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 |
9ef9acb0 MG |
162 | int page_group_by_mobility_disabled __read_mostly; |
163 | ||
b2a0ac88 MG |
164 | static inline int get_pageblock_migratetype(struct page *page) |
165 | { | |
9ef9acb0 MG |
166 | if (unlikely(page_group_by_mobility_disabled)) |
167 | return MIGRATE_UNMOVABLE; | |
168 | ||
b2a0ac88 MG |
169 | return get_pageblock_flags_group(page, PB_migrate, PB_migrate_end); |
170 | } | |
171 | ||
172 | static void set_pageblock_migratetype(struct page *page, int migratetype) | |
173 | { | |
174 | set_pageblock_flags_group(page, (unsigned long)migratetype, | |
175 | PB_migrate, PB_migrate_end); | |
176 | } | |
177 | ||
e010487d | 178 | static inline int allocflags_to_migratetype(gfp_t gfp_flags, int order) |
b2a0ac88 | 179 | { |
e12ba74d MG |
180 | WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK); |
181 | ||
9ef9acb0 MG |
182 | if (unlikely(page_group_by_mobility_disabled)) |
183 | return MIGRATE_UNMOVABLE; | |
184 | ||
e010487d MG |
185 | /* Cluster high-order atomic allocations together */ |
186 | if (unlikely(order > 0) && | |
187 | (!(gfp_flags & __GFP_WAIT) || in_interrupt())) | |
188 | return MIGRATE_HIGHATOMIC; | |
189 | ||
190 | /* Cluster based on mobility */ | |
e12ba74d MG |
191 | return (((gfp_flags & __GFP_MOVABLE) != 0) << 1) | |
192 | ((gfp_flags & __GFP_RECLAIMABLE) != 0); | |
b2a0ac88 MG |
193 | } |
194 | ||
b92a6edd MG |
195 | #else |
196 | static inline int get_pageblock_migratetype(struct page *page) | |
197 | { | |
198 | return MIGRATE_UNMOVABLE; | |
199 | } | |
200 | ||
201 | static void set_pageblock_migratetype(struct page *page, int migratetype) | |
202 | { | |
203 | } | |
204 | ||
e010487d | 205 | static inline int allocflags_to_migratetype(gfp_t gfp_flags, int order) |
b92a6edd MG |
206 | { |
207 | return MIGRATE_UNMOVABLE; | |
208 | } | |
209 | #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */ | |
210 | ||
13e7444b | 211 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 212 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 213 | { |
bdc8cb98 DH |
214 | int ret = 0; |
215 | unsigned seq; | |
216 | unsigned long pfn = page_to_pfn(page); | |
c6a57e19 | 217 | |
bdc8cb98 DH |
218 | do { |
219 | seq = zone_span_seqbegin(zone); | |
220 | if (pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
221 | ret = 1; | |
222 | else if (pfn < zone->zone_start_pfn) | |
223 | ret = 1; | |
224 | } while (zone_span_seqretry(zone, seq)); | |
225 | ||
226 | return ret; | |
c6a57e19 DH |
227 | } |
228 | ||
229 | static int page_is_consistent(struct zone *zone, struct page *page) | |
230 | { | |
14e07298 | 231 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 232 | return 0; |
1da177e4 | 233 | if (zone != page_zone(page)) |
c6a57e19 DH |
234 | return 0; |
235 | ||
236 | return 1; | |
237 | } | |
238 | /* | |
239 | * Temporary debugging check for pages not lying within a given zone. | |
240 | */ | |
241 | static int bad_range(struct zone *zone, struct page *page) | |
242 | { | |
243 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 244 | return 1; |
c6a57e19 DH |
245 | if (!page_is_consistent(zone, page)) |
246 | return 1; | |
247 | ||
1da177e4 LT |
248 | return 0; |
249 | } | |
13e7444b NP |
250 | #else |
251 | static inline int bad_range(struct zone *zone, struct page *page) | |
252 | { | |
253 | return 0; | |
254 | } | |
255 | #endif | |
256 | ||
224abf92 | 257 | static void bad_page(struct page *page) |
1da177e4 | 258 | { |
224abf92 | 259 | printk(KERN_EMERG "Bad page state in process '%s'\n" |
7365f3d1 HD |
260 | KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n" |
261 | KERN_EMERG "Trying to fix it up, but a reboot is needed\n" | |
262 | KERN_EMERG "Backtrace:\n", | |
224abf92 NP |
263 | current->comm, page, (int)(2*sizeof(unsigned long)), |
264 | (unsigned long)page->flags, page->mapping, | |
265 | page_mapcount(page), page_count(page)); | |
1da177e4 | 266 | dump_stack(); |
334795ec HD |
267 | page->flags &= ~(1 << PG_lru | |
268 | 1 << PG_private | | |
1da177e4 | 269 | 1 << PG_locked | |
1da177e4 LT |
270 | 1 << PG_active | |
271 | 1 << PG_dirty | | |
334795ec HD |
272 | 1 << PG_reclaim | |
273 | 1 << PG_slab | | |
1da177e4 | 274 | 1 << PG_swapcache | |
676165a8 NP |
275 | 1 << PG_writeback | |
276 | 1 << PG_buddy ); | |
1da177e4 LT |
277 | set_page_count(page, 0); |
278 | reset_page_mapcount(page); | |
279 | page->mapping = NULL; | |
9f158333 | 280 | add_taint(TAINT_BAD_PAGE); |
1da177e4 LT |
281 | } |
282 | ||
1da177e4 LT |
283 | /* |
284 | * Higher-order pages are called "compound pages". They are structured thusly: | |
285 | * | |
286 | * The first PAGE_SIZE page is called the "head page". | |
287 | * | |
288 | * The remaining PAGE_SIZE pages are called "tail pages". | |
289 | * | |
290 | * All pages have PG_compound set. All pages have their ->private pointing at | |
291 | * the head page (even the head page has this). | |
292 | * | |
41d78ba5 HD |
293 | * The first tail page's ->lru.next holds the address of the compound page's |
294 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
295 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 296 | */ |
d98c7a09 HD |
297 | |
298 | static void free_compound_page(struct page *page) | |
299 | { | |
d85f3385 | 300 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
301 | } |
302 | ||
1da177e4 LT |
303 | static void prep_compound_page(struct page *page, unsigned long order) |
304 | { | |
305 | int i; | |
306 | int nr_pages = 1 << order; | |
307 | ||
33f2ef89 | 308 | set_compound_page_dtor(page, free_compound_page); |
d85f3385 | 309 | set_compound_order(page, order); |
6d777953 | 310 | __SetPageHead(page); |
d85f3385 | 311 | for (i = 1; i < nr_pages; i++) { |
1da177e4 LT |
312 | struct page *p = page + i; |
313 | ||
d85f3385 | 314 | __SetPageTail(p); |
d85f3385 | 315 | p->first_page = page; |
1da177e4 LT |
316 | } |
317 | } | |
318 | ||
319 | static void destroy_compound_page(struct page *page, unsigned long order) | |
320 | { | |
321 | int i; | |
322 | int nr_pages = 1 << order; | |
323 | ||
d85f3385 | 324 | if (unlikely(compound_order(page) != order)) |
224abf92 | 325 | bad_page(page); |
1da177e4 | 326 | |
6d777953 | 327 | if (unlikely(!PageHead(page))) |
d85f3385 | 328 | bad_page(page); |
6d777953 | 329 | __ClearPageHead(page); |
d85f3385 | 330 | for (i = 1; i < nr_pages; i++) { |
1da177e4 LT |
331 | struct page *p = page + i; |
332 | ||
6d777953 | 333 | if (unlikely(!PageTail(p) | |
d85f3385 | 334 | (p->first_page != page))) |
224abf92 | 335 | bad_page(page); |
d85f3385 | 336 | __ClearPageTail(p); |
1da177e4 LT |
337 | } |
338 | } | |
1da177e4 | 339 | |
17cf4406 NP |
340 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) |
341 | { | |
342 | int i; | |
343 | ||
725d704e | 344 | VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM); |
6626c5d5 AM |
345 | /* |
346 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO | |
347 | * and __GFP_HIGHMEM from hard or soft interrupt context. | |
348 | */ | |
725d704e | 349 | VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); |
17cf4406 NP |
350 | for (i = 0; i < (1 << order); i++) |
351 | clear_highpage(page + i); | |
352 | } | |
353 | ||
1da177e4 LT |
354 | /* |
355 | * function for dealing with page's order in buddy system. | |
356 | * zone->lock is already acquired when we use these. | |
357 | * So, we don't need atomic page->flags operations here. | |
358 | */ | |
6aa3001b AM |
359 | static inline unsigned long page_order(struct page *page) |
360 | { | |
4c21e2f2 | 361 | return page_private(page); |
1da177e4 LT |
362 | } |
363 | ||
6aa3001b AM |
364 | static inline void set_page_order(struct page *page, int order) |
365 | { | |
4c21e2f2 | 366 | set_page_private(page, order); |
676165a8 | 367 | __SetPageBuddy(page); |
1da177e4 LT |
368 | } |
369 | ||
370 | static inline void rmv_page_order(struct page *page) | |
371 | { | |
676165a8 | 372 | __ClearPageBuddy(page); |
4c21e2f2 | 373 | set_page_private(page, 0); |
1da177e4 LT |
374 | } |
375 | ||
376 | /* | |
377 | * Locate the struct page for both the matching buddy in our | |
378 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
379 | * | |
380 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
381 | * the following equation: | |
382 | * B2 = B1 ^ (1 << O) | |
383 | * For example, if the starting buddy (buddy2) is #8 its order | |
384 | * 1 buddy is #10: | |
385 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
386 | * | |
387 | * 2) Any buddy B will have an order O+1 parent P which | |
388 | * satisfies the following equation: | |
389 | * P = B & ~(1 << O) | |
390 | * | |
d6e05edc | 391 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER |
1da177e4 LT |
392 | */ |
393 | static inline struct page * | |
394 | __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order) | |
395 | { | |
396 | unsigned long buddy_idx = page_idx ^ (1 << order); | |
397 | ||
398 | return page + (buddy_idx - page_idx); | |
399 | } | |
400 | ||
401 | static inline unsigned long | |
402 | __find_combined_index(unsigned long page_idx, unsigned int order) | |
403 | { | |
404 | return (page_idx & ~(1 << order)); | |
405 | } | |
406 | ||
407 | /* | |
408 | * This function checks whether a page is free && is the buddy | |
409 | * we can do coalesce a page and its buddy if | |
13e7444b | 410 | * (a) the buddy is not in a hole && |
676165a8 | 411 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
412 | * (c) a page and its buddy have the same order && |
413 | * (d) a page and its buddy are in the same zone. | |
676165a8 NP |
414 | * |
415 | * For recording whether a page is in the buddy system, we use PG_buddy. | |
416 | * Setting, clearing, and testing PG_buddy is serialized by zone->lock. | |
1da177e4 | 417 | * |
676165a8 | 418 | * For recording page's order, we use page_private(page). |
1da177e4 | 419 | */ |
cb2b95e1 AW |
420 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
421 | int order) | |
1da177e4 | 422 | { |
14e07298 | 423 | if (!pfn_valid_within(page_to_pfn(buddy))) |
13e7444b | 424 | return 0; |
13e7444b | 425 | |
cb2b95e1 AW |
426 | if (page_zone_id(page) != page_zone_id(buddy)) |
427 | return 0; | |
428 | ||
429 | if (PageBuddy(buddy) && page_order(buddy) == order) { | |
430 | BUG_ON(page_count(buddy) != 0); | |
6aa3001b | 431 | return 1; |
676165a8 | 432 | } |
6aa3001b | 433 | return 0; |
1da177e4 LT |
434 | } |
435 | ||
436 | /* | |
437 | * Freeing function for a buddy system allocator. | |
438 | * | |
439 | * The concept of a buddy system is to maintain direct-mapped table | |
440 | * (containing bit values) for memory blocks of various "orders". | |
441 | * The bottom level table contains the map for the smallest allocatable | |
442 | * units of memory (here, pages), and each level above it describes | |
443 | * pairs of units from the levels below, hence, "buddies". | |
444 | * At a high level, all that happens here is marking the table entry | |
445 | * at the bottom level available, and propagating the changes upward | |
446 | * as necessary, plus some accounting needed to play nicely with other | |
447 | * parts of the VM system. | |
448 | * At each level, we keep a list of pages, which are heads of continuous | |
676165a8 | 449 | * free pages of length of (1 << order) and marked with PG_buddy. Page's |
4c21e2f2 | 450 | * order is recorded in page_private(page) field. |
1da177e4 LT |
451 | * So when we are allocating or freeing one, we can derive the state of the |
452 | * other. That is, if we allocate a small block, and both were | |
453 | * free, the remainder of the region must be split into blocks. | |
454 | * If a block is freed, and its buddy is also free, then this | |
455 | * triggers coalescing into a block of larger size. | |
456 | * | |
457 | * -- wli | |
458 | */ | |
459 | ||
48db57f8 | 460 | static inline void __free_one_page(struct page *page, |
1da177e4 LT |
461 | struct zone *zone, unsigned int order) |
462 | { | |
463 | unsigned long page_idx; | |
464 | int order_size = 1 << order; | |
b2a0ac88 | 465 | int migratetype = get_pageblock_migratetype(page); |
1da177e4 | 466 | |
224abf92 | 467 | if (unlikely(PageCompound(page))) |
1da177e4 LT |
468 | destroy_compound_page(page, order); |
469 | ||
470 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); | |
471 | ||
725d704e NP |
472 | VM_BUG_ON(page_idx & (order_size - 1)); |
473 | VM_BUG_ON(bad_range(zone, page)); | |
1da177e4 | 474 | |
d23ad423 | 475 | __mod_zone_page_state(zone, NR_FREE_PAGES, order_size); |
1da177e4 LT |
476 | while (order < MAX_ORDER-1) { |
477 | unsigned long combined_idx; | |
1da177e4 LT |
478 | struct page *buddy; |
479 | ||
1da177e4 | 480 | buddy = __page_find_buddy(page, page_idx, order); |
cb2b95e1 | 481 | if (!page_is_buddy(page, buddy, order)) |
1da177e4 | 482 | break; /* Move the buddy up one level. */ |
13e7444b | 483 | |
1da177e4 | 484 | list_del(&buddy->lru); |
b2a0ac88 | 485 | zone->free_area[order].nr_free--; |
1da177e4 | 486 | rmv_page_order(buddy); |
13e7444b | 487 | combined_idx = __find_combined_index(page_idx, order); |
1da177e4 LT |
488 | page = page + (combined_idx - page_idx); |
489 | page_idx = combined_idx; | |
490 | order++; | |
491 | } | |
492 | set_page_order(page, order); | |
b2a0ac88 MG |
493 | list_add(&page->lru, |
494 | &zone->free_area[order].free_list[migratetype]); | |
1da177e4 LT |
495 | zone->free_area[order].nr_free++; |
496 | } | |
497 | ||
224abf92 | 498 | static inline int free_pages_check(struct page *page) |
1da177e4 | 499 | { |
92be2e33 NP |
500 | if (unlikely(page_mapcount(page) | |
501 | (page->mapping != NULL) | | |
502 | (page_count(page) != 0) | | |
1da177e4 LT |
503 | (page->flags & ( |
504 | 1 << PG_lru | | |
505 | 1 << PG_private | | |
506 | 1 << PG_locked | | |
507 | 1 << PG_active | | |
1da177e4 LT |
508 | 1 << PG_slab | |
509 | 1 << PG_swapcache | | |
b5810039 | 510 | 1 << PG_writeback | |
676165a8 NP |
511 | 1 << PG_reserved | |
512 | 1 << PG_buddy )))) | |
224abf92 | 513 | bad_page(page); |
1da177e4 | 514 | if (PageDirty(page)) |
242e5468 | 515 | __ClearPageDirty(page); |
689bcebf HD |
516 | /* |
517 | * For now, we report if PG_reserved was found set, but do not | |
518 | * clear it, and do not free the page. But we shall soon need | |
519 | * to do more, for when the ZERO_PAGE count wraps negative. | |
520 | */ | |
521 | return PageReserved(page); | |
1da177e4 LT |
522 | } |
523 | ||
524 | /* | |
525 | * Frees a list of pages. | |
526 | * Assumes all pages on list are in same zone, and of same order. | |
207f36ee | 527 | * count is the number of pages to free. |
1da177e4 LT |
528 | * |
529 | * If the zone was previously in an "all pages pinned" state then look to | |
530 | * see if this freeing clears that state. | |
531 | * | |
532 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
533 | * pinned" detection logic. | |
534 | */ | |
48db57f8 NP |
535 | static void free_pages_bulk(struct zone *zone, int count, |
536 | struct list_head *list, int order) | |
1da177e4 | 537 | { |
c54ad30c | 538 | spin_lock(&zone->lock); |
1da177e4 LT |
539 | zone->all_unreclaimable = 0; |
540 | zone->pages_scanned = 0; | |
48db57f8 NP |
541 | while (count--) { |
542 | struct page *page; | |
543 | ||
725d704e | 544 | VM_BUG_ON(list_empty(list)); |
1da177e4 | 545 | page = list_entry(list->prev, struct page, lru); |
48db57f8 | 546 | /* have to delete it as __free_one_page list manipulates */ |
1da177e4 | 547 | list_del(&page->lru); |
48db57f8 | 548 | __free_one_page(page, zone, order); |
1da177e4 | 549 | } |
c54ad30c | 550 | spin_unlock(&zone->lock); |
1da177e4 LT |
551 | } |
552 | ||
48db57f8 | 553 | static void free_one_page(struct zone *zone, struct page *page, int order) |
1da177e4 | 554 | { |
006d22d9 CL |
555 | spin_lock(&zone->lock); |
556 | zone->all_unreclaimable = 0; | |
557 | zone->pages_scanned = 0; | |
0798e519 | 558 | __free_one_page(page, zone, order); |
006d22d9 | 559 | spin_unlock(&zone->lock); |
48db57f8 NP |
560 | } |
561 | ||
562 | static void __free_pages_ok(struct page *page, unsigned int order) | |
563 | { | |
564 | unsigned long flags; | |
1da177e4 | 565 | int i; |
689bcebf | 566 | int reserved = 0; |
1da177e4 | 567 | |
1da177e4 | 568 | for (i = 0 ; i < (1 << order) ; ++i) |
224abf92 | 569 | reserved += free_pages_check(page + i); |
689bcebf HD |
570 | if (reserved) |
571 | return; | |
572 | ||
9858db50 NP |
573 | if (!PageHighMem(page)) |
574 | debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order); | |
dafb1367 | 575 | arch_free_page(page, order); |
48db57f8 | 576 | kernel_map_pages(page, 1 << order, 0); |
dafb1367 | 577 | |
c54ad30c | 578 | local_irq_save(flags); |
f8891e5e | 579 | __count_vm_events(PGFREE, 1 << order); |
48db57f8 | 580 | free_one_page(page_zone(page), page, order); |
c54ad30c | 581 | local_irq_restore(flags); |
1da177e4 LT |
582 | } |
583 | ||
a226f6c8 DH |
584 | /* |
585 | * permit the bootmem allocator to evade page validation on high-order frees | |
586 | */ | |
587 | void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order) | |
588 | { | |
589 | if (order == 0) { | |
590 | __ClearPageReserved(page); | |
591 | set_page_count(page, 0); | |
7835e98b | 592 | set_page_refcounted(page); |
545b1ea9 | 593 | __free_page(page); |
a226f6c8 | 594 | } else { |
a226f6c8 DH |
595 | int loop; |
596 | ||
545b1ea9 | 597 | prefetchw(page); |
a226f6c8 DH |
598 | for (loop = 0; loop < BITS_PER_LONG; loop++) { |
599 | struct page *p = &page[loop]; | |
600 | ||
545b1ea9 NP |
601 | if (loop + 1 < BITS_PER_LONG) |
602 | prefetchw(p + 1); | |
a226f6c8 DH |
603 | __ClearPageReserved(p); |
604 | set_page_count(p, 0); | |
605 | } | |
606 | ||
7835e98b | 607 | set_page_refcounted(page); |
545b1ea9 | 608 | __free_pages(page, order); |
a226f6c8 DH |
609 | } |
610 | } | |
611 | ||
1da177e4 LT |
612 | |
613 | /* | |
614 | * The order of subdivision here is critical for the IO subsystem. | |
615 | * Please do not alter this order without good reasons and regression | |
616 | * testing. Specifically, as large blocks of memory are subdivided, | |
617 | * the order in which smaller blocks are delivered depends on the order | |
618 | * they're subdivided in this function. This is the primary factor | |
619 | * influencing the order in which pages are delivered to the IO | |
620 | * subsystem according to empirical testing, and this is also justified | |
621 | * by considering the behavior of a buddy system containing a single | |
622 | * large block of memory acted on by a series of small allocations. | |
623 | * This behavior is a critical factor in sglist merging's success. | |
624 | * | |
625 | * -- wli | |
626 | */ | |
085cc7d5 | 627 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
628 | int low, int high, struct free_area *area, |
629 | int migratetype) | |
1da177e4 LT |
630 | { |
631 | unsigned long size = 1 << high; | |
632 | ||
633 | while (high > low) { | |
634 | area--; | |
635 | high--; | |
636 | size >>= 1; | |
725d704e | 637 | VM_BUG_ON(bad_range(zone, &page[size])); |
b2a0ac88 | 638 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
639 | area->nr_free++; |
640 | set_page_order(&page[size], high); | |
641 | } | |
1da177e4 LT |
642 | } |
643 | ||
1da177e4 LT |
644 | /* |
645 | * This page is about to be returned from the page allocator | |
646 | */ | |
17cf4406 | 647 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) |
1da177e4 | 648 | { |
92be2e33 NP |
649 | if (unlikely(page_mapcount(page) | |
650 | (page->mapping != NULL) | | |
651 | (page_count(page) != 0) | | |
334795ec HD |
652 | (page->flags & ( |
653 | 1 << PG_lru | | |
1da177e4 LT |
654 | 1 << PG_private | |
655 | 1 << PG_locked | | |
1da177e4 LT |
656 | 1 << PG_active | |
657 | 1 << PG_dirty | | |
334795ec | 658 | 1 << PG_slab | |
1da177e4 | 659 | 1 << PG_swapcache | |
b5810039 | 660 | 1 << PG_writeback | |
676165a8 NP |
661 | 1 << PG_reserved | |
662 | 1 << PG_buddy )))) | |
224abf92 | 663 | bad_page(page); |
1da177e4 | 664 | |
689bcebf HD |
665 | /* |
666 | * For now, we report if PG_reserved was found set, but do not | |
667 | * clear it, and do not allocate the page: as a safety net. | |
668 | */ | |
669 | if (PageReserved(page)) | |
670 | return 1; | |
671 | ||
d77c2d7c | 672 | page->flags &= ~(1 << PG_uptodate | 1 << PG_error | 1 << PG_readahead | |
1da177e4 | 673 | 1 << PG_referenced | 1 << PG_arch_1 | |
5409bae0 | 674 | 1 << PG_owner_priv_1 | 1 << PG_mappedtodisk); |
4c21e2f2 | 675 | set_page_private(page, 0); |
7835e98b | 676 | set_page_refcounted(page); |
cc102509 NP |
677 | |
678 | arch_alloc_page(page, order); | |
1da177e4 | 679 | kernel_map_pages(page, 1 << order, 1); |
17cf4406 NP |
680 | |
681 | if (gfp_flags & __GFP_ZERO) | |
682 | prep_zero_page(page, order, gfp_flags); | |
683 | ||
684 | if (order && (gfp_flags & __GFP_COMP)) | |
685 | prep_compound_page(page, order); | |
686 | ||
689bcebf | 687 | return 0; |
1da177e4 LT |
688 | } |
689 | ||
b92a6edd | 690 | #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY |
b2a0ac88 MG |
691 | /* |
692 | * This array describes the order lists are fallen back to when | |
693 | * the free lists for the desirable migrate type are depleted | |
694 | */ | |
695 | static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = { | |
e010487d MG |
696 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_HIGHATOMIC }, |
697 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_HIGHATOMIC }, | |
698 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE,MIGRATE_HIGHATOMIC }, | |
699 | [MIGRATE_HIGHATOMIC] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE,MIGRATE_MOVABLE}, | |
b2a0ac88 MG |
700 | }; |
701 | ||
c361be55 MG |
702 | /* |
703 | * Move the free pages in a range to the free lists of the requested type. | |
704 | * Note that start_page and end_pages are not aligned in a MAX_ORDER_NR_PAGES | |
705 | * boundary. If alignment is required, use move_freepages_block() | |
706 | */ | |
707 | int move_freepages(struct zone *zone, | |
708 | struct page *start_page, struct page *end_page, | |
709 | int migratetype) | |
710 | { | |
711 | struct page *page; | |
712 | unsigned long order; | |
713 | int blocks_moved = 0; | |
714 | ||
715 | #ifndef CONFIG_HOLES_IN_ZONE | |
716 | /* | |
717 | * page_zone is not safe to call in this context when | |
718 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
719 | * anyway as we check zone boundaries in move_freepages_block(). | |
720 | * Remove at a later date when no bug reports exist related to | |
721 | * CONFIG_PAGE_GROUP_BY_MOBILITY | |
722 | */ | |
723 | BUG_ON(page_zone(start_page) != page_zone(end_page)); | |
724 | #endif | |
725 | ||
726 | for (page = start_page; page <= end_page;) { | |
727 | if (!pfn_valid_within(page_to_pfn(page))) { | |
728 | page++; | |
729 | continue; | |
730 | } | |
731 | ||
732 | if (!PageBuddy(page)) { | |
733 | page++; | |
734 | continue; | |
735 | } | |
736 | ||
737 | order = page_order(page); | |
738 | list_del(&page->lru); | |
739 | list_add(&page->lru, | |
740 | &zone->free_area[order].free_list[migratetype]); | |
741 | page += 1 << order; | |
742 | blocks_moved++; | |
743 | } | |
744 | ||
745 | return blocks_moved; | |
746 | } | |
747 | ||
748 | int move_freepages_block(struct zone *zone, struct page *page, int migratetype) | |
749 | { | |
750 | unsigned long start_pfn, end_pfn; | |
751 | struct page *start_page, *end_page; | |
752 | ||
753 | start_pfn = page_to_pfn(page); | |
754 | start_pfn = start_pfn & ~(MAX_ORDER_NR_PAGES-1); | |
755 | start_page = pfn_to_page(start_pfn); | |
756 | end_page = start_page + MAX_ORDER_NR_PAGES - 1; | |
757 | end_pfn = start_pfn + MAX_ORDER_NR_PAGES - 1; | |
758 | ||
759 | /* Do not cross zone boundaries */ | |
760 | if (start_pfn < zone->zone_start_pfn) | |
761 | start_page = page; | |
762 | if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
763 | return 0; | |
764 | ||
765 | return move_freepages(zone, start_page, end_page, migratetype); | |
766 | } | |
767 | ||
5adc5be7 MG |
768 | /* Return the page with the lowest PFN in the list */ |
769 | static struct page *min_page(struct list_head *list) | |
770 | { | |
771 | unsigned long min_pfn = -1UL; | |
772 | struct page *min_page = NULL, *page;; | |
773 | ||
774 | list_for_each_entry(page, list, lru) { | |
775 | unsigned long pfn = page_to_pfn(page); | |
776 | if (pfn < min_pfn) { | |
777 | min_pfn = pfn; | |
778 | min_page = page; | |
779 | } | |
780 | } | |
781 | ||
782 | return min_page; | |
783 | } | |
784 | ||
b2a0ac88 MG |
785 | /* Remove an element from the buddy allocator from the fallback list */ |
786 | static struct page *__rmqueue_fallback(struct zone *zone, int order, | |
787 | int start_migratetype) | |
788 | { | |
789 | struct free_area * area; | |
790 | int current_order; | |
791 | struct page *page; | |
792 | int migratetype, i; | |
e010487d | 793 | int nonatomic_fallback_atomic = 0; |
b2a0ac88 | 794 | |
e010487d | 795 | retry: |
b2a0ac88 MG |
796 | /* Find the largest possible block of pages in the other list */ |
797 | for (current_order = MAX_ORDER-1; current_order >= order; | |
798 | --current_order) { | |
799 | for (i = 0; i < MIGRATE_TYPES - 1; i++) { | |
800 | migratetype = fallbacks[start_migratetype][i]; | |
801 | ||
e010487d MG |
802 | /* |
803 | * Make it hard to fallback to blocks used for | |
804 | * high-order atomic allocations | |
805 | */ | |
806 | if (migratetype == MIGRATE_HIGHATOMIC && | |
807 | start_migratetype != MIGRATE_UNMOVABLE && | |
808 | !nonatomic_fallback_atomic) | |
809 | continue; | |
810 | ||
b2a0ac88 MG |
811 | area = &(zone->free_area[current_order]); |
812 | if (list_empty(&area->free_list[migratetype])) | |
813 | continue; | |
814 | ||
5adc5be7 | 815 | /* Bias kernel allocations towards low pfns */ |
b2a0ac88 MG |
816 | page = list_entry(area->free_list[migratetype].next, |
817 | struct page, lru); | |
5adc5be7 MG |
818 | if (unlikely(start_migratetype != MIGRATE_MOVABLE)) |
819 | page = min_page(&area->free_list[migratetype]); | |
b2a0ac88 MG |
820 | area->nr_free--; |
821 | ||
822 | /* | |
c361be55 MG |
823 | * If breaking a large block of pages, move all free |
824 | * pages to the preferred allocation list | |
b2a0ac88 | 825 | */ |
c361be55 | 826 | if (unlikely(current_order >= MAX_ORDER / 2)) { |
b2a0ac88 | 827 | migratetype = start_migratetype; |
c361be55 MG |
828 | move_freepages_block(zone, page, migratetype); |
829 | } | |
b2a0ac88 MG |
830 | |
831 | /* Remove the page from the freelists */ | |
832 | list_del(&page->lru); | |
833 | rmv_page_order(page); | |
834 | __mod_zone_page_state(zone, NR_FREE_PAGES, | |
835 | -(1UL << order)); | |
836 | ||
837 | if (current_order == MAX_ORDER - 1) | |
838 | set_pageblock_migratetype(page, | |
839 | start_migratetype); | |
840 | ||
841 | expand(zone, page, order, current_order, area, migratetype); | |
842 | return page; | |
843 | } | |
844 | } | |
845 | ||
e010487d MG |
846 | /* Allow fallback to high-order atomic blocks if memory is that low */ |
847 | if (!nonatomic_fallback_atomic) { | |
848 | nonatomic_fallback_atomic = 1; | |
849 | goto retry; | |
850 | } | |
851 | ||
b2a0ac88 MG |
852 | return NULL; |
853 | } | |
b92a6edd MG |
854 | #else |
855 | static struct page *__rmqueue_fallback(struct zone *zone, int order, | |
856 | int start_migratetype) | |
857 | { | |
858 | return NULL; | |
859 | } | |
860 | #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */ | |
b2a0ac88 | 861 | |
1da177e4 LT |
862 | /* |
863 | * Do the hard work of removing an element from the buddy allocator. | |
864 | * Call me with the zone->lock already held. | |
865 | */ | |
b2a0ac88 MG |
866 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
867 | int migratetype) | |
1da177e4 LT |
868 | { |
869 | struct free_area * area; | |
870 | unsigned int current_order; | |
871 | struct page *page; | |
872 | ||
b2a0ac88 | 873 | /* Find a page of the appropriate size in the preferred list */ |
1da177e4 | 874 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { |
b2a0ac88 MG |
875 | area = &(zone->free_area[current_order]); |
876 | if (list_empty(&area->free_list[migratetype])) | |
1da177e4 LT |
877 | continue; |
878 | ||
b2a0ac88 MG |
879 | page = list_entry(area->free_list[migratetype].next, |
880 | struct page, lru); | |
1da177e4 LT |
881 | list_del(&page->lru); |
882 | rmv_page_order(page); | |
883 | area->nr_free--; | |
d23ad423 | 884 | __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order)); |
b2a0ac88 MG |
885 | expand(zone, page, order, current_order, area, migratetype); |
886 | goto got_page; | |
1da177e4 LT |
887 | } |
888 | ||
b2a0ac88 MG |
889 | page = __rmqueue_fallback(zone, order, migratetype); |
890 | ||
891 | got_page: | |
892 | ||
893 | return page; | |
1da177e4 LT |
894 | } |
895 | ||
896 | /* | |
897 | * Obtain a specified number of elements from the buddy allocator, all under | |
898 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
899 | * Returns the number of new pages which were placed at *list. | |
900 | */ | |
901 | static int rmqueue_bulk(struct zone *zone, unsigned int order, | |
b2a0ac88 MG |
902 | unsigned long count, struct list_head *list, |
903 | int migratetype) | |
1da177e4 | 904 | { |
1da177e4 | 905 | int i; |
1da177e4 | 906 | |
c54ad30c | 907 | spin_lock(&zone->lock); |
1da177e4 | 908 | for (i = 0; i < count; ++i) { |
b2a0ac88 | 909 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 910 | if (unlikely(page == NULL)) |
1da177e4 | 911 | break; |
535131e6 MG |
912 | list_add(&page->lru, list); |
913 | set_page_private(page, migratetype); | |
1da177e4 | 914 | } |
c54ad30c | 915 | spin_unlock(&zone->lock); |
085cc7d5 | 916 | return i; |
1da177e4 LT |
917 | } |
918 | ||
4ae7c039 | 919 | #ifdef CONFIG_NUMA |
8fce4d8e | 920 | /* |
4037d452 CL |
921 | * Called from the vmstat counter updater to drain pagesets of this |
922 | * currently executing processor on remote nodes after they have | |
923 | * expired. | |
924 | * | |
879336c3 CL |
925 | * Note that this function must be called with the thread pinned to |
926 | * a single processor. | |
8fce4d8e | 927 | */ |
4037d452 | 928 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 929 | { |
4ae7c039 | 930 | unsigned long flags; |
4037d452 | 931 | int to_drain; |
4ae7c039 | 932 | |
4037d452 CL |
933 | local_irq_save(flags); |
934 | if (pcp->count >= pcp->batch) | |
935 | to_drain = pcp->batch; | |
936 | else | |
937 | to_drain = pcp->count; | |
938 | free_pages_bulk(zone, to_drain, &pcp->list, 0); | |
939 | pcp->count -= to_drain; | |
940 | local_irq_restore(flags); | |
4ae7c039 CL |
941 | } |
942 | #endif | |
943 | ||
1da177e4 LT |
944 | static void __drain_pages(unsigned int cpu) |
945 | { | |
c54ad30c | 946 | unsigned long flags; |
1da177e4 LT |
947 | struct zone *zone; |
948 | int i; | |
949 | ||
950 | for_each_zone(zone) { | |
951 | struct per_cpu_pageset *pset; | |
952 | ||
f2e12bb2 CL |
953 | if (!populated_zone(zone)) |
954 | continue; | |
955 | ||
e7c8d5c9 | 956 | pset = zone_pcp(zone, cpu); |
1da177e4 LT |
957 | for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) { |
958 | struct per_cpu_pages *pcp; | |
959 | ||
960 | pcp = &pset->pcp[i]; | |
c54ad30c | 961 | local_irq_save(flags); |
48db57f8 NP |
962 | free_pages_bulk(zone, pcp->count, &pcp->list, 0); |
963 | pcp->count = 0; | |
c54ad30c | 964 | local_irq_restore(flags); |
1da177e4 LT |
965 | } |
966 | } | |
967 | } | |
1da177e4 | 968 | |
296699de | 969 | #ifdef CONFIG_HIBERNATION |
1da177e4 LT |
970 | |
971 | void mark_free_pages(struct zone *zone) | |
972 | { | |
f623f0db RW |
973 | unsigned long pfn, max_zone_pfn; |
974 | unsigned long flags; | |
b2a0ac88 | 975 | int order, t; |
1da177e4 LT |
976 | struct list_head *curr; |
977 | ||
978 | if (!zone->spanned_pages) | |
979 | return; | |
980 | ||
981 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db RW |
982 | |
983 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
984 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
985 | if (pfn_valid(pfn)) { | |
986 | struct page *page = pfn_to_page(pfn); | |
987 | ||
7be98234 RW |
988 | if (!swsusp_page_is_forbidden(page)) |
989 | swsusp_unset_page_free(page); | |
f623f0db | 990 | } |
1da177e4 | 991 | |
b2a0ac88 MG |
992 | for_each_migratetype_order(order, t) { |
993 | list_for_each(curr, &zone->free_area[order].free_list[t]) { | |
f623f0db | 994 | unsigned long i; |
1da177e4 | 995 | |
f623f0db RW |
996 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); |
997 | for (i = 0; i < (1UL << order); i++) | |
7be98234 | 998 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 999 | } |
b2a0ac88 | 1000 | } |
1da177e4 LT |
1001 | spin_unlock_irqrestore(&zone->lock, flags); |
1002 | } | |
e2c55dc8 | 1003 | #endif /* CONFIG_PM */ |
1da177e4 | 1004 | |
e2c55dc8 | 1005 | #if defined(CONFIG_HIBERNATION) || defined(CONFIG_PAGE_GROUP_BY_MOBILITY) |
1da177e4 LT |
1006 | /* |
1007 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
1008 | */ | |
1009 | void drain_local_pages(void) | |
1010 | { | |
1011 | unsigned long flags; | |
1012 | ||
1013 | local_irq_save(flags); | |
1014 | __drain_pages(smp_processor_id()); | |
1015 | local_irq_restore(flags); | |
1016 | } | |
e2c55dc8 MG |
1017 | |
1018 | void smp_drain_local_pages(void *arg) | |
1019 | { | |
1020 | drain_local_pages(); | |
1021 | } | |
1022 | ||
1023 | /* | |
1024 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator | |
1025 | */ | |
1026 | void drain_all_local_pages(void) | |
1027 | { | |
1028 | unsigned long flags; | |
1029 | ||
1030 | local_irq_save(flags); | |
1031 | __drain_pages(smp_processor_id()); | |
1032 | local_irq_restore(flags); | |
1033 | ||
1034 | smp_call_function(smp_drain_local_pages, NULL, 0, 1); | |
1035 | } | |
1036 | #else | |
1037 | void drain_all_local_pages(void) {} | |
1038 | #endif /* CONFIG_HIBERNATION || CONFIG_PAGE_GROUP_BY_MOBILITY */ | |
1da177e4 | 1039 | |
1da177e4 LT |
1040 | /* |
1041 | * Free a 0-order page | |
1042 | */ | |
1da177e4 LT |
1043 | static void fastcall free_hot_cold_page(struct page *page, int cold) |
1044 | { | |
1045 | struct zone *zone = page_zone(page); | |
1046 | struct per_cpu_pages *pcp; | |
1047 | unsigned long flags; | |
1048 | ||
1da177e4 LT |
1049 | if (PageAnon(page)) |
1050 | page->mapping = NULL; | |
224abf92 | 1051 | if (free_pages_check(page)) |
689bcebf HD |
1052 | return; |
1053 | ||
9858db50 NP |
1054 | if (!PageHighMem(page)) |
1055 | debug_check_no_locks_freed(page_address(page), PAGE_SIZE); | |
dafb1367 | 1056 | arch_free_page(page, 0); |
689bcebf HD |
1057 | kernel_map_pages(page, 1, 0); |
1058 | ||
e7c8d5c9 | 1059 | pcp = &zone_pcp(zone, get_cpu())->pcp[cold]; |
1da177e4 | 1060 | local_irq_save(flags); |
f8891e5e | 1061 | __count_vm_event(PGFREE); |
1da177e4 | 1062 | list_add(&page->lru, &pcp->list); |
535131e6 | 1063 | set_page_private(page, get_pageblock_migratetype(page)); |
1da177e4 | 1064 | pcp->count++; |
48db57f8 NP |
1065 | if (pcp->count >= pcp->high) { |
1066 | free_pages_bulk(zone, pcp->batch, &pcp->list, 0); | |
1067 | pcp->count -= pcp->batch; | |
1068 | } | |
1da177e4 LT |
1069 | local_irq_restore(flags); |
1070 | put_cpu(); | |
1071 | } | |
1072 | ||
1073 | void fastcall free_hot_page(struct page *page) | |
1074 | { | |
1075 | free_hot_cold_page(page, 0); | |
1076 | } | |
1077 | ||
1078 | void fastcall free_cold_page(struct page *page) | |
1079 | { | |
1080 | free_hot_cold_page(page, 1); | |
1081 | } | |
1082 | ||
8dfcc9ba NP |
1083 | /* |
1084 | * split_page takes a non-compound higher-order page, and splits it into | |
1085 | * n (1<<order) sub-pages: page[0..n] | |
1086 | * Each sub-page must be freed individually. | |
1087 | * | |
1088 | * Note: this is probably too low level an operation for use in drivers. | |
1089 | * Please consult with lkml before using this in your driver. | |
1090 | */ | |
1091 | void split_page(struct page *page, unsigned int order) | |
1092 | { | |
1093 | int i; | |
1094 | ||
725d704e NP |
1095 | VM_BUG_ON(PageCompound(page)); |
1096 | VM_BUG_ON(!page_count(page)); | |
7835e98b NP |
1097 | for (i = 1; i < (1 << order); i++) |
1098 | set_page_refcounted(page + i); | |
8dfcc9ba | 1099 | } |
8dfcc9ba | 1100 | |
1da177e4 LT |
1101 | /* |
1102 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
1103 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
1104 | * or two. | |
1105 | */ | |
a74609fa NP |
1106 | static struct page *buffered_rmqueue(struct zonelist *zonelist, |
1107 | struct zone *zone, int order, gfp_t gfp_flags) | |
1da177e4 LT |
1108 | { |
1109 | unsigned long flags; | |
689bcebf | 1110 | struct page *page; |
1da177e4 | 1111 | int cold = !!(gfp_flags & __GFP_COLD); |
a74609fa | 1112 | int cpu; |
e010487d | 1113 | int migratetype = allocflags_to_migratetype(gfp_flags, order); |
1da177e4 | 1114 | |
689bcebf | 1115 | again: |
a74609fa | 1116 | cpu = get_cpu(); |
48db57f8 | 1117 | if (likely(order == 0)) { |
1da177e4 LT |
1118 | struct per_cpu_pages *pcp; |
1119 | ||
a74609fa | 1120 | pcp = &zone_pcp(zone, cpu)->pcp[cold]; |
1da177e4 | 1121 | local_irq_save(flags); |
a74609fa | 1122 | if (!pcp->count) { |
941c7105 | 1123 | pcp->count = rmqueue_bulk(zone, 0, |
b2a0ac88 | 1124 | pcp->batch, &pcp->list, migratetype); |
a74609fa NP |
1125 | if (unlikely(!pcp->count)) |
1126 | goto failed; | |
1da177e4 | 1127 | } |
b92a6edd MG |
1128 | |
1129 | #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY | |
535131e6 | 1130 | /* Find a page of the appropriate migrate type */ |
b92a6edd MG |
1131 | list_for_each_entry(page, &pcp->list, lru) |
1132 | if (page_private(page) == migratetype) | |
535131e6 | 1133 | break; |
535131e6 | 1134 | |
b92a6edd MG |
1135 | /* Allocate more to the pcp list if necessary */ |
1136 | if (unlikely(&page->lru == &pcp->list)) { | |
535131e6 MG |
1137 | pcp->count += rmqueue_bulk(zone, 0, |
1138 | pcp->batch, &pcp->list, migratetype); | |
1139 | page = list_entry(pcp->list.next, struct page, lru); | |
535131e6 | 1140 | } |
b92a6edd MG |
1141 | #else |
1142 | page = list_entry(pcp->list.next, struct page, lru); | |
1143 | #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */ | |
1144 | ||
1145 | list_del(&page->lru); | |
1146 | pcp->count--; | |
7fb1d9fc | 1147 | } else { |
1da177e4 | 1148 | spin_lock_irqsave(&zone->lock, flags); |
b2a0ac88 | 1149 | page = __rmqueue(zone, order, migratetype); |
a74609fa NP |
1150 | spin_unlock(&zone->lock); |
1151 | if (!page) | |
1152 | goto failed; | |
1da177e4 LT |
1153 | } |
1154 | ||
f8891e5e | 1155 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
ca889e6c | 1156 | zone_statistics(zonelist, zone); |
a74609fa NP |
1157 | local_irq_restore(flags); |
1158 | put_cpu(); | |
1da177e4 | 1159 | |
725d704e | 1160 | VM_BUG_ON(bad_range(zone, page)); |
17cf4406 | 1161 | if (prep_new_page(page, order, gfp_flags)) |
a74609fa | 1162 | goto again; |
1da177e4 | 1163 | return page; |
a74609fa NP |
1164 | |
1165 | failed: | |
1166 | local_irq_restore(flags); | |
1167 | put_cpu(); | |
1168 | return NULL; | |
1da177e4 LT |
1169 | } |
1170 | ||
7fb1d9fc | 1171 | #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */ |
3148890b NP |
1172 | #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */ |
1173 | #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */ | |
1174 | #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */ | |
1175 | #define ALLOC_HARDER 0x10 /* try to alloc harder */ | |
1176 | #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */ | |
1177 | #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ | |
7fb1d9fc | 1178 | |
933e312e AM |
1179 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
1180 | ||
1181 | static struct fail_page_alloc_attr { | |
1182 | struct fault_attr attr; | |
1183 | ||
1184 | u32 ignore_gfp_highmem; | |
1185 | u32 ignore_gfp_wait; | |
54114994 | 1186 | u32 min_order; |
933e312e AM |
1187 | |
1188 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1189 | ||
1190 | struct dentry *ignore_gfp_highmem_file; | |
1191 | struct dentry *ignore_gfp_wait_file; | |
54114994 | 1192 | struct dentry *min_order_file; |
933e312e AM |
1193 | |
1194 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1195 | ||
1196 | } fail_page_alloc = { | |
1197 | .attr = FAULT_ATTR_INITIALIZER, | |
6b1b60f4 DM |
1198 | .ignore_gfp_wait = 1, |
1199 | .ignore_gfp_highmem = 1, | |
54114994 | 1200 | .min_order = 1, |
933e312e AM |
1201 | }; |
1202 | ||
1203 | static int __init setup_fail_page_alloc(char *str) | |
1204 | { | |
1205 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
1206 | } | |
1207 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
1208 | ||
1209 | static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
1210 | { | |
54114994 AM |
1211 | if (order < fail_page_alloc.min_order) |
1212 | return 0; | |
933e312e AM |
1213 | if (gfp_mask & __GFP_NOFAIL) |
1214 | return 0; | |
1215 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) | |
1216 | return 0; | |
1217 | if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT)) | |
1218 | return 0; | |
1219 | ||
1220 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
1221 | } | |
1222 | ||
1223 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1224 | ||
1225 | static int __init fail_page_alloc_debugfs(void) | |
1226 | { | |
1227 | mode_t mode = S_IFREG | S_IRUSR | S_IWUSR; | |
1228 | struct dentry *dir; | |
1229 | int err; | |
1230 | ||
1231 | err = init_fault_attr_dentries(&fail_page_alloc.attr, | |
1232 | "fail_page_alloc"); | |
1233 | if (err) | |
1234 | return err; | |
1235 | dir = fail_page_alloc.attr.dentries.dir; | |
1236 | ||
1237 | fail_page_alloc.ignore_gfp_wait_file = | |
1238 | debugfs_create_bool("ignore-gfp-wait", mode, dir, | |
1239 | &fail_page_alloc.ignore_gfp_wait); | |
1240 | ||
1241 | fail_page_alloc.ignore_gfp_highmem_file = | |
1242 | debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
1243 | &fail_page_alloc.ignore_gfp_highmem); | |
54114994 AM |
1244 | fail_page_alloc.min_order_file = |
1245 | debugfs_create_u32("min-order", mode, dir, | |
1246 | &fail_page_alloc.min_order); | |
933e312e AM |
1247 | |
1248 | if (!fail_page_alloc.ignore_gfp_wait_file || | |
54114994 AM |
1249 | !fail_page_alloc.ignore_gfp_highmem_file || |
1250 | !fail_page_alloc.min_order_file) { | |
933e312e AM |
1251 | err = -ENOMEM; |
1252 | debugfs_remove(fail_page_alloc.ignore_gfp_wait_file); | |
1253 | debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file); | |
54114994 | 1254 | debugfs_remove(fail_page_alloc.min_order_file); |
933e312e AM |
1255 | cleanup_fault_attr_dentries(&fail_page_alloc.attr); |
1256 | } | |
1257 | ||
1258 | return err; | |
1259 | } | |
1260 | ||
1261 | late_initcall(fail_page_alloc_debugfs); | |
1262 | ||
1263 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1264 | ||
1265 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
1266 | ||
1267 | static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
1268 | { | |
1269 | return 0; | |
1270 | } | |
1271 | ||
1272 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
1273 | ||
1da177e4 LT |
1274 | /* |
1275 | * Return 1 if free pages are above 'mark'. This takes into account the order | |
1276 | * of the allocation. | |
1277 | */ | |
1278 | int zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
7fb1d9fc | 1279 | int classzone_idx, int alloc_flags) |
1da177e4 LT |
1280 | { |
1281 | /* free_pages my go negative - that's OK */ | |
d23ad423 CL |
1282 | long min = mark; |
1283 | long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1; | |
1da177e4 LT |
1284 | int o; |
1285 | ||
7fb1d9fc | 1286 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 1287 | min -= min / 2; |
7fb1d9fc | 1288 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 LT |
1289 | min -= min / 4; |
1290 | ||
1291 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
1292 | return 0; | |
1293 | for (o = 0; o < order; o++) { | |
1294 | /* At the next order, this order's pages become unavailable */ | |
1295 | free_pages -= z->free_area[o].nr_free << o; | |
1296 | ||
1297 | /* Require fewer higher order pages to be free */ | |
1298 | min >>= 1; | |
1299 | ||
1300 | if (free_pages <= min) | |
1301 | return 0; | |
1302 | } | |
1303 | return 1; | |
1304 | } | |
1305 | ||
9276b1bc PJ |
1306 | #ifdef CONFIG_NUMA |
1307 | /* | |
1308 | * zlc_setup - Setup for "zonelist cache". Uses cached zone data to | |
1309 | * skip over zones that are not allowed by the cpuset, or that have | |
1310 | * been recently (in last second) found to be nearly full. See further | |
1311 | * comments in mmzone.h. Reduces cache footprint of zonelist scans | |
1312 | * that have to skip over alot of full or unallowed zones. | |
1313 | * | |
1314 | * If the zonelist cache is present in the passed in zonelist, then | |
1315 | * returns a pointer to the allowed node mask (either the current | |
37b07e41 | 1316 | * tasks mems_allowed, or node_states[N_HIGH_MEMORY].) |
9276b1bc PJ |
1317 | * |
1318 | * If the zonelist cache is not available for this zonelist, does | |
1319 | * nothing and returns NULL. | |
1320 | * | |
1321 | * If the fullzones BITMAP in the zonelist cache is stale (more than | |
1322 | * a second since last zap'd) then we zap it out (clear its bits.) | |
1323 | * | |
1324 | * We hold off even calling zlc_setup, until after we've checked the | |
1325 | * first zone in the zonelist, on the theory that most allocations will | |
1326 | * be satisfied from that first zone, so best to examine that zone as | |
1327 | * quickly as we can. | |
1328 | */ | |
1329 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1330 | { | |
1331 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1332 | nodemask_t *allowednodes; /* zonelist_cache approximation */ | |
1333 | ||
1334 | zlc = zonelist->zlcache_ptr; | |
1335 | if (!zlc) | |
1336 | return NULL; | |
1337 | ||
1338 | if (jiffies - zlc->last_full_zap > 1 * HZ) { | |
1339 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
1340 | zlc->last_full_zap = jiffies; | |
1341 | } | |
1342 | ||
1343 | allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ? | |
1344 | &cpuset_current_mems_allowed : | |
37b07e41 | 1345 | &node_states[N_HIGH_MEMORY]; |
9276b1bc PJ |
1346 | return allowednodes; |
1347 | } | |
1348 | ||
1349 | /* | |
1350 | * Given 'z' scanning a zonelist, run a couple of quick checks to see | |
1351 | * if it is worth looking at further for free memory: | |
1352 | * 1) Check that the zone isn't thought to be full (doesn't have its | |
1353 | * bit set in the zonelist_cache fullzones BITMAP). | |
1354 | * 2) Check that the zones node (obtained from the zonelist_cache | |
1355 | * z_to_n[] mapping) is allowed in the passed in allowednodes mask. | |
1356 | * Return true (non-zero) if zone is worth looking at further, or | |
1357 | * else return false (zero) if it is not. | |
1358 | * | |
1359 | * This check -ignores- the distinction between various watermarks, | |
1360 | * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is | |
1361 | * found to be full for any variation of these watermarks, it will | |
1362 | * be considered full for up to one second by all requests, unless | |
1363 | * we are so low on memory on all allowed nodes that we are forced | |
1364 | * into the second scan of the zonelist. | |
1365 | * | |
1366 | * In the second scan we ignore this zonelist cache and exactly | |
1367 | * apply the watermarks to all zones, even it is slower to do so. | |
1368 | * We are low on memory in the second scan, and should leave no stone | |
1369 | * unturned looking for a free page. | |
1370 | */ | |
1371 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z, | |
1372 | nodemask_t *allowednodes) | |
1373 | { | |
1374 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1375 | int i; /* index of *z in zonelist zones */ | |
1376 | int n; /* node that zone *z is on */ | |
1377 | ||
1378 | zlc = zonelist->zlcache_ptr; | |
1379 | if (!zlc) | |
1380 | return 1; | |
1381 | ||
1382 | i = z - zonelist->zones; | |
1383 | n = zlc->z_to_n[i]; | |
1384 | ||
1385 | /* This zone is worth trying if it is allowed but not full */ | |
1386 | return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones); | |
1387 | } | |
1388 | ||
1389 | /* | |
1390 | * Given 'z' scanning a zonelist, set the corresponding bit in | |
1391 | * zlc->fullzones, so that subsequent attempts to allocate a page | |
1392 | * from that zone don't waste time re-examining it. | |
1393 | */ | |
1394 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z) | |
1395 | { | |
1396 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1397 | int i; /* index of *z in zonelist zones */ | |
1398 | ||
1399 | zlc = zonelist->zlcache_ptr; | |
1400 | if (!zlc) | |
1401 | return; | |
1402 | ||
1403 | i = z - zonelist->zones; | |
1404 | ||
1405 | set_bit(i, zlc->fullzones); | |
1406 | } | |
1407 | ||
1408 | #else /* CONFIG_NUMA */ | |
1409 | ||
1410 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1411 | { | |
1412 | return NULL; | |
1413 | } | |
1414 | ||
1415 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z, | |
1416 | nodemask_t *allowednodes) | |
1417 | { | |
1418 | return 1; | |
1419 | } | |
1420 | ||
1421 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z) | |
1422 | { | |
1423 | } | |
1424 | #endif /* CONFIG_NUMA */ | |
1425 | ||
7fb1d9fc | 1426 | /* |
0798e519 | 1427 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
1428 | * a page. |
1429 | */ | |
1430 | static struct page * | |
1431 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, | |
1432 | struct zonelist *zonelist, int alloc_flags) | |
753ee728 | 1433 | { |
9276b1bc | 1434 | struct zone **z; |
7fb1d9fc | 1435 | struct page *page = NULL; |
9276b1bc | 1436 | int classzone_idx = zone_idx(zonelist->zones[0]); |
1192d526 | 1437 | struct zone *zone; |
9276b1bc PJ |
1438 | nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */ |
1439 | int zlc_active = 0; /* set if using zonelist_cache */ | |
1440 | int did_zlc_setup = 0; /* just call zlc_setup() one time */ | |
b377fd39 | 1441 | enum zone_type highest_zoneidx = -1; /* Gets set for policy zonelists */ |
7fb1d9fc | 1442 | |
9276b1bc | 1443 | zonelist_scan: |
7fb1d9fc | 1444 | /* |
9276b1bc | 1445 | * Scan zonelist, looking for a zone with enough free. |
7fb1d9fc RS |
1446 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1447 | */ | |
9276b1bc PJ |
1448 | z = zonelist->zones; |
1449 | ||
7fb1d9fc | 1450 | do { |
b377fd39 MG |
1451 | /* |
1452 | * In NUMA, this could be a policy zonelist which contains | |
1453 | * zones that may not be allowed by the current gfp_mask. | |
1454 | * Check the zone is allowed by the current flags | |
1455 | */ | |
1456 | if (unlikely(alloc_should_filter_zonelist(zonelist))) { | |
1457 | if (highest_zoneidx == -1) | |
1458 | highest_zoneidx = gfp_zone(gfp_mask); | |
1459 | if (zone_idx(*z) > highest_zoneidx) | |
1460 | continue; | |
1461 | } | |
1462 | ||
9276b1bc PJ |
1463 | if (NUMA_BUILD && zlc_active && |
1464 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) | |
1465 | continue; | |
1192d526 | 1466 | zone = *z; |
7fb1d9fc | 1467 | if ((alloc_flags & ALLOC_CPUSET) && |
02a0e53d | 1468 | !cpuset_zone_allowed_softwall(zone, gfp_mask)) |
9276b1bc | 1469 | goto try_next_zone; |
7fb1d9fc RS |
1470 | |
1471 | if (!(alloc_flags & ALLOC_NO_WATERMARKS)) { | |
3148890b NP |
1472 | unsigned long mark; |
1473 | if (alloc_flags & ALLOC_WMARK_MIN) | |
1192d526 | 1474 | mark = zone->pages_min; |
3148890b | 1475 | else if (alloc_flags & ALLOC_WMARK_LOW) |
1192d526 | 1476 | mark = zone->pages_low; |
3148890b | 1477 | else |
1192d526 | 1478 | mark = zone->pages_high; |
0798e519 PJ |
1479 | if (!zone_watermark_ok(zone, order, mark, |
1480 | classzone_idx, alloc_flags)) { | |
9eeff239 | 1481 | if (!zone_reclaim_mode || |
1192d526 | 1482 | !zone_reclaim(zone, gfp_mask, order)) |
9276b1bc | 1483 | goto this_zone_full; |
0798e519 | 1484 | } |
7fb1d9fc RS |
1485 | } |
1486 | ||
1192d526 | 1487 | page = buffered_rmqueue(zonelist, zone, order, gfp_mask); |
0798e519 | 1488 | if (page) |
7fb1d9fc | 1489 | break; |
9276b1bc PJ |
1490 | this_zone_full: |
1491 | if (NUMA_BUILD) | |
1492 | zlc_mark_zone_full(zonelist, z); | |
1493 | try_next_zone: | |
1494 | if (NUMA_BUILD && !did_zlc_setup) { | |
1495 | /* we do zlc_setup after the first zone is tried */ | |
1496 | allowednodes = zlc_setup(zonelist, alloc_flags); | |
1497 | zlc_active = 1; | |
1498 | did_zlc_setup = 1; | |
1499 | } | |
7fb1d9fc | 1500 | } while (*(++z) != NULL); |
9276b1bc PJ |
1501 | |
1502 | if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) { | |
1503 | /* Disable zlc cache for second zonelist scan */ | |
1504 | zlc_active = 0; | |
1505 | goto zonelist_scan; | |
1506 | } | |
7fb1d9fc | 1507 | return page; |
753ee728 MH |
1508 | } |
1509 | ||
1da177e4 LT |
1510 | /* |
1511 | * This is the 'heart' of the zoned buddy allocator. | |
1512 | */ | |
1513 | struct page * fastcall | |
dd0fc66f | 1514 | __alloc_pages(gfp_t gfp_mask, unsigned int order, |
1da177e4 LT |
1515 | struct zonelist *zonelist) |
1516 | { | |
260b2367 | 1517 | const gfp_t wait = gfp_mask & __GFP_WAIT; |
7fb1d9fc | 1518 | struct zone **z; |
1da177e4 LT |
1519 | struct page *page; |
1520 | struct reclaim_state reclaim_state; | |
1521 | struct task_struct *p = current; | |
1da177e4 | 1522 | int do_retry; |
7fb1d9fc | 1523 | int alloc_flags; |
1da177e4 LT |
1524 | int did_some_progress; |
1525 | ||
1526 | might_sleep_if(wait); | |
1527 | ||
933e312e AM |
1528 | if (should_fail_alloc_page(gfp_mask, order)) |
1529 | return NULL; | |
1530 | ||
6b1de916 | 1531 | restart: |
7fb1d9fc | 1532 | z = zonelist->zones; /* the list of zones suitable for gfp_mask */ |
1da177e4 | 1533 | |
7fb1d9fc | 1534 | if (unlikely(*z == NULL)) { |
523b9458 CL |
1535 | /* |
1536 | * Happens if we have an empty zonelist as a result of | |
1537 | * GFP_THISNODE being used on a memoryless node | |
1538 | */ | |
1da177e4 LT |
1539 | return NULL; |
1540 | } | |
6b1de916 | 1541 | |
7fb1d9fc | 1542 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, |
3148890b | 1543 | zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET); |
7fb1d9fc RS |
1544 | if (page) |
1545 | goto got_pg; | |
1da177e4 | 1546 | |
952f3b51 CL |
1547 | /* |
1548 | * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and | |
1549 | * __GFP_NOWARN set) should not cause reclaim since the subsystem | |
1550 | * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim | |
1551 | * using a larger set of nodes after it has established that the | |
1552 | * allowed per node queues are empty and that nodes are | |
1553 | * over allocated. | |
1554 | */ | |
1555 | if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE) | |
1556 | goto nopage; | |
1557 | ||
0798e519 | 1558 | for (z = zonelist->zones; *z; z++) |
43b0bc00 | 1559 | wakeup_kswapd(*z, order); |
1da177e4 | 1560 | |
9bf2229f | 1561 | /* |
7fb1d9fc RS |
1562 | * OK, we're below the kswapd watermark and have kicked background |
1563 | * reclaim. Now things get more complex, so set up alloc_flags according | |
1564 | * to how we want to proceed. | |
1565 | * | |
1566 | * The caller may dip into page reserves a bit more if the caller | |
1567 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
4eac915d PJ |
1568 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will |
1569 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). | |
9bf2229f | 1570 | */ |
3148890b | 1571 | alloc_flags = ALLOC_WMARK_MIN; |
7fb1d9fc RS |
1572 | if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait) |
1573 | alloc_flags |= ALLOC_HARDER; | |
1574 | if (gfp_mask & __GFP_HIGH) | |
1575 | alloc_flags |= ALLOC_HIGH; | |
bdd804f4 PJ |
1576 | if (wait) |
1577 | alloc_flags |= ALLOC_CPUSET; | |
1da177e4 LT |
1578 | |
1579 | /* | |
1580 | * Go through the zonelist again. Let __GFP_HIGH and allocations | |
7fb1d9fc | 1581 | * coming from realtime tasks go deeper into reserves. |
1da177e4 LT |
1582 | * |
1583 | * This is the last chance, in general, before the goto nopage. | |
1584 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. | |
9bf2229f | 1585 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1da177e4 | 1586 | */ |
7fb1d9fc RS |
1587 | page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags); |
1588 | if (page) | |
1589 | goto got_pg; | |
1da177e4 LT |
1590 | |
1591 | /* This allocation should allow future memory freeing. */ | |
b84a35be | 1592 | |
b43a57bb | 1593 | rebalance: |
b84a35be NP |
1594 | if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE))) |
1595 | && !in_interrupt()) { | |
1596 | if (!(gfp_mask & __GFP_NOMEMALLOC)) { | |
885036d3 | 1597 | nofail_alloc: |
b84a35be | 1598 | /* go through the zonelist yet again, ignoring mins */ |
7fb1d9fc | 1599 | page = get_page_from_freelist(gfp_mask, order, |
47f3a867 | 1600 | zonelist, ALLOC_NO_WATERMARKS); |
7fb1d9fc RS |
1601 | if (page) |
1602 | goto got_pg; | |
885036d3 | 1603 | if (gfp_mask & __GFP_NOFAIL) { |
3fcfab16 | 1604 | congestion_wait(WRITE, HZ/50); |
885036d3 KK |
1605 | goto nofail_alloc; |
1606 | } | |
1da177e4 LT |
1607 | } |
1608 | goto nopage; | |
1609 | } | |
1610 | ||
1611 | /* Atomic allocations - we can't balance anything */ | |
1612 | if (!wait) | |
1613 | goto nopage; | |
1614 | ||
1da177e4 LT |
1615 | cond_resched(); |
1616 | ||
1617 | /* We now go into synchronous reclaim */ | |
3e0d98b9 | 1618 | cpuset_memory_pressure_bump(); |
1da177e4 LT |
1619 | p->flags |= PF_MEMALLOC; |
1620 | reclaim_state.reclaimed_slab = 0; | |
1621 | p->reclaim_state = &reclaim_state; | |
1622 | ||
5ad333eb | 1623 | did_some_progress = try_to_free_pages(zonelist->zones, order, gfp_mask); |
1da177e4 LT |
1624 | |
1625 | p->reclaim_state = NULL; | |
1626 | p->flags &= ~PF_MEMALLOC; | |
1627 | ||
1628 | cond_resched(); | |
1629 | ||
e2c55dc8 MG |
1630 | if (order != 0) |
1631 | drain_all_local_pages(); | |
1632 | ||
1da177e4 | 1633 | if (likely(did_some_progress)) { |
7fb1d9fc RS |
1634 | page = get_page_from_freelist(gfp_mask, order, |
1635 | zonelist, alloc_flags); | |
1636 | if (page) | |
1637 | goto got_pg; | |
1da177e4 LT |
1638 | } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) { |
1639 | /* | |
1640 | * Go through the zonelist yet one more time, keep | |
1641 | * very high watermark here, this is only to catch | |
1642 | * a parallel oom killing, we must fail if we're still | |
1643 | * under heavy pressure. | |
1644 | */ | |
7fb1d9fc | 1645 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, |
3148890b | 1646 | zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET); |
7fb1d9fc RS |
1647 | if (page) |
1648 | goto got_pg; | |
1da177e4 | 1649 | |
a8bbf72a MG |
1650 | /* The OOM killer will not help higher order allocs so fail */ |
1651 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
1652 | goto nopage; | |
1653 | ||
9b0f8b04 | 1654 | out_of_memory(zonelist, gfp_mask, order); |
1da177e4 LT |
1655 | goto restart; |
1656 | } | |
1657 | ||
1658 | /* | |
1659 | * Don't let big-order allocations loop unless the caller explicitly | |
1660 | * requests that. Wait for some write requests to complete then retry. | |
1661 | * | |
1662 | * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order | |
1663 | * <= 3, but that may not be true in other implementations. | |
1664 | */ | |
1665 | do_retry = 0; | |
1666 | if (!(gfp_mask & __GFP_NORETRY)) { | |
5ad333eb AW |
1667 | if ((order <= PAGE_ALLOC_COSTLY_ORDER) || |
1668 | (gfp_mask & __GFP_REPEAT)) | |
1da177e4 LT |
1669 | do_retry = 1; |
1670 | if (gfp_mask & __GFP_NOFAIL) | |
1671 | do_retry = 1; | |
1672 | } | |
1673 | if (do_retry) { | |
3fcfab16 | 1674 | congestion_wait(WRITE, HZ/50); |
1da177e4 LT |
1675 | goto rebalance; |
1676 | } | |
1677 | ||
1678 | nopage: | |
1679 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) { | |
1680 | printk(KERN_WARNING "%s: page allocation failure." | |
1681 | " order:%d, mode:0x%x\n", | |
1682 | p->comm, order, gfp_mask); | |
1683 | dump_stack(); | |
578c2fd6 | 1684 | show_mem(); |
1da177e4 | 1685 | } |
1da177e4 | 1686 | got_pg: |
1da177e4 LT |
1687 | return page; |
1688 | } | |
1689 | ||
1690 | EXPORT_SYMBOL(__alloc_pages); | |
1691 | ||
1692 | /* | |
1693 | * Common helper functions. | |
1694 | */ | |
dd0fc66f | 1695 | fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 LT |
1696 | { |
1697 | struct page * page; | |
1698 | page = alloc_pages(gfp_mask, order); | |
1699 | if (!page) | |
1700 | return 0; | |
1701 | return (unsigned long) page_address(page); | |
1702 | } | |
1703 | ||
1704 | EXPORT_SYMBOL(__get_free_pages); | |
1705 | ||
dd0fc66f | 1706 | fastcall unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 LT |
1707 | { |
1708 | struct page * page; | |
1709 | ||
1710 | /* | |
1711 | * get_zeroed_page() returns a 32-bit address, which cannot represent | |
1712 | * a highmem page | |
1713 | */ | |
725d704e | 1714 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); |
1da177e4 LT |
1715 | |
1716 | page = alloc_pages(gfp_mask | __GFP_ZERO, 0); | |
1717 | if (page) | |
1718 | return (unsigned long) page_address(page); | |
1719 | return 0; | |
1720 | } | |
1721 | ||
1722 | EXPORT_SYMBOL(get_zeroed_page); | |
1723 | ||
1724 | void __pagevec_free(struct pagevec *pvec) | |
1725 | { | |
1726 | int i = pagevec_count(pvec); | |
1727 | ||
1728 | while (--i >= 0) | |
1729 | free_hot_cold_page(pvec->pages[i], pvec->cold); | |
1730 | } | |
1731 | ||
1732 | fastcall void __free_pages(struct page *page, unsigned int order) | |
1733 | { | |
b5810039 | 1734 | if (put_page_testzero(page)) { |
1da177e4 LT |
1735 | if (order == 0) |
1736 | free_hot_page(page); | |
1737 | else | |
1738 | __free_pages_ok(page, order); | |
1739 | } | |
1740 | } | |
1741 | ||
1742 | EXPORT_SYMBOL(__free_pages); | |
1743 | ||
1744 | fastcall void free_pages(unsigned long addr, unsigned int order) | |
1745 | { | |
1746 | if (addr != 0) { | |
725d704e | 1747 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
1748 | __free_pages(virt_to_page((void *)addr), order); |
1749 | } | |
1750 | } | |
1751 | ||
1752 | EXPORT_SYMBOL(free_pages); | |
1753 | ||
1da177e4 LT |
1754 | static unsigned int nr_free_zone_pages(int offset) |
1755 | { | |
e310fd43 MB |
1756 | /* Just pick one node, since fallback list is circular */ |
1757 | pg_data_t *pgdat = NODE_DATA(numa_node_id()); | |
1da177e4 LT |
1758 | unsigned int sum = 0; |
1759 | ||
e310fd43 MB |
1760 | struct zonelist *zonelist = pgdat->node_zonelists + offset; |
1761 | struct zone **zonep = zonelist->zones; | |
1762 | struct zone *zone; | |
1da177e4 | 1763 | |
e310fd43 MB |
1764 | for (zone = *zonep++; zone; zone = *zonep++) { |
1765 | unsigned long size = zone->present_pages; | |
1766 | unsigned long high = zone->pages_high; | |
1767 | if (size > high) | |
1768 | sum += size - high; | |
1da177e4 LT |
1769 | } |
1770 | ||
1771 | return sum; | |
1772 | } | |
1773 | ||
1774 | /* | |
1775 | * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL | |
1776 | */ | |
1777 | unsigned int nr_free_buffer_pages(void) | |
1778 | { | |
af4ca457 | 1779 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 1780 | } |
c2f1a551 | 1781 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 LT |
1782 | |
1783 | /* | |
1784 | * Amount of free RAM allocatable within all zones | |
1785 | */ | |
1786 | unsigned int nr_free_pagecache_pages(void) | |
1787 | { | |
2a1e274a | 1788 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 1789 | } |
08e0f6a9 CL |
1790 | |
1791 | static inline void show_node(struct zone *zone) | |
1da177e4 | 1792 | { |
08e0f6a9 | 1793 | if (NUMA_BUILD) |
25ba77c1 | 1794 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 1795 | } |
1da177e4 | 1796 | |
1da177e4 LT |
1797 | void si_meminfo(struct sysinfo *val) |
1798 | { | |
1799 | val->totalram = totalram_pages; | |
1800 | val->sharedram = 0; | |
d23ad423 | 1801 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 1802 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
1803 | val->totalhigh = totalhigh_pages; |
1804 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
1805 | val->mem_unit = PAGE_SIZE; |
1806 | } | |
1807 | ||
1808 | EXPORT_SYMBOL(si_meminfo); | |
1809 | ||
1810 | #ifdef CONFIG_NUMA | |
1811 | void si_meminfo_node(struct sysinfo *val, int nid) | |
1812 | { | |
1813 | pg_data_t *pgdat = NODE_DATA(nid); | |
1814 | ||
1815 | val->totalram = pgdat->node_present_pages; | |
d23ad423 | 1816 | val->freeram = node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 1817 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 1818 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
d23ad423 CL |
1819 | val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM], |
1820 | NR_FREE_PAGES); | |
98d2b0eb CL |
1821 | #else |
1822 | val->totalhigh = 0; | |
1823 | val->freehigh = 0; | |
1824 | #endif | |
1da177e4 LT |
1825 | val->mem_unit = PAGE_SIZE; |
1826 | } | |
1827 | #endif | |
1828 | ||
1829 | #define K(x) ((x) << (PAGE_SHIFT-10)) | |
1830 | ||
1831 | /* | |
1832 | * Show free area list (used inside shift_scroll-lock stuff) | |
1833 | * We also calculate the percentage fragmentation. We do this by counting the | |
1834 | * memory on each free list with the exception of the first item on the list. | |
1835 | */ | |
1836 | void show_free_areas(void) | |
1837 | { | |
c7241913 | 1838 | int cpu; |
1da177e4 LT |
1839 | struct zone *zone; |
1840 | ||
1841 | for_each_zone(zone) { | |
c7241913 | 1842 | if (!populated_zone(zone)) |
1da177e4 | 1843 | continue; |
c7241913 JS |
1844 | |
1845 | show_node(zone); | |
1846 | printk("%s per-cpu:\n", zone->name); | |
1da177e4 | 1847 | |
6b482c67 | 1848 | for_each_online_cpu(cpu) { |
1da177e4 LT |
1849 | struct per_cpu_pageset *pageset; |
1850 | ||
e7c8d5c9 | 1851 | pageset = zone_pcp(zone, cpu); |
1da177e4 | 1852 | |
c7241913 JS |
1853 | printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d " |
1854 | "Cold: hi:%5d, btch:%4d usd:%4d\n", | |
1855 | cpu, pageset->pcp[0].high, | |
1856 | pageset->pcp[0].batch, pageset->pcp[0].count, | |
1857 | pageset->pcp[1].high, pageset->pcp[1].batch, | |
1858 | pageset->pcp[1].count); | |
1da177e4 LT |
1859 | } |
1860 | } | |
1861 | ||
a25700a5 | 1862 | printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n" |
d23ad423 | 1863 | " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n", |
65e458d4 CL |
1864 | global_page_state(NR_ACTIVE), |
1865 | global_page_state(NR_INACTIVE), | |
b1e7a8fd | 1866 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 1867 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 1868 | global_page_state(NR_UNSTABLE_NFS), |
d23ad423 | 1869 | global_page_state(NR_FREE_PAGES), |
972d1a7b CL |
1870 | global_page_state(NR_SLAB_RECLAIMABLE) + |
1871 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 1872 | global_page_state(NR_FILE_MAPPED), |
a25700a5 AM |
1873 | global_page_state(NR_PAGETABLE), |
1874 | global_page_state(NR_BOUNCE)); | |
1da177e4 LT |
1875 | |
1876 | for_each_zone(zone) { | |
1877 | int i; | |
1878 | ||
c7241913 JS |
1879 | if (!populated_zone(zone)) |
1880 | continue; | |
1881 | ||
1da177e4 LT |
1882 | show_node(zone); |
1883 | printk("%s" | |
1884 | " free:%lukB" | |
1885 | " min:%lukB" | |
1886 | " low:%lukB" | |
1887 | " high:%lukB" | |
1888 | " active:%lukB" | |
1889 | " inactive:%lukB" | |
1890 | " present:%lukB" | |
1891 | " pages_scanned:%lu" | |
1892 | " all_unreclaimable? %s" | |
1893 | "\n", | |
1894 | zone->name, | |
d23ad423 | 1895 | K(zone_page_state(zone, NR_FREE_PAGES)), |
1da177e4 LT |
1896 | K(zone->pages_min), |
1897 | K(zone->pages_low), | |
1898 | K(zone->pages_high), | |
c8785385 CL |
1899 | K(zone_page_state(zone, NR_ACTIVE)), |
1900 | K(zone_page_state(zone, NR_INACTIVE)), | |
1da177e4 LT |
1901 | K(zone->present_pages), |
1902 | zone->pages_scanned, | |
1903 | (zone->all_unreclaimable ? "yes" : "no") | |
1904 | ); | |
1905 | printk("lowmem_reserve[]:"); | |
1906 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1907 | printk(" %lu", zone->lowmem_reserve[i]); | |
1908 | printk("\n"); | |
1909 | } | |
1910 | ||
1911 | for_each_zone(zone) { | |
8f9de51a | 1912 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
1da177e4 | 1913 | |
c7241913 JS |
1914 | if (!populated_zone(zone)) |
1915 | continue; | |
1916 | ||
1da177e4 LT |
1917 | show_node(zone); |
1918 | printk("%s: ", zone->name); | |
1da177e4 LT |
1919 | |
1920 | spin_lock_irqsave(&zone->lock, flags); | |
1921 | for (order = 0; order < MAX_ORDER; order++) { | |
8f9de51a KK |
1922 | nr[order] = zone->free_area[order].nr_free; |
1923 | total += nr[order] << order; | |
1da177e4 LT |
1924 | } |
1925 | spin_unlock_irqrestore(&zone->lock, flags); | |
8f9de51a KK |
1926 | for (order = 0; order < MAX_ORDER; order++) |
1927 | printk("%lu*%lukB ", nr[order], K(1UL) << order); | |
1da177e4 LT |
1928 | printk("= %lukB\n", K(total)); |
1929 | } | |
1930 | ||
1931 | show_swap_cache_info(); | |
1932 | } | |
1933 | ||
1934 | /* | |
1935 | * Builds allocation fallback zone lists. | |
1a93205b CL |
1936 | * |
1937 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 1938 | */ |
f0c0b2b8 KH |
1939 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
1940 | int nr_zones, enum zone_type zone_type) | |
1da177e4 | 1941 | { |
1a93205b CL |
1942 | struct zone *zone; |
1943 | ||
98d2b0eb | 1944 | BUG_ON(zone_type >= MAX_NR_ZONES); |
2f6726e5 | 1945 | zone_type++; |
02a68a5e CL |
1946 | |
1947 | do { | |
2f6726e5 | 1948 | zone_type--; |
070f8032 | 1949 | zone = pgdat->node_zones + zone_type; |
1a93205b | 1950 | if (populated_zone(zone)) { |
070f8032 CL |
1951 | zonelist->zones[nr_zones++] = zone; |
1952 | check_highest_zone(zone_type); | |
1da177e4 | 1953 | } |
02a68a5e | 1954 | |
2f6726e5 | 1955 | } while (zone_type); |
070f8032 | 1956 | return nr_zones; |
1da177e4 LT |
1957 | } |
1958 | ||
f0c0b2b8 KH |
1959 | |
1960 | /* | |
1961 | * zonelist_order: | |
1962 | * 0 = automatic detection of better ordering. | |
1963 | * 1 = order by ([node] distance, -zonetype) | |
1964 | * 2 = order by (-zonetype, [node] distance) | |
1965 | * | |
1966 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
1967 | * the same zonelist. So only NUMA can configure this param. | |
1968 | */ | |
1969 | #define ZONELIST_ORDER_DEFAULT 0 | |
1970 | #define ZONELIST_ORDER_NODE 1 | |
1971 | #define ZONELIST_ORDER_ZONE 2 | |
1972 | ||
1973 | /* zonelist order in the kernel. | |
1974 | * set_zonelist_order() will set this to NODE or ZONE. | |
1975 | */ | |
1976 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
1977 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
1978 | ||
1979 | ||
1da177e4 | 1980 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
1981 | /* The value user specified ....changed by config */ |
1982 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
1983 | /* string for sysctl */ | |
1984 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
1985 | char numa_zonelist_order[16] = "default"; | |
1986 | ||
1987 | /* | |
1988 | * interface for configure zonelist ordering. | |
1989 | * command line option "numa_zonelist_order" | |
1990 | * = "[dD]efault - default, automatic configuration. | |
1991 | * = "[nN]ode - order by node locality, then by zone within node | |
1992 | * = "[zZ]one - order by zone, then by locality within zone | |
1993 | */ | |
1994 | ||
1995 | static int __parse_numa_zonelist_order(char *s) | |
1996 | { | |
1997 | if (*s == 'd' || *s == 'D') { | |
1998 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
1999 | } else if (*s == 'n' || *s == 'N') { | |
2000 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
2001 | } else if (*s == 'z' || *s == 'Z') { | |
2002 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
2003 | } else { | |
2004 | printk(KERN_WARNING | |
2005 | "Ignoring invalid numa_zonelist_order value: " | |
2006 | "%s\n", s); | |
2007 | return -EINVAL; | |
2008 | } | |
2009 | return 0; | |
2010 | } | |
2011 | ||
2012 | static __init int setup_numa_zonelist_order(char *s) | |
2013 | { | |
2014 | if (s) | |
2015 | return __parse_numa_zonelist_order(s); | |
2016 | return 0; | |
2017 | } | |
2018 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
2019 | ||
2020 | /* | |
2021 | * sysctl handler for numa_zonelist_order | |
2022 | */ | |
2023 | int numa_zonelist_order_handler(ctl_table *table, int write, | |
2024 | struct file *file, void __user *buffer, size_t *length, | |
2025 | loff_t *ppos) | |
2026 | { | |
2027 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
2028 | int ret; | |
2029 | ||
2030 | if (write) | |
2031 | strncpy(saved_string, (char*)table->data, | |
2032 | NUMA_ZONELIST_ORDER_LEN); | |
2033 | ret = proc_dostring(table, write, file, buffer, length, ppos); | |
2034 | if (ret) | |
2035 | return ret; | |
2036 | if (write) { | |
2037 | int oldval = user_zonelist_order; | |
2038 | if (__parse_numa_zonelist_order((char*)table->data)) { | |
2039 | /* | |
2040 | * bogus value. restore saved string | |
2041 | */ | |
2042 | strncpy((char*)table->data, saved_string, | |
2043 | NUMA_ZONELIST_ORDER_LEN); | |
2044 | user_zonelist_order = oldval; | |
2045 | } else if (oldval != user_zonelist_order) | |
2046 | build_all_zonelists(); | |
2047 | } | |
2048 | return 0; | |
2049 | } | |
2050 | ||
2051 | ||
1da177e4 | 2052 | #define MAX_NODE_LOAD (num_online_nodes()) |
f0c0b2b8 KH |
2053 | static int node_load[MAX_NUMNODES]; |
2054 | ||
1da177e4 | 2055 | /** |
4dc3b16b | 2056 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
2057 | * @node: node whose fallback list we're appending |
2058 | * @used_node_mask: nodemask_t of already used nodes | |
2059 | * | |
2060 | * We use a number of factors to determine which is the next node that should | |
2061 | * appear on a given node's fallback list. The node should not have appeared | |
2062 | * already in @node's fallback list, and it should be the next closest node | |
2063 | * according to the distance array (which contains arbitrary distance values | |
2064 | * from each node to each node in the system), and should also prefer nodes | |
2065 | * with no CPUs, since presumably they'll have very little allocation pressure | |
2066 | * on them otherwise. | |
2067 | * It returns -1 if no node is found. | |
2068 | */ | |
f0c0b2b8 | 2069 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 2070 | { |
4cf808eb | 2071 | int n, val; |
1da177e4 LT |
2072 | int min_val = INT_MAX; |
2073 | int best_node = -1; | |
2074 | ||
4cf808eb LT |
2075 | /* Use the local node if we haven't already */ |
2076 | if (!node_isset(node, *used_node_mask)) { | |
2077 | node_set(node, *used_node_mask); | |
2078 | return node; | |
2079 | } | |
1da177e4 | 2080 | |
37b07e41 | 2081 | for_each_node_state(n, N_HIGH_MEMORY) { |
4cf808eb | 2082 | cpumask_t tmp; |
1da177e4 LT |
2083 | |
2084 | /* Don't want a node to appear more than once */ | |
2085 | if (node_isset(n, *used_node_mask)) | |
2086 | continue; | |
2087 | ||
1da177e4 LT |
2088 | /* Use the distance array to find the distance */ |
2089 | val = node_distance(node, n); | |
2090 | ||
4cf808eb LT |
2091 | /* Penalize nodes under us ("prefer the next node") */ |
2092 | val += (n < node); | |
2093 | ||
1da177e4 LT |
2094 | /* Give preference to headless and unused nodes */ |
2095 | tmp = node_to_cpumask(n); | |
2096 | if (!cpus_empty(tmp)) | |
2097 | val += PENALTY_FOR_NODE_WITH_CPUS; | |
2098 | ||
2099 | /* Slight preference for less loaded node */ | |
2100 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
2101 | val += node_load[n]; | |
2102 | ||
2103 | if (val < min_val) { | |
2104 | min_val = val; | |
2105 | best_node = n; | |
2106 | } | |
2107 | } | |
2108 | ||
2109 | if (best_node >= 0) | |
2110 | node_set(best_node, *used_node_mask); | |
2111 | ||
2112 | return best_node; | |
2113 | } | |
2114 | ||
f0c0b2b8 KH |
2115 | |
2116 | /* | |
2117 | * Build zonelists ordered by node and zones within node. | |
2118 | * This results in maximum locality--normal zone overflows into local | |
2119 | * DMA zone, if any--but risks exhausting DMA zone. | |
2120 | */ | |
2121 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 2122 | { |
19655d34 | 2123 | enum zone_type i; |
f0c0b2b8 | 2124 | int j; |
1da177e4 | 2125 | struct zonelist *zonelist; |
f0c0b2b8 KH |
2126 | |
2127 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2128 | zonelist = pgdat->node_zonelists + i; | |
2129 | for (j = 0; zonelist->zones[j] != NULL; j++) | |
2130 | ; | |
2131 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); | |
2132 | zonelist->zones[j] = NULL; | |
2133 | } | |
2134 | } | |
2135 | ||
523b9458 CL |
2136 | /* |
2137 | * Build gfp_thisnode zonelists | |
2138 | */ | |
2139 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
2140 | { | |
2141 | enum zone_type i; | |
2142 | int j; | |
2143 | struct zonelist *zonelist; | |
2144 | ||
2145 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2146 | zonelist = pgdat->node_zonelists + MAX_NR_ZONES + i; | |
2147 | j = build_zonelists_node(pgdat, zonelist, 0, i); | |
2148 | zonelist->zones[j] = NULL; | |
2149 | } | |
2150 | } | |
2151 | ||
f0c0b2b8 KH |
2152 | /* |
2153 | * Build zonelists ordered by zone and nodes within zones. | |
2154 | * This results in conserving DMA zone[s] until all Normal memory is | |
2155 | * exhausted, but results in overflowing to remote node while memory | |
2156 | * may still exist in local DMA zone. | |
2157 | */ | |
2158 | static int node_order[MAX_NUMNODES]; | |
2159 | ||
2160 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
2161 | { | |
2162 | enum zone_type i; | |
2163 | int pos, j, node; | |
2164 | int zone_type; /* needs to be signed */ | |
2165 | struct zone *z; | |
2166 | struct zonelist *zonelist; | |
2167 | ||
2168 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2169 | zonelist = pgdat->node_zonelists + i; | |
2170 | pos = 0; | |
2171 | for (zone_type = i; zone_type >= 0; zone_type--) { | |
2172 | for (j = 0; j < nr_nodes; j++) { | |
2173 | node = node_order[j]; | |
2174 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
2175 | if (populated_zone(z)) { | |
2176 | zonelist->zones[pos++] = z; | |
2177 | check_highest_zone(zone_type); | |
2178 | } | |
2179 | } | |
2180 | } | |
2181 | zonelist->zones[pos] = NULL; | |
2182 | } | |
2183 | } | |
2184 | ||
2185 | static int default_zonelist_order(void) | |
2186 | { | |
2187 | int nid, zone_type; | |
2188 | unsigned long low_kmem_size,total_size; | |
2189 | struct zone *z; | |
2190 | int average_size; | |
2191 | /* | |
2192 | * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem. | |
2193 | * If they are really small and used heavily, the system can fall | |
2194 | * into OOM very easily. | |
2195 | * This function detect ZONE_DMA/DMA32 size and confgigures zone order. | |
2196 | */ | |
2197 | /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */ | |
2198 | low_kmem_size = 0; | |
2199 | total_size = 0; | |
2200 | for_each_online_node(nid) { | |
2201 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
2202 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
2203 | if (populated_zone(z)) { | |
2204 | if (zone_type < ZONE_NORMAL) | |
2205 | low_kmem_size += z->present_pages; | |
2206 | total_size += z->present_pages; | |
2207 | } | |
2208 | } | |
2209 | } | |
2210 | if (!low_kmem_size || /* there are no DMA area. */ | |
2211 | low_kmem_size > total_size/2) /* DMA/DMA32 is big. */ | |
2212 | return ZONELIST_ORDER_NODE; | |
2213 | /* | |
2214 | * look into each node's config. | |
2215 | * If there is a node whose DMA/DMA32 memory is very big area on | |
2216 | * local memory, NODE_ORDER may be suitable. | |
2217 | */ | |
37b07e41 LS |
2218 | average_size = total_size / |
2219 | (nodes_weight(node_states[N_HIGH_MEMORY]) + 1); | |
f0c0b2b8 KH |
2220 | for_each_online_node(nid) { |
2221 | low_kmem_size = 0; | |
2222 | total_size = 0; | |
2223 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
2224 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
2225 | if (populated_zone(z)) { | |
2226 | if (zone_type < ZONE_NORMAL) | |
2227 | low_kmem_size += z->present_pages; | |
2228 | total_size += z->present_pages; | |
2229 | } | |
2230 | } | |
2231 | if (low_kmem_size && | |
2232 | total_size > average_size && /* ignore small node */ | |
2233 | low_kmem_size > total_size * 70/100) | |
2234 | return ZONELIST_ORDER_NODE; | |
2235 | } | |
2236 | return ZONELIST_ORDER_ZONE; | |
2237 | } | |
2238 | ||
2239 | static void set_zonelist_order(void) | |
2240 | { | |
2241 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
2242 | current_zonelist_order = default_zonelist_order(); | |
2243 | else | |
2244 | current_zonelist_order = user_zonelist_order; | |
2245 | } | |
2246 | ||
2247 | static void build_zonelists(pg_data_t *pgdat) | |
2248 | { | |
2249 | int j, node, load; | |
2250 | enum zone_type i; | |
1da177e4 | 2251 | nodemask_t used_mask; |
f0c0b2b8 KH |
2252 | int local_node, prev_node; |
2253 | struct zonelist *zonelist; | |
2254 | int order = current_zonelist_order; | |
1da177e4 LT |
2255 | |
2256 | /* initialize zonelists */ | |
523b9458 | 2257 | for (i = 0; i < MAX_ZONELISTS; i++) { |
1da177e4 LT |
2258 | zonelist = pgdat->node_zonelists + i; |
2259 | zonelist->zones[0] = NULL; | |
2260 | } | |
2261 | ||
2262 | /* NUMA-aware ordering of nodes */ | |
2263 | local_node = pgdat->node_id; | |
2264 | load = num_online_nodes(); | |
2265 | prev_node = local_node; | |
2266 | nodes_clear(used_mask); | |
f0c0b2b8 KH |
2267 | |
2268 | memset(node_load, 0, sizeof(node_load)); | |
2269 | memset(node_order, 0, sizeof(node_order)); | |
2270 | j = 0; | |
2271 | ||
1da177e4 | 2272 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
9eeff239 CL |
2273 | int distance = node_distance(local_node, node); |
2274 | ||
2275 | /* | |
2276 | * If another node is sufficiently far away then it is better | |
2277 | * to reclaim pages in a zone before going off node. | |
2278 | */ | |
2279 | if (distance > RECLAIM_DISTANCE) | |
2280 | zone_reclaim_mode = 1; | |
2281 | ||
1da177e4 LT |
2282 | /* |
2283 | * We don't want to pressure a particular node. | |
2284 | * So adding penalty to the first node in same | |
2285 | * distance group to make it round-robin. | |
2286 | */ | |
9eeff239 | 2287 | if (distance != node_distance(local_node, prev_node)) |
f0c0b2b8 KH |
2288 | node_load[node] = load; |
2289 | ||
1da177e4 LT |
2290 | prev_node = node; |
2291 | load--; | |
f0c0b2b8 KH |
2292 | if (order == ZONELIST_ORDER_NODE) |
2293 | build_zonelists_in_node_order(pgdat, node); | |
2294 | else | |
2295 | node_order[j++] = node; /* remember order */ | |
2296 | } | |
1da177e4 | 2297 | |
f0c0b2b8 KH |
2298 | if (order == ZONELIST_ORDER_ZONE) { |
2299 | /* calculate node order -- i.e., DMA last! */ | |
2300 | build_zonelists_in_zone_order(pgdat, j); | |
1da177e4 | 2301 | } |
523b9458 CL |
2302 | |
2303 | build_thisnode_zonelists(pgdat); | |
1da177e4 LT |
2304 | } |
2305 | ||
9276b1bc | 2306 | /* Construct the zonelist performance cache - see further mmzone.h */ |
f0c0b2b8 | 2307 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc PJ |
2308 | { |
2309 | int i; | |
2310 | ||
2311 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2312 | struct zonelist *zonelist; | |
2313 | struct zonelist_cache *zlc; | |
2314 | struct zone **z; | |
2315 | ||
2316 | zonelist = pgdat->node_zonelists + i; | |
2317 | zonelist->zlcache_ptr = zlc = &zonelist->zlcache; | |
2318 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
2319 | for (z = zonelist->zones; *z; z++) | |
2320 | zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z); | |
2321 | } | |
2322 | } | |
2323 | ||
f0c0b2b8 | 2324 | |
1da177e4 LT |
2325 | #else /* CONFIG_NUMA */ |
2326 | ||
f0c0b2b8 KH |
2327 | static void set_zonelist_order(void) |
2328 | { | |
2329 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
2330 | } | |
2331 | ||
2332 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 2333 | { |
19655d34 CL |
2334 | int node, local_node; |
2335 | enum zone_type i,j; | |
1da177e4 LT |
2336 | |
2337 | local_node = pgdat->node_id; | |
19655d34 | 2338 | for (i = 0; i < MAX_NR_ZONES; i++) { |
1da177e4 LT |
2339 | struct zonelist *zonelist; |
2340 | ||
2341 | zonelist = pgdat->node_zonelists + i; | |
2342 | ||
19655d34 | 2343 | j = build_zonelists_node(pgdat, zonelist, 0, i); |
1da177e4 LT |
2344 | /* |
2345 | * Now we build the zonelist so that it contains the zones | |
2346 | * of all the other nodes. | |
2347 | * We don't want to pressure a particular node, so when | |
2348 | * building the zones for node N, we make sure that the | |
2349 | * zones coming right after the local ones are those from | |
2350 | * node N+1 (modulo N) | |
2351 | */ | |
2352 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
2353 | if (!node_online(node)) | |
2354 | continue; | |
19655d34 | 2355 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); |
1da177e4 LT |
2356 | } |
2357 | for (node = 0; node < local_node; node++) { | |
2358 | if (!node_online(node)) | |
2359 | continue; | |
19655d34 | 2360 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); |
1da177e4 LT |
2361 | } |
2362 | ||
2363 | zonelist->zones[j] = NULL; | |
2364 | } | |
2365 | } | |
2366 | ||
9276b1bc | 2367 | /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */ |
f0c0b2b8 | 2368 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc PJ |
2369 | { |
2370 | int i; | |
2371 | ||
2372 | for (i = 0; i < MAX_NR_ZONES; i++) | |
2373 | pgdat->node_zonelists[i].zlcache_ptr = NULL; | |
2374 | } | |
2375 | ||
1da177e4 LT |
2376 | #endif /* CONFIG_NUMA */ |
2377 | ||
6811378e | 2378 | /* return values int ....just for stop_machine_run() */ |
f0c0b2b8 | 2379 | static int __build_all_zonelists(void *dummy) |
1da177e4 | 2380 | { |
6811378e | 2381 | int nid; |
9276b1bc PJ |
2382 | |
2383 | for_each_online_node(nid) { | |
7ea1530a CL |
2384 | pg_data_t *pgdat = NODE_DATA(nid); |
2385 | ||
2386 | build_zonelists(pgdat); | |
2387 | build_zonelist_cache(pgdat); | |
9276b1bc | 2388 | } |
6811378e YG |
2389 | return 0; |
2390 | } | |
2391 | ||
f0c0b2b8 | 2392 | void build_all_zonelists(void) |
6811378e | 2393 | { |
f0c0b2b8 KH |
2394 | set_zonelist_order(); |
2395 | ||
6811378e | 2396 | if (system_state == SYSTEM_BOOTING) { |
423b41d7 | 2397 | __build_all_zonelists(NULL); |
6811378e YG |
2398 | cpuset_init_current_mems_allowed(); |
2399 | } else { | |
2400 | /* we have to stop all cpus to guaranntee there is no user | |
2401 | of zonelist */ | |
2402 | stop_machine_run(__build_all_zonelists, NULL, NR_CPUS); | |
2403 | /* cpuset refresh routine should be here */ | |
2404 | } | |
bd1e22b8 | 2405 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
2406 | /* |
2407 | * Disable grouping by mobility if the number of pages in the | |
2408 | * system is too low to allow the mechanism to work. It would be | |
2409 | * more accurate, but expensive to check per-zone. This check is | |
2410 | * made on memory-hotadd so a system can start with mobility | |
2411 | * disabled and enable it later | |
2412 | */ | |
2413 | if (vm_total_pages < (MAX_ORDER_NR_PAGES * MIGRATE_TYPES)) | |
2414 | page_group_by_mobility_disabled = 1; | |
2415 | else | |
2416 | page_group_by_mobility_disabled = 0; | |
2417 | ||
2418 | printk("Built %i zonelists in %s order, mobility grouping %s. " | |
2419 | "Total pages: %ld\n", | |
f0c0b2b8 KH |
2420 | num_online_nodes(), |
2421 | zonelist_order_name[current_zonelist_order], | |
9ef9acb0 | 2422 | page_group_by_mobility_disabled ? "off" : "on", |
f0c0b2b8 KH |
2423 | vm_total_pages); |
2424 | #ifdef CONFIG_NUMA | |
2425 | printk("Policy zone: %s\n", zone_names[policy_zone]); | |
2426 | #endif | |
1da177e4 LT |
2427 | } |
2428 | ||
2429 | /* | |
2430 | * Helper functions to size the waitqueue hash table. | |
2431 | * Essentially these want to choose hash table sizes sufficiently | |
2432 | * large so that collisions trying to wait on pages are rare. | |
2433 | * But in fact, the number of active page waitqueues on typical | |
2434 | * systems is ridiculously low, less than 200. So this is even | |
2435 | * conservative, even though it seems large. | |
2436 | * | |
2437 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
2438 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
2439 | */ | |
2440 | #define PAGES_PER_WAITQUEUE 256 | |
2441 | ||
cca448fe | 2442 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 2443 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
2444 | { |
2445 | unsigned long size = 1; | |
2446 | ||
2447 | pages /= PAGES_PER_WAITQUEUE; | |
2448 | ||
2449 | while (size < pages) | |
2450 | size <<= 1; | |
2451 | ||
2452 | /* | |
2453 | * Once we have dozens or even hundreds of threads sleeping | |
2454 | * on IO we've got bigger problems than wait queue collision. | |
2455 | * Limit the size of the wait table to a reasonable size. | |
2456 | */ | |
2457 | size = min(size, 4096UL); | |
2458 | ||
2459 | return max(size, 4UL); | |
2460 | } | |
cca448fe YG |
2461 | #else |
2462 | /* | |
2463 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
2464 | * a suitable size for its wait_table. So we use the maximum size now. | |
2465 | * | |
2466 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
2467 | * | |
2468 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
2469 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
2470 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
2471 | * | |
2472 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
2473 | * or more by the traditional way. (See above). It equals: | |
2474 | * | |
2475 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
2476 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
2477 | * powerpc (64K page size) : = (32G +16M)byte. | |
2478 | */ | |
2479 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
2480 | { | |
2481 | return 4096UL; | |
2482 | } | |
2483 | #endif | |
1da177e4 LT |
2484 | |
2485 | /* | |
2486 | * This is an integer logarithm so that shifts can be used later | |
2487 | * to extract the more random high bits from the multiplicative | |
2488 | * hash function before the remainder is taken. | |
2489 | */ | |
2490 | static inline unsigned long wait_table_bits(unsigned long size) | |
2491 | { | |
2492 | return ffz(~size); | |
2493 | } | |
2494 | ||
2495 | #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1)) | |
2496 | ||
1da177e4 LT |
2497 | /* |
2498 | * Initially all pages are reserved - free ones are freed | |
2499 | * up by free_all_bootmem() once the early boot process is | |
2500 | * done. Non-atomic initialization, single-pass. | |
2501 | */ | |
c09b4240 | 2502 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 2503 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 2504 | { |
1da177e4 | 2505 | struct page *page; |
29751f69 AW |
2506 | unsigned long end_pfn = start_pfn + size; |
2507 | unsigned long pfn; | |
1da177e4 | 2508 | |
cbe8dd4a | 2509 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 DH |
2510 | /* |
2511 | * There can be holes in boot-time mem_map[]s | |
2512 | * handed to this function. They do not | |
2513 | * exist on hotplugged memory. | |
2514 | */ | |
2515 | if (context == MEMMAP_EARLY) { | |
2516 | if (!early_pfn_valid(pfn)) | |
2517 | continue; | |
2518 | if (!early_pfn_in_nid(pfn, nid)) | |
2519 | continue; | |
2520 | } | |
d41dee36 AW |
2521 | page = pfn_to_page(pfn); |
2522 | set_page_links(page, zone, nid, pfn); | |
7835e98b | 2523 | init_page_count(page); |
1da177e4 LT |
2524 | reset_page_mapcount(page); |
2525 | SetPageReserved(page); | |
b2a0ac88 MG |
2526 | |
2527 | /* | |
2528 | * Mark the block movable so that blocks are reserved for | |
2529 | * movable at startup. This will force kernel allocations | |
2530 | * to reserve their blocks rather than leaking throughout | |
2531 | * the address space during boot when many long-lived | |
2532 | * kernel allocations are made | |
2533 | */ | |
2534 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
2535 | ||
1da177e4 LT |
2536 | INIT_LIST_HEAD(&page->lru); |
2537 | #ifdef WANT_PAGE_VIRTUAL | |
2538 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
2539 | if (!is_highmem_idx(zone)) | |
3212c6be | 2540 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1da177e4 | 2541 | #endif |
1da177e4 LT |
2542 | } |
2543 | } | |
2544 | ||
6ea6e688 PM |
2545 | static void __meminit zone_init_free_lists(struct pglist_data *pgdat, |
2546 | struct zone *zone, unsigned long size) | |
1da177e4 | 2547 | { |
b2a0ac88 MG |
2548 | int order, t; |
2549 | for_each_migratetype_order(order, t) { | |
2550 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
2551 | zone->free_area[order].nr_free = 0; |
2552 | } | |
2553 | } | |
2554 | ||
2555 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
2556 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 2557 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
2558 | #endif |
2559 | ||
d09c6b80 | 2560 | static int __devinit zone_batchsize(struct zone *zone) |
e7c8d5c9 CL |
2561 | { |
2562 | int batch; | |
2563 | ||
2564 | /* | |
2565 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 2566 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
2567 | * |
2568 | * OK, so we don't know how big the cache is. So guess. | |
2569 | */ | |
2570 | batch = zone->present_pages / 1024; | |
ba56e91c SR |
2571 | if (batch * PAGE_SIZE > 512 * 1024) |
2572 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
2573 | batch /= 4; /* We effectively *= 4 below */ |
2574 | if (batch < 1) | |
2575 | batch = 1; | |
2576 | ||
2577 | /* | |
0ceaacc9 NP |
2578 | * Clamp the batch to a 2^n - 1 value. Having a power |
2579 | * of 2 value was found to be more likely to have | |
2580 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 2581 | * |
0ceaacc9 NP |
2582 | * For example if 2 tasks are alternately allocating |
2583 | * batches of pages, one task can end up with a lot | |
2584 | * of pages of one half of the possible page colors | |
2585 | * and the other with pages of the other colors. | |
e7c8d5c9 | 2586 | */ |
0ceaacc9 | 2587 | batch = (1 << (fls(batch + batch/2)-1)) - 1; |
ba56e91c | 2588 | |
e7c8d5c9 CL |
2589 | return batch; |
2590 | } | |
2591 | ||
2caaad41 CL |
2592 | inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
2593 | { | |
2594 | struct per_cpu_pages *pcp; | |
2595 | ||
1c6fe946 MD |
2596 | memset(p, 0, sizeof(*p)); |
2597 | ||
2caaad41 CL |
2598 | pcp = &p->pcp[0]; /* hot */ |
2599 | pcp->count = 0; | |
2caaad41 CL |
2600 | pcp->high = 6 * batch; |
2601 | pcp->batch = max(1UL, 1 * batch); | |
2602 | INIT_LIST_HEAD(&pcp->list); | |
2603 | ||
2604 | pcp = &p->pcp[1]; /* cold*/ | |
2605 | pcp->count = 0; | |
2caaad41 | 2606 | pcp->high = 2 * batch; |
e46a5e28 | 2607 | pcp->batch = max(1UL, batch/2); |
2caaad41 CL |
2608 | INIT_LIST_HEAD(&pcp->list); |
2609 | } | |
2610 | ||
8ad4b1fb RS |
2611 | /* |
2612 | * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist | |
2613 | * to the value high for the pageset p. | |
2614 | */ | |
2615 | ||
2616 | static void setup_pagelist_highmark(struct per_cpu_pageset *p, | |
2617 | unsigned long high) | |
2618 | { | |
2619 | struct per_cpu_pages *pcp; | |
2620 | ||
2621 | pcp = &p->pcp[0]; /* hot list */ | |
2622 | pcp->high = high; | |
2623 | pcp->batch = max(1UL, high/4); | |
2624 | if ((high/4) > (PAGE_SHIFT * 8)) | |
2625 | pcp->batch = PAGE_SHIFT * 8; | |
2626 | } | |
2627 | ||
2628 | ||
e7c8d5c9 CL |
2629 | #ifdef CONFIG_NUMA |
2630 | /* | |
2caaad41 CL |
2631 | * Boot pageset table. One per cpu which is going to be used for all |
2632 | * zones and all nodes. The parameters will be set in such a way | |
2633 | * that an item put on a list will immediately be handed over to | |
2634 | * the buddy list. This is safe since pageset manipulation is done | |
2635 | * with interrupts disabled. | |
2636 | * | |
2637 | * Some NUMA counter updates may also be caught by the boot pagesets. | |
b7c84c6a CL |
2638 | * |
2639 | * The boot_pagesets must be kept even after bootup is complete for | |
2640 | * unused processors and/or zones. They do play a role for bootstrapping | |
2641 | * hotplugged processors. | |
2642 | * | |
2643 | * zoneinfo_show() and maybe other functions do | |
2644 | * not check if the processor is online before following the pageset pointer. | |
2645 | * Other parts of the kernel may not check if the zone is available. | |
2caaad41 | 2646 | */ |
88a2a4ac | 2647 | static struct per_cpu_pageset boot_pageset[NR_CPUS]; |
2caaad41 CL |
2648 | |
2649 | /* | |
2650 | * Dynamically allocate memory for the | |
e7c8d5c9 CL |
2651 | * per cpu pageset array in struct zone. |
2652 | */ | |
6292d9aa | 2653 | static int __cpuinit process_zones(int cpu) |
e7c8d5c9 CL |
2654 | { |
2655 | struct zone *zone, *dzone; | |
37c0708d CL |
2656 | int node = cpu_to_node(cpu); |
2657 | ||
2658 | node_set_state(node, N_CPU); /* this node has a cpu */ | |
e7c8d5c9 CL |
2659 | |
2660 | for_each_zone(zone) { | |
e7c8d5c9 | 2661 | |
66a55030 CL |
2662 | if (!populated_zone(zone)) |
2663 | continue; | |
2664 | ||
23316bc8 | 2665 | zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset), |
37c0708d | 2666 | GFP_KERNEL, node); |
23316bc8 | 2667 | if (!zone_pcp(zone, cpu)) |
e7c8d5c9 | 2668 | goto bad; |
e7c8d5c9 | 2669 | |
23316bc8 | 2670 | setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone)); |
8ad4b1fb RS |
2671 | |
2672 | if (percpu_pagelist_fraction) | |
2673 | setup_pagelist_highmark(zone_pcp(zone, cpu), | |
2674 | (zone->present_pages / percpu_pagelist_fraction)); | |
e7c8d5c9 CL |
2675 | } |
2676 | ||
2677 | return 0; | |
2678 | bad: | |
2679 | for_each_zone(dzone) { | |
64191688 AM |
2680 | if (!populated_zone(dzone)) |
2681 | continue; | |
e7c8d5c9 CL |
2682 | if (dzone == zone) |
2683 | break; | |
23316bc8 NP |
2684 | kfree(zone_pcp(dzone, cpu)); |
2685 | zone_pcp(dzone, cpu) = NULL; | |
e7c8d5c9 CL |
2686 | } |
2687 | return -ENOMEM; | |
2688 | } | |
2689 | ||
2690 | static inline void free_zone_pagesets(int cpu) | |
2691 | { | |
e7c8d5c9 CL |
2692 | struct zone *zone; |
2693 | ||
2694 | for_each_zone(zone) { | |
2695 | struct per_cpu_pageset *pset = zone_pcp(zone, cpu); | |
2696 | ||
f3ef9ead DR |
2697 | /* Free per_cpu_pageset if it is slab allocated */ |
2698 | if (pset != &boot_pageset[cpu]) | |
2699 | kfree(pset); | |
e7c8d5c9 | 2700 | zone_pcp(zone, cpu) = NULL; |
e7c8d5c9 | 2701 | } |
e7c8d5c9 CL |
2702 | } |
2703 | ||
9c7b216d | 2704 | static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb, |
e7c8d5c9 CL |
2705 | unsigned long action, |
2706 | void *hcpu) | |
2707 | { | |
2708 | int cpu = (long)hcpu; | |
2709 | int ret = NOTIFY_OK; | |
2710 | ||
2711 | switch (action) { | |
ce421c79 | 2712 | case CPU_UP_PREPARE: |
8bb78442 | 2713 | case CPU_UP_PREPARE_FROZEN: |
ce421c79 AW |
2714 | if (process_zones(cpu)) |
2715 | ret = NOTIFY_BAD; | |
2716 | break; | |
2717 | case CPU_UP_CANCELED: | |
8bb78442 | 2718 | case CPU_UP_CANCELED_FROZEN: |
ce421c79 | 2719 | case CPU_DEAD: |
8bb78442 | 2720 | case CPU_DEAD_FROZEN: |
ce421c79 AW |
2721 | free_zone_pagesets(cpu); |
2722 | break; | |
2723 | default: | |
2724 | break; | |
e7c8d5c9 CL |
2725 | } |
2726 | return ret; | |
2727 | } | |
2728 | ||
74b85f37 | 2729 | static struct notifier_block __cpuinitdata pageset_notifier = |
e7c8d5c9 CL |
2730 | { &pageset_cpuup_callback, NULL, 0 }; |
2731 | ||
78d9955b | 2732 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 CL |
2733 | { |
2734 | int err; | |
2735 | ||
2736 | /* Initialize per_cpu_pageset for cpu 0. | |
2737 | * A cpuup callback will do this for every cpu | |
2738 | * as it comes online | |
2739 | */ | |
2740 | err = process_zones(smp_processor_id()); | |
2741 | BUG_ON(err); | |
2742 | register_cpu_notifier(&pageset_notifier); | |
2743 | } | |
2744 | ||
2745 | #endif | |
2746 | ||
577a32f6 | 2747 | static noinline __init_refok |
cca448fe | 2748 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
2749 | { |
2750 | int i; | |
2751 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe | 2752 | size_t alloc_size; |
ed8ece2e DH |
2753 | |
2754 | /* | |
2755 | * The per-page waitqueue mechanism uses hashed waitqueues | |
2756 | * per zone. | |
2757 | */ | |
02b694de YG |
2758 | zone->wait_table_hash_nr_entries = |
2759 | wait_table_hash_nr_entries(zone_size_pages); | |
2760 | zone->wait_table_bits = | |
2761 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
2762 | alloc_size = zone->wait_table_hash_nr_entries |
2763 | * sizeof(wait_queue_head_t); | |
2764 | ||
2765 | if (system_state == SYSTEM_BOOTING) { | |
2766 | zone->wait_table = (wait_queue_head_t *) | |
2767 | alloc_bootmem_node(pgdat, alloc_size); | |
2768 | } else { | |
2769 | /* | |
2770 | * This case means that a zone whose size was 0 gets new memory | |
2771 | * via memory hot-add. | |
2772 | * But it may be the case that a new node was hot-added. In | |
2773 | * this case vmalloc() will not be able to use this new node's | |
2774 | * memory - this wait_table must be initialized to use this new | |
2775 | * node itself as well. | |
2776 | * To use this new node's memory, further consideration will be | |
2777 | * necessary. | |
2778 | */ | |
8691f3a7 | 2779 | zone->wait_table = vmalloc(alloc_size); |
cca448fe YG |
2780 | } |
2781 | if (!zone->wait_table) | |
2782 | return -ENOMEM; | |
ed8ece2e | 2783 | |
02b694de | 2784 | for(i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 2785 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
2786 | |
2787 | return 0; | |
ed8ece2e DH |
2788 | } |
2789 | ||
c09b4240 | 2790 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e DH |
2791 | { |
2792 | int cpu; | |
2793 | unsigned long batch = zone_batchsize(zone); | |
2794 | ||
2795 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
2796 | #ifdef CONFIG_NUMA | |
2797 | /* Early boot. Slab allocator not functional yet */ | |
23316bc8 | 2798 | zone_pcp(zone, cpu) = &boot_pageset[cpu]; |
ed8ece2e DH |
2799 | setup_pageset(&boot_pageset[cpu],0); |
2800 | #else | |
2801 | setup_pageset(zone_pcp(zone,cpu), batch); | |
2802 | #endif | |
2803 | } | |
f5335c0f AB |
2804 | if (zone->present_pages) |
2805 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n", | |
2806 | zone->name, zone->present_pages, batch); | |
ed8ece2e DH |
2807 | } |
2808 | ||
718127cc YG |
2809 | __meminit int init_currently_empty_zone(struct zone *zone, |
2810 | unsigned long zone_start_pfn, | |
a2f3aa02 DH |
2811 | unsigned long size, |
2812 | enum memmap_context context) | |
ed8ece2e DH |
2813 | { |
2814 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
2815 | int ret; |
2816 | ret = zone_wait_table_init(zone, size); | |
2817 | if (ret) | |
2818 | return ret; | |
ed8ece2e DH |
2819 | pgdat->nr_zones = zone_idx(zone) + 1; |
2820 | ||
ed8ece2e DH |
2821 | zone->zone_start_pfn = zone_start_pfn; |
2822 | ||
2823 | memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn); | |
2824 | ||
2825 | zone_init_free_lists(pgdat, zone, zone->spanned_pages); | |
718127cc YG |
2826 | |
2827 | return 0; | |
ed8ece2e DH |
2828 | } |
2829 | ||
c713216d MG |
2830 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
2831 | /* | |
2832 | * Basic iterator support. Return the first range of PFNs for a node | |
2833 | * Note: nid == MAX_NUMNODES returns first region regardless of node | |
2834 | */ | |
a3142c8e | 2835 | static int __meminit first_active_region_index_in_nid(int nid) |
c713216d MG |
2836 | { |
2837 | int i; | |
2838 | ||
2839 | for (i = 0; i < nr_nodemap_entries; i++) | |
2840 | if (nid == MAX_NUMNODES || early_node_map[i].nid == nid) | |
2841 | return i; | |
2842 | ||
2843 | return -1; | |
2844 | } | |
2845 | ||
2846 | /* | |
2847 | * Basic iterator support. Return the next active range of PFNs for a node | |
2848 | * Note: nid == MAX_NUMNODES returns next region regardles of node | |
2849 | */ | |
a3142c8e | 2850 | static int __meminit next_active_region_index_in_nid(int index, int nid) |
c713216d MG |
2851 | { |
2852 | for (index = index + 1; index < nr_nodemap_entries; index++) | |
2853 | if (nid == MAX_NUMNODES || early_node_map[index].nid == nid) | |
2854 | return index; | |
2855 | ||
2856 | return -1; | |
2857 | } | |
2858 | ||
2859 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID | |
2860 | /* | |
2861 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
2862 | * Architectures may implement their own version but if add_active_range() | |
2863 | * was used and there are no special requirements, this is a convenient | |
2864 | * alternative | |
2865 | */ | |
6f076f5d | 2866 | int __meminit early_pfn_to_nid(unsigned long pfn) |
c713216d MG |
2867 | { |
2868 | int i; | |
2869 | ||
2870 | for (i = 0; i < nr_nodemap_entries; i++) { | |
2871 | unsigned long start_pfn = early_node_map[i].start_pfn; | |
2872 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
2873 | ||
2874 | if (start_pfn <= pfn && pfn < end_pfn) | |
2875 | return early_node_map[i].nid; | |
2876 | } | |
2877 | ||
2878 | return 0; | |
2879 | } | |
2880 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
2881 | ||
2882 | /* Basic iterator support to walk early_node_map[] */ | |
2883 | #define for_each_active_range_index_in_nid(i, nid) \ | |
2884 | for (i = first_active_region_index_in_nid(nid); i != -1; \ | |
2885 | i = next_active_region_index_in_nid(i, nid)) | |
2886 | ||
2887 | /** | |
2888 | * free_bootmem_with_active_regions - Call free_bootmem_node for each active range | |
88ca3b94 RD |
2889 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
2890 | * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node | |
c713216d MG |
2891 | * |
2892 | * If an architecture guarantees that all ranges registered with | |
2893 | * add_active_ranges() contain no holes and may be freed, this | |
2894 | * this function may be used instead of calling free_bootmem() manually. | |
2895 | */ | |
2896 | void __init free_bootmem_with_active_regions(int nid, | |
2897 | unsigned long max_low_pfn) | |
2898 | { | |
2899 | int i; | |
2900 | ||
2901 | for_each_active_range_index_in_nid(i, nid) { | |
2902 | unsigned long size_pages = 0; | |
2903 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
2904 | ||
2905 | if (early_node_map[i].start_pfn >= max_low_pfn) | |
2906 | continue; | |
2907 | ||
2908 | if (end_pfn > max_low_pfn) | |
2909 | end_pfn = max_low_pfn; | |
2910 | ||
2911 | size_pages = end_pfn - early_node_map[i].start_pfn; | |
2912 | free_bootmem_node(NODE_DATA(early_node_map[i].nid), | |
2913 | PFN_PHYS(early_node_map[i].start_pfn), | |
2914 | size_pages << PAGE_SHIFT); | |
2915 | } | |
2916 | } | |
2917 | ||
2918 | /** | |
2919 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 2920 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d MG |
2921 | * |
2922 | * If an architecture guarantees that all ranges registered with | |
2923 | * add_active_ranges() contain no holes and may be freed, this | |
88ca3b94 | 2924 | * function may be used instead of calling memory_present() manually. |
c713216d MG |
2925 | */ |
2926 | void __init sparse_memory_present_with_active_regions(int nid) | |
2927 | { | |
2928 | int i; | |
2929 | ||
2930 | for_each_active_range_index_in_nid(i, nid) | |
2931 | memory_present(early_node_map[i].nid, | |
2932 | early_node_map[i].start_pfn, | |
2933 | early_node_map[i].end_pfn); | |
2934 | } | |
2935 | ||
fb01439c MG |
2936 | /** |
2937 | * push_node_boundaries - Push node boundaries to at least the requested boundary | |
2938 | * @nid: The nid of the node to push the boundary for | |
2939 | * @start_pfn: The start pfn of the node | |
2940 | * @end_pfn: The end pfn of the node | |
2941 | * | |
2942 | * In reserve-based hot-add, mem_map is allocated that is unused until hotadd | |
2943 | * time. Specifically, on x86_64, SRAT will report ranges that can potentially | |
2944 | * be hotplugged even though no physical memory exists. This function allows | |
2945 | * an arch to push out the node boundaries so mem_map is allocated that can | |
2946 | * be used later. | |
2947 | */ | |
2948 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE | |
2949 | void __init push_node_boundaries(unsigned int nid, | |
2950 | unsigned long start_pfn, unsigned long end_pfn) | |
2951 | { | |
2952 | printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n", | |
2953 | nid, start_pfn, end_pfn); | |
2954 | ||
2955 | /* Initialise the boundary for this node if necessary */ | |
2956 | if (node_boundary_end_pfn[nid] == 0) | |
2957 | node_boundary_start_pfn[nid] = -1UL; | |
2958 | ||
2959 | /* Update the boundaries */ | |
2960 | if (node_boundary_start_pfn[nid] > start_pfn) | |
2961 | node_boundary_start_pfn[nid] = start_pfn; | |
2962 | if (node_boundary_end_pfn[nid] < end_pfn) | |
2963 | node_boundary_end_pfn[nid] = end_pfn; | |
2964 | } | |
2965 | ||
2966 | /* If necessary, push the node boundary out for reserve hotadd */ | |
98011f56 | 2967 | static void __meminit account_node_boundary(unsigned int nid, |
fb01439c MG |
2968 | unsigned long *start_pfn, unsigned long *end_pfn) |
2969 | { | |
2970 | printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n", | |
2971 | nid, *start_pfn, *end_pfn); | |
2972 | ||
2973 | /* Return if boundary information has not been provided */ | |
2974 | if (node_boundary_end_pfn[nid] == 0) | |
2975 | return; | |
2976 | ||
2977 | /* Check the boundaries and update if necessary */ | |
2978 | if (node_boundary_start_pfn[nid] < *start_pfn) | |
2979 | *start_pfn = node_boundary_start_pfn[nid]; | |
2980 | if (node_boundary_end_pfn[nid] > *end_pfn) | |
2981 | *end_pfn = node_boundary_end_pfn[nid]; | |
2982 | } | |
2983 | #else | |
2984 | void __init push_node_boundaries(unsigned int nid, | |
2985 | unsigned long start_pfn, unsigned long end_pfn) {} | |
2986 | ||
98011f56 | 2987 | static void __meminit account_node_boundary(unsigned int nid, |
fb01439c MG |
2988 | unsigned long *start_pfn, unsigned long *end_pfn) {} |
2989 | #endif | |
2990 | ||
2991 | ||
c713216d MG |
2992 | /** |
2993 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
2994 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
2995 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
2996 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
2997 | * |
2998 | * It returns the start and end page frame of a node based on information | |
2999 | * provided by an arch calling add_active_range(). If called for a node | |
3000 | * with no available memory, a warning is printed and the start and end | |
88ca3b94 | 3001 | * PFNs will be 0. |
c713216d | 3002 | */ |
a3142c8e | 3003 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
3004 | unsigned long *start_pfn, unsigned long *end_pfn) |
3005 | { | |
3006 | int i; | |
3007 | *start_pfn = -1UL; | |
3008 | *end_pfn = 0; | |
3009 | ||
3010 | for_each_active_range_index_in_nid(i, nid) { | |
3011 | *start_pfn = min(*start_pfn, early_node_map[i].start_pfn); | |
3012 | *end_pfn = max(*end_pfn, early_node_map[i].end_pfn); | |
3013 | } | |
3014 | ||
633c0666 | 3015 | if (*start_pfn == -1UL) |
c713216d | 3016 | *start_pfn = 0; |
fb01439c MG |
3017 | |
3018 | /* Push the node boundaries out if requested */ | |
3019 | account_node_boundary(nid, start_pfn, end_pfn); | |
c713216d MG |
3020 | } |
3021 | ||
2a1e274a MG |
3022 | /* |
3023 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
3024 | * assumption is made that zones within a node are ordered in monotonic | |
3025 | * increasing memory addresses so that the "highest" populated zone is used | |
3026 | */ | |
3027 | void __init find_usable_zone_for_movable(void) | |
3028 | { | |
3029 | int zone_index; | |
3030 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
3031 | if (zone_index == ZONE_MOVABLE) | |
3032 | continue; | |
3033 | ||
3034 | if (arch_zone_highest_possible_pfn[zone_index] > | |
3035 | arch_zone_lowest_possible_pfn[zone_index]) | |
3036 | break; | |
3037 | } | |
3038 | ||
3039 | VM_BUG_ON(zone_index == -1); | |
3040 | movable_zone = zone_index; | |
3041 | } | |
3042 | ||
3043 | /* | |
3044 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
3045 | * because it is sized independant of architecture. Unlike the other zones, | |
3046 | * the starting point for ZONE_MOVABLE is not fixed. It may be different | |
3047 | * in each node depending on the size of each node and how evenly kernelcore | |
3048 | * is distributed. This helper function adjusts the zone ranges | |
3049 | * provided by the architecture for a given node by using the end of the | |
3050 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
3051 | * zones within a node are in order of monotonic increases memory addresses | |
3052 | */ | |
3053 | void __meminit adjust_zone_range_for_zone_movable(int nid, | |
3054 | unsigned long zone_type, | |
3055 | unsigned long node_start_pfn, | |
3056 | unsigned long node_end_pfn, | |
3057 | unsigned long *zone_start_pfn, | |
3058 | unsigned long *zone_end_pfn) | |
3059 | { | |
3060 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
3061 | if (zone_movable_pfn[nid]) { | |
3062 | /* Size ZONE_MOVABLE */ | |
3063 | if (zone_type == ZONE_MOVABLE) { | |
3064 | *zone_start_pfn = zone_movable_pfn[nid]; | |
3065 | *zone_end_pfn = min(node_end_pfn, | |
3066 | arch_zone_highest_possible_pfn[movable_zone]); | |
3067 | ||
3068 | /* Adjust for ZONE_MOVABLE starting within this range */ | |
3069 | } else if (*zone_start_pfn < zone_movable_pfn[nid] && | |
3070 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
3071 | *zone_end_pfn = zone_movable_pfn[nid]; | |
3072 | ||
3073 | /* Check if this whole range is within ZONE_MOVABLE */ | |
3074 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
3075 | *zone_start_pfn = *zone_end_pfn; | |
3076 | } | |
3077 | } | |
3078 | ||
c713216d MG |
3079 | /* |
3080 | * Return the number of pages a zone spans in a node, including holes | |
3081 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
3082 | */ | |
6ea6e688 | 3083 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
3084 | unsigned long zone_type, |
3085 | unsigned long *ignored) | |
3086 | { | |
3087 | unsigned long node_start_pfn, node_end_pfn; | |
3088 | unsigned long zone_start_pfn, zone_end_pfn; | |
3089 | ||
3090 | /* Get the start and end of the node and zone */ | |
3091 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
3092 | zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; | |
3093 | zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
3094 | adjust_zone_range_for_zone_movable(nid, zone_type, |
3095 | node_start_pfn, node_end_pfn, | |
3096 | &zone_start_pfn, &zone_end_pfn); | |
c713216d MG |
3097 | |
3098 | /* Check that this node has pages within the zone's required range */ | |
3099 | if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn) | |
3100 | return 0; | |
3101 | ||
3102 | /* Move the zone boundaries inside the node if necessary */ | |
3103 | zone_end_pfn = min(zone_end_pfn, node_end_pfn); | |
3104 | zone_start_pfn = max(zone_start_pfn, node_start_pfn); | |
3105 | ||
3106 | /* Return the spanned pages */ | |
3107 | return zone_end_pfn - zone_start_pfn; | |
3108 | } | |
3109 | ||
3110 | /* | |
3111 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 3112 | * then all holes in the requested range will be accounted for. |
c713216d | 3113 | */ |
a3142c8e | 3114 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
3115 | unsigned long range_start_pfn, |
3116 | unsigned long range_end_pfn) | |
3117 | { | |
3118 | int i = 0; | |
3119 | unsigned long prev_end_pfn = 0, hole_pages = 0; | |
3120 | unsigned long start_pfn; | |
3121 | ||
3122 | /* Find the end_pfn of the first active range of pfns in the node */ | |
3123 | i = first_active_region_index_in_nid(nid); | |
3124 | if (i == -1) | |
3125 | return 0; | |
3126 | ||
b5445f95 MG |
3127 | prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn); |
3128 | ||
9c7cd687 MG |
3129 | /* Account for ranges before physical memory on this node */ |
3130 | if (early_node_map[i].start_pfn > range_start_pfn) | |
b5445f95 | 3131 | hole_pages = prev_end_pfn - range_start_pfn; |
c713216d MG |
3132 | |
3133 | /* Find all holes for the zone within the node */ | |
3134 | for (; i != -1; i = next_active_region_index_in_nid(i, nid)) { | |
3135 | ||
3136 | /* No need to continue if prev_end_pfn is outside the zone */ | |
3137 | if (prev_end_pfn >= range_end_pfn) | |
3138 | break; | |
3139 | ||
3140 | /* Make sure the end of the zone is not within the hole */ | |
3141 | start_pfn = min(early_node_map[i].start_pfn, range_end_pfn); | |
3142 | prev_end_pfn = max(prev_end_pfn, range_start_pfn); | |
3143 | ||
3144 | /* Update the hole size cound and move on */ | |
3145 | if (start_pfn > range_start_pfn) { | |
3146 | BUG_ON(prev_end_pfn > start_pfn); | |
3147 | hole_pages += start_pfn - prev_end_pfn; | |
3148 | } | |
3149 | prev_end_pfn = early_node_map[i].end_pfn; | |
3150 | } | |
3151 | ||
9c7cd687 MG |
3152 | /* Account for ranges past physical memory on this node */ |
3153 | if (range_end_pfn > prev_end_pfn) | |
0c6cb974 | 3154 | hole_pages += range_end_pfn - |
9c7cd687 MG |
3155 | max(range_start_pfn, prev_end_pfn); |
3156 | ||
c713216d MG |
3157 | return hole_pages; |
3158 | } | |
3159 | ||
3160 | /** | |
3161 | * absent_pages_in_range - Return number of page frames in holes within a range | |
3162 | * @start_pfn: The start PFN to start searching for holes | |
3163 | * @end_pfn: The end PFN to stop searching for holes | |
3164 | * | |
88ca3b94 | 3165 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
3166 | */ |
3167 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
3168 | unsigned long end_pfn) | |
3169 | { | |
3170 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
3171 | } | |
3172 | ||
3173 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 3174 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
3175 | unsigned long zone_type, |
3176 | unsigned long *ignored) | |
3177 | { | |
9c7cd687 MG |
3178 | unsigned long node_start_pfn, node_end_pfn; |
3179 | unsigned long zone_start_pfn, zone_end_pfn; | |
3180 | ||
3181 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
3182 | zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type], | |
3183 | node_start_pfn); | |
3184 | zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type], | |
3185 | node_end_pfn); | |
3186 | ||
2a1e274a MG |
3187 | adjust_zone_range_for_zone_movable(nid, zone_type, |
3188 | node_start_pfn, node_end_pfn, | |
3189 | &zone_start_pfn, &zone_end_pfn); | |
9c7cd687 | 3190 | return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
c713216d | 3191 | } |
0e0b864e | 3192 | |
c713216d | 3193 | #else |
6ea6e688 | 3194 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
3195 | unsigned long zone_type, |
3196 | unsigned long *zones_size) | |
3197 | { | |
3198 | return zones_size[zone_type]; | |
3199 | } | |
3200 | ||
6ea6e688 | 3201 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
3202 | unsigned long zone_type, |
3203 | unsigned long *zholes_size) | |
3204 | { | |
3205 | if (!zholes_size) | |
3206 | return 0; | |
3207 | ||
3208 | return zholes_size[zone_type]; | |
3209 | } | |
0e0b864e | 3210 | |
c713216d MG |
3211 | #endif |
3212 | ||
a3142c8e | 3213 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
c713216d MG |
3214 | unsigned long *zones_size, unsigned long *zholes_size) |
3215 | { | |
3216 | unsigned long realtotalpages, totalpages = 0; | |
3217 | enum zone_type i; | |
3218 | ||
3219 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3220 | totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, | |
3221 | zones_size); | |
3222 | pgdat->node_spanned_pages = totalpages; | |
3223 | ||
3224 | realtotalpages = totalpages; | |
3225 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3226 | realtotalpages -= | |
3227 | zone_absent_pages_in_node(pgdat->node_id, i, | |
3228 | zholes_size); | |
3229 | pgdat->node_present_pages = realtotalpages; | |
3230 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
3231 | realtotalpages); | |
3232 | } | |
3233 | ||
835c134e MG |
3234 | #ifndef CONFIG_SPARSEMEM |
3235 | /* | |
3236 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
3237 | * Start by making sure zonesize is a multiple of MAX_ORDER-1 by rounding up | |
3238 | * Then figure 1 NR_PAGEBLOCK_BITS worth of bits per MAX_ORDER-1, finally | |
3239 | * round what is now in bits to nearest long in bits, then return it in | |
3240 | * bytes. | |
3241 | */ | |
3242 | static unsigned long __init usemap_size(unsigned long zonesize) | |
3243 | { | |
3244 | unsigned long usemapsize; | |
3245 | ||
3246 | usemapsize = roundup(zonesize, MAX_ORDER_NR_PAGES); | |
3247 | usemapsize = usemapsize >> (MAX_ORDER-1); | |
3248 | usemapsize *= NR_PAGEBLOCK_BITS; | |
3249 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
3250 | ||
3251 | return usemapsize / 8; | |
3252 | } | |
3253 | ||
3254 | static void __init setup_usemap(struct pglist_data *pgdat, | |
3255 | struct zone *zone, unsigned long zonesize) | |
3256 | { | |
3257 | unsigned long usemapsize = usemap_size(zonesize); | |
3258 | zone->pageblock_flags = NULL; | |
3259 | if (usemapsize) { | |
3260 | zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize); | |
3261 | memset(zone->pageblock_flags, 0, usemapsize); | |
3262 | } | |
3263 | } | |
3264 | #else | |
3265 | static void inline setup_usemap(struct pglist_data *pgdat, | |
3266 | struct zone *zone, unsigned long zonesize) {} | |
3267 | #endif /* CONFIG_SPARSEMEM */ | |
3268 | ||
1da177e4 LT |
3269 | /* |
3270 | * Set up the zone data structures: | |
3271 | * - mark all pages reserved | |
3272 | * - mark all memory queues empty | |
3273 | * - clear the memory bitmaps | |
3274 | */ | |
86356ab1 | 3275 | static void __meminit free_area_init_core(struct pglist_data *pgdat, |
1da177e4 LT |
3276 | unsigned long *zones_size, unsigned long *zholes_size) |
3277 | { | |
2f1b6248 | 3278 | enum zone_type j; |
ed8ece2e | 3279 | int nid = pgdat->node_id; |
1da177e4 | 3280 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
718127cc | 3281 | int ret; |
1da177e4 | 3282 | |
208d54e5 | 3283 | pgdat_resize_init(pgdat); |
1da177e4 LT |
3284 | pgdat->nr_zones = 0; |
3285 | init_waitqueue_head(&pgdat->kswapd_wait); | |
3286 | pgdat->kswapd_max_order = 0; | |
3287 | ||
3288 | for (j = 0; j < MAX_NR_ZONES; j++) { | |
3289 | struct zone *zone = pgdat->node_zones + j; | |
0e0b864e | 3290 | unsigned long size, realsize, memmap_pages; |
1da177e4 | 3291 | |
c713216d MG |
3292 | size = zone_spanned_pages_in_node(nid, j, zones_size); |
3293 | realsize = size - zone_absent_pages_in_node(nid, j, | |
3294 | zholes_size); | |
1da177e4 | 3295 | |
0e0b864e MG |
3296 | /* |
3297 | * Adjust realsize so that it accounts for how much memory | |
3298 | * is used by this zone for memmap. This affects the watermark | |
3299 | * and per-cpu initialisations | |
3300 | */ | |
3301 | memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT; | |
3302 | if (realsize >= memmap_pages) { | |
3303 | realsize -= memmap_pages; | |
3304 | printk(KERN_DEBUG | |
3305 | " %s zone: %lu pages used for memmap\n", | |
3306 | zone_names[j], memmap_pages); | |
3307 | } else | |
3308 | printk(KERN_WARNING | |
3309 | " %s zone: %lu pages exceeds realsize %lu\n", | |
3310 | zone_names[j], memmap_pages, realsize); | |
3311 | ||
6267276f CL |
3312 | /* Account for reserved pages */ |
3313 | if (j == 0 && realsize > dma_reserve) { | |
0e0b864e | 3314 | realsize -= dma_reserve; |
6267276f CL |
3315 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
3316 | zone_names[0], dma_reserve); | |
0e0b864e MG |
3317 | } |
3318 | ||
98d2b0eb | 3319 | if (!is_highmem_idx(j)) |
1da177e4 LT |
3320 | nr_kernel_pages += realsize; |
3321 | nr_all_pages += realsize; | |
3322 | ||
3323 | zone->spanned_pages = size; | |
3324 | zone->present_pages = realsize; | |
9614634f | 3325 | #ifdef CONFIG_NUMA |
d5f541ed | 3326 | zone->node = nid; |
8417bba4 | 3327 | zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio) |
9614634f | 3328 | / 100; |
0ff38490 | 3329 | zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100; |
9614634f | 3330 | #endif |
1da177e4 LT |
3331 | zone->name = zone_names[j]; |
3332 | spin_lock_init(&zone->lock); | |
3333 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 3334 | zone_seqlock_init(zone); |
1da177e4 | 3335 | zone->zone_pgdat = pgdat; |
1da177e4 | 3336 | |
3bb1a852 | 3337 | zone->prev_priority = DEF_PRIORITY; |
1da177e4 | 3338 | |
ed8ece2e | 3339 | zone_pcp_init(zone); |
1da177e4 LT |
3340 | INIT_LIST_HEAD(&zone->active_list); |
3341 | INIT_LIST_HEAD(&zone->inactive_list); | |
3342 | zone->nr_scan_active = 0; | |
3343 | zone->nr_scan_inactive = 0; | |
2244b95a | 3344 | zap_zone_vm_stats(zone); |
53e9a615 | 3345 | atomic_set(&zone->reclaim_in_progress, 0); |
1da177e4 LT |
3346 | if (!size) |
3347 | continue; | |
3348 | ||
835c134e | 3349 | setup_usemap(pgdat, zone, size); |
a2f3aa02 DH |
3350 | ret = init_currently_empty_zone(zone, zone_start_pfn, |
3351 | size, MEMMAP_EARLY); | |
718127cc | 3352 | BUG_ON(ret); |
1da177e4 | 3353 | zone_start_pfn += size; |
1da177e4 LT |
3354 | } |
3355 | } | |
3356 | ||
577a32f6 | 3357 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 3358 | { |
1da177e4 LT |
3359 | /* Skip empty nodes */ |
3360 | if (!pgdat->node_spanned_pages) | |
3361 | return; | |
3362 | ||
d41dee36 | 3363 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
3364 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
3365 | if (!pgdat->node_mem_map) { | |
e984bb43 | 3366 | unsigned long size, start, end; |
d41dee36 AW |
3367 | struct page *map; |
3368 | ||
e984bb43 BP |
3369 | /* |
3370 | * The zone's endpoints aren't required to be MAX_ORDER | |
3371 | * aligned but the node_mem_map endpoints must be in order | |
3372 | * for the buddy allocator to function correctly. | |
3373 | */ | |
3374 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
3375 | end = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
3376 | end = ALIGN(end, MAX_ORDER_NR_PAGES); | |
3377 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
3378 | map = alloc_remap(pgdat->node_id, size); |
3379 | if (!map) | |
3380 | map = alloc_bootmem_node(pgdat, size); | |
e984bb43 | 3381 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
1da177e4 | 3382 | } |
12d810c1 | 3383 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
3384 | /* |
3385 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
3386 | */ | |
c713216d | 3387 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 3388 | mem_map = NODE_DATA(0)->node_mem_map; |
c713216d MG |
3389 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
3390 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) | |
3391 | mem_map -= pgdat->node_start_pfn; | |
3392 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ | |
3393 | } | |
1da177e4 | 3394 | #endif |
d41dee36 | 3395 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
3396 | } |
3397 | ||
86356ab1 | 3398 | void __meminit free_area_init_node(int nid, struct pglist_data *pgdat, |
1da177e4 LT |
3399 | unsigned long *zones_size, unsigned long node_start_pfn, |
3400 | unsigned long *zholes_size) | |
3401 | { | |
3402 | pgdat->node_id = nid; | |
3403 | pgdat->node_start_pfn = node_start_pfn; | |
c713216d | 3404 | calculate_node_totalpages(pgdat, zones_size, zholes_size); |
1da177e4 LT |
3405 | |
3406 | alloc_node_mem_map(pgdat); | |
3407 | ||
3408 | free_area_init_core(pgdat, zones_size, zholes_size); | |
3409 | } | |
3410 | ||
c713216d | 3411 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
418508c1 MS |
3412 | |
3413 | #if MAX_NUMNODES > 1 | |
3414 | /* | |
3415 | * Figure out the number of possible node ids. | |
3416 | */ | |
3417 | static void __init setup_nr_node_ids(void) | |
3418 | { | |
3419 | unsigned int node; | |
3420 | unsigned int highest = 0; | |
3421 | ||
3422 | for_each_node_mask(node, node_possible_map) | |
3423 | highest = node; | |
3424 | nr_node_ids = highest + 1; | |
3425 | } | |
3426 | #else | |
3427 | static inline void setup_nr_node_ids(void) | |
3428 | { | |
3429 | } | |
3430 | #endif | |
3431 | ||
c713216d MG |
3432 | /** |
3433 | * add_active_range - Register a range of PFNs backed by physical memory | |
3434 | * @nid: The node ID the range resides on | |
3435 | * @start_pfn: The start PFN of the available physical memory | |
3436 | * @end_pfn: The end PFN of the available physical memory | |
3437 | * | |
3438 | * These ranges are stored in an early_node_map[] and later used by | |
3439 | * free_area_init_nodes() to calculate zone sizes and holes. If the | |
3440 | * range spans a memory hole, it is up to the architecture to ensure | |
3441 | * the memory is not freed by the bootmem allocator. If possible | |
3442 | * the range being registered will be merged with existing ranges. | |
3443 | */ | |
3444 | void __init add_active_range(unsigned int nid, unsigned long start_pfn, | |
3445 | unsigned long end_pfn) | |
3446 | { | |
3447 | int i; | |
3448 | ||
3449 | printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) " | |
3450 | "%d entries of %d used\n", | |
3451 | nid, start_pfn, end_pfn, | |
3452 | nr_nodemap_entries, MAX_ACTIVE_REGIONS); | |
3453 | ||
3454 | /* Merge with existing active regions if possible */ | |
3455 | for (i = 0; i < nr_nodemap_entries; i++) { | |
3456 | if (early_node_map[i].nid != nid) | |
3457 | continue; | |
3458 | ||
3459 | /* Skip if an existing region covers this new one */ | |
3460 | if (start_pfn >= early_node_map[i].start_pfn && | |
3461 | end_pfn <= early_node_map[i].end_pfn) | |
3462 | return; | |
3463 | ||
3464 | /* Merge forward if suitable */ | |
3465 | if (start_pfn <= early_node_map[i].end_pfn && | |
3466 | end_pfn > early_node_map[i].end_pfn) { | |
3467 | early_node_map[i].end_pfn = end_pfn; | |
3468 | return; | |
3469 | } | |
3470 | ||
3471 | /* Merge backward if suitable */ | |
3472 | if (start_pfn < early_node_map[i].end_pfn && | |
3473 | end_pfn >= early_node_map[i].start_pfn) { | |
3474 | early_node_map[i].start_pfn = start_pfn; | |
3475 | return; | |
3476 | } | |
3477 | } | |
3478 | ||
3479 | /* Check that early_node_map is large enough */ | |
3480 | if (i >= MAX_ACTIVE_REGIONS) { | |
3481 | printk(KERN_CRIT "More than %d memory regions, truncating\n", | |
3482 | MAX_ACTIVE_REGIONS); | |
3483 | return; | |
3484 | } | |
3485 | ||
3486 | early_node_map[i].nid = nid; | |
3487 | early_node_map[i].start_pfn = start_pfn; | |
3488 | early_node_map[i].end_pfn = end_pfn; | |
3489 | nr_nodemap_entries = i + 1; | |
3490 | } | |
3491 | ||
3492 | /** | |
3493 | * shrink_active_range - Shrink an existing registered range of PFNs | |
3494 | * @nid: The node id the range is on that should be shrunk | |
3495 | * @old_end_pfn: The old end PFN of the range | |
3496 | * @new_end_pfn: The new PFN of the range | |
3497 | * | |
3498 | * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node. | |
3499 | * The map is kept at the end physical page range that has already been | |
3500 | * registered with add_active_range(). This function allows an arch to shrink | |
3501 | * an existing registered range. | |
3502 | */ | |
3503 | void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn, | |
3504 | unsigned long new_end_pfn) | |
3505 | { | |
3506 | int i; | |
3507 | ||
3508 | /* Find the old active region end and shrink */ | |
3509 | for_each_active_range_index_in_nid(i, nid) | |
3510 | if (early_node_map[i].end_pfn == old_end_pfn) { | |
3511 | early_node_map[i].end_pfn = new_end_pfn; | |
3512 | break; | |
3513 | } | |
3514 | } | |
3515 | ||
3516 | /** | |
3517 | * remove_all_active_ranges - Remove all currently registered regions | |
88ca3b94 | 3518 | * |
c713216d MG |
3519 | * During discovery, it may be found that a table like SRAT is invalid |
3520 | * and an alternative discovery method must be used. This function removes | |
3521 | * all currently registered regions. | |
3522 | */ | |
88ca3b94 | 3523 | void __init remove_all_active_ranges(void) |
c713216d MG |
3524 | { |
3525 | memset(early_node_map, 0, sizeof(early_node_map)); | |
3526 | nr_nodemap_entries = 0; | |
fb01439c MG |
3527 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE |
3528 | memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn)); | |
3529 | memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn)); | |
3530 | #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */ | |
c713216d MG |
3531 | } |
3532 | ||
3533 | /* Compare two active node_active_regions */ | |
3534 | static int __init cmp_node_active_region(const void *a, const void *b) | |
3535 | { | |
3536 | struct node_active_region *arange = (struct node_active_region *)a; | |
3537 | struct node_active_region *brange = (struct node_active_region *)b; | |
3538 | ||
3539 | /* Done this way to avoid overflows */ | |
3540 | if (arange->start_pfn > brange->start_pfn) | |
3541 | return 1; | |
3542 | if (arange->start_pfn < brange->start_pfn) | |
3543 | return -1; | |
3544 | ||
3545 | return 0; | |
3546 | } | |
3547 | ||
3548 | /* sort the node_map by start_pfn */ | |
3549 | static void __init sort_node_map(void) | |
3550 | { | |
3551 | sort(early_node_map, (size_t)nr_nodemap_entries, | |
3552 | sizeof(struct node_active_region), | |
3553 | cmp_node_active_region, NULL); | |
3554 | } | |
3555 | ||
a6af2bc3 | 3556 | /* Find the lowest pfn for a node */ |
c713216d MG |
3557 | unsigned long __init find_min_pfn_for_node(unsigned long nid) |
3558 | { | |
3559 | int i; | |
a6af2bc3 | 3560 | unsigned long min_pfn = ULONG_MAX; |
1abbfb41 | 3561 | |
c713216d MG |
3562 | /* Assuming a sorted map, the first range found has the starting pfn */ |
3563 | for_each_active_range_index_in_nid(i, nid) | |
a6af2bc3 | 3564 | min_pfn = min(min_pfn, early_node_map[i].start_pfn); |
c713216d | 3565 | |
a6af2bc3 MG |
3566 | if (min_pfn == ULONG_MAX) { |
3567 | printk(KERN_WARNING | |
3568 | "Could not find start_pfn for node %lu\n", nid); | |
3569 | return 0; | |
3570 | } | |
3571 | ||
3572 | return min_pfn; | |
c713216d MG |
3573 | } |
3574 | ||
3575 | /** | |
3576 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
3577 | * | |
3578 | * It returns the minimum PFN based on information provided via | |
88ca3b94 | 3579 | * add_active_range(). |
c713216d MG |
3580 | */ |
3581 | unsigned long __init find_min_pfn_with_active_regions(void) | |
3582 | { | |
3583 | return find_min_pfn_for_node(MAX_NUMNODES); | |
3584 | } | |
3585 | ||
3586 | /** | |
3587 | * find_max_pfn_with_active_regions - Find the maximum PFN registered | |
3588 | * | |
3589 | * It returns the maximum PFN based on information provided via | |
88ca3b94 | 3590 | * add_active_range(). |
c713216d MG |
3591 | */ |
3592 | unsigned long __init find_max_pfn_with_active_regions(void) | |
3593 | { | |
3594 | int i; | |
3595 | unsigned long max_pfn = 0; | |
3596 | ||
3597 | for (i = 0; i < nr_nodemap_entries; i++) | |
3598 | max_pfn = max(max_pfn, early_node_map[i].end_pfn); | |
3599 | ||
3600 | return max_pfn; | |
3601 | } | |
3602 | ||
37b07e41 LS |
3603 | /* |
3604 | * early_calculate_totalpages() | |
3605 | * Sum pages in active regions for movable zone. | |
3606 | * Populate N_HIGH_MEMORY for calculating usable_nodes. | |
3607 | */ | |
7e63efef MG |
3608 | unsigned long __init early_calculate_totalpages(void) |
3609 | { | |
3610 | int i; | |
3611 | unsigned long totalpages = 0; | |
3612 | ||
37b07e41 LS |
3613 | for (i = 0; i < nr_nodemap_entries; i++) { |
3614 | unsigned long pages = early_node_map[i].end_pfn - | |
7e63efef | 3615 | early_node_map[i].start_pfn; |
37b07e41 LS |
3616 | totalpages += pages; |
3617 | if (pages) | |
3618 | node_set_state(early_node_map[i].nid, N_HIGH_MEMORY); | |
3619 | } | |
3620 | return totalpages; | |
7e63efef MG |
3621 | } |
3622 | ||
2a1e274a MG |
3623 | /* |
3624 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
3625 | * is spread evenly between nodes as long as the nodes have enough | |
3626 | * memory. When they don't, some nodes will have more kernelcore than | |
3627 | * others | |
3628 | */ | |
3629 | void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn) | |
3630 | { | |
3631 | int i, nid; | |
3632 | unsigned long usable_startpfn; | |
3633 | unsigned long kernelcore_node, kernelcore_remaining; | |
37b07e41 LS |
3634 | unsigned long totalpages = early_calculate_totalpages(); |
3635 | int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]); | |
2a1e274a | 3636 | |
7e63efef MG |
3637 | /* |
3638 | * If movablecore was specified, calculate what size of | |
3639 | * kernelcore that corresponds so that memory usable for | |
3640 | * any allocation type is evenly spread. If both kernelcore | |
3641 | * and movablecore are specified, then the value of kernelcore | |
3642 | * will be used for required_kernelcore if it's greater than | |
3643 | * what movablecore would have allowed. | |
3644 | */ | |
3645 | if (required_movablecore) { | |
7e63efef MG |
3646 | unsigned long corepages; |
3647 | ||
3648 | /* | |
3649 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
3650 | * was requested by the user | |
3651 | */ | |
3652 | required_movablecore = | |
3653 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
3654 | corepages = totalpages - required_movablecore; | |
3655 | ||
3656 | required_kernelcore = max(required_kernelcore, corepages); | |
3657 | } | |
3658 | ||
2a1e274a MG |
3659 | /* If kernelcore was not specified, there is no ZONE_MOVABLE */ |
3660 | if (!required_kernelcore) | |
3661 | return; | |
3662 | ||
3663 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
3664 | find_usable_zone_for_movable(); | |
3665 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; | |
3666 | ||
3667 | restart: | |
3668 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
3669 | kernelcore_node = required_kernelcore / usable_nodes; | |
37b07e41 | 3670 | for_each_node_state(nid, N_HIGH_MEMORY) { |
2a1e274a MG |
3671 | /* |
3672 | * Recalculate kernelcore_node if the division per node | |
3673 | * now exceeds what is necessary to satisfy the requested | |
3674 | * amount of memory for the kernel | |
3675 | */ | |
3676 | if (required_kernelcore < kernelcore_node) | |
3677 | kernelcore_node = required_kernelcore / usable_nodes; | |
3678 | ||
3679 | /* | |
3680 | * As the map is walked, we track how much memory is usable | |
3681 | * by the kernel using kernelcore_remaining. When it is | |
3682 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
3683 | */ | |
3684 | kernelcore_remaining = kernelcore_node; | |
3685 | ||
3686 | /* Go through each range of PFNs within this node */ | |
3687 | for_each_active_range_index_in_nid(i, nid) { | |
3688 | unsigned long start_pfn, end_pfn; | |
3689 | unsigned long size_pages; | |
3690 | ||
3691 | start_pfn = max(early_node_map[i].start_pfn, | |
3692 | zone_movable_pfn[nid]); | |
3693 | end_pfn = early_node_map[i].end_pfn; | |
3694 | if (start_pfn >= end_pfn) | |
3695 | continue; | |
3696 | ||
3697 | /* Account for what is only usable for kernelcore */ | |
3698 | if (start_pfn < usable_startpfn) { | |
3699 | unsigned long kernel_pages; | |
3700 | kernel_pages = min(end_pfn, usable_startpfn) | |
3701 | - start_pfn; | |
3702 | ||
3703 | kernelcore_remaining -= min(kernel_pages, | |
3704 | kernelcore_remaining); | |
3705 | required_kernelcore -= min(kernel_pages, | |
3706 | required_kernelcore); | |
3707 | ||
3708 | /* Continue if range is now fully accounted */ | |
3709 | if (end_pfn <= usable_startpfn) { | |
3710 | ||
3711 | /* | |
3712 | * Push zone_movable_pfn to the end so | |
3713 | * that if we have to rebalance | |
3714 | * kernelcore across nodes, we will | |
3715 | * not double account here | |
3716 | */ | |
3717 | zone_movable_pfn[nid] = end_pfn; | |
3718 | continue; | |
3719 | } | |
3720 | start_pfn = usable_startpfn; | |
3721 | } | |
3722 | ||
3723 | /* | |
3724 | * The usable PFN range for ZONE_MOVABLE is from | |
3725 | * start_pfn->end_pfn. Calculate size_pages as the | |
3726 | * number of pages used as kernelcore | |
3727 | */ | |
3728 | size_pages = end_pfn - start_pfn; | |
3729 | if (size_pages > kernelcore_remaining) | |
3730 | size_pages = kernelcore_remaining; | |
3731 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
3732 | ||
3733 | /* | |
3734 | * Some kernelcore has been met, update counts and | |
3735 | * break if the kernelcore for this node has been | |
3736 | * satisified | |
3737 | */ | |
3738 | required_kernelcore -= min(required_kernelcore, | |
3739 | size_pages); | |
3740 | kernelcore_remaining -= size_pages; | |
3741 | if (!kernelcore_remaining) | |
3742 | break; | |
3743 | } | |
3744 | } | |
3745 | ||
3746 | /* | |
3747 | * If there is still required_kernelcore, we do another pass with one | |
3748 | * less node in the count. This will push zone_movable_pfn[nid] further | |
3749 | * along on the nodes that still have memory until kernelcore is | |
3750 | * satisified | |
3751 | */ | |
3752 | usable_nodes--; | |
3753 | if (usable_nodes && required_kernelcore > usable_nodes) | |
3754 | goto restart; | |
3755 | ||
3756 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ | |
3757 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
3758 | zone_movable_pfn[nid] = | |
3759 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
3760 | } | |
3761 | ||
37b07e41 LS |
3762 | /* Any regular memory on that node ? */ |
3763 | static void check_for_regular_memory(pg_data_t *pgdat) | |
3764 | { | |
3765 | #ifdef CONFIG_HIGHMEM | |
3766 | enum zone_type zone_type; | |
3767 | ||
3768 | for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) { | |
3769 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
3770 | if (zone->present_pages) | |
3771 | node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY); | |
3772 | } | |
3773 | #endif | |
3774 | } | |
3775 | ||
c713216d MG |
3776 | /** |
3777 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 3778 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
3779 | * |
3780 | * This will call free_area_init_node() for each active node in the system. | |
3781 | * Using the page ranges provided by add_active_range(), the size of each | |
3782 | * zone in each node and their holes is calculated. If the maximum PFN | |
3783 | * between two adjacent zones match, it is assumed that the zone is empty. | |
3784 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
3785 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
3786 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
3787 | * at arch_max_dma_pfn. | |
3788 | */ | |
3789 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
3790 | { | |
3791 | unsigned long nid; | |
3792 | enum zone_type i; | |
3793 | ||
a6af2bc3 MG |
3794 | /* Sort early_node_map as initialisation assumes it is sorted */ |
3795 | sort_node_map(); | |
3796 | ||
c713216d MG |
3797 | /* Record where the zone boundaries are */ |
3798 | memset(arch_zone_lowest_possible_pfn, 0, | |
3799 | sizeof(arch_zone_lowest_possible_pfn)); | |
3800 | memset(arch_zone_highest_possible_pfn, 0, | |
3801 | sizeof(arch_zone_highest_possible_pfn)); | |
3802 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
3803 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
3804 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
3805 | if (i == ZONE_MOVABLE) |
3806 | continue; | |
c713216d MG |
3807 | arch_zone_lowest_possible_pfn[i] = |
3808 | arch_zone_highest_possible_pfn[i-1]; | |
3809 | arch_zone_highest_possible_pfn[i] = | |
3810 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
3811 | } | |
2a1e274a MG |
3812 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
3813 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; | |
3814 | ||
3815 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
3816 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
3817 | find_zone_movable_pfns_for_nodes(zone_movable_pfn); | |
c713216d | 3818 | |
c713216d MG |
3819 | /* Print out the zone ranges */ |
3820 | printk("Zone PFN ranges:\n"); | |
2a1e274a MG |
3821 | for (i = 0; i < MAX_NR_ZONES; i++) { |
3822 | if (i == ZONE_MOVABLE) | |
3823 | continue; | |
c713216d MG |
3824 | printk(" %-8s %8lu -> %8lu\n", |
3825 | zone_names[i], | |
3826 | arch_zone_lowest_possible_pfn[i], | |
3827 | arch_zone_highest_possible_pfn[i]); | |
2a1e274a MG |
3828 | } |
3829 | ||
3830 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
3831 | printk("Movable zone start PFN for each node\n"); | |
3832 | for (i = 0; i < MAX_NUMNODES; i++) { | |
3833 | if (zone_movable_pfn[i]) | |
3834 | printk(" Node %d: %lu\n", i, zone_movable_pfn[i]); | |
3835 | } | |
c713216d MG |
3836 | |
3837 | /* Print out the early_node_map[] */ | |
3838 | printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries); | |
3839 | for (i = 0; i < nr_nodemap_entries; i++) | |
3840 | printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid, | |
3841 | early_node_map[i].start_pfn, | |
3842 | early_node_map[i].end_pfn); | |
3843 | ||
3844 | /* Initialise every node */ | |
8ef82866 | 3845 | setup_nr_node_ids(); |
c713216d MG |
3846 | for_each_online_node(nid) { |
3847 | pg_data_t *pgdat = NODE_DATA(nid); | |
3848 | free_area_init_node(nid, pgdat, NULL, | |
3849 | find_min_pfn_for_node(nid), NULL); | |
37b07e41 LS |
3850 | |
3851 | /* Any memory on that node */ | |
3852 | if (pgdat->node_present_pages) | |
3853 | node_set_state(nid, N_HIGH_MEMORY); | |
3854 | check_for_regular_memory(pgdat); | |
c713216d MG |
3855 | } |
3856 | } | |
2a1e274a | 3857 | |
7e63efef | 3858 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
3859 | { |
3860 | unsigned long long coremem; | |
3861 | if (!p) | |
3862 | return -EINVAL; | |
3863 | ||
3864 | coremem = memparse(p, &p); | |
7e63efef | 3865 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 3866 | |
7e63efef | 3867 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
3868 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
3869 | ||
3870 | return 0; | |
3871 | } | |
ed7ed365 | 3872 | |
7e63efef MG |
3873 | /* |
3874 | * kernelcore=size sets the amount of memory for use for allocations that | |
3875 | * cannot be reclaimed or migrated. | |
3876 | */ | |
3877 | static int __init cmdline_parse_kernelcore(char *p) | |
3878 | { | |
3879 | return cmdline_parse_core(p, &required_kernelcore); | |
3880 | } | |
3881 | ||
3882 | /* | |
3883 | * movablecore=size sets the amount of memory for use for allocations that | |
3884 | * can be reclaimed or migrated. | |
3885 | */ | |
3886 | static int __init cmdline_parse_movablecore(char *p) | |
3887 | { | |
3888 | return cmdline_parse_core(p, &required_movablecore); | |
3889 | } | |
3890 | ||
ed7ed365 | 3891 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 3892 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 3893 | |
c713216d MG |
3894 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
3895 | ||
0e0b864e | 3896 | /** |
88ca3b94 RD |
3897 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
3898 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e MG |
3899 | * |
3900 | * The per-cpu batchsize and zone watermarks are determined by present_pages. | |
3901 | * In the DMA zone, a significant percentage may be consumed by kernel image | |
3902 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
3903 | * function may optionally be used to account for unfreeable pages in the |
3904 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
3905 | * smaller per-cpu batchsize. | |
0e0b864e MG |
3906 | */ |
3907 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
3908 | { | |
3909 | dma_reserve = new_dma_reserve; | |
3910 | } | |
3911 | ||
93b7504e | 3912 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
3913 | static bootmem_data_t contig_bootmem_data; |
3914 | struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data }; | |
3915 | ||
3916 | EXPORT_SYMBOL(contig_page_data); | |
93b7504e | 3917 | #endif |
1da177e4 LT |
3918 | |
3919 | void __init free_area_init(unsigned long *zones_size) | |
3920 | { | |
93b7504e | 3921 | free_area_init_node(0, NODE_DATA(0), zones_size, |
1da177e4 LT |
3922 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
3923 | } | |
1da177e4 | 3924 | |
1da177e4 LT |
3925 | static int page_alloc_cpu_notify(struct notifier_block *self, |
3926 | unsigned long action, void *hcpu) | |
3927 | { | |
3928 | int cpu = (unsigned long)hcpu; | |
1da177e4 | 3929 | |
8bb78442 | 3930 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
1da177e4 LT |
3931 | local_irq_disable(); |
3932 | __drain_pages(cpu); | |
f8891e5e | 3933 | vm_events_fold_cpu(cpu); |
1da177e4 | 3934 | local_irq_enable(); |
2244b95a | 3935 | refresh_cpu_vm_stats(cpu); |
1da177e4 LT |
3936 | } |
3937 | return NOTIFY_OK; | |
3938 | } | |
1da177e4 LT |
3939 | |
3940 | void __init page_alloc_init(void) | |
3941 | { | |
3942 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
3943 | } | |
3944 | ||
cb45b0e9 HA |
3945 | /* |
3946 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio | |
3947 | * or min_free_kbytes changes. | |
3948 | */ | |
3949 | static void calculate_totalreserve_pages(void) | |
3950 | { | |
3951 | struct pglist_data *pgdat; | |
3952 | unsigned long reserve_pages = 0; | |
2f6726e5 | 3953 | enum zone_type i, j; |
cb45b0e9 HA |
3954 | |
3955 | for_each_online_pgdat(pgdat) { | |
3956 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
3957 | struct zone *zone = pgdat->node_zones + i; | |
3958 | unsigned long max = 0; | |
3959 | ||
3960 | /* Find valid and maximum lowmem_reserve in the zone */ | |
3961 | for (j = i; j < MAX_NR_ZONES; j++) { | |
3962 | if (zone->lowmem_reserve[j] > max) | |
3963 | max = zone->lowmem_reserve[j]; | |
3964 | } | |
3965 | ||
3966 | /* we treat pages_high as reserved pages. */ | |
3967 | max += zone->pages_high; | |
3968 | ||
3969 | if (max > zone->present_pages) | |
3970 | max = zone->present_pages; | |
3971 | reserve_pages += max; | |
3972 | } | |
3973 | } | |
3974 | totalreserve_pages = reserve_pages; | |
3975 | } | |
3976 | ||
1da177e4 LT |
3977 | /* |
3978 | * setup_per_zone_lowmem_reserve - called whenever | |
3979 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
3980 | * has a correct pages reserved value, so an adequate number of | |
3981 | * pages are left in the zone after a successful __alloc_pages(). | |
3982 | */ | |
3983 | static void setup_per_zone_lowmem_reserve(void) | |
3984 | { | |
3985 | struct pglist_data *pgdat; | |
2f6726e5 | 3986 | enum zone_type j, idx; |
1da177e4 | 3987 | |
ec936fc5 | 3988 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
3989 | for (j = 0; j < MAX_NR_ZONES; j++) { |
3990 | struct zone *zone = pgdat->node_zones + j; | |
3991 | unsigned long present_pages = zone->present_pages; | |
3992 | ||
3993 | zone->lowmem_reserve[j] = 0; | |
3994 | ||
2f6726e5 CL |
3995 | idx = j; |
3996 | while (idx) { | |
1da177e4 LT |
3997 | struct zone *lower_zone; |
3998 | ||
2f6726e5 CL |
3999 | idx--; |
4000 | ||
1da177e4 LT |
4001 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
4002 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
4003 | ||
4004 | lower_zone = pgdat->node_zones + idx; | |
4005 | lower_zone->lowmem_reserve[j] = present_pages / | |
4006 | sysctl_lowmem_reserve_ratio[idx]; | |
4007 | present_pages += lower_zone->present_pages; | |
4008 | } | |
4009 | } | |
4010 | } | |
cb45b0e9 HA |
4011 | |
4012 | /* update totalreserve_pages */ | |
4013 | calculate_totalreserve_pages(); | |
1da177e4 LT |
4014 | } |
4015 | ||
88ca3b94 RD |
4016 | /** |
4017 | * setup_per_zone_pages_min - called when min_free_kbytes changes. | |
4018 | * | |
4019 | * Ensures that the pages_{min,low,high} values for each zone are set correctly | |
4020 | * with respect to min_free_kbytes. | |
1da177e4 | 4021 | */ |
3947be19 | 4022 | void setup_per_zone_pages_min(void) |
1da177e4 LT |
4023 | { |
4024 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
4025 | unsigned long lowmem_pages = 0; | |
4026 | struct zone *zone; | |
4027 | unsigned long flags; | |
4028 | ||
4029 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
4030 | for_each_zone(zone) { | |
4031 | if (!is_highmem(zone)) | |
4032 | lowmem_pages += zone->present_pages; | |
4033 | } | |
4034 | ||
4035 | for_each_zone(zone) { | |
ac924c60 AM |
4036 | u64 tmp; |
4037 | ||
1da177e4 | 4038 | spin_lock_irqsave(&zone->lru_lock, flags); |
ac924c60 AM |
4039 | tmp = (u64)pages_min * zone->present_pages; |
4040 | do_div(tmp, lowmem_pages); | |
1da177e4 LT |
4041 | if (is_highmem(zone)) { |
4042 | /* | |
669ed175 NP |
4043 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
4044 | * need highmem pages, so cap pages_min to a small | |
4045 | * value here. | |
4046 | * | |
4047 | * The (pages_high-pages_low) and (pages_low-pages_min) | |
4048 | * deltas controls asynch page reclaim, and so should | |
4049 | * not be capped for highmem. | |
1da177e4 LT |
4050 | */ |
4051 | int min_pages; | |
4052 | ||
4053 | min_pages = zone->present_pages / 1024; | |
4054 | if (min_pages < SWAP_CLUSTER_MAX) | |
4055 | min_pages = SWAP_CLUSTER_MAX; | |
4056 | if (min_pages > 128) | |
4057 | min_pages = 128; | |
4058 | zone->pages_min = min_pages; | |
4059 | } else { | |
669ed175 NP |
4060 | /* |
4061 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
4062 | * proportionate to the zone's size. |
4063 | */ | |
669ed175 | 4064 | zone->pages_min = tmp; |
1da177e4 LT |
4065 | } |
4066 | ||
ac924c60 AM |
4067 | zone->pages_low = zone->pages_min + (tmp >> 2); |
4068 | zone->pages_high = zone->pages_min + (tmp >> 1); | |
1da177e4 LT |
4069 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
4070 | } | |
cb45b0e9 HA |
4071 | |
4072 | /* update totalreserve_pages */ | |
4073 | calculate_totalreserve_pages(); | |
1da177e4 LT |
4074 | } |
4075 | ||
4076 | /* | |
4077 | * Initialise min_free_kbytes. | |
4078 | * | |
4079 | * For small machines we want it small (128k min). For large machines | |
4080 | * we want it large (64MB max). But it is not linear, because network | |
4081 | * bandwidth does not increase linearly with machine size. We use | |
4082 | * | |
4083 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: | |
4084 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) | |
4085 | * | |
4086 | * which yields | |
4087 | * | |
4088 | * 16MB: 512k | |
4089 | * 32MB: 724k | |
4090 | * 64MB: 1024k | |
4091 | * 128MB: 1448k | |
4092 | * 256MB: 2048k | |
4093 | * 512MB: 2896k | |
4094 | * 1024MB: 4096k | |
4095 | * 2048MB: 5792k | |
4096 | * 4096MB: 8192k | |
4097 | * 8192MB: 11584k | |
4098 | * 16384MB: 16384k | |
4099 | */ | |
4100 | static int __init init_per_zone_pages_min(void) | |
4101 | { | |
4102 | unsigned long lowmem_kbytes; | |
4103 | ||
4104 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
4105 | ||
4106 | min_free_kbytes = int_sqrt(lowmem_kbytes * 16); | |
4107 | if (min_free_kbytes < 128) | |
4108 | min_free_kbytes = 128; | |
4109 | if (min_free_kbytes > 65536) | |
4110 | min_free_kbytes = 65536; | |
4111 | setup_per_zone_pages_min(); | |
4112 | setup_per_zone_lowmem_reserve(); | |
4113 | return 0; | |
4114 | } | |
4115 | module_init(init_per_zone_pages_min) | |
4116 | ||
4117 | /* | |
4118 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so | |
4119 | * that we can call two helper functions whenever min_free_kbytes | |
4120 | * changes. | |
4121 | */ | |
4122 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, | |
4123 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4124 | { | |
4125 | proc_dointvec(table, write, file, buffer, length, ppos); | |
3b1d92c5 MG |
4126 | if (write) |
4127 | setup_per_zone_pages_min(); | |
1da177e4 LT |
4128 | return 0; |
4129 | } | |
4130 | ||
9614634f CL |
4131 | #ifdef CONFIG_NUMA |
4132 | int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write, | |
4133 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4134 | { | |
4135 | struct zone *zone; | |
4136 | int rc; | |
4137 | ||
4138 | rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
4139 | if (rc) | |
4140 | return rc; | |
4141 | ||
4142 | for_each_zone(zone) | |
8417bba4 | 4143 | zone->min_unmapped_pages = (zone->present_pages * |
9614634f CL |
4144 | sysctl_min_unmapped_ratio) / 100; |
4145 | return 0; | |
4146 | } | |
0ff38490 CL |
4147 | |
4148 | int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write, | |
4149 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4150 | { | |
4151 | struct zone *zone; | |
4152 | int rc; | |
4153 | ||
4154 | rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
4155 | if (rc) | |
4156 | return rc; | |
4157 | ||
4158 | for_each_zone(zone) | |
4159 | zone->min_slab_pages = (zone->present_pages * | |
4160 | sysctl_min_slab_ratio) / 100; | |
4161 | return 0; | |
4162 | } | |
9614634f CL |
4163 | #endif |
4164 | ||
1da177e4 LT |
4165 | /* |
4166 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
4167 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
4168 | * whenever sysctl_lowmem_reserve_ratio changes. | |
4169 | * | |
4170 | * The reserve ratio obviously has absolutely no relation with the | |
4171 | * pages_min watermarks. The lowmem reserve ratio can only make sense | |
4172 | * if in function of the boot time zone sizes. | |
4173 | */ | |
4174 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, | |
4175 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4176 | { | |
4177 | proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
4178 | setup_per_zone_lowmem_reserve(); | |
4179 | return 0; | |
4180 | } | |
4181 | ||
8ad4b1fb RS |
4182 | /* |
4183 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
4184 | * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist | |
4185 | * can have before it gets flushed back to buddy allocator. | |
4186 | */ | |
4187 | ||
4188 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, | |
4189 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4190 | { | |
4191 | struct zone *zone; | |
4192 | unsigned int cpu; | |
4193 | int ret; | |
4194 | ||
4195 | ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
4196 | if (!write || (ret == -EINVAL)) | |
4197 | return ret; | |
4198 | for_each_zone(zone) { | |
4199 | for_each_online_cpu(cpu) { | |
4200 | unsigned long high; | |
4201 | high = zone->present_pages / percpu_pagelist_fraction; | |
4202 | setup_pagelist_highmark(zone_pcp(zone, cpu), high); | |
4203 | } | |
4204 | } | |
4205 | return 0; | |
4206 | } | |
4207 | ||
f034b5d4 | 4208 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 LT |
4209 | |
4210 | #ifdef CONFIG_NUMA | |
4211 | static int __init set_hashdist(char *str) | |
4212 | { | |
4213 | if (!str) | |
4214 | return 0; | |
4215 | hashdist = simple_strtoul(str, &str, 0); | |
4216 | return 1; | |
4217 | } | |
4218 | __setup("hashdist=", set_hashdist); | |
4219 | #endif | |
4220 | ||
4221 | /* | |
4222 | * allocate a large system hash table from bootmem | |
4223 | * - it is assumed that the hash table must contain an exact power-of-2 | |
4224 | * quantity of entries | |
4225 | * - limit is the number of hash buckets, not the total allocation size | |
4226 | */ | |
4227 | void *__init alloc_large_system_hash(const char *tablename, | |
4228 | unsigned long bucketsize, | |
4229 | unsigned long numentries, | |
4230 | int scale, | |
4231 | int flags, | |
4232 | unsigned int *_hash_shift, | |
4233 | unsigned int *_hash_mask, | |
4234 | unsigned long limit) | |
4235 | { | |
4236 | unsigned long long max = limit; | |
4237 | unsigned long log2qty, size; | |
4238 | void *table = NULL; | |
4239 | ||
4240 | /* allow the kernel cmdline to have a say */ | |
4241 | if (!numentries) { | |
4242 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 4243 | numentries = nr_kernel_pages; |
1da177e4 LT |
4244 | numentries += (1UL << (20 - PAGE_SHIFT)) - 1; |
4245 | numentries >>= 20 - PAGE_SHIFT; | |
4246 | numentries <<= 20 - PAGE_SHIFT; | |
4247 | ||
4248 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
4249 | if (scale > PAGE_SHIFT) | |
4250 | numentries >>= (scale - PAGE_SHIFT); | |
4251 | else | |
4252 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
4253 | |
4254 | /* Make sure we've got at least a 0-order allocation.. */ | |
4255 | if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
4256 | numentries = PAGE_SIZE / bucketsize; | |
1da177e4 | 4257 | } |
6e692ed3 | 4258 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
4259 | |
4260 | /* limit allocation size to 1/16 total memory by default */ | |
4261 | if (max == 0) { | |
4262 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
4263 | do_div(max, bucketsize); | |
4264 | } | |
4265 | ||
4266 | if (numentries > max) | |
4267 | numentries = max; | |
4268 | ||
f0d1b0b3 | 4269 | log2qty = ilog2(numentries); |
1da177e4 LT |
4270 | |
4271 | do { | |
4272 | size = bucketsize << log2qty; | |
4273 | if (flags & HASH_EARLY) | |
4274 | table = alloc_bootmem(size); | |
4275 | else if (hashdist) | |
4276 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
4277 | else { | |
4278 | unsigned long order; | |
4279 | for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++) | |
4280 | ; | |
4281 | table = (void*) __get_free_pages(GFP_ATOMIC, order); | |
1037b83b ED |
4282 | /* |
4283 | * If bucketsize is not a power-of-two, we may free | |
4284 | * some pages at the end of hash table. | |
4285 | */ | |
4286 | if (table) { | |
4287 | unsigned long alloc_end = (unsigned long)table + | |
4288 | (PAGE_SIZE << order); | |
4289 | unsigned long used = (unsigned long)table + | |
4290 | PAGE_ALIGN(size); | |
4291 | split_page(virt_to_page(table), order); | |
4292 | while (used < alloc_end) { | |
4293 | free_page(used); | |
4294 | used += PAGE_SIZE; | |
4295 | } | |
4296 | } | |
1da177e4 LT |
4297 | } |
4298 | } while (!table && size > PAGE_SIZE && --log2qty); | |
4299 | ||
4300 | if (!table) | |
4301 | panic("Failed to allocate %s hash table\n", tablename); | |
4302 | ||
b49ad484 | 4303 | printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n", |
1da177e4 LT |
4304 | tablename, |
4305 | (1U << log2qty), | |
f0d1b0b3 | 4306 | ilog2(size) - PAGE_SHIFT, |
1da177e4 LT |
4307 | size); |
4308 | ||
4309 | if (_hash_shift) | |
4310 | *_hash_shift = log2qty; | |
4311 | if (_hash_mask) | |
4312 | *_hash_mask = (1 << log2qty) - 1; | |
4313 | ||
4314 | return table; | |
4315 | } | |
a117e66e KH |
4316 | |
4317 | #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE | |
a117e66e KH |
4318 | struct page *pfn_to_page(unsigned long pfn) |
4319 | { | |
67de6482 | 4320 | return __pfn_to_page(pfn); |
a117e66e KH |
4321 | } |
4322 | unsigned long page_to_pfn(struct page *page) | |
4323 | { | |
67de6482 | 4324 | return __page_to_pfn(page); |
a117e66e | 4325 | } |
a117e66e KH |
4326 | EXPORT_SYMBOL(pfn_to_page); |
4327 | EXPORT_SYMBOL(page_to_pfn); | |
4328 | #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */ | |
6220ec78 | 4329 | |
835c134e MG |
4330 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
4331 | static inline unsigned long *get_pageblock_bitmap(struct zone *zone, | |
4332 | unsigned long pfn) | |
4333 | { | |
4334 | #ifdef CONFIG_SPARSEMEM | |
4335 | return __pfn_to_section(pfn)->pageblock_flags; | |
4336 | #else | |
4337 | return zone->pageblock_flags; | |
4338 | #endif /* CONFIG_SPARSEMEM */ | |
4339 | } | |
4340 | ||
4341 | static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn) | |
4342 | { | |
4343 | #ifdef CONFIG_SPARSEMEM | |
4344 | pfn &= (PAGES_PER_SECTION-1); | |
4345 | return (pfn >> (MAX_ORDER-1)) * NR_PAGEBLOCK_BITS; | |
4346 | #else | |
4347 | pfn = pfn - zone->zone_start_pfn; | |
4348 | return (pfn >> (MAX_ORDER-1)) * NR_PAGEBLOCK_BITS; | |
4349 | #endif /* CONFIG_SPARSEMEM */ | |
4350 | } | |
4351 | ||
4352 | /** | |
4353 | * get_pageblock_flags_group - Return the requested group of flags for the MAX_ORDER_NR_PAGES block of pages | |
4354 | * @page: The page within the block of interest | |
4355 | * @start_bitidx: The first bit of interest to retrieve | |
4356 | * @end_bitidx: The last bit of interest | |
4357 | * returns pageblock_bits flags | |
4358 | */ | |
4359 | unsigned long get_pageblock_flags_group(struct page *page, | |
4360 | int start_bitidx, int end_bitidx) | |
4361 | { | |
4362 | struct zone *zone; | |
4363 | unsigned long *bitmap; | |
4364 | unsigned long pfn, bitidx; | |
4365 | unsigned long flags = 0; | |
4366 | unsigned long value = 1; | |
4367 | ||
4368 | zone = page_zone(page); | |
4369 | pfn = page_to_pfn(page); | |
4370 | bitmap = get_pageblock_bitmap(zone, pfn); | |
4371 | bitidx = pfn_to_bitidx(zone, pfn); | |
4372 | ||
4373 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
4374 | if (test_bit(bitidx + start_bitidx, bitmap)) | |
4375 | flags |= value; | |
6220ec78 | 4376 | |
835c134e MG |
4377 | return flags; |
4378 | } | |
4379 | ||
4380 | /** | |
4381 | * set_pageblock_flags_group - Set the requested group of flags for a MAX_ORDER_NR_PAGES block of pages | |
4382 | * @page: The page within the block of interest | |
4383 | * @start_bitidx: The first bit of interest | |
4384 | * @end_bitidx: The last bit of interest | |
4385 | * @flags: The flags to set | |
4386 | */ | |
4387 | void set_pageblock_flags_group(struct page *page, unsigned long flags, | |
4388 | int start_bitidx, int end_bitidx) | |
4389 | { | |
4390 | struct zone *zone; | |
4391 | unsigned long *bitmap; | |
4392 | unsigned long pfn, bitidx; | |
4393 | unsigned long value = 1; | |
4394 | ||
4395 | zone = page_zone(page); | |
4396 | pfn = page_to_pfn(page); | |
4397 | bitmap = get_pageblock_bitmap(zone, pfn); | |
4398 | bitidx = pfn_to_bitidx(zone, pfn); | |
4399 | ||
4400 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
4401 | if (flags & value) | |
4402 | __set_bit(bitidx + start_bitidx, bitmap); | |
4403 | else | |
4404 | __clear_bit(bitidx + start_bitidx, bitmap); | |
4405 | } |