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 | ||
17 | #include <linux/config.h> | |
18 | #include <linux/stddef.h> | |
19 | #include <linux/mm.h> | |
20 | #include <linux/swap.h> | |
21 | #include <linux/interrupt.h> | |
22 | #include <linux/pagemap.h> | |
23 | #include <linux/bootmem.h> | |
24 | #include <linux/compiler.h> | |
9f158333 | 25 | #include <linux/kernel.h> |
1da177e4 LT |
26 | #include <linux/module.h> |
27 | #include <linux/suspend.h> | |
28 | #include <linux/pagevec.h> | |
29 | #include <linux/blkdev.h> | |
30 | #include <linux/slab.h> | |
31 | #include <linux/notifier.h> | |
32 | #include <linux/topology.h> | |
33 | #include <linux/sysctl.h> | |
34 | #include <linux/cpu.h> | |
35 | #include <linux/cpuset.h> | |
bdc8cb98 | 36 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
37 | #include <linux/nodemask.h> |
38 | #include <linux/vmalloc.h> | |
4be38e35 | 39 | #include <linux/mempolicy.h> |
6811378e | 40 | #include <linux/stop_machine.h> |
1da177e4 LT |
41 | |
42 | #include <asm/tlbflush.h> | |
ac924c60 | 43 | #include <asm/div64.h> |
1da177e4 LT |
44 | #include "internal.h" |
45 | ||
46 | /* | |
47 | * MCD - HACK: Find somewhere to initialize this EARLY, or make this | |
48 | * initializer cleaner | |
49 | */ | |
c3d8c141 | 50 | nodemask_t node_online_map __read_mostly = { { [0] = 1UL } }; |
7223a93a | 51 | EXPORT_SYMBOL(node_online_map); |
c3d8c141 | 52 | nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL; |
7223a93a | 53 | EXPORT_SYMBOL(node_possible_map); |
6c231b7b RT |
54 | unsigned long totalram_pages __read_mostly; |
55 | unsigned long totalhigh_pages __read_mostly; | |
cb45b0e9 | 56 | unsigned long totalreserve_pages __read_mostly; |
1da177e4 | 57 | long nr_swap_pages; |
8ad4b1fb | 58 | int percpu_pagelist_fraction; |
1da177e4 | 59 | |
d98c7a09 | 60 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 61 | |
1da177e4 LT |
62 | /* |
63 | * results with 256, 32 in the lowmem_reserve sysctl: | |
64 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
65 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
66 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
67 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
68 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA | |
a2f1b424 AK |
69 | * |
70 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
71 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 72 | */ |
a2f1b424 | 73 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { 256, 256, 32 }; |
1da177e4 LT |
74 | |
75 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 LT |
76 | |
77 | /* | |
78 | * Used by page_zone() to look up the address of the struct zone whose | |
79 | * id is encoded in the upper bits of page->flags | |
80 | */ | |
c3d8c141 | 81 | struct zone *zone_table[1 << ZONETABLE_SHIFT] __read_mostly; |
1da177e4 LT |
82 | EXPORT_SYMBOL(zone_table); |
83 | ||
a2f1b424 | 84 | static char *zone_names[MAX_NR_ZONES] = { "DMA", "DMA32", "Normal", "HighMem" }; |
1da177e4 LT |
85 | int min_free_kbytes = 1024; |
86 | ||
86356ab1 YG |
87 | unsigned long __meminitdata nr_kernel_pages; |
88 | unsigned long __meminitdata nr_all_pages; | |
1da177e4 | 89 | |
13e7444b | 90 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 91 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 92 | { |
bdc8cb98 DH |
93 | int ret = 0; |
94 | unsigned seq; | |
95 | unsigned long pfn = page_to_pfn(page); | |
c6a57e19 | 96 | |
bdc8cb98 DH |
97 | do { |
98 | seq = zone_span_seqbegin(zone); | |
99 | if (pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
100 | ret = 1; | |
101 | else if (pfn < zone->zone_start_pfn) | |
102 | ret = 1; | |
103 | } while (zone_span_seqretry(zone, seq)); | |
104 | ||
105 | return ret; | |
c6a57e19 DH |
106 | } |
107 | ||
108 | static int page_is_consistent(struct zone *zone, struct page *page) | |
109 | { | |
1da177e4 LT |
110 | #ifdef CONFIG_HOLES_IN_ZONE |
111 | if (!pfn_valid(page_to_pfn(page))) | |
c6a57e19 | 112 | return 0; |
1da177e4 LT |
113 | #endif |
114 | if (zone != page_zone(page)) | |
c6a57e19 DH |
115 | return 0; |
116 | ||
117 | return 1; | |
118 | } | |
119 | /* | |
120 | * Temporary debugging check for pages not lying within a given zone. | |
121 | */ | |
122 | static int bad_range(struct zone *zone, struct page *page) | |
123 | { | |
124 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 125 | return 1; |
c6a57e19 DH |
126 | if (!page_is_consistent(zone, page)) |
127 | return 1; | |
128 | ||
1da177e4 LT |
129 | return 0; |
130 | } | |
131 | ||
13e7444b NP |
132 | #else |
133 | static inline int bad_range(struct zone *zone, struct page *page) | |
134 | { | |
135 | return 0; | |
136 | } | |
137 | #endif | |
138 | ||
224abf92 | 139 | static void bad_page(struct page *page) |
1da177e4 | 140 | { |
224abf92 | 141 | printk(KERN_EMERG "Bad page state in process '%s'\n" |
7365f3d1 HD |
142 | KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n" |
143 | KERN_EMERG "Trying to fix it up, but a reboot is needed\n" | |
144 | KERN_EMERG "Backtrace:\n", | |
224abf92 NP |
145 | current->comm, page, (int)(2*sizeof(unsigned long)), |
146 | (unsigned long)page->flags, page->mapping, | |
147 | page_mapcount(page), page_count(page)); | |
1da177e4 | 148 | dump_stack(); |
334795ec HD |
149 | page->flags &= ~(1 << PG_lru | |
150 | 1 << PG_private | | |
1da177e4 | 151 | 1 << PG_locked | |
1da177e4 LT |
152 | 1 << PG_active | |
153 | 1 << PG_dirty | | |
334795ec HD |
154 | 1 << PG_reclaim | |
155 | 1 << PG_slab | | |
1da177e4 | 156 | 1 << PG_swapcache | |
676165a8 NP |
157 | 1 << PG_writeback | |
158 | 1 << PG_buddy ); | |
1da177e4 LT |
159 | set_page_count(page, 0); |
160 | reset_page_mapcount(page); | |
161 | page->mapping = NULL; | |
9f158333 | 162 | add_taint(TAINT_BAD_PAGE); |
1da177e4 LT |
163 | } |
164 | ||
1da177e4 LT |
165 | /* |
166 | * Higher-order pages are called "compound pages". They are structured thusly: | |
167 | * | |
168 | * The first PAGE_SIZE page is called the "head page". | |
169 | * | |
170 | * The remaining PAGE_SIZE pages are called "tail pages". | |
171 | * | |
172 | * All pages have PG_compound set. All pages have their ->private pointing at | |
173 | * the head page (even the head page has this). | |
174 | * | |
41d78ba5 HD |
175 | * The first tail page's ->lru.next holds the address of the compound page's |
176 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
177 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 178 | */ |
d98c7a09 HD |
179 | |
180 | static void free_compound_page(struct page *page) | |
181 | { | |
182 | __free_pages_ok(page, (unsigned long)page[1].lru.prev); | |
183 | } | |
184 | ||
1da177e4 LT |
185 | static void prep_compound_page(struct page *page, unsigned long order) |
186 | { | |
187 | int i; | |
188 | int nr_pages = 1 << order; | |
189 | ||
d98c7a09 | 190 | page[1].lru.next = (void *)free_compound_page; /* set dtor */ |
41d78ba5 | 191 | page[1].lru.prev = (void *)order; |
1da177e4 LT |
192 | for (i = 0; i < nr_pages; i++) { |
193 | struct page *p = page + i; | |
194 | ||
5e9dace8 | 195 | __SetPageCompound(p); |
4c21e2f2 | 196 | set_page_private(p, (unsigned long)page); |
1da177e4 LT |
197 | } |
198 | } | |
199 | ||
200 | static void destroy_compound_page(struct page *page, unsigned long order) | |
201 | { | |
202 | int i; | |
203 | int nr_pages = 1 << order; | |
204 | ||
41d78ba5 | 205 | if (unlikely((unsigned long)page[1].lru.prev != order)) |
224abf92 | 206 | bad_page(page); |
1da177e4 LT |
207 | |
208 | for (i = 0; i < nr_pages; i++) { | |
209 | struct page *p = page + i; | |
210 | ||
224abf92 NP |
211 | if (unlikely(!PageCompound(p) | |
212 | (page_private(p) != (unsigned long)page))) | |
213 | bad_page(page); | |
5e9dace8 | 214 | __ClearPageCompound(p); |
1da177e4 LT |
215 | } |
216 | } | |
1da177e4 | 217 | |
17cf4406 NP |
218 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) |
219 | { | |
220 | int i; | |
221 | ||
222 | BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM); | |
6626c5d5 AM |
223 | /* |
224 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO | |
225 | * and __GFP_HIGHMEM from hard or soft interrupt context. | |
226 | */ | |
227 | BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); | |
17cf4406 NP |
228 | for (i = 0; i < (1 << order); i++) |
229 | clear_highpage(page + i); | |
230 | } | |
231 | ||
1da177e4 LT |
232 | /* |
233 | * function for dealing with page's order in buddy system. | |
234 | * zone->lock is already acquired when we use these. | |
235 | * So, we don't need atomic page->flags operations here. | |
236 | */ | |
6aa3001b AM |
237 | static inline unsigned long page_order(struct page *page) |
238 | { | |
4c21e2f2 | 239 | return page_private(page); |
1da177e4 LT |
240 | } |
241 | ||
6aa3001b AM |
242 | static inline void set_page_order(struct page *page, int order) |
243 | { | |
4c21e2f2 | 244 | set_page_private(page, order); |
676165a8 | 245 | __SetPageBuddy(page); |
1da177e4 LT |
246 | } |
247 | ||
248 | static inline void rmv_page_order(struct page *page) | |
249 | { | |
676165a8 | 250 | __ClearPageBuddy(page); |
4c21e2f2 | 251 | set_page_private(page, 0); |
1da177e4 LT |
252 | } |
253 | ||
254 | /* | |
255 | * Locate the struct page for both the matching buddy in our | |
256 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
257 | * | |
258 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
259 | * the following equation: | |
260 | * B2 = B1 ^ (1 << O) | |
261 | * For example, if the starting buddy (buddy2) is #8 its order | |
262 | * 1 buddy is #10: | |
263 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
264 | * | |
265 | * 2) Any buddy B will have an order O+1 parent P which | |
266 | * satisfies the following equation: | |
267 | * P = B & ~(1 << O) | |
268 | * | |
d6e05edc | 269 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER |
1da177e4 LT |
270 | */ |
271 | static inline struct page * | |
272 | __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order) | |
273 | { | |
274 | unsigned long buddy_idx = page_idx ^ (1 << order); | |
275 | ||
276 | return page + (buddy_idx - page_idx); | |
277 | } | |
278 | ||
279 | static inline unsigned long | |
280 | __find_combined_index(unsigned long page_idx, unsigned int order) | |
281 | { | |
282 | return (page_idx & ~(1 << order)); | |
283 | } | |
284 | ||
285 | /* | |
286 | * This function checks whether a page is free && is the buddy | |
287 | * we can do coalesce a page and its buddy if | |
13e7444b | 288 | * (a) the buddy is not in a hole && |
676165a8 | 289 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
290 | * (c) a page and its buddy have the same order && |
291 | * (d) a page and its buddy are in the same zone. | |
676165a8 NP |
292 | * |
293 | * For recording whether a page is in the buddy system, we use PG_buddy. | |
294 | * Setting, clearing, and testing PG_buddy is serialized by zone->lock. | |
1da177e4 | 295 | * |
676165a8 | 296 | * For recording page's order, we use page_private(page). |
1da177e4 | 297 | */ |
cb2b95e1 AW |
298 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
299 | int order) | |
1da177e4 | 300 | { |
13e7444b | 301 | #ifdef CONFIG_HOLES_IN_ZONE |
cb2b95e1 | 302 | if (!pfn_valid(page_to_pfn(buddy))) |
13e7444b NP |
303 | return 0; |
304 | #endif | |
305 | ||
cb2b95e1 AW |
306 | if (page_zone_id(page) != page_zone_id(buddy)) |
307 | return 0; | |
308 | ||
309 | if (PageBuddy(buddy) && page_order(buddy) == order) { | |
310 | BUG_ON(page_count(buddy) != 0); | |
6aa3001b | 311 | return 1; |
676165a8 | 312 | } |
6aa3001b | 313 | return 0; |
1da177e4 LT |
314 | } |
315 | ||
316 | /* | |
317 | * Freeing function for a buddy system allocator. | |
318 | * | |
319 | * The concept of a buddy system is to maintain direct-mapped table | |
320 | * (containing bit values) for memory blocks of various "orders". | |
321 | * The bottom level table contains the map for the smallest allocatable | |
322 | * units of memory (here, pages), and each level above it describes | |
323 | * pairs of units from the levels below, hence, "buddies". | |
324 | * At a high level, all that happens here is marking the table entry | |
325 | * at the bottom level available, and propagating the changes upward | |
326 | * as necessary, plus some accounting needed to play nicely with other | |
327 | * parts of the VM system. | |
328 | * At each level, we keep a list of pages, which are heads of continuous | |
676165a8 | 329 | * free pages of length of (1 << order) and marked with PG_buddy. Page's |
4c21e2f2 | 330 | * order is recorded in page_private(page) field. |
1da177e4 LT |
331 | * So when we are allocating or freeing one, we can derive the state of the |
332 | * other. That is, if we allocate a small block, and both were | |
333 | * free, the remainder of the region must be split into blocks. | |
334 | * If a block is freed, and its buddy is also free, then this | |
335 | * triggers coalescing into a block of larger size. | |
336 | * | |
337 | * -- wli | |
338 | */ | |
339 | ||
48db57f8 | 340 | static inline void __free_one_page(struct page *page, |
1da177e4 LT |
341 | struct zone *zone, unsigned int order) |
342 | { | |
343 | unsigned long page_idx; | |
344 | int order_size = 1 << order; | |
345 | ||
224abf92 | 346 | if (unlikely(PageCompound(page))) |
1da177e4 LT |
347 | destroy_compound_page(page, order); |
348 | ||
349 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); | |
350 | ||
351 | BUG_ON(page_idx & (order_size - 1)); | |
352 | BUG_ON(bad_range(zone, page)); | |
353 | ||
354 | zone->free_pages += order_size; | |
355 | while (order < MAX_ORDER-1) { | |
356 | unsigned long combined_idx; | |
357 | struct free_area *area; | |
358 | struct page *buddy; | |
359 | ||
1da177e4 | 360 | buddy = __page_find_buddy(page, page_idx, order); |
cb2b95e1 | 361 | if (!page_is_buddy(page, buddy, order)) |
1da177e4 | 362 | break; /* Move the buddy up one level. */ |
13e7444b | 363 | |
1da177e4 LT |
364 | list_del(&buddy->lru); |
365 | area = zone->free_area + order; | |
366 | area->nr_free--; | |
367 | rmv_page_order(buddy); | |
13e7444b | 368 | combined_idx = __find_combined_index(page_idx, order); |
1da177e4 LT |
369 | page = page + (combined_idx - page_idx); |
370 | page_idx = combined_idx; | |
371 | order++; | |
372 | } | |
373 | set_page_order(page, order); | |
374 | list_add(&page->lru, &zone->free_area[order].free_list); | |
375 | zone->free_area[order].nr_free++; | |
376 | } | |
377 | ||
224abf92 | 378 | static inline int free_pages_check(struct page *page) |
1da177e4 | 379 | { |
92be2e33 NP |
380 | if (unlikely(page_mapcount(page) | |
381 | (page->mapping != NULL) | | |
382 | (page_count(page) != 0) | | |
1da177e4 LT |
383 | (page->flags & ( |
384 | 1 << PG_lru | | |
385 | 1 << PG_private | | |
386 | 1 << PG_locked | | |
387 | 1 << PG_active | | |
388 | 1 << PG_reclaim | | |
389 | 1 << PG_slab | | |
390 | 1 << PG_swapcache | | |
b5810039 | 391 | 1 << PG_writeback | |
676165a8 NP |
392 | 1 << PG_reserved | |
393 | 1 << PG_buddy )))) | |
224abf92 | 394 | bad_page(page); |
1da177e4 | 395 | if (PageDirty(page)) |
242e5468 | 396 | __ClearPageDirty(page); |
689bcebf HD |
397 | /* |
398 | * For now, we report if PG_reserved was found set, but do not | |
399 | * clear it, and do not free the page. But we shall soon need | |
400 | * to do more, for when the ZERO_PAGE count wraps negative. | |
401 | */ | |
402 | return PageReserved(page); | |
1da177e4 LT |
403 | } |
404 | ||
405 | /* | |
406 | * Frees a list of pages. | |
407 | * Assumes all pages on list are in same zone, and of same order. | |
207f36ee | 408 | * count is the number of pages to free. |
1da177e4 LT |
409 | * |
410 | * If the zone was previously in an "all pages pinned" state then look to | |
411 | * see if this freeing clears that state. | |
412 | * | |
413 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
414 | * pinned" detection logic. | |
415 | */ | |
48db57f8 NP |
416 | static void free_pages_bulk(struct zone *zone, int count, |
417 | struct list_head *list, int order) | |
1da177e4 | 418 | { |
c54ad30c | 419 | spin_lock(&zone->lock); |
1da177e4 LT |
420 | zone->all_unreclaimable = 0; |
421 | zone->pages_scanned = 0; | |
48db57f8 NP |
422 | while (count--) { |
423 | struct page *page; | |
424 | ||
425 | BUG_ON(list_empty(list)); | |
1da177e4 | 426 | page = list_entry(list->prev, struct page, lru); |
48db57f8 | 427 | /* have to delete it as __free_one_page list manipulates */ |
1da177e4 | 428 | list_del(&page->lru); |
48db57f8 | 429 | __free_one_page(page, zone, order); |
1da177e4 | 430 | } |
c54ad30c | 431 | spin_unlock(&zone->lock); |
1da177e4 LT |
432 | } |
433 | ||
48db57f8 | 434 | static void free_one_page(struct zone *zone, struct page *page, int order) |
1da177e4 LT |
435 | { |
436 | LIST_HEAD(list); | |
48db57f8 NP |
437 | list_add(&page->lru, &list); |
438 | free_pages_bulk(zone, 1, &list, order); | |
439 | } | |
440 | ||
441 | static void __free_pages_ok(struct page *page, unsigned int order) | |
442 | { | |
443 | unsigned long flags; | |
1da177e4 | 444 | int i; |
689bcebf | 445 | int reserved = 0; |
1da177e4 LT |
446 | |
447 | arch_free_page(page, order); | |
de5097c2 | 448 | if (!PageHighMem(page)) |
f9b8404c IM |
449 | debug_check_no_locks_freed(page_address(page), |
450 | PAGE_SIZE<<order); | |
1da177e4 | 451 | |
1da177e4 | 452 | for (i = 0 ; i < (1 << order) ; ++i) |
224abf92 | 453 | reserved += free_pages_check(page + i); |
689bcebf HD |
454 | if (reserved) |
455 | return; | |
456 | ||
48db57f8 | 457 | kernel_map_pages(page, 1 << order, 0); |
c54ad30c | 458 | local_irq_save(flags); |
a74609fa | 459 | __mod_page_state(pgfree, 1 << order); |
48db57f8 | 460 | free_one_page(page_zone(page), page, order); |
c54ad30c | 461 | local_irq_restore(flags); |
1da177e4 LT |
462 | } |
463 | ||
a226f6c8 DH |
464 | /* |
465 | * permit the bootmem allocator to evade page validation on high-order frees | |
466 | */ | |
467 | void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order) | |
468 | { | |
469 | if (order == 0) { | |
470 | __ClearPageReserved(page); | |
471 | set_page_count(page, 0); | |
7835e98b | 472 | set_page_refcounted(page); |
545b1ea9 | 473 | __free_page(page); |
a226f6c8 | 474 | } else { |
a226f6c8 DH |
475 | int loop; |
476 | ||
545b1ea9 | 477 | prefetchw(page); |
a226f6c8 DH |
478 | for (loop = 0; loop < BITS_PER_LONG; loop++) { |
479 | struct page *p = &page[loop]; | |
480 | ||
545b1ea9 NP |
481 | if (loop + 1 < BITS_PER_LONG) |
482 | prefetchw(p + 1); | |
a226f6c8 DH |
483 | __ClearPageReserved(p); |
484 | set_page_count(p, 0); | |
485 | } | |
486 | ||
7835e98b | 487 | set_page_refcounted(page); |
545b1ea9 | 488 | __free_pages(page, order); |
a226f6c8 DH |
489 | } |
490 | } | |
491 | ||
1da177e4 LT |
492 | |
493 | /* | |
494 | * The order of subdivision here is critical for the IO subsystem. | |
495 | * Please do not alter this order without good reasons and regression | |
496 | * testing. Specifically, as large blocks of memory are subdivided, | |
497 | * the order in which smaller blocks are delivered depends on the order | |
498 | * they're subdivided in this function. This is the primary factor | |
499 | * influencing the order in which pages are delivered to the IO | |
500 | * subsystem according to empirical testing, and this is also justified | |
501 | * by considering the behavior of a buddy system containing a single | |
502 | * large block of memory acted on by a series of small allocations. | |
503 | * This behavior is a critical factor in sglist merging's success. | |
504 | * | |
505 | * -- wli | |
506 | */ | |
085cc7d5 | 507 | static inline void expand(struct zone *zone, struct page *page, |
1da177e4 LT |
508 | int low, int high, struct free_area *area) |
509 | { | |
510 | unsigned long size = 1 << high; | |
511 | ||
512 | while (high > low) { | |
513 | area--; | |
514 | high--; | |
515 | size >>= 1; | |
516 | BUG_ON(bad_range(zone, &page[size])); | |
517 | list_add(&page[size].lru, &area->free_list); | |
518 | area->nr_free++; | |
519 | set_page_order(&page[size], high); | |
520 | } | |
1da177e4 LT |
521 | } |
522 | ||
1da177e4 LT |
523 | /* |
524 | * This page is about to be returned from the page allocator | |
525 | */ | |
17cf4406 | 526 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) |
1da177e4 | 527 | { |
92be2e33 NP |
528 | if (unlikely(page_mapcount(page) | |
529 | (page->mapping != NULL) | | |
530 | (page_count(page) != 0) | | |
334795ec HD |
531 | (page->flags & ( |
532 | 1 << PG_lru | | |
1da177e4 LT |
533 | 1 << PG_private | |
534 | 1 << PG_locked | | |
1da177e4 LT |
535 | 1 << PG_active | |
536 | 1 << PG_dirty | | |
537 | 1 << PG_reclaim | | |
334795ec | 538 | 1 << PG_slab | |
1da177e4 | 539 | 1 << PG_swapcache | |
b5810039 | 540 | 1 << PG_writeback | |
676165a8 NP |
541 | 1 << PG_reserved | |
542 | 1 << PG_buddy )))) | |
224abf92 | 543 | bad_page(page); |
1da177e4 | 544 | |
689bcebf HD |
545 | /* |
546 | * For now, we report if PG_reserved was found set, but do not | |
547 | * clear it, and do not allocate the page: as a safety net. | |
548 | */ | |
549 | if (PageReserved(page)) | |
550 | return 1; | |
551 | ||
1da177e4 LT |
552 | page->flags &= ~(1 << PG_uptodate | 1 << PG_error | |
553 | 1 << PG_referenced | 1 << PG_arch_1 | | |
554 | 1 << PG_checked | 1 << PG_mappedtodisk); | |
4c21e2f2 | 555 | set_page_private(page, 0); |
7835e98b | 556 | set_page_refcounted(page); |
1da177e4 | 557 | kernel_map_pages(page, 1 << order, 1); |
17cf4406 NP |
558 | |
559 | if (gfp_flags & __GFP_ZERO) | |
560 | prep_zero_page(page, order, gfp_flags); | |
561 | ||
562 | if (order && (gfp_flags & __GFP_COMP)) | |
563 | prep_compound_page(page, order); | |
564 | ||
689bcebf | 565 | return 0; |
1da177e4 LT |
566 | } |
567 | ||
568 | /* | |
569 | * Do the hard work of removing an element from the buddy allocator. | |
570 | * Call me with the zone->lock already held. | |
571 | */ | |
572 | static struct page *__rmqueue(struct zone *zone, unsigned int order) | |
573 | { | |
574 | struct free_area * area; | |
575 | unsigned int current_order; | |
576 | struct page *page; | |
577 | ||
578 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
579 | area = zone->free_area + current_order; | |
580 | if (list_empty(&area->free_list)) | |
581 | continue; | |
582 | ||
583 | page = list_entry(area->free_list.next, struct page, lru); | |
584 | list_del(&page->lru); | |
585 | rmv_page_order(page); | |
586 | area->nr_free--; | |
587 | zone->free_pages -= 1UL << order; | |
085cc7d5 NP |
588 | expand(zone, page, order, current_order, area); |
589 | return page; | |
1da177e4 LT |
590 | } |
591 | ||
592 | return NULL; | |
593 | } | |
594 | ||
595 | /* | |
596 | * Obtain a specified number of elements from the buddy allocator, all under | |
597 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
598 | * Returns the number of new pages which were placed at *list. | |
599 | */ | |
600 | static int rmqueue_bulk(struct zone *zone, unsigned int order, | |
601 | unsigned long count, struct list_head *list) | |
602 | { | |
1da177e4 | 603 | int i; |
1da177e4 | 604 | |
c54ad30c | 605 | spin_lock(&zone->lock); |
1da177e4 | 606 | for (i = 0; i < count; ++i) { |
085cc7d5 NP |
607 | struct page *page = __rmqueue(zone, order); |
608 | if (unlikely(page == NULL)) | |
1da177e4 | 609 | break; |
1da177e4 LT |
610 | list_add_tail(&page->lru, list); |
611 | } | |
c54ad30c | 612 | spin_unlock(&zone->lock); |
085cc7d5 | 613 | return i; |
1da177e4 LT |
614 | } |
615 | ||
4ae7c039 | 616 | #ifdef CONFIG_NUMA |
8fce4d8e CL |
617 | /* |
618 | * Called from the slab reaper to drain pagesets on a particular node that | |
619 | * belong to the currently executing processor. | |
879336c3 CL |
620 | * Note that this function must be called with the thread pinned to |
621 | * a single processor. | |
8fce4d8e CL |
622 | */ |
623 | void drain_node_pages(int nodeid) | |
4ae7c039 | 624 | { |
8fce4d8e | 625 | int i, z; |
4ae7c039 CL |
626 | unsigned long flags; |
627 | ||
8fce4d8e CL |
628 | for (z = 0; z < MAX_NR_ZONES; z++) { |
629 | struct zone *zone = NODE_DATA(nodeid)->node_zones + z; | |
4ae7c039 CL |
630 | struct per_cpu_pageset *pset; |
631 | ||
23316bc8 | 632 | pset = zone_pcp(zone, smp_processor_id()); |
4ae7c039 CL |
633 | for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) { |
634 | struct per_cpu_pages *pcp; | |
635 | ||
636 | pcp = &pset->pcp[i]; | |
879336c3 CL |
637 | if (pcp->count) { |
638 | local_irq_save(flags); | |
639 | free_pages_bulk(zone, pcp->count, &pcp->list, 0); | |
640 | pcp->count = 0; | |
641 | local_irq_restore(flags); | |
642 | } | |
4ae7c039 CL |
643 | } |
644 | } | |
4ae7c039 CL |
645 | } |
646 | #endif | |
647 | ||
1da177e4 LT |
648 | #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU) |
649 | static void __drain_pages(unsigned int cpu) | |
650 | { | |
c54ad30c | 651 | unsigned long flags; |
1da177e4 LT |
652 | struct zone *zone; |
653 | int i; | |
654 | ||
655 | for_each_zone(zone) { | |
656 | struct per_cpu_pageset *pset; | |
657 | ||
e7c8d5c9 | 658 | pset = zone_pcp(zone, cpu); |
1da177e4 LT |
659 | for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) { |
660 | struct per_cpu_pages *pcp; | |
661 | ||
662 | pcp = &pset->pcp[i]; | |
c54ad30c | 663 | local_irq_save(flags); |
48db57f8 NP |
664 | free_pages_bulk(zone, pcp->count, &pcp->list, 0); |
665 | pcp->count = 0; | |
c54ad30c | 666 | local_irq_restore(flags); |
1da177e4 LT |
667 | } |
668 | } | |
669 | } | |
670 | #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */ | |
671 | ||
672 | #ifdef CONFIG_PM | |
673 | ||
674 | void mark_free_pages(struct zone *zone) | |
675 | { | |
676 | unsigned long zone_pfn, flags; | |
677 | int order; | |
678 | struct list_head *curr; | |
679 | ||
680 | if (!zone->spanned_pages) | |
681 | return; | |
682 | ||
683 | spin_lock_irqsave(&zone->lock, flags); | |
684 | for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) | |
685 | ClearPageNosaveFree(pfn_to_page(zone_pfn + zone->zone_start_pfn)); | |
686 | ||
687 | for (order = MAX_ORDER - 1; order >= 0; --order) | |
688 | list_for_each(curr, &zone->free_area[order].free_list) { | |
689 | unsigned long start_pfn, i; | |
690 | ||
691 | start_pfn = page_to_pfn(list_entry(curr, struct page, lru)); | |
692 | ||
693 | for (i=0; i < (1<<order); i++) | |
694 | SetPageNosaveFree(pfn_to_page(start_pfn+i)); | |
695 | } | |
696 | spin_unlock_irqrestore(&zone->lock, flags); | |
697 | } | |
698 | ||
699 | /* | |
700 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
701 | */ | |
702 | void drain_local_pages(void) | |
703 | { | |
704 | unsigned long flags; | |
705 | ||
706 | local_irq_save(flags); | |
707 | __drain_pages(smp_processor_id()); | |
708 | local_irq_restore(flags); | |
709 | } | |
710 | #endif /* CONFIG_PM */ | |
711 | ||
1da177e4 LT |
712 | /* |
713 | * Free a 0-order page | |
714 | */ | |
1da177e4 LT |
715 | static void fastcall free_hot_cold_page(struct page *page, int cold) |
716 | { | |
717 | struct zone *zone = page_zone(page); | |
718 | struct per_cpu_pages *pcp; | |
719 | unsigned long flags; | |
720 | ||
721 | arch_free_page(page, 0); | |
722 | ||
1da177e4 LT |
723 | if (PageAnon(page)) |
724 | page->mapping = NULL; | |
224abf92 | 725 | if (free_pages_check(page)) |
689bcebf HD |
726 | return; |
727 | ||
689bcebf HD |
728 | kernel_map_pages(page, 1, 0); |
729 | ||
e7c8d5c9 | 730 | pcp = &zone_pcp(zone, get_cpu())->pcp[cold]; |
1da177e4 | 731 | local_irq_save(flags); |
a74609fa | 732 | __inc_page_state(pgfree); |
1da177e4 LT |
733 | list_add(&page->lru, &pcp->list); |
734 | pcp->count++; | |
48db57f8 NP |
735 | if (pcp->count >= pcp->high) { |
736 | free_pages_bulk(zone, pcp->batch, &pcp->list, 0); | |
737 | pcp->count -= pcp->batch; | |
738 | } | |
1da177e4 LT |
739 | local_irq_restore(flags); |
740 | put_cpu(); | |
741 | } | |
742 | ||
743 | void fastcall free_hot_page(struct page *page) | |
744 | { | |
745 | free_hot_cold_page(page, 0); | |
746 | } | |
747 | ||
748 | void fastcall free_cold_page(struct page *page) | |
749 | { | |
750 | free_hot_cold_page(page, 1); | |
751 | } | |
752 | ||
8dfcc9ba NP |
753 | /* |
754 | * split_page takes a non-compound higher-order page, and splits it into | |
755 | * n (1<<order) sub-pages: page[0..n] | |
756 | * Each sub-page must be freed individually. | |
757 | * | |
758 | * Note: this is probably too low level an operation for use in drivers. | |
759 | * Please consult with lkml before using this in your driver. | |
760 | */ | |
761 | void split_page(struct page *page, unsigned int order) | |
762 | { | |
763 | int i; | |
764 | ||
765 | BUG_ON(PageCompound(page)); | |
766 | BUG_ON(!page_count(page)); | |
7835e98b NP |
767 | for (i = 1; i < (1 << order); i++) |
768 | set_page_refcounted(page + i); | |
8dfcc9ba | 769 | } |
8dfcc9ba | 770 | |
1da177e4 LT |
771 | /* |
772 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
773 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
774 | * or two. | |
775 | */ | |
a74609fa NP |
776 | static struct page *buffered_rmqueue(struct zonelist *zonelist, |
777 | struct zone *zone, int order, gfp_t gfp_flags) | |
1da177e4 LT |
778 | { |
779 | unsigned long flags; | |
689bcebf | 780 | struct page *page; |
1da177e4 | 781 | int cold = !!(gfp_flags & __GFP_COLD); |
a74609fa | 782 | int cpu; |
1da177e4 | 783 | |
689bcebf | 784 | again: |
a74609fa | 785 | cpu = get_cpu(); |
48db57f8 | 786 | if (likely(order == 0)) { |
1da177e4 LT |
787 | struct per_cpu_pages *pcp; |
788 | ||
a74609fa | 789 | pcp = &zone_pcp(zone, cpu)->pcp[cold]; |
1da177e4 | 790 | local_irq_save(flags); |
a74609fa | 791 | if (!pcp->count) { |
1da177e4 LT |
792 | pcp->count += rmqueue_bulk(zone, 0, |
793 | pcp->batch, &pcp->list); | |
a74609fa NP |
794 | if (unlikely(!pcp->count)) |
795 | goto failed; | |
1da177e4 | 796 | } |
a74609fa NP |
797 | page = list_entry(pcp->list.next, struct page, lru); |
798 | list_del(&page->lru); | |
799 | pcp->count--; | |
7fb1d9fc | 800 | } else { |
1da177e4 LT |
801 | spin_lock_irqsave(&zone->lock, flags); |
802 | page = __rmqueue(zone, order); | |
a74609fa NP |
803 | spin_unlock(&zone->lock); |
804 | if (!page) | |
805 | goto failed; | |
1da177e4 LT |
806 | } |
807 | ||
a74609fa | 808 | __mod_page_state_zone(zone, pgalloc, 1 << order); |
ca889e6c | 809 | zone_statistics(zonelist, zone); |
a74609fa NP |
810 | local_irq_restore(flags); |
811 | put_cpu(); | |
1da177e4 | 812 | |
a74609fa | 813 | BUG_ON(bad_range(zone, page)); |
17cf4406 | 814 | if (prep_new_page(page, order, gfp_flags)) |
a74609fa | 815 | goto again; |
1da177e4 | 816 | return page; |
a74609fa NP |
817 | |
818 | failed: | |
819 | local_irq_restore(flags); | |
820 | put_cpu(); | |
821 | return NULL; | |
1da177e4 LT |
822 | } |
823 | ||
7fb1d9fc | 824 | #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */ |
3148890b NP |
825 | #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */ |
826 | #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */ | |
827 | #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */ | |
828 | #define ALLOC_HARDER 0x10 /* try to alloc harder */ | |
829 | #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */ | |
830 | #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ | |
7fb1d9fc | 831 | |
1da177e4 LT |
832 | /* |
833 | * Return 1 if free pages are above 'mark'. This takes into account the order | |
834 | * of the allocation. | |
835 | */ | |
836 | int zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
7fb1d9fc | 837 | int classzone_idx, int alloc_flags) |
1da177e4 LT |
838 | { |
839 | /* free_pages my go negative - that's OK */ | |
840 | long min = mark, free_pages = z->free_pages - (1 << order) + 1; | |
841 | int o; | |
842 | ||
7fb1d9fc | 843 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 844 | min -= min / 2; |
7fb1d9fc | 845 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 LT |
846 | min -= min / 4; |
847 | ||
848 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
849 | return 0; | |
850 | for (o = 0; o < order; o++) { | |
851 | /* At the next order, this order's pages become unavailable */ | |
852 | free_pages -= z->free_area[o].nr_free << o; | |
853 | ||
854 | /* Require fewer higher order pages to be free */ | |
855 | min >>= 1; | |
856 | ||
857 | if (free_pages <= min) | |
858 | return 0; | |
859 | } | |
860 | return 1; | |
861 | } | |
862 | ||
7fb1d9fc RS |
863 | /* |
864 | * get_page_from_freeliest goes through the zonelist trying to allocate | |
865 | * a page. | |
866 | */ | |
867 | static struct page * | |
868 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, | |
869 | struct zonelist *zonelist, int alloc_flags) | |
753ee728 | 870 | { |
7fb1d9fc RS |
871 | struct zone **z = zonelist->zones; |
872 | struct page *page = NULL; | |
873 | int classzone_idx = zone_idx(*z); | |
874 | ||
875 | /* | |
876 | * Go through the zonelist once, looking for a zone with enough free. | |
877 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. | |
878 | */ | |
879 | do { | |
880 | if ((alloc_flags & ALLOC_CPUSET) && | |
881 | !cpuset_zone_allowed(*z, gfp_mask)) | |
882 | continue; | |
883 | ||
884 | if (!(alloc_flags & ALLOC_NO_WATERMARKS)) { | |
3148890b NP |
885 | unsigned long mark; |
886 | if (alloc_flags & ALLOC_WMARK_MIN) | |
887 | mark = (*z)->pages_min; | |
888 | else if (alloc_flags & ALLOC_WMARK_LOW) | |
889 | mark = (*z)->pages_low; | |
890 | else | |
891 | mark = (*z)->pages_high; | |
892 | if (!zone_watermark_ok(*z, order, mark, | |
7fb1d9fc | 893 | classzone_idx, alloc_flags)) |
9eeff239 CL |
894 | if (!zone_reclaim_mode || |
895 | !zone_reclaim(*z, gfp_mask, order)) | |
896 | continue; | |
7fb1d9fc RS |
897 | } |
898 | ||
a74609fa | 899 | page = buffered_rmqueue(zonelist, *z, order, gfp_mask); |
7fb1d9fc | 900 | if (page) { |
7fb1d9fc RS |
901 | break; |
902 | } | |
903 | } while (*(++z) != NULL); | |
904 | return page; | |
753ee728 MH |
905 | } |
906 | ||
1da177e4 LT |
907 | /* |
908 | * This is the 'heart' of the zoned buddy allocator. | |
909 | */ | |
910 | struct page * fastcall | |
dd0fc66f | 911 | __alloc_pages(gfp_t gfp_mask, unsigned int order, |
1da177e4 LT |
912 | struct zonelist *zonelist) |
913 | { | |
260b2367 | 914 | const gfp_t wait = gfp_mask & __GFP_WAIT; |
7fb1d9fc | 915 | struct zone **z; |
1da177e4 LT |
916 | struct page *page; |
917 | struct reclaim_state reclaim_state; | |
918 | struct task_struct *p = current; | |
1da177e4 | 919 | int do_retry; |
7fb1d9fc | 920 | int alloc_flags; |
1da177e4 LT |
921 | int did_some_progress; |
922 | ||
923 | might_sleep_if(wait); | |
924 | ||
6b1de916 | 925 | restart: |
7fb1d9fc | 926 | z = zonelist->zones; /* the list of zones suitable for gfp_mask */ |
1da177e4 | 927 | |
7fb1d9fc | 928 | if (unlikely(*z == NULL)) { |
1da177e4 LT |
929 | /* Should this ever happen?? */ |
930 | return NULL; | |
931 | } | |
6b1de916 | 932 | |
7fb1d9fc | 933 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, |
3148890b | 934 | zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET); |
7fb1d9fc RS |
935 | if (page) |
936 | goto got_pg; | |
1da177e4 | 937 | |
6b1de916 | 938 | do { |
43b0bc00 | 939 | wakeup_kswapd(*z, order); |
6b1de916 | 940 | } while (*(++z)); |
1da177e4 | 941 | |
9bf2229f | 942 | /* |
7fb1d9fc RS |
943 | * OK, we're below the kswapd watermark and have kicked background |
944 | * reclaim. Now things get more complex, so set up alloc_flags according | |
945 | * to how we want to proceed. | |
946 | * | |
947 | * The caller may dip into page reserves a bit more if the caller | |
948 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
4eac915d PJ |
949 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will |
950 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). | |
9bf2229f | 951 | */ |
3148890b | 952 | alloc_flags = ALLOC_WMARK_MIN; |
7fb1d9fc RS |
953 | if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait) |
954 | alloc_flags |= ALLOC_HARDER; | |
955 | if (gfp_mask & __GFP_HIGH) | |
956 | alloc_flags |= ALLOC_HIGH; | |
bdd804f4 PJ |
957 | if (wait) |
958 | alloc_flags |= ALLOC_CPUSET; | |
1da177e4 LT |
959 | |
960 | /* | |
961 | * Go through the zonelist again. Let __GFP_HIGH and allocations | |
7fb1d9fc | 962 | * coming from realtime tasks go deeper into reserves. |
1da177e4 LT |
963 | * |
964 | * This is the last chance, in general, before the goto nopage. | |
965 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. | |
9bf2229f | 966 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1da177e4 | 967 | */ |
7fb1d9fc RS |
968 | page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags); |
969 | if (page) | |
970 | goto got_pg; | |
1da177e4 LT |
971 | |
972 | /* This allocation should allow future memory freeing. */ | |
b84a35be NP |
973 | |
974 | if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE))) | |
975 | && !in_interrupt()) { | |
976 | if (!(gfp_mask & __GFP_NOMEMALLOC)) { | |
885036d3 | 977 | nofail_alloc: |
b84a35be | 978 | /* go through the zonelist yet again, ignoring mins */ |
7fb1d9fc | 979 | page = get_page_from_freelist(gfp_mask, order, |
47f3a867 | 980 | zonelist, ALLOC_NO_WATERMARKS); |
7fb1d9fc RS |
981 | if (page) |
982 | goto got_pg; | |
885036d3 KK |
983 | if (gfp_mask & __GFP_NOFAIL) { |
984 | blk_congestion_wait(WRITE, HZ/50); | |
985 | goto nofail_alloc; | |
986 | } | |
1da177e4 LT |
987 | } |
988 | goto nopage; | |
989 | } | |
990 | ||
991 | /* Atomic allocations - we can't balance anything */ | |
992 | if (!wait) | |
993 | goto nopage; | |
994 | ||
995 | rebalance: | |
996 | cond_resched(); | |
997 | ||
998 | /* We now go into synchronous reclaim */ | |
3e0d98b9 | 999 | cpuset_memory_pressure_bump(); |
1da177e4 LT |
1000 | p->flags |= PF_MEMALLOC; |
1001 | reclaim_state.reclaimed_slab = 0; | |
1002 | p->reclaim_state = &reclaim_state; | |
1003 | ||
7fb1d9fc | 1004 | did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask); |
1da177e4 LT |
1005 | |
1006 | p->reclaim_state = NULL; | |
1007 | p->flags &= ~PF_MEMALLOC; | |
1008 | ||
1009 | cond_resched(); | |
1010 | ||
1011 | if (likely(did_some_progress)) { | |
7fb1d9fc RS |
1012 | page = get_page_from_freelist(gfp_mask, order, |
1013 | zonelist, alloc_flags); | |
1014 | if (page) | |
1015 | goto got_pg; | |
1da177e4 LT |
1016 | } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) { |
1017 | /* | |
1018 | * Go through the zonelist yet one more time, keep | |
1019 | * very high watermark here, this is only to catch | |
1020 | * a parallel oom killing, we must fail if we're still | |
1021 | * under heavy pressure. | |
1022 | */ | |
7fb1d9fc | 1023 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, |
3148890b | 1024 | zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET); |
7fb1d9fc RS |
1025 | if (page) |
1026 | goto got_pg; | |
1da177e4 | 1027 | |
9b0f8b04 | 1028 | out_of_memory(zonelist, gfp_mask, order); |
1da177e4 LT |
1029 | goto restart; |
1030 | } | |
1031 | ||
1032 | /* | |
1033 | * Don't let big-order allocations loop unless the caller explicitly | |
1034 | * requests that. Wait for some write requests to complete then retry. | |
1035 | * | |
1036 | * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order | |
1037 | * <= 3, but that may not be true in other implementations. | |
1038 | */ | |
1039 | do_retry = 0; | |
1040 | if (!(gfp_mask & __GFP_NORETRY)) { | |
1041 | if ((order <= 3) || (gfp_mask & __GFP_REPEAT)) | |
1042 | do_retry = 1; | |
1043 | if (gfp_mask & __GFP_NOFAIL) | |
1044 | do_retry = 1; | |
1045 | } | |
1046 | if (do_retry) { | |
1047 | blk_congestion_wait(WRITE, HZ/50); | |
1048 | goto rebalance; | |
1049 | } | |
1050 | ||
1051 | nopage: | |
1052 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) { | |
1053 | printk(KERN_WARNING "%s: page allocation failure." | |
1054 | " order:%d, mode:0x%x\n", | |
1055 | p->comm, order, gfp_mask); | |
1056 | dump_stack(); | |
578c2fd6 | 1057 | show_mem(); |
1da177e4 | 1058 | } |
1da177e4 | 1059 | got_pg: |
1da177e4 LT |
1060 | return page; |
1061 | } | |
1062 | ||
1063 | EXPORT_SYMBOL(__alloc_pages); | |
1064 | ||
1065 | /* | |
1066 | * Common helper functions. | |
1067 | */ | |
dd0fc66f | 1068 | fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 LT |
1069 | { |
1070 | struct page * page; | |
1071 | page = alloc_pages(gfp_mask, order); | |
1072 | if (!page) | |
1073 | return 0; | |
1074 | return (unsigned long) page_address(page); | |
1075 | } | |
1076 | ||
1077 | EXPORT_SYMBOL(__get_free_pages); | |
1078 | ||
dd0fc66f | 1079 | fastcall unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 LT |
1080 | { |
1081 | struct page * page; | |
1082 | ||
1083 | /* | |
1084 | * get_zeroed_page() returns a 32-bit address, which cannot represent | |
1085 | * a highmem page | |
1086 | */ | |
260b2367 | 1087 | BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); |
1da177e4 LT |
1088 | |
1089 | page = alloc_pages(gfp_mask | __GFP_ZERO, 0); | |
1090 | if (page) | |
1091 | return (unsigned long) page_address(page); | |
1092 | return 0; | |
1093 | } | |
1094 | ||
1095 | EXPORT_SYMBOL(get_zeroed_page); | |
1096 | ||
1097 | void __pagevec_free(struct pagevec *pvec) | |
1098 | { | |
1099 | int i = pagevec_count(pvec); | |
1100 | ||
1101 | while (--i >= 0) | |
1102 | free_hot_cold_page(pvec->pages[i], pvec->cold); | |
1103 | } | |
1104 | ||
1105 | fastcall void __free_pages(struct page *page, unsigned int order) | |
1106 | { | |
b5810039 | 1107 | if (put_page_testzero(page)) { |
1da177e4 LT |
1108 | if (order == 0) |
1109 | free_hot_page(page); | |
1110 | else | |
1111 | __free_pages_ok(page, order); | |
1112 | } | |
1113 | } | |
1114 | ||
1115 | EXPORT_SYMBOL(__free_pages); | |
1116 | ||
1117 | fastcall void free_pages(unsigned long addr, unsigned int order) | |
1118 | { | |
1119 | if (addr != 0) { | |
1120 | BUG_ON(!virt_addr_valid((void *)addr)); | |
1121 | __free_pages(virt_to_page((void *)addr), order); | |
1122 | } | |
1123 | } | |
1124 | ||
1125 | EXPORT_SYMBOL(free_pages); | |
1126 | ||
1127 | /* | |
1128 | * Total amount of free (allocatable) RAM: | |
1129 | */ | |
1130 | unsigned int nr_free_pages(void) | |
1131 | { | |
1132 | unsigned int sum = 0; | |
1133 | struct zone *zone; | |
1134 | ||
1135 | for_each_zone(zone) | |
1136 | sum += zone->free_pages; | |
1137 | ||
1138 | return sum; | |
1139 | } | |
1140 | ||
1141 | EXPORT_SYMBOL(nr_free_pages); | |
1142 | ||
1143 | #ifdef CONFIG_NUMA | |
1144 | unsigned int nr_free_pages_pgdat(pg_data_t *pgdat) | |
1145 | { | |
1146 | unsigned int i, sum = 0; | |
1147 | ||
1148 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1149 | sum += pgdat->node_zones[i].free_pages; | |
1150 | ||
1151 | return sum; | |
1152 | } | |
1153 | #endif | |
1154 | ||
1155 | static unsigned int nr_free_zone_pages(int offset) | |
1156 | { | |
e310fd43 MB |
1157 | /* Just pick one node, since fallback list is circular */ |
1158 | pg_data_t *pgdat = NODE_DATA(numa_node_id()); | |
1da177e4 LT |
1159 | unsigned int sum = 0; |
1160 | ||
e310fd43 MB |
1161 | struct zonelist *zonelist = pgdat->node_zonelists + offset; |
1162 | struct zone **zonep = zonelist->zones; | |
1163 | struct zone *zone; | |
1da177e4 | 1164 | |
e310fd43 MB |
1165 | for (zone = *zonep++; zone; zone = *zonep++) { |
1166 | unsigned long size = zone->present_pages; | |
1167 | unsigned long high = zone->pages_high; | |
1168 | if (size > high) | |
1169 | sum += size - high; | |
1da177e4 LT |
1170 | } |
1171 | ||
1172 | return sum; | |
1173 | } | |
1174 | ||
1175 | /* | |
1176 | * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL | |
1177 | */ | |
1178 | unsigned int nr_free_buffer_pages(void) | |
1179 | { | |
af4ca457 | 1180 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 LT |
1181 | } |
1182 | ||
1183 | /* | |
1184 | * Amount of free RAM allocatable within all zones | |
1185 | */ | |
1186 | unsigned int nr_free_pagecache_pages(void) | |
1187 | { | |
af4ca457 | 1188 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER)); |
1da177e4 LT |
1189 | } |
1190 | ||
1191 | #ifdef CONFIG_HIGHMEM | |
1192 | unsigned int nr_free_highpages (void) | |
1193 | { | |
1194 | pg_data_t *pgdat; | |
1195 | unsigned int pages = 0; | |
1196 | ||
ec936fc5 | 1197 | for_each_online_pgdat(pgdat) |
1da177e4 LT |
1198 | pages += pgdat->node_zones[ZONE_HIGHMEM].free_pages; |
1199 | ||
1200 | return pages; | |
1201 | } | |
1202 | #endif | |
1203 | ||
1204 | #ifdef CONFIG_NUMA | |
1205 | static void show_node(struct zone *zone) | |
1206 | { | |
1207 | printk("Node %d ", zone->zone_pgdat->node_id); | |
1208 | } | |
1209 | #else | |
1210 | #define show_node(zone) do { } while (0) | |
1211 | #endif | |
1212 | ||
1da177e4 LT |
1213 | void si_meminfo(struct sysinfo *val) |
1214 | { | |
1215 | val->totalram = totalram_pages; | |
1216 | val->sharedram = 0; | |
1217 | val->freeram = nr_free_pages(); | |
1218 | val->bufferram = nr_blockdev_pages(); | |
1219 | #ifdef CONFIG_HIGHMEM | |
1220 | val->totalhigh = totalhigh_pages; | |
1221 | val->freehigh = nr_free_highpages(); | |
1222 | #else | |
1223 | val->totalhigh = 0; | |
1224 | val->freehigh = 0; | |
1225 | #endif | |
1226 | val->mem_unit = PAGE_SIZE; | |
1227 | } | |
1228 | ||
1229 | EXPORT_SYMBOL(si_meminfo); | |
1230 | ||
1231 | #ifdef CONFIG_NUMA | |
1232 | void si_meminfo_node(struct sysinfo *val, int nid) | |
1233 | { | |
1234 | pg_data_t *pgdat = NODE_DATA(nid); | |
1235 | ||
1236 | val->totalram = pgdat->node_present_pages; | |
1237 | val->freeram = nr_free_pages_pgdat(pgdat); | |
1238 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages; | |
1239 | val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages; | |
1240 | val->mem_unit = PAGE_SIZE; | |
1241 | } | |
1242 | #endif | |
1243 | ||
1244 | #define K(x) ((x) << (PAGE_SHIFT-10)) | |
1245 | ||
1246 | /* | |
1247 | * Show free area list (used inside shift_scroll-lock stuff) | |
1248 | * We also calculate the percentage fragmentation. We do this by counting the | |
1249 | * memory on each free list with the exception of the first item on the list. | |
1250 | */ | |
1251 | void show_free_areas(void) | |
1252 | { | |
1da177e4 LT |
1253 | int cpu, temperature; |
1254 | unsigned long active; | |
1255 | unsigned long inactive; | |
1256 | unsigned long free; | |
1257 | struct zone *zone; | |
1258 | ||
1259 | for_each_zone(zone) { | |
1260 | show_node(zone); | |
1261 | printk("%s per-cpu:", zone->name); | |
1262 | ||
f3fe6512 | 1263 | if (!populated_zone(zone)) { |
1da177e4 LT |
1264 | printk(" empty\n"); |
1265 | continue; | |
1266 | } else | |
1267 | printk("\n"); | |
1268 | ||
6b482c67 | 1269 | for_each_online_cpu(cpu) { |
1da177e4 LT |
1270 | struct per_cpu_pageset *pageset; |
1271 | ||
e7c8d5c9 | 1272 | pageset = zone_pcp(zone, cpu); |
1da177e4 LT |
1273 | |
1274 | for (temperature = 0; temperature < 2; temperature++) | |
2d92c5c9 | 1275 | printk("cpu %d %s: high %d, batch %d used:%d\n", |
1da177e4 LT |
1276 | cpu, |
1277 | temperature ? "cold" : "hot", | |
1da177e4 | 1278 | pageset->pcp[temperature].high, |
4ae7c039 CL |
1279 | pageset->pcp[temperature].batch, |
1280 | pageset->pcp[temperature].count); | |
1da177e4 LT |
1281 | } |
1282 | } | |
1283 | ||
1da177e4 LT |
1284 | get_zone_counts(&active, &inactive, &free); |
1285 | ||
c0d62219 | 1286 | printk("Free pages: %11ukB (%ukB HighMem)\n", |
1da177e4 LT |
1287 | K(nr_free_pages()), |
1288 | K(nr_free_highpages())); | |
1289 | ||
1290 | printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu " | |
1291 | "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n", | |
1292 | active, | |
1293 | inactive, | |
b1e7a8fd | 1294 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 1295 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 1296 | global_page_state(NR_UNSTABLE_NFS), |
1da177e4 | 1297 | nr_free_pages(), |
9a865ffa | 1298 | global_page_state(NR_SLAB), |
65ba55f5 | 1299 | global_page_state(NR_FILE_MAPPED), |
df849a15 | 1300 | global_page_state(NR_PAGETABLE)); |
1da177e4 LT |
1301 | |
1302 | for_each_zone(zone) { | |
1303 | int i; | |
1304 | ||
1305 | show_node(zone); | |
1306 | printk("%s" | |
1307 | " free:%lukB" | |
1308 | " min:%lukB" | |
1309 | " low:%lukB" | |
1310 | " high:%lukB" | |
1311 | " active:%lukB" | |
1312 | " inactive:%lukB" | |
1313 | " present:%lukB" | |
1314 | " pages_scanned:%lu" | |
1315 | " all_unreclaimable? %s" | |
1316 | "\n", | |
1317 | zone->name, | |
1318 | K(zone->free_pages), | |
1319 | K(zone->pages_min), | |
1320 | K(zone->pages_low), | |
1321 | K(zone->pages_high), | |
1322 | K(zone->nr_active), | |
1323 | K(zone->nr_inactive), | |
1324 | K(zone->present_pages), | |
1325 | zone->pages_scanned, | |
1326 | (zone->all_unreclaimable ? "yes" : "no") | |
1327 | ); | |
1328 | printk("lowmem_reserve[]:"); | |
1329 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1330 | printk(" %lu", zone->lowmem_reserve[i]); | |
1331 | printk("\n"); | |
1332 | } | |
1333 | ||
1334 | for_each_zone(zone) { | |
8f9de51a | 1335 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
1da177e4 LT |
1336 | |
1337 | show_node(zone); | |
1338 | printk("%s: ", zone->name); | |
f3fe6512 | 1339 | if (!populated_zone(zone)) { |
1da177e4 LT |
1340 | printk("empty\n"); |
1341 | continue; | |
1342 | } | |
1343 | ||
1344 | spin_lock_irqsave(&zone->lock, flags); | |
1345 | for (order = 0; order < MAX_ORDER; order++) { | |
8f9de51a KK |
1346 | nr[order] = zone->free_area[order].nr_free; |
1347 | total += nr[order] << order; | |
1da177e4 LT |
1348 | } |
1349 | spin_unlock_irqrestore(&zone->lock, flags); | |
8f9de51a KK |
1350 | for (order = 0; order < MAX_ORDER; order++) |
1351 | printk("%lu*%lukB ", nr[order], K(1UL) << order); | |
1da177e4 LT |
1352 | printk("= %lukB\n", K(total)); |
1353 | } | |
1354 | ||
1355 | show_swap_cache_info(); | |
1356 | } | |
1357 | ||
1358 | /* | |
1359 | * Builds allocation fallback zone lists. | |
1a93205b CL |
1360 | * |
1361 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 1362 | */ |
86356ab1 | 1363 | static int __meminit build_zonelists_node(pg_data_t *pgdat, |
070f8032 | 1364 | struct zonelist *zonelist, int nr_zones, int zone_type) |
1da177e4 | 1365 | { |
1a93205b CL |
1366 | struct zone *zone; |
1367 | ||
070f8032 | 1368 | BUG_ON(zone_type > ZONE_HIGHMEM); |
02a68a5e CL |
1369 | |
1370 | do { | |
070f8032 | 1371 | zone = pgdat->node_zones + zone_type; |
1a93205b | 1372 | if (populated_zone(zone)) { |
1da177e4 | 1373 | #ifndef CONFIG_HIGHMEM |
070f8032 | 1374 | BUG_ON(zone_type > ZONE_NORMAL); |
1da177e4 | 1375 | #endif |
070f8032 CL |
1376 | zonelist->zones[nr_zones++] = zone; |
1377 | check_highest_zone(zone_type); | |
1da177e4 | 1378 | } |
070f8032 | 1379 | zone_type--; |
02a68a5e | 1380 | |
070f8032 CL |
1381 | } while (zone_type >= 0); |
1382 | return nr_zones; | |
1da177e4 LT |
1383 | } |
1384 | ||
260b2367 AV |
1385 | static inline int highest_zone(int zone_bits) |
1386 | { | |
1387 | int res = ZONE_NORMAL; | |
1388 | if (zone_bits & (__force int)__GFP_HIGHMEM) | |
1389 | res = ZONE_HIGHMEM; | |
a2f1b424 AK |
1390 | if (zone_bits & (__force int)__GFP_DMA32) |
1391 | res = ZONE_DMA32; | |
260b2367 AV |
1392 | if (zone_bits & (__force int)__GFP_DMA) |
1393 | res = ZONE_DMA; | |
1394 | return res; | |
1395 | } | |
1396 | ||
1da177e4 LT |
1397 | #ifdef CONFIG_NUMA |
1398 | #define MAX_NODE_LOAD (num_online_nodes()) | |
86356ab1 | 1399 | static int __meminitdata node_load[MAX_NUMNODES]; |
1da177e4 | 1400 | /** |
4dc3b16b | 1401 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
1402 | * @node: node whose fallback list we're appending |
1403 | * @used_node_mask: nodemask_t of already used nodes | |
1404 | * | |
1405 | * We use a number of factors to determine which is the next node that should | |
1406 | * appear on a given node's fallback list. The node should not have appeared | |
1407 | * already in @node's fallback list, and it should be the next closest node | |
1408 | * according to the distance array (which contains arbitrary distance values | |
1409 | * from each node to each node in the system), and should also prefer nodes | |
1410 | * with no CPUs, since presumably they'll have very little allocation pressure | |
1411 | * on them otherwise. | |
1412 | * It returns -1 if no node is found. | |
1413 | */ | |
86356ab1 | 1414 | static int __meminit find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 1415 | { |
4cf808eb | 1416 | int n, val; |
1da177e4 LT |
1417 | int min_val = INT_MAX; |
1418 | int best_node = -1; | |
1419 | ||
4cf808eb LT |
1420 | /* Use the local node if we haven't already */ |
1421 | if (!node_isset(node, *used_node_mask)) { | |
1422 | node_set(node, *used_node_mask); | |
1423 | return node; | |
1424 | } | |
1da177e4 | 1425 | |
4cf808eb LT |
1426 | for_each_online_node(n) { |
1427 | cpumask_t tmp; | |
1da177e4 LT |
1428 | |
1429 | /* Don't want a node to appear more than once */ | |
1430 | if (node_isset(n, *used_node_mask)) | |
1431 | continue; | |
1432 | ||
1da177e4 LT |
1433 | /* Use the distance array to find the distance */ |
1434 | val = node_distance(node, n); | |
1435 | ||
4cf808eb LT |
1436 | /* Penalize nodes under us ("prefer the next node") */ |
1437 | val += (n < node); | |
1438 | ||
1da177e4 LT |
1439 | /* Give preference to headless and unused nodes */ |
1440 | tmp = node_to_cpumask(n); | |
1441 | if (!cpus_empty(tmp)) | |
1442 | val += PENALTY_FOR_NODE_WITH_CPUS; | |
1443 | ||
1444 | /* Slight preference for less loaded node */ | |
1445 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
1446 | val += node_load[n]; | |
1447 | ||
1448 | if (val < min_val) { | |
1449 | min_val = val; | |
1450 | best_node = n; | |
1451 | } | |
1452 | } | |
1453 | ||
1454 | if (best_node >= 0) | |
1455 | node_set(best_node, *used_node_mask); | |
1456 | ||
1457 | return best_node; | |
1458 | } | |
1459 | ||
86356ab1 | 1460 | static void __meminit build_zonelists(pg_data_t *pgdat) |
1da177e4 LT |
1461 | { |
1462 | int i, j, k, node, local_node; | |
1463 | int prev_node, load; | |
1464 | struct zonelist *zonelist; | |
1465 | nodemask_t used_mask; | |
1466 | ||
1467 | /* initialize zonelists */ | |
1468 | for (i = 0; i < GFP_ZONETYPES; i++) { | |
1469 | zonelist = pgdat->node_zonelists + i; | |
1470 | zonelist->zones[0] = NULL; | |
1471 | } | |
1472 | ||
1473 | /* NUMA-aware ordering of nodes */ | |
1474 | local_node = pgdat->node_id; | |
1475 | load = num_online_nodes(); | |
1476 | prev_node = local_node; | |
1477 | nodes_clear(used_mask); | |
1478 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { | |
9eeff239 CL |
1479 | int distance = node_distance(local_node, node); |
1480 | ||
1481 | /* | |
1482 | * If another node is sufficiently far away then it is better | |
1483 | * to reclaim pages in a zone before going off node. | |
1484 | */ | |
1485 | if (distance > RECLAIM_DISTANCE) | |
1486 | zone_reclaim_mode = 1; | |
1487 | ||
1da177e4 LT |
1488 | /* |
1489 | * We don't want to pressure a particular node. | |
1490 | * So adding penalty to the first node in same | |
1491 | * distance group to make it round-robin. | |
1492 | */ | |
9eeff239 CL |
1493 | |
1494 | if (distance != node_distance(local_node, prev_node)) | |
1da177e4 LT |
1495 | node_load[node] += load; |
1496 | prev_node = node; | |
1497 | load--; | |
1498 | for (i = 0; i < GFP_ZONETYPES; i++) { | |
1499 | zonelist = pgdat->node_zonelists + i; | |
1500 | for (j = 0; zonelist->zones[j] != NULL; j++); | |
1501 | ||
260b2367 | 1502 | k = highest_zone(i); |
1da177e4 LT |
1503 | |
1504 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, k); | |
1505 | zonelist->zones[j] = NULL; | |
1506 | } | |
1507 | } | |
1508 | } | |
1509 | ||
1510 | #else /* CONFIG_NUMA */ | |
1511 | ||
86356ab1 | 1512 | static void __meminit build_zonelists(pg_data_t *pgdat) |
1da177e4 LT |
1513 | { |
1514 | int i, j, k, node, local_node; | |
1515 | ||
1516 | local_node = pgdat->node_id; | |
1517 | for (i = 0; i < GFP_ZONETYPES; i++) { | |
1518 | struct zonelist *zonelist; | |
1519 | ||
1520 | zonelist = pgdat->node_zonelists + i; | |
1521 | ||
1522 | j = 0; | |
260b2367 | 1523 | k = highest_zone(i); |
1da177e4 LT |
1524 | j = build_zonelists_node(pgdat, zonelist, j, k); |
1525 | /* | |
1526 | * Now we build the zonelist so that it contains the zones | |
1527 | * of all the other nodes. | |
1528 | * We don't want to pressure a particular node, so when | |
1529 | * building the zones for node N, we make sure that the | |
1530 | * zones coming right after the local ones are those from | |
1531 | * node N+1 (modulo N) | |
1532 | */ | |
1533 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
1534 | if (!node_online(node)) | |
1535 | continue; | |
1536 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, k); | |
1537 | } | |
1538 | for (node = 0; node < local_node; node++) { | |
1539 | if (!node_online(node)) | |
1540 | continue; | |
1541 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, k); | |
1542 | } | |
1543 | ||
1544 | zonelist->zones[j] = NULL; | |
1545 | } | |
1546 | } | |
1547 | ||
1548 | #endif /* CONFIG_NUMA */ | |
1549 | ||
6811378e YG |
1550 | /* return values int ....just for stop_machine_run() */ |
1551 | static int __meminit __build_all_zonelists(void *dummy) | |
1da177e4 | 1552 | { |
6811378e YG |
1553 | int nid; |
1554 | for_each_online_node(nid) | |
1555 | build_zonelists(NODE_DATA(nid)); | |
1556 | return 0; | |
1557 | } | |
1558 | ||
1559 | void __meminit build_all_zonelists(void) | |
1560 | { | |
1561 | if (system_state == SYSTEM_BOOTING) { | |
1562 | __build_all_zonelists(0); | |
1563 | cpuset_init_current_mems_allowed(); | |
1564 | } else { | |
1565 | /* we have to stop all cpus to guaranntee there is no user | |
1566 | of zonelist */ | |
1567 | stop_machine_run(__build_all_zonelists, NULL, NR_CPUS); | |
1568 | /* cpuset refresh routine should be here */ | |
1569 | } | |
bd1e22b8 AM |
1570 | vm_total_pages = nr_free_pagecache_pages(); |
1571 | printk("Built %i zonelists. Total pages: %ld\n", | |
1572 | num_online_nodes(), vm_total_pages); | |
1da177e4 LT |
1573 | } |
1574 | ||
1575 | /* | |
1576 | * Helper functions to size the waitqueue hash table. | |
1577 | * Essentially these want to choose hash table sizes sufficiently | |
1578 | * large so that collisions trying to wait on pages are rare. | |
1579 | * But in fact, the number of active page waitqueues on typical | |
1580 | * systems is ridiculously low, less than 200. So this is even | |
1581 | * conservative, even though it seems large. | |
1582 | * | |
1583 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
1584 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
1585 | */ | |
1586 | #define PAGES_PER_WAITQUEUE 256 | |
1587 | ||
cca448fe | 1588 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 1589 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
1590 | { |
1591 | unsigned long size = 1; | |
1592 | ||
1593 | pages /= PAGES_PER_WAITQUEUE; | |
1594 | ||
1595 | while (size < pages) | |
1596 | size <<= 1; | |
1597 | ||
1598 | /* | |
1599 | * Once we have dozens or even hundreds of threads sleeping | |
1600 | * on IO we've got bigger problems than wait queue collision. | |
1601 | * Limit the size of the wait table to a reasonable size. | |
1602 | */ | |
1603 | size = min(size, 4096UL); | |
1604 | ||
1605 | return max(size, 4UL); | |
1606 | } | |
cca448fe YG |
1607 | #else |
1608 | /* | |
1609 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
1610 | * a suitable size for its wait_table. So we use the maximum size now. | |
1611 | * | |
1612 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
1613 | * | |
1614 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
1615 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
1616 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
1617 | * | |
1618 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
1619 | * or more by the traditional way. (See above). It equals: | |
1620 | * | |
1621 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
1622 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
1623 | * powerpc (64K page size) : = (32G +16M)byte. | |
1624 | */ | |
1625 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
1626 | { | |
1627 | return 4096UL; | |
1628 | } | |
1629 | #endif | |
1da177e4 LT |
1630 | |
1631 | /* | |
1632 | * This is an integer logarithm so that shifts can be used later | |
1633 | * to extract the more random high bits from the multiplicative | |
1634 | * hash function before the remainder is taken. | |
1635 | */ | |
1636 | static inline unsigned long wait_table_bits(unsigned long size) | |
1637 | { | |
1638 | return ffz(~size); | |
1639 | } | |
1640 | ||
1641 | #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1)) | |
1642 | ||
1643 | static void __init calculate_zone_totalpages(struct pglist_data *pgdat, | |
1644 | unsigned long *zones_size, unsigned long *zholes_size) | |
1645 | { | |
1646 | unsigned long realtotalpages, totalpages = 0; | |
1647 | int i; | |
1648 | ||
1649 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1650 | totalpages += zones_size[i]; | |
1651 | pgdat->node_spanned_pages = totalpages; | |
1652 | ||
1653 | realtotalpages = totalpages; | |
1654 | if (zholes_size) | |
1655 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1656 | realtotalpages -= zholes_size[i]; | |
1657 | pgdat->node_present_pages = realtotalpages; | |
1658 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages); | |
1659 | } | |
1660 | ||
1661 | ||
1662 | /* | |
1663 | * Initially all pages are reserved - free ones are freed | |
1664 | * up by free_all_bootmem() once the early boot process is | |
1665 | * done. Non-atomic initialization, single-pass. | |
1666 | */ | |
c09b4240 | 1667 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
1da177e4 LT |
1668 | unsigned long start_pfn) |
1669 | { | |
1da177e4 | 1670 | struct page *page; |
29751f69 AW |
1671 | unsigned long end_pfn = start_pfn + size; |
1672 | unsigned long pfn; | |
1da177e4 | 1673 | |
cbe8dd4a | 1674 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
d41dee36 AW |
1675 | if (!early_pfn_valid(pfn)) |
1676 | continue; | |
1677 | page = pfn_to_page(pfn); | |
1678 | set_page_links(page, zone, nid, pfn); | |
7835e98b | 1679 | init_page_count(page); |
1da177e4 LT |
1680 | reset_page_mapcount(page); |
1681 | SetPageReserved(page); | |
1682 | INIT_LIST_HEAD(&page->lru); | |
1683 | #ifdef WANT_PAGE_VIRTUAL | |
1684 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
1685 | if (!is_highmem_idx(zone)) | |
3212c6be | 1686 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1da177e4 | 1687 | #endif |
1da177e4 LT |
1688 | } |
1689 | } | |
1690 | ||
1691 | void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone, | |
1692 | unsigned long size) | |
1693 | { | |
1694 | int order; | |
1695 | for (order = 0; order < MAX_ORDER ; order++) { | |
1696 | INIT_LIST_HEAD(&zone->free_area[order].free_list); | |
1697 | zone->free_area[order].nr_free = 0; | |
1698 | } | |
1699 | } | |
1700 | ||
d41dee36 AW |
1701 | #define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr) |
1702 | void zonetable_add(struct zone *zone, int nid, int zid, unsigned long pfn, | |
1703 | unsigned long size) | |
1704 | { | |
1705 | unsigned long snum = pfn_to_section_nr(pfn); | |
1706 | unsigned long end = pfn_to_section_nr(pfn + size); | |
1707 | ||
1708 | if (FLAGS_HAS_NODE) | |
1709 | zone_table[ZONETABLE_INDEX(nid, zid)] = zone; | |
1710 | else | |
1711 | for (; snum <= end; snum++) | |
1712 | zone_table[ZONETABLE_INDEX(snum, zid)] = zone; | |
1713 | } | |
1714 | ||
1da177e4 LT |
1715 | #ifndef __HAVE_ARCH_MEMMAP_INIT |
1716 | #define memmap_init(size, nid, zone, start_pfn) \ | |
1717 | memmap_init_zone((size), (nid), (zone), (start_pfn)) | |
1718 | #endif | |
1719 | ||
6292d9aa | 1720 | static int __cpuinit zone_batchsize(struct zone *zone) |
e7c8d5c9 CL |
1721 | { |
1722 | int batch; | |
1723 | ||
1724 | /* | |
1725 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 1726 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
1727 | * |
1728 | * OK, so we don't know how big the cache is. So guess. | |
1729 | */ | |
1730 | batch = zone->present_pages / 1024; | |
ba56e91c SR |
1731 | if (batch * PAGE_SIZE > 512 * 1024) |
1732 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
1733 | batch /= 4; /* We effectively *= 4 below */ |
1734 | if (batch < 1) | |
1735 | batch = 1; | |
1736 | ||
1737 | /* | |
0ceaacc9 NP |
1738 | * Clamp the batch to a 2^n - 1 value. Having a power |
1739 | * of 2 value was found to be more likely to have | |
1740 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 1741 | * |
0ceaacc9 NP |
1742 | * For example if 2 tasks are alternately allocating |
1743 | * batches of pages, one task can end up with a lot | |
1744 | * of pages of one half of the possible page colors | |
1745 | * and the other with pages of the other colors. | |
e7c8d5c9 | 1746 | */ |
0ceaacc9 | 1747 | batch = (1 << (fls(batch + batch/2)-1)) - 1; |
ba56e91c | 1748 | |
e7c8d5c9 CL |
1749 | return batch; |
1750 | } | |
1751 | ||
2caaad41 CL |
1752 | inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
1753 | { | |
1754 | struct per_cpu_pages *pcp; | |
1755 | ||
1c6fe946 MD |
1756 | memset(p, 0, sizeof(*p)); |
1757 | ||
2caaad41 CL |
1758 | pcp = &p->pcp[0]; /* hot */ |
1759 | pcp->count = 0; | |
2caaad41 CL |
1760 | pcp->high = 6 * batch; |
1761 | pcp->batch = max(1UL, 1 * batch); | |
1762 | INIT_LIST_HEAD(&pcp->list); | |
1763 | ||
1764 | pcp = &p->pcp[1]; /* cold*/ | |
1765 | pcp->count = 0; | |
2caaad41 | 1766 | pcp->high = 2 * batch; |
e46a5e28 | 1767 | pcp->batch = max(1UL, batch/2); |
2caaad41 CL |
1768 | INIT_LIST_HEAD(&pcp->list); |
1769 | } | |
1770 | ||
8ad4b1fb RS |
1771 | /* |
1772 | * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist | |
1773 | * to the value high for the pageset p. | |
1774 | */ | |
1775 | ||
1776 | static void setup_pagelist_highmark(struct per_cpu_pageset *p, | |
1777 | unsigned long high) | |
1778 | { | |
1779 | struct per_cpu_pages *pcp; | |
1780 | ||
1781 | pcp = &p->pcp[0]; /* hot list */ | |
1782 | pcp->high = high; | |
1783 | pcp->batch = max(1UL, high/4); | |
1784 | if ((high/4) > (PAGE_SHIFT * 8)) | |
1785 | pcp->batch = PAGE_SHIFT * 8; | |
1786 | } | |
1787 | ||
1788 | ||
e7c8d5c9 CL |
1789 | #ifdef CONFIG_NUMA |
1790 | /* | |
2caaad41 CL |
1791 | * Boot pageset table. One per cpu which is going to be used for all |
1792 | * zones and all nodes. The parameters will be set in such a way | |
1793 | * that an item put on a list will immediately be handed over to | |
1794 | * the buddy list. This is safe since pageset manipulation is done | |
1795 | * with interrupts disabled. | |
1796 | * | |
1797 | * Some NUMA counter updates may also be caught by the boot pagesets. | |
b7c84c6a CL |
1798 | * |
1799 | * The boot_pagesets must be kept even after bootup is complete for | |
1800 | * unused processors and/or zones. They do play a role for bootstrapping | |
1801 | * hotplugged processors. | |
1802 | * | |
1803 | * zoneinfo_show() and maybe other functions do | |
1804 | * not check if the processor is online before following the pageset pointer. | |
1805 | * Other parts of the kernel may not check if the zone is available. | |
2caaad41 | 1806 | */ |
88a2a4ac | 1807 | static struct per_cpu_pageset boot_pageset[NR_CPUS]; |
2caaad41 CL |
1808 | |
1809 | /* | |
1810 | * Dynamically allocate memory for the | |
e7c8d5c9 CL |
1811 | * per cpu pageset array in struct zone. |
1812 | */ | |
6292d9aa | 1813 | static int __cpuinit process_zones(int cpu) |
e7c8d5c9 CL |
1814 | { |
1815 | struct zone *zone, *dzone; | |
e7c8d5c9 CL |
1816 | |
1817 | for_each_zone(zone) { | |
e7c8d5c9 | 1818 | |
23316bc8 | 1819 | zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset), |
e7c8d5c9 | 1820 | GFP_KERNEL, cpu_to_node(cpu)); |
23316bc8 | 1821 | if (!zone_pcp(zone, cpu)) |
e7c8d5c9 | 1822 | goto bad; |
e7c8d5c9 | 1823 | |
23316bc8 | 1824 | setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone)); |
8ad4b1fb RS |
1825 | |
1826 | if (percpu_pagelist_fraction) | |
1827 | setup_pagelist_highmark(zone_pcp(zone, cpu), | |
1828 | (zone->present_pages / percpu_pagelist_fraction)); | |
e7c8d5c9 CL |
1829 | } |
1830 | ||
1831 | return 0; | |
1832 | bad: | |
1833 | for_each_zone(dzone) { | |
1834 | if (dzone == zone) | |
1835 | break; | |
23316bc8 NP |
1836 | kfree(zone_pcp(dzone, cpu)); |
1837 | zone_pcp(dzone, cpu) = NULL; | |
e7c8d5c9 CL |
1838 | } |
1839 | return -ENOMEM; | |
1840 | } | |
1841 | ||
1842 | static inline void free_zone_pagesets(int cpu) | |
1843 | { | |
e7c8d5c9 CL |
1844 | struct zone *zone; |
1845 | ||
1846 | for_each_zone(zone) { | |
1847 | struct per_cpu_pageset *pset = zone_pcp(zone, cpu); | |
1848 | ||
1849 | zone_pcp(zone, cpu) = NULL; | |
1850 | kfree(pset); | |
1851 | } | |
e7c8d5c9 CL |
1852 | } |
1853 | ||
9c7b216d | 1854 | static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb, |
e7c8d5c9 CL |
1855 | unsigned long action, |
1856 | void *hcpu) | |
1857 | { | |
1858 | int cpu = (long)hcpu; | |
1859 | int ret = NOTIFY_OK; | |
1860 | ||
1861 | switch (action) { | |
1862 | case CPU_UP_PREPARE: | |
1863 | if (process_zones(cpu)) | |
1864 | ret = NOTIFY_BAD; | |
1865 | break; | |
b0d41693 | 1866 | case CPU_UP_CANCELED: |
e7c8d5c9 CL |
1867 | case CPU_DEAD: |
1868 | free_zone_pagesets(cpu); | |
1869 | break; | |
e7c8d5c9 CL |
1870 | default: |
1871 | break; | |
1872 | } | |
1873 | return ret; | |
1874 | } | |
1875 | ||
74b85f37 | 1876 | static struct notifier_block __cpuinitdata pageset_notifier = |
e7c8d5c9 CL |
1877 | { &pageset_cpuup_callback, NULL, 0 }; |
1878 | ||
78d9955b | 1879 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 CL |
1880 | { |
1881 | int err; | |
1882 | ||
1883 | /* Initialize per_cpu_pageset for cpu 0. | |
1884 | * A cpuup callback will do this for every cpu | |
1885 | * as it comes online | |
1886 | */ | |
1887 | err = process_zones(smp_processor_id()); | |
1888 | BUG_ON(err); | |
1889 | register_cpu_notifier(&pageset_notifier); | |
1890 | } | |
1891 | ||
1892 | #endif | |
1893 | ||
c09b4240 | 1894 | static __meminit |
cca448fe | 1895 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
1896 | { |
1897 | int i; | |
1898 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe | 1899 | size_t alloc_size; |
ed8ece2e DH |
1900 | |
1901 | /* | |
1902 | * The per-page waitqueue mechanism uses hashed waitqueues | |
1903 | * per zone. | |
1904 | */ | |
02b694de YG |
1905 | zone->wait_table_hash_nr_entries = |
1906 | wait_table_hash_nr_entries(zone_size_pages); | |
1907 | zone->wait_table_bits = | |
1908 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
1909 | alloc_size = zone->wait_table_hash_nr_entries |
1910 | * sizeof(wait_queue_head_t); | |
1911 | ||
1912 | if (system_state == SYSTEM_BOOTING) { | |
1913 | zone->wait_table = (wait_queue_head_t *) | |
1914 | alloc_bootmem_node(pgdat, alloc_size); | |
1915 | } else { | |
1916 | /* | |
1917 | * This case means that a zone whose size was 0 gets new memory | |
1918 | * via memory hot-add. | |
1919 | * But it may be the case that a new node was hot-added. In | |
1920 | * this case vmalloc() will not be able to use this new node's | |
1921 | * memory - this wait_table must be initialized to use this new | |
1922 | * node itself as well. | |
1923 | * To use this new node's memory, further consideration will be | |
1924 | * necessary. | |
1925 | */ | |
1926 | zone->wait_table = (wait_queue_head_t *)vmalloc(alloc_size); | |
1927 | } | |
1928 | if (!zone->wait_table) | |
1929 | return -ENOMEM; | |
ed8ece2e | 1930 | |
02b694de | 1931 | for(i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 1932 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
1933 | |
1934 | return 0; | |
ed8ece2e DH |
1935 | } |
1936 | ||
c09b4240 | 1937 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e DH |
1938 | { |
1939 | int cpu; | |
1940 | unsigned long batch = zone_batchsize(zone); | |
1941 | ||
1942 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
1943 | #ifdef CONFIG_NUMA | |
1944 | /* Early boot. Slab allocator not functional yet */ | |
23316bc8 | 1945 | zone_pcp(zone, cpu) = &boot_pageset[cpu]; |
ed8ece2e DH |
1946 | setup_pageset(&boot_pageset[cpu],0); |
1947 | #else | |
1948 | setup_pageset(zone_pcp(zone,cpu), batch); | |
1949 | #endif | |
1950 | } | |
f5335c0f AB |
1951 | if (zone->present_pages) |
1952 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n", | |
1953 | zone->name, zone->present_pages, batch); | |
ed8ece2e DH |
1954 | } |
1955 | ||
718127cc YG |
1956 | __meminit int init_currently_empty_zone(struct zone *zone, |
1957 | unsigned long zone_start_pfn, | |
1958 | unsigned long size) | |
ed8ece2e DH |
1959 | { |
1960 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
1961 | int ret; |
1962 | ret = zone_wait_table_init(zone, size); | |
1963 | if (ret) | |
1964 | return ret; | |
ed8ece2e DH |
1965 | pgdat->nr_zones = zone_idx(zone) + 1; |
1966 | ||
ed8ece2e DH |
1967 | zone->zone_start_pfn = zone_start_pfn; |
1968 | ||
1969 | memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn); | |
1970 | ||
1971 | zone_init_free_lists(pgdat, zone, zone->spanned_pages); | |
718127cc YG |
1972 | |
1973 | return 0; | |
ed8ece2e DH |
1974 | } |
1975 | ||
1da177e4 LT |
1976 | /* |
1977 | * Set up the zone data structures: | |
1978 | * - mark all pages reserved | |
1979 | * - mark all memory queues empty | |
1980 | * - clear the memory bitmaps | |
1981 | */ | |
86356ab1 | 1982 | static void __meminit free_area_init_core(struct pglist_data *pgdat, |
1da177e4 LT |
1983 | unsigned long *zones_size, unsigned long *zholes_size) |
1984 | { | |
ed8ece2e DH |
1985 | unsigned long j; |
1986 | int nid = pgdat->node_id; | |
1da177e4 | 1987 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
718127cc | 1988 | int ret; |
1da177e4 | 1989 | |
208d54e5 | 1990 | pgdat_resize_init(pgdat); |
1da177e4 LT |
1991 | pgdat->nr_zones = 0; |
1992 | init_waitqueue_head(&pgdat->kswapd_wait); | |
1993 | pgdat->kswapd_max_order = 0; | |
1994 | ||
1995 | for (j = 0; j < MAX_NR_ZONES; j++) { | |
1996 | struct zone *zone = pgdat->node_zones + j; | |
1997 | unsigned long size, realsize; | |
1da177e4 | 1998 | |
1da177e4 LT |
1999 | realsize = size = zones_size[j]; |
2000 | if (zholes_size) | |
2001 | realsize -= zholes_size[j]; | |
2002 | ||
a2f1b424 | 2003 | if (j < ZONE_HIGHMEM) |
1da177e4 LT |
2004 | nr_kernel_pages += realsize; |
2005 | nr_all_pages += realsize; | |
2006 | ||
2007 | zone->spanned_pages = size; | |
2008 | zone->present_pages = realsize; | |
2009 | zone->name = zone_names[j]; | |
2010 | spin_lock_init(&zone->lock); | |
2011 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 2012 | zone_seqlock_init(zone); |
1da177e4 LT |
2013 | zone->zone_pgdat = pgdat; |
2014 | zone->free_pages = 0; | |
2015 | ||
2016 | zone->temp_priority = zone->prev_priority = DEF_PRIORITY; | |
2017 | ||
ed8ece2e | 2018 | zone_pcp_init(zone); |
1da177e4 LT |
2019 | INIT_LIST_HEAD(&zone->active_list); |
2020 | INIT_LIST_HEAD(&zone->inactive_list); | |
2021 | zone->nr_scan_active = 0; | |
2022 | zone->nr_scan_inactive = 0; | |
2023 | zone->nr_active = 0; | |
2024 | zone->nr_inactive = 0; | |
2244b95a | 2025 | zap_zone_vm_stats(zone); |
53e9a615 | 2026 | atomic_set(&zone->reclaim_in_progress, 0); |
1da177e4 LT |
2027 | if (!size) |
2028 | continue; | |
2029 | ||
d41dee36 | 2030 | zonetable_add(zone, nid, j, zone_start_pfn, size); |
718127cc YG |
2031 | ret = init_currently_empty_zone(zone, zone_start_pfn, size); |
2032 | BUG_ON(ret); | |
1da177e4 | 2033 | zone_start_pfn += size; |
1da177e4 LT |
2034 | } |
2035 | } | |
2036 | ||
2037 | static void __init alloc_node_mem_map(struct pglist_data *pgdat) | |
2038 | { | |
1da177e4 LT |
2039 | /* Skip empty nodes */ |
2040 | if (!pgdat->node_spanned_pages) | |
2041 | return; | |
2042 | ||
d41dee36 | 2043 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
2044 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
2045 | if (!pgdat->node_mem_map) { | |
e984bb43 | 2046 | unsigned long size, start, end; |
d41dee36 AW |
2047 | struct page *map; |
2048 | ||
e984bb43 BP |
2049 | /* |
2050 | * The zone's endpoints aren't required to be MAX_ORDER | |
2051 | * aligned but the node_mem_map endpoints must be in order | |
2052 | * for the buddy allocator to function correctly. | |
2053 | */ | |
2054 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
2055 | end = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
2056 | end = ALIGN(end, MAX_ORDER_NR_PAGES); | |
2057 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
2058 | map = alloc_remap(pgdat->node_id, size); |
2059 | if (!map) | |
2060 | map = alloc_bootmem_node(pgdat, size); | |
e984bb43 | 2061 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
1da177e4 | 2062 | } |
d41dee36 | 2063 | #ifdef CONFIG_FLATMEM |
1da177e4 LT |
2064 | /* |
2065 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
2066 | */ | |
2067 | if (pgdat == NODE_DATA(0)) | |
2068 | mem_map = NODE_DATA(0)->node_mem_map; | |
2069 | #endif | |
d41dee36 | 2070 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
2071 | } |
2072 | ||
86356ab1 | 2073 | void __meminit free_area_init_node(int nid, struct pglist_data *pgdat, |
1da177e4 LT |
2074 | unsigned long *zones_size, unsigned long node_start_pfn, |
2075 | unsigned long *zholes_size) | |
2076 | { | |
2077 | pgdat->node_id = nid; | |
2078 | pgdat->node_start_pfn = node_start_pfn; | |
2079 | calculate_zone_totalpages(pgdat, zones_size, zholes_size); | |
2080 | ||
2081 | alloc_node_mem_map(pgdat); | |
2082 | ||
2083 | free_area_init_core(pgdat, zones_size, zholes_size); | |
2084 | } | |
2085 | ||
93b7504e | 2086 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
2087 | static bootmem_data_t contig_bootmem_data; |
2088 | struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data }; | |
2089 | ||
2090 | EXPORT_SYMBOL(contig_page_data); | |
93b7504e | 2091 | #endif |
1da177e4 LT |
2092 | |
2093 | void __init free_area_init(unsigned long *zones_size) | |
2094 | { | |
93b7504e | 2095 | free_area_init_node(0, NODE_DATA(0), zones_size, |
1da177e4 LT |
2096 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
2097 | } | |
1da177e4 | 2098 | |
1da177e4 LT |
2099 | #ifdef CONFIG_HOTPLUG_CPU |
2100 | static int page_alloc_cpu_notify(struct notifier_block *self, | |
2101 | unsigned long action, void *hcpu) | |
2102 | { | |
2103 | int cpu = (unsigned long)hcpu; | |
1da177e4 LT |
2104 | unsigned long *src, *dest; |
2105 | ||
2106 | if (action == CPU_DEAD) { | |
2107 | int i; | |
2108 | ||
1da177e4 LT |
2109 | local_irq_disable(); |
2110 | __drain_pages(cpu); | |
2111 | ||
2112 | /* Add dead cpu's page_states to our own. */ | |
2113 | dest = (unsigned long *)&__get_cpu_var(page_states); | |
2114 | src = (unsigned long *)&per_cpu(page_states, cpu); | |
2115 | ||
2116 | for (i = 0; i < sizeof(struct page_state)/sizeof(unsigned long); | |
2117 | i++) { | |
2118 | dest[i] += src[i]; | |
2119 | src[i] = 0; | |
2120 | } | |
2121 | ||
2122 | local_irq_enable(); | |
2244b95a | 2123 | refresh_cpu_vm_stats(cpu); |
1da177e4 LT |
2124 | } |
2125 | return NOTIFY_OK; | |
2126 | } | |
2127 | #endif /* CONFIG_HOTPLUG_CPU */ | |
2128 | ||
2129 | void __init page_alloc_init(void) | |
2130 | { | |
2131 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
2132 | } | |
2133 | ||
cb45b0e9 HA |
2134 | /* |
2135 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio | |
2136 | * or min_free_kbytes changes. | |
2137 | */ | |
2138 | static void calculate_totalreserve_pages(void) | |
2139 | { | |
2140 | struct pglist_data *pgdat; | |
2141 | unsigned long reserve_pages = 0; | |
2142 | int i, j; | |
2143 | ||
2144 | for_each_online_pgdat(pgdat) { | |
2145 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2146 | struct zone *zone = pgdat->node_zones + i; | |
2147 | unsigned long max = 0; | |
2148 | ||
2149 | /* Find valid and maximum lowmem_reserve in the zone */ | |
2150 | for (j = i; j < MAX_NR_ZONES; j++) { | |
2151 | if (zone->lowmem_reserve[j] > max) | |
2152 | max = zone->lowmem_reserve[j]; | |
2153 | } | |
2154 | ||
2155 | /* we treat pages_high as reserved pages. */ | |
2156 | max += zone->pages_high; | |
2157 | ||
2158 | if (max > zone->present_pages) | |
2159 | max = zone->present_pages; | |
2160 | reserve_pages += max; | |
2161 | } | |
2162 | } | |
2163 | totalreserve_pages = reserve_pages; | |
2164 | } | |
2165 | ||
1da177e4 LT |
2166 | /* |
2167 | * setup_per_zone_lowmem_reserve - called whenever | |
2168 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
2169 | * has a correct pages reserved value, so an adequate number of | |
2170 | * pages are left in the zone after a successful __alloc_pages(). | |
2171 | */ | |
2172 | static void setup_per_zone_lowmem_reserve(void) | |
2173 | { | |
2174 | struct pglist_data *pgdat; | |
2175 | int j, idx; | |
2176 | ||
ec936fc5 | 2177 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
2178 | for (j = 0; j < MAX_NR_ZONES; j++) { |
2179 | struct zone *zone = pgdat->node_zones + j; | |
2180 | unsigned long present_pages = zone->present_pages; | |
2181 | ||
2182 | zone->lowmem_reserve[j] = 0; | |
2183 | ||
2184 | for (idx = j-1; idx >= 0; idx--) { | |
2185 | struct zone *lower_zone; | |
2186 | ||
2187 | if (sysctl_lowmem_reserve_ratio[idx] < 1) | |
2188 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
2189 | ||
2190 | lower_zone = pgdat->node_zones + idx; | |
2191 | lower_zone->lowmem_reserve[j] = present_pages / | |
2192 | sysctl_lowmem_reserve_ratio[idx]; | |
2193 | present_pages += lower_zone->present_pages; | |
2194 | } | |
2195 | } | |
2196 | } | |
cb45b0e9 HA |
2197 | |
2198 | /* update totalreserve_pages */ | |
2199 | calculate_totalreserve_pages(); | |
1da177e4 LT |
2200 | } |
2201 | ||
2202 | /* | |
2203 | * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures | |
2204 | * that the pages_{min,low,high} values for each zone are set correctly | |
2205 | * with respect to min_free_kbytes. | |
2206 | */ | |
3947be19 | 2207 | void setup_per_zone_pages_min(void) |
1da177e4 LT |
2208 | { |
2209 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
2210 | unsigned long lowmem_pages = 0; | |
2211 | struct zone *zone; | |
2212 | unsigned long flags; | |
2213 | ||
2214 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
2215 | for_each_zone(zone) { | |
2216 | if (!is_highmem(zone)) | |
2217 | lowmem_pages += zone->present_pages; | |
2218 | } | |
2219 | ||
2220 | for_each_zone(zone) { | |
ac924c60 AM |
2221 | u64 tmp; |
2222 | ||
1da177e4 | 2223 | spin_lock_irqsave(&zone->lru_lock, flags); |
ac924c60 AM |
2224 | tmp = (u64)pages_min * zone->present_pages; |
2225 | do_div(tmp, lowmem_pages); | |
1da177e4 LT |
2226 | if (is_highmem(zone)) { |
2227 | /* | |
669ed175 NP |
2228 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
2229 | * need highmem pages, so cap pages_min to a small | |
2230 | * value here. | |
2231 | * | |
2232 | * The (pages_high-pages_low) and (pages_low-pages_min) | |
2233 | * deltas controls asynch page reclaim, and so should | |
2234 | * not be capped for highmem. | |
1da177e4 LT |
2235 | */ |
2236 | int min_pages; | |
2237 | ||
2238 | min_pages = zone->present_pages / 1024; | |
2239 | if (min_pages < SWAP_CLUSTER_MAX) | |
2240 | min_pages = SWAP_CLUSTER_MAX; | |
2241 | if (min_pages > 128) | |
2242 | min_pages = 128; | |
2243 | zone->pages_min = min_pages; | |
2244 | } else { | |
669ed175 NP |
2245 | /* |
2246 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
2247 | * proportionate to the zone's size. |
2248 | */ | |
669ed175 | 2249 | zone->pages_min = tmp; |
1da177e4 LT |
2250 | } |
2251 | ||
ac924c60 AM |
2252 | zone->pages_low = zone->pages_min + (tmp >> 2); |
2253 | zone->pages_high = zone->pages_min + (tmp >> 1); | |
1da177e4 LT |
2254 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
2255 | } | |
cb45b0e9 HA |
2256 | |
2257 | /* update totalreserve_pages */ | |
2258 | calculate_totalreserve_pages(); | |
1da177e4 LT |
2259 | } |
2260 | ||
2261 | /* | |
2262 | * Initialise min_free_kbytes. | |
2263 | * | |
2264 | * For small machines we want it small (128k min). For large machines | |
2265 | * we want it large (64MB max). But it is not linear, because network | |
2266 | * bandwidth does not increase linearly with machine size. We use | |
2267 | * | |
2268 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: | |
2269 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) | |
2270 | * | |
2271 | * which yields | |
2272 | * | |
2273 | * 16MB: 512k | |
2274 | * 32MB: 724k | |
2275 | * 64MB: 1024k | |
2276 | * 128MB: 1448k | |
2277 | * 256MB: 2048k | |
2278 | * 512MB: 2896k | |
2279 | * 1024MB: 4096k | |
2280 | * 2048MB: 5792k | |
2281 | * 4096MB: 8192k | |
2282 | * 8192MB: 11584k | |
2283 | * 16384MB: 16384k | |
2284 | */ | |
2285 | static int __init init_per_zone_pages_min(void) | |
2286 | { | |
2287 | unsigned long lowmem_kbytes; | |
2288 | ||
2289 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
2290 | ||
2291 | min_free_kbytes = int_sqrt(lowmem_kbytes * 16); | |
2292 | if (min_free_kbytes < 128) | |
2293 | min_free_kbytes = 128; | |
2294 | if (min_free_kbytes > 65536) | |
2295 | min_free_kbytes = 65536; | |
2296 | setup_per_zone_pages_min(); | |
2297 | setup_per_zone_lowmem_reserve(); | |
2298 | return 0; | |
2299 | } | |
2300 | module_init(init_per_zone_pages_min) | |
2301 | ||
2302 | /* | |
2303 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so | |
2304 | * that we can call two helper functions whenever min_free_kbytes | |
2305 | * changes. | |
2306 | */ | |
2307 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, | |
2308 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
2309 | { | |
2310 | proc_dointvec(table, write, file, buffer, length, ppos); | |
2311 | setup_per_zone_pages_min(); | |
2312 | return 0; | |
2313 | } | |
2314 | ||
2315 | /* | |
2316 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
2317 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
2318 | * whenever sysctl_lowmem_reserve_ratio changes. | |
2319 | * | |
2320 | * The reserve ratio obviously has absolutely no relation with the | |
2321 | * pages_min watermarks. The lowmem reserve ratio can only make sense | |
2322 | * if in function of the boot time zone sizes. | |
2323 | */ | |
2324 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, | |
2325 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
2326 | { | |
2327 | proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
2328 | setup_per_zone_lowmem_reserve(); | |
2329 | return 0; | |
2330 | } | |
2331 | ||
8ad4b1fb RS |
2332 | /* |
2333 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
2334 | * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist | |
2335 | * can have before it gets flushed back to buddy allocator. | |
2336 | */ | |
2337 | ||
2338 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, | |
2339 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
2340 | { | |
2341 | struct zone *zone; | |
2342 | unsigned int cpu; | |
2343 | int ret; | |
2344 | ||
2345 | ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
2346 | if (!write || (ret == -EINVAL)) | |
2347 | return ret; | |
2348 | for_each_zone(zone) { | |
2349 | for_each_online_cpu(cpu) { | |
2350 | unsigned long high; | |
2351 | high = zone->present_pages / percpu_pagelist_fraction; | |
2352 | setup_pagelist_highmark(zone_pcp(zone, cpu), high); | |
2353 | } | |
2354 | } | |
2355 | return 0; | |
2356 | } | |
2357 | ||
1da177e4 LT |
2358 | __initdata int hashdist = HASHDIST_DEFAULT; |
2359 | ||
2360 | #ifdef CONFIG_NUMA | |
2361 | static int __init set_hashdist(char *str) | |
2362 | { | |
2363 | if (!str) | |
2364 | return 0; | |
2365 | hashdist = simple_strtoul(str, &str, 0); | |
2366 | return 1; | |
2367 | } | |
2368 | __setup("hashdist=", set_hashdist); | |
2369 | #endif | |
2370 | ||
2371 | /* | |
2372 | * allocate a large system hash table from bootmem | |
2373 | * - it is assumed that the hash table must contain an exact power-of-2 | |
2374 | * quantity of entries | |
2375 | * - limit is the number of hash buckets, not the total allocation size | |
2376 | */ | |
2377 | void *__init alloc_large_system_hash(const char *tablename, | |
2378 | unsigned long bucketsize, | |
2379 | unsigned long numentries, | |
2380 | int scale, | |
2381 | int flags, | |
2382 | unsigned int *_hash_shift, | |
2383 | unsigned int *_hash_mask, | |
2384 | unsigned long limit) | |
2385 | { | |
2386 | unsigned long long max = limit; | |
2387 | unsigned long log2qty, size; | |
2388 | void *table = NULL; | |
2389 | ||
2390 | /* allow the kernel cmdline to have a say */ | |
2391 | if (!numentries) { | |
2392 | /* round applicable memory size up to nearest megabyte */ | |
2393 | numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages; | |
2394 | numentries += (1UL << (20 - PAGE_SHIFT)) - 1; | |
2395 | numentries >>= 20 - PAGE_SHIFT; | |
2396 | numentries <<= 20 - PAGE_SHIFT; | |
2397 | ||
2398 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
2399 | if (scale > PAGE_SHIFT) | |
2400 | numentries >>= (scale - PAGE_SHIFT); | |
2401 | else | |
2402 | numentries <<= (PAGE_SHIFT - scale); | |
2403 | } | |
6e692ed3 | 2404 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
2405 | |
2406 | /* limit allocation size to 1/16 total memory by default */ | |
2407 | if (max == 0) { | |
2408 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
2409 | do_div(max, bucketsize); | |
2410 | } | |
2411 | ||
2412 | if (numentries > max) | |
2413 | numentries = max; | |
2414 | ||
2415 | log2qty = long_log2(numentries); | |
2416 | ||
2417 | do { | |
2418 | size = bucketsize << log2qty; | |
2419 | if (flags & HASH_EARLY) | |
2420 | table = alloc_bootmem(size); | |
2421 | else if (hashdist) | |
2422 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
2423 | else { | |
2424 | unsigned long order; | |
2425 | for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++) | |
2426 | ; | |
2427 | table = (void*) __get_free_pages(GFP_ATOMIC, order); | |
2428 | } | |
2429 | } while (!table && size > PAGE_SIZE && --log2qty); | |
2430 | ||
2431 | if (!table) | |
2432 | panic("Failed to allocate %s hash table\n", tablename); | |
2433 | ||
2434 | printk("%s hash table entries: %d (order: %d, %lu bytes)\n", | |
2435 | tablename, | |
2436 | (1U << log2qty), | |
2437 | long_log2(size) - PAGE_SHIFT, | |
2438 | size); | |
2439 | ||
2440 | if (_hash_shift) | |
2441 | *_hash_shift = log2qty; | |
2442 | if (_hash_mask) | |
2443 | *_hash_mask = (1 << log2qty) - 1; | |
2444 | ||
2445 | return table; | |
2446 | } | |
a117e66e KH |
2447 | |
2448 | #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE | |
a117e66e KH |
2449 | struct page *pfn_to_page(unsigned long pfn) |
2450 | { | |
67de6482 | 2451 | return __pfn_to_page(pfn); |
a117e66e KH |
2452 | } |
2453 | unsigned long page_to_pfn(struct page *page) | |
2454 | { | |
67de6482 | 2455 | return __page_to_pfn(page); |
a117e66e | 2456 | } |
a117e66e KH |
2457 | EXPORT_SYMBOL(pfn_to_page); |
2458 | EXPORT_SYMBOL(page_to_pfn); | |
2459 | #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */ |