blob: d03c946d556681ae9b67fafc857819d464f5b7a7 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/page_alloc.c
3 *
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
6 *
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
15 */
16
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/stddef.h>
18#include <linux/mm.h>
19#include <linux/swap.h>
20#include <linux/interrupt.h>
21#include <linux/pagemap.h>
KOSAKI Motohiro10ed2732008-03-04 14:28:32 -080022#include <linux/jiffies.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/bootmem.h>
24#include <linux/compiler.h>
Randy Dunlap9f158332005-09-13 01:25:16 -070025#include <linux/kernel.h>
Vegard Nossumb1eeab62008-11-25 16:55:53 +010026#include <linux/kmemcheck.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/module.h>
28#include <linux/suspend.h>
29#include <linux/pagevec.h>
30#include <linux/blkdev.h>
31#include <linux/slab.h>
David Rientjes5a3135c2007-10-16 23:25:53 -070032#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include <linux/notifier.h>
34#include <linux/topology.h>
35#include <linux/sysctl.h>
36#include <linux/cpu.h>
37#include <linux/cpuset.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070038#include <linux/memory_hotplug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include <linux/nodemask.h>
40#include <linux/vmalloc.h>
Christoph Lameter4be38e32006-01-06 00:11:17 -080041#include <linux/mempolicy.h>
Yasunori Goto68113782006-06-23 02:03:11 -070042#include <linux/stop_machine.h>
Mel Gormanc7132162006-09-27 01:49:43 -070043#include <linux/sort.h>
44#include <linux/pfn.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070045#include <linux/backing-dev.h>
Akinobu Mita933e3122006-12-08 02:39:45 -080046#include <linux/fault-inject.h>
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -070047#include <linux/page-isolation.h>
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070048#include <linux/page_cgroup.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070049#include <linux/debugobjects.h>
Catalin Marinasdbb1f812009-06-11 13:23:19 +010050#include <linux/kmemleak.h>
Robert Jennings925cc712009-12-17 14:44:38 +000051#include <linux/memory.h>
Mel Gorman0d3d0622009-09-21 17:02:44 -070052#include <trace/events/kmem.h>
Wu Fengguang718a3822010-03-10 15:20:43 -080053#include <linux/ftrace_event.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
55#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070056#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070057#include "internal.h"
58
59/*
Christoph Lameter13808912007-10-16 01:25:27 -070060 * Array of node states.
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 */
Christoph Lameter13808912007-10-16 01:25:27 -070062nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
63 [N_POSSIBLE] = NODE_MASK_ALL,
64 [N_ONLINE] = { { [0] = 1UL } },
65#ifndef CONFIG_NUMA
66 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
67#ifdef CONFIG_HIGHMEM
68 [N_HIGH_MEMORY] = { { [0] = 1UL } },
69#endif
70 [N_CPU] = { { [0] = 1UL } },
71#endif /* NUMA */
72};
73EXPORT_SYMBOL(node_states);
74
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070075unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -070076unsigned long totalreserve_pages __read_mostly;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -080077int percpu_pagelist_fraction;
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +100078gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -070079
Rafael J. Wysocki452aa692010-03-05 13:42:13 -080080#ifdef CONFIG_PM_SLEEP
81/*
82 * The following functions are used by the suspend/hibernate code to temporarily
83 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
84 * while devices are suspended. To avoid races with the suspend/hibernate code,
85 * they should always be called with pm_mutex held (gfp_allowed_mask also should
86 * only be modified with pm_mutex held, unless the suspend/hibernate code is
87 * guaranteed not to run in parallel with that modification).
88 */
89void set_gfp_allowed_mask(gfp_t mask)
90{
91 WARN_ON(!mutex_is_locked(&pm_mutex));
92 gfp_allowed_mask = mask;
93}
94
95gfp_t clear_gfp_allowed_mask(gfp_t mask)
96{
97 gfp_t ret = gfp_allowed_mask;
98
99 WARN_ON(!mutex_is_locked(&pm_mutex));
100 gfp_allowed_mask &= ~mask;
101 return ret;
102}
103#endif /* CONFIG_PM_SLEEP */
104
Mel Gormand9c23402007-10-16 01:26:01 -0700105#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
106int pageblock_order __read_mostly;
107#endif
108
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800109static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -0800110
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111/*
112 * results with 256, 32 in the lowmem_reserve sysctl:
113 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
114 * 1G machine -> (16M dma, 784M normal, 224M high)
115 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
116 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
117 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +0100118 *
119 * TBD: should special case ZONE_DMA32 machines here - in those we normally
120 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700122int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800123#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700124 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -0800125#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700126#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700127 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700128#endif
Christoph Lametere53ef382006-09-25 23:31:14 -0700129#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700130 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700131#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700132 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700133};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134
135EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136
Helge Deller15ad7cd2006-12-06 20:40:36 -0800137static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800138#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700139 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800140#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700141#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700142 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700143#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700144 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700145#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700146 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700147#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700148 "Movable",
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700149};
150
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151int min_free_kbytes = 1024;
152
Jan Beulich2c85f512009-09-21 17:03:07 -0700153static unsigned long __meminitdata nr_kernel_pages;
154static unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700155static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156
Mel Gormanc7132162006-09-27 01:49:43 -0700157#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
158 /*
Simon Arlott183ff222007-10-20 01:27:18 +0200159 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
Mel Gormanc7132162006-09-27 01:49:43 -0700160 * ranges of memory (RAM) that may be registered with add_active_range().
161 * Ranges passed to add_active_range() will be merged if possible
162 * so the number of times add_active_range() can be called is
163 * related to the number of nodes and the number of holes
164 */
165 #ifdef CONFIG_MAX_ACTIVE_REGIONS
166 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
167 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
168 #else
169 #if MAX_NUMNODES >= 32
170 /* If there can be many nodes, allow up to 50 holes per node */
171 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
172 #else
173 /* By default, allow up to 256 distinct regions */
174 #define MAX_ACTIVE_REGIONS 256
175 #endif
176 #endif
177
Jan Beulich98011f52007-07-15 23:38:17 -0700178 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
179 static int __meminitdata nr_nodemap_entries;
180 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
181 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
Adrian Bunkb69a7282008-07-23 21:28:12 -0700182 static unsigned long __initdata required_kernelcore;
Adrian Bunk484f51f2007-10-16 01:26:03 -0700183 static unsigned long __initdata required_movablecore;
Adrian Bunkb69a7282008-07-23 21:28:12 -0700184 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Mel Gorman2a1e2742007-07-17 04:03:12 -0700185
186 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
187 int movable_zone;
188 EXPORT_SYMBOL(movable_zone);
Mel Gormanc7132162006-09-27 01:49:43 -0700189#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
190
Miklos Szeredi418508c2007-05-23 13:57:55 -0700191#if MAX_NUMNODES > 1
192int nr_node_ids __read_mostly = MAX_NUMNODES;
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700193int nr_online_nodes __read_mostly = 1;
Miklos Szeredi418508c2007-05-23 13:57:55 -0700194EXPORT_SYMBOL(nr_node_ids);
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700195EXPORT_SYMBOL(nr_online_nodes);
Miklos Szeredi418508c2007-05-23 13:57:55 -0700196#endif
197
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700198int page_group_by_mobility_disabled __read_mostly;
199
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700200static void set_pageblock_migratetype(struct page *page, int migratetype)
201{
Mel Gorman49255c62009-06-16 15:31:58 -0700202
203 if (unlikely(page_group_by_mobility_disabled))
204 migratetype = MIGRATE_UNMOVABLE;
205
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700206 set_pageblock_flags_group(page, (unsigned long)migratetype,
207 PB_migrate, PB_migrate_end);
208}
209
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -0700210bool oom_killer_disabled __read_mostly;
211
Nick Piggin13e74442006-01-06 00:10:58 -0800212#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700213static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700215 int ret = 0;
216 unsigned seq;
217 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -0700218
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700219 do {
220 seq = zone_span_seqbegin(zone);
221 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
222 ret = 1;
223 else if (pfn < zone->zone_start_pfn)
224 ret = 1;
225 } while (zone_span_seqretry(zone, seq));
226
227 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700228}
229
230static int page_is_consistent(struct zone *zone, struct page *page)
231{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700232 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700233 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700235 return 0;
236
237 return 1;
238}
239/*
240 * Temporary debugging check for pages not lying within a given zone.
241 */
242static int bad_range(struct zone *zone, struct page *page)
243{
244 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700246 if (!page_is_consistent(zone, page))
247 return 1;
248
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 return 0;
250}
Nick Piggin13e74442006-01-06 00:10:58 -0800251#else
252static inline int bad_range(struct zone *zone, struct page *page)
253{
254 return 0;
255}
256#endif
257
Nick Piggin224abf92006-01-06 00:11:11 -0800258static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800260 static unsigned long resume;
261 static unsigned long nr_shown;
262 static unsigned long nr_unshown;
263
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200264 /* Don't complain about poisoned pages */
265 if (PageHWPoison(page)) {
266 __ClearPageBuddy(page);
267 return;
268 }
269
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800270 /*
271 * Allow a burst of 60 reports, then keep quiet for that minute;
272 * or allow a steady drip of one report per second.
273 */
274 if (nr_shown == 60) {
275 if (time_before(jiffies, resume)) {
276 nr_unshown++;
277 goto out;
278 }
279 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800280 printk(KERN_ALERT
281 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800282 nr_unshown);
283 nr_unshown = 0;
284 }
285 nr_shown = 0;
286 }
287 if (nr_shown++ == 0)
288 resume = jiffies + 60 * HZ;
289
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800290 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800291 current->comm, page_to_pfn(page));
Wu Fengguang718a3822010-03-10 15:20:43 -0800292 dump_page(page);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700293
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800295out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800296 /* Leave bad fields for debug, except PageBuddy could make trouble */
297 __ClearPageBuddy(page);
Randy Dunlap9f158332005-09-13 01:25:16 -0700298 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299}
300
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301/*
302 * Higher-order pages are called "compound pages". They are structured thusly:
303 *
304 * The first PAGE_SIZE page is called the "head page".
305 *
306 * The remaining PAGE_SIZE pages are called "tail pages".
307 *
308 * All pages have PG_compound set. All pages have their ->private pointing at
309 * the head page (even the head page has this).
310 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800311 * The first tail page's ->lru.next holds the address of the compound page's
312 * put_page() function. Its ->lru.prev holds the order of allocation.
313 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800315
316static void free_compound_page(struct page *page)
317{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700318 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800319}
320
Andi Kleen01ad1c02008-07-23 21:27:46 -0700321void prep_compound_page(struct page *page, unsigned long order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322{
323 int i;
324 int nr_pages = 1 << order;
325
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800326 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700327 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700328 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800329 for (i = 1; i < nr_pages; i++) {
330 struct page *p = page + i;
331
Christoph Lameterd85f3382007-05-06 14:49:39 -0700332 __SetPageTail(p);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700333 p->first_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 }
335}
336
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800337static int destroy_compound_page(struct page *page, unsigned long order)
Andy Whitcroft18229df2008-11-06 12:53:27 -0800338{
339 int i;
340 int nr_pages = 1 << order;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800341 int bad = 0;
Andy Whitcroft18229df2008-11-06 12:53:27 -0800342
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800343 if (unlikely(compound_order(page) != order) ||
344 unlikely(!PageHead(page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800345 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800346 bad++;
347 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348
Christoph Lameter6d777952007-05-06 14:49:40 -0700349 __ClearPageHead(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800350
Andy Whitcroft18229df2008-11-06 12:53:27 -0800351 for (i = 1; i < nr_pages; i++) {
352 struct page *p = page + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353
Alexey Zaytseve713a212009-01-10 02:47:57 +0300354 if (unlikely(!PageTail(p) || (p->first_page != page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800355 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800356 bad++;
357 }
Christoph Lameterd85f3382007-05-06 14:49:39 -0700358 __ClearPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800360
361 return bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363
Nick Piggin17cf4402006-03-22 00:08:41 -0800364static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
365{
366 int i;
367
Andrew Morton6626c5d2006-03-22 00:08:42 -0800368 /*
369 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
370 * and __GFP_HIGHMEM from hard or soft interrupt context.
371 */
Nick Piggin725d7042006-09-25 23:30:55 -0700372 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800373 for (i = 0; i < (1 << order); i++)
374 clear_highpage(page + i);
375}
376
Andrew Morton6aa30012006-04-18 22:20:52 -0700377static inline void set_page_order(struct page *page, int order)
378{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700379 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000380 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381}
382
383static inline void rmv_page_order(struct page *page)
384{
Nick Piggin676165a2006-04-10 11:21:48 +1000385 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700386 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387}
388
389/*
390 * Locate the struct page for both the matching buddy in our
391 * pair (buddy1) and the combined O(n+1) page they form (page).
392 *
393 * 1) Any buddy B1 will have an order O twin B2 which satisfies
394 * the following equation:
395 * B2 = B1 ^ (1 << O)
396 * For example, if the starting buddy (buddy2) is #8 its order
397 * 1 buddy is #10:
398 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
399 *
400 * 2) Any buddy B will have an order O+1 parent P which
401 * satisfies the following equation:
402 * P = B & ~(1 << O)
403 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200404 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 */
406static inline struct page *
407__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
408{
409 unsigned long buddy_idx = page_idx ^ (1 << order);
410
411 return page + (buddy_idx - page_idx);
412}
413
414static inline unsigned long
415__find_combined_index(unsigned long page_idx, unsigned int order)
416{
417 return (page_idx & ~(1 << order));
418}
419
420/*
421 * This function checks whether a page is free && is the buddy
422 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800423 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000424 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700425 * (c) a page and its buddy have the same order &&
426 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 *
Nick Piggin676165a2006-04-10 11:21:48 +1000428 * For recording whether a page is in the buddy system, we use PG_buddy.
429 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
430 *
431 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700433static inline int page_is_buddy(struct page *page, struct page *buddy,
434 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700436 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800437 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800438
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700439 if (page_zone_id(page) != page_zone_id(buddy))
440 return 0;
441
442 if (PageBuddy(buddy) && page_order(buddy) == order) {
Nick Piggina3af9c32009-06-16 15:32:10 -0700443 VM_BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa30012006-04-18 22:20:52 -0700444 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000445 }
Andrew Morton6aa30012006-04-18 22:20:52 -0700446 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447}
448
449/*
450 * Freeing function for a buddy system allocator.
451 *
452 * The concept of a buddy system is to maintain direct-mapped table
453 * (containing bit values) for memory blocks of various "orders".
454 * The bottom level table contains the map for the smallest allocatable
455 * units of memory (here, pages), and each level above it describes
456 * pairs of units from the levels below, hence, "buddies".
457 * At a high level, all that happens here is marking the table entry
458 * at the bottom level available, and propagating the changes upward
459 * as necessary, plus some accounting needed to play nicely with other
460 * parts of the VM system.
461 * At each level, we keep a list of pages, which are heads of continuous
Nick Piggin676165a2006-04-10 11:21:48 +1000462 * free pages of length of (1 << order) and marked with PG_buddy. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700463 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 * So when we are allocating or freeing one, we can derive the state of the
465 * other. That is, if we allocate a small block, and both were
466 * free, the remainder of the region must be split into blocks.
467 * If a block is freed, and its buddy is also free, then this
468 * triggers coalescing into a block of larger size.
469 *
470 * -- wli
471 */
472
Nick Piggin48db57f2006-01-08 01:00:42 -0800473static inline void __free_one_page(struct page *page,
Mel Gormaned0ae212009-06-16 15:32:07 -0700474 struct zone *zone, unsigned int order,
475 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476{
477 unsigned long page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478
Nick Piggin224abf92006-01-06 00:11:11 -0800479 if (unlikely(PageCompound(page)))
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800480 if (unlikely(destroy_compound_page(page, order)))
481 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482
Mel Gormaned0ae212009-06-16 15:32:07 -0700483 VM_BUG_ON(migratetype == -1);
484
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
486
Mel Gormanf2260e62009-06-16 15:32:13 -0700487 VM_BUG_ON(page_idx & ((1 << order) - 1));
Nick Piggin725d7042006-09-25 23:30:55 -0700488 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 while (order < MAX_ORDER-1) {
491 unsigned long combined_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 struct page *buddy;
493
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 buddy = __page_find_buddy(page, page_idx, order);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700495 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700496 break;
Nick Piggin13e74442006-01-06 00:10:58 -0800497
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700498 /* Our buddy is free, merge with it and move up one order. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 list_del(&buddy->lru);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700500 zone->free_area[order].nr_free--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 rmv_page_order(buddy);
Nick Piggin13e74442006-01-06 00:10:58 -0800502 combined_idx = __find_combined_index(page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 page = page + (combined_idx - page_idx);
504 page_idx = combined_idx;
505 order++;
506 }
507 set_page_order(page, order);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700508 list_add(&page->lru,
509 &zone->free_area[order].free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 zone->free_area[order].nr_free++;
511}
512
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700513/*
514 * free_page_mlock() -- clean up attempts to free and mlocked() page.
515 * Page should not be on lru, so no need to fix that up.
516 * free_pages_check() will verify...
517 */
518static inline void free_page_mlock(struct page *page)
519{
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700520 __dec_zone_page_state(page, NR_MLOCK);
521 __count_vm_event(UNEVICTABLE_MLOCKFREED);
522}
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700523
Nick Piggin224abf92006-01-06 00:11:11 -0800524static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525{
Nick Piggin92be2e32006-01-06 00:10:57 -0800526 if (unlikely(page_mapcount(page) |
527 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700528 (atomic_read(&page->_count) != 0) |
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800529 (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800530 bad_page(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800531 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800532 }
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800533 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
534 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
535 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536}
537
538/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700539 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700541 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 *
543 * If the zone was previously in an "all pages pinned" state then look to
544 * see if this freeing clears that state.
545 *
546 * And clear the zone's pages_scanned counter, to hold off the "all pages are
547 * pinned" detection logic.
548 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700549static void free_pcppages_bulk(struct zone *zone, int count,
550 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551{
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700552 int migratetype = 0;
Mel Gormana6f9edd2009-09-21 17:03:20 -0700553 int batch_free = 0;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700554
Nick Pigginc54ad302006-01-06 00:10:56 -0800555 spin_lock(&zone->lock);
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -0800556 zone->all_unreclaimable = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700558
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700559 __mod_zone_page_state(zone, NR_FREE_PAGES, count);
Mel Gormana6f9edd2009-09-21 17:03:20 -0700560 while (count) {
Nick Piggin48db57f2006-01-08 01:00:42 -0800561 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700562 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -0800563
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700564 /*
Mel Gormana6f9edd2009-09-21 17:03:20 -0700565 * Remove pages from lists in a round-robin fashion. A
566 * batch_free count is maintained that is incremented when an
567 * empty list is encountered. This is so more pages are freed
568 * off fuller lists instead of spinning excessively around empty
569 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700570 */
571 do {
Mel Gormana6f9edd2009-09-21 17:03:20 -0700572 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700573 if (++migratetype == MIGRATE_PCPTYPES)
574 migratetype = 0;
575 list = &pcp->lists[migratetype];
576 } while (list_empty(list));
577
Mel Gormana6f9edd2009-09-21 17:03:20 -0700578 do {
579 page = list_entry(list->prev, struct page, lru);
580 /* must delete as __free_one_page list manipulates */
581 list_del(&page->lru);
Hugh Dickinsa7016232010-01-29 17:46:34 +0000582 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
583 __free_one_page(page, zone, 0, page_private(page));
584 trace_mm_page_pcpu_drain(page, 0, page_private(page));
Mel Gormana6f9edd2009-09-21 17:03:20 -0700585 } while (--count && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800587 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588}
589
Mel Gormaned0ae212009-06-16 15:32:07 -0700590static void free_one_page(struct zone *zone, struct page *page, int order,
591 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800592{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700593 spin_lock(&zone->lock);
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -0800594 zone->all_unreclaimable = 0;
Christoph Lameter006d22d2006-09-25 23:31:48 -0700595 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700596
597 __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
Mel Gormaned0ae212009-06-16 15:32:07 -0700598 __free_one_page(page, zone, order, migratetype);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700599 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800600}
601
602static void __free_pages_ok(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603{
Nick Pigginc54ad302006-01-06 00:10:56 -0800604 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 int i;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800606 int bad = 0;
Johannes Weiner451ea252009-09-21 17:01:48 -0700607 int wasMlocked = __TestClearPageMlocked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608
Li Hongf6503162010-03-05 13:41:52 -0800609 trace_mm_page_free_direct(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100610 kmemcheck_free_shadow(page, order);
611
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 for (i = 0 ; i < (1 << order) ; ++i)
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800613 bad += free_pages_check(page + i);
614 if (bad)
Hugh Dickins689bceb2005-11-21 21:32:20 -0800615 return;
616
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700617 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -0700618 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700619 debug_check_no_obj_freed(page_address(page),
620 PAGE_SIZE << order);
621 }
Nick Piggindafb1362006-10-11 01:21:30 -0700622 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800623 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700624
Nick Pigginc54ad302006-01-06 00:10:56 -0800625 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +0200626 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -0700627 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700628 __count_vm_events(PGFREE, 1 << order);
Mel Gormaned0ae212009-06-16 15:32:07 -0700629 free_one_page(page_zone(page), page, order,
630 get_pageblock_migratetype(page));
Nick Pigginc54ad302006-01-06 00:10:56 -0800631 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632}
633
David Howellsa226f6c2006-01-06 00:11:08 -0800634/*
635 * permit the bootmem allocator to evade page validation on high-order frees
636 */
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700637void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800638{
639 if (order == 0) {
640 __ClearPageReserved(page);
641 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800642 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800643 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800644 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800645 int loop;
646
Nick Piggin545b1ea2006-03-22 00:08:07 -0800647 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800648 for (loop = 0; loop < BITS_PER_LONG; loop++) {
649 struct page *p = &page[loop];
650
Nick Piggin545b1ea2006-03-22 00:08:07 -0800651 if (loop + 1 < BITS_PER_LONG)
652 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800653 __ClearPageReserved(p);
654 set_page_count(p, 0);
655 }
656
Nick Piggin7835e982006-03-22 00:08:40 -0800657 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800658 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800659 }
660}
661
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662
663/*
664 * The order of subdivision here is critical for the IO subsystem.
665 * Please do not alter this order without good reasons and regression
666 * testing. Specifically, as large blocks of memory are subdivided,
667 * the order in which smaller blocks are delivered depends on the order
668 * they're subdivided in this function. This is the primary factor
669 * influencing the order in which pages are delivered to the IO
670 * subsystem according to empirical testing, and this is also justified
671 * by considering the behavior of a buddy system containing a single
672 * large block of memory acted on by a series of small allocations.
673 * This behavior is a critical factor in sglist merging's success.
674 *
675 * -- wli
676 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800677static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700678 int low, int high, struct free_area *area,
679 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680{
681 unsigned long size = 1 << high;
682
683 while (high > low) {
684 area--;
685 high--;
686 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700687 VM_BUG_ON(bad_range(zone, &page[size]));
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700688 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 area->nr_free++;
690 set_page_order(&page[size], high);
691 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692}
693
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694/*
695 * This page is about to be returned from the page allocator
696 */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200697static inline int check_new_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698{
Nick Piggin92be2e32006-01-06 00:10:57 -0800699 if (unlikely(page_mapcount(page) |
700 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700701 (atomic_read(&page->_count) != 0) |
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800702 (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800703 bad_page(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800704 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800705 }
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200706 return 0;
707}
708
709static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
710{
711 int i;
712
713 for (i = 0; i < (1 << order); i++) {
714 struct page *p = page + i;
715 if (unlikely(check_new_page(p)))
716 return 1;
717 }
Hugh Dickins689bceb2005-11-21 21:32:20 -0800718
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700719 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800720 set_page_refcounted(page);
Nick Piggincc102502006-12-06 20:32:00 -0800721
722 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800724
725 if (gfp_flags & __GFP_ZERO)
726 prep_zero_page(page, order, gfp_flags);
727
728 if (order && (gfp_flags & __GFP_COMP))
729 prep_compound_page(page, order);
730
Hugh Dickins689bceb2005-11-21 21:32:20 -0800731 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732}
733
Mel Gorman56fd56b2007-10-16 01:25:58 -0700734/*
735 * Go through the free lists for the given migratetype and remove
736 * the smallest available page from the freelists
737 */
Mel Gorman728ec982009-06-16 15:32:04 -0700738static inline
739struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -0700740 int migratetype)
741{
742 unsigned int current_order;
743 struct free_area * area;
744 struct page *page;
745
746 /* Find a page of the appropriate size in the preferred list */
747 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
748 area = &(zone->free_area[current_order]);
749 if (list_empty(&area->free_list[migratetype]))
750 continue;
751
752 page = list_entry(area->free_list[migratetype].next,
753 struct page, lru);
754 list_del(&page->lru);
755 rmv_page_order(page);
756 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -0700757 expand(zone, page, order, current_order, area, migratetype);
758 return page;
759 }
760
761 return NULL;
762}
763
764
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700765/*
766 * This array describes the order lists are fallen back to when
767 * the free lists for the desirable migrate type are depleted
768 */
769static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
Mel Gorman64c5e132007-10-16 01:25:59 -0700770 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
771 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
772 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
773 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700774};
775
Mel Gormanc361be52007-10-16 01:25:51 -0700776/*
777 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -0700778 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -0700779 * boundary. If alignment is required, use move_freepages_block()
780 */
Adrian Bunkb69a7282008-07-23 21:28:12 -0700781static int move_freepages(struct zone *zone,
782 struct page *start_page, struct page *end_page,
783 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700784{
785 struct page *page;
786 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -0700787 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -0700788
789#ifndef CONFIG_HOLES_IN_ZONE
790 /*
791 * page_zone is not safe to call in this context when
792 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
793 * anyway as we check zone boundaries in move_freepages_block().
794 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -0700795 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -0700796 */
797 BUG_ON(page_zone(start_page) != page_zone(end_page));
798#endif
799
800 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -0700801 /* Make sure we are not inadvertently changing nodes */
802 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
803
Mel Gormanc361be52007-10-16 01:25:51 -0700804 if (!pfn_valid_within(page_to_pfn(page))) {
805 page++;
806 continue;
807 }
808
809 if (!PageBuddy(page)) {
810 page++;
811 continue;
812 }
813
814 order = page_order(page);
815 list_del(&page->lru);
816 list_add(&page->lru,
817 &zone->free_area[order].free_list[migratetype]);
818 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -0700819 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -0700820 }
821
Mel Gormand1003132007-10-16 01:26:00 -0700822 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -0700823}
824
Adrian Bunkb69a7282008-07-23 21:28:12 -0700825static int move_freepages_block(struct zone *zone, struct page *page,
826 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700827{
828 unsigned long start_pfn, end_pfn;
829 struct page *start_page, *end_page;
830
831 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -0700832 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -0700833 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -0700834 end_page = start_page + pageblock_nr_pages - 1;
835 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -0700836
837 /* Do not cross zone boundaries */
838 if (start_pfn < zone->zone_start_pfn)
839 start_page = page;
840 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
841 return 0;
842
843 return move_freepages(zone, start_page, end_page, migratetype);
844}
845
Mel Gorman2f66a682009-09-21 17:02:31 -0700846static void change_pageblock_range(struct page *pageblock_page,
847 int start_order, int migratetype)
848{
849 int nr_pageblocks = 1 << (start_order - pageblock_order);
850
851 while (nr_pageblocks--) {
852 set_pageblock_migratetype(pageblock_page, migratetype);
853 pageblock_page += pageblock_nr_pages;
854 }
855}
856
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700857/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -0700858static inline struct page *
859__rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700860{
861 struct free_area * area;
862 int current_order;
863 struct page *page;
864 int migratetype, i;
865
866 /* Find the largest possible block of pages in the other list */
867 for (current_order = MAX_ORDER-1; current_order >= order;
868 --current_order) {
869 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
870 migratetype = fallbacks[start_migratetype][i];
871
Mel Gorman56fd56b2007-10-16 01:25:58 -0700872 /* MIGRATE_RESERVE handled later if necessary */
873 if (migratetype == MIGRATE_RESERVE)
874 continue;
Mel Gormane0104872007-10-16 01:25:53 -0700875
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700876 area = &(zone->free_area[current_order]);
877 if (list_empty(&area->free_list[migratetype]))
878 continue;
879
880 page = list_entry(area->free_list[migratetype].next,
881 struct page, lru);
882 area->nr_free--;
883
884 /*
Mel Gormanc361be52007-10-16 01:25:51 -0700885 * If breaking a large block of pages, move all free
Mel Gorman46dafbc2007-10-16 01:25:55 -0700886 * pages to the preferred allocation list. If falling
887 * back for a reclaimable kernel allocation, be more
888 * agressive about taking ownership of free pages
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700889 */
Mel Gormand9c23402007-10-16 01:26:01 -0700890 if (unlikely(current_order >= (pageblock_order >> 1)) ||
Mel Gormandd5d2412009-09-05 11:17:11 -0700891 start_migratetype == MIGRATE_RECLAIMABLE ||
892 page_group_by_mobility_disabled) {
Mel Gorman46dafbc2007-10-16 01:25:55 -0700893 unsigned long pages;
894 pages = move_freepages_block(zone, page,
895 start_migratetype);
896
897 /* Claim the whole block if over half of it is free */
Mel Gormandd5d2412009-09-05 11:17:11 -0700898 if (pages >= (1 << (pageblock_order-1)) ||
899 page_group_by_mobility_disabled)
Mel Gorman46dafbc2007-10-16 01:25:55 -0700900 set_pageblock_migratetype(page,
901 start_migratetype);
902
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700903 migratetype = start_migratetype;
Mel Gormanc361be52007-10-16 01:25:51 -0700904 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700905
906 /* Remove the page from the freelists */
907 list_del(&page->lru);
908 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700909
Mel Gorman2f66a682009-09-21 17:02:31 -0700910 /* Take ownership for orders >= pageblock_order */
911 if (current_order >= pageblock_order)
912 change_pageblock_range(page, current_order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700913 start_migratetype);
914
915 expand(zone, page, order, current_order, area, migratetype);
Mel Gormane0fff1b2009-09-21 17:02:42 -0700916
917 trace_mm_page_alloc_extfrag(page, order, current_order,
918 start_migratetype, migratetype);
919
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700920 return page;
921 }
922 }
923
Mel Gorman728ec982009-06-16 15:32:04 -0700924 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700925}
926
Mel Gorman56fd56b2007-10-16 01:25:58 -0700927/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 * Do the hard work of removing an element from the buddy allocator.
929 * Call me with the zone->lock already held.
930 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700931static struct page *__rmqueue(struct zone *zone, unsigned int order,
932 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 struct page *page;
935
Mel Gorman728ec982009-06-16 15:32:04 -0700936retry_reserve:
Mel Gorman56fd56b2007-10-16 01:25:58 -0700937 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938
Mel Gorman728ec982009-06-16 15:32:04 -0700939 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Mel Gorman56fd56b2007-10-16 01:25:58 -0700940 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700941
Mel Gorman728ec982009-06-16 15:32:04 -0700942 /*
943 * Use MIGRATE_RESERVE rather than fail an allocation. goto
944 * is used because __rmqueue_smallest is an inline function
945 * and we want just one call site
946 */
947 if (!page) {
948 migratetype = MIGRATE_RESERVE;
949 goto retry_reserve;
950 }
951 }
952
Mel Gorman0d3d0622009-09-21 17:02:44 -0700953 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700954 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955}
956
957/*
958 * Obtain a specified number of elements from the buddy allocator, all under
959 * a single hold of the lock, for efficiency. Add them to the supplied list.
960 * Returns the number of new pages which were placed at *list.
961 */
962static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700963 unsigned long count, struct list_head *list,
Mel Gormane084b2d2009-07-29 15:02:04 -0700964 int migratetype, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967
Nick Pigginc54ad302006-01-06 00:10:56 -0800968 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700970 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800971 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -0800973
974 /*
975 * Split buddy pages returned by expand() are received here
976 * in physical page order. The page is added to the callers and
977 * list and the list head then moves forward. From the callers
978 * perspective, the linked list is ordered by page number in
979 * some conditions. This is useful for IO devices that can
980 * merge IO requests if the physical pages are ordered
981 * properly.
982 */
Mel Gormane084b2d2009-07-29 15:02:04 -0700983 if (likely(cold == 0))
984 list_add(&page->lru, list);
985 else
986 list_add_tail(&page->lru, list);
Mel Gorman535131e2007-10-16 01:25:49 -0700987 set_page_private(page, migratetype);
Mel Gorman81eabcb2007-12-17 16:20:05 -0800988 list = &page->lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 }
Mel Gormanf2260e62009-06-16 15:32:13 -0700990 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -0800991 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800992 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993}
994
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700995#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800996/*
Christoph Lameter4037d452007-05-09 02:35:14 -0700997 * Called from the vmstat counter updater to drain pagesets of this
998 * currently executing processor on remote nodes after they have
999 * expired.
1000 *
Christoph Lameter879336c2006-03-22 00:09:08 -08001001 * Note that this function must be called with the thread pinned to
1002 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001003 */
Christoph Lameter4037d452007-05-09 02:35:14 -07001004void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001005{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001006 unsigned long flags;
Christoph Lameter4037d452007-05-09 02:35:14 -07001007 int to_drain;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001008
Christoph Lameter4037d452007-05-09 02:35:14 -07001009 local_irq_save(flags);
1010 if (pcp->count >= pcp->batch)
1011 to_drain = pcp->batch;
1012 else
1013 to_drain = pcp->count;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001014 free_pcppages_bulk(zone, to_drain, pcp);
Christoph Lameter4037d452007-05-09 02:35:14 -07001015 pcp->count -= to_drain;
1016 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001017}
1018#endif
1019
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001020/*
1021 * Drain pages of the indicated processor.
1022 *
1023 * The processor must either be the current processor and the
1024 * thread pinned to the current processor or a processor that
1025 * is not online.
1026 */
1027static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028{
Nick Pigginc54ad302006-01-06 00:10:56 -08001029 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07001032 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 struct per_cpu_pageset *pset;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001034 struct per_cpu_pages *pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001036 local_irq_save(flags);
1037 pset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001039 pcp = &pset->pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001040 free_pcppages_bulk(zone, pcp->count, pcp);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001041 pcp->count = 0;
1042 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 }
1044}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001046/*
1047 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1048 */
1049void drain_local_pages(void *arg)
1050{
1051 drain_pages(smp_processor_id());
1052}
1053
1054/*
1055 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
1056 */
1057void drain_all_pages(void)
1058{
Jens Axboe15c8b6c2008-05-09 09:39:44 +02001059 on_each_cpu(drain_local_pages, NULL, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001060}
1061
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02001062#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063
1064void mark_free_pages(struct zone *zone)
1065{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001066 unsigned long pfn, max_zone_pfn;
1067 unsigned long flags;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001068 int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 struct list_head *curr;
1070
1071 if (!zone->spanned_pages)
1072 return;
1073
1074 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001075
1076 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1077 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1078 if (pfn_valid(pfn)) {
1079 struct page *page = pfn_to_page(pfn);
1080
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001081 if (!swsusp_page_is_forbidden(page))
1082 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001083 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001085 for_each_migratetype_order(order, t) {
1086 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001087 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001089 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1090 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001091 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001092 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001093 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 spin_unlock_irqrestore(&zone->lock, flags);
1095}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001096#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097
1098/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 * Free a 0-order page
Li Hongfc916682010-03-05 13:41:54 -08001100 * cold == 1 ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 */
Li Hongfc916682010-03-05 13:41:54 -08001102void free_hot_cold_page(struct page *page, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103{
1104 struct zone *zone = page_zone(page);
1105 struct per_cpu_pages *pcp;
1106 unsigned long flags;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001107 int migratetype;
Johannes Weiner451ea252009-09-21 17:01:48 -07001108 int wasMlocked = __TestClearPageMlocked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109
Li Hongc475dab2010-03-05 13:41:53 -08001110 trace_mm_page_free_direct(page, 0);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001111 kmemcheck_free_shadow(page, 0);
1112
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 if (PageAnon(page))
1114 page->mapping = NULL;
Nick Piggin224abf92006-01-06 00:11:11 -08001115 if (free_pages_check(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001116 return;
1117
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001118 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -07001119 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001120 debug_check_no_obj_freed(page_address(page), PAGE_SIZE);
1121 }
Nick Piggindafb1362006-10-11 01:21:30 -07001122 arch_free_page(page, 0);
Hugh Dickins689bceb2005-11-21 21:32:20 -08001123 kernel_map_pages(page, 1, 0);
1124
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001125 migratetype = get_pageblock_migratetype(page);
1126 set_page_private(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +02001128 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -07001129 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001130 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001131
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001132 /*
1133 * We only track unmovable, reclaimable and movable on pcp lists.
1134 * Free ISOLATE pages back to the allocator because they are being
1135 * offlined but treat RESERVE as movable pages so we can get those
1136 * areas back if necessary. Otherwise, we may have to free
1137 * excessively into the page allocator
1138 */
1139 if (migratetype >= MIGRATE_PCPTYPES) {
1140 if (unlikely(migratetype == MIGRATE_ISOLATE)) {
1141 free_one_page(zone, page, 0, migratetype);
1142 goto out;
1143 }
1144 migratetype = MIGRATE_MOVABLE;
1145 }
1146
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001147 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001148 if (cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001149 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001150 else
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001151 list_add(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001153 if (pcp->count >= pcp->high) {
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001154 free_pcppages_bulk(zone, pcp->batch, pcp);
Nick Piggin48db57f2006-01-08 01:00:42 -08001155 pcp->count -= pcp->batch;
1156 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001157
1158out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160}
1161
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001162/*
1163 * split_page takes a non-compound higher-order page, and splits it into
1164 * n (1<<order) sub-pages: page[0..n]
1165 * Each sub-page must be freed individually.
1166 *
1167 * Note: this is probably too low level an operation for use in drivers.
1168 * Please consult with lkml before using this in your driver.
1169 */
1170void split_page(struct page *page, unsigned int order)
1171{
1172 int i;
1173
Nick Piggin725d7042006-09-25 23:30:55 -07001174 VM_BUG_ON(PageCompound(page));
1175 VM_BUG_ON(!page_count(page));
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001176
1177#ifdef CONFIG_KMEMCHECK
1178 /*
1179 * Split shadow pages too, because free(page[0]) would
1180 * otherwise free the whole shadow.
1181 */
1182 if (kmemcheck_page_is_tracked(page))
1183 split_page(virt_to_page(page[0].shadow), order);
1184#endif
1185
Nick Piggin7835e982006-03-22 00:08:40 -08001186 for (i = 1; i < (1 << order); i++)
1187 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001188}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001189
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190/*
1191 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1192 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1193 * or two.
1194 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07001195static inline
1196struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001197 struct zone *zone, int order, gfp_t gfp_flags,
1198 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199{
1200 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001201 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 int cold = !!(gfp_flags & __GFP_COLD);
1203
Hugh Dickins689bceb2005-11-21 21:32:20 -08001204again:
Nick Piggin48db57f2006-01-08 01:00:42 -08001205 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001207 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 local_irq_save(flags);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001210 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1211 list = &pcp->lists[migratetype];
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001212 if (list_empty(list)) {
Mel Gorman535131e2007-10-16 01:25:49 -07001213 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001214 pcp->batch, list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001215 migratetype, cold);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001216 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07001217 goto failed;
Mel Gorman535131e2007-10-16 01:25:49 -07001218 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001219
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001220 if (cold)
1221 page = list_entry(list->prev, struct page, lru);
1222 else
1223 page = list_entry(list->next, struct page, lru);
1224
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001225 list_del(&page->lru);
1226 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001227 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07001228 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1229 /*
1230 * __GFP_NOFAIL is not to be used in new code.
1231 *
1232 * All __GFP_NOFAIL callers should be fixed so that they
1233 * properly detect and handle allocation failures.
1234 *
1235 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07001236 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07001237 * __GFP_NOFAIL.
1238 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07001239 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07001240 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001242 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001243 spin_unlock(&zone->lock);
1244 if (!page)
1245 goto failed;
KOSAKI Motohiro6ccf80e2010-01-15 17:01:18 -08001246 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 }
1248
Christoph Lameterf8891e52006-06-30 01:55:45 -07001249 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Mel Gorman18ea7e72008-04-28 02:12:14 -07001250 zone_statistics(preferred_zone, zone);
Nick Piggina74609f2006-01-06 00:11:20 -08001251 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252
Nick Piggin725d7042006-09-25 23:30:55 -07001253 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -08001254 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001255 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001257
1258failed:
1259 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001260 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261}
1262
Mel Gorman41858962009-06-16 15:32:12 -07001263/* The ALLOC_WMARK bits are used as an index to zone->watermark */
1264#define ALLOC_WMARK_MIN WMARK_MIN
1265#define ALLOC_WMARK_LOW WMARK_LOW
1266#define ALLOC_WMARK_HIGH WMARK_HIGH
1267#define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1268
1269/* Mask to get the watermark bits */
1270#define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1271
Nick Piggin31488902005-11-28 13:44:03 -08001272#define ALLOC_HARDER 0x10 /* try to alloc harder */
1273#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1274#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001275
Akinobu Mita933e3122006-12-08 02:39:45 -08001276#ifdef CONFIG_FAIL_PAGE_ALLOC
1277
1278static struct fail_page_alloc_attr {
1279 struct fault_attr attr;
1280
1281 u32 ignore_gfp_highmem;
1282 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001283 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001284
1285#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1286
1287 struct dentry *ignore_gfp_highmem_file;
1288 struct dentry *ignore_gfp_wait_file;
Akinobu Mita54114992007-07-15 23:40:23 -07001289 struct dentry *min_order_file;
Akinobu Mita933e3122006-12-08 02:39:45 -08001290
1291#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1292
1293} fail_page_alloc = {
1294 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001295 .ignore_gfp_wait = 1,
1296 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001297 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001298};
1299
1300static int __init setup_fail_page_alloc(char *str)
1301{
1302 return setup_fault_attr(&fail_page_alloc.attr, str);
1303}
1304__setup("fail_page_alloc=", setup_fail_page_alloc);
1305
1306static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1307{
Akinobu Mita54114992007-07-15 23:40:23 -07001308 if (order < fail_page_alloc.min_order)
1309 return 0;
Akinobu Mita933e3122006-12-08 02:39:45 -08001310 if (gfp_mask & __GFP_NOFAIL)
1311 return 0;
1312 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1313 return 0;
1314 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1315 return 0;
1316
1317 return should_fail(&fail_page_alloc.attr, 1 << order);
1318}
1319
1320#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1321
1322static int __init fail_page_alloc_debugfs(void)
1323{
1324 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1325 struct dentry *dir;
1326 int err;
1327
1328 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1329 "fail_page_alloc");
1330 if (err)
1331 return err;
1332 dir = fail_page_alloc.attr.dentries.dir;
1333
1334 fail_page_alloc.ignore_gfp_wait_file =
1335 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1336 &fail_page_alloc.ignore_gfp_wait);
1337
1338 fail_page_alloc.ignore_gfp_highmem_file =
1339 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1340 &fail_page_alloc.ignore_gfp_highmem);
Akinobu Mita54114992007-07-15 23:40:23 -07001341 fail_page_alloc.min_order_file =
1342 debugfs_create_u32("min-order", mode, dir,
1343 &fail_page_alloc.min_order);
Akinobu Mita933e3122006-12-08 02:39:45 -08001344
1345 if (!fail_page_alloc.ignore_gfp_wait_file ||
Akinobu Mita54114992007-07-15 23:40:23 -07001346 !fail_page_alloc.ignore_gfp_highmem_file ||
1347 !fail_page_alloc.min_order_file) {
Akinobu Mita933e3122006-12-08 02:39:45 -08001348 err = -ENOMEM;
1349 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1350 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
Akinobu Mita54114992007-07-15 23:40:23 -07001351 debugfs_remove(fail_page_alloc.min_order_file);
Akinobu Mita933e3122006-12-08 02:39:45 -08001352 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1353 }
1354
1355 return err;
1356}
1357
1358late_initcall(fail_page_alloc_debugfs);
1359
1360#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1361
1362#else /* CONFIG_FAIL_PAGE_ALLOC */
1363
1364static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1365{
1366 return 0;
1367}
1368
1369#endif /* CONFIG_FAIL_PAGE_ALLOC */
1370
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371/*
1372 * Return 1 if free pages are above 'mark'. This takes into account the order
1373 * of the allocation.
1374 */
1375int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001376 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377{
1378 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001379 long min = mark;
1380 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 int o;
1382
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001383 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001385 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 min -= min / 4;
1387
1388 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1389 return 0;
1390 for (o = 0; o < order; o++) {
1391 /* At the next order, this order's pages become unavailable */
1392 free_pages -= z->free_area[o].nr_free << o;
1393
1394 /* Require fewer higher order pages to be free */
1395 min >>= 1;
1396
1397 if (free_pages <= min)
1398 return 0;
1399 }
1400 return 1;
1401}
1402
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001403#ifdef CONFIG_NUMA
1404/*
1405 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1406 * skip over zones that are not allowed by the cpuset, or that have
1407 * been recently (in last second) found to be nearly full. See further
1408 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001409 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001410 *
1411 * If the zonelist cache is present in the passed in zonelist, then
1412 * returns a pointer to the allowed node mask (either the current
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001413 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001414 *
1415 * If the zonelist cache is not available for this zonelist, does
1416 * nothing and returns NULL.
1417 *
1418 * If the fullzones BITMAP in the zonelist cache is stale (more than
1419 * a second since last zap'd) then we zap it out (clear its bits.)
1420 *
1421 * We hold off even calling zlc_setup, until after we've checked the
1422 * first zone in the zonelist, on the theory that most allocations will
1423 * be satisfied from that first zone, so best to examine that zone as
1424 * quickly as we can.
1425 */
1426static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1427{
1428 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1429 nodemask_t *allowednodes; /* zonelist_cache approximation */
1430
1431 zlc = zonelist->zlcache_ptr;
1432 if (!zlc)
1433 return NULL;
1434
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001435 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001436 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1437 zlc->last_full_zap = jiffies;
1438 }
1439
1440 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1441 &cpuset_current_mems_allowed :
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001442 &node_states[N_HIGH_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001443 return allowednodes;
1444}
1445
1446/*
1447 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1448 * if it is worth looking at further for free memory:
1449 * 1) Check that the zone isn't thought to be full (doesn't have its
1450 * bit set in the zonelist_cache fullzones BITMAP).
1451 * 2) Check that the zones node (obtained from the zonelist_cache
1452 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1453 * Return true (non-zero) if zone is worth looking at further, or
1454 * else return false (zero) if it is not.
1455 *
1456 * This check -ignores- the distinction between various watermarks,
1457 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1458 * found to be full for any variation of these watermarks, it will
1459 * be considered full for up to one second by all requests, unless
1460 * we are so low on memory on all allowed nodes that we are forced
1461 * into the second scan of the zonelist.
1462 *
1463 * In the second scan we ignore this zonelist cache and exactly
1464 * apply the watermarks to all zones, even it is slower to do so.
1465 * We are low on memory in the second scan, and should leave no stone
1466 * unturned looking for a free page.
1467 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001468static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001469 nodemask_t *allowednodes)
1470{
1471 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1472 int i; /* index of *z in zonelist zones */
1473 int n; /* node that zone *z is on */
1474
1475 zlc = zonelist->zlcache_ptr;
1476 if (!zlc)
1477 return 1;
1478
Mel Gormandd1a2392008-04-28 02:12:17 -07001479 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001480 n = zlc->z_to_n[i];
1481
1482 /* This zone is worth trying if it is allowed but not full */
1483 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1484}
1485
1486/*
1487 * Given 'z' scanning a zonelist, set the corresponding bit in
1488 * zlc->fullzones, so that subsequent attempts to allocate a page
1489 * from that zone don't waste time re-examining it.
1490 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001491static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001492{
1493 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1494 int i; /* index of *z in zonelist zones */
1495
1496 zlc = zonelist->zlcache_ptr;
1497 if (!zlc)
1498 return;
1499
Mel Gormandd1a2392008-04-28 02:12:17 -07001500 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001501
1502 set_bit(i, zlc->fullzones);
1503}
1504
1505#else /* CONFIG_NUMA */
1506
1507static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1508{
1509 return NULL;
1510}
1511
Mel Gormandd1a2392008-04-28 02:12:17 -07001512static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001513 nodemask_t *allowednodes)
1514{
1515 return 1;
1516}
1517
Mel Gormandd1a2392008-04-28 02:12:17 -07001518static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001519{
1520}
1521#endif /* CONFIG_NUMA */
1522
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001523/*
Paul Jackson0798e512006-12-06 20:31:38 -08001524 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001525 * a page.
1526 */
1527static struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001528get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
Mel Gorman5117f452009-06-16 15:31:59 -07001529 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
Mel Gorman3dd28262009-06-16 15:32:00 -07001530 struct zone *preferred_zone, int migratetype)
Martin Hicks753ee722005-06-21 17:14:41 -07001531{
Mel Gormandd1a2392008-04-28 02:12:17 -07001532 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001533 struct page *page = NULL;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001534 int classzone_idx;
Mel Gorman5117f452009-06-16 15:31:59 -07001535 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001536 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1537 int zlc_active = 0; /* set if using zonelist_cache */
1538 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gorman54a6eb52008-04-28 02:12:16 -07001539
Mel Gorman19770b32008-04-28 02:12:18 -07001540 classzone_idx = zone_idx(preferred_zone);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001541zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001542 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001543 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001544 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1545 */
Mel Gorman19770b32008-04-28 02:12:18 -07001546 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1547 high_zoneidx, nodemask) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001548 if (NUMA_BUILD && zlc_active &&
1549 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1550 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001551 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001552 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001553 goto try_next_zone;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001554
Mel Gorman41858962009-06-16 15:32:12 -07001555 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001556 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001557 unsigned long mark;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001558 int ret;
1559
Mel Gorman41858962009-06-16 15:32:12 -07001560 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
Mel Gormanfa5e0842009-06-16 15:33:22 -07001561 if (zone_watermark_ok(zone, order, mark,
1562 classzone_idx, alloc_flags))
1563 goto try_this_zone;
1564
1565 if (zone_reclaim_mode == 0)
1566 goto this_zone_full;
1567
1568 ret = zone_reclaim(zone, gfp_mask, order);
1569 switch (ret) {
1570 case ZONE_RECLAIM_NOSCAN:
1571 /* did not scan */
1572 goto try_next_zone;
1573 case ZONE_RECLAIM_FULL:
1574 /* scanned but unreclaimable */
1575 goto this_zone_full;
1576 default:
1577 /* did we reclaim enough */
1578 if (!zone_watermark_ok(zone, order, mark,
1579 classzone_idx, alloc_flags))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001580 goto this_zone_full;
Paul Jackson0798e512006-12-06 20:31:38 -08001581 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001582 }
1583
Mel Gormanfa5e0842009-06-16 15:33:22 -07001584try_this_zone:
Mel Gorman3dd28262009-06-16 15:32:00 -07001585 page = buffered_rmqueue(preferred_zone, zone, order,
1586 gfp_mask, migratetype);
Paul Jackson0798e512006-12-06 20:31:38 -08001587 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001588 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001589this_zone_full:
1590 if (NUMA_BUILD)
1591 zlc_mark_zone_full(zonelist, z);
1592try_next_zone:
Christoph Lameter62bc62a2009-06-16 15:32:15 -07001593 if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) {
Mel Gormand395b732009-06-16 15:32:09 -07001594 /*
1595 * we do zlc_setup after the first zone is tried but only
1596 * if there are multiple nodes make it worthwhile
1597 */
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001598 allowednodes = zlc_setup(zonelist, alloc_flags);
1599 zlc_active = 1;
1600 did_zlc_setup = 1;
1601 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07001602 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001603
1604 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1605 /* Disable zlc cache for second zonelist scan */
1606 zlc_active = 0;
1607 goto zonelist_scan;
1608 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001609 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07001610}
1611
Mel Gorman11e33f62009-06-16 15:31:57 -07001612static inline int
1613should_alloc_retry(gfp_t gfp_mask, unsigned int order,
1614 unsigned long pages_reclaimed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615{
Mel Gorman11e33f62009-06-16 15:31:57 -07001616 /* Do not loop if specifically requested */
1617 if (gfp_mask & __GFP_NORETRY)
1618 return 0;
1619
1620 /*
1621 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1622 * means __GFP_NOFAIL, but that may not be true in other
1623 * implementations.
1624 */
1625 if (order <= PAGE_ALLOC_COSTLY_ORDER)
1626 return 1;
1627
1628 /*
1629 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1630 * specified, then we retry until we no longer reclaim any pages
1631 * (above), or we've reclaimed an order of pages at least as
1632 * large as the allocation's order. In both cases, if the
1633 * allocation still fails, we stop retrying.
1634 */
1635 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
1636 return 1;
1637
1638 /*
1639 * Don't let big-order allocations loop unless the caller
1640 * explicitly requests that.
1641 */
1642 if (gfp_mask & __GFP_NOFAIL)
1643 return 1;
1644
1645 return 0;
1646}
1647
1648static inline struct page *
1649__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1650 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001651 nodemask_t *nodemask, struct zone *preferred_zone,
1652 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001653{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655
Mel Gorman11e33f62009-06-16 15:31:57 -07001656 /* Acquire the OOM killer lock for the zones in zonelist */
1657 if (!try_set_zone_oom(zonelist, gfp_mask)) {
1658 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 return NULL;
1660 }
Jens Axboe6b1de912005-11-17 21:35:02 +01001661
Mel Gorman11e33f62009-06-16 15:31:57 -07001662 /*
1663 * Go through the zonelist yet one more time, keep very high watermark
1664 * here, this is only to catch a parallel oom killing, we must fail if
1665 * we're still under heavy pressure.
1666 */
1667 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1668 order, zonelist, high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001669 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07001670 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001671 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07001672 goto out;
1673
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001674 if (!(gfp_mask & __GFP_NOFAIL)) {
1675 /* The OOM killer will not help higher order allocs */
1676 if (order > PAGE_ALLOC_COSTLY_ORDER)
1677 goto out;
1678 /*
1679 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
1680 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
1681 * The caller should handle page allocation failure by itself if
1682 * it specifies __GFP_THISNODE.
1683 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
1684 */
1685 if (gfp_mask & __GFP_THISNODE)
1686 goto out;
1687 }
Mel Gorman11e33f62009-06-16 15:31:57 -07001688 /* Exhausted what can be done so it's blamo time */
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001689 out_of_memory(zonelist, gfp_mask, order, nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07001690
1691out:
1692 clear_zonelist_oom(zonelist, gfp_mask);
1693 return page;
1694}
1695
1696/* The really slow allocator path where we enter direct reclaim */
1697static inline struct page *
1698__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
1699 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001700 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001701 int migratetype, unsigned long *did_some_progress)
Mel Gorman11e33f62009-06-16 15:31:57 -07001702{
1703 struct page *page = NULL;
1704 struct reclaim_state reclaim_state;
1705 struct task_struct *p = current;
1706
1707 cond_resched();
1708
1709 /* We now go into synchronous reclaim */
1710 cpuset_memory_pressure_bump();
Mel Gorman11e33f62009-06-16 15:31:57 -07001711 p->flags |= PF_MEMALLOC;
1712 lockdep_set_current_reclaim_state(gfp_mask);
1713 reclaim_state.reclaimed_slab = 0;
1714 p->reclaim_state = &reclaim_state;
1715
1716 *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
1717
1718 p->reclaim_state = NULL;
1719 lockdep_clear_current_reclaim_state();
1720 p->flags &= ~PF_MEMALLOC;
1721
1722 cond_resched();
1723
1724 if (order != 0)
1725 drain_all_pages();
1726
1727 if (likely(*did_some_progress))
1728 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001729 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001730 alloc_flags, preferred_zone,
1731 migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001732 return page;
1733}
1734
Mel Gorman11e33f62009-06-16 15:31:57 -07001735/*
1736 * This is called in the allocator slow-path if the allocation request is of
1737 * sufficient urgency to ignore watermarks and take other desperate measures
1738 */
1739static inline struct page *
1740__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
1741 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001742 nodemask_t *nodemask, struct zone *preferred_zone,
1743 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001744{
1745 struct page *page;
1746
1747 do {
1748 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001749 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
Mel Gorman3dd28262009-06-16 15:32:00 -07001750 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001751
1752 if (!page && gfp_mask & __GFP_NOFAIL)
Jens Axboe8aa7e842009-07-09 14:52:32 +02001753 congestion_wait(BLK_RW_ASYNC, HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07001754 } while (!page && (gfp_mask & __GFP_NOFAIL));
1755
1756 return page;
1757}
1758
1759static inline
1760void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
1761 enum zone_type high_zoneidx)
1762{
1763 struct zoneref *z;
1764 struct zone *zone;
1765
1766 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
1767 wakeup_kswapd(zone, order);
1768}
1769
Peter Zijlstra341ce062009-06-16 15:32:02 -07001770static inline int
1771gfp_to_alloc_flags(gfp_t gfp_mask)
1772{
1773 struct task_struct *p = current;
1774 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
1775 const gfp_t wait = gfp_mask & __GFP_WAIT;
1776
Mel Gormana56f57f2009-06-16 15:32:02 -07001777 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
1778 BUILD_BUG_ON(__GFP_HIGH != ALLOC_HIGH);
1779
Peter Zijlstra341ce062009-06-16 15:32:02 -07001780 /*
1781 * The caller may dip into page reserves a bit more if the caller
1782 * cannot run direct reclaim, or if the caller has realtime scheduling
1783 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1784 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1785 */
Mel Gormana56f57f2009-06-16 15:32:02 -07001786 alloc_flags |= (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07001787
1788 if (!wait) {
1789 alloc_flags |= ALLOC_HARDER;
1790 /*
1791 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1792 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1793 */
1794 alloc_flags &= ~ALLOC_CPUSET;
Mel Gorman9d0ed602009-11-11 14:26:17 -08001795 } else if (unlikely(rt_task(p)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07001796 alloc_flags |= ALLOC_HARDER;
1797
1798 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
1799 if (!in_interrupt() &&
1800 ((p->flags & PF_MEMALLOC) ||
1801 unlikely(test_thread_flag(TIF_MEMDIE))))
1802 alloc_flags |= ALLOC_NO_WATERMARKS;
1803 }
1804
1805 return alloc_flags;
1806}
1807
Mel Gorman11e33f62009-06-16 15:31:57 -07001808static inline struct page *
1809__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
1810 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001811 nodemask_t *nodemask, struct zone *preferred_zone,
1812 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001813{
1814 const gfp_t wait = gfp_mask & __GFP_WAIT;
1815 struct page *page = NULL;
1816 int alloc_flags;
1817 unsigned long pages_reclaimed = 0;
1818 unsigned long did_some_progress;
1819 struct task_struct *p = current;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001820
Christoph Lameter952f3b52006-12-06 20:33:26 -08001821 /*
Mel Gorman72807a72009-06-16 15:32:18 -07001822 * In the slowpath, we sanity check order to avoid ever trying to
1823 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
1824 * be using allocators in order of preference for an area that is
1825 * too large.
1826 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07001827 if (order >= MAX_ORDER) {
1828 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07001829 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07001830 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001831
Christoph Lameter952f3b52006-12-06 20:33:26 -08001832 /*
1833 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1834 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1835 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1836 * using a larger set of nodes after it has established that the
1837 * allowed per node queues are empty and that nodes are
1838 * over allocated.
1839 */
1840 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1841 goto nopage;
1842
Mel Gormancc4a6852009-11-11 14:26:14 -08001843restart:
Mel Gorman11e33f62009-06-16 15:31:57 -07001844 wake_all_kswapd(order, zonelist, high_zoneidx);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001845
Paul Jackson9bf22292005-09-06 15:18:12 -07001846 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001847 * OK, we're below the kswapd watermark and have kicked background
1848 * reclaim. Now things get more complex, so set up alloc_flags according
1849 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07001850 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07001851 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852
Peter Zijlstra341ce062009-06-16 15:32:02 -07001853 /* This is the last chance, in general, before the goto nopage. */
Mel Gorman19770b32008-04-28 02:12:18 -07001854 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
Peter Zijlstra341ce062009-06-16 15:32:02 -07001855 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
1856 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001857 if (page)
1858 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859
Kirill Korotaevb43a57b2006-12-06 20:32:27 -08001860rebalance:
Mel Gorman11e33f62009-06-16 15:31:57 -07001861 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07001862 if (alloc_flags & ALLOC_NO_WATERMARKS) {
1863 page = __alloc_pages_high_priority(gfp_mask, order,
1864 zonelist, high_zoneidx, nodemask,
1865 preferred_zone, migratetype);
1866 if (page)
1867 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 }
1869
1870 /* Atomic allocations - we can't balance anything */
1871 if (!wait)
1872 goto nopage;
1873
Peter Zijlstra341ce062009-06-16 15:32:02 -07001874 /* Avoid recursion of direct reclaim */
1875 if (p->flags & PF_MEMALLOC)
1876 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877
David Rientjes6583bb62009-07-29 15:02:06 -07001878 /* Avoid allocations with no watermarks from looping endlessly */
1879 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
1880 goto nopage;
1881
Mel Gorman11e33f62009-06-16 15:31:57 -07001882 /* Try direct reclaim and then allocating */
1883 page = __alloc_pages_direct_reclaim(gfp_mask, order,
1884 zonelist, high_zoneidx,
1885 nodemask,
Mel Gorman5117f452009-06-16 15:31:59 -07001886 alloc_flags, preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001887 migratetype, &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07001888 if (page)
1889 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891 /*
Mel Gorman11e33f62009-06-16 15:31:57 -07001892 * If we failed to make any progress reclaiming, then we are
1893 * running out of options and have to consider going OOM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 */
Mel Gorman11e33f62009-06-16 15:31:57 -07001895 if (!did_some_progress) {
1896 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -07001897 if (oom_killer_disabled)
1898 goto nopage;
Mel Gorman11e33f62009-06-16 15:31:57 -07001899 page = __alloc_pages_may_oom(gfp_mask, order,
1900 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001901 nodemask, preferred_zone,
1902 migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001903 if (page)
1904 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905
Mel Gorman11e33f62009-06-16 15:31:57 -07001906 /*
David Rientjes82553a92009-06-16 15:32:58 -07001907 * The OOM killer does not trigger for high-order
1908 * ~__GFP_NOFAIL allocations so if no progress is being
1909 * made, there are no other options and retrying is
1910 * unlikely to help.
Mel Gorman11e33f62009-06-16 15:31:57 -07001911 */
David Rientjes82553a92009-06-16 15:32:58 -07001912 if (order > PAGE_ALLOC_COSTLY_ORDER &&
1913 !(gfp_mask & __GFP_NOFAIL))
Mel Gorman11e33f62009-06-16 15:31:57 -07001914 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916 goto restart;
1917 }
1918 }
1919
Mel Gorman11e33f62009-06-16 15:31:57 -07001920 /* Check if we should retry the allocation */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07001921 pages_reclaimed += did_some_progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07001922 if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
1923 /* Wait for some write requests to complete then retry */
Jens Axboe8aa7e842009-07-09 14:52:32 +02001924 congestion_wait(BLK_RW_ASYNC, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 goto rebalance;
1926 }
1927
1928nopage:
1929 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1930 printk(KERN_WARNING "%s: page allocation failure."
1931 " order:%d, mode:0x%x\n",
1932 p->comm, order, gfp_mask);
1933 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001934 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 }
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001936 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937got_pg:
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001938 if (kmemcheck_enabled)
1939 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940 return page;
Mel Gorman11e33f62009-06-16 15:31:57 -07001941
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942}
Mel Gorman11e33f62009-06-16 15:31:57 -07001943
1944/*
1945 * This is the 'heart' of the zoned buddy allocator.
1946 */
1947struct page *
1948__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
1949 struct zonelist *zonelist, nodemask_t *nodemask)
1950{
1951 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Mel Gorman5117f452009-06-16 15:31:59 -07001952 struct zone *preferred_zone;
Mel Gorman11e33f62009-06-16 15:31:57 -07001953 struct page *page;
Mel Gorman3dd28262009-06-16 15:32:00 -07001954 int migratetype = allocflags_to_migratetype(gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07001955
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10001956 gfp_mask &= gfp_allowed_mask;
1957
Mel Gorman11e33f62009-06-16 15:31:57 -07001958 lockdep_trace_alloc(gfp_mask);
1959
1960 might_sleep_if(gfp_mask & __GFP_WAIT);
1961
1962 if (should_fail_alloc_page(gfp_mask, order))
1963 return NULL;
1964
1965 /*
1966 * Check the zones suitable for the gfp_mask contain at least one
1967 * valid zone. It's possible to have an empty zonelist as a result
1968 * of GFP_THISNODE and a memoryless node
1969 */
1970 if (unlikely(!zonelist->_zonerefs->zone))
1971 return NULL;
1972
Mel Gorman5117f452009-06-16 15:31:59 -07001973 /* The preferred zone is used for statistics later */
1974 first_zones_zonelist(zonelist, high_zoneidx, nodemask, &preferred_zone);
1975 if (!preferred_zone)
1976 return NULL;
1977
1978 /* First allocation attempt */
Mel Gorman11e33f62009-06-16 15:31:57 -07001979 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001980 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07001981 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001982 if (unlikely(!page))
1983 page = __alloc_pages_slowpath(gfp_mask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001984 zonelist, high_zoneidx, nodemask,
Mel Gorman3dd28262009-06-16 15:32:00 -07001985 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001986
Mel Gorman4b4f2782009-09-21 17:02:41 -07001987 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001988 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989}
Mel Gormand2391712009-06-16 15:31:52 -07001990EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991
1992/*
1993 * Common helper functions.
1994 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001995unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996{
Akinobu Mita945a1112009-09-21 17:01:47 -07001997 struct page *page;
1998
1999 /*
2000 * __get_free_pages() returns a 32-bit address, which cannot represent
2001 * a highmem page
2002 */
2003 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2004
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005 page = alloc_pages(gfp_mask, order);
2006 if (!page)
2007 return 0;
2008 return (unsigned long) page_address(page);
2009}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010EXPORT_SYMBOL(__get_free_pages);
2011
Harvey Harrison920c7a52008-02-04 22:29:26 -08002012unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013{
Akinobu Mita945a1112009-09-21 17:01:47 -07002014 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016EXPORT_SYMBOL(get_zeroed_page);
2017
2018void __pagevec_free(struct pagevec *pvec)
2019{
2020 int i = pagevec_count(pvec);
2021
Mel Gorman4b4f2782009-09-21 17:02:41 -07002022 while (--i >= 0) {
2023 trace_mm_pagevec_free(pvec->pages[i], pvec->cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024 free_hot_cold_page(pvec->pages[i], pvec->cold);
Mel Gorman4b4f2782009-09-21 17:02:41 -07002025 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026}
2027
Harvey Harrison920c7a52008-02-04 22:29:26 -08002028void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029{
Nick Pigginb5810032005-10-29 18:16:12 -07002030 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 if (order == 0)
Li Hongfc916682010-03-05 13:41:54 -08002032 free_hot_cold_page(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 else
2034 __free_pages_ok(page, order);
2035 }
2036}
2037
2038EXPORT_SYMBOL(__free_pages);
2039
Harvey Harrison920c7a52008-02-04 22:29:26 -08002040void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041{
2042 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07002043 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 __free_pages(virt_to_page((void *)addr), order);
2045 }
2046}
2047
2048EXPORT_SYMBOL(free_pages);
2049
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002050/**
2051 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2052 * @size: the number of bytes to allocate
2053 * @gfp_mask: GFP flags for the allocation
2054 *
2055 * This function is similar to alloc_pages(), except that it allocates the
2056 * minimum number of pages to satisfy the request. alloc_pages() can only
2057 * allocate memory in power-of-two pages.
2058 *
2059 * This function is also limited by MAX_ORDER.
2060 *
2061 * Memory allocated by this function must be released by free_pages_exact().
2062 */
2063void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2064{
2065 unsigned int order = get_order(size);
2066 unsigned long addr;
2067
2068 addr = __get_free_pages(gfp_mask, order);
2069 if (addr) {
2070 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2071 unsigned long used = addr + PAGE_ALIGN(size);
2072
Kevin Cernekee5bfd7562009-07-05 12:08:19 -07002073 split_page(virt_to_page((void *)addr), order);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002074 while (used < alloc_end) {
2075 free_page(used);
2076 used += PAGE_SIZE;
2077 }
2078 }
2079
2080 return (void *)addr;
2081}
2082EXPORT_SYMBOL(alloc_pages_exact);
2083
2084/**
2085 * free_pages_exact - release memory allocated via alloc_pages_exact()
2086 * @virt: the value returned by alloc_pages_exact.
2087 * @size: size of allocation, same value as passed to alloc_pages_exact().
2088 *
2089 * Release the memory allocated by a previous call to alloc_pages_exact.
2090 */
2091void free_pages_exact(void *virt, size_t size)
2092{
2093 unsigned long addr = (unsigned long)virt;
2094 unsigned long end = addr + PAGE_ALIGN(size);
2095
2096 while (addr < end) {
2097 free_page(addr);
2098 addr += PAGE_SIZE;
2099 }
2100}
2101EXPORT_SYMBOL(free_pages_exact);
2102
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103static unsigned int nr_free_zone_pages(int offset)
2104{
Mel Gormandd1a2392008-04-28 02:12:17 -07002105 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002106 struct zone *zone;
2107
Martin J. Blighe310fd42005-07-29 22:59:18 -07002108 /* Just pick one node, since fallback list is circular */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 unsigned int sum = 0;
2110
Mel Gorman0e884602008-04-28 02:12:14 -07002111 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112
Mel Gorman54a6eb52008-04-28 02:12:16 -07002113 for_each_zone_zonelist(zone, z, zonelist, offset) {
Martin J. Blighe310fd42005-07-29 22:59:18 -07002114 unsigned long size = zone->present_pages;
Mel Gorman41858962009-06-16 15:32:12 -07002115 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07002116 if (size > high)
2117 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118 }
2119
2120 return sum;
2121}
2122
2123/*
2124 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2125 */
2126unsigned int nr_free_buffer_pages(void)
2127{
Al Viroaf4ca452005-10-21 02:55:38 -04002128 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129}
Meelap Shahc2f1a552007-07-17 04:04:39 -07002130EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131
2132/*
2133 * Amount of free RAM allocatable within all zones
2134 */
2135unsigned int nr_free_pagecache_pages(void)
2136{
Mel Gorman2a1e2742007-07-17 04:03:12 -07002137 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002139
2140static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141{
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002142 if (NUMA_BUILD)
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08002143 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146void si_meminfo(struct sysinfo *val)
2147{
2148 val->totalram = totalram_pages;
2149 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002150 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 val->totalhigh = totalhigh_pages;
2153 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 val->mem_unit = PAGE_SIZE;
2155}
2156
2157EXPORT_SYMBOL(si_meminfo);
2158
2159#ifdef CONFIG_NUMA
2160void si_meminfo_node(struct sysinfo *val, int nid)
2161{
2162 pg_data_t *pgdat = NODE_DATA(nid);
2163
2164 val->totalram = pgdat->node_present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002165 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002166#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002168 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2169 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002170#else
2171 val->totalhigh = 0;
2172 val->freehigh = 0;
2173#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 val->mem_unit = PAGE_SIZE;
2175}
2176#endif
2177
2178#define K(x) ((x) << (PAGE_SHIFT-10))
2179
2180/*
2181 * Show free area list (used inside shift_scroll-lock stuff)
2182 * We also calculate the percentage fragmentation. We do this by counting the
2183 * memory on each free list with the exception of the first item on the list.
2184 */
2185void show_free_areas(void)
2186{
Jes Sorensenc7241912006-09-27 01:50:05 -07002187 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 struct zone *zone;
2189
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002190 for_each_populated_zone(zone) {
Jes Sorensenc7241912006-09-27 01:50:05 -07002191 show_node(zone);
2192 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193
Dave Jones6b482c62005-11-10 15:45:56 -05002194 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 struct per_cpu_pageset *pageset;
2196
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002197 pageset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002199 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2200 cpu, pageset->pcp.high,
2201 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202 }
2203 }
2204
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002205 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2206 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002207 " unevictable:%lu"
Andrew Mortonb76146e2009-10-26 16:49:52 -07002208 " dirty:%lu writeback:%lu unstable:%lu\n"
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07002209 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002210 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002211 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002212 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002213 global_page_state(NR_ISOLATED_ANON),
2214 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002215 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002216 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002217 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07002218 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07002219 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07002220 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08002221 global_page_state(NR_FREE_PAGES),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07002222 global_page_state(NR_SLAB_RECLAIMABLE),
2223 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07002224 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002225 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08002226 global_page_state(NR_PAGETABLE),
2227 global_page_state(NR_BOUNCE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002229 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230 int i;
2231
2232 show_node(zone);
2233 printk("%s"
2234 " free:%lukB"
2235 " min:%lukB"
2236 " low:%lukB"
2237 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002238 " active_anon:%lukB"
2239 " inactive_anon:%lukB"
2240 " active_file:%lukB"
2241 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002242 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002243 " isolated(anon):%lukB"
2244 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 " present:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002246 " mlocked:%lukB"
2247 " dirty:%lukB"
2248 " writeback:%lukB"
2249 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002250 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002251 " slab_reclaimable:%lukB"
2252 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07002253 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002254 " pagetables:%lukB"
2255 " unstable:%lukB"
2256 " bounce:%lukB"
2257 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 " pages_scanned:%lu"
2259 " all_unreclaimable? %s"
2260 "\n",
2261 zone->name,
Christoph Lameterd23ad422007-02-10 01:43:02 -08002262 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07002263 K(min_wmark_pages(zone)),
2264 K(low_wmark_pages(zone)),
2265 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002266 K(zone_page_state(zone, NR_ACTIVE_ANON)),
2267 K(zone_page_state(zone, NR_INACTIVE_ANON)),
2268 K(zone_page_state(zone, NR_ACTIVE_FILE)),
2269 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002270 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002271 K(zone_page_state(zone, NR_ISOLATED_ANON)),
2272 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273 K(zone->present_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002274 K(zone_page_state(zone, NR_MLOCK)),
2275 K(zone_page_state(zone, NR_FILE_DIRTY)),
2276 K(zone_page_state(zone, NR_WRITEBACK)),
2277 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002278 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002279 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
2280 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07002281 zone_page_state(zone, NR_KERNEL_STACK) *
2282 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002283 K(zone_page_state(zone, NR_PAGETABLE)),
2284 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
2285 K(zone_page_state(zone, NR_BOUNCE)),
2286 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 zone->pages_scanned,
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -08002288 (zone->all_unreclaimable ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289 );
2290 printk("lowmem_reserve[]:");
2291 for (i = 0; i < MAX_NR_ZONES; i++)
2292 printk(" %lu", zone->lowmem_reserve[i]);
2293 printk("\n");
2294 }
2295
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002296 for_each_populated_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002297 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298
2299 show_node(zone);
2300 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301
2302 spin_lock_irqsave(&zone->lock, flags);
2303 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002304 nr[order] = zone->free_area[order].nr_free;
2305 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306 }
2307 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002308 for (order = 0; order < MAX_ORDER; order++)
2309 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310 printk("= %lukB\n", K(total));
2311 }
2312
Larry Woodmane6f36022008-02-04 22:29:30 -08002313 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2314
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 show_swap_cache_info();
2316}
2317
Mel Gorman19770b32008-04-28 02:12:18 -07002318static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2319{
2320 zoneref->zone = zone;
2321 zoneref->zone_idx = zone_idx(zone);
2322}
2323
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324/*
2325 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08002326 *
2327 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002329static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2330 int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331{
Christoph Lameter1a932052006-01-06 00:11:16 -08002332 struct zone *zone;
2333
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002334 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002335 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08002336
2337 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002338 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08002339 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08002340 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002341 zoneref_set_zone(zone,
2342 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08002343 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08002345
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002346 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08002347 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348}
2349
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002350
2351/*
2352 * zonelist_order:
2353 * 0 = automatic detection of better ordering.
2354 * 1 = order by ([node] distance, -zonetype)
2355 * 2 = order by (-zonetype, [node] distance)
2356 *
2357 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2358 * the same zonelist. So only NUMA can configure this param.
2359 */
2360#define ZONELIST_ORDER_DEFAULT 0
2361#define ZONELIST_ORDER_NODE 1
2362#define ZONELIST_ORDER_ZONE 2
2363
2364/* zonelist order in the kernel.
2365 * set_zonelist_order() will set this to NODE or ZONE.
2366 */
2367static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2368static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2369
2370
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002372/* The value user specified ....changed by config */
2373static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2374/* string for sysctl */
2375#define NUMA_ZONELIST_ORDER_LEN 16
2376char numa_zonelist_order[16] = "default";
2377
2378/*
2379 * interface for configure zonelist ordering.
2380 * command line option "numa_zonelist_order"
2381 * = "[dD]efault - default, automatic configuration.
2382 * = "[nN]ode - order by node locality, then by zone within node
2383 * = "[zZ]one - order by zone, then by locality within zone
2384 */
2385
2386static int __parse_numa_zonelist_order(char *s)
2387{
2388 if (*s == 'd' || *s == 'D') {
2389 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2390 } else if (*s == 'n' || *s == 'N') {
2391 user_zonelist_order = ZONELIST_ORDER_NODE;
2392 } else if (*s == 'z' || *s == 'Z') {
2393 user_zonelist_order = ZONELIST_ORDER_ZONE;
2394 } else {
2395 printk(KERN_WARNING
2396 "Ignoring invalid numa_zonelist_order value: "
2397 "%s\n", s);
2398 return -EINVAL;
2399 }
2400 return 0;
2401}
2402
2403static __init int setup_numa_zonelist_order(char *s)
2404{
2405 if (s)
2406 return __parse_numa_zonelist_order(s);
2407 return 0;
2408}
2409early_param("numa_zonelist_order", setup_numa_zonelist_order);
2410
2411/*
2412 * sysctl handler for numa_zonelist_order
2413 */
2414int numa_zonelist_order_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07002415 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002416 loff_t *ppos)
2417{
2418 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2419 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01002420 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002421
Andi Kleen443c6f12009-12-23 21:00:47 +01002422 mutex_lock(&zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002423 if (write)
Andi Kleen443c6f12009-12-23 21:00:47 +01002424 strcpy(saved_string, (char*)table->data);
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07002425 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002426 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01002427 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002428 if (write) {
2429 int oldval = user_zonelist_order;
2430 if (__parse_numa_zonelist_order((char*)table->data)) {
2431 /*
2432 * bogus value. restore saved string
2433 */
2434 strncpy((char*)table->data, saved_string,
2435 NUMA_ZONELIST_ORDER_LEN);
2436 user_zonelist_order = oldval;
2437 } else if (oldval != user_zonelist_order)
2438 build_all_zonelists();
2439 }
Andi Kleen443c6f12009-12-23 21:00:47 +01002440out:
2441 mutex_unlock(&zl_order_mutex);
2442 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002443}
2444
2445
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002446#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002447static int node_load[MAX_NUMNODES];
2448
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07002450 * 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 -07002451 * @node: node whose fallback list we're appending
2452 * @used_node_mask: nodemask_t of already used nodes
2453 *
2454 * We use a number of factors to determine which is the next node that should
2455 * appear on a given node's fallback list. The node should not have appeared
2456 * already in @node's fallback list, and it should be the next closest node
2457 * according to the distance array (which contains arbitrary distance values
2458 * from each node to each node in the system), and should also prefer nodes
2459 * with no CPUs, since presumably they'll have very little allocation pressure
2460 * on them otherwise.
2461 * It returns -1 if no node is found.
2462 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002463static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002465 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466 int min_val = INT_MAX;
2467 int best_node = -1;
Rusty Russella70f7302009-03-13 14:49:46 +10302468 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002470 /* Use the local node if we haven't already */
2471 if (!node_isset(node, *used_node_mask)) {
2472 node_set(node, *used_node_mask);
2473 return node;
2474 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002476 for_each_node_state(n, N_HIGH_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477
2478 /* Don't want a node to appear more than once */
2479 if (node_isset(n, *used_node_mask))
2480 continue;
2481
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482 /* Use the distance array to find the distance */
2483 val = node_distance(node, n);
2484
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002485 /* Penalize nodes under us ("prefer the next node") */
2486 val += (n < node);
2487
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10302489 tmp = cpumask_of_node(n);
2490 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491 val += PENALTY_FOR_NODE_WITH_CPUS;
2492
2493 /* Slight preference for less loaded node */
2494 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2495 val += node_load[n];
2496
2497 if (val < min_val) {
2498 min_val = val;
2499 best_node = n;
2500 }
2501 }
2502
2503 if (best_node >= 0)
2504 node_set(best_node, *used_node_mask);
2505
2506 return best_node;
2507}
2508
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002509
2510/*
2511 * Build zonelists ordered by node and zones within node.
2512 * This results in maximum locality--normal zone overflows into local
2513 * DMA zone, if any--but risks exhausting DMA zone.
2514 */
2515static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002517 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002519
Mel Gorman54a6eb52008-04-28 02:12:16 -07002520 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07002521 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07002522 ;
2523 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2524 MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002525 zonelist->_zonerefs[j].zone = NULL;
2526 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002527}
2528
2529/*
Christoph Lameter523b9452007-10-16 01:25:37 -07002530 * Build gfp_thisnode zonelists
2531 */
2532static void build_thisnode_zonelists(pg_data_t *pgdat)
2533{
Christoph Lameter523b9452007-10-16 01:25:37 -07002534 int j;
2535 struct zonelist *zonelist;
2536
Mel Gorman54a6eb52008-04-28 02:12:16 -07002537 zonelist = &pgdat->node_zonelists[1];
2538 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002539 zonelist->_zonerefs[j].zone = NULL;
2540 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07002541}
2542
2543/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002544 * Build zonelists ordered by zone and nodes within zones.
2545 * This results in conserving DMA zone[s] until all Normal memory is
2546 * exhausted, but results in overflowing to remote node while memory
2547 * may still exist in local DMA zone.
2548 */
2549static int node_order[MAX_NUMNODES];
2550
2551static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2552{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002553 int pos, j, node;
2554 int zone_type; /* needs to be signed */
2555 struct zone *z;
2556 struct zonelist *zonelist;
2557
Mel Gorman54a6eb52008-04-28 02:12:16 -07002558 zonelist = &pgdat->node_zonelists[0];
2559 pos = 0;
2560 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2561 for (j = 0; j < nr_nodes; j++) {
2562 node = node_order[j];
2563 z = &NODE_DATA(node)->node_zones[zone_type];
2564 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002565 zoneref_set_zone(z,
2566 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002567 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002568 }
2569 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002570 }
Mel Gormandd1a2392008-04-28 02:12:17 -07002571 zonelist->_zonerefs[pos].zone = NULL;
2572 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002573}
2574
2575static int default_zonelist_order(void)
2576{
2577 int nid, zone_type;
2578 unsigned long low_kmem_size,total_size;
2579 struct zone *z;
2580 int average_size;
2581 /*
2582 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2583 * If they are really small and used heavily, the system can fall
2584 * into OOM very easily.
2585 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2586 */
2587 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2588 low_kmem_size = 0;
2589 total_size = 0;
2590 for_each_online_node(nid) {
2591 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2592 z = &NODE_DATA(nid)->node_zones[zone_type];
2593 if (populated_zone(z)) {
2594 if (zone_type < ZONE_NORMAL)
2595 low_kmem_size += z->present_pages;
2596 total_size += z->present_pages;
2597 }
2598 }
2599 }
2600 if (!low_kmem_size || /* there are no DMA area. */
2601 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2602 return ZONELIST_ORDER_NODE;
2603 /*
2604 * look into each node's config.
2605 * If there is a node whose DMA/DMA32 memory is very big area on
2606 * local memory, NODE_ORDER may be suitable.
2607 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002608 average_size = total_size /
2609 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002610 for_each_online_node(nid) {
2611 low_kmem_size = 0;
2612 total_size = 0;
2613 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2614 z = &NODE_DATA(nid)->node_zones[zone_type];
2615 if (populated_zone(z)) {
2616 if (zone_type < ZONE_NORMAL)
2617 low_kmem_size += z->present_pages;
2618 total_size += z->present_pages;
2619 }
2620 }
2621 if (low_kmem_size &&
2622 total_size > average_size && /* ignore small node */
2623 low_kmem_size > total_size * 70/100)
2624 return ZONELIST_ORDER_NODE;
2625 }
2626 return ZONELIST_ORDER_ZONE;
2627}
2628
2629static void set_zonelist_order(void)
2630{
2631 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2632 current_zonelist_order = default_zonelist_order();
2633 else
2634 current_zonelist_order = user_zonelist_order;
2635}
2636
2637static void build_zonelists(pg_data_t *pgdat)
2638{
2639 int j, node, load;
2640 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002642 int local_node, prev_node;
2643 struct zonelist *zonelist;
2644 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645
2646 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07002647 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07002649 zonelist->_zonerefs[0].zone = NULL;
2650 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651 }
2652
2653 /* NUMA-aware ordering of nodes */
2654 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002655 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656 prev_node = local_node;
2657 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002658
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002659 memset(node_order, 0, sizeof(node_order));
2660 j = 0;
2661
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08002663 int distance = node_distance(local_node, node);
2664
2665 /*
2666 * If another node is sufficiently far away then it is better
2667 * to reclaim pages in a zone before going off node.
2668 */
2669 if (distance > RECLAIM_DISTANCE)
2670 zone_reclaim_mode = 1;
2671
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672 /*
2673 * We don't want to pressure a particular node.
2674 * So adding penalty to the first node in same
2675 * distance group to make it round-robin.
2676 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08002677 if (distance != node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002678 node_load[node] = load;
2679
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680 prev_node = node;
2681 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002682 if (order == ZONELIST_ORDER_NODE)
2683 build_zonelists_in_node_order(pgdat, node);
2684 else
2685 node_order[j++] = node; /* remember order */
2686 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002688 if (order == ZONELIST_ORDER_ZONE) {
2689 /* calculate node order -- i.e., DMA last! */
2690 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691 }
Christoph Lameter523b9452007-10-16 01:25:37 -07002692
2693 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694}
2695
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002696/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002697static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002698{
Mel Gorman54a6eb52008-04-28 02:12:16 -07002699 struct zonelist *zonelist;
2700 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07002701 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002702
Mel Gorman54a6eb52008-04-28 02:12:16 -07002703 zonelist = &pgdat->node_zonelists[0];
2704 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2705 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07002706 for (z = zonelist->_zonerefs; z->zone; z++)
2707 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002708}
2709
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002710
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711#else /* CONFIG_NUMA */
2712
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002713static void set_zonelist_order(void)
2714{
2715 current_zonelist_order = ZONELIST_ORDER_ZONE;
2716}
2717
2718static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719{
Christoph Lameter19655d32006-09-25 23:31:19 -07002720 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002721 enum zone_type j;
2722 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723
2724 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725
Mel Gorman54a6eb52008-04-28 02:12:16 -07002726 zonelist = &pgdat->node_zonelists[0];
2727 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728
Mel Gorman54a6eb52008-04-28 02:12:16 -07002729 /*
2730 * Now we build the zonelist so that it contains the zones
2731 * of all the other nodes.
2732 * We don't want to pressure a particular node, so when
2733 * building the zones for node N, we make sure that the
2734 * zones coming right after the local ones are those from
2735 * node N+1 (modulo N)
2736 */
2737 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2738 if (!node_online(node))
2739 continue;
2740 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2741 MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07002743 for (node = 0; node < local_node; node++) {
2744 if (!node_online(node))
2745 continue;
2746 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2747 MAX_NR_ZONES - 1);
2748 }
2749
Mel Gormandd1a2392008-04-28 02:12:17 -07002750 zonelist->_zonerefs[j].zone = NULL;
2751 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752}
2753
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002754/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002755static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002756{
Mel Gorman54a6eb52008-04-28 02:12:16 -07002757 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002758}
2759
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760#endif /* CONFIG_NUMA */
2761
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002762/*
2763 * Boot pageset table. One per cpu which is going to be used for all
2764 * zones and all nodes. The parameters will be set in such a way
2765 * that an item put on a list will immediately be handed over to
2766 * the buddy list. This is safe since pageset manipulation is done
2767 * with interrupts disabled.
2768 *
2769 * The boot_pagesets must be kept even after bootup is complete for
2770 * unused processors and/or zones. They do play a role for bootstrapping
2771 * hotplugged processors.
2772 *
2773 * zoneinfo_show() and maybe other functions do
2774 * not check if the processor is online before following the pageset pointer.
2775 * Other parts of the kernel may not check if the zone is available.
2776 */
2777static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
2778static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
2779
Rusty Russell9b1a4d32008-07-28 12:16:30 -05002780/* return values int ....just for stop_machine() */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002781static int __build_all_zonelists(void *dummy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782{
Yasunori Goto68113782006-06-23 02:03:11 -07002783 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002784 int cpu;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002785
Bo Liu7f9cfb32009-08-18 14:11:19 -07002786#ifdef CONFIG_NUMA
2787 memset(node_load, 0, sizeof(node_load));
2788#endif
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002789 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07002790 pg_data_t *pgdat = NODE_DATA(nid);
2791
2792 build_zonelists(pgdat);
2793 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002794 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002795
2796 /*
2797 * Initialize the boot_pagesets that are going to be used
2798 * for bootstrapping processors. The real pagesets for
2799 * each zone will be allocated later when the per cpu
2800 * allocator is available.
2801 *
2802 * boot_pagesets are used also for bootstrapping offline
2803 * cpus if the system is already booted because the pagesets
2804 * are needed to initialize allocators on a specific cpu too.
2805 * F.e. the percpu allocator needs the page allocator which
2806 * needs the percpu allocator in order to allocate its pagesets
2807 * (a chicken-egg dilemma).
2808 */
2809 for_each_possible_cpu(cpu)
2810 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
2811
Yasunori Goto68113782006-06-23 02:03:11 -07002812 return 0;
2813}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002815void build_all_zonelists(void)
Yasunori Goto68113782006-06-23 02:03:11 -07002816{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002817 set_zonelist_order();
2818
Yasunori Goto68113782006-06-23 02:03:11 -07002819 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07002820 __build_all_zonelists(NULL);
Mel Gorman68ad8df2008-07-23 21:26:52 -07002821 mminit_verify_zonelist();
Yasunori Goto68113782006-06-23 02:03:11 -07002822 cpuset_init_current_mems_allowed();
2823 } else {
Simon Arlott183ff222007-10-20 01:27:18 +02002824 /* we have to stop all cpus to guarantee there is no user
Yasunori Goto68113782006-06-23 02:03:11 -07002825 of zonelist */
Rusty Russell9b1a4d32008-07-28 12:16:30 -05002826 stop_machine(__build_all_zonelists, NULL, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07002827 /* cpuset refresh routine should be here */
2828 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07002829 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002830 /*
2831 * Disable grouping by mobility if the number of pages in the
2832 * system is too low to allow the mechanism to work. It would be
2833 * more accurate, but expensive to check per-zone. This check is
2834 * made on memory-hotadd so a system can start with mobility
2835 * disabled and enable it later
2836 */
Mel Gormand9c23402007-10-16 01:26:01 -07002837 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002838 page_group_by_mobility_disabled = 1;
2839 else
2840 page_group_by_mobility_disabled = 0;
2841
2842 printk("Built %i zonelists in %s order, mobility grouping %s. "
2843 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002844 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002845 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002846 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002847 vm_total_pages);
2848#ifdef CONFIG_NUMA
2849 printk("Policy zone: %s\n", zone_names[policy_zone]);
2850#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851}
2852
2853/*
2854 * Helper functions to size the waitqueue hash table.
2855 * Essentially these want to choose hash table sizes sufficiently
2856 * large so that collisions trying to wait on pages are rare.
2857 * But in fact, the number of active page waitqueues on typical
2858 * systems is ridiculously low, less than 200. So this is even
2859 * conservative, even though it seems large.
2860 *
2861 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2862 * waitqueues, i.e. the size of the waitq table given the number of pages.
2863 */
2864#define PAGES_PER_WAITQUEUE 256
2865
Yasunori Gotocca448f2006-06-23 02:03:10 -07002866#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07002867static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868{
2869 unsigned long size = 1;
2870
2871 pages /= PAGES_PER_WAITQUEUE;
2872
2873 while (size < pages)
2874 size <<= 1;
2875
2876 /*
2877 * Once we have dozens or even hundreds of threads sleeping
2878 * on IO we've got bigger problems than wait queue collision.
2879 * Limit the size of the wait table to a reasonable size.
2880 */
2881 size = min(size, 4096UL);
2882
2883 return max(size, 4UL);
2884}
Yasunori Gotocca448f2006-06-23 02:03:10 -07002885#else
2886/*
2887 * A zone's size might be changed by hot-add, so it is not possible to determine
2888 * a suitable size for its wait_table. So we use the maximum size now.
2889 *
2890 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2891 *
2892 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2893 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2894 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2895 *
2896 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2897 * or more by the traditional way. (See above). It equals:
2898 *
2899 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2900 * ia64(16K page size) : = ( 8G + 4M)byte.
2901 * powerpc (64K page size) : = (32G +16M)byte.
2902 */
2903static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2904{
2905 return 4096UL;
2906}
2907#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908
2909/*
2910 * This is an integer logarithm so that shifts can be used later
2911 * to extract the more random high bits from the multiplicative
2912 * hash function before the remainder is taken.
2913 */
2914static inline unsigned long wait_table_bits(unsigned long size)
2915{
2916 return ffz(~size);
2917}
2918
2919#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2920
Mel Gorman56fd56b2007-10-16 01:25:58 -07002921/*
Mel Gormand9c23402007-10-16 01:26:01 -07002922 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07002923 * of blocks reserved is based on min_wmark_pages(zone). The memory within
2924 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07002925 * higher will lead to a bigger reserve which will get freed as contiguous
2926 * blocks as reclaim kicks in
2927 */
2928static void setup_zone_migrate_reserve(struct zone *zone)
2929{
2930 unsigned long start_pfn, pfn, end_pfn;
2931 struct page *page;
Mel Gorman78986a62009-09-21 17:03:02 -07002932 unsigned long block_migratetype;
2933 int reserve;
Mel Gorman56fd56b2007-10-16 01:25:58 -07002934
2935 /* Get the start pfn, end pfn and the number of blocks to reserve */
2936 start_pfn = zone->zone_start_pfn;
2937 end_pfn = start_pfn + zone->spanned_pages;
Mel Gorman41858962009-06-16 15:32:12 -07002938 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07002939 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07002940
Mel Gorman78986a62009-09-21 17:03:02 -07002941 /*
2942 * Reserve blocks are generally in place to help high-order atomic
2943 * allocations that are short-lived. A min_free_kbytes value that
2944 * would result in more than 2 reserve blocks for atomic allocations
2945 * is assumed to be in place to help anti-fragmentation for the
2946 * future allocation of hugepages at runtime.
2947 */
2948 reserve = min(2, reserve);
2949
Mel Gormand9c23402007-10-16 01:26:01 -07002950 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07002951 if (!pfn_valid(pfn))
2952 continue;
2953 page = pfn_to_page(pfn);
2954
Adam Litke344c7902008-09-02 14:35:38 -07002955 /* Watch out for overlapping nodes */
2956 if (page_to_nid(page) != zone_to_nid(zone))
2957 continue;
2958
Mel Gorman56fd56b2007-10-16 01:25:58 -07002959 /* Blocks with reserved pages will never free, skip them. */
2960 if (PageReserved(page))
2961 continue;
2962
2963 block_migratetype = get_pageblock_migratetype(page);
2964
2965 /* If this block is reserved, account for it */
2966 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
2967 reserve--;
2968 continue;
2969 }
2970
2971 /* Suitable for reserving if this block is movable */
2972 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
2973 set_pageblock_migratetype(page, MIGRATE_RESERVE);
2974 move_freepages_block(zone, page, MIGRATE_RESERVE);
2975 reserve--;
2976 continue;
2977 }
2978
2979 /*
2980 * If the reserve is met and this is a previous reserved block,
2981 * take it back
2982 */
2983 if (block_migratetype == MIGRATE_RESERVE) {
2984 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2985 move_freepages_block(zone, page, MIGRATE_MOVABLE);
2986 }
2987 }
2988}
Mel Gormanac0e5b72007-10-16 01:25:58 -07002989
Linus Torvalds1da177e2005-04-16 15:20:36 -07002990/*
2991 * Initially all pages are reserved - free ones are freed
2992 * up by free_all_bootmem() once the early boot process is
2993 * done. Non-atomic initialization, single-pass.
2994 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002995void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa02007-01-10 23:15:30 -08002996 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002997{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07002999 unsigned long end_pfn = start_pfn + size;
3000 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003001 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002
Hugh Dickins22b31ee2009-01-06 14:40:09 -08003003 if (highest_memmap_pfn < end_pfn - 1)
3004 highest_memmap_pfn = end_pfn - 1;
3005
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003006 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08003007 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa02007-01-10 23:15:30 -08003008 /*
3009 * There can be holes in boot-time mem_map[]s
3010 * handed to this function. They do not
3011 * exist on hotplugged memory.
3012 */
3013 if (context == MEMMAP_EARLY) {
3014 if (!early_pfn_valid(pfn))
3015 continue;
3016 if (!early_pfn_in_nid(pfn, nid))
3017 continue;
3018 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003019 page = pfn_to_page(pfn);
3020 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07003021 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08003022 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003023 reset_page_mapcount(page);
3024 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003025 /*
3026 * Mark the block movable so that blocks are reserved for
3027 * movable at startup. This will force kernel allocations
3028 * to reserve their blocks rather than leaking throughout
3029 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07003030 * kernel allocations are made. Later some blocks near
3031 * the start are marked MIGRATE_RESERVE by
3032 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003033 *
3034 * bitmap is created for zone's valid pfn range. but memmap
3035 * can be created for invalid pages (for alignment)
3036 * check here not to call set_pageblock_migratetype() against
3037 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003038 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003039 if ((z->zone_start_pfn <= pfn)
3040 && (pfn < z->zone_start_pfn + z->spanned_pages)
3041 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07003042 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003043
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044 INIT_LIST_HEAD(&page->lru);
3045#ifdef WANT_PAGE_VIRTUAL
3046 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3047 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07003048 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003049#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003050 }
3051}
3052
Andi Kleen1e548de2008-02-04 22:29:26 -08003053static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003054{
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003055 int order, t;
3056 for_each_migratetype_order(order, t) {
3057 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058 zone->free_area[order].nr_free = 0;
3059 }
3060}
3061
3062#ifndef __HAVE_ARCH_MEMMAP_INIT
3063#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa02007-01-10 23:15:30 -08003064 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003065#endif
3066
Sam Ravnborg1d6f4e62008-01-17 15:21:12 -08003067static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003068{
David Howells3a6be872009-05-06 16:03:03 -07003069#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003070 int batch;
3071
3072 /*
3073 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07003074 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003075 *
3076 * OK, so we don't know how big the cache is. So guess.
3077 */
3078 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07003079 if (batch * PAGE_SIZE > 512 * 1024)
3080 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003081 batch /= 4; /* We effectively *= 4 below */
3082 if (batch < 1)
3083 batch = 1;
3084
3085 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11003086 * Clamp the batch to a 2^n - 1 value. Having a power
3087 * of 2 value was found to be more likely to have
3088 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003089 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11003090 * For example if 2 tasks are alternately allocating
3091 * batches of pages, one task can end up with a lot
3092 * of pages of one half of the possible page colors
3093 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003094 */
David Howells91552032009-05-06 16:03:02 -07003095 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07003096
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003097 return batch;
David Howells3a6be872009-05-06 16:03:03 -07003098
3099#else
3100 /* The deferral and batching of frees should be suppressed under NOMMU
3101 * conditions.
3102 *
3103 * The problem is that NOMMU needs to be able to allocate large chunks
3104 * of contiguous memory as there's no hardware page translation to
3105 * assemble apparent contiguous memory from discontiguous pages.
3106 *
3107 * Queueing large contiguous runs of pages for batching, however,
3108 * causes the pages to actually be freed in smaller chunks. As there
3109 * can be a significant delay between the individual batches being
3110 * recycled, this leads to the once large chunks of space being
3111 * fragmented and becoming unavailable for high-order allocations.
3112 */
3113 return 0;
3114#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003115}
3116
Adrian Bunkb69a7282008-07-23 21:28:12 -07003117static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
Christoph Lameter2caaad42005-06-21 17:15:00 -07003118{
3119 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003120 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003121
Magnus Damm1c6fe942005-10-26 01:58:59 -07003122 memset(p, 0, sizeof(*p));
3123
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003124 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003125 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003126 pcp->high = 6 * batch;
3127 pcp->batch = max(1UL, 1 * batch);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003128 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
3129 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07003130}
3131
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003132/*
3133 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3134 * to the value high for the pageset p.
3135 */
3136
3137static void setup_pagelist_highmark(struct per_cpu_pageset *p,
3138 unsigned long high)
3139{
3140 struct per_cpu_pages *pcp;
3141
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003142 pcp = &p->pcp;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003143 pcp->high = high;
3144 pcp->batch = max(1UL, high/4);
3145 if ((high/4) > (PAGE_SHIFT * 8))
3146 pcp->batch = PAGE_SHIFT * 8;
3147}
3148
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003149/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003150 * Allocate per cpu pagesets and initialize them.
3151 * Before this call only boot pagesets were available.
3152 * Boot pagesets will no longer be used by this processorr
3153 * after setup_per_cpu_pageset().
Christoph Lameter2caaad42005-06-21 17:15:00 -07003154 */
Al Viro78d99552005-12-15 09:18:25 +00003155void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003156{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003157 struct zone *zone;
3158 int cpu;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003159
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003160 for_each_populated_zone(zone) {
3161 zone->pageset = alloc_percpu(struct per_cpu_pageset);
3162
3163 for_each_possible_cpu(cpu) {
3164 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
3165
3166 setup_pageset(pcp, zone_batchsize(zone));
3167
3168 if (percpu_pagelist_fraction)
3169 setup_pagelist_highmark(pcp,
3170 (zone->present_pages /
3171 percpu_pagelist_fraction));
3172 }
3173 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003174}
3175
Sam Ravnborg577a32f2007-05-17 23:29:25 +02003176static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07003177int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07003178{
3179 int i;
3180 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003181 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07003182
3183 /*
3184 * The per-page waitqueue mechanism uses hashed waitqueues
3185 * per zone.
3186 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07003187 zone->wait_table_hash_nr_entries =
3188 wait_table_hash_nr_entries(zone_size_pages);
3189 zone->wait_table_bits =
3190 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003191 alloc_size = zone->wait_table_hash_nr_entries
3192 * sizeof(wait_queue_head_t);
3193
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07003194 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07003195 zone->wait_table = (wait_queue_head_t *)
3196 alloc_bootmem_node(pgdat, alloc_size);
3197 } else {
3198 /*
3199 * This case means that a zone whose size was 0 gets new memory
3200 * via memory hot-add.
3201 * But it may be the case that a new node was hot-added. In
3202 * this case vmalloc() will not be able to use this new node's
3203 * memory - this wait_table must be initialized to use this new
3204 * node itself as well.
3205 * To use this new node's memory, further consideration will be
3206 * necessary.
3207 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07003208 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003209 }
3210 if (!zone->wait_table)
3211 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07003212
Yasunori Goto02b694d2006-06-23 02:03:08 -07003213 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07003214 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003215
3216 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003217}
3218
Shaohua Li112067f2009-09-21 17:01:16 -07003219static int __zone_pcp_update(void *data)
3220{
3221 struct zone *zone = data;
3222 int cpu;
3223 unsigned long batch = zone_batchsize(zone), flags;
3224
Thomas Gleixner2d30a1f2010-03-10 15:20:40 -08003225 for_each_possible_cpu(cpu) {
Shaohua Li112067f2009-09-21 17:01:16 -07003226 struct per_cpu_pageset *pset;
3227 struct per_cpu_pages *pcp;
3228
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003229 pset = per_cpu_ptr(zone->pageset, cpu);
Shaohua Li112067f2009-09-21 17:01:16 -07003230 pcp = &pset->pcp;
3231
3232 local_irq_save(flags);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003233 free_pcppages_bulk(zone, pcp->count, pcp);
Shaohua Li112067f2009-09-21 17:01:16 -07003234 setup_pageset(pset, batch);
3235 local_irq_restore(flags);
3236 }
3237 return 0;
3238}
3239
3240void zone_pcp_update(struct zone *zone)
3241{
3242 stop_machine(__zone_pcp_update, zone, NULL);
3243}
3244
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003245static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07003246{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003247 /*
3248 * per cpu subsystem is not up at this point. The following code
3249 * relies on the ability of the linker to provide the
3250 * offset of a (static) per cpu variable into the per cpu area.
3251 */
3252 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07003253
Anton Blanchardf5335c02006-03-25 03:06:49 -08003254 if (zone->present_pages)
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003255 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
3256 zone->name, zone->present_pages,
3257 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07003258}
3259
Yasunori Goto718127c2006-06-23 02:03:10 -07003260__meminit int init_currently_empty_zone(struct zone *zone,
3261 unsigned long zone_start_pfn,
Dave Hansena2f3aa02007-01-10 23:15:30 -08003262 unsigned long size,
3263 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07003264{
3265 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003266 int ret;
3267 ret = zone_wait_table_init(zone, size);
3268 if (ret)
3269 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07003270 pgdat->nr_zones = zone_idx(zone) + 1;
3271
Dave Hansened8ece22005-10-29 18:16:50 -07003272 zone->zone_start_pfn = zone_start_pfn;
3273
Mel Gorman708614e2008-07-23 21:26:51 -07003274 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3275 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3276 pgdat->node_id,
3277 (unsigned long)zone_idx(zone),
3278 zone_start_pfn, (zone_start_pfn + size));
3279
Andi Kleen1e548de2008-02-04 22:29:26 -08003280 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07003281
3282 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003283}
3284
Mel Gormanc7132162006-09-27 01:49:43 -07003285#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3286/*
3287 * Basic iterator support. Return the first range of PFNs for a node
3288 * Note: nid == MAX_NUMNODES returns first region regardless of node
3289 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003290static int __meminit first_active_region_index_in_nid(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003291{
3292 int i;
3293
3294 for (i = 0; i < nr_nodemap_entries; i++)
3295 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3296 return i;
3297
3298 return -1;
3299}
3300
3301/*
3302 * Basic iterator support. Return the next active range of PFNs for a node
Simon Arlott183ff222007-10-20 01:27:18 +02003303 * Note: nid == MAX_NUMNODES returns next region regardless of node
Mel Gormanc7132162006-09-27 01:49:43 -07003304 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003305static int __meminit next_active_region_index_in_nid(int index, int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003306{
3307 for (index = index + 1; index < nr_nodemap_entries; index++)
3308 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3309 return index;
3310
3311 return -1;
3312}
3313
3314#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3315/*
3316 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3317 * Architectures may implement their own version but if add_active_range()
3318 * was used and there are no special requirements, this is a convenient
3319 * alternative
3320 */
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003321int __meminit __early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07003322{
3323 int i;
3324
3325 for (i = 0; i < nr_nodemap_entries; i++) {
3326 unsigned long start_pfn = early_node_map[i].start_pfn;
3327 unsigned long end_pfn = early_node_map[i].end_pfn;
3328
3329 if (start_pfn <= pfn && pfn < end_pfn)
3330 return early_node_map[i].nid;
3331 }
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003332 /* This is a memory hole */
3333 return -1;
Mel Gormanc7132162006-09-27 01:49:43 -07003334}
3335#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3336
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003337int __meminit early_pfn_to_nid(unsigned long pfn)
3338{
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003339 int nid;
3340
3341 nid = __early_pfn_to_nid(pfn);
3342 if (nid >= 0)
3343 return nid;
3344 /* just returns 0 */
3345 return 0;
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003346}
3347
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003348#ifdef CONFIG_NODES_SPAN_OTHER_NODES
3349bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3350{
3351 int nid;
3352
3353 nid = __early_pfn_to_nid(pfn);
3354 if (nid >= 0 && nid != node)
3355 return false;
3356 return true;
3357}
3358#endif
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003359
Mel Gormanc7132162006-09-27 01:49:43 -07003360/* Basic iterator support to walk early_node_map[] */
3361#define for_each_active_range_index_in_nid(i, nid) \
3362 for (i = first_active_region_index_in_nid(nid); i != -1; \
3363 i = next_active_region_index_in_nid(i, nid))
3364
3365/**
3366 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003367 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3368 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07003369 *
3370 * If an architecture guarantees that all ranges registered with
3371 * add_active_ranges() contain no holes and may be freed, this
3372 * this function may be used instead of calling free_bootmem() manually.
3373 */
3374void __init free_bootmem_with_active_regions(int nid,
3375 unsigned long max_low_pfn)
3376{
3377 int i;
3378
3379 for_each_active_range_index_in_nid(i, nid) {
3380 unsigned long size_pages = 0;
3381 unsigned long end_pfn = early_node_map[i].end_pfn;
3382
3383 if (early_node_map[i].start_pfn >= max_low_pfn)
3384 continue;
3385
3386 if (end_pfn > max_low_pfn)
3387 end_pfn = max_low_pfn;
3388
3389 size_pages = end_pfn - early_node_map[i].start_pfn;
3390 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3391 PFN_PHYS(early_node_map[i].start_pfn),
3392 size_pages << PAGE_SHIFT);
3393 }
3394}
3395
Yinghai Lu08677212010-02-10 01:20:20 -08003396int __init add_from_early_node_map(struct range *range, int az,
3397 int nr_range, int nid)
3398{
3399 int i;
3400 u64 start, end;
3401
3402 /* need to go over early_node_map to find out good range for node */
3403 for_each_active_range_index_in_nid(i, nid) {
3404 start = early_node_map[i].start_pfn;
3405 end = early_node_map[i].end_pfn;
3406 nr_range = add_range(range, az, nr_range, start, end);
3407 }
3408 return nr_range;
3409}
3410
Yinghai Lu2ee78f72010-02-17 11:29:49 -08003411#ifdef CONFIG_NO_BOOTMEM
Yinghai Lu08677212010-02-10 01:20:20 -08003412void * __init __alloc_memory_core_early(int nid, u64 size, u64 align,
3413 u64 goal, u64 limit)
3414{
3415 int i;
3416 void *ptr;
3417
3418 /* need to go over early_node_map to find out good range for node */
3419 for_each_active_range_index_in_nid(i, nid) {
3420 u64 addr;
3421 u64 ei_start, ei_last;
3422
3423 ei_last = early_node_map[i].end_pfn;
3424 ei_last <<= PAGE_SHIFT;
3425 ei_start = early_node_map[i].start_pfn;
3426 ei_start <<= PAGE_SHIFT;
3427 addr = find_early_area(ei_start, ei_last,
3428 goal, limit, size, align);
3429
3430 if (addr == -1ULL)
3431 continue;
3432
3433#if 0
3434 printk(KERN_DEBUG "alloc (nid=%d %llx - %llx) (%llx - %llx) %llx %llx => %llx\n",
3435 nid,
3436 ei_start, ei_last, goal, limit, size,
3437 align, addr);
3438#endif
3439
3440 ptr = phys_to_virt(addr);
3441 memset(ptr, 0, size);
3442 reserve_early_without_check(addr, addr + size, "BOOTMEM");
3443 return ptr;
3444 }
3445
3446 return NULL;
3447}
Yinghai Lu2ee78f72010-02-17 11:29:49 -08003448#endif
Yinghai Lu08677212010-02-10 01:20:20 -08003449
3450
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003451void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3452{
3453 int i;
Yinghai Lud52d53b2008-06-16 20:10:55 -07003454 int ret;
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003455
Yinghai Lud52d53b2008-06-16 20:10:55 -07003456 for_each_active_range_index_in_nid(i, nid) {
3457 ret = work_fn(early_node_map[i].start_pfn,
3458 early_node_map[i].end_pfn, data);
3459 if (ret)
3460 break;
3461 }
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003462}
Mel Gormanc7132162006-09-27 01:49:43 -07003463/**
3464 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003465 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07003466 *
3467 * If an architecture guarantees that all ranges registered with
3468 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003469 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07003470 */
3471void __init sparse_memory_present_with_active_regions(int nid)
3472{
3473 int i;
3474
3475 for_each_active_range_index_in_nid(i, nid)
3476 memory_present(early_node_map[i].nid,
3477 early_node_map[i].start_pfn,
3478 early_node_map[i].end_pfn);
3479}
3480
3481/**
3482 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003483 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3484 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3485 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07003486 *
3487 * It returns the start and end page frame of a node based on information
3488 * provided by an arch calling add_active_range(). If called for a node
3489 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003490 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07003491 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003492void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003493 unsigned long *start_pfn, unsigned long *end_pfn)
3494{
3495 int i;
3496 *start_pfn = -1UL;
3497 *end_pfn = 0;
3498
3499 for_each_active_range_index_in_nid(i, nid) {
3500 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3501 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3502 }
3503
Christoph Lameter633c0662007-10-16 01:25:37 -07003504 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07003505 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07003506}
3507
3508/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07003509 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3510 * assumption is made that zones within a node are ordered in monotonic
3511 * increasing memory addresses so that the "highest" populated zone is used
3512 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003513static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07003514{
3515 int zone_index;
3516 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3517 if (zone_index == ZONE_MOVABLE)
3518 continue;
3519
3520 if (arch_zone_highest_possible_pfn[zone_index] >
3521 arch_zone_lowest_possible_pfn[zone_index])
3522 break;
3523 }
3524
3525 VM_BUG_ON(zone_index == -1);
3526 movable_zone = zone_index;
3527}
3528
3529/*
3530 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3531 * because it is sized independant of architecture. Unlike the other zones,
3532 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3533 * in each node depending on the size of each node and how evenly kernelcore
3534 * is distributed. This helper function adjusts the zone ranges
3535 * provided by the architecture for a given node by using the end of the
3536 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3537 * zones within a node are in order of monotonic increases memory addresses
3538 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003539static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07003540 unsigned long zone_type,
3541 unsigned long node_start_pfn,
3542 unsigned long node_end_pfn,
3543 unsigned long *zone_start_pfn,
3544 unsigned long *zone_end_pfn)
3545{
3546 /* Only adjust if ZONE_MOVABLE is on this node */
3547 if (zone_movable_pfn[nid]) {
3548 /* Size ZONE_MOVABLE */
3549 if (zone_type == ZONE_MOVABLE) {
3550 *zone_start_pfn = zone_movable_pfn[nid];
3551 *zone_end_pfn = min(node_end_pfn,
3552 arch_zone_highest_possible_pfn[movable_zone]);
3553
3554 /* Adjust for ZONE_MOVABLE starting within this range */
3555 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3556 *zone_end_pfn > zone_movable_pfn[nid]) {
3557 *zone_end_pfn = zone_movable_pfn[nid];
3558
3559 /* Check if this whole range is within ZONE_MOVABLE */
3560 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3561 *zone_start_pfn = *zone_end_pfn;
3562 }
3563}
3564
3565/*
Mel Gormanc7132162006-09-27 01:49:43 -07003566 * Return the number of pages a zone spans in a node, including holes
3567 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3568 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003569static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003570 unsigned long zone_type,
3571 unsigned long *ignored)
3572{
3573 unsigned long node_start_pfn, node_end_pfn;
3574 unsigned long zone_start_pfn, zone_end_pfn;
3575
3576 /* Get the start and end of the node and zone */
3577 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3578 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3579 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07003580 adjust_zone_range_for_zone_movable(nid, zone_type,
3581 node_start_pfn, node_end_pfn,
3582 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003583
3584 /* Check that this node has pages within the zone's required range */
3585 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3586 return 0;
3587
3588 /* Move the zone boundaries inside the node if necessary */
3589 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3590 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3591
3592 /* Return the spanned pages */
3593 return zone_end_pfn - zone_start_pfn;
3594}
3595
3596/*
3597 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003598 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07003599 */
Yinghai Lu32996252009-12-15 17:59:02 -08003600unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003601 unsigned long range_start_pfn,
3602 unsigned long range_end_pfn)
3603{
3604 int i = 0;
3605 unsigned long prev_end_pfn = 0, hole_pages = 0;
3606 unsigned long start_pfn;
3607
3608 /* Find the end_pfn of the first active range of pfns in the node */
3609 i = first_active_region_index_in_nid(nid);
3610 if (i == -1)
3611 return 0;
3612
Mel Gormanb5445f92007-07-26 10:41:18 -07003613 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3614
Mel Gorman9c7cd682006-09-27 01:49:58 -07003615 /* Account for ranges before physical memory on this node */
3616 if (early_node_map[i].start_pfn > range_start_pfn)
Mel Gormanb5445f92007-07-26 10:41:18 -07003617 hole_pages = prev_end_pfn - range_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003618
3619 /* Find all holes for the zone within the node */
3620 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3621
3622 /* No need to continue if prev_end_pfn is outside the zone */
3623 if (prev_end_pfn >= range_end_pfn)
3624 break;
3625
3626 /* Make sure the end of the zone is not within the hole */
3627 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3628 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3629
3630 /* Update the hole size cound and move on */
3631 if (start_pfn > range_start_pfn) {
3632 BUG_ON(prev_end_pfn > start_pfn);
3633 hole_pages += start_pfn - prev_end_pfn;
3634 }
3635 prev_end_pfn = early_node_map[i].end_pfn;
3636 }
3637
Mel Gorman9c7cd682006-09-27 01:49:58 -07003638 /* Account for ranges past physical memory on this node */
3639 if (range_end_pfn > prev_end_pfn)
Mel Gorman0c6cb972006-10-28 10:38:59 -07003640 hole_pages += range_end_pfn -
Mel Gorman9c7cd682006-09-27 01:49:58 -07003641 max(range_start_pfn, prev_end_pfn);
3642
Mel Gormanc7132162006-09-27 01:49:43 -07003643 return hole_pages;
3644}
3645
3646/**
3647 * absent_pages_in_range - Return number of page frames in holes within a range
3648 * @start_pfn: The start PFN to start searching for holes
3649 * @end_pfn: The end PFN to stop searching for holes
3650 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003651 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07003652 */
3653unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3654 unsigned long end_pfn)
3655{
3656 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3657}
3658
3659/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003660static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003661 unsigned long zone_type,
3662 unsigned long *ignored)
3663{
Mel Gorman9c7cd682006-09-27 01:49:58 -07003664 unsigned long node_start_pfn, node_end_pfn;
3665 unsigned long zone_start_pfn, zone_end_pfn;
3666
3667 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3668 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3669 node_start_pfn);
3670 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3671 node_end_pfn);
3672
Mel Gorman2a1e2742007-07-17 04:03:12 -07003673 adjust_zone_range_for_zone_movable(nid, zone_type,
3674 node_start_pfn, node_end_pfn,
3675 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07003676 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003677}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003678
Mel Gormanc7132162006-09-27 01:49:43 -07003679#else
Paul Mundt6ea6e682007-07-15 23:38:20 -07003680static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003681 unsigned long zone_type,
3682 unsigned long *zones_size)
3683{
3684 return zones_size[zone_type];
3685}
3686
Paul Mundt6ea6e682007-07-15 23:38:20 -07003687static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003688 unsigned long zone_type,
3689 unsigned long *zholes_size)
3690{
3691 if (!zholes_size)
3692 return 0;
3693
3694 return zholes_size[zone_type];
3695}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003696
Mel Gormanc7132162006-09-27 01:49:43 -07003697#endif
3698
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003699static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Mel Gormanc7132162006-09-27 01:49:43 -07003700 unsigned long *zones_size, unsigned long *zholes_size)
3701{
3702 unsigned long realtotalpages, totalpages = 0;
3703 enum zone_type i;
3704
3705 for (i = 0; i < MAX_NR_ZONES; i++)
3706 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3707 zones_size);
3708 pgdat->node_spanned_pages = totalpages;
3709
3710 realtotalpages = totalpages;
3711 for (i = 0; i < MAX_NR_ZONES; i++)
3712 realtotalpages -=
3713 zone_absent_pages_in_node(pgdat->node_id, i,
3714 zholes_size);
3715 pgdat->node_present_pages = realtotalpages;
3716 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3717 realtotalpages);
3718}
3719
Mel Gorman835c1342007-10-16 01:25:47 -07003720#ifndef CONFIG_SPARSEMEM
3721/*
3722 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07003723 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3724 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07003725 * round what is now in bits to nearest long in bits, then return it in
3726 * bytes.
3727 */
3728static unsigned long __init usemap_size(unsigned long zonesize)
3729{
3730 unsigned long usemapsize;
3731
Mel Gormand9c23402007-10-16 01:26:01 -07003732 usemapsize = roundup(zonesize, pageblock_nr_pages);
3733 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07003734 usemapsize *= NR_PAGEBLOCK_BITS;
3735 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3736
3737 return usemapsize / 8;
3738}
3739
3740static void __init setup_usemap(struct pglist_data *pgdat,
3741 struct zone *zone, unsigned long zonesize)
3742{
3743 unsigned long usemapsize = usemap_size(zonesize);
3744 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08003745 if (usemapsize)
Mel Gorman835c1342007-10-16 01:25:47 -07003746 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
Mel Gorman835c1342007-10-16 01:25:47 -07003747}
3748#else
3749static void inline setup_usemap(struct pglist_data *pgdat,
3750 struct zone *zone, unsigned long zonesize) {}
3751#endif /* CONFIG_SPARSEMEM */
3752
Mel Gormand9c23402007-10-16 01:26:01 -07003753#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08003754
3755/* Return a sensible default order for the pageblock size. */
3756static inline int pageblock_default_order(void)
3757{
3758 if (HPAGE_SHIFT > PAGE_SHIFT)
3759 return HUGETLB_PAGE_ORDER;
3760
3761 return MAX_ORDER-1;
3762}
3763
Mel Gormand9c23402007-10-16 01:26:01 -07003764/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3765static inline void __init set_pageblock_order(unsigned int order)
3766{
3767 /* Check that pageblock_nr_pages has not already been setup */
3768 if (pageblock_order)
3769 return;
3770
3771 /*
3772 * Assume the largest contiguous order of interest is a huge page.
3773 * This value may be variable depending on boot parameters on IA64
3774 */
3775 pageblock_order = order;
3776}
3777#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3778
Mel Gormanba72cb82007-11-28 16:21:13 -08003779/*
3780 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3781 * and pageblock_default_order() are unused as pageblock_order is set
3782 * at compile-time. See include/linux/pageblock-flags.h for the values of
3783 * pageblock_order based on the kernel config
3784 */
3785static inline int pageblock_default_order(unsigned int order)
3786{
3787 return MAX_ORDER-1;
3788}
Mel Gormand9c23402007-10-16 01:26:01 -07003789#define set_pageblock_order(x) do {} while (0)
3790
3791#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3792
Linus Torvalds1da177e2005-04-16 15:20:36 -07003793/*
3794 * Set up the zone data structures:
3795 * - mark all pages reserved
3796 * - mark all memory queues empty
3797 * - clear the memory bitmaps
3798 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08003799static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003800 unsigned long *zones_size, unsigned long *zholes_size)
3801{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07003802 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07003803 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003804 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07003805 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003806
Dave Hansen208d54e2005-10-29 18:16:52 -07003807 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003808 pgdat->nr_zones = 0;
3809 init_waitqueue_head(&pgdat->kswapd_wait);
3810 pgdat->kswapd_max_order = 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003811 pgdat_page_cgroup_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003812
3813 for (j = 0; j < MAX_NR_ZONES; j++) {
3814 struct zone *zone = pgdat->node_zones + j;
Mel Gorman0e0b8642006-09-27 01:49:56 -07003815 unsigned long size, realsize, memmap_pages;
Christoph Lameterb69408e2008-10-18 20:26:14 -07003816 enum lru_list l;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003817
Mel Gormanc7132162006-09-27 01:49:43 -07003818 size = zone_spanned_pages_in_node(nid, j, zones_size);
3819 realsize = size - zone_absent_pages_in_node(nid, j,
3820 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003821
Mel Gorman0e0b8642006-09-27 01:49:56 -07003822 /*
3823 * Adjust realsize so that it accounts for how much memory
3824 * is used by this zone for memmap. This affects the watermark
3825 * and per-cpu initialisations
3826 */
Johannes Weinerf7232152008-05-23 13:04:21 -07003827 memmap_pages =
3828 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
Mel Gorman0e0b8642006-09-27 01:49:56 -07003829 if (realsize >= memmap_pages) {
3830 realsize -= memmap_pages;
Yinghai Lu5594c8c2009-01-06 14:39:14 -08003831 if (memmap_pages)
3832 printk(KERN_DEBUG
3833 " %s zone: %lu pages used for memmap\n",
3834 zone_names[j], memmap_pages);
Mel Gorman0e0b8642006-09-27 01:49:56 -07003835 } else
3836 printk(KERN_WARNING
3837 " %s zone: %lu pages exceeds realsize %lu\n",
3838 zone_names[j], memmap_pages, realsize);
3839
Christoph Lameter62672762007-02-10 01:43:07 -08003840 /* Account for reserved pages */
3841 if (j == 0 && realsize > dma_reserve) {
Mel Gorman0e0b8642006-09-27 01:49:56 -07003842 realsize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07003843 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08003844 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07003845 }
3846
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003847 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003848 nr_kernel_pages += realsize;
3849 nr_all_pages += realsize;
3850
3851 zone->spanned_pages = size;
3852 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07003853#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07003854 zone->node = nid;
Christoph Lameter8417bba2006-09-25 23:31:51 -07003855 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07003856 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07003857 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07003858#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003859 zone->name = zone_names[j];
3860 spin_lock_init(&zone->lock);
3861 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07003862 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003863 zone->zone_pgdat = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003864
Martin Bligh3bb1a852006-10-28 10:38:24 -07003865 zone->prev_priority = DEF_PRIORITY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003866
Dave Hansened8ece22005-10-29 18:16:50 -07003867 zone_pcp_init(zone);
Christoph Lameterb69408e2008-10-18 20:26:14 -07003868 for_each_lru(l) {
3869 INIT_LIST_HEAD(&zone->lru[l].list);
Wu Fengguangf8629632009-09-21 17:03:11 -07003870 zone->reclaim_stat.nr_saved_scan[l] = 0;
Christoph Lameterb69408e2008-10-18 20:26:14 -07003871 }
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08003872 zone->reclaim_stat.recent_rotated[0] = 0;
3873 zone->reclaim_stat.recent_rotated[1] = 0;
3874 zone->reclaim_stat.recent_scanned[0] = 0;
3875 zone->reclaim_stat.recent_scanned[1] = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07003876 zap_zone_vm_stats(zone);
David Rientjese815af92007-10-16 23:25:54 -07003877 zone->flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003878 if (!size)
3879 continue;
3880
Mel Gormanba72cb82007-11-28 16:21:13 -08003881 set_pageblock_order(pageblock_default_order());
Mel Gorman835c1342007-10-16 01:25:47 -07003882 setup_usemap(pgdat, zone, size);
Dave Hansena2f3aa02007-01-10 23:15:30 -08003883 ret = init_currently_empty_zone(zone, zone_start_pfn,
3884 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07003885 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07003886 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003887 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003888 }
3889}
3890
Sam Ravnborg577a32f2007-05-17 23:29:25 +02003891static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003892{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003893 /* Skip empty nodes */
3894 if (!pgdat->node_spanned_pages)
3895 return;
3896
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003897#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07003898 /* ia64 gets its own node_mem_map, before this, without bootmem */
3899 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07003900 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003901 struct page *map;
3902
Bob Piccoe984bb42006-05-20 15:00:31 -07003903 /*
3904 * The zone's endpoints aren't required to be MAX_ORDER
3905 * aligned but the node_mem_map endpoints must be in order
3906 * for the buddy allocator to function correctly.
3907 */
3908 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3909 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3910 end = ALIGN(end, MAX_ORDER_NR_PAGES);
3911 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07003912 map = alloc_remap(pgdat->node_id, size);
3913 if (!map)
3914 map = alloc_bootmem_node(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07003915 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003916 }
Roman Zippel12d810c2007-05-31 00:40:54 -07003917#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07003918 /*
3919 * With no DISCONTIG, the global mem_map is just set as node 0's
3920 */
Mel Gormanc7132162006-09-27 01:49:43 -07003921 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003922 mem_map = NODE_DATA(0)->node_mem_map;
Mel Gormanc7132162006-09-27 01:49:43 -07003923#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3924 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08003925 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Mel Gormanc7132162006-09-27 01:49:43 -07003926#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3927 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003928#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003929#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003930}
3931
Johannes Weiner9109fb72008-07-23 21:27:20 -07003932void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
3933 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003934{
Johannes Weiner9109fb72008-07-23 21:27:20 -07003935 pg_data_t *pgdat = NODE_DATA(nid);
3936
Linus Torvalds1da177e2005-04-16 15:20:36 -07003937 pgdat->node_id = nid;
3938 pgdat->node_start_pfn = node_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003939 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003940
3941 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07003942#ifdef CONFIG_FLAT_NODE_MEM_MAP
3943 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3944 nid, (unsigned long)pgdat,
3945 (unsigned long)pgdat->node_mem_map);
3946#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003947
3948 free_area_init_core(pgdat, zones_size, zholes_size);
3949}
3950
Mel Gormanc7132162006-09-27 01:49:43 -07003951#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07003952
3953#if MAX_NUMNODES > 1
3954/*
3955 * Figure out the number of possible node ids.
3956 */
3957static void __init setup_nr_node_ids(void)
3958{
3959 unsigned int node;
3960 unsigned int highest = 0;
3961
3962 for_each_node_mask(node, node_possible_map)
3963 highest = node;
3964 nr_node_ids = highest + 1;
3965}
3966#else
3967static inline void setup_nr_node_ids(void)
3968{
3969}
3970#endif
3971
Mel Gormanc7132162006-09-27 01:49:43 -07003972/**
3973 * add_active_range - Register a range of PFNs backed by physical memory
3974 * @nid: The node ID the range resides on
3975 * @start_pfn: The start PFN of the available physical memory
3976 * @end_pfn: The end PFN of the available physical memory
3977 *
3978 * These ranges are stored in an early_node_map[] and later used by
3979 * free_area_init_nodes() to calculate zone sizes and holes. If the
3980 * range spans a memory hole, it is up to the architecture to ensure
3981 * the memory is not freed by the bootmem allocator. If possible
3982 * the range being registered will be merged with existing ranges.
3983 */
3984void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3985 unsigned long end_pfn)
3986{
3987 int i;
3988
Mel Gorman6b74ab92008-07-23 21:26:49 -07003989 mminit_dprintk(MMINIT_TRACE, "memory_register",
3990 "Entering add_active_range(%d, %#lx, %#lx) "
3991 "%d entries of %d used\n",
3992 nid, start_pfn, end_pfn,
3993 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
Mel Gormanc7132162006-09-27 01:49:43 -07003994
Mel Gorman2dbb51c2008-07-23 21:26:52 -07003995 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
3996
Mel Gormanc7132162006-09-27 01:49:43 -07003997 /* Merge with existing active regions if possible */
3998 for (i = 0; i < nr_nodemap_entries; i++) {
3999 if (early_node_map[i].nid != nid)
4000 continue;
4001
4002 /* Skip if an existing region covers this new one */
4003 if (start_pfn >= early_node_map[i].start_pfn &&
4004 end_pfn <= early_node_map[i].end_pfn)
4005 return;
4006
4007 /* Merge forward if suitable */
4008 if (start_pfn <= early_node_map[i].end_pfn &&
4009 end_pfn > early_node_map[i].end_pfn) {
4010 early_node_map[i].end_pfn = end_pfn;
4011 return;
4012 }
4013
4014 /* Merge backward if suitable */
Kazuhisa Ichikawad2dbe082010-01-15 17:01:20 -08004015 if (start_pfn < early_node_map[i].start_pfn &&
Mel Gormanc7132162006-09-27 01:49:43 -07004016 end_pfn >= early_node_map[i].start_pfn) {
4017 early_node_map[i].start_pfn = start_pfn;
4018 return;
4019 }
4020 }
4021
4022 /* Check that early_node_map is large enough */
4023 if (i >= MAX_ACTIVE_REGIONS) {
4024 printk(KERN_CRIT "More than %d memory regions, truncating\n",
4025 MAX_ACTIVE_REGIONS);
4026 return;
4027 }
4028
4029 early_node_map[i].nid = nid;
4030 early_node_map[i].start_pfn = start_pfn;
4031 early_node_map[i].end_pfn = end_pfn;
4032 nr_nodemap_entries = i + 1;
4033}
4034
4035/**
Yinghai Lucc1050b2008-06-13 19:08:52 -07004036 * remove_active_range - Shrink an existing registered range of PFNs
Mel Gormanc7132162006-09-27 01:49:43 -07004037 * @nid: The node id the range is on that should be shrunk
Yinghai Lucc1050b2008-06-13 19:08:52 -07004038 * @start_pfn: The new PFN of the range
4039 * @end_pfn: The new PFN of the range
Mel Gormanc7132162006-09-27 01:49:43 -07004040 *
4041 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004042 * The map is kept near the end physical page range that has already been
4043 * registered. This function allows an arch to shrink an existing registered
4044 * range.
Mel Gormanc7132162006-09-27 01:49:43 -07004045 */
Yinghai Lucc1050b2008-06-13 19:08:52 -07004046void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
4047 unsigned long end_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004048{
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004049 int i, j;
4050 int removed = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07004051
Yinghai Lucc1050b2008-06-13 19:08:52 -07004052 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
4053 nid, start_pfn, end_pfn);
4054
Mel Gormanc7132162006-09-27 01:49:43 -07004055 /* Find the old active region end and shrink */
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004056 for_each_active_range_index_in_nid(i, nid) {
Yinghai Lucc1050b2008-06-13 19:08:52 -07004057 if (early_node_map[i].start_pfn >= start_pfn &&
4058 early_node_map[i].end_pfn <= end_pfn) {
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004059 /* clear it */
Yinghai Lucc1050b2008-06-13 19:08:52 -07004060 early_node_map[i].start_pfn = 0;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004061 early_node_map[i].end_pfn = 0;
4062 removed = 1;
4063 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004064 }
Yinghai Lucc1050b2008-06-13 19:08:52 -07004065 if (early_node_map[i].start_pfn < start_pfn &&
4066 early_node_map[i].end_pfn > start_pfn) {
4067 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
4068 early_node_map[i].end_pfn = start_pfn;
4069 if (temp_end_pfn > end_pfn)
4070 add_active_range(nid, end_pfn, temp_end_pfn);
4071 continue;
4072 }
4073 if (early_node_map[i].start_pfn >= start_pfn &&
4074 early_node_map[i].end_pfn > end_pfn &&
4075 early_node_map[i].start_pfn < end_pfn) {
4076 early_node_map[i].start_pfn = end_pfn;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004077 continue;
4078 }
4079 }
4080
4081 if (!removed)
4082 return;
4083
4084 /* remove the blank ones */
4085 for (i = nr_nodemap_entries - 1; i > 0; i--) {
4086 if (early_node_map[i].nid != nid)
4087 continue;
4088 if (early_node_map[i].end_pfn)
4089 continue;
4090 /* we found it, get rid of it */
4091 for (j = i; j < nr_nodemap_entries - 1; j++)
4092 memcpy(&early_node_map[j], &early_node_map[j+1],
4093 sizeof(early_node_map[j]));
4094 j = nr_nodemap_entries - 1;
4095 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
4096 nr_nodemap_entries--;
4097 }
Mel Gormanc7132162006-09-27 01:49:43 -07004098}
4099
4100/**
4101 * remove_all_active_ranges - Remove all currently registered regions
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004102 *
Mel Gormanc7132162006-09-27 01:49:43 -07004103 * During discovery, it may be found that a table like SRAT is invalid
4104 * and an alternative discovery method must be used. This function removes
4105 * all currently registered regions.
4106 */
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004107void __init remove_all_active_ranges(void)
Mel Gormanc7132162006-09-27 01:49:43 -07004108{
4109 memset(early_node_map, 0, sizeof(early_node_map));
4110 nr_nodemap_entries = 0;
4111}
4112
4113/* Compare two active node_active_regions */
4114static int __init cmp_node_active_region(const void *a, const void *b)
4115{
4116 struct node_active_region *arange = (struct node_active_region *)a;
4117 struct node_active_region *brange = (struct node_active_region *)b;
4118
4119 /* Done this way to avoid overflows */
4120 if (arange->start_pfn > brange->start_pfn)
4121 return 1;
4122 if (arange->start_pfn < brange->start_pfn)
4123 return -1;
4124
4125 return 0;
4126}
4127
4128/* sort the node_map by start_pfn */
Yinghai Lu32996252009-12-15 17:59:02 -08004129void __init sort_node_map(void)
Mel Gormanc7132162006-09-27 01:49:43 -07004130{
4131 sort(early_node_map, (size_t)nr_nodemap_entries,
4132 sizeof(struct node_active_region),
4133 cmp_node_active_region, NULL);
4134}
4135
Mel Gormana6af2bc2007-02-10 01:42:57 -08004136/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004137static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07004138{
4139 int i;
Mel Gormana6af2bc2007-02-10 01:42:57 -08004140 unsigned long min_pfn = ULONG_MAX;
Mel Gorman1abbfb42006-11-23 12:01:41 +00004141
Mel Gormanc7132162006-09-27 01:49:43 -07004142 /* Assuming a sorted map, the first range found has the starting pfn */
4143 for_each_active_range_index_in_nid(i, nid)
Mel Gormana6af2bc2007-02-10 01:42:57 -08004144 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004145
Mel Gormana6af2bc2007-02-10 01:42:57 -08004146 if (min_pfn == ULONG_MAX) {
4147 printk(KERN_WARNING
Paul Jackson2bc0d262008-06-22 07:22:17 -07004148 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08004149 return 0;
4150 }
4151
4152 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004153}
4154
4155/**
4156 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4157 *
4158 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004159 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07004160 */
4161unsigned long __init find_min_pfn_with_active_regions(void)
4162{
4163 return find_min_pfn_for_node(MAX_NUMNODES);
4164}
4165
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004166/*
4167 * early_calculate_totalpages()
4168 * Sum pages in active regions for movable zone.
4169 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4170 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07004171static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07004172{
4173 int i;
4174 unsigned long totalpages = 0;
4175
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004176 for (i = 0; i < nr_nodemap_entries; i++) {
4177 unsigned long pages = early_node_map[i].end_pfn -
Mel Gorman7e63efe2007-07-17 04:03:15 -07004178 early_node_map[i].start_pfn;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004179 totalpages += pages;
4180 if (pages)
4181 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
4182 }
4183 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07004184}
4185
Mel Gorman2a1e2742007-07-17 04:03:12 -07004186/*
4187 * Find the PFN the Movable zone begins in each node. Kernel memory
4188 * is spread evenly between nodes as long as the nodes have enough
4189 * memory. When they don't, some nodes will have more kernelcore than
4190 * others
4191 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004192static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004193{
4194 int i, nid;
4195 unsigned long usable_startpfn;
4196 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07004197 /* save the state before borrow the nodemask */
4198 nodemask_t saved_node_state = node_states[N_HIGH_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004199 unsigned long totalpages = early_calculate_totalpages();
4200 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004201
Mel Gorman7e63efe2007-07-17 04:03:15 -07004202 /*
4203 * If movablecore was specified, calculate what size of
4204 * kernelcore that corresponds so that memory usable for
4205 * any allocation type is evenly spread. If both kernelcore
4206 * and movablecore are specified, then the value of kernelcore
4207 * will be used for required_kernelcore if it's greater than
4208 * what movablecore would have allowed.
4209 */
4210 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07004211 unsigned long corepages;
4212
4213 /*
4214 * Round-up so that ZONE_MOVABLE is at least as large as what
4215 * was requested by the user
4216 */
4217 required_movablecore =
4218 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
4219 corepages = totalpages - required_movablecore;
4220
4221 required_kernelcore = max(required_kernelcore, corepages);
4222 }
4223
Mel Gorman2a1e2742007-07-17 04:03:12 -07004224 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4225 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07004226 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004227
4228 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4229 find_usable_zone_for_movable();
4230 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4231
4232restart:
4233 /* Spread kernelcore memory as evenly as possible throughout nodes */
4234 kernelcore_node = required_kernelcore / usable_nodes;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004235 for_each_node_state(nid, N_HIGH_MEMORY) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004236 /*
4237 * Recalculate kernelcore_node if the division per node
4238 * now exceeds what is necessary to satisfy the requested
4239 * amount of memory for the kernel
4240 */
4241 if (required_kernelcore < kernelcore_node)
4242 kernelcore_node = required_kernelcore / usable_nodes;
4243
4244 /*
4245 * As the map is walked, we track how much memory is usable
4246 * by the kernel using kernelcore_remaining. When it is
4247 * 0, the rest of the node is usable by ZONE_MOVABLE
4248 */
4249 kernelcore_remaining = kernelcore_node;
4250
4251 /* Go through each range of PFNs within this node */
4252 for_each_active_range_index_in_nid(i, nid) {
4253 unsigned long start_pfn, end_pfn;
4254 unsigned long size_pages;
4255
4256 start_pfn = max(early_node_map[i].start_pfn,
4257 zone_movable_pfn[nid]);
4258 end_pfn = early_node_map[i].end_pfn;
4259 if (start_pfn >= end_pfn)
4260 continue;
4261
4262 /* Account for what is only usable for kernelcore */
4263 if (start_pfn < usable_startpfn) {
4264 unsigned long kernel_pages;
4265 kernel_pages = min(end_pfn, usable_startpfn)
4266 - start_pfn;
4267
4268 kernelcore_remaining -= min(kernel_pages,
4269 kernelcore_remaining);
4270 required_kernelcore -= min(kernel_pages,
4271 required_kernelcore);
4272
4273 /* Continue if range is now fully accounted */
4274 if (end_pfn <= usable_startpfn) {
4275
4276 /*
4277 * Push zone_movable_pfn to the end so
4278 * that if we have to rebalance
4279 * kernelcore across nodes, we will
4280 * not double account here
4281 */
4282 zone_movable_pfn[nid] = end_pfn;
4283 continue;
4284 }
4285 start_pfn = usable_startpfn;
4286 }
4287
4288 /*
4289 * The usable PFN range for ZONE_MOVABLE is from
4290 * start_pfn->end_pfn. Calculate size_pages as the
4291 * number of pages used as kernelcore
4292 */
4293 size_pages = end_pfn - start_pfn;
4294 if (size_pages > kernelcore_remaining)
4295 size_pages = kernelcore_remaining;
4296 zone_movable_pfn[nid] = start_pfn + size_pages;
4297
4298 /*
4299 * Some kernelcore has been met, update counts and
4300 * break if the kernelcore for this node has been
4301 * satisified
4302 */
4303 required_kernelcore -= min(required_kernelcore,
4304 size_pages);
4305 kernelcore_remaining -= size_pages;
4306 if (!kernelcore_remaining)
4307 break;
4308 }
4309 }
4310
4311 /*
4312 * If there is still required_kernelcore, we do another pass with one
4313 * less node in the count. This will push zone_movable_pfn[nid] further
4314 * along on the nodes that still have memory until kernelcore is
4315 * satisified
4316 */
4317 usable_nodes--;
4318 if (usable_nodes && required_kernelcore > usable_nodes)
4319 goto restart;
4320
4321 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4322 for (nid = 0; nid < MAX_NUMNODES; nid++)
4323 zone_movable_pfn[nid] =
4324 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07004325
4326out:
4327 /* restore the node_state */
4328 node_states[N_HIGH_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004329}
4330
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004331/* Any regular memory on that node ? */
4332static void check_for_regular_memory(pg_data_t *pgdat)
4333{
4334#ifdef CONFIG_HIGHMEM
4335 enum zone_type zone_type;
4336
4337 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4338 struct zone *zone = &pgdat->node_zones[zone_type];
4339 if (zone->present_pages)
4340 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4341 }
4342#endif
4343}
4344
Mel Gormanc7132162006-09-27 01:49:43 -07004345/**
4346 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004347 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07004348 *
4349 * This will call free_area_init_node() for each active node in the system.
4350 * Using the page ranges provided by add_active_range(), the size of each
4351 * zone in each node and their holes is calculated. If the maximum PFN
4352 * between two adjacent zones match, it is assumed that the zone is empty.
4353 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4354 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4355 * starts where the previous one ended. For example, ZONE_DMA32 starts
4356 * at arch_max_dma_pfn.
4357 */
4358void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4359{
4360 unsigned long nid;
Andrew Mortondb991002008-10-15 22:01:07 -07004361 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004362
Mel Gormana6af2bc2007-02-10 01:42:57 -08004363 /* Sort early_node_map as initialisation assumes it is sorted */
4364 sort_node_map();
4365
Mel Gormanc7132162006-09-27 01:49:43 -07004366 /* Record where the zone boundaries are */
4367 memset(arch_zone_lowest_possible_pfn, 0,
4368 sizeof(arch_zone_lowest_possible_pfn));
4369 memset(arch_zone_highest_possible_pfn, 0,
4370 sizeof(arch_zone_highest_possible_pfn));
4371 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4372 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4373 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004374 if (i == ZONE_MOVABLE)
4375 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004376 arch_zone_lowest_possible_pfn[i] =
4377 arch_zone_highest_possible_pfn[i-1];
4378 arch_zone_highest_possible_pfn[i] =
4379 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4380 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07004381 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4382 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4383
4384 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4385 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4386 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004387
Mel Gormanc7132162006-09-27 01:49:43 -07004388 /* Print out the zone ranges */
4389 printk("Zone PFN ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07004390 for (i = 0; i < MAX_NR_ZONES; i++) {
4391 if (i == ZONE_MOVABLE)
4392 continue;
David Rientjes72f0ba02010-03-05 13:42:14 -08004393 printk(" %-8s ", zone_names[i]);
4394 if (arch_zone_lowest_possible_pfn[i] ==
4395 arch_zone_highest_possible_pfn[i])
4396 printk("empty\n");
4397 else
4398 printk("%0#10lx -> %0#10lx\n",
Mel Gormanc7132162006-09-27 01:49:43 -07004399 arch_zone_lowest_possible_pfn[i],
4400 arch_zone_highest_possible_pfn[i]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004401 }
4402
4403 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4404 printk("Movable zone start PFN for each node\n");
4405 for (i = 0; i < MAX_NUMNODES; i++) {
4406 if (zone_movable_pfn[i])
4407 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4408 }
Mel Gormanc7132162006-09-27 01:49:43 -07004409
4410 /* Print out the early_node_map[] */
4411 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4412 for (i = 0; i < nr_nodemap_entries; i++)
Paul Jackson5dab8ec2008-06-25 05:44:40 -07004413 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004414 early_node_map[i].start_pfn,
4415 early_node_map[i].end_pfn);
4416
4417 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07004418 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08004419 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07004420 for_each_online_node(nid) {
4421 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07004422 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07004423 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004424
4425 /* Any memory on that node */
4426 if (pgdat->node_present_pages)
4427 node_set_state(nid, N_HIGH_MEMORY);
4428 check_for_regular_memory(pgdat);
Mel Gormanc7132162006-09-27 01:49:43 -07004429 }
4430}
Mel Gorman2a1e2742007-07-17 04:03:12 -07004431
Mel Gorman7e63efe2007-07-17 04:03:15 -07004432static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004433{
4434 unsigned long long coremem;
4435 if (!p)
4436 return -EINVAL;
4437
4438 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004439 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004440
Mel Gorman7e63efe2007-07-17 04:03:15 -07004441 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07004442 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4443
4444 return 0;
4445}
Mel Gormaned7ed362007-07-17 04:03:14 -07004446
Mel Gorman7e63efe2007-07-17 04:03:15 -07004447/*
4448 * kernelcore=size sets the amount of memory for use for allocations that
4449 * cannot be reclaimed or migrated.
4450 */
4451static int __init cmdline_parse_kernelcore(char *p)
4452{
4453 return cmdline_parse_core(p, &required_kernelcore);
4454}
4455
4456/*
4457 * movablecore=size sets the amount of memory for use for allocations that
4458 * can be reclaimed or migrated.
4459 */
4460static int __init cmdline_parse_movablecore(char *p)
4461{
4462 return cmdline_parse_core(p, &required_movablecore);
4463}
4464
Mel Gormaned7ed362007-07-17 04:03:14 -07004465early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004466early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07004467
Mel Gormanc7132162006-09-27 01:49:43 -07004468#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4469
Mel Gorman0e0b8642006-09-27 01:49:56 -07004470/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004471 * set_dma_reserve - set the specified number of pages reserved in the first zone
4472 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07004473 *
4474 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4475 * In the DMA zone, a significant percentage may be consumed by kernel image
4476 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004477 * function may optionally be used to account for unfreeable pages in the
4478 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4479 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07004480 */
4481void __init set_dma_reserve(unsigned long new_dma_reserve)
4482{
4483 dma_reserve = new_dma_reserve;
4484}
4485
Dave Hansen93b75042005-06-23 00:07:47 -07004486#ifndef CONFIG_NEED_MULTIPLE_NODES
Yinghai Lu08677212010-02-10 01:20:20 -08004487struct pglist_data __refdata contig_page_data = {
4488#ifndef CONFIG_NO_BOOTMEM
4489 .bdata = &bootmem_node_data[0]
4490#endif
4491 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07004492EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07004493#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004494
4495void __init free_area_init(unsigned long *zones_size)
4496{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004497 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004498 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4499}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004500
Linus Torvalds1da177e2005-04-16 15:20:36 -07004501static int page_alloc_cpu_notify(struct notifier_block *self,
4502 unsigned long action, void *hcpu)
4503{
4504 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004505
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07004506 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004507 drain_pages(cpu);
4508
4509 /*
4510 * Spill the event counters of the dead processor
4511 * into the current processors event counters.
4512 * This artificially elevates the count of the current
4513 * processor.
4514 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07004515 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004516
4517 /*
4518 * Zero the differential counters of the dead processor
4519 * so that the vm statistics are consistent.
4520 *
4521 * This is only okay since the processor is dead and cannot
4522 * race with what we are doing.
4523 */
Christoph Lameter2244b952006-06-30 01:55:33 -07004524 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004525 }
4526 return NOTIFY_OK;
4527}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004528
4529void __init page_alloc_init(void)
4530{
4531 hotcpu_notifier(page_alloc_cpu_notify, 0);
4532}
4533
4534/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004535 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4536 * or min_free_kbytes changes.
4537 */
4538static void calculate_totalreserve_pages(void)
4539{
4540 struct pglist_data *pgdat;
4541 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004542 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004543
4544 for_each_online_pgdat(pgdat) {
4545 for (i = 0; i < MAX_NR_ZONES; i++) {
4546 struct zone *zone = pgdat->node_zones + i;
4547 unsigned long max = 0;
4548
4549 /* Find valid and maximum lowmem_reserve in the zone */
4550 for (j = i; j < MAX_NR_ZONES; j++) {
4551 if (zone->lowmem_reserve[j] > max)
4552 max = zone->lowmem_reserve[j];
4553 }
4554
Mel Gorman41858962009-06-16 15:32:12 -07004555 /* we treat the high watermark as reserved pages. */
4556 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004557
4558 if (max > zone->present_pages)
4559 max = zone->present_pages;
4560 reserve_pages += max;
4561 }
4562 }
4563 totalreserve_pages = reserve_pages;
4564}
4565
4566/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07004567 * setup_per_zone_lowmem_reserve - called whenever
4568 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4569 * has a correct pages reserved value, so an adequate number of
4570 * pages are left in the zone after a successful __alloc_pages().
4571 */
4572static void setup_per_zone_lowmem_reserve(void)
4573{
4574 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004575 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004576
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08004577 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004578 for (j = 0; j < MAX_NR_ZONES; j++) {
4579 struct zone *zone = pgdat->node_zones + j;
4580 unsigned long present_pages = zone->present_pages;
4581
4582 zone->lowmem_reserve[j] = 0;
4583
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004584 idx = j;
4585 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004586 struct zone *lower_zone;
4587
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004588 idx--;
4589
Linus Torvalds1da177e2005-04-16 15:20:36 -07004590 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4591 sysctl_lowmem_reserve_ratio[idx] = 1;
4592
4593 lower_zone = pgdat->node_zones + idx;
4594 lower_zone->lowmem_reserve[j] = present_pages /
4595 sysctl_lowmem_reserve_ratio[idx];
4596 present_pages += lower_zone->present_pages;
4597 }
4598 }
4599 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004600
4601 /* update totalreserve_pages */
4602 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004603}
4604
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004605/**
Minchan Kimbc75d332009-06-16 15:32:48 -07004606 * setup_per_zone_wmarks - called when min_free_kbytes changes
Minchan Kimbce73942009-06-16 15:32:50 -07004607 * or when memory is hot-{added|removed}
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004608 *
Minchan Kimbc75d332009-06-16 15:32:48 -07004609 * Ensures that the watermark[min,low,high] values for each zone are set
4610 * correctly with respect to min_free_kbytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004611 */
Minchan Kimbc75d332009-06-16 15:32:48 -07004612void setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004613{
4614 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4615 unsigned long lowmem_pages = 0;
4616 struct zone *zone;
4617 unsigned long flags;
4618
4619 /* Calculate total number of !ZONE_HIGHMEM pages */
4620 for_each_zone(zone) {
4621 if (!is_highmem(zone))
4622 lowmem_pages += zone->present_pages;
4623 }
4624
4625 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07004626 u64 tmp;
4627
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07004628 spin_lock_irqsave(&zone->lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07004629 tmp = (u64)pages_min * zone->present_pages;
4630 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004631 if (is_highmem(zone)) {
4632 /*
Nick Piggin669ed172005-11-13 16:06:45 -08004633 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4634 * need highmem pages, so cap pages_min to a small
4635 * value here.
4636 *
Mel Gorman41858962009-06-16 15:32:12 -07004637 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Nick Piggin669ed172005-11-13 16:06:45 -08004638 * deltas controls asynch page reclaim, and so should
4639 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004640 */
4641 int min_pages;
4642
4643 min_pages = zone->present_pages / 1024;
4644 if (min_pages < SWAP_CLUSTER_MAX)
4645 min_pages = SWAP_CLUSTER_MAX;
4646 if (min_pages > 128)
4647 min_pages = 128;
Mel Gorman41858962009-06-16 15:32:12 -07004648 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004649 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08004650 /*
4651 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07004652 * proportionate to the zone's size.
4653 */
Mel Gorman41858962009-06-16 15:32:12 -07004654 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004655 }
4656
Mel Gorman41858962009-06-16 15:32:12 -07004657 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
4658 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Mel Gorman56fd56b2007-10-16 01:25:58 -07004659 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07004660 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004661 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004662
4663 /* update totalreserve_pages */
4664 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004665}
4666
Randy Dunlap55a44622009-09-21 17:01:20 -07004667/*
Rik van Riel556adec2008-10-18 20:26:34 -07004668 * The inactive anon list should be small enough that the VM never has to
4669 * do too much work, but large enough that each inactive page has a chance
4670 * to be referenced again before it is swapped out.
4671 *
4672 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4673 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4674 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4675 * the anonymous pages are kept on the inactive list.
4676 *
4677 * total target max
4678 * memory ratio inactive anon
4679 * -------------------------------------
4680 * 10MB 1 5MB
4681 * 100MB 1 50MB
4682 * 1GB 3 250MB
4683 * 10GB 10 0.9GB
4684 * 100GB 31 3GB
4685 * 1TB 101 10GB
4686 * 10TB 320 32GB
4687 */
Minchan Kim96cb4df2009-06-16 15:32:49 -07004688void calculate_zone_inactive_ratio(struct zone *zone)
4689{
4690 unsigned int gb, ratio;
4691
4692 /* Zone size in gigabytes */
4693 gb = zone->present_pages >> (30 - PAGE_SHIFT);
4694 if (gb)
4695 ratio = int_sqrt(10 * gb);
4696 else
4697 ratio = 1;
4698
4699 zone->inactive_ratio = ratio;
4700}
4701
Cyrill Gorcunove9bb35d2009-06-16 15:32:32 -07004702static void __init setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07004703{
4704 struct zone *zone;
4705
Minchan Kim96cb4df2009-06-16 15:32:49 -07004706 for_each_zone(zone)
4707 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07004708}
4709
Linus Torvalds1da177e2005-04-16 15:20:36 -07004710/*
4711 * Initialise min_free_kbytes.
4712 *
4713 * For small machines we want it small (128k min). For large machines
4714 * we want it large (64MB max). But it is not linear, because network
4715 * bandwidth does not increase linearly with machine size. We use
4716 *
4717 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4718 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4719 *
4720 * which yields
4721 *
4722 * 16MB: 512k
4723 * 32MB: 724k
4724 * 64MB: 1024k
4725 * 128MB: 1448k
4726 * 256MB: 2048k
4727 * 512MB: 2896k
4728 * 1024MB: 4096k
4729 * 2048MB: 5792k
4730 * 4096MB: 8192k
4731 * 8192MB: 11584k
4732 * 16384MB: 16384k
4733 */
Minchan Kimbc75d332009-06-16 15:32:48 -07004734static int __init init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004735{
4736 unsigned long lowmem_kbytes;
4737
4738 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4739
4740 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4741 if (min_free_kbytes < 128)
4742 min_free_kbytes = 128;
4743 if (min_free_kbytes > 65536)
4744 min_free_kbytes = 65536;
Minchan Kimbc75d332009-06-16 15:32:48 -07004745 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004746 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07004747 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004748 return 0;
4749}
Minchan Kimbc75d332009-06-16 15:32:48 -07004750module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004751
4752/*
4753 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4754 * that we can call two helper functions whenever min_free_kbytes
4755 * changes.
4756 */
4757int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004758 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004759{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004760 proc_dointvec(table, write, buffer, length, ppos);
Mel Gorman3b1d92c2007-05-06 14:49:30 -07004761 if (write)
Minchan Kimbc75d332009-06-16 15:32:48 -07004762 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004763 return 0;
4764}
4765
Christoph Lameter96146342006-07-03 00:24:13 -07004766#ifdef CONFIG_NUMA
4767int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004768 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07004769{
4770 struct zone *zone;
4771 int rc;
4772
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004773 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07004774 if (rc)
4775 return rc;
4776
4777 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07004778 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07004779 sysctl_min_unmapped_ratio) / 100;
4780 return 0;
4781}
Christoph Lameter0ff38492006-09-25 23:31:52 -07004782
4783int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004784 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07004785{
4786 struct zone *zone;
4787 int rc;
4788
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004789 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07004790 if (rc)
4791 return rc;
4792
4793 for_each_zone(zone)
4794 zone->min_slab_pages = (zone->present_pages *
4795 sysctl_min_slab_ratio) / 100;
4796 return 0;
4797}
Christoph Lameter96146342006-07-03 00:24:13 -07004798#endif
4799
Linus Torvalds1da177e2005-04-16 15:20:36 -07004800/*
4801 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4802 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4803 * whenever sysctl_lowmem_reserve_ratio changes.
4804 *
4805 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07004806 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07004807 * if in function of the boot time zone sizes.
4808 */
4809int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004810 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004811{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004812 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004813 setup_per_zone_lowmem_reserve();
4814 return 0;
4815}
4816
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004817/*
4818 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4819 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4820 * can have before it gets flushed back to buddy allocator.
4821 */
4822
4823int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004824 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004825{
4826 struct zone *zone;
4827 unsigned int cpu;
4828 int ret;
4829
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004830 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004831 if (!write || (ret == -EINVAL))
4832 return ret;
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07004833 for_each_populated_zone(zone) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004834 for_each_possible_cpu(cpu) {
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004835 unsigned long high;
4836 high = zone->present_pages / percpu_pagelist_fraction;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004837 setup_pagelist_highmark(
4838 per_cpu_ptr(zone->pageset, cpu), high);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004839 }
4840 }
4841 return 0;
4842}
4843
David S. Millerf034b5d2006-08-24 03:08:07 -07004844int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004845
4846#ifdef CONFIG_NUMA
4847static int __init set_hashdist(char *str)
4848{
4849 if (!str)
4850 return 0;
4851 hashdist = simple_strtoul(str, &str, 0);
4852 return 1;
4853}
4854__setup("hashdist=", set_hashdist);
4855#endif
4856
4857/*
4858 * allocate a large system hash table from bootmem
4859 * - it is assumed that the hash table must contain an exact power-of-2
4860 * quantity of entries
4861 * - limit is the number of hash buckets, not the total allocation size
4862 */
4863void *__init alloc_large_system_hash(const char *tablename,
4864 unsigned long bucketsize,
4865 unsigned long numentries,
4866 int scale,
4867 int flags,
4868 unsigned int *_hash_shift,
4869 unsigned int *_hash_mask,
4870 unsigned long limit)
4871{
4872 unsigned long long max = limit;
4873 unsigned long log2qty, size;
4874 void *table = NULL;
4875
4876 /* allow the kernel cmdline to have a say */
4877 if (!numentries) {
4878 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08004879 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004880 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4881 numentries >>= 20 - PAGE_SHIFT;
4882 numentries <<= 20 - PAGE_SHIFT;
4883
4884 /* limit to 1 bucket per 2^scale bytes of low memory */
4885 if (scale > PAGE_SHIFT)
4886 numentries >>= (scale - PAGE_SHIFT);
4887 else
4888 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08004889
4890 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07004891 if (unlikely(flags & HASH_SMALL)) {
4892 /* Makes no sense without HASH_EARLY */
4893 WARN_ON(!(flags & HASH_EARLY));
4894 if (!(numentries >> *_hash_shift)) {
4895 numentries = 1UL << *_hash_shift;
4896 BUG_ON(!numentries);
4897 }
4898 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08004899 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004900 }
John Hawkes6e692ed2006-03-25 03:08:02 -08004901 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004902
4903 /* limit allocation size to 1/16 total memory by default */
4904 if (max == 0) {
4905 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4906 do_div(max, bucketsize);
4907 }
4908
4909 if (numentries > max)
4910 numentries = max;
4911
David Howellsf0d1b0b2006-12-08 02:37:49 -08004912 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004913
4914 do {
4915 size = bucketsize << log2qty;
4916 if (flags & HASH_EARLY)
Jan Beulich74768ed2008-08-12 15:08:39 -07004917 table = alloc_bootmem_nopanic(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004918 else if (hashdist)
4919 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4920 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07004921 /*
4922 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07004923 * some pages at the end of hash table which
4924 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07004925 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01004926 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07004927 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01004928 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
4929 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004930 }
4931 } while (!table && size > PAGE_SIZE && --log2qty);
4932
4933 if (!table)
4934 panic("Failed to allocate %s hash table\n", tablename);
4935
Dan Alonib49ad482007-07-15 23:38:23 -07004936 printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07004937 tablename,
4938 (1U << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08004939 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004940 size);
4941
4942 if (_hash_shift)
4943 *_hash_shift = log2qty;
4944 if (_hash_mask)
4945 *_hash_mask = (1 << log2qty) - 1;
4946
4947 return table;
4948}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004949
Mel Gorman835c1342007-10-16 01:25:47 -07004950/* Return a pointer to the bitmap storing bits affecting a block of pages */
4951static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4952 unsigned long pfn)
4953{
4954#ifdef CONFIG_SPARSEMEM
4955 return __pfn_to_section(pfn)->pageblock_flags;
4956#else
4957 return zone->pageblock_flags;
4958#endif /* CONFIG_SPARSEMEM */
4959}
Andrew Morton6220ec72006-10-19 23:29:05 -07004960
Mel Gorman835c1342007-10-16 01:25:47 -07004961static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4962{
4963#ifdef CONFIG_SPARSEMEM
4964 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07004965 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07004966#else
4967 pfn = pfn - zone->zone_start_pfn;
Mel Gormand9c23402007-10-16 01:26:01 -07004968 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07004969#endif /* CONFIG_SPARSEMEM */
4970}
4971
4972/**
Mel Gormand9c23402007-10-16 01:26:01 -07004973 * 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 -07004974 * @page: The page within the block of interest
4975 * @start_bitidx: The first bit of interest to retrieve
4976 * @end_bitidx: The last bit of interest
4977 * returns pageblock_bits flags
4978 */
4979unsigned long get_pageblock_flags_group(struct page *page,
4980 int start_bitidx, int end_bitidx)
4981{
4982 struct zone *zone;
4983 unsigned long *bitmap;
4984 unsigned long pfn, bitidx;
4985 unsigned long flags = 0;
4986 unsigned long value = 1;
4987
4988 zone = page_zone(page);
4989 pfn = page_to_pfn(page);
4990 bitmap = get_pageblock_bitmap(zone, pfn);
4991 bitidx = pfn_to_bitidx(zone, pfn);
4992
4993 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4994 if (test_bit(bitidx + start_bitidx, bitmap))
4995 flags |= value;
4996
4997 return flags;
4998}
4999
5000/**
Mel Gormand9c23402007-10-16 01:26:01 -07005001 * 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 -07005002 * @page: The page within the block of interest
5003 * @start_bitidx: The first bit of interest
5004 * @end_bitidx: The last bit of interest
5005 * @flags: The flags to set
5006 */
5007void set_pageblock_flags_group(struct page *page, unsigned long flags,
5008 int start_bitidx, int end_bitidx)
5009{
5010 struct zone *zone;
5011 unsigned long *bitmap;
5012 unsigned long pfn, bitidx;
5013 unsigned long value = 1;
5014
5015 zone = page_zone(page);
5016 pfn = page_to_pfn(page);
5017 bitmap = get_pageblock_bitmap(zone, pfn);
5018 bitidx = pfn_to_bitidx(zone, pfn);
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07005019 VM_BUG_ON(pfn < zone->zone_start_pfn);
5020 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
Mel Gorman835c1342007-10-16 01:25:47 -07005021
5022 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5023 if (flags & value)
5024 __set_bit(bitidx + start_bitidx, bitmap);
5025 else
5026 __clear_bit(bitidx + start_bitidx, bitmap);
5027}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005028
5029/*
5030 * This is designed as sub function...plz see page_isolation.c also.
5031 * set/clear page block's type to be ISOLATE.
5032 * page allocater never alloc memory from ISOLATE block.
5033 */
5034
5035int set_migratetype_isolate(struct page *page)
5036{
5037 struct zone *zone;
Robert Jennings925cc712009-12-17 14:44:38 +00005038 struct page *curr_page;
5039 unsigned long flags, pfn, iter;
5040 unsigned long immobile = 0;
5041 struct memory_isolate_notify arg;
5042 int notifier_ret;
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005043 int ret = -EBUSY;
Shaohua Li8e7e40d2009-09-21 17:01:18 -07005044 int zone_idx;
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005045
5046 zone = page_zone(page);
Shaohua Li8e7e40d2009-09-21 17:01:18 -07005047 zone_idx = zone_idx(zone);
Robert Jennings925cc712009-12-17 14:44:38 +00005048
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005049 spin_lock_irqsave(&zone->lock, flags);
Robert Jennings925cc712009-12-17 14:44:38 +00005050 if (get_pageblock_migratetype(page) == MIGRATE_MOVABLE ||
5051 zone_idx == ZONE_MOVABLE) {
5052 ret = 0;
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005053 goto out;
Robert Jennings925cc712009-12-17 14:44:38 +00005054 }
5055
5056 pfn = page_to_pfn(page);
5057 arg.start_pfn = pfn;
5058 arg.nr_pages = pageblock_nr_pages;
5059 arg.pages_found = 0;
5060
5061 /*
5062 * It may be possible to isolate a pageblock even if the
5063 * migratetype is not MIGRATE_MOVABLE. The memory isolation
5064 * notifier chain is used by balloon drivers to return the
5065 * number of pages in a range that are held by the balloon
5066 * driver to shrink memory. If all the pages are accounted for
5067 * by balloons, are free, or on the LRU, isolation can continue.
5068 * Later, for example, when memory hotplug notifier runs, these
5069 * pages reported as "can be isolated" should be isolated(freed)
5070 * by the balloon driver through the memory notifier chain.
5071 */
5072 notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg);
5073 notifier_ret = notifier_to_errno(notifier_ret);
5074 if (notifier_ret || !arg.pages_found)
5075 goto out;
5076
5077 for (iter = pfn; iter < (pfn + pageblock_nr_pages); iter++) {
5078 if (!pfn_valid_within(pfn))
5079 continue;
5080
5081 curr_page = pfn_to_page(iter);
5082 if (!page_count(curr_page) || PageLRU(curr_page))
5083 continue;
5084
5085 immobile++;
5086 }
5087
5088 if (arg.pages_found == immobile)
5089 ret = 0;
5090
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005091out:
Robert Jennings925cc712009-12-17 14:44:38 +00005092 if (!ret) {
5093 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
5094 move_freepages_block(zone, page, MIGRATE_ISOLATE);
5095 }
5096
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005097 spin_unlock_irqrestore(&zone->lock, flags);
5098 if (!ret)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005099 drain_all_pages();
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005100 return ret;
5101}
5102
5103void unset_migratetype_isolate(struct page *page)
5104{
5105 struct zone *zone;
5106 unsigned long flags;
5107 zone = page_zone(page);
5108 spin_lock_irqsave(&zone->lock, flags);
5109 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
5110 goto out;
5111 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5112 move_freepages_block(zone, page, MIGRATE_MOVABLE);
5113out:
5114 spin_unlock_irqrestore(&zone->lock, flags);
5115}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07005116
5117#ifdef CONFIG_MEMORY_HOTREMOVE
5118/*
5119 * All pages in the range must be isolated before calling this.
5120 */
5121void
5122__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
5123{
5124 struct page *page;
5125 struct zone *zone;
5126 int order, i;
5127 unsigned long pfn;
5128 unsigned long flags;
5129 /* find the first valid pfn */
5130 for (pfn = start_pfn; pfn < end_pfn; pfn++)
5131 if (pfn_valid(pfn))
5132 break;
5133 if (pfn == end_pfn)
5134 return;
5135 zone = page_zone(pfn_to_page(pfn));
5136 spin_lock_irqsave(&zone->lock, flags);
5137 pfn = start_pfn;
5138 while (pfn < end_pfn) {
5139 if (!pfn_valid(pfn)) {
5140 pfn++;
5141 continue;
5142 }
5143 page = pfn_to_page(pfn);
5144 BUG_ON(page_count(page));
5145 BUG_ON(!PageBuddy(page));
5146 order = page_order(page);
5147#ifdef CONFIG_DEBUG_VM
5148 printk(KERN_INFO "remove from free list %lx %d %lx\n",
5149 pfn, 1 << order, end_pfn);
5150#endif
5151 list_del(&page->lru);
5152 rmv_page_order(page);
5153 zone->free_area[order].nr_free--;
5154 __mod_zone_page_state(zone, NR_FREE_PAGES,
5155 - (1UL << order));
5156 for (i = 0; i < (1 << order); i++)
5157 SetPageReserved((page+i));
5158 pfn += (1 << order);
5159 }
5160 spin_unlock_irqrestore(&zone->lock, flags);
5161}
5162#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01005163
5164#ifdef CONFIG_MEMORY_FAILURE
5165bool is_free_buddy_page(struct page *page)
5166{
5167 struct zone *zone = page_zone(page);
5168 unsigned long pfn = page_to_pfn(page);
5169 unsigned long flags;
5170 int order;
5171
5172 spin_lock_irqsave(&zone->lock, flags);
5173 for (order = 0; order < MAX_ORDER; order++) {
5174 struct page *page_head = page - (pfn & ((1 << order) - 1));
5175
5176 if (PageBuddy(page_head) && page_order(page_head) >= order)
5177 break;
5178 }
5179 spin_unlock_irqrestore(&zone->lock, flags);
5180
5181 return order < MAX_ORDER;
5182}
5183#endif
Wu Fengguang718a3822010-03-10 15:20:43 -08005184
5185static struct trace_print_flags pageflag_names[] = {
5186 {1UL << PG_locked, "locked" },
5187 {1UL << PG_error, "error" },
5188 {1UL << PG_referenced, "referenced" },
5189 {1UL << PG_uptodate, "uptodate" },
5190 {1UL << PG_dirty, "dirty" },
5191 {1UL << PG_lru, "lru" },
5192 {1UL << PG_active, "active" },
5193 {1UL << PG_slab, "slab" },
5194 {1UL << PG_owner_priv_1, "owner_priv_1" },
5195 {1UL << PG_arch_1, "arch_1" },
5196 {1UL << PG_reserved, "reserved" },
5197 {1UL << PG_private, "private" },
5198 {1UL << PG_private_2, "private_2" },
5199 {1UL << PG_writeback, "writeback" },
5200#ifdef CONFIG_PAGEFLAGS_EXTENDED
5201 {1UL << PG_head, "head" },
5202 {1UL << PG_tail, "tail" },
5203#else
5204 {1UL << PG_compound, "compound" },
5205#endif
5206 {1UL << PG_swapcache, "swapcache" },
5207 {1UL << PG_mappedtodisk, "mappedtodisk" },
5208 {1UL << PG_reclaim, "reclaim" },
5209 {1UL << PG_buddy, "buddy" },
5210 {1UL << PG_swapbacked, "swapbacked" },
5211 {1UL << PG_unevictable, "unevictable" },
5212#ifdef CONFIG_MMU
5213 {1UL << PG_mlocked, "mlocked" },
5214#endif
5215#ifdef CONFIG_ARCH_USES_PG_UNCACHED
5216 {1UL << PG_uncached, "uncached" },
5217#endif
5218#ifdef CONFIG_MEMORY_FAILURE
5219 {1UL << PG_hwpoison, "hwpoison" },
5220#endif
5221 {-1UL, NULL },
5222};
5223
5224static void dump_page_flags(unsigned long flags)
5225{
5226 const char *delim = "";
5227 unsigned long mask;
5228 int i;
5229
5230 printk(KERN_ALERT "page flags: %#lx(", flags);
5231
5232 /* remove zone id */
5233 flags &= (1UL << NR_PAGEFLAGS) - 1;
5234
5235 for (i = 0; pageflag_names[i].name && flags; i++) {
5236
5237 mask = pageflag_names[i].mask;
5238 if ((flags & mask) != mask)
5239 continue;
5240
5241 flags &= ~mask;
5242 printk("%s%s", delim, pageflag_names[i].name);
5243 delim = "|";
5244 }
5245
5246 /* check for left over flags */
5247 if (flags)
5248 printk("%s%#lx", delim, flags);
5249
5250 printk(")\n");
5251}
5252
5253void dump_page(struct page *page)
5254{
5255 printk(KERN_ALERT
5256 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
5257 page, page_count(page), page_mapcount(page),
5258 page->mapping, page->index);
5259 dump_page_flags(page->flags);
5260}