Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/page_alloc.c | |
3 | * | |
4 | * Manages the free list, the system allocates free pages here. | |
5 | * Note that kmalloc() lives in slab.c | |
6 | * | |
7 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
8 | * Swap reorganised 29.12.95, Stephen Tweedie | |
9 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 | |
10 | * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999 | |
11 | * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999 | |
12 | * Zone balancing, Kanoj Sarcar, SGI, Jan 2000 | |
13 | * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002 | |
14 | * (lots of bits borrowed from Ingo Molnar & Andrew Morton) | |
15 | */ | |
16 | ||
17 | #include <linux/config.h> | |
18 | #include <linux/stddef.h> | |
19 | #include <linux/mm.h> | |
20 | #include <linux/swap.h> | |
21 | #include <linux/interrupt.h> | |
22 | #include <linux/pagemap.h> | |
23 | #include <linux/bootmem.h> | |
24 | #include <linux/compiler.h> | |
9f158333 | 25 | #include <linux/kernel.h> |
1da177e4 LT |
26 | #include <linux/module.h> |
27 | #include <linux/suspend.h> | |
28 | #include <linux/pagevec.h> | |
29 | #include <linux/blkdev.h> | |
30 | #include <linux/slab.h> | |
31 | #include <linux/notifier.h> | |
32 | #include <linux/topology.h> | |
33 | #include <linux/sysctl.h> | |
34 | #include <linux/cpu.h> | |
35 | #include <linux/cpuset.h> | |
bdc8cb98 | 36 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
37 | #include <linux/nodemask.h> |
38 | #include <linux/vmalloc.h> | |
4be38e35 | 39 | #include <linux/mempolicy.h> |
6811378e | 40 | #include <linux/stop_machine.h> |
1da177e4 LT |
41 | |
42 | #include <asm/tlbflush.h> | |
ac924c60 | 43 | #include <asm/div64.h> |
1da177e4 LT |
44 | #include "internal.h" |
45 | ||
46 | /* | |
47 | * MCD - HACK: Find somewhere to initialize this EARLY, or make this | |
48 | * initializer cleaner | |
49 | */ | |
c3d8c141 | 50 | nodemask_t node_online_map __read_mostly = { { [0] = 1UL } }; |
7223a93a | 51 | EXPORT_SYMBOL(node_online_map); |
c3d8c141 | 52 | nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL; |
7223a93a | 53 | EXPORT_SYMBOL(node_possible_map); |
6c231b7b RT |
54 | unsigned long totalram_pages __read_mostly; |
55 | unsigned long totalhigh_pages __read_mostly; | |
cb45b0e9 | 56 | unsigned long totalreserve_pages __read_mostly; |
1da177e4 | 57 | long nr_swap_pages; |
8ad4b1fb | 58 | int percpu_pagelist_fraction; |
1da177e4 | 59 | |
d98c7a09 | 60 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 61 | |
1da177e4 LT |
62 | /* |
63 | * results with 256, 32 in the lowmem_reserve sysctl: | |
64 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
65 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
66 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
67 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
68 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA | |
a2f1b424 AK |
69 | * |
70 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
71 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 72 | */ |
a2f1b424 | 73 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { 256, 256, 32 }; |
1da177e4 LT |
74 | |
75 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 LT |
76 | |
77 | /* | |
78 | * Used by page_zone() to look up the address of the struct zone whose | |
79 | * id is encoded in the upper bits of page->flags | |
80 | */ | |
c3d8c141 | 81 | struct zone *zone_table[1 << ZONETABLE_SHIFT] __read_mostly; |
1da177e4 LT |
82 | EXPORT_SYMBOL(zone_table); |
83 | ||
a2f1b424 | 84 | static char *zone_names[MAX_NR_ZONES] = { "DMA", "DMA32", "Normal", "HighMem" }; |
1da177e4 LT |
85 | int min_free_kbytes = 1024; |
86 | ||
86356ab1 YG |
87 | unsigned long __meminitdata nr_kernel_pages; |
88 | unsigned long __meminitdata nr_all_pages; | |
1da177e4 | 89 | |
13e7444b | 90 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 91 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 92 | { |
bdc8cb98 DH |
93 | int ret = 0; |
94 | unsigned seq; | |
95 | unsigned long pfn = page_to_pfn(page); | |
c6a57e19 | 96 | |
bdc8cb98 DH |
97 | do { |
98 | seq = zone_span_seqbegin(zone); | |
99 | if (pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
100 | ret = 1; | |
101 | else if (pfn < zone->zone_start_pfn) | |
102 | ret = 1; | |
103 | } while (zone_span_seqretry(zone, seq)); | |
104 | ||
105 | return ret; | |
c6a57e19 DH |
106 | } |
107 | ||
108 | static int page_is_consistent(struct zone *zone, struct page *page) | |
109 | { | |
1da177e4 LT |
110 | #ifdef CONFIG_HOLES_IN_ZONE |
111 | if (!pfn_valid(page_to_pfn(page))) | |
c6a57e19 | 112 | return 0; |
1da177e4 LT |
113 | #endif |
114 | if (zone != page_zone(page)) | |
c6a57e19 DH |
115 | return 0; |
116 | ||
117 | return 1; | |
118 | } | |
119 | /* | |
120 | * Temporary debugging check for pages not lying within a given zone. | |
121 | */ | |
122 | static int bad_range(struct zone *zone, struct page *page) | |
123 | { | |
124 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 125 | return 1; |
c6a57e19 DH |
126 | if (!page_is_consistent(zone, page)) |
127 | return 1; | |
128 | ||
1da177e4 LT |
129 | return 0; |
130 | } | |
131 | ||
13e7444b NP |
132 | #else |
133 | static inline int bad_range(struct zone *zone, struct page *page) | |
134 | { | |
135 | return 0; | |
136 | } | |
137 | #endif | |
138 | ||
224abf92 | 139 | static void bad_page(struct page *page) |
1da177e4 | 140 | { |
224abf92 | 141 | printk(KERN_EMERG "Bad page state in process '%s'\n" |
7365f3d1 HD |
142 | KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n" |
143 | KERN_EMERG "Trying to fix it up, but a reboot is needed\n" | |
144 | KERN_EMERG "Backtrace:\n", | |
224abf92 NP |
145 | current->comm, page, (int)(2*sizeof(unsigned long)), |
146 | (unsigned long)page->flags, page->mapping, | |
147 | page_mapcount(page), page_count(page)); | |
1da177e4 | 148 | dump_stack(); |
334795ec HD |
149 | page->flags &= ~(1 << PG_lru | |
150 | 1 << PG_private | | |
1da177e4 | 151 | 1 << PG_locked | |
1da177e4 LT |
152 | 1 << PG_active | |
153 | 1 << PG_dirty | | |
334795ec HD |
154 | 1 << PG_reclaim | |
155 | 1 << PG_slab | | |
1da177e4 | 156 | 1 << PG_swapcache | |
676165a8 NP |
157 | 1 << PG_writeback | |
158 | 1 << PG_buddy ); | |
1da177e4 LT |
159 | set_page_count(page, 0); |
160 | reset_page_mapcount(page); | |
161 | page->mapping = NULL; | |
9f158333 | 162 | add_taint(TAINT_BAD_PAGE); |
1da177e4 LT |
163 | } |
164 | ||
1da177e4 LT |
165 | /* |
166 | * Higher-order pages are called "compound pages". They are structured thusly: | |
167 | * | |
168 | * The first PAGE_SIZE page is called the "head page". | |
169 | * | |
170 | * The remaining PAGE_SIZE pages are called "tail pages". | |
171 | * | |
172 | * All pages have PG_compound set. All pages have their ->private pointing at | |
173 | * the head page (even the head page has this). | |
174 | * | |
41d78ba5 HD |
175 | * The first tail page's ->lru.next holds the address of the compound page's |
176 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
177 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 178 | */ |
d98c7a09 HD |
179 | |
180 | static void free_compound_page(struct page *page) | |
181 | { | |
182 | __free_pages_ok(page, (unsigned long)page[1].lru.prev); | |
183 | } | |
184 | ||
1da177e4 LT |
185 | static void prep_compound_page(struct page *page, unsigned long order) |
186 | { | |
187 | int i; | |
188 | int nr_pages = 1 << order; | |
189 | ||
d98c7a09 | 190 | page[1].lru.next = (void *)free_compound_page; /* set dtor */ |
41d78ba5 | 191 | page[1].lru.prev = (void *)order; |
1da177e4 LT |
192 | for (i = 0; i < nr_pages; i++) { |
193 | struct page *p = page + i; | |
194 | ||
5e9dace8 | 195 | __SetPageCompound(p); |
4c21e2f2 | 196 | set_page_private(p, (unsigned long)page); |
1da177e4 LT |
197 | } |
198 | } | |
199 | ||
200 | static void destroy_compound_page(struct page *page, unsigned long order) | |
201 | { | |
202 | int i; | |
203 | int nr_pages = 1 << order; | |
204 | ||
41d78ba5 | 205 | if (unlikely((unsigned long)page[1].lru.prev != order)) |
224abf92 | 206 | bad_page(page); |
1da177e4 LT |
207 | |
208 | for (i = 0; i < nr_pages; i++) { | |
209 | struct page *p = page + i; | |
210 | ||
224abf92 NP |
211 | if (unlikely(!PageCompound(p) | |
212 | (page_private(p) != (unsigned long)page))) | |
213 | bad_page(page); | |
5e9dace8 | 214 | __ClearPageCompound(p); |
1da177e4 LT |
215 | } |
216 | } | |
1da177e4 | 217 | |
17cf4406 NP |
218 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) |
219 | { | |
220 | int i; | |
221 | ||
222 | BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM); | |
6626c5d5 AM |
223 | /* |
224 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO | |
225 | * and __GFP_HIGHMEM from hard or soft interrupt context. | |
226 | */ | |
227 | BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); | |
17cf4406 NP |
228 | for (i = 0; i < (1 << order); i++) |
229 | clear_highpage(page + i); | |
230 | } | |
231 | ||
1da177e4 LT |
232 | /* |
233 | * function for dealing with page's order in buddy system. | |
234 | * zone->lock is already acquired when we use these. | |
235 | * So, we don't need atomic page->flags operations here. | |
236 | */ | |
6aa3001b AM |
237 | static inline unsigned long page_order(struct page *page) |
238 | { | |
4c21e2f2 | 239 | return page_private(page); |
1da177e4 LT |
240 | } |
241 | ||
6aa3001b AM |
242 | static inline void set_page_order(struct page *page, int order) |
243 | { | |
4c21e2f2 | 244 | set_page_private(page, order); |
676165a8 | 245 | __SetPageBuddy(page); |
1da177e4 LT |
246 | } |
247 | ||
248 | static inline void rmv_page_order(struct page *page) | |
249 | { | |
676165a8 | 250 | __ClearPageBuddy(page); |
4c21e2f2 | 251 | set_page_private(page, 0); |
1da177e4 LT |
252 | } |
253 | ||
254 | /* | |
255 | * Locate the struct page for both the matching buddy in our | |
256 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
257 | * | |
258 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
259 | * the following equation: | |
260 | * B2 = B1 ^ (1 << O) | |
261 | * For example, if the starting buddy (buddy2) is #8 its order | |
262 | * 1 buddy is #10: | |
263 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
264 | * | |
265 | * 2) Any buddy B will have an order O+1 parent P which | |
266 | * satisfies the following equation: | |
267 | * P = B & ~(1 << O) | |
268 | * | |
d6e05edc | 269 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER |
1da177e4 LT |
270 | */ |
271 | static inline struct page * | |
272 | __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order) | |
273 | { | |
274 | unsigned long buddy_idx = page_idx ^ (1 << order); | |
275 | ||
276 | return page + (buddy_idx - page_idx); | |
277 | } | |
278 | ||
279 | static inline unsigned long | |
280 | __find_combined_index(unsigned long page_idx, unsigned int order) | |
281 | { | |
282 | return (page_idx & ~(1 << order)); | |
283 | } | |
284 | ||
285 | /* | |
286 | * This function checks whether a page is free && is the buddy | |
287 | * we can do coalesce a page and its buddy if | |
13e7444b | 288 | * (a) the buddy is not in a hole && |
676165a8 | 289 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
290 | * (c) a page and its buddy have the same order && |
291 | * (d) a page and its buddy are in the same zone. | |
676165a8 NP |
292 | * |
293 | * For recording whether a page is in the buddy system, we use PG_buddy. | |
294 | * Setting, clearing, and testing PG_buddy is serialized by zone->lock. | |
1da177e4 | 295 | * |
676165a8 | 296 | * For recording page's order, we use page_private(page). |
1da177e4 | 297 | */ |
cb2b95e1 AW |
298 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
299 | int order) | |
1da177e4 | 300 | { |
13e7444b | 301 | #ifdef CONFIG_HOLES_IN_ZONE |
cb2b95e1 | 302 | if (!pfn_valid(page_to_pfn(buddy))) |
13e7444b NP |
303 | return 0; |
304 | #endif | |
305 | ||
cb2b95e1 AW |
306 | if (page_zone_id(page) != page_zone_id(buddy)) |
307 | return 0; | |
308 | ||
309 | if (PageBuddy(buddy) && page_order(buddy) == order) { | |
310 | BUG_ON(page_count(buddy) != 0); | |
6aa3001b | 311 | return 1; |
676165a8 | 312 | } |
6aa3001b | 313 | return 0; |
1da177e4 LT |
314 | } |
315 | ||
316 | /* | |
317 | * Freeing function for a buddy system allocator. | |
318 | * | |
319 | * The concept of a buddy system is to maintain direct-mapped table | |
320 | * (containing bit values) for memory blocks of various "orders". | |
321 | * The bottom level table contains the map for the smallest allocatable | |
322 | * units of memory (here, pages), and each level above it describes | |
323 | * pairs of units from the levels below, hence, "buddies". | |
324 | * At a high level, all that happens here is marking the table entry | |
325 | * at the bottom level available, and propagating the changes upward | |
326 | * as necessary, plus some accounting needed to play nicely with other | |
327 | * parts of the VM system. | |
328 | * At each level, we keep a list of pages, which are heads of continuous | |
676165a8 | 329 | * free pages of length of (1 << order) and marked with PG_buddy. Page's |
4c21e2f2 | 330 | * order is recorded in page_private(page) field. |
1da177e4 LT |
331 | * So when we are allocating or freeing one, we can derive the state of the |
332 | * other. That is, if we allocate a small block, and both were | |
333 | * free, the remainder of the region must be split into blocks. | |
334 | * If a block is freed, and its buddy is also free, then this | |
335 | * triggers coalescing into a block of larger size. | |
336 | * | |
337 | * -- wli | |
338 | */ | |
339 | ||
48db57f8 | 340 | static inline void __free_one_page(struct page *page, |
1da177e4 LT |
341 | struct zone *zone, unsigned int order) |
342 | { | |
343 | unsigned long page_idx; | |
344 | int order_size = 1 << order; | |
345 | ||
224abf92 | 346 | if (unlikely(PageCompound(page))) |
1da177e4 LT |
347 | destroy_compound_page(page, order); |
348 | ||
349 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); | |
350 | ||
351 | BUG_ON(page_idx & (order_size - 1)); | |
352 | BUG_ON(bad_range(zone, page)); | |
353 | ||
354 | zone->free_pages += order_size; | |
355 | while (order < MAX_ORDER-1) { | |
356 | unsigned long combined_idx; | |
357 | struct free_area *area; | |
358 | struct page *buddy; | |
359 | ||
1da177e4 | 360 | buddy = __page_find_buddy(page, page_idx, order); |
cb2b95e1 | 361 | if (!page_is_buddy(page, buddy, order)) |
1da177e4 | 362 | break; /* Move the buddy up one level. */ |
13e7444b | 363 | |
1da177e4 LT |
364 | list_del(&buddy->lru); |
365 | area = zone->free_area + order; | |
366 | area->nr_free--; | |
367 | rmv_page_order(buddy); | |
13e7444b | 368 | combined_idx = __find_combined_index(page_idx, order); |
1da177e4 LT |
369 | page = page + (combined_idx - page_idx); |
370 | page_idx = combined_idx; | |
371 | order++; | |
372 | } | |
373 | set_page_order(page, order); | |
374 | list_add(&page->lru, &zone->free_area[order].free_list); | |
375 | zone->free_area[order].nr_free++; | |
376 | } | |
377 | ||
224abf92 | 378 | static inline int free_pages_check(struct page *page) |
1da177e4 | 379 | { |
92be2e33 NP |
380 | if (unlikely(page_mapcount(page) | |
381 | (page->mapping != NULL) | | |
382 | (page_count(page) != 0) | | |
1da177e4 LT |
383 | (page->flags & ( |
384 | 1 << PG_lru | | |
385 | 1 << PG_private | | |
386 | 1 << PG_locked | | |
387 | 1 << PG_active | | |
388 | 1 << PG_reclaim | | |
389 | 1 << PG_slab | | |
390 | 1 << PG_swapcache | | |
b5810039 | 391 | 1 << PG_writeback | |
676165a8 NP |
392 | 1 << PG_reserved | |
393 | 1 << PG_buddy )))) | |
224abf92 | 394 | bad_page(page); |
1da177e4 | 395 | if (PageDirty(page)) |
242e5468 | 396 | __ClearPageDirty(page); |
689bcebf HD |
397 | /* |
398 | * For now, we report if PG_reserved was found set, but do not | |
399 | * clear it, and do not free the page. But we shall soon need | |
400 | * to do more, for when the ZERO_PAGE count wraps negative. | |
401 | */ | |
402 | return PageReserved(page); | |
1da177e4 LT |
403 | } |
404 | ||
405 | /* | |
406 | * Frees a list of pages. | |
407 | * Assumes all pages on list are in same zone, and of same order. | |
207f36ee | 408 | * count is the number of pages to free. |
1da177e4 LT |
409 | * |
410 | * If the zone was previously in an "all pages pinned" state then look to | |
411 | * see if this freeing clears that state. | |
412 | * | |
413 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
414 | * pinned" detection logic. | |
415 | */ | |
48db57f8 NP |
416 | static void free_pages_bulk(struct zone *zone, int count, |
417 | struct list_head *list, int order) | |
1da177e4 | 418 | { |
c54ad30c | 419 | spin_lock(&zone->lock); |
1da177e4 LT |
420 | zone->all_unreclaimable = 0; |
421 | zone->pages_scanned = 0; | |
48db57f8 NP |
422 | while (count--) { |
423 | struct page *page; | |
424 | ||
425 | BUG_ON(list_empty(list)); | |
1da177e4 | 426 | page = list_entry(list->prev, struct page, lru); |
48db57f8 | 427 | /* have to delete it as __free_one_page list manipulates */ |
1da177e4 | 428 | list_del(&page->lru); |
48db57f8 | 429 | __free_one_page(page, zone, order); |
1da177e4 | 430 | } |
c54ad30c | 431 | spin_unlock(&zone->lock); |
1da177e4 LT |
432 | } |
433 | ||
48db57f8 | 434 | static void free_one_page(struct zone *zone, struct page *page, int order) |
1da177e4 LT |
435 | { |
436 | LIST_HEAD(list); | |
48db57f8 NP |
437 | list_add(&page->lru, &list); |
438 | free_pages_bulk(zone, 1, &list, order); | |
439 | } | |
440 | ||
441 | static void __free_pages_ok(struct page *page, unsigned int order) | |
442 | { | |
443 | unsigned long flags; | |
1da177e4 | 444 | int i; |
689bcebf | 445 | int reserved = 0; |
1da177e4 LT |
446 | |
447 | arch_free_page(page, order); | |
de5097c2 | 448 | if (!PageHighMem(page)) |
f9b8404c IM |
449 | debug_check_no_locks_freed(page_address(page), |
450 | PAGE_SIZE<<order); | |
1da177e4 | 451 | |
1da177e4 | 452 | for (i = 0 ; i < (1 << order) ; ++i) |
224abf92 | 453 | reserved += free_pages_check(page + i); |
689bcebf HD |
454 | if (reserved) |
455 | return; | |
456 | ||
48db57f8 | 457 | kernel_map_pages(page, 1 << order, 0); |
c54ad30c | 458 | local_irq_save(flags); |
a74609fa | 459 | __mod_page_state(pgfree, 1 << order); |
48db57f8 | 460 | free_one_page(page_zone(page), page, order); |
c54ad30c | 461 | local_irq_restore(flags); |
1da177e4 LT |
462 | } |
463 | ||
a226f6c8 DH |
464 | /* |
465 | * permit the bootmem allocator to evade page validation on high-order frees | |
466 | */ | |
467 | void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order) | |
468 | { | |
469 | if (order == 0) { | |
470 | __ClearPageReserved(page); | |
471 | set_page_count(page, 0); | |
7835e98b | 472 | set_page_refcounted(page); |
545b1ea9 | 473 | __free_page(page); |
a226f6c8 | 474 | } else { |
a226f6c8 DH |
475 | int loop; |
476 | ||
545b1ea9 | 477 | prefetchw(page); |
a226f6c8 DH |
478 | for (loop = 0; loop < BITS_PER_LONG; loop++) { |
479 | struct page *p = &page[loop]; | |
480 | ||
545b1ea9 NP |
481 | if (loop + 1 < BITS_PER_LONG) |
482 | prefetchw(p + 1); | |
a226f6c8 DH |
483 | __ClearPageReserved(p); |
484 | set_page_count(p, 0); | |
485 | } | |
486 | ||
7835e98b | 487 | set_page_refcounted(page); |
545b1ea9 | 488 | __free_pages(page, order); |
a226f6c8 DH |
489 | } |
490 | } | |
491 | ||
1da177e4 LT |
492 | |
493 | /* | |
494 | * The order of subdivision here is critical for the IO subsystem. | |
495 | * Please do not alter this order without good reasons and regression | |
496 | * testing. Specifically, as large blocks of memory are subdivided, | |
497 | * the order in which smaller blocks are delivered depends on the order | |
498 | * they're subdivided in this function. This is the primary factor | |
499 | * influencing the order in which pages are delivered to the IO | |
500 | * subsystem according to empirical testing, and this is also justified | |
501 | * by considering the behavior of a buddy system containing a single | |
502 | * large block of memory acted on by a series of small allocations. | |
503 | * This behavior is a critical factor in sglist merging's success. | |
504 | * | |
505 | * -- wli | |
506 | */ | |
085cc7d5 | 507 | static inline void expand(struct zone *zone, struct page *page, |
1da177e4 LT |
508 | int low, int high, struct free_area *area) |
509 | { | |
510 | unsigned long size = 1 << high; | |
511 | ||
512 | while (high > low) { | |
513 | area--; | |
514 | high--; | |
515 | size >>= 1; | |
516 | BUG_ON(bad_range(zone, &page[size])); | |
517 | list_add(&page[size].lru, &area->free_list); | |
518 | area->nr_free++; | |
519 | set_page_order(&page[size], high); | |
520 | } | |
1da177e4 LT |
521 | } |
522 | ||
1da177e4 LT |
523 | /* |
524 | * This page is about to be returned from the page allocator | |
525 | */ | |
17cf4406 | 526 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) |
1da177e4 | 527 | { |
92be2e33 NP |
528 | if (unlikely(page_mapcount(page) | |
529 | (page->mapping != NULL) | | |
530 | (page_count(page) != 0) | | |
334795ec HD |
531 | (page->flags & ( |
532 | 1 << PG_lru | | |
1da177e4 LT |
533 | 1 << PG_private | |
534 | 1 << PG_locked | | |
1da177e4 LT |
535 | 1 << PG_active | |
536 | 1 << PG_dirty | | |
537 | 1 << PG_reclaim | | |
334795ec | 538 | 1 << PG_slab | |
1da177e4 | 539 | 1 << PG_swapcache | |
b5810039 | 540 | 1 << PG_writeback | |
676165a8 NP |
541 | 1 << PG_reserved | |
542 | 1 << PG_buddy )))) | |
224abf92 | 543 | bad_page(page); |
1da177e4 | 544 | |
689bcebf HD |
545 | /* |
546 | * For now, we report if PG_reserved was found set, but do not | |
547 | * clear it, and do not allocate the page: as a safety net. | |
548 | */ | |
549 | if (PageReserved(page)) | |
550 | return 1; | |
551 | ||
1da177e4 LT |
552 | page->flags &= ~(1 << PG_uptodate | 1 << PG_error | |
553 | 1 << PG_referenced | 1 << PG_arch_1 | | |
554 | 1 << PG_checked | 1 << PG_mappedtodisk); | |
4c21e2f2 | 555 | set_page_private(page, 0); |
7835e98b | 556 | set_page_refcounted(page); |
1da177e4 | 557 | kernel_map_pages(page, 1 << order, 1); |
17cf4406 NP |
558 | |
559 | if (gfp_flags & __GFP_ZERO) | |
560 | prep_zero_page(page, order, gfp_flags); | |
561 | ||
562 | if (order && (gfp_flags & __GFP_COMP)) | |
563 | prep_compound_page(page, order); | |
564 | ||
689bcebf | 565 | return 0; |
1da177e4 LT |
566 | } |
567 | ||
568 | /* | |
569 | * Do the hard work of removing an element from the buddy allocator. | |
570 | * Call me with the zone->lock already held. | |
571 | */ | |
572 | static struct page *__rmqueue(struct zone *zone, unsigned int order) | |
573 | { | |
574 | struct free_area * area; | |
575 | unsigned int current_order; | |
576 | struct page *page; | |
577 | ||
578 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
579 | area = zone->free_area + current_order; | |
580 | if (list_empty(&area->free_list)) | |
581 | continue; | |
582 | ||
583 | page = list_entry(area->free_list.next, struct page, lru); | |
584 | list_del(&page->lru); | |
585 | rmv_page_order(page); | |
586 | area->nr_free--; | |
587 | zone->free_pages -= 1UL << order; | |
085cc7d5 NP |
588 | expand(zone, page, order, current_order, area); |
589 | return page; | |
1da177e4 LT |
590 | } |
591 | ||
592 | return NULL; | |
593 | } | |
594 | ||
595 | /* | |
596 | * Obtain a specified number of elements from the buddy allocator, all under | |
597 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
598 | * Returns the number of new pages which were placed at *list. | |
599 | */ | |
600 | static int rmqueue_bulk(struct zone *zone, unsigned int order, | |
601 | unsigned long count, struct list_head *list) | |
602 | { | |
1da177e4 | 603 | int i; |
1da177e4 | 604 | |
c54ad30c | 605 | spin_lock(&zone->lock); |
1da177e4 | 606 | for (i = 0; i < count; ++i) { |
085cc7d5 NP |
607 | struct page *page = __rmqueue(zone, order); |
608 | if (unlikely(page == NULL)) | |
1da177e4 | 609 | break; |
1da177e4 LT |
610 | list_add_tail(&page->lru, list); |
611 | } | |
c54ad30c | 612 | spin_unlock(&zone->lock); |
085cc7d5 | 613 | return i; |
1da177e4 LT |
614 | } |
615 | ||
4ae7c039 | 616 | #ifdef CONFIG_NUMA |
8fce4d8e CL |
617 | /* |
618 | * Called from the slab reaper to drain pagesets on a particular node that | |
619 | * belong to the currently executing processor. | |
879336c3 CL |
620 | * Note that this function must be called with the thread pinned to |
621 | * a single processor. | |
8fce4d8e CL |
622 | */ |
623 | void drain_node_pages(int nodeid) | |
4ae7c039 | 624 | { |
8fce4d8e | 625 | int i, z; |
4ae7c039 CL |
626 | unsigned long flags; |
627 | ||
8fce4d8e CL |
628 | for (z = 0; z < MAX_NR_ZONES; z++) { |
629 | struct zone *zone = NODE_DATA(nodeid)->node_zones + z; | |
4ae7c039 CL |
630 | struct per_cpu_pageset *pset; |
631 | ||
23316bc8 | 632 | pset = zone_pcp(zone, smp_processor_id()); |
4ae7c039 CL |
633 | for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) { |
634 | struct per_cpu_pages *pcp; | |
635 | ||
636 | pcp = &pset->pcp[i]; | |
879336c3 CL |
637 | if (pcp->count) { |
638 | local_irq_save(flags); | |
639 | free_pages_bulk(zone, pcp->count, &pcp->list, 0); | |
640 | pcp->count = 0; | |
641 | local_irq_restore(flags); | |
642 | } | |
4ae7c039 CL |
643 | } |
644 | } | |
4ae7c039 CL |
645 | } |
646 | #endif | |
647 | ||
1da177e4 LT |
648 | #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU) |
649 | static void __drain_pages(unsigned int cpu) | |
650 | { | |
c54ad30c | 651 | unsigned long flags; |
1da177e4 LT |
652 | struct zone *zone; |
653 | int i; | |
654 | ||
655 | for_each_zone(zone) { | |
656 | struct per_cpu_pageset *pset; | |
657 | ||
e7c8d5c9 | 658 | pset = zone_pcp(zone, cpu); |
1da177e4 LT |
659 | for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) { |
660 | struct per_cpu_pages *pcp; | |
661 | ||
662 | pcp = &pset->pcp[i]; | |
c54ad30c | 663 | local_irq_save(flags); |
48db57f8 NP |
664 | free_pages_bulk(zone, pcp->count, &pcp->list, 0); |
665 | pcp->count = 0; | |
c54ad30c | 666 | local_irq_restore(flags); |
1da177e4 LT |
667 | } |
668 | } | |
669 | } | |
670 | #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */ | |
671 | ||
672 | #ifdef CONFIG_PM | |
673 | ||
674 | void mark_free_pages(struct zone *zone) | |
675 | { | |
676 | unsigned long zone_pfn, flags; | |
677 | int order; | |
678 | struct list_head *curr; | |
679 | ||
680 | if (!zone->spanned_pages) | |
681 | return; | |
682 | ||
683 | spin_lock_irqsave(&zone->lock, flags); | |
684 | for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) | |
685 | ClearPageNosaveFree(pfn_to_page(zone_pfn + zone->zone_start_pfn)); | |
686 | ||
687 | for (order = MAX_ORDER - 1; order >= 0; --order) | |
688 | list_for_each(curr, &zone->free_area[order].free_list) { | |
689 | unsigned long start_pfn, i; | |
690 | ||
691 | start_pfn = page_to_pfn(list_entry(curr, struct page, lru)); | |
692 | ||
693 | for (i=0; i < (1<<order); i++) | |
694 | SetPageNosaveFree(pfn_to_page(start_pfn+i)); | |
695 | } | |
696 | spin_unlock_irqrestore(&zone->lock, flags); | |
697 | } | |
698 | ||
699 | /* | |
700 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
701 | */ | |
702 | void drain_local_pages(void) | |
703 | { | |
704 | unsigned long flags; | |
705 | ||
706 | local_irq_save(flags); | |
707 | __drain_pages(smp_processor_id()); | |
708 | local_irq_restore(flags); | |
709 | } | |
710 | #endif /* CONFIG_PM */ | |
711 | ||
a74609fa | 712 | static void zone_statistics(struct zonelist *zonelist, struct zone *z, int cpu) |
1da177e4 LT |
713 | { |
714 | #ifdef CONFIG_NUMA | |
1da177e4 LT |
715 | pg_data_t *pg = z->zone_pgdat; |
716 | pg_data_t *orig = zonelist->zones[0]->zone_pgdat; | |
717 | struct per_cpu_pageset *p; | |
718 | ||
a74609fa | 719 | p = zone_pcp(z, cpu); |
1da177e4 | 720 | if (pg == orig) { |
e7c8d5c9 | 721 | p->numa_hit++; |
1da177e4 LT |
722 | } else { |
723 | p->numa_miss++; | |
e7c8d5c9 | 724 | zone_pcp(zonelist->zones[0], cpu)->numa_foreign++; |
1da177e4 LT |
725 | } |
726 | if (pg == NODE_DATA(numa_node_id())) | |
727 | p->local_node++; | |
728 | else | |
729 | p->other_node++; | |
1da177e4 LT |
730 | #endif |
731 | } | |
732 | ||
733 | /* | |
734 | * Free a 0-order page | |
735 | */ | |
1da177e4 LT |
736 | static void fastcall free_hot_cold_page(struct page *page, int cold) |
737 | { | |
738 | struct zone *zone = page_zone(page); | |
739 | struct per_cpu_pages *pcp; | |
740 | unsigned long flags; | |
741 | ||
742 | arch_free_page(page, 0); | |
743 | ||
1da177e4 LT |
744 | if (PageAnon(page)) |
745 | page->mapping = NULL; | |
224abf92 | 746 | if (free_pages_check(page)) |
689bcebf HD |
747 | return; |
748 | ||
689bcebf HD |
749 | kernel_map_pages(page, 1, 0); |
750 | ||
e7c8d5c9 | 751 | pcp = &zone_pcp(zone, get_cpu())->pcp[cold]; |
1da177e4 | 752 | local_irq_save(flags); |
a74609fa | 753 | __inc_page_state(pgfree); |
1da177e4 LT |
754 | list_add(&page->lru, &pcp->list); |
755 | pcp->count++; | |
48db57f8 NP |
756 | if (pcp->count >= pcp->high) { |
757 | free_pages_bulk(zone, pcp->batch, &pcp->list, 0); | |
758 | pcp->count -= pcp->batch; | |
759 | } | |
1da177e4 LT |
760 | local_irq_restore(flags); |
761 | put_cpu(); | |
762 | } | |
763 | ||
764 | void fastcall free_hot_page(struct page *page) | |
765 | { | |
766 | free_hot_cold_page(page, 0); | |
767 | } | |
768 | ||
769 | void fastcall free_cold_page(struct page *page) | |
770 | { | |
771 | free_hot_cold_page(page, 1); | |
772 | } | |
773 | ||
8dfcc9ba NP |
774 | /* |
775 | * split_page takes a non-compound higher-order page, and splits it into | |
776 | * n (1<<order) sub-pages: page[0..n] | |
777 | * Each sub-page must be freed individually. | |
778 | * | |
779 | * Note: this is probably too low level an operation for use in drivers. | |
780 | * Please consult with lkml before using this in your driver. | |
781 | */ | |
782 | void split_page(struct page *page, unsigned int order) | |
783 | { | |
784 | int i; | |
785 | ||
786 | BUG_ON(PageCompound(page)); | |
787 | BUG_ON(!page_count(page)); | |
7835e98b NP |
788 | for (i = 1; i < (1 << order); i++) |
789 | set_page_refcounted(page + i); | |
8dfcc9ba | 790 | } |
8dfcc9ba | 791 | |
1da177e4 LT |
792 | /* |
793 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
794 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
795 | * or two. | |
796 | */ | |
a74609fa NP |
797 | static struct page *buffered_rmqueue(struct zonelist *zonelist, |
798 | struct zone *zone, int order, gfp_t gfp_flags) | |
1da177e4 LT |
799 | { |
800 | unsigned long flags; | |
689bcebf | 801 | struct page *page; |
1da177e4 | 802 | int cold = !!(gfp_flags & __GFP_COLD); |
a74609fa | 803 | int cpu; |
1da177e4 | 804 | |
689bcebf | 805 | again: |
a74609fa | 806 | cpu = get_cpu(); |
48db57f8 | 807 | if (likely(order == 0)) { |
1da177e4 LT |
808 | struct per_cpu_pages *pcp; |
809 | ||
a74609fa | 810 | pcp = &zone_pcp(zone, cpu)->pcp[cold]; |
1da177e4 | 811 | local_irq_save(flags); |
a74609fa | 812 | if (!pcp->count) { |
1da177e4 LT |
813 | pcp->count += rmqueue_bulk(zone, 0, |
814 | pcp->batch, &pcp->list); | |
a74609fa NP |
815 | if (unlikely(!pcp->count)) |
816 | goto failed; | |
1da177e4 | 817 | } |
a74609fa NP |
818 | page = list_entry(pcp->list.next, struct page, lru); |
819 | list_del(&page->lru); | |
820 | pcp->count--; | |
7fb1d9fc | 821 | } else { |
1da177e4 LT |
822 | spin_lock_irqsave(&zone->lock, flags); |
823 | page = __rmqueue(zone, order); | |
a74609fa NP |
824 | spin_unlock(&zone->lock); |
825 | if (!page) | |
826 | goto failed; | |
1da177e4 LT |
827 | } |
828 | ||
a74609fa NP |
829 | __mod_page_state_zone(zone, pgalloc, 1 << order); |
830 | zone_statistics(zonelist, zone, cpu); | |
831 | local_irq_restore(flags); | |
832 | put_cpu(); | |
1da177e4 | 833 | |
a74609fa | 834 | BUG_ON(bad_range(zone, page)); |
17cf4406 | 835 | if (prep_new_page(page, order, gfp_flags)) |
a74609fa | 836 | goto again; |
1da177e4 | 837 | return page; |
a74609fa NP |
838 | |
839 | failed: | |
840 | local_irq_restore(flags); | |
841 | put_cpu(); | |
842 | return NULL; | |
1da177e4 LT |
843 | } |
844 | ||
7fb1d9fc | 845 | #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */ |
3148890b NP |
846 | #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */ |
847 | #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */ | |
848 | #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */ | |
849 | #define ALLOC_HARDER 0x10 /* try to alloc harder */ | |
850 | #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */ | |
851 | #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ | |
7fb1d9fc | 852 | |
1da177e4 LT |
853 | /* |
854 | * Return 1 if free pages are above 'mark'. This takes into account the order | |
855 | * of the allocation. | |
856 | */ | |
857 | int zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
7fb1d9fc | 858 | int classzone_idx, int alloc_flags) |
1da177e4 LT |
859 | { |
860 | /* free_pages my go negative - that's OK */ | |
861 | long min = mark, free_pages = z->free_pages - (1 << order) + 1; | |
862 | int o; | |
863 | ||
7fb1d9fc | 864 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 865 | min -= min / 2; |
7fb1d9fc | 866 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 LT |
867 | min -= min / 4; |
868 | ||
869 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
870 | return 0; | |
871 | for (o = 0; o < order; o++) { | |
872 | /* At the next order, this order's pages become unavailable */ | |
873 | free_pages -= z->free_area[o].nr_free << o; | |
874 | ||
875 | /* Require fewer higher order pages to be free */ | |
876 | min >>= 1; | |
877 | ||
878 | if (free_pages <= min) | |
879 | return 0; | |
880 | } | |
881 | return 1; | |
882 | } | |
883 | ||
7fb1d9fc RS |
884 | /* |
885 | * get_page_from_freeliest goes through the zonelist trying to allocate | |
886 | * a page. | |
887 | */ | |
888 | static struct page * | |
889 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, | |
890 | struct zonelist *zonelist, int alloc_flags) | |
753ee728 | 891 | { |
7fb1d9fc RS |
892 | struct zone **z = zonelist->zones; |
893 | struct page *page = NULL; | |
894 | int classzone_idx = zone_idx(*z); | |
895 | ||
896 | /* | |
897 | * Go through the zonelist once, looking for a zone with enough free. | |
898 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. | |
899 | */ | |
900 | do { | |
901 | if ((alloc_flags & ALLOC_CPUSET) && | |
902 | !cpuset_zone_allowed(*z, gfp_mask)) | |
903 | continue; | |
904 | ||
905 | if (!(alloc_flags & ALLOC_NO_WATERMARKS)) { | |
3148890b NP |
906 | unsigned long mark; |
907 | if (alloc_flags & ALLOC_WMARK_MIN) | |
908 | mark = (*z)->pages_min; | |
909 | else if (alloc_flags & ALLOC_WMARK_LOW) | |
910 | mark = (*z)->pages_low; | |
911 | else | |
912 | mark = (*z)->pages_high; | |
913 | if (!zone_watermark_ok(*z, order, mark, | |
7fb1d9fc | 914 | classzone_idx, alloc_flags)) |
9eeff239 CL |
915 | if (!zone_reclaim_mode || |
916 | !zone_reclaim(*z, gfp_mask, order)) | |
917 | continue; | |
7fb1d9fc RS |
918 | } |
919 | ||
a74609fa | 920 | page = buffered_rmqueue(zonelist, *z, order, gfp_mask); |
7fb1d9fc | 921 | if (page) { |
7fb1d9fc RS |
922 | break; |
923 | } | |
924 | } while (*(++z) != NULL); | |
925 | return page; | |
753ee728 MH |
926 | } |
927 | ||
1da177e4 LT |
928 | /* |
929 | * This is the 'heart' of the zoned buddy allocator. | |
930 | */ | |
931 | struct page * fastcall | |
dd0fc66f | 932 | __alloc_pages(gfp_t gfp_mask, unsigned int order, |
1da177e4 LT |
933 | struct zonelist *zonelist) |
934 | { | |
260b2367 | 935 | const gfp_t wait = gfp_mask & __GFP_WAIT; |
7fb1d9fc | 936 | struct zone **z; |
1da177e4 LT |
937 | struct page *page; |
938 | struct reclaim_state reclaim_state; | |
939 | struct task_struct *p = current; | |
1da177e4 | 940 | int do_retry; |
7fb1d9fc | 941 | int alloc_flags; |
1da177e4 LT |
942 | int did_some_progress; |
943 | ||
944 | might_sleep_if(wait); | |
945 | ||
6b1de916 | 946 | restart: |
7fb1d9fc | 947 | z = zonelist->zones; /* the list of zones suitable for gfp_mask */ |
1da177e4 | 948 | |
7fb1d9fc | 949 | if (unlikely(*z == NULL)) { |
1da177e4 LT |
950 | /* Should this ever happen?? */ |
951 | return NULL; | |
952 | } | |
6b1de916 | 953 | |
7fb1d9fc | 954 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, |
3148890b | 955 | zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET); |
7fb1d9fc RS |
956 | if (page) |
957 | goto got_pg; | |
1da177e4 | 958 | |
6b1de916 | 959 | do { |
43b0bc00 | 960 | wakeup_kswapd(*z, order); |
6b1de916 | 961 | } while (*(++z)); |
1da177e4 | 962 | |
9bf2229f | 963 | /* |
7fb1d9fc RS |
964 | * OK, we're below the kswapd watermark and have kicked background |
965 | * reclaim. Now things get more complex, so set up alloc_flags according | |
966 | * to how we want to proceed. | |
967 | * | |
968 | * The caller may dip into page reserves a bit more if the caller | |
969 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
4eac915d PJ |
970 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will |
971 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). | |
9bf2229f | 972 | */ |
3148890b | 973 | alloc_flags = ALLOC_WMARK_MIN; |
7fb1d9fc RS |
974 | if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait) |
975 | alloc_flags |= ALLOC_HARDER; | |
976 | if (gfp_mask & __GFP_HIGH) | |
977 | alloc_flags |= ALLOC_HIGH; | |
bdd804f4 PJ |
978 | if (wait) |
979 | alloc_flags |= ALLOC_CPUSET; | |
1da177e4 LT |
980 | |
981 | /* | |
982 | * Go through the zonelist again. Let __GFP_HIGH and allocations | |
7fb1d9fc | 983 | * coming from realtime tasks go deeper into reserves. |
1da177e4 LT |
984 | * |
985 | * This is the last chance, in general, before the goto nopage. | |
986 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. | |
9bf2229f | 987 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1da177e4 | 988 | */ |
7fb1d9fc RS |
989 | page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags); |
990 | if (page) | |
991 | goto got_pg; | |
1da177e4 LT |
992 | |
993 | /* This allocation should allow future memory freeing. */ | |
b84a35be NP |
994 | |
995 | if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE))) | |
996 | && !in_interrupt()) { | |
997 | if (!(gfp_mask & __GFP_NOMEMALLOC)) { | |
885036d3 | 998 | nofail_alloc: |
b84a35be | 999 | /* go through the zonelist yet again, ignoring mins */ |
7fb1d9fc | 1000 | page = get_page_from_freelist(gfp_mask, order, |
47f3a867 | 1001 | zonelist, ALLOC_NO_WATERMARKS); |
7fb1d9fc RS |
1002 | if (page) |
1003 | goto got_pg; | |
885036d3 KK |
1004 | if (gfp_mask & __GFP_NOFAIL) { |
1005 | blk_congestion_wait(WRITE, HZ/50); | |
1006 | goto nofail_alloc; | |
1007 | } | |
1da177e4 LT |
1008 | } |
1009 | goto nopage; | |
1010 | } | |
1011 | ||
1012 | /* Atomic allocations - we can't balance anything */ | |
1013 | if (!wait) | |
1014 | goto nopage; | |
1015 | ||
1016 | rebalance: | |
1017 | cond_resched(); | |
1018 | ||
1019 | /* We now go into synchronous reclaim */ | |
3e0d98b9 | 1020 | cpuset_memory_pressure_bump(); |
1da177e4 LT |
1021 | p->flags |= PF_MEMALLOC; |
1022 | reclaim_state.reclaimed_slab = 0; | |
1023 | p->reclaim_state = &reclaim_state; | |
1024 | ||
7fb1d9fc | 1025 | did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask); |
1da177e4 LT |
1026 | |
1027 | p->reclaim_state = NULL; | |
1028 | p->flags &= ~PF_MEMALLOC; | |
1029 | ||
1030 | cond_resched(); | |
1031 | ||
1032 | if (likely(did_some_progress)) { | |
7fb1d9fc RS |
1033 | page = get_page_from_freelist(gfp_mask, order, |
1034 | zonelist, alloc_flags); | |
1035 | if (page) | |
1036 | goto got_pg; | |
1da177e4 LT |
1037 | } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) { |
1038 | /* | |
1039 | * Go through the zonelist yet one more time, keep | |
1040 | * very high watermark here, this is only to catch | |
1041 | * a parallel oom killing, we must fail if we're still | |
1042 | * under heavy pressure. | |
1043 | */ | |
7fb1d9fc | 1044 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, |
3148890b | 1045 | zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET); |
7fb1d9fc RS |
1046 | if (page) |
1047 | goto got_pg; | |
1da177e4 | 1048 | |
9b0f8b04 | 1049 | out_of_memory(zonelist, gfp_mask, order); |
1da177e4 LT |
1050 | goto restart; |
1051 | } | |
1052 | ||
1053 | /* | |
1054 | * Don't let big-order allocations loop unless the caller explicitly | |
1055 | * requests that. Wait for some write requests to complete then retry. | |
1056 | * | |
1057 | * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order | |
1058 | * <= 3, but that may not be true in other implementations. | |
1059 | */ | |
1060 | do_retry = 0; | |
1061 | if (!(gfp_mask & __GFP_NORETRY)) { | |
1062 | if ((order <= 3) || (gfp_mask & __GFP_REPEAT)) | |
1063 | do_retry = 1; | |
1064 | if (gfp_mask & __GFP_NOFAIL) | |
1065 | do_retry = 1; | |
1066 | } | |
1067 | if (do_retry) { | |
1068 | blk_congestion_wait(WRITE, HZ/50); | |
1069 | goto rebalance; | |
1070 | } | |
1071 | ||
1072 | nopage: | |
1073 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) { | |
1074 | printk(KERN_WARNING "%s: page allocation failure." | |
1075 | " order:%d, mode:0x%x\n", | |
1076 | p->comm, order, gfp_mask); | |
1077 | dump_stack(); | |
578c2fd6 | 1078 | show_mem(); |
1da177e4 | 1079 | } |
1da177e4 | 1080 | got_pg: |
1da177e4 LT |
1081 | return page; |
1082 | } | |
1083 | ||
1084 | EXPORT_SYMBOL(__alloc_pages); | |
1085 | ||
1086 | /* | |
1087 | * Common helper functions. | |
1088 | */ | |
dd0fc66f | 1089 | fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 LT |
1090 | { |
1091 | struct page * page; | |
1092 | page = alloc_pages(gfp_mask, order); | |
1093 | if (!page) | |
1094 | return 0; | |
1095 | return (unsigned long) page_address(page); | |
1096 | } | |
1097 | ||
1098 | EXPORT_SYMBOL(__get_free_pages); | |
1099 | ||
dd0fc66f | 1100 | fastcall unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 LT |
1101 | { |
1102 | struct page * page; | |
1103 | ||
1104 | /* | |
1105 | * get_zeroed_page() returns a 32-bit address, which cannot represent | |
1106 | * a highmem page | |
1107 | */ | |
260b2367 | 1108 | BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); |
1da177e4 LT |
1109 | |
1110 | page = alloc_pages(gfp_mask | __GFP_ZERO, 0); | |
1111 | if (page) | |
1112 | return (unsigned long) page_address(page); | |
1113 | return 0; | |
1114 | } | |
1115 | ||
1116 | EXPORT_SYMBOL(get_zeroed_page); | |
1117 | ||
1118 | void __pagevec_free(struct pagevec *pvec) | |
1119 | { | |
1120 | int i = pagevec_count(pvec); | |
1121 | ||
1122 | while (--i >= 0) | |
1123 | free_hot_cold_page(pvec->pages[i], pvec->cold); | |
1124 | } | |
1125 | ||
1126 | fastcall void __free_pages(struct page *page, unsigned int order) | |
1127 | { | |
b5810039 | 1128 | if (put_page_testzero(page)) { |
1da177e4 LT |
1129 | if (order == 0) |
1130 | free_hot_page(page); | |
1131 | else | |
1132 | __free_pages_ok(page, order); | |
1133 | } | |
1134 | } | |
1135 | ||
1136 | EXPORT_SYMBOL(__free_pages); | |
1137 | ||
1138 | fastcall void free_pages(unsigned long addr, unsigned int order) | |
1139 | { | |
1140 | if (addr != 0) { | |
1141 | BUG_ON(!virt_addr_valid((void *)addr)); | |
1142 | __free_pages(virt_to_page((void *)addr), order); | |
1143 | } | |
1144 | } | |
1145 | ||
1146 | EXPORT_SYMBOL(free_pages); | |
1147 | ||
1148 | /* | |
1149 | * Total amount of free (allocatable) RAM: | |
1150 | */ | |
1151 | unsigned int nr_free_pages(void) | |
1152 | { | |
1153 | unsigned int sum = 0; | |
1154 | struct zone *zone; | |
1155 | ||
1156 | for_each_zone(zone) | |
1157 | sum += zone->free_pages; | |
1158 | ||
1159 | return sum; | |
1160 | } | |
1161 | ||
1162 | EXPORT_SYMBOL(nr_free_pages); | |
1163 | ||
1164 | #ifdef CONFIG_NUMA | |
1165 | unsigned int nr_free_pages_pgdat(pg_data_t *pgdat) | |
1166 | { | |
1167 | unsigned int i, sum = 0; | |
1168 | ||
1169 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1170 | sum += pgdat->node_zones[i].free_pages; | |
1171 | ||
1172 | return sum; | |
1173 | } | |
1174 | #endif | |
1175 | ||
1176 | static unsigned int nr_free_zone_pages(int offset) | |
1177 | { | |
e310fd43 MB |
1178 | /* Just pick one node, since fallback list is circular */ |
1179 | pg_data_t *pgdat = NODE_DATA(numa_node_id()); | |
1da177e4 LT |
1180 | unsigned int sum = 0; |
1181 | ||
e310fd43 MB |
1182 | struct zonelist *zonelist = pgdat->node_zonelists + offset; |
1183 | struct zone **zonep = zonelist->zones; | |
1184 | struct zone *zone; | |
1da177e4 | 1185 | |
e310fd43 MB |
1186 | for (zone = *zonep++; zone; zone = *zonep++) { |
1187 | unsigned long size = zone->present_pages; | |
1188 | unsigned long high = zone->pages_high; | |
1189 | if (size > high) | |
1190 | sum += size - high; | |
1da177e4 LT |
1191 | } |
1192 | ||
1193 | return sum; | |
1194 | } | |
1195 | ||
1196 | /* | |
1197 | * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL | |
1198 | */ | |
1199 | unsigned int nr_free_buffer_pages(void) | |
1200 | { | |
af4ca457 | 1201 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 LT |
1202 | } |
1203 | ||
1204 | /* | |
1205 | * Amount of free RAM allocatable within all zones | |
1206 | */ | |
1207 | unsigned int nr_free_pagecache_pages(void) | |
1208 | { | |
af4ca457 | 1209 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER)); |
1da177e4 LT |
1210 | } |
1211 | ||
1212 | #ifdef CONFIG_HIGHMEM | |
1213 | unsigned int nr_free_highpages (void) | |
1214 | { | |
1215 | pg_data_t *pgdat; | |
1216 | unsigned int pages = 0; | |
1217 | ||
ec936fc5 | 1218 | for_each_online_pgdat(pgdat) |
1da177e4 LT |
1219 | pages += pgdat->node_zones[ZONE_HIGHMEM].free_pages; |
1220 | ||
1221 | return pages; | |
1222 | } | |
1223 | #endif | |
1224 | ||
1225 | #ifdef CONFIG_NUMA | |
1226 | static void show_node(struct zone *zone) | |
1227 | { | |
1228 | printk("Node %d ", zone->zone_pgdat->node_id); | |
1229 | } | |
1230 | #else | |
1231 | #define show_node(zone) do { } while (0) | |
1232 | #endif | |
1233 | ||
1234 | /* | |
1235 | * Accumulate the page_state information across all CPUs. | |
1236 | * The result is unavoidably approximate - it can change | |
1237 | * during and after execution of this function. | |
1238 | */ | |
1239 | static DEFINE_PER_CPU(struct page_state, page_states) = {0}; | |
1240 | ||
1241 | atomic_t nr_pagecache = ATOMIC_INIT(0); | |
1242 | EXPORT_SYMBOL(nr_pagecache); | |
1243 | #ifdef CONFIG_SMP | |
1244 | DEFINE_PER_CPU(long, nr_pagecache_local) = 0; | |
1245 | #endif | |
1246 | ||
a86b1f53 | 1247 | static void __get_page_state(struct page_state *ret, int nr, cpumask_t *cpumask) |
1da177e4 | 1248 | { |
b40607fc | 1249 | unsigned cpu; |
1da177e4 | 1250 | |
88a2a4ac | 1251 | memset(ret, 0, nr * sizeof(unsigned long)); |
84c2008a | 1252 | cpus_and(*cpumask, *cpumask, cpu_online_map); |
1da177e4 | 1253 | |
b40607fc AM |
1254 | for_each_cpu_mask(cpu, *cpumask) { |
1255 | unsigned long *in; | |
1256 | unsigned long *out; | |
1257 | unsigned off; | |
1258 | unsigned next_cpu; | |
88a2a4ac | 1259 | |
1da177e4 LT |
1260 | in = (unsigned long *)&per_cpu(page_states, cpu); |
1261 | ||
b40607fc AM |
1262 | next_cpu = next_cpu(cpu, *cpumask); |
1263 | if (likely(next_cpu < NR_CPUS)) | |
1264 | prefetch(&per_cpu(page_states, next_cpu)); | |
1da177e4 LT |
1265 | |
1266 | out = (unsigned long *)ret; | |
1267 | for (off = 0; off < nr; off++) | |
1268 | *out++ += *in++; | |
1269 | } | |
1270 | } | |
1271 | ||
c07e02db MH |
1272 | void get_page_state_node(struct page_state *ret, int node) |
1273 | { | |
1274 | int nr; | |
1275 | cpumask_t mask = node_to_cpumask(node); | |
1276 | ||
1277 | nr = offsetof(struct page_state, GET_PAGE_STATE_LAST); | |
1278 | nr /= sizeof(unsigned long); | |
1279 | ||
1280 | __get_page_state(ret, nr+1, &mask); | |
1281 | } | |
1282 | ||
1da177e4 LT |
1283 | void get_page_state(struct page_state *ret) |
1284 | { | |
1285 | int nr; | |
c07e02db | 1286 | cpumask_t mask = CPU_MASK_ALL; |
1da177e4 LT |
1287 | |
1288 | nr = offsetof(struct page_state, GET_PAGE_STATE_LAST); | |
1289 | nr /= sizeof(unsigned long); | |
1290 | ||
c07e02db | 1291 | __get_page_state(ret, nr + 1, &mask); |
1da177e4 LT |
1292 | } |
1293 | ||
1294 | void get_full_page_state(struct page_state *ret) | |
1295 | { | |
c07e02db MH |
1296 | cpumask_t mask = CPU_MASK_ALL; |
1297 | ||
1298 | __get_page_state(ret, sizeof(*ret) / sizeof(unsigned long), &mask); | |
1da177e4 LT |
1299 | } |
1300 | ||
a74609fa | 1301 | unsigned long read_page_state_offset(unsigned long offset) |
1da177e4 LT |
1302 | { |
1303 | unsigned long ret = 0; | |
1304 | int cpu; | |
1305 | ||
84c2008a | 1306 | for_each_online_cpu(cpu) { |
1da177e4 LT |
1307 | unsigned long in; |
1308 | ||
1309 | in = (unsigned long)&per_cpu(page_states, cpu) + offset; | |
1310 | ret += *((unsigned long *)in); | |
1311 | } | |
1312 | return ret; | |
1313 | } | |
1314 | ||
a74609fa NP |
1315 | void __mod_page_state_offset(unsigned long offset, unsigned long delta) |
1316 | { | |
1317 | void *ptr; | |
1318 | ||
1319 | ptr = &__get_cpu_var(page_states); | |
1320 | *(unsigned long *)(ptr + offset) += delta; | |
1321 | } | |
1322 | EXPORT_SYMBOL(__mod_page_state_offset); | |
1323 | ||
1324 | void mod_page_state_offset(unsigned long offset, unsigned long delta) | |
1da177e4 LT |
1325 | { |
1326 | unsigned long flags; | |
a74609fa | 1327 | void *ptr; |
1da177e4 LT |
1328 | |
1329 | local_irq_save(flags); | |
1330 | ptr = &__get_cpu_var(page_states); | |
a74609fa | 1331 | *(unsigned long *)(ptr + offset) += delta; |
1da177e4 LT |
1332 | local_irq_restore(flags); |
1333 | } | |
a74609fa | 1334 | EXPORT_SYMBOL(mod_page_state_offset); |
1da177e4 LT |
1335 | |
1336 | void __get_zone_counts(unsigned long *active, unsigned long *inactive, | |
1337 | unsigned long *free, struct pglist_data *pgdat) | |
1338 | { | |
1339 | struct zone *zones = pgdat->node_zones; | |
1340 | int i; | |
1341 | ||
1342 | *active = 0; | |
1343 | *inactive = 0; | |
1344 | *free = 0; | |
1345 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
1346 | *active += zones[i].nr_active; | |
1347 | *inactive += zones[i].nr_inactive; | |
1348 | *free += zones[i].free_pages; | |
1349 | } | |
1350 | } | |
1351 | ||
1352 | void get_zone_counts(unsigned long *active, | |
1353 | unsigned long *inactive, unsigned long *free) | |
1354 | { | |
1355 | struct pglist_data *pgdat; | |
1356 | ||
1357 | *active = 0; | |
1358 | *inactive = 0; | |
1359 | *free = 0; | |
ec936fc5 | 1360 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
1361 | unsigned long l, m, n; |
1362 | __get_zone_counts(&l, &m, &n, pgdat); | |
1363 | *active += l; | |
1364 | *inactive += m; | |
1365 | *free += n; | |
1366 | } | |
1367 | } | |
1368 | ||
1369 | void si_meminfo(struct sysinfo *val) | |
1370 | { | |
1371 | val->totalram = totalram_pages; | |
1372 | val->sharedram = 0; | |
1373 | val->freeram = nr_free_pages(); | |
1374 | val->bufferram = nr_blockdev_pages(); | |
1375 | #ifdef CONFIG_HIGHMEM | |
1376 | val->totalhigh = totalhigh_pages; | |
1377 | val->freehigh = nr_free_highpages(); | |
1378 | #else | |
1379 | val->totalhigh = 0; | |
1380 | val->freehigh = 0; | |
1381 | #endif | |
1382 | val->mem_unit = PAGE_SIZE; | |
1383 | } | |
1384 | ||
1385 | EXPORT_SYMBOL(si_meminfo); | |
1386 | ||
1387 | #ifdef CONFIG_NUMA | |
1388 | void si_meminfo_node(struct sysinfo *val, int nid) | |
1389 | { | |
1390 | pg_data_t *pgdat = NODE_DATA(nid); | |
1391 | ||
1392 | val->totalram = pgdat->node_present_pages; | |
1393 | val->freeram = nr_free_pages_pgdat(pgdat); | |
1394 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages; | |
1395 | val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages; | |
1396 | val->mem_unit = PAGE_SIZE; | |
1397 | } | |
1398 | #endif | |
1399 | ||
1400 | #define K(x) ((x) << (PAGE_SHIFT-10)) | |
1401 | ||
1402 | /* | |
1403 | * Show free area list (used inside shift_scroll-lock stuff) | |
1404 | * We also calculate the percentage fragmentation. We do this by counting the | |
1405 | * memory on each free list with the exception of the first item on the list. | |
1406 | */ | |
1407 | void show_free_areas(void) | |
1408 | { | |
1409 | struct page_state ps; | |
1410 | int cpu, temperature; | |
1411 | unsigned long active; | |
1412 | unsigned long inactive; | |
1413 | unsigned long free; | |
1414 | struct zone *zone; | |
1415 | ||
1416 | for_each_zone(zone) { | |
1417 | show_node(zone); | |
1418 | printk("%s per-cpu:", zone->name); | |
1419 | ||
f3fe6512 | 1420 | if (!populated_zone(zone)) { |
1da177e4 LT |
1421 | printk(" empty\n"); |
1422 | continue; | |
1423 | } else | |
1424 | printk("\n"); | |
1425 | ||
6b482c67 | 1426 | for_each_online_cpu(cpu) { |
1da177e4 LT |
1427 | struct per_cpu_pageset *pageset; |
1428 | ||
e7c8d5c9 | 1429 | pageset = zone_pcp(zone, cpu); |
1da177e4 LT |
1430 | |
1431 | for (temperature = 0; temperature < 2; temperature++) | |
2d92c5c9 | 1432 | printk("cpu %d %s: high %d, batch %d used:%d\n", |
1da177e4 LT |
1433 | cpu, |
1434 | temperature ? "cold" : "hot", | |
1da177e4 | 1435 | pageset->pcp[temperature].high, |
4ae7c039 CL |
1436 | pageset->pcp[temperature].batch, |
1437 | pageset->pcp[temperature].count); | |
1da177e4 LT |
1438 | } |
1439 | } | |
1440 | ||
1441 | get_page_state(&ps); | |
1442 | get_zone_counts(&active, &inactive, &free); | |
1443 | ||
c0d62219 | 1444 | printk("Free pages: %11ukB (%ukB HighMem)\n", |
1da177e4 LT |
1445 | K(nr_free_pages()), |
1446 | K(nr_free_highpages())); | |
1447 | ||
1448 | printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu " | |
1449 | "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n", | |
1450 | active, | |
1451 | inactive, | |
1452 | ps.nr_dirty, | |
1453 | ps.nr_writeback, | |
1454 | ps.nr_unstable, | |
1455 | nr_free_pages(), | |
1456 | ps.nr_slab, | |
1457 | ps.nr_mapped, | |
1458 | ps.nr_page_table_pages); | |
1459 | ||
1460 | for_each_zone(zone) { | |
1461 | int i; | |
1462 | ||
1463 | show_node(zone); | |
1464 | printk("%s" | |
1465 | " free:%lukB" | |
1466 | " min:%lukB" | |
1467 | " low:%lukB" | |
1468 | " high:%lukB" | |
1469 | " active:%lukB" | |
1470 | " inactive:%lukB" | |
1471 | " present:%lukB" | |
1472 | " pages_scanned:%lu" | |
1473 | " all_unreclaimable? %s" | |
1474 | "\n", | |
1475 | zone->name, | |
1476 | K(zone->free_pages), | |
1477 | K(zone->pages_min), | |
1478 | K(zone->pages_low), | |
1479 | K(zone->pages_high), | |
1480 | K(zone->nr_active), | |
1481 | K(zone->nr_inactive), | |
1482 | K(zone->present_pages), | |
1483 | zone->pages_scanned, | |
1484 | (zone->all_unreclaimable ? "yes" : "no") | |
1485 | ); | |
1486 | printk("lowmem_reserve[]:"); | |
1487 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1488 | printk(" %lu", zone->lowmem_reserve[i]); | |
1489 | printk("\n"); | |
1490 | } | |
1491 | ||
1492 | for_each_zone(zone) { | |
8f9de51a | 1493 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
1da177e4 LT |
1494 | |
1495 | show_node(zone); | |
1496 | printk("%s: ", zone->name); | |
f3fe6512 | 1497 | if (!populated_zone(zone)) { |
1da177e4 LT |
1498 | printk("empty\n"); |
1499 | continue; | |
1500 | } | |
1501 | ||
1502 | spin_lock_irqsave(&zone->lock, flags); | |
1503 | for (order = 0; order < MAX_ORDER; order++) { | |
8f9de51a KK |
1504 | nr[order] = zone->free_area[order].nr_free; |
1505 | total += nr[order] << order; | |
1da177e4 LT |
1506 | } |
1507 | spin_unlock_irqrestore(&zone->lock, flags); | |
8f9de51a KK |
1508 | for (order = 0; order < MAX_ORDER; order++) |
1509 | printk("%lu*%lukB ", nr[order], K(1UL) << order); | |
1da177e4 LT |
1510 | printk("= %lukB\n", K(total)); |
1511 | } | |
1512 | ||
1513 | show_swap_cache_info(); | |
1514 | } | |
1515 | ||
1516 | /* | |
1517 | * Builds allocation fallback zone lists. | |
1a93205b CL |
1518 | * |
1519 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 1520 | */ |
86356ab1 | 1521 | static int __meminit build_zonelists_node(pg_data_t *pgdat, |
070f8032 | 1522 | struct zonelist *zonelist, int nr_zones, int zone_type) |
1da177e4 | 1523 | { |
1a93205b CL |
1524 | struct zone *zone; |
1525 | ||
070f8032 | 1526 | BUG_ON(zone_type > ZONE_HIGHMEM); |
02a68a5e CL |
1527 | |
1528 | do { | |
070f8032 | 1529 | zone = pgdat->node_zones + zone_type; |
1a93205b | 1530 | if (populated_zone(zone)) { |
1da177e4 | 1531 | #ifndef CONFIG_HIGHMEM |
070f8032 | 1532 | BUG_ON(zone_type > ZONE_NORMAL); |
1da177e4 | 1533 | #endif |
070f8032 CL |
1534 | zonelist->zones[nr_zones++] = zone; |
1535 | check_highest_zone(zone_type); | |
1da177e4 | 1536 | } |
070f8032 | 1537 | zone_type--; |
02a68a5e | 1538 | |
070f8032 CL |
1539 | } while (zone_type >= 0); |
1540 | return nr_zones; | |
1da177e4 LT |
1541 | } |
1542 | ||
260b2367 AV |
1543 | static inline int highest_zone(int zone_bits) |
1544 | { | |
1545 | int res = ZONE_NORMAL; | |
1546 | if (zone_bits & (__force int)__GFP_HIGHMEM) | |
1547 | res = ZONE_HIGHMEM; | |
a2f1b424 AK |
1548 | if (zone_bits & (__force int)__GFP_DMA32) |
1549 | res = ZONE_DMA32; | |
260b2367 AV |
1550 | if (zone_bits & (__force int)__GFP_DMA) |
1551 | res = ZONE_DMA; | |
1552 | return res; | |
1553 | } | |
1554 | ||
1da177e4 LT |
1555 | #ifdef CONFIG_NUMA |
1556 | #define MAX_NODE_LOAD (num_online_nodes()) | |
86356ab1 | 1557 | static int __meminitdata node_load[MAX_NUMNODES]; |
1da177e4 | 1558 | /** |
4dc3b16b | 1559 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
1560 | * @node: node whose fallback list we're appending |
1561 | * @used_node_mask: nodemask_t of already used nodes | |
1562 | * | |
1563 | * We use a number of factors to determine which is the next node that should | |
1564 | * appear on a given node's fallback list. The node should not have appeared | |
1565 | * already in @node's fallback list, and it should be the next closest node | |
1566 | * according to the distance array (which contains arbitrary distance values | |
1567 | * from each node to each node in the system), and should also prefer nodes | |
1568 | * with no CPUs, since presumably they'll have very little allocation pressure | |
1569 | * on them otherwise. | |
1570 | * It returns -1 if no node is found. | |
1571 | */ | |
86356ab1 | 1572 | static int __meminit find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 1573 | { |
4cf808eb | 1574 | int n, val; |
1da177e4 LT |
1575 | int min_val = INT_MAX; |
1576 | int best_node = -1; | |
1577 | ||
4cf808eb LT |
1578 | /* Use the local node if we haven't already */ |
1579 | if (!node_isset(node, *used_node_mask)) { | |
1580 | node_set(node, *used_node_mask); | |
1581 | return node; | |
1582 | } | |
1da177e4 | 1583 | |
4cf808eb LT |
1584 | for_each_online_node(n) { |
1585 | cpumask_t tmp; | |
1da177e4 LT |
1586 | |
1587 | /* Don't want a node to appear more than once */ | |
1588 | if (node_isset(n, *used_node_mask)) | |
1589 | continue; | |
1590 | ||
1da177e4 LT |
1591 | /* Use the distance array to find the distance */ |
1592 | val = node_distance(node, n); | |
1593 | ||
4cf808eb LT |
1594 | /* Penalize nodes under us ("prefer the next node") */ |
1595 | val += (n < node); | |
1596 | ||
1da177e4 LT |
1597 | /* Give preference to headless and unused nodes */ |
1598 | tmp = node_to_cpumask(n); | |
1599 | if (!cpus_empty(tmp)) | |
1600 | val += PENALTY_FOR_NODE_WITH_CPUS; | |
1601 | ||
1602 | /* Slight preference for less loaded node */ | |
1603 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
1604 | val += node_load[n]; | |
1605 | ||
1606 | if (val < min_val) { | |
1607 | min_val = val; | |
1608 | best_node = n; | |
1609 | } | |
1610 | } | |
1611 | ||
1612 | if (best_node >= 0) | |
1613 | node_set(best_node, *used_node_mask); | |
1614 | ||
1615 | return best_node; | |
1616 | } | |
1617 | ||
86356ab1 | 1618 | static void __meminit build_zonelists(pg_data_t *pgdat) |
1da177e4 LT |
1619 | { |
1620 | int i, j, k, node, local_node; | |
1621 | int prev_node, load; | |
1622 | struct zonelist *zonelist; | |
1623 | nodemask_t used_mask; | |
1624 | ||
1625 | /* initialize zonelists */ | |
1626 | for (i = 0; i < GFP_ZONETYPES; i++) { | |
1627 | zonelist = pgdat->node_zonelists + i; | |
1628 | zonelist->zones[0] = NULL; | |
1629 | } | |
1630 | ||
1631 | /* NUMA-aware ordering of nodes */ | |
1632 | local_node = pgdat->node_id; | |
1633 | load = num_online_nodes(); | |
1634 | prev_node = local_node; | |
1635 | nodes_clear(used_mask); | |
1636 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { | |
9eeff239 CL |
1637 | int distance = node_distance(local_node, node); |
1638 | ||
1639 | /* | |
1640 | * If another node is sufficiently far away then it is better | |
1641 | * to reclaim pages in a zone before going off node. | |
1642 | */ | |
1643 | if (distance > RECLAIM_DISTANCE) | |
1644 | zone_reclaim_mode = 1; | |
1645 | ||
1da177e4 LT |
1646 | /* |
1647 | * We don't want to pressure a particular node. | |
1648 | * So adding penalty to the first node in same | |
1649 | * distance group to make it round-robin. | |
1650 | */ | |
9eeff239 CL |
1651 | |
1652 | if (distance != node_distance(local_node, prev_node)) | |
1da177e4 LT |
1653 | node_load[node] += load; |
1654 | prev_node = node; | |
1655 | load--; | |
1656 | for (i = 0; i < GFP_ZONETYPES; i++) { | |
1657 | zonelist = pgdat->node_zonelists + i; | |
1658 | for (j = 0; zonelist->zones[j] != NULL; j++); | |
1659 | ||
260b2367 | 1660 | k = highest_zone(i); |
1da177e4 LT |
1661 | |
1662 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, k); | |
1663 | zonelist->zones[j] = NULL; | |
1664 | } | |
1665 | } | |
1666 | } | |
1667 | ||
1668 | #else /* CONFIG_NUMA */ | |
1669 | ||
86356ab1 | 1670 | static void __meminit build_zonelists(pg_data_t *pgdat) |
1da177e4 LT |
1671 | { |
1672 | int i, j, k, node, local_node; | |
1673 | ||
1674 | local_node = pgdat->node_id; | |
1675 | for (i = 0; i < GFP_ZONETYPES; i++) { | |
1676 | struct zonelist *zonelist; | |
1677 | ||
1678 | zonelist = pgdat->node_zonelists + i; | |
1679 | ||
1680 | j = 0; | |
260b2367 | 1681 | k = highest_zone(i); |
1da177e4 LT |
1682 | j = build_zonelists_node(pgdat, zonelist, j, k); |
1683 | /* | |
1684 | * Now we build the zonelist so that it contains the zones | |
1685 | * of all the other nodes. | |
1686 | * We don't want to pressure a particular node, so when | |
1687 | * building the zones for node N, we make sure that the | |
1688 | * zones coming right after the local ones are those from | |
1689 | * node N+1 (modulo N) | |
1690 | */ | |
1691 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
1692 | if (!node_online(node)) | |
1693 | continue; | |
1694 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, k); | |
1695 | } | |
1696 | for (node = 0; node < local_node; node++) { | |
1697 | if (!node_online(node)) | |
1698 | continue; | |
1699 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, k); | |
1700 | } | |
1701 | ||
1702 | zonelist->zones[j] = NULL; | |
1703 | } | |
1704 | } | |
1705 | ||
1706 | #endif /* CONFIG_NUMA */ | |
1707 | ||
6811378e YG |
1708 | /* return values int ....just for stop_machine_run() */ |
1709 | static int __meminit __build_all_zonelists(void *dummy) | |
1da177e4 | 1710 | { |
6811378e YG |
1711 | int nid; |
1712 | for_each_online_node(nid) | |
1713 | build_zonelists(NODE_DATA(nid)); | |
1714 | return 0; | |
1715 | } | |
1716 | ||
1717 | void __meminit build_all_zonelists(void) | |
1718 | { | |
1719 | if (system_state == SYSTEM_BOOTING) { | |
1720 | __build_all_zonelists(0); | |
1721 | cpuset_init_current_mems_allowed(); | |
1722 | } else { | |
1723 | /* we have to stop all cpus to guaranntee there is no user | |
1724 | of zonelist */ | |
1725 | stop_machine_run(__build_all_zonelists, NULL, NR_CPUS); | |
1726 | /* cpuset refresh routine should be here */ | |
1727 | } | |
bd1e22b8 AM |
1728 | vm_total_pages = nr_free_pagecache_pages(); |
1729 | printk("Built %i zonelists. Total pages: %ld\n", | |
1730 | num_online_nodes(), vm_total_pages); | |
1da177e4 LT |
1731 | } |
1732 | ||
1733 | /* | |
1734 | * Helper functions to size the waitqueue hash table. | |
1735 | * Essentially these want to choose hash table sizes sufficiently | |
1736 | * large so that collisions trying to wait on pages are rare. | |
1737 | * But in fact, the number of active page waitqueues on typical | |
1738 | * systems is ridiculously low, less than 200. So this is even | |
1739 | * conservative, even though it seems large. | |
1740 | * | |
1741 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
1742 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
1743 | */ | |
1744 | #define PAGES_PER_WAITQUEUE 256 | |
1745 | ||
cca448fe | 1746 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 1747 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
1748 | { |
1749 | unsigned long size = 1; | |
1750 | ||
1751 | pages /= PAGES_PER_WAITQUEUE; | |
1752 | ||
1753 | while (size < pages) | |
1754 | size <<= 1; | |
1755 | ||
1756 | /* | |
1757 | * Once we have dozens or even hundreds of threads sleeping | |
1758 | * on IO we've got bigger problems than wait queue collision. | |
1759 | * Limit the size of the wait table to a reasonable size. | |
1760 | */ | |
1761 | size = min(size, 4096UL); | |
1762 | ||
1763 | return max(size, 4UL); | |
1764 | } | |
cca448fe YG |
1765 | #else |
1766 | /* | |
1767 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
1768 | * a suitable size for its wait_table. So we use the maximum size now. | |
1769 | * | |
1770 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
1771 | * | |
1772 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
1773 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
1774 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
1775 | * | |
1776 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
1777 | * or more by the traditional way. (See above). It equals: | |
1778 | * | |
1779 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
1780 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
1781 | * powerpc (64K page size) : = (32G +16M)byte. | |
1782 | */ | |
1783 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
1784 | { | |
1785 | return 4096UL; | |
1786 | } | |
1787 | #endif | |
1da177e4 LT |
1788 | |
1789 | /* | |
1790 | * This is an integer logarithm so that shifts can be used later | |
1791 | * to extract the more random high bits from the multiplicative | |
1792 | * hash function before the remainder is taken. | |
1793 | */ | |
1794 | static inline unsigned long wait_table_bits(unsigned long size) | |
1795 | { | |
1796 | return ffz(~size); | |
1797 | } | |
1798 | ||
1799 | #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1)) | |
1800 | ||
1801 | static void __init calculate_zone_totalpages(struct pglist_data *pgdat, | |
1802 | unsigned long *zones_size, unsigned long *zholes_size) | |
1803 | { | |
1804 | unsigned long realtotalpages, totalpages = 0; | |
1805 | int i; | |
1806 | ||
1807 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1808 | totalpages += zones_size[i]; | |
1809 | pgdat->node_spanned_pages = totalpages; | |
1810 | ||
1811 | realtotalpages = totalpages; | |
1812 | if (zholes_size) | |
1813 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1814 | realtotalpages -= zholes_size[i]; | |
1815 | pgdat->node_present_pages = realtotalpages; | |
1816 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages); | |
1817 | } | |
1818 | ||
1819 | ||
1820 | /* | |
1821 | * Initially all pages are reserved - free ones are freed | |
1822 | * up by free_all_bootmem() once the early boot process is | |
1823 | * done. Non-atomic initialization, single-pass. | |
1824 | */ | |
c09b4240 | 1825 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
1da177e4 LT |
1826 | unsigned long start_pfn) |
1827 | { | |
1da177e4 | 1828 | struct page *page; |
29751f69 AW |
1829 | unsigned long end_pfn = start_pfn + size; |
1830 | unsigned long pfn; | |
1da177e4 | 1831 | |
cbe8dd4a | 1832 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
d41dee36 AW |
1833 | if (!early_pfn_valid(pfn)) |
1834 | continue; | |
1835 | page = pfn_to_page(pfn); | |
1836 | set_page_links(page, zone, nid, pfn); | |
7835e98b | 1837 | init_page_count(page); |
1da177e4 LT |
1838 | reset_page_mapcount(page); |
1839 | SetPageReserved(page); | |
1840 | INIT_LIST_HEAD(&page->lru); | |
1841 | #ifdef WANT_PAGE_VIRTUAL | |
1842 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
1843 | if (!is_highmem_idx(zone)) | |
3212c6be | 1844 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1da177e4 | 1845 | #endif |
1da177e4 LT |
1846 | } |
1847 | } | |
1848 | ||
1849 | void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone, | |
1850 | unsigned long size) | |
1851 | { | |
1852 | int order; | |
1853 | for (order = 0; order < MAX_ORDER ; order++) { | |
1854 | INIT_LIST_HEAD(&zone->free_area[order].free_list); | |
1855 | zone->free_area[order].nr_free = 0; | |
1856 | } | |
1857 | } | |
1858 | ||
d41dee36 AW |
1859 | #define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr) |
1860 | void zonetable_add(struct zone *zone, int nid, int zid, unsigned long pfn, | |
1861 | unsigned long size) | |
1862 | { | |
1863 | unsigned long snum = pfn_to_section_nr(pfn); | |
1864 | unsigned long end = pfn_to_section_nr(pfn + size); | |
1865 | ||
1866 | if (FLAGS_HAS_NODE) | |
1867 | zone_table[ZONETABLE_INDEX(nid, zid)] = zone; | |
1868 | else | |
1869 | for (; snum <= end; snum++) | |
1870 | zone_table[ZONETABLE_INDEX(snum, zid)] = zone; | |
1871 | } | |
1872 | ||
1da177e4 LT |
1873 | #ifndef __HAVE_ARCH_MEMMAP_INIT |
1874 | #define memmap_init(size, nid, zone, start_pfn) \ | |
1875 | memmap_init_zone((size), (nid), (zone), (start_pfn)) | |
1876 | #endif | |
1877 | ||
6292d9aa | 1878 | static int __cpuinit zone_batchsize(struct zone *zone) |
e7c8d5c9 CL |
1879 | { |
1880 | int batch; | |
1881 | ||
1882 | /* | |
1883 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 1884 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
1885 | * |
1886 | * OK, so we don't know how big the cache is. So guess. | |
1887 | */ | |
1888 | batch = zone->present_pages / 1024; | |
ba56e91c SR |
1889 | if (batch * PAGE_SIZE > 512 * 1024) |
1890 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
1891 | batch /= 4; /* We effectively *= 4 below */ |
1892 | if (batch < 1) | |
1893 | batch = 1; | |
1894 | ||
1895 | /* | |
0ceaacc9 NP |
1896 | * Clamp the batch to a 2^n - 1 value. Having a power |
1897 | * of 2 value was found to be more likely to have | |
1898 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 1899 | * |
0ceaacc9 NP |
1900 | * For example if 2 tasks are alternately allocating |
1901 | * batches of pages, one task can end up with a lot | |
1902 | * of pages of one half of the possible page colors | |
1903 | * and the other with pages of the other colors. | |
e7c8d5c9 | 1904 | */ |
0ceaacc9 | 1905 | batch = (1 << (fls(batch + batch/2)-1)) - 1; |
ba56e91c | 1906 | |
e7c8d5c9 CL |
1907 | return batch; |
1908 | } | |
1909 | ||
2caaad41 CL |
1910 | inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
1911 | { | |
1912 | struct per_cpu_pages *pcp; | |
1913 | ||
1c6fe946 MD |
1914 | memset(p, 0, sizeof(*p)); |
1915 | ||
2caaad41 CL |
1916 | pcp = &p->pcp[0]; /* hot */ |
1917 | pcp->count = 0; | |
2caaad41 CL |
1918 | pcp->high = 6 * batch; |
1919 | pcp->batch = max(1UL, 1 * batch); | |
1920 | INIT_LIST_HEAD(&pcp->list); | |
1921 | ||
1922 | pcp = &p->pcp[1]; /* cold*/ | |
1923 | pcp->count = 0; | |
2caaad41 | 1924 | pcp->high = 2 * batch; |
e46a5e28 | 1925 | pcp->batch = max(1UL, batch/2); |
2caaad41 CL |
1926 | INIT_LIST_HEAD(&pcp->list); |
1927 | } | |
1928 | ||
8ad4b1fb RS |
1929 | /* |
1930 | * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist | |
1931 | * to the value high for the pageset p. | |
1932 | */ | |
1933 | ||
1934 | static void setup_pagelist_highmark(struct per_cpu_pageset *p, | |
1935 | unsigned long high) | |
1936 | { | |
1937 | struct per_cpu_pages *pcp; | |
1938 | ||
1939 | pcp = &p->pcp[0]; /* hot list */ | |
1940 | pcp->high = high; | |
1941 | pcp->batch = max(1UL, high/4); | |
1942 | if ((high/4) > (PAGE_SHIFT * 8)) | |
1943 | pcp->batch = PAGE_SHIFT * 8; | |
1944 | } | |
1945 | ||
1946 | ||
e7c8d5c9 CL |
1947 | #ifdef CONFIG_NUMA |
1948 | /* | |
2caaad41 CL |
1949 | * Boot pageset table. One per cpu which is going to be used for all |
1950 | * zones and all nodes. The parameters will be set in such a way | |
1951 | * that an item put on a list will immediately be handed over to | |
1952 | * the buddy list. This is safe since pageset manipulation is done | |
1953 | * with interrupts disabled. | |
1954 | * | |
1955 | * Some NUMA counter updates may also be caught by the boot pagesets. | |
b7c84c6a CL |
1956 | * |
1957 | * The boot_pagesets must be kept even after bootup is complete for | |
1958 | * unused processors and/or zones. They do play a role for bootstrapping | |
1959 | * hotplugged processors. | |
1960 | * | |
1961 | * zoneinfo_show() and maybe other functions do | |
1962 | * not check if the processor is online before following the pageset pointer. | |
1963 | * Other parts of the kernel may not check if the zone is available. | |
2caaad41 | 1964 | */ |
88a2a4ac | 1965 | static struct per_cpu_pageset boot_pageset[NR_CPUS]; |
2caaad41 CL |
1966 | |
1967 | /* | |
1968 | * Dynamically allocate memory for the | |
e7c8d5c9 CL |
1969 | * per cpu pageset array in struct zone. |
1970 | */ | |
6292d9aa | 1971 | static int __cpuinit process_zones(int cpu) |
e7c8d5c9 CL |
1972 | { |
1973 | struct zone *zone, *dzone; | |
e7c8d5c9 CL |
1974 | |
1975 | for_each_zone(zone) { | |
e7c8d5c9 | 1976 | |
23316bc8 | 1977 | zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset), |
e7c8d5c9 | 1978 | GFP_KERNEL, cpu_to_node(cpu)); |
23316bc8 | 1979 | if (!zone_pcp(zone, cpu)) |
e7c8d5c9 | 1980 | goto bad; |
e7c8d5c9 | 1981 | |
23316bc8 | 1982 | setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone)); |
8ad4b1fb RS |
1983 | |
1984 | if (percpu_pagelist_fraction) | |
1985 | setup_pagelist_highmark(zone_pcp(zone, cpu), | |
1986 | (zone->present_pages / percpu_pagelist_fraction)); | |
e7c8d5c9 CL |
1987 | } |
1988 | ||
1989 | return 0; | |
1990 | bad: | |
1991 | for_each_zone(dzone) { | |
1992 | if (dzone == zone) | |
1993 | break; | |
23316bc8 NP |
1994 | kfree(zone_pcp(dzone, cpu)); |
1995 | zone_pcp(dzone, cpu) = NULL; | |
e7c8d5c9 CL |
1996 | } |
1997 | return -ENOMEM; | |
1998 | } | |
1999 | ||
2000 | static inline void free_zone_pagesets(int cpu) | |
2001 | { | |
e7c8d5c9 CL |
2002 | struct zone *zone; |
2003 | ||
2004 | for_each_zone(zone) { | |
2005 | struct per_cpu_pageset *pset = zone_pcp(zone, cpu); | |
2006 | ||
2007 | zone_pcp(zone, cpu) = NULL; | |
2008 | kfree(pset); | |
2009 | } | |
e7c8d5c9 CL |
2010 | } |
2011 | ||
9c7b216d | 2012 | static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb, |
e7c8d5c9 CL |
2013 | unsigned long action, |
2014 | void *hcpu) | |
2015 | { | |
2016 | int cpu = (long)hcpu; | |
2017 | int ret = NOTIFY_OK; | |
2018 | ||
2019 | switch (action) { | |
2020 | case CPU_UP_PREPARE: | |
2021 | if (process_zones(cpu)) | |
2022 | ret = NOTIFY_BAD; | |
2023 | break; | |
b0d41693 | 2024 | case CPU_UP_CANCELED: |
e7c8d5c9 CL |
2025 | case CPU_DEAD: |
2026 | free_zone_pagesets(cpu); | |
2027 | break; | |
e7c8d5c9 CL |
2028 | default: |
2029 | break; | |
2030 | } | |
2031 | return ret; | |
2032 | } | |
2033 | ||
74b85f37 | 2034 | static struct notifier_block __cpuinitdata pageset_notifier = |
e7c8d5c9 CL |
2035 | { &pageset_cpuup_callback, NULL, 0 }; |
2036 | ||
78d9955b | 2037 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 CL |
2038 | { |
2039 | int err; | |
2040 | ||
2041 | /* Initialize per_cpu_pageset for cpu 0. | |
2042 | * A cpuup callback will do this for every cpu | |
2043 | * as it comes online | |
2044 | */ | |
2045 | err = process_zones(smp_processor_id()); | |
2046 | BUG_ON(err); | |
2047 | register_cpu_notifier(&pageset_notifier); | |
2048 | } | |
2049 | ||
2050 | #endif | |
2051 | ||
c09b4240 | 2052 | static __meminit |
cca448fe | 2053 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
2054 | { |
2055 | int i; | |
2056 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe | 2057 | size_t alloc_size; |
ed8ece2e DH |
2058 | |
2059 | /* | |
2060 | * The per-page waitqueue mechanism uses hashed waitqueues | |
2061 | * per zone. | |
2062 | */ | |
02b694de YG |
2063 | zone->wait_table_hash_nr_entries = |
2064 | wait_table_hash_nr_entries(zone_size_pages); | |
2065 | zone->wait_table_bits = | |
2066 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
2067 | alloc_size = zone->wait_table_hash_nr_entries |
2068 | * sizeof(wait_queue_head_t); | |
2069 | ||
2070 | if (system_state == SYSTEM_BOOTING) { | |
2071 | zone->wait_table = (wait_queue_head_t *) | |
2072 | alloc_bootmem_node(pgdat, alloc_size); | |
2073 | } else { | |
2074 | /* | |
2075 | * This case means that a zone whose size was 0 gets new memory | |
2076 | * via memory hot-add. | |
2077 | * But it may be the case that a new node was hot-added. In | |
2078 | * this case vmalloc() will not be able to use this new node's | |
2079 | * memory - this wait_table must be initialized to use this new | |
2080 | * node itself as well. | |
2081 | * To use this new node's memory, further consideration will be | |
2082 | * necessary. | |
2083 | */ | |
2084 | zone->wait_table = (wait_queue_head_t *)vmalloc(alloc_size); | |
2085 | } | |
2086 | if (!zone->wait_table) | |
2087 | return -ENOMEM; | |
ed8ece2e | 2088 | |
02b694de | 2089 | for(i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 2090 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
2091 | |
2092 | return 0; | |
ed8ece2e DH |
2093 | } |
2094 | ||
c09b4240 | 2095 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e DH |
2096 | { |
2097 | int cpu; | |
2098 | unsigned long batch = zone_batchsize(zone); | |
2099 | ||
2100 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
2101 | #ifdef CONFIG_NUMA | |
2102 | /* Early boot. Slab allocator not functional yet */ | |
23316bc8 | 2103 | zone_pcp(zone, cpu) = &boot_pageset[cpu]; |
ed8ece2e DH |
2104 | setup_pageset(&boot_pageset[cpu],0); |
2105 | #else | |
2106 | setup_pageset(zone_pcp(zone,cpu), batch); | |
2107 | #endif | |
2108 | } | |
f5335c0f AB |
2109 | if (zone->present_pages) |
2110 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n", | |
2111 | zone->name, zone->present_pages, batch); | |
ed8ece2e DH |
2112 | } |
2113 | ||
718127cc YG |
2114 | __meminit int init_currently_empty_zone(struct zone *zone, |
2115 | unsigned long zone_start_pfn, | |
2116 | unsigned long size) | |
ed8ece2e DH |
2117 | { |
2118 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
2119 | int ret; |
2120 | ret = zone_wait_table_init(zone, size); | |
2121 | if (ret) | |
2122 | return ret; | |
ed8ece2e DH |
2123 | pgdat->nr_zones = zone_idx(zone) + 1; |
2124 | ||
ed8ece2e DH |
2125 | zone->zone_start_pfn = zone_start_pfn; |
2126 | ||
2127 | memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn); | |
2128 | ||
2129 | zone_init_free_lists(pgdat, zone, zone->spanned_pages); | |
718127cc YG |
2130 | |
2131 | return 0; | |
ed8ece2e DH |
2132 | } |
2133 | ||
1da177e4 LT |
2134 | /* |
2135 | * Set up the zone data structures: | |
2136 | * - mark all pages reserved | |
2137 | * - mark all memory queues empty | |
2138 | * - clear the memory bitmaps | |
2139 | */ | |
86356ab1 | 2140 | static void __meminit free_area_init_core(struct pglist_data *pgdat, |
1da177e4 LT |
2141 | unsigned long *zones_size, unsigned long *zholes_size) |
2142 | { | |
ed8ece2e DH |
2143 | unsigned long j; |
2144 | int nid = pgdat->node_id; | |
1da177e4 | 2145 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
718127cc | 2146 | int ret; |
1da177e4 | 2147 | |
208d54e5 | 2148 | pgdat_resize_init(pgdat); |
1da177e4 LT |
2149 | pgdat->nr_zones = 0; |
2150 | init_waitqueue_head(&pgdat->kswapd_wait); | |
2151 | pgdat->kswapd_max_order = 0; | |
2152 | ||
2153 | for (j = 0; j < MAX_NR_ZONES; j++) { | |
2154 | struct zone *zone = pgdat->node_zones + j; | |
2155 | unsigned long size, realsize; | |
1da177e4 | 2156 | |
1da177e4 LT |
2157 | realsize = size = zones_size[j]; |
2158 | if (zholes_size) | |
2159 | realsize -= zholes_size[j]; | |
2160 | ||
a2f1b424 | 2161 | if (j < ZONE_HIGHMEM) |
1da177e4 LT |
2162 | nr_kernel_pages += realsize; |
2163 | nr_all_pages += realsize; | |
2164 | ||
2165 | zone->spanned_pages = size; | |
2166 | zone->present_pages = realsize; | |
2167 | zone->name = zone_names[j]; | |
2168 | spin_lock_init(&zone->lock); | |
2169 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 2170 | zone_seqlock_init(zone); |
1da177e4 LT |
2171 | zone->zone_pgdat = pgdat; |
2172 | zone->free_pages = 0; | |
2173 | ||
2174 | zone->temp_priority = zone->prev_priority = DEF_PRIORITY; | |
2175 | ||
ed8ece2e | 2176 | zone_pcp_init(zone); |
1da177e4 LT |
2177 | INIT_LIST_HEAD(&zone->active_list); |
2178 | INIT_LIST_HEAD(&zone->inactive_list); | |
2179 | zone->nr_scan_active = 0; | |
2180 | zone->nr_scan_inactive = 0; | |
2181 | zone->nr_active = 0; | |
2182 | zone->nr_inactive = 0; | |
53e9a615 | 2183 | atomic_set(&zone->reclaim_in_progress, 0); |
1da177e4 LT |
2184 | if (!size) |
2185 | continue; | |
2186 | ||
d41dee36 | 2187 | zonetable_add(zone, nid, j, zone_start_pfn, size); |
718127cc YG |
2188 | ret = init_currently_empty_zone(zone, zone_start_pfn, size); |
2189 | BUG_ON(ret); | |
1da177e4 | 2190 | zone_start_pfn += size; |
1da177e4 LT |
2191 | } |
2192 | } | |
2193 | ||
2194 | static void __init alloc_node_mem_map(struct pglist_data *pgdat) | |
2195 | { | |
1da177e4 LT |
2196 | /* Skip empty nodes */ |
2197 | if (!pgdat->node_spanned_pages) | |
2198 | return; | |
2199 | ||
d41dee36 | 2200 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
2201 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
2202 | if (!pgdat->node_mem_map) { | |
e984bb43 | 2203 | unsigned long size, start, end; |
d41dee36 AW |
2204 | struct page *map; |
2205 | ||
e984bb43 BP |
2206 | /* |
2207 | * The zone's endpoints aren't required to be MAX_ORDER | |
2208 | * aligned but the node_mem_map endpoints must be in order | |
2209 | * for the buddy allocator to function correctly. | |
2210 | */ | |
2211 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
2212 | end = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
2213 | end = ALIGN(end, MAX_ORDER_NR_PAGES); | |
2214 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
2215 | map = alloc_remap(pgdat->node_id, size); |
2216 | if (!map) | |
2217 | map = alloc_bootmem_node(pgdat, size); | |
e984bb43 | 2218 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
1da177e4 | 2219 | } |
d41dee36 | 2220 | #ifdef CONFIG_FLATMEM |
1da177e4 LT |
2221 | /* |
2222 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
2223 | */ | |
2224 | if (pgdat == NODE_DATA(0)) | |
2225 | mem_map = NODE_DATA(0)->node_mem_map; | |
2226 | #endif | |
d41dee36 | 2227 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
2228 | } |
2229 | ||
86356ab1 | 2230 | void __meminit free_area_init_node(int nid, struct pglist_data *pgdat, |
1da177e4 LT |
2231 | unsigned long *zones_size, unsigned long node_start_pfn, |
2232 | unsigned long *zholes_size) | |
2233 | { | |
2234 | pgdat->node_id = nid; | |
2235 | pgdat->node_start_pfn = node_start_pfn; | |
2236 | calculate_zone_totalpages(pgdat, zones_size, zholes_size); | |
2237 | ||
2238 | alloc_node_mem_map(pgdat); | |
2239 | ||
2240 | free_area_init_core(pgdat, zones_size, zholes_size); | |
2241 | } | |
2242 | ||
93b7504e | 2243 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
2244 | static bootmem_data_t contig_bootmem_data; |
2245 | struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data }; | |
2246 | ||
2247 | EXPORT_SYMBOL(contig_page_data); | |
93b7504e | 2248 | #endif |
1da177e4 LT |
2249 | |
2250 | void __init free_area_init(unsigned long *zones_size) | |
2251 | { | |
93b7504e | 2252 | free_area_init_node(0, NODE_DATA(0), zones_size, |
1da177e4 LT |
2253 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
2254 | } | |
1da177e4 LT |
2255 | |
2256 | #ifdef CONFIG_PROC_FS | |
2257 | ||
2258 | #include <linux/seq_file.h> | |
2259 | ||
2260 | static void *frag_start(struct seq_file *m, loff_t *pos) | |
2261 | { | |
2262 | pg_data_t *pgdat; | |
2263 | loff_t node = *pos; | |
ae0f15fb KH |
2264 | for (pgdat = first_online_pgdat(); |
2265 | pgdat && node; | |
2266 | pgdat = next_online_pgdat(pgdat)) | |
1da177e4 LT |
2267 | --node; |
2268 | ||
2269 | return pgdat; | |
2270 | } | |
2271 | ||
2272 | static void *frag_next(struct seq_file *m, void *arg, loff_t *pos) | |
2273 | { | |
2274 | pg_data_t *pgdat = (pg_data_t *)arg; | |
2275 | ||
2276 | (*pos)++; | |
ae0f15fb | 2277 | return next_online_pgdat(pgdat); |
1da177e4 LT |
2278 | } |
2279 | ||
2280 | static void frag_stop(struct seq_file *m, void *arg) | |
2281 | { | |
2282 | } | |
2283 | ||
2284 | /* | |
2285 | * This walks the free areas for each zone. | |
2286 | */ | |
2287 | static int frag_show(struct seq_file *m, void *arg) | |
2288 | { | |
2289 | pg_data_t *pgdat = (pg_data_t *)arg; | |
2290 | struct zone *zone; | |
2291 | struct zone *node_zones = pgdat->node_zones; | |
2292 | unsigned long flags; | |
2293 | int order; | |
2294 | ||
2295 | for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) { | |
f3fe6512 | 2296 | if (!populated_zone(zone)) |
1da177e4 LT |
2297 | continue; |
2298 | ||
2299 | spin_lock_irqsave(&zone->lock, flags); | |
2300 | seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name); | |
2301 | for (order = 0; order < MAX_ORDER; ++order) | |
2302 | seq_printf(m, "%6lu ", zone->free_area[order].nr_free); | |
2303 | spin_unlock_irqrestore(&zone->lock, flags); | |
2304 | seq_putc(m, '\n'); | |
2305 | } | |
2306 | return 0; | |
2307 | } | |
2308 | ||
2309 | struct seq_operations fragmentation_op = { | |
2310 | .start = frag_start, | |
2311 | .next = frag_next, | |
2312 | .stop = frag_stop, | |
2313 | .show = frag_show, | |
2314 | }; | |
2315 | ||
295ab934 ND |
2316 | /* |
2317 | * Output information about zones in @pgdat. | |
2318 | */ | |
2319 | static int zoneinfo_show(struct seq_file *m, void *arg) | |
2320 | { | |
2321 | pg_data_t *pgdat = arg; | |
2322 | struct zone *zone; | |
2323 | struct zone *node_zones = pgdat->node_zones; | |
2324 | unsigned long flags; | |
2325 | ||
2326 | for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; zone++) { | |
2327 | int i; | |
2328 | ||
f3fe6512 | 2329 | if (!populated_zone(zone)) |
295ab934 ND |
2330 | continue; |
2331 | ||
2332 | spin_lock_irqsave(&zone->lock, flags); | |
2333 | seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name); | |
2334 | seq_printf(m, | |
2335 | "\n pages free %lu" | |
2336 | "\n min %lu" | |
2337 | "\n low %lu" | |
2338 | "\n high %lu" | |
2339 | "\n active %lu" | |
2340 | "\n inactive %lu" | |
2341 | "\n scanned %lu (a: %lu i: %lu)" | |
2342 | "\n spanned %lu" | |
2343 | "\n present %lu", | |
2344 | zone->free_pages, | |
2345 | zone->pages_min, | |
2346 | zone->pages_low, | |
2347 | zone->pages_high, | |
2348 | zone->nr_active, | |
2349 | zone->nr_inactive, | |
2350 | zone->pages_scanned, | |
2351 | zone->nr_scan_active, zone->nr_scan_inactive, | |
2352 | zone->spanned_pages, | |
2353 | zone->present_pages); | |
2354 | seq_printf(m, | |
2355 | "\n protection: (%lu", | |
2356 | zone->lowmem_reserve[0]); | |
2357 | for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++) | |
2358 | seq_printf(m, ", %lu", zone->lowmem_reserve[i]); | |
2359 | seq_printf(m, | |
2360 | ")" | |
2361 | "\n pagesets"); | |
23316bc8 | 2362 | for_each_online_cpu(i) { |
295ab934 ND |
2363 | struct per_cpu_pageset *pageset; |
2364 | int j; | |
2365 | ||
e7c8d5c9 | 2366 | pageset = zone_pcp(zone, i); |
295ab934 ND |
2367 | for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) { |
2368 | if (pageset->pcp[j].count) | |
2369 | break; | |
2370 | } | |
2371 | if (j == ARRAY_SIZE(pageset->pcp)) | |
2372 | continue; | |
2373 | for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) { | |
2374 | seq_printf(m, | |
2375 | "\n cpu: %i pcp: %i" | |
2376 | "\n count: %i" | |
295ab934 ND |
2377 | "\n high: %i" |
2378 | "\n batch: %i", | |
2379 | i, j, | |
2380 | pageset->pcp[j].count, | |
295ab934 ND |
2381 | pageset->pcp[j].high, |
2382 | pageset->pcp[j].batch); | |
2383 | } | |
2384 | #ifdef CONFIG_NUMA | |
2385 | seq_printf(m, | |
2386 | "\n numa_hit: %lu" | |
2387 | "\n numa_miss: %lu" | |
2388 | "\n numa_foreign: %lu" | |
2389 | "\n interleave_hit: %lu" | |
2390 | "\n local_node: %lu" | |
2391 | "\n other_node: %lu", | |
2392 | pageset->numa_hit, | |
2393 | pageset->numa_miss, | |
2394 | pageset->numa_foreign, | |
2395 | pageset->interleave_hit, | |
2396 | pageset->local_node, | |
2397 | pageset->other_node); | |
2398 | #endif | |
2399 | } | |
2400 | seq_printf(m, | |
2401 | "\n all_unreclaimable: %u" | |
2402 | "\n prev_priority: %i" | |
2403 | "\n temp_priority: %i" | |
2404 | "\n start_pfn: %lu", | |
2405 | zone->all_unreclaimable, | |
2406 | zone->prev_priority, | |
2407 | zone->temp_priority, | |
2408 | zone->zone_start_pfn); | |
2409 | spin_unlock_irqrestore(&zone->lock, flags); | |
2410 | seq_putc(m, '\n'); | |
2411 | } | |
2412 | return 0; | |
2413 | } | |
2414 | ||
2415 | struct seq_operations zoneinfo_op = { | |
2416 | .start = frag_start, /* iterate over all zones. The same as in | |
2417 | * fragmentation. */ | |
2418 | .next = frag_next, | |
2419 | .stop = frag_stop, | |
2420 | .show = zoneinfo_show, | |
2421 | }; | |
2422 | ||
1da177e4 LT |
2423 | static char *vmstat_text[] = { |
2424 | "nr_dirty", | |
2425 | "nr_writeback", | |
2426 | "nr_unstable", | |
2427 | "nr_page_table_pages", | |
2428 | "nr_mapped", | |
2429 | "nr_slab", | |
2430 | ||
2431 | "pgpgin", | |
2432 | "pgpgout", | |
2433 | "pswpin", | |
2434 | "pswpout", | |
1da177e4 | 2435 | |
9328b8fa | 2436 | "pgalloc_high", |
1da177e4 | 2437 | "pgalloc_normal", |
9328b8fa | 2438 | "pgalloc_dma32", |
1da177e4 | 2439 | "pgalloc_dma", |
9328b8fa | 2440 | |
1da177e4 LT |
2441 | "pgfree", |
2442 | "pgactivate", | |
2443 | "pgdeactivate", | |
2444 | ||
2445 | "pgfault", | |
2446 | "pgmajfault", | |
9328b8fa | 2447 | |
1da177e4 LT |
2448 | "pgrefill_high", |
2449 | "pgrefill_normal", | |
9328b8fa | 2450 | "pgrefill_dma32", |
1da177e4 LT |
2451 | "pgrefill_dma", |
2452 | ||
2453 | "pgsteal_high", | |
2454 | "pgsteal_normal", | |
9328b8fa | 2455 | "pgsteal_dma32", |
1da177e4 | 2456 | "pgsteal_dma", |
9328b8fa | 2457 | |
1da177e4 LT |
2458 | "pgscan_kswapd_high", |
2459 | "pgscan_kswapd_normal", | |
9328b8fa | 2460 | "pgscan_kswapd_dma32", |
1da177e4 | 2461 | "pgscan_kswapd_dma", |
9328b8fa | 2462 | |
1da177e4 LT |
2463 | "pgscan_direct_high", |
2464 | "pgscan_direct_normal", | |
9328b8fa | 2465 | "pgscan_direct_dma32", |
1da177e4 | 2466 | "pgscan_direct_dma", |
1da177e4 | 2467 | |
9328b8fa | 2468 | "pginodesteal", |
1da177e4 LT |
2469 | "slabs_scanned", |
2470 | "kswapd_steal", | |
2471 | "kswapd_inodesteal", | |
2472 | "pageoutrun", | |
2473 | "allocstall", | |
2474 | ||
2475 | "pgrotated", | |
edfbe2b0 | 2476 | "nr_bounce", |
1da177e4 LT |
2477 | }; |
2478 | ||
2479 | static void *vmstat_start(struct seq_file *m, loff_t *pos) | |
2480 | { | |
2481 | struct page_state *ps; | |
2482 | ||
2483 | if (*pos >= ARRAY_SIZE(vmstat_text)) | |
2484 | return NULL; | |
2485 | ||
2486 | ps = kmalloc(sizeof(*ps), GFP_KERNEL); | |
2487 | m->private = ps; | |
2488 | if (!ps) | |
2489 | return ERR_PTR(-ENOMEM); | |
2490 | get_full_page_state(ps); | |
2491 | ps->pgpgin /= 2; /* sectors -> kbytes */ | |
2492 | ps->pgpgout /= 2; | |
2493 | return (unsigned long *)ps + *pos; | |
2494 | } | |
2495 | ||
2496 | static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos) | |
2497 | { | |
2498 | (*pos)++; | |
2499 | if (*pos >= ARRAY_SIZE(vmstat_text)) | |
2500 | return NULL; | |
2501 | return (unsigned long *)m->private + *pos; | |
2502 | } | |
2503 | ||
2504 | static int vmstat_show(struct seq_file *m, void *arg) | |
2505 | { | |
2506 | unsigned long *l = arg; | |
2507 | unsigned long off = l - (unsigned long *)m->private; | |
2508 | ||
2509 | seq_printf(m, "%s %lu\n", vmstat_text[off], *l); | |
2510 | return 0; | |
2511 | } | |
2512 | ||
2513 | static void vmstat_stop(struct seq_file *m, void *arg) | |
2514 | { | |
2515 | kfree(m->private); | |
2516 | m->private = NULL; | |
2517 | } | |
2518 | ||
2519 | struct seq_operations vmstat_op = { | |
2520 | .start = vmstat_start, | |
2521 | .next = vmstat_next, | |
2522 | .stop = vmstat_stop, | |
2523 | .show = vmstat_show, | |
2524 | }; | |
2525 | ||
2526 | #endif /* CONFIG_PROC_FS */ | |
2527 | ||
2528 | #ifdef CONFIG_HOTPLUG_CPU | |
2529 | static int page_alloc_cpu_notify(struct notifier_block *self, | |
2530 | unsigned long action, void *hcpu) | |
2531 | { | |
2532 | int cpu = (unsigned long)hcpu; | |
2533 | long *count; | |
2534 | unsigned long *src, *dest; | |
2535 | ||
2536 | if (action == CPU_DEAD) { | |
2537 | int i; | |
2538 | ||
2539 | /* Drain local pagecache count. */ | |
2540 | count = &per_cpu(nr_pagecache_local, cpu); | |
2541 | atomic_add(*count, &nr_pagecache); | |
2542 | *count = 0; | |
2543 | local_irq_disable(); | |
2544 | __drain_pages(cpu); | |
2545 | ||
2546 | /* Add dead cpu's page_states to our own. */ | |
2547 | dest = (unsigned long *)&__get_cpu_var(page_states); | |
2548 | src = (unsigned long *)&per_cpu(page_states, cpu); | |
2549 | ||
2550 | for (i = 0; i < sizeof(struct page_state)/sizeof(unsigned long); | |
2551 | i++) { | |
2552 | dest[i] += src[i]; | |
2553 | src[i] = 0; | |
2554 | } | |
2555 | ||
2556 | local_irq_enable(); | |
2557 | } | |
2558 | return NOTIFY_OK; | |
2559 | } | |
2560 | #endif /* CONFIG_HOTPLUG_CPU */ | |
2561 | ||
2562 | void __init page_alloc_init(void) | |
2563 | { | |
2564 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
2565 | } | |
2566 | ||
cb45b0e9 HA |
2567 | /* |
2568 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio | |
2569 | * or min_free_kbytes changes. | |
2570 | */ | |
2571 | static void calculate_totalreserve_pages(void) | |
2572 | { | |
2573 | struct pglist_data *pgdat; | |
2574 | unsigned long reserve_pages = 0; | |
2575 | int i, j; | |
2576 | ||
2577 | for_each_online_pgdat(pgdat) { | |
2578 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2579 | struct zone *zone = pgdat->node_zones + i; | |
2580 | unsigned long max = 0; | |
2581 | ||
2582 | /* Find valid and maximum lowmem_reserve in the zone */ | |
2583 | for (j = i; j < MAX_NR_ZONES; j++) { | |
2584 | if (zone->lowmem_reserve[j] > max) | |
2585 | max = zone->lowmem_reserve[j]; | |
2586 | } | |
2587 | ||
2588 | /* we treat pages_high as reserved pages. */ | |
2589 | max += zone->pages_high; | |
2590 | ||
2591 | if (max > zone->present_pages) | |
2592 | max = zone->present_pages; | |
2593 | reserve_pages += max; | |
2594 | } | |
2595 | } | |
2596 | totalreserve_pages = reserve_pages; | |
2597 | } | |
2598 | ||
1da177e4 LT |
2599 | /* |
2600 | * setup_per_zone_lowmem_reserve - called whenever | |
2601 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
2602 | * has a correct pages reserved value, so an adequate number of | |
2603 | * pages are left in the zone after a successful __alloc_pages(). | |
2604 | */ | |
2605 | static void setup_per_zone_lowmem_reserve(void) | |
2606 | { | |
2607 | struct pglist_data *pgdat; | |
2608 | int j, idx; | |
2609 | ||
ec936fc5 | 2610 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
2611 | for (j = 0; j < MAX_NR_ZONES; j++) { |
2612 | struct zone *zone = pgdat->node_zones + j; | |
2613 | unsigned long present_pages = zone->present_pages; | |
2614 | ||
2615 | zone->lowmem_reserve[j] = 0; | |
2616 | ||
2617 | for (idx = j-1; idx >= 0; idx--) { | |
2618 | struct zone *lower_zone; | |
2619 | ||
2620 | if (sysctl_lowmem_reserve_ratio[idx] < 1) | |
2621 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
2622 | ||
2623 | lower_zone = pgdat->node_zones + idx; | |
2624 | lower_zone->lowmem_reserve[j] = present_pages / | |
2625 | sysctl_lowmem_reserve_ratio[idx]; | |
2626 | present_pages += lower_zone->present_pages; | |
2627 | } | |
2628 | } | |
2629 | } | |
cb45b0e9 HA |
2630 | |
2631 | /* update totalreserve_pages */ | |
2632 | calculate_totalreserve_pages(); | |
1da177e4 LT |
2633 | } |
2634 | ||
2635 | /* | |
2636 | * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures | |
2637 | * that the pages_{min,low,high} values for each zone are set correctly | |
2638 | * with respect to min_free_kbytes. | |
2639 | */ | |
3947be19 | 2640 | void setup_per_zone_pages_min(void) |
1da177e4 LT |
2641 | { |
2642 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
2643 | unsigned long lowmem_pages = 0; | |
2644 | struct zone *zone; | |
2645 | unsigned long flags; | |
2646 | ||
2647 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
2648 | for_each_zone(zone) { | |
2649 | if (!is_highmem(zone)) | |
2650 | lowmem_pages += zone->present_pages; | |
2651 | } | |
2652 | ||
2653 | for_each_zone(zone) { | |
ac924c60 AM |
2654 | u64 tmp; |
2655 | ||
1da177e4 | 2656 | spin_lock_irqsave(&zone->lru_lock, flags); |
ac924c60 AM |
2657 | tmp = (u64)pages_min * zone->present_pages; |
2658 | do_div(tmp, lowmem_pages); | |
1da177e4 LT |
2659 | if (is_highmem(zone)) { |
2660 | /* | |
669ed175 NP |
2661 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
2662 | * need highmem pages, so cap pages_min to a small | |
2663 | * value here. | |
2664 | * | |
2665 | * The (pages_high-pages_low) and (pages_low-pages_min) | |
2666 | * deltas controls asynch page reclaim, and so should | |
2667 | * not be capped for highmem. | |
1da177e4 LT |
2668 | */ |
2669 | int min_pages; | |
2670 | ||
2671 | min_pages = zone->present_pages / 1024; | |
2672 | if (min_pages < SWAP_CLUSTER_MAX) | |
2673 | min_pages = SWAP_CLUSTER_MAX; | |
2674 | if (min_pages > 128) | |
2675 | min_pages = 128; | |
2676 | zone->pages_min = min_pages; | |
2677 | } else { | |
669ed175 NP |
2678 | /* |
2679 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
2680 | * proportionate to the zone's size. |
2681 | */ | |
669ed175 | 2682 | zone->pages_min = tmp; |
1da177e4 LT |
2683 | } |
2684 | ||
ac924c60 AM |
2685 | zone->pages_low = zone->pages_min + (tmp >> 2); |
2686 | zone->pages_high = zone->pages_min + (tmp >> 1); | |
1da177e4 LT |
2687 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
2688 | } | |
cb45b0e9 HA |
2689 | |
2690 | /* update totalreserve_pages */ | |
2691 | calculate_totalreserve_pages(); | |
1da177e4 LT |
2692 | } |
2693 | ||
2694 | /* | |
2695 | * Initialise min_free_kbytes. | |
2696 | * | |
2697 | * For small machines we want it small (128k min). For large machines | |
2698 | * we want it large (64MB max). But it is not linear, because network | |
2699 | * bandwidth does not increase linearly with machine size. We use | |
2700 | * | |
2701 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: | |
2702 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) | |
2703 | * | |
2704 | * which yields | |
2705 | * | |
2706 | * 16MB: 512k | |
2707 | * 32MB: 724k | |
2708 | * 64MB: 1024k | |
2709 | * 128MB: 1448k | |
2710 | * 256MB: 2048k | |
2711 | * 512MB: 2896k | |
2712 | * 1024MB: 4096k | |
2713 | * 2048MB: 5792k | |
2714 | * 4096MB: 8192k | |
2715 | * 8192MB: 11584k | |
2716 | * 16384MB: 16384k | |
2717 | */ | |
2718 | static int __init init_per_zone_pages_min(void) | |
2719 | { | |
2720 | unsigned long lowmem_kbytes; | |
2721 | ||
2722 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
2723 | ||
2724 | min_free_kbytes = int_sqrt(lowmem_kbytes * 16); | |
2725 | if (min_free_kbytes < 128) | |
2726 | min_free_kbytes = 128; | |
2727 | if (min_free_kbytes > 65536) | |
2728 | min_free_kbytes = 65536; | |
2729 | setup_per_zone_pages_min(); | |
2730 | setup_per_zone_lowmem_reserve(); | |
2731 | return 0; | |
2732 | } | |
2733 | module_init(init_per_zone_pages_min) | |
2734 | ||
2735 | /* | |
2736 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so | |
2737 | * that we can call two helper functions whenever min_free_kbytes | |
2738 | * changes. | |
2739 | */ | |
2740 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, | |
2741 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
2742 | { | |
2743 | proc_dointvec(table, write, file, buffer, length, ppos); | |
2744 | setup_per_zone_pages_min(); | |
2745 | return 0; | |
2746 | } | |
2747 | ||
2748 | /* | |
2749 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
2750 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
2751 | * whenever sysctl_lowmem_reserve_ratio changes. | |
2752 | * | |
2753 | * The reserve ratio obviously has absolutely no relation with the | |
2754 | * pages_min watermarks. The lowmem reserve ratio can only make sense | |
2755 | * if in function of the boot time zone sizes. | |
2756 | */ | |
2757 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, | |
2758 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
2759 | { | |
2760 | proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
2761 | setup_per_zone_lowmem_reserve(); | |
2762 | return 0; | |
2763 | } | |
2764 | ||
8ad4b1fb RS |
2765 | /* |
2766 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
2767 | * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist | |
2768 | * can have before it gets flushed back to buddy allocator. | |
2769 | */ | |
2770 | ||
2771 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, | |
2772 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
2773 | { | |
2774 | struct zone *zone; | |
2775 | unsigned int cpu; | |
2776 | int ret; | |
2777 | ||
2778 | ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
2779 | if (!write || (ret == -EINVAL)) | |
2780 | return ret; | |
2781 | for_each_zone(zone) { | |
2782 | for_each_online_cpu(cpu) { | |
2783 | unsigned long high; | |
2784 | high = zone->present_pages / percpu_pagelist_fraction; | |
2785 | setup_pagelist_highmark(zone_pcp(zone, cpu), high); | |
2786 | } | |
2787 | } | |
2788 | return 0; | |
2789 | } | |
2790 | ||
1da177e4 LT |
2791 | __initdata int hashdist = HASHDIST_DEFAULT; |
2792 | ||
2793 | #ifdef CONFIG_NUMA | |
2794 | static int __init set_hashdist(char *str) | |
2795 | { | |
2796 | if (!str) | |
2797 | return 0; | |
2798 | hashdist = simple_strtoul(str, &str, 0); | |
2799 | return 1; | |
2800 | } | |
2801 | __setup("hashdist=", set_hashdist); | |
2802 | #endif | |
2803 | ||
2804 | /* | |
2805 | * allocate a large system hash table from bootmem | |
2806 | * - it is assumed that the hash table must contain an exact power-of-2 | |
2807 | * quantity of entries | |
2808 | * - limit is the number of hash buckets, not the total allocation size | |
2809 | */ | |
2810 | void *__init alloc_large_system_hash(const char *tablename, | |
2811 | unsigned long bucketsize, | |
2812 | unsigned long numentries, | |
2813 | int scale, | |
2814 | int flags, | |
2815 | unsigned int *_hash_shift, | |
2816 | unsigned int *_hash_mask, | |
2817 | unsigned long limit) | |
2818 | { | |
2819 | unsigned long long max = limit; | |
2820 | unsigned long log2qty, size; | |
2821 | void *table = NULL; | |
2822 | ||
2823 | /* allow the kernel cmdline to have a say */ | |
2824 | if (!numentries) { | |
2825 | /* round applicable memory size up to nearest megabyte */ | |
2826 | numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages; | |
2827 | numentries += (1UL << (20 - PAGE_SHIFT)) - 1; | |
2828 | numentries >>= 20 - PAGE_SHIFT; | |
2829 | numentries <<= 20 - PAGE_SHIFT; | |
2830 | ||
2831 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
2832 | if (scale > PAGE_SHIFT) | |
2833 | numentries >>= (scale - PAGE_SHIFT); | |
2834 | else | |
2835 | numentries <<= (PAGE_SHIFT - scale); | |
2836 | } | |
6e692ed3 | 2837 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
2838 | |
2839 | /* limit allocation size to 1/16 total memory by default */ | |
2840 | if (max == 0) { | |
2841 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
2842 | do_div(max, bucketsize); | |
2843 | } | |
2844 | ||
2845 | if (numentries > max) | |
2846 | numentries = max; | |
2847 | ||
2848 | log2qty = long_log2(numentries); | |
2849 | ||
2850 | do { | |
2851 | size = bucketsize << log2qty; | |
2852 | if (flags & HASH_EARLY) | |
2853 | table = alloc_bootmem(size); | |
2854 | else if (hashdist) | |
2855 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
2856 | else { | |
2857 | unsigned long order; | |
2858 | for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++) | |
2859 | ; | |
2860 | table = (void*) __get_free_pages(GFP_ATOMIC, order); | |
2861 | } | |
2862 | } while (!table && size > PAGE_SIZE && --log2qty); | |
2863 | ||
2864 | if (!table) | |
2865 | panic("Failed to allocate %s hash table\n", tablename); | |
2866 | ||
2867 | printk("%s hash table entries: %d (order: %d, %lu bytes)\n", | |
2868 | tablename, | |
2869 | (1U << log2qty), | |
2870 | long_log2(size) - PAGE_SHIFT, | |
2871 | size); | |
2872 | ||
2873 | if (_hash_shift) | |
2874 | *_hash_shift = log2qty; | |
2875 | if (_hash_mask) | |
2876 | *_hash_mask = (1 << log2qty) - 1; | |
2877 | ||
2878 | return table; | |
2879 | } | |
a117e66e KH |
2880 | |
2881 | #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE | |
a117e66e KH |
2882 | struct page *pfn_to_page(unsigned long pfn) |
2883 | { | |
67de6482 | 2884 | return __pfn_to_page(pfn); |
a117e66e KH |
2885 | } |
2886 | unsigned long page_to_pfn(struct page *page) | |
2887 | { | |
67de6482 | 2888 | return __page_to_pfn(page); |
a117e66e | 2889 | } |
a117e66e KH |
2890 | EXPORT_SYMBOL(pfn_to_page); |
2891 | EXPORT_SYMBOL(page_to_pfn); | |
2892 | #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */ |