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> |
1da177e4 LT |
40 | |
41 | #include <asm/tlbflush.h> | |
42 | #include "internal.h" | |
43 | ||
44 | /* | |
45 | * MCD - HACK: Find somewhere to initialize this EARLY, or make this | |
46 | * initializer cleaner | |
47 | */ | |
c3d8c141 | 48 | nodemask_t node_online_map __read_mostly = { { [0] = 1UL } }; |
7223a93a | 49 | EXPORT_SYMBOL(node_online_map); |
c3d8c141 | 50 | nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL; |
7223a93a | 51 | EXPORT_SYMBOL(node_possible_map); |
c3d8c141 | 52 | struct pglist_data *pgdat_list __read_mostly; |
6c231b7b RT |
53 | unsigned long totalram_pages __read_mostly; |
54 | unsigned long totalhigh_pages __read_mostly; | |
1da177e4 | 55 | long nr_swap_pages; |
8ad4b1fb | 56 | int percpu_pagelist_fraction; |
1da177e4 | 57 | |
d98c7a09 | 58 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 59 | |
1da177e4 LT |
60 | /* |
61 | * results with 256, 32 in the lowmem_reserve sysctl: | |
62 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
63 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
64 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
65 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
66 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA | |
a2f1b424 AK |
67 | * |
68 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
69 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 70 | */ |
a2f1b424 | 71 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { 256, 256, 32 }; |
1da177e4 LT |
72 | |
73 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 LT |
74 | |
75 | /* | |
76 | * Used by page_zone() to look up the address of the struct zone whose | |
77 | * id is encoded in the upper bits of page->flags | |
78 | */ | |
c3d8c141 | 79 | struct zone *zone_table[1 << ZONETABLE_SHIFT] __read_mostly; |
1da177e4 LT |
80 | EXPORT_SYMBOL(zone_table); |
81 | ||
a2f1b424 | 82 | static char *zone_names[MAX_NR_ZONES] = { "DMA", "DMA32", "Normal", "HighMem" }; |
1da177e4 LT |
83 | int min_free_kbytes = 1024; |
84 | ||
85 | unsigned long __initdata nr_kernel_pages; | |
86 | unsigned long __initdata nr_all_pages; | |
87 | ||
13e7444b | 88 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 89 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 90 | { |
bdc8cb98 DH |
91 | int ret = 0; |
92 | unsigned seq; | |
93 | unsigned long pfn = page_to_pfn(page); | |
c6a57e19 | 94 | |
bdc8cb98 DH |
95 | do { |
96 | seq = zone_span_seqbegin(zone); | |
97 | if (pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
98 | ret = 1; | |
99 | else if (pfn < zone->zone_start_pfn) | |
100 | ret = 1; | |
101 | } while (zone_span_seqretry(zone, seq)); | |
102 | ||
103 | return ret; | |
c6a57e19 DH |
104 | } |
105 | ||
106 | static int page_is_consistent(struct zone *zone, struct page *page) | |
107 | { | |
1da177e4 LT |
108 | #ifdef CONFIG_HOLES_IN_ZONE |
109 | if (!pfn_valid(page_to_pfn(page))) | |
c6a57e19 | 110 | return 0; |
1da177e4 LT |
111 | #endif |
112 | if (zone != page_zone(page)) | |
c6a57e19 DH |
113 | return 0; |
114 | ||
115 | return 1; | |
116 | } | |
117 | /* | |
118 | * Temporary debugging check for pages not lying within a given zone. | |
119 | */ | |
120 | static int bad_range(struct zone *zone, struct page *page) | |
121 | { | |
122 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 123 | return 1; |
c6a57e19 DH |
124 | if (!page_is_consistent(zone, page)) |
125 | return 1; | |
126 | ||
1da177e4 LT |
127 | return 0; |
128 | } | |
129 | ||
13e7444b NP |
130 | #else |
131 | static inline int bad_range(struct zone *zone, struct page *page) | |
132 | { | |
133 | return 0; | |
134 | } | |
135 | #endif | |
136 | ||
224abf92 | 137 | static void bad_page(struct page *page) |
1da177e4 | 138 | { |
224abf92 | 139 | printk(KERN_EMERG "Bad page state in process '%s'\n" |
7365f3d1 HD |
140 | KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n" |
141 | KERN_EMERG "Trying to fix it up, but a reboot is needed\n" | |
142 | KERN_EMERG "Backtrace:\n", | |
224abf92 NP |
143 | current->comm, page, (int)(2*sizeof(unsigned long)), |
144 | (unsigned long)page->flags, page->mapping, | |
145 | page_mapcount(page), page_count(page)); | |
1da177e4 | 146 | dump_stack(); |
334795ec HD |
147 | page->flags &= ~(1 << PG_lru | |
148 | 1 << PG_private | | |
1da177e4 | 149 | 1 << PG_locked | |
1da177e4 LT |
150 | 1 << PG_active | |
151 | 1 << PG_dirty | | |
334795ec HD |
152 | 1 << PG_reclaim | |
153 | 1 << PG_slab | | |
1da177e4 | 154 | 1 << PG_swapcache | |
689bcebf | 155 | 1 << PG_writeback ); |
1da177e4 LT |
156 | set_page_count(page, 0); |
157 | reset_page_mapcount(page); | |
158 | page->mapping = NULL; | |
9f158333 | 159 | add_taint(TAINT_BAD_PAGE); |
1da177e4 LT |
160 | } |
161 | ||
1da177e4 LT |
162 | /* |
163 | * Higher-order pages are called "compound pages". They are structured thusly: | |
164 | * | |
165 | * The first PAGE_SIZE page is called the "head page". | |
166 | * | |
167 | * The remaining PAGE_SIZE pages are called "tail pages". | |
168 | * | |
169 | * All pages have PG_compound set. All pages have their ->private pointing at | |
170 | * the head page (even the head page has this). | |
171 | * | |
41d78ba5 HD |
172 | * The first tail page's ->lru.next holds the address of the compound page's |
173 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
174 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 175 | */ |
d98c7a09 HD |
176 | |
177 | static void free_compound_page(struct page *page) | |
178 | { | |
179 | __free_pages_ok(page, (unsigned long)page[1].lru.prev); | |
180 | } | |
181 | ||
1da177e4 LT |
182 | static void prep_compound_page(struct page *page, unsigned long order) |
183 | { | |
184 | int i; | |
185 | int nr_pages = 1 << order; | |
186 | ||
d98c7a09 | 187 | page[1].lru.next = (void *)free_compound_page; /* set dtor */ |
41d78ba5 | 188 | page[1].lru.prev = (void *)order; |
1da177e4 LT |
189 | for (i = 0; i < nr_pages; i++) { |
190 | struct page *p = page + i; | |
191 | ||
5e9dace8 | 192 | __SetPageCompound(p); |
4c21e2f2 | 193 | set_page_private(p, (unsigned long)page); |
1da177e4 LT |
194 | } |
195 | } | |
196 | ||
197 | static void destroy_compound_page(struct page *page, unsigned long order) | |
198 | { | |
199 | int i; | |
200 | int nr_pages = 1 << order; | |
201 | ||
41d78ba5 | 202 | if (unlikely((unsigned long)page[1].lru.prev != order)) |
224abf92 | 203 | bad_page(page); |
1da177e4 LT |
204 | |
205 | for (i = 0; i < nr_pages; i++) { | |
206 | struct page *p = page + i; | |
207 | ||
224abf92 NP |
208 | if (unlikely(!PageCompound(p) | |
209 | (page_private(p) != (unsigned long)page))) | |
210 | bad_page(page); | |
5e9dace8 | 211 | __ClearPageCompound(p); |
1da177e4 LT |
212 | } |
213 | } | |
1da177e4 | 214 | |
17cf4406 NP |
215 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) |
216 | { | |
217 | int i; | |
218 | ||
219 | BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM); | |
220 | for (i = 0; i < (1 << order); i++) | |
221 | clear_highpage(page + i); | |
222 | } | |
223 | ||
1da177e4 LT |
224 | /* |
225 | * function for dealing with page's order in buddy system. | |
226 | * zone->lock is already acquired when we use these. | |
227 | * So, we don't need atomic page->flags operations here. | |
228 | */ | |
229 | static inline unsigned long page_order(struct page *page) { | |
4c21e2f2 | 230 | return page_private(page); |
1da177e4 LT |
231 | } |
232 | ||
233 | static inline void set_page_order(struct page *page, int order) { | |
4c21e2f2 | 234 | set_page_private(page, order); |
1da177e4 LT |
235 | __SetPagePrivate(page); |
236 | } | |
237 | ||
238 | static inline void rmv_page_order(struct page *page) | |
239 | { | |
240 | __ClearPagePrivate(page); | |
4c21e2f2 | 241 | set_page_private(page, 0); |
1da177e4 LT |
242 | } |
243 | ||
244 | /* | |
245 | * Locate the struct page for both the matching buddy in our | |
246 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
247 | * | |
248 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
249 | * the following equation: | |
250 | * B2 = B1 ^ (1 << O) | |
251 | * For example, if the starting buddy (buddy2) is #8 its order | |
252 | * 1 buddy is #10: | |
253 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
254 | * | |
255 | * 2) Any buddy B will have an order O+1 parent P which | |
256 | * satisfies the following equation: | |
257 | * P = B & ~(1 << O) | |
258 | * | |
259 | * Assumption: *_mem_map is contigious at least up to MAX_ORDER | |
260 | */ | |
261 | static inline struct page * | |
262 | __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order) | |
263 | { | |
264 | unsigned long buddy_idx = page_idx ^ (1 << order); | |
265 | ||
266 | return page + (buddy_idx - page_idx); | |
267 | } | |
268 | ||
269 | static inline unsigned long | |
270 | __find_combined_index(unsigned long page_idx, unsigned int order) | |
271 | { | |
272 | return (page_idx & ~(1 << order)); | |
273 | } | |
274 | ||
275 | /* | |
276 | * This function checks whether a page is free && is the buddy | |
277 | * we can do coalesce a page and its buddy if | |
13e7444b NP |
278 | * (a) the buddy is not in a hole && |
279 | * (b) the buddy is free && | |
280 | * (c) the buddy is on the buddy system && | |
281 | * (d) a page and its buddy have the same order. | |
4c21e2f2 | 282 | * for recording page's order, we use page_private(page) and PG_private. |
1da177e4 LT |
283 | * |
284 | */ | |
285 | static inline int page_is_buddy(struct page *page, int order) | |
286 | { | |
13e7444b NP |
287 | #ifdef CONFIG_HOLES_IN_ZONE |
288 | if (!pfn_valid(page_to_pfn(page))) | |
289 | return 0; | |
290 | #endif | |
291 | ||
1da177e4 LT |
292 | if (PagePrivate(page) && |
293 | (page_order(page) == order) && | |
1da177e4 LT |
294 | page_count(page) == 0) |
295 | return 1; | |
296 | return 0; | |
297 | } | |
298 | ||
299 | /* | |
300 | * Freeing function for a buddy system allocator. | |
301 | * | |
302 | * The concept of a buddy system is to maintain direct-mapped table | |
303 | * (containing bit values) for memory blocks of various "orders". | |
304 | * The bottom level table contains the map for the smallest allocatable | |
305 | * units of memory (here, pages), and each level above it describes | |
306 | * pairs of units from the levels below, hence, "buddies". | |
307 | * At a high level, all that happens here is marking the table entry | |
308 | * at the bottom level available, and propagating the changes upward | |
309 | * as necessary, plus some accounting needed to play nicely with other | |
310 | * parts of the VM system. | |
311 | * At each level, we keep a list of pages, which are heads of continuous | |
312 | * free pages of length of (1 << order) and marked with PG_Private.Page's | |
4c21e2f2 | 313 | * order is recorded in page_private(page) field. |
1da177e4 LT |
314 | * So when we are allocating or freeing one, we can derive the state of the |
315 | * other. That is, if we allocate a small block, and both were | |
316 | * free, the remainder of the region must be split into blocks. | |
317 | * If a block is freed, and its buddy is also free, then this | |
318 | * triggers coalescing into a block of larger size. | |
319 | * | |
320 | * -- wli | |
321 | */ | |
322 | ||
48db57f8 | 323 | static inline void __free_one_page(struct page *page, |
1da177e4 LT |
324 | struct zone *zone, unsigned int order) |
325 | { | |
326 | unsigned long page_idx; | |
327 | int order_size = 1 << order; | |
328 | ||
224abf92 | 329 | if (unlikely(PageCompound(page))) |
1da177e4 LT |
330 | destroy_compound_page(page, order); |
331 | ||
332 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); | |
333 | ||
334 | BUG_ON(page_idx & (order_size - 1)); | |
335 | BUG_ON(bad_range(zone, page)); | |
336 | ||
337 | zone->free_pages += order_size; | |
338 | while (order < MAX_ORDER-1) { | |
339 | unsigned long combined_idx; | |
340 | struct free_area *area; | |
341 | struct page *buddy; | |
342 | ||
1da177e4 | 343 | buddy = __page_find_buddy(page, page_idx, order); |
1da177e4 LT |
344 | if (!page_is_buddy(buddy, order)) |
345 | break; /* Move the buddy up one level. */ | |
13e7444b | 346 | |
1da177e4 LT |
347 | list_del(&buddy->lru); |
348 | area = zone->free_area + order; | |
349 | area->nr_free--; | |
350 | rmv_page_order(buddy); | |
13e7444b | 351 | combined_idx = __find_combined_index(page_idx, order); |
1da177e4 LT |
352 | page = page + (combined_idx - page_idx); |
353 | page_idx = combined_idx; | |
354 | order++; | |
355 | } | |
356 | set_page_order(page, order); | |
357 | list_add(&page->lru, &zone->free_area[order].free_list); | |
358 | zone->free_area[order].nr_free++; | |
359 | } | |
360 | ||
224abf92 | 361 | static inline int free_pages_check(struct page *page) |
1da177e4 | 362 | { |
92be2e33 NP |
363 | if (unlikely(page_mapcount(page) | |
364 | (page->mapping != NULL) | | |
365 | (page_count(page) != 0) | | |
1da177e4 LT |
366 | (page->flags & ( |
367 | 1 << PG_lru | | |
368 | 1 << PG_private | | |
369 | 1 << PG_locked | | |
370 | 1 << PG_active | | |
371 | 1 << PG_reclaim | | |
372 | 1 << PG_slab | | |
373 | 1 << PG_swapcache | | |
b5810039 | 374 | 1 << PG_writeback | |
92be2e33 | 375 | 1 << PG_reserved )))) |
224abf92 | 376 | bad_page(page); |
1da177e4 | 377 | if (PageDirty(page)) |
242e5468 | 378 | __ClearPageDirty(page); |
689bcebf HD |
379 | /* |
380 | * For now, we report if PG_reserved was found set, but do not | |
381 | * clear it, and do not free the page. But we shall soon need | |
382 | * to do more, for when the ZERO_PAGE count wraps negative. | |
383 | */ | |
384 | return PageReserved(page); | |
1da177e4 LT |
385 | } |
386 | ||
387 | /* | |
388 | * Frees a list of pages. | |
389 | * Assumes all pages on list are in same zone, and of same order. | |
207f36ee | 390 | * count is the number of pages to free. |
1da177e4 LT |
391 | * |
392 | * If the zone was previously in an "all pages pinned" state then look to | |
393 | * see if this freeing clears that state. | |
394 | * | |
395 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
396 | * pinned" detection logic. | |
397 | */ | |
48db57f8 NP |
398 | static void free_pages_bulk(struct zone *zone, int count, |
399 | struct list_head *list, int order) | |
1da177e4 | 400 | { |
c54ad30c | 401 | spin_lock(&zone->lock); |
1da177e4 LT |
402 | zone->all_unreclaimable = 0; |
403 | zone->pages_scanned = 0; | |
48db57f8 NP |
404 | while (count--) { |
405 | struct page *page; | |
406 | ||
407 | BUG_ON(list_empty(list)); | |
1da177e4 | 408 | page = list_entry(list->prev, struct page, lru); |
48db57f8 | 409 | /* have to delete it as __free_one_page list manipulates */ |
1da177e4 | 410 | list_del(&page->lru); |
48db57f8 | 411 | __free_one_page(page, zone, order); |
1da177e4 | 412 | } |
c54ad30c | 413 | spin_unlock(&zone->lock); |
1da177e4 LT |
414 | } |
415 | ||
48db57f8 | 416 | static void free_one_page(struct zone *zone, struct page *page, int order) |
1da177e4 LT |
417 | { |
418 | LIST_HEAD(list); | |
48db57f8 NP |
419 | list_add(&page->lru, &list); |
420 | free_pages_bulk(zone, 1, &list, order); | |
421 | } | |
422 | ||
423 | static void __free_pages_ok(struct page *page, unsigned int order) | |
424 | { | |
425 | unsigned long flags; | |
1da177e4 | 426 | int i; |
689bcebf | 427 | int reserved = 0; |
1da177e4 LT |
428 | |
429 | arch_free_page(page, order); | |
de5097c2 IM |
430 | if (!PageHighMem(page)) |
431 | mutex_debug_check_no_locks_freed(page_address(page), | |
a4fc7ab1 | 432 | PAGE_SIZE<<order); |
1da177e4 | 433 | |
1da177e4 | 434 | for (i = 0 ; i < (1 << order) ; ++i) |
224abf92 | 435 | reserved += free_pages_check(page + i); |
689bcebf HD |
436 | if (reserved) |
437 | return; | |
438 | ||
48db57f8 | 439 | kernel_map_pages(page, 1 << order, 0); |
c54ad30c | 440 | local_irq_save(flags); |
a74609fa | 441 | __mod_page_state(pgfree, 1 << order); |
48db57f8 | 442 | free_one_page(page_zone(page), page, order); |
c54ad30c | 443 | local_irq_restore(flags); |
1da177e4 LT |
444 | } |
445 | ||
a226f6c8 DH |
446 | /* |
447 | * permit the bootmem allocator to evade page validation on high-order frees | |
448 | */ | |
449 | void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order) | |
450 | { | |
451 | if (order == 0) { | |
452 | __ClearPageReserved(page); | |
453 | set_page_count(page, 0); | |
7835e98b | 454 | set_page_refcounted(page); |
545b1ea9 | 455 | __free_page(page); |
a226f6c8 | 456 | } else { |
a226f6c8 DH |
457 | int loop; |
458 | ||
545b1ea9 | 459 | prefetchw(page); |
a226f6c8 DH |
460 | for (loop = 0; loop < BITS_PER_LONG; loop++) { |
461 | struct page *p = &page[loop]; | |
462 | ||
545b1ea9 NP |
463 | if (loop + 1 < BITS_PER_LONG) |
464 | prefetchw(p + 1); | |
a226f6c8 DH |
465 | __ClearPageReserved(p); |
466 | set_page_count(p, 0); | |
467 | } | |
468 | ||
7835e98b | 469 | set_page_refcounted(page); |
545b1ea9 | 470 | __free_pages(page, order); |
a226f6c8 DH |
471 | } |
472 | } | |
473 | ||
1da177e4 LT |
474 | |
475 | /* | |
476 | * The order of subdivision here is critical for the IO subsystem. | |
477 | * Please do not alter this order without good reasons and regression | |
478 | * testing. Specifically, as large blocks of memory are subdivided, | |
479 | * the order in which smaller blocks are delivered depends on the order | |
480 | * they're subdivided in this function. This is the primary factor | |
481 | * influencing the order in which pages are delivered to the IO | |
482 | * subsystem according to empirical testing, and this is also justified | |
483 | * by considering the behavior of a buddy system containing a single | |
484 | * large block of memory acted on by a series of small allocations. | |
485 | * This behavior is a critical factor in sglist merging's success. | |
486 | * | |
487 | * -- wli | |
488 | */ | |
085cc7d5 | 489 | static inline void expand(struct zone *zone, struct page *page, |
1da177e4 LT |
490 | int low, int high, struct free_area *area) |
491 | { | |
492 | unsigned long size = 1 << high; | |
493 | ||
494 | while (high > low) { | |
495 | area--; | |
496 | high--; | |
497 | size >>= 1; | |
498 | BUG_ON(bad_range(zone, &page[size])); | |
499 | list_add(&page[size].lru, &area->free_list); | |
500 | area->nr_free++; | |
501 | set_page_order(&page[size], high); | |
502 | } | |
1da177e4 LT |
503 | } |
504 | ||
1da177e4 LT |
505 | /* |
506 | * This page is about to be returned from the page allocator | |
507 | */ | |
17cf4406 | 508 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) |
1da177e4 | 509 | { |
92be2e33 NP |
510 | if (unlikely(page_mapcount(page) | |
511 | (page->mapping != NULL) | | |
512 | (page_count(page) != 0) | | |
334795ec HD |
513 | (page->flags & ( |
514 | 1 << PG_lru | | |
1da177e4 LT |
515 | 1 << PG_private | |
516 | 1 << PG_locked | | |
1da177e4 LT |
517 | 1 << PG_active | |
518 | 1 << PG_dirty | | |
519 | 1 << PG_reclaim | | |
334795ec | 520 | 1 << PG_slab | |
1da177e4 | 521 | 1 << PG_swapcache | |
b5810039 | 522 | 1 << PG_writeback | |
92be2e33 | 523 | 1 << PG_reserved )))) |
224abf92 | 524 | bad_page(page); |
1da177e4 | 525 | |
689bcebf HD |
526 | /* |
527 | * For now, we report if PG_reserved was found set, but do not | |
528 | * clear it, and do not allocate the page: as a safety net. | |
529 | */ | |
530 | if (PageReserved(page)) | |
531 | return 1; | |
532 | ||
1da177e4 LT |
533 | page->flags &= ~(1 << PG_uptodate | 1 << PG_error | |
534 | 1 << PG_referenced | 1 << PG_arch_1 | | |
535 | 1 << PG_checked | 1 << PG_mappedtodisk); | |
4c21e2f2 | 536 | set_page_private(page, 0); |
7835e98b | 537 | set_page_refcounted(page); |
1da177e4 | 538 | kernel_map_pages(page, 1 << order, 1); |
17cf4406 NP |
539 | |
540 | if (gfp_flags & __GFP_ZERO) | |
541 | prep_zero_page(page, order, gfp_flags); | |
542 | ||
543 | if (order && (gfp_flags & __GFP_COMP)) | |
544 | prep_compound_page(page, order); | |
545 | ||
689bcebf | 546 | return 0; |
1da177e4 LT |
547 | } |
548 | ||
549 | /* | |
550 | * Do the hard work of removing an element from the buddy allocator. | |
551 | * Call me with the zone->lock already held. | |
552 | */ | |
553 | static struct page *__rmqueue(struct zone *zone, unsigned int order) | |
554 | { | |
555 | struct free_area * area; | |
556 | unsigned int current_order; | |
557 | struct page *page; | |
558 | ||
559 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
560 | area = zone->free_area + current_order; | |
561 | if (list_empty(&area->free_list)) | |
562 | continue; | |
563 | ||
564 | page = list_entry(area->free_list.next, struct page, lru); | |
565 | list_del(&page->lru); | |
566 | rmv_page_order(page); | |
567 | area->nr_free--; | |
568 | zone->free_pages -= 1UL << order; | |
085cc7d5 NP |
569 | expand(zone, page, order, current_order, area); |
570 | return page; | |
1da177e4 LT |
571 | } |
572 | ||
573 | return NULL; | |
574 | } | |
575 | ||
576 | /* | |
577 | * Obtain a specified number of elements from the buddy allocator, all under | |
578 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
579 | * Returns the number of new pages which were placed at *list. | |
580 | */ | |
581 | static int rmqueue_bulk(struct zone *zone, unsigned int order, | |
582 | unsigned long count, struct list_head *list) | |
583 | { | |
1da177e4 | 584 | int i; |
1da177e4 | 585 | |
c54ad30c | 586 | spin_lock(&zone->lock); |
1da177e4 | 587 | for (i = 0; i < count; ++i) { |
085cc7d5 NP |
588 | struct page *page = __rmqueue(zone, order); |
589 | if (unlikely(page == NULL)) | |
1da177e4 | 590 | break; |
1da177e4 LT |
591 | list_add_tail(&page->lru, list); |
592 | } | |
c54ad30c | 593 | spin_unlock(&zone->lock); |
085cc7d5 | 594 | return i; |
1da177e4 LT |
595 | } |
596 | ||
4ae7c039 | 597 | #ifdef CONFIG_NUMA |
8fce4d8e CL |
598 | /* |
599 | * Called from the slab reaper to drain pagesets on a particular node that | |
600 | * belong to the currently executing processor. | |
601 | */ | |
602 | void drain_node_pages(int nodeid) | |
4ae7c039 | 603 | { |
8fce4d8e | 604 | int i, z; |
4ae7c039 CL |
605 | unsigned long flags; |
606 | ||
607 | local_irq_save(flags); | |
8fce4d8e CL |
608 | for (z = 0; z < MAX_NR_ZONES; z++) { |
609 | struct zone *zone = NODE_DATA(nodeid)->node_zones + z; | |
4ae7c039 CL |
610 | struct per_cpu_pageset *pset; |
611 | ||
23316bc8 | 612 | pset = zone_pcp(zone, smp_processor_id()); |
4ae7c039 CL |
613 | for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) { |
614 | struct per_cpu_pages *pcp; | |
615 | ||
616 | pcp = &pset->pcp[i]; | |
48db57f8 NP |
617 | free_pages_bulk(zone, pcp->count, &pcp->list, 0); |
618 | pcp->count = 0; | |
4ae7c039 CL |
619 | } |
620 | } | |
621 | local_irq_restore(flags); | |
622 | } | |
623 | #endif | |
624 | ||
1da177e4 LT |
625 | #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU) |
626 | static void __drain_pages(unsigned int cpu) | |
627 | { | |
c54ad30c | 628 | unsigned long flags; |
1da177e4 LT |
629 | struct zone *zone; |
630 | int i; | |
631 | ||
632 | for_each_zone(zone) { | |
633 | struct per_cpu_pageset *pset; | |
634 | ||
e7c8d5c9 | 635 | pset = zone_pcp(zone, cpu); |
1da177e4 LT |
636 | for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) { |
637 | struct per_cpu_pages *pcp; | |
638 | ||
639 | pcp = &pset->pcp[i]; | |
c54ad30c | 640 | local_irq_save(flags); |
48db57f8 NP |
641 | free_pages_bulk(zone, pcp->count, &pcp->list, 0); |
642 | pcp->count = 0; | |
c54ad30c | 643 | local_irq_restore(flags); |
1da177e4 LT |
644 | } |
645 | } | |
646 | } | |
647 | #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */ | |
648 | ||
649 | #ifdef CONFIG_PM | |
650 | ||
651 | void mark_free_pages(struct zone *zone) | |
652 | { | |
653 | unsigned long zone_pfn, flags; | |
654 | int order; | |
655 | struct list_head *curr; | |
656 | ||
657 | if (!zone->spanned_pages) | |
658 | return; | |
659 | ||
660 | spin_lock_irqsave(&zone->lock, flags); | |
661 | for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) | |
662 | ClearPageNosaveFree(pfn_to_page(zone_pfn + zone->zone_start_pfn)); | |
663 | ||
664 | for (order = MAX_ORDER - 1; order >= 0; --order) | |
665 | list_for_each(curr, &zone->free_area[order].free_list) { | |
666 | unsigned long start_pfn, i; | |
667 | ||
668 | start_pfn = page_to_pfn(list_entry(curr, struct page, lru)); | |
669 | ||
670 | for (i=0; i < (1<<order); i++) | |
671 | SetPageNosaveFree(pfn_to_page(start_pfn+i)); | |
672 | } | |
673 | spin_unlock_irqrestore(&zone->lock, flags); | |
674 | } | |
675 | ||
676 | /* | |
677 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
678 | */ | |
679 | void drain_local_pages(void) | |
680 | { | |
681 | unsigned long flags; | |
682 | ||
683 | local_irq_save(flags); | |
684 | __drain_pages(smp_processor_id()); | |
685 | local_irq_restore(flags); | |
686 | } | |
687 | #endif /* CONFIG_PM */ | |
688 | ||
a74609fa | 689 | static void zone_statistics(struct zonelist *zonelist, struct zone *z, int cpu) |
1da177e4 LT |
690 | { |
691 | #ifdef CONFIG_NUMA | |
1da177e4 LT |
692 | pg_data_t *pg = z->zone_pgdat; |
693 | pg_data_t *orig = zonelist->zones[0]->zone_pgdat; | |
694 | struct per_cpu_pageset *p; | |
695 | ||
a74609fa | 696 | p = zone_pcp(z, cpu); |
1da177e4 | 697 | if (pg == orig) { |
e7c8d5c9 | 698 | p->numa_hit++; |
1da177e4 LT |
699 | } else { |
700 | p->numa_miss++; | |
e7c8d5c9 | 701 | zone_pcp(zonelist->zones[0], cpu)->numa_foreign++; |
1da177e4 LT |
702 | } |
703 | if (pg == NODE_DATA(numa_node_id())) | |
704 | p->local_node++; | |
705 | else | |
706 | p->other_node++; | |
1da177e4 LT |
707 | #endif |
708 | } | |
709 | ||
710 | /* | |
711 | * Free a 0-order page | |
712 | */ | |
1da177e4 LT |
713 | static void fastcall free_hot_cold_page(struct page *page, int cold) |
714 | { | |
715 | struct zone *zone = page_zone(page); | |
716 | struct per_cpu_pages *pcp; | |
717 | unsigned long flags; | |
718 | ||
719 | arch_free_page(page, 0); | |
720 | ||
1da177e4 LT |
721 | if (PageAnon(page)) |
722 | page->mapping = NULL; | |
224abf92 | 723 | if (free_pages_check(page)) |
689bcebf HD |
724 | return; |
725 | ||
689bcebf HD |
726 | kernel_map_pages(page, 1, 0); |
727 | ||
e7c8d5c9 | 728 | pcp = &zone_pcp(zone, get_cpu())->pcp[cold]; |
1da177e4 | 729 | local_irq_save(flags); |
a74609fa | 730 | __inc_page_state(pgfree); |
1da177e4 LT |
731 | list_add(&page->lru, &pcp->list); |
732 | pcp->count++; | |
48db57f8 NP |
733 | if (pcp->count >= pcp->high) { |
734 | free_pages_bulk(zone, pcp->batch, &pcp->list, 0); | |
735 | pcp->count -= pcp->batch; | |
736 | } | |
1da177e4 LT |
737 | local_irq_restore(flags); |
738 | put_cpu(); | |
739 | } | |
740 | ||
741 | void fastcall free_hot_page(struct page *page) | |
742 | { | |
743 | free_hot_cold_page(page, 0); | |
744 | } | |
745 | ||
746 | void fastcall free_cold_page(struct page *page) | |
747 | { | |
748 | free_hot_cold_page(page, 1); | |
749 | } | |
750 | ||
8dfcc9ba NP |
751 | /* |
752 | * split_page takes a non-compound higher-order page, and splits it into | |
753 | * n (1<<order) sub-pages: page[0..n] | |
754 | * Each sub-page must be freed individually. | |
755 | * | |
756 | * Note: this is probably too low level an operation for use in drivers. | |
757 | * Please consult with lkml before using this in your driver. | |
758 | */ | |
759 | void split_page(struct page *page, unsigned int order) | |
760 | { | |
761 | int i; | |
762 | ||
763 | BUG_ON(PageCompound(page)); | |
764 | BUG_ON(!page_count(page)); | |
7835e98b NP |
765 | for (i = 1; i < (1 << order); i++) |
766 | set_page_refcounted(page + i); | |
8dfcc9ba | 767 | } |
8dfcc9ba | 768 | |
1da177e4 LT |
769 | /* |
770 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
771 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
772 | * or two. | |
773 | */ | |
a74609fa NP |
774 | static struct page *buffered_rmqueue(struct zonelist *zonelist, |
775 | struct zone *zone, int order, gfp_t gfp_flags) | |
1da177e4 LT |
776 | { |
777 | unsigned long flags; | |
689bcebf | 778 | struct page *page; |
1da177e4 | 779 | int cold = !!(gfp_flags & __GFP_COLD); |
a74609fa | 780 | int cpu; |
1da177e4 | 781 | |
689bcebf | 782 | again: |
a74609fa | 783 | cpu = get_cpu(); |
48db57f8 | 784 | if (likely(order == 0)) { |
1da177e4 LT |
785 | struct per_cpu_pages *pcp; |
786 | ||
a74609fa | 787 | pcp = &zone_pcp(zone, cpu)->pcp[cold]; |
1da177e4 | 788 | local_irq_save(flags); |
a74609fa | 789 | if (!pcp->count) { |
1da177e4 LT |
790 | pcp->count += rmqueue_bulk(zone, 0, |
791 | pcp->batch, &pcp->list); | |
a74609fa NP |
792 | if (unlikely(!pcp->count)) |
793 | goto failed; | |
1da177e4 | 794 | } |
a74609fa NP |
795 | page = list_entry(pcp->list.next, struct page, lru); |
796 | list_del(&page->lru); | |
797 | pcp->count--; | |
7fb1d9fc | 798 | } else { |
1da177e4 LT |
799 | spin_lock_irqsave(&zone->lock, flags); |
800 | page = __rmqueue(zone, order); | |
a74609fa NP |
801 | spin_unlock(&zone->lock); |
802 | if (!page) | |
803 | goto failed; | |
1da177e4 LT |
804 | } |
805 | ||
a74609fa NP |
806 | __mod_page_state_zone(zone, pgalloc, 1 << order); |
807 | zone_statistics(zonelist, zone, cpu); | |
808 | local_irq_restore(flags); | |
809 | put_cpu(); | |
1da177e4 | 810 | |
a74609fa | 811 | BUG_ON(bad_range(zone, page)); |
17cf4406 | 812 | if (prep_new_page(page, order, gfp_flags)) |
a74609fa | 813 | goto again; |
1da177e4 | 814 | return page; |
a74609fa NP |
815 | |
816 | failed: | |
817 | local_irq_restore(flags); | |
818 | put_cpu(); | |
819 | return NULL; | |
1da177e4 LT |
820 | } |
821 | ||
7fb1d9fc | 822 | #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */ |
3148890b NP |
823 | #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */ |
824 | #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */ | |
825 | #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */ | |
826 | #define ALLOC_HARDER 0x10 /* try to alloc harder */ | |
827 | #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */ | |
828 | #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ | |
7fb1d9fc | 829 | |
1da177e4 LT |
830 | /* |
831 | * Return 1 if free pages are above 'mark'. This takes into account the order | |
832 | * of the allocation. | |
833 | */ | |
834 | int zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
7fb1d9fc | 835 | int classzone_idx, int alloc_flags) |
1da177e4 LT |
836 | { |
837 | /* free_pages my go negative - that's OK */ | |
838 | long min = mark, free_pages = z->free_pages - (1 << order) + 1; | |
839 | int o; | |
840 | ||
7fb1d9fc | 841 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 842 | min -= min / 2; |
7fb1d9fc | 843 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 LT |
844 | min -= min / 4; |
845 | ||
846 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
847 | return 0; | |
848 | for (o = 0; o < order; o++) { | |
849 | /* At the next order, this order's pages become unavailable */ | |
850 | free_pages -= z->free_area[o].nr_free << o; | |
851 | ||
852 | /* Require fewer higher order pages to be free */ | |
853 | min >>= 1; | |
854 | ||
855 | if (free_pages <= min) | |
856 | return 0; | |
857 | } | |
858 | return 1; | |
859 | } | |
860 | ||
7fb1d9fc RS |
861 | /* |
862 | * get_page_from_freeliest goes through the zonelist trying to allocate | |
863 | * a page. | |
864 | */ | |
865 | static struct page * | |
866 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, | |
867 | struct zonelist *zonelist, int alloc_flags) | |
753ee728 | 868 | { |
7fb1d9fc RS |
869 | struct zone **z = zonelist->zones; |
870 | struct page *page = NULL; | |
871 | int classzone_idx = zone_idx(*z); | |
872 | ||
873 | /* | |
874 | * Go through the zonelist once, looking for a zone with enough free. | |
875 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. | |
876 | */ | |
877 | do { | |
878 | if ((alloc_flags & ALLOC_CPUSET) && | |
879 | !cpuset_zone_allowed(*z, gfp_mask)) | |
880 | continue; | |
881 | ||
882 | if (!(alloc_flags & ALLOC_NO_WATERMARKS)) { | |
3148890b NP |
883 | unsigned long mark; |
884 | if (alloc_flags & ALLOC_WMARK_MIN) | |
885 | mark = (*z)->pages_min; | |
886 | else if (alloc_flags & ALLOC_WMARK_LOW) | |
887 | mark = (*z)->pages_low; | |
888 | else | |
889 | mark = (*z)->pages_high; | |
890 | if (!zone_watermark_ok(*z, order, mark, | |
7fb1d9fc | 891 | classzone_idx, alloc_flags)) |
9eeff239 CL |
892 | if (!zone_reclaim_mode || |
893 | !zone_reclaim(*z, gfp_mask, order)) | |
894 | continue; | |
7fb1d9fc RS |
895 | } |
896 | ||
a74609fa | 897 | page = buffered_rmqueue(zonelist, *z, order, gfp_mask); |
7fb1d9fc | 898 | if (page) { |
7fb1d9fc RS |
899 | break; |
900 | } | |
901 | } while (*(++z) != NULL); | |
902 | return page; | |
753ee728 MH |
903 | } |
904 | ||
1da177e4 LT |
905 | /* |
906 | * This is the 'heart' of the zoned buddy allocator. | |
907 | */ | |
908 | struct page * fastcall | |
dd0fc66f | 909 | __alloc_pages(gfp_t gfp_mask, unsigned int order, |
1da177e4 LT |
910 | struct zonelist *zonelist) |
911 | { | |
260b2367 | 912 | const gfp_t wait = gfp_mask & __GFP_WAIT; |
7fb1d9fc | 913 | struct zone **z; |
1da177e4 LT |
914 | struct page *page; |
915 | struct reclaim_state reclaim_state; | |
916 | struct task_struct *p = current; | |
1da177e4 | 917 | int do_retry; |
7fb1d9fc | 918 | int alloc_flags; |
1da177e4 LT |
919 | int did_some_progress; |
920 | ||
921 | might_sleep_if(wait); | |
922 | ||
6b1de916 | 923 | restart: |
7fb1d9fc | 924 | z = zonelist->zones; /* the list of zones suitable for gfp_mask */ |
1da177e4 | 925 | |
7fb1d9fc | 926 | if (unlikely(*z == NULL)) { |
1da177e4 LT |
927 | /* Should this ever happen?? */ |
928 | return NULL; | |
929 | } | |
6b1de916 | 930 | |
7fb1d9fc | 931 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, |
3148890b | 932 | zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET); |
7fb1d9fc RS |
933 | if (page) |
934 | goto got_pg; | |
1da177e4 | 935 | |
6b1de916 | 936 | do { |
7fb1d9fc | 937 | wakeup_kswapd(*z, order); |
6b1de916 | 938 | } while (*(++z)); |
1da177e4 | 939 | |
9bf2229f | 940 | /* |
7fb1d9fc RS |
941 | * OK, we're below the kswapd watermark and have kicked background |
942 | * reclaim. Now things get more complex, so set up alloc_flags according | |
943 | * to how we want to proceed. | |
944 | * | |
945 | * The caller may dip into page reserves a bit more if the caller | |
946 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
4eac915d PJ |
947 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will |
948 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). | |
9bf2229f | 949 | */ |
3148890b | 950 | alloc_flags = ALLOC_WMARK_MIN; |
7fb1d9fc RS |
951 | if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait) |
952 | alloc_flags |= ALLOC_HARDER; | |
953 | if (gfp_mask & __GFP_HIGH) | |
954 | alloc_flags |= ALLOC_HIGH; | |
47f3a867 | 955 | alloc_flags |= ALLOC_CPUSET; |
1da177e4 LT |
956 | |
957 | /* | |
958 | * Go through the zonelist again. Let __GFP_HIGH and allocations | |
7fb1d9fc | 959 | * coming from realtime tasks go deeper into reserves. |
1da177e4 LT |
960 | * |
961 | * This is the last chance, in general, before the goto nopage. | |
962 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. | |
9bf2229f | 963 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1da177e4 | 964 | */ |
7fb1d9fc RS |
965 | page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags); |
966 | if (page) | |
967 | goto got_pg; | |
1da177e4 LT |
968 | |
969 | /* This allocation should allow future memory freeing. */ | |
b84a35be NP |
970 | |
971 | if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE))) | |
972 | && !in_interrupt()) { | |
973 | if (!(gfp_mask & __GFP_NOMEMALLOC)) { | |
885036d3 | 974 | nofail_alloc: |
b84a35be | 975 | /* go through the zonelist yet again, ignoring mins */ |
7fb1d9fc | 976 | page = get_page_from_freelist(gfp_mask, order, |
47f3a867 | 977 | zonelist, ALLOC_NO_WATERMARKS); |
7fb1d9fc RS |
978 | if (page) |
979 | goto got_pg; | |
885036d3 KK |
980 | if (gfp_mask & __GFP_NOFAIL) { |
981 | blk_congestion_wait(WRITE, HZ/50); | |
982 | goto nofail_alloc; | |
983 | } | |
1da177e4 LT |
984 | } |
985 | goto nopage; | |
986 | } | |
987 | ||
988 | /* Atomic allocations - we can't balance anything */ | |
989 | if (!wait) | |
990 | goto nopage; | |
991 | ||
992 | rebalance: | |
993 | cond_resched(); | |
994 | ||
995 | /* We now go into synchronous reclaim */ | |
3e0d98b9 | 996 | cpuset_memory_pressure_bump(); |
1da177e4 LT |
997 | p->flags |= PF_MEMALLOC; |
998 | reclaim_state.reclaimed_slab = 0; | |
999 | p->reclaim_state = &reclaim_state; | |
1000 | ||
7fb1d9fc | 1001 | did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask); |
1da177e4 LT |
1002 | |
1003 | p->reclaim_state = NULL; | |
1004 | p->flags &= ~PF_MEMALLOC; | |
1005 | ||
1006 | cond_resched(); | |
1007 | ||
1008 | if (likely(did_some_progress)) { | |
7fb1d9fc RS |
1009 | page = get_page_from_freelist(gfp_mask, order, |
1010 | zonelist, alloc_flags); | |
1011 | if (page) | |
1012 | goto got_pg; | |
1da177e4 LT |
1013 | } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) { |
1014 | /* | |
1015 | * Go through the zonelist yet one more time, keep | |
1016 | * very high watermark here, this is only to catch | |
1017 | * a parallel oom killing, we must fail if we're still | |
1018 | * under heavy pressure. | |
1019 | */ | |
7fb1d9fc | 1020 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, |
3148890b | 1021 | zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET); |
7fb1d9fc RS |
1022 | if (page) |
1023 | goto got_pg; | |
1da177e4 | 1024 | |
9b0f8b04 | 1025 | out_of_memory(zonelist, gfp_mask, order); |
1da177e4 LT |
1026 | goto restart; |
1027 | } | |
1028 | ||
1029 | /* | |
1030 | * Don't let big-order allocations loop unless the caller explicitly | |
1031 | * requests that. Wait for some write requests to complete then retry. | |
1032 | * | |
1033 | * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order | |
1034 | * <= 3, but that may not be true in other implementations. | |
1035 | */ | |
1036 | do_retry = 0; | |
1037 | if (!(gfp_mask & __GFP_NORETRY)) { | |
1038 | if ((order <= 3) || (gfp_mask & __GFP_REPEAT)) | |
1039 | do_retry = 1; | |
1040 | if (gfp_mask & __GFP_NOFAIL) | |
1041 | do_retry = 1; | |
1042 | } | |
1043 | if (do_retry) { | |
1044 | blk_congestion_wait(WRITE, HZ/50); | |
1045 | goto rebalance; | |
1046 | } | |
1047 | ||
1048 | nopage: | |
1049 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) { | |
1050 | printk(KERN_WARNING "%s: page allocation failure." | |
1051 | " order:%d, mode:0x%x\n", | |
1052 | p->comm, order, gfp_mask); | |
1053 | dump_stack(); | |
578c2fd6 | 1054 | show_mem(); |
1da177e4 | 1055 | } |
1da177e4 | 1056 | got_pg: |
1da177e4 LT |
1057 | return page; |
1058 | } | |
1059 | ||
1060 | EXPORT_SYMBOL(__alloc_pages); | |
1061 | ||
1062 | /* | |
1063 | * Common helper functions. | |
1064 | */ | |
dd0fc66f | 1065 | fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 LT |
1066 | { |
1067 | struct page * page; | |
1068 | page = alloc_pages(gfp_mask, order); | |
1069 | if (!page) | |
1070 | return 0; | |
1071 | return (unsigned long) page_address(page); | |
1072 | } | |
1073 | ||
1074 | EXPORT_SYMBOL(__get_free_pages); | |
1075 | ||
dd0fc66f | 1076 | fastcall unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 LT |
1077 | { |
1078 | struct page * page; | |
1079 | ||
1080 | /* | |
1081 | * get_zeroed_page() returns a 32-bit address, which cannot represent | |
1082 | * a highmem page | |
1083 | */ | |
260b2367 | 1084 | BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); |
1da177e4 LT |
1085 | |
1086 | page = alloc_pages(gfp_mask | __GFP_ZERO, 0); | |
1087 | if (page) | |
1088 | return (unsigned long) page_address(page); | |
1089 | return 0; | |
1090 | } | |
1091 | ||
1092 | EXPORT_SYMBOL(get_zeroed_page); | |
1093 | ||
1094 | void __pagevec_free(struct pagevec *pvec) | |
1095 | { | |
1096 | int i = pagevec_count(pvec); | |
1097 | ||
1098 | while (--i >= 0) | |
1099 | free_hot_cold_page(pvec->pages[i], pvec->cold); | |
1100 | } | |
1101 | ||
1102 | fastcall void __free_pages(struct page *page, unsigned int order) | |
1103 | { | |
b5810039 | 1104 | if (put_page_testzero(page)) { |
1da177e4 LT |
1105 | if (order == 0) |
1106 | free_hot_page(page); | |
1107 | else | |
1108 | __free_pages_ok(page, order); | |
1109 | } | |
1110 | } | |
1111 | ||
1112 | EXPORT_SYMBOL(__free_pages); | |
1113 | ||
1114 | fastcall void free_pages(unsigned long addr, unsigned int order) | |
1115 | { | |
1116 | if (addr != 0) { | |
1117 | BUG_ON(!virt_addr_valid((void *)addr)); | |
1118 | __free_pages(virt_to_page((void *)addr), order); | |
1119 | } | |
1120 | } | |
1121 | ||
1122 | EXPORT_SYMBOL(free_pages); | |
1123 | ||
1124 | /* | |
1125 | * Total amount of free (allocatable) RAM: | |
1126 | */ | |
1127 | unsigned int nr_free_pages(void) | |
1128 | { | |
1129 | unsigned int sum = 0; | |
1130 | struct zone *zone; | |
1131 | ||
1132 | for_each_zone(zone) | |
1133 | sum += zone->free_pages; | |
1134 | ||
1135 | return sum; | |
1136 | } | |
1137 | ||
1138 | EXPORT_SYMBOL(nr_free_pages); | |
1139 | ||
1140 | #ifdef CONFIG_NUMA | |
1141 | unsigned int nr_free_pages_pgdat(pg_data_t *pgdat) | |
1142 | { | |
1143 | unsigned int i, sum = 0; | |
1144 | ||
1145 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1146 | sum += pgdat->node_zones[i].free_pages; | |
1147 | ||
1148 | return sum; | |
1149 | } | |
1150 | #endif | |
1151 | ||
1152 | static unsigned int nr_free_zone_pages(int offset) | |
1153 | { | |
e310fd43 MB |
1154 | /* Just pick one node, since fallback list is circular */ |
1155 | pg_data_t *pgdat = NODE_DATA(numa_node_id()); | |
1da177e4 LT |
1156 | unsigned int sum = 0; |
1157 | ||
e310fd43 MB |
1158 | struct zonelist *zonelist = pgdat->node_zonelists + offset; |
1159 | struct zone **zonep = zonelist->zones; | |
1160 | struct zone *zone; | |
1da177e4 | 1161 | |
e310fd43 MB |
1162 | for (zone = *zonep++; zone; zone = *zonep++) { |
1163 | unsigned long size = zone->present_pages; | |
1164 | unsigned long high = zone->pages_high; | |
1165 | if (size > high) | |
1166 | sum += size - high; | |
1da177e4 LT |
1167 | } |
1168 | ||
1169 | return sum; | |
1170 | } | |
1171 | ||
1172 | /* | |
1173 | * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL | |
1174 | */ | |
1175 | unsigned int nr_free_buffer_pages(void) | |
1176 | { | |
af4ca457 | 1177 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 LT |
1178 | } |
1179 | ||
1180 | /* | |
1181 | * Amount of free RAM allocatable within all zones | |
1182 | */ | |
1183 | unsigned int nr_free_pagecache_pages(void) | |
1184 | { | |
af4ca457 | 1185 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER)); |
1da177e4 LT |
1186 | } |
1187 | ||
1188 | #ifdef CONFIG_HIGHMEM | |
1189 | unsigned int nr_free_highpages (void) | |
1190 | { | |
1191 | pg_data_t *pgdat; | |
1192 | unsigned int pages = 0; | |
1193 | ||
1194 | for_each_pgdat(pgdat) | |
1195 | pages += pgdat->node_zones[ZONE_HIGHMEM].free_pages; | |
1196 | ||
1197 | return pages; | |
1198 | } | |
1199 | #endif | |
1200 | ||
1201 | #ifdef CONFIG_NUMA | |
1202 | static void show_node(struct zone *zone) | |
1203 | { | |
1204 | printk("Node %d ", zone->zone_pgdat->node_id); | |
1205 | } | |
1206 | #else | |
1207 | #define show_node(zone) do { } while (0) | |
1208 | #endif | |
1209 | ||
1210 | /* | |
1211 | * Accumulate the page_state information across all CPUs. | |
1212 | * The result is unavoidably approximate - it can change | |
1213 | * during and after execution of this function. | |
1214 | */ | |
1215 | static DEFINE_PER_CPU(struct page_state, page_states) = {0}; | |
1216 | ||
1217 | atomic_t nr_pagecache = ATOMIC_INIT(0); | |
1218 | EXPORT_SYMBOL(nr_pagecache); | |
1219 | #ifdef CONFIG_SMP | |
1220 | DEFINE_PER_CPU(long, nr_pagecache_local) = 0; | |
1221 | #endif | |
1222 | ||
a86b1f53 | 1223 | static void __get_page_state(struct page_state *ret, int nr, cpumask_t *cpumask) |
1da177e4 | 1224 | { |
b40607fc | 1225 | unsigned cpu; |
1da177e4 | 1226 | |
88a2a4ac | 1227 | memset(ret, 0, nr * sizeof(unsigned long)); |
84c2008a | 1228 | cpus_and(*cpumask, *cpumask, cpu_online_map); |
1da177e4 | 1229 | |
b40607fc AM |
1230 | for_each_cpu_mask(cpu, *cpumask) { |
1231 | unsigned long *in; | |
1232 | unsigned long *out; | |
1233 | unsigned off; | |
1234 | unsigned next_cpu; | |
88a2a4ac | 1235 | |
1da177e4 LT |
1236 | in = (unsigned long *)&per_cpu(page_states, cpu); |
1237 | ||
b40607fc AM |
1238 | next_cpu = next_cpu(cpu, *cpumask); |
1239 | if (likely(next_cpu < NR_CPUS)) | |
1240 | prefetch(&per_cpu(page_states, next_cpu)); | |
1da177e4 LT |
1241 | |
1242 | out = (unsigned long *)ret; | |
1243 | for (off = 0; off < nr; off++) | |
1244 | *out++ += *in++; | |
1245 | } | |
1246 | } | |
1247 | ||
c07e02db MH |
1248 | void get_page_state_node(struct page_state *ret, int node) |
1249 | { | |
1250 | int nr; | |
1251 | cpumask_t mask = node_to_cpumask(node); | |
1252 | ||
1253 | nr = offsetof(struct page_state, GET_PAGE_STATE_LAST); | |
1254 | nr /= sizeof(unsigned long); | |
1255 | ||
1256 | __get_page_state(ret, nr+1, &mask); | |
1257 | } | |
1258 | ||
1da177e4 LT |
1259 | void get_page_state(struct page_state *ret) |
1260 | { | |
1261 | int nr; | |
c07e02db | 1262 | cpumask_t mask = CPU_MASK_ALL; |
1da177e4 LT |
1263 | |
1264 | nr = offsetof(struct page_state, GET_PAGE_STATE_LAST); | |
1265 | nr /= sizeof(unsigned long); | |
1266 | ||
c07e02db | 1267 | __get_page_state(ret, nr + 1, &mask); |
1da177e4 LT |
1268 | } |
1269 | ||
1270 | void get_full_page_state(struct page_state *ret) | |
1271 | { | |
c07e02db MH |
1272 | cpumask_t mask = CPU_MASK_ALL; |
1273 | ||
1274 | __get_page_state(ret, sizeof(*ret) / sizeof(unsigned long), &mask); | |
1da177e4 LT |
1275 | } |
1276 | ||
a74609fa | 1277 | unsigned long read_page_state_offset(unsigned long offset) |
1da177e4 LT |
1278 | { |
1279 | unsigned long ret = 0; | |
1280 | int cpu; | |
1281 | ||
84c2008a | 1282 | for_each_online_cpu(cpu) { |
1da177e4 LT |
1283 | unsigned long in; |
1284 | ||
1285 | in = (unsigned long)&per_cpu(page_states, cpu) + offset; | |
1286 | ret += *((unsigned long *)in); | |
1287 | } | |
1288 | return ret; | |
1289 | } | |
1290 | ||
a74609fa NP |
1291 | void __mod_page_state_offset(unsigned long offset, unsigned long delta) |
1292 | { | |
1293 | void *ptr; | |
1294 | ||
1295 | ptr = &__get_cpu_var(page_states); | |
1296 | *(unsigned long *)(ptr + offset) += delta; | |
1297 | } | |
1298 | EXPORT_SYMBOL(__mod_page_state_offset); | |
1299 | ||
1300 | void mod_page_state_offset(unsigned long offset, unsigned long delta) | |
1da177e4 LT |
1301 | { |
1302 | unsigned long flags; | |
a74609fa | 1303 | void *ptr; |
1da177e4 LT |
1304 | |
1305 | local_irq_save(flags); | |
1306 | ptr = &__get_cpu_var(page_states); | |
a74609fa | 1307 | *(unsigned long *)(ptr + offset) += delta; |
1da177e4 LT |
1308 | local_irq_restore(flags); |
1309 | } | |
a74609fa | 1310 | EXPORT_SYMBOL(mod_page_state_offset); |
1da177e4 LT |
1311 | |
1312 | void __get_zone_counts(unsigned long *active, unsigned long *inactive, | |
1313 | unsigned long *free, struct pglist_data *pgdat) | |
1314 | { | |
1315 | struct zone *zones = pgdat->node_zones; | |
1316 | int i; | |
1317 | ||
1318 | *active = 0; | |
1319 | *inactive = 0; | |
1320 | *free = 0; | |
1321 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
1322 | *active += zones[i].nr_active; | |
1323 | *inactive += zones[i].nr_inactive; | |
1324 | *free += zones[i].free_pages; | |
1325 | } | |
1326 | } | |
1327 | ||
1328 | void get_zone_counts(unsigned long *active, | |
1329 | unsigned long *inactive, unsigned long *free) | |
1330 | { | |
1331 | struct pglist_data *pgdat; | |
1332 | ||
1333 | *active = 0; | |
1334 | *inactive = 0; | |
1335 | *free = 0; | |
1336 | for_each_pgdat(pgdat) { | |
1337 | unsigned long l, m, n; | |
1338 | __get_zone_counts(&l, &m, &n, pgdat); | |
1339 | *active += l; | |
1340 | *inactive += m; | |
1341 | *free += n; | |
1342 | } | |
1343 | } | |
1344 | ||
1345 | void si_meminfo(struct sysinfo *val) | |
1346 | { | |
1347 | val->totalram = totalram_pages; | |
1348 | val->sharedram = 0; | |
1349 | val->freeram = nr_free_pages(); | |
1350 | val->bufferram = nr_blockdev_pages(); | |
1351 | #ifdef CONFIG_HIGHMEM | |
1352 | val->totalhigh = totalhigh_pages; | |
1353 | val->freehigh = nr_free_highpages(); | |
1354 | #else | |
1355 | val->totalhigh = 0; | |
1356 | val->freehigh = 0; | |
1357 | #endif | |
1358 | val->mem_unit = PAGE_SIZE; | |
1359 | } | |
1360 | ||
1361 | EXPORT_SYMBOL(si_meminfo); | |
1362 | ||
1363 | #ifdef CONFIG_NUMA | |
1364 | void si_meminfo_node(struct sysinfo *val, int nid) | |
1365 | { | |
1366 | pg_data_t *pgdat = NODE_DATA(nid); | |
1367 | ||
1368 | val->totalram = pgdat->node_present_pages; | |
1369 | val->freeram = nr_free_pages_pgdat(pgdat); | |
1370 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages; | |
1371 | val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages; | |
1372 | val->mem_unit = PAGE_SIZE; | |
1373 | } | |
1374 | #endif | |
1375 | ||
1376 | #define K(x) ((x) << (PAGE_SHIFT-10)) | |
1377 | ||
1378 | /* | |
1379 | * Show free area list (used inside shift_scroll-lock stuff) | |
1380 | * We also calculate the percentage fragmentation. We do this by counting the | |
1381 | * memory on each free list with the exception of the first item on the list. | |
1382 | */ | |
1383 | void show_free_areas(void) | |
1384 | { | |
1385 | struct page_state ps; | |
1386 | int cpu, temperature; | |
1387 | unsigned long active; | |
1388 | unsigned long inactive; | |
1389 | unsigned long free; | |
1390 | struct zone *zone; | |
1391 | ||
1392 | for_each_zone(zone) { | |
1393 | show_node(zone); | |
1394 | printk("%s per-cpu:", zone->name); | |
1395 | ||
f3fe6512 | 1396 | if (!populated_zone(zone)) { |
1da177e4 LT |
1397 | printk(" empty\n"); |
1398 | continue; | |
1399 | } else | |
1400 | printk("\n"); | |
1401 | ||
6b482c67 | 1402 | for_each_online_cpu(cpu) { |
1da177e4 LT |
1403 | struct per_cpu_pageset *pageset; |
1404 | ||
e7c8d5c9 | 1405 | pageset = zone_pcp(zone, cpu); |
1da177e4 LT |
1406 | |
1407 | for (temperature = 0; temperature < 2; temperature++) | |
2d92c5c9 | 1408 | printk("cpu %d %s: high %d, batch %d used:%d\n", |
1da177e4 LT |
1409 | cpu, |
1410 | temperature ? "cold" : "hot", | |
1da177e4 | 1411 | pageset->pcp[temperature].high, |
4ae7c039 CL |
1412 | pageset->pcp[temperature].batch, |
1413 | pageset->pcp[temperature].count); | |
1da177e4 LT |
1414 | } |
1415 | } | |
1416 | ||
1417 | get_page_state(&ps); | |
1418 | get_zone_counts(&active, &inactive, &free); | |
1419 | ||
c0d62219 | 1420 | printk("Free pages: %11ukB (%ukB HighMem)\n", |
1da177e4 LT |
1421 | K(nr_free_pages()), |
1422 | K(nr_free_highpages())); | |
1423 | ||
1424 | printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu " | |
1425 | "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n", | |
1426 | active, | |
1427 | inactive, | |
1428 | ps.nr_dirty, | |
1429 | ps.nr_writeback, | |
1430 | ps.nr_unstable, | |
1431 | nr_free_pages(), | |
1432 | ps.nr_slab, | |
1433 | ps.nr_mapped, | |
1434 | ps.nr_page_table_pages); | |
1435 | ||
1436 | for_each_zone(zone) { | |
1437 | int i; | |
1438 | ||
1439 | show_node(zone); | |
1440 | printk("%s" | |
1441 | " free:%lukB" | |
1442 | " min:%lukB" | |
1443 | " low:%lukB" | |
1444 | " high:%lukB" | |
1445 | " active:%lukB" | |
1446 | " inactive:%lukB" | |
1447 | " present:%lukB" | |
1448 | " pages_scanned:%lu" | |
1449 | " all_unreclaimable? %s" | |
1450 | "\n", | |
1451 | zone->name, | |
1452 | K(zone->free_pages), | |
1453 | K(zone->pages_min), | |
1454 | K(zone->pages_low), | |
1455 | K(zone->pages_high), | |
1456 | K(zone->nr_active), | |
1457 | K(zone->nr_inactive), | |
1458 | K(zone->present_pages), | |
1459 | zone->pages_scanned, | |
1460 | (zone->all_unreclaimable ? "yes" : "no") | |
1461 | ); | |
1462 | printk("lowmem_reserve[]:"); | |
1463 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1464 | printk(" %lu", zone->lowmem_reserve[i]); | |
1465 | printk("\n"); | |
1466 | } | |
1467 | ||
1468 | for_each_zone(zone) { | |
1469 | unsigned long nr, flags, order, total = 0; | |
1470 | ||
1471 | show_node(zone); | |
1472 | printk("%s: ", zone->name); | |
f3fe6512 | 1473 | if (!populated_zone(zone)) { |
1da177e4 LT |
1474 | printk("empty\n"); |
1475 | continue; | |
1476 | } | |
1477 | ||
1478 | spin_lock_irqsave(&zone->lock, flags); | |
1479 | for (order = 0; order < MAX_ORDER; order++) { | |
1480 | nr = zone->free_area[order].nr_free; | |
1481 | total += nr << order; | |
1482 | printk("%lu*%lukB ", nr, K(1UL) << order); | |
1483 | } | |
1484 | spin_unlock_irqrestore(&zone->lock, flags); | |
1485 | printk("= %lukB\n", K(total)); | |
1486 | } | |
1487 | ||
1488 | show_swap_cache_info(); | |
1489 | } | |
1490 | ||
1491 | /* | |
1492 | * Builds allocation fallback zone lists. | |
1a93205b CL |
1493 | * |
1494 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 1495 | */ |
1a93205b | 1496 | static int __init build_zonelists_node(pg_data_t *pgdat, |
070f8032 | 1497 | struct zonelist *zonelist, int nr_zones, int zone_type) |
1da177e4 | 1498 | { |
1a93205b CL |
1499 | struct zone *zone; |
1500 | ||
070f8032 | 1501 | BUG_ON(zone_type > ZONE_HIGHMEM); |
02a68a5e CL |
1502 | |
1503 | do { | |
070f8032 | 1504 | zone = pgdat->node_zones + zone_type; |
1a93205b | 1505 | if (populated_zone(zone)) { |
1da177e4 | 1506 | #ifndef CONFIG_HIGHMEM |
070f8032 | 1507 | BUG_ON(zone_type > ZONE_NORMAL); |
1da177e4 | 1508 | #endif |
070f8032 CL |
1509 | zonelist->zones[nr_zones++] = zone; |
1510 | check_highest_zone(zone_type); | |
1da177e4 | 1511 | } |
070f8032 | 1512 | zone_type--; |
02a68a5e | 1513 | |
070f8032 CL |
1514 | } while (zone_type >= 0); |
1515 | return nr_zones; | |
1da177e4 LT |
1516 | } |
1517 | ||
260b2367 AV |
1518 | static inline int highest_zone(int zone_bits) |
1519 | { | |
1520 | int res = ZONE_NORMAL; | |
1521 | if (zone_bits & (__force int)__GFP_HIGHMEM) | |
1522 | res = ZONE_HIGHMEM; | |
a2f1b424 AK |
1523 | if (zone_bits & (__force int)__GFP_DMA32) |
1524 | res = ZONE_DMA32; | |
260b2367 AV |
1525 | if (zone_bits & (__force int)__GFP_DMA) |
1526 | res = ZONE_DMA; | |
1527 | return res; | |
1528 | } | |
1529 | ||
1da177e4 LT |
1530 | #ifdef CONFIG_NUMA |
1531 | #define MAX_NODE_LOAD (num_online_nodes()) | |
1532 | static int __initdata node_load[MAX_NUMNODES]; | |
1533 | /** | |
4dc3b16b | 1534 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
1535 | * @node: node whose fallback list we're appending |
1536 | * @used_node_mask: nodemask_t of already used nodes | |
1537 | * | |
1538 | * We use a number of factors to determine which is the next node that should | |
1539 | * appear on a given node's fallback list. The node should not have appeared | |
1540 | * already in @node's fallback list, and it should be the next closest node | |
1541 | * according to the distance array (which contains arbitrary distance values | |
1542 | * from each node to each node in the system), and should also prefer nodes | |
1543 | * with no CPUs, since presumably they'll have very little allocation pressure | |
1544 | * on them otherwise. | |
1545 | * It returns -1 if no node is found. | |
1546 | */ | |
1547 | static int __init find_next_best_node(int node, nodemask_t *used_node_mask) | |
1548 | { | |
4cf808eb | 1549 | int n, val; |
1da177e4 LT |
1550 | int min_val = INT_MAX; |
1551 | int best_node = -1; | |
1552 | ||
4cf808eb LT |
1553 | /* Use the local node if we haven't already */ |
1554 | if (!node_isset(node, *used_node_mask)) { | |
1555 | node_set(node, *used_node_mask); | |
1556 | return node; | |
1557 | } | |
1da177e4 | 1558 | |
4cf808eb LT |
1559 | for_each_online_node(n) { |
1560 | cpumask_t tmp; | |
1da177e4 LT |
1561 | |
1562 | /* Don't want a node to appear more than once */ | |
1563 | if (node_isset(n, *used_node_mask)) | |
1564 | continue; | |
1565 | ||
1da177e4 LT |
1566 | /* Use the distance array to find the distance */ |
1567 | val = node_distance(node, n); | |
1568 | ||
4cf808eb LT |
1569 | /* Penalize nodes under us ("prefer the next node") */ |
1570 | val += (n < node); | |
1571 | ||
1da177e4 LT |
1572 | /* Give preference to headless and unused nodes */ |
1573 | tmp = node_to_cpumask(n); | |
1574 | if (!cpus_empty(tmp)) | |
1575 | val += PENALTY_FOR_NODE_WITH_CPUS; | |
1576 | ||
1577 | /* Slight preference for less loaded node */ | |
1578 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
1579 | val += node_load[n]; | |
1580 | ||
1581 | if (val < min_val) { | |
1582 | min_val = val; | |
1583 | best_node = n; | |
1584 | } | |
1585 | } | |
1586 | ||
1587 | if (best_node >= 0) | |
1588 | node_set(best_node, *used_node_mask); | |
1589 | ||
1590 | return best_node; | |
1591 | } | |
1592 | ||
1593 | static void __init build_zonelists(pg_data_t *pgdat) | |
1594 | { | |
1595 | int i, j, k, node, local_node; | |
1596 | int prev_node, load; | |
1597 | struct zonelist *zonelist; | |
1598 | nodemask_t used_mask; | |
1599 | ||
1600 | /* initialize zonelists */ | |
1601 | for (i = 0; i < GFP_ZONETYPES; i++) { | |
1602 | zonelist = pgdat->node_zonelists + i; | |
1603 | zonelist->zones[0] = NULL; | |
1604 | } | |
1605 | ||
1606 | /* NUMA-aware ordering of nodes */ | |
1607 | local_node = pgdat->node_id; | |
1608 | load = num_online_nodes(); | |
1609 | prev_node = local_node; | |
1610 | nodes_clear(used_mask); | |
1611 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { | |
9eeff239 CL |
1612 | int distance = node_distance(local_node, node); |
1613 | ||
1614 | /* | |
1615 | * If another node is sufficiently far away then it is better | |
1616 | * to reclaim pages in a zone before going off node. | |
1617 | */ | |
1618 | if (distance > RECLAIM_DISTANCE) | |
1619 | zone_reclaim_mode = 1; | |
1620 | ||
1da177e4 LT |
1621 | /* |
1622 | * We don't want to pressure a particular node. | |
1623 | * So adding penalty to the first node in same | |
1624 | * distance group to make it round-robin. | |
1625 | */ | |
9eeff239 CL |
1626 | |
1627 | if (distance != node_distance(local_node, prev_node)) | |
1da177e4 LT |
1628 | node_load[node] += load; |
1629 | prev_node = node; | |
1630 | load--; | |
1631 | for (i = 0; i < GFP_ZONETYPES; i++) { | |
1632 | zonelist = pgdat->node_zonelists + i; | |
1633 | for (j = 0; zonelist->zones[j] != NULL; j++); | |
1634 | ||
260b2367 | 1635 | k = highest_zone(i); |
1da177e4 LT |
1636 | |
1637 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, k); | |
1638 | zonelist->zones[j] = NULL; | |
1639 | } | |
1640 | } | |
1641 | } | |
1642 | ||
1643 | #else /* CONFIG_NUMA */ | |
1644 | ||
1645 | static void __init build_zonelists(pg_data_t *pgdat) | |
1646 | { | |
1647 | int i, j, k, node, local_node; | |
1648 | ||
1649 | local_node = pgdat->node_id; | |
1650 | for (i = 0; i < GFP_ZONETYPES; i++) { | |
1651 | struct zonelist *zonelist; | |
1652 | ||
1653 | zonelist = pgdat->node_zonelists + i; | |
1654 | ||
1655 | j = 0; | |
260b2367 | 1656 | k = highest_zone(i); |
1da177e4 LT |
1657 | j = build_zonelists_node(pgdat, zonelist, j, k); |
1658 | /* | |
1659 | * Now we build the zonelist so that it contains the zones | |
1660 | * of all the other nodes. | |
1661 | * We don't want to pressure a particular node, so when | |
1662 | * building the zones for node N, we make sure that the | |
1663 | * zones coming right after the local ones are those from | |
1664 | * node N+1 (modulo N) | |
1665 | */ | |
1666 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
1667 | if (!node_online(node)) | |
1668 | continue; | |
1669 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, k); | |
1670 | } | |
1671 | for (node = 0; node < local_node; node++) { | |
1672 | if (!node_online(node)) | |
1673 | continue; | |
1674 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, k); | |
1675 | } | |
1676 | ||
1677 | zonelist->zones[j] = NULL; | |
1678 | } | |
1679 | } | |
1680 | ||
1681 | #endif /* CONFIG_NUMA */ | |
1682 | ||
1683 | void __init build_all_zonelists(void) | |
1684 | { | |
1685 | int i; | |
1686 | ||
1687 | for_each_online_node(i) | |
1688 | build_zonelists(NODE_DATA(i)); | |
1689 | printk("Built %i zonelists\n", num_online_nodes()); | |
1690 | cpuset_init_current_mems_allowed(); | |
1691 | } | |
1692 | ||
1693 | /* | |
1694 | * Helper functions to size the waitqueue hash table. | |
1695 | * Essentially these want to choose hash table sizes sufficiently | |
1696 | * large so that collisions trying to wait on pages are rare. | |
1697 | * But in fact, the number of active page waitqueues on typical | |
1698 | * systems is ridiculously low, less than 200. So this is even | |
1699 | * conservative, even though it seems large. | |
1700 | * | |
1701 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
1702 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
1703 | */ | |
1704 | #define PAGES_PER_WAITQUEUE 256 | |
1705 | ||
1706 | static inline unsigned long wait_table_size(unsigned long pages) | |
1707 | { | |
1708 | unsigned long size = 1; | |
1709 | ||
1710 | pages /= PAGES_PER_WAITQUEUE; | |
1711 | ||
1712 | while (size < pages) | |
1713 | size <<= 1; | |
1714 | ||
1715 | /* | |
1716 | * Once we have dozens or even hundreds of threads sleeping | |
1717 | * on IO we've got bigger problems than wait queue collision. | |
1718 | * Limit the size of the wait table to a reasonable size. | |
1719 | */ | |
1720 | size = min(size, 4096UL); | |
1721 | ||
1722 | return max(size, 4UL); | |
1723 | } | |
1724 | ||
1725 | /* | |
1726 | * This is an integer logarithm so that shifts can be used later | |
1727 | * to extract the more random high bits from the multiplicative | |
1728 | * hash function before the remainder is taken. | |
1729 | */ | |
1730 | static inline unsigned long wait_table_bits(unsigned long size) | |
1731 | { | |
1732 | return ffz(~size); | |
1733 | } | |
1734 | ||
1735 | #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1)) | |
1736 | ||
1737 | static void __init calculate_zone_totalpages(struct pglist_data *pgdat, | |
1738 | unsigned long *zones_size, unsigned long *zholes_size) | |
1739 | { | |
1740 | unsigned long realtotalpages, totalpages = 0; | |
1741 | int i; | |
1742 | ||
1743 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1744 | totalpages += zones_size[i]; | |
1745 | pgdat->node_spanned_pages = totalpages; | |
1746 | ||
1747 | realtotalpages = totalpages; | |
1748 | if (zholes_size) | |
1749 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1750 | realtotalpages -= zholes_size[i]; | |
1751 | pgdat->node_present_pages = realtotalpages; | |
1752 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages); | |
1753 | } | |
1754 | ||
1755 | ||
1756 | /* | |
1757 | * Initially all pages are reserved - free ones are freed | |
1758 | * up by free_all_bootmem() once the early boot process is | |
1759 | * done. Non-atomic initialization, single-pass. | |
1760 | */ | |
c09b4240 | 1761 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
1da177e4 LT |
1762 | unsigned long start_pfn) |
1763 | { | |
1da177e4 | 1764 | struct page *page; |
29751f69 AW |
1765 | unsigned long end_pfn = start_pfn + size; |
1766 | unsigned long pfn; | |
1da177e4 | 1767 | |
cbe8dd4a | 1768 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
d41dee36 AW |
1769 | if (!early_pfn_valid(pfn)) |
1770 | continue; | |
1771 | page = pfn_to_page(pfn); | |
1772 | set_page_links(page, zone, nid, pfn); | |
7835e98b | 1773 | init_page_count(page); |
1da177e4 LT |
1774 | reset_page_mapcount(page); |
1775 | SetPageReserved(page); | |
1776 | INIT_LIST_HEAD(&page->lru); | |
1777 | #ifdef WANT_PAGE_VIRTUAL | |
1778 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
1779 | if (!is_highmem_idx(zone)) | |
3212c6be | 1780 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1da177e4 | 1781 | #endif |
1da177e4 LT |
1782 | } |
1783 | } | |
1784 | ||
1785 | void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone, | |
1786 | unsigned long size) | |
1787 | { | |
1788 | int order; | |
1789 | for (order = 0; order < MAX_ORDER ; order++) { | |
1790 | INIT_LIST_HEAD(&zone->free_area[order].free_list); | |
1791 | zone->free_area[order].nr_free = 0; | |
1792 | } | |
1793 | } | |
1794 | ||
d41dee36 AW |
1795 | #define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr) |
1796 | void zonetable_add(struct zone *zone, int nid, int zid, unsigned long pfn, | |
1797 | unsigned long size) | |
1798 | { | |
1799 | unsigned long snum = pfn_to_section_nr(pfn); | |
1800 | unsigned long end = pfn_to_section_nr(pfn + size); | |
1801 | ||
1802 | if (FLAGS_HAS_NODE) | |
1803 | zone_table[ZONETABLE_INDEX(nid, zid)] = zone; | |
1804 | else | |
1805 | for (; snum <= end; snum++) | |
1806 | zone_table[ZONETABLE_INDEX(snum, zid)] = zone; | |
1807 | } | |
1808 | ||
1da177e4 LT |
1809 | #ifndef __HAVE_ARCH_MEMMAP_INIT |
1810 | #define memmap_init(size, nid, zone, start_pfn) \ | |
1811 | memmap_init_zone((size), (nid), (zone), (start_pfn)) | |
1812 | #endif | |
1813 | ||
6292d9aa | 1814 | static int __cpuinit zone_batchsize(struct zone *zone) |
e7c8d5c9 CL |
1815 | { |
1816 | int batch; | |
1817 | ||
1818 | /* | |
1819 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 1820 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
1821 | * |
1822 | * OK, so we don't know how big the cache is. So guess. | |
1823 | */ | |
1824 | batch = zone->present_pages / 1024; | |
ba56e91c SR |
1825 | if (batch * PAGE_SIZE > 512 * 1024) |
1826 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
1827 | batch /= 4; /* We effectively *= 4 below */ |
1828 | if (batch < 1) | |
1829 | batch = 1; | |
1830 | ||
1831 | /* | |
0ceaacc9 NP |
1832 | * Clamp the batch to a 2^n - 1 value. Having a power |
1833 | * of 2 value was found to be more likely to have | |
1834 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 1835 | * |
0ceaacc9 NP |
1836 | * For example if 2 tasks are alternately allocating |
1837 | * batches of pages, one task can end up with a lot | |
1838 | * of pages of one half of the possible page colors | |
1839 | * and the other with pages of the other colors. | |
e7c8d5c9 | 1840 | */ |
0ceaacc9 | 1841 | batch = (1 << (fls(batch + batch/2)-1)) - 1; |
ba56e91c | 1842 | |
e7c8d5c9 CL |
1843 | return batch; |
1844 | } | |
1845 | ||
2caaad41 CL |
1846 | inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
1847 | { | |
1848 | struct per_cpu_pages *pcp; | |
1849 | ||
1c6fe946 MD |
1850 | memset(p, 0, sizeof(*p)); |
1851 | ||
2caaad41 CL |
1852 | pcp = &p->pcp[0]; /* hot */ |
1853 | pcp->count = 0; | |
2caaad41 CL |
1854 | pcp->high = 6 * batch; |
1855 | pcp->batch = max(1UL, 1 * batch); | |
1856 | INIT_LIST_HEAD(&pcp->list); | |
1857 | ||
1858 | pcp = &p->pcp[1]; /* cold*/ | |
1859 | pcp->count = 0; | |
2caaad41 | 1860 | pcp->high = 2 * batch; |
e46a5e28 | 1861 | pcp->batch = max(1UL, batch/2); |
2caaad41 CL |
1862 | INIT_LIST_HEAD(&pcp->list); |
1863 | } | |
1864 | ||
8ad4b1fb RS |
1865 | /* |
1866 | * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist | |
1867 | * to the value high for the pageset p. | |
1868 | */ | |
1869 | ||
1870 | static void setup_pagelist_highmark(struct per_cpu_pageset *p, | |
1871 | unsigned long high) | |
1872 | { | |
1873 | struct per_cpu_pages *pcp; | |
1874 | ||
1875 | pcp = &p->pcp[0]; /* hot list */ | |
1876 | pcp->high = high; | |
1877 | pcp->batch = max(1UL, high/4); | |
1878 | if ((high/4) > (PAGE_SHIFT * 8)) | |
1879 | pcp->batch = PAGE_SHIFT * 8; | |
1880 | } | |
1881 | ||
1882 | ||
e7c8d5c9 CL |
1883 | #ifdef CONFIG_NUMA |
1884 | /* | |
2caaad41 CL |
1885 | * Boot pageset table. One per cpu which is going to be used for all |
1886 | * zones and all nodes. The parameters will be set in such a way | |
1887 | * that an item put on a list will immediately be handed over to | |
1888 | * the buddy list. This is safe since pageset manipulation is done | |
1889 | * with interrupts disabled. | |
1890 | * | |
1891 | * Some NUMA counter updates may also be caught by the boot pagesets. | |
b7c84c6a CL |
1892 | * |
1893 | * The boot_pagesets must be kept even after bootup is complete for | |
1894 | * unused processors and/or zones. They do play a role for bootstrapping | |
1895 | * hotplugged processors. | |
1896 | * | |
1897 | * zoneinfo_show() and maybe other functions do | |
1898 | * not check if the processor is online before following the pageset pointer. | |
1899 | * Other parts of the kernel may not check if the zone is available. | |
2caaad41 | 1900 | */ |
88a2a4ac | 1901 | static struct per_cpu_pageset boot_pageset[NR_CPUS]; |
2caaad41 CL |
1902 | |
1903 | /* | |
1904 | * Dynamically allocate memory for the | |
e7c8d5c9 CL |
1905 | * per cpu pageset array in struct zone. |
1906 | */ | |
6292d9aa | 1907 | static int __cpuinit process_zones(int cpu) |
e7c8d5c9 CL |
1908 | { |
1909 | struct zone *zone, *dzone; | |
e7c8d5c9 CL |
1910 | |
1911 | for_each_zone(zone) { | |
e7c8d5c9 | 1912 | |
23316bc8 | 1913 | zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset), |
e7c8d5c9 | 1914 | GFP_KERNEL, cpu_to_node(cpu)); |
23316bc8 | 1915 | if (!zone_pcp(zone, cpu)) |
e7c8d5c9 | 1916 | goto bad; |
e7c8d5c9 | 1917 | |
23316bc8 | 1918 | setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone)); |
8ad4b1fb RS |
1919 | |
1920 | if (percpu_pagelist_fraction) | |
1921 | setup_pagelist_highmark(zone_pcp(zone, cpu), | |
1922 | (zone->present_pages / percpu_pagelist_fraction)); | |
e7c8d5c9 CL |
1923 | } |
1924 | ||
1925 | return 0; | |
1926 | bad: | |
1927 | for_each_zone(dzone) { | |
1928 | if (dzone == zone) | |
1929 | break; | |
23316bc8 NP |
1930 | kfree(zone_pcp(dzone, cpu)); |
1931 | zone_pcp(dzone, cpu) = NULL; | |
e7c8d5c9 CL |
1932 | } |
1933 | return -ENOMEM; | |
1934 | } | |
1935 | ||
1936 | static inline void free_zone_pagesets(int cpu) | |
1937 | { | |
e7c8d5c9 CL |
1938 | struct zone *zone; |
1939 | ||
1940 | for_each_zone(zone) { | |
1941 | struct per_cpu_pageset *pset = zone_pcp(zone, cpu); | |
1942 | ||
1943 | zone_pcp(zone, cpu) = NULL; | |
1944 | kfree(pset); | |
1945 | } | |
e7c8d5c9 CL |
1946 | } |
1947 | ||
6292d9aa | 1948 | static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb, |
e7c8d5c9 CL |
1949 | unsigned long action, |
1950 | void *hcpu) | |
1951 | { | |
1952 | int cpu = (long)hcpu; | |
1953 | int ret = NOTIFY_OK; | |
1954 | ||
1955 | switch (action) { | |
1956 | case CPU_UP_PREPARE: | |
1957 | if (process_zones(cpu)) | |
1958 | ret = NOTIFY_BAD; | |
1959 | break; | |
b0d41693 | 1960 | case CPU_UP_CANCELED: |
e7c8d5c9 CL |
1961 | case CPU_DEAD: |
1962 | free_zone_pagesets(cpu); | |
1963 | break; | |
e7c8d5c9 CL |
1964 | default: |
1965 | break; | |
1966 | } | |
1967 | return ret; | |
1968 | } | |
1969 | ||
1970 | static struct notifier_block pageset_notifier = | |
1971 | { &pageset_cpuup_callback, NULL, 0 }; | |
1972 | ||
78d9955b | 1973 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 CL |
1974 | { |
1975 | int err; | |
1976 | ||
1977 | /* Initialize per_cpu_pageset for cpu 0. | |
1978 | * A cpuup callback will do this for every cpu | |
1979 | * as it comes online | |
1980 | */ | |
1981 | err = process_zones(smp_processor_id()); | |
1982 | BUG_ON(err); | |
1983 | register_cpu_notifier(&pageset_notifier); | |
1984 | } | |
1985 | ||
1986 | #endif | |
1987 | ||
c09b4240 | 1988 | static __meminit |
ed8ece2e DH |
1989 | void zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
1990 | { | |
1991 | int i; | |
1992 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1993 | ||
1994 | /* | |
1995 | * The per-page waitqueue mechanism uses hashed waitqueues | |
1996 | * per zone. | |
1997 | */ | |
1998 | zone->wait_table_size = wait_table_size(zone_size_pages); | |
1999 | zone->wait_table_bits = wait_table_bits(zone->wait_table_size); | |
2000 | zone->wait_table = (wait_queue_head_t *) | |
2001 | alloc_bootmem_node(pgdat, zone->wait_table_size | |
2002 | * sizeof(wait_queue_head_t)); | |
2003 | ||
2004 | for(i = 0; i < zone->wait_table_size; ++i) | |
2005 | init_waitqueue_head(zone->wait_table + i); | |
2006 | } | |
2007 | ||
c09b4240 | 2008 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e DH |
2009 | { |
2010 | int cpu; | |
2011 | unsigned long batch = zone_batchsize(zone); | |
2012 | ||
2013 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
2014 | #ifdef CONFIG_NUMA | |
2015 | /* Early boot. Slab allocator not functional yet */ | |
23316bc8 | 2016 | zone_pcp(zone, cpu) = &boot_pageset[cpu]; |
ed8ece2e DH |
2017 | setup_pageset(&boot_pageset[cpu],0); |
2018 | #else | |
2019 | setup_pageset(zone_pcp(zone,cpu), batch); | |
2020 | #endif | |
2021 | } | |
2022 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n", | |
2023 | zone->name, zone->present_pages, batch); | |
2024 | } | |
2025 | ||
c09b4240 | 2026 | static __meminit void init_currently_empty_zone(struct zone *zone, |
ed8ece2e DH |
2027 | unsigned long zone_start_pfn, unsigned long size) |
2028 | { | |
2029 | struct pglist_data *pgdat = zone->zone_pgdat; | |
2030 | ||
2031 | zone_wait_table_init(zone, size); | |
2032 | pgdat->nr_zones = zone_idx(zone) + 1; | |
2033 | ||
2034 | zone->zone_mem_map = pfn_to_page(zone_start_pfn); | |
2035 | zone->zone_start_pfn = zone_start_pfn; | |
2036 | ||
2037 | memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn); | |
2038 | ||
2039 | zone_init_free_lists(pgdat, zone, zone->spanned_pages); | |
2040 | } | |
2041 | ||
1da177e4 LT |
2042 | /* |
2043 | * Set up the zone data structures: | |
2044 | * - mark all pages reserved | |
2045 | * - mark all memory queues empty | |
2046 | * - clear the memory bitmaps | |
2047 | */ | |
2048 | static void __init free_area_init_core(struct pglist_data *pgdat, | |
2049 | unsigned long *zones_size, unsigned long *zholes_size) | |
2050 | { | |
ed8ece2e DH |
2051 | unsigned long j; |
2052 | int nid = pgdat->node_id; | |
1da177e4 LT |
2053 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
2054 | ||
208d54e5 | 2055 | pgdat_resize_init(pgdat); |
1da177e4 LT |
2056 | pgdat->nr_zones = 0; |
2057 | init_waitqueue_head(&pgdat->kswapd_wait); | |
2058 | pgdat->kswapd_max_order = 0; | |
2059 | ||
2060 | for (j = 0; j < MAX_NR_ZONES; j++) { | |
2061 | struct zone *zone = pgdat->node_zones + j; | |
2062 | unsigned long size, realsize; | |
1da177e4 | 2063 | |
1da177e4 LT |
2064 | realsize = size = zones_size[j]; |
2065 | if (zholes_size) | |
2066 | realsize -= zholes_size[j]; | |
2067 | ||
a2f1b424 | 2068 | if (j < ZONE_HIGHMEM) |
1da177e4 LT |
2069 | nr_kernel_pages += realsize; |
2070 | nr_all_pages += realsize; | |
2071 | ||
2072 | zone->spanned_pages = size; | |
2073 | zone->present_pages = realsize; | |
2074 | zone->name = zone_names[j]; | |
2075 | spin_lock_init(&zone->lock); | |
2076 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 2077 | zone_seqlock_init(zone); |
1da177e4 LT |
2078 | zone->zone_pgdat = pgdat; |
2079 | zone->free_pages = 0; | |
2080 | ||
2081 | zone->temp_priority = zone->prev_priority = DEF_PRIORITY; | |
2082 | ||
ed8ece2e | 2083 | zone_pcp_init(zone); |
1da177e4 LT |
2084 | INIT_LIST_HEAD(&zone->active_list); |
2085 | INIT_LIST_HEAD(&zone->inactive_list); | |
2086 | zone->nr_scan_active = 0; | |
2087 | zone->nr_scan_inactive = 0; | |
2088 | zone->nr_active = 0; | |
2089 | zone->nr_inactive = 0; | |
53e9a615 | 2090 | atomic_set(&zone->reclaim_in_progress, 0); |
1da177e4 LT |
2091 | if (!size) |
2092 | continue; | |
2093 | ||
d41dee36 | 2094 | zonetable_add(zone, nid, j, zone_start_pfn, size); |
ed8ece2e | 2095 | init_currently_empty_zone(zone, zone_start_pfn, size); |
1da177e4 | 2096 | zone_start_pfn += size; |
1da177e4 LT |
2097 | } |
2098 | } | |
2099 | ||
2100 | static void __init alloc_node_mem_map(struct pglist_data *pgdat) | |
2101 | { | |
1da177e4 LT |
2102 | /* Skip empty nodes */ |
2103 | if (!pgdat->node_spanned_pages) | |
2104 | return; | |
2105 | ||
d41dee36 | 2106 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
2107 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
2108 | if (!pgdat->node_mem_map) { | |
d41dee36 AW |
2109 | unsigned long size; |
2110 | struct page *map; | |
2111 | ||
1da177e4 | 2112 | size = (pgdat->node_spanned_pages + 1) * sizeof(struct page); |
6f167ec7 DH |
2113 | map = alloc_remap(pgdat->node_id, size); |
2114 | if (!map) | |
2115 | map = alloc_bootmem_node(pgdat, size); | |
2116 | pgdat->node_mem_map = map; | |
1da177e4 | 2117 | } |
d41dee36 | 2118 | #ifdef CONFIG_FLATMEM |
1da177e4 LT |
2119 | /* |
2120 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
2121 | */ | |
2122 | if (pgdat == NODE_DATA(0)) | |
2123 | mem_map = NODE_DATA(0)->node_mem_map; | |
2124 | #endif | |
d41dee36 | 2125 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
2126 | } |
2127 | ||
2128 | void __init free_area_init_node(int nid, struct pglist_data *pgdat, | |
2129 | unsigned long *zones_size, unsigned long node_start_pfn, | |
2130 | unsigned long *zholes_size) | |
2131 | { | |
2132 | pgdat->node_id = nid; | |
2133 | pgdat->node_start_pfn = node_start_pfn; | |
2134 | calculate_zone_totalpages(pgdat, zones_size, zholes_size); | |
2135 | ||
2136 | alloc_node_mem_map(pgdat); | |
2137 | ||
2138 | free_area_init_core(pgdat, zones_size, zholes_size); | |
2139 | } | |
2140 | ||
93b7504e | 2141 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
2142 | static bootmem_data_t contig_bootmem_data; |
2143 | struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data }; | |
2144 | ||
2145 | EXPORT_SYMBOL(contig_page_data); | |
93b7504e | 2146 | #endif |
1da177e4 LT |
2147 | |
2148 | void __init free_area_init(unsigned long *zones_size) | |
2149 | { | |
93b7504e | 2150 | free_area_init_node(0, NODE_DATA(0), zones_size, |
1da177e4 LT |
2151 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
2152 | } | |
1da177e4 LT |
2153 | |
2154 | #ifdef CONFIG_PROC_FS | |
2155 | ||
2156 | #include <linux/seq_file.h> | |
2157 | ||
2158 | static void *frag_start(struct seq_file *m, loff_t *pos) | |
2159 | { | |
2160 | pg_data_t *pgdat; | |
2161 | loff_t node = *pos; | |
2162 | ||
2163 | for (pgdat = pgdat_list; pgdat && node; pgdat = pgdat->pgdat_next) | |
2164 | --node; | |
2165 | ||
2166 | return pgdat; | |
2167 | } | |
2168 | ||
2169 | static void *frag_next(struct seq_file *m, void *arg, loff_t *pos) | |
2170 | { | |
2171 | pg_data_t *pgdat = (pg_data_t *)arg; | |
2172 | ||
2173 | (*pos)++; | |
2174 | return pgdat->pgdat_next; | |
2175 | } | |
2176 | ||
2177 | static void frag_stop(struct seq_file *m, void *arg) | |
2178 | { | |
2179 | } | |
2180 | ||
2181 | /* | |
2182 | * This walks the free areas for each zone. | |
2183 | */ | |
2184 | static int frag_show(struct seq_file *m, void *arg) | |
2185 | { | |
2186 | pg_data_t *pgdat = (pg_data_t *)arg; | |
2187 | struct zone *zone; | |
2188 | struct zone *node_zones = pgdat->node_zones; | |
2189 | unsigned long flags; | |
2190 | int order; | |
2191 | ||
2192 | for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) { | |
f3fe6512 | 2193 | if (!populated_zone(zone)) |
1da177e4 LT |
2194 | continue; |
2195 | ||
2196 | spin_lock_irqsave(&zone->lock, flags); | |
2197 | seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name); | |
2198 | for (order = 0; order < MAX_ORDER; ++order) | |
2199 | seq_printf(m, "%6lu ", zone->free_area[order].nr_free); | |
2200 | spin_unlock_irqrestore(&zone->lock, flags); | |
2201 | seq_putc(m, '\n'); | |
2202 | } | |
2203 | return 0; | |
2204 | } | |
2205 | ||
2206 | struct seq_operations fragmentation_op = { | |
2207 | .start = frag_start, | |
2208 | .next = frag_next, | |
2209 | .stop = frag_stop, | |
2210 | .show = frag_show, | |
2211 | }; | |
2212 | ||
295ab934 ND |
2213 | /* |
2214 | * Output information about zones in @pgdat. | |
2215 | */ | |
2216 | static int zoneinfo_show(struct seq_file *m, void *arg) | |
2217 | { | |
2218 | pg_data_t *pgdat = arg; | |
2219 | struct zone *zone; | |
2220 | struct zone *node_zones = pgdat->node_zones; | |
2221 | unsigned long flags; | |
2222 | ||
2223 | for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; zone++) { | |
2224 | int i; | |
2225 | ||
f3fe6512 | 2226 | if (!populated_zone(zone)) |
295ab934 ND |
2227 | continue; |
2228 | ||
2229 | spin_lock_irqsave(&zone->lock, flags); | |
2230 | seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name); | |
2231 | seq_printf(m, | |
2232 | "\n pages free %lu" | |
2233 | "\n min %lu" | |
2234 | "\n low %lu" | |
2235 | "\n high %lu" | |
2236 | "\n active %lu" | |
2237 | "\n inactive %lu" | |
2238 | "\n scanned %lu (a: %lu i: %lu)" | |
2239 | "\n spanned %lu" | |
2240 | "\n present %lu", | |
2241 | zone->free_pages, | |
2242 | zone->pages_min, | |
2243 | zone->pages_low, | |
2244 | zone->pages_high, | |
2245 | zone->nr_active, | |
2246 | zone->nr_inactive, | |
2247 | zone->pages_scanned, | |
2248 | zone->nr_scan_active, zone->nr_scan_inactive, | |
2249 | zone->spanned_pages, | |
2250 | zone->present_pages); | |
2251 | seq_printf(m, | |
2252 | "\n protection: (%lu", | |
2253 | zone->lowmem_reserve[0]); | |
2254 | for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++) | |
2255 | seq_printf(m, ", %lu", zone->lowmem_reserve[i]); | |
2256 | seq_printf(m, | |
2257 | ")" | |
2258 | "\n pagesets"); | |
23316bc8 | 2259 | for_each_online_cpu(i) { |
295ab934 ND |
2260 | struct per_cpu_pageset *pageset; |
2261 | int j; | |
2262 | ||
e7c8d5c9 | 2263 | pageset = zone_pcp(zone, i); |
295ab934 ND |
2264 | for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) { |
2265 | if (pageset->pcp[j].count) | |
2266 | break; | |
2267 | } | |
2268 | if (j == ARRAY_SIZE(pageset->pcp)) | |
2269 | continue; | |
2270 | for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) { | |
2271 | seq_printf(m, | |
2272 | "\n cpu: %i pcp: %i" | |
2273 | "\n count: %i" | |
295ab934 ND |
2274 | "\n high: %i" |
2275 | "\n batch: %i", | |
2276 | i, j, | |
2277 | pageset->pcp[j].count, | |
295ab934 ND |
2278 | pageset->pcp[j].high, |
2279 | pageset->pcp[j].batch); | |
2280 | } | |
2281 | #ifdef CONFIG_NUMA | |
2282 | seq_printf(m, | |
2283 | "\n numa_hit: %lu" | |
2284 | "\n numa_miss: %lu" | |
2285 | "\n numa_foreign: %lu" | |
2286 | "\n interleave_hit: %lu" | |
2287 | "\n local_node: %lu" | |
2288 | "\n other_node: %lu", | |
2289 | pageset->numa_hit, | |
2290 | pageset->numa_miss, | |
2291 | pageset->numa_foreign, | |
2292 | pageset->interleave_hit, | |
2293 | pageset->local_node, | |
2294 | pageset->other_node); | |
2295 | #endif | |
2296 | } | |
2297 | seq_printf(m, | |
2298 | "\n all_unreclaimable: %u" | |
2299 | "\n prev_priority: %i" | |
2300 | "\n temp_priority: %i" | |
2301 | "\n start_pfn: %lu", | |
2302 | zone->all_unreclaimable, | |
2303 | zone->prev_priority, | |
2304 | zone->temp_priority, | |
2305 | zone->zone_start_pfn); | |
2306 | spin_unlock_irqrestore(&zone->lock, flags); | |
2307 | seq_putc(m, '\n'); | |
2308 | } | |
2309 | return 0; | |
2310 | } | |
2311 | ||
2312 | struct seq_operations zoneinfo_op = { | |
2313 | .start = frag_start, /* iterate over all zones. The same as in | |
2314 | * fragmentation. */ | |
2315 | .next = frag_next, | |
2316 | .stop = frag_stop, | |
2317 | .show = zoneinfo_show, | |
2318 | }; | |
2319 | ||
1da177e4 LT |
2320 | static char *vmstat_text[] = { |
2321 | "nr_dirty", | |
2322 | "nr_writeback", | |
2323 | "nr_unstable", | |
2324 | "nr_page_table_pages", | |
2325 | "nr_mapped", | |
2326 | "nr_slab", | |
2327 | ||
2328 | "pgpgin", | |
2329 | "pgpgout", | |
2330 | "pswpin", | |
2331 | "pswpout", | |
1da177e4 | 2332 | |
9328b8fa | 2333 | "pgalloc_high", |
1da177e4 | 2334 | "pgalloc_normal", |
9328b8fa | 2335 | "pgalloc_dma32", |
1da177e4 | 2336 | "pgalloc_dma", |
9328b8fa | 2337 | |
1da177e4 LT |
2338 | "pgfree", |
2339 | "pgactivate", | |
2340 | "pgdeactivate", | |
2341 | ||
2342 | "pgfault", | |
2343 | "pgmajfault", | |
9328b8fa | 2344 | |
1da177e4 LT |
2345 | "pgrefill_high", |
2346 | "pgrefill_normal", | |
9328b8fa | 2347 | "pgrefill_dma32", |
1da177e4 LT |
2348 | "pgrefill_dma", |
2349 | ||
2350 | "pgsteal_high", | |
2351 | "pgsteal_normal", | |
9328b8fa | 2352 | "pgsteal_dma32", |
1da177e4 | 2353 | "pgsteal_dma", |
9328b8fa | 2354 | |
1da177e4 LT |
2355 | "pgscan_kswapd_high", |
2356 | "pgscan_kswapd_normal", | |
9328b8fa | 2357 | "pgscan_kswapd_dma32", |
1da177e4 | 2358 | "pgscan_kswapd_dma", |
9328b8fa | 2359 | |
1da177e4 LT |
2360 | "pgscan_direct_high", |
2361 | "pgscan_direct_normal", | |
9328b8fa | 2362 | "pgscan_direct_dma32", |
1da177e4 | 2363 | "pgscan_direct_dma", |
1da177e4 | 2364 | |
9328b8fa | 2365 | "pginodesteal", |
1da177e4 LT |
2366 | "slabs_scanned", |
2367 | "kswapd_steal", | |
2368 | "kswapd_inodesteal", | |
2369 | "pageoutrun", | |
2370 | "allocstall", | |
2371 | ||
2372 | "pgrotated", | |
edfbe2b0 | 2373 | "nr_bounce", |
1da177e4 LT |
2374 | }; |
2375 | ||
2376 | static void *vmstat_start(struct seq_file *m, loff_t *pos) | |
2377 | { | |
2378 | struct page_state *ps; | |
2379 | ||
2380 | if (*pos >= ARRAY_SIZE(vmstat_text)) | |
2381 | return NULL; | |
2382 | ||
2383 | ps = kmalloc(sizeof(*ps), GFP_KERNEL); | |
2384 | m->private = ps; | |
2385 | if (!ps) | |
2386 | return ERR_PTR(-ENOMEM); | |
2387 | get_full_page_state(ps); | |
2388 | ps->pgpgin /= 2; /* sectors -> kbytes */ | |
2389 | ps->pgpgout /= 2; | |
2390 | return (unsigned long *)ps + *pos; | |
2391 | } | |
2392 | ||
2393 | static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos) | |
2394 | { | |
2395 | (*pos)++; | |
2396 | if (*pos >= ARRAY_SIZE(vmstat_text)) | |
2397 | return NULL; | |
2398 | return (unsigned long *)m->private + *pos; | |
2399 | } | |
2400 | ||
2401 | static int vmstat_show(struct seq_file *m, void *arg) | |
2402 | { | |
2403 | unsigned long *l = arg; | |
2404 | unsigned long off = l - (unsigned long *)m->private; | |
2405 | ||
2406 | seq_printf(m, "%s %lu\n", vmstat_text[off], *l); | |
2407 | return 0; | |
2408 | } | |
2409 | ||
2410 | static void vmstat_stop(struct seq_file *m, void *arg) | |
2411 | { | |
2412 | kfree(m->private); | |
2413 | m->private = NULL; | |
2414 | } | |
2415 | ||
2416 | struct seq_operations vmstat_op = { | |
2417 | .start = vmstat_start, | |
2418 | .next = vmstat_next, | |
2419 | .stop = vmstat_stop, | |
2420 | .show = vmstat_show, | |
2421 | }; | |
2422 | ||
2423 | #endif /* CONFIG_PROC_FS */ | |
2424 | ||
2425 | #ifdef CONFIG_HOTPLUG_CPU | |
2426 | static int page_alloc_cpu_notify(struct notifier_block *self, | |
2427 | unsigned long action, void *hcpu) | |
2428 | { | |
2429 | int cpu = (unsigned long)hcpu; | |
2430 | long *count; | |
2431 | unsigned long *src, *dest; | |
2432 | ||
2433 | if (action == CPU_DEAD) { | |
2434 | int i; | |
2435 | ||
2436 | /* Drain local pagecache count. */ | |
2437 | count = &per_cpu(nr_pagecache_local, cpu); | |
2438 | atomic_add(*count, &nr_pagecache); | |
2439 | *count = 0; | |
2440 | local_irq_disable(); | |
2441 | __drain_pages(cpu); | |
2442 | ||
2443 | /* Add dead cpu's page_states to our own. */ | |
2444 | dest = (unsigned long *)&__get_cpu_var(page_states); | |
2445 | src = (unsigned long *)&per_cpu(page_states, cpu); | |
2446 | ||
2447 | for (i = 0; i < sizeof(struct page_state)/sizeof(unsigned long); | |
2448 | i++) { | |
2449 | dest[i] += src[i]; | |
2450 | src[i] = 0; | |
2451 | } | |
2452 | ||
2453 | local_irq_enable(); | |
2454 | } | |
2455 | return NOTIFY_OK; | |
2456 | } | |
2457 | #endif /* CONFIG_HOTPLUG_CPU */ | |
2458 | ||
2459 | void __init page_alloc_init(void) | |
2460 | { | |
2461 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
2462 | } | |
2463 | ||
2464 | /* | |
2465 | * setup_per_zone_lowmem_reserve - called whenever | |
2466 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
2467 | * has a correct pages reserved value, so an adequate number of | |
2468 | * pages are left in the zone after a successful __alloc_pages(). | |
2469 | */ | |
2470 | static void setup_per_zone_lowmem_reserve(void) | |
2471 | { | |
2472 | struct pglist_data *pgdat; | |
2473 | int j, idx; | |
2474 | ||
2475 | for_each_pgdat(pgdat) { | |
2476 | for (j = 0; j < MAX_NR_ZONES; j++) { | |
2477 | struct zone *zone = pgdat->node_zones + j; | |
2478 | unsigned long present_pages = zone->present_pages; | |
2479 | ||
2480 | zone->lowmem_reserve[j] = 0; | |
2481 | ||
2482 | for (idx = j-1; idx >= 0; idx--) { | |
2483 | struct zone *lower_zone; | |
2484 | ||
2485 | if (sysctl_lowmem_reserve_ratio[idx] < 1) | |
2486 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
2487 | ||
2488 | lower_zone = pgdat->node_zones + idx; | |
2489 | lower_zone->lowmem_reserve[j] = present_pages / | |
2490 | sysctl_lowmem_reserve_ratio[idx]; | |
2491 | present_pages += lower_zone->present_pages; | |
2492 | } | |
2493 | } | |
2494 | } | |
2495 | } | |
2496 | ||
2497 | /* | |
2498 | * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures | |
2499 | * that the pages_{min,low,high} values for each zone are set correctly | |
2500 | * with respect to min_free_kbytes. | |
2501 | */ | |
3947be19 | 2502 | void setup_per_zone_pages_min(void) |
1da177e4 LT |
2503 | { |
2504 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
2505 | unsigned long lowmem_pages = 0; | |
2506 | struct zone *zone; | |
2507 | unsigned long flags; | |
2508 | ||
2509 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
2510 | for_each_zone(zone) { | |
2511 | if (!is_highmem(zone)) | |
2512 | lowmem_pages += zone->present_pages; | |
2513 | } | |
2514 | ||
2515 | for_each_zone(zone) { | |
669ed175 | 2516 | unsigned long tmp; |
1da177e4 | 2517 | spin_lock_irqsave(&zone->lru_lock, flags); |
669ed175 | 2518 | tmp = (pages_min * zone->present_pages) / lowmem_pages; |
1da177e4 LT |
2519 | if (is_highmem(zone)) { |
2520 | /* | |
669ed175 NP |
2521 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
2522 | * need highmem pages, so cap pages_min to a small | |
2523 | * value here. | |
2524 | * | |
2525 | * The (pages_high-pages_low) and (pages_low-pages_min) | |
2526 | * deltas controls asynch page reclaim, and so should | |
2527 | * not be capped for highmem. | |
1da177e4 LT |
2528 | */ |
2529 | int min_pages; | |
2530 | ||
2531 | min_pages = zone->present_pages / 1024; | |
2532 | if (min_pages < SWAP_CLUSTER_MAX) | |
2533 | min_pages = SWAP_CLUSTER_MAX; | |
2534 | if (min_pages > 128) | |
2535 | min_pages = 128; | |
2536 | zone->pages_min = min_pages; | |
2537 | } else { | |
669ed175 NP |
2538 | /* |
2539 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
2540 | * proportionate to the zone's size. |
2541 | */ | |
669ed175 | 2542 | zone->pages_min = tmp; |
1da177e4 LT |
2543 | } |
2544 | ||
669ed175 NP |
2545 | zone->pages_low = zone->pages_min + tmp / 4; |
2546 | zone->pages_high = zone->pages_min + tmp / 2; | |
1da177e4 LT |
2547 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
2548 | } | |
2549 | } | |
2550 | ||
2551 | /* | |
2552 | * Initialise min_free_kbytes. | |
2553 | * | |
2554 | * For small machines we want it small (128k min). For large machines | |
2555 | * we want it large (64MB max). But it is not linear, because network | |
2556 | * bandwidth does not increase linearly with machine size. We use | |
2557 | * | |
2558 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: | |
2559 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) | |
2560 | * | |
2561 | * which yields | |
2562 | * | |
2563 | * 16MB: 512k | |
2564 | * 32MB: 724k | |
2565 | * 64MB: 1024k | |
2566 | * 128MB: 1448k | |
2567 | * 256MB: 2048k | |
2568 | * 512MB: 2896k | |
2569 | * 1024MB: 4096k | |
2570 | * 2048MB: 5792k | |
2571 | * 4096MB: 8192k | |
2572 | * 8192MB: 11584k | |
2573 | * 16384MB: 16384k | |
2574 | */ | |
2575 | static int __init init_per_zone_pages_min(void) | |
2576 | { | |
2577 | unsigned long lowmem_kbytes; | |
2578 | ||
2579 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
2580 | ||
2581 | min_free_kbytes = int_sqrt(lowmem_kbytes * 16); | |
2582 | if (min_free_kbytes < 128) | |
2583 | min_free_kbytes = 128; | |
2584 | if (min_free_kbytes > 65536) | |
2585 | min_free_kbytes = 65536; | |
2586 | setup_per_zone_pages_min(); | |
2587 | setup_per_zone_lowmem_reserve(); | |
2588 | return 0; | |
2589 | } | |
2590 | module_init(init_per_zone_pages_min) | |
2591 | ||
2592 | /* | |
2593 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so | |
2594 | * that we can call two helper functions whenever min_free_kbytes | |
2595 | * changes. | |
2596 | */ | |
2597 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, | |
2598 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
2599 | { | |
2600 | proc_dointvec(table, write, file, buffer, length, ppos); | |
2601 | setup_per_zone_pages_min(); | |
2602 | return 0; | |
2603 | } | |
2604 | ||
2605 | /* | |
2606 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
2607 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
2608 | * whenever sysctl_lowmem_reserve_ratio changes. | |
2609 | * | |
2610 | * The reserve ratio obviously has absolutely no relation with the | |
2611 | * pages_min watermarks. The lowmem reserve ratio can only make sense | |
2612 | * if in function of the boot time zone sizes. | |
2613 | */ | |
2614 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, | |
2615 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
2616 | { | |
2617 | proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
2618 | setup_per_zone_lowmem_reserve(); | |
2619 | return 0; | |
2620 | } | |
2621 | ||
8ad4b1fb RS |
2622 | /* |
2623 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
2624 | * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist | |
2625 | * can have before it gets flushed back to buddy allocator. | |
2626 | */ | |
2627 | ||
2628 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, | |
2629 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
2630 | { | |
2631 | struct zone *zone; | |
2632 | unsigned int cpu; | |
2633 | int ret; | |
2634 | ||
2635 | ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
2636 | if (!write || (ret == -EINVAL)) | |
2637 | return ret; | |
2638 | for_each_zone(zone) { | |
2639 | for_each_online_cpu(cpu) { | |
2640 | unsigned long high; | |
2641 | high = zone->present_pages / percpu_pagelist_fraction; | |
2642 | setup_pagelist_highmark(zone_pcp(zone, cpu), high); | |
2643 | } | |
2644 | } | |
2645 | return 0; | |
2646 | } | |
2647 | ||
1da177e4 LT |
2648 | __initdata int hashdist = HASHDIST_DEFAULT; |
2649 | ||
2650 | #ifdef CONFIG_NUMA | |
2651 | static int __init set_hashdist(char *str) | |
2652 | { | |
2653 | if (!str) | |
2654 | return 0; | |
2655 | hashdist = simple_strtoul(str, &str, 0); | |
2656 | return 1; | |
2657 | } | |
2658 | __setup("hashdist=", set_hashdist); | |
2659 | #endif | |
2660 | ||
2661 | /* | |
2662 | * allocate a large system hash table from bootmem | |
2663 | * - it is assumed that the hash table must contain an exact power-of-2 | |
2664 | * quantity of entries | |
2665 | * - limit is the number of hash buckets, not the total allocation size | |
2666 | */ | |
2667 | void *__init alloc_large_system_hash(const char *tablename, | |
2668 | unsigned long bucketsize, | |
2669 | unsigned long numentries, | |
2670 | int scale, | |
2671 | int flags, | |
2672 | unsigned int *_hash_shift, | |
2673 | unsigned int *_hash_mask, | |
2674 | unsigned long limit) | |
2675 | { | |
2676 | unsigned long long max = limit; | |
2677 | unsigned long log2qty, size; | |
2678 | void *table = NULL; | |
2679 | ||
2680 | /* allow the kernel cmdline to have a say */ | |
2681 | if (!numentries) { | |
2682 | /* round applicable memory size up to nearest megabyte */ | |
2683 | numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages; | |
2684 | numentries += (1UL << (20 - PAGE_SHIFT)) - 1; | |
2685 | numentries >>= 20 - PAGE_SHIFT; | |
2686 | numentries <<= 20 - PAGE_SHIFT; | |
2687 | ||
2688 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
2689 | if (scale > PAGE_SHIFT) | |
2690 | numentries >>= (scale - PAGE_SHIFT); | |
2691 | else | |
2692 | numentries <<= (PAGE_SHIFT - scale); | |
2693 | } | |
2694 | /* rounded up to nearest power of 2 in size */ | |
2695 | numentries = 1UL << (long_log2(numentries) + 1); | |
2696 | ||
2697 | /* limit allocation size to 1/16 total memory by default */ | |
2698 | if (max == 0) { | |
2699 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
2700 | do_div(max, bucketsize); | |
2701 | } | |
2702 | ||
2703 | if (numentries > max) | |
2704 | numentries = max; | |
2705 | ||
2706 | log2qty = long_log2(numentries); | |
2707 | ||
2708 | do { | |
2709 | size = bucketsize << log2qty; | |
2710 | if (flags & HASH_EARLY) | |
2711 | table = alloc_bootmem(size); | |
2712 | else if (hashdist) | |
2713 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
2714 | else { | |
2715 | unsigned long order; | |
2716 | for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++) | |
2717 | ; | |
2718 | table = (void*) __get_free_pages(GFP_ATOMIC, order); | |
2719 | } | |
2720 | } while (!table && size > PAGE_SIZE && --log2qty); | |
2721 | ||
2722 | if (!table) | |
2723 | panic("Failed to allocate %s hash table\n", tablename); | |
2724 | ||
2725 | printk("%s hash table entries: %d (order: %d, %lu bytes)\n", | |
2726 | tablename, | |
2727 | (1U << log2qty), | |
2728 | long_log2(size) - PAGE_SHIFT, | |
2729 | size); | |
2730 | ||
2731 | if (_hash_shift) | |
2732 | *_hash_shift = log2qty; | |
2733 | if (_hash_mask) | |
2734 | *_hash_mask = (1 << log2qty) - 1; | |
2735 | ||
2736 | return table; | |
2737 | } |