blob: 54a4f5375bbaeaaa9af8faef441e4a758b89d12e [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/stddef.h>
18#include <linux/mm.h>
19#include <linux/swap.h>
20#include <linux/interrupt.h>
21#include <linux/pagemap.h>
22#include <linux/bootmem.h>
23#include <linux/compiler.h>
Randy Dunlap9f158332005-09-13 01:25:16 -070024#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/module.h>
26#include <linux/suspend.h>
27#include <linux/pagevec.h>
28#include <linux/blkdev.h>
29#include <linux/slab.h>
30#include <linux/notifier.h>
31#include <linux/topology.h>
32#include <linux/sysctl.h>
33#include <linux/cpu.h>
34#include <linux/cpuset.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070035#include <linux/memory_hotplug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <linux/nodemask.h>
37#include <linux/vmalloc.h>
Christoph Lameter4be38e32006-01-06 00:11:17 -080038#include <linux/mempolicy.h>
Yasunori Goto68113782006-06-23 02:03:11 -070039#include <linux/stop_machine.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
41#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070042#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include "internal.h"
44
45/*
46 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
47 * initializer cleaner
48 */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070049nodemask_t node_online_map __read_mostly = { { [0] = 1UL } };
Dean Nelson7223a932005-03-23 19:00:00 -070050EXPORT_SYMBOL(node_online_map);
Christoph Lameterc3d8c142005-09-06 15:16:33 -070051nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL;
Dean Nelson7223a932005-03-23 19:00:00 -070052EXPORT_SYMBOL(node_possible_map);
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070053unsigned long totalram_pages __read_mostly;
54unsigned long totalhigh_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -070055unsigned long totalreserve_pages __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070056long nr_swap_pages;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -080057int percpu_pagelist_fraction;
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
Hugh Dickinsd98c7a02006-02-14 13:52:59 -080059static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -080060
Linus Torvalds1da177e2005-04-16 15:20:36 -070061/*
62 * results with 256, 32 in the lowmem_reserve sysctl:
63 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
64 * 1G machine -> (16M dma, 784M normal, 224M high)
65 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
66 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
67 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +010068 *
69 * TBD: should special case ZONE_DMA32 machines here - in those we normally
70 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 */
Andi Kleena2f1b422005-11-05 17:25:53 +010072int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { 256, 256, 32 };
Linus Torvalds1da177e2005-04-16 15:20:36 -070073
74EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -070075
76/*
77 * Used by page_zone() to look up the address of the struct zone whose
78 * id is encoded in the upper bits of page->flags
79 */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070080struct zone *zone_table[1 << ZONETABLE_SHIFT] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070081EXPORT_SYMBOL(zone_table);
82
Andi Kleena2f1b422005-11-05 17:25:53 +010083static char *zone_names[MAX_NR_ZONES] = { "DMA", "DMA32", "Normal", "HighMem" };
Linus Torvalds1da177e2005-04-16 15:20:36 -070084int min_free_kbytes = 1024;
85
Yasunori Goto86356ab2006-06-23 02:03:09 -070086unsigned long __meminitdata nr_kernel_pages;
87unsigned long __meminitdata nr_all_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Nick Piggin13e74442006-01-06 00:10:58 -080089#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -070090static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -070091{
Dave Hansenbdc8cb92005-10-29 18:16:53 -070092 int ret = 0;
93 unsigned seq;
94 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -070095
Dave Hansenbdc8cb92005-10-29 18:16:53 -070096 do {
97 seq = zone_span_seqbegin(zone);
98 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
99 ret = 1;
100 else if (pfn < zone->zone_start_pfn)
101 ret = 1;
102 } while (zone_span_seqretry(zone, seq));
103
104 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700105}
106
107static int page_is_consistent(struct zone *zone, struct page *page)
108{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109#ifdef CONFIG_HOLES_IN_ZONE
110 if (!pfn_valid(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700111 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112#endif
113 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700114 return 0;
115
116 return 1;
117}
118/*
119 * Temporary debugging check for pages not lying within a given zone.
120 */
121static int bad_range(struct zone *zone, struct page *page)
122{
123 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700125 if (!page_is_consistent(zone, page))
126 return 1;
127
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 return 0;
129}
130
Nick Piggin13e74442006-01-06 00:10:58 -0800131#else
132static inline int bad_range(struct zone *zone, struct page *page)
133{
134 return 0;
135}
136#endif
137
Nick Piggin224abf92006-01-06 00:11:11 -0800138static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139{
Nick Piggin224abf92006-01-06 00:11:11 -0800140 printk(KERN_EMERG "Bad page state in process '%s'\n"
Hugh Dickins7365f3d2006-01-11 12:17:18 -0800141 KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
142 KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
143 KERN_EMERG "Backtrace:\n",
Nick Piggin224abf92006-01-06 00:11:11 -0800144 current->comm, page, (int)(2*sizeof(unsigned long)),
145 (unsigned long)page->flags, page->mapping,
146 page_mapcount(page), page_count(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 dump_stack();
Hugh Dickins334795e2005-06-21 17:15:08 -0700148 page->flags &= ~(1 << PG_lru |
149 1 << PG_private |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 1 << PG_active |
152 1 << PG_dirty |
Hugh Dickins334795e2005-06-21 17:15:08 -0700153 1 << PG_reclaim |
154 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 1 << PG_swapcache |
Nick Piggin676165a2006-04-10 11:21:48 +1000156 1 << PG_writeback |
157 1 << PG_buddy );
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 set_page_count(page, 0);
159 reset_page_mapcount(page);
160 page->mapping = NULL;
Randy Dunlap9f158332005-09-13 01:25:16 -0700161 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162}
163
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164/*
165 * Higher-order pages are called "compound pages". They are structured thusly:
166 *
167 * The first PAGE_SIZE page is called the "head page".
168 *
169 * The remaining PAGE_SIZE pages are called "tail pages".
170 *
171 * All pages have PG_compound set. All pages have their ->private pointing at
172 * the head page (even the head page has this).
173 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800174 * The first tail page's ->lru.next holds the address of the compound page's
175 * put_page() function. Its ->lru.prev holds the order of allocation.
176 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800178
179static void free_compound_page(struct page *page)
180{
181 __free_pages_ok(page, (unsigned long)page[1].lru.prev);
182}
183
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184static void prep_compound_page(struct page *page, unsigned long order)
185{
186 int i;
187 int nr_pages = 1 << order;
188
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800189 page[1].lru.next = (void *)free_compound_page; /* set dtor */
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800190 page[1].lru.prev = (void *)order;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 for (i = 0; i < nr_pages; i++) {
192 struct page *p = page + i;
193
Nick Piggin5e9dace2006-03-22 00:08:01 -0800194 __SetPageCompound(p);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700195 set_page_private(p, (unsigned long)page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 }
197}
198
199static void destroy_compound_page(struct page *page, unsigned long order)
200{
201 int i;
202 int nr_pages = 1 << order;
203
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800204 if (unlikely((unsigned long)page[1].lru.prev != order))
Nick Piggin224abf92006-01-06 00:11:11 -0800205 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206
207 for (i = 0; i < nr_pages; i++) {
208 struct page *p = page + i;
209
Nick Piggin224abf92006-01-06 00:11:11 -0800210 if (unlikely(!PageCompound(p) |
211 (page_private(p) != (unsigned long)page)))
212 bad_page(page);
Nick Piggin5e9dace2006-03-22 00:08:01 -0800213 __ClearPageCompound(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 }
215}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216
Nick Piggin17cf4402006-03-22 00:08:41 -0800217static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
218{
219 int i;
220
221 BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
Andrew Morton6626c5d2006-03-22 00:08:42 -0800222 /*
223 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
224 * and __GFP_HIGHMEM from hard or soft interrupt context.
225 */
226 BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800227 for (i = 0; i < (1 << order); i++)
228 clear_highpage(page + i);
229}
230
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231/*
232 * function for dealing with page's order in buddy system.
233 * zone->lock is already acquired when we use these.
234 * So, we don't need atomic page->flags operations here.
235 */
Andrew Morton6aa30012006-04-18 22:20:52 -0700236static inline unsigned long page_order(struct page *page)
237{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700238 return page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239}
240
Andrew Morton6aa30012006-04-18 22:20:52 -0700241static inline void set_page_order(struct page *page, int order)
242{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700243 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000244 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245}
246
247static inline void rmv_page_order(struct page *page)
248{
Nick Piggin676165a2006-04-10 11:21:48 +1000249 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700250 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251}
252
253/*
254 * Locate the struct page for both the matching buddy in our
255 * pair (buddy1) and the combined O(n+1) page they form (page).
256 *
257 * 1) Any buddy B1 will have an order O twin B2 which satisfies
258 * the following equation:
259 * B2 = B1 ^ (1 << O)
260 * For example, if the starting buddy (buddy2) is #8 its order
261 * 1 buddy is #10:
262 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
263 *
264 * 2) Any buddy B will have an order O+1 parent P which
265 * satisfies the following equation:
266 * P = B & ~(1 << O)
267 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200268 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 */
270static inline struct page *
271__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
272{
273 unsigned long buddy_idx = page_idx ^ (1 << order);
274
275 return page + (buddy_idx - page_idx);
276}
277
278static inline unsigned long
279__find_combined_index(unsigned long page_idx, unsigned int order)
280{
281 return (page_idx & ~(1 << order));
282}
283
284/*
285 * This function checks whether a page is free && is the buddy
286 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800287 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000288 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700289 * (c) a page and its buddy have the same order &&
290 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 *
Nick Piggin676165a2006-04-10 11:21:48 +1000292 * For recording whether a page is in the buddy system, we use PG_buddy.
293 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
294 *
295 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700297static inline int page_is_buddy(struct page *page, struct page *buddy,
298 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299{
Nick Piggin13e74442006-01-06 00:10:58 -0800300#ifdef CONFIG_HOLES_IN_ZONE
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700301 if (!pfn_valid(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800302 return 0;
303#endif
304
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700305 if (page_zone_id(page) != page_zone_id(buddy))
306 return 0;
307
308 if (PageBuddy(buddy) && page_order(buddy) == order) {
309 BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa30012006-04-18 22:20:52 -0700310 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000311 }
Andrew Morton6aa30012006-04-18 22:20:52 -0700312 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313}
314
315/*
316 * Freeing function for a buddy system allocator.
317 *
318 * The concept of a buddy system is to maintain direct-mapped table
319 * (containing bit values) for memory blocks of various "orders".
320 * The bottom level table contains the map for the smallest allocatable
321 * units of memory (here, pages), and each level above it describes
322 * pairs of units from the levels below, hence, "buddies".
323 * At a high level, all that happens here is marking the table entry
324 * at the bottom level available, and propagating the changes upward
325 * as necessary, plus some accounting needed to play nicely with other
326 * parts of the VM system.
327 * At each level, we keep a list of pages, which are heads of continuous
Nick Piggin676165a2006-04-10 11:21:48 +1000328 * free pages of length of (1 << order) and marked with PG_buddy. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700329 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 * So when we are allocating or freeing one, we can derive the state of the
331 * other. That is, if we allocate a small block, and both were
332 * free, the remainder of the region must be split into blocks.
333 * If a block is freed, and its buddy is also free, then this
334 * triggers coalescing into a block of larger size.
335 *
336 * -- wli
337 */
338
Nick Piggin48db57f2006-01-08 01:00:42 -0800339static inline void __free_one_page(struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 struct zone *zone, unsigned int order)
341{
342 unsigned long page_idx;
343 int order_size = 1 << order;
344
Nick Piggin224abf92006-01-06 00:11:11 -0800345 if (unlikely(PageCompound(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 destroy_compound_page(page, order);
347
348 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
349
350 BUG_ON(page_idx & (order_size - 1));
351 BUG_ON(bad_range(zone, page));
352
353 zone->free_pages += order_size;
354 while (order < MAX_ORDER-1) {
355 unsigned long combined_idx;
356 struct free_area *area;
357 struct page *buddy;
358
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 buddy = __page_find_buddy(page, page_idx, order);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700360 if (!page_is_buddy(page, buddy, order))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 break; /* Move the buddy up one level. */
Nick Piggin13e74442006-01-06 00:10:58 -0800362
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 list_del(&buddy->lru);
364 area = zone->free_area + order;
365 area->nr_free--;
366 rmv_page_order(buddy);
Nick Piggin13e74442006-01-06 00:10:58 -0800367 combined_idx = __find_combined_index(page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 page = page + (combined_idx - page_idx);
369 page_idx = combined_idx;
370 order++;
371 }
372 set_page_order(page, order);
373 list_add(&page->lru, &zone->free_area[order].free_list);
374 zone->free_area[order].nr_free++;
375}
376
Nick Piggin224abf92006-01-06 00:11:11 -0800377static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378{
Nick Piggin92be2e32006-01-06 00:10:57 -0800379 if (unlikely(page_mapcount(page) |
380 (page->mapping != NULL) |
381 (page_count(page) != 0) |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 (page->flags & (
383 1 << PG_lru |
384 1 << PG_private |
385 1 << PG_locked |
386 1 << PG_active |
387 1 << PG_reclaim |
388 1 << PG_slab |
389 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700390 1 << PG_writeback |
Nick Piggin676165a2006-04-10 11:21:48 +1000391 1 << PG_reserved |
392 1 << PG_buddy ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800393 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 if (PageDirty(page))
Nick Piggin242e5462005-09-03 15:54:50 -0700395 __ClearPageDirty(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800396 /*
397 * For now, we report if PG_reserved was found set, but do not
398 * clear it, and do not free the page. But we shall soon need
399 * to do more, for when the ZERO_PAGE count wraps negative.
400 */
401 return PageReserved(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402}
403
404/*
405 * Frees a list of pages.
406 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700407 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 *
409 * If the zone was previously in an "all pages pinned" state then look to
410 * see if this freeing clears that state.
411 *
412 * And clear the zone's pages_scanned counter, to hold off the "all pages are
413 * pinned" detection logic.
414 */
Nick Piggin48db57f2006-01-08 01:00:42 -0800415static void free_pages_bulk(struct zone *zone, int count,
416 struct list_head *list, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417{
Nick Pigginc54ad302006-01-06 00:10:56 -0800418 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 zone->all_unreclaimable = 0;
420 zone->pages_scanned = 0;
Nick Piggin48db57f2006-01-08 01:00:42 -0800421 while (count--) {
422 struct page *page;
423
424 BUG_ON(list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 page = list_entry(list->prev, struct page, lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800426 /* have to delete it as __free_one_page list manipulates */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 list_del(&page->lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800428 __free_one_page(page, zone, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800430 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431}
432
Nick Piggin48db57f2006-01-08 01:00:42 -0800433static void free_one_page(struct zone *zone, struct page *page, int order)
434{
435 LIST_HEAD(list);
436 list_add(&page->lru, &list);
437 free_pages_bulk(zone, 1, &list, order);
438}
439
440static void __free_pages_ok(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441{
Nick Pigginc54ad302006-01-06 00:10:56 -0800442 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 int i;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800444 int reserved = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445
446 arch_free_page(page, order);
Ingo Molnarde5097c2006-01-09 15:59:21 -0800447 if (!PageHighMem(page))
Ingo Molnarf9b84042006-06-27 02:54:49 -0700448 debug_check_no_locks_freed(page_address(page),
449 PAGE_SIZE<<order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 for (i = 0 ; i < (1 << order) ; ++i)
Nick Piggin224abf92006-01-06 00:11:11 -0800452 reserved += free_pages_check(page + i);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800453 if (reserved)
454 return;
455
Nick Piggin48db57f2006-01-08 01:00:42 -0800456 kernel_map_pages(page, 1 << order, 0);
Nick Pigginc54ad302006-01-06 00:10:56 -0800457 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700458 __count_vm_events(PGFREE, 1 << order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800459 free_one_page(page_zone(page), page, order);
Nick Pigginc54ad302006-01-06 00:10:56 -0800460 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461}
462
David Howellsa226f6c2006-01-06 00:11:08 -0800463/*
464 * permit the bootmem allocator to evade page validation on high-order frees
465 */
466void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
467{
468 if (order == 0) {
469 __ClearPageReserved(page);
470 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800471 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800472 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800473 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800474 int loop;
475
Nick Piggin545b1ea2006-03-22 00:08:07 -0800476 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800477 for (loop = 0; loop < BITS_PER_LONG; loop++) {
478 struct page *p = &page[loop];
479
Nick Piggin545b1ea2006-03-22 00:08:07 -0800480 if (loop + 1 < BITS_PER_LONG)
481 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800482 __ClearPageReserved(p);
483 set_page_count(p, 0);
484 }
485
Nick Piggin7835e982006-03-22 00:08:40 -0800486 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800487 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800488 }
489}
490
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491
492/*
493 * The order of subdivision here is critical for the IO subsystem.
494 * Please do not alter this order without good reasons and regression
495 * testing. Specifically, as large blocks of memory are subdivided,
496 * the order in which smaller blocks are delivered depends on the order
497 * they're subdivided in this function. This is the primary factor
498 * influencing the order in which pages are delivered to the IO
499 * subsystem according to empirical testing, and this is also justified
500 * by considering the behavior of a buddy system containing a single
501 * large block of memory acted on by a series of small allocations.
502 * This behavior is a critical factor in sglist merging's success.
503 *
504 * -- wli
505 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800506static inline void expand(struct zone *zone, struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 int low, int high, struct free_area *area)
508{
509 unsigned long size = 1 << high;
510
511 while (high > low) {
512 area--;
513 high--;
514 size >>= 1;
515 BUG_ON(bad_range(zone, &page[size]));
516 list_add(&page[size].lru, &area->free_list);
517 area->nr_free++;
518 set_page_order(&page[size], high);
519 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520}
521
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522/*
523 * This page is about to be returned from the page allocator
524 */
Nick Piggin17cf4402006-03-22 00:08:41 -0800525static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526{
Nick Piggin92be2e32006-01-06 00:10:57 -0800527 if (unlikely(page_mapcount(page) |
528 (page->mapping != NULL) |
529 (page_count(page) != 0) |
Hugh Dickins334795e2005-06-21 17:15:08 -0700530 (page->flags & (
531 1 << PG_lru |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 1 << PG_private |
533 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 1 << PG_active |
535 1 << PG_dirty |
536 1 << PG_reclaim |
Hugh Dickins334795e2005-06-21 17:15:08 -0700537 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700539 1 << PG_writeback |
Nick Piggin676165a2006-04-10 11:21:48 +1000540 1 << PG_reserved |
541 1 << PG_buddy ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800542 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543
Hugh Dickins689bceb2005-11-21 21:32:20 -0800544 /*
545 * For now, we report if PG_reserved was found set, but do not
546 * clear it, and do not allocate the page: as a safety net.
547 */
548 if (PageReserved(page))
549 return 1;
550
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
552 1 << PG_referenced | 1 << PG_arch_1 |
553 1 << PG_checked | 1 << PG_mappedtodisk);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700554 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800555 set_page_refcounted(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800557
558 if (gfp_flags & __GFP_ZERO)
559 prep_zero_page(page, order, gfp_flags);
560
561 if (order && (gfp_flags & __GFP_COMP))
562 prep_compound_page(page, order);
563
Hugh Dickins689bceb2005-11-21 21:32:20 -0800564 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565}
566
567/*
568 * Do the hard work of removing an element from the buddy allocator.
569 * Call me with the zone->lock already held.
570 */
571static struct page *__rmqueue(struct zone *zone, unsigned int order)
572{
573 struct free_area * area;
574 unsigned int current_order;
575 struct page *page;
576
577 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
578 area = zone->free_area + current_order;
579 if (list_empty(&area->free_list))
580 continue;
581
582 page = list_entry(area->free_list.next, struct page, lru);
583 list_del(&page->lru);
584 rmv_page_order(page);
585 area->nr_free--;
586 zone->free_pages -= 1UL << order;
Nick Piggin085cc7d2006-01-06 00:11:01 -0800587 expand(zone, page, order, current_order, area);
588 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 }
590
591 return NULL;
592}
593
594/*
595 * Obtain a specified number of elements from the buddy allocator, all under
596 * a single hold of the lock, for efficiency. Add them to the supplied list.
597 * Returns the number of new pages which were placed at *list.
598 */
599static int rmqueue_bulk(struct zone *zone, unsigned int order,
600 unsigned long count, struct list_head *list)
601{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603
Nick Pigginc54ad302006-01-06 00:10:56 -0800604 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 for (i = 0; i < count; ++i) {
Nick Piggin085cc7d2006-01-06 00:11:01 -0800606 struct page *page = __rmqueue(zone, order);
607 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 list_add_tail(&page->lru, list);
610 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800611 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800612 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613}
614
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700615#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800616/*
617 * Called from the slab reaper to drain pagesets on a particular node that
618 * belong to the currently executing processor.
Christoph Lameter879336c2006-03-22 00:09:08 -0800619 * Note that this function must be called with the thread pinned to
620 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800621 */
622void drain_node_pages(int nodeid)
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700623{
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800624 int i, z;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700625 unsigned long flags;
626
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800627 for (z = 0; z < MAX_NR_ZONES; z++) {
628 struct zone *zone = NODE_DATA(nodeid)->node_zones + z;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700629 struct per_cpu_pageset *pset;
630
Nick Piggin23316bc2006-01-08 01:00:41 -0800631 pset = zone_pcp(zone, smp_processor_id());
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700632 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
633 struct per_cpu_pages *pcp;
634
635 pcp = &pset->pcp[i];
Christoph Lameter879336c2006-03-22 00:09:08 -0800636 if (pcp->count) {
637 local_irq_save(flags);
638 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
639 pcp->count = 0;
640 local_irq_restore(flags);
641 }
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700642 }
643 }
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700644}
645#endif
646
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647#if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
648static void __drain_pages(unsigned int cpu)
649{
Nick Pigginc54ad302006-01-06 00:10:56 -0800650 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 struct zone *zone;
652 int i;
653
654 for_each_zone(zone) {
655 struct per_cpu_pageset *pset;
656
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700657 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
659 struct per_cpu_pages *pcp;
660
661 pcp = &pset->pcp[i];
Nick Pigginc54ad302006-01-06 00:10:56 -0800662 local_irq_save(flags);
Nick Piggin48db57f2006-01-08 01:00:42 -0800663 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
664 pcp->count = 0;
Nick Pigginc54ad302006-01-06 00:10:56 -0800665 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 }
667 }
668}
669#endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
670
671#ifdef CONFIG_PM
672
673void mark_free_pages(struct zone *zone)
674{
675 unsigned long zone_pfn, flags;
676 int order;
677 struct list_head *curr;
678
679 if (!zone->spanned_pages)
680 return;
681
682 spin_lock_irqsave(&zone->lock, flags);
683 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
684 ClearPageNosaveFree(pfn_to_page(zone_pfn + zone->zone_start_pfn));
685
686 for (order = MAX_ORDER - 1; order >= 0; --order)
687 list_for_each(curr, &zone->free_area[order].free_list) {
688 unsigned long start_pfn, i;
689
690 start_pfn = page_to_pfn(list_entry(curr, struct page, lru));
691
692 for (i=0; i < (1<<order); i++)
693 SetPageNosaveFree(pfn_to_page(start_pfn+i));
694 }
695 spin_unlock_irqrestore(&zone->lock, flags);
696}
697
698/*
699 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
700 */
701void drain_local_pages(void)
702{
703 unsigned long flags;
704
705 local_irq_save(flags);
706 __drain_pages(smp_processor_id());
707 local_irq_restore(flags);
708}
709#endif /* CONFIG_PM */
710
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711/*
712 * Free a 0-order page
713 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714static void fastcall free_hot_cold_page(struct page *page, int cold)
715{
716 struct zone *zone = page_zone(page);
717 struct per_cpu_pages *pcp;
718 unsigned long flags;
719
720 arch_free_page(page, 0);
721
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 if (PageAnon(page))
723 page->mapping = NULL;
Nick Piggin224abf92006-01-06 00:11:11 -0800724 if (free_pages_check(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -0800725 return;
726
Hugh Dickins689bceb2005-11-21 21:32:20 -0800727 kernel_map_pages(page, 1, 0);
728
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700729 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700731 __count_vm_event(PGFREE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 list_add(&page->lru, &pcp->list);
733 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -0800734 if (pcp->count >= pcp->high) {
735 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
736 pcp->count -= pcp->batch;
737 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 local_irq_restore(flags);
739 put_cpu();
740}
741
742void fastcall free_hot_page(struct page *page)
743{
744 free_hot_cold_page(page, 0);
745}
746
747void fastcall free_cold_page(struct page *page)
748{
749 free_hot_cold_page(page, 1);
750}
751
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800752/*
753 * split_page takes a non-compound higher-order page, and splits it into
754 * n (1<<order) sub-pages: page[0..n]
755 * Each sub-page must be freed individually.
756 *
757 * Note: this is probably too low level an operation for use in drivers.
758 * Please consult with lkml before using this in your driver.
759 */
760void split_page(struct page *page, unsigned int order)
761{
762 int i;
763
764 BUG_ON(PageCompound(page));
765 BUG_ON(!page_count(page));
Nick Piggin7835e982006-03-22 00:08:40 -0800766 for (i = 1; i < (1 << order); i++)
767 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800768}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800769
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770/*
771 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
772 * we cheat by calling it from here, in the order > 0 path. Saves a branch
773 * or two.
774 */
Nick Piggina74609f2006-01-06 00:11:20 -0800775static struct page *buffered_rmqueue(struct zonelist *zonelist,
776 struct zone *zone, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777{
778 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800779 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 int cold = !!(gfp_flags & __GFP_COLD);
Nick Piggina74609f2006-01-06 00:11:20 -0800781 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782
Hugh Dickins689bceb2005-11-21 21:32:20 -0800783again:
Nick Piggina74609f2006-01-06 00:11:20 -0800784 cpu = get_cpu();
Nick Piggin48db57f2006-01-08 01:00:42 -0800785 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 struct per_cpu_pages *pcp;
787
Nick Piggina74609f2006-01-06 00:11:20 -0800788 pcp = &zone_pcp(zone, cpu)->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -0800790 if (!pcp->count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 pcp->count += rmqueue_bulk(zone, 0,
792 pcp->batch, &pcp->list);
Nick Piggina74609f2006-01-06 00:11:20 -0800793 if (unlikely(!pcp->count))
794 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 }
Nick Piggina74609f2006-01-06 00:11:20 -0800796 page = list_entry(pcp->list.next, struct page, lru);
797 list_del(&page->lru);
798 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800799 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 spin_lock_irqsave(&zone->lock, flags);
801 page = __rmqueue(zone, order);
Nick Piggina74609f2006-01-06 00:11:20 -0800802 spin_unlock(&zone->lock);
803 if (!page)
804 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 }
806
Christoph Lameterf8891e52006-06-30 01:55:45 -0700807 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Christoph Lameterca889e62006-06-30 01:55:44 -0700808 zone_statistics(zonelist, zone);
Nick Piggina74609f2006-01-06 00:11:20 -0800809 local_irq_restore(flags);
810 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811
Nick Piggina74609f2006-01-06 00:11:20 -0800812 BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -0800813 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -0800814 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 return page;
Nick Piggina74609f2006-01-06 00:11:20 -0800816
817failed:
818 local_irq_restore(flags);
819 put_cpu();
820 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821}
822
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800823#define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
Nick Piggin31488902005-11-28 13:44:03 -0800824#define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
825#define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
826#define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
827#define ALLOC_HARDER 0x10 /* try to alloc harder */
828#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
829#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800830
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831/*
832 * Return 1 if free pages are above 'mark'. This takes into account the order
833 * of the allocation.
834 */
835int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800836 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837{
838 /* free_pages my go negative - that's OK */
839 long min = mark, free_pages = z->free_pages - (1 << order) + 1;
840 int o;
841
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800842 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800844 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 min -= min / 4;
846
847 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
848 return 0;
849 for (o = 0; o < order; o++) {
850 /* At the next order, this order's pages become unavailable */
851 free_pages -= z->free_area[o].nr_free << o;
852
853 /* Require fewer higher order pages to be free */
854 min >>= 1;
855
856 if (free_pages <= min)
857 return 0;
858 }
859 return 1;
860}
861
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800862/*
863 * get_page_from_freeliest goes through the zonelist trying to allocate
864 * a page.
865 */
866static struct page *
867get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
868 struct zonelist *zonelist, int alloc_flags)
Martin Hicks753ee722005-06-21 17:14:41 -0700869{
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800870 struct zone **z = zonelist->zones;
871 struct page *page = NULL;
872 int classzone_idx = zone_idx(*z);
873
874 /*
875 * Go through the zonelist once, looking for a zone with enough free.
876 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
877 */
878 do {
879 if ((alloc_flags & ALLOC_CPUSET) &&
880 !cpuset_zone_allowed(*z, gfp_mask))
881 continue;
882
883 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -0800884 unsigned long mark;
885 if (alloc_flags & ALLOC_WMARK_MIN)
886 mark = (*z)->pages_min;
887 else if (alloc_flags & ALLOC_WMARK_LOW)
888 mark = (*z)->pages_low;
889 else
890 mark = (*z)->pages_high;
891 if (!zone_watermark_ok(*z, order, mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800892 classzone_idx, alloc_flags))
Christoph Lameter9eeff232006-01-18 17:42:31 -0800893 if (!zone_reclaim_mode ||
894 !zone_reclaim(*z, gfp_mask, order))
895 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800896 }
897
Nick Piggina74609f2006-01-06 00:11:20 -0800898 page = buffered_rmqueue(zonelist, *z, order, gfp_mask);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800899 if (page) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800900 break;
901 }
902 } while (*(++z) != NULL);
903 return page;
Martin Hicks753ee722005-06-21 17:14:41 -0700904}
905
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906/*
907 * This is the 'heart' of the zoned buddy allocator.
908 */
909struct page * fastcall
Al Virodd0fc662005-10-07 07:46:04 +0100910__alloc_pages(gfp_t gfp_mask, unsigned int order,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 struct zonelist *zonelist)
912{
Al Viro260b2362005-10-21 03:22:44 -0400913 const gfp_t wait = gfp_mask & __GFP_WAIT;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800914 struct zone **z;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 struct page *page;
916 struct reclaim_state reclaim_state;
917 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 int do_retry;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800919 int alloc_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 int did_some_progress;
921
922 might_sleep_if(wait);
923
Jens Axboe6b1de912005-11-17 21:35:02 +0100924restart:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800925 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800927 if (unlikely(*z == NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 /* Should this ever happen?? */
929 return NULL;
930 }
Jens Axboe6b1de912005-11-17 21:35:02 +0100931
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800932 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -0800933 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800934 if (page)
935 goto got_pg;
936
Jens Axboe6b1de912005-11-17 21:35:02 +0100937 do {
Chris Wright43b0bc02006-06-25 05:47:55 -0700938 wakeup_kswapd(*z, order);
Jens Axboe6b1de912005-11-17 21:35:02 +0100939 } while (*(++z));
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800940
Paul Jackson9bf22292005-09-06 15:18:12 -0700941 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800942 * OK, we're below the kswapd watermark and have kicked background
943 * reclaim. Now things get more complex, so set up alloc_flags according
944 * to how we want to proceed.
945 *
946 * The caller may dip into page reserves a bit more if the caller
947 * cannot run direct reclaim, or if the caller has realtime scheduling
Paul Jackson4eac9152006-01-11 12:17:19 -0800948 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
949 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
Paul Jackson9bf22292005-09-06 15:18:12 -0700950 */
Nick Piggin31488902005-11-28 13:44:03 -0800951 alloc_flags = ALLOC_WMARK_MIN;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800952 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
953 alloc_flags |= ALLOC_HARDER;
954 if (gfp_mask & __GFP_HIGH)
955 alloc_flags |= ALLOC_HIGH;
Paul Jacksonbdd804f2006-05-20 15:00:09 -0700956 if (wait)
957 alloc_flags |= ALLOC_CPUSET;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958
959 /*
960 * Go through the zonelist again. Let __GFP_HIGH and allocations
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800961 * coming from realtime tasks go deeper into reserves.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 *
963 * This is the last chance, in general, before the goto nopage.
964 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
Paul Jackson9bf22292005-09-06 15:18:12 -0700965 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800967 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
968 if (page)
969 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970
971 /* This allocation should allow future memory freeing. */
Nick Pigginb84a35b2005-05-01 08:58:36 -0700972
973 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
974 && !in_interrupt()) {
975 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
Kirill Korotaev885036d2005-11-13 16:06:41 -0800976nofail_alloc:
Nick Pigginb84a35b2005-05-01 08:58:36 -0700977 /* go through the zonelist yet again, ignoring mins */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800978 page = get_page_from_freelist(gfp_mask, order,
Paul Jackson47f3a862006-01-06 00:10:32 -0800979 zonelist, ALLOC_NO_WATERMARKS);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800980 if (page)
981 goto got_pg;
Kirill Korotaev885036d2005-11-13 16:06:41 -0800982 if (gfp_mask & __GFP_NOFAIL) {
983 blk_congestion_wait(WRITE, HZ/50);
984 goto nofail_alloc;
985 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 }
987 goto nopage;
988 }
989
990 /* Atomic allocations - we can't balance anything */
991 if (!wait)
992 goto nopage;
993
994rebalance:
995 cond_resched();
996
997 /* We now go into synchronous reclaim */
Paul Jackson3e0d98b2006-01-08 01:01:49 -0800998 cpuset_memory_pressure_bump();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 p->flags |= PF_MEMALLOC;
1000 reclaim_state.reclaimed_slab = 0;
1001 p->reclaim_state = &reclaim_state;
1002
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001003 did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004
1005 p->reclaim_state = NULL;
1006 p->flags &= ~PF_MEMALLOC;
1007
1008 cond_resched();
1009
1010 if (likely(did_some_progress)) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001011 page = get_page_from_freelist(gfp_mask, order,
1012 zonelist, alloc_flags);
1013 if (page)
1014 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1016 /*
1017 * Go through the zonelist yet one more time, keep
1018 * very high watermark here, this is only to catch
1019 * a parallel oom killing, we must fail if we're still
1020 * under heavy pressure.
1021 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001022 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -08001023 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001024 if (page)
1025 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026
Christoph Lameter9b0f8b02006-02-20 18:27:52 -08001027 out_of_memory(zonelist, gfp_mask, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 goto restart;
1029 }
1030
1031 /*
1032 * Don't let big-order allocations loop unless the caller explicitly
1033 * requests that. Wait for some write requests to complete then retry.
1034 *
1035 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1036 * <= 3, but that may not be true in other implementations.
1037 */
1038 do_retry = 0;
1039 if (!(gfp_mask & __GFP_NORETRY)) {
1040 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
1041 do_retry = 1;
1042 if (gfp_mask & __GFP_NOFAIL)
1043 do_retry = 1;
1044 }
1045 if (do_retry) {
1046 blk_congestion_wait(WRITE, HZ/50);
1047 goto rebalance;
1048 }
1049
1050nopage:
1051 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1052 printk(KERN_WARNING "%s: page allocation failure."
1053 " order:%d, mode:0x%x\n",
1054 p->comm, order, gfp_mask);
1055 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001056 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058got_pg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 return page;
1060}
1061
1062EXPORT_SYMBOL(__alloc_pages);
1063
1064/*
1065 * Common helper functions.
1066 */
Al Virodd0fc662005-10-07 07:46:04 +01001067fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068{
1069 struct page * page;
1070 page = alloc_pages(gfp_mask, order);
1071 if (!page)
1072 return 0;
1073 return (unsigned long) page_address(page);
1074}
1075
1076EXPORT_SYMBOL(__get_free_pages);
1077
Al Virodd0fc662005-10-07 07:46:04 +01001078fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079{
1080 struct page * page;
1081
1082 /*
1083 * get_zeroed_page() returns a 32-bit address, which cannot represent
1084 * a highmem page
1085 */
Al Viro260b2362005-10-21 03:22:44 -04001086 BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087
1088 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1089 if (page)
1090 return (unsigned long) page_address(page);
1091 return 0;
1092}
1093
1094EXPORT_SYMBOL(get_zeroed_page);
1095
1096void __pagevec_free(struct pagevec *pvec)
1097{
1098 int i = pagevec_count(pvec);
1099
1100 while (--i >= 0)
1101 free_hot_cold_page(pvec->pages[i], pvec->cold);
1102}
1103
1104fastcall void __free_pages(struct page *page, unsigned int order)
1105{
Nick Pigginb5810032005-10-29 18:16:12 -07001106 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 if (order == 0)
1108 free_hot_page(page);
1109 else
1110 __free_pages_ok(page, order);
1111 }
1112}
1113
1114EXPORT_SYMBOL(__free_pages);
1115
1116fastcall void free_pages(unsigned long addr, unsigned int order)
1117{
1118 if (addr != 0) {
1119 BUG_ON(!virt_addr_valid((void *)addr));
1120 __free_pages(virt_to_page((void *)addr), order);
1121 }
1122}
1123
1124EXPORT_SYMBOL(free_pages);
1125
1126/*
1127 * Total amount of free (allocatable) RAM:
1128 */
1129unsigned int nr_free_pages(void)
1130{
1131 unsigned int sum = 0;
1132 struct zone *zone;
1133
1134 for_each_zone(zone)
1135 sum += zone->free_pages;
1136
1137 return sum;
1138}
1139
1140EXPORT_SYMBOL(nr_free_pages);
1141
1142#ifdef CONFIG_NUMA
1143unsigned int nr_free_pages_pgdat(pg_data_t *pgdat)
1144{
1145 unsigned int i, sum = 0;
1146
1147 for (i = 0; i < MAX_NR_ZONES; i++)
1148 sum += pgdat->node_zones[i].free_pages;
1149
1150 return sum;
1151}
1152#endif
1153
1154static unsigned int nr_free_zone_pages(int offset)
1155{
Martin J. Blighe310fd42005-07-29 22:59:18 -07001156 /* Just pick one node, since fallback list is circular */
1157 pg_data_t *pgdat = NODE_DATA(numa_node_id());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 unsigned int sum = 0;
1159
Martin J. Blighe310fd42005-07-29 22:59:18 -07001160 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1161 struct zone **zonep = zonelist->zones;
1162 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163
Martin J. Blighe310fd42005-07-29 22:59:18 -07001164 for (zone = *zonep++; zone; zone = *zonep++) {
1165 unsigned long size = zone->present_pages;
1166 unsigned long high = zone->pages_high;
1167 if (size > high)
1168 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 }
1170
1171 return sum;
1172}
1173
1174/*
1175 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1176 */
1177unsigned int nr_free_buffer_pages(void)
1178{
Al Viroaf4ca452005-10-21 02:55:38 -04001179 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180}
1181
1182/*
1183 * Amount of free RAM allocatable within all zones
1184 */
1185unsigned int nr_free_pagecache_pages(void)
1186{
Al Viroaf4ca452005-10-21 02:55:38 -04001187 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188}
1189
1190#ifdef CONFIG_HIGHMEM
1191unsigned int nr_free_highpages (void)
1192{
1193 pg_data_t *pgdat;
1194 unsigned int pages = 0;
1195
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08001196 for_each_online_pgdat(pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 pages += pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1198
1199 return pages;
1200}
1201#endif
1202
1203#ifdef CONFIG_NUMA
1204static void show_node(struct zone *zone)
1205{
1206 printk("Node %d ", zone->zone_pgdat->node_id);
1207}
1208#else
1209#define show_node(zone) do { } while (0)
1210#endif
1211
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212void si_meminfo(struct sysinfo *val)
1213{
1214 val->totalram = totalram_pages;
1215 val->sharedram = 0;
1216 val->freeram = nr_free_pages();
1217 val->bufferram = nr_blockdev_pages();
1218#ifdef CONFIG_HIGHMEM
1219 val->totalhigh = totalhigh_pages;
1220 val->freehigh = nr_free_highpages();
1221#else
1222 val->totalhigh = 0;
1223 val->freehigh = 0;
1224#endif
1225 val->mem_unit = PAGE_SIZE;
1226}
1227
1228EXPORT_SYMBOL(si_meminfo);
1229
1230#ifdef CONFIG_NUMA
1231void si_meminfo_node(struct sysinfo *val, int nid)
1232{
1233 pg_data_t *pgdat = NODE_DATA(nid);
1234
1235 val->totalram = pgdat->node_present_pages;
1236 val->freeram = nr_free_pages_pgdat(pgdat);
1237 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1238 val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1239 val->mem_unit = PAGE_SIZE;
1240}
1241#endif
1242
1243#define K(x) ((x) << (PAGE_SHIFT-10))
1244
1245/*
1246 * Show free area list (used inside shift_scroll-lock stuff)
1247 * We also calculate the percentage fragmentation. We do this by counting the
1248 * memory on each free list with the exception of the first item on the list.
1249 */
1250void show_free_areas(void)
1251{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 int cpu, temperature;
1253 unsigned long active;
1254 unsigned long inactive;
1255 unsigned long free;
1256 struct zone *zone;
1257
1258 for_each_zone(zone) {
1259 show_node(zone);
1260 printk("%s per-cpu:", zone->name);
1261
Con Kolivasf3fe6512006-01-06 00:11:15 -08001262 if (!populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 printk(" empty\n");
1264 continue;
1265 } else
1266 printk("\n");
1267
Dave Jones6b482c62005-11-10 15:45:56 -05001268 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 struct per_cpu_pageset *pageset;
1270
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001271 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272
1273 for (temperature = 0; temperature < 2; temperature++)
Nick Piggin2d92c5c2006-01-06 00:10:59 -08001274 printk("cpu %d %s: high %d, batch %d used:%d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 cpu,
1276 temperature ? "cold" : "hot",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 pageset->pcp[temperature].high,
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001278 pageset->pcp[temperature].batch,
1279 pageset->pcp[temperature].count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 }
1281 }
1282
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 get_zone_counts(&active, &inactive, &free);
1284
Denis Vlasenkoc0d62212005-06-21 17:15:14 -07001285 printk("Free pages: %11ukB (%ukB HighMem)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 K(nr_free_pages()),
1287 K(nr_free_highpages()));
1288
1289 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1290 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1291 active,
1292 inactive,
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07001293 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07001294 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07001295 global_page_state(NR_UNSTABLE_NFS),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 nr_free_pages(),
Christoph Lameter9a865ff2006-06-30 01:55:38 -07001297 global_page_state(NR_SLAB),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07001298 global_page_state(NR_FILE_MAPPED),
Christoph Lameterdf849a12006-06-30 01:55:38 -07001299 global_page_state(NR_PAGETABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300
1301 for_each_zone(zone) {
1302 int i;
1303
1304 show_node(zone);
1305 printk("%s"
1306 " free:%lukB"
1307 " min:%lukB"
1308 " low:%lukB"
1309 " high:%lukB"
1310 " active:%lukB"
1311 " inactive:%lukB"
1312 " present:%lukB"
1313 " pages_scanned:%lu"
1314 " all_unreclaimable? %s"
1315 "\n",
1316 zone->name,
1317 K(zone->free_pages),
1318 K(zone->pages_min),
1319 K(zone->pages_low),
1320 K(zone->pages_high),
1321 K(zone->nr_active),
1322 K(zone->nr_inactive),
1323 K(zone->present_pages),
1324 zone->pages_scanned,
1325 (zone->all_unreclaimable ? "yes" : "no")
1326 );
1327 printk("lowmem_reserve[]:");
1328 for (i = 0; i < MAX_NR_ZONES; i++)
1329 printk(" %lu", zone->lowmem_reserve[i]);
1330 printk("\n");
1331 }
1332
1333 for_each_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001334 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335
1336 show_node(zone);
1337 printk("%s: ", zone->name);
Con Kolivasf3fe6512006-01-06 00:11:15 -08001338 if (!populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 printk("empty\n");
1340 continue;
1341 }
1342
1343 spin_lock_irqsave(&zone->lock, flags);
1344 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001345 nr[order] = zone->free_area[order].nr_free;
1346 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 }
1348 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001349 for (order = 0; order < MAX_ORDER; order++)
1350 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 printk("= %lukB\n", K(total));
1352 }
1353
1354 show_swap_cache_info();
1355}
1356
1357/*
1358 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08001359 *
1360 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07001362static int __meminit build_zonelists_node(pg_data_t *pgdat,
Christoph Lameter070f8032006-01-06 00:11:19 -08001363 struct zonelist *zonelist, int nr_zones, int zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364{
Christoph Lameter1a932052006-01-06 00:11:16 -08001365 struct zone *zone;
1366
Christoph Lameter070f8032006-01-06 00:11:19 -08001367 BUG_ON(zone_type > ZONE_HIGHMEM);
Christoph Lameter02a68a52006-01-06 00:11:18 -08001368
1369 do {
Christoph Lameter070f8032006-01-06 00:11:19 -08001370 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08001371 if (populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372#ifndef CONFIG_HIGHMEM
Christoph Lameter070f8032006-01-06 00:11:19 -08001373 BUG_ON(zone_type > ZONE_NORMAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374#endif
Christoph Lameter070f8032006-01-06 00:11:19 -08001375 zonelist->zones[nr_zones++] = zone;
1376 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 }
Christoph Lameter070f8032006-01-06 00:11:19 -08001378 zone_type--;
Christoph Lameter02a68a52006-01-06 00:11:18 -08001379
Christoph Lameter070f8032006-01-06 00:11:19 -08001380 } while (zone_type >= 0);
1381 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382}
1383
Al Viro260b2362005-10-21 03:22:44 -04001384static inline int highest_zone(int zone_bits)
1385{
1386 int res = ZONE_NORMAL;
1387 if (zone_bits & (__force int)__GFP_HIGHMEM)
1388 res = ZONE_HIGHMEM;
Andi Kleena2f1b422005-11-05 17:25:53 +01001389 if (zone_bits & (__force int)__GFP_DMA32)
1390 res = ZONE_DMA32;
Al Viro260b2362005-10-21 03:22:44 -04001391 if (zone_bits & (__force int)__GFP_DMA)
1392 res = ZONE_DMA;
1393 return res;
1394}
1395
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396#ifdef CONFIG_NUMA
1397#define MAX_NODE_LOAD (num_online_nodes())
Yasunori Goto86356ab2006-06-23 02:03:09 -07001398static int __meminitdata node_load[MAX_NUMNODES];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001400 * find_next_best_node - find the next node that should appear in a given node's fallback list
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 * @node: node whose fallback list we're appending
1402 * @used_node_mask: nodemask_t of already used nodes
1403 *
1404 * We use a number of factors to determine which is the next node that should
1405 * appear on a given node's fallback list. The node should not have appeared
1406 * already in @node's fallback list, and it should be the next closest node
1407 * according to the distance array (which contains arbitrary distance values
1408 * from each node to each node in the system), and should also prefer nodes
1409 * with no CPUs, since presumably they'll have very little allocation pressure
1410 * on them otherwise.
1411 * It returns -1 if no node is found.
1412 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07001413static int __meminit find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001415 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 int min_val = INT_MAX;
1417 int best_node = -1;
1418
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001419 /* Use the local node if we haven't already */
1420 if (!node_isset(node, *used_node_mask)) {
1421 node_set(node, *used_node_mask);
1422 return node;
1423 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001425 for_each_online_node(n) {
1426 cpumask_t tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427
1428 /* Don't want a node to appear more than once */
1429 if (node_isset(n, *used_node_mask))
1430 continue;
1431
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 /* Use the distance array to find the distance */
1433 val = node_distance(node, n);
1434
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001435 /* Penalize nodes under us ("prefer the next node") */
1436 val += (n < node);
1437
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 /* Give preference to headless and unused nodes */
1439 tmp = node_to_cpumask(n);
1440 if (!cpus_empty(tmp))
1441 val += PENALTY_FOR_NODE_WITH_CPUS;
1442
1443 /* Slight preference for less loaded node */
1444 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1445 val += node_load[n];
1446
1447 if (val < min_val) {
1448 min_val = val;
1449 best_node = n;
1450 }
1451 }
1452
1453 if (best_node >= 0)
1454 node_set(best_node, *used_node_mask);
1455
1456 return best_node;
1457}
1458
Yasunori Goto86356ab2006-06-23 02:03:09 -07001459static void __meminit build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460{
1461 int i, j, k, node, local_node;
1462 int prev_node, load;
1463 struct zonelist *zonelist;
1464 nodemask_t used_mask;
1465
1466 /* initialize zonelists */
1467 for (i = 0; i < GFP_ZONETYPES; i++) {
1468 zonelist = pgdat->node_zonelists + i;
1469 zonelist->zones[0] = NULL;
1470 }
1471
1472 /* NUMA-aware ordering of nodes */
1473 local_node = pgdat->node_id;
1474 load = num_online_nodes();
1475 prev_node = local_node;
1476 nodes_clear(used_mask);
1477 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08001478 int distance = node_distance(local_node, node);
1479
1480 /*
1481 * If another node is sufficiently far away then it is better
1482 * to reclaim pages in a zone before going off node.
1483 */
1484 if (distance > RECLAIM_DISTANCE)
1485 zone_reclaim_mode = 1;
1486
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 /*
1488 * We don't want to pressure a particular node.
1489 * So adding penalty to the first node in same
1490 * distance group to make it round-robin.
1491 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08001492
1493 if (distance != node_distance(local_node, prev_node))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 node_load[node] += load;
1495 prev_node = node;
1496 load--;
1497 for (i = 0; i < GFP_ZONETYPES; i++) {
1498 zonelist = pgdat->node_zonelists + i;
1499 for (j = 0; zonelist->zones[j] != NULL; j++);
1500
Al Viro260b2362005-10-21 03:22:44 -04001501 k = highest_zone(i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502
1503 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1504 zonelist->zones[j] = NULL;
1505 }
1506 }
1507}
1508
1509#else /* CONFIG_NUMA */
1510
Yasunori Goto86356ab2006-06-23 02:03:09 -07001511static void __meminit build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512{
1513 int i, j, k, node, local_node;
1514
1515 local_node = pgdat->node_id;
1516 for (i = 0; i < GFP_ZONETYPES; i++) {
1517 struct zonelist *zonelist;
1518
1519 zonelist = pgdat->node_zonelists + i;
1520
1521 j = 0;
Al Viro260b2362005-10-21 03:22:44 -04001522 k = highest_zone(i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 j = build_zonelists_node(pgdat, zonelist, j, k);
1524 /*
1525 * Now we build the zonelist so that it contains the zones
1526 * of all the other nodes.
1527 * We don't want to pressure a particular node, so when
1528 * building the zones for node N, we make sure that the
1529 * zones coming right after the local ones are those from
1530 * node N+1 (modulo N)
1531 */
1532 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1533 if (!node_online(node))
1534 continue;
1535 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1536 }
1537 for (node = 0; node < local_node; node++) {
1538 if (!node_online(node))
1539 continue;
1540 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1541 }
1542
1543 zonelist->zones[j] = NULL;
1544 }
1545}
1546
1547#endif /* CONFIG_NUMA */
1548
Yasunori Goto68113782006-06-23 02:03:11 -07001549/* return values int ....just for stop_machine_run() */
1550static int __meminit __build_all_zonelists(void *dummy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551{
Yasunori Goto68113782006-06-23 02:03:11 -07001552 int nid;
1553 for_each_online_node(nid)
1554 build_zonelists(NODE_DATA(nid));
1555 return 0;
1556}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557
Yasunori Goto68113782006-06-23 02:03:11 -07001558void __meminit build_all_zonelists(void)
1559{
1560 if (system_state == SYSTEM_BOOTING) {
1561 __build_all_zonelists(0);
1562 cpuset_init_current_mems_allowed();
1563 } else {
1564 /* we have to stop all cpus to guaranntee there is no user
1565 of zonelist */
1566 stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
1567 /* cpuset refresh routine should be here */
1568 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07001569 vm_total_pages = nr_free_pagecache_pages();
1570 printk("Built %i zonelists. Total pages: %ld\n",
1571 num_online_nodes(), vm_total_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572}
1573
1574/*
1575 * Helper functions to size the waitqueue hash table.
1576 * Essentially these want to choose hash table sizes sufficiently
1577 * large so that collisions trying to wait on pages are rare.
1578 * But in fact, the number of active page waitqueues on typical
1579 * systems is ridiculously low, less than 200. So this is even
1580 * conservative, even though it seems large.
1581 *
1582 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1583 * waitqueues, i.e. the size of the waitq table given the number of pages.
1584 */
1585#define PAGES_PER_WAITQUEUE 256
1586
Yasunori Gotocca448f2006-06-23 02:03:10 -07001587#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07001588static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589{
1590 unsigned long size = 1;
1591
1592 pages /= PAGES_PER_WAITQUEUE;
1593
1594 while (size < pages)
1595 size <<= 1;
1596
1597 /*
1598 * Once we have dozens or even hundreds of threads sleeping
1599 * on IO we've got bigger problems than wait queue collision.
1600 * Limit the size of the wait table to a reasonable size.
1601 */
1602 size = min(size, 4096UL);
1603
1604 return max(size, 4UL);
1605}
Yasunori Gotocca448f2006-06-23 02:03:10 -07001606#else
1607/*
1608 * A zone's size might be changed by hot-add, so it is not possible to determine
1609 * a suitable size for its wait_table. So we use the maximum size now.
1610 *
1611 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
1612 *
1613 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
1614 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
1615 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
1616 *
1617 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
1618 * or more by the traditional way. (See above). It equals:
1619 *
1620 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
1621 * ia64(16K page size) : = ( 8G + 4M)byte.
1622 * powerpc (64K page size) : = (32G +16M)byte.
1623 */
1624static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
1625{
1626 return 4096UL;
1627}
1628#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629
1630/*
1631 * This is an integer logarithm so that shifts can be used later
1632 * to extract the more random high bits from the multiplicative
1633 * hash function before the remainder is taken.
1634 */
1635static inline unsigned long wait_table_bits(unsigned long size)
1636{
1637 return ffz(~size);
1638}
1639
1640#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1641
1642static void __init calculate_zone_totalpages(struct pglist_data *pgdat,
1643 unsigned long *zones_size, unsigned long *zholes_size)
1644{
1645 unsigned long realtotalpages, totalpages = 0;
1646 int i;
1647
1648 for (i = 0; i < MAX_NR_ZONES; i++)
1649 totalpages += zones_size[i];
1650 pgdat->node_spanned_pages = totalpages;
1651
1652 realtotalpages = totalpages;
1653 if (zholes_size)
1654 for (i = 0; i < MAX_NR_ZONES; i++)
1655 realtotalpages -= zholes_size[i];
1656 pgdat->node_present_pages = realtotalpages;
1657 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
1658}
1659
1660
1661/*
1662 * Initially all pages are reserved - free ones are freed
1663 * up by free_all_bootmem() once the early boot process is
1664 * done. Non-atomic initialization, single-pass.
1665 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001666void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 unsigned long start_pfn)
1668{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07001670 unsigned long end_pfn = start_pfn + size;
1671 unsigned long pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672
Greg Ungerercbe8dd42006-01-12 01:05:24 -08001673 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001674 if (!early_pfn_valid(pfn))
1675 continue;
1676 page = pfn_to_page(pfn);
1677 set_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08001678 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 reset_page_mapcount(page);
1680 SetPageReserved(page);
1681 INIT_LIST_HEAD(&page->lru);
1682#ifdef WANT_PAGE_VIRTUAL
1683 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1684 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07001685 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 }
1688}
1689
1690void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1691 unsigned long size)
1692{
1693 int order;
1694 for (order = 0; order < MAX_ORDER ; order++) {
1695 INIT_LIST_HEAD(&zone->free_area[order].free_list);
1696 zone->free_area[order].nr_free = 0;
1697 }
1698}
1699
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001700#define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1701void zonetable_add(struct zone *zone, int nid, int zid, unsigned long pfn,
1702 unsigned long size)
1703{
1704 unsigned long snum = pfn_to_section_nr(pfn);
1705 unsigned long end = pfn_to_section_nr(pfn + size);
1706
1707 if (FLAGS_HAS_NODE)
1708 zone_table[ZONETABLE_INDEX(nid, zid)] = zone;
1709 else
1710 for (; snum <= end; snum++)
1711 zone_table[ZONETABLE_INDEX(snum, zid)] = zone;
1712}
1713
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714#ifndef __HAVE_ARCH_MEMMAP_INIT
1715#define memmap_init(size, nid, zone, start_pfn) \
1716 memmap_init_zone((size), (nid), (zone), (start_pfn))
1717#endif
1718
Ashok Raj6292d9a2006-02-01 03:04:44 -08001719static int __cpuinit zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001720{
1721 int batch;
1722
1723 /*
1724 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07001725 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001726 *
1727 * OK, so we don't know how big the cache is. So guess.
1728 */
1729 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07001730 if (batch * PAGE_SIZE > 512 * 1024)
1731 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001732 batch /= 4; /* We effectively *= 4 below */
1733 if (batch < 1)
1734 batch = 1;
1735
1736 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001737 * Clamp the batch to a 2^n - 1 value. Having a power
1738 * of 2 value was found to be more likely to have
1739 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001740 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001741 * For example if 2 tasks are alternately allocating
1742 * batches of pages, one task can end up with a lot
1743 * of pages of one half of the possible page colors
1744 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001745 */
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001746 batch = (1 << (fls(batch + batch/2)-1)) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07001747
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001748 return batch;
1749}
1750
Christoph Lameter2caaad42005-06-21 17:15:00 -07001751inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
1752{
1753 struct per_cpu_pages *pcp;
1754
Magnus Damm1c6fe942005-10-26 01:58:59 -07001755 memset(p, 0, sizeof(*p));
1756
Christoph Lameter2caaad42005-06-21 17:15:00 -07001757 pcp = &p->pcp[0]; /* hot */
1758 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001759 pcp->high = 6 * batch;
1760 pcp->batch = max(1UL, 1 * batch);
1761 INIT_LIST_HEAD(&pcp->list);
1762
1763 pcp = &p->pcp[1]; /* cold*/
1764 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001765 pcp->high = 2 * batch;
Seth, Rohite46a5e22005-10-29 18:15:48 -07001766 pcp->batch = max(1UL, batch/2);
Christoph Lameter2caaad42005-06-21 17:15:00 -07001767 INIT_LIST_HEAD(&pcp->list);
1768}
1769
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08001770/*
1771 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
1772 * to the value high for the pageset p.
1773 */
1774
1775static void setup_pagelist_highmark(struct per_cpu_pageset *p,
1776 unsigned long high)
1777{
1778 struct per_cpu_pages *pcp;
1779
1780 pcp = &p->pcp[0]; /* hot list */
1781 pcp->high = high;
1782 pcp->batch = max(1UL, high/4);
1783 if ((high/4) > (PAGE_SHIFT * 8))
1784 pcp->batch = PAGE_SHIFT * 8;
1785}
1786
1787
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001788#ifdef CONFIG_NUMA
1789/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07001790 * Boot pageset table. One per cpu which is going to be used for all
1791 * zones and all nodes. The parameters will be set in such a way
1792 * that an item put on a list will immediately be handed over to
1793 * the buddy list. This is safe since pageset manipulation is done
1794 * with interrupts disabled.
1795 *
1796 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07001797 *
1798 * The boot_pagesets must be kept even after bootup is complete for
1799 * unused processors and/or zones. They do play a role for bootstrapping
1800 * hotplugged processors.
1801 *
1802 * zoneinfo_show() and maybe other functions do
1803 * not check if the processor is online before following the pageset pointer.
1804 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07001805 */
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08001806static struct per_cpu_pageset boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07001807
1808/*
1809 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001810 * per cpu pageset array in struct zone.
1811 */
Ashok Raj6292d9a2006-02-01 03:04:44 -08001812static int __cpuinit process_zones(int cpu)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001813{
1814 struct zone *zone, *dzone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001815
1816 for_each_zone(zone) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001817
Nick Piggin23316bc2006-01-08 01:00:41 -08001818 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001819 GFP_KERNEL, cpu_to_node(cpu));
Nick Piggin23316bc2006-01-08 01:00:41 -08001820 if (!zone_pcp(zone, cpu))
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001821 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001822
Nick Piggin23316bc2006-01-08 01:00:41 -08001823 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08001824
1825 if (percpu_pagelist_fraction)
1826 setup_pagelist_highmark(zone_pcp(zone, cpu),
1827 (zone->present_pages / percpu_pagelist_fraction));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001828 }
1829
1830 return 0;
1831bad:
1832 for_each_zone(dzone) {
1833 if (dzone == zone)
1834 break;
Nick Piggin23316bc2006-01-08 01:00:41 -08001835 kfree(zone_pcp(dzone, cpu));
1836 zone_pcp(dzone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001837 }
1838 return -ENOMEM;
1839}
1840
1841static inline void free_zone_pagesets(int cpu)
1842{
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001843 struct zone *zone;
1844
1845 for_each_zone(zone) {
1846 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
1847
1848 zone_pcp(zone, cpu) = NULL;
1849 kfree(pset);
1850 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001851}
1852
Chandra Seetharaman9c7b2162006-06-27 02:54:07 -07001853static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001854 unsigned long action,
1855 void *hcpu)
1856{
1857 int cpu = (long)hcpu;
1858 int ret = NOTIFY_OK;
1859
1860 switch (action) {
1861 case CPU_UP_PREPARE:
1862 if (process_zones(cpu))
1863 ret = NOTIFY_BAD;
1864 break;
Andi Kleenb0d41692005-11-05 17:25:53 +01001865 case CPU_UP_CANCELED:
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001866 case CPU_DEAD:
1867 free_zone_pagesets(cpu);
1868 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001869 default:
1870 break;
1871 }
1872 return ret;
1873}
1874
Chandra Seetharaman74b85f32006-06-27 02:54:09 -07001875static struct notifier_block __cpuinitdata pageset_notifier =
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001876 { &pageset_cpuup_callback, NULL, 0 };
1877
Al Viro78d99552005-12-15 09:18:25 +00001878void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001879{
1880 int err;
1881
1882 /* Initialize per_cpu_pageset for cpu 0.
1883 * A cpuup callback will do this for every cpu
1884 * as it comes online
1885 */
1886 err = process_zones(smp_processor_id());
1887 BUG_ON(err);
1888 register_cpu_notifier(&pageset_notifier);
1889}
1890
1891#endif
1892
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001893static __meminit
Yasunori Gotocca448f2006-06-23 02:03:10 -07001894int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07001895{
1896 int i;
1897 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07001898 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07001899
1900 /*
1901 * The per-page waitqueue mechanism uses hashed waitqueues
1902 * per zone.
1903 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07001904 zone->wait_table_hash_nr_entries =
1905 wait_table_hash_nr_entries(zone_size_pages);
1906 zone->wait_table_bits =
1907 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07001908 alloc_size = zone->wait_table_hash_nr_entries
1909 * sizeof(wait_queue_head_t);
1910
1911 if (system_state == SYSTEM_BOOTING) {
1912 zone->wait_table = (wait_queue_head_t *)
1913 alloc_bootmem_node(pgdat, alloc_size);
1914 } else {
1915 /*
1916 * This case means that a zone whose size was 0 gets new memory
1917 * via memory hot-add.
1918 * But it may be the case that a new node was hot-added. In
1919 * this case vmalloc() will not be able to use this new node's
1920 * memory - this wait_table must be initialized to use this new
1921 * node itself as well.
1922 * To use this new node's memory, further consideration will be
1923 * necessary.
1924 */
1925 zone->wait_table = (wait_queue_head_t *)vmalloc(alloc_size);
1926 }
1927 if (!zone->wait_table)
1928 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07001929
Yasunori Goto02b694d2006-06-23 02:03:08 -07001930 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07001931 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07001932
1933 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07001934}
1935
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001936static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07001937{
1938 int cpu;
1939 unsigned long batch = zone_batchsize(zone);
1940
1941 for (cpu = 0; cpu < NR_CPUS; cpu++) {
1942#ifdef CONFIG_NUMA
1943 /* Early boot. Slab allocator not functional yet */
Nick Piggin23316bc2006-01-08 01:00:41 -08001944 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Dave Hansened8ece22005-10-29 18:16:50 -07001945 setup_pageset(&boot_pageset[cpu],0);
1946#else
1947 setup_pageset(zone_pcp(zone,cpu), batch);
1948#endif
1949 }
Anton Blanchardf5335c02006-03-25 03:06:49 -08001950 if (zone->present_pages)
1951 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
1952 zone->name, zone->present_pages, batch);
Dave Hansened8ece22005-10-29 18:16:50 -07001953}
1954
Yasunori Goto718127c2006-06-23 02:03:10 -07001955__meminit int init_currently_empty_zone(struct zone *zone,
1956 unsigned long zone_start_pfn,
1957 unsigned long size)
Dave Hansened8ece22005-10-29 18:16:50 -07001958{
1959 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07001960 int ret;
1961 ret = zone_wait_table_init(zone, size);
1962 if (ret)
1963 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07001964 pgdat->nr_zones = zone_idx(zone) + 1;
1965
Dave Hansened8ece22005-10-29 18:16:50 -07001966 zone->zone_start_pfn = zone_start_pfn;
1967
1968 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
1969
1970 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
Yasunori Goto718127c2006-06-23 02:03:10 -07001971
1972 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07001973}
1974
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975/*
1976 * Set up the zone data structures:
1977 * - mark all pages reserved
1978 * - mark all memory queues empty
1979 * - clear the memory bitmaps
1980 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07001981static void __meminit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 unsigned long *zones_size, unsigned long *zholes_size)
1983{
Dave Hansened8ece22005-10-29 18:16:50 -07001984 unsigned long j;
1985 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07001987 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988
Dave Hansen208d54e2005-10-29 18:16:52 -07001989 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 pgdat->nr_zones = 0;
1991 init_waitqueue_head(&pgdat->kswapd_wait);
1992 pgdat->kswapd_max_order = 0;
1993
1994 for (j = 0; j < MAX_NR_ZONES; j++) {
1995 struct zone *zone = pgdat->node_zones + j;
1996 unsigned long size, realsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 realsize = size = zones_size[j];
1999 if (zholes_size)
2000 realsize -= zholes_size[j];
2001
Andi Kleena2f1b422005-11-05 17:25:53 +01002002 if (j < ZONE_HIGHMEM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003 nr_kernel_pages += realsize;
2004 nr_all_pages += realsize;
2005
2006 zone->spanned_pages = size;
2007 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07002008#ifdef CONFIG_NUMA
2009 zone->min_unmapped_ratio = (realsize*sysctl_min_unmapped_ratio)
2010 / 100;
2011#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 zone->name = zone_names[j];
2013 spin_lock_init(&zone->lock);
2014 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07002015 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 zone->zone_pgdat = pgdat;
2017 zone->free_pages = 0;
2018
2019 zone->temp_priority = zone->prev_priority = DEF_PRIORITY;
2020
Dave Hansened8ece22005-10-29 18:16:50 -07002021 zone_pcp_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 INIT_LIST_HEAD(&zone->active_list);
2023 INIT_LIST_HEAD(&zone->inactive_list);
2024 zone->nr_scan_active = 0;
2025 zone->nr_scan_inactive = 0;
2026 zone->nr_active = 0;
2027 zone->nr_inactive = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07002028 zap_zone_vm_stats(zone);
Martin Hicks53e9a612005-09-03 15:54:51 -07002029 atomic_set(&zone->reclaim_in_progress, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 if (!size)
2031 continue;
2032
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002033 zonetable_add(zone, nid, j, zone_start_pfn, size);
Yasunori Goto718127c2006-06-23 02:03:10 -07002034 ret = init_currently_empty_zone(zone, zone_start_pfn, size);
2035 BUG_ON(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 }
2038}
2039
2040static void __init alloc_node_mem_map(struct pglist_data *pgdat)
2041{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 /* Skip empty nodes */
2043 if (!pgdat->node_spanned_pages)
2044 return;
2045
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002046#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 /* ia64 gets its own node_mem_map, before this, without bootmem */
2048 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07002049 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002050 struct page *map;
2051
Bob Piccoe984bb42006-05-20 15:00:31 -07002052 /*
2053 * The zone's endpoints aren't required to be MAX_ORDER
2054 * aligned but the node_mem_map endpoints must be in order
2055 * for the buddy allocator to function correctly.
2056 */
2057 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
2058 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
2059 end = ALIGN(end, MAX_ORDER_NR_PAGES);
2060 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07002061 map = alloc_remap(pgdat->node_id, size);
2062 if (!map)
2063 map = alloc_bootmem_node(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07002064 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002066#ifdef CONFIG_FLATMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 /*
2068 * With no DISCONTIG, the global mem_map is just set as node 0's
2069 */
2070 if (pgdat == NODE_DATA(0))
2071 mem_map = NODE_DATA(0)->node_mem_map;
2072#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002073#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074}
2075
Yasunori Goto86356ab2006-06-23 02:03:09 -07002076void __meminit free_area_init_node(int nid, struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077 unsigned long *zones_size, unsigned long node_start_pfn,
2078 unsigned long *zholes_size)
2079{
2080 pgdat->node_id = nid;
2081 pgdat->node_start_pfn = node_start_pfn;
2082 calculate_zone_totalpages(pgdat, zones_size, zholes_size);
2083
2084 alloc_node_mem_map(pgdat);
2085
2086 free_area_init_core(pgdat, zones_size, zholes_size);
2087}
2088
Dave Hansen93b75042005-06-23 00:07:47 -07002089#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090static bootmem_data_t contig_bootmem_data;
2091struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
2092
2093EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07002094#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095
2096void __init free_area_init(unsigned long *zones_size)
2097{
Dave Hansen93b75042005-06-23 00:07:47 -07002098 free_area_init_node(0, NODE_DATA(0), zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
2100}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102#ifdef CONFIG_HOTPLUG_CPU
2103static int page_alloc_cpu_notify(struct notifier_block *self,
2104 unsigned long action, void *hcpu)
2105{
2106 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107
2108 if (action == CPU_DEAD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 local_irq_disable();
2110 __drain_pages(cpu);
Christoph Lameterf8891e52006-06-30 01:55:45 -07002111 vm_events_fold_cpu(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 local_irq_enable();
Christoph Lameter2244b952006-06-30 01:55:33 -07002113 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 }
2115 return NOTIFY_OK;
2116}
2117#endif /* CONFIG_HOTPLUG_CPU */
2118
2119void __init page_alloc_init(void)
2120{
2121 hotcpu_notifier(page_alloc_cpu_notify, 0);
2122}
2123
2124/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002125 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
2126 * or min_free_kbytes changes.
2127 */
2128static void calculate_totalreserve_pages(void)
2129{
2130 struct pglist_data *pgdat;
2131 unsigned long reserve_pages = 0;
2132 int i, j;
2133
2134 for_each_online_pgdat(pgdat) {
2135 for (i = 0; i < MAX_NR_ZONES; i++) {
2136 struct zone *zone = pgdat->node_zones + i;
2137 unsigned long max = 0;
2138
2139 /* Find valid and maximum lowmem_reserve in the zone */
2140 for (j = i; j < MAX_NR_ZONES; j++) {
2141 if (zone->lowmem_reserve[j] > max)
2142 max = zone->lowmem_reserve[j];
2143 }
2144
2145 /* we treat pages_high as reserved pages. */
2146 max += zone->pages_high;
2147
2148 if (max > zone->present_pages)
2149 max = zone->present_pages;
2150 reserve_pages += max;
2151 }
2152 }
2153 totalreserve_pages = reserve_pages;
2154}
2155
2156/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157 * setup_per_zone_lowmem_reserve - called whenever
2158 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2159 * has a correct pages reserved value, so an adequate number of
2160 * pages are left in the zone after a successful __alloc_pages().
2161 */
2162static void setup_per_zone_lowmem_reserve(void)
2163{
2164 struct pglist_data *pgdat;
2165 int j, idx;
2166
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08002167 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 for (j = 0; j < MAX_NR_ZONES; j++) {
2169 struct zone *zone = pgdat->node_zones + j;
2170 unsigned long present_pages = zone->present_pages;
2171
2172 zone->lowmem_reserve[j] = 0;
2173
2174 for (idx = j-1; idx >= 0; idx--) {
2175 struct zone *lower_zone;
2176
2177 if (sysctl_lowmem_reserve_ratio[idx] < 1)
2178 sysctl_lowmem_reserve_ratio[idx] = 1;
2179
2180 lower_zone = pgdat->node_zones + idx;
2181 lower_zone->lowmem_reserve[j] = present_pages /
2182 sysctl_lowmem_reserve_ratio[idx];
2183 present_pages += lower_zone->present_pages;
2184 }
2185 }
2186 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002187
2188 /* update totalreserve_pages */
2189 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190}
2191
2192/*
2193 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2194 * that the pages_{min,low,high} values for each zone are set correctly
2195 * with respect to min_free_kbytes.
2196 */
Dave Hansen3947be12005-10-29 18:16:54 -07002197void setup_per_zone_pages_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198{
2199 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
2200 unsigned long lowmem_pages = 0;
2201 struct zone *zone;
2202 unsigned long flags;
2203
2204 /* Calculate total number of !ZONE_HIGHMEM pages */
2205 for_each_zone(zone) {
2206 if (!is_highmem(zone))
2207 lowmem_pages += zone->present_pages;
2208 }
2209
2210 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07002211 u64 tmp;
2212
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213 spin_lock_irqsave(&zone->lru_lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07002214 tmp = (u64)pages_min * zone->present_pages;
2215 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216 if (is_highmem(zone)) {
2217 /*
Nick Piggin669ed172005-11-13 16:06:45 -08002218 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2219 * need highmem pages, so cap pages_min to a small
2220 * value here.
2221 *
2222 * The (pages_high-pages_low) and (pages_low-pages_min)
2223 * deltas controls asynch page reclaim, and so should
2224 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 */
2226 int min_pages;
2227
2228 min_pages = zone->present_pages / 1024;
2229 if (min_pages < SWAP_CLUSTER_MAX)
2230 min_pages = SWAP_CLUSTER_MAX;
2231 if (min_pages > 128)
2232 min_pages = 128;
2233 zone->pages_min = min_pages;
2234 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08002235 /*
2236 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 * proportionate to the zone's size.
2238 */
Nick Piggin669ed172005-11-13 16:06:45 -08002239 zone->pages_min = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 }
2241
Andrew Mortonac924c62006-05-15 09:43:59 -07002242 zone->pages_low = zone->pages_min + (tmp >> 2);
2243 zone->pages_high = zone->pages_min + (tmp >> 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 spin_unlock_irqrestore(&zone->lru_lock, flags);
2245 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002246
2247 /* update totalreserve_pages */
2248 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249}
2250
2251/*
2252 * Initialise min_free_kbytes.
2253 *
2254 * For small machines we want it small (128k min). For large machines
2255 * we want it large (64MB max). But it is not linear, because network
2256 * bandwidth does not increase linearly with machine size. We use
2257 *
2258 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2259 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2260 *
2261 * which yields
2262 *
2263 * 16MB: 512k
2264 * 32MB: 724k
2265 * 64MB: 1024k
2266 * 128MB: 1448k
2267 * 256MB: 2048k
2268 * 512MB: 2896k
2269 * 1024MB: 4096k
2270 * 2048MB: 5792k
2271 * 4096MB: 8192k
2272 * 8192MB: 11584k
2273 * 16384MB: 16384k
2274 */
2275static int __init init_per_zone_pages_min(void)
2276{
2277 unsigned long lowmem_kbytes;
2278
2279 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
2280
2281 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
2282 if (min_free_kbytes < 128)
2283 min_free_kbytes = 128;
2284 if (min_free_kbytes > 65536)
2285 min_free_kbytes = 65536;
2286 setup_per_zone_pages_min();
2287 setup_per_zone_lowmem_reserve();
2288 return 0;
2289}
2290module_init(init_per_zone_pages_min)
2291
2292/*
2293 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2294 * that we can call two helper functions whenever min_free_kbytes
2295 * changes.
2296 */
2297int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
2298 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2299{
2300 proc_dointvec(table, write, file, buffer, length, ppos);
2301 setup_per_zone_pages_min();
2302 return 0;
2303}
2304
Christoph Lameter96146342006-07-03 00:24:13 -07002305#ifdef CONFIG_NUMA
2306int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
2307 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2308{
2309 struct zone *zone;
2310 int rc;
2311
2312 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2313 if (rc)
2314 return rc;
2315
2316 for_each_zone(zone)
2317 zone->min_unmapped_ratio = (zone->present_pages *
2318 sysctl_min_unmapped_ratio) / 100;
2319 return 0;
2320}
2321#endif
2322
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323/*
2324 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2325 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2326 * whenever sysctl_lowmem_reserve_ratio changes.
2327 *
2328 * The reserve ratio obviously has absolutely no relation with the
2329 * pages_min watermarks. The lowmem reserve ratio can only make sense
2330 * if in function of the boot time zone sizes.
2331 */
2332int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
2333 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2334{
2335 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2336 setup_per_zone_lowmem_reserve();
2337 return 0;
2338}
2339
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002340/*
2341 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
2342 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
2343 * can have before it gets flushed back to buddy allocator.
2344 */
2345
2346int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
2347 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2348{
2349 struct zone *zone;
2350 unsigned int cpu;
2351 int ret;
2352
2353 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2354 if (!write || (ret == -EINVAL))
2355 return ret;
2356 for_each_zone(zone) {
2357 for_each_online_cpu(cpu) {
2358 unsigned long high;
2359 high = zone->present_pages / percpu_pagelist_fraction;
2360 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
2361 }
2362 }
2363 return 0;
2364}
2365
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366__initdata int hashdist = HASHDIST_DEFAULT;
2367
2368#ifdef CONFIG_NUMA
2369static int __init set_hashdist(char *str)
2370{
2371 if (!str)
2372 return 0;
2373 hashdist = simple_strtoul(str, &str, 0);
2374 return 1;
2375}
2376__setup("hashdist=", set_hashdist);
2377#endif
2378
2379/*
2380 * allocate a large system hash table from bootmem
2381 * - it is assumed that the hash table must contain an exact power-of-2
2382 * quantity of entries
2383 * - limit is the number of hash buckets, not the total allocation size
2384 */
2385void *__init alloc_large_system_hash(const char *tablename,
2386 unsigned long bucketsize,
2387 unsigned long numentries,
2388 int scale,
2389 int flags,
2390 unsigned int *_hash_shift,
2391 unsigned int *_hash_mask,
2392 unsigned long limit)
2393{
2394 unsigned long long max = limit;
2395 unsigned long log2qty, size;
2396 void *table = NULL;
2397
2398 /* allow the kernel cmdline to have a say */
2399 if (!numentries) {
2400 /* round applicable memory size up to nearest megabyte */
2401 numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages;
2402 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
2403 numentries >>= 20 - PAGE_SHIFT;
2404 numentries <<= 20 - PAGE_SHIFT;
2405
2406 /* limit to 1 bucket per 2^scale bytes of low memory */
2407 if (scale > PAGE_SHIFT)
2408 numentries >>= (scale - PAGE_SHIFT);
2409 else
2410 numentries <<= (PAGE_SHIFT - scale);
2411 }
John Hawkes6e692ed2006-03-25 03:08:02 -08002412 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413
2414 /* limit allocation size to 1/16 total memory by default */
2415 if (max == 0) {
2416 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
2417 do_div(max, bucketsize);
2418 }
2419
2420 if (numentries > max)
2421 numentries = max;
2422
2423 log2qty = long_log2(numentries);
2424
2425 do {
2426 size = bucketsize << log2qty;
2427 if (flags & HASH_EARLY)
2428 table = alloc_bootmem(size);
2429 else if (hashdist)
2430 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
2431 else {
2432 unsigned long order;
2433 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
2434 ;
2435 table = (void*) __get_free_pages(GFP_ATOMIC, order);
2436 }
2437 } while (!table && size > PAGE_SIZE && --log2qty);
2438
2439 if (!table)
2440 panic("Failed to allocate %s hash table\n", tablename);
2441
2442 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2443 tablename,
2444 (1U << log2qty),
2445 long_log2(size) - PAGE_SHIFT,
2446 size);
2447
2448 if (_hash_shift)
2449 *_hash_shift = log2qty;
2450 if (_hash_mask)
2451 *_hash_mask = (1 << log2qty) - 1;
2452
2453 return table;
2454}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08002455
2456#ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08002457struct page *pfn_to_page(unsigned long pfn)
2458{
Andy Whitcroft67de6482006-06-23 02:03:12 -07002459 return __pfn_to_page(pfn);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08002460}
2461unsigned long page_to_pfn(struct page *page)
2462{
Andy Whitcroft67de6482006-06-23 02:03:12 -07002463 return __page_to_pfn(page);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08002464}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08002465EXPORT_SYMBOL(pfn_to_page);
2466EXPORT_SYMBOL(page_to_pfn);
2467#endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */