blob: 8deb9d0fd5b1781cac582487fbbb11bb28129b0b [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>
KOSAKI Motohiro10ed2732008-03-04 14:28:32 -080022#include <linux/jiffies.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/bootmem.h>
24#include <linux/compiler.h>
Randy Dunlap9f158332005-09-13 01:25:16 -070025#include <linux/kernel.h>
Vegard Nossumb1eeab62008-11-25 16:55:53 +010026#include <linux/kmemcheck.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/module.h>
28#include <linux/suspend.h>
29#include <linux/pagevec.h>
30#include <linux/blkdev.h>
31#include <linux/slab.h>
David Rientjes5a3135c2007-10-16 23:25:53 -070032#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include <linux/notifier.h>
34#include <linux/topology.h>
35#include <linux/sysctl.h>
36#include <linux/cpu.h>
37#include <linux/cpuset.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070038#include <linux/memory_hotplug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include <linux/nodemask.h>
40#include <linux/vmalloc.h>
Christoph Lameter4be38e32006-01-06 00:11:17 -080041#include <linux/mempolicy.h>
Yasunori Goto68113782006-06-23 02:03:11 -070042#include <linux/stop_machine.h>
Mel Gormanc7132162006-09-27 01:49:43 -070043#include <linux/sort.h>
44#include <linux/pfn.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070045#include <linux/backing-dev.h>
Akinobu Mita933e3122006-12-08 02:39:45 -080046#include <linux/fault-inject.h>
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -070047#include <linux/page-isolation.h>
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070048#include <linux/page_cgroup.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070049#include <linux/debugobjects.h>
Catalin Marinasdbb1f812009-06-11 13:23:19 +010050#include <linux/kmemleak.h>
Robert Jennings925cc712009-12-17 14:44:38 +000051#include <linux/memory.h>
Mel Gorman0d3d0622009-09-21 17:02:44 -070052#include <trace/events/kmem.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
54#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070055#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070056#include "internal.h"
57
58/*
Christoph Lameter13808912007-10-16 01:25:27 -070059 * Array of node states.
Linus Torvalds1da177e2005-04-16 15:20:36 -070060 */
Christoph Lameter13808912007-10-16 01:25:27 -070061nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
62 [N_POSSIBLE] = NODE_MASK_ALL,
63 [N_ONLINE] = { { [0] = 1UL } },
64#ifndef CONFIG_NUMA
65 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
66#ifdef CONFIG_HIGHMEM
67 [N_HIGH_MEMORY] = { { [0] = 1UL } },
68#endif
69 [N_CPU] = { { [0] = 1UL } },
70#endif /* NUMA */
71};
72EXPORT_SYMBOL(node_states);
73
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070074unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -070075unsigned long totalreserve_pages __read_mostly;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -080076int percpu_pagelist_fraction;
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +100077gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -070078
Mel Gormand9c23402007-10-16 01:26:01 -070079#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
80int pageblock_order __read_mostly;
81#endif
82
Hugh Dickinsd98c7a02006-02-14 13:52:59 -080083static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -080084
Linus Torvalds1da177e2005-04-16 15:20:36 -070085/*
86 * results with 256, 32 in the lowmem_reserve sysctl:
87 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
88 * 1G machine -> (16M dma, 784M normal, 224M high)
89 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
90 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
91 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +010092 *
93 * TBD: should special case ZONE_DMA32 machines here - in those we normally
94 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -070096int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -080097#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -070098 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -080099#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700100#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700101 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700102#endif
Christoph Lametere53ef382006-09-25 23:31:14 -0700103#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700104 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700105#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700106 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700107};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108
109EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110
Helge Deller15ad7cd2006-12-06 20:40:36 -0800111static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800112#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700113 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800114#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700115#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700116 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700117#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700118 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700119#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700120 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700121#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700122 "Movable",
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700123};
124
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125int min_free_kbytes = 1024;
126
Jan Beulich2c85f512009-09-21 17:03:07 -0700127static unsigned long __meminitdata nr_kernel_pages;
128static unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700129static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130
Mel Gormanc7132162006-09-27 01:49:43 -0700131#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
132 /*
Simon Arlott183ff222007-10-20 01:27:18 +0200133 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
Mel Gormanc7132162006-09-27 01:49:43 -0700134 * ranges of memory (RAM) that may be registered with add_active_range().
135 * Ranges passed to add_active_range() will be merged if possible
136 * so the number of times add_active_range() can be called is
137 * related to the number of nodes and the number of holes
138 */
139 #ifdef CONFIG_MAX_ACTIVE_REGIONS
140 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
141 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
142 #else
143 #if MAX_NUMNODES >= 32
144 /* If there can be many nodes, allow up to 50 holes per node */
145 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
146 #else
147 /* By default, allow up to 256 distinct regions */
148 #define MAX_ACTIVE_REGIONS 256
149 #endif
150 #endif
151
Jan Beulich98011f52007-07-15 23:38:17 -0700152 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
153 static int __meminitdata nr_nodemap_entries;
154 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
155 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
Adrian Bunkb69a7282008-07-23 21:28:12 -0700156 static unsigned long __initdata required_kernelcore;
Adrian Bunk484f51f2007-10-16 01:26:03 -0700157 static unsigned long __initdata required_movablecore;
Adrian Bunkb69a7282008-07-23 21:28:12 -0700158 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Mel Gorman2a1e2742007-07-17 04:03:12 -0700159
160 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
161 int movable_zone;
162 EXPORT_SYMBOL(movable_zone);
Mel Gormanc7132162006-09-27 01:49:43 -0700163#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
164
Miklos Szeredi418508c2007-05-23 13:57:55 -0700165#if MAX_NUMNODES > 1
166int nr_node_ids __read_mostly = MAX_NUMNODES;
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700167int nr_online_nodes __read_mostly = 1;
Miklos Szeredi418508c2007-05-23 13:57:55 -0700168EXPORT_SYMBOL(nr_node_ids);
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700169EXPORT_SYMBOL(nr_online_nodes);
Miklos Szeredi418508c2007-05-23 13:57:55 -0700170#endif
171
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700172int page_group_by_mobility_disabled __read_mostly;
173
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700174static void set_pageblock_migratetype(struct page *page, int migratetype)
175{
Mel Gorman49255c62009-06-16 15:31:58 -0700176
177 if (unlikely(page_group_by_mobility_disabled))
178 migratetype = MIGRATE_UNMOVABLE;
179
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700180 set_pageblock_flags_group(page, (unsigned long)migratetype,
181 PB_migrate, PB_migrate_end);
182}
183
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -0700184bool oom_killer_disabled __read_mostly;
185
Nick Piggin13e74442006-01-06 00:10:58 -0800186#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700187static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700189 int ret = 0;
190 unsigned seq;
191 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -0700192
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700193 do {
194 seq = zone_span_seqbegin(zone);
195 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
196 ret = 1;
197 else if (pfn < zone->zone_start_pfn)
198 ret = 1;
199 } while (zone_span_seqretry(zone, seq));
200
201 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700202}
203
204static int page_is_consistent(struct zone *zone, struct page *page)
205{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700206 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700207 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700209 return 0;
210
211 return 1;
212}
213/*
214 * Temporary debugging check for pages not lying within a given zone.
215 */
216static int bad_range(struct zone *zone, struct page *page)
217{
218 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700220 if (!page_is_consistent(zone, page))
221 return 1;
222
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 return 0;
224}
Nick Piggin13e74442006-01-06 00:10:58 -0800225#else
226static inline int bad_range(struct zone *zone, struct page *page)
227{
228 return 0;
229}
230#endif
231
Nick Piggin224abf92006-01-06 00:11:11 -0800232static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800234 static unsigned long resume;
235 static unsigned long nr_shown;
236 static unsigned long nr_unshown;
237
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200238 /* Don't complain about poisoned pages */
239 if (PageHWPoison(page)) {
240 __ClearPageBuddy(page);
241 return;
242 }
243
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800244 /*
245 * Allow a burst of 60 reports, then keep quiet for that minute;
246 * or allow a steady drip of one report per second.
247 */
248 if (nr_shown == 60) {
249 if (time_before(jiffies, resume)) {
250 nr_unshown++;
251 goto out;
252 }
253 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800254 printk(KERN_ALERT
255 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800256 nr_unshown);
257 nr_unshown = 0;
258 }
259 nr_shown = 0;
260 }
261 if (nr_shown++ == 0)
262 resume = jiffies + 60 * HZ;
263
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800264 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800265 current->comm, page_to_pfn(page));
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800266 printk(KERN_ALERT
Hugh Dickins3dc14742009-01-06 14:40:08 -0800267 "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n",
268 page, (void *)page->flags, page_count(page),
269 page_mapcount(page), page->mapping, page->index);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700270
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800272out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800273 /* Leave bad fields for debug, except PageBuddy could make trouble */
274 __ClearPageBuddy(page);
Randy Dunlap9f158332005-09-13 01:25:16 -0700275 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276}
277
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278/*
279 * Higher-order pages are called "compound pages". They are structured thusly:
280 *
281 * The first PAGE_SIZE page is called the "head page".
282 *
283 * The remaining PAGE_SIZE pages are called "tail pages".
284 *
285 * All pages have PG_compound set. All pages have their ->private pointing at
286 * the head page (even the head page has this).
287 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800288 * The first tail page's ->lru.next holds the address of the compound page's
289 * put_page() function. Its ->lru.prev holds the order of allocation.
290 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800292
293static void free_compound_page(struct page *page)
294{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700295 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800296}
297
Andi Kleen01ad1c02008-07-23 21:27:46 -0700298void prep_compound_page(struct page *page, unsigned long order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299{
300 int i;
301 int nr_pages = 1 << order;
302
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800303 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700304 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700305 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800306 for (i = 1; i < nr_pages; i++) {
307 struct page *p = page + i;
308
Christoph Lameterd85f3382007-05-06 14:49:39 -0700309 __SetPageTail(p);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700310 p->first_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 }
312}
313
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800314static int destroy_compound_page(struct page *page, unsigned long order)
Andy Whitcroft18229df2008-11-06 12:53:27 -0800315{
316 int i;
317 int nr_pages = 1 << order;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800318 int bad = 0;
Andy Whitcroft18229df2008-11-06 12:53:27 -0800319
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800320 if (unlikely(compound_order(page) != order) ||
321 unlikely(!PageHead(page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800322 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800323 bad++;
324 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325
Christoph Lameter6d777952007-05-06 14:49:40 -0700326 __ClearPageHead(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800327
Andy Whitcroft18229df2008-11-06 12:53:27 -0800328 for (i = 1; i < nr_pages; i++) {
329 struct page *p = page + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330
Alexey Zaytseve713a212009-01-10 02:47:57 +0300331 if (unlikely(!PageTail(p) || (p->first_page != page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800332 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800333 bad++;
334 }
Christoph Lameterd85f3382007-05-06 14:49:39 -0700335 __ClearPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800337
338 return bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340
Nick Piggin17cf4402006-03-22 00:08:41 -0800341static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
342{
343 int i;
344
Andrew Morton6626c5d2006-03-22 00:08:42 -0800345 /*
346 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
347 * and __GFP_HIGHMEM from hard or soft interrupt context.
348 */
Nick Piggin725d7042006-09-25 23:30:55 -0700349 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800350 for (i = 0; i < (1 << order); i++)
351 clear_highpage(page + i);
352}
353
Andrew Morton6aa30012006-04-18 22:20:52 -0700354static inline void set_page_order(struct page *page, int order)
355{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700356 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000357 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358}
359
360static inline void rmv_page_order(struct page *page)
361{
Nick Piggin676165a2006-04-10 11:21:48 +1000362 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700363 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364}
365
366/*
367 * Locate the struct page for both the matching buddy in our
368 * pair (buddy1) and the combined O(n+1) page they form (page).
369 *
370 * 1) Any buddy B1 will have an order O twin B2 which satisfies
371 * the following equation:
372 * B2 = B1 ^ (1 << O)
373 * For example, if the starting buddy (buddy2) is #8 its order
374 * 1 buddy is #10:
375 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
376 *
377 * 2) Any buddy B will have an order O+1 parent P which
378 * satisfies the following equation:
379 * P = B & ~(1 << O)
380 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200381 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 */
383static inline struct page *
384__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
385{
386 unsigned long buddy_idx = page_idx ^ (1 << order);
387
388 return page + (buddy_idx - page_idx);
389}
390
391static inline unsigned long
392__find_combined_index(unsigned long page_idx, unsigned int order)
393{
394 return (page_idx & ~(1 << order));
395}
396
397/*
398 * This function checks whether a page is free && is the buddy
399 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800400 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000401 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700402 * (c) a page and its buddy have the same order &&
403 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 *
Nick Piggin676165a2006-04-10 11:21:48 +1000405 * For recording whether a page is in the buddy system, we use PG_buddy.
406 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
407 *
408 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700410static inline int page_is_buddy(struct page *page, struct page *buddy,
411 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700413 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800414 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800415
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700416 if (page_zone_id(page) != page_zone_id(buddy))
417 return 0;
418
419 if (PageBuddy(buddy) && page_order(buddy) == order) {
Nick Piggina3af9c32009-06-16 15:32:10 -0700420 VM_BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa30012006-04-18 22:20:52 -0700421 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000422 }
Andrew Morton6aa30012006-04-18 22:20:52 -0700423 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424}
425
426/*
427 * Freeing function for a buddy system allocator.
428 *
429 * The concept of a buddy system is to maintain direct-mapped table
430 * (containing bit values) for memory blocks of various "orders".
431 * The bottom level table contains the map for the smallest allocatable
432 * units of memory (here, pages), and each level above it describes
433 * pairs of units from the levels below, hence, "buddies".
434 * At a high level, all that happens here is marking the table entry
435 * at the bottom level available, and propagating the changes upward
436 * as necessary, plus some accounting needed to play nicely with other
437 * parts of the VM system.
438 * At each level, we keep a list of pages, which are heads of continuous
Nick Piggin676165a2006-04-10 11:21:48 +1000439 * free pages of length of (1 << order) and marked with PG_buddy. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700440 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 * So when we are allocating or freeing one, we can derive the state of the
442 * other. That is, if we allocate a small block, and both were
443 * free, the remainder of the region must be split into blocks.
444 * If a block is freed, and its buddy is also free, then this
445 * triggers coalescing into a block of larger size.
446 *
447 * -- wli
448 */
449
Nick Piggin48db57f2006-01-08 01:00:42 -0800450static inline void __free_one_page(struct page *page,
Mel Gormaned0ae212009-06-16 15:32:07 -0700451 struct zone *zone, unsigned int order,
452 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453{
454 unsigned long page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455
Nick Piggin224abf92006-01-06 00:11:11 -0800456 if (unlikely(PageCompound(page)))
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800457 if (unlikely(destroy_compound_page(page, order)))
458 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459
Mel Gormaned0ae212009-06-16 15:32:07 -0700460 VM_BUG_ON(migratetype == -1);
461
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
463
Mel Gormanf2260e62009-06-16 15:32:13 -0700464 VM_BUG_ON(page_idx & ((1 << order) - 1));
Nick Piggin725d7042006-09-25 23:30:55 -0700465 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 while (order < MAX_ORDER-1) {
468 unsigned long combined_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 struct page *buddy;
470
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 buddy = __page_find_buddy(page, page_idx, order);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700472 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700473 break;
Nick Piggin13e74442006-01-06 00:10:58 -0800474
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700475 /* Our buddy is free, merge with it and move up one order. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 list_del(&buddy->lru);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700477 zone->free_area[order].nr_free--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 rmv_page_order(buddy);
Nick Piggin13e74442006-01-06 00:10:58 -0800479 combined_idx = __find_combined_index(page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 page = page + (combined_idx - page_idx);
481 page_idx = combined_idx;
482 order++;
483 }
484 set_page_order(page, order);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700485 list_add(&page->lru,
486 &zone->free_area[order].free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 zone->free_area[order].nr_free++;
488}
489
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700490/*
491 * free_page_mlock() -- clean up attempts to free and mlocked() page.
492 * Page should not be on lru, so no need to fix that up.
493 * free_pages_check() will verify...
494 */
495static inline void free_page_mlock(struct page *page)
496{
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700497 __dec_zone_page_state(page, NR_MLOCK);
498 __count_vm_event(UNEVICTABLE_MLOCKFREED);
499}
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700500
Nick Piggin224abf92006-01-06 00:11:11 -0800501static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502{
Nick Piggin92be2e32006-01-06 00:10:57 -0800503 if (unlikely(page_mapcount(page) |
504 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700505 (atomic_read(&page->_count) != 0) |
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800506 (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800507 bad_page(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800508 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800509 }
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800510 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
511 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
512 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513}
514
515/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700516 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700518 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 *
520 * If the zone was previously in an "all pages pinned" state then look to
521 * see if this freeing clears that state.
522 *
523 * And clear the zone's pages_scanned counter, to hold off the "all pages are
524 * pinned" detection logic.
525 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700526static void free_pcppages_bulk(struct zone *zone, int count,
527 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528{
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700529 int migratetype = 0;
Mel Gormana6f9edd2009-09-21 17:03:20 -0700530 int batch_free = 0;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700531
Nick Pigginc54ad302006-01-06 00:10:56 -0800532 spin_lock(&zone->lock);
David Rientjese815af92007-10-16 23:25:54 -0700533 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700535
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700536 __mod_zone_page_state(zone, NR_FREE_PAGES, count);
Mel Gormana6f9edd2009-09-21 17:03:20 -0700537 while (count) {
Nick Piggin48db57f2006-01-08 01:00:42 -0800538 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700539 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -0800540
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700541 /*
Mel Gormana6f9edd2009-09-21 17:03:20 -0700542 * Remove pages from lists in a round-robin fashion. A
543 * batch_free count is maintained that is incremented when an
544 * empty list is encountered. This is so more pages are freed
545 * off fuller lists instead of spinning excessively around empty
546 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700547 */
548 do {
Mel Gormana6f9edd2009-09-21 17:03:20 -0700549 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700550 if (++migratetype == MIGRATE_PCPTYPES)
551 migratetype = 0;
552 list = &pcp->lists[migratetype];
553 } while (list_empty(list));
554
Mel Gormana6f9edd2009-09-21 17:03:20 -0700555 do {
556 page = list_entry(list->prev, struct page, lru);
557 /* must delete as __free_one_page list manipulates */
558 list_del(&page->lru);
Hugh Dickinsa7016232010-01-29 17:46:34 +0000559 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
560 __free_one_page(page, zone, 0, page_private(page));
561 trace_mm_page_pcpu_drain(page, 0, page_private(page));
Mel Gormana6f9edd2009-09-21 17:03:20 -0700562 } while (--count && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800564 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565}
566
Mel Gormaned0ae212009-06-16 15:32:07 -0700567static void free_one_page(struct zone *zone, struct page *page, int order,
568 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800569{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700570 spin_lock(&zone->lock);
David Rientjese815af92007-10-16 23:25:54 -0700571 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700572 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700573
574 __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
Mel Gormaned0ae212009-06-16 15:32:07 -0700575 __free_one_page(page, zone, order, migratetype);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700576 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800577}
578
579static void __free_pages_ok(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580{
Nick Pigginc54ad302006-01-06 00:10:56 -0800581 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 int i;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800583 int bad = 0;
Johannes Weiner451ea252009-09-21 17:01:48 -0700584 int wasMlocked = __TestClearPageMlocked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100586 kmemcheck_free_shadow(page, order);
587
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 for (i = 0 ; i < (1 << order) ; ++i)
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800589 bad += free_pages_check(page + i);
590 if (bad)
Hugh Dickins689bceb2005-11-21 21:32:20 -0800591 return;
592
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700593 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -0700594 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700595 debug_check_no_obj_freed(page_address(page),
596 PAGE_SIZE << order);
597 }
Nick Piggindafb1362006-10-11 01:21:30 -0700598 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800599 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700600
Nick Pigginc54ad302006-01-06 00:10:56 -0800601 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +0200602 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -0700603 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700604 __count_vm_events(PGFREE, 1 << order);
Mel Gormaned0ae212009-06-16 15:32:07 -0700605 free_one_page(page_zone(page), page, order,
606 get_pageblock_migratetype(page));
Nick Pigginc54ad302006-01-06 00:10:56 -0800607 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608}
609
David Howellsa226f6c2006-01-06 00:11:08 -0800610/*
611 * permit the bootmem allocator to evade page validation on high-order frees
612 */
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700613void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800614{
615 if (order == 0) {
616 __ClearPageReserved(page);
617 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800618 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800619 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800620 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800621 int loop;
622
Nick Piggin545b1ea2006-03-22 00:08:07 -0800623 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800624 for (loop = 0; loop < BITS_PER_LONG; loop++) {
625 struct page *p = &page[loop];
626
Nick Piggin545b1ea2006-03-22 00:08:07 -0800627 if (loop + 1 < BITS_PER_LONG)
628 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800629 __ClearPageReserved(p);
630 set_page_count(p, 0);
631 }
632
Nick Piggin7835e982006-03-22 00:08:40 -0800633 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800634 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800635 }
636}
637
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638
639/*
640 * The order of subdivision here is critical for the IO subsystem.
641 * Please do not alter this order without good reasons and regression
642 * testing. Specifically, as large blocks of memory are subdivided,
643 * the order in which smaller blocks are delivered depends on the order
644 * they're subdivided in this function. This is the primary factor
645 * influencing the order in which pages are delivered to the IO
646 * subsystem according to empirical testing, and this is also justified
647 * by considering the behavior of a buddy system containing a single
648 * large block of memory acted on by a series of small allocations.
649 * This behavior is a critical factor in sglist merging's success.
650 *
651 * -- wli
652 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800653static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700654 int low, int high, struct free_area *area,
655 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656{
657 unsigned long size = 1 << high;
658
659 while (high > low) {
660 area--;
661 high--;
662 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700663 VM_BUG_ON(bad_range(zone, &page[size]));
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700664 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 area->nr_free++;
666 set_page_order(&page[size], high);
667 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668}
669
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670/*
671 * This page is about to be returned from the page allocator
672 */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200673static inline int check_new_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674{
Nick Piggin92be2e32006-01-06 00:10:57 -0800675 if (unlikely(page_mapcount(page) |
676 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700677 (atomic_read(&page->_count) != 0) |
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800678 (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800679 bad_page(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800680 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800681 }
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200682 return 0;
683}
684
685static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
686{
687 int i;
688
689 for (i = 0; i < (1 << order); i++) {
690 struct page *p = page + i;
691 if (unlikely(check_new_page(p)))
692 return 1;
693 }
Hugh Dickins689bceb2005-11-21 21:32:20 -0800694
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700695 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800696 set_page_refcounted(page);
Nick Piggincc102502006-12-06 20:32:00 -0800697
698 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800700
701 if (gfp_flags & __GFP_ZERO)
702 prep_zero_page(page, order, gfp_flags);
703
704 if (order && (gfp_flags & __GFP_COMP))
705 prep_compound_page(page, order);
706
Hugh Dickins689bceb2005-11-21 21:32:20 -0800707 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708}
709
Mel Gorman56fd56b2007-10-16 01:25:58 -0700710/*
711 * Go through the free lists for the given migratetype and remove
712 * the smallest available page from the freelists
713 */
Mel Gorman728ec982009-06-16 15:32:04 -0700714static inline
715struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -0700716 int migratetype)
717{
718 unsigned int current_order;
719 struct free_area * area;
720 struct page *page;
721
722 /* Find a page of the appropriate size in the preferred list */
723 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
724 area = &(zone->free_area[current_order]);
725 if (list_empty(&area->free_list[migratetype]))
726 continue;
727
728 page = list_entry(area->free_list[migratetype].next,
729 struct page, lru);
730 list_del(&page->lru);
731 rmv_page_order(page);
732 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -0700733 expand(zone, page, order, current_order, area, migratetype);
734 return page;
735 }
736
737 return NULL;
738}
739
740
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700741/*
742 * This array describes the order lists are fallen back to when
743 * the free lists for the desirable migrate type are depleted
744 */
745static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
Mel Gorman64c5e132007-10-16 01:25:59 -0700746 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
747 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
748 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
749 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700750};
751
Mel Gormanc361be52007-10-16 01:25:51 -0700752/*
753 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -0700754 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -0700755 * boundary. If alignment is required, use move_freepages_block()
756 */
Adrian Bunkb69a7282008-07-23 21:28:12 -0700757static int move_freepages(struct zone *zone,
758 struct page *start_page, struct page *end_page,
759 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700760{
761 struct page *page;
762 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -0700763 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -0700764
765#ifndef CONFIG_HOLES_IN_ZONE
766 /*
767 * page_zone is not safe to call in this context when
768 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
769 * anyway as we check zone boundaries in move_freepages_block().
770 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -0700771 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -0700772 */
773 BUG_ON(page_zone(start_page) != page_zone(end_page));
774#endif
775
776 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -0700777 /* Make sure we are not inadvertently changing nodes */
778 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
779
Mel Gormanc361be52007-10-16 01:25:51 -0700780 if (!pfn_valid_within(page_to_pfn(page))) {
781 page++;
782 continue;
783 }
784
785 if (!PageBuddy(page)) {
786 page++;
787 continue;
788 }
789
790 order = page_order(page);
791 list_del(&page->lru);
792 list_add(&page->lru,
793 &zone->free_area[order].free_list[migratetype]);
794 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -0700795 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -0700796 }
797
Mel Gormand1003132007-10-16 01:26:00 -0700798 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -0700799}
800
Adrian Bunkb69a7282008-07-23 21:28:12 -0700801static int move_freepages_block(struct zone *zone, struct page *page,
802 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700803{
804 unsigned long start_pfn, end_pfn;
805 struct page *start_page, *end_page;
806
807 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -0700808 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -0700809 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -0700810 end_page = start_page + pageblock_nr_pages - 1;
811 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -0700812
813 /* Do not cross zone boundaries */
814 if (start_pfn < zone->zone_start_pfn)
815 start_page = page;
816 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
817 return 0;
818
819 return move_freepages(zone, start_page, end_page, migratetype);
820}
821
Mel Gorman2f66a682009-09-21 17:02:31 -0700822static void change_pageblock_range(struct page *pageblock_page,
823 int start_order, int migratetype)
824{
825 int nr_pageblocks = 1 << (start_order - pageblock_order);
826
827 while (nr_pageblocks--) {
828 set_pageblock_migratetype(pageblock_page, migratetype);
829 pageblock_page += pageblock_nr_pages;
830 }
831}
832
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700833/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -0700834static inline struct page *
835__rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700836{
837 struct free_area * area;
838 int current_order;
839 struct page *page;
840 int migratetype, i;
841
842 /* Find the largest possible block of pages in the other list */
843 for (current_order = MAX_ORDER-1; current_order >= order;
844 --current_order) {
845 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
846 migratetype = fallbacks[start_migratetype][i];
847
Mel Gorman56fd56b2007-10-16 01:25:58 -0700848 /* MIGRATE_RESERVE handled later if necessary */
849 if (migratetype == MIGRATE_RESERVE)
850 continue;
Mel Gormane0104872007-10-16 01:25:53 -0700851
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700852 area = &(zone->free_area[current_order]);
853 if (list_empty(&area->free_list[migratetype]))
854 continue;
855
856 page = list_entry(area->free_list[migratetype].next,
857 struct page, lru);
858 area->nr_free--;
859
860 /*
Mel Gormanc361be52007-10-16 01:25:51 -0700861 * If breaking a large block of pages, move all free
Mel Gorman46dafbc2007-10-16 01:25:55 -0700862 * pages to the preferred allocation list. If falling
863 * back for a reclaimable kernel allocation, be more
864 * agressive about taking ownership of free pages
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700865 */
Mel Gormand9c23402007-10-16 01:26:01 -0700866 if (unlikely(current_order >= (pageblock_order >> 1)) ||
Mel Gormandd5d2412009-09-05 11:17:11 -0700867 start_migratetype == MIGRATE_RECLAIMABLE ||
868 page_group_by_mobility_disabled) {
Mel Gorman46dafbc2007-10-16 01:25:55 -0700869 unsigned long pages;
870 pages = move_freepages_block(zone, page,
871 start_migratetype);
872
873 /* Claim the whole block if over half of it is free */
Mel Gormandd5d2412009-09-05 11:17:11 -0700874 if (pages >= (1 << (pageblock_order-1)) ||
875 page_group_by_mobility_disabled)
Mel Gorman46dafbc2007-10-16 01:25:55 -0700876 set_pageblock_migratetype(page,
877 start_migratetype);
878
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700879 migratetype = start_migratetype;
Mel Gormanc361be52007-10-16 01:25:51 -0700880 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700881
882 /* Remove the page from the freelists */
883 list_del(&page->lru);
884 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700885
Mel Gorman2f66a682009-09-21 17:02:31 -0700886 /* Take ownership for orders >= pageblock_order */
887 if (current_order >= pageblock_order)
888 change_pageblock_range(page, current_order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700889 start_migratetype);
890
891 expand(zone, page, order, current_order, area, migratetype);
Mel Gormane0fff1b2009-09-21 17:02:42 -0700892
893 trace_mm_page_alloc_extfrag(page, order, current_order,
894 start_migratetype, migratetype);
895
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700896 return page;
897 }
898 }
899
Mel Gorman728ec982009-06-16 15:32:04 -0700900 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700901}
902
Mel Gorman56fd56b2007-10-16 01:25:58 -0700903/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 * Do the hard work of removing an element from the buddy allocator.
905 * Call me with the zone->lock already held.
906 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700907static struct page *__rmqueue(struct zone *zone, unsigned int order,
908 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 struct page *page;
911
Mel Gorman728ec982009-06-16 15:32:04 -0700912retry_reserve:
Mel Gorman56fd56b2007-10-16 01:25:58 -0700913 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914
Mel Gorman728ec982009-06-16 15:32:04 -0700915 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Mel Gorman56fd56b2007-10-16 01:25:58 -0700916 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700917
Mel Gorman728ec982009-06-16 15:32:04 -0700918 /*
919 * Use MIGRATE_RESERVE rather than fail an allocation. goto
920 * is used because __rmqueue_smallest is an inline function
921 * and we want just one call site
922 */
923 if (!page) {
924 migratetype = MIGRATE_RESERVE;
925 goto retry_reserve;
926 }
927 }
928
Mel Gorman0d3d0622009-09-21 17:02:44 -0700929 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700930 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931}
932
933/*
934 * Obtain a specified number of elements from the buddy allocator, all under
935 * a single hold of the lock, for efficiency. Add them to the supplied list.
936 * Returns the number of new pages which were placed at *list.
937 */
938static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700939 unsigned long count, struct list_head *list,
Mel Gormane084b2d2009-07-29 15:02:04 -0700940 int migratetype, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943
Nick Pigginc54ad302006-01-06 00:10:56 -0800944 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700946 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800947 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -0800949
950 /*
951 * Split buddy pages returned by expand() are received here
952 * in physical page order. The page is added to the callers and
953 * list and the list head then moves forward. From the callers
954 * perspective, the linked list is ordered by page number in
955 * some conditions. This is useful for IO devices that can
956 * merge IO requests if the physical pages are ordered
957 * properly.
958 */
Mel Gormane084b2d2009-07-29 15:02:04 -0700959 if (likely(cold == 0))
960 list_add(&page->lru, list);
961 else
962 list_add_tail(&page->lru, list);
Mel Gorman535131e2007-10-16 01:25:49 -0700963 set_page_private(page, migratetype);
Mel Gorman81eabcb2007-12-17 16:20:05 -0800964 list = &page->lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 }
Mel Gormanf2260e62009-06-16 15:32:13 -0700966 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -0800967 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800968 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969}
970
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700971#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800972/*
Christoph Lameter4037d452007-05-09 02:35:14 -0700973 * Called from the vmstat counter updater to drain pagesets of this
974 * currently executing processor on remote nodes after they have
975 * expired.
976 *
Christoph Lameter879336c2006-03-22 00:09:08 -0800977 * Note that this function must be called with the thread pinned to
978 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800979 */
Christoph Lameter4037d452007-05-09 02:35:14 -0700980void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700981{
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700982 unsigned long flags;
Christoph Lameter4037d452007-05-09 02:35:14 -0700983 int to_drain;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700984
Christoph Lameter4037d452007-05-09 02:35:14 -0700985 local_irq_save(flags);
986 if (pcp->count >= pcp->batch)
987 to_drain = pcp->batch;
988 else
989 to_drain = pcp->count;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700990 free_pcppages_bulk(zone, to_drain, pcp);
Christoph Lameter4037d452007-05-09 02:35:14 -0700991 pcp->count -= to_drain;
992 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700993}
994#endif
995
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800996/*
997 * Drain pages of the indicated processor.
998 *
999 * The processor must either be the current processor and the
1000 * thread pinned to the current processor or a processor that
1001 * is not online.
1002 */
1003static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004{
Nick Pigginc54ad302006-01-06 00:10:56 -08001005 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07001008 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 struct per_cpu_pageset *pset;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001010 struct per_cpu_pages *pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001012 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001014 pcp = &pset->pcp;
1015 local_irq_save(flags);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001016 free_pcppages_bulk(zone, pcp->count, pcp);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001017 pcp->count = 0;
1018 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 }
1020}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001022/*
1023 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1024 */
1025void drain_local_pages(void *arg)
1026{
1027 drain_pages(smp_processor_id());
1028}
1029
1030/*
1031 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
1032 */
1033void drain_all_pages(void)
1034{
Jens Axboe15c8b6c2008-05-09 09:39:44 +02001035 on_each_cpu(drain_local_pages, NULL, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001036}
1037
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02001038#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039
1040void mark_free_pages(struct zone *zone)
1041{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001042 unsigned long pfn, max_zone_pfn;
1043 unsigned long flags;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001044 int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 struct list_head *curr;
1046
1047 if (!zone->spanned_pages)
1048 return;
1049
1050 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001051
1052 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1053 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1054 if (pfn_valid(pfn)) {
1055 struct page *page = pfn_to_page(pfn);
1056
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001057 if (!swsusp_page_is_forbidden(page))
1058 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001059 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001061 for_each_migratetype_order(order, t) {
1062 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001063 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001065 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1066 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001067 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001068 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001069 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 spin_unlock_irqrestore(&zone->lock, flags);
1071}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001072#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073
1074/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 * Free a 0-order page
1076 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001077static void free_hot_cold_page(struct page *page, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078{
1079 struct zone *zone = page_zone(page);
1080 struct per_cpu_pages *pcp;
1081 unsigned long flags;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001082 int migratetype;
Johannes Weiner451ea252009-09-21 17:01:48 -07001083 int wasMlocked = __TestClearPageMlocked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001085 kmemcheck_free_shadow(page, 0);
1086
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 if (PageAnon(page))
1088 page->mapping = NULL;
Nick Piggin224abf92006-01-06 00:11:11 -08001089 if (free_pages_check(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001090 return;
1091
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001092 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -07001093 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001094 debug_check_no_obj_freed(page_address(page), PAGE_SIZE);
1095 }
Nick Piggindafb1362006-10-11 01:21:30 -07001096 arch_free_page(page, 0);
Hugh Dickins689bceb2005-11-21 21:32:20 -08001097 kernel_map_pages(page, 1, 0);
1098
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001099 pcp = &zone_pcp(zone, get_cpu())->pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001100 migratetype = get_pageblock_migratetype(page);
1101 set_page_private(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +02001103 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -07001104 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001105 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001106
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001107 /*
1108 * We only track unmovable, reclaimable and movable on pcp lists.
1109 * Free ISOLATE pages back to the allocator because they are being
1110 * offlined but treat RESERVE as movable pages so we can get those
1111 * areas back if necessary. Otherwise, we may have to free
1112 * excessively into the page allocator
1113 */
1114 if (migratetype >= MIGRATE_PCPTYPES) {
1115 if (unlikely(migratetype == MIGRATE_ISOLATE)) {
1116 free_one_page(zone, page, 0, migratetype);
1117 goto out;
1118 }
1119 migratetype = MIGRATE_MOVABLE;
1120 }
1121
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001122 if (cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001123 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001124 else
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001125 list_add(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001127 if (pcp->count >= pcp->high) {
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001128 free_pcppages_bulk(zone, pcp->batch, pcp);
Nick Piggin48db57f2006-01-08 01:00:42 -08001129 pcp->count -= pcp->batch;
1130 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001131
1132out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 local_irq_restore(flags);
1134 put_cpu();
1135}
1136
Harvey Harrison920c7a52008-02-04 22:29:26 -08001137void free_hot_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138{
Mel Gorman4b4f2782009-09-21 17:02:41 -07001139 trace_mm_page_free_direct(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 free_hot_cold_page(page, 0);
1141}
1142
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001143/*
1144 * split_page takes a non-compound higher-order page, and splits it into
1145 * n (1<<order) sub-pages: page[0..n]
1146 * Each sub-page must be freed individually.
1147 *
1148 * Note: this is probably too low level an operation for use in drivers.
1149 * Please consult with lkml before using this in your driver.
1150 */
1151void split_page(struct page *page, unsigned int order)
1152{
1153 int i;
1154
Nick Piggin725d7042006-09-25 23:30:55 -07001155 VM_BUG_ON(PageCompound(page));
1156 VM_BUG_ON(!page_count(page));
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001157
1158#ifdef CONFIG_KMEMCHECK
1159 /*
1160 * Split shadow pages too, because free(page[0]) would
1161 * otherwise free the whole shadow.
1162 */
1163 if (kmemcheck_page_is_tracked(page))
1164 split_page(virt_to_page(page[0].shadow), order);
1165#endif
1166
Nick Piggin7835e982006-03-22 00:08:40 -08001167 for (i = 1; i < (1 << order); i++)
1168 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001169}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001170
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171/*
1172 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1173 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1174 * or two.
1175 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07001176static inline
1177struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001178 struct zone *zone, int order, gfp_t gfp_flags,
1179 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180{
1181 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001182 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 int cold = !!(gfp_flags & __GFP_COLD);
Nick Piggina74609f2006-01-06 00:11:20 -08001184 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185
Hugh Dickins689bceb2005-11-21 21:32:20 -08001186again:
Nick Piggina74609f2006-01-06 00:11:20 -08001187 cpu = get_cpu();
Nick Piggin48db57f2006-01-08 01:00:42 -08001188 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001190 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001192 pcp = &zone_pcp(zone, cpu)->pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001193 list = &pcp->lists[migratetype];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 local_irq_save(flags);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001195 if (list_empty(list)) {
Mel Gorman535131e2007-10-16 01:25:49 -07001196 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001197 pcp->batch, list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001198 migratetype, cold);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001199 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07001200 goto failed;
Mel Gorman535131e2007-10-16 01:25:49 -07001201 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001202
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001203 if (cold)
1204 page = list_entry(list->prev, struct page, lru);
1205 else
1206 page = list_entry(list->next, struct page, lru);
1207
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001208 list_del(&page->lru);
1209 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001210 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07001211 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1212 /*
1213 * __GFP_NOFAIL is not to be used in new code.
1214 *
1215 * All __GFP_NOFAIL callers should be fixed so that they
1216 * properly detect and handle allocation failures.
1217 *
1218 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07001219 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07001220 * __GFP_NOFAIL.
1221 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07001222 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07001223 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001225 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001226 spin_unlock(&zone->lock);
1227 if (!page)
1228 goto failed;
KOSAKI Motohiro6ccf80e2010-01-15 17:01:18 -08001229 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 }
1231
Christoph Lameterf8891e52006-06-30 01:55:45 -07001232 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Mel Gorman18ea7e72008-04-28 02:12:14 -07001233 zone_statistics(preferred_zone, zone);
Nick Piggina74609f2006-01-06 00:11:20 -08001234 local_irq_restore(flags);
1235 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236
Nick Piggin725d7042006-09-25 23:30:55 -07001237 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -08001238 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001239 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001241
1242failed:
1243 local_irq_restore(flags);
1244 put_cpu();
1245 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246}
1247
Mel Gorman41858962009-06-16 15:32:12 -07001248/* The ALLOC_WMARK bits are used as an index to zone->watermark */
1249#define ALLOC_WMARK_MIN WMARK_MIN
1250#define ALLOC_WMARK_LOW WMARK_LOW
1251#define ALLOC_WMARK_HIGH WMARK_HIGH
1252#define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1253
1254/* Mask to get the watermark bits */
1255#define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1256
Nick Piggin31488902005-11-28 13:44:03 -08001257#define ALLOC_HARDER 0x10 /* try to alloc harder */
1258#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1259#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001260
Akinobu Mita933e3122006-12-08 02:39:45 -08001261#ifdef CONFIG_FAIL_PAGE_ALLOC
1262
1263static struct fail_page_alloc_attr {
1264 struct fault_attr attr;
1265
1266 u32 ignore_gfp_highmem;
1267 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001268 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001269
1270#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1271
1272 struct dentry *ignore_gfp_highmem_file;
1273 struct dentry *ignore_gfp_wait_file;
Akinobu Mita54114992007-07-15 23:40:23 -07001274 struct dentry *min_order_file;
Akinobu Mita933e3122006-12-08 02:39:45 -08001275
1276#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1277
1278} fail_page_alloc = {
1279 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001280 .ignore_gfp_wait = 1,
1281 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001282 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001283};
1284
1285static int __init setup_fail_page_alloc(char *str)
1286{
1287 return setup_fault_attr(&fail_page_alloc.attr, str);
1288}
1289__setup("fail_page_alloc=", setup_fail_page_alloc);
1290
1291static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1292{
Akinobu Mita54114992007-07-15 23:40:23 -07001293 if (order < fail_page_alloc.min_order)
1294 return 0;
Akinobu Mita933e3122006-12-08 02:39:45 -08001295 if (gfp_mask & __GFP_NOFAIL)
1296 return 0;
1297 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1298 return 0;
1299 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1300 return 0;
1301
1302 return should_fail(&fail_page_alloc.attr, 1 << order);
1303}
1304
1305#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1306
1307static int __init fail_page_alloc_debugfs(void)
1308{
1309 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1310 struct dentry *dir;
1311 int err;
1312
1313 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1314 "fail_page_alloc");
1315 if (err)
1316 return err;
1317 dir = fail_page_alloc.attr.dentries.dir;
1318
1319 fail_page_alloc.ignore_gfp_wait_file =
1320 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1321 &fail_page_alloc.ignore_gfp_wait);
1322
1323 fail_page_alloc.ignore_gfp_highmem_file =
1324 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1325 &fail_page_alloc.ignore_gfp_highmem);
Akinobu Mita54114992007-07-15 23:40:23 -07001326 fail_page_alloc.min_order_file =
1327 debugfs_create_u32("min-order", mode, dir,
1328 &fail_page_alloc.min_order);
Akinobu Mita933e3122006-12-08 02:39:45 -08001329
1330 if (!fail_page_alloc.ignore_gfp_wait_file ||
Akinobu Mita54114992007-07-15 23:40:23 -07001331 !fail_page_alloc.ignore_gfp_highmem_file ||
1332 !fail_page_alloc.min_order_file) {
Akinobu Mita933e3122006-12-08 02:39:45 -08001333 err = -ENOMEM;
1334 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1335 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
Akinobu Mita54114992007-07-15 23:40:23 -07001336 debugfs_remove(fail_page_alloc.min_order_file);
Akinobu Mita933e3122006-12-08 02:39:45 -08001337 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1338 }
1339
1340 return err;
1341}
1342
1343late_initcall(fail_page_alloc_debugfs);
1344
1345#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1346
1347#else /* CONFIG_FAIL_PAGE_ALLOC */
1348
1349static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1350{
1351 return 0;
1352}
1353
1354#endif /* CONFIG_FAIL_PAGE_ALLOC */
1355
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356/*
1357 * Return 1 if free pages are above 'mark'. This takes into account the order
1358 * of the allocation.
1359 */
1360int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001361 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362{
1363 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001364 long min = mark;
1365 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 int o;
1367
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001368 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001370 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 min -= min / 4;
1372
1373 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1374 return 0;
1375 for (o = 0; o < order; o++) {
1376 /* At the next order, this order's pages become unavailable */
1377 free_pages -= z->free_area[o].nr_free << o;
1378
1379 /* Require fewer higher order pages to be free */
1380 min >>= 1;
1381
1382 if (free_pages <= min)
1383 return 0;
1384 }
1385 return 1;
1386}
1387
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001388#ifdef CONFIG_NUMA
1389/*
1390 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1391 * skip over zones that are not allowed by the cpuset, or that have
1392 * been recently (in last second) found to be nearly full. See further
1393 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001394 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001395 *
1396 * If the zonelist cache is present in the passed in zonelist, then
1397 * returns a pointer to the allowed node mask (either the current
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001398 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001399 *
1400 * If the zonelist cache is not available for this zonelist, does
1401 * nothing and returns NULL.
1402 *
1403 * If the fullzones BITMAP in the zonelist cache is stale (more than
1404 * a second since last zap'd) then we zap it out (clear its bits.)
1405 *
1406 * We hold off even calling zlc_setup, until after we've checked the
1407 * first zone in the zonelist, on the theory that most allocations will
1408 * be satisfied from that first zone, so best to examine that zone as
1409 * quickly as we can.
1410 */
1411static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1412{
1413 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1414 nodemask_t *allowednodes; /* zonelist_cache approximation */
1415
1416 zlc = zonelist->zlcache_ptr;
1417 if (!zlc)
1418 return NULL;
1419
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001420 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001421 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1422 zlc->last_full_zap = jiffies;
1423 }
1424
1425 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1426 &cpuset_current_mems_allowed :
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001427 &node_states[N_HIGH_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001428 return allowednodes;
1429}
1430
1431/*
1432 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1433 * if it is worth looking at further for free memory:
1434 * 1) Check that the zone isn't thought to be full (doesn't have its
1435 * bit set in the zonelist_cache fullzones BITMAP).
1436 * 2) Check that the zones node (obtained from the zonelist_cache
1437 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1438 * Return true (non-zero) if zone is worth looking at further, or
1439 * else return false (zero) if it is not.
1440 *
1441 * This check -ignores- the distinction between various watermarks,
1442 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1443 * found to be full for any variation of these watermarks, it will
1444 * be considered full for up to one second by all requests, unless
1445 * we are so low on memory on all allowed nodes that we are forced
1446 * into the second scan of the zonelist.
1447 *
1448 * In the second scan we ignore this zonelist cache and exactly
1449 * apply the watermarks to all zones, even it is slower to do so.
1450 * We are low on memory in the second scan, and should leave no stone
1451 * unturned looking for a free page.
1452 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001453static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001454 nodemask_t *allowednodes)
1455{
1456 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1457 int i; /* index of *z in zonelist zones */
1458 int n; /* node that zone *z is on */
1459
1460 zlc = zonelist->zlcache_ptr;
1461 if (!zlc)
1462 return 1;
1463
Mel Gormandd1a2392008-04-28 02:12:17 -07001464 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001465 n = zlc->z_to_n[i];
1466
1467 /* This zone is worth trying if it is allowed but not full */
1468 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1469}
1470
1471/*
1472 * Given 'z' scanning a zonelist, set the corresponding bit in
1473 * zlc->fullzones, so that subsequent attempts to allocate a page
1474 * from that zone don't waste time re-examining it.
1475 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001476static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001477{
1478 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1479 int i; /* index of *z in zonelist zones */
1480
1481 zlc = zonelist->zlcache_ptr;
1482 if (!zlc)
1483 return;
1484
Mel Gormandd1a2392008-04-28 02:12:17 -07001485 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001486
1487 set_bit(i, zlc->fullzones);
1488}
1489
1490#else /* CONFIG_NUMA */
1491
1492static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1493{
1494 return NULL;
1495}
1496
Mel Gormandd1a2392008-04-28 02:12:17 -07001497static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001498 nodemask_t *allowednodes)
1499{
1500 return 1;
1501}
1502
Mel Gormandd1a2392008-04-28 02:12:17 -07001503static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001504{
1505}
1506#endif /* CONFIG_NUMA */
1507
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001508/*
Paul Jackson0798e512006-12-06 20:31:38 -08001509 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001510 * a page.
1511 */
1512static struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001513get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
Mel Gorman5117f452009-06-16 15:31:59 -07001514 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
Mel Gorman3dd28262009-06-16 15:32:00 -07001515 struct zone *preferred_zone, int migratetype)
Martin Hicks753ee722005-06-21 17:14:41 -07001516{
Mel Gormandd1a2392008-04-28 02:12:17 -07001517 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001518 struct page *page = NULL;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001519 int classzone_idx;
Mel Gorman5117f452009-06-16 15:31:59 -07001520 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001521 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1522 int zlc_active = 0; /* set if using zonelist_cache */
1523 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gorman54a6eb52008-04-28 02:12:16 -07001524
Mel Gorman19770b32008-04-28 02:12:18 -07001525 classzone_idx = zone_idx(preferred_zone);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001526zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001527 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001528 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001529 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1530 */
Mel Gorman19770b32008-04-28 02:12:18 -07001531 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1532 high_zoneidx, nodemask) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001533 if (NUMA_BUILD && zlc_active &&
1534 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1535 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001536 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001537 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001538 goto try_next_zone;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001539
Mel Gorman41858962009-06-16 15:32:12 -07001540 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001541 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001542 unsigned long mark;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001543 int ret;
1544
Mel Gorman41858962009-06-16 15:32:12 -07001545 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
Mel Gormanfa5e0842009-06-16 15:33:22 -07001546 if (zone_watermark_ok(zone, order, mark,
1547 classzone_idx, alloc_flags))
1548 goto try_this_zone;
1549
1550 if (zone_reclaim_mode == 0)
1551 goto this_zone_full;
1552
1553 ret = zone_reclaim(zone, gfp_mask, order);
1554 switch (ret) {
1555 case ZONE_RECLAIM_NOSCAN:
1556 /* did not scan */
1557 goto try_next_zone;
1558 case ZONE_RECLAIM_FULL:
1559 /* scanned but unreclaimable */
1560 goto this_zone_full;
1561 default:
1562 /* did we reclaim enough */
1563 if (!zone_watermark_ok(zone, order, mark,
1564 classzone_idx, alloc_flags))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001565 goto this_zone_full;
Paul Jackson0798e512006-12-06 20:31:38 -08001566 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001567 }
1568
Mel Gormanfa5e0842009-06-16 15:33:22 -07001569try_this_zone:
Mel Gorman3dd28262009-06-16 15:32:00 -07001570 page = buffered_rmqueue(preferred_zone, zone, order,
1571 gfp_mask, migratetype);
Paul Jackson0798e512006-12-06 20:31:38 -08001572 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001573 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001574this_zone_full:
1575 if (NUMA_BUILD)
1576 zlc_mark_zone_full(zonelist, z);
1577try_next_zone:
Christoph Lameter62bc62a2009-06-16 15:32:15 -07001578 if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) {
Mel Gormand395b732009-06-16 15:32:09 -07001579 /*
1580 * we do zlc_setup after the first zone is tried but only
1581 * if there are multiple nodes make it worthwhile
1582 */
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001583 allowednodes = zlc_setup(zonelist, alloc_flags);
1584 zlc_active = 1;
1585 did_zlc_setup = 1;
1586 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07001587 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001588
1589 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1590 /* Disable zlc cache for second zonelist scan */
1591 zlc_active = 0;
1592 goto zonelist_scan;
1593 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001594 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07001595}
1596
Mel Gorman11e33f62009-06-16 15:31:57 -07001597static inline int
1598should_alloc_retry(gfp_t gfp_mask, unsigned int order,
1599 unsigned long pages_reclaimed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600{
Mel Gorman11e33f62009-06-16 15:31:57 -07001601 /* Do not loop if specifically requested */
1602 if (gfp_mask & __GFP_NORETRY)
1603 return 0;
1604
1605 /*
1606 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1607 * means __GFP_NOFAIL, but that may not be true in other
1608 * implementations.
1609 */
1610 if (order <= PAGE_ALLOC_COSTLY_ORDER)
1611 return 1;
1612
1613 /*
1614 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1615 * specified, then we retry until we no longer reclaim any pages
1616 * (above), or we've reclaimed an order of pages at least as
1617 * large as the allocation's order. In both cases, if the
1618 * allocation still fails, we stop retrying.
1619 */
1620 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
1621 return 1;
1622
1623 /*
1624 * Don't let big-order allocations loop unless the caller
1625 * explicitly requests that.
1626 */
1627 if (gfp_mask & __GFP_NOFAIL)
1628 return 1;
1629
1630 return 0;
1631}
1632
1633static inline struct page *
1634__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1635 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001636 nodemask_t *nodemask, struct zone *preferred_zone,
1637 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001638{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640
Mel Gorman11e33f62009-06-16 15:31:57 -07001641 /* Acquire the OOM killer lock for the zones in zonelist */
1642 if (!try_set_zone_oom(zonelist, gfp_mask)) {
1643 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 return NULL;
1645 }
Jens Axboe6b1de912005-11-17 21:35:02 +01001646
Mel Gorman11e33f62009-06-16 15:31:57 -07001647 /*
1648 * Go through the zonelist yet one more time, keep very high watermark
1649 * here, this is only to catch a parallel oom killing, we must fail if
1650 * we're still under heavy pressure.
1651 */
1652 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1653 order, zonelist, high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001654 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07001655 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001656 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07001657 goto out;
1658
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001659 if (!(gfp_mask & __GFP_NOFAIL)) {
1660 /* The OOM killer will not help higher order allocs */
1661 if (order > PAGE_ALLOC_COSTLY_ORDER)
1662 goto out;
1663 /*
1664 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
1665 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
1666 * The caller should handle page allocation failure by itself if
1667 * it specifies __GFP_THISNODE.
1668 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
1669 */
1670 if (gfp_mask & __GFP_THISNODE)
1671 goto out;
1672 }
Mel Gorman11e33f62009-06-16 15:31:57 -07001673 /* Exhausted what can be done so it's blamo time */
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001674 out_of_memory(zonelist, gfp_mask, order, nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07001675
1676out:
1677 clear_zonelist_oom(zonelist, gfp_mask);
1678 return page;
1679}
1680
1681/* The really slow allocator path where we enter direct reclaim */
1682static inline struct page *
1683__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
1684 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001685 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001686 int migratetype, unsigned long *did_some_progress)
Mel Gorman11e33f62009-06-16 15:31:57 -07001687{
1688 struct page *page = NULL;
1689 struct reclaim_state reclaim_state;
1690 struct task_struct *p = current;
1691
1692 cond_resched();
1693
1694 /* We now go into synchronous reclaim */
1695 cpuset_memory_pressure_bump();
Mel Gorman11e33f62009-06-16 15:31:57 -07001696 p->flags |= PF_MEMALLOC;
1697 lockdep_set_current_reclaim_state(gfp_mask);
1698 reclaim_state.reclaimed_slab = 0;
1699 p->reclaim_state = &reclaim_state;
1700
1701 *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
1702
1703 p->reclaim_state = NULL;
1704 lockdep_clear_current_reclaim_state();
1705 p->flags &= ~PF_MEMALLOC;
1706
1707 cond_resched();
1708
1709 if (order != 0)
1710 drain_all_pages();
1711
1712 if (likely(*did_some_progress))
1713 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001714 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001715 alloc_flags, preferred_zone,
1716 migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001717 return page;
1718}
1719
Mel Gorman11e33f62009-06-16 15:31:57 -07001720/*
1721 * This is called in the allocator slow-path if the allocation request is of
1722 * sufficient urgency to ignore watermarks and take other desperate measures
1723 */
1724static inline struct page *
1725__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
1726 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001727 nodemask_t *nodemask, struct zone *preferred_zone,
1728 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001729{
1730 struct page *page;
1731
1732 do {
1733 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001734 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
Mel Gorman3dd28262009-06-16 15:32:00 -07001735 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001736
1737 if (!page && gfp_mask & __GFP_NOFAIL)
Jens Axboe8aa7e842009-07-09 14:52:32 +02001738 congestion_wait(BLK_RW_ASYNC, HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07001739 } while (!page && (gfp_mask & __GFP_NOFAIL));
1740
1741 return page;
1742}
1743
1744static inline
1745void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
1746 enum zone_type high_zoneidx)
1747{
1748 struct zoneref *z;
1749 struct zone *zone;
1750
1751 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
1752 wakeup_kswapd(zone, order);
1753}
1754
Peter Zijlstra341ce062009-06-16 15:32:02 -07001755static inline int
1756gfp_to_alloc_flags(gfp_t gfp_mask)
1757{
1758 struct task_struct *p = current;
1759 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
1760 const gfp_t wait = gfp_mask & __GFP_WAIT;
1761
Mel Gormana56f57f2009-06-16 15:32:02 -07001762 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
1763 BUILD_BUG_ON(__GFP_HIGH != ALLOC_HIGH);
1764
Peter Zijlstra341ce062009-06-16 15:32:02 -07001765 /*
1766 * The caller may dip into page reserves a bit more if the caller
1767 * cannot run direct reclaim, or if the caller has realtime scheduling
1768 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1769 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1770 */
Mel Gormana56f57f2009-06-16 15:32:02 -07001771 alloc_flags |= (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07001772
1773 if (!wait) {
1774 alloc_flags |= ALLOC_HARDER;
1775 /*
1776 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1777 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1778 */
1779 alloc_flags &= ~ALLOC_CPUSET;
Mel Gorman9d0ed602009-11-11 14:26:17 -08001780 } else if (unlikely(rt_task(p)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07001781 alloc_flags |= ALLOC_HARDER;
1782
1783 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
1784 if (!in_interrupt() &&
1785 ((p->flags & PF_MEMALLOC) ||
1786 unlikely(test_thread_flag(TIF_MEMDIE))))
1787 alloc_flags |= ALLOC_NO_WATERMARKS;
1788 }
1789
1790 return alloc_flags;
1791}
1792
Mel Gorman11e33f62009-06-16 15:31:57 -07001793static inline struct page *
1794__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
1795 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001796 nodemask_t *nodemask, struct zone *preferred_zone,
1797 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001798{
1799 const gfp_t wait = gfp_mask & __GFP_WAIT;
1800 struct page *page = NULL;
1801 int alloc_flags;
1802 unsigned long pages_reclaimed = 0;
1803 unsigned long did_some_progress;
1804 struct task_struct *p = current;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001805
Christoph Lameter952f3b52006-12-06 20:33:26 -08001806 /*
Mel Gorman72807a72009-06-16 15:32:18 -07001807 * In the slowpath, we sanity check order to avoid ever trying to
1808 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
1809 * be using allocators in order of preference for an area that is
1810 * too large.
1811 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07001812 if (order >= MAX_ORDER) {
1813 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07001814 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07001815 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001816
Christoph Lameter952f3b52006-12-06 20:33:26 -08001817 /*
1818 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1819 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1820 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1821 * using a larger set of nodes after it has established that the
1822 * allowed per node queues are empty and that nodes are
1823 * over allocated.
1824 */
1825 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1826 goto nopage;
1827
Mel Gormancc4a6852009-11-11 14:26:14 -08001828restart:
Mel Gorman11e33f62009-06-16 15:31:57 -07001829 wake_all_kswapd(order, zonelist, high_zoneidx);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001830
Paul Jackson9bf22292005-09-06 15:18:12 -07001831 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001832 * OK, we're below the kswapd watermark and have kicked background
1833 * reclaim. Now things get more complex, so set up alloc_flags according
1834 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07001835 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07001836 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837
Peter Zijlstra341ce062009-06-16 15:32:02 -07001838 /* This is the last chance, in general, before the goto nopage. */
Mel Gorman19770b32008-04-28 02:12:18 -07001839 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
Peter Zijlstra341ce062009-06-16 15:32:02 -07001840 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
1841 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001842 if (page)
1843 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844
Kirill Korotaevb43a57b2006-12-06 20:32:27 -08001845rebalance:
Mel Gorman11e33f62009-06-16 15:31:57 -07001846 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07001847 if (alloc_flags & ALLOC_NO_WATERMARKS) {
1848 page = __alloc_pages_high_priority(gfp_mask, order,
1849 zonelist, high_zoneidx, nodemask,
1850 preferred_zone, migratetype);
1851 if (page)
1852 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 }
1854
1855 /* Atomic allocations - we can't balance anything */
1856 if (!wait)
1857 goto nopage;
1858
Peter Zijlstra341ce062009-06-16 15:32:02 -07001859 /* Avoid recursion of direct reclaim */
1860 if (p->flags & PF_MEMALLOC)
1861 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862
David Rientjes6583bb62009-07-29 15:02:06 -07001863 /* Avoid allocations with no watermarks from looping endlessly */
1864 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
1865 goto nopage;
1866
Mel Gorman11e33f62009-06-16 15:31:57 -07001867 /* Try direct reclaim and then allocating */
1868 page = __alloc_pages_direct_reclaim(gfp_mask, order,
1869 zonelist, high_zoneidx,
1870 nodemask,
Mel Gorman5117f452009-06-16 15:31:59 -07001871 alloc_flags, preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001872 migratetype, &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07001873 if (page)
1874 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 /*
Mel Gorman11e33f62009-06-16 15:31:57 -07001877 * If we failed to make any progress reclaiming, then we are
1878 * running out of options and have to consider going OOM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 */
Mel Gorman11e33f62009-06-16 15:31:57 -07001880 if (!did_some_progress) {
1881 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -07001882 if (oom_killer_disabled)
1883 goto nopage;
Mel Gorman11e33f62009-06-16 15:31:57 -07001884 page = __alloc_pages_may_oom(gfp_mask, order,
1885 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001886 nodemask, preferred_zone,
1887 migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001888 if (page)
1889 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890
Mel Gorman11e33f62009-06-16 15:31:57 -07001891 /*
David Rientjes82553a92009-06-16 15:32:58 -07001892 * The OOM killer does not trigger for high-order
1893 * ~__GFP_NOFAIL allocations so if no progress is being
1894 * made, there are no other options and retrying is
1895 * unlikely to help.
Mel Gorman11e33f62009-06-16 15:31:57 -07001896 */
David Rientjes82553a92009-06-16 15:32:58 -07001897 if (order > PAGE_ALLOC_COSTLY_ORDER &&
1898 !(gfp_mask & __GFP_NOFAIL))
Mel Gorman11e33f62009-06-16 15:31:57 -07001899 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 goto restart;
1902 }
1903 }
1904
Mel Gorman11e33f62009-06-16 15:31:57 -07001905 /* Check if we should retry the allocation */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07001906 pages_reclaimed += did_some_progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07001907 if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
1908 /* Wait for some write requests to complete then retry */
Jens Axboe8aa7e842009-07-09 14:52:32 +02001909 congestion_wait(BLK_RW_ASYNC, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 goto rebalance;
1911 }
1912
1913nopage:
1914 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1915 printk(KERN_WARNING "%s: page allocation failure."
1916 " order:%d, mode:0x%x\n",
1917 p->comm, order, gfp_mask);
1918 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001919 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 }
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001921 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922got_pg:
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001923 if (kmemcheck_enabled)
1924 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 return page;
Mel Gorman11e33f62009-06-16 15:31:57 -07001926
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927}
Mel Gorman11e33f62009-06-16 15:31:57 -07001928
1929/*
1930 * This is the 'heart' of the zoned buddy allocator.
1931 */
1932struct page *
1933__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
1934 struct zonelist *zonelist, nodemask_t *nodemask)
1935{
1936 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Mel Gorman5117f452009-06-16 15:31:59 -07001937 struct zone *preferred_zone;
Mel Gorman11e33f62009-06-16 15:31:57 -07001938 struct page *page;
Mel Gorman3dd28262009-06-16 15:32:00 -07001939 int migratetype = allocflags_to_migratetype(gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07001940
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10001941 gfp_mask &= gfp_allowed_mask;
1942
Mel Gorman11e33f62009-06-16 15:31:57 -07001943 lockdep_trace_alloc(gfp_mask);
1944
1945 might_sleep_if(gfp_mask & __GFP_WAIT);
1946
1947 if (should_fail_alloc_page(gfp_mask, order))
1948 return NULL;
1949
1950 /*
1951 * Check the zones suitable for the gfp_mask contain at least one
1952 * valid zone. It's possible to have an empty zonelist as a result
1953 * of GFP_THISNODE and a memoryless node
1954 */
1955 if (unlikely(!zonelist->_zonerefs->zone))
1956 return NULL;
1957
Mel Gorman5117f452009-06-16 15:31:59 -07001958 /* The preferred zone is used for statistics later */
1959 first_zones_zonelist(zonelist, high_zoneidx, nodemask, &preferred_zone);
1960 if (!preferred_zone)
1961 return NULL;
1962
1963 /* First allocation attempt */
Mel Gorman11e33f62009-06-16 15:31:57 -07001964 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001965 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07001966 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001967 if (unlikely(!page))
1968 page = __alloc_pages_slowpath(gfp_mask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001969 zonelist, high_zoneidx, nodemask,
Mel Gorman3dd28262009-06-16 15:32:00 -07001970 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001971
Mel Gorman4b4f2782009-09-21 17:02:41 -07001972 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001973 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974}
Mel Gormand2391712009-06-16 15:31:52 -07001975EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976
1977/*
1978 * Common helper functions.
1979 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001980unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981{
Akinobu Mita945a1112009-09-21 17:01:47 -07001982 struct page *page;
1983
1984 /*
1985 * __get_free_pages() returns a 32-bit address, which cannot represent
1986 * a highmem page
1987 */
1988 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
1989
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 page = alloc_pages(gfp_mask, order);
1991 if (!page)
1992 return 0;
1993 return (unsigned long) page_address(page);
1994}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995EXPORT_SYMBOL(__get_free_pages);
1996
Harvey Harrison920c7a52008-02-04 22:29:26 -08001997unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998{
Akinobu Mita945a1112009-09-21 17:01:47 -07001999 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001EXPORT_SYMBOL(get_zeroed_page);
2002
2003void __pagevec_free(struct pagevec *pvec)
2004{
2005 int i = pagevec_count(pvec);
2006
Mel Gorman4b4f2782009-09-21 17:02:41 -07002007 while (--i >= 0) {
2008 trace_mm_pagevec_free(pvec->pages[i], pvec->cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 free_hot_cold_page(pvec->pages[i], pvec->cold);
Mel Gorman4b4f2782009-09-21 17:02:41 -07002010 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011}
2012
Harvey Harrison920c7a52008-02-04 22:29:26 -08002013void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014{
Nick Pigginb5810032005-10-29 18:16:12 -07002015 if (put_page_testzero(page)) {
Mel Gorman4b4f2782009-09-21 17:02:41 -07002016 trace_mm_page_free_direct(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 if (order == 0)
2018 free_hot_page(page);
2019 else
2020 __free_pages_ok(page, order);
2021 }
2022}
2023
2024EXPORT_SYMBOL(__free_pages);
2025
Harvey Harrison920c7a52008-02-04 22:29:26 -08002026void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027{
2028 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07002029 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 __free_pages(virt_to_page((void *)addr), order);
2031 }
2032}
2033
2034EXPORT_SYMBOL(free_pages);
2035
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002036/**
2037 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2038 * @size: the number of bytes to allocate
2039 * @gfp_mask: GFP flags for the allocation
2040 *
2041 * This function is similar to alloc_pages(), except that it allocates the
2042 * minimum number of pages to satisfy the request. alloc_pages() can only
2043 * allocate memory in power-of-two pages.
2044 *
2045 * This function is also limited by MAX_ORDER.
2046 *
2047 * Memory allocated by this function must be released by free_pages_exact().
2048 */
2049void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2050{
2051 unsigned int order = get_order(size);
2052 unsigned long addr;
2053
2054 addr = __get_free_pages(gfp_mask, order);
2055 if (addr) {
2056 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2057 unsigned long used = addr + PAGE_ALIGN(size);
2058
Kevin Cernekee5bfd7562009-07-05 12:08:19 -07002059 split_page(virt_to_page((void *)addr), order);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002060 while (used < alloc_end) {
2061 free_page(used);
2062 used += PAGE_SIZE;
2063 }
2064 }
2065
2066 return (void *)addr;
2067}
2068EXPORT_SYMBOL(alloc_pages_exact);
2069
2070/**
2071 * free_pages_exact - release memory allocated via alloc_pages_exact()
2072 * @virt: the value returned by alloc_pages_exact.
2073 * @size: size of allocation, same value as passed to alloc_pages_exact().
2074 *
2075 * Release the memory allocated by a previous call to alloc_pages_exact.
2076 */
2077void free_pages_exact(void *virt, size_t size)
2078{
2079 unsigned long addr = (unsigned long)virt;
2080 unsigned long end = addr + PAGE_ALIGN(size);
2081
2082 while (addr < end) {
2083 free_page(addr);
2084 addr += PAGE_SIZE;
2085 }
2086}
2087EXPORT_SYMBOL(free_pages_exact);
2088
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089static unsigned int nr_free_zone_pages(int offset)
2090{
Mel Gormandd1a2392008-04-28 02:12:17 -07002091 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002092 struct zone *zone;
2093
Martin J. Blighe310fd42005-07-29 22:59:18 -07002094 /* Just pick one node, since fallback list is circular */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 unsigned int sum = 0;
2096
Mel Gorman0e884602008-04-28 02:12:14 -07002097 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098
Mel Gorman54a6eb52008-04-28 02:12:16 -07002099 for_each_zone_zonelist(zone, z, zonelist, offset) {
Martin J. Blighe310fd42005-07-29 22:59:18 -07002100 unsigned long size = zone->present_pages;
Mel Gorman41858962009-06-16 15:32:12 -07002101 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07002102 if (size > high)
2103 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104 }
2105
2106 return sum;
2107}
2108
2109/*
2110 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2111 */
2112unsigned int nr_free_buffer_pages(void)
2113{
Al Viroaf4ca452005-10-21 02:55:38 -04002114 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115}
Meelap Shahc2f1a552007-07-17 04:04:39 -07002116EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117
2118/*
2119 * Amount of free RAM allocatable within all zones
2120 */
2121unsigned int nr_free_pagecache_pages(void)
2122{
Mel Gorman2a1e2742007-07-17 04:03:12 -07002123 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002125
2126static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127{
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002128 if (NUMA_BUILD)
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08002129 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132void si_meminfo(struct sysinfo *val)
2133{
2134 val->totalram = totalram_pages;
2135 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002136 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 val->totalhigh = totalhigh_pages;
2139 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 val->mem_unit = PAGE_SIZE;
2141}
2142
2143EXPORT_SYMBOL(si_meminfo);
2144
2145#ifdef CONFIG_NUMA
2146void si_meminfo_node(struct sysinfo *val, int nid)
2147{
2148 pg_data_t *pgdat = NODE_DATA(nid);
2149
2150 val->totalram = pgdat->node_present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002151 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002152#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002154 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2155 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002156#else
2157 val->totalhigh = 0;
2158 val->freehigh = 0;
2159#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 val->mem_unit = PAGE_SIZE;
2161}
2162#endif
2163
2164#define K(x) ((x) << (PAGE_SHIFT-10))
2165
2166/*
2167 * Show free area list (used inside shift_scroll-lock stuff)
2168 * We also calculate the percentage fragmentation. We do this by counting the
2169 * memory on each free list with the exception of the first item on the list.
2170 */
2171void show_free_areas(void)
2172{
Jes Sorensenc7241912006-09-27 01:50:05 -07002173 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 struct zone *zone;
2175
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002176 for_each_populated_zone(zone) {
Jes Sorensenc7241912006-09-27 01:50:05 -07002177 show_node(zone);
2178 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179
Dave Jones6b482c62005-11-10 15:45:56 -05002180 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181 struct per_cpu_pageset *pageset;
2182
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002183 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002185 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2186 cpu, pageset->pcp.high,
2187 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 }
2189 }
2190
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002191 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2192 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002193 " unevictable:%lu"
Andrew Mortonb76146e2009-10-26 16:49:52 -07002194 " dirty:%lu writeback:%lu unstable:%lu\n"
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07002195 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002196 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002197 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002198 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002199 global_page_state(NR_ISOLATED_ANON),
2200 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002201 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002202 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002203 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07002204 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07002205 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07002206 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08002207 global_page_state(NR_FREE_PAGES),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07002208 global_page_state(NR_SLAB_RECLAIMABLE),
2209 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07002210 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002211 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08002212 global_page_state(NR_PAGETABLE),
2213 global_page_state(NR_BOUNCE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002215 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216 int i;
2217
2218 show_node(zone);
2219 printk("%s"
2220 " free:%lukB"
2221 " min:%lukB"
2222 " low:%lukB"
2223 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002224 " active_anon:%lukB"
2225 " inactive_anon:%lukB"
2226 " active_file:%lukB"
2227 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002228 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002229 " isolated(anon):%lukB"
2230 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231 " present:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002232 " mlocked:%lukB"
2233 " dirty:%lukB"
2234 " writeback:%lukB"
2235 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002236 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002237 " slab_reclaimable:%lukB"
2238 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07002239 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002240 " pagetables:%lukB"
2241 " unstable:%lukB"
2242 " bounce:%lukB"
2243 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 " pages_scanned:%lu"
2245 " all_unreclaimable? %s"
2246 "\n",
2247 zone->name,
Christoph Lameterd23ad422007-02-10 01:43:02 -08002248 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07002249 K(min_wmark_pages(zone)),
2250 K(low_wmark_pages(zone)),
2251 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002252 K(zone_page_state(zone, NR_ACTIVE_ANON)),
2253 K(zone_page_state(zone, NR_INACTIVE_ANON)),
2254 K(zone_page_state(zone, NR_ACTIVE_FILE)),
2255 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002256 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002257 K(zone_page_state(zone, NR_ISOLATED_ANON)),
2258 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259 K(zone->present_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002260 K(zone_page_state(zone, NR_MLOCK)),
2261 K(zone_page_state(zone, NR_FILE_DIRTY)),
2262 K(zone_page_state(zone, NR_WRITEBACK)),
2263 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002264 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002265 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
2266 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07002267 zone_page_state(zone, NR_KERNEL_STACK) *
2268 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002269 K(zone_page_state(zone, NR_PAGETABLE)),
2270 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
2271 K(zone_page_state(zone, NR_BOUNCE)),
2272 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273 zone->pages_scanned,
David Rientjese815af92007-10-16 23:25:54 -07002274 (zone_is_all_unreclaimable(zone) ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 );
2276 printk("lowmem_reserve[]:");
2277 for (i = 0; i < MAX_NR_ZONES; i++)
2278 printk(" %lu", zone->lowmem_reserve[i]);
2279 printk("\n");
2280 }
2281
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002282 for_each_populated_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002283 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284
2285 show_node(zone);
2286 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287
2288 spin_lock_irqsave(&zone->lock, flags);
2289 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002290 nr[order] = zone->free_area[order].nr_free;
2291 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 }
2293 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002294 for (order = 0; order < MAX_ORDER; order++)
2295 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296 printk("= %lukB\n", K(total));
2297 }
2298
Larry Woodmane6f36022008-02-04 22:29:30 -08002299 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2300
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301 show_swap_cache_info();
2302}
2303
Mel Gorman19770b32008-04-28 02:12:18 -07002304static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2305{
2306 zoneref->zone = zone;
2307 zoneref->zone_idx = zone_idx(zone);
2308}
2309
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310/*
2311 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08002312 *
2313 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002315static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2316 int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317{
Christoph Lameter1a932052006-01-06 00:11:16 -08002318 struct zone *zone;
2319
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002320 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002321 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08002322
2323 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002324 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08002325 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08002326 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002327 zoneref_set_zone(zone,
2328 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08002329 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08002331
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002332 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08002333 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334}
2335
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002336
2337/*
2338 * zonelist_order:
2339 * 0 = automatic detection of better ordering.
2340 * 1 = order by ([node] distance, -zonetype)
2341 * 2 = order by (-zonetype, [node] distance)
2342 *
2343 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2344 * the same zonelist. So only NUMA can configure this param.
2345 */
2346#define ZONELIST_ORDER_DEFAULT 0
2347#define ZONELIST_ORDER_NODE 1
2348#define ZONELIST_ORDER_ZONE 2
2349
2350/* zonelist order in the kernel.
2351 * set_zonelist_order() will set this to NODE or ZONE.
2352 */
2353static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2354static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2355
2356
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002358/* The value user specified ....changed by config */
2359static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2360/* string for sysctl */
2361#define NUMA_ZONELIST_ORDER_LEN 16
2362char numa_zonelist_order[16] = "default";
2363
2364/*
2365 * interface for configure zonelist ordering.
2366 * command line option "numa_zonelist_order"
2367 * = "[dD]efault - default, automatic configuration.
2368 * = "[nN]ode - order by node locality, then by zone within node
2369 * = "[zZ]one - order by zone, then by locality within zone
2370 */
2371
2372static int __parse_numa_zonelist_order(char *s)
2373{
2374 if (*s == 'd' || *s == 'D') {
2375 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2376 } else if (*s == 'n' || *s == 'N') {
2377 user_zonelist_order = ZONELIST_ORDER_NODE;
2378 } else if (*s == 'z' || *s == 'Z') {
2379 user_zonelist_order = ZONELIST_ORDER_ZONE;
2380 } else {
2381 printk(KERN_WARNING
2382 "Ignoring invalid numa_zonelist_order value: "
2383 "%s\n", s);
2384 return -EINVAL;
2385 }
2386 return 0;
2387}
2388
2389static __init int setup_numa_zonelist_order(char *s)
2390{
2391 if (s)
2392 return __parse_numa_zonelist_order(s);
2393 return 0;
2394}
2395early_param("numa_zonelist_order", setup_numa_zonelist_order);
2396
2397/*
2398 * sysctl handler for numa_zonelist_order
2399 */
2400int numa_zonelist_order_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07002401 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002402 loff_t *ppos)
2403{
2404 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2405 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01002406 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002407
Andi Kleen443c6f12009-12-23 21:00:47 +01002408 mutex_lock(&zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002409 if (write)
Andi Kleen443c6f12009-12-23 21:00:47 +01002410 strcpy(saved_string, (char*)table->data);
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07002411 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002412 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01002413 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002414 if (write) {
2415 int oldval = user_zonelist_order;
2416 if (__parse_numa_zonelist_order((char*)table->data)) {
2417 /*
2418 * bogus value. restore saved string
2419 */
2420 strncpy((char*)table->data, saved_string,
2421 NUMA_ZONELIST_ORDER_LEN);
2422 user_zonelist_order = oldval;
2423 } else if (oldval != user_zonelist_order)
2424 build_all_zonelists();
2425 }
Andi Kleen443c6f12009-12-23 21:00:47 +01002426out:
2427 mutex_unlock(&zl_order_mutex);
2428 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002429}
2430
2431
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002432#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002433static int node_load[MAX_NUMNODES];
2434
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07002436 * 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 -07002437 * @node: node whose fallback list we're appending
2438 * @used_node_mask: nodemask_t of already used nodes
2439 *
2440 * We use a number of factors to determine which is the next node that should
2441 * appear on a given node's fallback list. The node should not have appeared
2442 * already in @node's fallback list, and it should be the next closest node
2443 * according to the distance array (which contains arbitrary distance values
2444 * from each node to each node in the system), and should also prefer nodes
2445 * with no CPUs, since presumably they'll have very little allocation pressure
2446 * on them otherwise.
2447 * It returns -1 if no node is found.
2448 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002449static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002451 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452 int min_val = INT_MAX;
2453 int best_node = -1;
Rusty Russella70f7302009-03-13 14:49:46 +10302454 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002456 /* Use the local node if we haven't already */
2457 if (!node_isset(node, *used_node_mask)) {
2458 node_set(node, *used_node_mask);
2459 return node;
2460 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002462 for_each_node_state(n, N_HIGH_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463
2464 /* Don't want a node to appear more than once */
2465 if (node_isset(n, *used_node_mask))
2466 continue;
2467
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468 /* Use the distance array to find the distance */
2469 val = node_distance(node, n);
2470
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002471 /* Penalize nodes under us ("prefer the next node") */
2472 val += (n < node);
2473
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10302475 tmp = cpumask_of_node(n);
2476 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477 val += PENALTY_FOR_NODE_WITH_CPUS;
2478
2479 /* Slight preference for less loaded node */
2480 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2481 val += node_load[n];
2482
2483 if (val < min_val) {
2484 min_val = val;
2485 best_node = n;
2486 }
2487 }
2488
2489 if (best_node >= 0)
2490 node_set(best_node, *used_node_mask);
2491
2492 return best_node;
2493}
2494
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002495
2496/*
2497 * Build zonelists ordered by node and zones within node.
2498 * This results in maximum locality--normal zone overflows into local
2499 * DMA zone, if any--but risks exhausting DMA zone.
2500 */
2501static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002503 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002505
Mel Gorman54a6eb52008-04-28 02:12:16 -07002506 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07002507 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07002508 ;
2509 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2510 MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002511 zonelist->_zonerefs[j].zone = NULL;
2512 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002513}
2514
2515/*
Christoph Lameter523b9452007-10-16 01:25:37 -07002516 * Build gfp_thisnode zonelists
2517 */
2518static void build_thisnode_zonelists(pg_data_t *pgdat)
2519{
Christoph Lameter523b9452007-10-16 01:25:37 -07002520 int j;
2521 struct zonelist *zonelist;
2522
Mel Gorman54a6eb52008-04-28 02:12:16 -07002523 zonelist = &pgdat->node_zonelists[1];
2524 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002525 zonelist->_zonerefs[j].zone = NULL;
2526 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07002527}
2528
2529/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002530 * Build zonelists ordered by zone and nodes within zones.
2531 * This results in conserving DMA zone[s] until all Normal memory is
2532 * exhausted, but results in overflowing to remote node while memory
2533 * may still exist in local DMA zone.
2534 */
2535static int node_order[MAX_NUMNODES];
2536
2537static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2538{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002539 int pos, j, node;
2540 int zone_type; /* needs to be signed */
2541 struct zone *z;
2542 struct zonelist *zonelist;
2543
Mel Gorman54a6eb52008-04-28 02:12:16 -07002544 zonelist = &pgdat->node_zonelists[0];
2545 pos = 0;
2546 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2547 for (j = 0; j < nr_nodes; j++) {
2548 node = node_order[j];
2549 z = &NODE_DATA(node)->node_zones[zone_type];
2550 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002551 zoneref_set_zone(z,
2552 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002553 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002554 }
2555 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002556 }
Mel Gormandd1a2392008-04-28 02:12:17 -07002557 zonelist->_zonerefs[pos].zone = NULL;
2558 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002559}
2560
2561static int default_zonelist_order(void)
2562{
2563 int nid, zone_type;
2564 unsigned long low_kmem_size,total_size;
2565 struct zone *z;
2566 int average_size;
2567 /*
2568 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2569 * If they are really small and used heavily, the system can fall
2570 * into OOM very easily.
2571 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2572 */
2573 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2574 low_kmem_size = 0;
2575 total_size = 0;
2576 for_each_online_node(nid) {
2577 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2578 z = &NODE_DATA(nid)->node_zones[zone_type];
2579 if (populated_zone(z)) {
2580 if (zone_type < ZONE_NORMAL)
2581 low_kmem_size += z->present_pages;
2582 total_size += z->present_pages;
2583 }
2584 }
2585 }
2586 if (!low_kmem_size || /* there are no DMA area. */
2587 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2588 return ZONELIST_ORDER_NODE;
2589 /*
2590 * look into each node's config.
2591 * If there is a node whose DMA/DMA32 memory is very big area on
2592 * local memory, NODE_ORDER may be suitable.
2593 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002594 average_size = total_size /
2595 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002596 for_each_online_node(nid) {
2597 low_kmem_size = 0;
2598 total_size = 0;
2599 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2600 z = &NODE_DATA(nid)->node_zones[zone_type];
2601 if (populated_zone(z)) {
2602 if (zone_type < ZONE_NORMAL)
2603 low_kmem_size += z->present_pages;
2604 total_size += z->present_pages;
2605 }
2606 }
2607 if (low_kmem_size &&
2608 total_size > average_size && /* ignore small node */
2609 low_kmem_size > total_size * 70/100)
2610 return ZONELIST_ORDER_NODE;
2611 }
2612 return ZONELIST_ORDER_ZONE;
2613}
2614
2615static void set_zonelist_order(void)
2616{
2617 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2618 current_zonelist_order = default_zonelist_order();
2619 else
2620 current_zonelist_order = user_zonelist_order;
2621}
2622
2623static void build_zonelists(pg_data_t *pgdat)
2624{
2625 int j, node, load;
2626 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002628 int local_node, prev_node;
2629 struct zonelist *zonelist;
2630 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631
2632 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07002633 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07002635 zonelist->_zonerefs[0].zone = NULL;
2636 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637 }
2638
2639 /* NUMA-aware ordering of nodes */
2640 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002641 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642 prev_node = local_node;
2643 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002644
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002645 memset(node_order, 0, sizeof(node_order));
2646 j = 0;
2647
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08002649 int distance = node_distance(local_node, node);
2650
2651 /*
2652 * If another node is sufficiently far away then it is better
2653 * to reclaim pages in a zone before going off node.
2654 */
2655 if (distance > RECLAIM_DISTANCE)
2656 zone_reclaim_mode = 1;
2657
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658 /*
2659 * We don't want to pressure a particular node.
2660 * So adding penalty to the first node in same
2661 * distance group to make it round-robin.
2662 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08002663 if (distance != node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002664 node_load[node] = load;
2665
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 prev_node = node;
2667 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002668 if (order == ZONELIST_ORDER_NODE)
2669 build_zonelists_in_node_order(pgdat, node);
2670 else
2671 node_order[j++] = node; /* remember order */
2672 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002674 if (order == ZONELIST_ORDER_ZONE) {
2675 /* calculate node order -- i.e., DMA last! */
2676 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 }
Christoph Lameter523b9452007-10-16 01:25:37 -07002678
2679 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680}
2681
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002682/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002683static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002684{
Mel Gorman54a6eb52008-04-28 02:12:16 -07002685 struct zonelist *zonelist;
2686 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07002687 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002688
Mel Gorman54a6eb52008-04-28 02:12:16 -07002689 zonelist = &pgdat->node_zonelists[0];
2690 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2691 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07002692 for (z = zonelist->_zonerefs; z->zone; z++)
2693 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002694}
2695
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002696
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697#else /* CONFIG_NUMA */
2698
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002699static void set_zonelist_order(void)
2700{
2701 current_zonelist_order = ZONELIST_ORDER_ZONE;
2702}
2703
2704static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705{
Christoph Lameter19655d32006-09-25 23:31:19 -07002706 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002707 enum zone_type j;
2708 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709
2710 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711
Mel Gorman54a6eb52008-04-28 02:12:16 -07002712 zonelist = &pgdat->node_zonelists[0];
2713 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714
Mel Gorman54a6eb52008-04-28 02:12:16 -07002715 /*
2716 * Now we build the zonelist so that it contains the zones
2717 * of all the other nodes.
2718 * We don't want to pressure a particular node, so when
2719 * building the zones for node N, we make sure that the
2720 * zones coming right after the local ones are those from
2721 * node N+1 (modulo N)
2722 */
2723 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2724 if (!node_online(node))
2725 continue;
2726 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2727 MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07002729 for (node = 0; node < local_node; node++) {
2730 if (!node_online(node))
2731 continue;
2732 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2733 MAX_NR_ZONES - 1);
2734 }
2735
Mel Gormandd1a2392008-04-28 02:12:17 -07002736 zonelist->_zonerefs[j].zone = NULL;
2737 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002738}
2739
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002740/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002741static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002742{
Mel Gorman54a6eb52008-04-28 02:12:16 -07002743 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002744}
2745
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746#endif /* CONFIG_NUMA */
2747
Rusty Russell9b1a4d32008-07-28 12:16:30 -05002748/* return values int ....just for stop_machine() */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002749static int __build_all_zonelists(void *dummy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750{
Yasunori Goto68113782006-06-23 02:03:11 -07002751 int nid;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002752
Bo Liu7f9cfb32009-08-18 14:11:19 -07002753#ifdef CONFIG_NUMA
2754 memset(node_load, 0, sizeof(node_load));
2755#endif
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002756 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07002757 pg_data_t *pgdat = NODE_DATA(nid);
2758
2759 build_zonelists(pgdat);
2760 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002761 }
Yasunori Goto68113782006-06-23 02:03:11 -07002762 return 0;
2763}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002764
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002765void build_all_zonelists(void)
Yasunori Goto68113782006-06-23 02:03:11 -07002766{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002767 set_zonelist_order();
2768
Yasunori Goto68113782006-06-23 02:03:11 -07002769 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07002770 __build_all_zonelists(NULL);
Mel Gorman68ad8df2008-07-23 21:26:52 -07002771 mminit_verify_zonelist();
Yasunori Goto68113782006-06-23 02:03:11 -07002772 cpuset_init_current_mems_allowed();
2773 } else {
Simon Arlott183ff222007-10-20 01:27:18 +02002774 /* we have to stop all cpus to guarantee there is no user
Yasunori Goto68113782006-06-23 02:03:11 -07002775 of zonelist */
Rusty Russell9b1a4d32008-07-28 12:16:30 -05002776 stop_machine(__build_all_zonelists, NULL, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07002777 /* cpuset refresh routine should be here */
2778 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07002779 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002780 /*
2781 * Disable grouping by mobility if the number of pages in the
2782 * system is too low to allow the mechanism to work. It would be
2783 * more accurate, but expensive to check per-zone. This check is
2784 * made on memory-hotadd so a system can start with mobility
2785 * disabled and enable it later
2786 */
Mel Gormand9c23402007-10-16 01:26:01 -07002787 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002788 page_group_by_mobility_disabled = 1;
2789 else
2790 page_group_by_mobility_disabled = 0;
2791
2792 printk("Built %i zonelists in %s order, mobility grouping %s. "
2793 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002794 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002795 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002796 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002797 vm_total_pages);
2798#ifdef CONFIG_NUMA
2799 printk("Policy zone: %s\n", zone_names[policy_zone]);
2800#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801}
2802
2803/*
2804 * Helper functions to size the waitqueue hash table.
2805 * Essentially these want to choose hash table sizes sufficiently
2806 * large so that collisions trying to wait on pages are rare.
2807 * But in fact, the number of active page waitqueues on typical
2808 * systems is ridiculously low, less than 200. So this is even
2809 * conservative, even though it seems large.
2810 *
2811 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2812 * waitqueues, i.e. the size of the waitq table given the number of pages.
2813 */
2814#define PAGES_PER_WAITQUEUE 256
2815
Yasunori Gotocca448f2006-06-23 02:03:10 -07002816#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07002817static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818{
2819 unsigned long size = 1;
2820
2821 pages /= PAGES_PER_WAITQUEUE;
2822
2823 while (size < pages)
2824 size <<= 1;
2825
2826 /*
2827 * Once we have dozens or even hundreds of threads sleeping
2828 * on IO we've got bigger problems than wait queue collision.
2829 * Limit the size of the wait table to a reasonable size.
2830 */
2831 size = min(size, 4096UL);
2832
2833 return max(size, 4UL);
2834}
Yasunori Gotocca448f2006-06-23 02:03:10 -07002835#else
2836/*
2837 * A zone's size might be changed by hot-add, so it is not possible to determine
2838 * a suitable size for its wait_table. So we use the maximum size now.
2839 *
2840 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2841 *
2842 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2843 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2844 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2845 *
2846 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2847 * or more by the traditional way. (See above). It equals:
2848 *
2849 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2850 * ia64(16K page size) : = ( 8G + 4M)byte.
2851 * powerpc (64K page size) : = (32G +16M)byte.
2852 */
2853static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2854{
2855 return 4096UL;
2856}
2857#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858
2859/*
2860 * This is an integer logarithm so that shifts can be used later
2861 * to extract the more random high bits from the multiplicative
2862 * hash function before the remainder is taken.
2863 */
2864static inline unsigned long wait_table_bits(unsigned long size)
2865{
2866 return ffz(~size);
2867}
2868
2869#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2870
Mel Gorman56fd56b2007-10-16 01:25:58 -07002871/*
Mel Gormand9c23402007-10-16 01:26:01 -07002872 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07002873 * of blocks reserved is based on min_wmark_pages(zone). The memory within
2874 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07002875 * higher will lead to a bigger reserve which will get freed as contiguous
2876 * blocks as reclaim kicks in
2877 */
2878static void setup_zone_migrate_reserve(struct zone *zone)
2879{
2880 unsigned long start_pfn, pfn, end_pfn;
2881 struct page *page;
Mel Gorman78986a62009-09-21 17:03:02 -07002882 unsigned long block_migratetype;
2883 int reserve;
Mel Gorman56fd56b2007-10-16 01:25:58 -07002884
2885 /* Get the start pfn, end pfn and the number of blocks to reserve */
2886 start_pfn = zone->zone_start_pfn;
2887 end_pfn = start_pfn + zone->spanned_pages;
Mel Gorman41858962009-06-16 15:32:12 -07002888 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07002889 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07002890
Mel Gorman78986a62009-09-21 17:03:02 -07002891 /*
2892 * Reserve blocks are generally in place to help high-order atomic
2893 * allocations that are short-lived. A min_free_kbytes value that
2894 * would result in more than 2 reserve blocks for atomic allocations
2895 * is assumed to be in place to help anti-fragmentation for the
2896 * future allocation of hugepages at runtime.
2897 */
2898 reserve = min(2, reserve);
2899
Mel Gormand9c23402007-10-16 01:26:01 -07002900 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07002901 if (!pfn_valid(pfn))
2902 continue;
2903 page = pfn_to_page(pfn);
2904
Adam Litke344c7902008-09-02 14:35:38 -07002905 /* Watch out for overlapping nodes */
2906 if (page_to_nid(page) != zone_to_nid(zone))
2907 continue;
2908
Mel Gorman56fd56b2007-10-16 01:25:58 -07002909 /* Blocks with reserved pages will never free, skip them. */
2910 if (PageReserved(page))
2911 continue;
2912
2913 block_migratetype = get_pageblock_migratetype(page);
2914
2915 /* If this block is reserved, account for it */
2916 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
2917 reserve--;
2918 continue;
2919 }
2920
2921 /* Suitable for reserving if this block is movable */
2922 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
2923 set_pageblock_migratetype(page, MIGRATE_RESERVE);
2924 move_freepages_block(zone, page, MIGRATE_RESERVE);
2925 reserve--;
2926 continue;
2927 }
2928
2929 /*
2930 * If the reserve is met and this is a previous reserved block,
2931 * take it back
2932 */
2933 if (block_migratetype == MIGRATE_RESERVE) {
2934 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2935 move_freepages_block(zone, page, MIGRATE_MOVABLE);
2936 }
2937 }
2938}
Mel Gormanac0e5b72007-10-16 01:25:58 -07002939
Linus Torvalds1da177e2005-04-16 15:20:36 -07002940/*
2941 * Initially all pages are reserved - free ones are freed
2942 * up by free_all_bootmem() once the early boot process is
2943 * done. Non-atomic initialization, single-pass.
2944 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002945void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa02007-01-10 23:15:30 -08002946 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002947{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002948 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07002949 unsigned long end_pfn = start_pfn + size;
2950 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002951 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002952
Hugh Dickins22b31ee2009-01-06 14:40:09 -08002953 if (highest_memmap_pfn < end_pfn - 1)
2954 highest_memmap_pfn = end_pfn - 1;
2955
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002956 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08002957 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa02007-01-10 23:15:30 -08002958 /*
2959 * There can be holes in boot-time mem_map[]s
2960 * handed to this function. They do not
2961 * exist on hotplugged memory.
2962 */
2963 if (context == MEMMAP_EARLY) {
2964 if (!early_pfn_valid(pfn))
2965 continue;
2966 if (!early_pfn_in_nid(pfn, nid))
2967 continue;
2968 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002969 page = pfn_to_page(pfn);
2970 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07002971 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08002972 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002973 reset_page_mapcount(page);
2974 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002975 /*
2976 * Mark the block movable so that blocks are reserved for
2977 * movable at startup. This will force kernel allocations
2978 * to reserve their blocks rather than leaking throughout
2979 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07002980 * kernel allocations are made. Later some blocks near
2981 * the start are marked MIGRATE_RESERVE by
2982 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002983 *
2984 * bitmap is created for zone's valid pfn range. but memmap
2985 * can be created for invalid pages (for alignment)
2986 * check here not to call set_pageblock_migratetype() against
2987 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002988 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002989 if ((z->zone_start_pfn <= pfn)
2990 && (pfn < z->zone_start_pfn + z->spanned_pages)
2991 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07002992 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002993
Linus Torvalds1da177e2005-04-16 15:20:36 -07002994 INIT_LIST_HEAD(&page->lru);
2995#ifdef WANT_PAGE_VIRTUAL
2996 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2997 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07002998 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002999#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003000 }
3001}
3002
Andi Kleen1e548de2008-02-04 22:29:26 -08003003static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004{
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003005 int order, t;
3006 for_each_migratetype_order(order, t) {
3007 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003008 zone->free_area[order].nr_free = 0;
3009 }
3010}
3011
3012#ifndef __HAVE_ARCH_MEMMAP_INIT
3013#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa02007-01-10 23:15:30 -08003014 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003015#endif
3016
Sam Ravnborg1d6f4e62008-01-17 15:21:12 -08003017static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003018{
David Howells3a6be872009-05-06 16:03:03 -07003019#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003020 int batch;
3021
3022 /*
3023 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07003024 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003025 *
3026 * OK, so we don't know how big the cache is. So guess.
3027 */
3028 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07003029 if (batch * PAGE_SIZE > 512 * 1024)
3030 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003031 batch /= 4; /* We effectively *= 4 below */
3032 if (batch < 1)
3033 batch = 1;
3034
3035 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11003036 * Clamp the batch to a 2^n - 1 value. Having a power
3037 * of 2 value was found to be more likely to have
3038 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003039 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11003040 * For example if 2 tasks are alternately allocating
3041 * batches of pages, one task can end up with a lot
3042 * of pages of one half of the possible page colors
3043 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003044 */
David Howells91552032009-05-06 16:03:02 -07003045 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07003046
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003047 return batch;
David Howells3a6be872009-05-06 16:03:03 -07003048
3049#else
3050 /* The deferral and batching of frees should be suppressed under NOMMU
3051 * conditions.
3052 *
3053 * The problem is that NOMMU needs to be able to allocate large chunks
3054 * of contiguous memory as there's no hardware page translation to
3055 * assemble apparent contiguous memory from discontiguous pages.
3056 *
3057 * Queueing large contiguous runs of pages for batching, however,
3058 * causes the pages to actually be freed in smaller chunks. As there
3059 * can be a significant delay between the individual batches being
3060 * recycled, this leads to the once large chunks of space being
3061 * fragmented and becoming unavailable for high-order allocations.
3062 */
3063 return 0;
3064#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003065}
3066
Adrian Bunkb69a7282008-07-23 21:28:12 -07003067static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
Christoph Lameter2caaad42005-06-21 17:15:00 -07003068{
3069 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003070 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003071
Magnus Damm1c6fe942005-10-26 01:58:59 -07003072 memset(p, 0, sizeof(*p));
3073
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003074 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003075 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003076 pcp->high = 6 * batch;
3077 pcp->batch = max(1UL, 1 * batch);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003078 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
3079 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07003080}
3081
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003082/*
3083 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3084 * to the value high for the pageset p.
3085 */
3086
3087static void setup_pagelist_highmark(struct per_cpu_pageset *p,
3088 unsigned long high)
3089{
3090 struct per_cpu_pages *pcp;
3091
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003092 pcp = &p->pcp;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003093 pcp->high = high;
3094 pcp->batch = max(1UL, high/4);
3095 if ((high/4) > (PAGE_SHIFT * 8))
3096 pcp->batch = PAGE_SHIFT * 8;
3097}
3098
3099
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003100#ifdef CONFIG_NUMA
3101/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07003102 * Boot pageset table. One per cpu which is going to be used for all
3103 * zones and all nodes. The parameters will be set in such a way
3104 * that an item put on a list will immediately be handed over to
3105 * the buddy list. This is safe since pageset manipulation is done
3106 * with interrupts disabled.
3107 *
3108 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07003109 *
3110 * The boot_pagesets must be kept even after bootup is complete for
3111 * unused processors and/or zones. They do play a role for bootstrapping
3112 * hotplugged processors.
3113 *
3114 * zoneinfo_show() and maybe other functions do
3115 * not check if the processor is online before following the pageset pointer.
3116 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07003117 */
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08003118static struct per_cpu_pageset boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07003119
3120/*
3121 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003122 * per cpu pageset array in struct zone.
3123 */
Ashok Raj6292d9a2006-02-01 03:04:44 -08003124static int __cpuinit process_zones(int cpu)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003125{
3126 struct zone *zone, *dzone;
Christoph Lameter37c07082007-10-16 01:25:36 -07003127 int node = cpu_to_node(cpu);
3128
3129 node_set_state(node, N_CPU); /* this node has a cpu */
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003130
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003131 for_each_populated_zone(zone) {
Nick Piggin23316bc2006-01-08 01:00:41 -08003132 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lameter37c07082007-10-16 01:25:36 -07003133 GFP_KERNEL, node);
Nick Piggin23316bc2006-01-08 01:00:41 -08003134 if (!zone_pcp(zone, cpu))
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003135 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003136
Nick Piggin23316bc2006-01-08 01:00:41 -08003137 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003138
3139 if (percpu_pagelist_fraction)
3140 setup_pagelist_highmark(zone_pcp(zone, cpu),
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08003141 (zone->present_pages / percpu_pagelist_fraction));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003142 }
3143
3144 return 0;
3145bad:
3146 for_each_zone(dzone) {
Andrew Morton64191682007-08-30 23:56:17 -07003147 if (!populated_zone(dzone))
3148 continue;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003149 if (dzone == zone)
3150 break;
Nick Piggin23316bc2006-01-08 01:00:41 -08003151 kfree(zone_pcp(dzone, cpu));
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07003152 zone_pcp(dzone, cpu) = &boot_pageset[cpu];
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003153 }
3154 return -ENOMEM;
3155}
3156
3157static inline void free_zone_pagesets(int cpu)
3158{
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003159 struct zone *zone;
3160
3161 for_each_zone(zone) {
3162 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
3163
David Rientjesf3ef9ea2006-09-25 16:24:57 -07003164 /* Free per_cpu_pageset if it is slab allocated */
3165 if (pset != &boot_pageset[cpu])
3166 kfree(pset);
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07003167 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003168 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003169}
3170
Chandra Seetharaman9c7b2162006-06-27 02:54:07 -07003171static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003172 unsigned long action,
3173 void *hcpu)
3174{
3175 int cpu = (long)hcpu;
3176 int ret = NOTIFY_OK;
3177
3178 switch (action) {
Andy Whitcroftce421c72006-12-06 20:33:08 -08003179 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003180 case CPU_UP_PREPARE_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08003181 if (process_zones(cpu))
3182 ret = NOTIFY_BAD;
3183 break;
3184 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003185 case CPU_UP_CANCELED_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08003186 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003187 case CPU_DEAD_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08003188 free_zone_pagesets(cpu);
3189 break;
3190 default:
3191 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003192 }
3193 return ret;
3194}
3195
Chandra Seetharaman74b85f32006-06-27 02:54:09 -07003196static struct notifier_block __cpuinitdata pageset_notifier =
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003197 { &pageset_cpuup_callback, NULL, 0 };
3198
Al Viro78d99552005-12-15 09:18:25 +00003199void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003200{
3201 int err;
3202
3203 /* Initialize per_cpu_pageset for cpu 0.
3204 * A cpuup callback will do this for every cpu
3205 * as it comes online
3206 */
3207 err = process_zones(smp_processor_id());
3208 BUG_ON(err);
3209 register_cpu_notifier(&pageset_notifier);
3210}
3211
3212#endif
3213
Sam Ravnborg577a32f2007-05-17 23:29:25 +02003214static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07003215int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07003216{
3217 int i;
3218 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003219 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07003220
3221 /*
3222 * The per-page waitqueue mechanism uses hashed waitqueues
3223 * per zone.
3224 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07003225 zone->wait_table_hash_nr_entries =
3226 wait_table_hash_nr_entries(zone_size_pages);
3227 zone->wait_table_bits =
3228 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003229 alloc_size = zone->wait_table_hash_nr_entries
3230 * sizeof(wait_queue_head_t);
3231
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07003232 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07003233 zone->wait_table = (wait_queue_head_t *)
3234 alloc_bootmem_node(pgdat, alloc_size);
3235 } else {
3236 /*
3237 * This case means that a zone whose size was 0 gets new memory
3238 * via memory hot-add.
3239 * But it may be the case that a new node was hot-added. In
3240 * this case vmalloc() will not be able to use this new node's
3241 * memory - this wait_table must be initialized to use this new
3242 * node itself as well.
3243 * To use this new node's memory, further consideration will be
3244 * necessary.
3245 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07003246 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003247 }
3248 if (!zone->wait_table)
3249 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07003250
Yasunori Goto02b694d2006-06-23 02:03:08 -07003251 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07003252 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003253
3254 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003255}
3256
Shaohua Li112067f2009-09-21 17:01:16 -07003257static int __zone_pcp_update(void *data)
3258{
3259 struct zone *zone = data;
3260 int cpu;
3261 unsigned long batch = zone_batchsize(zone), flags;
3262
3263 for (cpu = 0; cpu < NR_CPUS; cpu++) {
3264 struct per_cpu_pageset *pset;
3265 struct per_cpu_pages *pcp;
3266
3267 pset = zone_pcp(zone, cpu);
3268 pcp = &pset->pcp;
3269
3270 local_irq_save(flags);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003271 free_pcppages_bulk(zone, pcp->count, pcp);
Shaohua Li112067f2009-09-21 17:01:16 -07003272 setup_pageset(pset, batch);
3273 local_irq_restore(flags);
3274 }
3275 return 0;
3276}
3277
3278void zone_pcp_update(struct zone *zone)
3279{
3280 stop_machine(__zone_pcp_update, zone, NULL);
3281}
3282
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003283static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07003284{
3285 int cpu;
3286 unsigned long batch = zone_batchsize(zone);
3287
3288 for (cpu = 0; cpu < NR_CPUS; cpu++) {
3289#ifdef CONFIG_NUMA
3290 /* Early boot. Slab allocator not functional yet */
Nick Piggin23316bc2006-01-08 01:00:41 -08003291 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Dave Hansened8ece22005-10-29 18:16:50 -07003292 setup_pageset(&boot_pageset[cpu],0);
3293#else
3294 setup_pageset(zone_pcp(zone,cpu), batch);
3295#endif
3296 }
Anton Blanchardf5335c02006-03-25 03:06:49 -08003297 if (zone->present_pages)
3298 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
3299 zone->name, zone->present_pages, batch);
Dave Hansened8ece22005-10-29 18:16:50 -07003300}
3301
Yasunori Goto718127c2006-06-23 02:03:10 -07003302__meminit int init_currently_empty_zone(struct zone *zone,
3303 unsigned long zone_start_pfn,
Dave Hansena2f3aa02007-01-10 23:15:30 -08003304 unsigned long size,
3305 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07003306{
3307 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003308 int ret;
3309 ret = zone_wait_table_init(zone, size);
3310 if (ret)
3311 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07003312 pgdat->nr_zones = zone_idx(zone) + 1;
3313
Dave Hansened8ece22005-10-29 18:16:50 -07003314 zone->zone_start_pfn = zone_start_pfn;
3315
Mel Gorman708614e2008-07-23 21:26:51 -07003316 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3317 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3318 pgdat->node_id,
3319 (unsigned long)zone_idx(zone),
3320 zone_start_pfn, (zone_start_pfn + size));
3321
Andi Kleen1e548de2008-02-04 22:29:26 -08003322 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07003323
3324 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003325}
3326
Mel Gormanc7132162006-09-27 01:49:43 -07003327#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3328/*
3329 * Basic iterator support. Return the first range of PFNs for a node
3330 * Note: nid == MAX_NUMNODES returns first region regardless of node
3331 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003332static int __meminit first_active_region_index_in_nid(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003333{
3334 int i;
3335
3336 for (i = 0; i < nr_nodemap_entries; i++)
3337 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3338 return i;
3339
3340 return -1;
3341}
3342
3343/*
3344 * Basic iterator support. Return the next active range of PFNs for a node
Simon Arlott183ff222007-10-20 01:27:18 +02003345 * Note: nid == MAX_NUMNODES returns next region regardless of node
Mel Gormanc7132162006-09-27 01:49:43 -07003346 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003347static int __meminit next_active_region_index_in_nid(int index, int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003348{
3349 for (index = index + 1; index < nr_nodemap_entries; index++)
3350 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3351 return index;
3352
3353 return -1;
3354}
3355
3356#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3357/*
3358 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3359 * Architectures may implement their own version but if add_active_range()
3360 * was used and there are no special requirements, this is a convenient
3361 * alternative
3362 */
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003363int __meminit __early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07003364{
3365 int i;
3366
3367 for (i = 0; i < nr_nodemap_entries; i++) {
3368 unsigned long start_pfn = early_node_map[i].start_pfn;
3369 unsigned long end_pfn = early_node_map[i].end_pfn;
3370
3371 if (start_pfn <= pfn && pfn < end_pfn)
3372 return early_node_map[i].nid;
3373 }
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003374 /* This is a memory hole */
3375 return -1;
Mel Gormanc7132162006-09-27 01:49:43 -07003376}
3377#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3378
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003379int __meminit early_pfn_to_nid(unsigned long pfn)
3380{
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003381 int nid;
3382
3383 nid = __early_pfn_to_nid(pfn);
3384 if (nid >= 0)
3385 return nid;
3386 /* just returns 0 */
3387 return 0;
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003388}
3389
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003390#ifdef CONFIG_NODES_SPAN_OTHER_NODES
3391bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3392{
3393 int nid;
3394
3395 nid = __early_pfn_to_nid(pfn);
3396 if (nid >= 0 && nid != node)
3397 return false;
3398 return true;
3399}
3400#endif
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003401
Mel Gormanc7132162006-09-27 01:49:43 -07003402/* Basic iterator support to walk early_node_map[] */
3403#define for_each_active_range_index_in_nid(i, nid) \
3404 for (i = first_active_region_index_in_nid(nid); i != -1; \
3405 i = next_active_region_index_in_nid(i, nid))
3406
3407/**
3408 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003409 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3410 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07003411 *
3412 * If an architecture guarantees that all ranges registered with
3413 * add_active_ranges() contain no holes and may be freed, this
3414 * this function may be used instead of calling free_bootmem() manually.
3415 */
3416void __init free_bootmem_with_active_regions(int nid,
3417 unsigned long max_low_pfn)
3418{
3419 int i;
3420
3421 for_each_active_range_index_in_nid(i, nid) {
3422 unsigned long size_pages = 0;
3423 unsigned long end_pfn = early_node_map[i].end_pfn;
3424
3425 if (early_node_map[i].start_pfn >= max_low_pfn)
3426 continue;
3427
3428 if (end_pfn > max_low_pfn)
3429 end_pfn = max_low_pfn;
3430
3431 size_pages = end_pfn - early_node_map[i].start_pfn;
3432 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3433 PFN_PHYS(early_node_map[i].start_pfn),
3434 size_pages << PAGE_SHIFT);
3435 }
3436}
3437
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003438void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3439{
3440 int i;
Yinghai Lud52d53b2008-06-16 20:10:55 -07003441 int ret;
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003442
Yinghai Lud52d53b2008-06-16 20:10:55 -07003443 for_each_active_range_index_in_nid(i, nid) {
3444 ret = work_fn(early_node_map[i].start_pfn,
3445 early_node_map[i].end_pfn, data);
3446 if (ret)
3447 break;
3448 }
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003449}
Mel Gormanc7132162006-09-27 01:49:43 -07003450/**
3451 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003452 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07003453 *
3454 * If an architecture guarantees that all ranges registered with
3455 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003456 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07003457 */
3458void __init sparse_memory_present_with_active_regions(int nid)
3459{
3460 int i;
3461
3462 for_each_active_range_index_in_nid(i, nid)
3463 memory_present(early_node_map[i].nid,
3464 early_node_map[i].start_pfn,
3465 early_node_map[i].end_pfn);
3466}
3467
3468/**
3469 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003470 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3471 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3472 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07003473 *
3474 * It returns the start and end page frame of a node based on information
3475 * provided by an arch calling add_active_range(). If called for a node
3476 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003477 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07003478 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003479void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003480 unsigned long *start_pfn, unsigned long *end_pfn)
3481{
3482 int i;
3483 *start_pfn = -1UL;
3484 *end_pfn = 0;
3485
3486 for_each_active_range_index_in_nid(i, nid) {
3487 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3488 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3489 }
3490
Christoph Lameter633c0662007-10-16 01:25:37 -07003491 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07003492 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07003493}
3494
3495/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07003496 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3497 * assumption is made that zones within a node are ordered in monotonic
3498 * increasing memory addresses so that the "highest" populated zone is used
3499 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003500static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07003501{
3502 int zone_index;
3503 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3504 if (zone_index == ZONE_MOVABLE)
3505 continue;
3506
3507 if (arch_zone_highest_possible_pfn[zone_index] >
3508 arch_zone_lowest_possible_pfn[zone_index])
3509 break;
3510 }
3511
3512 VM_BUG_ON(zone_index == -1);
3513 movable_zone = zone_index;
3514}
3515
3516/*
3517 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3518 * because it is sized independant of architecture. Unlike the other zones,
3519 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3520 * in each node depending on the size of each node and how evenly kernelcore
3521 * is distributed. This helper function adjusts the zone ranges
3522 * provided by the architecture for a given node by using the end of the
3523 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3524 * zones within a node are in order of monotonic increases memory addresses
3525 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003526static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07003527 unsigned long zone_type,
3528 unsigned long node_start_pfn,
3529 unsigned long node_end_pfn,
3530 unsigned long *zone_start_pfn,
3531 unsigned long *zone_end_pfn)
3532{
3533 /* Only adjust if ZONE_MOVABLE is on this node */
3534 if (zone_movable_pfn[nid]) {
3535 /* Size ZONE_MOVABLE */
3536 if (zone_type == ZONE_MOVABLE) {
3537 *zone_start_pfn = zone_movable_pfn[nid];
3538 *zone_end_pfn = min(node_end_pfn,
3539 arch_zone_highest_possible_pfn[movable_zone]);
3540
3541 /* Adjust for ZONE_MOVABLE starting within this range */
3542 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3543 *zone_end_pfn > zone_movable_pfn[nid]) {
3544 *zone_end_pfn = zone_movable_pfn[nid];
3545
3546 /* Check if this whole range is within ZONE_MOVABLE */
3547 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3548 *zone_start_pfn = *zone_end_pfn;
3549 }
3550}
3551
3552/*
Mel Gormanc7132162006-09-27 01:49:43 -07003553 * Return the number of pages a zone spans in a node, including holes
3554 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3555 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003556static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003557 unsigned long zone_type,
3558 unsigned long *ignored)
3559{
3560 unsigned long node_start_pfn, node_end_pfn;
3561 unsigned long zone_start_pfn, zone_end_pfn;
3562
3563 /* Get the start and end of the node and zone */
3564 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3565 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3566 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07003567 adjust_zone_range_for_zone_movable(nid, zone_type,
3568 node_start_pfn, node_end_pfn,
3569 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003570
3571 /* Check that this node has pages within the zone's required range */
3572 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3573 return 0;
3574
3575 /* Move the zone boundaries inside the node if necessary */
3576 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3577 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3578
3579 /* Return the spanned pages */
3580 return zone_end_pfn - zone_start_pfn;
3581}
3582
3583/*
3584 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003585 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07003586 */
Yinghai Lu32996252009-12-15 17:59:02 -08003587unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003588 unsigned long range_start_pfn,
3589 unsigned long range_end_pfn)
3590{
3591 int i = 0;
3592 unsigned long prev_end_pfn = 0, hole_pages = 0;
3593 unsigned long start_pfn;
3594
3595 /* Find the end_pfn of the first active range of pfns in the node */
3596 i = first_active_region_index_in_nid(nid);
3597 if (i == -1)
3598 return 0;
3599
Mel Gormanb5445f92007-07-26 10:41:18 -07003600 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3601
Mel Gorman9c7cd682006-09-27 01:49:58 -07003602 /* Account for ranges before physical memory on this node */
3603 if (early_node_map[i].start_pfn > range_start_pfn)
Mel Gormanb5445f92007-07-26 10:41:18 -07003604 hole_pages = prev_end_pfn - range_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003605
3606 /* Find all holes for the zone within the node */
3607 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3608
3609 /* No need to continue if prev_end_pfn is outside the zone */
3610 if (prev_end_pfn >= range_end_pfn)
3611 break;
3612
3613 /* Make sure the end of the zone is not within the hole */
3614 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3615 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3616
3617 /* Update the hole size cound and move on */
3618 if (start_pfn > range_start_pfn) {
3619 BUG_ON(prev_end_pfn > start_pfn);
3620 hole_pages += start_pfn - prev_end_pfn;
3621 }
3622 prev_end_pfn = early_node_map[i].end_pfn;
3623 }
3624
Mel Gorman9c7cd682006-09-27 01:49:58 -07003625 /* Account for ranges past physical memory on this node */
3626 if (range_end_pfn > prev_end_pfn)
Mel Gorman0c6cb972006-10-28 10:38:59 -07003627 hole_pages += range_end_pfn -
Mel Gorman9c7cd682006-09-27 01:49:58 -07003628 max(range_start_pfn, prev_end_pfn);
3629
Mel Gormanc7132162006-09-27 01:49:43 -07003630 return hole_pages;
3631}
3632
3633/**
3634 * absent_pages_in_range - Return number of page frames in holes within a range
3635 * @start_pfn: The start PFN to start searching for holes
3636 * @end_pfn: The end PFN to stop searching for holes
3637 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003638 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07003639 */
3640unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3641 unsigned long end_pfn)
3642{
3643 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3644}
3645
3646/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003647static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003648 unsigned long zone_type,
3649 unsigned long *ignored)
3650{
Mel Gorman9c7cd682006-09-27 01:49:58 -07003651 unsigned long node_start_pfn, node_end_pfn;
3652 unsigned long zone_start_pfn, zone_end_pfn;
3653
3654 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3655 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3656 node_start_pfn);
3657 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3658 node_end_pfn);
3659
Mel Gorman2a1e2742007-07-17 04:03:12 -07003660 adjust_zone_range_for_zone_movable(nid, zone_type,
3661 node_start_pfn, node_end_pfn,
3662 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07003663 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003664}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003665
Mel Gormanc7132162006-09-27 01:49:43 -07003666#else
Paul Mundt6ea6e682007-07-15 23:38:20 -07003667static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003668 unsigned long zone_type,
3669 unsigned long *zones_size)
3670{
3671 return zones_size[zone_type];
3672}
3673
Paul Mundt6ea6e682007-07-15 23:38:20 -07003674static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003675 unsigned long zone_type,
3676 unsigned long *zholes_size)
3677{
3678 if (!zholes_size)
3679 return 0;
3680
3681 return zholes_size[zone_type];
3682}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003683
Mel Gormanc7132162006-09-27 01:49:43 -07003684#endif
3685
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003686static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Mel Gormanc7132162006-09-27 01:49:43 -07003687 unsigned long *zones_size, unsigned long *zholes_size)
3688{
3689 unsigned long realtotalpages, totalpages = 0;
3690 enum zone_type i;
3691
3692 for (i = 0; i < MAX_NR_ZONES; i++)
3693 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3694 zones_size);
3695 pgdat->node_spanned_pages = totalpages;
3696
3697 realtotalpages = totalpages;
3698 for (i = 0; i < MAX_NR_ZONES; i++)
3699 realtotalpages -=
3700 zone_absent_pages_in_node(pgdat->node_id, i,
3701 zholes_size);
3702 pgdat->node_present_pages = realtotalpages;
3703 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3704 realtotalpages);
3705}
3706
Mel Gorman835c1342007-10-16 01:25:47 -07003707#ifndef CONFIG_SPARSEMEM
3708/*
3709 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07003710 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3711 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07003712 * round what is now in bits to nearest long in bits, then return it in
3713 * bytes.
3714 */
3715static unsigned long __init usemap_size(unsigned long zonesize)
3716{
3717 unsigned long usemapsize;
3718
Mel Gormand9c23402007-10-16 01:26:01 -07003719 usemapsize = roundup(zonesize, pageblock_nr_pages);
3720 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07003721 usemapsize *= NR_PAGEBLOCK_BITS;
3722 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3723
3724 return usemapsize / 8;
3725}
3726
3727static void __init setup_usemap(struct pglist_data *pgdat,
3728 struct zone *zone, unsigned long zonesize)
3729{
3730 unsigned long usemapsize = usemap_size(zonesize);
3731 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08003732 if (usemapsize)
Mel Gorman835c1342007-10-16 01:25:47 -07003733 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
Mel Gorman835c1342007-10-16 01:25:47 -07003734}
3735#else
3736static void inline setup_usemap(struct pglist_data *pgdat,
3737 struct zone *zone, unsigned long zonesize) {}
3738#endif /* CONFIG_SPARSEMEM */
3739
Mel Gormand9c23402007-10-16 01:26:01 -07003740#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08003741
3742/* Return a sensible default order for the pageblock size. */
3743static inline int pageblock_default_order(void)
3744{
3745 if (HPAGE_SHIFT > PAGE_SHIFT)
3746 return HUGETLB_PAGE_ORDER;
3747
3748 return MAX_ORDER-1;
3749}
3750
Mel Gormand9c23402007-10-16 01:26:01 -07003751/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3752static inline void __init set_pageblock_order(unsigned int order)
3753{
3754 /* Check that pageblock_nr_pages has not already been setup */
3755 if (pageblock_order)
3756 return;
3757
3758 /*
3759 * Assume the largest contiguous order of interest is a huge page.
3760 * This value may be variable depending on boot parameters on IA64
3761 */
3762 pageblock_order = order;
3763}
3764#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3765
Mel Gormanba72cb82007-11-28 16:21:13 -08003766/*
3767 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3768 * and pageblock_default_order() are unused as pageblock_order is set
3769 * at compile-time. See include/linux/pageblock-flags.h for the values of
3770 * pageblock_order based on the kernel config
3771 */
3772static inline int pageblock_default_order(unsigned int order)
3773{
3774 return MAX_ORDER-1;
3775}
Mel Gormand9c23402007-10-16 01:26:01 -07003776#define set_pageblock_order(x) do {} while (0)
3777
3778#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3779
Linus Torvalds1da177e2005-04-16 15:20:36 -07003780/*
3781 * Set up the zone data structures:
3782 * - mark all pages reserved
3783 * - mark all memory queues empty
3784 * - clear the memory bitmaps
3785 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08003786static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003787 unsigned long *zones_size, unsigned long *zholes_size)
3788{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07003789 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07003790 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003791 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07003792 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003793
Dave Hansen208d54e2005-10-29 18:16:52 -07003794 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003795 pgdat->nr_zones = 0;
3796 init_waitqueue_head(&pgdat->kswapd_wait);
3797 pgdat->kswapd_max_order = 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003798 pgdat_page_cgroup_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003799
3800 for (j = 0; j < MAX_NR_ZONES; j++) {
3801 struct zone *zone = pgdat->node_zones + j;
Mel Gorman0e0b8642006-09-27 01:49:56 -07003802 unsigned long size, realsize, memmap_pages;
Christoph Lameterb69408e2008-10-18 20:26:14 -07003803 enum lru_list l;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003804
Mel Gormanc7132162006-09-27 01:49:43 -07003805 size = zone_spanned_pages_in_node(nid, j, zones_size);
3806 realsize = size - zone_absent_pages_in_node(nid, j,
3807 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003808
Mel Gorman0e0b8642006-09-27 01:49:56 -07003809 /*
3810 * Adjust realsize so that it accounts for how much memory
3811 * is used by this zone for memmap. This affects the watermark
3812 * and per-cpu initialisations
3813 */
Johannes Weinerf7232152008-05-23 13:04:21 -07003814 memmap_pages =
3815 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
Mel Gorman0e0b8642006-09-27 01:49:56 -07003816 if (realsize >= memmap_pages) {
3817 realsize -= memmap_pages;
Yinghai Lu5594c8c2009-01-06 14:39:14 -08003818 if (memmap_pages)
3819 printk(KERN_DEBUG
3820 " %s zone: %lu pages used for memmap\n",
3821 zone_names[j], memmap_pages);
Mel Gorman0e0b8642006-09-27 01:49:56 -07003822 } else
3823 printk(KERN_WARNING
3824 " %s zone: %lu pages exceeds realsize %lu\n",
3825 zone_names[j], memmap_pages, realsize);
3826
Christoph Lameter62672762007-02-10 01:43:07 -08003827 /* Account for reserved pages */
3828 if (j == 0 && realsize > dma_reserve) {
Mel Gorman0e0b8642006-09-27 01:49:56 -07003829 realsize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07003830 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08003831 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07003832 }
3833
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003834 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003835 nr_kernel_pages += realsize;
3836 nr_all_pages += realsize;
3837
3838 zone->spanned_pages = size;
3839 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07003840#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07003841 zone->node = nid;
Christoph Lameter8417bba2006-09-25 23:31:51 -07003842 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07003843 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07003844 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07003845#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003846 zone->name = zone_names[j];
3847 spin_lock_init(&zone->lock);
3848 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07003849 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003850 zone->zone_pgdat = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003851
Martin Bligh3bb1a852006-10-28 10:38:24 -07003852 zone->prev_priority = DEF_PRIORITY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003853
Dave Hansened8ece22005-10-29 18:16:50 -07003854 zone_pcp_init(zone);
Christoph Lameterb69408e2008-10-18 20:26:14 -07003855 for_each_lru(l) {
3856 INIT_LIST_HEAD(&zone->lru[l].list);
Wu Fengguangf8629632009-09-21 17:03:11 -07003857 zone->reclaim_stat.nr_saved_scan[l] = 0;
Christoph Lameterb69408e2008-10-18 20:26:14 -07003858 }
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08003859 zone->reclaim_stat.recent_rotated[0] = 0;
3860 zone->reclaim_stat.recent_rotated[1] = 0;
3861 zone->reclaim_stat.recent_scanned[0] = 0;
3862 zone->reclaim_stat.recent_scanned[1] = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07003863 zap_zone_vm_stats(zone);
David Rientjese815af92007-10-16 23:25:54 -07003864 zone->flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003865 if (!size)
3866 continue;
3867
Mel Gormanba72cb82007-11-28 16:21:13 -08003868 set_pageblock_order(pageblock_default_order());
Mel Gorman835c1342007-10-16 01:25:47 -07003869 setup_usemap(pgdat, zone, size);
Dave Hansena2f3aa02007-01-10 23:15:30 -08003870 ret = init_currently_empty_zone(zone, zone_start_pfn,
3871 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07003872 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07003873 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003874 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003875 }
3876}
3877
Sam Ravnborg577a32f2007-05-17 23:29:25 +02003878static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003879{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003880 /* Skip empty nodes */
3881 if (!pgdat->node_spanned_pages)
3882 return;
3883
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003884#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07003885 /* ia64 gets its own node_mem_map, before this, without bootmem */
3886 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07003887 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003888 struct page *map;
3889
Bob Piccoe984bb42006-05-20 15:00:31 -07003890 /*
3891 * The zone's endpoints aren't required to be MAX_ORDER
3892 * aligned but the node_mem_map endpoints must be in order
3893 * for the buddy allocator to function correctly.
3894 */
3895 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3896 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3897 end = ALIGN(end, MAX_ORDER_NR_PAGES);
3898 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07003899 map = alloc_remap(pgdat->node_id, size);
3900 if (!map)
3901 map = alloc_bootmem_node(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07003902 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003903 }
Roman Zippel12d810c2007-05-31 00:40:54 -07003904#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07003905 /*
3906 * With no DISCONTIG, the global mem_map is just set as node 0's
3907 */
Mel Gormanc7132162006-09-27 01:49:43 -07003908 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003909 mem_map = NODE_DATA(0)->node_mem_map;
Mel Gormanc7132162006-09-27 01:49:43 -07003910#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3911 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08003912 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Mel Gormanc7132162006-09-27 01:49:43 -07003913#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3914 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003915#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003916#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003917}
3918
Johannes Weiner9109fb72008-07-23 21:27:20 -07003919void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
3920 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003921{
Johannes Weiner9109fb72008-07-23 21:27:20 -07003922 pg_data_t *pgdat = NODE_DATA(nid);
3923
Linus Torvalds1da177e2005-04-16 15:20:36 -07003924 pgdat->node_id = nid;
3925 pgdat->node_start_pfn = node_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003926 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003927
3928 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07003929#ifdef CONFIG_FLAT_NODE_MEM_MAP
3930 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3931 nid, (unsigned long)pgdat,
3932 (unsigned long)pgdat->node_mem_map);
3933#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003934
3935 free_area_init_core(pgdat, zones_size, zholes_size);
3936}
3937
Mel Gormanc7132162006-09-27 01:49:43 -07003938#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07003939
3940#if MAX_NUMNODES > 1
3941/*
3942 * Figure out the number of possible node ids.
3943 */
3944static void __init setup_nr_node_ids(void)
3945{
3946 unsigned int node;
3947 unsigned int highest = 0;
3948
3949 for_each_node_mask(node, node_possible_map)
3950 highest = node;
3951 nr_node_ids = highest + 1;
3952}
3953#else
3954static inline void setup_nr_node_ids(void)
3955{
3956}
3957#endif
3958
Mel Gormanc7132162006-09-27 01:49:43 -07003959/**
3960 * add_active_range - Register a range of PFNs backed by physical memory
3961 * @nid: The node ID the range resides on
3962 * @start_pfn: The start PFN of the available physical memory
3963 * @end_pfn: The end PFN of the available physical memory
3964 *
3965 * These ranges are stored in an early_node_map[] and later used by
3966 * free_area_init_nodes() to calculate zone sizes and holes. If the
3967 * range spans a memory hole, it is up to the architecture to ensure
3968 * the memory is not freed by the bootmem allocator. If possible
3969 * the range being registered will be merged with existing ranges.
3970 */
3971void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3972 unsigned long end_pfn)
3973{
3974 int i;
3975
Mel Gorman6b74ab92008-07-23 21:26:49 -07003976 mminit_dprintk(MMINIT_TRACE, "memory_register",
3977 "Entering add_active_range(%d, %#lx, %#lx) "
3978 "%d entries of %d used\n",
3979 nid, start_pfn, end_pfn,
3980 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
Mel Gormanc7132162006-09-27 01:49:43 -07003981
Mel Gorman2dbb51c2008-07-23 21:26:52 -07003982 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
3983
Mel Gormanc7132162006-09-27 01:49:43 -07003984 /* Merge with existing active regions if possible */
3985 for (i = 0; i < nr_nodemap_entries; i++) {
3986 if (early_node_map[i].nid != nid)
3987 continue;
3988
3989 /* Skip if an existing region covers this new one */
3990 if (start_pfn >= early_node_map[i].start_pfn &&
3991 end_pfn <= early_node_map[i].end_pfn)
3992 return;
3993
3994 /* Merge forward if suitable */
3995 if (start_pfn <= early_node_map[i].end_pfn &&
3996 end_pfn > early_node_map[i].end_pfn) {
3997 early_node_map[i].end_pfn = end_pfn;
3998 return;
3999 }
4000
4001 /* Merge backward if suitable */
Kazuhisa Ichikawad2dbe082010-01-15 17:01:20 -08004002 if (start_pfn < early_node_map[i].start_pfn &&
Mel Gormanc7132162006-09-27 01:49:43 -07004003 end_pfn >= early_node_map[i].start_pfn) {
4004 early_node_map[i].start_pfn = start_pfn;
4005 return;
4006 }
4007 }
4008
4009 /* Check that early_node_map is large enough */
4010 if (i >= MAX_ACTIVE_REGIONS) {
4011 printk(KERN_CRIT "More than %d memory regions, truncating\n",
4012 MAX_ACTIVE_REGIONS);
4013 return;
4014 }
4015
4016 early_node_map[i].nid = nid;
4017 early_node_map[i].start_pfn = start_pfn;
4018 early_node_map[i].end_pfn = end_pfn;
4019 nr_nodemap_entries = i + 1;
4020}
4021
4022/**
Yinghai Lucc1050b2008-06-13 19:08:52 -07004023 * remove_active_range - Shrink an existing registered range of PFNs
Mel Gormanc7132162006-09-27 01:49:43 -07004024 * @nid: The node id the range is on that should be shrunk
Yinghai Lucc1050b2008-06-13 19:08:52 -07004025 * @start_pfn: The new PFN of the range
4026 * @end_pfn: The new PFN of the range
Mel Gormanc7132162006-09-27 01:49:43 -07004027 *
4028 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004029 * The map is kept near the end physical page range that has already been
4030 * registered. This function allows an arch to shrink an existing registered
4031 * range.
Mel Gormanc7132162006-09-27 01:49:43 -07004032 */
Yinghai Lucc1050b2008-06-13 19:08:52 -07004033void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
4034 unsigned long end_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004035{
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004036 int i, j;
4037 int removed = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07004038
Yinghai Lucc1050b2008-06-13 19:08:52 -07004039 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
4040 nid, start_pfn, end_pfn);
4041
Mel Gormanc7132162006-09-27 01:49:43 -07004042 /* Find the old active region end and shrink */
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004043 for_each_active_range_index_in_nid(i, nid) {
Yinghai Lucc1050b2008-06-13 19:08:52 -07004044 if (early_node_map[i].start_pfn >= start_pfn &&
4045 early_node_map[i].end_pfn <= end_pfn) {
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004046 /* clear it */
Yinghai Lucc1050b2008-06-13 19:08:52 -07004047 early_node_map[i].start_pfn = 0;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004048 early_node_map[i].end_pfn = 0;
4049 removed = 1;
4050 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004051 }
Yinghai Lucc1050b2008-06-13 19:08:52 -07004052 if (early_node_map[i].start_pfn < start_pfn &&
4053 early_node_map[i].end_pfn > start_pfn) {
4054 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
4055 early_node_map[i].end_pfn = start_pfn;
4056 if (temp_end_pfn > end_pfn)
4057 add_active_range(nid, end_pfn, temp_end_pfn);
4058 continue;
4059 }
4060 if (early_node_map[i].start_pfn >= start_pfn &&
4061 early_node_map[i].end_pfn > end_pfn &&
4062 early_node_map[i].start_pfn < end_pfn) {
4063 early_node_map[i].start_pfn = end_pfn;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004064 continue;
4065 }
4066 }
4067
4068 if (!removed)
4069 return;
4070
4071 /* remove the blank ones */
4072 for (i = nr_nodemap_entries - 1; i > 0; i--) {
4073 if (early_node_map[i].nid != nid)
4074 continue;
4075 if (early_node_map[i].end_pfn)
4076 continue;
4077 /* we found it, get rid of it */
4078 for (j = i; j < nr_nodemap_entries - 1; j++)
4079 memcpy(&early_node_map[j], &early_node_map[j+1],
4080 sizeof(early_node_map[j]));
4081 j = nr_nodemap_entries - 1;
4082 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
4083 nr_nodemap_entries--;
4084 }
Mel Gormanc7132162006-09-27 01:49:43 -07004085}
4086
4087/**
4088 * remove_all_active_ranges - Remove all currently registered regions
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004089 *
Mel Gormanc7132162006-09-27 01:49:43 -07004090 * During discovery, it may be found that a table like SRAT is invalid
4091 * and an alternative discovery method must be used. This function removes
4092 * all currently registered regions.
4093 */
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004094void __init remove_all_active_ranges(void)
Mel Gormanc7132162006-09-27 01:49:43 -07004095{
4096 memset(early_node_map, 0, sizeof(early_node_map));
4097 nr_nodemap_entries = 0;
4098}
4099
4100/* Compare two active node_active_regions */
4101static int __init cmp_node_active_region(const void *a, const void *b)
4102{
4103 struct node_active_region *arange = (struct node_active_region *)a;
4104 struct node_active_region *brange = (struct node_active_region *)b;
4105
4106 /* Done this way to avoid overflows */
4107 if (arange->start_pfn > brange->start_pfn)
4108 return 1;
4109 if (arange->start_pfn < brange->start_pfn)
4110 return -1;
4111
4112 return 0;
4113}
4114
4115/* sort the node_map by start_pfn */
Yinghai Lu32996252009-12-15 17:59:02 -08004116void __init sort_node_map(void)
Mel Gormanc7132162006-09-27 01:49:43 -07004117{
4118 sort(early_node_map, (size_t)nr_nodemap_entries,
4119 sizeof(struct node_active_region),
4120 cmp_node_active_region, NULL);
4121}
4122
Mel Gormana6af2bc2007-02-10 01:42:57 -08004123/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004124static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07004125{
4126 int i;
Mel Gormana6af2bc2007-02-10 01:42:57 -08004127 unsigned long min_pfn = ULONG_MAX;
Mel Gorman1abbfb42006-11-23 12:01:41 +00004128
Mel Gormanc7132162006-09-27 01:49:43 -07004129 /* Assuming a sorted map, the first range found has the starting pfn */
4130 for_each_active_range_index_in_nid(i, nid)
Mel Gormana6af2bc2007-02-10 01:42:57 -08004131 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004132
Mel Gormana6af2bc2007-02-10 01:42:57 -08004133 if (min_pfn == ULONG_MAX) {
4134 printk(KERN_WARNING
Paul Jackson2bc0d262008-06-22 07:22:17 -07004135 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08004136 return 0;
4137 }
4138
4139 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004140}
4141
4142/**
4143 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4144 *
4145 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004146 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07004147 */
4148unsigned long __init find_min_pfn_with_active_regions(void)
4149{
4150 return find_min_pfn_for_node(MAX_NUMNODES);
4151}
4152
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004153/*
4154 * early_calculate_totalpages()
4155 * Sum pages in active regions for movable zone.
4156 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4157 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07004158static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07004159{
4160 int i;
4161 unsigned long totalpages = 0;
4162
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004163 for (i = 0; i < nr_nodemap_entries; i++) {
4164 unsigned long pages = early_node_map[i].end_pfn -
Mel Gorman7e63efe2007-07-17 04:03:15 -07004165 early_node_map[i].start_pfn;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004166 totalpages += pages;
4167 if (pages)
4168 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
4169 }
4170 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07004171}
4172
Mel Gorman2a1e2742007-07-17 04:03:12 -07004173/*
4174 * Find the PFN the Movable zone begins in each node. Kernel memory
4175 * is spread evenly between nodes as long as the nodes have enough
4176 * memory. When they don't, some nodes will have more kernelcore than
4177 * others
4178 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004179static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004180{
4181 int i, nid;
4182 unsigned long usable_startpfn;
4183 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07004184 /* save the state before borrow the nodemask */
4185 nodemask_t saved_node_state = node_states[N_HIGH_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004186 unsigned long totalpages = early_calculate_totalpages();
4187 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004188
Mel Gorman7e63efe2007-07-17 04:03:15 -07004189 /*
4190 * If movablecore was specified, calculate what size of
4191 * kernelcore that corresponds so that memory usable for
4192 * any allocation type is evenly spread. If both kernelcore
4193 * and movablecore are specified, then the value of kernelcore
4194 * will be used for required_kernelcore if it's greater than
4195 * what movablecore would have allowed.
4196 */
4197 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07004198 unsigned long corepages;
4199
4200 /*
4201 * Round-up so that ZONE_MOVABLE is at least as large as what
4202 * was requested by the user
4203 */
4204 required_movablecore =
4205 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
4206 corepages = totalpages - required_movablecore;
4207
4208 required_kernelcore = max(required_kernelcore, corepages);
4209 }
4210
Mel Gorman2a1e2742007-07-17 04:03:12 -07004211 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4212 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07004213 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004214
4215 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4216 find_usable_zone_for_movable();
4217 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4218
4219restart:
4220 /* Spread kernelcore memory as evenly as possible throughout nodes */
4221 kernelcore_node = required_kernelcore / usable_nodes;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004222 for_each_node_state(nid, N_HIGH_MEMORY) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004223 /*
4224 * Recalculate kernelcore_node if the division per node
4225 * now exceeds what is necessary to satisfy the requested
4226 * amount of memory for the kernel
4227 */
4228 if (required_kernelcore < kernelcore_node)
4229 kernelcore_node = required_kernelcore / usable_nodes;
4230
4231 /*
4232 * As the map is walked, we track how much memory is usable
4233 * by the kernel using kernelcore_remaining. When it is
4234 * 0, the rest of the node is usable by ZONE_MOVABLE
4235 */
4236 kernelcore_remaining = kernelcore_node;
4237
4238 /* Go through each range of PFNs within this node */
4239 for_each_active_range_index_in_nid(i, nid) {
4240 unsigned long start_pfn, end_pfn;
4241 unsigned long size_pages;
4242
4243 start_pfn = max(early_node_map[i].start_pfn,
4244 zone_movable_pfn[nid]);
4245 end_pfn = early_node_map[i].end_pfn;
4246 if (start_pfn >= end_pfn)
4247 continue;
4248
4249 /* Account for what is only usable for kernelcore */
4250 if (start_pfn < usable_startpfn) {
4251 unsigned long kernel_pages;
4252 kernel_pages = min(end_pfn, usable_startpfn)
4253 - start_pfn;
4254
4255 kernelcore_remaining -= min(kernel_pages,
4256 kernelcore_remaining);
4257 required_kernelcore -= min(kernel_pages,
4258 required_kernelcore);
4259
4260 /* Continue if range is now fully accounted */
4261 if (end_pfn <= usable_startpfn) {
4262
4263 /*
4264 * Push zone_movable_pfn to the end so
4265 * that if we have to rebalance
4266 * kernelcore across nodes, we will
4267 * not double account here
4268 */
4269 zone_movable_pfn[nid] = end_pfn;
4270 continue;
4271 }
4272 start_pfn = usable_startpfn;
4273 }
4274
4275 /*
4276 * The usable PFN range for ZONE_MOVABLE is from
4277 * start_pfn->end_pfn. Calculate size_pages as the
4278 * number of pages used as kernelcore
4279 */
4280 size_pages = end_pfn - start_pfn;
4281 if (size_pages > kernelcore_remaining)
4282 size_pages = kernelcore_remaining;
4283 zone_movable_pfn[nid] = start_pfn + size_pages;
4284
4285 /*
4286 * Some kernelcore has been met, update counts and
4287 * break if the kernelcore for this node has been
4288 * satisified
4289 */
4290 required_kernelcore -= min(required_kernelcore,
4291 size_pages);
4292 kernelcore_remaining -= size_pages;
4293 if (!kernelcore_remaining)
4294 break;
4295 }
4296 }
4297
4298 /*
4299 * If there is still required_kernelcore, we do another pass with one
4300 * less node in the count. This will push zone_movable_pfn[nid] further
4301 * along on the nodes that still have memory until kernelcore is
4302 * satisified
4303 */
4304 usable_nodes--;
4305 if (usable_nodes && required_kernelcore > usable_nodes)
4306 goto restart;
4307
4308 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4309 for (nid = 0; nid < MAX_NUMNODES; nid++)
4310 zone_movable_pfn[nid] =
4311 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07004312
4313out:
4314 /* restore the node_state */
4315 node_states[N_HIGH_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004316}
4317
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004318/* Any regular memory on that node ? */
4319static void check_for_regular_memory(pg_data_t *pgdat)
4320{
4321#ifdef CONFIG_HIGHMEM
4322 enum zone_type zone_type;
4323
4324 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4325 struct zone *zone = &pgdat->node_zones[zone_type];
4326 if (zone->present_pages)
4327 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4328 }
4329#endif
4330}
4331
Mel Gormanc7132162006-09-27 01:49:43 -07004332/**
4333 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004334 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07004335 *
4336 * This will call free_area_init_node() for each active node in the system.
4337 * Using the page ranges provided by add_active_range(), the size of each
4338 * zone in each node and their holes is calculated. If the maximum PFN
4339 * between two adjacent zones match, it is assumed that the zone is empty.
4340 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4341 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4342 * starts where the previous one ended. For example, ZONE_DMA32 starts
4343 * at arch_max_dma_pfn.
4344 */
4345void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4346{
4347 unsigned long nid;
Andrew Mortondb991002008-10-15 22:01:07 -07004348 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004349
Mel Gormana6af2bc2007-02-10 01:42:57 -08004350 /* Sort early_node_map as initialisation assumes it is sorted */
4351 sort_node_map();
4352
Mel Gormanc7132162006-09-27 01:49:43 -07004353 /* Record where the zone boundaries are */
4354 memset(arch_zone_lowest_possible_pfn, 0,
4355 sizeof(arch_zone_lowest_possible_pfn));
4356 memset(arch_zone_highest_possible_pfn, 0,
4357 sizeof(arch_zone_highest_possible_pfn));
4358 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4359 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4360 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004361 if (i == ZONE_MOVABLE)
4362 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004363 arch_zone_lowest_possible_pfn[i] =
4364 arch_zone_highest_possible_pfn[i-1];
4365 arch_zone_highest_possible_pfn[i] =
4366 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4367 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07004368 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4369 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4370
4371 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4372 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4373 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004374
Mel Gormanc7132162006-09-27 01:49:43 -07004375 /* Print out the zone ranges */
4376 printk("Zone PFN ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07004377 for (i = 0; i < MAX_NR_ZONES; i++) {
4378 if (i == ZONE_MOVABLE)
4379 continue;
Paul Jackson5dab8ec2008-06-25 05:44:40 -07004380 printk(" %-8s %0#10lx -> %0#10lx\n",
Mel Gormanc7132162006-09-27 01:49:43 -07004381 zone_names[i],
4382 arch_zone_lowest_possible_pfn[i],
4383 arch_zone_highest_possible_pfn[i]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004384 }
4385
4386 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4387 printk("Movable zone start PFN for each node\n");
4388 for (i = 0; i < MAX_NUMNODES; i++) {
4389 if (zone_movable_pfn[i])
4390 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4391 }
Mel Gormanc7132162006-09-27 01:49:43 -07004392
4393 /* Print out the early_node_map[] */
4394 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4395 for (i = 0; i < nr_nodemap_entries; i++)
Paul Jackson5dab8ec2008-06-25 05:44:40 -07004396 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004397 early_node_map[i].start_pfn,
4398 early_node_map[i].end_pfn);
4399
4400 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07004401 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08004402 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07004403 for_each_online_node(nid) {
4404 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07004405 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07004406 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004407
4408 /* Any memory on that node */
4409 if (pgdat->node_present_pages)
4410 node_set_state(nid, N_HIGH_MEMORY);
4411 check_for_regular_memory(pgdat);
Mel Gormanc7132162006-09-27 01:49:43 -07004412 }
4413}
Mel Gorman2a1e2742007-07-17 04:03:12 -07004414
Mel Gorman7e63efe2007-07-17 04:03:15 -07004415static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004416{
4417 unsigned long long coremem;
4418 if (!p)
4419 return -EINVAL;
4420
4421 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004422 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004423
Mel Gorman7e63efe2007-07-17 04:03:15 -07004424 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07004425 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4426
4427 return 0;
4428}
Mel Gormaned7ed362007-07-17 04:03:14 -07004429
Mel Gorman7e63efe2007-07-17 04:03:15 -07004430/*
4431 * kernelcore=size sets the amount of memory for use for allocations that
4432 * cannot be reclaimed or migrated.
4433 */
4434static int __init cmdline_parse_kernelcore(char *p)
4435{
4436 return cmdline_parse_core(p, &required_kernelcore);
4437}
4438
4439/*
4440 * movablecore=size sets the amount of memory for use for allocations that
4441 * can be reclaimed or migrated.
4442 */
4443static int __init cmdline_parse_movablecore(char *p)
4444{
4445 return cmdline_parse_core(p, &required_movablecore);
4446}
4447
Mel Gormaned7ed362007-07-17 04:03:14 -07004448early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004449early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07004450
Mel Gormanc7132162006-09-27 01:49:43 -07004451#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4452
Mel Gorman0e0b8642006-09-27 01:49:56 -07004453/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004454 * set_dma_reserve - set the specified number of pages reserved in the first zone
4455 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07004456 *
4457 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4458 * In the DMA zone, a significant percentage may be consumed by kernel image
4459 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004460 * function may optionally be used to account for unfreeable pages in the
4461 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4462 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07004463 */
4464void __init set_dma_reserve(unsigned long new_dma_reserve)
4465{
4466 dma_reserve = new_dma_reserve;
4467}
4468
Dave Hansen93b75042005-06-23 00:07:47 -07004469#ifndef CONFIG_NEED_MULTIPLE_NODES
Marcin Slusarz52765582008-09-02 14:35:41 -07004470struct pglist_data __refdata contig_page_data = { .bdata = &bootmem_node_data[0] };
Linus Torvalds1da177e2005-04-16 15:20:36 -07004471EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07004472#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004473
4474void __init free_area_init(unsigned long *zones_size)
4475{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004476 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004477 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4478}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004479
Linus Torvalds1da177e2005-04-16 15:20:36 -07004480static int page_alloc_cpu_notify(struct notifier_block *self,
4481 unsigned long action, void *hcpu)
4482{
4483 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004484
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07004485 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004486 drain_pages(cpu);
4487
4488 /*
4489 * Spill the event counters of the dead processor
4490 * into the current processors event counters.
4491 * This artificially elevates the count of the current
4492 * processor.
4493 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07004494 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004495
4496 /*
4497 * Zero the differential counters of the dead processor
4498 * so that the vm statistics are consistent.
4499 *
4500 * This is only okay since the processor is dead and cannot
4501 * race with what we are doing.
4502 */
Christoph Lameter2244b952006-06-30 01:55:33 -07004503 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004504 }
4505 return NOTIFY_OK;
4506}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004507
4508void __init page_alloc_init(void)
4509{
4510 hotcpu_notifier(page_alloc_cpu_notify, 0);
4511}
4512
4513/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004514 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4515 * or min_free_kbytes changes.
4516 */
4517static void calculate_totalreserve_pages(void)
4518{
4519 struct pglist_data *pgdat;
4520 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004521 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004522
4523 for_each_online_pgdat(pgdat) {
4524 for (i = 0; i < MAX_NR_ZONES; i++) {
4525 struct zone *zone = pgdat->node_zones + i;
4526 unsigned long max = 0;
4527
4528 /* Find valid and maximum lowmem_reserve in the zone */
4529 for (j = i; j < MAX_NR_ZONES; j++) {
4530 if (zone->lowmem_reserve[j] > max)
4531 max = zone->lowmem_reserve[j];
4532 }
4533
Mel Gorman41858962009-06-16 15:32:12 -07004534 /* we treat the high watermark as reserved pages. */
4535 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004536
4537 if (max > zone->present_pages)
4538 max = zone->present_pages;
4539 reserve_pages += max;
4540 }
4541 }
4542 totalreserve_pages = reserve_pages;
4543}
4544
4545/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07004546 * setup_per_zone_lowmem_reserve - called whenever
4547 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4548 * has a correct pages reserved value, so an adequate number of
4549 * pages are left in the zone after a successful __alloc_pages().
4550 */
4551static void setup_per_zone_lowmem_reserve(void)
4552{
4553 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004554 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004555
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08004556 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004557 for (j = 0; j < MAX_NR_ZONES; j++) {
4558 struct zone *zone = pgdat->node_zones + j;
4559 unsigned long present_pages = zone->present_pages;
4560
4561 zone->lowmem_reserve[j] = 0;
4562
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004563 idx = j;
4564 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004565 struct zone *lower_zone;
4566
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004567 idx--;
4568
Linus Torvalds1da177e2005-04-16 15:20:36 -07004569 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4570 sysctl_lowmem_reserve_ratio[idx] = 1;
4571
4572 lower_zone = pgdat->node_zones + idx;
4573 lower_zone->lowmem_reserve[j] = present_pages /
4574 sysctl_lowmem_reserve_ratio[idx];
4575 present_pages += lower_zone->present_pages;
4576 }
4577 }
4578 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004579
4580 /* update totalreserve_pages */
4581 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004582}
4583
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004584/**
Minchan Kimbc75d332009-06-16 15:32:48 -07004585 * setup_per_zone_wmarks - called when min_free_kbytes changes
Minchan Kimbce73942009-06-16 15:32:50 -07004586 * or when memory is hot-{added|removed}
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004587 *
Minchan Kimbc75d332009-06-16 15:32:48 -07004588 * Ensures that the watermark[min,low,high] values for each zone are set
4589 * correctly with respect to min_free_kbytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004590 */
Minchan Kimbc75d332009-06-16 15:32:48 -07004591void setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004592{
4593 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4594 unsigned long lowmem_pages = 0;
4595 struct zone *zone;
4596 unsigned long flags;
4597
4598 /* Calculate total number of !ZONE_HIGHMEM pages */
4599 for_each_zone(zone) {
4600 if (!is_highmem(zone))
4601 lowmem_pages += zone->present_pages;
4602 }
4603
4604 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07004605 u64 tmp;
4606
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07004607 spin_lock_irqsave(&zone->lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07004608 tmp = (u64)pages_min * zone->present_pages;
4609 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004610 if (is_highmem(zone)) {
4611 /*
Nick Piggin669ed172005-11-13 16:06:45 -08004612 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4613 * need highmem pages, so cap pages_min to a small
4614 * value here.
4615 *
Mel Gorman41858962009-06-16 15:32:12 -07004616 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Nick Piggin669ed172005-11-13 16:06:45 -08004617 * deltas controls asynch page reclaim, and so should
4618 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004619 */
4620 int min_pages;
4621
4622 min_pages = zone->present_pages / 1024;
4623 if (min_pages < SWAP_CLUSTER_MAX)
4624 min_pages = SWAP_CLUSTER_MAX;
4625 if (min_pages > 128)
4626 min_pages = 128;
Mel Gorman41858962009-06-16 15:32:12 -07004627 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004628 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08004629 /*
4630 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07004631 * proportionate to the zone's size.
4632 */
Mel Gorman41858962009-06-16 15:32:12 -07004633 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004634 }
4635
Mel Gorman41858962009-06-16 15:32:12 -07004636 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
4637 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Mel Gorman56fd56b2007-10-16 01:25:58 -07004638 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07004639 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004640 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004641
4642 /* update totalreserve_pages */
4643 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004644}
4645
Randy Dunlap55a44622009-09-21 17:01:20 -07004646/*
Rik van Riel556adec2008-10-18 20:26:34 -07004647 * The inactive anon list should be small enough that the VM never has to
4648 * do too much work, but large enough that each inactive page has a chance
4649 * to be referenced again before it is swapped out.
4650 *
4651 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4652 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4653 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4654 * the anonymous pages are kept on the inactive list.
4655 *
4656 * total target max
4657 * memory ratio inactive anon
4658 * -------------------------------------
4659 * 10MB 1 5MB
4660 * 100MB 1 50MB
4661 * 1GB 3 250MB
4662 * 10GB 10 0.9GB
4663 * 100GB 31 3GB
4664 * 1TB 101 10GB
4665 * 10TB 320 32GB
4666 */
Minchan Kim96cb4df2009-06-16 15:32:49 -07004667void calculate_zone_inactive_ratio(struct zone *zone)
4668{
4669 unsigned int gb, ratio;
4670
4671 /* Zone size in gigabytes */
4672 gb = zone->present_pages >> (30 - PAGE_SHIFT);
4673 if (gb)
4674 ratio = int_sqrt(10 * gb);
4675 else
4676 ratio = 1;
4677
4678 zone->inactive_ratio = ratio;
4679}
4680
Cyrill Gorcunove9bb35d2009-06-16 15:32:32 -07004681static void __init setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07004682{
4683 struct zone *zone;
4684
Minchan Kim96cb4df2009-06-16 15:32:49 -07004685 for_each_zone(zone)
4686 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07004687}
4688
Linus Torvalds1da177e2005-04-16 15:20:36 -07004689/*
4690 * Initialise min_free_kbytes.
4691 *
4692 * For small machines we want it small (128k min). For large machines
4693 * we want it large (64MB max). But it is not linear, because network
4694 * bandwidth does not increase linearly with machine size. We use
4695 *
4696 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4697 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4698 *
4699 * which yields
4700 *
4701 * 16MB: 512k
4702 * 32MB: 724k
4703 * 64MB: 1024k
4704 * 128MB: 1448k
4705 * 256MB: 2048k
4706 * 512MB: 2896k
4707 * 1024MB: 4096k
4708 * 2048MB: 5792k
4709 * 4096MB: 8192k
4710 * 8192MB: 11584k
4711 * 16384MB: 16384k
4712 */
Minchan Kimbc75d332009-06-16 15:32:48 -07004713static int __init init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004714{
4715 unsigned long lowmem_kbytes;
4716
4717 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4718
4719 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4720 if (min_free_kbytes < 128)
4721 min_free_kbytes = 128;
4722 if (min_free_kbytes > 65536)
4723 min_free_kbytes = 65536;
Minchan Kimbc75d332009-06-16 15:32:48 -07004724 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004725 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07004726 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004727 return 0;
4728}
Minchan Kimbc75d332009-06-16 15:32:48 -07004729module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004730
4731/*
4732 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4733 * that we can call two helper functions whenever min_free_kbytes
4734 * changes.
4735 */
4736int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004737 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004738{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004739 proc_dointvec(table, write, buffer, length, ppos);
Mel Gorman3b1d92c2007-05-06 14:49:30 -07004740 if (write)
Minchan Kimbc75d332009-06-16 15:32:48 -07004741 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004742 return 0;
4743}
4744
Christoph Lameter96146342006-07-03 00:24:13 -07004745#ifdef CONFIG_NUMA
4746int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004747 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07004748{
4749 struct zone *zone;
4750 int rc;
4751
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004752 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07004753 if (rc)
4754 return rc;
4755
4756 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07004757 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07004758 sysctl_min_unmapped_ratio) / 100;
4759 return 0;
4760}
Christoph Lameter0ff38492006-09-25 23:31:52 -07004761
4762int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004763 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07004764{
4765 struct zone *zone;
4766 int rc;
4767
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004768 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07004769 if (rc)
4770 return rc;
4771
4772 for_each_zone(zone)
4773 zone->min_slab_pages = (zone->present_pages *
4774 sysctl_min_slab_ratio) / 100;
4775 return 0;
4776}
Christoph Lameter96146342006-07-03 00:24:13 -07004777#endif
4778
Linus Torvalds1da177e2005-04-16 15:20:36 -07004779/*
4780 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4781 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4782 * whenever sysctl_lowmem_reserve_ratio changes.
4783 *
4784 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07004785 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07004786 * if in function of the boot time zone sizes.
4787 */
4788int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004789 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004790{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004791 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004792 setup_per_zone_lowmem_reserve();
4793 return 0;
4794}
4795
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004796/*
4797 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4798 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4799 * can have before it gets flushed back to buddy allocator.
4800 */
4801
4802int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004803 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004804{
4805 struct zone *zone;
4806 unsigned int cpu;
4807 int ret;
4808
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004809 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004810 if (!write || (ret == -EINVAL))
4811 return ret;
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07004812 for_each_populated_zone(zone) {
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004813 for_each_online_cpu(cpu) {
4814 unsigned long high;
4815 high = zone->present_pages / percpu_pagelist_fraction;
4816 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4817 }
4818 }
4819 return 0;
4820}
4821
David S. Millerf034b5d2006-08-24 03:08:07 -07004822int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004823
4824#ifdef CONFIG_NUMA
4825static int __init set_hashdist(char *str)
4826{
4827 if (!str)
4828 return 0;
4829 hashdist = simple_strtoul(str, &str, 0);
4830 return 1;
4831}
4832__setup("hashdist=", set_hashdist);
4833#endif
4834
4835/*
4836 * allocate a large system hash table from bootmem
4837 * - it is assumed that the hash table must contain an exact power-of-2
4838 * quantity of entries
4839 * - limit is the number of hash buckets, not the total allocation size
4840 */
4841void *__init alloc_large_system_hash(const char *tablename,
4842 unsigned long bucketsize,
4843 unsigned long numentries,
4844 int scale,
4845 int flags,
4846 unsigned int *_hash_shift,
4847 unsigned int *_hash_mask,
4848 unsigned long limit)
4849{
4850 unsigned long long max = limit;
4851 unsigned long log2qty, size;
4852 void *table = NULL;
4853
4854 /* allow the kernel cmdline to have a say */
4855 if (!numentries) {
4856 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08004857 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004858 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4859 numentries >>= 20 - PAGE_SHIFT;
4860 numentries <<= 20 - PAGE_SHIFT;
4861
4862 /* limit to 1 bucket per 2^scale bytes of low memory */
4863 if (scale > PAGE_SHIFT)
4864 numentries >>= (scale - PAGE_SHIFT);
4865 else
4866 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08004867
4868 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07004869 if (unlikely(flags & HASH_SMALL)) {
4870 /* Makes no sense without HASH_EARLY */
4871 WARN_ON(!(flags & HASH_EARLY));
4872 if (!(numentries >> *_hash_shift)) {
4873 numentries = 1UL << *_hash_shift;
4874 BUG_ON(!numentries);
4875 }
4876 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08004877 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004878 }
John Hawkes6e692ed2006-03-25 03:08:02 -08004879 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004880
4881 /* limit allocation size to 1/16 total memory by default */
4882 if (max == 0) {
4883 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4884 do_div(max, bucketsize);
4885 }
4886
4887 if (numentries > max)
4888 numentries = max;
4889
David Howellsf0d1b0b2006-12-08 02:37:49 -08004890 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004891
4892 do {
4893 size = bucketsize << log2qty;
4894 if (flags & HASH_EARLY)
Jan Beulich74768ed2008-08-12 15:08:39 -07004895 table = alloc_bootmem_nopanic(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004896 else if (hashdist)
4897 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4898 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07004899 /*
4900 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07004901 * some pages at the end of hash table which
4902 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07004903 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01004904 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07004905 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01004906 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
4907 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004908 }
4909 } while (!table && size > PAGE_SIZE && --log2qty);
4910
4911 if (!table)
4912 panic("Failed to allocate %s hash table\n", tablename);
4913
Dan Alonib49ad482007-07-15 23:38:23 -07004914 printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07004915 tablename,
4916 (1U << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08004917 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004918 size);
4919
4920 if (_hash_shift)
4921 *_hash_shift = log2qty;
4922 if (_hash_mask)
4923 *_hash_mask = (1 << log2qty) - 1;
4924
4925 return table;
4926}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004927
Mel Gorman835c1342007-10-16 01:25:47 -07004928/* Return a pointer to the bitmap storing bits affecting a block of pages */
4929static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4930 unsigned long pfn)
4931{
4932#ifdef CONFIG_SPARSEMEM
4933 return __pfn_to_section(pfn)->pageblock_flags;
4934#else
4935 return zone->pageblock_flags;
4936#endif /* CONFIG_SPARSEMEM */
4937}
Andrew Morton6220ec72006-10-19 23:29:05 -07004938
Mel Gorman835c1342007-10-16 01:25:47 -07004939static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4940{
4941#ifdef CONFIG_SPARSEMEM
4942 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07004943 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07004944#else
4945 pfn = pfn - zone->zone_start_pfn;
Mel Gormand9c23402007-10-16 01:26:01 -07004946 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07004947#endif /* CONFIG_SPARSEMEM */
4948}
4949
4950/**
Mel Gormand9c23402007-10-16 01:26:01 -07004951 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
Mel Gorman835c1342007-10-16 01:25:47 -07004952 * @page: The page within the block of interest
4953 * @start_bitidx: The first bit of interest to retrieve
4954 * @end_bitidx: The last bit of interest
4955 * returns pageblock_bits flags
4956 */
4957unsigned long get_pageblock_flags_group(struct page *page,
4958 int start_bitidx, int end_bitidx)
4959{
4960 struct zone *zone;
4961 unsigned long *bitmap;
4962 unsigned long pfn, bitidx;
4963 unsigned long flags = 0;
4964 unsigned long value = 1;
4965
4966 zone = page_zone(page);
4967 pfn = page_to_pfn(page);
4968 bitmap = get_pageblock_bitmap(zone, pfn);
4969 bitidx = pfn_to_bitidx(zone, pfn);
4970
4971 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4972 if (test_bit(bitidx + start_bitidx, bitmap))
4973 flags |= value;
4974
4975 return flags;
4976}
4977
4978/**
Mel Gormand9c23402007-10-16 01:26:01 -07004979 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
Mel Gorman835c1342007-10-16 01:25:47 -07004980 * @page: The page within the block of interest
4981 * @start_bitidx: The first bit of interest
4982 * @end_bitidx: The last bit of interest
4983 * @flags: The flags to set
4984 */
4985void set_pageblock_flags_group(struct page *page, unsigned long flags,
4986 int start_bitidx, int end_bitidx)
4987{
4988 struct zone *zone;
4989 unsigned long *bitmap;
4990 unsigned long pfn, bitidx;
4991 unsigned long value = 1;
4992
4993 zone = page_zone(page);
4994 pfn = page_to_pfn(page);
4995 bitmap = get_pageblock_bitmap(zone, pfn);
4996 bitidx = pfn_to_bitidx(zone, pfn);
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004997 VM_BUG_ON(pfn < zone->zone_start_pfn);
4998 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
Mel Gorman835c1342007-10-16 01:25:47 -07004999
5000 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5001 if (flags & value)
5002 __set_bit(bitidx + start_bitidx, bitmap);
5003 else
5004 __clear_bit(bitidx + start_bitidx, bitmap);
5005}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005006
5007/*
5008 * This is designed as sub function...plz see page_isolation.c also.
5009 * set/clear page block's type to be ISOLATE.
5010 * page allocater never alloc memory from ISOLATE block.
5011 */
5012
5013int set_migratetype_isolate(struct page *page)
5014{
5015 struct zone *zone;
Robert Jennings925cc712009-12-17 14:44:38 +00005016 struct page *curr_page;
5017 unsigned long flags, pfn, iter;
5018 unsigned long immobile = 0;
5019 struct memory_isolate_notify arg;
5020 int notifier_ret;
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005021 int ret = -EBUSY;
Shaohua Li8e7e40d2009-09-21 17:01:18 -07005022 int zone_idx;
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005023
5024 zone = page_zone(page);
Shaohua Li8e7e40d2009-09-21 17:01:18 -07005025 zone_idx = zone_idx(zone);
Robert Jennings925cc712009-12-17 14:44:38 +00005026
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005027 spin_lock_irqsave(&zone->lock, flags);
Robert Jennings925cc712009-12-17 14:44:38 +00005028 if (get_pageblock_migratetype(page) == MIGRATE_MOVABLE ||
5029 zone_idx == ZONE_MOVABLE) {
5030 ret = 0;
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005031 goto out;
Robert Jennings925cc712009-12-17 14:44:38 +00005032 }
5033
5034 pfn = page_to_pfn(page);
5035 arg.start_pfn = pfn;
5036 arg.nr_pages = pageblock_nr_pages;
5037 arg.pages_found = 0;
5038
5039 /*
5040 * It may be possible to isolate a pageblock even if the
5041 * migratetype is not MIGRATE_MOVABLE. The memory isolation
5042 * notifier chain is used by balloon drivers to return the
5043 * number of pages in a range that are held by the balloon
5044 * driver to shrink memory. If all the pages are accounted for
5045 * by balloons, are free, or on the LRU, isolation can continue.
5046 * Later, for example, when memory hotplug notifier runs, these
5047 * pages reported as "can be isolated" should be isolated(freed)
5048 * by the balloon driver through the memory notifier chain.
5049 */
5050 notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg);
5051 notifier_ret = notifier_to_errno(notifier_ret);
5052 if (notifier_ret || !arg.pages_found)
5053 goto out;
5054
5055 for (iter = pfn; iter < (pfn + pageblock_nr_pages); iter++) {
5056 if (!pfn_valid_within(pfn))
5057 continue;
5058
5059 curr_page = pfn_to_page(iter);
5060 if (!page_count(curr_page) || PageLRU(curr_page))
5061 continue;
5062
5063 immobile++;
5064 }
5065
5066 if (arg.pages_found == immobile)
5067 ret = 0;
5068
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005069out:
Robert Jennings925cc712009-12-17 14:44:38 +00005070 if (!ret) {
5071 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
5072 move_freepages_block(zone, page, MIGRATE_ISOLATE);
5073 }
5074
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005075 spin_unlock_irqrestore(&zone->lock, flags);
5076 if (!ret)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005077 drain_all_pages();
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005078 return ret;
5079}
5080
5081void unset_migratetype_isolate(struct page *page)
5082{
5083 struct zone *zone;
5084 unsigned long flags;
5085 zone = page_zone(page);
5086 spin_lock_irqsave(&zone->lock, flags);
5087 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
5088 goto out;
5089 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5090 move_freepages_block(zone, page, MIGRATE_MOVABLE);
5091out:
5092 spin_unlock_irqrestore(&zone->lock, flags);
5093}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07005094
5095#ifdef CONFIG_MEMORY_HOTREMOVE
5096/*
5097 * All pages in the range must be isolated before calling this.
5098 */
5099void
5100__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
5101{
5102 struct page *page;
5103 struct zone *zone;
5104 int order, i;
5105 unsigned long pfn;
5106 unsigned long flags;
5107 /* find the first valid pfn */
5108 for (pfn = start_pfn; pfn < end_pfn; pfn++)
5109 if (pfn_valid(pfn))
5110 break;
5111 if (pfn == end_pfn)
5112 return;
5113 zone = page_zone(pfn_to_page(pfn));
5114 spin_lock_irqsave(&zone->lock, flags);
5115 pfn = start_pfn;
5116 while (pfn < end_pfn) {
5117 if (!pfn_valid(pfn)) {
5118 pfn++;
5119 continue;
5120 }
5121 page = pfn_to_page(pfn);
5122 BUG_ON(page_count(page));
5123 BUG_ON(!PageBuddy(page));
5124 order = page_order(page);
5125#ifdef CONFIG_DEBUG_VM
5126 printk(KERN_INFO "remove from free list %lx %d %lx\n",
5127 pfn, 1 << order, end_pfn);
5128#endif
5129 list_del(&page->lru);
5130 rmv_page_order(page);
5131 zone->free_area[order].nr_free--;
5132 __mod_zone_page_state(zone, NR_FREE_PAGES,
5133 - (1UL << order));
5134 for (i = 0; i < (1 << order); i++)
5135 SetPageReserved((page+i));
5136 pfn += (1 << order);
5137 }
5138 spin_unlock_irqrestore(&zone->lock, flags);
5139}
5140#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01005141
5142#ifdef CONFIG_MEMORY_FAILURE
5143bool is_free_buddy_page(struct page *page)
5144{
5145 struct zone *zone = page_zone(page);
5146 unsigned long pfn = page_to_pfn(page);
5147 unsigned long flags;
5148 int order;
5149
5150 spin_lock_irqsave(&zone->lock, flags);
5151 for (order = 0; order < MAX_ORDER; order++) {
5152 struct page *page_head = page - (pfn & ((1 << order) - 1));
5153
5154 if (PageBuddy(page_head) && page_order(page_head) >= order)
5155 break;
5156 }
5157 spin_unlock_irqrestore(&zone->lock, flags);
5158
5159 return order < MAX_ORDER;
5160}
5161#endif