blob: bf720550b44d85adc294f7fd0b8ede38f73a8902 [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 Rientjes5a3135c22007-10-16 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 Hiroyukia5d76b542007-10-16 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>
Mel Gorman0d3d0622009-09-21 17:02:44 -070051#include <trace/events/kmem.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
53#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070054#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070055#include "internal.h"
56
57/*
Christoph Lameter13808912007-10-16 01:25:27 -070058 * Array of node states.
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 */
Christoph Lameter13808912007-10-16 01:25:27 -070060nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
61 [N_POSSIBLE] = NODE_MASK_ALL,
62 [N_ONLINE] = { { [0] = 1UL } },
63#ifndef CONFIG_NUMA
64 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
65#ifdef CONFIG_HIGHMEM
66 [N_HIGH_MEMORY] = { { [0] = 1UL } },
67#endif
68 [N_CPU] = { { [0] = 1UL } },
69#endif /* NUMA */
70};
71EXPORT_SYMBOL(node_states);
72
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070073unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -070074unsigned long totalreserve_pages __read_mostly;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -080075int percpu_pagelist_fraction;
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +100076gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -070077
Mel Gormand9c23402007-10-16 01:26:01 -070078#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
79int pageblock_order __read_mostly;
80#endif
81
Hugh Dickinsd98c7a02006-02-14 13:52:59 -080082static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -080083
Linus Torvalds1da177e2005-04-16 15:20:36 -070084/*
85 * results with 256, 32 in the lowmem_reserve sysctl:
86 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
87 * 1G machine -> (16M dma, 784M normal, 224M high)
88 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
89 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
90 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +010091 *
92 * TBD: should special case ZONE_DMA32 machines here - in those we normally
93 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -070095int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -080096#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -070097 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -080098#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -070099#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700100 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700101#endif
Christoph Lametere53ef382006-09-25 23:31:14 -0700102#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700103 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700104#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700105 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700106};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107
108EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109
Helge Deller15ad7cd2006-12-06 20:40:36 -0800110static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800111#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700112 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800113#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700114#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700115 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700116#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700117 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700118#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700119 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700120#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700121 "Movable",
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700122};
123
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124int min_free_kbytes = 1024;
125
Jan Beulich2c85f512009-09-21 17:03:07 -0700126static unsigned long __meminitdata nr_kernel_pages;
127static unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700128static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129
Mel Gormanc7132162006-09-27 01:49:43 -0700130#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
131 /*
Simon Arlott183ff222007-10-20 01:27:18 +0200132 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
Mel Gormanc7132162006-09-27 01:49:43 -0700133 * ranges of memory (RAM) that may be registered with add_active_range().
134 * Ranges passed to add_active_range() will be merged if possible
135 * so the number of times add_active_range() can be called is
136 * related to the number of nodes and the number of holes
137 */
138 #ifdef CONFIG_MAX_ACTIVE_REGIONS
139 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
140 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
141 #else
142 #if MAX_NUMNODES >= 32
143 /* If there can be many nodes, allow up to 50 holes per node */
144 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
145 #else
146 /* By default, allow up to 256 distinct regions */
147 #define MAX_ACTIVE_REGIONS 256
148 #endif
149 #endif
150
Jan Beulich98011f52007-07-15 23:38:17 -0700151 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
152 static int __meminitdata nr_nodemap_entries;
153 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
154 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
Adrian Bunkb69a7282008-07-23 21:28:12 -0700155 static unsigned long __initdata required_kernelcore;
Adrian Bunk484f51f2007-10-16 01:26:03 -0700156 static unsigned long __initdata required_movablecore;
Adrian Bunkb69a7282008-07-23 21:28:12 -0700157 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Mel Gorman2a1e2742007-07-17 04:03:12 -0700158
159 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
160 int movable_zone;
161 EXPORT_SYMBOL(movable_zone);
Mel Gormanc7132162006-09-27 01:49:43 -0700162#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
163
Miklos Szeredi418508c2007-05-23 13:57:55 -0700164#if MAX_NUMNODES > 1
165int nr_node_ids __read_mostly = MAX_NUMNODES;
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700166int nr_online_nodes __read_mostly = 1;
Miklos Szeredi418508c2007-05-23 13:57:55 -0700167EXPORT_SYMBOL(nr_node_ids);
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700168EXPORT_SYMBOL(nr_online_nodes);
Miklos Szeredi418508c2007-05-23 13:57:55 -0700169#endif
170
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700171int page_group_by_mobility_disabled __read_mostly;
172
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700173static void set_pageblock_migratetype(struct page *page, int migratetype)
174{
Mel Gorman49255c62009-06-16 15:31:58 -0700175
176 if (unlikely(page_group_by_mobility_disabled))
177 migratetype = MIGRATE_UNMOVABLE;
178
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700179 set_pageblock_flags_group(page, (unsigned long)migratetype,
180 PB_migrate, PB_migrate_end);
181}
182
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -0700183bool oom_killer_disabled __read_mostly;
184
Nick Piggin13e74442006-01-06 00:10:58 -0800185#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700186static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700188 int ret = 0;
189 unsigned seq;
190 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -0700191
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700192 do {
193 seq = zone_span_seqbegin(zone);
194 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
195 ret = 1;
196 else if (pfn < zone->zone_start_pfn)
197 ret = 1;
198 } while (zone_span_seqretry(zone, seq));
199
200 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700201}
202
203static int page_is_consistent(struct zone *zone, struct page *page)
204{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700205 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700206 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700208 return 0;
209
210 return 1;
211}
212/*
213 * Temporary debugging check for pages not lying within a given zone.
214 */
215static int bad_range(struct zone *zone, struct page *page)
216{
217 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700219 if (!page_is_consistent(zone, page))
220 return 1;
221
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 return 0;
223}
Nick Piggin13e74442006-01-06 00:10:58 -0800224#else
225static inline int bad_range(struct zone *zone, struct page *page)
226{
227 return 0;
228}
229#endif
230
Nick Piggin224abf92006-01-06 00:11:11 -0800231static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800233 static unsigned long resume;
234 static unsigned long nr_shown;
235 static unsigned long nr_unshown;
236
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200237 /* Don't complain about poisoned pages */
238 if (PageHWPoison(page)) {
239 __ClearPageBuddy(page);
240 return;
241 }
242
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800243 /*
244 * Allow a burst of 60 reports, then keep quiet for that minute;
245 * or allow a steady drip of one report per second.
246 */
247 if (nr_shown == 60) {
248 if (time_before(jiffies, resume)) {
249 nr_unshown++;
250 goto out;
251 }
252 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800253 printk(KERN_ALERT
254 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800255 nr_unshown);
256 nr_unshown = 0;
257 }
258 nr_shown = 0;
259 }
260 if (nr_shown++ == 0)
261 resume = jiffies + 60 * HZ;
262
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800263 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800264 current->comm, page_to_pfn(page));
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800265 printk(KERN_ALERT
Hugh Dickins3dc14742009-01-06 14:40:08 -0800266 "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n",
267 page, (void *)page->flags, page_count(page),
268 page_mapcount(page), page->mapping, page->index);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700269
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800271out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800272 /* Leave bad fields for debug, except PageBuddy could make trouble */
273 __ClearPageBuddy(page);
Randy Dunlap9f158332005-09-13 01:25:16 -0700274 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275}
276
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277/*
278 * Higher-order pages are called "compound pages". They are structured thusly:
279 *
280 * The first PAGE_SIZE page is called the "head page".
281 *
282 * The remaining PAGE_SIZE pages are called "tail pages".
283 *
284 * All pages have PG_compound set. All pages have their ->private pointing at
285 * the head page (even the head page has this).
286 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800287 * The first tail page's ->lru.next holds the address of the compound page's
288 * put_page() function. Its ->lru.prev holds the order of allocation.
289 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800291
292static void free_compound_page(struct page *page)
293{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700294 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800295}
296
Andi Kleen01ad1c02008-07-23 21:27:46 -0700297void prep_compound_page(struct page *page, unsigned long order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298{
299 int i;
300 int nr_pages = 1 << order;
301
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800302 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700303 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700304 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800305 for (i = 1; i < nr_pages; i++) {
306 struct page *p = page + i;
307
Christoph Lameterd85f3382007-05-06 14:49:39 -0700308 __SetPageTail(p);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700309 p->first_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 }
311}
312
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800313static int destroy_compound_page(struct page *page, unsigned long order)
Andy Whitcroft18229df2008-11-06 12:53:27 -0800314{
315 int i;
316 int nr_pages = 1 << order;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800317 int bad = 0;
Andy Whitcroft18229df2008-11-06 12:53:27 -0800318
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800319 if (unlikely(compound_order(page) != order) ||
320 unlikely(!PageHead(page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800321 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800322 bad++;
323 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324
Christoph Lameter6d777952007-05-06 14:49:40 -0700325 __ClearPageHead(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800326
Andy Whitcroft18229df2008-11-06 12:53:27 -0800327 for (i = 1; i < nr_pages; i++) {
328 struct page *p = page + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329
Alexey Zaytseve713a212009-01-10 02:47:57 +0300330 if (unlikely(!PageTail(p) || (p->first_page != page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800331 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800332 bad++;
333 }
Christoph Lameterd85f3382007-05-06 14:49:39 -0700334 __ClearPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800336
337 return bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339
Nick Piggin17cf4402006-03-22 00:08:41 -0800340static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
341{
342 int i;
343
Andrew Morton6626c5d2006-03-22 00:08:42 -0800344 /*
345 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
346 * and __GFP_HIGHMEM from hard or soft interrupt context.
347 */
Nick Piggin725d7042006-09-25 23:30:55 -0700348 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800349 for (i = 0; i < (1 << order); i++)
350 clear_highpage(page + i);
351}
352
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700353static inline void set_page_order(struct page *page, int order)
354{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700355 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000356 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357}
358
359static inline void rmv_page_order(struct page *page)
360{
Nick Piggin676165a2006-04-10 11:21:48 +1000361 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700362 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363}
364
365/*
366 * Locate the struct page for both the matching buddy in our
367 * pair (buddy1) and the combined O(n+1) page they form (page).
368 *
369 * 1) Any buddy B1 will have an order O twin B2 which satisfies
370 * the following equation:
371 * B2 = B1 ^ (1 << O)
372 * For example, if the starting buddy (buddy2) is #8 its order
373 * 1 buddy is #10:
374 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
375 *
376 * 2) Any buddy B will have an order O+1 parent P which
377 * satisfies the following equation:
378 * P = B & ~(1 << O)
379 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200380 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 */
382static inline struct page *
383__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
384{
385 unsigned long buddy_idx = page_idx ^ (1 << order);
386
387 return page + (buddy_idx - page_idx);
388}
389
390static inline unsigned long
391__find_combined_index(unsigned long page_idx, unsigned int order)
392{
393 return (page_idx & ~(1 << order));
394}
395
396/*
397 * This function checks whether a page is free && is the buddy
398 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800399 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000400 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700401 * (c) a page and its buddy have the same order &&
402 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 *
Nick Piggin676165a2006-04-10 11:21:48 +1000404 * For recording whether a page is in the buddy system, we use PG_buddy.
405 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
406 *
407 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700409static inline int page_is_buddy(struct page *page, struct page *buddy,
410 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700412 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800413 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800414
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700415 if (page_zone_id(page) != page_zone_id(buddy))
416 return 0;
417
418 if (PageBuddy(buddy) && page_order(buddy) == order) {
Nick Piggina3af9c32009-06-16 15:32:10 -0700419 VM_BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700420 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000421 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700422 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423}
424
425/*
426 * Freeing function for a buddy system allocator.
427 *
428 * The concept of a buddy system is to maintain direct-mapped table
429 * (containing bit values) for memory blocks of various "orders".
430 * The bottom level table contains the map for the smallest allocatable
431 * units of memory (here, pages), and each level above it describes
432 * pairs of units from the levels below, hence, "buddies".
433 * At a high level, all that happens here is marking the table entry
434 * at the bottom level available, and propagating the changes upward
435 * as necessary, plus some accounting needed to play nicely with other
436 * parts of the VM system.
437 * At each level, we keep a list of pages, which are heads of continuous
Nick Piggin676165a2006-04-10 11:21:48 +1000438 * free pages of length of (1 << order) and marked with PG_buddy. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700439 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 * So when we are allocating or freeing one, we can derive the state of the
441 * other. That is, if we allocate a small block, and both were
442 * free, the remainder of the region must be split into blocks.
443 * If a block is freed, and its buddy is also free, then this
444 * triggers coalescing into a block of larger size.
445 *
446 * -- wli
447 */
448
Nick Piggin48db57f2006-01-08 01:00:42 -0800449static inline void __free_one_page(struct page *page,
Mel Gormaned0ae212009-06-16 15:32:07 -0700450 struct zone *zone, unsigned int order,
451 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452{
453 unsigned long page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454
Nick Piggin224abf92006-01-06 00:11:11 -0800455 if (unlikely(PageCompound(page)))
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800456 if (unlikely(destroy_compound_page(page, order)))
457 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458
Mel Gormaned0ae212009-06-16 15:32:07 -0700459 VM_BUG_ON(migratetype == -1);
460
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
462
Mel Gormanf2260e62009-06-16 15:32:13 -0700463 VM_BUG_ON(page_idx & ((1 << order) - 1));
Nick Piggin725d7042006-09-25 23:30:55 -0700464 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 while (order < MAX_ORDER-1) {
467 unsigned long combined_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 struct page *buddy;
469
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 buddy = __page_find_buddy(page, page_idx, order);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700471 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700472 break;
Nick Piggin13e74442006-01-06 00:10:58 -0800473
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700474 /* Our buddy is free, merge with it and move up one order. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 list_del(&buddy->lru);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700476 zone->free_area[order].nr_free--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 rmv_page_order(buddy);
Nick Piggin13e74442006-01-06 00:10:58 -0800478 combined_idx = __find_combined_index(page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 page = page + (combined_idx - page_idx);
480 page_idx = combined_idx;
481 order++;
482 }
483 set_page_order(page, order);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700484 list_add(&page->lru,
485 &zone->free_area[order].free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 zone->free_area[order].nr_free++;
487}
488
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700489#ifdef CONFIG_HAVE_MLOCKED_PAGE_BIT
490/*
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}
500#else
501static void free_page_mlock(struct page *page) { }
502#endif
503
Nick Piggin224abf92006-01-06 00:11:11 -0800504static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505{
Nick Piggin92be2e332006-01-06 00:10:57 -0800506 if (unlikely(page_mapcount(page) |
507 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700508 (atomic_read(&page->_count) != 0) |
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800509 (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800510 bad_page(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800511 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800512 }
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800513 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
514 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
515 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516}
517
518/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700519 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700521 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 *
523 * If the zone was previously in an "all pages pinned" state then look to
524 * see if this freeing clears that state.
525 *
526 * And clear the zone's pages_scanned counter, to hold off the "all pages are
527 * pinned" detection logic.
528 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700529static void free_pcppages_bulk(struct zone *zone, int count,
530 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531{
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700532 int migratetype = 0;
Mel Gormana6f9edd62009-09-21 17:03:20 -0700533 int batch_free = 0;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700534
Nick Pigginc54ad302006-01-06 00:10:56 -0800535 spin_lock(&zone->lock);
David Rientjese815af92007-10-16 23:25:54 -0700536 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700538
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700539 __mod_zone_page_state(zone, NR_FREE_PAGES, count);
Mel Gormana6f9edd62009-09-21 17:03:20 -0700540 while (count) {
Nick Piggin48db57f2006-01-08 01:00:42 -0800541 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700542 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -0800543
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700544 /*
Mel Gormana6f9edd62009-09-21 17:03:20 -0700545 * Remove pages from lists in a round-robin fashion. A
546 * batch_free count is maintained that is incremented when an
547 * empty list is encountered. This is so more pages are freed
548 * off fuller lists instead of spinning excessively around empty
549 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700550 */
551 do {
Mel Gormana6f9edd62009-09-21 17:03:20 -0700552 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700553 if (++migratetype == MIGRATE_PCPTYPES)
554 migratetype = 0;
555 list = &pcp->lists[migratetype];
556 } while (list_empty(list));
557
Mel Gormana6f9edd62009-09-21 17:03:20 -0700558 do {
559 page = list_entry(list->prev, struct page, lru);
560 /* must delete as __free_one_page list manipulates */
561 list_del(&page->lru);
562 __free_one_page(page, zone, 0, migratetype);
563 trace_mm_page_pcpu_drain(page, 0, migratetype);
564 } while (--count && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800566 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567}
568
Mel Gormaned0ae212009-06-16 15:32:07 -0700569static void free_one_page(struct zone *zone, struct page *page, int order,
570 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800571{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700572 spin_lock(&zone->lock);
David Rientjese815af92007-10-16 23:25:54 -0700573 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700574 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700575
576 __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
Mel Gormaned0ae212009-06-16 15:32:07 -0700577 __free_one_page(page, zone, order, migratetype);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700578 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800579}
580
581static void __free_pages_ok(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582{
Nick Pigginc54ad302006-01-06 00:10:56 -0800583 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 int i;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800585 int bad = 0;
Johannes Weiner451ea252009-09-21 17:01:48 -0700586 int wasMlocked = __TestClearPageMlocked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100588 kmemcheck_free_shadow(page, order);
589
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 for (i = 0 ; i < (1 << order) ; ++i)
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800591 bad += free_pages_check(page + i);
592 if (bad)
Hugh Dickins689bceb2005-11-21 21:32:20 -0800593 return;
594
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700595 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -0700596 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700597 debug_check_no_obj_freed(page_address(page),
598 PAGE_SIZE << order);
599 }
Nick Piggindafb1362006-10-11 01:21:30 -0700600 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800601 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700602
Nick Pigginc54ad302006-01-06 00:10:56 -0800603 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +0200604 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -0700605 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700606 __count_vm_events(PGFREE, 1 << order);
Mel Gormaned0ae212009-06-16 15:32:07 -0700607 free_one_page(page_zone(page), page, order,
608 get_pageblock_migratetype(page));
Nick Pigginc54ad302006-01-06 00:10:56 -0800609 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610}
611
David Howellsa226f6c2006-01-06 00:11:08 -0800612/*
613 * permit the bootmem allocator to evade page validation on high-order frees
614 */
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700615void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800616{
617 if (order == 0) {
618 __ClearPageReserved(page);
619 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800620 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800621 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800622 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800623 int loop;
624
Nick Piggin545b1ea2006-03-22 00:08:07 -0800625 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800626 for (loop = 0; loop < BITS_PER_LONG; loop++) {
627 struct page *p = &page[loop];
628
Nick Piggin545b1ea2006-03-22 00:08:07 -0800629 if (loop + 1 < BITS_PER_LONG)
630 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800631 __ClearPageReserved(p);
632 set_page_count(p, 0);
633 }
634
Nick Piggin7835e982006-03-22 00:08:40 -0800635 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800636 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800637 }
638}
639
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640
641/*
642 * The order of subdivision here is critical for the IO subsystem.
643 * Please do not alter this order without good reasons and regression
644 * testing. Specifically, as large blocks of memory are subdivided,
645 * the order in which smaller blocks are delivered depends on the order
646 * they're subdivided in this function. This is the primary factor
647 * influencing the order in which pages are delivered to the IO
648 * subsystem according to empirical testing, and this is also justified
649 * by considering the behavior of a buddy system containing a single
650 * large block of memory acted on by a series of small allocations.
651 * This behavior is a critical factor in sglist merging's success.
652 *
653 * -- wli
654 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800655static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700656 int low, int high, struct free_area *area,
657 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658{
659 unsigned long size = 1 << high;
660
661 while (high > low) {
662 area--;
663 high--;
664 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700665 VM_BUG_ON(bad_range(zone, &page[size]));
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700666 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 area->nr_free++;
668 set_page_order(&page[size], high);
669 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670}
671
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672/*
673 * This page is about to be returned from the page allocator
674 */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200675static inline int check_new_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676{
Nick Piggin92be2e332006-01-06 00:10:57 -0800677 if (unlikely(page_mapcount(page) |
678 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700679 (atomic_read(&page->_count) != 0) |
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800680 (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800681 bad_page(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800682 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800683 }
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200684 return 0;
685}
686
687static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
688{
689 int i;
690
691 for (i = 0; i < (1 << order); i++) {
692 struct page *p = page + i;
693 if (unlikely(check_new_page(p)))
694 return 1;
695 }
Hugh Dickins689bceb2005-11-21 21:32:20 -0800696
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700697 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800698 set_page_refcounted(page);
Nick Piggincc1025092006-12-06 20:32:00 -0800699
700 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800702
703 if (gfp_flags & __GFP_ZERO)
704 prep_zero_page(page, order, gfp_flags);
705
706 if (order && (gfp_flags & __GFP_COMP))
707 prep_compound_page(page, order);
708
Hugh Dickins689bceb2005-11-21 21:32:20 -0800709 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710}
711
Mel Gorman56fd56b2007-10-16 01:25:58 -0700712/*
713 * Go through the free lists for the given migratetype and remove
714 * the smallest available page from the freelists
715 */
Mel Gorman728ec982009-06-16 15:32:04 -0700716static inline
717struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -0700718 int migratetype)
719{
720 unsigned int current_order;
721 struct free_area * area;
722 struct page *page;
723
724 /* Find a page of the appropriate size in the preferred list */
725 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
726 area = &(zone->free_area[current_order]);
727 if (list_empty(&area->free_list[migratetype]))
728 continue;
729
730 page = list_entry(area->free_list[migratetype].next,
731 struct page, lru);
732 list_del(&page->lru);
733 rmv_page_order(page);
734 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -0700735 expand(zone, page, order, current_order, area, migratetype);
736 return page;
737 }
738
739 return NULL;
740}
741
742
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700743/*
744 * This array describes the order lists are fallen back to when
745 * the free lists for the desirable migrate type are depleted
746 */
747static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
Mel Gorman64c5e132007-10-16 01:25:59 -0700748 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
749 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
750 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
751 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700752};
753
Mel Gormanc361be52007-10-16 01:25:51 -0700754/*
755 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -0700756 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -0700757 * boundary. If alignment is required, use move_freepages_block()
758 */
Adrian Bunkb69a7282008-07-23 21:28:12 -0700759static int move_freepages(struct zone *zone,
760 struct page *start_page, struct page *end_page,
761 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700762{
763 struct page *page;
764 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -0700765 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -0700766
767#ifndef CONFIG_HOLES_IN_ZONE
768 /*
769 * page_zone is not safe to call in this context when
770 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
771 * anyway as we check zone boundaries in move_freepages_block().
772 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -0700773 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -0700774 */
775 BUG_ON(page_zone(start_page) != page_zone(end_page));
776#endif
777
778 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -0700779 /* Make sure we are not inadvertently changing nodes */
780 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
781
Mel Gormanc361be52007-10-16 01:25:51 -0700782 if (!pfn_valid_within(page_to_pfn(page))) {
783 page++;
784 continue;
785 }
786
787 if (!PageBuddy(page)) {
788 page++;
789 continue;
790 }
791
792 order = page_order(page);
793 list_del(&page->lru);
794 list_add(&page->lru,
795 &zone->free_area[order].free_list[migratetype]);
796 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -0700797 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -0700798 }
799
Mel Gormand1003132007-10-16 01:26:00 -0700800 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -0700801}
802
Adrian Bunkb69a7282008-07-23 21:28:12 -0700803static int move_freepages_block(struct zone *zone, struct page *page,
804 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700805{
806 unsigned long start_pfn, end_pfn;
807 struct page *start_page, *end_page;
808
809 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -0700810 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -0700811 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -0700812 end_page = start_page + pageblock_nr_pages - 1;
813 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -0700814
815 /* Do not cross zone boundaries */
816 if (start_pfn < zone->zone_start_pfn)
817 start_page = page;
818 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
819 return 0;
820
821 return move_freepages(zone, start_page, end_page, migratetype);
822}
823
Mel Gorman2f66a682009-09-21 17:02:31 -0700824static void change_pageblock_range(struct page *pageblock_page,
825 int start_order, int migratetype)
826{
827 int nr_pageblocks = 1 << (start_order - pageblock_order);
828
829 while (nr_pageblocks--) {
830 set_pageblock_migratetype(pageblock_page, migratetype);
831 pageblock_page += pageblock_nr_pages;
832 }
833}
834
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700835/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -0700836static inline struct page *
837__rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700838{
839 struct free_area * area;
840 int current_order;
841 struct page *page;
842 int migratetype, i;
843
844 /* Find the largest possible block of pages in the other list */
845 for (current_order = MAX_ORDER-1; current_order >= order;
846 --current_order) {
847 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
848 migratetype = fallbacks[start_migratetype][i];
849
Mel Gorman56fd56b2007-10-16 01:25:58 -0700850 /* MIGRATE_RESERVE handled later if necessary */
851 if (migratetype == MIGRATE_RESERVE)
852 continue;
Mel Gormane0104872007-10-16 01:25:53 -0700853
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700854 area = &(zone->free_area[current_order]);
855 if (list_empty(&area->free_list[migratetype]))
856 continue;
857
858 page = list_entry(area->free_list[migratetype].next,
859 struct page, lru);
860 area->nr_free--;
861
862 /*
Mel Gormanc361be52007-10-16 01:25:51 -0700863 * If breaking a large block of pages, move all free
Mel Gorman46dafbc2007-10-16 01:25:55 -0700864 * pages to the preferred allocation list. If falling
865 * back for a reclaimable kernel allocation, be more
866 * agressive about taking ownership of free pages
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700867 */
Mel Gormand9c23402007-10-16 01:26:01 -0700868 if (unlikely(current_order >= (pageblock_order >> 1)) ||
Mel Gormandd5d2412009-09-05 11:17:11 -0700869 start_migratetype == MIGRATE_RECLAIMABLE ||
870 page_group_by_mobility_disabled) {
Mel Gorman46dafbc2007-10-16 01:25:55 -0700871 unsigned long pages;
872 pages = move_freepages_block(zone, page,
873 start_migratetype);
874
875 /* Claim the whole block if over half of it is free */
Mel Gormandd5d2412009-09-05 11:17:11 -0700876 if (pages >= (1 << (pageblock_order-1)) ||
877 page_group_by_mobility_disabled)
Mel Gorman46dafbc2007-10-16 01:25:55 -0700878 set_pageblock_migratetype(page,
879 start_migratetype);
880
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700881 migratetype = start_migratetype;
Mel Gormanc361be52007-10-16 01:25:51 -0700882 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700883
884 /* Remove the page from the freelists */
885 list_del(&page->lru);
886 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700887
Mel Gorman2f66a682009-09-21 17:02:31 -0700888 /* Take ownership for orders >= pageblock_order */
889 if (current_order >= pageblock_order)
890 change_pageblock_range(page, current_order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700891 start_migratetype);
892
893 expand(zone, page, order, current_order, area, migratetype);
Mel Gormane0fff1b2009-09-21 17:02:42 -0700894
895 trace_mm_page_alloc_extfrag(page, order, current_order,
896 start_migratetype, migratetype);
897
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700898 return page;
899 }
900 }
901
Mel Gorman728ec982009-06-16 15:32:04 -0700902 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700903}
904
Mel Gorman56fd56b2007-10-16 01:25:58 -0700905/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 * Do the hard work of removing an element from the buddy allocator.
907 * Call me with the zone->lock already held.
908 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700909static struct page *__rmqueue(struct zone *zone, unsigned int order,
910 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 struct page *page;
913
Mel Gorman728ec982009-06-16 15:32:04 -0700914retry_reserve:
Mel Gorman56fd56b2007-10-16 01:25:58 -0700915 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916
Mel Gorman728ec982009-06-16 15:32:04 -0700917 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Mel Gorman56fd56b2007-10-16 01:25:58 -0700918 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700919
Mel Gorman728ec982009-06-16 15:32:04 -0700920 /*
921 * Use MIGRATE_RESERVE rather than fail an allocation. goto
922 * is used because __rmqueue_smallest is an inline function
923 * and we want just one call site
924 */
925 if (!page) {
926 migratetype = MIGRATE_RESERVE;
927 goto retry_reserve;
928 }
929 }
930
Mel Gorman0d3d0622009-09-21 17:02:44 -0700931 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700932 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933}
934
935/*
936 * Obtain a specified number of elements from the buddy allocator, all under
937 * a single hold of the lock, for efficiency. Add them to the supplied list.
938 * Returns the number of new pages which were placed at *list.
939 */
940static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700941 unsigned long count, struct list_head *list,
Mel Gormane084b2d2009-07-29 15:02:04 -0700942 int migratetype, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945
Nick Pigginc54ad302006-01-06 00:10:56 -0800946 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700948 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800949 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -0800951
952 /*
953 * Split buddy pages returned by expand() are received here
954 * in physical page order. The page is added to the callers and
955 * list and the list head then moves forward. From the callers
956 * perspective, the linked list is ordered by page number in
957 * some conditions. This is useful for IO devices that can
958 * merge IO requests if the physical pages are ordered
959 * properly.
960 */
Mel Gormane084b2d2009-07-29 15:02:04 -0700961 if (likely(cold == 0))
962 list_add(&page->lru, list);
963 else
964 list_add_tail(&page->lru, list);
Mel Gorman535131e62007-10-16 01:25:49 -0700965 set_page_private(page, migratetype);
Mel Gorman81eabcb2007-12-17 16:20:05 -0800966 list = &page->lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 }
Mel Gormanf2260e62009-06-16 15:32:13 -0700968 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -0800969 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800970 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971}
972
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700973#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800974/*
Christoph Lameter4037d452007-05-09 02:35:14 -0700975 * Called from the vmstat counter updater to drain pagesets of this
976 * currently executing processor on remote nodes after they have
977 * expired.
978 *
Christoph Lameter879336c2006-03-22 00:09:08 -0800979 * Note that this function must be called with the thread pinned to
980 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800981 */
Christoph Lameter4037d452007-05-09 02:35:14 -0700982void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700983{
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700984 unsigned long flags;
Christoph Lameter4037d452007-05-09 02:35:14 -0700985 int to_drain;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700986
Christoph Lameter4037d452007-05-09 02:35:14 -0700987 local_irq_save(flags);
988 if (pcp->count >= pcp->batch)
989 to_drain = pcp->batch;
990 else
991 to_drain = pcp->count;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700992 free_pcppages_bulk(zone, to_drain, pcp);
Christoph Lameter4037d452007-05-09 02:35:14 -0700993 pcp->count -= to_drain;
994 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700995}
996#endif
997
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800998/*
999 * Drain pages of the indicated processor.
1000 *
1001 * The processor must either be the current processor and the
1002 * thread pinned to the current processor or a processor that
1003 * is not online.
1004 */
1005static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006{
Nick Pigginc54ad302006-01-06 00:10:56 -08001007 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07001010 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 struct per_cpu_pageset *pset;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001012 struct per_cpu_pages *pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001014 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001016 pcp = &pset->pcp;
1017 local_irq_save(flags);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001018 free_pcppages_bulk(zone, pcp->count, pcp);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001019 pcp->count = 0;
1020 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 }
1022}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001024/*
1025 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1026 */
1027void drain_local_pages(void *arg)
1028{
1029 drain_pages(smp_processor_id());
1030}
1031
1032/*
1033 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
1034 */
1035void drain_all_pages(void)
1036{
Jens Axboe15c8b6c2008-05-09 09:39:44 +02001037 on_each_cpu(drain_local_pages, NULL, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001038}
1039
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02001040#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041
1042void mark_free_pages(struct zone *zone)
1043{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001044 unsigned long pfn, max_zone_pfn;
1045 unsigned long flags;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001046 int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 struct list_head *curr;
1048
1049 if (!zone->spanned_pages)
1050 return;
1051
1052 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001053
1054 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1055 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1056 if (pfn_valid(pfn)) {
1057 struct page *page = pfn_to_page(pfn);
1058
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001059 if (!swsusp_page_is_forbidden(page))
1060 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001061 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001063 for_each_migratetype_order(order, t) {
1064 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001065 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001067 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1068 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001069 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001070 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001071 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 spin_unlock_irqrestore(&zone->lock, flags);
1073}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001074#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075
1076/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 * Free a 0-order page
1078 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001079static void free_hot_cold_page(struct page *page, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080{
1081 struct zone *zone = page_zone(page);
1082 struct per_cpu_pages *pcp;
1083 unsigned long flags;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001084 int migratetype;
Johannes Weiner451ea252009-09-21 17:01:48 -07001085 int wasMlocked = __TestClearPageMlocked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001087 kmemcheck_free_shadow(page, 0);
1088
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 if (PageAnon(page))
1090 page->mapping = NULL;
Nick Piggin224abf92006-01-06 00:11:11 -08001091 if (free_pages_check(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001092 return;
1093
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001094 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -07001095 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001096 debug_check_no_obj_freed(page_address(page), PAGE_SIZE);
1097 }
Nick Piggindafb1362006-10-11 01:21:30 -07001098 arch_free_page(page, 0);
Hugh Dickins689bceb2005-11-21 21:32:20 -08001099 kernel_map_pages(page, 1, 0);
1100
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001101 pcp = &zone_pcp(zone, get_cpu())->pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001102 migratetype = get_pageblock_migratetype(page);
1103 set_page_private(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +02001105 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -07001106 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001107 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001108
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001109 /*
1110 * We only track unmovable, reclaimable and movable on pcp lists.
1111 * Free ISOLATE pages back to the allocator because they are being
1112 * offlined but treat RESERVE as movable pages so we can get those
1113 * areas back if necessary. Otherwise, we may have to free
1114 * excessively into the page allocator
1115 */
1116 if (migratetype >= MIGRATE_PCPTYPES) {
1117 if (unlikely(migratetype == MIGRATE_ISOLATE)) {
1118 free_one_page(zone, page, 0, migratetype);
1119 goto out;
1120 }
1121 migratetype = MIGRATE_MOVABLE;
1122 }
1123
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001124 if (cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001125 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001126 else
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001127 list_add(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001129 if (pcp->count >= pcp->high) {
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001130 free_pcppages_bulk(zone, pcp->batch, pcp);
Nick Piggin48db57f2006-01-08 01:00:42 -08001131 pcp->count -= pcp->batch;
1132 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001133
1134out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 local_irq_restore(flags);
1136 put_cpu();
1137}
1138
Harvey Harrison920c7a52008-02-04 22:29:26 -08001139void free_hot_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140{
Mel Gorman4b4f2782009-09-21 17:02:41 -07001141 trace_mm_page_free_direct(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 free_hot_cold_page(page, 0);
1143}
1144
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001145/*
1146 * split_page takes a non-compound higher-order page, and splits it into
1147 * n (1<<order) sub-pages: page[0..n]
1148 * Each sub-page must be freed individually.
1149 *
1150 * Note: this is probably too low level an operation for use in drivers.
1151 * Please consult with lkml before using this in your driver.
1152 */
1153void split_page(struct page *page, unsigned int order)
1154{
1155 int i;
1156
Nick Piggin725d7042006-09-25 23:30:55 -07001157 VM_BUG_ON(PageCompound(page));
1158 VM_BUG_ON(!page_count(page));
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001159
1160#ifdef CONFIG_KMEMCHECK
1161 /*
1162 * Split shadow pages too, because free(page[0]) would
1163 * otherwise free the whole shadow.
1164 */
1165 if (kmemcheck_page_is_tracked(page))
1166 split_page(virt_to_page(page[0].shadow), order);
1167#endif
1168
Nick Piggin7835e982006-03-22 00:08:40 -08001169 for (i = 1; i < (1 << order); i++)
1170 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001171}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001172
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173/*
1174 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1175 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1176 * or two.
1177 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07001178static inline
1179struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001180 struct zone *zone, int order, gfp_t gfp_flags,
1181 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182{
1183 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001184 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 int cold = !!(gfp_flags & __GFP_COLD);
Nick Piggina74609f2006-01-06 00:11:20 -08001186 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187
Hugh Dickins689bceb2005-11-21 21:32:20 -08001188again:
Nick Piggina74609f2006-01-06 00:11:20 -08001189 cpu = get_cpu();
Nick Piggin48db57f2006-01-08 01:00:42 -08001190 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001192 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001194 pcp = &zone_pcp(zone, cpu)->pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001195 list = &pcp->lists[migratetype];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 local_irq_save(flags);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001197 if (list_empty(list)) {
Mel Gorman535131e62007-10-16 01:25:49 -07001198 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001199 pcp->batch, list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001200 migratetype, cold);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001201 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07001202 goto failed;
Mel Gorman535131e62007-10-16 01:25:49 -07001203 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001204
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001205 if (cold)
1206 page = list_entry(list->prev, struct page, lru);
1207 else
1208 page = list_entry(list->next, struct page, lru);
1209
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001210 list_del(&page->lru);
1211 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001212 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07001213 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1214 /*
1215 * __GFP_NOFAIL is not to be used in new code.
1216 *
1217 * All __GFP_NOFAIL callers should be fixed so that they
1218 * properly detect and handle allocation failures.
1219 *
1220 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07001221 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07001222 * __GFP_NOFAIL.
1223 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07001224 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07001225 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001227 page = __rmqueue(zone, order, migratetype);
Mel Gormanf2260e62009-06-16 15:32:13 -07001228 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order));
Nick Piggina74609f2006-01-06 00:11:20 -08001229 spin_unlock(&zone->lock);
1230 if (!page)
1231 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 }
1233
Christoph Lameterf8891e52006-06-30 01:55:45 -07001234 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Mel Gorman18ea7e72008-04-28 02:12:14 -07001235 zone_statistics(preferred_zone, zone);
Nick Piggina74609f2006-01-06 00:11:20 -08001236 local_irq_restore(flags);
1237 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238
Nick Piggin725d7042006-09-25 23:30:55 -07001239 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -08001240 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001241 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001243
1244failed:
1245 local_irq_restore(flags);
1246 put_cpu();
1247 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248}
1249
Mel Gorman41858962009-06-16 15:32:12 -07001250/* The ALLOC_WMARK bits are used as an index to zone->watermark */
1251#define ALLOC_WMARK_MIN WMARK_MIN
1252#define ALLOC_WMARK_LOW WMARK_LOW
1253#define ALLOC_WMARK_HIGH WMARK_HIGH
1254#define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1255
1256/* Mask to get the watermark bits */
1257#define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1258
Nick Piggin31488902005-11-28 13:44:03 -08001259#define ALLOC_HARDER 0x10 /* try to alloc harder */
1260#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1261#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001262
Akinobu Mita933e3122006-12-08 02:39:45 -08001263#ifdef CONFIG_FAIL_PAGE_ALLOC
1264
1265static struct fail_page_alloc_attr {
1266 struct fault_attr attr;
1267
1268 u32 ignore_gfp_highmem;
1269 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001270 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001271
1272#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1273
1274 struct dentry *ignore_gfp_highmem_file;
1275 struct dentry *ignore_gfp_wait_file;
Akinobu Mita54114992007-07-15 23:40:23 -07001276 struct dentry *min_order_file;
Akinobu Mita933e3122006-12-08 02:39:45 -08001277
1278#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1279
1280} fail_page_alloc = {
1281 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001282 .ignore_gfp_wait = 1,
1283 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001284 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001285};
1286
1287static int __init setup_fail_page_alloc(char *str)
1288{
1289 return setup_fault_attr(&fail_page_alloc.attr, str);
1290}
1291__setup("fail_page_alloc=", setup_fail_page_alloc);
1292
1293static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1294{
Akinobu Mita54114992007-07-15 23:40:23 -07001295 if (order < fail_page_alloc.min_order)
1296 return 0;
Akinobu Mita933e3122006-12-08 02:39:45 -08001297 if (gfp_mask & __GFP_NOFAIL)
1298 return 0;
1299 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1300 return 0;
1301 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1302 return 0;
1303
1304 return should_fail(&fail_page_alloc.attr, 1 << order);
1305}
1306
1307#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1308
1309static int __init fail_page_alloc_debugfs(void)
1310{
1311 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1312 struct dentry *dir;
1313 int err;
1314
1315 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1316 "fail_page_alloc");
1317 if (err)
1318 return err;
1319 dir = fail_page_alloc.attr.dentries.dir;
1320
1321 fail_page_alloc.ignore_gfp_wait_file =
1322 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1323 &fail_page_alloc.ignore_gfp_wait);
1324
1325 fail_page_alloc.ignore_gfp_highmem_file =
1326 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1327 &fail_page_alloc.ignore_gfp_highmem);
Akinobu Mita54114992007-07-15 23:40:23 -07001328 fail_page_alloc.min_order_file =
1329 debugfs_create_u32("min-order", mode, dir,
1330 &fail_page_alloc.min_order);
Akinobu Mita933e3122006-12-08 02:39:45 -08001331
1332 if (!fail_page_alloc.ignore_gfp_wait_file ||
Akinobu Mita54114992007-07-15 23:40:23 -07001333 !fail_page_alloc.ignore_gfp_highmem_file ||
1334 !fail_page_alloc.min_order_file) {
Akinobu Mita933e3122006-12-08 02:39:45 -08001335 err = -ENOMEM;
1336 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1337 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
Akinobu Mita54114992007-07-15 23:40:23 -07001338 debugfs_remove(fail_page_alloc.min_order_file);
Akinobu Mita933e3122006-12-08 02:39:45 -08001339 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1340 }
1341
1342 return err;
1343}
1344
1345late_initcall(fail_page_alloc_debugfs);
1346
1347#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1348
1349#else /* CONFIG_FAIL_PAGE_ALLOC */
1350
1351static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1352{
1353 return 0;
1354}
1355
1356#endif /* CONFIG_FAIL_PAGE_ALLOC */
1357
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358/*
1359 * Return 1 if free pages are above 'mark'. This takes into account the order
1360 * of the allocation.
1361 */
1362int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001363 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364{
1365 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001366 long min = mark;
1367 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 int o;
1369
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001370 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001372 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 min -= min / 4;
1374
1375 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1376 return 0;
1377 for (o = 0; o < order; o++) {
1378 /* At the next order, this order's pages become unavailable */
1379 free_pages -= z->free_area[o].nr_free << o;
1380
1381 /* Require fewer higher order pages to be free */
1382 min >>= 1;
1383
1384 if (free_pages <= min)
1385 return 0;
1386 }
1387 return 1;
1388}
1389
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001390#ifdef CONFIG_NUMA
1391/*
1392 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1393 * skip over zones that are not allowed by the cpuset, or that have
1394 * been recently (in last second) found to be nearly full. See further
1395 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001396 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001397 *
1398 * If the zonelist cache is present in the passed in zonelist, then
1399 * returns a pointer to the allowed node mask (either the current
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001400 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001401 *
1402 * If the zonelist cache is not available for this zonelist, does
1403 * nothing and returns NULL.
1404 *
1405 * If the fullzones BITMAP in the zonelist cache is stale (more than
1406 * a second since last zap'd) then we zap it out (clear its bits.)
1407 *
1408 * We hold off even calling zlc_setup, until after we've checked the
1409 * first zone in the zonelist, on the theory that most allocations will
1410 * be satisfied from that first zone, so best to examine that zone as
1411 * quickly as we can.
1412 */
1413static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1414{
1415 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1416 nodemask_t *allowednodes; /* zonelist_cache approximation */
1417
1418 zlc = zonelist->zlcache_ptr;
1419 if (!zlc)
1420 return NULL;
1421
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001422 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001423 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1424 zlc->last_full_zap = jiffies;
1425 }
1426
1427 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1428 &cpuset_current_mems_allowed :
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001429 &node_states[N_HIGH_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001430 return allowednodes;
1431}
1432
1433/*
1434 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1435 * if it is worth looking at further for free memory:
1436 * 1) Check that the zone isn't thought to be full (doesn't have its
1437 * bit set in the zonelist_cache fullzones BITMAP).
1438 * 2) Check that the zones node (obtained from the zonelist_cache
1439 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1440 * Return true (non-zero) if zone is worth looking at further, or
1441 * else return false (zero) if it is not.
1442 *
1443 * This check -ignores- the distinction between various watermarks,
1444 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1445 * found to be full for any variation of these watermarks, it will
1446 * be considered full for up to one second by all requests, unless
1447 * we are so low on memory on all allowed nodes that we are forced
1448 * into the second scan of the zonelist.
1449 *
1450 * In the second scan we ignore this zonelist cache and exactly
1451 * apply the watermarks to all zones, even it is slower to do so.
1452 * We are low on memory in the second scan, and should leave no stone
1453 * unturned looking for a free page.
1454 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001455static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001456 nodemask_t *allowednodes)
1457{
1458 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1459 int i; /* index of *z in zonelist zones */
1460 int n; /* node that zone *z is on */
1461
1462 zlc = zonelist->zlcache_ptr;
1463 if (!zlc)
1464 return 1;
1465
Mel Gormandd1a2392008-04-28 02:12:17 -07001466 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001467 n = zlc->z_to_n[i];
1468
1469 /* This zone is worth trying if it is allowed but not full */
1470 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1471}
1472
1473/*
1474 * Given 'z' scanning a zonelist, set the corresponding bit in
1475 * zlc->fullzones, so that subsequent attempts to allocate a page
1476 * from that zone don't waste time re-examining it.
1477 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001478static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001479{
1480 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1481 int i; /* index of *z in zonelist zones */
1482
1483 zlc = zonelist->zlcache_ptr;
1484 if (!zlc)
1485 return;
1486
Mel Gormandd1a2392008-04-28 02:12:17 -07001487 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001488
1489 set_bit(i, zlc->fullzones);
1490}
1491
1492#else /* CONFIG_NUMA */
1493
1494static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1495{
1496 return NULL;
1497}
1498
Mel Gormandd1a2392008-04-28 02:12:17 -07001499static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001500 nodemask_t *allowednodes)
1501{
1502 return 1;
1503}
1504
Mel Gormandd1a2392008-04-28 02:12:17 -07001505static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001506{
1507}
1508#endif /* CONFIG_NUMA */
1509
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001510/*
Paul Jackson0798e512006-12-06 20:31:38 -08001511 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001512 * a page.
1513 */
1514static struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001515get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
Mel Gorman5117f452009-06-16 15:31:59 -07001516 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
Mel Gorman3dd28262009-06-16 15:32:00 -07001517 struct zone *preferred_zone, int migratetype)
Martin Hicks753ee722005-06-21 17:14:41 -07001518{
Mel Gormandd1a2392008-04-28 02:12:17 -07001519 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001520 struct page *page = NULL;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001521 int classzone_idx;
Mel Gorman5117f452009-06-16 15:31:59 -07001522 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001523 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1524 int zlc_active = 0; /* set if using zonelist_cache */
1525 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gorman54a6eb52008-04-28 02:12:16 -07001526
Mel Gorman19770b32008-04-28 02:12:18 -07001527 classzone_idx = zone_idx(preferred_zone);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001528zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001529 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001530 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001531 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1532 */
Mel Gorman19770b32008-04-28 02:12:18 -07001533 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1534 high_zoneidx, nodemask) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001535 if (NUMA_BUILD && zlc_active &&
1536 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1537 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001538 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001539 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001540 goto try_next_zone;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001541
Mel Gorman41858962009-06-16 15:32:12 -07001542 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001543 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001544 unsigned long mark;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001545 int ret;
1546
Mel Gorman41858962009-06-16 15:32:12 -07001547 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
Mel Gormanfa5e0842009-06-16 15:33:22 -07001548 if (zone_watermark_ok(zone, order, mark,
1549 classzone_idx, alloc_flags))
1550 goto try_this_zone;
1551
1552 if (zone_reclaim_mode == 0)
1553 goto this_zone_full;
1554
1555 ret = zone_reclaim(zone, gfp_mask, order);
1556 switch (ret) {
1557 case ZONE_RECLAIM_NOSCAN:
1558 /* did not scan */
1559 goto try_next_zone;
1560 case ZONE_RECLAIM_FULL:
1561 /* scanned but unreclaimable */
1562 goto this_zone_full;
1563 default:
1564 /* did we reclaim enough */
1565 if (!zone_watermark_ok(zone, order, mark,
1566 classzone_idx, alloc_flags))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001567 goto this_zone_full;
Paul Jackson0798e512006-12-06 20:31:38 -08001568 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001569 }
1570
Mel Gormanfa5e0842009-06-16 15:33:22 -07001571try_this_zone:
Mel Gorman3dd28262009-06-16 15:32:00 -07001572 page = buffered_rmqueue(preferred_zone, zone, order,
1573 gfp_mask, migratetype);
Paul Jackson0798e512006-12-06 20:31:38 -08001574 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001575 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001576this_zone_full:
1577 if (NUMA_BUILD)
1578 zlc_mark_zone_full(zonelist, z);
1579try_next_zone:
Christoph Lameter62bc62a2009-06-16 15:32:15 -07001580 if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) {
Mel Gormand395b732009-06-16 15:32:09 -07001581 /*
1582 * we do zlc_setup after the first zone is tried but only
1583 * if there are multiple nodes make it worthwhile
1584 */
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001585 allowednodes = zlc_setup(zonelist, alloc_flags);
1586 zlc_active = 1;
1587 did_zlc_setup = 1;
1588 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07001589 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001590
1591 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1592 /* Disable zlc cache for second zonelist scan */
1593 zlc_active = 0;
1594 goto zonelist_scan;
1595 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001596 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07001597}
1598
Mel Gorman11e33f62009-06-16 15:31:57 -07001599static inline int
1600should_alloc_retry(gfp_t gfp_mask, unsigned int order,
1601 unsigned long pages_reclaimed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602{
Mel Gorman11e33f62009-06-16 15:31:57 -07001603 /* Do not loop if specifically requested */
1604 if (gfp_mask & __GFP_NORETRY)
1605 return 0;
1606
1607 /*
1608 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1609 * means __GFP_NOFAIL, but that may not be true in other
1610 * implementations.
1611 */
1612 if (order <= PAGE_ALLOC_COSTLY_ORDER)
1613 return 1;
1614
1615 /*
1616 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1617 * specified, then we retry until we no longer reclaim any pages
1618 * (above), or we've reclaimed an order of pages at least as
1619 * large as the allocation's order. In both cases, if the
1620 * allocation still fails, we stop retrying.
1621 */
1622 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
1623 return 1;
1624
1625 /*
1626 * Don't let big-order allocations loop unless the caller
1627 * explicitly requests that.
1628 */
1629 if (gfp_mask & __GFP_NOFAIL)
1630 return 1;
1631
1632 return 0;
1633}
1634
1635static inline struct page *
1636__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1637 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001638 nodemask_t *nodemask, struct zone *preferred_zone,
1639 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001640{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642
Mel Gorman11e33f62009-06-16 15:31:57 -07001643 /* Acquire the OOM killer lock for the zones in zonelist */
1644 if (!try_set_zone_oom(zonelist, gfp_mask)) {
1645 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 return NULL;
1647 }
Jens Axboe6b1de912005-11-17 21:35:02 +01001648
Mel Gorman11e33f62009-06-16 15:31:57 -07001649 /*
1650 * Go through the zonelist yet one more time, keep very high watermark
1651 * here, this is only to catch a parallel oom killing, we must fail if
1652 * we're still under heavy pressure.
1653 */
1654 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1655 order, zonelist, high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001656 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07001657 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001658 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07001659 goto out;
1660
1661 /* The OOM killer will not help higher order allocs */
David Rientjes82553a92009-06-16 15:32:58 -07001662 if (order > PAGE_ALLOC_COSTLY_ORDER && !(gfp_mask & __GFP_NOFAIL))
Mel Gorman11e33f62009-06-16 15:31:57 -07001663 goto out;
1664
1665 /* Exhausted what can be done so it's blamo time */
1666 out_of_memory(zonelist, gfp_mask, order);
1667
1668out:
1669 clear_zonelist_oom(zonelist, gfp_mask);
1670 return page;
1671}
1672
1673/* The really slow allocator path where we enter direct reclaim */
1674static inline struct page *
1675__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
1676 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001677 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001678 int migratetype, unsigned long *did_some_progress)
Mel Gorman11e33f62009-06-16 15:31:57 -07001679{
1680 struct page *page = NULL;
1681 struct reclaim_state reclaim_state;
1682 struct task_struct *p = current;
1683
1684 cond_resched();
1685
1686 /* We now go into synchronous reclaim */
1687 cpuset_memory_pressure_bump();
Mel Gorman11e33f62009-06-16 15:31:57 -07001688 p->flags |= PF_MEMALLOC;
1689 lockdep_set_current_reclaim_state(gfp_mask);
1690 reclaim_state.reclaimed_slab = 0;
1691 p->reclaim_state = &reclaim_state;
1692
1693 *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
1694
1695 p->reclaim_state = NULL;
1696 lockdep_clear_current_reclaim_state();
1697 p->flags &= ~PF_MEMALLOC;
1698
1699 cond_resched();
1700
1701 if (order != 0)
1702 drain_all_pages();
1703
1704 if (likely(*did_some_progress))
1705 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001706 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001707 alloc_flags, preferred_zone,
1708 migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001709 return page;
1710}
1711
Mel Gorman11e33f62009-06-16 15:31:57 -07001712/*
1713 * This is called in the allocator slow-path if the allocation request is of
1714 * sufficient urgency to ignore watermarks and take other desperate measures
1715 */
1716static inline struct page *
1717__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
1718 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001719 nodemask_t *nodemask, struct zone *preferred_zone,
1720 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001721{
1722 struct page *page;
1723
1724 do {
1725 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001726 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
Mel Gorman3dd28262009-06-16 15:32:00 -07001727 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001728
1729 if (!page && gfp_mask & __GFP_NOFAIL)
Jens Axboe8aa7e842009-07-09 14:52:32 +02001730 congestion_wait(BLK_RW_ASYNC, HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07001731 } while (!page && (gfp_mask & __GFP_NOFAIL));
1732
1733 return page;
1734}
1735
1736static inline
1737void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
1738 enum zone_type high_zoneidx)
1739{
1740 struct zoneref *z;
1741 struct zone *zone;
1742
1743 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
1744 wakeup_kswapd(zone, order);
1745}
1746
Peter Zijlstra341ce062009-06-16 15:32:02 -07001747static inline int
1748gfp_to_alloc_flags(gfp_t gfp_mask)
1749{
1750 struct task_struct *p = current;
1751 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
1752 const gfp_t wait = gfp_mask & __GFP_WAIT;
1753
Mel Gormana56f57f2009-06-16 15:32:02 -07001754 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
1755 BUILD_BUG_ON(__GFP_HIGH != ALLOC_HIGH);
1756
Peter Zijlstra341ce062009-06-16 15:32:02 -07001757 /*
1758 * The caller may dip into page reserves a bit more if the caller
1759 * cannot run direct reclaim, or if the caller has realtime scheduling
1760 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1761 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1762 */
Mel Gormana56f57f2009-06-16 15:32:02 -07001763 alloc_flags |= (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07001764
1765 if (!wait) {
1766 alloc_flags |= ALLOC_HARDER;
1767 /*
1768 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1769 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1770 */
1771 alloc_flags &= ~ALLOC_CPUSET;
1772 } else if (unlikely(rt_task(p)))
1773 alloc_flags |= ALLOC_HARDER;
1774
1775 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
1776 if (!in_interrupt() &&
1777 ((p->flags & PF_MEMALLOC) ||
1778 unlikely(test_thread_flag(TIF_MEMDIE))))
1779 alloc_flags |= ALLOC_NO_WATERMARKS;
1780 }
1781
1782 return alloc_flags;
1783}
1784
Mel Gorman11e33f62009-06-16 15:31:57 -07001785static inline struct page *
1786__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
1787 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001788 nodemask_t *nodemask, struct zone *preferred_zone,
1789 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001790{
1791 const gfp_t wait = gfp_mask & __GFP_WAIT;
1792 struct page *page = NULL;
1793 int alloc_flags;
1794 unsigned long pages_reclaimed = 0;
1795 unsigned long did_some_progress;
1796 struct task_struct *p = current;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001797
Christoph Lameter952f3b52006-12-06 20:33:26 -08001798 /*
Mel Gorman72807a72009-06-16 15:32:18 -07001799 * In the slowpath, we sanity check order to avoid ever trying to
1800 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
1801 * be using allocators in order of preference for an area that is
1802 * too large.
1803 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07001804 if (order >= MAX_ORDER) {
1805 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07001806 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07001807 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001808
Christoph Lameter952f3b52006-12-06 20:33:26 -08001809 /*
1810 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1811 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1812 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1813 * using a larger set of nodes after it has established that the
1814 * allowed per node queues are empty and that nodes are
1815 * over allocated.
1816 */
1817 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1818 goto nopage;
1819
Mel Gorman11e33f62009-06-16 15:31:57 -07001820 wake_all_kswapd(order, zonelist, high_zoneidx);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001821
David Rientjesb259fbd2009-09-21 17:01:34 -07001822restart:
Paul Jackson9bf22292005-09-06 15:18:12 -07001823 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001824 * OK, we're below the kswapd watermark and have kicked background
1825 * reclaim. Now things get more complex, so set up alloc_flags according
1826 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07001827 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07001828 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829
Peter Zijlstra341ce062009-06-16 15:32:02 -07001830 /* This is the last chance, in general, before the goto nopage. */
Mel Gorman19770b32008-04-28 02:12:18 -07001831 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
Peter Zijlstra341ce062009-06-16 15:32:02 -07001832 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
1833 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001834 if (page)
1835 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836
Kirill Korotaevb43a57b2006-12-06 20:32:27 -08001837rebalance:
Mel Gorman11e33f62009-06-16 15:31:57 -07001838 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07001839 if (alloc_flags & ALLOC_NO_WATERMARKS) {
1840 page = __alloc_pages_high_priority(gfp_mask, order,
1841 zonelist, high_zoneidx, nodemask,
1842 preferred_zone, migratetype);
1843 if (page)
1844 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 }
1846
1847 /* Atomic allocations - we can't balance anything */
1848 if (!wait)
1849 goto nopage;
1850
Peter Zijlstra341ce062009-06-16 15:32:02 -07001851 /* Avoid recursion of direct reclaim */
1852 if (p->flags & PF_MEMALLOC)
1853 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854
David Rientjes6583bb62009-07-29 15:02:06 -07001855 /* Avoid allocations with no watermarks from looping endlessly */
1856 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
1857 goto nopage;
1858
Mel Gorman11e33f62009-06-16 15:31:57 -07001859 /* Try direct reclaim and then allocating */
1860 page = __alloc_pages_direct_reclaim(gfp_mask, order,
1861 zonelist, high_zoneidx,
1862 nodemask,
Mel Gorman5117f452009-06-16 15:31:59 -07001863 alloc_flags, preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001864 migratetype, &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07001865 if (page)
1866 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 /*
Mel Gorman11e33f62009-06-16 15:31:57 -07001869 * If we failed to make any progress reclaiming, then we are
1870 * running out of options and have to consider going OOM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 */
Mel Gorman11e33f62009-06-16 15:31:57 -07001872 if (!did_some_progress) {
1873 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -07001874 if (oom_killer_disabled)
1875 goto nopage;
Mel Gorman11e33f62009-06-16 15:31:57 -07001876 page = __alloc_pages_may_oom(gfp_mask, order,
1877 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001878 nodemask, preferred_zone,
1879 migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001880 if (page)
1881 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882
Mel Gorman11e33f62009-06-16 15:31:57 -07001883 /*
David Rientjes82553a92009-06-16 15:32:58 -07001884 * The OOM killer does not trigger for high-order
1885 * ~__GFP_NOFAIL allocations so if no progress is being
1886 * made, there are no other options and retrying is
1887 * unlikely to help.
Mel Gorman11e33f62009-06-16 15:31:57 -07001888 */
David Rientjes82553a92009-06-16 15:32:58 -07001889 if (order > PAGE_ALLOC_COSTLY_ORDER &&
1890 !(gfp_mask & __GFP_NOFAIL))
Mel Gorman11e33f62009-06-16 15:31:57 -07001891 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 goto restart;
1894 }
1895 }
1896
Mel Gorman11e33f62009-06-16 15:31:57 -07001897 /* Check if we should retry the allocation */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07001898 pages_reclaimed += did_some_progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07001899 if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
1900 /* Wait for some write requests to complete then retry */
Jens Axboe8aa7e842009-07-09 14:52:32 +02001901 congestion_wait(BLK_RW_ASYNC, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902 goto rebalance;
1903 }
1904
1905nopage:
1906 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1907 printk(KERN_WARNING "%s: page allocation failure."
1908 " order:%d, mode:0x%x\n",
1909 p->comm, order, gfp_mask);
1910 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001911 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 }
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001913 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914got_pg:
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001915 if (kmemcheck_enabled)
1916 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 return page;
Mel Gorman11e33f62009-06-16 15:31:57 -07001918
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919}
Mel Gorman11e33f62009-06-16 15:31:57 -07001920
1921/*
1922 * This is the 'heart' of the zoned buddy allocator.
1923 */
1924struct page *
1925__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
1926 struct zonelist *zonelist, nodemask_t *nodemask)
1927{
1928 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Mel Gorman5117f452009-06-16 15:31:59 -07001929 struct zone *preferred_zone;
Mel Gorman11e33f62009-06-16 15:31:57 -07001930 struct page *page;
Mel Gorman3dd28262009-06-16 15:32:00 -07001931 int migratetype = allocflags_to_migratetype(gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07001932
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10001933 gfp_mask &= gfp_allowed_mask;
1934
Mel Gorman11e33f62009-06-16 15:31:57 -07001935 lockdep_trace_alloc(gfp_mask);
1936
1937 might_sleep_if(gfp_mask & __GFP_WAIT);
1938
1939 if (should_fail_alloc_page(gfp_mask, order))
1940 return NULL;
1941
1942 /*
1943 * Check the zones suitable for the gfp_mask contain at least one
1944 * valid zone. It's possible to have an empty zonelist as a result
1945 * of GFP_THISNODE and a memoryless node
1946 */
1947 if (unlikely(!zonelist->_zonerefs->zone))
1948 return NULL;
1949
Mel Gorman5117f452009-06-16 15:31:59 -07001950 /* The preferred zone is used for statistics later */
1951 first_zones_zonelist(zonelist, high_zoneidx, nodemask, &preferred_zone);
1952 if (!preferred_zone)
1953 return NULL;
1954
1955 /* First allocation attempt */
Mel Gorman11e33f62009-06-16 15:31:57 -07001956 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001957 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07001958 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001959 if (unlikely(!page))
1960 page = __alloc_pages_slowpath(gfp_mask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001961 zonelist, high_zoneidx, nodemask,
Mel Gorman3dd28262009-06-16 15:32:00 -07001962 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001963
Mel Gorman4b4f2782009-09-21 17:02:41 -07001964 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001965 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966}
Mel Gormand2391712009-06-16 15:31:52 -07001967EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968
1969/*
1970 * Common helper functions.
1971 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001972unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973{
Akinobu Mita945a1112009-09-21 17:01:47 -07001974 struct page *page;
1975
1976 /*
1977 * __get_free_pages() returns a 32-bit address, which cannot represent
1978 * a highmem page
1979 */
1980 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
1981
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 page = alloc_pages(gfp_mask, order);
1983 if (!page)
1984 return 0;
1985 return (unsigned long) page_address(page);
1986}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987EXPORT_SYMBOL(__get_free_pages);
1988
Harvey Harrison920c7a52008-02-04 22:29:26 -08001989unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990{
Akinobu Mita945a1112009-09-21 17:01:47 -07001991 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993EXPORT_SYMBOL(get_zeroed_page);
1994
1995void __pagevec_free(struct pagevec *pvec)
1996{
1997 int i = pagevec_count(pvec);
1998
Mel Gorman4b4f2782009-09-21 17:02:41 -07001999 while (--i >= 0) {
2000 trace_mm_pagevec_free(pvec->pages[i], pvec->cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 free_hot_cold_page(pvec->pages[i], pvec->cold);
Mel Gorman4b4f2782009-09-21 17:02:41 -07002002 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003}
2004
Harvey Harrison920c7a52008-02-04 22:29:26 -08002005void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006{
Nick Pigginb5810032005-10-29 18:16:12 -07002007 if (put_page_testzero(page)) {
Mel Gorman4b4f2782009-09-21 17:02:41 -07002008 trace_mm_page_free_direct(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 if (order == 0)
2010 free_hot_page(page);
2011 else
2012 __free_pages_ok(page, order);
2013 }
2014}
2015
2016EXPORT_SYMBOL(__free_pages);
2017
Harvey Harrison920c7a52008-02-04 22:29:26 -08002018void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019{
2020 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07002021 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 __free_pages(virt_to_page((void *)addr), order);
2023 }
2024}
2025
2026EXPORT_SYMBOL(free_pages);
2027
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002028/**
2029 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2030 * @size: the number of bytes to allocate
2031 * @gfp_mask: GFP flags for the allocation
2032 *
2033 * This function is similar to alloc_pages(), except that it allocates the
2034 * minimum number of pages to satisfy the request. alloc_pages() can only
2035 * allocate memory in power-of-two pages.
2036 *
2037 * This function is also limited by MAX_ORDER.
2038 *
2039 * Memory allocated by this function must be released by free_pages_exact().
2040 */
2041void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2042{
2043 unsigned int order = get_order(size);
2044 unsigned long addr;
2045
2046 addr = __get_free_pages(gfp_mask, order);
2047 if (addr) {
2048 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2049 unsigned long used = addr + PAGE_ALIGN(size);
2050
Kevin Cernekee5bfd7562009-07-05 12:08:19 -07002051 split_page(virt_to_page((void *)addr), order);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002052 while (used < alloc_end) {
2053 free_page(used);
2054 used += PAGE_SIZE;
2055 }
2056 }
2057
2058 return (void *)addr;
2059}
2060EXPORT_SYMBOL(alloc_pages_exact);
2061
2062/**
2063 * free_pages_exact - release memory allocated via alloc_pages_exact()
2064 * @virt: the value returned by alloc_pages_exact.
2065 * @size: size of allocation, same value as passed to alloc_pages_exact().
2066 *
2067 * Release the memory allocated by a previous call to alloc_pages_exact.
2068 */
2069void free_pages_exact(void *virt, size_t size)
2070{
2071 unsigned long addr = (unsigned long)virt;
2072 unsigned long end = addr + PAGE_ALIGN(size);
2073
2074 while (addr < end) {
2075 free_page(addr);
2076 addr += PAGE_SIZE;
2077 }
2078}
2079EXPORT_SYMBOL(free_pages_exact);
2080
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081static unsigned int nr_free_zone_pages(int offset)
2082{
Mel Gormandd1a2392008-04-28 02:12:17 -07002083 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002084 struct zone *zone;
2085
Martin J. Blighe310fd42005-07-29 22:59:18 -07002086 /* Just pick one node, since fallback list is circular */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 unsigned int sum = 0;
2088
Mel Gorman0e884602008-04-28 02:12:14 -07002089 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090
Mel Gorman54a6eb52008-04-28 02:12:16 -07002091 for_each_zone_zonelist(zone, z, zonelist, offset) {
Martin J. Blighe310fd42005-07-29 22:59:18 -07002092 unsigned long size = zone->present_pages;
Mel Gorman41858962009-06-16 15:32:12 -07002093 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07002094 if (size > high)
2095 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 }
2097
2098 return sum;
2099}
2100
2101/*
2102 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2103 */
2104unsigned int nr_free_buffer_pages(void)
2105{
Al Viroaf4ca452005-10-21 02:55:38 -04002106 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107}
Meelap Shahc2f1a552007-07-17 04:04:39 -07002108EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109
2110/*
2111 * Amount of free RAM allocatable within all zones
2112 */
2113unsigned int nr_free_pagecache_pages(void)
2114{
Mel Gorman2a1e2742007-07-17 04:03:12 -07002115 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002117
2118static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119{
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002120 if (NUMA_BUILD)
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08002121 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124void si_meminfo(struct sysinfo *val)
2125{
2126 val->totalram = totalram_pages;
2127 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002128 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 val->totalhigh = totalhigh_pages;
2131 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 val->mem_unit = PAGE_SIZE;
2133}
2134
2135EXPORT_SYMBOL(si_meminfo);
2136
2137#ifdef CONFIG_NUMA
2138void si_meminfo_node(struct sysinfo *val, int nid)
2139{
2140 pg_data_t *pgdat = NODE_DATA(nid);
2141
2142 val->totalram = pgdat->node_present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002143 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002144#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002146 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2147 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002148#else
2149 val->totalhigh = 0;
2150 val->freehigh = 0;
2151#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 val->mem_unit = PAGE_SIZE;
2153}
2154#endif
2155
2156#define K(x) ((x) << (PAGE_SHIFT-10))
2157
2158/*
2159 * Show free area list (used inside shift_scroll-lock stuff)
2160 * We also calculate the percentage fragmentation. We do this by counting the
2161 * memory on each free list with the exception of the first item on the list.
2162 */
2163void show_free_areas(void)
2164{
Jes Sorensenc7241912006-09-27 01:50:05 -07002165 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166 struct zone *zone;
2167
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002168 for_each_populated_zone(zone) {
Jes Sorensenc7241912006-09-27 01:50:05 -07002169 show_node(zone);
2170 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171
Dave Jones6b482c62005-11-10 15:45:56 -05002172 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173 struct per_cpu_pageset *pageset;
2174
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002175 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002177 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2178 cpu, pageset->pcp.high,
2179 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 }
2181 }
2182
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002183 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2184 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002185 " unevictable:%lu"
KOSAKI Motohiro71de1cc2009-09-21 17:01:31 -07002186 " dirty:%lu writeback:%lu unstable:%lu buffer:%lu\n"
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07002187 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002188 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002189 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002190 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002191 global_page_state(NR_ISOLATED_ANON),
2192 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002193 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002194 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002195 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07002196 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07002197 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07002198 global_page_state(NR_UNSTABLE_NFS),
KOSAKI Motohiro71de1cc2009-09-21 17:01:31 -07002199 nr_blockdev_pages(),
Christoph Lameterd23ad422007-02-10 01:43:02 -08002200 global_page_state(NR_FREE_PAGES),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07002201 global_page_state(NR_SLAB_RECLAIMABLE),
2202 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07002203 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002204 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08002205 global_page_state(NR_PAGETABLE),
2206 global_page_state(NR_BOUNCE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002208 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 int i;
2210
2211 show_node(zone);
2212 printk("%s"
2213 " free:%lukB"
2214 " min:%lukB"
2215 " low:%lukB"
2216 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002217 " active_anon:%lukB"
2218 " inactive_anon:%lukB"
2219 " active_file:%lukB"
2220 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002221 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002222 " isolated(anon):%lukB"
2223 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 " present:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002225 " mlocked:%lukB"
2226 " dirty:%lukB"
2227 " writeback:%lukB"
2228 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002229 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002230 " slab_reclaimable:%lukB"
2231 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07002232 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002233 " pagetables:%lukB"
2234 " unstable:%lukB"
2235 " bounce:%lukB"
2236 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 " pages_scanned:%lu"
2238 " all_unreclaimable? %s"
2239 "\n",
2240 zone->name,
Christoph Lameterd23ad422007-02-10 01:43:02 -08002241 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07002242 K(min_wmark_pages(zone)),
2243 K(low_wmark_pages(zone)),
2244 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002245 K(zone_page_state(zone, NR_ACTIVE_ANON)),
2246 K(zone_page_state(zone, NR_INACTIVE_ANON)),
2247 K(zone_page_state(zone, NR_ACTIVE_FILE)),
2248 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002249 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002250 K(zone_page_state(zone, NR_ISOLATED_ANON)),
2251 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 K(zone->present_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002253 K(zone_page_state(zone, NR_MLOCK)),
2254 K(zone_page_state(zone, NR_FILE_DIRTY)),
2255 K(zone_page_state(zone, NR_WRITEBACK)),
2256 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002257 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002258 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
2259 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07002260 zone_page_state(zone, NR_KERNEL_STACK) *
2261 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002262 K(zone_page_state(zone, NR_PAGETABLE)),
2263 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
2264 K(zone_page_state(zone, NR_BOUNCE)),
2265 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266 zone->pages_scanned,
David Rientjese815af92007-10-16 23:25:54 -07002267 (zone_is_all_unreclaimable(zone) ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268 );
2269 printk("lowmem_reserve[]:");
2270 for (i = 0; i < MAX_NR_ZONES; i++)
2271 printk(" %lu", zone->lowmem_reserve[i]);
2272 printk("\n");
2273 }
2274
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002275 for_each_populated_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002276 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277
2278 show_node(zone);
2279 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280
2281 spin_lock_irqsave(&zone->lock, flags);
2282 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002283 nr[order] = zone->free_area[order].nr_free;
2284 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 }
2286 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002287 for (order = 0; order < MAX_ORDER; order++)
2288 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289 printk("= %lukB\n", K(total));
2290 }
2291
Larry Woodmane6f36022008-02-04 22:29:30 -08002292 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2293
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 show_swap_cache_info();
2295}
2296
Mel Gorman19770b32008-04-28 02:12:18 -07002297static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2298{
2299 zoneref->zone = zone;
2300 zoneref->zone_idx = zone_idx(zone);
2301}
2302
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303/*
2304 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08002305 *
2306 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002308static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2309 int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310{
Christoph Lameter1a932052006-01-06 00:11:16 -08002311 struct zone *zone;
2312
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002313 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002314 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08002315
2316 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002317 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08002318 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08002319 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002320 zoneref_set_zone(zone,
2321 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08002322 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08002324
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002325 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08002326 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327}
2328
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002329
2330/*
2331 * zonelist_order:
2332 * 0 = automatic detection of better ordering.
2333 * 1 = order by ([node] distance, -zonetype)
2334 * 2 = order by (-zonetype, [node] distance)
2335 *
2336 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2337 * the same zonelist. So only NUMA can configure this param.
2338 */
2339#define ZONELIST_ORDER_DEFAULT 0
2340#define ZONELIST_ORDER_NODE 1
2341#define ZONELIST_ORDER_ZONE 2
2342
2343/* zonelist order in the kernel.
2344 * set_zonelist_order() will set this to NODE or ZONE.
2345 */
2346static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2347static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2348
2349
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002351/* The value user specified ....changed by config */
2352static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2353/* string for sysctl */
2354#define NUMA_ZONELIST_ORDER_LEN 16
2355char numa_zonelist_order[16] = "default";
2356
2357/*
2358 * interface for configure zonelist ordering.
2359 * command line option "numa_zonelist_order"
2360 * = "[dD]efault - default, automatic configuration.
2361 * = "[nN]ode - order by node locality, then by zone within node
2362 * = "[zZ]one - order by zone, then by locality within zone
2363 */
2364
2365static int __parse_numa_zonelist_order(char *s)
2366{
2367 if (*s == 'd' || *s == 'D') {
2368 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2369 } else if (*s == 'n' || *s == 'N') {
2370 user_zonelist_order = ZONELIST_ORDER_NODE;
2371 } else if (*s == 'z' || *s == 'Z') {
2372 user_zonelist_order = ZONELIST_ORDER_ZONE;
2373 } else {
2374 printk(KERN_WARNING
2375 "Ignoring invalid numa_zonelist_order value: "
2376 "%s\n", s);
2377 return -EINVAL;
2378 }
2379 return 0;
2380}
2381
2382static __init int setup_numa_zonelist_order(char *s)
2383{
2384 if (s)
2385 return __parse_numa_zonelist_order(s);
2386 return 0;
2387}
2388early_param("numa_zonelist_order", setup_numa_zonelist_order);
2389
2390/*
2391 * sysctl handler for numa_zonelist_order
2392 */
2393int numa_zonelist_order_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07002394 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002395 loff_t *ppos)
2396{
2397 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2398 int ret;
2399
2400 if (write)
2401 strncpy(saved_string, (char*)table->data,
2402 NUMA_ZONELIST_ORDER_LEN);
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07002403 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002404 if (ret)
2405 return ret;
2406 if (write) {
2407 int oldval = user_zonelist_order;
2408 if (__parse_numa_zonelist_order((char*)table->data)) {
2409 /*
2410 * bogus value. restore saved string
2411 */
2412 strncpy((char*)table->data, saved_string,
2413 NUMA_ZONELIST_ORDER_LEN);
2414 user_zonelist_order = oldval;
2415 } else if (oldval != user_zonelist_order)
2416 build_all_zonelists();
2417 }
2418 return 0;
2419}
2420
2421
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002422#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002423static int node_load[MAX_NUMNODES];
2424
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07002426 * 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 -07002427 * @node: node whose fallback list we're appending
2428 * @used_node_mask: nodemask_t of already used nodes
2429 *
2430 * We use a number of factors to determine which is the next node that should
2431 * appear on a given node's fallback list. The node should not have appeared
2432 * already in @node's fallback list, and it should be the next closest node
2433 * according to the distance array (which contains arbitrary distance values
2434 * from each node to each node in the system), and should also prefer nodes
2435 * with no CPUs, since presumably they'll have very little allocation pressure
2436 * on them otherwise.
2437 * It returns -1 if no node is found.
2438 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002439static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440{
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01002441 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442 int min_val = INT_MAX;
2443 int best_node = -1;
Rusty Russella70f7302009-03-13 14:49:46 +10302444 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01002446 /* Use the local node if we haven't already */
2447 if (!node_isset(node, *used_node_mask)) {
2448 node_set(node, *used_node_mask);
2449 return node;
2450 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002452 for_each_node_state(n, N_HIGH_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453
2454 /* Don't want a node to appear more than once */
2455 if (node_isset(n, *used_node_mask))
2456 continue;
2457
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458 /* Use the distance array to find the distance */
2459 val = node_distance(node, n);
2460
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01002461 /* Penalize nodes under us ("prefer the next node") */
2462 val += (n < node);
2463
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10302465 tmp = cpumask_of_node(n);
2466 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467 val += PENALTY_FOR_NODE_WITH_CPUS;
2468
2469 /* Slight preference for less loaded node */
2470 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2471 val += node_load[n];
2472
2473 if (val < min_val) {
2474 min_val = val;
2475 best_node = n;
2476 }
2477 }
2478
2479 if (best_node >= 0)
2480 node_set(best_node, *used_node_mask);
2481
2482 return best_node;
2483}
2484
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002485
2486/*
2487 * Build zonelists ordered by node and zones within node.
2488 * This results in maximum locality--normal zone overflows into local
2489 * DMA zone, if any--but risks exhausting DMA zone.
2490 */
2491static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002493 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002495
Mel Gorman54a6eb52008-04-28 02:12:16 -07002496 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07002497 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07002498 ;
2499 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2500 MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002501 zonelist->_zonerefs[j].zone = NULL;
2502 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002503}
2504
2505/*
Christoph Lameter523b9452007-10-16 01:25:37 -07002506 * Build gfp_thisnode zonelists
2507 */
2508static void build_thisnode_zonelists(pg_data_t *pgdat)
2509{
Christoph Lameter523b9452007-10-16 01:25:37 -07002510 int j;
2511 struct zonelist *zonelist;
2512
Mel Gorman54a6eb52008-04-28 02:12:16 -07002513 zonelist = &pgdat->node_zonelists[1];
2514 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002515 zonelist->_zonerefs[j].zone = NULL;
2516 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07002517}
2518
2519/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002520 * Build zonelists ordered by zone and nodes within zones.
2521 * This results in conserving DMA zone[s] until all Normal memory is
2522 * exhausted, but results in overflowing to remote node while memory
2523 * may still exist in local DMA zone.
2524 */
2525static int node_order[MAX_NUMNODES];
2526
2527static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2528{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002529 int pos, j, node;
2530 int zone_type; /* needs to be signed */
2531 struct zone *z;
2532 struct zonelist *zonelist;
2533
Mel Gorman54a6eb52008-04-28 02:12:16 -07002534 zonelist = &pgdat->node_zonelists[0];
2535 pos = 0;
2536 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2537 for (j = 0; j < nr_nodes; j++) {
2538 node = node_order[j];
2539 z = &NODE_DATA(node)->node_zones[zone_type];
2540 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002541 zoneref_set_zone(z,
2542 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002543 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002544 }
2545 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002546 }
Mel Gormandd1a2392008-04-28 02:12:17 -07002547 zonelist->_zonerefs[pos].zone = NULL;
2548 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002549}
2550
2551static int default_zonelist_order(void)
2552{
2553 int nid, zone_type;
2554 unsigned long low_kmem_size,total_size;
2555 struct zone *z;
2556 int average_size;
2557 /*
2558 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2559 * If they are really small and used heavily, the system can fall
2560 * into OOM very easily.
2561 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2562 */
2563 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2564 low_kmem_size = 0;
2565 total_size = 0;
2566 for_each_online_node(nid) {
2567 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2568 z = &NODE_DATA(nid)->node_zones[zone_type];
2569 if (populated_zone(z)) {
2570 if (zone_type < ZONE_NORMAL)
2571 low_kmem_size += z->present_pages;
2572 total_size += z->present_pages;
2573 }
2574 }
2575 }
2576 if (!low_kmem_size || /* there are no DMA area. */
2577 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2578 return ZONELIST_ORDER_NODE;
2579 /*
2580 * look into each node's config.
2581 * If there is a node whose DMA/DMA32 memory is very big area on
2582 * local memory, NODE_ORDER may be suitable.
2583 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002584 average_size = total_size /
2585 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002586 for_each_online_node(nid) {
2587 low_kmem_size = 0;
2588 total_size = 0;
2589 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2590 z = &NODE_DATA(nid)->node_zones[zone_type];
2591 if (populated_zone(z)) {
2592 if (zone_type < ZONE_NORMAL)
2593 low_kmem_size += z->present_pages;
2594 total_size += z->present_pages;
2595 }
2596 }
2597 if (low_kmem_size &&
2598 total_size > average_size && /* ignore small node */
2599 low_kmem_size > total_size * 70/100)
2600 return ZONELIST_ORDER_NODE;
2601 }
2602 return ZONELIST_ORDER_ZONE;
2603}
2604
2605static void set_zonelist_order(void)
2606{
2607 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2608 current_zonelist_order = default_zonelist_order();
2609 else
2610 current_zonelist_order = user_zonelist_order;
2611}
2612
2613static void build_zonelists(pg_data_t *pgdat)
2614{
2615 int j, node, load;
2616 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002618 int local_node, prev_node;
2619 struct zonelist *zonelist;
2620 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621
2622 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07002623 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07002625 zonelist->_zonerefs[0].zone = NULL;
2626 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 }
2628
2629 /* NUMA-aware ordering of nodes */
2630 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002631 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632 prev_node = local_node;
2633 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002634
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002635 memset(node_order, 0, sizeof(node_order));
2636 j = 0;
2637
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08002639 int distance = node_distance(local_node, node);
2640
2641 /*
2642 * If another node is sufficiently far away then it is better
2643 * to reclaim pages in a zone before going off node.
2644 */
2645 if (distance > RECLAIM_DISTANCE)
2646 zone_reclaim_mode = 1;
2647
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 /*
2649 * We don't want to pressure a particular node.
2650 * So adding penalty to the first node in same
2651 * distance group to make it round-robin.
2652 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08002653 if (distance != node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002654 node_load[node] = load;
2655
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656 prev_node = node;
2657 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002658 if (order == ZONELIST_ORDER_NODE)
2659 build_zonelists_in_node_order(pgdat, node);
2660 else
2661 node_order[j++] = node; /* remember order */
2662 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002664 if (order == ZONELIST_ORDER_ZONE) {
2665 /* calculate node order -- i.e., DMA last! */
2666 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667 }
Christoph Lameter523b9452007-10-16 01:25:37 -07002668
2669 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670}
2671
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002672/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002673static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002674{
Mel Gorman54a6eb52008-04-28 02:12:16 -07002675 struct zonelist *zonelist;
2676 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07002677 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002678
Mel Gorman54a6eb52008-04-28 02:12:16 -07002679 zonelist = &pgdat->node_zonelists[0];
2680 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2681 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07002682 for (z = zonelist->_zonerefs; z->zone; z++)
2683 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002684}
2685
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002686
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687#else /* CONFIG_NUMA */
2688
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002689static void set_zonelist_order(void)
2690{
2691 current_zonelist_order = ZONELIST_ORDER_ZONE;
2692}
2693
2694static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695{
Christoph Lameter19655d32006-09-25 23:31:19 -07002696 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002697 enum zone_type j;
2698 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699
2700 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002701
Mel Gorman54a6eb52008-04-28 02:12:16 -07002702 zonelist = &pgdat->node_zonelists[0];
2703 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704
Mel Gorman54a6eb52008-04-28 02:12:16 -07002705 /*
2706 * Now we build the zonelist so that it contains the zones
2707 * of all the other nodes.
2708 * We don't want to pressure a particular node, so when
2709 * building the zones for node N, we make sure that the
2710 * zones coming right after the local ones are those from
2711 * node N+1 (modulo N)
2712 */
2713 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2714 if (!node_online(node))
2715 continue;
2716 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2717 MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07002719 for (node = 0; node < local_node; node++) {
2720 if (!node_online(node))
2721 continue;
2722 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2723 MAX_NR_ZONES - 1);
2724 }
2725
Mel Gormandd1a2392008-04-28 02:12:17 -07002726 zonelist->_zonerefs[j].zone = NULL;
2727 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728}
2729
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002730/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002731static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002732{
Mel Gorman54a6eb52008-04-28 02:12:16 -07002733 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002734}
2735
Linus Torvalds1da177e2005-04-16 15:20:36 -07002736#endif /* CONFIG_NUMA */
2737
Rusty Russell9b1a4d32008-07-28 12:16:30 -05002738/* return values int ....just for stop_machine() */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002739static int __build_all_zonelists(void *dummy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740{
Yasunori Goto68113782006-06-23 02:03:11 -07002741 int nid;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002742
Bo Liu7f9cfb32009-08-18 14:11:19 -07002743#ifdef CONFIG_NUMA
2744 memset(node_load, 0, sizeof(node_load));
2745#endif
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002746 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07002747 pg_data_t *pgdat = NODE_DATA(nid);
2748
2749 build_zonelists(pgdat);
2750 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002751 }
Yasunori Goto68113782006-06-23 02:03:11 -07002752 return 0;
2753}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002754
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002755void build_all_zonelists(void)
Yasunori Goto68113782006-06-23 02:03:11 -07002756{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002757 set_zonelist_order();
2758
Yasunori Goto68113782006-06-23 02:03:11 -07002759 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07002760 __build_all_zonelists(NULL);
Mel Gorman68ad8df2008-07-23 21:26:52 -07002761 mminit_verify_zonelist();
Yasunori Goto68113782006-06-23 02:03:11 -07002762 cpuset_init_current_mems_allowed();
2763 } else {
Simon Arlott183ff222007-10-20 01:27:18 +02002764 /* we have to stop all cpus to guarantee there is no user
Yasunori Goto68113782006-06-23 02:03:11 -07002765 of zonelist */
Rusty Russell9b1a4d32008-07-28 12:16:30 -05002766 stop_machine(__build_all_zonelists, NULL, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07002767 /* cpuset refresh routine should be here */
2768 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07002769 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002770 /*
2771 * Disable grouping by mobility if the number of pages in the
2772 * system is too low to allow the mechanism to work. It would be
2773 * more accurate, but expensive to check per-zone. This check is
2774 * made on memory-hotadd so a system can start with mobility
2775 * disabled and enable it later
2776 */
Mel Gormand9c23402007-10-16 01:26:01 -07002777 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002778 page_group_by_mobility_disabled = 1;
2779 else
2780 page_group_by_mobility_disabled = 0;
2781
2782 printk("Built %i zonelists in %s order, mobility grouping %s. "
2783 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002784 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002785 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002786 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002787 vm_total_pages);
2788#ifdef CONFIG_NUMA
2789 printk("Policy zone: %s\n", zone_names[policy_zone]);
2790#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791}
2792
2793/*
2794 * Helper functions to size the waitqueue hash table.
2795 * Essentially these want to choose hash table sizes sufficiently
2796 * large so that collisions trying to wait on pages are rare.
2797 * But in fact, the number of active page waitqueues on typical
2798 * systems is ridiculously low, less than 200. So this is even
2799 * conservative, even though it seems large.
2800 *
2801 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2802 * waitqueues, i.e. the size of the waitq table given the number of pages.
2803 */
2804#define PAGES_PER_WAITQUEUE 256
2805
Yasunori Gotocca448f2006-06-23 02:03:10 -07002806#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07002807static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808{
2809 unsigned long size = 1;
2810
2811 pages /= PAGES_PER_WAITQUEUE;
2812
2813 while (size < pages)
2814 size <<= 1;
2815
2816 /*
2817 * Once we have dozens or even hundreds of threads sleeping
2818 * on IO we've got bigger problems than wait queue collision.
2819 * Limit the size of the wait table to a reasonable size.
2820 */
2821 size = min(size, 4096UL);
2822
2823 return max(size, 4UL);
2824}
Yasunori Gotocca448f2006-06-23 02:03:10 -07002825#else
2826/*
2827 * A zone's size might be changed by hot-add, so it is not possible to determine
2828 * a suitable size for its wait_table. So we use the maximum size now.
2829 *
2830 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2831 *
2832 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2833 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2834 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2835 *
2836 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2837 * or more by the traditional way. (See above). It equals:
2838 *
2839 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2840 * ia64(16K page size) : = ( 8G + 4M)byte.
2841 * powerpc (64K page size) : = (32G +16M)byte.
2842 */
2843static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2844{
2845 return 4096UL;
2846}
2847#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848
2849/*
2850 * This is an integer logarithm so that shifts can be used later
2851 * to extract the more random high bits from the multiplicative
2852 * hash function before the remainder is taken.
2853 */
2854static inline unsigned long wait_table_bits(unsigned long size)
2855{
2856 return ffz(~size);
2857}
2858
2859#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2860
Mel Gorman56fd56b2007-10-16 01:25:58 -07002861/*
Mel Gormand9c23402007-10-16 01:26:01 -07002862 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07002863 * of blocks reserved is based on min_wmark_pages(zone). The memory within
2864 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07002865 * higher will lead to a bigger reserve which will get freed as contiguous
2866 * blocks as reclaim kicks in
2867 */
2868static void setup_zone_migrate_reserve(struct zone *zone)
2869{
2870 unsigned long start_pfn, pfn, end_pfn;
2871 struct page *page;
Mel Gorman78986a62009-09-21 17:03:02 -07002872 unsigned long block_migratetype;
2873 int reserve;
Mel Gorman56fd56b2007-10-16 01:25:58 -07002874
2875 /* Get the start pfn, end pfn and the number of blocks to reserve */
2876 start_pfn = zone->zone_start_pfn;
2877 end_pfn = start_pfn + zone->spanned_pages;
Mel Gorman41858962009-06-16 15:32:12 -07002878 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07002879 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07002880
Mel Gorman78986a62009-09-21 17:03:02 -07002881 /*
2882 * Reserve blocks are generally in place to help high-order atomic
2883 * allocations that are short-lived. A min_free_kbytes value that
2884 * would result in more than 2 reserve blocks for atomic allocations
2885 * is assumed to be in place to help anti-fragmentation for the
2886 * future allocation of hugepages at runtime.
2887 */
2888 reserve = min(2, reserve);
2889
Mel Gormand9c23402007-10-16 01:26:01 -07002890 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07002891 if (!pfn_valid(pfn))
2892 continue;
2893 page = pfn_to_page(pfn);
2894
Adam Litke344c7902008-09-02 14:35:38 -07002895 /* Watch out for overlapping nodes */
2896 if (page_to_nid(page) != zone_to_nid(zone))
2897 continue;
2898
Mel Gorman56fd56b2007-10-16 01:25:58 -07002899 /* Blocks with reserved pages will never free, skip them. */
2900 if (PageReserved(page))
2901 continue;
2902
2903 block_migratetype = get_pageblock_migratetype(page);
2904
2905 /* If this block is reserved, account for it */
2906 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
2907 reserve--;
2908 continue;
2909 }
2910
2911 /* Suitable for reserving if this block is movable */
2912 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
2913 set_pageblock_migratetype(page, MIGRATE_RESERVE);
2914 move_freepages_block(zone, page, MIGRATE_RESERVE);
2915 reserve--;
2916 continue;
2917 }
2918
2919 /*
2920 * If the reserve is met and this is a previous reserved block,
2921 * take it back
2922 */
2923 if (block_migratetype == MIGRATE_RESERVE) {
2924 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2925 move_freepages_block(zone, page, MIGRATE_MOVABLE);
2926 }
2927 }
2928}
Mel Gormanac0e5b72007-10-16 01:25:58 -07002929
Linus Torvalds1da177e2005-04-16 15:20:36 -07002930/*
2931 * Initially all pages are reserved - free ones are freed
2932 * up by free_all_bootmem() once the early boot process is
2933 * done. Non-atomic initialization, single-pass.
2934 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002935void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08002936 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002937{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002938 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07002939 unsigned long end_pfn = start_pfn + size;
2940 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002941 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002942
Hugh Dickins22b31ee2009-01-06 14:40:09 -08002943 if (highest_memmap_pfn < end_pfn - 1)
2944 highest_memmap_pfn = end_pfn - 1;
2945
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002946 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08002947 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08002948 /*
2949 * There can be holes in boot-time mem_map[]s
2950 * handed to this function. They do not
2951 * exist on hotplugged memory.
2952 */
2953 if (context == MEMMAP_EARLY) {
2954 if (!early_pfn_valid(pfn))
2955 continue;
2956 if (!early_pfn_in_nid(pfn, nid))
2957 continue;
2958 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002959 page = pfn_to_page(pfn);
2960 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07002961 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08002962 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002963 reset_page_mapcount(page);
2964 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002965 /*
2966 * Mark the block movable so that blocks are reserved for
2967 * movable at startup. This will force kernel allocations
2968 * to reserve their blocks rather than leaking throughout
2969 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07002970 * kernel allocations are made. Later some blocks near
2971 * the start are marked MIGRATE_RESERVE by
2972 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002973 *
2974 * bitmap is created for zone's valid pfn range. but memmap
2975 * can be created for invalid pages (for alignment)
2976 * check here not to call set_pageblock_migratetype() against
2977 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002978 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002979 if ((z->zone_start_pfn <= pfn)
2980 && (pfn < z->zone_start_pfn + z->spanned_pages)
2981 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07002982 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002983
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984 INIT_LIST_HEAD(&page->lru);
2985#ifdef WANT_PAGE_VIRTUAL
2986 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2987 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07002988 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002990 }
2991}
2992
Andi Kleen1e548de2008-02-04 22:29:26 -08002993static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002994{
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002995 int order, t;
2996 for_each_migratetype_order(order, t) {
2997 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998 zone->free_area[order].nr_free = 0;
2999 }
3000}
3001
3002#ifndef __HAVE_ARCH_MEMMAP_INIT
3003#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08003004 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005#endif
3006
Sam Ravnborg1d6f4e62008-01-17 15:21:12 -08003007static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003008{
David Howells3a6be872009-05-06 16:03:03 -07003009#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003010 int batch;
3011
3012 /*
3013 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07003014 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003015 *
3016 * OK, so we don't know how big the cache is. So guess.
3017 */
3018 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07003019 if (batch * PAGE_SIZE > 512 * 1024)
3020 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003021 batch /= 4; /* We effectively *= 4 below */
3022 if (batch < 1)
3023 batch = 1;
3024
3025 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11003026 * Clamp the batch to a 2^n - 1 value. Having a power
3027 * of 2 value was found to be more likely to have
3028 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003029 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11003030 * For example if 2 tasks are alternately allocating
3031 * batches of pages, one task can end up with a lot
3032 * of pages of one half of the possible page colors
3033 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003034 */
David Howells91552032009-05-06 16:03:02 -07003035 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07003036
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003037 return batch;
David Howells3a6be872009-05-06 16:03:03 -07003038
3039#else
3040 /* The deferral and batching of frees should be suppressed under NOMMU
3041 * conditions.
3042 *
3043 * The problem is that NOMMU needs to be able to allocate large chunks
3044 * of contiguous memory as there's no hardware page translation to
3045 * assemble apparent contiguous memory from discontiguous pages.
3046 *
3047 * Queueing large contiguous runs of pages for batching, however,
3048 * causes the pages to actually be freed in smaller chunks. As there
3049 * can be a significant delay between the individual batches being
3050 * recycled, this leads to the once large chunks of space being
3051 * fragmented and becoming unavailable for high-order allocations.
3052 */
3053 return 0;
3054#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003055}
3056
Adrian Bunkb69a7282008-07-23 21:28:12 -07003057static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
Christoph Lameter2caaad42005-06-21 17:15:00 -07003058{
3059 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003060 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003061
Magnus Damm1c6fe942005-10-26 01:58:59 -07003062 memset(p, 0, sizeof(*p));
3063
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003064 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003065 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003066 pcp->high = 6 * batch;
3067 pcp->batch = max(1UL, 1 * batch);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003068 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
3069 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07003070}
3071
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003072/*
3073 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3074 * to the value high for the pageset p.
3075 */
3076
3077static void setup_pagelist_highmark(struct per_cpu_pageset *p,
3078 unsigned long high)
3079{
3080 struct per_cpu_pages *pcp;
3081
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003082 pcp = &p->pcp;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003083 pcp->high = high;
3084 pcp->batch = max(1UL, high/4);
3085 if ((high/4) > (PAGE_SHIFT * 8))
3086 pcp->batch = PAGE_SHIFT * 8;
3087}
3088
3089
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003090#ifdef CONFIG_NUMA
3091/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07003092 * Boot pageset table. One per cpu which is going to be used for all
3093 * zones and all nodes. The parameters will be set in such a way
3094 * that an item put on a list will immediately be handed over to
3095 * the buddy list. This is safe since pageset manipulation is done
3096 * with interrupts disabled.
3097 *
3098 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07003099 *
3100 * The boot_pagesets must be kept even after bootup is complete for
3101 * unused processors and/or zones. They do play a role for bootstrapping
3102 * hotplugged processors.
3103 *
3104 * zoneinfo_show() and maybe other functions do
3105 * not check if the processor is online before following the pageset pointer.
3106 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07003107 */
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08003108static struct per_cpu_pageset boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07003109
3110/*
3111 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003112 * per cpu pageset array in struct zone.
3113 */
Ashok Raj6292d9a2006-02-01 03:04:44 -08003114static int __cpuinit process_zones(int cpu)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003115{
3116 struct zone *zone, *dzone;
Christoph Lameter37c07082007-10-16 01:25:36 -07003117 int node = cpu_to_node(cpu);
3118
3119 node_set_state(node, N_CPU); /* this node has a cpu */
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003120
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003121 for_each_populated_zone(zone) {
Nick Piggin23316bc2006-01-08 01:00:41 -08003122 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lameter37c07082007-10-16 01:25:36 -07003123 GFP_KERNEL, node);
Nick Piggin23316bc2006-01-08 01:00:41 -08003124 if (!zone_pcp(zone, cpu))
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003125 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003126
Nick Piggin23316bc2006-01-08 01:00:41 -08003127 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003128
3129 if (percpu_pagelist_fraction)
3130 setup_pagelist_highmark(zone_pcp(zone, cpu),
3131 (zone->present_pages / percpu_pagelist_fraction));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003132 }
3133
3134 return 0;
3135bad:
3136 for_each_zone(dzone) {
Andrew Morton64191682007-08-30 23:56:17 -07003137 if (!populated_zone(dzone))
3138 continue;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003139 if (dzone == zone)
3140 break;
Nick Piggin23316bc2006-01-08 01:00:41 -08003141 kfree(zone_pcp(dzone, cpu));
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07003142 zone_pcp(dzone, cpu) = &boot_pageset[cpu];
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003143 }
3144 return -ENOMEM;
3145}
3146
3147static inline void free_zone_pagesets(int cpu)
3148{
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003149 struct zone *zone;
3150
3151 for_each_zone(zone) {
3152 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
3153
David Rientjesf3ef9ea2006-09-25 16:24:57 -07003154 /* Free per_cpu_pageset if it is slab allocated */
3155 if (pset != &boot_pageset[cpu])
3156 kfree(pset);
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07003157 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003158 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003159}
3160
Chandra Seetharaman9c7b2162006-06-27 02:54:07 -07003161static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003162 unsigned long action,
3163 void *hcpu)
3164{
3165 int cpu = (long)hcpu;
3166 int ret = NOTIFY_OK;
3167
3168 switch (action) {
Andy Whitcroftce421c72006-12-06 20:33:08 -08003169 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003170 case CPU_UP_PREPARE_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08003171 if (process_zones(cpu))
3172 ret = NOTIFY_BAD;
3173 break;
3174 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003175 case CPU_UP_CANCELED_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08003176 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003177 case CPU_DEAD_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08003178 free_zone_pagesets(cpu);
3179 break;
3180 default:
3181 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003182 }
3183 return ret;
3184}
3185
Chandra Seetharaman74b85f32006-06-27 02:54:09 -07003186static struct notifier_block __cpuinitdata pageset_notifier =
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003187 { &pageset_cpuup_callback, NULL, 0 };
3188
Al Viro78d99552005-12-15 09:18:25 +00003189void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003190{
3191 int err;
3192
3193 /* Initialize per_cpu_pageset for cpu 0.
3194 * A cpuup callback will do this for every cpu
3195 * as it comes online
3196 */
3197 err = process_zones(smp_processor_id());
3198 BUG_ON(err);
3199 register_cpu_notifier(&pageset_notifier);
3200}
3201
3202#endif
3203
Sam Ravnborg577a32f2007-05-17 23:29:25 +02003204static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07003205int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07003206{
3207 int i;
3208 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003209 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07003210
3211 /*
3212 * The per-page waitqueue mechanism uses hashed waitqueues
3213 * per zone.
3214 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07003215 zone->wait_table_hash_nr_entries =
3216 wait_table_hash_nr_entries(zone_size_pages);
3217 zone->wait_table_bits =
3218 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003219 alloc_size = zone->wait_table_hash_nr_entries
3220 * sizeof(wait_queue_head_t);
3221
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07003222 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07003223 zone->wait_table = (wait_queue_head_t *)
3224 alloc_bootmem_node(pgdat, alloc_size);
3225 } else {
3226 /*
3227 * This case means that a zone whose size was 0 gets new memory
3228 * via memory hot-add.
3229 * But it may be the case that a new node was hot-added. In
3230 * this case vmalloc() will not be able to use this new node's
3231 * memory - this wait_table must be initialized to use this new
3232 * node itself as well.
3233 * To use this new node's memory, further consideration will be
3234 * necessary.
3235 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07003236 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003237 }
3238 if (!zone->wait_table)
3239 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07003240
Yasunori Goto02b694d2006-06-23 02:03:08 -07003241 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07003242 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003243
3244 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003245}
3246
Shaohua Li112067f2009-09-21 17:01:16 -07003247static int __zone_pcp_update(void *data)
3248{
3249 struct zone *zone = data;
3250 int cpu;
3251 unsigned long batch = zone_batchsize(zone), flags;
3252
3253 for (cpu = 0; cpu < NR_CPUS; cpu++) {
3254 struct per_cpu_pageset *pset;
3255 struct per_cpu_pages *pcp;
3256
3257 pset = zone_pcp(zone, cpu);
3258 pcp = &pset->pcp;
3259
3260 local_irq_save(flags);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003261 free_pcppages_bulk(zone, pcp->count, pcp);
Shaohua Li112067f2009-09-21 17:01:16 -07003262 setup_pageset(pset, batch);
3263 local_irq_restore(flags);
3264 }
3265 return 0;
3266}
3267
3268void zone_pcp_update(struct zone *zone)
3269{
3270 stop_machine(__zone_pcp_update, zone, NULL);
3271}
3272
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003273static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07003274{
3275 int cpu;
3276 unsigned long batch = zone_batchsize(zone);
3277
3278 for (cpu = 0; cpu < NR_CPUS; cpu++) {
3279#ifdef CONFIG_NUMA
3280 /* Early boot. Slab allocator not functional yet */
Nick Piggin23316bc2006-01-08 01:00:41 -08003281 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Dave Hansened8ece22005-10-29 18:16:50 -07003282 setup_pageset(&boot_pageset[cpu],0);
3283#else
3284 setup_pageset(zone_pcp(zone,cpu), batch);
3285#endif
3286 }
Anton Blanchardf5335c02006-03-25 03:06:49 -08003287 if (zone->present_pages)
3288 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
3289 zone->name, zone->present_pages, batch);
Dave Hansened8ece22005-10-29 18:16:50 -07003290}
3291
Yasunori Goto718127c2006-06-23 02:03:10 -07003292__meminit int init_currently_empty_zone(struct zone *zone,
3293 unsigned long zone_start_pfn,
Dave Hansena2f3aa022007-01-10 23:15:30 -08003294 unsigned long size,
3295 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07003296{
3297 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003298 int ret;
3299 ret = zone_wait_table_init(zone, size);
3300 if (ret)
3301 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07003302 pgdat->nr_zones = zone_idx(zone) + 1;
3303
Dave Hansened8ece22005-10-29 18:16:50 -07003304 zone->zone_start_pfn = zone_start_pfn;
3305
Mel Gorman708614e2008-07-23 21:26:51 -07003306 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3307 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3308 pgdat->node_id,
3309 (unsigned long)zone_idx(zone),
3310 zone_start_pfn, (zone_start_pfn + size));
3311
Andi Kleen1e548de2008-02-04 22:29:26 -08003312 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07003313
3314 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003315}
3316
Mel Gormanc7132162006-09-27 01:49:43 -07003317#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3318/*
3319 * Basic iterator support. Return the first range of PFNs for a node
3320 * Note: nid == MAX_NUMNODES returns first region regardless of node
3321 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003322static int __meminit first_active_region_index_in_nid(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003323{
3324 int i;
3325
3326 for (i = 0; i < nr_nodemap_entries; i++)
3327 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3328 return i;
3329
3330 return -1;
3331}
3332
3333/*
3334 * Basic iterator support. Return the next active range of PFNs for a node
Simon Arlott183ff222007-10-20 01:27:18 +02003335 * Note: nid == MAX_NUMNODES returns next region regardless of node
Mel Gormanc7132162006-09-27 01:49:43 -07003336 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003337static int __meminit next_active_region_index_in_nid(int index, int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003338{
3339 for (index = index + 1; index < nr_nodemap_entries; index++)
3340 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3341 return index;
3342
3343 return -1;
3344}
3345
3346#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3347/*
3348 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3349 * Architectures may implement their own version but if add_active_range()
3350 * was used and there are no special requirements, this is a convenient
3351 * alternative
3352 */
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003353int __meminit __early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07003354{
3355 int i;
3356
3357 for (i = 0; i < nr_nodemap_entries; i++) {
3358 unsigned long start_pfn = early_node_map[i].start_pfn;
3359 unsigned long end_pfn = early_node_map[i].end_pfn;
3360
3361 if (start_pfn <= pfn && pfn < end_pfn)
3362 return early_node_map[i].nid;
3363 }
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003364 /* This is a memory hole */
3365 return -1;
Mel Gormanc7132162006-09-27 01:49:43 -07003366}
3367#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3368
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003369int __meminit early_pfn_to_nid(unsigned long pfn)
3370{
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003371 int nid;
3372
3373 nid = __early_pfn_to_nid(pfn);
3374 if (nid >= 0)
3375 return nid;
3376 /* just returns 0 */
3377 return 0;
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003378}
3379
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003380#ifdef CONFIG_NODES_SPAN_OTHER_NODES
3381bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3382{
3383 int nid;
3384
3385 nid = __early_pfn_to_nid(pfn);
3386 if (nid >= 0 && nid != node)
3387 return false;
3388 return true;
3389}
3390#endif
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003391
Mel Gormanc7132162006-09-27 01:49:43 -07003392/* Basic iterator support to walk early_node_map[] */
3393#define for_each_active_range_index_in_nid(i, nid) \
3394 for (i = first_active_region_index_in_nid(nid); i != -1; \
3395 i = next_active_region_index_in_nid(i, nid))
3396
3397/**
3398 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003399 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3400 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07003401 *
3402 * If an architecture guarantees that all ranges registered with
3403 * add_active_ranges() contain no holes and may be freed, this
3404 * this function may be used instead of calling free_bootmem() manually.
3405 */
3406void __init free_bootmem_with_active_regions(int nid,
3407 unsigned long max_low_pfn)
3408{
3409 int i;
3410
3411 for_each_active_range_index_in_nid(i, nid) {
3412 unsigned long size_pages = 0;
3413 unsigned long end_pfn = early_node_map[i].end_pfn;
3414
3415 if (early_node_map[i].start_pfn >= max_low_pfn)
3416 continue;
3417
3418 if (end_pfn > max_low_pfn)
3419 end_pfn = max_low_pfn;
3420
3421 size_pages = end_pfn - early_node_map[i].start_pfn;
3422 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3423 PFN_PHYS(early_node_map[i].start_pfn),
3424 size_pages << PAGE_SHIFT);
3425 }
3426}
3427
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003428void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3429{
3430 int i;
Yinghai Lud52d53b2008-06-16 20:10:55 -07003431 int ret;
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003432
Yinghai Lud52d53b2008-06-16 20:10:55 -07003433 for_each_active_range_index_in_nid(i, nid) {
3434 ret = work_fn(early_node_map[i].start_pfn,
3435 early_node_map[i].end_pfn, data);
3436 if (ret)
3437 break;
3438 }
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003439}
Mel Gormanc7132162006-09-27 01:49:43 -07003440/**
3441 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003442 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07003443 *
3444 * If an architecture guarantees that all ranges registered with
3445 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003446 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07003447 */
3448void __init sparse_memory_present_with_active_regions(int nid)
3449{
3450 int i;
3451
3452 for_each_active_range_index_in_nid(i, nid)
3453 memory_present(early_node_map[i].nid,
3454 early_node_map[i].start_pfn,
3455 early_node_map[i].end_pfn);
3456}
3457
3458/**
3459 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003460 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3461 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3462 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07003463 *
3464 * It returns the start and end page frame of a node based on information
3465 * provided by an arch calling add_active_range(). If called for a node
3466 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003467 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07003468 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003469void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003470 unsigned long *start_pfn, unsigned long *end_pfn)
3471{
3472 int i;
3473 *start_pfn = -1UL;
3474 *end_pfn = 0;
3475
3476 for_each_active_range_index_in_nid(i, nid) {
3477 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3478 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3479 }
3480
Christoph Lameter633c0662007-10-16 01:25:37 -07003481 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07003482 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07003483}
3484
3485/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07003486 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3487 * assumption is made that zones within a node are ordered in monotonic
3488 * increasing memory addresses so that the "highest" populated zone is used
3489 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003490static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07003491{
3492 int zone_index;
3493 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3494 if (zone_index == ZONE_MOVABLE)
3495 continue;
3496
3497 if (arch_zone_highest_possible_pfn[zone_index] >
3498 arch_zone_lowest_possible_pfn[zone_index])
3499 break;
3500 }
3501
3502 VM_BUG_ON(zone_index == -1);
3503 movable_zone = zone_index;
3504}
3505
3506/*
3507 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3508 * because it is sized independant of architecture. Unlike the other zones,
3509 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3510 * in each node depending on the size of each node and how evenly kernelcore
3511 * is distributed. This helper function adjusts the zone ranges
3512 * provided by the architecture for a given node by using the end of the
3513 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3514 * zones within a node are in order of monotonic increases memory addresses
3515 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003516static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07003517 unsigned long zone_type,
3518 unsigned long node_start_pfn,
3519 unsigned long node_end_pfn,
3520 unsigned long *zone_start_pfn,
3521 unsigned long *zone_end_pfn)
3522{
3523 /* Only adjust if ZONE_MOVABLE is on this node */
3524 if (zone_movable_pfn[nid]) {
3525 /* Size ZONE_MOVABLE */
3526 if (zone_type == ZONE_MOVABLE) {
3527 *zone_start_pfn = zone_movable_pfn[nid];
3528 *zone_end_pfn = min(node_end_pfn,
3529 arch_zone_highest_possible_pfn[movable_zone]);
3530
3531 /* Adjust for ZONE_MOVABLE starting within this range */
3532 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3533 *zone_end_pfn > zone_movable_pfn[nid]) {
3534 *zone_end_pfn = zone_movable_pfn[nid];
3535
3536 /* Check if this whole range is within ZONE_MOVABLE */
3537 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3538 *zone_start_pfn = *zone_end_pfn;
3539 }
3540}
3541
3542/*
Mel Gormanc7132162006-09-27 01:49:43 -07003543 * Return the number of pages a zone spans in a node, including holes
3544 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3545 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003546static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003547 unsigned long zone_type,
3548 unsigned long *ignored)
3549{
3550 unsigned long node_start_pfn, node_end_pfn;
3551 unsigned long zone_start_pfn, zone_end_pfn;
3552
3553 /* Get the start and end of the node and zone */
3554 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3555 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3556 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07003557 adjust_zone_range_for_zone_movable(nid, zone_type,
3558 node_start_pfn, node_end_pfn,
3559 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003560
3561 /* Check that this node has pages within the zone's required range */
3562 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3563 return 0;
3564
3565 /* Move the zone boundaries inside the node if necessary */
3566 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3567 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3568
3569 /* Return the spanned pages */
3570 return zone_end_pfn - zone_start_pfn;
3571}
3572
3573/*
3574 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003575 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07003576 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003577static unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003578 unsigned long range_start_pfn,
3579 unsigned long range_end_pfn)
3580{
3581 int i = 0;
3582 unsigned long prev_end_pfn = 0, hole_pages = 0;
3583 unsigned long start_pfn;
3584
3585 /* Find the end_pfn of the first active range of pfns in the node */
3586 i = first_active_region_index_in_nid(nid);
3587 if (i == -1)
3588 return 0;
3589
Mel Gormanb5445f92007-07-26 10:41:18 -07003590 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3591
Mel Gorman9c7cd682006-09-27 01:49:58 -07003592 /* Account for ranges before physical memory on this node */
3593 if (early_node_map[i].start_pfn > range_start_pfn)
Mel Gormanb5445f92007-07-26 10:41:18 -07003594 hole_pages = prev_end_pfn - range_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003595
3596 /* Find all holes for the zone within the node */
3597 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3598
3599 /* No need to continue if prev_end_pfn is outside the zone */
3600 if (prev_end_pfn >= range_end_pfn)
3601 break;
3602
3603 /* Make sure the end of the zone is not within the hole */
3604 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3605 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3606
3607 /* Update the hole size cound and move on */
3608 if (start_pfn > range_start_pfn) {
3609 BUG_ON(prev_end_pfn > start_pfn);
3610 hole_pages += start_pfn - prev_end_pfn;
3611 }
3612 prev_end_pfn = early_node_map[i].end_pfn;
3613 }
3614
Mel Gorman9c7cd682006-09-27 01:49:58 -07003615 /* Account for ranges past physical memory on this node */
3616 if (range_end_pfn > prev_end_pfn)
Mel Gorman0c6cb972006-10-28 10:38:59 -07003617 hole_pages += range_end_pfn -
Mel Gorman9c7cd682006-09-27 01:49:58 -07003618 max(range_start_pfn, prev_end_pfn);
3619
Mel Gormanc7132162006-09-27 01:49:43 -07003620 return hole_pages;
3621}
3622
3623/**
3624 * absent_pages_in_range - Return number of page frames in holes within a range
3625 * @start_pfn: The start PFN to start searching for holes
3626 * @end_pfn: The end PFN to stop searching for holes
3627 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003628 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07003629 */
3630unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3631 unsigned long end_pfn)
3632{
3633 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3634}
3635
3636/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003637static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003638 unsigned long zone_type,
3639 unsigned long *ignored)
3640{
Mel Gorman9c7cd682006-09-27 01:49:58 -07003641 unsigned long node_start_pfn, node_end_pfn;
3642 unsigned long zone_start_pfn, zone_end_pfn;
3643
3644 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3645 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3646 node_start_pfn);
3647 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3648 node_end_pfn);
3649
Mel Gorman2a1e2742007-07-17 04:03:12 -07003650 adjust_zone_range_for_zone_movable(nid, zone_type,
3651 node_start_pfn, node_end_pfn,
3652 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07003653 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003654}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003655
Mel Gormanc7132162006-09-27 01:49:43 -07003656#else
Paul Mundt6ea6e682007-07-15 23:38:20 -07003657static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003658 unsigned long zone_type,
3659 unsigned long *zones_size)
3660{
3661 return zones_size[zone_type];
3662}
3663
Paul Mundt6ea6e682007-07-15 23:38:20 -07003664static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003665 unsigned long zone_type,
3666 unsigned long *zholes_size)
3667{
3668 if (!zholes_size)
3669 return 0;
3670
3671 return zholes_size[zone_type];
3672}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003673
Mel Gormanc7132162006-09-27 01:49:43 -07003674#endif
3675
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003676static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Mel Gormanc7132162006-09-27 01:49:43 -07003677 unsigned long *zones_size, unsigned long *zholes_size)
3678{
3679 unsigned long realtotalpages, totalpages = 0;
3680 enum zone_type i;
3681
3682 for (i = 0; i < MAX_NR_ZONES; i++)
3683 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3684 zones_size);
3685 pgdat->node_spanned_pages = totalpages;
3686
3687 realtotalpages = totalpages;
3688 for (i = 0; i < MAX_NR_ZONES; i++)
3689 realtotalpages -=
3690 zone_absent_pages_in_node(pgdat->node_id, i,
3691 zholes_size);
3692 pgdat->node_present_pages = realtotalpages;
3693 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3694 realtotalpages);
3695}
3696
Mel Gorman835c1342007-10-16 01:25:47 -07003697#ifndef CONFIG_SPARSEMEM
3698/*
3699 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07003700 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3701 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07003702 * round what is now in bits to nearest long in bits, then return it in
3703 * bytes.
3704 */
3705static unsigned long __init usemap_size(unsigned long zonesize)
3706{
3707 unsigned long usemapsize;
3708
Mel Gormand9c23402007-10-16 01:26:01 -07003709 usemapsize = roundup(zonesize, pageblock_nr_pages);
3710 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07003711 usemapsize *= NR_PAGEBLOCK_BITS;
3712 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3713
3714 return usemapsize / 8;
3715}
3716
3717static void __init setup_usemap(struct pglist_data *pgdat,
3718 struct zone *zone, unsigned long zonesize)
3719{
3720 unsigned long usemapsize = usemap_size(zonesize);
3721 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08003722 if (usemapsize)
Mel Gorman835c1342007-10-16 01:25:47 -07003723 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
Mel Gorman835c1342007-10-16 01:25:47 -07003724}
3725#else
3726static void inline setup_usemap(struct pglist_data *pgdat,
3727 struct zone *zone, unsigned long zonesize) {}
3728#endif /* CONFIG_SPARSEMEM */
3729
Mel Gormand9c23402007-10-16 01:26:01 -07003730#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08003731
3732/* Return a sensible default order for the pageblock size. */
3733static inline int pageblock_default_order(void)
3734{
3735 if (HPAGE_SHIFT > PAGE_SHIFT)
3736 return HUGETLB_PAGE_ORDER;
3737
3738 return MAX_ORDER-1;
3739}
3740
Mel Gormand9c23402007-10-16 01:26:01 -07003741/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3742static inline void __init set_pageblock_order(unsigned int order)
3743{
3744 /* Check that pageblock_nr_pages has not already been setup */
3745 if (pageblock_order)
3746 return;
3747
3748 /*
3749 * Assume the largest contiguous order of interest is a huge page.
3750 * This value may be variable depending on boot parameters on IA64
3751 */
3752 pageblock_order = order;
3753}
3754#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3755
Mel Gormanba72cb82007-11-28 16:21:13 -08003756/*
3757 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3758 * and pageblock_default_order() are unused as pageblock_order is set
3759 * at compile-time. See include/linux/pageblock-flags.h for the values of
3760 * pageblock_order based on the kernel config
3761 */
3762static inline int pageblock_default_order(unsigned int order)
3763{
3764 return MAX_ORDER-1;
3765}
Mel Gormand9c23402007-10-16 01:26:01 -07003766#define set_pageblock_order(x) do {} while (0)
3767
3768#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3769
Linus Torvalds1da177e2005-04-16 15:20:36 -07003770/*
3771 * Set up the zone data structures:
3772 * - mark all pages reserved
3773 * - mark all memory queues empty
3774 * - clear the memory bitmaps
3775 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08003776static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003777 unsigned long *zones_size, unsigned long *zholes_size)
3778{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07003779 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07003780 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003781 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07003782 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003783
Dave Hansen208d54e2005-10-29 18:16:52 -07003784 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003785 pgdat->nr_zones = 0;
3786 init_waitqueue_head(&pgdat->kswapd_wait);
3787 pgdat->kswapd_max_order = 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003788 pgdat_page_cgroup_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003789
3790 for (j = 0; j < MAX_NR_ZONES; j++) {
3791 struct zone *zone = pgdat->node_zones + j;
Mel Gorman0e0b8642006-09-27 01:49:56 -07003792 unsigned long size, realsize, memmap_pages;
Christoph Lameterb69408e2008-10-18 20:26:14 -07003793 enum lru_list l;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003794
Mel Gormanc7132162006-09-27 01:49:43 -07003795 size = zone_spanned_pages_in_node(nid, j, zones_size);
3796 realsize = size - zone_absent_pages_in_node(nid, j,
3797 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003798
Mel Gorman0e0b8642006-09-27 01:49:56 -07003799 /*
3800 * Adjust realsize so that it accounts for how much memory
3801 * is used by this zone for memmap. This affects the watermark
3802 * and per-cpu initialisations
3803 */
Johannes Weinerf7232152008-05-23 13:04:21 -07003804 memmap_pages =
3805 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
Mel Gorman0e0b8642006-09-27 01:49:56 -07003806 if (realsize >= memmap_pages) {
3807 realsize -= memmap_pages;
Yinghai Lu5594c8c2009-01-06 14:39:14 -08003808 if (memmap_pages)
3809 printk(KERN_DEBUG
3810 " %s zone: %lu pages used for memmap\n",
3811 zone_names[j], memmap_pages);
Mel Gorman0e0b8642006-09-27 01:49:56 -07003812 } else
3813 printk(KERN_WARNING
3814 " %s zone: %lu pages exceeds realsize %lu\n",
3815 zone_names[j], memmap_pages, realsize);
3816
Christoph Lameter62672762007-02-10 01:43:07 -08003817 /* Account for reserved pages */
3818 if (j == 0 && realsize > dma_reserve) {
Mel Gorman0e0b8642006-09-27 01:49:56 -07003819 realsize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07003820 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08003821 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07003822 }
3823
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003824 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003825 nr_kernel_pages += realsize;
3826 nr_all_pages += realsize;
3827
3828 zone->spanned_pages = size;
3829 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07003830#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07003831 zone->node = nid;
Christoph Lameter8417bba2006-09-25 23:31:51 -07003832 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07003833 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07003834 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07003835#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003836 zone->name = zone_names[j];
3837 spin_lock_init(&zone->lock);
3838 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07003839 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003840 zone->zone_pgdat = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003841
Martin Bligh3bb1a8522006-10-28 10:38:24 -07003842 zone->prev_priority = DEF_PRIORITY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003843
Dave Hansened8ece22005-10-29 18:16:50 -07003844 zone_pcp_init(zone);
Christoph Lameterb69408e2008-10-18 20:26:14 -07003845 for_each_lru(l) {
3846 INIT_LIST_HEAD(&zone->lru[l].list);
Wu Fengguangf8629632009-09-21 17:03:11 -07003847 zone->reclaim_stat.nr_saved_scan[l] = 0;
Christoph Lameterb69408e2008-10-18 20:26:14 -07003848 }
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08003849 zone->reclaim_stat.recent_rotated[0] = 0;
3850 zone->reclaim_stat.recent_rotated[1] = 0;
3851 zone->reclaim_stat.recent_scanned[0] = 0;
3852 zone->reclaim_stat.recent_scanned[1] = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07003853 zap_zone_vm_stats(zone);
David Rientjese815af92007-10-16 23:25:54 -07003854 zone->flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003855 if (!size)
3856 continue;
3857
Mel Gormanba72cb82007-11-28 16:21:13 -08003858 set_pageblock_order(pageblock_default_order());
Mel Gorman835c1342007-10-16 01:25:47 -07003859 setup_usemap(pgdat, zone, size);
Dave Hansena2f3aa022007-01-10 23:15:30 -08003860 ret = init_currently_empty_zone(zone, zone_start_pfn,
3861 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07003862 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07003863 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003864 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003865 }
3866}
3867
Sam Ravnborg577a32f2007-05-17 23:29:25 +02003868static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003869{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003870 /* Skip empty nodes */
3871 if (!pgdat->node_spanned_pages)
3872 return;
3873
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003874#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07003875 /* ia64 gets its own node_mem_map, before this, without bootmem */
3876 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07003877 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003878 struct page *map;
3879
Bob Piccoe984bb42006-05-20 15:00:31 -07003880 /*
3881 * The zone's endpoints aren't required to be MAX_ORDER
3882 * aligned but the node_mem_map endpoints must be in order
3883 * for the buddy allocator to function correctly.
3884 */
3885 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3886 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3887 end = ALIGN(end, MAX_ORDER_NR_PAGES);
3888 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07003889 map = alloc_remap(pgdat->node_id, size);
3890 if (!map)
3891 map = alloc_bootmem_node(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07003892 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003893 }
Roman Zippel12d810c2007-05-31 00:40:54 -07003894#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07003895 /*
3896 * With no DISCONTIG, the global mem_map is just set as node 0's
3897 */
Mel Gormanc7132162006-09-27 01:49:43 -07003898 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003899 mem_map = NODE_DATA(0)->node_mem_map;
Mel Gormanc7132162006-09-27 01:49:43 -07003900#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3901 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08003902 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Mel Gormanc7132162006-09-27 01:49:43 -07003903#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3904 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003905#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003906#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003907}
3908
Johannes Weiner9109fb72008-07-23 21:27:20 -07003909void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
3910 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003911{
Johannes Weiner9109fb72008-07-23 21:27:20 -07003912 pg_data_t *pgdat = NODE_DATA(nid);
3913
Linus Torvalds1da177e2005-04-16 15:20:36 -07003914 pgdat->node_id = nid;
3915 pgdat->node_start_pfn = node_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003916 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003917
3918 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07003919#ifdef CONFIG_FLAT_NODE_MEM_MAP
3920 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3921 nid, (unsigned long)pgdat,
3922 (unsigned long)pgdat->node_mem_map);
3923#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003924
3925 free_area_init_core(pgdat, zones_size, zholes_size);
3926}
3927
Mel Gormanc7132162006-09-27 01:49:43 -07003928#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07003929
3930#if MAX_NUMNODES > 1
3931/*
3932 * Figure out the number of possible node ids.
3933 */
3934static void __init setup_nr_node_ids(void)
3935{
3936 unsigned int node;
3937 unsigned int highest = 0;
3938
3939 for_each_node_mask(node, node_possible_map)
3940 highest = node;
3941 nr_node_ids = highest + 1;
3942}
3943#else
3944static inline void setup_nr_node_ids(void)
3945{
3946}
3947#endif
3948
Mel Gormanc7132162006-09-27 01:49:43 -07003949/**
3950 * add_active_range - Register a range of PFNs backed by physical memory
3951 * @nid: The node ID the range resides on
3952 * @start_pfn: The start PFN of the available physical memory
3953 * @end_pfn: The end PFN of the available physical memory
3954 *
3955 * These ranges are stored in an early_node_map[] and later used by
3956 * free_area_init_nodes() to calculate zone sizes and holes. If the
3957 * range spans a memory hole, it is up to the architecture to ensure
3958 * the memory is not freed by the bootmem allocator. If possible
3959 * the range being registered will be merged with existing ranges.
3960 */
3961void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3962 unsigned long end_pfn)
3963{
3964 int i;
3965
Mel Gorman6b74ab92008-07-23 21:26:49 -07003966 mminit_dprintk(MMINIT_TRACE, "memory_register",
3967 "Entering add_active_range(%d, %#lx, %#lx) "
3968 "%d entries of %d used\n",
3969 nid, start_pfn, end_pfn,
3970 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
Mel Gormanc7132162006-09-27 01:49:43 -07003971
Mel Gorman2dbb51c2008-07-23 21:26:52 -07003972 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
3973
Mel Gormanc7132162006-09-27 01:49:43 -07003974 /* Merge with existing active regions if possible */
3975 for (i = 0; i < nr_nodemap_entries; i++) {
3976 if (early_node_map[i].nid != nid)
3977 continue;
3978
3979 /* Skip if an existing region covers this new one */
3980 if (start_pfn >= early_node_map[i].start_pfn &&
3981 end_pfn <= early_node_map[i].end_pfn)
3982 return;
3983
3984 /* Merge forward if suitable */
3985 if (start_pfn <= early_node_map[i].end_pfn &&
3986 end_pfn > early_node_map[i].end_pfn) {
3987 early_node_map[i].end_pfn = end_pfn;
3988 return;
3989 }
3990
3991 /* Merge backward if suitable */
3992 if (start_pfn < early_node_map[i].end_pfn &&
3993 end_pfn >= early_node_map[i].start_pfn) {
3994 early_node_map[i].start_pfn = start_pfn;
3995 return;
3996 }
3997 }
3998
3999 /* Check that early_node_map is large enough */
4000 if (i >= MAX_ACTIVE_REGIONS) {
4001 printk(KERN_CRIT "More than %d memory regions, truncating\n",
4002 MAX_ACTIVE_REGIONS);
4003 return;
4004 }
4005
4006 early_node_map[i].nid = nid;
4007 early_node_map[i].start_pfn = start_pfn;
4008 early_node_map[i].end_pfn = end_pfn;
4009 nr_nodemap_entries = i + 1;
4010}
4011
4012/**
Yinghai Lucc1050b2008-06-13 19:08:52 -07004013 * remove_active_range - Shrink an existing registered range of PFNs
Mel Gormanc7132162006-09-27 01:49:43 -07004014 * @nid: The node id the range is on that should be shrunk
Yinghai Lucc1050b2008-06-13 19:08:52 -07004015 * @start_pfn: The new PFN of the range
4016 * @end_pfn: The new PFN of the range
Mel Gormanc7132162006-09-27 01:49:43 -07004017 *
4018 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004019 * The map is kept near the end physical page range that has already been
4020 * registered. This function allows an arch to shrink an existing registered
4021 * range.
Mel Gormanc7132162006-09-27 01:49:43 -07004022 */
Yinghai Lucc1050b2008-06-13 19:08:52 -07004023void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
4024 unsigned long end_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004025{
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004026 int i, j;
4027 int removed = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07004028
Yinghai Lucc1050b2008-06-13 19:08:52 -07004029 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
4030 nid, start_pfn, end_pfn);
4031
Mel Gormanc7132162006-09-27 01:49:43 -07004032 /* Find the old active region end and shrink */
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004033 for_each_active_range_index_in_nid(i, nid) {
Yinghai Lucc1050b2008-06-13 19:08:52 -07004034 if (early_node_map[i].start_pfn >= start_pfn &&
4035 early_node_map[i].end_pfn <= end_pfn) {
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004036 /* clear it */
Yinghai Lucc1050b2008-06-13 19:08:52 -07004037 early_node_map[i].start_pfn = 0;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004038 early_node_map[i].end_pfn = 0;
4039 removed = 1;
4040 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004041 }
Yinghai Lucc1050b2008-06-13 19:08:52 -07004042 if (early_node_map[i].start_pfn < start_pfn &&
4043 early_node_map[i].end_pfn > start_pfn) {
4044 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
4045 early_node_map[i].end_pfn = start_pfn;
4046 if (temp_end_pfn > end_pfn)
4047 add_active_range(nid, end_pfn, temp_end_pfn);
4048 continue;
4049 }
4050 if (early_node_map[i].start_pfn >= start_pfn &&
4051 early_node_map[i].end_pfn > end_pfn &&
4052 early_node_map[i].start_pfn < end_pfn) {
4053 early_node_map[i].start_pfn = end_pfn;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004054 continue;
4055 }
4056 }
4057
4058 if (!removed)
4059 return;
4060
4061 /* remove the blank ones */
4062 for (i = nr_nodemap_entries - 1; i > 0; i--) {
4063 if (early_node_map[i].nid != nid)
4064 continue;
4065 if (early_node_map[i].end_pfn)
4066 continue;
4067 /* we found it, get rid of it */
4068 for (j = i; j < nr_nodemap_entries - 1; j++)
4069 memcpy(&early_node_map[j], &early_node_map[j+1],
4070 sizeof(early_node_map[j]));
4071 j = nr_nodemap_entries - 1;
4072 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
4073 nr_nodemap_entries--;
4074 }
Mel Gormanc7132162006-09-27 01:49:43 -07004075}
4076
4077/**
4078 * remove_all_active_ranges - Remove all currently registered regions
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004079 *
Mel Gormanc7132162006-09-27 01:49:43 -07004080 * During discovery, it may be found that a table like SRAT is invalid
4081 * and an alternative discovery method must be used. This function removes
4082 * all currently registered regions.
4083 */
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004084void __init remove_all_active_ranges(void)
Mel Gormanc7132162006-09-27 01:49:43 -07004085{
4086 memset(early_node_map, 0, sizeof(early_node_map));
4087 nr_nodemap_entries = 0;
4088}
4089
4090/* Compare two active node_active_regions */
4091static int __init cmp_node_active_region(const void *a, const void *b)
4092{
4093 struct node_active_region *arange = (struct node_active_region *)a;
4094 struct node_active_region *brange = (struct node_active_region *)b;
4095
4096 /* Done this way to avoid overflows */
4097 if (arange->start_pfn > brange->start_pfn)
4098 return 1;
4099 if (arange->start_pfn < brange->start_pfn)
4100 return -1;
4101
4102 return 0;
4103}
4104
4105/* sort the node_map by start_pfn */
4106static void __init sort_node_map(void)
4107{
4108 sort(early_node_map, (size_t)nr_nodemap_entries,
4109 sizeof(struct node_active_region),
4110 cmp_node_active_region, NULL);
4111}
4112
Mel Gormana6af2bc2007-02-10 01:42:57 -08004113/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004114static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07004115{
4116 int i;
Mel Gormana6af2bc2007-02-10 01:42:57 -08004117 unsigned long min_pfn = ULONG_MAX;
Mel Gorman1abbfb42006-11-23 12:01:41 +00004118
Mel Gormanc7132162006-09-27 01:49:43 -07004119 /* Assuming a sorted map, the first range found has the starting pfn */
4120 for_each_active_range_index_in_nid(i, nid)
Mel Gormana6af2bc2007-02-10 01:42:57 -08004121 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004122
Mel Gormana6af2bc2007-02-10 01:42:57 -08004123 if (min_pfn == ULONG_MAX) {
4124 printk(KERN_WARNING
Paul Jackson2bc0d2612008-06-22 07:22:17 -07004125 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08004126 return 0;
4127 }
4128
4129 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004130}
4131
4132/**
4133 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4134 *
4135 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004136 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07004137 */
4138unsigned long __init find_min_pfn_with_active_regions(void)
4139{
4140 return find_min_pfn_for_node(MAX_NUMNODES);
4141}
4142
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004143/*
4144 * early_calculate_totalpages()
4145 * Sum pages in active regions for movable zone.
4146 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4147 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07004148static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07004149{
4150 int i;
4151 unsigned long totalpages = 0;
4152
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004153 for (i = 0; i < nr_nodemap_entries; i++) {
4154 unsigned long pages = early_node_map[i].end_pfn -
Mel Gorman7e63efe2007-07-17 04:03:15 -07004155 early_node_map[i].start_pfn;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004156 totalpages += pages;
4157 if (pages)
4158 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
4159 }
4160 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07004161}
4162
Mel Gorman2a1e2742007-07-17 04:03:12 -07004163/*
4164 * Find the PFN the Movable zone begins in each node. Kernel memory
4165 * is spread evenly between nodes as long as the nodes have enough
4166 * memory. When they don't, some nodes will have more kernelcore than
4167 * others
4168 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004169static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004170{
4171 int i, nid;
4172 unsigned long usable_startpfn;
4173 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07004174 /* save the state before borrow the nodemask */
4175 nodemask_t saved_node_state = node_states[N_HIGH_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004176 unsigned long totalpages = early_calculate_totalpages();
4177 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004178
Mel Gorman7e63efe2007-07-17 04:03:15 -07004179 /*
4180 * If movablecore was specified, calculate what size of
4181 * kernelcore that corresponds so that memory usable for
4182 * any allocation type is evenly spread. If both kernelcore
4183 * and movablecore are specified, then the value of kernelcore
4184 * will be used for required_kernelcore if it's greater than
4185 * what movablecore would have allowed.
4186 */
4187 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07004188 unsigned long corepages;
4189
4190 /*
4191 * Round-up so that ZONE_MOVABLE is at least as large as what
4192 * was requested by the user
4193 */
4194 required_movablecore =
4195 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
4196 corepages = totalpages - required_movablecore;
4197
4198 required_kernelcore = max(required_kernelcore, corepages);
4199 }
4200
Mel Gorman2a1e2742007-07-17 04:03:12 -07004201 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4202 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07004203 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004204
4205 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4206 find_usable_zone_for_movable();
4207 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4208
4209restart:
4210 /* Spread kernelcore memory as evenly as possible throughout nodes */
4211 kernelcore_node = required_kernelcore / usable_nodes;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004212 for_each_node_state(nid, N_HIGH_MEMORY) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004213 /*
4214 * Recalculate kernelcore_node if the division per node
4215 * now exceeds what is necessary to satisfy the requested
4216 * amount of memory for the kernel
4217 */
4218 if (required_kernelcore < kernelcore_node)
4219 kernelcore_node = required_kernelcore / usable_nodes;
4220
4221 /*
4222 * As the map is walked, we track how much memory is usable
4223 * by the kernel using kernelcore_remaining. When it is
4224 * 0, the rest of the node is usable by ZONE_MOVABLE
4225 */
4226 kernelcore_remaining = kernelcore_node;
4227
4228 /* Go through each range of PFNs within this node */
4229 for_each_active_range_index_in_nid(i, nid) {
4230 unsigned long start_pfn, end_pfn;
4231 unsigned long size_pages;
4232
4233 start_pfn = max(early_node_map[i].start_pfn,
4234 zone_movable_pfn[nid]);
4235 end_pfn = early_node_map[i].end_pfn;
4236 if (start_pfn >= end_pfn)
4237 continue;
4238
4239 /* Account for what is only usable for kernelcore */
4240 if (start_pfn < usable_startpfn) {
4241 unsigned long kernel_pages;
4242 kernel_pages = min(end_pfn, usable_startpfn)
4243 - start_pfn;
4244
4245 kernelcore_remaining -= min(kernel_pages,
4246 kernelcore_remaining);
4247 required_kernelcore -= min(kernel_pages,
4248 required_kernelcore);
4249
4250 /* Continue if range is now fully accounted */
4251 if (end_pfn <= usable_startpfn) {
4252
4253 /*
4254 * Push zone_movable_pfn to the end so
4255 * that if we have to rebalance
4256 * kernelcore across nodes, we will
4257 * not double account here
4258 */
4259 zone_movable_pfn[nid] = end_pfn;
4260 continue;
4261 }
4262 start_pfn = usable_startpfn;
4263 }
4264
4265 /*
4266 * The usable PFN range for ZONE_MOVABLE is from
4267 * start_pfn->end_pfn. Calculate size_pages as the
4268 * number of pages used as kernelcore
4269 */
4270 size_pages = end_pfn - start_pfn;
4271 if (size_pages > kernelcore_remaining)
4272 size_pages = kernelcore_remaining;
4273 zone_movable_pfn[nid] = start_pfn + size_pages;
4274
4275 /*
4276 * Some kernelcore has been met, update counts and
4277 * break if the kernelcore for this node has been
4278 * satisified
4279 */
4280 required_kernelcore -= min(required_kernelcore,
4281 size_pages);
4282 kernelcore_remaining -= size_pages;
4283 if (!kernelcore_remaining)
4284 break;
4285 }
4286 }
4287
4288 /*
4289 * If there is still required_kernelcore, we do another pass with one
4290 * less node in the count. This will push zone_movable_pfn[nid] further
4291 * along on the nodes that still have memory until kernelcore is
4292 * satisified
4293 */
4294 usable_nodes--;
4295 if (usable_nodes && required_kernelcore > usable_nodes)
4296 goto restart;
4297
4298 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4299 for (nid = 0; nid < MAX_NUMNODES; nid++)
4300 zone_movable_pfn[nid] =
4301 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07004302
4303out:
4304 /* restore the node_state */
4305 node_states[N_HIGH_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004306}
4307
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004308/* Any regular memory on that node ? */
4309static void check_for_regular_memory(pg_data_t *pgdat)
4310{
4311#ifdef CONFIG_HIGHMEM
4312 enum zone_type zone_type;
4313
4314 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4315 struct zone *zone = &pgdat->node_zones[zone_type];
4316 if (zone->present_pages)
4317 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4318 }
4319#endif
4320}
4321
Mel Gormanc7132162006-09-27 01:49:43 -07004322/**
4323 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004324 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07004325 *
4326 * This will call free_area_init_node() for each active node in the system.
4327 * Using the page ranges provided by add_active_range(), the size of each
4328 * zone in each node and their holes is calculated. If the maximum PFN
4329 * between two adjacent zones match, it is assumed that the zone is empty.
4330 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4331 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4332 * starts where the previous one ended. For example, ZONE_DMA32 starts
4333 * at arch_max_dma_pfn.
4334 */
4335void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4336{
4337 unsigned long nid;
Andrew Mortondb991002008-10-15 22:01:07 -07004338 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004339
Mel Gormana6af2bc2007-02-10 01:42:57 -08004340 /* Sort early_node_map as initialisation assumes it is sorted */
4341 sort_node_map();
4342
Mel Gormanc7132162006-09-27 01:49:43 -07004343 /* Record where the zone boundaries are */
4344 memset(arch_zone_lowest_possible_pfn, 0,
4345 sizeof(arch_zone_lowest_possible_pfn));
4346 memset(arch_zone_highest_possible_pfn, 0,
4347 sizeof(arch_zone_highest_possible_pfn));
4348 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4349 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4350 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004351 if (i == ZONE_MOVABLE)
4352 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004353 arch_zone_lowest_possible_pfn[i] =
4354 arch_zone_highest_possible_pfn[i-1];
4355 arch_zone_highest_possible_pfn[i] =
4356 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4357 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07004358 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4359 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4360
4361 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4362 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4363 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004364
Mel Gormanc7132162006-09-27 01:49:43 -07004365 /* Print out the zone ranges */
4366 printk("Zone PFN ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07004367 for (i = 0; i < MAX_NR_ZONES; i++) {
4368 if (i == ZONE_MOVABLE)
4369 continue;
Paul Jackson5dab8ec2008-06-25 05:44:40 -07004370 printk(" %-8s %0#10lx -> %0#10lx\n",
Mel Gormanc7132162006-09-27 01:49:43 -07004371 zone_names[i],
4372 arch_zone_lowest_possible_pfn[i],
4373 arch_zone_highest_possible_pfn[i]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004374 }
4375
4376 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4377 printk("Movable zone start PFN for each node\n");
4378 for (i = 0; i < MAX_NUMNODES; i++) {
4379 if (zone_movable_pfn[i])
4380 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4381 }
Mel Gormanc7132162006-09-27 01:49:43 -07004382
4383 /* Print out the early_node_map[] */
4384 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4385 for (i = 0; i < nr_nodemap_entries; i++)
Paul Jackson5dab8ec2008-06-25 05:44:40 -07004386 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004387 early_node_map[i].start_pfn,
4388 early_node_map[i].end_pfn);
4389
4390 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07004391 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08004392 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07004393 for_each_online_node(nid) {
4394 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07004395 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07004396 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004397
4398 /* Any memory on that node */
4399 if (pgdat->node_present_pages)
4400 node_set_state(nid, N_HIGH_MEMORY);
4401 check_for_regular_memory(pgdat);
Mel Gormanc7132162006-09-27 01:49:43 -07004402 }
4403}
Mel Gorman2a1e2742007-07-17 04:03:12 -07004404
Mel Gorman7e63efe2007-07-17 04:03:15 -07004405static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004406{
4407 unsigned long long coremem;
4408 if (!p)
4409 return -EINVAL;
4410
4411 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004412 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004413
Mel Gorman7e63efe2007-07-17 04:03:15 -07004414 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07004415 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4416
4417 return 0;
4418}
Mel Gormaned7ed362007-07-17 04:03:14 -07004419
Mel Gorman7e63efe2007-07-17 04:03:15 -07004420/*
4421 * kernelcore=size sets the amount of memory for use for allocations that
4422 * cannot be reclaimed or migrated.
4423 */
4424static int __init cmdline_parse_kernelcore(char *p)
4425{
4426 return cmdline_parse_core(p, &required_kernelcore);
4427}
4428
4429/*
4430 * movablecore=size sets the amount of memory for use for allocations that
4431 * can be reclaimed or migrated.
4432 */
4433static int __init cmdline_parse_movablecore(char *p)
4434{
4435 return cmdline_parse_core(p, &required_movablecore);
4436}
4437
Mel Gormaned7ed362007-07-17 04:03:14 -07004438early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004439early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07004440
Mel Gormanc7132162006-09-27 01:49:43 -07004441#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4442
Mel Gorman0e0b8642006-09-27 01:49:56 -07004443/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004444 * set_dma_reserve - set the specified number of pages reserved in the first zone
4445 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07004446 *
4447 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4448 * In the DMA zone, a significant percentage may be consumed by kernel image
4449 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004450 * function may optionally be used to account for unfreeable pages in the
4451 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4452 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07004453 */
4454void __init set_dma_reserve(unsigned long new_dma_reserve)
4455{
4456 dma_reserve = new_dma_reserve;
4457}
4458
Dave Hansen93b75042005-06-23 00:07:47 -07004459#ifndef CONFIG_NEED_MULTIPLE_NODES
Marcin Slusarz52765582008-09-02 14:35:41 -07004460struct pglist_data __refdata contig_page_data = { .bdata = &bootmem_node_data[0] };
Linus Torvalds1da177e2005-04-16 15:20:36 -07004461EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07004462#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004463
4464void __init free_area_init(unsigned long *zones_size)
4465{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004466 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004467 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4468}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004469
Linus Torvalds1da177e2005-04-16 15:20:36 -07004470static int page_alloc_cpu_notify(struct notifier_block *self,
4471 unsigned long action, void *hcpu)
4472{
4473 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004474
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07004475 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004476 drain_pages(cpu);
4477
4478 /*
4479 * Spill the event counters of the dead processor
4480 * into the current processors event counters.
4481 * This artificially elevates the count of the current
4482 * processor.
4483 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07004484 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004485
4486 /*
4487 * Zero the differential counters of the dead processor
4488 * so that the vm statistics are consistent.
4489 *
4490 * This is only okay since the processor is dead and cannot
4491 * race with what we are doing.
4492 */
Christoph Lameter2244b952006-06-30 01:55:33 -07004493 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004494 }
4495 return NOTIFY_OK;
4496}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004497
4498void __init page_alloc_init(void)
4499{
4500 hotcpu_notifier(page_alloc_cpu_notify, 0);
4501}
4502
4503/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004504 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4505 * or min_free_kbytes changes.
4506 */
4507static void calculate_totalreserve_pages(void)
4508{
4509 struct pglist_data *pgdat;
4510 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004511 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004512
4513 for_each_online_pgdat(pgdat) {
4514 for (i = 0; i < MAX_NR_ZONES; i++) {
4515 struct zone *zone = pgdat->node_zones + i;
4516 unsigned long max = 0;
4517
4518 /* Find valid and maximum lowmem_reserve in the zone */
4519 for (j = i; j < MAX_NR_ZONES; j++) {
4520 if (zone->lowmem_reserve[j] > max)
4521 max = zone->lowmem_reserve[j];
4522 }
4523
Mel Gorman41858962009-06-16 15:32:12 -07004524 /* we treat the high watermark as reserved pages. */
4525 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004526
4527 if (max > zone->present_pages)
4528 max = zone->present_pages;
4529 reserve_pages += max;
4530 }
4531 }
4532 totalreserve_pages = reserve_pages;
4533}
4534
4535/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07004536 * setup_per_zone_lowmem_reserve - called whenever
4537 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4538 * has a correct pages reserved value, so an adequate number of
4539 * pages are left in the zone after a successful __alloc_pages().
4540 */
4541static void setup_per_zone_lowmem_reserve(void)
4542{
4543 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004544 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004545
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08004546 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004547 for (j = 0; j < MAX_NR_ZONES; j++) {
4548 struct zone *zone = pgdat->node_zones + j;
4549 unsigned long present_pages = zone->present_pages;
4550
4551 zone->lowmem_reserve[j] = 0;
4552
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004553 idx = j;
4554 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004555 struct zone *lower_zone;
4556
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004557 idx--;
4558
Linus Torvalds1da177e2005-04-16 15:20:36 -07004559 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4560 sysctl_lowmem_reserve_ratio[idx] = 1;
4561
4562 lower_zone = pgdat->node_zones + idx;
4563 lower_zone->lowmem_reserve[j] = present_pages /
4564 sysctl_lowmem_reserve_ratio[idx];
4565 present_pages += lower_zone->present_pages;
4566 }
4567 }
4568 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004569
4570 /* update totalreserve_pages */
4571 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004572}
4573
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004574/**
Minchan Kimbc75d332009-06-16 15:32:48 -07004575 * setup_per_zone_wmarks - called when min_free_kbytes changes
Minchan Kimbce73942009-06-16 15:32:50 -07004576 * or when memory is hot-{added|removed}
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004577 *
Minchan Kimbc75d332009-06-16 15:32:48 -07004578 * Ensures that the watermark[min,low,high] values for each zone are set
4579 * correctly with respect to min_free_kbytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004580 */
Minchan Kimbc75d332009-06-16 15:32:48 -07004581void setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004582{
4583 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4584 unsigned long lowmem_pages = 0;
4585 struct zone *zone;
4586 unsigned long flags;
4587
4588 /* Calculate total number of !ZONE_HIGHMEM pages */
4589 for_each_zone(zone) {
4590 if (!is_highmem(zone))
4591 lowmem_pages += zone->present_pages;
4592 }
4593
4594 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07004595 u64 tmp;
4596
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07004597 spin_lock_irqsave(&zone->lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07004598 tmp = (u64)pages_min * zone->present_pages;
4599 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004600 if (is_highmem(zone)) {
4601 /*
Nick Piggin669ed172005-11-13 16:06:45 -08004602 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4603 * need highmem pages, so cap pages_min to a small
4604 * value here.
4605 *
Mel Gorman41858962009-06-16 15:32:12 -07004606 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Nick Piggin669ed172005-11-13 16:06:45 -08004607 * deltas controls asynch page reclaim, and so should
4608 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004609 */
4610 int min_pages;
4611
4612 min_pages = zone->present_pages / 1024;
4613 if (min_pages < SWAP_CLUSTER_MAX)
4614 min_pages = SWAP_CLUSTER_MAX;
4615 if (min_pages > 128)
4616 min_pages = 128;
Mel Gorman41858962009-06-16 15:32:12 -07004617 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004618 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08004619 /*
4620 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07004621 * proportionate to the zone's size.
4622 */
Mel Gorman41858962009-06-16 15:32:12 -07004623 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004624 }
4625
Mel Gorman41858962009-06-16 15:32:12 -07004626 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
4627 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Mel Gorman56fd56b2007-10-16 01:25:58 -07004628 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07004629 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004630 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004631
4632 /* update totalreserve_pages */
4633 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004634}
4635
Randy Dunlap55a44622009-09-21 17:01:20 -07004636/*
Rik van Riel556adec2008-10-18 20:26:34 -07004637 * The inactive anon list should be small enough that the VM never has to
4638 * do too much work, but large enough that each inactive page has a chance
4639 * to be referenced again before it is swapped out.
4640 *
4641 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4642 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4643 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4644 * the anonymous pages are kept on the inactive list.
4645 *
4646 * total target max
4647 * memory ratio inactive anon
4648 * -------------------------------------
4649 * 10MB 1 5MB
4650 * 100MB 1 50MB
4651 * 1GB 3 250MB
4652 * 10GB 10 0.9GB
4653 * 100GB 31 3GB
4654 * 1TB 101 10GB
4655 * 10TB 320 32GB
4656 */
Minchan Kim96cb4df2009-06-16 15:32:49 -07004657void calculate_zone_inactive_ratio(struct zone *zone)
4658{
4659 unsigned int gb, ratio;
4660
4661 /* Zone size in gigabytes */
4662 gb = zone->present_pages >> (30 - PAGE_SHIFT);
4663 if (gb)
4664 ratio = int_sqrt(10 * gb);
4665 else
4666 ratio = 1;
4667
4668 zone->inactive_ratio = ratio;
4669}
4670
Cyrill Gorcunove9bb35d2009-06-16 15:32:32 -07004671static void __init setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07004672{
4673 struct zone *zone;
4674
Minchan Kim96cb4df2009-06-16 15:32:49 -07004675 for_each_zone(zone)
4676 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07004677}
4678
Linus Torvalds1da177e2005-04-16 15:20:36 -07004679/*
4680 * Initialise min_free_kbytes.
4681 *
4682 * For small machines we want it small (128k min). For large machines
4683 * we want it large (64MB max). But it is not linear, because network
4684 * bandwidth does not increase linearly with machine size. We use
4685 *
4686 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4687 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4688 *
4689 * which yields
4690 *
4691 * 16MB: 512k
4692 * 32MB: 724k
4693 * 64MB: 1024k
4694 * 128MB: 1448k
4695 * 256MB: 2048k
4696 * 512MB: 2896k
4697 * 1024MB: 4096k
4698 * 2048MB: 5792k
4699 * 4096MB: 8192k
4700 * 8192MB: 11584k
4701 * 16384MB: 16384k
4702 */
Minchan Kimbc75d332009-06-16 15:32:48 -07004703static int __init init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004704{
4705 unsigned long lowmem_kbytes;
4706
4707 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4708
4709 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4710 if (min_free_kbytes < 128)
4711 min_free_kbytes = 128;
4712 if (min_free_kbytes > 65536)
4713 min_free_kbytes = 65536;
Minchan Kimbc75d332009-06-16 15:32:48 -07004714 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004715 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07004716 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004717 return 0;
4718}
Minchan Kimbc75d332009-06-16 15:32:48 -07004719module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004720
4721/*
4722 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4723 * that we can call two helper functions whenever min_free_kbytes
4724 * changes.
4725 */
4726int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004727 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004728{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004729 proc_dointvec(table, write, buffer, length, ppos);
Mel Gorman3b1d92c2007-05-06 14:49:30 -07004730 if (write)
Minchan Kimbc75d332009-06-16 15:32:48 -07004731 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004732 return 0;
4733}
4734
Christoph Lameter96146342006-07-03 00:24:13 -07004735#ifdef CONFIG_NUMA
4736int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004737 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07004738{
4739 struct zone *zone;
4740 int rc;
4741
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004742 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07004743 if (rc)
4744 return rc;
4745
4746 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07004747 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07004748 sysctl_min_unmapped_ratio) / 100;
4749 return 0;
4750}
Christoph Lameter0ff38492006-09-25 23:31:52 -07004751
4752int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004753 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07004754{
4755 struct zone *zone;
4756 int rc;
4757
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004758 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07004759 if (rc)
4760 return rc;
4761
4762 for_each_zone(zone)
4763 zone->min_slab_pages = (zone->present_pages *
4764 sysctl_min_slab_ratio) / 100;
4765 return 0;
4766}
Christoph Lameter96146342006-07-03 00:24:13 -07004767#endif
4768
Linus Torvalds1da177e2005-04-16 15:20:36 -07004769/*
4770 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4771 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4772 * whenever sysctl_lowmem_reserve_ratio changes.
4773 *
4774 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07004775 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07004776 * if in function of the boot time zone sizes.
4777 */
4778int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004779 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004780{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004781 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004782 setup_per_zone_lowmem_reserve();
4783 return 0;
4784}
4785
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004786/*
4787 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4788 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4789 * can have before it gets flushed back to buddy allocator.
4790 */
4791
4792int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004793 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004794{
4795 struct zone *zone;
4796 unsigned int cpu;
4797 int ret;
4798
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004799 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004800 if (!write || (ret == -EINVAL))
4801 return ret;
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07004802 for_each_populated_zone(zone) {
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004803 for_each_online_cpu(cpu) {
4804 unsigned long high;
4805 high = zone->present_pages / percpu_pagelist_fraction;
4806 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4807 }
4808 }
4809 return 0;
4810}
4811
David S. Millerf034b5d2006-08-24 03:08:07 -07004812int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004813
4814#ifdef CONFIG_NUMA
4815static int __init set_hashdist(char *str)
4816{
4817 if (!str)
4818 return 0;
4819 hashdist = simple_strtoul(str, &str, 0);
4820 return 1;
4821}
4822__setup("hashdist=", set_hashdist);
4823#endif
4824
4825/*
4826 * allocate a large system hash table from bootmem
4827 * - it is assumed that the hash table must contain an exact power-of-2
4828 * quantity of entries
4829 * - limit is the number of hash buckets, not the total allocation size
4830 */
4831void *__init alloc_large_system_hash(const char *tablename,
4832 unsigned long bucketsize,
4833 unsigned long numentries,
4834 int scale,
4835 int flags,
4836 unsigned int *_hash_shift,
4837 unsigned int *_hash_mask,
4838 unsigned long limit)
4839{
4840 unsigned long long max = limit;
4841 unsigned long log2qty, size;
4842 void *table = NULL;
4843
4844 /* allow the kernel cmdline to have a say */
4845 if (!numentries) {
4846 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08004847 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004848 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4849 numentries >>= 20 - PAGE_SHIFT;
4850 numentries <<= 20 - PAGE_SHIFT;
4851
4852 /* limit to 1 bucket per 2^scale bytes of low memory */
4853 if (scale > PAGE_SHIFT)
4854 numentries >>= (scale - PAGE_SHIFT);
4855 else
4856 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08004857
4858 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07004859 if (unlikely(flags & HASH_SMALL)) {
4860 /* Makes no sense without HASH_EARLY */
4861 WARN_ON(!(flags & HASH_EARLY));
4862 if (!(numentries >> *_hash_shift)) {
4863 numentries = 1UL << *_hash_shift;
4864 BUG_ON(!numentries);
4865 }
4866 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08004867 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004868 }
John Hawkes6e692ed2006-03-25 03:08:02 -08004869 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004870
4871 /* limit allocation size to 1/16 total memory by default */
4872 if (max == 0) {
4873 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4874 do_div(max, bucketsize);
4875 }
4876
4877 if (numentries > max)
4878 numentries = max;
4879
David Howellsf0d1b0b2006-12-08 02:37:49 -08004880 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004881
4882 do {
4883 size = bucketsize << log2qty;
4884 if (flags & HASH_EARLY)
Jan Beulich74768ed2008-08-12 15:08:39 -07004885 table = alloc_bootmem_nopanic(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004886 else if (hashdist)
4887 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4888 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07004889 /*
4890 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07004891 * some pages at the end of hash table which
4892 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07004893 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01004894 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07004895 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01004896 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
4897 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004898 }
4899 } while (!table && size > PAGE_SIZE && --log2qty);
4900
4901 if (!table)
4902 panic("Failed to allocate %s hash table\n", tablename);
4903
Dan Alonib49ad482007-07-15 23:38:23 -07004904 printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07004905 tablename,
4906 (1U << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08004907 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004908 size);
4909
4910 if (_hash_shift)
4911 *_hash_shift = log2qty;
4912 if (_hash_mask)
4913 *_hash_mask = (1 << log2qty) - 1;
4914
4915 return table;
4916}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004917
Mel Gorman835c1342007-10-16 01:25:47 -07004918/* Return a pointer to the bitmap storing bits affecting a block of pages */
4919static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4920 unsigned long pfn)
4921{
4922#ifdef CONFIG_SPARSEMEM
4923 return __pfn_to_section(pfn)->pageblock_flags;
4924#else
4925 return zone->pageblock_flags;
4926#endif /* CONFIG_SPARSEMEM */
4927}
Andrew Morton6220ec72006-10-19 23:29:05 -07004928
Mel Gorman835c1342007-10-16 01:25:47 -07004929static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4930{
4931#ifdef CONFIG_SPARSEMEM
4932 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07004933 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07004934#else
4935 pfn = pfn - zone->zone_start_pfn;
Mel Gormand9c23402007-10-16 01:26:01 -07004936 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07004937#endif /* CONFIG_SPARSEMEM */
4938}
4939
4940/**
Mel Gormand9c23402007-10-16 01:26:01 -07004941 * 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 -07004942 * @page: The page within the block of interest
4943 * @start_bitidx: The first bit of interest to retrieve
4944 * @end_bitidx: The last bit of interest
4945 * returns pageblock_bits flags
4946 */
4947unsigned long get_pageblock_flags_group(struct page *page,
4948 int start_bitidx, int end_bitidx)
4949{
4950 struct zone *zone;
4951 unsigned long *bitmap;
4952 unsigned long pfn, bitidx;
4953 unsigned long flags = 0;
4954 unsigned long value = 1;
4955
4956 zone = page_zone(page);
4957 pfn = page_to_pfn(page);
4958 bitmap = get_pageblock_bitmap(zone, pfn);
4959 bitidx = pfn_to_bitidx(zone, pfn);
4960
4961 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4962 if (test_bit(bitidx + start_bitidx, bitmap))
4963 flags |= value;
4964
4965 return flags;
4966}
4967
4968/**
Mel Gormand9c23402007-10-16 01:26:01 -07004969 * 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 -07004970 * @page: The page within the block of interest
4971 * @start_bitidx: The first bit of interest
4972 * @end_bitidx: The last bit of interest
4973 * @flags: The flags to set
4974 */
4975void set_pageblock_flags_group(struct page *page, unsigned long flags,
4976 int start_bitidx, int end_bitidx)
4977{
4978 struct zone *zone;
4979 unsigned long *bitmap;
4980 unsigned long pfn, bitidx;
4981 unsigned long value = 1;
4982
4983 zone = page_zone(page);
4984 pfn = page_to_pfn(page);
4985 bitmap = get_pageblock_bitmap(zone, pfn);
4986 bitidx = pfn_to_bitidx(zone, pfn);
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004987 VM_BUG_ON(pfn < zone->zone_start_pfn);
4988 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
Mel Gorman835c1342007-10-16 01:25:47 -07004989
4990 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4991 if (flags & value)
4992 __set_bit(bitidx + start_bitidx, bitmap);
4993 else
4994 __clear_bit(bitidx + start_bitidx, bitmap);
4995}
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07004996
4997/*
4998 * This is designed as sub function...plz see page_isolation.c also.
4999 * set/clear page block's type to be ISOLATE.
5000 * page allocater never alloc memory from ISOLATE block.
5001 */
5002
5003int set_migratetype_isolate(struct page *page)
5004{
5005 struct zone *zone;
5006 unsigned long flags;
5007 int ret = -EBUSY;
Shaohua Li8e7e40d2009-09-21 17:01:18 -07005008 int zone_idx;
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07005009
5010 zone = page_zone(page);
Shaohua Li8e7e40d2009-09-21 17:01:18 -07005011 zone_idx = zone_idx(zone);
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07005012 spin_lock_irqsave(&zone->lock, flags);
5013 /*
5014 * In future, more migrate types will be able to be isolation target.
5015 */
Shaohua Li8e7e40d2009-09-21 17:01:18 -07005016 if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE &&
5017 zone_idx != ZONE_MOVABLE)
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07005018 goto out;
5019 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
5020 move_freepages_block(zone, page, MIGRATE_ISOLATE);
5021 ret = 0;
5022out:
5023 spin_unlock_irqrestore(&zone->lock, flags);
5024 if (!ret)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005025 drain_all_pages();
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07005026 return ret;
5027}
5028
5029void unset_migratetype_isolate(struct page *page)
5030{
5031 struct zone *zone;
5032 unsigned long flags;
5033 zone = page_zone(page);
5034 spin_lock_irqsave(&zone->lock, flags);
5035 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
5036 goto out;
5037 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5038 move_freepages_block(zone, page, MIGRATE_MOVABLE);
5039out:
5040 spin_unlock_irqrestore(&zone->lock, flags);
5041}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07005042
5043#ifdef CONFIG_MEMORY_HOTREMOVE
5044/*
5045 * All pages in the range must be isolated before calling this.
5046 */
5047void
5048__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
5049{
5050 struct page *page;
5051 struct zone *zone;
5052 int order, i;
5053 unsigned long pfn;
5054 unsigned long flags;
5055 /* find the first valid pfn */
5056 for (pfn = start_pfn; pfn < end_pfn; pfn++)
5057 if (pfn_valid(pfn))
5058 break;
5059 if (pfn == end_pfn)
5060 return;
5061 zone = page_zone(pfn_to_page(pfn));
5062 spin_lock_irqsave(&zone->lock, flags);
5063 pfn = start_pfn;
5064 while (pfn < end_pfn) {
5065 if (!pfn_valid(pfn)) {
5066 pfn++;
5067 continue;
5068 }
5069 page = pfn_to_page(pfn);
5070 BUG_ON(page_count(page));
5071 BUG_ON(!PageBuddy(page));
5072 order = page_order(page);
5073#ifdef CONFIG_DEBUG_VM
5074 printk(KERN_INFO "remove from free list %lx %d %lx\n",
5075 pfn, 1 << order, end_pfn);
5076#endif
5077 list_del(&page->lru);
5078 rmv_page_order(page);
5079 zone->free_area[order].nr_free--;
5080 __mod_zone_page_state(zone, NR_FREE_PAGES,
5081 - (1UL << order));
5082 for (i = 0; i < (1 << order); i++)
5083 SetPageReserved((page+i));
5084 pfn += (1 << order);
5085 }
5086 spin_unlock_irqrestore(&zone->lock, flags);
5087}
5088#endif