Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/page_alloc.c | |
3 | * | |
4 | * Manages the free list, the system allocates free pages here. | |
5 | * Note that kmalloc() lives in slab.c | |
6 | * | |
7 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
8 | * Swap reorganised 29.12.95, Stephen Tweedie | |
9 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 | |
10 | * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999 | |
11 | * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999 | |
12 | * Zone balancing, Kanoj Sarcar, SGI, Jan 2000 | |
13 | * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002 | |
14 | * (lots of bits borrowed from Ingo Molnar & Andrew Morton) | |
15 | */ | |
16 | ||
1da177e4 LT |
17 | #include <linux/stddef.h> |
18 | #include <linux/mm.h> | |
19 | #include <linux/swap.h> | |
20 | #include <linux/interrupt.h> | |
21 | #include <linux/pagemap.h> | |
22 | #include <linux/bootmem.h> | |
23 | #include <linux/compiler.h> | |
9f158333 | 24 | #include <linux/kernel.h> |
1da177e4 LT |
25 | #include <linux/module.h> |
26 | #include <linux/suspend.h> | |
27 | #include <linux/pagevec.h> | |
28 | #include <linux/blkdev.h> | |
29 | #include <linux/slab.h> | |
30 | #include <linux/notifier.h> | |
31 | #include <linux/topology.h> | |
32 | #include <linux/sysctl.h> | |
33 | #include <linux/cpu.h> | |
34 | #include <linux/cpuset.h> | |
bdc8cb98 | 35 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
36 | #include <linux/nodemask.h> |
37 | #include <linux/vmalloc.h> | |
4be38e35 | 38 | #include <linux/mempolicy.h> |
6811378e | 39 | #include <linux/stop_machine.h> |
1da177e4 LT |
40 | |
41 | #include <asm/tlbflush.h> | |
ac924c60 | 42 | #include <asm/div64.h> |
1da177e4 LT |
43 | #include "internal.h" |
44 | ||
45 | /* | |
46 | * MCD - HACK: Find somewhere to initialize this EARLY, or make this | |
47 | * initializer cleaner | |
48 | */ | |
c3d8c141 | 49 | nodemask_t node_online_map __read_mostly = { { [0] = 1UL } }; |
7223a93a | 50 | EXPORT_SYMBOL(node_online_map); |
c3d8c141 | 51 | nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL; |
7223a93a | 52 | EXPORT_SYMBOL(node_possible_map); |
6c231b7b | 53 | unsigned long totalram_pages __read_mostly; |
cb45b0e9 | 54 | unsigned long totalreserve_pages __read_mostly; |
1da177e4 | 55 | long nr_swap_pages; |
8ad4b1fb | 56 | int percpu_pagelist_fraction; |
1da177e4 | 57 | |
d98c7a09 | 58 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 59 | |
1da177e4 LT |
60 | /* |
61 | * results with 256, 32 in the lowmem_reserve sysctl: | |
62 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
63 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
64 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
65 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
66 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA | |
a2f1b424 AK |
67 | * |
68 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
69 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 70 | */ |
2f1b6248 CL |
71 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
72 | 256, | |
fb0e7942 | 73 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 74 | 256, |
fb0e7942 | 75 | #endif |
e53ef38d | 76 | #ifdef CONFIG_HIGHMEM |
2f1b6248 | 77 | 32 |
e53ef38d | 78 | #endif |
2f1b6248 | 79 | }; |
1da177e4 LT |
80 | |
81 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 LT |
82 | |
83 | /* | |
84 | * Used by page_zone() to look up the address of the struct zone whose | |
85 | * id is encoded in the upper bits of page->flags | |
86 | */ | |
c3d8c141 | 87 | struct zone *zone_table[1 << ZONETABLE_SHIFT] __read_mostly; |
1da177e4 LT |
88 | EXPORT_SYMBOL(zone_table); |
89 | ||
2f1b6248 CL |
90 | static char *zone_names[MAX_NR_ZONES] = { |
91 | "DMA", | |
fb0e7942 | 92 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 93 | "DMA32", |
fb0e7942 | 94 | #endif |
2f1b6248 | 95 | "Normal", |
e53ef38d | 96 | #ifdef CONFIG_HIGHMEM |
2f1b6248 | 97 | "HighMem" |
e53ef38d | 98 | #endif |
2f1b6248 CL |
99 | }; |
100 | ||
1da177e4 LT |
101 | int min_free_kbytes = 1024; |
102 | ||
86356ab1 YG |
103 | unsigned long __meminitdata nr_kernel_pages; |
104 | unsigned long __meminitdata nr_all_pages; | |
1da177e4 | 105 | |
13e7444b | 106 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 107 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 108 | { |
bdc8cb98 DH |
109 | int ret = 0; |
110 | unsigned seq; | |
111 | unsigned long pfn = page_to_pfn(page); | |
c6a57e19 | 112 | |
bdc8cb98 DH |
113 | do { |
114 | seq = zone_span_seqbegin(zone); | |
115 | if (pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
116 | ret = 1; | |
117 | else if (pfn < zone->zone_start_pfn) | |
118 | ret = 1; | |
119 | } while (zone_span_seqretry(zone, seq)); | |
120 | ||
121 | return ret; | |
c6a57e19 DH |
122 | } |
123 | ||
124 | static int page_is_consistent(struct zone *zone, struct page *page) | |
125 | { | |
1da177e4 LT |
126 | #ifdef CONFIG_HOLES_IN_ZONE |
127 | if (!pfn_valid(page_to_pfn(page))) | |
c6a57e19 | 128 | return 0; |
1da177e4 LT |
129 | #endif |
130 | if (zone != page_zone(page)) | |
c6a57e19 DH |
131 | return 0; |
132 | ||
133 | return 1; | |
134 | } | |
135 | /* | |
136 | * Temporary debugging check for pages not lying within a given zone. | |
137 | */ | |
138 | static int bad_range(struct zone *zone, struct page *page) | |
139 | { | |
140 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 141 | return 1; |
c6a57e19 DH |
142 | if (!page_is_consistent(zone, page)) |
143 | return 1; | |
144 | ||
1da177e4 LT |
145 | return 0; |
146 | } | |
13e7444b NP |
147 | #else |
148 | static inline int bad_range(struct zone *zone, struct page *page) | |
149 | { | |
150 | return 0; | |
151 | } | |
152 | #endif | |
153 | ||
224abf92 | 154 | static void bad_page(struct page *page) |
1da177e4 | 155 | { |
224abf92 | 156 | printk(KERN_EMERG "Bad page state in process '%s'\n" |
7365f3d1 HD |
157 | KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n" |
158 | KERN_EMERG "Trying to fix it up, but a reboot is needed\n" | |
159 | KERN_EMERG "Backtrace:\n", | |
224abf92 NP |
160 | current->comm, page, (int)(2*sizeof(unsigned long)), |
161 | (unsigned long)page->flags, page->mapping, | |
162 | page_mapcount(page), page_count(page)); | |
1da177e4 | 163 | dump_stack(); |
334795ec HD |
164 | page->flags &= ~(1 << PG_lru | |
165 | 1 << PG_private | | |
1da177e4 | 166 | 1 << PG_locked | |
1da177e4 LT |
167 | 1 << PG_active | |
168 | 1 << PG_dirty | | |
334795ec HD |
169 | 1 << PG_reclaim | |
170 | 1 << PG_slab | | |
1da177e4 | 171 | 1 << PG_swapcache | |
676165a8 NP |
172 | 1 << PG_writeback | |
173 | 1 << PG_buddy ); | |
1da177e4 LT |
174 | set_page_count(page, 0); |
175 | reset_page_mapcount(page); | |
176 | page->mapping = NULL; | |
9f158333 | 177 | add_taint(TAINT_BAD_PAGE); |
1da177e4 LT |
178 | } |
179 | ||
1da177e4 LT |
180 | /* |
181 | * Higher-order pages are called "compound pages". They are structured thusly: | |
182 | * | |
183 | * The first PAGE_SIZE page is called the "head page". | |
184 | * | |
185 | * The remaining PAGE_SIZE pages are called "tail pages". | |
186 | * | |
187 | * All pages have PG_compound set. All pages have their ->private pointing at | |
188 | * the head page (even the head page has this). | |
189 | * | |
41d78ba5 HD |
190 | * The first tail page's ->lru.next holds the address of the compound page's |
191 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
192 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 193 | */ |
d98c7a09 HD |
194 | |
195 | static void free_compound_page(struct page *page) | |
196 | { | |
197 | __free_pages_ok(page, (unsigned long)page[1].lru.prev); | |
198 | } | |
199 | ||
1da177e4 LT |
200 | static void prep_compound_page(struct page *page, unsigned long order) |
201 | { | |
202 | int i; | |
203 | int nr_pages = 1 << order; | |
204 | ||
d98c7a09 | 205 | page[1].lru.next = (void *)free_compound_page; /* set dtor */ |
41d78ba5 | 206 | page[1].lru.prev = (void *)order; |
1da177e4 LT |
207 | for (i = 0; i < nr_pages; i++) { |
208 | struct page *p = page + i; | |
209 | ||
5e9dace8 | 210 | __SetPageCompound(p); |
4c21e2f2 | 211 | set_page_private(p, (unsigned long)page); |
1da177e4 LT |
212 | } |
213 | } | |
214 | ||
215 | static void destroy_compound_page(struct page *page, unsigned long order) | |
216 | { | |
217 | int i; | |
218 | int nr_pages = 1 << order; | |
219 | ||
41d78ba5 | 220 | if (unlikely((unsigned long)page[1].lru.prev != order)) |
224abf92 | 221 | bad_page(page); |
1da177e4 LT |
222 | |
223 | for (i = 0; i < nr_pages; i++) { | |
224 | struct page *p = page + i; | |
225 | ||
224abf92 NP |
226 | if (unlikely(!PageCompound(p) | |
227 | (page_private(p) != (unsigned long)page))) | |
228 | bad_page(page); | |
5e9dace8 | 229 | __ClearPageCompound(p); |
1da177e4 LT |
230 | } |
231 | } | |
1da177e4 | 232 | |
17cf4406 NP |
233 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) |
234 | { | |
235 | int i; | |
236 | ||
725d704e | 237 | VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM); |
6626c5d5 AM |
238 | /* |
239 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO | |
240 | * and __GFP_HIGHMEM from hard or soft interrupt context. | |
241 | */ | |
725d704e | 242 | VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); |
17cf4406 NP |
243 | for (i = 0; i < (1 << order); i++) |
244 | clear_highpage(page + i); | |
245 | } | |
246 | ||
1da177e4 LT |
247 | /* |
248 | * function for dealing with page's order in buddy system. | |
249 | * zone->lock is already acquired when we use these. | |
250 | * So, we don't need atomic page->flags operations here. | |
251 | */ | |
6aa3001b AM |
252 | static inline unsigned long page_order(struct page *page) |
253 | { | |
4c21e2f2 | 254 | return page_private(page); |
1da177e4 LT |
255 | } |
256 | ||
6aa3001b AM |
257 | static inline void set_page_order(struct page *page, int order) |
258 | { | |
4c21e2f2 | 259 | set_page_private(page, order); |
676165a8 | 260 | __SetPageBuddy(page); |
1da177e4 LT |
261 | } |
262 | ||
263 | static inline void rmv_page_order(struct page *page) | |
264 | { | |
676165a8 | 265 | __ClearPageBuddy(page); |
4c21e2f2 | 266 | set_page_private(page, 0); |
1da177e4 LT |
267 | } |
268 | ||
269 | /* | |
270 | * Locate the struct page for both the matching buddy in our | |
271 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
272 | * | |
273 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
274 | * the following equation: | |
275 | * B2 = B1 ^ (1 << O) | |
276 | * For example, if the starting buddy (buddy2) is #8 its order | |
277 | * 1 buddy is #10: | |
278 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
279 | * | |
280 | * 2) Any buddy B will have an order O+1 parent P which | |
281 | * satisfies the following equation: | |
282 | * P = B & ~(1 << O) | |
283 | * | |
d6e05edc | 284 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER |
1da177e4 LT |
285 | */ |
286 | static inline struct page * | |
287 | __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order) | |
288 | { | |
289 | unsigned long buddy_idx = page_idx ^ (1 << order); | |
290 | ||
291 | return page + (buddy_idx - page_idx); | |
292 | } | |
293 | ||
294 | static inline unsigned long | |
295 | __find_combined_index(unsigned long page_idx, unsigned int order) | |
296 | { | |
297 | return (page_idx & ~(1 << order)); | |
298 | } | |
299 | ||
300 | /* | |
301 | * This function checks whether a page is free && is the buddy | |
302 | * we can do coalesce a page and its buddy if | |
13e7444b | 303 | * (a) the buddy is not in a hole && |
676165a8 | 304 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
305 | * (c) a page and its buddy have the same order && |
306 | * (d) a page and its buddy are in the same zone. | |
676165a8 NP |
307 | * |
308 | * For recording whether a page is in the buddy system, we use PG_buddy. | |
309 | * Setting, clearing, and testing PG_buddy is serialized by zone->lock. | |
1da177e4 | 310 | * |
676165a8 | 311 | * For recording page's order, we use page_private(page). |
1da177e4 | 312 | */ |
cb2b95e1 AW |
313 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
314 | int order) | |
1da177e4 | 315 | { |
13e7444b | 316 | #ifdef CONFIG_HOLES_IN_ZONE |
cb2b95e1 | 317 | if (!pfn_valid(page_to_pfn(buddy))) |
13e7444b NP |
318 | return 0; |
319 | #endif | |
320 | ||
cb2b95e1 AW |
321 | if (page_zone_id(page) != page_zone_id(buddy)) |
322 | return 0; | |
323 | ||
324 | if (PageBuddy(buddy) && page_order(buddy) == order) { | |
325 | BUG_ON(page_count(buddy) != 0); | |
6aa3001b | 326 | return 1; |
676165a8 | 327 | } |
6aa3001b | 328 | return 0; |
1da177e4 LT |
329 | } |
330 | ||
331 | /* | |
332 | * Freeing function for a buddy system allocator. | |
333 | * | |
334 | * The concept of a buddy system is to maintain direct-mapped table | |
335 | * (containing bit values) for memory blocks of various "orders". | |
336 | * The bottom level table contains the map for the smallest allocatable | |
337 | * units of memory (here, pages), and each level above it describes | |
338 | * pairs of units from the levels below, hence, "buddies". | |
339 | * At a high level, all that happens here is marking the table entry | |
340 | * at the bottom level available, and propagating the changes upward | |
341 | * as necessary, plus some accounting needed to play nicely with other | |
342 | * parts of the VM system. | |
343 | * At each level, we keep a list of pages, which are heads of continuous | |
676165a8 | 344 | * free pages of length of (1 << order) and marked with PG_buddy. Page's |
4c21e2f2 | 345 | * order is recorded in page_private(page) field. |
1da177e4 LT |
346 | * So when we are allocating or freeing one, we can derive the state of the |
347 | * other. That is, if we allocate a small block, and both were | |
348 | * free, the remainder of the region must be split into blocks. | |
349 | * If a block is freed, and its buddy is also free, then this | |
350 | * triggers coalescing into a block of larger size. | |
351 | * | |
352 | * -- wli | |
353 | */ | |
354 | ||
48db57f8 | 355 | static inline void __free_one_page(struct page *page, |
1da177e4 LT |
356 | struct zone *zone, unsigned int order) |
357 | { | |
358 | unsigned long page_idx; | |
359 | int order_size = 1 << order; | |
360 | ||
224abf92 | 361 | if (unlikely(PageCompound(page))) |
1da177e4 LT |
362 | destroy_compound_page(page, order); |
363 | ||
364 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); | |
365 | ||
725d704e NP |
366 | VM_BUG_ON(page_idx & (order_size - 1)); |
367 | VM_BUG_ON(bad_range(zone, page)); | |
1da177e4 LT |
368 | |
369 | zone->free_pages += order_size; | |
370 | while (order < MAX_ORDER-1) { | |
371 | unsigned long combined_idx; | |
372 | struct free_area *area; | |
373 | struct page *buddy; | |
374 | ||
1da177e4 | 375 | buddy = __page_find_buddy(page, page_idx, order); |
cb2b95e1 | 376 | if (!page_is_buddy(page, buddy, order)) |
1da177e4 | 377 | break; /* Move the buddy up one level. */ |
13e7444b | 378 | |
1da177e4 LT |
379 | list_del(&buddy->lru); |
380 | area = zone->free_area + order; | |
381 | area->nr_free--; | |
382 | rmv_page_order(buddy); | |
13e7444b | 383 | combined_idx = __find_combined_index(page_idx, order); |
1da177e4 LT |
384 | page = page + (combined_idx - page_idx); |
385 | page_idx = combined_idx; | |
386 | order++; | |
387 | } | |
388 | set_page_order(page, order); | |
389 | list_add(&page->lru, &zone->free_area[order].free_list); | |
390 | zone->free_area[order].nr_free++; | |
391 | } | |
392 | ||
224abf92 | 393 | static inline int free_pages_check(struct page *page) |
1da177e4 | 394 | { |
92be2e33 NP |
395 | if (unlikely(page_mapcount(page) | |
396 | (page->mapping != NULL) | | |
397 | (page_count(page) != 0) | | |
1da177e4 LT |
398 | (page->flags & ( |
399 | 1 << PG_lru | | |
400 | 1 << PG_private | | |
401 | 1 << PG_locked | | |
402 | 1 << PG_active | | |
403 | 1 << PG_reclaim | | |
404 | 1 << PG_slab | | |
405 | 1 << PG_swapcache | | |
b5810039 | 406 | 1 << PG_writeback | |
676165a8 NP |
407 | 1 << PG_reserved | |
408 | 1 << PG_buddy )))) | |
224abf92 | 409 | bad_page(page); |
1da177e4 | 410 | if (PageDirty(page)) |
242e5468 | 411 | __ClearPageDirty(page); |
689bcebf HD |
412 | /* |
413 | * For now, we report if PG_reserved was found set, but do not | |
414 | * clear it, and do not free the page. But we shall soon need | |
415 | * to do more, for when the ZERO_PAGE count wraps negative. | |
416 | */ | |
417 | return PageReserved(page); | |
1da177e4 LT |
418 | } |
419 | ||
420 | /* | |
421 | * Frees a list of pages. | |
422 | * Assumes all pages on list are in same zone, and of same order. | |
207f36ee | 423 | * count is the number of pages to free. |
1da177e4 LT |
424 | * |
425 | * If the zone was previously in an "all pages pinned" state then look to | |
426 | * see if this freeing clears that state. | |
427 | * | |
428 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
429 | * pinned" detection logic. | |
430 | */ | |
48db57f8 NP |
431 | static void free_pages_bulk(struct zone *zone, int count, |
432 | struct list_head *list, int order) | |
1da177e4 | 433 | { |
c54ad30c | 434 | spin_lock(&zone->lock); |
1da177e4 LT |
435 | zone->all_unreclaimable = 0; |
436 | zone->pages_scanned = 0; | |
48db57f8 NP |
437 | while (count--) { |
438 | struct page *page; | |
439 | ||
725d704e | 440 | VM_BUG_ON(list_empty(list)); |
1da177e4 | 441 | page = list_entry(list->prev, struct page, lru); |
48db57f8 | 442 | /* have to delete it as __free_one_page list manipulates */ |
1da177e4 | 443 | list_del(&page->lru); |
48db57f8 | 444 | __free_one_page(page, zone, order); |
1da177e4 | 445 | } |
c54ad30c | 446 | spin_unlock(&zone->lock); |
1da177e4 LT |
447 | } |
448 | ||
48db57f8 | 449 | static void free_one_page(struct zone *zone, struct page *page, int order) |
1da177e4 | 450 | { |
006d22d9 CL |
451 | spin_lock(&zone->lock); |
452 | zone->all_unreclaimable = 0; | |
453 | zone->pages_scanned = 0; | |
454 | __free_one_page(page, zone ,order); | |
455 | spin_unlock(&zone->lock); | |
48db57f8 NP |
456 | } |
457 | ||
458 | static void __free_pages_ok(struct page *page, unsigned int order) | |
459 | { | |
460 | unsigned long flags; | |
1da177e4 | 461 | int i; |
689bcebf | 462 | int reserved = 0; |
1da177e4 LT |
463 | |
464 | arch_free_page(page, order); | |
de5097c2 | 465 | if (!PageHighMem(page)) |
f9b8404c IM |
466 | debug_check_no_locks_freed(page_address(page), |
467 | PAGE_SIZE<<order); | |
1da177e4 | 468 | |
1da177e4 | 469 | for (i = 0 ; i < (1 << order) ; ++i) |
224abf92 | 470 | reserved += free_pages_check(page + i); |
689bcebf HD |
471 | if (reserved) |
472 | return; | |
473 | ||
48db57f8 | 474 | kernel_map_pages(page, 1 << order, 0); |
c54ad30c | 475 | local_irq_save(flags); |
f8891e5e | 476 | __count_vm_events(PGFREE, 1 << order); |
48db57f8 | 477 | free_one_page(page_zone(page), page, order); |
c54ad30c | 478 | local_irq_restore(flags); |
1da177e4 LT |
479 | } |
480 | ||
a226f6c8 DH |
481 | /* |
482 | * permit the bootmem allocator to evade page validation on high-order frees | |
483 | */ | |
484 | void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order) | |
485 | { | |
486 | if (order == 0) { | |
487 | __ClearPageReserved(page); | |
488 | set_page_count(page, 0); | |
7835e98b | 489 | set_page_refcounted(page); |
545b1ea9 | 490 | __free_page(page); |
a226f6c8 | 491 | } else { |
a226f6c8 DH |
492 | int loop; |
493 | ||
545b1ea9 | 494 | prefetchw(page); |
a226f6c8 DH |
495 | for (loop = 0; loop < BITS_PER_LONG; loop++) { |
496 | struct page *p = &page[loop]; | |
497 | ||
545b1ea9 NP |
498 | if (loop + 1 < BITS_PER_LONG) |
499 | prefetchw(p + 1); | |
a226f6c8 DH |
500 | __ClearPageReserved(p); |
501 | set_page_count(p, 0); | |
502 | } | |
503 | ||
7835e98b | 504 | set_page_refcounted(page); |
545b1ea9 | 505 | __free_pages(page, order); |
a226f6c8 DH |
506 | } |
507 | } | |
508 | ||
1da177e4 LT |
509 | |
510 | /* | |
511 | * The order of subdivision here is critical for the IO subsystem. | |
512 | * Please do not alter this order without good reasons and regression | |
513 | * testing. Specifically, as large blocks of memory are subdivided, | |
514 | * the order in which smaller blocks are delivered depends on the order | |
515 | * they're subdivided in this function. This is the primary factor | |
516 | * influencing the order in which pages are delivered to the IO | |
517 | * subsystem according to empirical testing, and this is also justified | |
518 | * by considering the behavior of a buddy system containing a single | |
519 | * large block of memory acted on by a series of small allocations. | |
520 | * This behavior is a critical factor in sglist merging's success. | |
521 | * | |
522 | * -- wli | |
523 | */ | |
085cc7d5 | 524 | static inline void expand(struct zone *zone, struct page *page, |
1da177e4 LT |
525 | int low, int high, struct free_area *area) |
526 | { | |
527 | unsigned long size = 1 << high; | |
528 | ||
529 | while (high > low) { | |
530 | area--; | |
531 | high--; | |
532 | size >>= 1; | |
725d704e | 533 | VM_BUG_ON(bad_range(zone, &page[size])); |
1da177e4 LT |
534 | list_add(&page[size].lru, &area->free_list); |
535 | area->nr_free++; | |
536 | set_page_order(&page[size], high); | |
537 | } | |
1da177e4 LT |
538 | } |
539 | ||
1da177e4 LT |
540 | /* |
541 | * This page is about to be returned from the page allocator | |
542 | */ | |
17cf4406 | 543 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) |
1da177e4 | 544 | { |
92be2e33 NP |
545 | if (unlikely(page_mapcount(page) | |
546 | (page->mapping != NULL) | | |
547 | (page_count(page) != 0) | | |
334795ec HD |
548 | (page->flags & ( |
549 | 1 << PG_lru | | |
1da177e4 LT |
550 | 1 << PG_private | |
551 | 1 << PG_locked | | |
1da177e4 LT |
552 | 1 << PG_active | |
553 | 1 << PG_dirty | | |
554 | 1 << PG_reclaim | | |
334795ec | 555 | 1 << PG_slab | |
1da177e4 | 556 | 1 << PG_swapcache | |
b5810039 | 557 | 1 << PG_writeback | |
676165a8 NP |
558 | 1 << PG_reserved | |
559 | 1 << PG_buddy )))) | |
224abf92 | 560 | bad_page(page); |
1da177e4 | 561 | |
689bcebf HD |
562 | /* |
563 | * For now, we report if PG_reserved was found set, but do not | |
564 | * clear it, and do not allocate the page: as a safety net. | |
565 | */ | |
566 | if (PageReserved(page)) | |
567 | return 1; | |
568 | ||
1da177e4 LT |
569 | page->flags &= ~(1 << PG_uptodate | 1 << PG_error | |
570 | 1 << PG_referenced | 1 << PG_arch_1 | | |
571 | 1 << PG_checked | 1 << PG_mappedtodisk); | |
4c21e2f2 | 572 | set_page_private(page, 0); |
7835e98b | 573 | set_page_refcounted(page); |
1da177e4 | 574 | kernel_map_pages(page, 1 << order, 1); |
17cf4406 NP |
575 | |
576 | if (gfp_flags & __GFP_ZERO) | |
577 | prep_zero_page(page, order, gfp_flags); | |
578 | ||
579 | if (order && (gfp_flags & __GFP_COMP)) | |
580 | prep_compound_page(page, order); | |
581 | ||
689bcebf | 582 | return 0; |
1da177e4 LT |
583 | } |
584 | ||
585 | /* | |
586 | * Do the hard work of removing an element from the buddy allocator. | |
587 | * Call me with the zone->lock already held. | |
588 | */ | |
589 | static struct page *__rmqueue(struct zone *zone, unsigned int order) | |
590 | { | |
591 | struct free_area * area; | |
592 | unsigned int current_order; | |
593 | struct page *page; | |
594 | ||
595 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
596 | area = zone->free_area + current_order; | |
597 | if (list_empty(&area->free_list)) | |
598 | continue; | |
599 | ||
600 | page = list_entry(area->free_list.next, struct page, lru); | |
601 | list_del(&page->lru); | |
602 | rmv_page_order(page); | |
603 | area->nr_free--; | |
604 | zone->free_pages -= 1UL << order; | |
085cc7d5 NP |
605 | expand(zone, page, order, current_order, area); |
606 | return page; | |
1da177e4 LT |
607 | } |
608 | ||
609 | return NULL; | |
610 | } | |
611 | ||
612 | /* | |
613 | * Obtain a specified number of elements from the buddy allocator, all under | |
614 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
615 | * Returns the number of new pages which were placed at *list. | |
616 | */ | |
617 | static int rmqueue_bulk(struct zone *zone, unsigned int order, | |
618 | unsigned long count, struct list_head *list) | |
619 | { | |
1da177e4 | 620 | int i; |
1da177e4 | 621 | |
c54ad30c | 622 | spin_lock(&zone->lock); |
1da177e4 | 623 | for (i = 0; i < count; ++i) { |
085cc7d5 NP |
624 | struct page *page = __rmqueue(zone, order); |
625 | if (unlikely(page == NULL)) | |
1da177e4 | 626 | break; |
1da177e4 LT |
627 | list_add_tail(&page->lru, list); |
628 | } | |
c54ad30c | 629 | spin_unlock(&zone->lock); |
085cc7d5 | 630 | return i; |
1da177e4 LT |
631 | } |
632 | ||
4ae7c039 | 633 | #ifdef CONFIG_NUMA |
8fce4d8e CL |
634 | /* |
635 | * Called from the slab reaper to drain pagesets on a particular node that | |
39bbcb8f | 636 | * belongs to the currently executing processor. |
879336c3 CL |
637 | * Note that this function must be called with the thread pinned to |
638 | * a single processor. | |
8fce4d8e CL |
639 | */ |
640 | void drain_node_pages(int nodeid) | |
4ae7c039 | 641 | { |
2f6726e5 CL |
642 | int i; |
643 | enum zone_type z; | |
4ae7c039 CL |
644 | unsigned long flags; |
645 | ||
8fce4d8e CL |
646 | for (z = 0; z < MAX_NR_ZONES; z++) { |
647 | struct zone *zone = NODE_DATA(nodeid)->node_zones + z; | |
4ae7c039 CL |
648 | struct per_cpu_pageset *pset; |
649 | ||
39bbcb8f CL |
650 | if (!populated_zone(zone)) |
651 | continue; | |
652 | ||
23316bc8 | 653 | pset = zone_pcp(zone, smp_processor_id()); |
4ae7c039 CL |
654 | for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) { |
655 | struct per_cpu_pages *pcp; | |
656 | ||
657 | pcp = &pset->pcp[i]; | |
879336c3 CL |
658 | if (pcp->count) { |
659 | local_irq_save(flags); | |
660 | free_pages_bulk(zone, pcp->count, &pcp->list, 0); | |
661 | pcp->count = 0; | |
662 | local_irq_restore(flags); | |
663 | } | |
4ae7c039 CL |
664 | } |
665 | } | |
4ae7c039 CL |
666 | } |
667 | #endif | |
668 | ||
1da177e4 LT |
669 | #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU) |
670 | static void __drain_pages(unsigned int cpu) | |
671 | { | |
c54ad30c | 672 | unsigned long flags; |
1da177e4 LT |
673 | struct zone *zone; |
674 | int i; | |
675 | ||
676 | for_each_zone(zone) { | |
677 | struct per_cpu_pageset *pset; | |
678 | ||
e7c8d5c9 | 679 | pset = zone_pcp(zone, cpu); |
1da177e4 LT |
680 | for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) { |
681 | struct per_cpu_pages *pcp; | |
682 | ||
683 | pcp = &pset->pcp[i]; | |
c54ad30c | 684 | local_irq_save(flags); |
48db57f8 NP |
685 | free_pages_bulk(zone, pcp->count, &pcp->list, 0); |
686 | pcp->count = 0; | |
c54ad30c | 687 | local_irq_restore(flags); |
1da177e4 LT |
688 | } |
689 | } | |
690 | } | |
691 | #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */ | |
692 | ||
693 | #ifdef CONFIG_PM | |
694 | ||
695 | void mark_free_pages(struct zone *zone) | |
696 | { | |
f623f0db RW |
697 | unsigned long pfn, max_zone_pfn; |
698 | unsigned long flags; | |
1da177e4 LT |
699 | int order; |
700 | struct list_head *curr; | |
701 | ||
702 | if (!zone->spanned_pages) | |
703 | return; | |
704 | ||
705 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db RW |
706 | |
707 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
708 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
709 | if (pfn_valid(pfn)) { | |
710 | struct page *page = pfn_to_page(pfn); | |
711 | ||
712 | if (!PageNosave(page)) | |
713 | ClearPageNosaveFree(page); | |
714 | } | |
1da177e4 LT |
715 | |
716 | for (order = MAX_ORDER - 1; order >= 0; --order) | |
717 | list_for_each(curr, &zone->free_area[order].free_list) { | |
f623f0db | 718 | unsigned long i; |
1da177e4 | 719 | |
f623f0db RW |
720 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); |
721 | for (i = 0; i < (1UL << order); i++) | |
722 | SetPageNosaveFree(pfn_to_page(pfn + i)); | |
723 | } | |
1da177e4 | 724 | |
1da177e4 LT |
725 | spin_unlock_irqrestore(&zone->lock, flags); |
726 | } | |
727 | ||
728 | /* | |
729 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
730 | */ | |
731 | void drain_local_pages(void) | |
732 | { | |
733 | unsigned long flags; | |
734 | ||
735 | local_irq_save(flags); | |
736 | __drain_pages(smp_processor_id()); | |
737 | local_irq_restore(flags); | |
738 | } | |
739 | #endif /* CONFIG_PM */ | |
740 | ||
1da177e4 LT |
741 | /* |
742 | * Free a 0-order page | |
743 | */ | |
1da177e4 LT |
744 | static void fastcall free_hot_cold_page(struct page *page, int cold) |
745 | { | |
746 | struct zone *zone = page_zone(page); | |
747 | struct per_cpu_pages *pcp; | |
748 | unsigned long flags; | |
749 | ||
750 | arch_free_page(page, 0); | |
751 | ||
1da177e4 LT |
752 | if (PageAnon(page)) |
753 | page->mapping = NULL; | |
224abf92 | 754 | if (free_pages_check(page)) |
689bcebf HD |
755 | return; |
756 | ||
689bcebf HD |
757 | kernel_map_pages(page, 1, 0); |
758 | ||
e7c8d5c9 | 759 | pcp = &zone_pcp(zone, get_cpu())->pcp[cold]; |
1da177e4 | 760 | local_irq_save(flags); |
f8891e5e | 761 | __count_vm_event(PGFREE); |
1da177e4 LT |
762 | list_add(&page->lru, &pcp->list); |
763 | pcp->count++; | |
48db57f8 NP |
764 | if (pcp->count >= pcp->high) { |
765 | free_pages_bulk(zone, pcp->batch, &pcp->list, 0); | |
766 | pcp->count -= pcp->batch; | |
767 | } | |
1da177e4 LT |
768 | local_irq_restore(flags); |
769 | put_cpu(); | |
770 | } | |
771 | ||
772 | void fastcall free_hot_page(struct page *page) | |
773 | { | |
774 | free_hot_cold_page(page, 0); | |
775 | } | |
776 | ||
777 | void fastcall free_cold_page(struct page *page) | |
778 | { | |
779 | free_hot_cold_page(page, 1); | |
780 | } | |
781 | ||
8dfcc9ba NP |
782 | /* |
783 | * split_page takes a non-compound higher-order page, and splits it into | |
784 | * n (1<<order) sub-pages: page[0..n] | |
785 | * Each sub-page must be freed individually. | |
786 | * | |
787 | * Note: this is probably too low level an operation for use in drivers. | |
788 | * Please consult with lkml before using this in your driver. | |
789 | */ | |
790 | void split_page(struct page *page, unsigned int order) | |
791 | { | |
792 | int i; | |
793 | ||
725d704e NP |
794 | VM_BUG_ON(PageCompound(page)); |
795 | VM_BUG_ON(!page_count(page)); | |
7835e98b NP |
796 | for (i = 1; i < (1 << order); i++) |
797 | set_page_refcounted(page + i); | |
8dfcc9ba | 798 | } |
8dfcc9ba | 799 | |
1da177e4 LT |
800 | /* |
801 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
802 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
803 | * or two. | |
804 | */ | |
a74609fa NP |
805 | static struct page *buffered_rmqueue(struct zonelist *zonelist, |
806 | struct zone *zone, int order, gfp_t gfp_flags) | |
1da177e4 LT |
807 | { |
808 | unsigned long flags; | |
689bcebf | 809 | struct page *page; |
1da177e4 | 810 | int cold = !!(gfp_flags & __GFP_COLD); |
a74609fa | 811 | int cpu; |
1da177e4 | 812 | |
689bcebf | 813 | again: |
a74609fa | 814 | cpu = get_cpu(); |
48db57f8 | 815 | if (likely(order == 0)) { |
1da177e4 LT |
816 | struct per_cpu_pages *pcp; |
817 | ||
a74609fa | 818 | pcp = &zone_pcp(zone, cpu)->pcp[cold]; |
1da177e4 | 819 | local_irq_save(flags); |
a74609fa | 820 | if (!pcp->count) { |
1da177e4 LT |
821 | pcp->count += rmqueue_bulk(zone, 0, |
822 | pcp->batch, &pcp->list); | |
a74609fa NP |
823 | if (unlikely(!pcp->count)) |
824 | goto failed; | |
1da177e4 | 825 | } |
a74609fa NP |
826 | page = list_entry(pcp->list.next, struct page, lru); |
827 | list_del(&page->lru); | |
828 | pcp->count--; | |
7fb1d9fc | 829 | } else { |
1da177e4 LT |
830 | spin_lock_irqsave(&zone->lock, flags); |
831 | page = __rmqueue(zone, order); | |
a74609fa NP |
832 | spin_unlock(&zone->lock); |
833 | if (!page) | |
834 | goto failed; | |
1da177e4 LT |
835 | } |
836 | ||
f8891e5e | 837 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
ca889e6c | 838 | zone_statistics(zonelist, zone); |
a74609fa NP |
839 | local_irq_restore(flags); |
840 | put_cpu(); | |
1da177e4 | 841 | |
725d704e | 842 | VM_BUG_ON(bad_range(zone, page)); |
17cf4406 | 843 | if (prep_new_page(page, order, gfp_flags)) |
a74609fa | 844 | goto again; |
1da177e4 | 845 | return page; |
a74609fa NP |
846 | |
847 | failed: | |
848 | local_irq_restore(flags); | |
849 | put_cpu(); | |
850 | return NULL; | |
1da177e4 LT |
851 | } |
852 | ||
7fb1d9fc | 853 | #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */ |
3148890b NP |
854 | #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */ |
855 | #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */ | |
856 | #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */ | |
857 | #define ALLOC_HARDER 0x10 /* try to alloc harder */ | |
858 | #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */ | |
859 | #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ | |
7fb1d9fc | 860 | |
1da177e4 LT |
861 | /* |
862 | * Return 1 if free pages are above 'mark'. This takes into account the order | |
863 | * of the allocation. | |
864 | */ | |
865 | int zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
7fb1d9fc | 866 | int classzone_idx, int alloc_flags) |
1da177e4 LT |
867 | { |
868 | /* free_pages my go negative - that's OK */ | |
869 | long min = mark, free_pages = z->free_pages - (1 << order) + 1; | |
870 | int o; | |
871 | ||
7fb1d9fc | 872 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 873 | min -= min / 2; |
7fb1d9fc | 874 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 LT |
875 | min -= min / 4; |
876 | ||
877 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
878 | return 0; | |
879 | for (o = 0; o < order; o++) { | |
880 | /* At the next order, this order's pages become unavailable */ | |
881 | free_pages -= z->free_area[o].nr_free << o; | |
882 | ||
883 | /* Require fewer higher order pages to be free */ | |
884 | min >>= 1; | |
885 | ||
886 | if (free_pages <= min) | |
887 | return 0; | |
888 | } | |
889 | return 1; | |
890 | } | |
891 | ||
7fb1d9fc RS |
892 | /* |
893 | * get_page_from_freeliest goes through the zonelist trying to allocate | |
894 | * a page. | |
895 | */ | |
896 | static struct page * | |
897 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, | |
898 | struct zonelist *zonelist, int alloc_flags) | |
753ee728 | 899 | { |
7fb1d9fc RS |
900 | struct zone **z = zonelist->zones; |
901 | struct page *page = NULL; | |
902 | int classzone_idx = zone_idx(*z); | |
1192d526 | 903 | struct zone *zone; |
7fb1d9fc RS |
904 | |
905 | /* | |
906 | * Go through the zonelist once, looking for a zone with enough free. | |
907 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. | |
908 | */ | |
909 | do { | |
1192d526 | 910 | zone = *z; |
9b819d20 | 911 | if (unlikely((gfp_mask & __GFP_THISNODE) && |
1192d526 | 912 | zone->zone_pgdat != zonelist->zones[0]->zone_pgdat)) |
9b819d20 | 913 | break; |
7fb1d9fc | 914 | if ((alloc_flags & ALLOC_CPUSET) && |
1192d526 | 915 | !cpuset_zone_allowed(zone, gfp_mask)) |
7fb1d9fc RS |
916 | continue; |
917 | ||
918 | if (!(alloc_flags & ALLOC_NO_WATERMARKS)) { | |
3148890b NP |
919 | unsigned long mark; |
920 | if (alloc_flags & ALLOC_WMARK_MIN) | |
1192d526 | 921 | mark = zone->pages_min; |
3148890b | 922 | else if (alloc_flags & ALLOC_WMARK_LOW) |
1192d526 | 923 | mark = zone->pages_low; |
3148890b | 924 | else |
1192d526 CL |
925 | mark = zone->pages_high; |
926 | if (!zone_watermark_ok(zone , order, mark, | |
7fb1d9fc | 927 | classzone_idx, alloc_flags)) |
9eeff239 | 928 | if (!zone_reclaim_mode || |
1192d526 | 929 | !zone_reclaim(zone, gfp_mask, order)) |
9eeff239 | 930 | continue; |
7fb1d9fc RS |
931 | } |
932 | ||
1192d526 | 933 | page = buffered_rmqueue(zonelist, zone, order, gfp_mask); |
7fb1d9fc | 934 | if (page) { |
7fb1d9fc RS |
935 | break; |
936 | } | |
937 | } while (*(++z) != NULL); | |
938 | return page; | |
753ee728 MH |
939 | } |
940 | ||
1da177e4 LT |
941 | /* |
942 | * This is the 'heart' of the zoned buddy allocator. | |
943 | */ | |
944 | struct page * fastcall | |
dd0fc66f | 945 | __alloc_pages(gfp_t gfp_mask, unsigned int order, |
1da177e4 LT |
946 | struct zonelist *zonelist) |
947 | { | |
260b2367 | 948 | const gfp_t wait = gfp_mask & __GFP_WAIT; |
7fb1d9fc | 949 | struct zone **z; |
1da177e4 LT |
950 | struct page *page; |
951 | struct reclaim_state reclaim_state; | |
952 | struct task_struct *p = current; | |
1da177e4 | 953 | int do_retry; |
7fb1d9fc | 954 | int alloc_flags; |
1da177e4 LT |
955 | int did_some_progress; |
956 | ||
957 | might_sleep_if(wait); | |
958 | ||
6b1de916 | 959 | restart: |
7fb1d9fc | 960 | z = zonelist->zones; /* the list of zones suitable for gfp_mask */ |
1da177e4 | 961 | |
7fb1d9fc | 962 | if (unlikely(*z == NULL)) { |
1da177e4 LT |
963 | /* Should this ever happen?? */ |
964 | return NULL; | |
965 | } | |
6b1de916 | 966 | |
7fb1d9fc | 967 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, |
3148890b | 968 | zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET); |
7fb1d9fc RS |
969 | if (page) |
970 | goto got_pg; | |
1da177e4 | 971 | |
6b1de916 | 972 | do { |
43b0bc00 | 973 | wakeup_kswapd(*z, order); |
6b1de916 | 974 | } while (*(++z)); |
1da177e4 | 975 | |
9bf2229f | 976 | /* |
7fb1d9fc RS |
977 | * OK, we're below the kswapd watermark and have kicked background |
978 | * reclaim. Now things get more complex, so set up alloc_flags according | |
979 | * to how we want to proceed. | |
980 | * | |
981 | * The caller may dip into page reserves a bit more if the caller | |
982 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
4eac915d PJ |
983 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will |
984 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). | |
9bf2229f | 985 | */ |
3148890b | 986 | alloc_flags = ALLOC_WMARK_MIN; |
7fb1d9fc RS |
987 | if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait) |
988 | alloc_flags |= ALLOC_HARDER; | |
989 | if (gfp_mask & __GFP_HIGH) | |
990 | alloc_flags |= ALLOC_HIGH; | |
bdd804f4 PJ |
991 | if (wait) |
992 | alloc_flags |= ALLOC_CPUSET; | |
1da177e4 LT |
993 | |
994 | /* | |
995 | * Go through the zonelist again. Let __GFP_HIGH and allocations | |
7fb1d9fc | 996 | * coming from realtime tasks go deeper into reserves. |
1da177e4 LT |
997 | * |
998 | * This is the last chance, in general, before the goto nopage. | |
999 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. | |
9bf2229f | 1000 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1da177e4 | 1001 | */ |
7fb1d9fc RS |
1002 | page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags); |
1003 | if (page) | |
1004 | goto got_pg; | |
1da177e4 LT |
1005 | |
1006 | /* This allocation should allow future memory freeing. */ | |
b84a35be NP |
1007 | |
1008 | if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE))) | |
1009 | && !in_interrupt()) { | |
1010 | if (!(gfp_mask & __GFP_NOMEMALLOC)) { | |
885036d3 | 1011 | nofail_alloc: |
b84a35be | 1012 | /* go through the zonelist yet again, ignoring mins */ |
7fb1d9fc | 1013 | page = get_page_from_freelist(gfp_mask, order, |
47f3a867 | 1014 | zonelist, ALLOC_NO_WATERMARKS); |
7fb1d9fc RS |
1015 | if (page) |
1016 | goto got_pg; | |
885036d3 KK |
1017 | if (gfp_mask & __GFP_NOFAIL) { |
1018 | blk_congestion_wait(WRITE, HZ/50); | |
1019 | goto nofail_alloc; | |
1020 | } | |
1da177e4 LT |
1021 | } |
1022 | goto nopage; | |
1023 | } | |
1024 | ||
1025 | /* Atomic allocations - we can't balance anything */ | |
1026 | if (!wait) | |
1027 | goto nopage; | |
1028 | ||
1029 | rebalance: | |
1030 | cond_resched(); | |
1031 | ||
1032 | /* We now go into synchronous reclaim */ | |
3e0d98b9 | 1033 | cpuset_memory_pressure_bump(); |
1da177e4 LT |
1034 | p->flags |= PF_MEMALLOC; |
1035 | reclaim_state.reclaimed_slab = 0; | |
1036 | p->reclaim_state = &reclaim_state; | |
1037 | ||
7fb1d9fc | 1038 | did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask); |
1da177e4 LT |
1039 | |
1040 | p->reclaim_state = NULL; | |
1041 | p->flags &= ~PF_MEMALLOC; | |
1042 | ||
1043 | cond_resched(); | |
1044 | ||
1045 | if (likely(did_some_progress)) { | |
7fb1d9fc RS |
1046 | page = get_page_from_freelist(gfp_mask, order, |
1047 | zonelist, alloc_flags); | |
1048 | if (page) | |
1049 | goto got_pg; | |
1da177e4 LT |
1050 | } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) { |
1051 | /* | |
1052 | * Go through the zonelist yet one more time, keep | |
1053 | * very high watermark here, this is only to catch | |
1054 | * a parallel oom killing, we must fail if we're still | |
1055 | * under heavy pressure. | |
1056 | */ | |
7fb1d9fc | 1057 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, |
3148890b | 1058 | zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET); |
7fb1d9fc RS |
1059 | if (page) |
1060 | goto got_pg; | |
1da177e4 | 1061 | |
9b0f8b04 | 1062 | out_of_memory(zonelist, gfp_mask, order); |
1da177e4 LT |
1063 | goto restart; |
1064 | } | |
1065 | ||
1066 | /* | |
1067 | * Don't let big-order allocations loop unless the caller explicitly | |
1068 | * requests that. Wait for some write requests to complete then retry. | |
1069 | * | |
1070 | * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order | |
1071 | * <= 3, but that may not be true in other implementations. | |
1072 | */ | |
1073 | do_retry = 0; | |
1074 | if (!(gfp_mask & __GFP_NORETRY)) { | |
1075 | if ((order <= 3) || (gfp_mask & __GFP_REPEAT)) | |
1076 | do_retry = 1; | |
1077 | if (gfp_mask & __GFP_NOFAIL) | |
1078 | do_retry = 1; | |
1079 | } | |
1080 | if (do_retry) { | |
1081 | blk_congestion_wait(WRITE, HZ/50); | |
1082 | goto rebalance; | |
1083 | } | |
1084 | ||
1085 | nopage: | |
1086 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) { | |
1087 | printk(KERN_WARNING "%s: page allocation failure." | |
1088 | " order:%d, mode:0x%x\n", | |
1089 | p->comm, order, gfp_mask); | |
1090 | dump_stack(); | |
578c2fd6 | 1091 | show_mem(); |
1da177e4 | 1092 | } |
1da177e4 | 1093 | got_pg: |
1da177e4 LT |
1094 | return page; |
1095 | } | |
1096 | ||
1097 | EXPORT_SYMBOL(__alloc_pages); | |
1098 | ||
1099 | /* | |
1100 | * Common helper functions. | |
1101 | */ | |
dd0fc66f | 1102 | fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 LT |
1103 | { |
1104 | struct page * page; | |
1105 | page = alloc_pages(gfp_mask, order); | |
1106 | if (!page) | |
1107 | return 0; | |
1108 | return (unsigned long) page_address(page); | |
1109 | } | |
1110 | ||
1111 | EXPORT_SYMBOL(__get_free_pages); | |
1112 | ||
dd0fc66f | 1113 | fastcall unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 LT |
1114 | { |
1115 | struct page * page; | |
1116 | ||
1117 | /* | |
1118 | * get_zeroed_page() returns a 32-bit address, which cannot represent | |
1119 | * a highmem page | |
1120 | */ | |
725d704e | 1121 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); |
1da177e4 LT |
1122 | |
1123 | page = alloc_pages(gfp_mask | __GFP_ZERO, 0); | |
1124 | if (page) | |
1125 | return (unsigned long) page_address(page); | |
1126 | return 0; | |
1127 | } | |
1128 | ||
1129 | EXPORT_SYMBOL(get_zeroed_page); | |
1130 | ||
1131 | void __pagevec_free(struct pagevec *pvec) | |
1132 | { | |
1133 | int i = pagevec_count(pvec); | |
1134 | ||
1135 | while (--i >= 0) | |
1136 | free_hot_cold_page(pvec->pages[i], pvec->cold); | |
1137 | } | |
1138 | ||
1139 | fastcall void __free_pages(struct page *page, unsigned int order) | |
1140 | { | |
b5810039 | 1141 | if (put_page_testzero(page)) { |
1da177e4 LT |
1142 | if (order == 0) |
1143 | free_hot_page(page); | |
1144 | else | |
1145 | __free_pages_ok(page, order); | |
1146 | } | |
1147 | } | |
1148 | ||
1149 | EXPORT_SYMBOL(__free_pages); | |
1150 | ||
1151 | fastcall void free_pages(unsigned long addr, unsigned int order) | |
1152 | { | |
1153 | if (addr != 0) { | |
725d704e | 1154 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
1155 | __free_pages(virt_to_page((void *)addr), order); |
1156 | } | |
1157 | } | |
1158 | ||
1159 | EXPORT_SYMBOL(free_pages); | |
1160 | ||
1161 | /* | |
1162 | * Total amount of free (allocatable) RAM: | |
1163 | */ | |
1164 | unsigned int nr_free_pages(void) | |
1165 | { | |
1166 | unsigned int sum = 0; | |
1167 | struct zone *zone; | |
1168 | ||
1169 | for_each_zone(zone) | |
1170 | sum += zone->free_pages; | |
1171 | ||
1172 | return sum; | |
1173 | } | |
1174 | ||
1175 | EXPORT_SYMBOL(nr_free_pages); | |
1176 | ||
1177 | #ifdef CONFIG_NUMA | |
1178 | unsigned int nr_free_pages_pgdat(pg_data_t *pgdat) | |
1179 | { | |
2f6726e5 CL |
1180 | unsigned int sum = 0; |
1181 | enum zone_type i; | |
1da177e4 LT |
1182 | |
1183 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1184 | sum += pgdat->node_zones[i].free_pages; | |
1185 | ||
1186 | return sum; | |
1187 | } | |
1188 | #endif | |
1189 | ||
1190 | static unsigned int nr_free_zone_pages(int offset) | |
1191 | { | |
e310fd43 MB |
1192 | /* Just pick one node, since fallback list is circular */ |
1193 | pg_data_t *pgdat = NODE_DATA(numa_node_id()); | |
1da177e4 LT |
1194 | unsigned int sum = 0; |
1195 | ||
e310fd43 MB |
1196 | struct zonelist *zonelist = pgdat->node_zonelists + offset; |
1197 | struct zone **zonep = zonelist->zones; | |
1198 | struct zone *zone; | |
1da177e4 | 1199 | |
e310fd43 MB |
1200 | for (zone = *zonep++; zone; zone = *zonep++) { |
1201 | unsigned long size = zone->present_pages; | |
1202 | unsigned long high = zone->pages_high; | |
1203 | if (size > high) | |
1204 | sum += size - high; | |
1da177e4 LT |
1205 | } |
1206 | ||
1207 | return sum; | |
1208 | } | |
1209 | ||
1210 | /* | |
1211 | * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL | |
1212 | */ | |
1213 | unsigned int nr_free_buffer_pages(void) | |
1214 | { | |
af4ca457 | 1215 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 LT |
1216 | } |
1217 | ||
1218 | /* | |
1219 | * Amount of free RAM allocatable within all zones | |
1220 | */ | |
1221 | unsigned int nr_free_pagecache_pages(void) | |
1222 | { | |
af4ca457 | 1223 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER)); |
1da177e4 | 1224 | } |
1da177e4 LT |
1225 | #ifdef CONFIG_NUMA |
1226 | static void show_node(struct zone *zone) | |
1227 | { | |
89fa3024 | 1228 | printk("Node %ld ", zone_to_nid(zone)); |
1da177e4 LT |
1229 | } |
1230 | #else | |
1231 | #define show_node(zone) do { } while (0) | |
1232 | #endif | |
1233 | ||
1da177e4 LT |
1234 | void si_meminfo(struct sysinfo *val) |
1235 | { | |
1236 | val->totalram = totalram_pages; | |
1237 | val->sharedram = 0; | |
1238 | val->freeram = nr_free_pages(); | |
1239 | val->bufferram = nr_blockdev_pages(); | |
1da177e4 LT |
1240 | val->totalhigh = totalhigh_pages; |
1241 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
1242 | val->mem_unit = PAGE_SIZE; |
1243 | } | |
1244 | ||
1245 | EXPORT_SYMBOL(si_meminfo); | |
1246 | ||
1247 | #ifdef CONFIG_NUMA | |
1248 | void si_meminfo_node(struct sysinfo *val, int nid) | |
1249 | { | |
1250 | pg_data_t *pgdat = NODE_DATA(nid); | |
1251 | ||
1252 | val->totalram = pgdat->node_present_pages; | |
1253 | val->freeram = nr_free_pages_pgdat(pgdat); | |
98d2b0eb | 1254 | #ifdef CONFIG_HIGHMEM |
1da177e4 LT |
1255 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
1256 | val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages; | |
98d2b0eb CL |
1257 | #else |
1258 | val->totalhigh = 0; | |
1259 | val->freehigh = 0; | |
1260 | #endif | |
1da177e4 LT |
1261 | val->mem_unit = PAGE_SIZE; |
1262 | } | |
1263 | #endif | |
1264 | ||
1265 | #define K(x) ((x) << (PAGE_SHIFT-10)) | |
1266 | ||
1267 | /* | |
1268 | * Show free area list (used inside shift_scroll-lock stuff) | |
1269 | * We also calculate the percentage fragmentation. We do this by counting the | |
1270 | * memory on each free list with the exception of the first item on the list. | |
1271 | */ | |
1272 | void show_free_areas(void) | |
1273 | { | |
1da177e4 LT |
1274 | int cpu, temperature; |
1275 | unsigned long active; | |
1276 | unsigned long inactive; | |
1277 | unsigned long free; | |
1278 | struct zone *zone; | |
1279 | ||
1280 | for_each_zone(zone) { | |
1281 | show_node(zone); | |
1282 | printk("%s per-cpu:", zone->name); | |
1283 | ||
f3fe6512 | 1284 | if (!populated_zone(zone)) { |
1da177e4 LT |
1285 | printk(" empty\n"); |
1286 | continue; | |
1287 | } else | |
1288 | printk("\n"); | |
1289 | ||
6b482c67 | 1290 | for_each_online_cpu(cpu) { |
1da177e4 LT |
1291 | struct per_cpu_pageset *pageset; |
1292 | ||
e7c8d5c9 | 1293 | pageset = zone_pcp(zone, cpu); |
1da177e4 LT |
1294 | |
1295 | for (temperature = 0; temperature < 2; temperature++) | |
2d92c5c9 | 1296 | printk("cpu %d %s: high %d, batch %d used:%d\n", |
1da177e4 LT |
1297 | cpu, |
1298 | temperature ? "cold" : "hot", | |
1da177e4 | 1299 | pageset->pcp[temperature].high, |
4ae7c039 CL |
1300 | pageset->pcp[temperature].batch, |
1301 | pageset->pcp[temperature].count); | |
1da177e4 LT |
1302 | } |
1303 | } | |
1304 | ||
1da177e4 LT |
1305 | get_zone_counts(&active, &inactive, &free); |
1306 | ||
1da177e4 LT |
1307 | printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu " |
1308 | "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n", | |
1309 | active, | |
1310 | inactive, | |
b1e7a8fd | 1311 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 1312 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 1313 | global_page_state(NR_UNSTABLE_NFS), |
1da177e4 | 1314 | nr_free_pages(), |
972d1a7b CL |
1315 | global_page_state(NR_SLAB_RECLAIMABLE) + |
1316 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 1317 | global_page_state(NR_FILE_MAPPED), |
df849a15 | 1318 | global_page_state(NR_PAGETABLE)); |
1da177e4 LT |
1319 | |
1320 | for_each_zone(zone) { | |
1321 | int i; | |
1322 | ||
1323 | show_node(zone); | |
1324 | printk("%s" | |
1325 | " free:%lukB" | |
1326 | " min:%lukB" | |
1327 | " low:%lukB" | |
1328 | " high:%lukB" | |
1329 | " active:%lukB" | |
1330 | " inactive:%lukB" | |
1331 | " present:%lukB" | |
1332 | " pages_scanned:%lu" | |
1333 | " all_unreclaimable? %s" | |
1334 | "\n", | |
1335 | zone->name, | |
1336 | K(zone->free_pages), | |
1337 | K(zone->pages_min), | |
1338 | K(zone->pages_low), | |
1339 | K(zone->pages_high), | |
1340 | K(zone->nr_active), | |
1341 | K(zone->nr_inactive), | |
1342 | K(zone->present_pages), | |
1343 | zone->pages_scanned, | |
1344 | (zone->all_unreclaimable ? "yes" : "no") | |
1345 | ); | |
1346 | printk("lowmem_reserve[]:"); | |
1347 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1348 | printk(" %lu", zone->lowmem_reserve[i]); | |
1349 | printk("\n"); | |
1350 | } | |
1351 | ||
1352 | for_each_zone(zone) { | |
8f9de51a | 1353 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
1da177e4 LT |
1354 | |
1355 | show_node(zone); | |
1356 | printk("%s: ", zone->name); | |
f3fe6512 | 1357 | if (!populated_zone(zone)) { |
1da177e4 LT |
1358 | printk("empty\n"); |
1359 | continue; | |
1360 | } | |
1361 | ||
1362 | spin_lock_irqsave(&zone->lock, flags); | |
1363 | for (order = 0; order < MAX_ORDER; order++) { | |
8f9de51a KK |
1364 | nr[order] = zone->free_area[order].nr_free; |
1365 | total += nr[order] << order; | |
1da177e4 LT |
1366 | } |
1367 | spin_unlock_irqrestore(&zone->lock, flags); | |
8f9de51a KK |
1368 | for (order = 0; order < MAX_ORDER; order++) |
1369 | printk("%lu*%lukB ", nr[order], K(1UL) << order); | |
1da177e4 LT |
1370 | printk("= %lukB\n", K(total)); |
1371 | } | |
1372 | ||
1373 | show_swap_cache_info(); | |
1374 | } | |
1375 | ||
1376 | /* | |
1377 | * Builds allocation fallback zone lists. | |
1a93205b CL |
1378 | * |
1379 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 1380 | */ |
86356ab1 | 1381 | static int __meminit build_zonelists_node(pg_data_t *pgdat, |
2f6726e5 | 1382 | struct zonelist *zonelist, int nr_zones, enum zone_type zone_type) |
1da177e4 | 1383 | { |
1a93205b CL |
1384 | struct zone *zone; |
1385 | ||
98d2b0eb | 1386 | BUG_ON(zone_type >= MAX_NR_ZONES); |
2f6726e5 | 1387 | zone_type++; |
02a68a5e CL |
1388 | |
1389 | do { | |
2f6726e5 | 1390 | zone_type--; |
070f8032 | 1391 | zone = pgdat->node_zones + zone_type; |
1a93205b | 1392 | if (populated_zone(zone)) { |
070f8032 CL |
1393 | zonelist->zones[nr_zones++] = zone; |
1394 | check_highest_zone(zone_type); | |
1da177e4 | 1395 | } |
02a68a5e | 1396 | |
2f6726e5 | 1397 | } while (zone_type); |
070f8032 | 1398 | return nr_zones; |
1da177e4 LT |
1399 | } |
1400 | ||
1401 | #ifdef CONFIG_NUMA | |
1402 | #define MAX_NODE_LOAD (num_online_nodes()) | |
86356ab1 | 1403 | static int __meminitdata node_load[MAX_NUMNODES]; |
1da177e4 | 1404 | /** |
4dc3b16b | 1405 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
1406 | * @node: node whose fallback list we're appending |
1407 | * @used_node_mask: nodemask_t of already used nodes | |
1408 | * | |
1409 | * We use a number of factors to determine which is the next node that should | |
1410 | * appear on a given node's fallback list. The node should not have appeared | |
1411 | * already in @node's fallback list, and it should be the next closest node | |
1412 | * according to the distance array (which contains arbitrary distance values | |
1413 | * from each node to each node in the system), and should also prefer nodes | |
1414 | * with no CPUs, since presumably they'll have very little allocation pressure | |
1415 | * on them otherwise. | |
1416 | * It returns -1 if no node is found. | |
1417 | */ | |
86356ab1 | 1418 | static int __meminit find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 1419 | { |
4cf808eb | 1420 | int n, val; |
1da177e4 LT |
1421 | int min_val = INT_MAX; |
1422 | int best_node = -1; | |
1423 | ||
4cf808eb LT |
1424 | /* Use the local node if we haven't already */ |
1425 | if (!node_isset(node, *used_node_mask)) { | |
1426 | node_set(node, *used_node_mask); | |
1427 | return node; | |
1428 | } | |
1da177e4 | 1429 | |
4cf808eb LT |
1430 | for_each_online_node(n) { |
1431 | cpumask_t tmp; | |
1da177e4 LT |
1432 | |
1433 | /* Don't want a node to appear more than once */ | |
1434 | if (node_isset(n, *used_node_mask)) | |
1435 | continue; | |
1436 | ||
1da177e4 LT |
1437 | /* Use the distance array to find the distance */ |
1438 | val = node_distance(node, n); | |
1439 | ||
4cf808eb LT |
1440 | /* Penalize nodes under us ("prefer the next node") */ |
1441 | val += (n < node); | |
1442 | ||
1da177e4 LT |
1443 | /* Give preference to headless and unused nodes */ |
1444 | tmp = node_to_cpumask(n); | |
1445 | if (!cpus_empty(tmp)) | |
1446 | val += PENALTY_FOR_NODE_WITH_CPUS; | |
1447 | ||
1448 | /* Slight preference for less loaded node */ | |
1449 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
1450 | val += node_load[n]; | |
1451 | ||
1452 | if (val < min_val) { | |
1453 | min_val = val; | |
1454 | best_node = n; | |
1455 | } | |
1456 | } | |
1457 | ||
1458 | if (best_node >= 0) | |
1459 | node_set(best_node, *used_node_mask); | |
1460 | ||
1461 | return best_node; | |
1462 | } | |
1463 | ||
86356ab1 | 1464 | static void __meminit build_zonelists(pg_data_t *pgdat) |
1da177e4 | 1465 | { |
19655d34 CL |
1466 | int j, node, local_node; |
1467 | enum zone_type i; | |
1da177e4 LT |
1468 | int prev_node, load; |
1469 | struct zonelist *zonelist; | |
1470 | nodemask_t used_mask; | |
1471 | ||
1472 | /* initialize zonelists */ | |
19655d34 | 1473 | for (i = 0; i < MAX_NR_ZONES; i++) { |
1da177e4 LT |
1474 | zonelist = pgdat->node_zonelists + i; |
1475 | zonelist->zones[0] = NULL; | |
1476 | } | |
1477 | ||
1478 | /* NUMA-aware ordering of nodes */ | |
1479 | local_node = pgdat->node_id; | |
1480 | load = num_online_nodes(); | |
1481 | prev_node = local_node; | |
1482 | nodes_clear(used_mask); | |
1483 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { | |
9eeff239 CL |
1484 | int distance = node_distance(local_node, node); |
1485 | ||
1486 | /* | |
1487 | * If another node is sufficiently far away then it is better | |
1488 | * to reclaim pages in a zone before going off node. | |
1489 | */ | |
1490 | if (distance > RECLAIM_DISTANCE) | |
1491 | zone_reclaim_mode = 1; | |
1492 | ||
1da177e4 LT |
1493 | /* |
1494 | * We don't want to pressure a particular node. | |
1495 | * So adding penalty to the first node in same | |
1496 | * distance group to make it round-robin. | |
1497 | */ | |
9eeff239 CL |
1498 | |
1499 | if (distance != node_distance(local_node, prev_node)) | |
1da177e4 LT |
1500 | node_load[node] += load; |
1501 | prev_node = node; | |
1502 | load--; | |
19655d34 | 1503 | for (i = 0; i < MAX_NR_ZONES; i++) { |
1da177e4 LT |
1504 | zonelist = pgdat->node_zonelists + i; |
1505 | for (j = 0; zonelist->zones[j] != NULL; j++); | |
1506 | ||
19655d34 | 1507 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); |
1da177e4 LT |
1508 | zonelist->zones[j] = NULL; |
1509 | } | |
1510 | } | |
1511 | } | |
1512 | ||
1513 | #else /* CONFIG_NUMA */ | |
1514 | ||
86356ab1 | 1515 | static void __meminit build_zonelists(pg_data_t *pgdat) |
1da177e4 | 1516 | { |
19655d34 CL |
1517 | int node, local_node; |
1518 | enum zone_type i,j; | |
1da177e4 LT |
1519 | |
1520 | local_node = pgdat->node_id; | |
19655d34 | 1521 | for (i = 0; i < MAX_NR_ZONES; i++) { |
1da177e4 LT |
1522 | struct zonelist *zonelist; |
1523 | ||
1524 | zonelist = pgdat->node_zonelists + i; | |
1525 | ||
19655d34 | 1526 | j = build_zonelists_node(pgdat, zonelist, 0, i); |
1da177e4 LT |
1527 | /* |
1528 | * Now we build the zonelist so that it contains the zones | |
1529 | * of all the other nodes. | |
1530 | * We don't want to pressure a particular node, so when | |
1531 | * building the zones for node N, we make sure that the | |
1532 | * zones coming right after the local ones are those from | |
1533 | * node N+1 (modulo N) | |
1534 | */ | |
1535 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
1536 | if (!node_online(node)) | |
1537 | continue; | |
19655d34 | 1538 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); |
1da177e4 LT |
1539 | } |
1540 | for (node = 0; node < local_node; node++) { | |
1541 | if (!node_online(node)) | |
1542 | continue; | |
19655d34 | 1543 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); |
1da177e4 LT |
1544 | } |
1545 | ||
1546 | zonelist->zones[j] = NULL; | |
1547 | } | |
1548 | } | |
1549 | ||
1550 | #endif /* CONFIG_NUMA */ | |
1551 | ||
6811378e YG |
1552 | /* return values int ....just for stop_machine_run() */ |
1553 | static int __meminit __build_all_zonelists(void *dummy) | |
1da177e4 | 1554 | { |
6811378e YG |
1555 | int nid; |
1556 | for_each_online_node(nid) | |
1557 | build_zonelists(NODE_DATA(nid)); | |
1558 | return 0; | |
1559 | } | |
1560 | ||
1561 | void __meminit build_all_zonelists(void) | |
1562 | { | |
1563 | if (system_state == SYSTEM_BOOTING) { | |
1564 | __build_all_zonelists(0); | |
1565 | cpuset_init_current_mems_allowed(); | |
1566 | } else { | |
1567 | /* we have to stop all cpus to guaranntee there is no user | |
1568 | of zonelist */ | |
1569 | stop_machine_run(__build_all_zonelists, NULL, NR_CPUS); | |
1570 | /* cpuset refresh routine should be here */ | |
1571 | } | |
bd1e22b8 AM |
1572 | vm_total_pages = nr_free_pagecache_pages(); |
1573 | printk("Built %i zonelists. Total pages: %ld\n", | |
1574 | num_online_nodes(), vm_total_pages); | |
1da177e4 LT |
1575 | } |
1576 | ||
1577 | /* | |
1578 | * Helper functions to size the waitqueue hash table. | |
1579 | * Essentially these want to choose hash table sizes sufficiently | |
1580 | * large so that collisions trying to wait on pages are rare. | |
1581 | * But in fact, the number of active page waitqueues on typical | |
1582 | * systems is ridiculously low, less than 200. So this is even | |
1583 | * conservative, even though it seems large. | |
1584 | * | |
1585 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
1586 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
1587 | */ | |
1588 | #define PAGES_PER_WAITQUEUE 256 | |
1589 | ||
cca448fe | 1590 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 1591 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
1592 | { |
1593 | unsigned long size = 1; | |
1594 | ||
1595 | pages /= PAGES_PER_WAITQUEUE; | |
1596 | ||
1597 | while (size < pages) | |
1598 | size <<= 1; | |
1599 | ||
1600 | /* | |
1601 | * Once we have dozens or even hundreds of threads sleeping | |
1602 | * on IO we've got bigger problems than wait queue collision. | |
1603 | * Limit the size of the wait table to a reasonable size. | |
1604 | */ | |
1605 | size = min(size, 4096UL); | |
1606 | ||
1607 | return max(size, 4UL); | |
1608 | } | |
cca448fe YG |
1609 | #else |
1610 | /* | |
1611 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
1612 | * a suitable size for its wait_table. So we use the maximum size now. | |
1613 | * | |
1614 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
1615 | * | |
1616 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
1617 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
1618 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
1619 | * | |
1620 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
1621 | * or more by the traditional way. (See above). It equals: | |
1622 | * | |
1623 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
1624 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
1625 | * powerpc (64K page size) : = (32G +16M)byte. | |
1626 | */ | |
1627 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
1628 | { | |
1629 | return 4096UL; | |
1630 | } | |
1631 | #endif | |
1da177e4 LT |
1632 | |
1633 | /* | |
1634 | * This is an integer logarithm so that shifts can be used later | |
1635 | * to extract the more random high bits from the multiplicative | |
1636 | * hash function before the remainder is taken. | |
1637 | */ | |
1638 | static inline unsigned long wait_table_bits(unsigned long size) | |
1639 | { | |
1640 | return ffz(~size); | |
1641 | } | |
1642 | ||
1643 | #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1)) | |
1644 | ||
1645 | static void __init calculate_zone_totalpages(struct pglist_data *pgdat, | |
1646 | unsigned long *zones_size, unsigned long *zholes_size) | |
1647 | { | |
1648 | unsigned long realtotalpages, totalpages = 0; | |
2f6726e5 | 1649 | enum zone_type i; |
1da177e4 LT |
1650 | |
1651 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1652 | totalpages += zones_size[i]; | |
1653 | pgdat->node_spanned_pages = totalpages; | |
1654 | ||
1655 | realtotalpages = totalpages; | |
1656 | if (zholes_size) | |
1657 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1658 | realtotalpages -= zholes_size[i]; | |
1659 | pgdat->node_present_pages = realtotalpages; | |
1660 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages); | |
1661 | } | |
1662 | ||
1663 | ||
1664 | /* | |
1665 | * Initially all pages are reserved - free ones are freed | |
1666 | * up by free_all_bootmem() once the early boot process is | |
1667 | * done. Non-atomic initialization, single-pass. | |
1668 | */ | |
c09b4240 | 1669 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
1da177e4 LT |
1670 | unsigned long start_pfn) |
1671 | { | |
1da177e4 | 1672 | struct page *page; |
29751f69 AW |
1673 | unsigned long end_pfn = start_pfn + size; |
1674 | unsigned long pfn; | |
1da177e4 | 1675 | |
cbe8dd4a | 1676 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
d41dee36 AW |
1677 | if (!early_pfn_valid(pfn)) |
1678 | continue; | |
1679 | page = pfn_to_page(pfn); | |
1680 | set_page_links(page, zone, nid, pfn); | |
7835e98b | 1681 | init_page_count(page); |
1da177e4 LT |
1682 | reset_page_mapcount(page); |
1683 | SetPageReserved(page); | |
1684 | INIT_LIST_HEAD(&page->lru); | |
1685 | #ifdef WANT_PAGE_VIRTUAL | |
1686 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
1687 | if (!is_highmem_idx(zone)) | |
3212c6be | 1688 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1da177e4 | 1689 | #endif |
1da177e4 LT |
1690 | } |
1691 | } | |
1692 | ||
1693 | void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone, | |
1694 | unsigned long size) | |
1695 | { | |
1696 | int order; | |
1697 | for (order = 0; order < MAX_ORDER ; order++) { | |
1698 | INIT_LIST_HEAD(&zone->free_area[order].free_list); | |
1699 | zone->free_area[order].nr_free = 0; | |
1700 | } | |
1701 | } | |
1702 | ||
d41dee36 | 1703 | #define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr) |
2f1b6248 CL |
1704 | void zonetable_add(struct zone *zone, int nid, enum zone_type zid, |
1705 | unsigned long pfn, unsigned long size) | |
d41dee36 AW |
1706 | { |
1707 | unsigned long snum = pfn_to_section_nr(pfn); | |
1708 | unsigned long end = pfn_to_section_nr(pfn + size); | |
1709 | ||
1710 | if (FLAGS_HAS_NODE) | |
1711 | zone_table[ZONETABLE_INDEX(nid, zid)] = zone; | |
1712 | else | |
1713 | for (; snum <= end; snum++) | |
1714 | zone_table[ZONETABLE_INDEX(snum, zid)] = zone; | |
1715 | } | |
1716 | ||
1da177e4 LT |
1717 | #ifndef __HAVE_ARCH_MEMMAP_INIT |
1718 | #define memmap_init(size, nid, zone, start_pfn) \ | |
1719 | memmap_init_zone((size), (nid), (zone), (start_pfn)) | |
1720 | #endif | |
1721 | ||
6292d9aa | 1722 | static int __cpuinit zone_batchsize(struct zone *zone) |
e7c8d5c9 CL |
1723 | { |
1724 | int batch; | |
1725 | ||
1726 | /* | |
1727 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 1728 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
1729 | * |
1730 | * OK, so we don't know how big the cache is. So guess. | |
1731 | */ | |
1732 | batch = zone->present_pages / 1024; | |
ba56e91c SR |
1733 | if (batch * PAGE_SIZE > 512 * 1024) |
1734 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
1735 | batch /= 4; /* We effectively *= 4 below */ |
1736 | if (batch < 1) | |
1737 | batch = 1; | |
1738 | ||
1739 | /* | |
0ceaacc9 NP |
1740 | * Clamp the batch to a 2^n - 1 value. Having a power |
1741 | * of 2 value was found to be more likely to have | |
1742 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 1743 | * |
0ceaacc9 NP |
1744 | * For example if 2 tasks are alternately allocating |
1745 | * batches of pages, one task can end up with a lot | |
1746 | * of pages of one half of the possible page colors | |
1747 | * and the other with pages of the other colors. | |
e7c8d5c9 | 1748 | */ |
0ceaacc9 | 1749 | batch = (1 << (fls(batch + batch/2)-1)) - 1; |
ba56e91c | 1750 | |
e7c8d5c9 CL |
1751 | return batch; |
1752 | } | |
1753 | ||
2caaad41 CL |
1754 | inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
1755 | { | |
1756 | struct per_cpu_pages *pcp; | |
1757 | ||
1c6fe946 MD |
1758 | memset(p, 0, sizeof(*p)); |
1759 | ||
2caaad41 CL |
1760 | pcp = &p->pcp[0]; /* hot */ |
1761 | pcp->count = 0; | |
2caaad41 CL |
1762 | pcp->high = 6 * batch; |
1763 | pcp->batch = max(1UL, 1 * batch); | |
1764 | INIT_LIST_HEAD(&pcp->list); | |
1765 | ||
1766 | pcp = &p->pcp[1]; /* cold*/ | |
1767 | pcp->count = 0; | |
2caaad41 | 1768 | pcp->high = 2 * batch; |
e46a5e28 | 1769 | pcp->batch = max(1UL, batch/2); |
2caaad41 CL |
1770 | INIT_LIST_HEAD(&pcp->list); |
1771 | } | |
1772 | ||
8ad4b1fb RS |
1773 | /* |
1774 | * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist | |
1775 | * to the value high for the pageset p. | |
1776 | */ | |
1777 | ||
1778 | static void setup_pagelist_highmark(struct per_cpu_pageset *p, | |
1779 | unsigned long high) | |
1780 | { | |
1781 | struct per_cpu_pages *pcp; | |
1782 | ||
1783 | pcp = &p->pcp[0]; /* hot list */ | |
1784 | pcp->high = high; | |
1785 | pcp->batch = max(1UL, high/4); | |
1786 | if ((high/4) > (PAGE_SHIFT * 8)) | |
1787 | pcp->batch = PAGE_SHIFT * 8; | |
1788 | } | |
1789 | ||
1790 | ||
e7c8d5c9 CL |
1791 | #ifdef CONFIG_NUMA |
1792 | /* | |
2caaad41 CL |
1793 | * Boot pageset table. One per cpu which is going to be used for all |
1794 | * zones and all nodes. The parameters will be set in such a way | |
1795 | * that an item put on a list will immediately be handed over to | |
1796 | * the buddy list. This is safe since pageset manipulation is done | |
1797 | * with interrupts disabled. | |
1798 | * | |
1799 | * Some NUMA counter updates may also be caught by the boot pagesets. | |
b7c84c6a CL |
1800 | * |
1801 | * The boot_pagesets must be kept even after bootup is complete for | |
1802 | * unused processors and/or zones. They do play a role for bootstrapping | |
1803 | * hotplugged processors. | |
1804 | * | |
1805 | * zoneinfo_show() and maybe other functions do | |
1806 | * not check if the processor is online before following the pageset pointer. | |
1807 | * Other parts of the kernel may not check if the zone is available. | |
2caaad41 | 1808 | */ |
88a2a4ac | 1809 | static struct per_cpu_pageset boot_pageset[NR_CPUS]; |
2caaad41 CL |
1810 | |
1811 | /* | |
1812 | * Dynamically allocate memory for the | |
e7c8d5c9 CL |
1813 | * per cpu pageset array in struct zone. |
1814 | */ | |
6292d9aa | 1815 | static int __cpuinit process_zones(int cpu) |
e7c8d5c9 CL |
1816 | { |
1817 | struct zone *zone, *dzone; | |
e7c8d5c9 CL |
1818 | |
1819 | for_each_zone(zone) { | |
e7c8d5c9 | 1820 | |
23316bc8 | 1821 | zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset), |
e7c8d5c9 | 1822 | GFP_KERNEL, cpu_to_node(cpu)); |
23316bc8 | 1823 | if (!zone_pcp(zone, cpu)) |
e7c8d5c9 | 1824 | goto bad; |
e7c8d5c9 | 1825 | |
23316bc8 | 1826 | setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone)); |
8ad4b1fb RS |
1827 | |
1828 | if (percpu_pagelist_fraction) | |
1829 | setup_pagelist_highmark(zone_pcp(zone, cpu), | |
1830 | (zone->present_pages / percpu_pagelist_fraction)); | |
e7c8d5c9 CL |
1831 | } |
1832 | ||
1833 | return 0; | |
1834 | bad: | |
1835 | for_each_zone(dzone) { | |
1836 | if (dzone == zone) | |
1837 | break; | |
23316bc8 NP |
1838 | kfree(zone_pcp(dzone, cpu)); |
1839 | zone_pcp(dzone, cpu) = NULL; | |
e7c8d5c9 CL |
1840 | } |
1841 | return -ENOMEM; | |
1842 | } | |
1843 | ||
1844 | static inline void free_zone_pagesets(int cpu) | |
1845 | { | |
e7c8d5c9 CL |
1846 | struct zone *zone; |
1847 | ||
1848 | for_each_zone(zone) { | |
1849 | struct per_cpu_pageset *pset = zone_pcp(zone, cpu); | |
1850 | ||
f3ef9ead DR |
1851 | /* Free per_cpu_pageset if it is slab allocated */ |
1852 | if (pset != &boot_pageset[cpu]) | |
1853 | kfree(pset); | |
e7c8d5c9 | 1854 | zone_pcp(zone, cpu) = NULL; |
e7c8d5c9 | 1855 | } |
e7c8d5c9 CL |
1856 | } |
1857 | ||
9c7b216d | 1858 | static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb, |
e7c8d5c9 CL |
1859 | unsigned long action, |
1860 | void *hcpu) | |
1861 | { | |
1862 | int cpu = (long)hcpu; | |
1863 | int ret = NOTIFY_OK; | |
1864 | ||
1865 | switch (action) { | |
1866 | case CPU_UP_PREPARE: | |
1867 | if (process_zones(cpu)) | |
1868 | ret = NOTIFY_BAD; | |
1869 | break; | |
b0d41693 | 1870 | case CPU_UP_CANCELED: |
e7c8d5c9 CL |
1871 | case CPU_DEAD: |
1872 | free_zone_pagesets(cpu); | |
1873 | break; | |
e7c8d5c9 CL |
1874 | default: |
1875 | break; | |
1876 | } | |
1877 | return ret; | |
1878 | } | |
1879 | ||
74b85f37 | 1880 | static struct notifier_block __cpuinitdata pageset_notifier = |
e7c8d5c9 CL |
1881 | { &pageset_cpuup_callback, NULL, 0 }; |
1882 | ||
78d9955b | 1883 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 CL |
1884 | { |
1885 | int err; | |
1886 | ||
1887 | /* Initialize per_cpu_pageset for cpu 0. | |
1888 | * A cpuup callback will do this for every cpu | |
1889 | * as it comes online | |
1890 | */ | |
1891 | err = process_zones(smp_processor_id()); | |
1892 | BUG_ON(err); | |
1893 | register_cpu_notifier(&pageset_notifier); | |
1894 | } | |
1895 | ||
1896 | #endif | |
1897 | ||
c09b4240 | 1898 | static __meminit |
cca448fe | 1899 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
1900 | { |
1901 | int i; | |
1902 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe | 1903 | size_t alloc_size; |
ed8ece2e DH |
1904 | |
1905 | /* | |
1906 | * The per-page waitqueue mechanism uses hashed waitqueues | |
1907 | * per zone. | |
1908 | */ | |
02b694de YG |
1909 | zone->wait_table_hash_nr_entries = |
1910 | wait_table_hash_nr_entries(zone_size_pages); | |
1911 | zone->wait_table_bits = | |
1912 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
1913 | alloc_size = zone->wait_table_hash_nr_entries |
1914 | * sizeof(wait_queue_head_t); | |
1915 | ||
1916 | if (system_state == SYSTEM_BOOTING) { | |
1917 | zone->wait_table = (wait_queue_head_t *) | |
1918 | alloc_bootmem_node(pgdat, alloc_size); | |
1919 | } else { | |
1920 | /* | |
1921 | * This case means that a zone whose size was 0 gets new memory | |
1922 | * via memory hot-add. | |
1923 | * But it may be the case that a new node was hot-added. In | |
1924 | * this case vmalloc() will not be able to use this new node's | |
1925 | * memory - this wait_table must be initialized to use this new | |
1926 | * node itself as well. | |
1927 | * To use this new node's memory, further consideration will be | |
1928 | * necessary. | |
1929 | */ | |
1930 | zone->wait_table = (wait_queue_head_t *)vmalloc(alloc_size); | |
1931 | } | |
1932 | if (!zone->wait_table) | |
1933 | return -ENOMEM; | |
ed8ece2e | 1934 | |
02b694de | 1935 | for(i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 1936 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
1937 | |
1938 | return 0; | |
ed8ece2e DH |
1939 | } |
1940 | ||
c09b4240 | 1941 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e DH |
1942 | { |
1943 | int cpu; | |
1944 | unsigned long batch = zone_batchsize(zone); | |
1945 | ||
1946 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
1947 | #ifdef CONFIG_NUMA | |
1948 | /* Early boot. Slab allocator not functional yet */ | |
23316bc8 | 1949 | zone_pcp(zone, cpu) = &boot_pageset[cpu]; |
ed8ece2e DH |
1950 | setup_pageset(&boot_pageset[cpu],0); |
1951 | #else | |
1952 | setup_pageset(zone_pcp(zone,cpu), batch); | |
1953 | #endif | |
1954 | } | |
f5335c0f AB |
1955 | if (zone->present_pages) |
1956 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n", | |
1957 | zone->name, zone->present_pages, batch); | |
ed8ece2e DH |
1958 | } |
1959 | ||
718127cc YG |
1960 | __meminit int init_currently_empty_zone(struct zone *zone, |
1961 | unsigned long zone_start_pfn, | |
1962 | unsigned long size) | |
ed8ece2e DH |
1963 | { |
1964 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
1965 | int ret; |
1966 | ret = zone_wait_table_init(zone, size); | |
1967 | if (ret) | |
1968 | return ret; | |
ed8ece2e DH |
1969 | pgdat->nr_zones = zone_idx(zone) + 1; |
1970 | ||
ed8ece2e DH |
1971 | zone->zone_start_pfn = zone_start_pfn; |
1972 | ||
1973 | memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn); | |
1974 | ||
1975 | zone_init_free_lists(pgdat, zone, zone->spanned_pages); | |
718127cc YG |
1976 | |
1977 | return 0; | |
ed8ece2e DH |
1978 | } |
1979 | ||
1da177e4 LT |
1980 | /* |
1981 | * Set up the zone data structures: | |
1982 | * - mark all pages reserved | |
1983 | * - mark all memory queues empty | |
1984 | * - clear the memory bitmaps | |
1985 | */ | |
86356ab1 | 1986 | static void __meminit free_area_init_core(struct pglist_data *pgdat, |
1da177e4 LT |
1987 | unsigned long *zones_size, unsigned long *zholes_size) |
1988 | { | |
2f1b6248 | 1989 | enum zone_type j; |
ed8ece2e | 1990 | int nid = pgdat->node_id; |
1da177e4 | 1991 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
718127cc | 1992 | int ret; |
1da177e4 | 1993 | |
208d54e5 | 1994 | pgdat_resize_init(pgdat); |
1da177e4 LT |
1995 | pgdat->nr_zones = 0; |
1996 | init_waitqueue_head(&pgdat->kswapd_wait); | |
1997 | pgdat->kswapd_max_order = 0; | |
1998 | ||
1999 | for (j = 0; j < MAX_NR_ZONES; j++) { | |
2000 | struct zone *zone = pgdat->node_zones + j; | |
2001 | unsigned long size, realsize; | |
1da177e4 | 2002 | |
1da177e4 LT |
2003 | realsize = size = zones_size[j]; |
2004 | if (zholes_size) | |
2005 | realsize -= zholes_size[j]; | |
2006 | ||
98d2b0eb | 2007 | if (!is_highmem_idx(j)) |
1da177e4 LT |
2008 | nr_kernel_pages += realsize; |
2009 | nr_all_pages += realsize; | |
2010 | ||
2011 | zone->spanned_pages = size; | |
2012 | zone->present_pages = realsize; | |
9614634f | 2013 | #ifdef CONFIG_NUMA |
8417bba4 | 2014 | zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio) |
9614634f | 2015 | / 100; |
0ff38490 | 2016 | zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100; |
9614634f | 2017 | #endif |
1da177e4 LT |
2018 | zone->name = zone_names[j]; |
2019 | spin_lock_init(&zone->lock); | |
2020 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 2021 | zone_seqlock_init(zone); |
1da177e4 LT |
2022 | zone->zone_pgdat = pgdat; |
2023 | zone->free_pages = 0; | |
2024 | ||
2025 | zone->temp_priority = zone->prev_priority = DEF_PRIORITY; | |
2026 | ||
ed8ece2e | 2027 | zone_pcp_init(zone); |
1da177e4 LT |
2028 | INIT_LIST_HEAD(&zone->active_list); |
2029 | INIT_LIST_HEAD(&zone->inactive_list); | |
2030 | zone->nr_scan_active = 0; | |
2031 | zone->nr_scan_inactive = 0; | |
2032 | zone->nr_active = 0; | |
2033 | zone->nr_inactive = 0; | |
2244b95a | 2034 | zap_zone_vm_stats(zone); |
53e9a615 | 2035 | atomic_set(&zone->reclaim_in_progress, 0); |
1da177e4 LT |
2036 | if (!size) |
2037 | continue; | |
2038 | ||
d41dee36 | 2039 | zonetable_add(zone, nid, j, zone_start_pfn, size); |
718127cc YG |
2040 | ret = init_currently_empty_zone(zone, zone_start_pfn, size); |
2041 | BUG_ON(ret); | |
1da177e4 | 2042 | zone_start_pfn += size; |
1da177e4 LT |
2043 | } |
2044 | } | |
2045 | ||
2046 | static void __init alloc_node_mem_map(struct pglist_data *pgdat) | |
2047 | { | |
1da177e4 LT |
2048 | /* Skip empty nodes */ |
2049 | if (!pgdat->node_spanned_pages) | |
2050 | return; | |
2051 | ||
d41dee36 | 2052 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
2053 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
2054 | if (!pgdat->node_mem_map) { | |
e984bb43 | 2055 | unsigned long size, start, end; |
d41dee36 AW |
2056 | struct page *map; |
2057 | ||
e984bb43 BP |
2058 | /* |
2059 | * The zone's endpoints aren't required to be MAX_ORDER | |
2060 | * aligned but the node_mem_map endpoints must be in order | |
2061 | * for the buddy allocator to function correctly. | |
2062 | */ | |
2063 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
2064 | end = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
2065 | end = ALIGN(end, MAX_ORDER_NR_PAGES); | |
2066 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
2067 | map = alloc_remap(pgdat->node_id, size); |
2068 | if (!map) | |
2069 | map = alloc_bootmem_node(pgdat, size); | |
e984bb43 | 2070 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
1da177e4 | 2071 | } |
d41dee36 | 2072 | #ifdef CONFIG_FLATMEM |
1da177e4 LT |
2073 | /* |
2074 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
2075 | */ | |
2076 | if (pgdat == NODE_DATA(0)) | |
2077 | mem_map = NODE_DATA(0)->node_mem_map; | |
2078 | #endif | |
d41dee36 | 2079 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
2080 | } |
2081 | ||
86356ab1 | 2082 | void __meminit free_area_init_node(int nid, struct pglist_data *pgdat, |
1da177e4 LT |
2083 | unsigned long *zones_size, unsigned long node_start_pfn, |
2084 | unsigned long *zholes_size) | |
2085 | { | |
2086 | pgdat->node_id = nid; | |
2087 | pgdat->node_start_pfn = node_start_pfn; | |
2088 | calculate_zone_totalpages(pgdat, zones_size, zholes_size); | |
2089 | ||
2090 | alloc_node_mem_map(pgdat); | |
2091 | ||
2092 | free_area_init_core(pgdat, zones_size, zholes_size); | |
2093 | } | |
2094 | ||
93b7504e | 2095 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
2096 | static bootmem_data_t contig_bootmem_data; |
2097 | struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data }; | |
2098 | ||
2099 | EXPORT_SYMBOL(contig_page_data); | |
93b7504e | 2100 | #endif |
1da177e4 LT |
2101 | |
2102 | void __init free_area_init(unsigned long *zones_size) | |
2103 | { | |
93b7504e | 2104 | free_area_init_node(0, NODE_DATA(0), zones_size, |
1da177e4 LT |
2105 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
2106 | } | |
1da177e4 | 2107 | |
1da177e4 LT |
2108 | #ifdef CONFIG_HOTPLUG_CPU |
2109 | static int page_alloc_cpu_notify(struct notifier_block *self, | |
2110 | unsigned long action, void *hcpu) | |
2111 | { | |
2112 | int cpu = (unsigned long)hcpu; | |
1da177e4 LT |
2113 | |
2114 | if (action == CPU_DEAD) { | |
1da177e4 LT |
2115 | local_irq_disable(); |
2116 | __drain_pages(cpu); | |
f8891e5e | 2117 | vm_events_fold_cpu(cpu); |
1da177e4 | 2118 | local_irq_enable(); |
2244b95a | 2119 | refresh_cpu_vm_stats(cpu); |
1da177e4 LT |
2120 | } |
2121 | return NOTIFY_OK; | |
2122 | } | |
2123 | #endif /* CONFIG_HOTPLUG_CPU */ | |
2124 | ||
2125 | void __init page_alloc_init(void) | |
2126 | { | |
2127 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
2128 | } | |
2129 | ||
cb45b0e9 HA |
2130 | /* |
2131 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio | |
2132 | * or min_free_kbytes changes. | |
2133 | */ | |
2134 | static void calculate_totalreserve_pages(void) | |
2135 | { | |
2136 | struct pglist_data *pgdat; | |
2137 | unsigned long reserve_pages = 0; | |
2f6726e5 | 2138 | enum zone_type i, j; |
cb45b0e9 HA |
2139 | |
2140 | for_each_online_pgdat(pgdat) { | |
2141 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2142 | struct zone *zone = pgdat->node_zones + i; | |
2143 | unsigned long max = 0; | |
2144 | ||
2145 | /* Find valid and maximum lowmem_reserve in the zone */ | |
2146 | for (j = i; j < MAX_NR_ZONES; j++) { | |
2147 | if (zone->lowmem_reserve[j] > max) | |
2148 | max = zone->lowmem_reserve[j]; | |
2149 | } | |
2150 | ||
2151 | /* we treat pages_high as reserved pages. */ | |
2152 | max += zone->pages_high; | |
2153 | ||
2154 | if (max > zone->present_pages) | |
2155 | max = zone->present_pages; | |
2156 | reserve_pages += max; | |
2157 | } | |
2158 | } | |
2159 | totalreserve_pages = reserve_pages; | |
2160 | } | |
2161 | ||
1da177e4 LT |
2162 | /* |
2163 | * setup_per_zone_lowmem_reserve - called whenever | |
2164 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
2165 | * has a correct pages reserved value, so an adequate number of | |
2166 | * pages are left in the zone after a successful __alloc_pages(). | |
2167 | */ | |
2168 | static void setup_per_zone_lowmem_reserve(void) | |
2169 | { | |
2170 | struct pglist_data *pgdat; | |
2f6726e5 | 2171 | enum zone_type j, idx; |
1da177e4 | 2172 | |
ec936fc5 | 2173 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
2174 | for (j = 0; j < MAX_NR_ZONES; j++) { |
2175 | struct zone *zone = pgdat->node_zones + j; | |
2176 | unsigned long present_pages = zone->present_pages; | |
2177 | ||
2178 | zone->lowmem_reserve[j] = 0; | |
2179 | ||
2f6726e5 CL |
2180 | idx = j; |
2181 | while (idx) { | |
1da177e4 LT |
2182 | struct zone *lower_zone; |
2183 | ||
2f6726e5 CL |
2184 | idx--; |
2185 | ||
1da177e4 LT |
2186 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
2187 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
2188 | ||
2189 | lower_zone = pgdat->node_zones + idx; | |
2190 | lower_zone->lowmem_reserve[j] = present_pages / | |
2191 | sysctl_lowmem_reserve_ratio[idx]; | |
2192 | present_pages += lower_zone->present_pages; | |
2193 | } | |
2194 | } | |
2195 | } | |
cb45b0e9 HA |
2196 | |
2197 | /* update totalreserve_pages */ | |
2198 | calculate_totalreserve_pages(); | |
1da177e4 LT |
2199 | } |
2200 | ||
2201 | /* | |
2202 | * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures | |
2203 | * that the pages_{min,low,high} values for each zone are set correctly | |
2204 | * with respect to min_free_kbytes. | |
2205 | */ | |
3947be19 | 2206 | void setup_per_zone_pages_min(void) |
1da177e4 LT |
2207 | { |
2208 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
2209 | unsigned long lowmem_pages = 0; | |
2210 | struct zone *zone; | |
2211 | unsigned long flags; | |
2212 | ||
2213 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
2214 | for_each_zone(zone) { | |
2215 | if (!is_highmem(zone)) | |
2216 | lowmem_pages += zone->present_pages; | |
2217 | } | |
2218 | ||
2219 | for_each_zone(zone) { | |
ac924c60 AM |
2220 | u64 tmp; |
2221 | ||
1da177e4 | 2222 | spin_lock_irqsave(&zone->lru_lock, flags); |
ac924c60 AM |
2223 | tmp = (u64)pages_min * zone->present_pages; |
2224 | do_div(tmp, lowmem_pages); | |
1da177e4 LT |
2225 | if (is_highmem(zone)) { |
2226 | /* | |
669ed175 NP |
2227 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
2228 | * need highmem pages, so cap pages_min to a small | |
2229 | * value here. | |
2230 | * | |
2231 | * The (pages_high-pages_low) and (pages_low-pages_min) | |
2232 | * deltas controls asynch page reclaim, and so should | |
2233 | * not be capped for highmem. | |
1da177e4 LT |
2234 | */ |
2235 | int min_pages; | |
2236 | ||
2237 | min_pages = zone->present_pages / 1024; | |
2238 | if (min_pages < SWAP_CLUSTER_MAX) | |
2239 | min_pages = SWAP_CLUSTER_MAX; | |
2240 | if (min_pages > 128) | |
2241 | min_pages = 128; | |
2242 | zone->pages_min = min_pages; | |
2243 | } else { | |
669ed175 NP |
2244 | /* |
2245 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
2246 | * proportionate to the zone's size. |
2247 | */ | |
669ed175 | 2248 | zone->pages_min = tmp; |
1da177e4 LT |
2249 | } |
2250 | ||
ac924c60 AM |
2251 | zone->pages_low = zone->pages_min + (tmp >> 2); |
2252 | zone->pages_high = zone->pages_min + (tmp >> 1); | |
1da177e4 LT |
2253 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
2254 | } | |
cb45b0e9 HA |
2255 | |
2256 | /* update totalreserve_pages */ | |
2257 | calculate_totalreserve_pages(); | |
1da177e4 LT |
2258 | } |
2259 | ||
2260 | /* | |
2261 | * Initialise min_free_kbytes. | |
2262 | * | |
2263 | * For small machines we want it small (128k min). For large machines | |
2264 | * we want it large (64MB max). But it is not linear, because network | |
2265 | * bandwidth does not increase linearly with machine size. We use | |
2266 | * | |
2267 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: | |
2268 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) | |
2269 | * | |
2270 | * which yields | |
2271 | * | |
2272 | * 16MB: 512k | |
2273 | * 32MB: 724k | |
2274 | * 64MB: 1024k | |
2275 | * 128MB: 1448k | |
2276 | * 256MB: 2048k | |
2277 | * 512MB: 2896k | |
2278 | * 1024MB: 4096k | |
2279 | * 2048MB: 5792k | |
2280 | * 4096MB: 8192k | |
2281 | * 8192MB: 11584k | |
2282 | * 16384MB: 16384k | |
2283 | */ | |
2284 | static int __init init_per_zone_pages_min(void) | |
2285 | { | |
2286 | unsigned long lowmem_kbytes; | |
2287 | ||
2288 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
2289 | ||
2290 | min_free_kbytes = int_sqrt(lowmem_kbytes * 16); | |
2291 | if (min_free_kbytes < 128) | |
2292 | min_free_kbytes = 128; | |
2293 | if (min_free_kbytes > 65536) | |
2294 | min_free_kbytes = 65536; | |
2295 | setup_per_zone_pages_min(); | |
2296 | setup_per_zone_lowmem_reserve(); | |
2297 | return 0; | |
2298 | } | |
2299 | module_init(init_per_zone_pages_min) | |
2300 | ||
2301 | /* | |
2302 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so | |
2303 | * that we can call two helper functions whenever min_free_kbytes | |
2304 | * changes. | |
2305 | */ | |
2306 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, | |
2307 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
2308 | { | |
2309 | proc_dointvec(table, write, file, buffer, length, ppos); | |
2310 | setup_per_zone_pages_min(); | |
2311 | return 0; | |
2312 | } | |
2313 | ||
9614634f CL |
2314 | #ifdef CONFIG_NUMA |
2315 | int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write, | |
2316 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
2317 | { | |
2318 | struct zone *zone; | |
2319 | int rc; | |
2320 | ||
2321 | rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
2322 | if (rc) | |
2323 | return rc; | |
2324 | ||
2325 | for_each_zone(zone) | |
8417bba4 | 2326 | zone->min_unmapped_pages = (zone->present_pages * |
9614634f CL |
2327 | sysctl_min_unmapped_ratio) / 100; |
2328 | return 0; | |
2329 | } | |
0ff38490 CL |
2330 | |
2331 | int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write, | |
2332 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
2333 | { | |
2334 | struct zone *zone; | |
2335 | int rc; | |
2336 | ||
2337 | rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
2338 | if (rc) | |
2339 | return rc; | |
2340 | ||
2341 | for_each_zone(zone) | |
2342 | zone->min_slab_pages = (zone->present_pages * | |
2343 | sysctl_min_slab_ratio) / 100; | |
2344 | return 0; | |
2345 | } | |
9614634f CL |
2346 | #endif |
2347 | ||
1da177e4 LT |
2348 | /* |
2349 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
2350 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
2351 | * whenever sysctl_lowmem_reserve_ratio changes. | |
2352 | * | |
2353 | * The reserve ratio obviously has absolutely no relation with the | |
2354 | * pages_min watermarks. The lowmem reserve ratio can only make sense | |
2355 | * if in function of the boot time zone sizes. | |
2356 | */ | |
2357 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, | |
2358 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
2359 | { | |
2360 | proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
2361 | setup_per_zone_lowmem_reserve(); | |
2362 | return 0; | |
2363 | } | |
2364 | ||
8ad4b1fb RS |
2365 | /* |
2366 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
2367 | * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist | |
2368 | * can have before it gets flushed back to buddy allocator. | |
2369 | */ | |
2370 | ||
2371 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, | |
2372 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
2373 | { | |
2374 | struct zone *zone; | |
2375 | unsigned int cpu; | |
2376 | int ret; | |
2377 | ||
2378 | ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
2379 | if (!write || (ret == -EINVAL)) | |
2380 | return ret; | |
2381 | for_each_zone(zone) { | |
2382 | for_each_online_cpu(cpu) { | |
2383 | unsigned long high; | |
2384 | high = zone->present_pages / percpu_pagelist_fraction; | |
2385 | setup_pagelist_highmark(zone_pcp(zone, cpu), high); | |
2386 | } | |
2387 | } | |
2388 | return 0; | |
2389 | } | |
2390 | ||
f034b5d4 | 2391 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 LT |
2392 | |
2393 | #ifdef CONFIG_NUMA | |
2394 | static int __init set_hashdist(char *str) | |
2395 | { | |
2396 | if (!str) | |
2397 | return 0; | |
2398 | hashdist = simple_strtoul(str, &str, 0); | |
2399 | return 1; | |
2400 | } | |
2401 | __setup("hashdist=", set_hashdist); | |
2402 | #endif | |
2403 | ||
2404 | /* | |
2405 | * allocate a large system hash table from bootmem | |
2406 | * - it is assumed that the hash table must contain an exact power-of-2 | |
2407 | * quantity of entries | |
2408 | * - limit is the number of hash buckets, not the total allocation size | |
2409 | */ | |
2410 | void *__init alloc_large_system_hash(const char *tablename, | |
2411 | unsigned long bucketsize, | |
2412 | unsigned long numentries, | |
2413 | int scale, | |
2414 | int flags, | |
2415 | unsigned int *_hash_shift, | |
2416 | unsigned int *_hash_mask, | |
2417 | unsigned long limit) | |
2418 | { | |
2419 | unsigned long long max = limit; | |
2420 | unsigned long log2qty, size; | |
2421 | void *table = NULL; | |
2422 | ||
2423 | /* allow the kernel cmdline to have a say */ | |
2424 | if (!numentries) { | |
2425 | /* round applicable memory size up to nearest megabyte */ | |
2426 | numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages; | |
2427 | numentries += (1UL << (20 - PAGE_SHIFT)) - 1; | |
2428 | numentries >>= 20 - PAGE_SHIFT; | |
2429 | numentries <<= 20 - PAGE_SHIFT; | |
2430 | ||
2431 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
2432 | if (scale > PAGE_SHIFT) | |
2433 | numentries >>= (scale - PAGE_SHIFT); | |
2434 | else | |
2435 | numentries <<= (PAGE_SHIFT - scale); | |
2436 | } | |
6e692ed3 | 2437 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
2438 | |
2439 | /* limit allocation size to 1/16 total memory by default */ | |
2440 | if (max == 0) { | |
2441 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
2442 | do_div(max, bucketsize); | |
2443 | } | |
2444 | ||
2445 | if (numentries > max) | |
2446 | numentries = max; | |
2447 | ||
2448 | log2qty = long_log2(numentries); | |
2449 | ||
2450 | do { | |
2451 | size = bucketsize << log2qty; | |
2452 | if (flags & HASH_EARLY) | |
2453 | table = alloc_bootmem(size); | |
2454 | else if (hashdist) | |
2455 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
2456 | else { | |
2457 | unsigned long order; | |
2458 | for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++) | |
2459 | ; | |
2460 | table = (void*) __get_free_pages(GFP_ATOMIC, order); | |
2461 | } | |
2462 | } while (!table && size > PAGE_SIZE && --log2qty); | |
2463 | ||
2464 | if (!table) | |
2465 | panic("Failed to allocate %s hash table\n", tablename); | |
2466 | ||
2467 | printk("%s hash table entries: %d (order: %d, %lu bytes)\n", | |
2468 | tablename, | |
2469 | (1U << log2qty), | |
2470 | long_log2(size) - PAGE_SHIFT, | |
2471 | size); | |
2472 | ||
2473 | if (_hash_shift) | |
2474 | *_hash_shift = log2qty; | |
2475 | if (_hash_mask) | |
2476 | *_hash_mask = (1 << log2qty) - 1; | |
2477 | ||
2478 | return table; | |
2479 | } | |
a117e66e KH |
2480 | |
2481 | #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE | |
a117e66e KH |
2482 | struct page *pfn_to_page(unsigned long pfn) |
2483 | { | |
67de6482 | 2484 | return __pfn_to_page(pfn); |
a117e66e KH |
2485 | } |
2486 | unsigned long page_to_pfn(struct page *page) | |
2487 | { | |
67de6482 | 2488 | return __page_to_pfn(page); |
a117e66e | 2489 | } |
a117e66e KH |
2490 | EXPORT_SYMBOL(pfn_to_page); |
2491 | EXPORT_SYMBOL(page_to_pfn); | |
2492 | #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */ |