blob: 208a6b03aa78b8765bb0226b56b01f97e3f26983 [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;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -070054unsigned long totalreserve_pages __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070055long nr_swap_pages;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -080056int percpu_pagelist_fraction;
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
Hugh Dickinsd98c7a02006-02-14 13:52:59 -080058static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -080059
Linus Torvalds1da177e2005-04-16 15:20:36 -070060/*
61 * results with 256, 32 in the lowmem_reserve sysctl:
62 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
63 * 1G machine -> (16M dma, 784M normal, 224M high)
64 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
65 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
66 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +010067 *
68 * TBD: should special case ZONE_DMA32 machines here - in those we normally
69 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -070071int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
72 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -070073#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -070074 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -070075#endif
Christoph Lametere53ef382006-09-25 23:31:14 -070076#ifdef CONFIG_HIGHMEM
Christoph Lameter2f1b6242006-09-25 23:31:13 -070077 32
Christoph Lametere53ef382006-09-25 23:31:14 -070078#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -070079};
Linus Torvalds1da177e2005-04-16 15:20:36 -070080
81EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -070082
83/*
84 * Used by page_zone() to look up the address of the struct zone whose
85 * id is encoded in the upper bits of page->flags
86 */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070087struct zone *zone_table[1 << ZONETABLE_SHIFT] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070088EXPORT_SYMBOL(zone_table);
89
Christoph Lameter2f1b6242006-09-25 23:31:13 -070090static char *zone_names[MAX_NR_ZONES] = {
91 "DMA",
Christoph Lameterfb0e7942006-09-25 23:31:13 -070092#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -070093 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -070094#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -070095 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -070096#ifdef CONFIG_HIGHMEM
Christoph Lameter2f1b6242006-09-25 23:31:13 -070097 "HighMem"
Christoph Lametere53ef382006-09-25 23:31:14 -070098#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -070099};
100
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101int min_free_kbytes = 1024;
102
Yasunori Goto86356ab2006-06-23 02:03:09 -0700103unsigned long __meminitdata nr_kernel_pages;
104unsigned long __meminitdata nr_all_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105
Nick Piggin13e74442006-01-06 00:10:58 -0800106#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700107static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700109 int ret = 0;
110 unsigned seq;
111 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -0700112
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700113 do {
114 seq = zone_span_seqbegin(zone);
115 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
116 ret = 1;
117 else if (pfn < zone->zone_start_pfn)
118 ret = 1;
119 } while (zone_span_seqretry(zone, seq));
120
121 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700122}
123
124static int page_is_consistent(struct zone *zone, struct page *page)
125{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126#ifdef CONFIG_HOLES_IN_ZONE
127 if (!pfn_valid(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700128 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129#endif
130 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700131 return 0;
132
133 return 1;
134}
135/*
136 * Temporary debugging check for pages not lying within a given zone.
137 */
138static int bad_range(struct zone *zone, struct page *page)
139{
140 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700142 if (!page_is_consistent(zone, page))
143 return 1;
144
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 return 0;
146}
Nick Piggin13e74442006-01-06 00:10:58 -0800147#else
148static inline int bad_range(struct zone *zone, struct page *page)
149{
150 return 0;
151}
152#endif
153
Nick Piggin224abf92006-01-06 00:11:11 -0800154static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155{
Nick Piggin224abf92006-01-06 00:11:11 -0800156 printk(KERN_EMERG "Bad page state in process '%s'\n"
Hugh Dickins7365f3d2006-01-11 12:17:18 -0800157 KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
158 KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
159 KERN_EMERG "Backtrace:\n",
Nick Piggin224abf92006-01-06 00:11:11 -0800160 current->comm, page, (int)(2*sizeof(unsigned long)),
161 (unsigned long)page->flags, page->mapping,
162 page_mapcount(page), page_count(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 dump_stack();
Hugh Dickins334795e2005-06-21 17:15:08 -0700164 page->flags &= ~(1 << PG_lru |
165 1 << PG_private |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 1 << PG_active |
168 1 << PG_dirty |
Hugh Dickins334795e2005-06-21 17:15:08 -0700169 1 << PG_reclaim |
170 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 1 << PG_swapcache |
Nick Piggin676165a2006-04-10 11:21:48 +1000172 1 << PG_writeback |
173 1 << PG_buddy );
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 set_page_count(page, 0);
175 reset_page_mapcount(page);
176 page->mapping = NULL;
Randy Dunlap9f158332005-09-13 01:25:16 -0700177 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178}
179
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180/*
181 * Higher-order pages are called "compound pages". They are structured thusly:
182 *
183 * The first PAGE_SIZE page is called the "head page".
184 *
185 * The remaining PAGE_SIZE pages are called "tail pages".
186 *
187 * All pages have PG_compound set. All pages have their ->private pointing at
188 * the head page (even the head page has this).
189 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800190 * The first tail page's ->lru.next holds the address of the compound page's
191 * put_page() function. Its ->lru.prev holds the order of allocation.
192 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800194
195static void free_compound_page(struct page *page)
196{
197 __free_pages_ok(page, (unsigned long)page[1].lru.prev);
198}
199
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200static void prep_compound_page(struct page *page, unsigned long order)
201{
202 int i;
203 int nr_pages = 1 << order;
204
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800205 page[1].lru.next = (void *)free_compound_page; /* set dtor */
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800206 page[1].lru.prev = (void *)order;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 for (i = 0; i < nr_pages; i++) {
208 struct page *p = page + i;
209
Nick Piggin5e9dace2006-03-22 00:08:01 -0800210 __SetPageCompound(p);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700211 set_page_private(p, (unsigned long)page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 }
213}
214
215static void destroy_compound_page(struct page *page, unsigned long order)
216{
217 int i;
218 int nr_pages = 1 << order;
219
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800220 if (unlikely((unsigned long)page[1].lru.prev != order))
Nick Piggin224abf92006-01-06 00:11:11 -0800221 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222
223 for (i = 0; i < nr_pages; i++) {
224 struct page *p = page + i;
225
Nick Piggin224abf92006-01-06 00:11:11 -0800226 if (unlikely(!PageCompound(p) |
227 (page_private(p) != (unsigned long)page)))
228 bad_page(page);
Nick Piggin5e9dace2006-03-22 00:08:01 -0800229 __ClearPageCompound(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 }
231}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232
Nick Piggin17cf4402006-03-22 00:08:41 -0800233static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
234{
235 int i;
236
Nick Piggin725d7042006-09-25 23:30:55 -0700237 VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
Andrew Morton6626c5d2006-03-22 00:08:42 -0800238 /*
239 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
240 * and __GFP_HIGHMEM from hard or soft interrupt context.
241 */
Nick Piggin725d7042006-09-25 23:30:55 -0700242 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800243 for (i = 0; i < (1 << order); i++)
244 clear_highpage(page + i);
245}
246
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247/*
248 * function for dealing with page's order in buddy system.
249 * zone->lock is already acquired when we use these.
250 * So, we don't need atomic page->flags operations here.
251 */
Andrew Morton6aa30012006-04-18 22:20:52 -0700252static inline unsigned long page_order(struct page *page)
253{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700254 return page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255}
256
Andrew Morton6aa30012006-04-18 22:20:52 -0700257static inline void set_page_order(struct page *page, int order)
258{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700259 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000260 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261}
262
263static inline void rmv_page_order(struct page *page)
264{
Nick Piggin676165a2006-04-10 11:21:48 +1000265 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700266 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267}
268
269/*
270 * Locate the struct page for both the matching buddy in our
271 * pair (buddy1) and the combined O(n+1) page they form (page).
272 *
273 * 1) Any buddy B1 will have an order O twin B2 which satisfies
274 * the following equation:
275 * B2 = B1 ^ (1 << O)
276 * For example, if the starting buddy (buddy2) is #8 its order
277 * 1 buddy is #10:
278 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
279 *
280 * 2) Any buddy B will have an order O+1 parent P which
281 * satisfies the following equation:
282 * P = B & ~(1 << O)
283 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200284 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 */
286static inline struct page *
287__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
288{
289 unsigned long buddy_idx = page_idx ^ (1 << order);
290
291 return page + (buddy_idx - page_idx);
292}
293
294static inline unsigned long
295__find_combined_index(unsigned long page_idx, unsigned int order)
296{
297 return (page_idx & ~(1 << order));
298}
299
300/*
301 * This function checks whether a page is free && is the buddy
302 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800303 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000304 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700305 * (c) a page and its buddy have the same order &&
306 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 *
Nick Piggin676165a2006-04-10 11:21:48 +1000308 * For recording whether a page is in the buddy system, we use PG_buddy.
309 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
310 *
311 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700313static inline int page_is_buddy(struct page *page, struct page *buddy,
314 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315{
Nick Piggin13e74442006-01-06 00:10:58 -0800316#ifdef CONFIG_HOLES_IN_ZONE
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700317 if (!pfn_valid(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800318 return 0;
319#endif
320
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700321 if (page_zone_id(page) != page_zone_id(buddy))
322 return 0;
323
324 if (PageBuddy(buddy) && page_order(buddy) == order) {
325 BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa30012006-04-18 22:20:52 -0700326 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000327 }
Andrew Morton6aa30012006-04-18 22:20:52 -0700328 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329}
330
331/*
332 * Freeing function for a buddy system allocator.
333 *
334 * The concept of a buddy system is to maintain direct-mapped table
335 * (containing bit values) for memory blocks of various "orders".
336 * The bottom level table contains the map for the smallest allocatable
337 * units of memory (here, pages), and each level above it describes
338 * pairs of units from the levels below, hence, "buddies".
339 * At a high level, all that happens here is marking the table entry
340 * at the bottom level available, and propagating the changes upward
341 * as necessary, plus some accounting needed to play nicely with other
342 * parts of the VM system.
