blob: 51070b6d593f9b71d12a05d7246a5323b44ea3f3 [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{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700451 spin_lock(&zone->lock);
452 zone->all_unreclaimable = 0;
453 zone->pages_scanned = 0;
454 __free_one_page(page, zone ,order);
455 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800456}
457
458static void __free_pages_ok(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459{
Nick Pigginc54ad302006-01-06 00:10:56 -0800460 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 int i;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800462 int reserved = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463
464 arch_free_page(page, order);
Ingo Molnarde5097c2006-01-09 15:59:21 -0800465 if (!PageHighMem(page))
Ingo Molnarf9b84042006-06-27 02:54:49 -0700466 debug_check_no_locks_freed(page_address(page),
467 PAGE_SIZE<<order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 for (i = 0 ; i < (1 << order) ; ++i)
Nick Piggin224abf92006-01-06 00:11:11 -0800470 reserved += free_pages_check(page + i);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800471 if (reserved)
472 return;
473
Nick Piggin48db57f2006-01-08 01:00:42 -0800474 kernel_map_pages(page, 1 << order, 0);
Nick Pigginc54ad302006-01-06 00:10:56 -0800475 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700476 __count_vm_events(PGFREE, 1 << order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800477 free_one_page(page_zone(page), page, order);
Nick Pigginc54ad302006-01-06 00:10:56 -0800478 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479}
480
David Howellsa226f6c2006-01-06 00:11:08 -0800481/*
482 * permit the bootmem allocator to evade page validation on high-order frees
483 */
484void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
485{
486 if (order == 0) {
487 __ClearPageReserved(page);
488 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800489 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800490 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800491 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800492 int loop;
493
Nick Piggin545b1ea2006-03-22 00:08:07 -0800494 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800495 for (loop = 0; loop < BITS_PER_LONG; loop++) {
496 struct page *p = &page[loop];
497
Nick Piggin545b1ea2006-03-22 00:08:07 -0800498 if (loop + 1 < BITS_PER_LONG)
499 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800500 __ClearPageReserved(p);
501 set_page_count(p, 0);
502 }
503
Nick Piggin7835e982006-03-22 00:08:40 -0800504 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800505 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800506 }
507}
508
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509
510/*
511 * The order of subdivision here is critical for the IO subsystem.
512 * Please do not alter this order without good reasons and regression
513 * testing. Specifically, as large blocks of memory are subdivided,
514 * the order in which smaller blocks are delivered depends on the order
515 * they're subdivided in this function. This is the primary factor
516 * influencing the order in which pages are delivered to the IO
517 * subsystem according to empirical testing, and this is also justified
518 * by considering the behavior of a buddy system containing a single
519 * large block of memory acted on by a series of small allocations.
520 * This behavior is a critical factor in sglist merging's success.
521 *
522 * -- wli
523 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800524static inline void expand(struct zone *zone, struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 int low, int high, struct free_area *area)
526{
527 unsigned long size = 1 << high;
528
529 while (high > low) {
530 area--;
531 high--;
532 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700533 VM_BUG_ON(bad_range(zone, &page[size]));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 list_add(&page[size].lru, &area->free_list);
535 area->nr_free++;
536 set_page_order(&page[size], high);
537 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538}
539
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540/*
541 * This page is about to be returned from the page allocator
542 */
Nick Piggin17cf4402006-03-22 00:08:41 -0800543static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544{
Nick Piggin92be2e32006-01-06 00:10:57 -0800545 if (unlikely(page_mapcount(page) |
546 (page->mapping != NULL) |
547 (page_count(page) != 0) |
Hugh Dickins334795e2005-06-21 17:15:08 -0700548 (page->flags & (
549 1 << PG_lru |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 1 << PG_private |
551 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 1 << PG_active |
553 1 << PG_dirty |
554 1 << PG_reclaim |
Hugh Dickins334795e2005-06-21 17:15:08 -0700555 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700557 1 << PG_writeback |
Nick Piggin676165a2006-04-10 11:21:48 +1000558 1 << PG_reserved |
559 1 << PG_buddy ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800560 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561
Hugh Dickins689bceb2005-11-21 21:32:20 -0800562 /*
563 * For now, we report if PG_reserved was found set, but do not
564 * clear it, and do not allocate the page: as a safety net.
565 */
566 if (PageReserved(page))
567 return 1;
568
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
570 1 << PG_referenced | 1 << PG_arch_1 |
571 1 << PG_checked | 1 << PG_mappedtodisk);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700572 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800573 set_page_refcounted(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800575
576 if (gfp_flags & __GFP_ZERO)
577 prep_zero_page(page, order, gfp_flags);
578
579 if (order && (gfp_flags & __GFP_COMP))
580 prep_compound_page(page, order);
581
Hugh Dickins689bceb2005-11-21 21:32:20 -0800582 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583}
584
585/*
586 * Do the hard work of removing an element from the buddy allocator.
587 * Call me with the zone->lock already held.
588 */
589static struct page *__rmqueue(struct zone *zone, unsigned int order)
590{
591 struct free_area * area;
592 unsigned int current_order;
593 struct page *page;
594
595 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
596 area = zone->free_area + current_order;
597 if (list_empty(&area->free_list))
598 continue;
599
600 page = list_entry(area->free_list.next, struct page, lru);
601 list_del(&page->lru);
602 rmv_page_order(page);
603 area->nr_free--;
604 zone->free_pages -= 1UL << order;
Nick Piggin085cc7d2006-01-06 00:11:01 -0800605 expand(zone, page, order, current_order, area);
606 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 }
608
609 return NULL;
610}
611
612/*
613 * Obtain a specified number of elements from the buddy allocator, all under
614 * a single hold of the lock, for efficiency. Add them to the supplied list.
615 * Returns the number of new pages which were placed at *list.
616 */
617static int rmqueue_bulk(struct zone *zone, unsigned int order,
618 unsigned long count, struct list_head *list)
619{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621
Nick Pigginc54ad302006-01-06 00:10:56 -0800622 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 for (i = 0; i < count; ++i) {
Nick Piggin085cc7d2006-01-06 00:11:01 -0800624 struct page *page = __rmqueue(zone, order);
625 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 list_add_tail(&page->lru, list);
628 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800629 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800630 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631}
632
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700633#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800634/*
635 * Called from the slab reaper to drain pagesets on a particular node that
Christoph Lameter39bbcb82006-09-25 23:31:49 -0700636 * belongs to the currently executing processor.
Christoph Lameter879336c2006-03-22 00:09:08 -0800637 * Note that this function must be called with the thread pinned to
638 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800639 */
640void drain_node_pages(int nodeid)
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700641{
Christoph Lameter2f6726e2006-09-25 23:31:18 -0700642 int i;
643 enum zone_type z;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700644 unsigned long flags;
645
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800646 for (z = 0; z < MAX_NR_ZONES; z++) {
647 struct zone *zone = NODE_DATA(nodeid)->node_zones + z;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700648 struct per_cpu_pageset *pset;
649
Christoph Lameter39bbcb82006-09-25 23:31:49 -0700650 if (!populated_zone(zone))
651 continue;
652
Nick Piggin23316bc2006-01-08 01:00:41 -0800653 pset = zone_pcp(zone, smp_processor_id());
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700654 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
655 struct per_cpu_pages *pcp;
656
657 pcp = &pset->pcp[i];
Christoph Lameter879336c2006-03-22 00:09:08 -0800658 if (pcp->count) {
659 local_irq_save(flags);
660 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
661 pcp->count = 0;
662 local_irq_restore(flags);
663 }
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700664 }
665 }
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700666}
667#endif
668
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669#if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
670static void __drain_pages(unsigned int cpu)
671{
Nick Pigginc54ad302006-01-06 00:10:56 -0800672 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 struct zone *zone;
674 int i;
675
676 for_each_zone(zone) {
677 struct per_cpu_pageset *pset;
678
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700679 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
681 struct per_cpu_pages *pcp;
682
683 pcp = &pset->pcp[i];
Nick Pigginc54ad302006-01-06 00:10:56 -0800684 local_irq_save(flags);
Nick Piggin48db57f2006-01-08 01:00:42 -0800685 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
686 pcp->count = 0;
Nick Pigginc54ad302006-01-06 00:10:56 -0800687 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 }
689 }
690}
691#endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
692
693#ifdef CONFIG_PM
694
695void mark_free_pages(struct zone *zone)
696{
697 unsigned long zone_pfn, flags;
698 int order;
699 struct list_head *curr;
700
701 if (!zone->spanned_pages)
702 return;
703
704 spin_lock_irqsave(&zone->lock, flags);
705 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
706 ClearPageNosaveFree(pfn_to_page(zone_pfn + zone->zone_start_pfn));
707
708 for (order = MAX_ORDER - 1; order >= 0; --order)
709 list_for_each(curr, &zone->free_area[order].free_list) {
710 unsigned long start_pfn, i;
711
712 start_pfn = page_to_pfn(list_entry(curr, struct page, lru));
713
714 for (i=0; i < (1<<order); i++)
715 SetPageNosaveFree(pfn_to_page(start_pfn+i));
716 }
717 spin_unlock_irqrestore(&zone->lock, flags);
718}
719
720/*
721 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
722 */
723void drain_local_pages(void)
724{
725 unsigned long flags;
726
727 local_irq_save(flags);
728 __drain_pages(smp_processor_id());
729 local_irq_restore(flags);
730}
731#endif /* CONFIG_PM */
732
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733/*
734 * Free a 0-order page
735 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736static void fastcall free_hot_cold_page(struct page *page, int cold)
737{
738 struct zone *zone = page_zone(page);
739 struct per_cpu_pages *pcp;
740 unsigned long flags;
741
742 arch_free_page(page, 0);
743
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 if (PageAnon(page))
745 page->mapping = NULL;
Nick Piggin224abf92006-01-06 00:11:11 -0800746 if (free_pages_check(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -0800747 return;
748
Hugh Dickins689bceb2005-11-21 21:32:20 -0800749 kernel_map_pages(page, 1, 0);
750
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700751 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700753 __count_vm_event(PGFREE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 list_add(&page->lru, &pcp->list);
755 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -0800756 if (pcp->count >= pcp->high) {
757 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
758 pcp->count -= pcp->batch;
759 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 local_irq_restore(flags);
761 put_cpu();
762}
763
764void fastcall free_hot_page(struct page *page)
765{
766 free_hot_cold_page(page, 0);
767}
768
769void fastcall free_cold_page(struct page *page)
770{
771 free_hot_cold_page(page, 1);
772}
773
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800774/*
775 * split_page takes a non-compound higher-order page, and splits it into
776 * n (1<<order) sub-pages: page[0..n]
777 * Each sub-page must be freed individually.
778 *
779 * Note: this is probably too low level an operation for use in drivers.
780 * Please consult with lkml before using this in your driver.
781 */
782void split_page(struct page *page, unsigned int order)
783{
784 int i;
785
Nick Piggin725d7042006-09-25 23:30:55 -0700786 VM_BUG_ON(PageCompound(page));
787 VM_BUG_ON(!page_count(page));
Nick Piggin7835e982006-03-22 00:08:40 -0800788 for (i = 1; i < (1 << order); i++)
789 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800790}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800791
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792/*
793 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
794 * we cheat by calling it from here, in the order > 0 path. Saves a branch
795 * or two.
796 */
Nick Piggina74609f2006-01-06 00:11:20 -0800797static struct page *buffered_rmqueue(struct zonelist *zonelist,
798 struct zone *zone, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799{
800 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800801 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 int cold = !!(gfp_flags & __GFP_COLD);
Nick Piggina74609f2006-01-06 00:11:20 -0800803 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804
Hugh Dickins689bceb2005-11-21 21:32:20 -0800805again:
Nick Piggina74609f2006-01-06 00:11:20 -0800806 cpu = get_cpu();
Nick Piggin48db57f2006-01-08 01:00:42 -0800807 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 struct per_cpu_pages *pcp;
809
Nick Piggina74609f2006-01-06 00:11:20 -0800810 pcp = &zone_pcp(zone, cpu)->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -0800812 if (!pcp->count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 pcp->count += rmqueue_bulk(zone, 0,
814 pcp->batch, &pcp->list);
Nick Piggina74609f2006-01-06 00:11:20 -0800815 if (unlikely(!pcp->count))
816 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 }
Nick Piggina74609f2006-01-06 00:11:20 -0800818 page = list_entry(pcp->list.next, struct page, lru);
819 list_del(&page->lru);
820 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800821 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 spin_lock_irqsave(&zone->lock, flags);
823 page = __rmqueue(zone, order);
Nick Piggina74609f2006-01-06 00:11:20 -0800824 spin_unlock(&zone->lock);
825 if (!page)
826 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 }
828
Christoph Lameterf8891e52006-06-30 01:55:45 -0700829 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Christoph Lameterca889e62006-06-30 01:55:44 -0700830 zone_statistics(zonelist, zone);
Nick Piggina74609f2006-01-06 00:11:20 -0800831 local_irq_restore(flags);
832 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833
Nick Piggin725d7042006-09-25 23:30:55 -0700834 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -0800835 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -0800836 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 return page;
Nick Piggina74609f2006-01-06 00:11:20 -0800838
839failed:
840 local_irq_restore(flags);
841 put_cpu();
842 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843}
844
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800845#define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
Nick Piggin31488902005-11-28 13:44:03 -0800846#define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
847#define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
848#define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
849#define ALLOC_HARDER 0x10 /* try to alloc harder */
850#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
851#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800852
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853/*
854 * Return 1 if free pages are above 'mark'. This takes into account the order
855 * of the allocation.
856 */
857int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800858 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859{
860 /* free_pages my go negative - that's OK */
861 long min = mark, free_pages = z->free_pages - (1 << order) + 1;
862 int o;
863
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800864 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800866 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 min -= min / 4;
868
869 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
870 return 0;
871 for (o = 0; o < order; o++) {
872 /* At the next order, this order's pages become unavailable */
873 free_pages -= z->free_area[o].nr_free << o;
874
875 /* Require fewer higher order pages to be free */
876 min >>= 1;
877
878 if (free_pages <= min)
879 return 0;
880 }
881 return 1;
882}
883
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800884/*
885 * get_page_from_freeliest goes through the zonelist trying to allocate
886 * a page.
887 */
888static struct page *
889get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
890 struct zonelist *zonelist, int alloc_flags)
Martin Hicks753ee722005-06-21 17:14:41 -0700891{
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800892 struct zone **z = zonelist->zones;
893 struct page *page = NULL;
894 int classzone_idx = zone_idx(*z);
Christoph Lameter1192d522006-09-25 23:31:45 -0700895 struct zone *zone;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800896
897 /*
898 * Go through the zonelist once, looking for a zone with enough free.
899 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
900 */
901 do {
Christoph Lameter1192d522006-09-25 23:31:45 -0700902 zone = *z;
Christoph Lameter9b819d22006-09-25 23:31:40 -0700903 if (unlikely((gfp_mask & __GFP_THISNODE) &&
Christoph Lameter1192d522006-09-25 23:31:45 -0700904 zone->zone_pgdat != zonelist->zones[0]->zone_pgdat))
Christoph Lameter9b819d22006-09-25 23:31:40 -0700905 break;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800906 if ((alloc_flags & ALLOC_CPUSET) &&
Christoph Lameter1192d522006-09-25 23:31:45 -0700907 !cpuset_zone_allowed(zone, gfp_mask))
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800908 continue;
909
910 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -0800911 unsigned long mark;
912 if (alloc_flags & ALLOC_WMARK_MIN)
Christoph Lameter1192d522006-09-25 23:31:45 -0700913 mark = zone->pages_min;
Nick Piggin31488902005-11-28 13:44:03 -0800914 else if (alloc_flags & ALLOC_WMARK_LOW)
Christoph Lameter1192d522006-09-25 23:31:45 -0700915 mark = zone->pages_low;
Nick Piggin31488902005-11-28 13:44:03 -0800916 else
Christoph Lameter1192d522006-09-25 23:31:45 -0700917 mark = zone->pages_high;
918 if (!zone_watermark_ok(zone , order, mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800919 classzone_idx, alloc_flags))
Christoph Lameter9eeff232006-01-18 17:42:31 -0800920 if (!zone_reclaim_mode ||
Christoph Lameter1192d522006-09-25 23:31:45 -0700921 !zone_reclaim(zone, gfp_mask, order))
Christoph Lameter9eeff232006-01-18 17:42:31 -0800922 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800923 }
924
Christoph Lameter1192d522006-09-25 23:31:45 -0700925 page = buffered_rmqueue(zonelist, zone, order, gfp_mask);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800926 if (page) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800927 break;
928 }
929 } while (*(++z) != NULL);
930 return page;
Martin Hicks753ee722005-06-21 17:14:41 -0700931}
932
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933/*
934 * This is the 'heart' of the zoned buddy allocator.
935 */
936struct page * fastcall
Al Virodd0fc662005-10-07 07:46:04 +0100937__alloc_pages(gfp_t gfp_mask, unsigned int order,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 struct zonelist *zonelist)
939{
Al Viro260b2362005-10-21 03:22:44 -0400940 const gfp_t wait = gfp_mask & __GFP_WAIT;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800941 struct zone **z;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 struct page *page;
943 struct reclaim_state reclaim_state;
944 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 int do_retry;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800946 int alloc_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 int did_some_progress;
948
949 might_sleep_if(wait);
950
Jens Axboe6b1de912005-11-17 21:35:02 +0100951restart:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800952 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800954 if (unlikely(*z == NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 /* Should this ever happen?? */
956 return NULL;
957 }
Jens Axboe6b1de912005-11-17 21:35:02 +0100958
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800959 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -0800960 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800961 if (page)
962 goto got_pg;
963
Jens Axboe6b1de912005-11-17 21:35:02 +0100964 do {
Chris Wright43b0bc02006-06-25 05:47:55 -0700965 wakeup_kswapd(*z, order);
Jens Axboe6b1de912005-11-17 21:35:02 +0100966 } while (*(++z));
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800967
Paul Jackson9bf22292005-09-06 15:18:12 -0700968 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800969 * OK, we're below the kswapd watermark and have kicked background
970 * reclaim. Now things get more complex, so set up alloc_flags according
971 * to how we want to proceed.
972 *
973 * The caller may dip into page reserves a bit more if the caller
974 * cannot run direct reclaim, or if the caller has realtime scheduling
Paul Jackson4eac9152006-01-11 12:17:19 -0800975 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
976 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
Paul Jackson9bf22292005-09-06 15:18:12 -0700977 */
Nick Piggin31488902005-11-28 13:44:03 -0800978 alloc_flags = ALLOC_WMARK_MIN;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800979 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
980 alloc_flags |= ALLOC_HARDER;
981 if (gfp_mask & __GFP_HIGH)
982 alloc_flags |= ALLOC_HIGH;
Paul Jacksonbdd804f2006-05-20 15:00:09 -0700983 if (wait)
984 alloc_flags |= ALLOC_CPUSET;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985
986 /*
987 * Go through the zonelist again. Let __GFP_HIGH and allocations
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800988 * coming from realtime tasks go deeper into reserves.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 *
990 * This is the last chance, in general, before the goto nopage.
991 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
Paul Jackson9bf22292005-09-06 15:18:12 -0700992 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800994 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
995 if (page)
996 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
998 /* This allocation should allow future memory freeing. */
Nick Pigginb84a35b2005-05-01 08:58:36 -0700999
1000 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1001 && !in_interrupt()) {
1002 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
Kirill Korotaev885036d2005-11-13 16:06:41 -08001003nofail_alloc:
Nick Pigginb84a35b2005-05-01 08:58:36 -07001004 /* go through the zonelist yet again, ignoring mins */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001005 page = get_page_from_freelist(gfp_mask, order,
Paul Jackson47f3a862006-01-06 00:10:32 -08001006 zonelist, ALLOC_NO_WATERMARKS);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001007 if (page)
1008 goto got_pg;
Kirill Korotaev885036d2005-11-13 16:06:41 -08001009 if (gfp_mask & __GFP_NOFAIL) {
1010 blk_congestion_wait(WRITE, HZ/50);
1011 goto nofail_alloc;
1012 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 }
1014 goto nopage;
1015 }
1016
1017 /* Atomic allocations - we can't balance anything */
1018 if (!wait)
1019 goto nopage;
1020
1021rebalance:
1022 cond_resched();
1023
1024 /* We now go into synchronous reclaim */
Paul Jackson3e0d98b2006-01-08 01:01:49 -08001025 cpuset_memory_pressure_bump();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 p->flags |= PF_MEMALLOC;
1027 reclaim_state.reclaimed_slab = 0;
1028 p->reclaim_state = &reclaim_state;
1029
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001030 did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031
1032 p->reclaim_state = NULL;
1033 p->flags &= ~PF_MEMALLOC;
1034
1035 cond_resched();
1036
1037 if (likely(did_some_progress)) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001038 page = get_page_from_freelist(gfp_mask, order,
1039 zonelist, alloc_flags);
1040 if (page)
1041 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1043 /*
1044 * Go through the zonelist yet one more time, keep
1045 * very high watermark here, this is only to catch
1046 * a parallel oom killing, we must fail if we're still
1047 * under heavy pressure.
1048 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001049 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -08001050 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001051 if (page)
1052 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053
Christoph Lameter9b0f8b02006-02-20 18:27:52 -08001054 out_of_memory(zonelist, gfp_mask, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 goto restart;
1056 }
1057
1058 /*
1059 * Don't let big-order allocations loop unless the caller explicitly
1060 * requests that. Wait for some write requests to complete then retry.
1061 *
1062 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1063 * <= 3, but that may not be true in other implementations.
1064 */
1065 do_retry = 0;
1066 if (!(gfp_mask & __GFP_NORETRY)) {
1067 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
1068 do_retry = 1;
1069 if (gfp_mask & __GFP_NOFAIL)
1070 do_retry = 1;
1071 }
1072 if (do_retry) {
1073 blk_congestion_wait(WRITE, HZ/50);
1074 goto rebalance;
1075 }
1076
1077nopage:
1078 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1079 printk(KERN_WARNING "%s: page allocation failure."
1080 " order:%d, mode:0x%x\n",
1081 p->comm, order, gfp_mask);
1082 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001083 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085got_pg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 return page;
1087}
1088
1089EXPORT_SYMBOL(__alloc_pages);
1090
1091/*
1092 * Common helper functions.
1093 */
Al Virodd0fc662005-10-07 07:46:04 +01001094fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095{
1096 struct page * page;
1097 page = alloc_pages(gfp_mask, order);
1098 if (!page)
1099 return 0;
1100 return (unsigned long) page_address(page);
1101}
1102
1103EXPORT_SYMBOL(__get_free_pages);
1104
Al Virodd0fc662005-10-07 07:46:04 +01001105fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106{
1107 struct page * page;
1108
1109 /*
1110 * get_zeroed_page() returns a 32-bit address, which cannot represent
1111 * a highmem page
1112 */
Nick Piggin725d7042006-09-25 23:30:55 -07001113 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114
1115 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1116 if (page)
1117 return (unsigned long) page_address(page);
1118 return 0;
1119}
1120
1121EXPORT_SYMBOL(get_zeroed_page);
1122
1123void __pagevec_free(struct pagevec *pvec)
1124{
1125 int i = pagevec_count(pvec);
1126
1127 while (--i >= 0)
1128 free_hot_cold_page(pvec->pages[i], pvec->cold);
1129}
1130
1131fastcall void __free_pages(struct page *page, unsigned int order)
1132{
Nick Pigginb5810032005-10-29 18:16:12 -07001133 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 if (order == 0)
1135 free_hot_page(page);
1136 else
1137 __free_pages_ok(page, order);
1138 }
1139}
1140
1141EXPORT_SYMBOL(__free_pages);
1142
1143fastcall void free_pages(unsigned long addr, unsigned int order)
1144{
1145 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07001146 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 __free_pages(virt_to_page((void *)addr), order);
1148 }
1149}
1150
1151EXPORT_SYMBOL(free_pages);
1152
1153/*
1154 * Total amount of free (allocatable) RAM:
1155 */
1156unsigned int nr_free_pages(void)
1157{
1158 unsigned int sum = 0;
1159 struct zone *zone;
1160
1161 for_each_zone(zone)
1162 sum += zone->free_pages;
1163
1164 return sum;
1165}
1166
1167EXPORT_SYMBOL(nr_free_pages);
1168
1169#ifdef CONFIG_NUMA
1170unsigned int nr_free_pages_pgdat(pg_data_t *pgdat)
1171{
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001172 unsigned int sum = 0;
1173 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174
1175 for (i = 0; i < MAX_NR_ZONES; i++)
1176 sum += pgdat->node_zones[i].free_pages;
1177
1178 return sum;
1179}
1180#endif
1181
1182static unsigned int nr_free_zone_pages(int offset)
1183{
Martin J. Blighe310fd42005-07-29 22:59:18 -07001184 /* Just pick one node, since fallback list is circular */
1185 pg_data_t *pgdat = NODE_DATA(numa_node_id());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 unsigned int sum = 0;
1187
Martin J. Blighe310fd42005-07-29 22:59:18 -07001188 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1189 struct zone **zonep = zonelist->zones;
1190 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191
Martin J. Blighe310fd42005-07-29 22:59:18 -07001192 for (zone = *zonep++; zone; zone = *zonep++) {
1193 unsigned long size = zone->present_pages;
1194 unsigned long high = zone->pages_high;
1195 if (size > high)
1196 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 }
1198
1199 return sum;
1200}
1201
1202/*
1203 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1204 */
1205unsigned int nr_free_buffer_pages(void)
1206{
Al Viroaf4ca452005-10-21 02:55:38 -04001207 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208}
1209
1210/*
1211 * Amount of free RAM allocatable within all zones
1212 */
1213unsigned int nr_free_pagecache_pages(void)
1214{
Al Viroaf4ca452005-10-21 02:55:38 -04001215 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217#ifdef CONFIG_NUMA
1218static void show_node(struct zone *zone)
1219{
Christoph Lameter89fa3022006-09-25 23:31:55 -07001220 printk("Node %ld ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221}
1222#else
1223#define show_node(zone) do { } while (0)
1224#endif
1225
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226void si_meminfo(struct sysinfo *val)
1227{
1228 val->totalram = totalram_pages;
1229 val->sharedram = 0;
1230 val->freeram = nr_free_pages();
1231 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 val->totalhigh = totalhigh_pages;
1233 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 val->mem_unit = PAGE_SIZE;
1235}
1236
1237EXPORT_SYMBOL(si_meminfo);
1238
1239#ifdef CONFIG_NUMA
1240void si_meminfo_node(struct sysinfo *val, int nid)
1241{
1242 pg_data_t *pgdat = NODE_DATA(nid);
1243
1244 val->totalram = pgdat->node_present_pages;
1245 val->freeram = nr_free_pages_pgdat(pgdat);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001246#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1248 val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages;
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001249#else
1250 val->totalhigh = 0;
1251 val->freehigh = 0;
1252#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 val->mem_unit = PAGE_SIZE;
1254}
1255#endif
1256
1257#define K(x) ((x) << (PAGE_SHIFT-10))
1258
1259/*
1260 * Show free area list (used inside shift_scroll-lock stuff)
1261 * We also calculate the percentage fragmentation. We do this by counting the
1262 * memory on each free list with the exception of the first item on the list.
1263 */
1264void show_free_areas(void)
1265{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 int cpu, temperature;
1267 unsigned long active;
1268 unsigned long inactive;
1269 unsigned long free;
1270 struct zone *zone;
1271
1272 for_each_zone(zone) {
1273 show_node(zone);
1274 printk("%s per-cpu:", zone->name);
1275
Con Kolivasf3fe6512006-01-06 00:11:15 -08001276 if (!populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 printk(" empty\n");
1278 continue;
1279 } else
1280 printk("\n");
1281
Dave Jones6b482c62005-11-10 15:45:56 -05001282 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 struct per_cpu_pageset *pageset;
1284
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001285 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286
1287 for (temperature = 0; temperature < 2; temperature++)
Nick Piggin2d92c5c2006-01-06 00:10:59 -08001288 printk("cpu %d %s: high %d, batch %d used:%d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 cpu,
1290 temperature ? "cold" : "hot",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 pageset->pcp[temperature].high,
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001292 pageset->pcp[temperature].batch,
1293 pageset->pcp[temperature].count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 }
1295 }
1296
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 get_zone_counts(&active, &inactive, &free);
1298
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1300 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1301 active,
1302 inactive,
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07001303 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07001304 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07001305 global_page_state(NR_UNSTABLE_NFS),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 nr_free_pages(),
Christoph Lameter972d1a72006-09-25 23:31:51 -07001307 global_page_state(NR_SLAB_RECLAIMABLE) +
1308 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07001309 global_page_state(NR_FILE_MAPPED),
Christoph Lameterdf849a12006-06-30 01:55:38 -07001310 global_page_state(NR_PAGETABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311
1312 for_each_zone(zone) {
1313 int i;
1314
1315 show_node(zone);
1316 printk("%s"
1317 " free:%lukB"
1318 " min:%lukB"
1319 " low:%lukB"
1320 " high:%lukB"
1321 " active:%lukB"
1322 " inactive:%lukB"
1323 " present:%lukB"
1324 " pages_scanned:%lu"
1325 " all_unreclaimable? %s"
1326 "\n",
1327 zone->name,
1328 K(zone->free_pages),
1329 K(zone->pages_min),
1330 K(zone->pages_low),
1331 K(zone->pages_high),
1332 K(zone->nr_active),
1333 K(zone->nr_inactive),
1334 K(zone->present_pages),
1335 zone->pages_scanned,
1336 (zone->all_unreclaimable ? "yes" : "no")
1337 );
1338 printk("lowmem_reserve[]:");
1339 for (i = 0; i < MAX_NR_ZONES; i++)
1340 printk(" %lu", zone->lowmem_reserve[i]);
1341 printk("\n");
1342 }
1343
1344 for_each_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001345 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346
1347 show_node(zone);
1348 printk("%s: ", zone->name);
Con Kolivasf3fe6512006-01-06 00:11:15 -08001349 if (!populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 printk("empty\n");
1351 continue;
1352 }
1353
1354 spin_lock_irqsave(&zone->lock, flags);
1355 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001356 nr[order] = zone->free_area[order].nr_free;
1357 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 }
1359 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001360 for (order = 0; order < MAX_ORDER; order++)
1361 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 printk("= %lukB\n", K(total));
1363 }
1364
1365 show_swap_cache_info();
1366}
1367
1368/*
1369 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08001370 *
1371 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07001373static int __meminit build_zonelists_node(pg_data_t *pgdat,
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001374 struct zonelist *zonelist, int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375{
Christoph Lameter1a932052006-01-06 00:11:16 -08001376 struct zone *zone;
1377
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001378 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001379 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08001380
1381 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001382 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08001383 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08001384 if (populated_zone(zone)) {
Christoph Lameter070f8032006-01-06 00:11:19 -08001385 zonelist->zones[nr_zones++] = zone;
1386 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08001388
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001389 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08001390 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391}
1392
1393#ifdef CONFIG_NUMA
1394#define MAX_NODE_LOAD (num_online_nodes())
Yasunori Goto86356ab2006-06-23 02:03:09 -07001395static int __meminitdata node_load[MAX_NUMNODES];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001397 * 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 -07001398 * @node: node whose fallback list we're appending
1399 * @used_node_mask: nodemask_t of already used nodes
1400 *
1401 * We use a number of factors to determine which is the next node that should
1402 * appear on a given node's fallback list. The node should not have appeared
1403 * already in @node's fallback list, and it should be the next closest node
1404 * according to the distance array (which contains arbitrary distance values
1405 * from each node to each node in the system), and should also prefer nodes
1406 * with no CPUs, since presumably they'll have very little allocation pressure
1407 * on them otherwise.
1408 * It returns -1 if no node is found.
1409 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07001410static int __meminit find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001412 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 int min_val = INT_MAX;
1414 int best_node = -1;
1415
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001416 /* Use the local node if we haven't already */
1417 if (!node_isset(node, *used_node_mask)) {
1418 node_set(node, *used_node_mask);
1419 return node;
1420 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001422 for_each_online_node(n) {
1423 cpumask_t tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424
1425 /* Don't want a node to appear more than once */
1426 if (node_isset(n, *used_node_mask))
1427 continue;
1428
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 /* Use the distance array to find the distance */
1430 val = node_distance(node, n);
1431
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001432 /* Penalize nodes under us ("prefer the next node") */
1433 val += (n < node);
1434
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 /* Give preference to headless and unused nodes */
1436 tmp = node_to_cpumask(n);
1437 if (!cpus_empty(tmp))
1438 val += PENALTY_FOR_NODE_WITH_CPUS;
1439
1440 /* Slight preference for less loaded node */
1441 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1442 val += node_load[n];
1443
1444 if (val < min_val) {
1445 min_val = val;
1446 best_node = n;
1447 }
1448 }
1449
1450 if (best_node >= 0)
1451 node_set(best_node, *used_node_mask);
1452
1453 return best_node;
1454}
1455
Yasunori Goto86356ab2006-06-23 02:03:09 -07001456static void __meminit build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457{
Christoph Lameter19655d32006-09-25 23:31:19 -07001458 int j, node, local_node;
1459 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 int prev_node, load;
1461 struct zonelist *zonelist;
1462 nodemask_t used_mask;
1463
1464 /* initialize zonelists */
Christoph Lameter19655d32006-09-25 23:31:19 -07001465 for (i = 0; i < MAX_NR_ZONES; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 zonelist = pgdat->node_zonelists + i;
1467 zonelist->zones[0] = NULL;
1468 }
1469
1470 /* NUMA-aware ordering of nodes */
1471 local_node = pgdat->node_id;
1472 load = num_online_nodes();
1473 prev_node = local_node;
1474 nodes_clear(used_mask);
1475 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08001476 int distance = node_distance(local_node, node);
1477
1478 /*
1479 * If another node is sufficiently far away then it is better
1480 * to reclaim pages in a zone before going off node.
1481 */
1482 if (distance > RECLAIM_DISTANCE)
1483 zone_reclaim_mode = 1;
1484
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 /*
1486 * We don't want to pressure a particular node.
1487 * So adding penalty to the first node in same
1488 * distance group to make it round-robin.
1489 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08001490
1491 if (distance != node_distance(local_node, prev_node))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 node_load[node] += load;
1493 prev_node = node;
1494 load--;
Christoph Lameter19655d32006-09-25 23:31:19 -07001495 for (i = 0; i < MAX_NR_ZONES; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 zonelist = pgdat->node_zonelists + i;
1497 for (j = 0; zonelist->zones[j] != NULL; j++);
1498
Christoph Lameter19655d32006-09-25 23:31:19 -07001499 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 zonelist->zones[j] = NULL;
1501 }
1502 }
1503}
1504
1505#else /* CONFIG_NUMA */
1506
Yasunori Goto86356ab2006-06-23 02:03:09 -07001507static void __meminit build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508{
Christoph Lameter19655d32006-09-25 23:31:19 -07001509 int node, local_node;
1510 enum zone_type i,j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511
1512 local_node = pgdat->node_id;
Christoph Lameter19655d32006-09-25 23:31:19 -07001513 for (i = 0; i < MAX_NR_ZONES; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 struct zonelist *zonelist;
1515
1516 zonelist = pgdat->node_zonelists + i;
1517
Christoph Lameter19655d32006-09-25 23:31:19 -07001518 j = build_zonelists_node(pgdat, zonelist, 0, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 /*
1520 * Now we build the zonelist so that it contains the zones
1521 * of all the other nodes.
1522 * We don't want to pressure a particular node, so when
1523 * building the zones for node N, we make sure that the
1524 * zones coming right after the local ones are those from
1525 * node N+1 (modulo N)
1526 */
1527 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1528 if (!node_online(node))
1529 continue;
Christoph Lameter19655d32006-09-25 23:31:19 -07001530 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 }
1532 for (node = 0; node < local_node; node++) {
1533 if (!node_online(node))
1534 continue;
Christoph Lameter19655d32006-09-25 23:31:19 -07001535 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 }
1537
1538 zonelist->zones[j] = NULL;
1539 }
1540}
1541
1542#endif /* CONFIG_NUMA */
1543
Yasunori Goto68113782006-06-23 02:03:11 -07001544/* return values int ....just for stop_machine_run() */
1545static int __meminit __build_all_zonelists(void *dummy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546{
Yasunori Goto68113782006-06-23 02:03:11 -07001547 int nid;
1548 for_each_online_node(nid)
1549 build_zonelists(NODE_DATA(nid));
1550 return 0;
1551}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552
Yasunori Goto68113782006-06-23 02:03:11 -07001553void __meminit build_all_zonelists(void)
1554{
1555 if (system_state == SYSTEM_BOOTING) {
1556 __build_all_zonelists(0);
1557 cpuset_init_current_mems_allowed();
1558 } else {
1559 /* we have to stop all cpus to guaranntee there is no user
1560 of zonelist */
1561 stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
1562 /* cpuset refresh routine should be here */
1563 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07001564 vm_total_pages = nr_free_pagecache_pages();
1565 printk("Built %i zonelists. Total pages: %ld\n",
1566 num_online_nodes(), vm_total_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567}
1568
1569/*
1570 * Helper functions to size the waitqueue hash table.
1571 * Essentially these want to choose hash table sizes sufficiently
1572 * large so that collisions trying to wait on pages are rare.
1573 * But in fact, the number of active page waitqueues on typical
1574 * systems is ridiculously low, less than 200. So this is even
1575 * conservative, even though it seems large.
1576 *
1577 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1578 * waitqueues, i.e. the size of the waitq table given the number of pages.
1579 */
1580#define PAGES_PER_WAITQUEUE 256
1581
Yasunori Gotocca448f2006-06-23 02:03:10 -07001582#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07001583static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584{
1585 unsigned long size = 1;
1586
1587 pages /= PAGES_PER_WAITQUEUE;
1588
1589 while (size < pages)
1590 size <<= 1;
1591
1592 /*
1593 * Once we have dozens or even hundreds of threads sleeping
1594 * on IO we've got bigger problems than wait queue collision.
1595 * Limit the size of the wait table to a reasonable size.
1596 */
1597 size = min(size, 4096UL);
1598
1599 return max(size, 4UL);
1600}
Yasunori Gotocca448f2006-06-23 02:03:10 -07001601#else
1602/*
1603 * A zone's size might be changed by hot-add, so it is not possible to determine
1604 * a suitable size for its wait_table. So we use the maximum size now.
1605 *
1606 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
1607 *
1608 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
1609 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
1610 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
1611 *
1612 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
1613 * or more by the traditional way. (See above). It equals:
1614 *
1615 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
1616 * ia64(16K page size) : = ( 8G + 4M)byte.
1617 * powerpc (64K page size) : = (32G +16M)byte.
1618 */
1619static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
1620{
1621 return 4096UL;
1622}
1623#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624
1625/*
1626 * This is an integer logarithm so that shifts can be used later
1627 * to extract the more random high bits from the multiplicative
1628 * hash function before the remainder is taken.
1629 */
1630static inline unsigned long wait_table_bits(unsigned long size)
1631{
1632 return ffz(~size);
1633}
1634
1635#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1636
1637static void __init calculate_zone_totalpages(struct pglist_data *pgdat,
1638 unsigned long *zones_size, unsigned long *zholes_size)
1639{
1640 unsigned long realtotalpages, totalpages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001641 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642
1643 for (i = 0; i < MAX_NR_ZONES; i++)
1644 totalpages += zones_size[i];
1645 pgdat->node_spanned_pages = totalpages;
1646
1647 realtotalpages = totalpages;
1648 if (zholes_size)
1649 for (i = 0; i < MAX_NR_ZONES; i++)
1650 realtotalpages -= zholes_size[i];
1651 pgdat->node_present_pages = realtotalpages;
1652 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
1653}
1654
1655
1656/*
1657 * Initially all pages are reserved - free ones are freed
1658 * up by free_all_bootmem() once the early boot process is
1659 * done. Non-atomic initialization, single-pass.
1660 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001661void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 unsigned long start_pfn)
1663{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07001665 unsigned long end_pfn = start_pfn + size;
1666 unsigned long pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667
Greg Ungerercbe8dd42006-01-12 01:05:24 -08001668 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001669 if (!early_pfn_valid(pfn))
1670 continue;
1671 page = pfn_to_page(pfn);
1672 set_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08001673 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 reset_page_mapcount(page);
1675 SetPageReserved(page);
1676 INIT_LIST_HEAD(&page->lru);
1677#ifdef WANT_PAGE_VIRTUAL
1678 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1679 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07001680 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 }
1683}
1684
1685void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1686 unsigned long size)
1687{
1688 int order;
1689 for (order = 0; order < MAX_ORDER ; order++) {
1690 INIT_LIST_HEAD(&zone->free_area[order].free_list);
1691 zone->free_area[order].nr_free = 0;
1692 }
1693}
1694
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001695#define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
Christoph Lameter2f1b6242006-09-25 23:31:13 -07001696void zonetable_add(struct zone *zone, int nid, enum zone_type zid,
1697 unsigned long pfn, unsigned long size)
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001698{
1699 unsigned long snum = pfn_to_section_nr(pfn);
1700 unsigned long end = pfn_to_section_nr(pfn + size);
1701
1702 if (FLAGS_HAS_NODE)
1703 zone_table[ZONETABLE_INDEX(nid, zid)] = zone;
1704 else
1705 for (; snum <= end; snum++)
1706 zone_table[ZONETABLE_INDEX(snum, zid)] = zone;
1707}
1708
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709#ifndef __HAVE_ARCH_MEMMAP_INIT
1710#define memmap_init(size, nid, zone, start_pfn) \
1711 memmap_init_zone((size), (nid), (zone), (start_pfn))
1712#endif
1713
Ashok Raj6292d9a2006-02-01 03:04:44 -08001714static int __cpuinit zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001715{
1716 int batch;
1717
1718 /*
1719 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07001720 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001721 *
1722 * OK, so we don't know how big the cache is. So guess.
1723 */
1724 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07001725 if (batch * PAGE_SIZE > 512 * 1024)
1726 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001727 batch /= 4; /* We effectively *= 4 below */
1728 if (batch < 1)
1729 batch = 1;
1730
1731 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001732 * Clamp the batch to a 2^n - 1 value. Having a power
1733 * of 2 value was found to be more likely to have
1734 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001735 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001736 * For example if 2 tasks are alternately allocating
1737 * batches of pages, one task can end up with a lot
1738 * of pages of one half of the possible page colors
1739 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001740 */
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001741 batch = (1 << (fls(batch + batch/2)-1)) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07001742
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001743 return batch;
1744}
1745
Christoph Lameter2caaad42005-06-21 17:15:00 -07001746inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
1747{
1748 struct per_cpu_pages *pcp;
1749
Magnus Damm1c6fe942005-10-26 01:58:59 -07001750 memset(p, 0, sizeof(*p));
1751
Christoph Lameter2caaad42005-06-21 17:15:00 -07001752 pcp = &p->pcp[0]; /* hot */
1753 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001754 pcp->high = 6 * batch;
1755 pcp->batch = max(1UL, 1 * batch);
1756 INIT_LIST_HEAD(&pcp->list);
1757
1758 pcp = &p->pcp[1]; /* cold*/
1759 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001760 pcp->high = 2 * batch;
Seth, Rohite46a5e22005-10-29 18:15:48 -07001761 pcp->batch = max(1UL, batch/2);
Christoph Lameter2caaad42005-06-21 17:15:00 -07001762 INIT_LIST_HEAD(&pcp->list);
1763}
1764
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08001765/*
1766 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
1767 * to the value high for the pageset p.
1768 */
1769
1770static void setup_pagelist_highmark(struct per_cpu_pageset *p,
1771 unsigned long high)
1772{
1773 struct per_cpu_pages *pcp;
1774
1775 pcp = &p->pcp[0]; /* hot list */
1776 pcp->high = high;
1777 pcp->batch = max(1UL, high/4);
1778 if ((high/4) > (PAGE_SHIFT * 8))
1779 pcp->batch = PAGE_SHIFT * 8;
1780}
1781
1782
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001783#ifdef CONFIG_NUMA
1784/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07001785 * Boot pageset table. One per cpu which is going to be used for all
1786 * zones and all nodes. The parameters will be set in such a way
1787 * that an item put on a list will immediately be handed over to
1788 * the buddy list. This is safe since pageset manipulation is done
1789 * with interrupts disabled.
1790 *
1791 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07001792 *
1793 * The boot_pagesets must be kept even after bootup is complete for
1794 * unused processors and/or zones. They do play a role for bootstrapping
1795 * hotplugged processors.
1796 *
1797 * zoneinfo_show() and maybe other functions do
1798 * not check if the processor is online before following the pageset pointer.
1799 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07001800 */
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08001801static struct per_cpu_pageset boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07001802
1803/*
1804 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001805 * per cpu pageset array in struct zone.
1806 */
Ashok Raj6292d9a2006-02-01 03:04:44 -08001807static int __cpuinit process_zones(int cpu)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001808{
1809 struct zone *zone, *dzone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001810
1811 for_each_zone(zone) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001812
Nick Piggin23316bc2006-01-08 01:00:41 -08001813 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001814 GFP_KERNEL, cpu_to_node(cpu));
Nick Piggin23316bc2006-01-08 01:00:41 -08001815 if (!zone_pcp(zone, cpu))
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001816 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001817
Nick Piggin23316bc2006-01-08 01:00:41 -08001818 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08001819
1820 if (percpu_pagelist_fraction)
1821 setup_pagelist_highmark(zone_pcp(zone, cpu),
1822 (zone->present_pages / percpu_pagelist_fraction));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001823 }
1824
1825 return 0;
1826bad:
1827 for_each_zone(dzone) {
1828 if (dzone == zone)
1829 break;
Nick Piggin23316bc2006-01-08 01:00:41 -08001830 kfree(zone_pcp(dzone, cpu));
1831 zone_pcp(dzone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001832 }
1833 return -ENOMEM;
1834}
1835
1836static inline void free_zone_pagesets(int cpu)
1837{
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001838 struct zone *zone;
1839
1840 for_each_zone(zone) {
1841 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
1842
David Rientjesf3ef9ea2006-09-25 16:24:57 -07001843 /* Free per_cpu_pageset if it is slab allocated */
1844 if (pset != &boot_pageset[cpu])
1845 kfree(pset);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001846 zone_pcp(zone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001847 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001848}
1849
Chandra Seetharaman9c7b2162006-06-27 02:54:07 -07001850static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001851 unsigned long action,
1852 void *hcpu)
1853{
1854 int cpu = (long)hcpu;
1855 int ret = NOTIFY_OK;
1856
1857 switch (action) {
1858 case CPU_UP_PREPARE:
1859 if (process_zones(cpu))
1860 ret = NOTIFY_BAD;
1861 break;
Andi Kleenb0d41692005-11-05 17:25:53 +01001862 case CPU_UP_CANCELED:
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001863 case CPU_DEAD:
1864 free_zone_pagesets(cpu);
1865 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001866 default:
1867 break;
1868 }
1869 return ret;
1870}
1871
Chandra Seetharaman74b85f32006-06-27 02:54:09 -07001872static struct notifier_block __cpuinitdata pageset_notifier =
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001873 { &pageset_cpuup_callback, NULL, 0 };
1874
Al Viro78d99552005-12-15 09:18:25 +00001875void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001876{
1877 int err;
1878
1879 /* Initialize per_cpu_pageset for cpu 0.
1880 * A cpuup callback will do this for every cpu
1881 * as it comes online
1882 */
1883 err = process_zones(smp_processor_id());
1884 BUG_ON(err);
1885 register_cpu_notifier(&pageset_notifier);
1886}
1887
1888#endif
1889
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001890static __meminit
Yasunori Gotocca448f2006-06-23 02:03:10 -07001891int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07001892{
1893 int i;
1894 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07001895 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07001896
1897 /*
1898 * The per-page waitqueue mechanism uses hashed waitqueues
1899 * per zone.
1900 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07001901 zone->wait_table_hash_nr_entries =
1902 wait_table_hash_nr_entries(zone_size_pages);
1903 zone->wait_table_bits =
1904 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07001905 alloc_size = zone->wait_table_hash_nr_entries
1906 * sizeof(wait_queue_head_t);
1907
1908 if (system_state == SYSTEM_BOOTING) {
1909 zone->wait_table = (wait_queue_head_t *)
1910 alloc_bootmem_node(pgdat, alloc_size);
1911 } else {
1912 /*
1913 * This case means that a zone whose size was 0 gets new memory
1914 * via memory hot-add.
1915 * But it may be the case that a new node was hot-added. In
1916 * this case vmalloc() will not be able to use this new node's
1917 * memory - this wait_table must be initialized to use this new
1918 * node itself as well.
1919 * To use this new node's memory, further consideration will be
1920 * necessary.
1921 */
1922 zone->wait_table = (wait_queue_head_t *)vmalloc(alloc_size);
1923 }
1924 if (!zone->wait_table)
1925 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07001926
Yasunori Goto02b694d2006-06-23 02:03:08 -07001927 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07001928 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07001929
1930 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07001931}
1932
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001933static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07001934{
1935 int cpu;
1936 unsigned long batch = zone_batchsize(zone);
1937
1938 for (cpu = 0; cpu < NR_CPUS; cpu++) {
1939#ifdef CONFIG_NUMA
1940 /* Early boot. Slab allocator not functional yet */
Nick Piggin23316bc2006-01-08 01:00:41 -08001941 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Dave Hansened8ece22005-10-29 18:16:50 -07001942 setup_pageset(&boot_pageset[cpu],0);
1943#else
1944 setup_pageset(zone_pcp(zone,cpu), batch);
1945#endif
1946 }
Anton Blanchardf5335c02006-03-25 03:06:49 -08001947 if (zone->present_pages)
1948 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
1949 zone->name, zone->present_pages, batch);
Dave Hansened8ece22005-10-29 18:16:50 -07001950}
1951
Yasunori Goto718127c2006-06-23 02:03:10 -07001952__meminit int init_currently_empty_zone(struct zone *zone,
1953 unsigned long zone_start_pfn,
1954 unsigned long size)
Dave Hansened8ece22005-10-29 18:16:50 -07001955{
1956 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07001957 int ret;
1958 ret = zone_wait_table_init(zone, size);
1959 if (ret)
1960 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07001961 pgdat->nr_zones = zone_idx(zone) + 1;
1962
Dave Hansened8ece22005-10-29 18:16:50 -07001963 zone->zone_start_pfn = zone_start_pfn;
1964
1965 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
1966
1967 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
Yasunori Goto718127c2006-06-23 02:03:10 -07001968
1969 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07001970}
1971
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972/*
1973 * Set up the zone data structures:
1974 * - mark all pages reserved
1975 * - mark all memory queues empty
1976 * - clear the memory bitmaps
1977 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07001978static void __meminit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 unsigned long *zones_size, unsigned long *zholes_size)
1980{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07001981 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07001982 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07001984 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985
Dave Hansen208d54e2005-10-29 18:16:52 -07001986 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 pgdat->nr_zones = 0;
1988 init_waitqueue_head(&pgdat->kswapd_wait);
1989 pgdat->kswapd_max_order = 0;
1990
1991 for (j = 0; j < MAX_NR_ZONES; j++) {
1992 struct zone *zone = pgdat->node_zones + j;
1993 unsigned long size, realsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 realsize = size = zones_size[j];
1996 if (zholes_size)
1997 realsize -= zholes_size[j];
1998
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001999 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 nr_kernel_pages += realsize;
2001 nr_all_pages += realsize;
2002
2003 zone->spanned_pages = size;
2004 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07002005#ifdef CONFIG_NUMA
Christoph Lameter8417bba2006-09-25 23:31:51 -07002006 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07002007 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07002008 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07002009#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010 zone->name = zone_names[j];
2011 spin_lock_init(&zone->lock);
2012 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07002013 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 zone->zone_pgdat = pgdat;
2015 zone->free_pages = 0;
2016
2017 zone->temp_priority = zone->prev_priority = DEF_PRIORITY;
2018
Dave Hansened8ece22005-10-29 18:16:50 -07002019 zone_pcp_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 INIT_LIST_HEAD(&zone->active_list);
2021 INIT_LIST_HEAD(&zone->inactive_list);
2022 zone->nr_scan_active = 0;
2023 zone->nr_scan_inactive = 0;
2024 zone->nr_active = 0;
2025 zone->nr_inactive = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07002026 zap_zone_vm_stats(zone);
Martin Hicks53e9a612005-09-03 15:54:51 -07002027 atomic_set(&zone->reclaim_in_progress, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028 if (!size)
2029 continue;
2030
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002031 zonetable_add(zone, nid, j, zone_start_pfn, size);
Yasunori Goto718127c2006-06-23 02:03:10 -07002032 ret = init_currently_empty_zone(zone, zone_start_pfn, size);
2033 BUG_ON(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 }
2036}
2037
2038static void __init alloc_node_mem_map(struct pglist_data *pgdat)
2039{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 /* Skip empty nodes */
2041 if (!pgdat->node_spanned_pages)
2042 return;
2043
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002044#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 /* ia64 gets its own node_mem_map, before this, without bootmem */
2046 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07002047 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002048 struct page *map;
2049
Bob Piccoe984bb42006-05-20 15:00:31 -07002050 /*
2051 * The zone's endpoints aren't required to be MAX_ORDER
2052 * aligned but the node_mem_map endpoints must be in order
2053 * for the buddy allocator to function correctly.
2054 */
2055 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
2056 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
2057 end = ALIGN(end, MAX_ORDER_NR_PAGES);
2058 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07002059 map = alloc_remap(pgdat->node_id, size);
2060 if (!map)
2061 map = alloc_bootmem_node(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07002062 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002064#ifdef CONFIG_FLATMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 /*
2066 * With no DISCONTIG, the global mem_map is just set as node 0's
2067 */
2068 if (pgdat == NODE_DATA(0))
2069 mem_map = NODE_DATA(0)->node_mem_map;
2070#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002071#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072}
2073
Yasunori Goto86356ab2006-06-23 02:03:09 -07002074void __meminit free_area_init_node(int nid, struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 unsigned long *zones_size, unsigned long node_start_pfn,
2076 unsigned long *zholes_size)
2077{
2078 pgdat->node_id = nid;
2079 pgdat->node_start_pfn = node_start_pfn;
2080 calculate_zone_totalpages(pgdat, zones_size, zholes_size);
2081
2082 alloc_node_mem_map(pgdat);
2083
2084 free_area_init_core(pgdat, zones_size, zholes_size);
2085}
2086
Dave Hansen93b75042005-06-23 00:07:47 -07002087#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088static bootmem_data_t contig_bootmem_data;
2089struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
2090
2091EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07002092#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093
2094void __init free_area_init(unsigned long *zones_size)
2095{
Dave Hansen93b75042005-06-23 00:07:47 -07002096 free_area_init_node(0, NODE_DATA(0), zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
2098}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100#ifdef CONFIG_HOTPLUG_CPU
2101static int page_alloc_cpu_notify(struct notifier_block *self,
2102 unsigned long action, void *hcpu)
2103{
2104 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105
2106 if (action == CPU_DEAD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107 local_irq_disable();
2108 __drain_pages(cpu);
Christoph Lameterf8891e52006-06-30 01:55:45 -07002109 vm_events_fold_cpu(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 local_irq_enable();
Christoph Lameter2244b952006-06-30 01:55:33 -07002111 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 }
2113 return NOTIFY_OK;
2114}
2115#endif /* CONFIG_HOTPLUG_CPU */
2116
2117void __init page_alloc_init(void)
2118{
2119 hotcpu_notifier(page_alloc_cpu_notify, 0);
2120}
2121
2122/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002123 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
2124 * or min_free_kbytes changes.
2125 */
2126static void calculate_totalreserve_pages(void)
2127{
2128 struct pglist_data *pgdat;
2129 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002130 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002131
2132 for_each_online_pgdat(pgdat) {
2133 for (i = 0; i < MAX_NR_ZONES; i++) {
2134 struct zone *zone = pgdat->node_zones + i;
2135 unsigned long max = 0;
2136
2137 /* Find valid and maximum lowmem_reserve in the zone */
2138 for (j = i; j < MAX_NR_ZONES; j++) {
2139 if (zone->lowmem_reserve[j] > max)
2140 max = zone->lowmem_reserve[j];
2141 }
2142
2143 /* we treat pages_high as reserved pages. */
2144 max += zone->pages_high;
2145
2146 if (max > zone->present_pages)
2147 max = zone->present_pages;
2148 reserve_pages += max;
2149 }
2150 }
2151 totalreserve_pages = reserve_pages;
2152}
2153
2154/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155 * setup_per_zone_lowmem_reserve - called whenever
2156 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2157 * has a correct pages reserved value, so an adequate number of
2158 * pages are left in the zone after a successful __alloc_pages().
2159 */
2160static void setup_per_zone_lowmem_reserve(void)
2161{
2162 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002163 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08002165 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166 for (j = 0; j < MAX_NR_ZONES; j++) {
2167 struct zone *zone = pgdat->node_zones + j;
2168 unsigned long present_pages = zone->present_pages;
2169
2170 zone->lowmem_reserve[j] = 0;
2171
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002172 idx = j;
2173 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 struct zone *lower_zone;
2175
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002176 idx--;
2177
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178 if (sysctl_lowmem_reserve_ratio[idx] < 1)
2179 sysctl_lowmem_reserve_ratio[idx] = 1;
2180
2181 lower_zone = pgdat->node_zones + idx;
2182 lower_zone->lowmem_reserve[j] = present_pages /
2183 sysctl_lowmem_reserve_ratio[idx];
2184 present_pages += lower_zone->present_pages;
2185 }
2186 }
2187 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002188
2189 /* update totalreserve_pages */
2190 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191}
2192
2193/*
2194 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2195 * that the pages_{min,low,high} values for each zone are set correctly
2196 * with respect to min_free_kbytes.
2197 */
Dave Hansen3947be12005-10-29 18:16:54 -07002198void setup_per_zone_pages_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199{
2200 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
2201 unsigned long lowmem_pages = 0;
2202 struct zone *zone;
2203 unsigned long flags;
2204
2205 /* Calculate total number of !ZONE_HIGHMEM pages */
2206 for_each_zone(zone) {
2207 if (!is_highmem(zone))
2208 lowmem_pages += zone->present_pages;
2209 }
2210
2211 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07002212 u64 tmp;
2213
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 spin_lock_irqsave(&zone->lru_lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07002215 tmp = (u64)pages_min * zone->present_pages;
2216 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217 if (is_highmem(zone)) {
2218 /*
Nick Piggin669ed172005-11-13 16:06:45 -08002219 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2220 * need highmem pages, so cap pages_min to a small
2221 * value here.
2222 *
2223 * The (pages_high-pages_low) and (pages_low-pages_min)
2224 * deltas controls asynch page reclaim, and so should
2225 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 */
2227 int min_pages;
2228
2229 min_pages = zone->present_pages / 1024;
2230 if (min_pages < SWAP_CLUSTER_MAX)
2231 min_pages = SWAP_CLUSTER_MAX;
2232 if (min_pages > 128)
2233 min_pages = 128;
2234 zone->pages_min = min_pages;
2235 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08002236 /*
2237 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 * proportionate to the zone's size.
2239 */
Nick Piggin669ed172005-11-13 16:06:45 -08002240 zone->pages_min = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241 }
2242
Andrew Mortonac924c62006-05-15 09:43:59 -07002243 zone->pages_low = zone->pages_min + (tmp >> 2);
2244 zone->pages_high = zone->pages_min + (tmp >> 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 spin_unlock_irqrestore(&zone->lru_lock, flags);
2246 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002247
2248 /* update totalreserve_pages */
2249 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250}
2251
2252/*
2253 * Initialise min_free_kbytes.
2254 *
2255 * For small machines we want it small (128k min). For large machines
2256 * we want it large (64MB max). But it is not linear, because network
2257 * bandwidth does not increase linearly with machine size. We use
2258 *
2259 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2260 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2261 *
2262 * which yields
2263 *
2264 * 16MB: 512k
2265 * 32MB: 724k
2266 * 64MB: 1024k
2267 * 128MB: 1448k
2268 * 256MB: 2048k
2269 * 512MB: 2896k
2270 * 1024MB: 4096k
2271 * 2048MB: 5792k
2272 * 4096MB: 8192k
2273 * 8192MB: 11584k
2274 * 16384MB: 16384k
2275 */
2276static int __init init_per_zone_pages_min(void)
2277{
2278 unsigned long lowmem_kbytes;
2279
2280 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
2281
2282 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
2283 if (min_free_kbytes < 128)
2284 min_free_kbytes = 128;
2285 if (min_free_kbytes > 65536)
2286 min_free_kbytes = 65536;
2287 setup_per_zone_pages_min();
2288 setup_per_zone_lowmem_reserve();
2289 return 0;
2290}
2291module_init(init_per_zone_pages_min)
2292
2293/*
2294 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2295 * that we can call two helper functions whenever min_free_kbytes
2296 * changes.
2297 */
2298int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
2299 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2300{
2301 proc_dointvec(table, write, file, buffer, length, ppos);
2302 setup_per_zone_pages_min();
2303 return 0;
2304}
2305
Christoph Lameter96146342006-07-03 00:24:13 -07002306#ifdef CONFIG_NUMA
2307int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
2308 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2309{
2310 struct zone *zone;
2311 int rc;
2312
2313 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2314 if (rc)
2315 return rc;
2316
2317 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07002318 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07002319 sysctl_min_unmapped_ratio) / 100;
2320 return 0;
2321}
Christoph Lameter0ff38492006-09-25 23:31:52 -07002322
2323int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
2324 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2325{
2326 struct zone *zone;
2327 int rc;
2328
2329 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2330 if (rc)
2331 return rc;
2332
2333 for_each_zone(zone)
2334 zone->min_slab_pages = (zone->present_pages *
2335 sysctl_min_slab_ratio) / 100;
2336 return 0;
2337}
Christoph Lameter96146342006-07-03 00:24:13 -07002338#endif
2339
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340/*
2341 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2342 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2343 * whenever sysctl_lowmem_reserve_ratio changes.
2344 *
2345 * The reserve ratio obviously has absolutely no relation with the
2346 * pages_min watermarks. The lowmem reserve ratio can only make sense
2347 * if in function of the boot time zone sizes.
2348 */
2349int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
2350 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2351{
2352 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2353 setup_per_zone_lowmem_reserve();
2354 return 0;
2355}
2356
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002357/*
2358 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
2359 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
2360 * can have before it gets flushed back to buddy allocator.
2361 */
2362
2363int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
2364 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2365{
2366 struct zone *zone;
2367 unsigned int cpu;
2368 int ret;
2369
2370 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2371 if (!write || (ret == -EINVAL))
2372 return ret;
2373 for_each_zone(zone) {
2374 for_each_online_cpu(cpu) {
2375 unsigned long high;
2376 high = zone->present_pages / percpu_pagelist_fraction;
2377 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
2378 }
2379 }
2380 return 0;
2381}
2382
David S. Millerf034b5d2006-08-24 03:08:07 -07002383int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384
2385#ifdef CONFIG_NUMA
2386static int __init set_hashdist(char *str)
2387{
2388 if (!str)
2389 return 0;
2390 hashdist = simple_strtoul(str, &str, 0);
2391 return 1;
2392}
2393__setup("hashdist=", set_hashdist);
2394#endif
2395
2396/*
2397 * allocate a large system hash table from bootmem
2398 * - it is assumed that the hash table must contain an exact power-of-2
2399 * quantity of entries
2400 * - limit is the number of hash buckets, not the total allocation size
2401 */
2402void *__init alloc_large_system_hash(const char *tablename,
2403 unsigned long bucketsize,
2404 unsigned long numentries,
2405 int scale,
2406 int flags,
2407 unsigned int *_hash_shift,
2408 unsigned int *_hash_mask,
2409 unsigned long limit)
2410{
2411 unsigned long long max = limit;
2412 unsigned long log2qty, size;
2413 void *table = NULL;
2414
2415 /* allow the kernel cmdline to have a say */
2416 if (!numentries) {
2417 /* round applicable memory size up to nearest megabyte */
2418 numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages;
2419 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
2420 numentries >>= 20 - PAGE_SHIFT;
2421 numentries <<= 20 - PAGE_SHIFT;
2422
2423 /* limit to 1 bucket per 2^scale bytes of low memory */
2424 if (scale > PAGE_SHIFT)
2425 numentries >>= (scale - PAGE_SHIFT);
2426 else
2427 numentries <<= (PAGE_SHIFT - scale);
2428 }
John Hawkes6e692ed2006-03-25 03:08:02 -08002429 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430
2431 /* limit allocation size to 1/16 total memory by default */
2432 if (max == 0) {
2433 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
2434 do_div(max, bucketsize);
2435 }
2436
2437 if (numentries > max)
2438 numentries = max;
2439
2440 log2qty = long_log2(numentries);
2441
2442 do {
2443 size = bucketsize << log2qty;
2444 if (flags & HASH_EARLY)
2445 table = alloc_bootmem(size);
2446 else if (hashdist)
2447 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
2448 else {
2449 unsigned long order;
2450 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
2451 ;
2452 table = (void*) __get_free_pages(GFP_ATOMIC, order);
2453 }
2454 } while (!table && size > PAGE_SIZE && --log2qty);
2455
2456 if (!table)
2457 panic("Failed to allocate %s hash table\n", tablename);
2458
2459 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2460 tablename,
2461 (1U << log2qty),
2462 long_log2(size) - PAGE_SHIFT,
2463 size);
2464
2465 if (_hash_shift)
2466 *_hash_shift = log2qty;
2467 if (_hash_mask)
2468 *_hash_mask = (1 << log2qty) - 1;
2469
2470 return table;
2471}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08002472
2473#ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08002474struct page *pfn_to_page(unsigned long pfn)
2475{
Andy Whitcroft67de6482006-06-23 02:03:12 -07002476 return __pfn_to_page(pfn);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08002477}
2478unsigned long page_to_pfn(struct page *page)
2479{
Andy Whitcroft67de6482006-06-23 02:03:12 -07002480 return __page_to_pfn(page);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08002481}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08002482EXPORT_SYMBOL(pfn_to_page);
2483EXPORT_SYMBOL(page_to_pfn);
2484#endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */