) | |
4074 | return rc; | |
4075 | ||
4076 | for_each_zone(zone) | |
8417bba4 | 4077 | zone->min_unmapped_pages = (zone->present_pages * |
9614634f CL |
4078 | sysctl_min_unmapped_ratio) / 100; |
4079 | return 0; | |
4080 | } | |
0ff38490 CL |
4081 | |
4082 | int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write, | |
4083 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4084 | { | |
4085 | struct zone *zone; | |
4086 | int rc; | |
4087 | ||
4088 | rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
4089 | if (rc) | |
4090 | return rc; | |
4091 | ||
4092 | for_each_zone(zone) | |
4093 | zone->min_slab_pages = (zone->present_pages * | |
4094 | sysctl_min_slab_ratio) / 100; | |
4095 | return 0; | |
4096 | } | |
9614634f CL |
4097 | #endif |
4098 | ||
1da177e4 LT |
4099 | /* |
4100 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
4101 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
4102 | * whenever sysctl_lowmem_reserve_ratio changes. | |
4103 | * | |
4104 | * The reserve ratio obviously has absolutely no relation with the | |
4105 | * pages_min watermarks. The lowmem reserve ratio can only make sense | |
4106 | * if in function of the boot time zone sizes. | |
4107 | */ | |
4108 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, | |
4109 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4110 | { | |
4111 | proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
4112 | setup_per_zone_lowmem_reserve(); | |
4113 | return 0; | |
4114 | } | |
4115 | ||
8ad4b1fb RS |
4116 | /* |
4117 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
4118 | * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist | |
4119 | * can have before it gets flushed back to buddy allocator. | |
4120 | */ | |
4121 | ||
4122 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, | |
4123 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4124 | { | |
4125 | struct zone *zone; | |
4126 | unsigned int cpu; | |
4127 | int ret; | |
4128 | ||
4129 | ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
4130 | if (!write || (ret == -EINVAL)) | |
4131 | return ret; | |
4132 | for_each_zone(zone) { | |
4133 | for_each_online_cpu(cpu) { | |
4134 | unsigned long high; | |
4135 | high = zone->present_pages / percpu_pagelist_fraction; | |
4136 | setup_pagelist_highmark(zone_pcp(zone, cpu), high); | |
4137 | } | |
4138 | } | |
4139 | return 0; | |
4140 | } | |
4141 | ||
f034b5d4 | 4142 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 LT |
4143 | |
4144 | #ifdef CONFIG_NUMA | |
4145 | static int __init set_hashdist(char *str) | |
4146 | { | |
4147 | if (!str) | |
4148 | return 0; | |
4149 | hashdist = simple_strtoul(str, &str, 0); | |
4150 | return 1; | |
4151 | } | |
4152 | __setup("hashdist=", set_hashdist); | |
4153 | #endif | |
4154 | ||
4155 | /* | |
4156 | * allocate a large system hash table from bootmem | |
4157 | * - it is assumed that the hash table must contain an exact power-of-2 | |
4158 | * quantity of entries | |
4159 | * - limit is the number of hash buckets, not the total allocation size | |
4160 | */ | |
4161 | void *__init alloc_large_system_hash(const char *tablename, | |
4162 | unsigned long bucketsize, | |
4163 | unsigned long numentries, | |
4164 | int scale, | |
4165 | int flags, | |
4166 | unsigned int *_hash_shift, | |
4167 | unsigned int *_hash_mask, | |
4168 | unsigned long limit) | |
4169 | { | |
4170 | unsigned long long max = limit; | |
4171 | unsigned long log2qty, size; | |
4172 | void *table = NULL; | |
4173 | ||
4174 | /* allow the kernel cmdline to have a say */ | |
4175 | if (!numentries) { | |
4176 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 4177 | numentries = nr_kernel_pages; |
1da177e4 LT |
4178 | numentries += (1UL << (20 - PAGE_SHIFT)) - 1; |
4179 | numentries >>= 20 - PAGE_SHIFT; | |
4180 | numentries <<= 20 - PAGE_SHIFT; | |
4181 | ||
4182 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
4183 | if (scale > PAGE_SHIFT) | |
4184 | numentries >>= (scale - PAGE_SHIFT); | |
4185 | else | |
4186 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
4187 | |
4188 | /* Make sure we've got at least a 0-order allocation.. */ | |
4189 | if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
4190 | numentries = PAGE_SIZE / bucketsize; | |
1da177e4 | 4191 | } |
6e692ed3 | 4192 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
4193 | |
4194 | /* limit allocation size to 1/16 total memory by default */ | |
4195 | if (max == 0) { | |
4196 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
4197 | do_div(max, bucketsize); | |
4198 | } | |
4199 | ||
4200 | if (numentries > max) | |
4201 | numentries = max; | |
4202 | ||
f0d1b0b3 | 4203 | log2qty = ilog2(numentries); |
1da177e4 LT |
4204 | |
4205 | do { | |
4206 | size = bucketsize << log2qty; | |
4207 | if (flags & HASH_EARLY) | |
4208 | table = alloc_bootmem(size); | |
4209 | else if (hashdist) | |
4210 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
4211 | else { | |
4212 | unsigned long order; | |
4213 | for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++) | |
4214 | ; | |
4215 | table = (void*) __get_free_pages(GFP_ATOMIC, order); | |
1037b83b ED |
4216 | /* |
4217 | * If bucketsize is not a power-of-two, we may free | |
4218 | * some pages at the end of hash table. | |
4219 | */ | |
4220 | if (table) { | |
4221 | unsigned long alloc_end = (unsigned long)table + | |
4222 | (PAGE_SIZE << order); | |
4223 | unsigned long used = (unsigned long)table + | |
4224 | PAGE_ALIGN(size); | |
4225 | split_page(virt_to_page(table), order); | |
4226 | while (used < alloc_end) { | |
4227 | free_page(used); | |
4228 | used += PAGE_SIZE; | |
4229 | } | |
4230 | } | |
1da177e4 LT |
4231 | } |
4232 | } while (!table && size > PAGE_SIZE && --log2qty); | |
4233 | ||
4234 | if (!table) | |
4235 | panic("Failed to allocate %s hash table\n", tablename); | |
4236 | ||
b49ad484 | 4237 | printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n", |
1da177e4 LT |
4238 | tablename, |
4239 | (1U << log2qty), | |
f0d1b0b3 | 4240 | ilog2(size) - PAGE_SHIFT, |
1da177e4 LT |
4241 | size); |
4242 | ||
4243 | if (_hash_shift) | |
4244 | *_hash_shift = log2qty; | |
4245 | if (_hash_mask) | |
4246 | *_hash_mask = (1 << log2qty) - 1; | |
4247 | ||
4248 | return table; | |
4249 | } | |
a117e66e KH |
4250 | |
4251 | #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE | |
a117e66e KH |
4252 | struct page *pfn_to_page(unsigned long pfn) |
4253 | { | |
67de6482 | 4254 | return __pfn_to_page(pfn); |
a117e66e KH |
4255 | } |
4256 | unsigned long page_to_pfn(struct page *page) | |
4257 | { | |
67de6482 | 4258 | return __page_to_pfn(page); |
a117e66e | 4259 | } |
a117e66e KH |
4260 | EXPORT_SYMBOL(pfn_to_page); |
4261 | EXPORT_SYMBOL(page_to_pfn); | |
4262 | #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */ | |
6220ec78 | 4263 | |
835c134e MG |
4264 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
4265 | static inline unsigned long *get_pageblock_bitmap(struct zone *zone, | |
4266 | unsigned long pfn) | |
4267 | { | |
4268 | #ifdef CONFIG_SPARSEMEM | |
4269 | return __pfn_to_section(pfn)->pageblock_flags; | |
4270 | #else | |
4271 | return zone->pageblock_flags; | |
4272 | #endif /* CONFIG_SPARSEMEM */ | |
4273 | } | |
4274 | ||
4275 | static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn) | |
4276 | { | |
4277 | #ifdef CONFIG_SPARSEMEM | |
4278 | pfn &= (PAGES_PER_SECTION-1); | |
4279 | return (pfn >> (MAX_ORDER-1)) * NR_PAGEBLOCK_BITS; | |
4280 | #else | |
4281 | pfn = pfn - zone->zone_start_pfn; | |
4282 | return (pfn >> (MAX_ORDER-1)) * NR_PAGEBLOCK_BITS; | |
4283 | #endif /* CONFIG_SPARSEMEM */ | |
4284 | } | |
4285 | ||
4286 | /** | |
4287 | * get_pageblock_flags_group - Return the requested group of flags for the MAX_ORDER_NR_PAGES block of pages | |
4288 | * @page: The page within the block of interest | |
4289 | * @start_bitidx: The first bit of interest to retrieve | |
4290 | * @end_bitidx: The last bit of interest | |
4291 | * returns pageblock_bits flags | |
4292 | */ | |
4293 | unsigned long get_pageblock_flags_group(struct page *page, | |
4294 | int start_bitidx, int end_bitidx) | |
4295 | { | |
4296 | struct zone *zone; | |
4297 | unsigned long *bitmap; | |
4298 | unsigned long pfn, bitidx; | |
4299 | unsigned long flags = 0; | |
4300 | unsigned long value = 1; | |
4301 | ||
4302 | zone = page_zone(page); | |
4303 | pfn = page_to_pfn(page); | |
4304 | bitmap = get_pageblock_bitmap(zone, pfn); | |
4305 | bitidx = pfn_to_bitidx(zone, pfn); | |
4306 | ||
4307 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
4308 | if (test_bit(bitidx + start_bitidx, bitmap)) | |
4309 | flags |= value; | |
6220ec78 | 4310 | |
835c134e MG |
4311 | return flags; |
4312 | } | |
4313 | ||
4314 | /** | |
4315 | * set_pageblock_flags_group - Set the requested group of flags for a MAX_ORDER_NR_PAGES block of pages | |
4316 | * @page: The page within the block of interest | |
4317 | * @start_bitidx: The first bit of interest | |
4318 | * @end_bitidx: The last bit of interest | |
4319 | * @flags: The flags to set | |
4320 | */ | |
4321 | void set_pageblock_flags_group(struct page *page, unsigned long flags, | |
4322 | int start_bitidx, int end_bitidx) | |
4323 | { | |
4324 | struct zone *zone; | |
4325 | unsigned long *bitmap; | |
4326 | unsigned long pfn, bitidx; | |
4327 | unsigned long value = 1; | |
4328 | ||
4329 | zone = page_zone(page); | |
4330 | pfn = page_to_pfn(page); | |
4331 | bitmap = get_pageblock_bitmap(zone, pfn); | |
4332 | bitidx = pfn_to_bitidx(zone, pfn); | |
4333 | ||
4334 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
4335 | if (flags & value) | |
4336 | __set_bit(bitidx + start_bitidx, bitmap); | |
4337 | else | |
4338 | __clear_bit(bitidx + start_bitidx, bitmap); | |
4339 | } |