343 * At each level, we keep a list of pages, which are heads of continuous
Nick Piggin676165a2006-04-10 11:21:48 +1000344 * free pages of length of (1 << order) and marked with PG_buddy. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700345 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 * So when we are allocating or freeing one, we can derive the state of the
347 * other. That is, if we allocate a small block, and both were
348 * free, the remainder of the region must be split into blocks.
349 * If a block is freed, and its buddy is also free, then this
350 * triggers coalescing into a block of larger size.
351 *
352 * -- wli
353 */
354
Nick Piggin48db57f2006-01-08 01:00:42 -0800355static inline void __free_one_page(struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 struct zone *zone, unsigned int order)
357{
358 unsigned long page_idx;
359 int order_size = 1 << order;
360
Nick Piggin224abf92006-01-06 00:11:11 -0800361 if (unlikely(PageCompound(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 destroy_compound_page(page, order);
363
364 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
365
Nick Piggin725d7042006-09-25 23:30:55 -0700366 VM_BUG_ON(page_idx & (order_size - 1));
367 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368
369 zone->free_pages += order_size;
370 while (order < MAX_ORDER-1) {
371 unsigned long combined_idx;
372 struct free_area *area;
373 struct page *buddy;
374
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 buddy = __page_find_buddy(page, page_idx, order);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700376 if (!page_is_buddy(page, buddy, order))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 break; /* Move the buddy up one level. */
Nick Piggin13e74442006-01-06 00:10:58 -0800378
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 list_del(&buddy->lru);
380 area = zone->free_area + order;
381 area->nr_free--;
382 rmv_page_order(buddy);
Nick Piggin13e74442006-01-06 00:10:58 -0800383 combined_idx = __find_combined_index(page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 page = page + (combined_idx - page_idx);
385 page_idx = combined_idx;
386 order++;
387 }
388 set_page_order(page, order);
389 list_add(&page->lru, &zone->free_area[order].free_list);
390 zone->free_area[order].nr_free++;
391}
392
Nick Piggin224abf92006-01-06 00:11:11 -0800393static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394{
Nick Piggin92be2e32006-01-06 00:10:57 -0800395 if (unlikely(page_mapcount(page) |
396 (page->mapping != NULL) |
397 (page_count(page) != 0) |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 (page->flags & (
399 1 << PG_lru |
400 1 << PG_private |
401 1 << PG_locked |
402 1 << PG_active |
403 1 << PG_reclaim |
404 1 << PG_slab |
405 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700406 1 << PG_writeback |
Nick Piggin676165a2006-04-10 11:21:48 +1000407 1 << PG_reserved |
408 1 << PG_buddy ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800409 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 if (PageDirty(page))
Nick Piggin242e5462005-09-03 15:54:50 -0700411 __ClearPageDirty(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800412 /*
413 * For now, we report if PG_reserved was found set, but do not
414 * clear it, and do not free the page. But we shall soon need
415 * to do more, for when the ZERO_PAGE count wraps negative.
416 */
417 return PageReserved(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418}
419
420/*
421 * Frees a list of pages.
422 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700423 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 *
425 * If the zone was previously in an "all pages pinned" state then look to
426 * see if this freeing clears that state.
427 *
428 * And clear the zone's pages_scanned counter, to hold off the "all pages are
429 * pinned" detection logic.
430 */
Nick Piggin48db57f2006-01-08 01:00:42 -0800431static void free_pages_bulk(struct zone *zone, int count,
432 struct list_head *list, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433{
Nick Pigginc54ad302006-01-06 00:10:56 -0800434 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 zone->all_unreclaimable = 0;
436 zone->pages_scanned = 0;
Nick Piggin48db57f2006-01-08 01:00:42 -0800437 while (count--) {
438 struct page *page;
439
Nick Piggin725d7042006-09-25 23:30:55 -0700440 VM_BUG_ON(list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 page = list_entry(list->prev, struct page, lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800442 /* have to delete it as __free_one_page list manipulates */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 list_del(&page->lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800444 __free_one_page(page, zone, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800446 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447}
448
Nick Piggin48db57f2006-01-08 01:00:42 -0800449static void free_one_page(struct zone *zone, struct page *page, int order)
450{
451 LIST_HEAD(list);
452 list_add(&page->lru, &list);
453 free_pages_bulk(zone, 1, &list, order);
454}
455
456static void __free_pages_ok(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457{
Nick Pigginc54ad302006-01-06 00:10:56 -0800458 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 int i;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800460 int reserved = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461
462 arch_free_page(page, order);
Ingo Molnarde5097c2006-01-09 15:59:21 -0800463 if (!PageHighMem(page))
Ingo Molnarf9b84042006-06-27 02:54:49 -0700464 debug_check_no_locks_freed(page_address(page),
465 PAGE_SIZE<<order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 for (i = 0 ; i < (1 << order) ; ++i)
Nick Piggin224abf92006-01-06 00:11:11 -0800468 reserved += free_pages_check(page + i);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800469 if (reserved)
470 return;
471
Nick Piggin48db57f2006-01-08 01:00:42 -0800472 kernel_map_pages(page, 1 << order, 0);
Nick Pigginc54ad302006-01-06 00:10:56 -0800473 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700474 __count_vm_events(PGFREE, 1 << order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800475 free_one_page(page_zone(page), page, order);
Nick Pigginc54ad302006-01-06 00:10:56 -0800476 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477}
478
David Howellsa226f6c2006-01-06 00:11:08 -0800479/*
480 * permit the bootmem allocator to evade page validation on high-order frees
481 */
482void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
483{
484 if (order == 0) {
485 __ClearPageReserved(page);
486 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800487 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800488 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800489 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800490 int loop;
491
Nick Piggin545b1ea2006-03-22 00:08:07 -0800492 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800493 for (loop = 0; loop < BITS_PER_LONG; loop++) {
494 struct page *p = &page[loop];
495
Nick Piggin545b1ea2006-03-22 00:08:07 -0800496 if (loop + 1 < BITS_PER_LONG)
497 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800498 __ClearPageReserved(p);
499 set_page_count(p, 0);
500 }
501
Nick Piggin7835e982006-03-22 00:08:40 -0800502 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800503 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800504 }
505}
506
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507
508/*
509 * The order of subdivision here is critical for the IO subsystem.
510 * Please do not alter this order without good reasons and regression
511 * testing. Specifically, as large blocks of memory are subdivided,
512 * the order in which smaller blocks are delivered depends on the order
513 * they're subdivided in this function. This is the primary factor
514 * influencing the order in which pages are delivered to the IO
515 * subsystem according to empirical testing, and this is also justified
516 * by considering the behavior of a buddy system containing a single
517 * large block of memory acted on by a series of small allocations.
518 * This behavior is a critical factor in sglist merging's success.
519 *
520 * -- wli
521 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800522static inline void expand(struct zone *zone, struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 int low, int high, struct free_area *area)
524{
525 unsigned long size = 1 << high;
526
527 while (high > low) {
528 area--;
529 high--;
530 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700531 VM_BUG_ON(bad_range(zone, &page[size]));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 list_add(&page[size].lru, &area->free_list);
533 area->nr_free++;
534 set_page_order(&page[size], high);
535 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536}
537
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538/*
539 * This page is about to be returned from the page allocator
540 */
Nick Piggin17cf4402006-03-22 00:08:41 -0800541static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542{
Nick Piggin92be2e32006-01-06 00:10:57 -0800543 if (unlikely(page_mapcount(page) |
544 (page->mapping != NULL) |
545 (page_count(page) != 0) |
Hugh Dickins334795e2005-06-21 17:15:08 -0700546 (page->flags & (
547 1 << PG_lru |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 1 << PG_private |
549 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 1 << PG_active |
551 1 << PG_dirty |
552 1 << PG_reclaim |
Hugh Dickins334795e2005-06-21 17:15:08 -0700553 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700555 1 << PG_writeback |
Nick Piggin676165a2006-04-10 11:21:48 +1000556 1 << PG_reserved |
557 1 << PG_buddy ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800558 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559
Hugh Dickins689bceb2005-11-21 21:32:20 -0800560 /*
561 * For now, we report if PG_reserved was found set, but do not
562 * clear it, and do not allocate the page: as a safety net.
563 */
564 if (PageReserved(page))
565 return 1;
566
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
568 1 << PG_referenced | 1 << PG_arch_1 |
569 1 << PG_checked | 1 << PG_mappedtodisk);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700570 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800571 set_page_refcounted(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800573
574 if (gfp_flags & __GFP_ZERO)
575 prep_zero_page(page, order, gfp_flags);
576
577 if (order && (gfp_flags & __GFP_COMP))
578 prep_compound_page(page, order);
579
Hugh Dickins689bceb2005-11-21 21:32:20 -0800580 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581}
582
583/*
584 * Do the hard work of removing an element from the buddy allocator.
585 * Call me with the zone->lock already held.
586 */
587static struct page *__rmqueue(struct zone *zone, unsigned int order)
588{
589 struct free_area * area;
590 unsigned int current_order;
591 struct page *page;
592
593 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
594 area = zone->free_area + current_order;
595 if (list_empty(&area->free_list))
596 continue;
597
598 page = list_entry(area->free_list.next, struct page, lru);
599 list_del(&page->lru);
600 rmv_page_order(page);
601 area->nr_free--;
602 zone->free_pages -= 1UL << order;
Nick Piggin085cc7d2006-01-06 00:11:01 -0800603 expand(zone, page, order, current_order, area);
604 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 }
606
607 return NULL;
608}
609
610/*
611 * Obtain a specified number of elements from the buddy allocator, all under
612 * a single hold of the lock, for efficiency. Add them to the supplied list.
613 * Returns the number of new pages which were placed at *list.
614 */
615static int rmqueue_bulk(struct zone *zone, unsigned int order,
616 unsigned long count, struct list_head *list)
617{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619
Nick Pigginc54ad302006-01-06 00:10:56 -0800620 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 for (i = 0; i < count; ++i) {
Nick Piggin085cc7d2006-01-06 00:11:01 -0800622 struct page *page = __rmqueue(zone, order);
623 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 list_add_tail(&page->lru, list);
626 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800627 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800628 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629}
630
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700631#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800632/*
633 * Called from the slab reaper to drain pagesets on a particular node that
634 * belong to the currently executing processor.
Christoph Lameter879336c2006-03-22 00:09:08 -0800635 * Note that this function must be called with the thread pinned to
636 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800637 */
638void drain_node_pages(int nodeid)
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700639{
Christoph Lameter2f6726e2006-09-25 23:31:18 -0700640 int i;
641 enum zone_type z;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700642 unsigned long flags;
643
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800644 for (z = 0; z < MAX_NR_ZONES; z++) {
645 struct zone *zone = NODE_DATA(nodeid)->node_zones + z;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700646 struct per_cpu_pageset *pset;
647
Nick Piggin23316bc2006-01-08 01:00:41 -0800648 pset = zone_pcp(zone, smp_processor_id());
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700649 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
650 struct per_cpu_pages *pcp;
651
652 pcp = &pset->pcp[i];
Christoph Lameter879336c2006-03-22 00:09:08 -0800653 if (pcp->count) {
654 local_irq_save(flags);
655 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
656 pcp->count = 0;
657 local_irq_restore(flags);
658 }
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700659 }
660 }
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700661}
662#endif
663
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664#if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
665static void __drain_pages(unsigned int cpu)
666{
Nick Pigginc54ad302006-01-06 00:10:56 -0800667 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 struct zone *zone;
669 int i;
670
671 for_each_zone(zone) {
672 struct per_cpu_pageset *pset;
673
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700674 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
676 struct per_cpu_pages *pcp;
677
678 pcp = &pset->pcp[i];
Nick Pigginc54ad302006-01-06 00:10:56 -0800679 local_irq_save(flags);
Nick Piggin48db57f2006-01-08 01:00:42 -0800680 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
681 pcp->count = 0;
Nick Pigginc54ad302006-01-06 00:10:56 -0800682 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 }
684 }
685}
686#endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
687
688#ifdef CONFIG_PM
689
690void mark_free_pages(struct zone *zone)
691{
692 unsigned long zone_pfn, flags;
693 int order;
694 struct list_head *curr;
695
696 if (!zone->spanned_pages)
697 return;
698
699 spin_lock_irqsave(&zone->lock, flags);
700 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
701 ClearPageNosaveFree(pfn_to_page(zone_pfn + zone->zone_start_pfn));
702
703 for (order = MAX_ORDER - 1; order >= 0; --order)
704 list_for_each(curr, &zone->free_area[order].free_list) {
705 unsigned long start_pfn, i;
706
707 start_pfn = page_to_pfn(list_entry(curr, struct page, lru));
708
709 for (i=0; i < (1<<order); i++)
710 SetPageNosaveFree(pfn_to_page(start_pfn+i));
711 }
712 spin_unlock_irqrestore(&zone->lock, flags);
713}
714
715/*
716 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
717 */
718void drain_local_pages(void)
719{
720 unsigned long flags;
721
722 local_irq_save(flags);
723 __drain_pages(smp_processor_id());
724 local_irq_restore(flags);
725}
726#endif /* CONFIG_PM */
727
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728/*
729 * Free a 0-order page
730 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731static void fastcall free_hot_cold_page(struct page *page, int cold)
732{
733 struct zone *zone = page_zone(page);
734 struct per_cpu_pages *pcp;
735 unsigned long flags;
736
737 arch_free_page(page, 0);
738
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 if (PageAnon(page))
740 page->mapping = NULL;
Nick Piggin224abf92006-01-06 00:11:11 -0800741 if (free_pages_check(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -0800742 return;
743
Hugh Dickins689bceb2005-11-21 21:32:20 -0800744 kernel_map_pages(page, 1, 0);
745
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700746 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700748 __count_vm_event(PGFREE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 list_add(&page->lru, &pcp->list);
750 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -0800751 if (pcp->count >= pcp->high) {
752 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
753 pcp->count -= pcp->batch;
754 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 local_irq_restore(flags);
756 put_cpu();
757}
758
759void fastcall free_hot_page(struct page *page)
760{
761 free_hot_cold_page(page, 0);
762}
763
764void fastcall free_cold_page(struct page *page)
765{
766 free_hot_cold_page(page, 1);
767}
768
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800769/*
770 * split_page takes a non-compound higher-order page, and splits it into
771 * n (1<<order) sub-pages: page[0..n]
772 * Each sub-page must be freed individually.
773 *
774 * Note: this is probably too low level an operation for use in drivers.
775 * Please consult with lkml before using this in your driver.
776 */
777void split_page(struct page *page, unsigned int order)
778{
779 int i;
780
Nick Piggin725d7042006-09-25 23:30:55 -0700781 VM_BUG_ON(PageCompound(page));
782 VM_BUG_ON(!page_count(page));
Nick Piggin7835e982006-03-22 00:08:40 -0800783 for (i = 1; i < (1 << order); i++)
784 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800785}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800786
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787/*
788 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
789 * we cheat by calling it from here, in the order > 0 path. Saves a branch
790 * or two.
791 */
Nick Piggina74609f2006-01-06 00:11:20 -0800792static struct page *buffered_rmqueue(struct zonelist *zonelist,
793 struct zone *zone, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794{
795 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800796 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 int cold = !!(gfp_flags & __GFP_COLD);
Nick Piggina74609f2006-01-06 00:11:20 -0800798 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799
Hugh Dickins689bceb2005-11-21 21:32:20 -0800800again:
Nick Piggina74609f2006-01-06 00:11:20 -0800801 cpu = get_cpu();
Nick Piggin48db57f2006-01-08 01:00:42 -0800802 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 struct per_cpu_pages *pcp;
804
Nick Piggina74609f2006-01-06 00:11:20 -0800805 pcp = &zone_pcp(zone, cpu)->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -0800807 if (!pcp->count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 pcp->count += rmqueue_bulk(zone, 0,
809 pcp->batch, &pcp->list);
Nick Piggina74609f2006-01-06 00:11:20 -0800810 if (unlikely(!pcp->count))
811 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 }
Nick Piggina74609f2006-01-06 00:11:20 -0800813 page = list_entry(pcp->list.next, struct page, lru);
814 list_del(&page->lru);
815 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800816 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 spin_lock_irqsave(&zone->lock, flags);
818 page = __rmqueue(zone, order);
Nick Piggina74609f2006-01-06 00:11:20 -0800819 spin_unlock(&zone->lock);
820 if (!page)
821 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 }
823
Christoph Lameterf8891e52006-06-30 01:55:45 -0700824 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Christoph Lameterca889e62006-06-30 01:55:44 -0700825 zone_statistics(zonelist, zone);
Nick Piggina74609f2006-01-06 00:11:20 -0800826 local_irq_restore(flags);
827 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828
Nick Piggin725d7042006-09-25 23:30:55 -0700829 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -0800830 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -0800831 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 return page;
Nick Piggina74609f2006-01-06 00:11:20 -0800833
834failed:
835 local_irq_restore(flags);
836 put_cpu();
837 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838}
839
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800840#define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
Nick Piggin31488902005-11-28 13:44:03 -0800841#define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
842#define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
843#define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
844#define ALLOC_HARDER 0x10 /* try to alloc harder */
845#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
846#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800847
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848/*
849 * Return 1 if free pages are above 'mark'. This takes into account the order
850 * of the allocation.
851 */
852int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800853 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854{
855 /* free_pages my go negative - that's OK */
856 long min = mark, free_pages = z->free_pages - (1 << order) + 1;
857 int o;
858
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800859 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800861 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 min -= min / 4;
863
864 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
865 return 0;
866 for (o = 0; o < order; o++) {
867 /* At the next order, this order's pages become unavailable */
868 free_pages -= z->free_area[o].nr_free << o;
869
870 /* Require fewer higher order pages to be free */
871 min >>= 1;
872
873 if (free_pages <= min)
874 return 0;
875 }
876 return 1;
877}
878
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800879/*
880 * get_page_from_freeliest goes through the zonelist trying to allocate
881 * a page.
882 */
883static struct page *
884get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
885 struct zonelist *zonelist, int alloc_flags)
Martin Hicks753ee722005-06-21 17:14:41 -0700886{
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800887 struct zone **z = zonelist->zones;
888 struct page *page = NULL;
889 int classzone_idx = zone_idx(*z);
890
891 /*
892 * Go through the zonelist once, looking for a zone with enough free.
893 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
894 */
895 do {
896 if ((alloc_flags & ALLOC_CPUSET) &&
897 !cpuset_zone_allowed(*z, gfp_mask))
898 continue;
899
900 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -0800901 unsigned long mark;
902 if (alloc_flags & ALLOC_WMARK_MIN)
903 mark = (*z)->pages_min;
904 else if (alloc_flags & ALLOC_WMARK_LOW)
905 mark = (*z)->pages_low;
906 else
907 mark = (*z)->pages_high;
908 if (!zone_watermark_ok(*z, order, mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800909 classzone_idx, alloc_flags))
Christoph Lameter9eeff232006-01-18 17:42:31 -0800910 if (!zone_reclaim_mode ||
911 !zone_reclaim(*z, gfp_mask, order))
912 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800913 }
914
Nick Piggina74609f2006-01-06 00:11:20 -0800915 page = buffered_rmqueue(zonelist, *z, order, gfp_mask);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800916 if (page) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800917 break;
918 }
919 } while (*(++z) != NULL);
920 return page;
Martin Hicks753ee722005-06-21 17:14:41 -0700921}
922
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923/*
924 * This is the 'heart' of the zoned buddy allocator.
925 */
926struct page * fastcall
Al Virodd0fc662005-10-07 07:46:04 +0100927__alloc_pages(gfp_t gfp_mask, unsigned int order,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 struct zonelist *zonelist)
929{
Al Viro260b2362005-10-21 03:22:44 -0400930 const gfp_t wait = gfp_mask & __GFP_WAIT;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800931 struct zone **z;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 struct page *page;
933 struct reclaim_state reclaim_state;
934 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 int do_retry;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800936 int alloc_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 int did_some_progress;
938
939 might_sleep_if(wait);
940
Jens Axboe6b1de912005-11-17 21:35:02 +0100941restart:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800942 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800944 if (unlikely(*z == NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 /* Should this ever happen?? */
946 return NULL;
947 }
Jens Axboe6b1de912005-11-17 21:35:02 +0100948
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800949 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -0800950 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800951 if (page)
952 goto got_pg;
953
Jens Axboe6b1de912005-11-17 21:35:02 +0100954 do {
Chris Wright43b0bc02006-06-25 05:47:55 -0700955 wakeup_kswapd(*z, order);
Jens Axboe6b1de912005-11-17 21:35:02 +0100956 } while (*(++z));
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800957
Paul Jackson9bf22292005-09-06 15:18:12 -0700958 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800959 * OK, we're below the kswapd watermark and have kicked background
960 * reclaim. Now things get more complex, so set up alloc_flags according
961 * to how we want to proceed.
962 *
963 * The caller may dip into page reserves a bit more if the caller
964 * cannot run direct reclaim, or if the caller has realtime scheduling
Paul Jackson4eac9152006-01-11 12:17:19 -0800965 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
966 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
Paul Jackson9bf22292005-09-06 15:18:12 -0700967 */
Nick Piggin31488902005-11-28 13:44:03 -0800968 alloc_flags = ALLOC_WMARK_MIN;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800969 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
970 alloc_flags |= ALLOC_HARDER;
971 if (gfp_mask & __GFP_HIGH)
972 alloc_flags |= ALLOC_HIGH;
Paul Jacksonbdd804f2006-05-20 15:00:09 -0700973 if (wait)
974 alloc_flags |= ALLOC_CPUSET;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975
976 /*
977 * Go through the zonelist again. Let __GFP_HIGH and allocations
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800978 * coming from realtime tasks go deeper into reserves.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 *
980 * This is the last chance, in general, before the goto nopage.
981 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
Paul Jackson9bf22292005-09-06 15:18:12 -0700982 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800984 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
985 if (page)
986 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987
988 /* This allocation should allow future memory freeing. */
Nick Pigginb84a35b2005-05-01 08:58:36 -0700989
990 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
991 && !in_interrupt()) {
992 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
Kirill Korotaev885036d2005-11-13 16:06:41 -0800993nofail_alloc:
Nick Pigginb84a35b2005-05-01 08:58:36 -0700994 /* go through the zonelist yet again, ignoring mins */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800995 page = get_page_from_freelist(gfp_mask, order,
Paul Jackson47f3a862006-01-06 00:10:32 -0800996 zonelist, ALLOC_NO_WATERMARKS);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800997 if (page)
998 goto got_pg;
Kirill Korotaev885036d2005-11-13 16:06:41 -0800999 if (gfp_mask & __GFP_NOFAIL) {
1000 blk_congestion_wait(WRITE, HZ/50);
1001 goto nofail_alloc;
1002 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 }
1004 goto nopage;
1005 }
1006
1007 /* Atomic allocations - we can't balance anything */
1008 if (!wait)
1009 goto nopage;
1010
1011rebalance:
1012 cond_resched();
1013
1014 /* We now go into synchronous reclaim */
Paul Jackson3e0d98b2006-01-08 01:01:49 -08001015 cpuset_memory_pressure_bump();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 p->flags |= PF_MEMALLOC;
1017 reclaim_state.reclaimed_slab = 0;
1018 p->reclaim_state = &reclaim_state;
1019
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001020 did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021
1022 p->reclaim_state = NULL;
1023 p->flags &= ~PF_MEMALLOC;
1024
1025 cond_resched();
1026
1027 if (likely(did_some_progress)) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001028 page = get_page_from_freelist(gfp_mask, order,
1029 zonelist, alloc_flags);
1030 if (page)
1031 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1033 /*
1034 * Go through the zonelist yet one more time, keep
1035 * very high watermark here, this is only to catch
1036 * a parallel oom killing, we must fail if we're still
1037 * under heavy pressure.
1038 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001039 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -08001040 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001041 if (page)
1042 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043
Christoph Lameter9b0f8b02006-02-20 18:27:52 -08001044 out_of_memory(zonelist, gfp_mask, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 goto restart;
1046 }
1047
1048 /*
1049 * Don't let big-order allocations loop unless the caller explicitly
1050 * requests that. Wait for some write requests to complete then retry.
1051 *
1052 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1053 * <= 3, but that may not be true in other implementations.
1054 */
1055 do_retry = 0;
1056 if (!(gfp_mask & __GFP_NORETRY)) {
1057 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
1058 do_retry = 1;
1059 if (gfp_mask & __GFP_NOFAIL)
1060 do_retry = 1;
1061 }
1062 if (do_retry) {
1063 blk_congestion_wait(WRITE, HZ/50);
1064 goto rebalance;
1065 }
1066
1067nopage:
1068 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1069 printk(KERN_WARNING "%s: page allocation failure."
1070 " order:%d, mode:0x%x\n",
1071 p->comm, order, gfp_mask);
1072 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001073 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075got_pg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 return page;
1077}
1078
1079EXPORT_SYMBOL(__alloc_pages);
1080
1081/*
1082 * Common helper functions.
1083 */
Al Virodd0fc662005-10-07 07:46:04 +01001084fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085{
1086 struct page * page;
1087 page = alloc_pages(gfp_mask, order);
1088 if (!page)
1089 return 0;
1090 return (unsigned long) page_address(page);
1091}
1092
1093EXPORT_SYMBOL(__get_free_pages);
1094
Al Virodd0fc662005-10-07 07:46:04 +01001095fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096{
1097 struct page * page;
1098
1099 /*
1100 * get_zeroed_page() returns a 32-bit address, which cannot represent
1101 * a highmem page
1102 */
Nick Piggin725d7042006-09-25 23:30:55 -07001103 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104
1105 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1106 if (page)
1107 return (unsigned long) page_address(page);
1108 return 0;
1109}
1110
1111EXPORT_SYMBOL(get_zeroed_page);
1112
1113void __pagevec_free(struct pagevec *pvec)
1114{
1115 int i = pagevec_count(pvec);
1116
1117 while (--i >= 0)
1118 free_hot_cold_page(pvec->pages[i], pvec->cold);
1119}
1120
1121fastcall void __free_pages(struct page *page, unsigned int order)
1122{
Nick Pigginb5810032005-10-29 18:16:12 -07001123 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 if (order == 0)
1125 free_hot_page(page);
1126 else
1127 __free_pages_ok(page, order);
1128 }
1129}
1130
1131EXPORT_SYMBOL(__free_pages);
1132
1133fastcall void free_pages(unsigned long addr, unsigned int order)
1134{
1135 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07001136 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 __free_pages(virt_to_page((void *)addr), order);
1138 }
1139}
1140
1141EXPORT_SYMBOL(free_pages);
1142
1143/*
1144 * Total amount of free (allocatable) RAM:
1145 */
1146unsigned int nr_free_pages(void)
1147{
1148 unsigned int sum = 0;
1149 struct zone *zone;
1150
1151 for_each_zone(zone)
1152 sum += zone->free_pages;
1153
1154 return sum;
1155}
1156
1157EXPORT_SYMBOL(nr_free_pages);
1158
1159#ifdef CONFIG_NUMA
1160unsigned int nr_free_pages_pgdat(pg_data_t *pgdat)
1161{
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001162 unsigned int sum = 0;
1163 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164
1165 for (i = 0; i < MAX_NR_ZONES; i++)
1166 sum += pgdat->node_zones[i].free_pages;
1167
1168 return sum;
1169}
1170#endif
1171
1172static unsigned int nr_free_zone_pages(int offset)
1173{
Martin J. Blighe310fd42005-07-29 22:59:18 -07001174 /* Just pick one node, since fallback list is circular */
1175 pg_data_t *pgdat = NODE_DATA(numa_node_id());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 unsigned int sum = 0;
1177
Martin J. Blighe310fd42005-07-29 22:59:18 -07001178 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1179 struct zone **zonep = zonelist->zones;
1180 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181
Martin J. Blighe310fd42005-07-29 22:59:18 -07001182 for (zone = *zonep++; zone; zone = *zonep++) {
1183 unsigned long size = zone->present_pages;
1184 unsigned long high = zone->pages_high;
1185 if (size > high)
1186 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 }
1188
1189 return sum;
1190}
1191
1192/*
1193 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1194 */
1195unsigned int nr_free_buffer_pages(void)
1196{
Al Viroaf4ca452005-10-21 02:55:38 -04001197 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198}
1199
1200/*
1201 * Amount of free RAM allocatable within all zones
1202 */
1203unsigned int nr_free_pagecache_pages(void)
1204{
Al Viroaf4ca452005-10-21 02:55:38 -04001205 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207#ifdef CONFIG_NUMA
1208static void show_node(struct zone *zone)
1209{
1210 printk("Node %d ", zone->zone_pgdat->node_id);
1211}
1212#else
1213#define show_node(zone) do { } while (0)
1214#endif
1215
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216void si_meminfo(struct sysinfo *val)
1217{
1218 val->totalram = totalram_pages;
1219 val->sharedram = 0;
1220 val->freeram = nr_free_pages();
1221 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 val->totalhigh = totalhigh_pages;
1223 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 val->mem_unit = PAGE_SIZE;
1225}
1226
1227EXPORT_SYMBOL(si_meminfo);
1228
1229#ifdef CONFIG_NUMA
1230void si_meminfo_node(struct sysinfo *val, int nid)
1231{
1232 pg_data_t *pgdat = NODE_DATA(nid);
1233
1234 val->totalram = pgdat->node_present_pages;
1235 val->freeram = nr_free_pages_pgdat(pgdat);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001236#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1238 val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages;
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001239#else
1240 val->totalhigh = 0;
1241 val->freehigh = 0;
1242#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 val->mem_unit = PAGE_SIZE;
1244}
1245#endif
1246
1247#define K(x) ((x) << (PAGE_SHIFT-10))
1248
1249/*
1250 * Show free area list (used inside shift_scroll-lock stuff)
1251 * We also calculate the percentage fragmentation. We do this by counting the
1252 * memory on each free list with the exception of the first item on the list.
1253 */
1254void show_free_areas(void)
1255{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 int cpu, temperature;
1257 unsigned long active;
1258 unsigned long inactive;
1259 unsigned long free;
1260 struct zone *zone;
1261
1262 for_each_zone(zone) {
1263 show_node(zone);
1264 printk("%s per-cpu:", zone->name);
1265
Con Kolivasf3fe6512006-01-06 00:11:15 -08001266 if (!populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 printk(" empty\n");
1268 continue;
1269 } else
1270 printk("\n");
1271
Dave Jones6b482c62005-11-10 15:45:56 -05001272 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 struct per_cpu_pageset *pageset;
1274
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001275 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276
1277 for (temperature = 0; temperature < 2; temperature++)
Nick Piggin2d92c5c2006-01-06 00:10:59 -08001278 printk("cpu %d %s: high %d, batch %d used:%d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 cpu,
1280 temperature ? "cold" : "hot",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 pageset->pcp[temperature].high,
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001282 pageset->pcp[temperature].batch,
1283 pageset->pcp[temperature].count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 }
1285 }
1286
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 get_zone_counts(&active, &inactive, &free);
1288
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 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 Lameter2f6726e2006-09-25 23:31:18 -07001363 struct zonelist *zonelist, int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364{
Christoph Lameter1a932052006-01-06 00:11:16 -08001365 struct zone *zone;
1366
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001367 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001368 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08001369
1370 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001371 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08001372 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08001373 if (populated_zone(zone)) {
Christoph Lameter070f8032006-01-06 00:11:19 -08001374 zonelist->zones[nr_zones++] = zone;
1375 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08001377
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001378 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08001379 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380}
1381
1382#ifdef CONFIG_NUMA
1383#define MAX_NODE_LOAD (num_online_nodes())
Yasunori Goto86356ab2006-06-23 02:03:09 -07001384static int __meminitdata node_load[MAX_NUMNODES];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001386 * 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 -07001387 * @node: node whose fallback list we're appending
1388 * @used_node_mask: nodemask_t of already used nodes
1389 *
1390 * We use a number of factors to determine which is the next node that should
1391 * appear on a given node's fallback list. The node should not have appeared
1392 * already in @node's fallback list, and it should be the next closest node
1393 * according to the distance array (which contains arbitrary distance values
1394 * from each node to each node in the system), and should also prefer nodes
1395 * with no CPUs, since presumably they'll have very little allocation pressure
1396 * on them otherwise.
1397 * It returns -1 if no node is found.
1398 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07001399static int __meminit find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001401 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 int min_val = INT_MAX;
1403 int best_node = -1;
1404
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001405 /* Use the local node if we haven't already */
1406 if (!node_isset(node, *used_node_mask)) {
1407 node_set(node, *used_node_mask);
1408 return node;
1409 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001411 for_each_online_node(n) {
1412 cpumask_t tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413
1414 /* Don't want a node to appear more than once */
1415 if (node_isset(n, *used_node_mask))
1416 continue;
1417
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 /* Use the distance array to find the distance */
1419 val = node_distance(node, n);
1420
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001421 /* Penalize nodes under us ("prefer the next node") */
1422 val += (n < node);
1423
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 /* Give preference to headless and unused nodes */
1425 tmp = node_to_cpumask(n);
1426 if (!cpus_empty(tmp))
1427 val += PENALTY_FOR_NODE_WITH_CPUS;
1428
1429 /* Slight preference for less loaded node */
1430 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1431 val += node_load[n];
1432
1433 if (val < min_val) {
1434 min_val = val;
1435 best_node = n;
1436 }
1437 }
1438
1439 if (best_node >= 0)
1440 node_set(best_node, *used_node_mask);
1441
1442 return best_node;
1443}
1444
Yasunori Goto86356ab2006-06-23 02:03:09 -07001445static void __meminit build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446{
Christoph Lameter19655d32006-09-25 23:31:19 -07001447 int j, node, local_node;
1448 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 int prev_node, load;
1450 struct zonelist *zonelist;
1451 nodemask_t used_mask;
1452
1453 /* initialize zonelists */
Christoph Lameter19655d32006-09-25 23:31:19 -07001454 for (i = 0; i < MAX_NR_ZONES; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 zonelist = pgdat->node_zonelists + i;
1456 zonelist->zones[0] = NULL;
1457 }
1458
1459 /* NUMA-aware ordering of nodes */
1460 local_node = pgdat->node_id;
1461 load = num_online_nodes();
1462 prev_node = local_node;
1463 nodes_clear(used_mask);
1464 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08001465 int distance = node_distance(local_node, node);
1466
1467 /*
1468 * If another node is sufficiently far away then it is better
1469 * to reclaim pages in a zone before going off node.
1470 */
1471 if (distance > RECLAIM_DISTANCE)
1472 zone_reclaim_mode = 1;
1473
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 /*
1475 * We don't want to pressure a particular node.
1476 * So adding penalty to the first node in same
1477 * distance group to make it round-robin.
1478 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08001479
1480 if (distance != node_distance(local_node, prev_node))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 node_load[node] += load;
1482 prev_node = node;
1483 load--;
Christoph Lameter19655d32006-09-25 23:31:19 -07001484 for (i = 0; i < MAX_NR_ZONES; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 zonelist = pgdat->node_zonelists + i;
1486 for (j = 0; zonelist->zones[j] != NULL; j++);
1487
Christoph Lameter19655d32006-09-25 23:31:19 -07001488 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 zonelist->zones[j] = NULL;
1490 }
1491 }
1492}
1493
1494#else /* CONFIG_NUMA */
1495
Yasunori Goto86356ab2006-06-23 02:03:09 -07001496static void __meminit build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497{
Christoph Lameter19655d32006-09-25 23:31:19 -07001498 int node, local_node;
1499 enum zone_type i,j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500
1501 local_node = pgdat->node_id;
Christoph Lameter19655d32006-09-25 23:31:19 -07001502 for (i = 0; i < MAX_NR_ZONES; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 struct zonelist *zonelist;
1504
1505 zonelist = pgdat->node_zonelists + i;
1506
Christoph Lameter19655d32006-09-25 23:31:19 -07001507 j = build_zonelists_node(pgdat, zonelist, 0, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 /*
1509 * Now we build the zonelist so that it contains the zones
1510 * of all the other nodes.
1511 * We don't want to pressure a particular node, so when
1512 * building the zones for node N, we make sure that the
1513 * zones coming right after the local ones are those from
1514 * node N+1 (modulo N)
1515 */
1516 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1517 if (!node_online(node))
1518 continue;
Christoph Lameter19655d32006-09-25 23:31:19 -07001519 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 }
1521 for (node = 0; node < local_node; node++) {
1522 if (!node_online(node))
1523 continue;
Christoph Lameter19655d32006-09-25 23:31:19 -07001524 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 }
1526
1527 zonelist->zones[j] = NULL;
1528 }
1529}
1530
1531#endif /* CONFIG_NUMA */
1532
Yasunori Goto68113782006-06-23 02:03:11 -07001533/* return values int ....just for stop_machine_run() */
1534static int __meminit __build_all_zonelists(void *dummy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535{
Yasunori Goto68113782006-06-23 02:03:11 -07001536 int nid;
1537 for_each_online_node(nid)
1538 build_zonelists(NODE_DATA(nid));
1539 return 0;
1540}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541
Yasunori Goto68113782006-06-23 02:03:11 -07001542void __meminit build_all_zonelists(void)
1543{
1544 if (system_state == SYSTEM_BOOTING) {
1545 __build_all_zonelists(0);
1546 cpuset_init_current_mems_allowed();
1547 } else {
1548 /* we have to stop all cpus to guaranntee there is no user
1549 of zonelist */
1550 stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
1551 /* cpuset refresh routine should be here */
1552 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07001553 vm_total_pages = nr_free_pagecache_pages();
1554 printk("Built %i zonelists. Total pages: %ld\n",
1555 num_online_nodes(), vm_total_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556}
1557
1558/*
1559 * Helper functions to size the waitqueue hash table.
1560 * Essentially these want to choose hash table sizes sufficiently
1561 * large so that collisions trying to wait on pages are rare.
1562 * But in fact, the number of active page waitqueues on typical
1563 * systems is ridiculously low, less than 200. So this is even
1564 * conservative, even though it seems large.
1565 *
1566 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1567 * waitqueues, i.e. the size of the waitq table given the number of pages.
1568 */
1569#define PAGES_PER_WAITQUEUE 256
1570
Yasunori Gotocca448f2006-06-23 02:03:10 -07001571#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07001572static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573{
1574 unsigned long size = 1;
1575
1576 pages /= PAGES_PER_WAITQUEUE;
1577
1578 while (size < pages)
1579 size <<= 1;
1580
1581 /*
1582 * Once we have dozens or even hundreds of threads sleeping
1583 * on IO we've got bigger problems than wait queue collision.
1584 * Limit the size of the wait table to a reasonable size.
1585 */
1586 size = min(size, 4096UL);
1587
1588 return max(size, 4UL);
1589}
Yasunori Gotocca448f2006-06-23 02:03:10 -07001590#else
1591/*
1592 * A zone's size might be changed by hot-add, so it is not possible to determine
1593 * a suitable size for its wait_table. So we use the maximum size now.
1594 *
1595 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
1596 *
1597 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
1598 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
1599 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
1600 *
1601 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
1602 * or more by the traditional way. (See above). It equals:
1603 *
1604 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
1605 * ia64(16K page size) : = ( 8G + 4M)byte.
1606 * powerpc (64K page size) : = (32G +16M)byte.
1607 */
1608static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
1609{
1610 return 4096UL;
1611}
1612#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613
1614/*
1615 * This is an integer logarithm so that shifts can be used later
1616 * to extract the more random high bits from the multiplicative
1617 * hash function before the remainder is taken.
1618 */
1619static inline unsigned long wait_table_bits(unsigned long size)
1620{
1621 return ffz(~size);
1622}
1623
1624#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1625
1626static void __init calculate_zone_totalpages(struct pglist_data *pgdat,
1627 unsigned long *zones_size, unsigned long *zholes_size)
1628{
1629 unsigned long realtotalpages, totalpages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001630 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631
1632 for (i = 0; i < MAX_NR_ZONES; i++)
1633 totalpages += zones_size[i];
1634 pgdat->node_spanned_pages = totalpages;
1635
1636 realtotalpages = totalpages;
1637 if (zholes_size)
1638 for (i = 0; i < MAX_NR_ZONES; i++)
1639 realtotalpages -= zholes_size[i];
1640 pgdat->node_present_pages = realtotalpages;
1641 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
1642}
1643
1644
1645/*
1646 * Initially all pages are reserved - free ones are freed
1647 * up by free_all_bootmem() once the early boot process is
1648 * done. Non-atomic initialization, single-pass.
1649 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001650void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 unsigned long start_pfn)
1652{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07001654 unsigned long end_pfn = start_pfn + size;
1655 unsigned long pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656
Greg Ungerercbe8dd42006-01-12 01:05:24 -08001657 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001658 if (!early_pfn_valid(pfn))
1659 continue;
1660 page = pfn_to_page(pfn);
1661 set_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08001662 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 reset_page_mapcount(page);
1664 SetPageReserved(page);
1665 INIT_LIST_HEAD(&page->lru);
1666#ifdef WANT_PAGE_VIRTUAL
1667 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1668 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07001669 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 }
1672}
1673
1674void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1675 unsigned long size)
1676{
1677 int order;
1678 for (order = 0; order < MAX_ORDER ; order++) {
1679 INIT_LIST_HEAD(&zone->free_area[order].free_list);
1680 zone->free_area[order].nr_free = 0;
1681 }
1682}
1683
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001684#define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
Christoph Lameter2f1b6242006-09-25 23:31:13 -07001685void zonetable_add(struct zone *zone, int nid, enum zone_type zid,
1686 unsigned long pfn, unsigned long size)
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001687{
1688 unsigned long snum = pfn_to_section_nr(pfn);
1689 unsigned long end = pfn_to_section_nr(pfn + size);
1690
1691 if (FLAGS_HAS_NODE)
1692 zone_table[ZONETABLE_INDEX(nid, zid)] = zone;
1693 else
1694 for (; snum <= end; snum++)
1695 zone_table[ZONETABLE_INDEX(snum, zid)] = zone;
1696}
1697
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698#ifndef __HAVE_ARCH_MEMMAP_INIT
1699#define memmap_init(size, nid, zone, start_pfn) \
1700 memmap_init_zone((size), (nid), (zone), (start_pfn))
1701#endif
1702
Ashok Raj6292d9a2006-02-01 03:04:44 -08001703static int __cpuinit zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001704{
1705 int batch;
1706
1707 /*
1708 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07001709 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001710 *
1711 * OK, so we don't know how big the cache is. So guess.
1712 */
1713 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07001714 if (batch * PAGE_SIZE > 512 * 1024)
1715 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001716 batch /= 4; /* We effectively *= 4 below */
1717 if (batch < 1)
1718 batch = 1;
1719
1720 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001721 * Clamp the batch to a 2^n - 1 value. Having a power
1722 * of 2 value was found to be more likely to have
1723 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001724 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001725 * For example if 2 tasks are alternately allocating
1726 * batches of pages, one task can end up with a lot
1727 * of pages of one half of the possible page colors
1728 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001729 */
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001730 batch = (1 << (fls(batch + batch/2)-1)) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07001731
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001732 return batch;
1733}
1734
Christoph Lameter2caaad42005-06-21 17:15:00 -07001735inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
1736{
1737 struct per_cpu_pages *pcp;
1738
Magnus Damm1c6fe942005-10-26 01:58:59 -07001739 memset(p, 0, sizeof(*p));
1740
Christoph Lameter2caaad42005-06-21 17:15:00 -07001741 pcp = &p->pcp[0]; /* hot */
1742 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001743 pcp->high = 6 * batch;
1744 pcp->batch = max(1UL, 1 * batch);
1745 INIT_LIST_HEAD(&pcp->list);
1746
1747 pcp = &p->pcp[1]; /* cold*/
1748 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001749 pcp->high = 2 * batch;
Seth, Rohite46a5e22005-10-29 18:15:48 -07001750 pcp->batch = max(1UL, batch/2);
Christoph Lameter2caaad42005-06-21 17:15:00 -07001751 INIT_LIST_HEAD(&pcp->list);
1752}
1753
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08001754/*
1755 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
1756 * to the value high for the pageset p.
1757 */
1758
1759static void setup_pagelist_highmark(struct per_cpu_pageset *p,
1760 unsigned long high)
1761{
1762 struct per_cpu_pages *pcp;
1763
1764 pcp = &p->pcp[0]; /* hot list */
1765 pcp->high = high;
1766 pcp->batch = max(1UL, high/4);
1767 if ((high/4) > (PAGE_SHIFT * 8))
1768 pcp->batch = PAGE_SHIFT * 8;
1769}
1770
1771
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001772#ifdef CONFIG_NUMA
1773/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07001774 * Boot pageset table. One per cpu which is going to be used for all
1775 * zones and all nodes. The parameters will be set in such a way
1776 * that an item put on a list will immediately be handed over to
1777 * the buddy list. This is safe since pageset manipulation is done
1778 * with interrupts disabled.
1779 *
1780 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07001781 *
1782 * The boot_pagesets must be kept even after bootup is complete for
1783 * unused processors and/or zones. They do play a role for bootstrapping
1784 * hotplugged processors.
1785 *
1786 * zoneinfo_show() and maybe other functions do
1787 * not check if the processor is online before following the pageset pointer.
1788 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07001789 */
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08001790static struct per_cpu_pageset boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07001791
1792/*
1793 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001794 * per cpu pageset array in struct zone.
1795 */
Ashok Raj6292d9a2006-02-01 03:04:44 -08001796static int __cpuinit process_zones(int cpu)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001797{
1798 struct zone *zone, *dzone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001799
1800 for_each_zone(zone) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001801
Nick Piggin23316bc2006-01-08 01:00:41 -08001802 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001803 GFP_KERNEL, cpu_to_node(cpu));
Nick Piggin23316bc2006-01-08 01:00:41 -08001804 if (!zone_pcp(zone, cpu))
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001805 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001806
Nick Piggin23316bc2006-01-08 01:00:41 -08001807 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08001808
1809 if (percpu_pagelist_fraction)
1810 setup_pagelist_highmark(zone_pcp(zone, cpu),
1811 (zone->present_pages / percpu_pagelist_fraction));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001812 }
1813
1814 return 0;
1815bad:
1816 for_each_zone(dzone) {
1817 if (dzone == zone)
1818 break;
Nick Piggin23316bc2006-01-08 01:00:41 -08001819 kfree(zone_pcp(dzone, cpu));
1820 zone_pcp(dzone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001821 }
1822 return -ENOMEM;
1823}
1824
1825static inline void free_zone_pagesets(int cpu)
1826{
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001827 struct zone *zone;
1828
1829 for_each_zone(zone) {
1830 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
1831
David Rientjesf3ef9ea2006-09-25 16:24:57 -07001832 /* Free per_cpu_pageset if it is slab allocated */
1833 if (pset != &boot_pageset[cpu])
1834 kfree(pset);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001835 zone_pcp(zone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001836 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001837}
1838
Chandra Seetharaman9c7b2162006-06-27 02:54:07 -07001839static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001840 unsigned long action,
1841 void *hcpu)
1842{
1843 int cpu = (long)hcpu;
1844 int ret = NOTIFY_OK;
1845
1846 switch (action) {
1847 case CPU_UP_PREPARE:
1848 if (process_zones(cpu))
1849 ret = NOTIFY_BAD;
1850 break;
Andi Kleenb0d41692005-11-05 17:25:53 +01001851 case CPU_UP_CANCELED:
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001852 case CPU_DEAD:
1853 free_zone_pagesets(cpu);
1854 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001855 default:
1856 break;
1857 }
1858 return ret;
1859}
1860
Chandra Seetharaman74b85f32006-06-27 02:54:09 -07001861static struct notifier_block __cpuinitdata pageset_notifier =
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001862 { &pageset_cpuup_callback, NULL, 0 };
1863
Al Viro78d99552005-12-15 09:18:25 +00001864void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001865{
1866 int err;
1867
1868 /* Initialize per_cpu_pageset for cpu 0.
1869 * A cpuup callback will do this for every cpu
1870 * as it comes online
1871 */
1872 err = process_zones(smp_processor_id());
1873 BUG_ON(err);
1874 register_cpu_notifier(&pageset_notifier);
1875}
1876
1877#endif
1878
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001879static __meminit
Yasunori Gotocca448f2006-06-23 02:03:10 -07001880int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07001881{
1882 int i;
1883 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07001884 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07001885
1886 /*
1887 * The per-page waitqueue mechanism uses hashed waitqueues
1888 * per zone.
1889 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07001890 zone->wait_table_hash_nr_entries =
1891 wait_table_hash_nr_entries(zone_size_pages);
1892 zone->wait_table_bits =
1893 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07001894 alloc_size = zone->wait_table_hash_nr_entries
1895 * sizeof(wait_queue_head_t);
1896
1897 if (system_state == SYSTEM_BOOTING) {
1898 zone->wait_table = (wait_queue_head_t *)
1899 alloc_bootmem_node(pgdat, alloc_size);
1900 } else {
1901 /*
1902 * This case means that a zone whose size was 0 gets new memory
1903 * via memory hot-add.
1904 * But it may be the case that a new node was hot-added. In
1905 * this case vmalloc() will not be able to use this new node's
1906 * memory - this wait_table must be initialized to use this new
1907 * node itself as well.
1908 * To use this new node's memory, further consideration will be
1909 * necessary.
1910 */
1911 zone->wait_table = (wait_queue_head_t *)vmalloc(alloc_size);
1912 }
1913 if (!zone->wait_table)
1914 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07001915
Yasunori Goto02b694d2006-06-23 02:03:08 -07001916 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07001917 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07001918
1919 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07001920}
1921
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001922static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07001923{
1924 int cpu;
1925 unsigned long batch = zone_batchsize(zone);
1926
1927 for (cpu = 0; cpu < NR_CPUS; cpu++) {
1928#ifdef CONFIG_NUMA
1929 /* Early boot. Slab allocator not functional yet */
Nick Piggin23316bc2006-01-08 01:00:41 -08001930 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Dave Hansened8ece22005-10-29 18:16:50 -07001931 setup_pageset(&boot_pageset[cpu],0);
1932#else
1933 setup_pageset(zone_pcp(zone,cpu), batch);
1934#endif
1935 }
Anton Blanchardf5335c02006-03-25 03:06:49 -08001936 if (zone->present_pages)
1937 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
1938 zone->name, zone->present_pages, batch);
Dave Hansened8ece22005-10-29 18:16:50 -07001939}
1940
Yasunori Goto718127c2006-06-23 02:03:10 -07001941__meminit int init_currently_empty_zone(struct zone *zone,
1942 unsigned long zone_start_pfn,
1943 unsigned long size)
Dave Hansened8ece22005-10-29 18:16:50 -07001944{
1945 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07001946 int ret;
1947 ret = zone_wait_table_init(zone, size);
1948 if (ret)
1949 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07001950 pgdat->nr_zones = zone_idx(zone) + 1;
1951
Dave Hansened8ece22005-10-29 18:16:50 -07001952 zone->zone_start_pfn = zone_start_pfn;
1953
1954 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
1955
1956 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
Yasunori Goto718127c2006-06-23 02:03:10 -07001957
1958 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07001959}
1960
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961/*
1962 * Set up the zone data structures:
1963 * - mark all pages reserved
1964 * - mark all memory queues empty
1965 * - clear the memory bitmaps
1966 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07001967static void __meminit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 unsigned long *zones_size, unsigned long *zholes_size)
1969{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07001970 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07001971 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07001973 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974
Dave Hansen208d54e2005-10-29 18:16:52 -07001975 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 pgdat->nr_zones = 0;
1977 init_waitqueue_head(&pgdat->kswapd_wait);
1978 pgdat->kswapd_max_order = 0;
1979
1980 for (j = 0; j < MAX_NR_ZONES; j++) {
1981 struct zone *zone = pgdat->node_zones + j;
1982 unsigned long size, realsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 realsize = size = zones_size[j];
1985 if (zholes_size)
1986 realsize -= zholes_size[j];
1987
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001988 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 nr_kernel_pages += realsize;
1990 nr_all_pages += realsize;
1991
1992 zone->spanned_pages = size;
1993 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07001994#ifdef CONFIG_NUMA
1995 zone->min_unmapped_ratio = (realsize*sysctl_min_unmapped_ratio)
1996 / 100;
1997#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 zone->name = zone_names[j];
1999 spin_lock_init(&zone->lock);
2000 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07002001 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 zone->zone_pgdat = pgdat;
2003 zone->free_pages = 0;
2004
2005 zone->temp_priority = zone->prev_priority = DEF_PRIORITY;
2006
Dave Hansened8ece22005-10-29 18:16:50 -07002007 zone_pcp_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 INIT_LIST_HEAD(&zone->active_list);
2009 INIT_LIST_HEAD(&zone->inactive_list);
2010 zone->nr_scan_active = 0;
2011 zone->nr_scan_inactive = 0;
2012 zone->nr_active = 0;
2013 zone->nr_inactive = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07002014 zap_zone_vm_stats(zone);
Martin Hicks53e9a612005-09-03 15:54:51 -07002015 atomic_set(&zone->reclaim_in_progress, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 if (!size)
2017 continue;
2018
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002019 zonetable_add(zone, nid, j, zone_start_pfn, size);
Yasunori Goto718127c2006-06-23 02:03:10 -07002020 ret = init_currently_empty_zone(zone, zone_start_pfn, size);
2021 BUG_ON(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 }
2024}
2025
2026static void __init alloc_node_mem_map(struct pglist_data *pgdat)
2027{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028 /* Skip empty nodes */
2029 if (!pgdat->node_spanned_pages)
2030 return;
2031
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002032#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 /* ia64 gets its own node_mem_map, before this, without bootmem */
2034 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07002035 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002036 struct page *map;
2037
Bob Piccoe984bb42006-05-20 15:00:31 -07002038 /*
2039 * The zone's endpoints aren't required to be MAX_ORDER
2040 * aligned but the node_mem_map endpoints must be in order
2041 * for the buddy allocator to function correctly.
2042 */
2043 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
2044 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
2045 end = ALIGN(end, MAX_ORDER_NR_PAGES);
2046 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07002047 map = alloc_remap(pgdat->node_id, size);
2048 if (!map)
2049 map = alloc_bootmem_node(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07002050 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002052#ifdef CONFIG_FLATMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 /*
2054 * With no DISCONTIG, the global mem_map is just set as node 0's
2055 */
2056 if (pgdat == NODE_DATA(0))
2057 mem_map = NODE_DATA(0)->node_mem_map;
2058#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002059#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060}
2061
Yasunori Goto86356ab2006-06-23 02:03:09 -07002062void __meminit free_area_init_node(int nid, struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 unsigned long *zones_size, unsigned long node_start_pfn,
2064 unsigned long *zholes_size)
2065{
2066 pgdat->node_id = nid;
2067 pgdat->node_start_pfn = node_start_pfn;
2068 calculate_zone_totalpages(pgdat, zones_size, zholes_size);
2069
2070 alloc_node_mem_map(pgdat);
2071
2072 free_area_init_core(pgdat, zones_size, zholes_size);
2073}
2074
Dave Hansen93b75042005-06-23 00:07:47 -07002075#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076static bootmem_data_t contig_bootmem_data;
2077struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
2078
2079EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07002080#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081
2082void __init free_area_init(unsigned long *zones_size)
2083{
Dave Hansen93b75042005-06-23 00:07:47 -07002084 free_area_init_node(0, NODE_DATA(0), zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
2086}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088#ifdef CONFIG_HOTPLUG_CPU
2089static int page_alloc_cpu_notify(struct notifier_block *self,
2090 unsigned long action, void *hcpu)
2091{
2092 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093
2094 if (action == CPU_DEAD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 local_irq_disable();
2096 __drain_pages(cpu);
Christoph Lameterf8891e52006-06-30 01:55:45 -07002097 vm_events_fold_cpu(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098 local_irq_enable();
Christoph Lameter2244b952006-06-30 01:55:33 -07002099 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 }
2101 return NOTIFY_OK;
2102}
2103#endif /* CONFIG_HOTPLUG_CPU */
2104
2105void __init page_alloc_init(void)
2106{
2107 hotcpu_notifier(page_alloc_cpu_notify, 0);
2108}
2109
2110/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002111 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
2112 * or min_free_kbytes changes.
2113 */
2114static void calculate_totalreserve_pages(void)
2115{
2116 struct pglist_data *pgdat;
2117 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002118 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002119
2120 for_each_online_pgdat(pgdat) {
2121 for (i = 0; i < MAX_NR_ZONES; i++) {
2122 struct zone *zone = pgdat->node_zones + i;
2123 unsigned long max = 0;
2124
2125 /* Find valid and maximum lowmem_reserve in the zone */
2126 for (j = i; j < MAX_NR_ZONES; j++) {
2127 if (zone->lowmem_reserve[j] > max)
2128 max = zone->lowmem_reserve[j];
2129 }
2130
2131 /* we treat pages_high as reserved pages. */
2132 max += zone->pages_high;
2133
2134 if (max > zone->present_pages)
2135 max = zone->present_pages;
2136 reserve_pages += max;
2137 }
2138 }
2139 totalreserve_pages = reserve_pages;
2140}
2141
2142/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143 * setup_per_zone_lowmem_reserve - called whenever
2144 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2145 * has a correct pages reserved value, so an adequate number of
2146 * pages are left in the zone after a successful __alloc_pages().
2147 */
2148static void setup_per_zone_lowmem_reserve(void)
2149{
2150 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002151 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08002153 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 for (j = 0; j < MAX_NR_ZONES; j++) {
2155 struct zone *zone = pgdat->node_zones + j;
2156 unsigned long present_pages = zone->present_pages;
2157
2158 zone->lowmem_reserve[j] = 0;
2159
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002160 idx = j;
2161 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 struct zone *lower_zone;
2163
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002164 idx--;
2165
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166 if (sysctl_lowmem_reserve_ratio[idx] < 1)
2167 sysctl_lowmem_reserve_ratio[idx] = 1;
2168
2169 lower_zone = pgdat->node_zones + idx;
2170 lower_zone->lowmem_reserve[j] = present_pages /
2171 sysctl_lowmem_reserve_ratio[idx];
2172 present_pages += lower_zone->present_pages;
2173 }
2174 }
2175 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002176
2177 /* update totalreserve_pages */
2178 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179}
2180
2181/*
2182 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2183 * that the pages_{min,low,high} values for each zone are set correctly
2184 * with respect to min_free_kbytes.
2185 */
Dave Hansen3947be12005-10-29 18:16:54 -07002186void setup_per_zone_pages_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187{
2188 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
2189 unsigned long lowmem_pages = 0;
2190 struct zone *zone;
2191 unsigned long flags;
2192
2193 /* Calculate total number of !ZONE_HIGHMEM pages */
2194 for_each_zone(zone) {
2195 if (!is_highmem(zone))
2196 lowmem_pages += zone->present_pages;
2197 }
2198
2199 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07002200 u64 tmp;
2201
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202 spin_lock_irqsave(&zone->lru_lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07002203 tmp = (u64)pages_min * zone->present_pages;
2204 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 if (is_highmem(zone)) {
2206 /*
Nick Piggin669ed172005-11-13 16:06:45 -08002207 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2208 * need highmem pages, so cap pages_min to a small
2209 * value here.
2210 *
2211 * The (pages_high-pages_low) and (pages_low-pages_min)
2212 * deltas controls asynch page reclaim, and so should
2213 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 */
2215 int min_pages;
2216
2217 min_pages = zone->present_pages / 1024;
2218 if (min_pages < SWAP_CLUSTER_MAX)
2219 min_pages = SWAP_CLUSTER_MAX;
2220 if (min_pages > 128)
2221 min_pages = 128;
2222 zone->pages_min = min_pages;
2223 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08002224 /*
2225 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 * proportionate to the zone's size.
2227 */
Nick Piggin669ed172005-11-13 16:06:45 -08002228 zone->pages_min = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 }
2230
Andrew Mortonac924c62006-05-15 09:43:59 -07002231 zone->pages_low = zone->pages_min + (tmp >> 2);
2232 zone->pages_high = zone->pages_min + (tmp >> 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 spin_unlock_irqrestore(&zone->lru_lock, flags);
2234 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002235
2236 /* update totalreserve_pages */
2237 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238}
2239
2240/*
2241 * Initialise min_free_kbytes.
2242 *
2243 * For small machines we want it small (128k min). For large machines
2244 * we want it large (64MB max). But it is not linear, because network
2245 * bandwidth does not increase linearly with machine size. We use
2246 *
2247 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2248 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2249 *
2250 * which yields
2251 *
2252 * 16MB: 512k
2253 * 32MB: 724k
2254 * 64MB: 1024k
2255 * 128MB: 1448k
2256 * 256MB: 2048k
2257 * 512MB: 2896k
2258 * 1024MB: 4096k
2259 * 2048MB: 5792k
2260 * 4096MB: 8192k
2261 * 8192MB: 11584k
2262 * 16384MB: 16384k
2263 */
2264static int __init init_per_zone_pages_min(void)
2265{
2266 unsigned long lowmem_kbytes;
2267
2268 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
2269
2270 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
2271 if (min_free_kbytes < 128)
2272 min_free_kbytes = 128;
2273 if (min_free_kbytes > 65536)
2274 min_free_kbytes = 65536;
2275 setup_per_zone_pages_min();
2276 setup_per_zone_lowmem_reserve();
2277 return 0;
2278}
2279module_init(init_per_zone_pages_min)
2280
2281/*
2282 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2283 * that we can call two helper functions whenever min_free_kbytes
2284 * changes.
2285 */
2286int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
2287 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2288{
2289 proc_dointvec(table, write, file, buffer, length, ppos);
2290 setup_per_zone_pages_min();
2291 return 0;
2292}
2293
Christoph Lameter96146342006-07-03 00:24:13 -07002294#ifdef CONFIG_NUMA
2295int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
2296 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2297{
2298 struct zone *zone;
2299 int rc;
2300
2301 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2302 if (rc)
2303 return rc;
2304
2305 for_each_zone(zone)
2306 zone->min_unmapped_ratio = (zone->present_pages *
2307 sysctl_min_unmapped_ratio) / 100;
2308 return 0;
2309}
2310#endif
2311
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312/*
2313 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2314 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2315 * whenever sysctl_lowmem_reserve_ratio changes.
2316 *
2317 * The reserve ratio obviously has absolutely no relation with the
2318 * pages_min watermarks. The lowmem reserve ratio can only make sense
2319 * if in function of the boot time zone sizes.
2320 */
2321int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
2322 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2323{
2324 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2325 setup_per_zone_lowmem_reserve();
2326 return 0;
2327}
2328
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002329/*
2330 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
2331 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
2332 * can have before it gets flushed back to buddy allocator.
2333 */
2334
2335int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
2336 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2337{
2338 struct zone *zone;
2339 unsigned int cpu;
2340 int ret;
2341
2342 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2343 if (!write || (ret == -EINVAL))
2344 return ret;
2345 for_each_zone(zone) {
2346 for_each_online_cpu(cpu) {
2347 unsigned long high;
2348 high = zone->present_pages / percpu_pagelist_fraction;
2349 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
2350 }
2351 }
2352 return 0;
2353}
2354
David S. Millerf034b5d2006-08-24 03:08:07 -07002355int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356
2357#ifdef CONFIG_NUMA
2358static int __init set_hashdist(char *str)
2359{
2360 if (!str)
2361 return 0;
2362 hashdist = simple_strtoul(str, &str, 0);
2363 return 1;
2364}
2365__setup("hashdist=", set_hashdist);
2366#endif
2367
2368/*
2369 * allocate a large system hash table from bootmem
2370 * - it is assumed that the hash table must contain an exact power-of-2
2371 * quantity of entries
2372 * - limit is the number of hash buckets, not the total allocation size
2373 */
2374void *__init alloc_large_system_hash(const char *tablename,
2375 unsigned long bucketsize,
2376 unsigned long numentries,
2377 int scale,
2378 int flags,
2379 unsigned int *_hash_shift,
2380 unsigned int *_hash_mask,
2381 unsigned long limit)
2382{
2383 unsigned long long max = limit;
2384 unsigned long log2qty, size;
2385 void *table = NULL;
2386
2387 /* allow the kernel cmdline to have a say */
2388 if (!numentries) {
2389 /* round applicable memory size up to nearest megabyte */
2390 numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages;
2391 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
2392 numentries >>= 20 - PAGE_SHIFT;
2393 numentries <<= 20 - PAGE_SHIFT;
2394
2395 /* limit to 1 bucket per 2^scale bytes of low memory */
2396 if (scale > PAGE_SHIFT)
2397 numentries >>= (scale - PAGE_SHIFT);
2398 else
2399 numentries <<= (PAGE_SHIFT - scale);
2400 }
John Hawkes6e692ed2006-03-25 03:08:02 -08002401 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402
2403 /* limit allocation size to 1/16 total memory by default */
2404 if (max == 0) {
2405 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
2406 do_div(max, bucketsize);
2407 }
2408
2409 if (numentries > max)
2410 numentries = max;
2411
2412 log2qty = long_log2(numentries);
2413
2414 do {
2415 size = bucketsize << log2qty;
2416 if (flags & HASH_EARLY)
2417 table = alloc_bootmem(size);
2418 else if (hashdist)
2419 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
2420 else {
2421 unsigned long order;
2422 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
2423 ;
2424 table = (void*) __get_free_pages(GFP_ATOMIC, order);
2425 }
2426 } while (!table && size > PAGE_SIZE && --log2qty);
2427
2428 if (!table)
2429 panic("Failed to allocate %s hash table\n", tablename);
2430
2431 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2432 tablename,
2433 (1U << log2qty),
2434 long_log2(size) - PAGE_SHIFT,
2435 size);
2436
2437 if (_hash_shift)
2438 *_hash_shift = log2qty;
2439 if (_hash_mask)
2440 *_hash_mask = (1 << log2qty) - 1;
2441
2442 return table;
2443}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08002444
2445#ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08002446struct page *pfn_to_page(unsigned long pfn)
2447{
Andy Whitcroft67de6482006-06-23 02:03:12 -07002448 return __pfn_to_page(pfn);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08002449}
2450unsigned long page_to_pfn(struct page *page)
2451{
Andy Whitcroft67de6482006-06-23 02:03:12 -07002452 return __page_to_pfn(page);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08002453}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08002454EXPORT_SYMBOL(pfn_to_page);
2455EXPORT_SYMBOL(page_to_pfn);
2456#endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */