blob: a0de15f46987f65249bd60d4cad45e129fbddf29 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
52#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070053#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070054#include "internal.h"
55
56/*
Christoph Lameter13808912007-10-16 01:25:27 -070057 * Array of node states.
Linus Torvalds1da177e2005-04-16 15:20:36 -070058 */
Christoph Lameter13808912007-10-16 01:25:27 -070059nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
60 [N_POSSIBLE] = NODE_MASK_ALL,
61 [N_ONLINE] = { { [0] = 1UL } },
62#ifndef CONFIG_NUMA
63 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
64#ifdef CONFIG_HIGHMEM
65 [N_HIGH_MEMORY] = { { [0] = 1UL } },
66#endif
67 [N_CPU] = { { [0] = 1UL } },
68#endif /* NUMA */
69};
70EXPORT_SYMBOL(node_states);
71
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070072unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -070073unsigned long totalreserve_pages __read_mostly;
Hugh Dickins22b31ee2009-01-06 14:40:09 -080074unsigned long highest_memmap_pfn __read_mostly;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -080075int percpu_pagelist_fraction;
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +100076gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -070077
Mel Gormand9c23402007-10-16 01:26:01 -070078#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
79int pageblock_order __read_mostly;
80#endif
81
Hugh Dickinsd98c7a02006-02-14 13:52:59 -080082static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -080083
Linus Torvalds1da177e2005-04-16 15:20:36 -070084/*
85 * results with 256, 32 in the lowmem_reserve sysctl:
86 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
87 * 1G machine -> (16M dma, 784M normal, 224M high)
88 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
89 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
90 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +010091 *
92 * TBD: should special case ZONE_DMA32 machines here - in those we normally
93 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -070095int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -080096#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -070097 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -080098#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -070099#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700100 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700101#endif
Christoph Lametere53ef382006-09-25 23:31:14 -0700102#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700103 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700104#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700105 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700106};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107
108EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109
Helge Deller15ad7cd2006-12-06 20:40:36 -0800110static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800111#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700112 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800113#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700114#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700115 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700116#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700117 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700118#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700119 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700120#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700121 "Movable",
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700122};
123
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124int min_free_kbytes = 1024;
125
Yasunori Goto86356ab2006-06-23 02:03:09 -0700126unsigned long __meminitdata nr_kernel_pages;
127unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700128static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129
Mel Gormanc7132162006-09-27 01:49:43 -0700130#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
131 /*
Simon Arlott183ff222007-10-20 01:27:18 +0200132 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
Mel Gormanc7132162006-09-27 01:49:43 -0700133 * ranges of memory (RAM) that may be registered with add_active_range().
134 * Ranges passed to add_active_range() will be merged if possible
135 * so the number of times add_active_range() can be called is
136 * related to the number of nodes and the number of holes
137 */
138 #ifdef CONFIG_MAX_ACTIVE_REGIONS
139 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
140 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
141 #else
142 #if MAX_NUMNODES >= 32
143 /* If there can be many nodes, allow up to 50 holes per node */
144 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
145 #else
146 /* By default, allow up to 256 distinct regions */
147 #define MAX_ACTIVE_REGIONS 256
148 #endif
149 #endif
150
Jan Beulich98011f52007-07-15 23:38:17 -0700151 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
152 static int __meminitdata nr_nodemap_entries;
153 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
154 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
Adrian Bunkb69a7282008-07-23 21:28:12 -0700155 static unsigned long __initdata required_kernelcore;
Adrian Bunk484f51f2007-10-16 01:26:03 -0700156 static unsigned long __initdata required_movablecore;
Adrian Bunkb69a7282008-07-23 21:28:12 -0700157 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Mel Gorman2a1e2742007-07-17 04:03:12 -0700158
159 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
160 int movable_zone;
161 EXPORT_SYMBOL(movable_zone);
Mel Gormanc7132162006-09-27 01:49:43 -0700162#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
163
Miklos Szeredi418508c2007-05-23 13:57:55 -0700164#if MAX_NUMNODES > 1
165int nr_node_ids __read_mostly = MAX_NUMNODES;
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700166int nr_online_nodes __read_mostly = 1;
Miklos Szeredi418508c2007-05-23 13:57:55 -0700167EXPORT_SYMBOL(nr_node_ids);
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700168EXPORT_SYMBOL(nr_online_nodes);
Miklos Szeredi418508c2007-05-23 13:57:55 -0700169#endif
170
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700171int page_group_by_mobility_disabled __read_mostly;
172
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700173static void set_pageblock_migratetype(struct page *page, int migratetype)
174{
Mel Gorman49255c62009-06-16 15:31:58 -0700175
176 if (unlikely(page_group_by_mobility_disabled))
177 migratetype = MIGRATE_UNMOVABLE;
178
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700179 set_pageblock_flags_group(page, (unsigned long)migratetype,
180 PB_migrate, PB_migrate_end);
181}
182
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -0700183bool oom_killer_disabled __read_mostly;
184
Nick Piggin13e74442006-01-06 00:10:58 -0800185#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700186static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700188 int ret = 0;
189 unsigned seq;
190 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -0700191
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700192 do {
193 seq = zone_span_seqbegin(zone);
194 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
195 ret = 1;
196 else if (pfn < zone->zone_start_pfn)
197 ret = 1;
198 } while (zone_span_seqretry(zone, seq));
199
200 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700201}
202
203static int page_is_consistent(struct zone *zone, struct page *page)
204{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700205 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700206 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700208 return 0;
209
210 return 1;
211}
212/*
213 * Temporary debugging check for pages not lying within a given zone.
214 */
215static int bad_range(struct zone *zone, struct page *page)
216{
217 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700219 if (!page_is_consistent(zone, page))
220 return 1;
221
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 return 0;
223}
Nick Piggin13e74442006-01-06 00:10:58 -0800224#else
225static inline int bad_range(struct zone *zone, struct page *page)
226{
227 return 0;
228}
229#endif
230
Nick Piggin224abf92006-01-06 00:11:11 -0800231static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800233 static unsigned long resume;
234 static unsigned long nr_shown;
235 static unsigned long nr_unshown;
236
237 /*
238 * Allow a burst of 60 reports, then keep quiet for that minute;
239 * or allow a steady drip of one report per second.
240 */
241 if (nr_shown == 60) {
242 if (time_before(jiffies, resume)) {
243 nr_unshown++;
244 goto out;
245 }
246 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800247 printk(KERN_ALERT
248 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800249 nr_unshown);
250 nr_unshown = 0;
251 }
252 nr_shown = 0;
253 }
254 if (nr_shown++ == 0)
255 resume = jiffies + 60 * HZ;
256
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800257 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800258 current->comm, page_to_pfn(page));
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800259 printk(KERN_ALERT
Hugh Dickins3dc14742009-01-06 14:40:08 -0800260 "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n",
261 page, (void *)page->flags, page_count(page),
262 page_mapcount(page), page->mapping, page->index);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700263
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800265out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800266 /* Leave bad fields for debug, except PageBuddy could make trouble */
267 __ClearPageBuddy(page);
Randy Dunlap9f158332005-09-13 01:25:16 -0700268 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269}
270
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271/*
272 * Higher-order pages are called "compound pages". They are structured thusly:
273 *
274 * The first PAGE_SIZE page is called the "head page".
275 *
276 * The remaining PAGE_SIZE pages are called "tail pages".
277 *
278 * All pages have PG_compound set. All pages have their ->private pointing at
279 * the head page (even the head page has this).
280 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800281 * The first tail page's ->lru.next holds the address of the compound page's
282 * put_page() function. Its ->lru.prev holds the order of allocation.
283 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800285
286static void free_compound_page(struct page *page)
287{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700288 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800289}
290
Andi Kleen01ad1c02008-07-23 21:27:46 -0700291void prep_compound_page(struct page *page, unsigned long order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292{
293 int i;
294 int nr_pages = 1 << order;
295
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800296 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700297 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700298 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800299 for (i = 1; i < nr_pages; i++) {
300 struct page *p = page + i;
301
Christoph Lameterd85f3382007-05-06 14:49:39 -0700302 __SetPageTail(p);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700303 p->first_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 }
305}
306
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800307static int destroy_compound_page(struct page *page, unsigned long order)
Andy Whitcroft18229df2008-11-06 12:53:27 -0800308{
309 int i;
310 int nr_pages = 1 << order;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800311 int bad = 0;
Andy Whitcroft18229df2008-11-06 12:53:27 -0800312
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800313 if (unlikely(compound_order(page) != order) ||
314 unlikely(!PageHead(page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800315 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800316 bad++;
317 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318
Christoph Lameter6d777952007-05-06 14:49:40 -0700319 __ClearPageHead(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800320
Andy Whitcroft18229df2008-11-06 12:53:27 -0800321 for (i = 1; i < nr_pages; i++) {
322 struct page *p = page + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323
Alexey Zaytseve713a212009-01-10 02:47:57 +0300324 if (unlikely(!PageTail(p) || (p->first_page != page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800325 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800326 bad++;
327 }
Christoph Lameterd85f3382007-05-06 14:49:39 -0700328 __ClearPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800330
331 return bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333
Nick Piggin17cf4402006-03-22 00:08:41 -0800334static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
335{
336 int i;
337
Andrew Morton6626c5d2006-03-22 00:08:42 -0800338 /*
339 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
340 * and __GFP_HIGHMEM from hard or soft interrupt context.
341 */
Nick Piggin725d7042006-09-25 23:30:55 -0700342 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800343 for (i = 0; i < (1 << order); i++)
344 clear_highpage(page + i);
345}
346
Andrew Morton6aa30012006-04-18 22:20:52 -0700347static inline void set_page_order(struct page *page, int order)
348{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700349 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000350 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351}
352
353static inline void rmv_page_order(struct page *page)
354{
Nick Piggin676165a2006-04-10 11:21:48 +1000355 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700356 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357}
358
359/*
360 * Locate the struct page for both the matching buddy in our
361 * pair (buddy1) and the combined O(n+1) page they form (page).
362 *
363 * 1) Any buddy B1 will have an order O twin B2 which satisfies
364 * the following equation:
365 * B2 = B1 ^ (1 << O)
366 * For example, if the starting buddy (buddy2) is #8 its order
367 * 1 buddy is #10:
368 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
369 *
370 * 2) Any buddy B will have an order O+1 parent P which
371 * satisfies the following equation:
372 * P = B & ~(1 << O)
373 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200374 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 */
376static inline struct page *
377__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
378{
379 unsigned long buddy_idx = page_idx ^ (1 << order);
380
381 return page + (buddy_idx - page_idx);
382}
383
384static inline unsigned long
385__find_combined_index(unsigned long page_idx, unsigned int order)
386{
387 return (page_idx & ~(1 << order));
388}
389
390/*
391 * This function checks whether a page is free && is the buddy
392 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800393 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000394 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700395 * (c) a page and its buddy have the same order &&
396 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 *
Nick Piggin676165a2006-04-10 11:21:48 +1000398 * For recording whether a page is in the buddy system, we use PG_buddy.
399 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
400 *
401 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700403static inline int page_is_buddy(struct page *page, struct page *buddy,
404 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700406 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800407 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800408
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700409 if (page_zone_id(page) != page_zone_id(buddy))
410 return 0;
411
412 if (PageBuddy(buddy) && page_order(buddy) == order) {
Nick Piggina3af9c32009-06-16 15:32:10 -0700413 VM_BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa30012006-04-18 22:20:52 -0700414 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000415 }
Andrew Morton6aa30012006-04-18 22:20:52 -0700416 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417}
418
419/*
420 * Freeing function for a buddy system allocator.
421 *
422 * The concept of a buddy system is to maintain direct-mapped table
423 * (containing bit values) for memory blocks of various "orders".
424 * The bottom level table contains the map for the smallest allocatable
425 * units of memory (here, pages), and each level above it describes
426 * pairs of units from the levels below, hence, "buddies".
427 * At a high level, all that happens here is marking the table entry
428 * at the bottom level available, and propagating the changes upward
429 * as necessary, plus some accounting needed to play nicely with other
430 * parts of the VM system.
431 * At each level, we keep a list of pages, which are heads of continuous
Nick Piggin676165a2006-04-10 11:21:48 +1000432 * free pages of length of (1 << order) and marked with PG_buddy. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700433 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 * So when we are allocating or freeing one, we can derive the state of the
435 * other. That is, if we allocate a small block, and both were
436 * free, the remainder of the region must be split into blocks.
437 * If a block is freed, and its buddy is also free, then this
438 * triggers coalescing into a block of larger size.
439 *
440 * -- wli
441 */
442
Nick Piggin48db57f2006-01-08 01:00:42 -0800443static inline void __free_one_page(struct page *page,
Mel Gormaned0ae212009-06-16 15:32:07 -0700444 struct zone *zone, unsigned int order,
445 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446{
447 unsigned long page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448
Nick Piggin224abf92006-01-06 00:11:11 -0800449 if (unlikely(PageCompound(page)))
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800450 if (unlikely(destroy_compound_page(page, order)))
451 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452
Mel Gormaned0ae212009-06-16 15:32:07 -0700453 VM_BUG_ON(migratetype == -1);
454
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
456
Mel Gormanf2260e62009-06-16 15:32:13 -0700457 VM_BUG_ON(page_idx & ((1 << order) - 1));
Nick Piggin725d7042006-09-25 23:30:55 -0700458 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 while (order < MAX_ORDER-1) {
461 unsigned long combined_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 struct page *buddy;
463
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 buddy = __page_find_buddy(page, page_idx, order);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700465 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700466 break;
Nick Piggin13e74442006-01-06 00:10:58 -0800467
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700468 /* Our buddy is free, merge with it and move up one order. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 list_del(&buddy->lru);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700470 zone->free_area[order].nr_free--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 rmv_page_order(buddy);
Nick Piggin13e74442006-01-06 00:10:58 -0800472 combined_idx = __find_combined_index(page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 page = page + (combined_idx - page_idx);
474 page_idx = combined_idx;
475 order++;
476 }
477 set_page_order(page, order);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700478 list_add(&page->lru,
479 &zone->free_area[order].free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 zone->free_area[order].nr_free++;
481}
482
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700483#ifdef CONFIG_HAVE_MLOCKED_PAGE_BIT
484/*
485 * free_page_mlock() -- clean up attempts to free and mlocked() page.
486 * Page should not be on lru, so no need to fix that up.
487 * free_pages_check() will verify...
488 */
489static inline void free_page_mlock(struct page *page)
490{
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700491 __dec_zone_page_state(page, NR_MLOCK);
492 __count_vm_event(UNEVICTABLE_MLOCKFREED);
493}
494#else
495static void free_page_mlock(struct page *page) { }
496#endif
497
Nick Piggin224abf92006-01-06 00:11:11 -0800498static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499{
Nick Piggin92be2e32006-01-06 00:10:57 -0800500 if (unlikely(page_mapcount(page) |
501 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700502 (atomic_read(&page->_count) != 0) |
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800503 (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800504 bad_page(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800505 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800506 }
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800507 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
508 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
509 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510}
511
512/*
513 * Frees a list of pages.
514 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700515 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 *
517 * If the zone was previously in an "all pages pinned" state then look to
518 * see if this freeing clears that state.
519 *
520 * And clear the zone's pages_scanned counter, to hold off the "all pages are
521 * pinned" detection logic.
522 */
Nick Piggin48db57f2006-01-08 01:00:42 -0800523static void free_pages_bulk(struct zone *zone, int count,
524 struct list_head *list, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525{
Nick Pigginc54ad302006-01-06 00:10:56 -0800526 spin_lock(&zone->lock);
David Rientjese815af92007-10-16 23:25:54 -0700527 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700529
530 __mod_zone_page_state(zone, NR_FREE_PAGES, count << order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800531 while (count--) {
532 struct page *page;
533
Nick Piggin725d7042006-09-25 23:30:55 -0700534 VM_BUG_ON(list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 page = list_entry(list->prev, struct page, lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800536 /* have to delete it as __free_one_page list manipulates */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 list_del(&page->lru);
Mel Gormaned0ae212009-06-16 15:32:07 -0700538 __free_one_page(page, zone, order, page_private(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800540 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541}
542
Mel Gormaned0ae212009-06-16 15:32:07 -0700543static void free_one_page(struct zone *zone, struct page *page, int order,
544 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800545{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700546 spin_lock(&zone->lock);
David Rientjese815af92007-10-16 23:25:54 -0700547 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700548 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700549
550 __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
Mel Gormaned0ae212009-06-16 15:32:07 -0700551 __free_one_page(page, zone, order, migratetype);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700552 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800553}
554
555static void __free_pages_ok(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556{
Nick Pigginc54ad302006-01-06 00:10:56 -0800557 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 int i;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800559 int bad = 0;
Johannes Weinerc2773312009-06-19 19:30:56 +0200560 int wasMlocked = TestClearPageMlocked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100562 kmemcheck_free_shadow(page, order);
563
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 for (i = 0 ; i < (1 << order) ; ++i)
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800565 bad += free_pages_check(page + i);
566 if (bad)
Hugh Dickins689bceb2005-11-21 21:32:20 -0800567 return;
568
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700569 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -0700570 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700571 debug_check_no_obj_freed(page_address(page),
572 PAGE_SIZE << order);
573 }
Nick Piggindafb1362006-10-11 01:21:30 -0700574 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800575 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700576
Nick Pigginc54ad302006-01-06 00:10:56 -0800577 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +0200578 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -0700579 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700580 __count_vm_events(PGFREE, 1 << order);
Mel Gormaned0ae212009-06-16 15:32:07 -0700581 free_one_page(page_zone(page), page, order,
582 get_pageblock_migratetype(page));
Nick Pigginc54ad302006-01-06 00:10:56 -0800583 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584}
585
David Howellsa226f6c2006-01-06 00:11:08 -0800586/*
587 * permit the bootmem allocator to evade page validation on high-order frees
588 */
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700589void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800590{
591 if (order == 0) {
592 __ClearPageReserved(page);
593 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800594 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800595 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800596 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800597 int loop;
598
Nick Piggin545b1ea2006-03-22 00:08:07 -0800599 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800600 for (loop = 0; loop < BITS_PER_LONG; loop++) {
601 struct page *p = &page[loop];
602
Nick Piggin545b1ea2006-03-22 00:08:07 -0800603 if (loop + 1 < BITS_PER_LONG)
604 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800605 __ClearPageReserved(p);
606 set_page_count(p, 0);
607 }
608
Nick Piggin7835e982006-03-22 00:08:40 -0800609 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800610 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800611 }
612}
613
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614
615/*
616 * The order of subdivision here is critical for the IO subsystem.
617 * Please do not alter this order without good reasons and regression
618 * testing. Specifically, as large blocks of memory are subdivided,
619 * the order in which smaller blocks are delivered depends on the order
620 * they're subdivided in this function. This is the primary factor
621 * influencing the order in which pages are delivered to the IO
622 * subsystem according to empirical testing, and this is also justified
623 * by considering the behavior of a buddy system containing a single
624 * large block of memory acted on by a series of small allocations.
625 * This behavior is a critical factor in sglist merging's success.
626 *
627 * -- wli
628 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800629static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700630 int low, int high, struct free_area *area,
631 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632{
633 unsigned long size = 1 << high;
634
635 while (high > low) {
636 area--;
637 high--;
638 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700639 VM_BUG_ON(bad_range(zone, &page[size]));
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700640 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 area->nr_free++;
642 set_page_order(&page[size], high);
643 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644}
645
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646/*
647 * This page is about to be returned from the page allocator
648 */
Nick Piggin17cf4402006-03-22 00:08:41 -0800649static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650{
Nick Piggin92be2e32006-01-06 00:10:57 -0800651 if (unlikely(page_mapcount(page) |
652 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700653 (atomic_read(&page->_count) != 0) |
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800654 (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800655 bad_page(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800656 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800657 }
Hugh Dickins689bceb2005-11-21 21:32:20 -0800658
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700659 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800660 set_page_refcounted(page);
Nick Piggincc102502006-12-06 20:32:00 -0800661
662 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800664
665 if (gfp_flags & __GFP_ZERO)
666 prep_zero_page(page, order, gfp_flags);
667
668 if (order && (gfp_flags & __GFP_COMP))
669 prep_compound_page(page, order);
670
Hugh Dickins689bceb2005-11-21 21:32:20 -0800671 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672}
673
Mel Gorman56fd56b2007-10-16 01:25:58 -0700674/*
675 * Go through the free lists for the given migratetype and remove
676 * the smallest available page from the freelists
677 */
Mel Gorman728ec982009-06-16 15:32:04 -0700678static inline
679struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -0700680 int migratetype)
681{
682 unsigned int current_order;
683 struct free_area * area;
684 struct page *page;
685
686 /* Find a page of the appropriate size in the preferred list */
687 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
688 area = &(zone->free_area[current_order]);
689 if (list_empty(&area->free_list[migratetype]))
690 continue;
691
692 page = list_entry(area->free_list[migratetype].next,
693 struct page, lru);
694 list_del(&page->lru);
695 rmv_page_order(page);
696 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -0700697 expand(zone, page, order, current_order, area, migratetype);
698 return page;
699 }
700
701 return NULL;
702}
703
704
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700705/*
706 * This array describes the order lists are fallen back to when
707 * the free lists for the desirable migrate type are depleted
708 */
709static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
Mel Gorman64c5e132007-10-16 01:25:59 -0700710 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
711 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
712 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
713 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700714};
715
Mel Gormanc361be52007-10-16 01:25:51 -0700716/*
717 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -0700718 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -0700719 * boundary. If alignment is required, use move_freepages_block()
720 */
Adrian Bunkb69a7282008-07-23 21:28:12 -0700721static int move_freepages(struct zone *zone,
722 struct page *start_page, struct page *end_page,
723 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700724{
725 struct page *page;
726 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -0700727 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -0700728
729#ifndef CONFIG_HOLES_IN_ZONE
730 /*
731 * page_zone is not safe to call in this context when
732 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
733 * anyway as we check zone boundaries in move_freepages_block().
734 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -0700735 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -0700736 */
737 BUG_ON(page_zone(start_page) != page_zone(end_page));
738#endif
739
740 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -0700741 /* Make sure we are not inadvertently changing nodes */
742 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
743
Mel Gormanc361be52007-10-16 01:25:51 -0700744 if (!pfn_valid_within(page_to_pfn(page))) {
745 page++;
746 continue;
747 }
748
749 if (!PageBuddy(page)) {
750 page++;
751 continue;
752 }
753
754 order = page_order(page);
755 list_del(&page->lru);
756 list_add(&page->lru,
757 &zone->free_area[order].free_list[migratetype]);
758 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -0700759 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -0700760 }
761
Mel Gormand1003132007-10-16 01:26:00 -0700762 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -0700763}
764
Adrian Bunkb69a7282008-07-23 21:28:12 -0700765static int move_freepages_block(struct zone *zone, struct page *page,
766 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700767{
768 unsigned long start_pfn, end_pfn;
769 struct page *start_page, *end_page;
770
771 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -0700772 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -0700773 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -0700774 end_page = start_page + pageblock_nr_pages - 1;
775 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -0700776
777 /* Do not cross zone boundaries */
778 if (start_pfn < zone->zone_start_pfn)
779 start_page = page;
780 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
781 return 0;
782
783 return move_freepages(zone, start_page, end_page, migratetype);
784}
785
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700786/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -0700787static inline struct page *
788__rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700789{
790 struct free_area * area;
791 int current_order;
792 struct page *page;
793 int migratetype, i;
794
795 /* Find the largest possible block of pages in the other list */
796 for (current_order = MAX_ORDER-1; current_order >= order;
797 --current_order) {
798 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
799 migratetype = fallbacks[start_migratetype][i];
800
Mel Gorman56fd56b2007-10-16 01:25:58 -0700801 /* MIGRATE_RESERVE handled later if necessary */
802 if (migratetype == MIGRATE_RESERVE)
803 continue;
Mel Gormane0104872007-10-16 01:25:53 -0700804
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700805 area = &(zone->free_area[current_order]);
806 if (list_empty(&area->free_list[migratetype]))
807 continue;
808
809 page = list_entry(area->free_list[migratetype].next,
810 struct page, lru);
811 area->nr_free--;
812
813 /*
Mel Gormanc361be52007-10-16 01:25:51 -0700814 * If breaking a large block of pages, move all free
Mel Gorman46dafbc2007-10-16 01:25:55 -0700815 * pages to the preferred allocation list. If falling
816 * back for a reclaimable kernel allocation, be more
817 * agressive about taking ownership of free pages
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700818 */
Mel Gormand9c23402007-10-16 01:26:01 -0700819 if (unlikely(current_order >= (pageblock_order >> 1)) ||
Mel Gormandd5d2412009-09-05 11:17:11 -0700820 start_migratetype == MIGRATE_RECLAIMABLE ||
821 page_group_by_mobility_disabled) {
Mel Gorman46dafbc2007-10-16 01:25:55 -0700822 unsigned long pages;
823 pages = move_freepages_block(zone, page,
824 start_migratetype);
825
826 /* Claim the whole block if over half of it is free */
Mel Gormandd5d2412009-09-05 11:17:11 -0700827 if (pages >= (1 << (pageblock_order-1)) ||
828 page_group_by_mobility_disabled)
Mel Gorman46dafbc2007-10-16 01:25:55 -0700829 set_pageblock_migratetype(page,
830 start_migratetype);
831
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700832 migratetype = start_migratetype;
Mel Gormanc361be52007-10-16 01:25:51 -0700833 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700834
835 /* Remove the page from the freelists */
836 list_del(&page->lru);
837 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700838
Mel Gormand9c23402007-10-16 01:26:01 -0700839 if (current_order == pageblock_order)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700840 set_pageblock_migratetype(page,
841 start_migratetype);
842
843 expand(zone, page, order, current_order, area, migratetype);
844 return page;
845 }
846 }
847
Mel Gorman728ec982009-06-16 15:32:04 -0700848 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700849}
850
Mel Gorman56fd56b2007-10-16 01:25:58 -0700851/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 * Do the hard work of removing an element from the buddy allocator.
853 * Call me with the zone->lock already held.
854 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700855static struct page *__rmqueue(struct zone *zone, unsigned int order,
856 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 struct page *page;
859
Mel Gorman728ec982009-06-16 15:32:04 -0700860retry_reserve:
Mel Gorman56fd56b2007-10-16 01:25:58 -0700861 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862
Mel Gorman728ec982009-06-16 15:32:04 -0700863 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Mel Gorman56fd56b2007-10-16 01:25:58 -0700864 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700865
Mel Gorman728ec982009-06-16 15:32:04 -0700866 /*
867 * Use MIGRATE_RESERVE rather than fail an allocation. goto
868 * is used because __rmqueue_smallest is an inline function
869 * and we want just one call site
870 */
871 if (!page) {
872 migratetype = MIGRATE_RESERVE;
873 goto retry_reserve;
874 }
875 }
876
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700877 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878}
879
880/*
881 * Obtain a specified number of elements from the buddy allocator, all under
882 * a single hold of the lock, for efficiency. Add them to the supplied list.
883 * Returns the number of new pages which were placed at *list.
884 */
885static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700886 unsigned long count, struct list_head *list,
Mel Gormane084b2d2009-07-29 15:02:04 -0700887 int migratetype, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890
Nick Pigginc54ad302006-01-06 00:10:56 -0800891 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700893 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800894 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -0800896
897 /*
898 * Split buddy pages returned by expand() are received here
899 * in physical page order. The page is added to the callers and
900 * list and the list head then moves forward. From the callers
901 * perspective, the linked list is ordered by page number in
902 * some conditions. This is useful for IO devices that can
903 * merge IO requests if the physical pages are ordered
904 * properly.
905 */
Mel Gormane084b2d2009-07-29 15:02:04 -0700906 if (likely(cold == 0))
907 list_add(&page->lru, list);
908 else
909 list_add_tail(&page->lru, list);
Mel Gorman535131e2007-10-16 01:25:49 -0700910 set_page_private(page, migratetype);
Mel Gorman81eabcb2007-12-17 16:20:05 -0800911 list = &page->lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 }
Mel Gormanf2260e62009-06-16 15:32:13 -0700913 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -0800914 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800915 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916}
917
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700918#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800919/*
Christoph Lameter4037d452007-05-09 02:35:14 -0700920 * Called from the vmstat counter updater to drain pagesets of this
921 * currently executing processor on remote nodes after they have
922 * expired.
923 *
Christoph Lameter879336c2006-03-22 00:09:08 -0800924 * Note that this function must be called with the thread pinned to
925 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800926 */
Christoph Lameter4037d452007-05-09 02:35:14 -0700927void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700928{
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700929 unsigned long flags;
Christoph Lameter4037d452007-05-09 02:35:14 -0700930 int to_drain;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700931
Christoph Lameter4037d452007-05-09 02:35:14 -0700932 local_irq_save(flags);
933 if (pcp->count >= pcp->batch)
934 to_drain = pcp->batch;
935 else
936 to_drain = pcp->count;
937 free_pages_bulk(zone, to_drain, &pcp->list, 0);
938 pcp->count -= to_drain;
939 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700940}
941#endif
942
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800943/*
944 * Drain pages of the indicated processor.
945 *
946 * The processor must either be the current processor and the
947 * thread pinned to the current processor or a processor that
948 * is not online.
949 */
950static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951{
Nick Pigginc54ad302006-01-06 00:10:56 -0800952 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954
KOSAKI Motohiroee99c712009-03-31 15:19:31 -0700955 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 struct per_cpu_pageset *pset;
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800957 struct per_cpu_pages *pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700959 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800961 pcp = &pset->pcp;
962 local_irq_save(flags);
963 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
964 pcp->count = 0;
965 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 }
967}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800969/*
970 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
971 */
972void drain_local_pages(void *arg)
973{
974 drain_pages(smp_processor_id());
975}
976
977/*
978 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
979 */
980void drain_all_pages(void)
981{
Jens Axboe15c8b6c2008-05-09 09:39:44 +0200982 on_each_cpu(drain_local_pages, NULL, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800983}
984
Rafael J. Wysocki296699d2007-07-29 23:27:18 +0200985#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986
987void mark_free_pages(struct zone *zone)
988{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700989 unsigned long pfn, max_zone_pfn;
990 unsigned long flags;
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700991 int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 struct list_head *curr;
993
994 if (!zone->spanned_pages)
995 return;
996
997 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700998
999 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1000 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1001 if (pfn_valid(pfn)) {
1002 struct page *page = pfn_to_page(pfn);
1003
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001004 if (!swsusp_page_is_forbidden(page))
1005 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001006 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001008 for_each_migratetype_order(order, t) {
1009 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001010 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001012 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1013 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001014 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001015 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001016 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 spin_unlock_irqrestore(&zone->lock, flags);
1018}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001019#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020
1021/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 * Free a 0-order page
1023 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001024static void free_hot_cold_page(struct page *page, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025{
1026 struct zone *zone = page_zone(page);
1027 struct per_cpu_pages *pcp;
1028 unsigned long flags;
Johannes Weinerc2773312009-06-19 19:30:56 +02001029 int wasMlocked = TestClearPageMlocked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001031 kmemcheck_free_shadow(page, 0);
1032
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 if (PageAnon(page))
1034 page->mapping = NULL;
Nick Piggin224abf92006-01-06 00:11:11 -08001035 if (free_pages_check(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001036 return;
1037
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001038 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -07001039 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001040 debug_check_no_obj_freed(page_address(page), PAGE_SIZE);
1041 }
Nick Piggindafb1362006-10-11 01:21:30 -07001042 arch_free_page(page, 0);
Hugh Dickins689bceb2005-11-21 21:32:20 -08001043 kernel_map_pages(page, 1, 0);
1044
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001045 pcp = &zone_pcp(zone, get_cpu())->pcp;
Mel Gorman974709b2009-06-16 15:32:14 -07001046 set_page_private(page, get_pageblock_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +02001048 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -07001049 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001050 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001051
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001052 if (cold)
1053 list_add_tail(&page->lru, &pcp->list);
1054 else
1055 list_add(&page->lru, &pcp->list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001057 if (pcp->count >= pcp->high) {
1058 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
1059 pcp->count -= pcp->batch;
1060 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 local_irq_restore(flags);
1062 put_cpu();
1063}
1064
Harvey Harrison920c7a52008-02-04 22:29:26 -08001065void free_hot_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066{
1067 free_hot_cold_page(page, 0);
1068}
1069
Harvey Harrison920c7a52008-02-04 22:29:26 -08001070void free_cold_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071{
1072 free_hot_cold_page(page, 1);
1073}
1074
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001075/*
1076 * split_page takes a non-compound higher-order page, and splits it into
1077 * n (1<<order) sub-pages: page[0..n]
1078 * Each sub-page must be freed individually.
1079 *
1080 * Note: this is probably too low level an operation for use in drivers.
1081 * Please consult with lkml before using this in your driver.
1082 */
1083void split_page(struct page *page, unsigned int order)
1084{
1085 int i;
1086
Nick Piggin725d7042006-09-25 23:30:55 -07001087 VM_BUG_ON(PageCompound(page));
1088 VM_BUG_ON(!page_count(page));
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001089
1090#ifdef CONFIG_KMEMCHECK
1091 /*
1092 * Split shadow pages too, because free(page[0]) would
1093 * otherwise free the whole shadow.
1094 */
1095 if (kmemcheck_page_is_tracked(page))
1096 split_page(virt_to_page(page[0].shadow), order);
1097#endif
1098
Nick Piggin7835e982006-03-22 00:08:40 -08001099 for (i = 1; i < (1 << order); i++)
1100 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001101}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001102
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103/*
1104 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1105 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1106 * or two.
1107 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07001108static inline
1109struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001110 struct zone *zone, int order, gfp_t gfp_flags,
1111 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112{
1113 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001114 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 int cold = !!(gfp_flags & __GFP_COLD);
Nick Piggina74609f2006-01-06 00:11:20 -08001116 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117
Hugh Dickins689bceb2005-11-21 21:32:20 -08001118again:
Nick Piggina74609f2006-01-06 00:11:20 -08001119 cpu = get_cpu();
Nick Piggin48db57f2006-01-08 01:00:42 -08001120 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 struct per_cpu_pages *pcp;
1122
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001123 pcp = &zone_pcp(zone, cpu)->pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001125 if (!pcp->count) {
nkalmala941c7102006-11-02 22:07:04 -08001126 pcp->count = rmqueue_bulk(zone, 0,
Mel Gormane084b2d2009-07-29 15:02:04 -07001127 pcp->batch, &pcp->list,
1128 migratetype, cold);
Nick Piggina74609f2006-01-06 00:11:20 -08001129 if (unlikely(!pcp->count))
1130 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 }
Mel Gorman535131e2007-10-16 01:25:49 -07001132
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001133 /* Find a page of the appropriate migrate type */
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001134 if (cold) {
1135 list_for_each_entry_reverse(page, &pcp->list, lru)
1136 if (page_private(page) == migratetype)
1137 break;
1138 } else {
1139 list_for_each_entry(page, &pcp->list, lru)
1140 if (page_private(page) == migratetype)
1141 break;
1142 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001143
1144 /* Allocate more to the pcp list if necessary */
1145 if (unlikely(&page->lru == &pcp->list)) {
Mel Gorman535131e2007-10-16 01:25:49 -07001146 pcp->count += rmqueue_bulk(zone, 0,
Mel Gormane084b2d2009-07-29 15:02:04 -07001147 pcp->batch, &pcp->list,
1148 migratetype, cold);
Mel Gorman535131e2007-10-16 01:25:49 -07001149 page = list_entry(pcp->list.next, struct page, lru);
Mel Gorman535131e2007-10-16 01:25:49 -07001150 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001151
1152 list_del(&page->lru);
1153 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001154 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07001155 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1156 /*
1157 * __GFP_NOFAIL is not to be used in new code.
1158 *
1159 * All __GFP_NOFAIL callers should be fixed so that they
1160 * properly detect and handle allocation failures.
1161 *
1162 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07001163 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07001164 * __GFP_NOFAIL.
1165 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07001166 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07001167 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001169 page = __rmqueue(zone, order, migratetype);
Mel Gormanf2260e62009-06-16 15:32:13 -07001170 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order));
Nick Piggina74609f2006-01-06 00:11:20 -08001171 spin_unlock(&zone->lock);
1172 if (!page)
1173 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 }
1175
Christoph Lameterf8891e52006-06-30 01:55:45 -07001176 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Mel Gorman18ea7e72008-04-28 02:12:14 -07001177 zone_statistics(preferred_zone, zone);
Nick Piggina74609f2006-01-06 00:11:20 -08001178 local_irq_restore(flags);
1179 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180
Nick Piggin725d7042006-09-25 23:30:55 -07001181 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -08001182 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001183 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001185
1186failed:
1187 local_irq_restore(flags);
1188 put_cpu();
1189 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190}
1191
Mel Gorman41858962009-06-16 15:32:12 -07001192/* The ALLOC_WMARK bits are used as an index to zone->watermark */
1193#define ALLOC_WMARK_MIN WMARK_MIN
1194#define ALLOC_WMARK_LOW WMARK_LOW
1195#define ALLOC_WMARK_HIGH WMARK_HIGH
1196#define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1197
1198/* Mask to get the watermark bits */
1199#define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1200
Nick Piggin31488902005-11-28 13:44:03 -08001201#define ALLOC_HARDER 0x10 /* try to alloc harder */
1202#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1203#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001204
Akinobu Mita933e3122006-12-08 02:39:45 -08001205#ifdef CONFIG_FAIL_PAGE_ALLOC
1206
1207static struct fail_page_alloc_attr {
1208 struct fault_attr attr;
1209
1210 u32 ignore_gfp_highmem;
1211 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001212 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001213
1214#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1215
1216 struct dentry *ignore_gfp_highmem_file;
1217 struct dentry *ignore_gfp_wait_file;
Akinobu Mita54114992007-07-15 23:40:23 -07001218 struct dentry *min_order_file;
Akinobu Mita933e3122006-12-08 02:39:45 -08001219
1220#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1221
1222} fail_page_alloc = {
1223 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001224 .ignore_gfp_wait = 1,
1225 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001226 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001227};
1228
1229static int __init setup_fail_page_alloc(char *str)
1230{
1231 return setup_fault_attr(&fail_page_alloc.attr, str);
1232}
1233__setup("fail_page_alloc=", setup_fail_page_alloc);
1234
1235static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1236{
Akinobu Mita54114992007-07-15 23:40:23 -07001237 if (order < fail_page_alloc.min_order)
1238 return 0;
Akinobu Mita933e3122006-12-08 02:39:45 -08001239 if (gfp_mask & __GFP_NOFAIL)
1240 return 0;
1241 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1242 return 0;
1243 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1244 return 0;
1245
1246 return should_fail(&fail_page_alloc.attr, 1 << order);
1247}
1248
1249#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1250
1251static int __init fail_page_alloc_debugfs(void)
1252{
1253 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1254 struct dentry *dir;
1255 int err;
1256
1257 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1258 "fail_page_alloc");
1259 if (err)
1260 return err;
1261 dir = fail_page_alloc.attr.dentries.dir;
1262
1263 fail_page_alloc.ignore_gfp_wait_file =
1264 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1265 &fail_page_alloc.ignore_gfp_wait);
1266
1267 fail_page_alloc.ignore_gfp_highmem_file =
1268 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1269 &fail_page_alloc.ignore_gfp_highmem);
Akinobu Mita54114992007-07-15 23:40:23 -07001270 fail_page_alloc.min_order_file =
1271 debugfs_create_u32("min-order", mode, dir,
1272 &fail_page_alloc.min_order);
Akinobu Mita933e3122006-12-08 02:39:45 -08001273
1274 if (!fail_page_alloc.ignore_gfp_wait_file ||
Akinobu Mita54114992007-07-15 23:40:23 -07001275 !fail_page_alloc.ignore_gfp_highmem_file ||
1276 !fail_page_alloc.min_order_file) {
Akinobu Mita933e3122006-12-08 02:39:45 -08001277 err = -ENOMEM;
1278 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1279 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
Akinobu Mita54114992007-07-15 23:40:23 -07001280 debugfs_remove(fail_page_alloc.min_order_file);
Akinobu Mita933e3122006-12-08 02:39:45 -08001281 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1282 }
1283
1284 return err;
1285}
1286
1287late_initcall(fail_page_alloc_debugfs);
1288
1289#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1290
1291#else /* CONFIG_FAIL_PAGE_ALLOC */
1292
1293static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1294{
1295 return 0;
1296}
1297
1298#endif /* CONFIG_FAIL_PAGE_ALLOC */
1299
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300/*
1301 * Return 1 if free pages are above 'mark'. This takes into account the order
1302 * of the allocation.
1303 */
1304int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001305 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306{
1307 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001308 long min = mark;
1309 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 int o;
1311
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001312 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001314 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 min -= min / 4;
1316
1317 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1318 return 0;
1319 for (o = 0; o < order; o++) {
1320 /* At the next order, this order's pages become unavailable */
1321 free_pages -= z->free_area[o].nr_free << o;
1322
1323 /* Require fewer higher order pages to be free */
1324 min >>= 1;
1325
1326 if (free_pages <= min)
1327 return 0;
1328 }
1329 return 1;
1330}
1331
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001332#ifdef CONFIG_NUMA
1333/*
1334 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1335 * skip over zones that are not allowed by the cpuset, or that have
1336 * been recently (in last second) found to be nearly full. See further
1337 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001338 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001339 *
1340 * If the zonelist cache is present in the passed in zonelist, then
1341 * returns a pointer to the allowed node mask (either the current
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001342 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001343 *
1344 * If the zonelist cache is not available for this zonelist, does
1345 * nothing and returns NULL.
1346 *
1347 * If the fullzones BITMAP in the zonelist cache is stale (more than
1348 * a second since last zap'd) then we zap it out (clear its bits.)
1349 *
1350 * We hold off even calling zlc_setup, until after we've checked the
1351 * first zone in the zonelist, on the theory that most allocations will
1352 * be satisfied from that first zone, so best to examine that zone as
1353 * quickly as we can.
1354 */
1355static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1356{
1357 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1358 nodemask_t *allowednodes; /* zonelist_cache approximation */
1359
1360 zlc = zonelist->zlcache_ptr;
1361 if (!zlc)
1362 return NULL;
1363
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001364 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001365 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1366 zlc->last_full_zap = jiffies;
1367 }
1368
1369 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1370 &cpuset_current_mems_allowed :
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001371 &node_states[N_HIGH_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001372 return allowednodes;
1373}
1374
1375/*
1376 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1377 * if it is worth looking at further for free memory:
1378 * 1) Check that the zone isn't thought to be full (doesn't have its
1379 * bit set in the zonelist_cache fullzones BITMAP).
1380 * 2) Check that the zones node (obtained from the zonelist_cache
1381 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1382 * Return true (non-zero) if zone is worth looking at further, or
1383 * else return false (zero) if it is not.
1384 *
1385 * This check -ignores- the distinction between various watermarks,
1386 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1387 * found to be full for any variation of these watermarks, it will
1388 * be considered full for up to one second by all requests, unless
1389 * we are so low on memory on all allowed nodes that we are forced
1390 * into the second scan of the zonelist.
1391 *
1392 * In the second scan we ignore this zonelist cache and exactly
1393 * apply the watermarks to all zones, even it is slower to do so.
1394 * We are low on memory in the second scan, and should leave no stone
1395 * unturned looking for a free page.
1396 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001397static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001398 nodemask_t *allowednodes)
1399{
1400 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1401 int i; /* index of *z in zonelist zones */
1402 int n; /* node that zone *z is on */
1403
1404 zlc = zonelist->zlcache_ptr;
1405 if (!zlc)
1406 return 1;
1407
Mel Gormandd1a2392008-04-28 02:12:17 -07001408 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001409 n = zlc->z_to_n[i];
1410
1411 /* This zone is worth trying if it is allowed but not full */
1412 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1413}
1414
1415/*
1416 * Given 'z' scanning a zonelist, set the corresponding bit in
1417 * zlc->fullzones, so that subsequent attempts to allocate a page
1418 * from that zone don't waste time re-examining it.
1419 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001420static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001421{
1422 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1423 int i; /* index of *z in zonelist zones */
1424
1425 zlc = zonelist->zlcache_ptr;
1426 if (!zlc)
1427 return;
1428
Mel Gormandd1a2392008-04-28 02:12:17 -07001429 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001430
1431 set_bit(i, zlc->fullzones);
1432}
1433
1434#else /* CONFIG_NUMA */
1435
1436static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1437{
1438 return NULL;
1439}
1440
Mel Gormandd1a2392008-04-28 02:12:17 -07001441static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001442 nodemask_t *allowednodes)
1443{
1444 return 1;
1445}
1446
Mel Gormandd1a2392008-04-28 02:12:17 -07001447static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001448{
1449}
1450#endif /* CONFIG_NUMA */
1451
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001452/*
Paul Jackson0798e512006-12-06 20:31:38 -08001453 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001454 * a page.
1455 */
1456static struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001457get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
Mel Gorman5117f452009-06-16 15:31:59 -07001458 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
Mel Gorman3dd28262009-06-16 15:32:00 -07001459 struct zone *preferred_zone, int migratetype)
Martin Hicks753ee722005-06-21 17:14:41 -07001460{
Mel Gormandd1a2392008-04-28 02:12:17 -07001461 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001462 struct page *page = NULL;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001463 int classzone_idx;
Mel Gorman5117f452009-06-16 15:31:59 -07001464 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001465 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1466 int zlc_active = 0; /* set if using zonelist_cache */
1467 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gorman54a6eb52008-04-28 02:12:16 -07001468
Mel Gorman19770b32008-04-28 02:12:18 -07001469 classzone_idx = zone_idx(preferred_zone);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001470zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001471 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001472 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001473 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1474 */
Mel Gorman19770b32008-04-28 02:12:18 -07001475 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1476 high_zoneidx, nodemask) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001477 if (NUMA_BUILD && zlc_active &&
1478 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1479 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001480 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001481 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001482 goto try_next_zone;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001483
Mel Gorman41858962009-06-16 15:32:12 -07001484 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001485 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001486 unsigned long mark;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001487 int ret;
1488
Mel Gorman41858962009-06-16 15:32:12 -07001489 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
Mel Gormanfa5e0842009-06-16 15:33:22 -07001490 if (zone_watermark_ok(zone, order, mark,
1491 classzone_idx, alloc_flags))
1492 goto try_this_zone;
1493
1494 if (zone_reclaim_mode == 0)
1495 goto this_zone_full;
1496
1497 ret = zone_reclaim(zone, gfp_mask, order);
1498 switch (ret) {
1499 case ZONE_RECLAIM_NOSCAN:
1500 /* did not scan */
1501 goto try_next_zone;
1502 case ZONE_RECLAIM_FULL:
1503 /* scanned but unreclaimable */
1504 goto this_zone_full;
1505 default:
1506 /* did we reclaim enough */
1507 if (!zone_watermark_ok(zone, order, mark,
1508 classzone_idx, alloc_flags))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001509 goto this_zone_full;
Paul Jackson0798e512006-12-06 20:31:38 -08001510 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001511 }
1512
Mel Gormanfa5e0842009-06-16 15:33:22 -07001513try_this_zone:
Mel Gorman3dd28262009-06-16 15:32:00 -07001514 page = buffered_rmqueue(preferred_zone, zone, order,
1515 gfp_mask, migratetype);
Paul Jackson0798e512006-12-06 20:31:38 -08001516 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001517 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001518this_zone_full:
1519 if (NUMA_BUILD)
1520 zlc_mark_zone_full(zonelist, z);
1521try_next_zone:
Christoph Lameter62bc62a2009-06-16 15:32:15 -07001522 if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) {
Mel Gormand395b732009-06-16 15:32:09 -07001523 /*
1524 * we do zlc_setup after the first zone is tried but only
1525 * if there are multiple nodes make it worthwhile
1526 */
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001527 allowednodes = zlc_setup(zonelist, alloc_flags);
1528 zlc_active = 1;
1529 did_zlc_setup = 1;
1530 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07001531 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001532
1533 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1534 /* Disable zlc cache for second zonelist scan */
1535 zlc_active = 0;
1536 goto zonelist_scan;
1537 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001538 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07001539}
1540
Mel Gorman11e33f62009-06-16 15:31:57 -07001541static inline int
1542should_alloc_retry(gfp_t gfp_mask, unsigned int order,
1543 unsigned long pages_reclaimed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544{
Mel Gorman11e33f62009-06-16 15:31:57 -07001545 /* Do not loop if specifically requested */
1546 if (gfp_mask & __GFP_NORETRY)
1547 return 0;
1548
1549 /*
1550 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1551 * means __GFP_NOFAIL, but that may not be true in other
1552 * implementations.
1553 */
1554 if (order <= PAGE_ALLOC_COSTLY_ORDER)
1555 return 1;
1556
1557 /*
1558 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1559 * specified, then we retry until we no longer reclaim any pages
1560 * (above), or we've reclaimed an order of pages at least as
1561 * large as the allocation's order. In both cases, if the
1562 * allocation still fails, we stop retrying.
1563 */
1564 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
1565 return 1;
1566
1567 /*
1568 * Don't let big-order allocations loop unless the caller
1569 * explicitly requests that.
1570 */
1571 if (gfp_mask & __GFP_NOFAIL)
1572 return 1;
1573
1574 return 0;
1575}
1576
1577static inline struct page *
1578__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1579 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001580 nodemask_t *nodemask, struct zone *preferred_zone,
1581 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001582{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584
Mel Gorman11e33f62009-06-16 15:31:57 -07001585 /* Acquire the OOM killer lock for the zones in zonelist */
1586 if (!try_set_zone_oom(zonelist, gfp_mask)) {
1587 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 return NULL;
1589 }
Jens Axboe6b1de912005-11-17 21:35:02 +01001590
Mel Gorman11e33f62009-06-16 15:31:57 -07001591 /*
1592 * Go through the zonelist yet one more time, keep very high watermark
1593 * here, this is only to catch a parallel oom killing, we must fail if
1594 * we're still under heavy pressure.
1595 */
1596 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1597 order, zonelist, high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001598 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07001599 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001600 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07001601 goto out;
1602
1603 /* The OOM killer will not help higher order allocs */
David Rientjes82553a92009-06-16 15:32:58 -07001604 if (order > PAGE_ALLOC_COSTLY_ORDER && !(gfp_mask & __GFP_NOFAIL))
Mel Gorman11e33f62009-06-16 15:31:57 -07001605 goto out;
1606
1607 /* Exhausted what can be done so it's blamo time */
1608 out_of_memory(zonelist, gfp_mask, order);
1609
1610out:
1611 clear_zonelist_oom(zonelist, gfp_mask);
1612 return page;
1613}
1614
1615/* The really slow allocator path where we enter direct reclaim */
1616static inline struct page *
1617__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
1618 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001619 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001620 int migratetype, unsigned long *did_some_progress)
Mel Gorman11e33f62009-06-16 15:31:57 -07001621{
1622 struct page *page = NULL;
1623 struct reclaim_state reclaim_state;
1624 struct task_struct *p = current;
1625
1626 cond_resched();
1627
1628 /* We now go into synchronous reclaim */
1629 cpuset_memory_pressure_bump();
1630
1631 /*
1632 * The task's cpuset might have expanded its set of allowable nodes
1633 */
1634 p->flags |= PF_MEMALLOC;
1635 lockdep_set_current_reclaim_state(gfp_mask);
1636 reclaim_state.reclaimed_slab = 0;
1637 p->reclaim_state = &reclaim_state;
1638
1639 *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
1640
1641 p->reclaim_state = NULL;
1642 lockdep_clear_current_reclaim_state();
1643 p->flags &= ~PF_MEMALLOC;
1644
1645 cond_resched();
1646
1647 if (order != 0)
1648 drain_all_pages();
1649
1650 if (likely(*did_some_progress))
1651 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001652 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001653 alloc_flags, preferred_zone,
1654 migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001655 return page;
1656}
1657
Mel Gorman11e33f62009-06-16 15:31:57 -07001658/*
1659 * This is called in the allocator slow-path if the allocation request is of
1660 * sufficient urgency to ignore watermarks and take other desperate measures
1661 */
1662static inline struct page *
1663__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
1664 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001665 nodemask_t *nodemask, struct zone *preferred_zone,
1666 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001667{
1668 struct page *page;
1669
1670 do {
1671 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001672 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
Mel Gorman3dd28262009-06-16 15:32:00 -07001673 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001674
1675 if (!page && gfp_mask & __GFP_NOFAIL)
Jens Axboe8aa7e842009-07-09 14:52:32 +02001676 congestion_wait(BLK_RW_ASYNC, HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07001677 } while (!page && (gfp_mask & __GFP_NOFAIL));
1678
1679 return page;
1680}
1681
1682static inline
1683void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
1684 enum zone_type high_zoneidx)
1685{
1686 struct zoneref *z;
1687 struct zone *zone;
1688
1689 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
1690 wakeup_kswapd(zone, order);
1691}
1692
Peter Zijlstra341ce062009-06-16 15:32:02 -07001693static inline int
1694gfp_to_alloc_flags(gfp_t gfp_mask)
1695{
1696 struct task_struct *p = current;
1697 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
1698 const gfp_t wait = gfp_mask & __GFP_WAIT;
1699
Mel Gormana56f57f2009-06-16 15:32:02 -07001700 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
1701 BUILD_BUG_ON(__GFP_HIGH != ALLOC_HIGH);
1702
Peter Zijlstra341ce062009-06-16 15:32:02 -07001703 /*
1704 * The caller may dip into page reserves a bit more if the caller
1705 * cannot run direct reclaim, or if the caller has realtime scheduling
1706 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1707 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1708 */
Mel Gormana56f57f2009-06-16 15:32:02 -07001709 alloc_flags |= (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07001710
1711 if (!wait) {
1712 alloc_flags |= ALLOC_HARDER;
1713 /*
1714 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1715 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1716 */
1717 alloc_flags &= ~ALLOC_CPUSET;
1718 } else if (unlikely(rt_task(p)))
1719 alloc_flags |= ALLOC_HARDER;
1720
1721 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
1722 if (!in_interrupt() &&
1723 ((p->flags & PF_MEMALLOC) ||
1724 unlikely(test_thread_flag(TIF_MEMDIE))))
1725 alloc_flags |= ALLOC_NO_WATERMARKS;
1726 }
1727
1728 return alloc_flags;
1729}
1730
Mel Gorman11e33f62009-06-16 15:31:57 -07001731static inline struct page *
1732__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
1733 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001734 nodemask_t *nodemask, struct zone *preferred_zone,
1735 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001736{
1737 const gfp_t wait = gfp_mask & __GFP_WAIT;
1738 struct page *page = NULL;
1739 int alloc_flags;
1740 unsigned long pages_reclaimed = 0;
1741 unsigned long did_some_progress;
1742 struct task_struct *p = current;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001743
Christoph Lameter952f3b52006-12-06 20:33:26 -08001744 /*
Mel Gorman72807a72009-06-16 15:32:18 -07001745 * In the slowpath, we sanity check order to avoid ever trying to
1746 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
1747 * be using allocators in order of preference for an area that is
1748 * too large.
1749 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07001750 if (order >= MAX_ORDER) {
1751 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07001752 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07001753 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001754
Christoph Lameter952f3b52006-12-06 20:33:26 -08001755 /*
1756 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1757 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1758 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1759 * using a larger set of nodes after it has established that the
1760 * allowed per node queues are empty and that nodes are
1761 * over allocated.
1762 */
1763 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1764 goto nopage;
1765
Mel Gorman11e33f62009-06-16 15:31:57 -07001766 wake_all_kswapd(order, zonelist, high_zoneidx);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001767
Paul Jackson9bf22292005-09-06 15:18:12 -07001768 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001769 * OK, we're below the kswapd watermark and have kicked background
1770 * reclaim. Now things get more complex, so set up alloc_flags according
1771 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07001772 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07001773 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774
Mel Gorman11e33f62009-06-16 15:31:57 -07001775restart:
Peter Zijlstra341ce062009-06-16 15:32:02 -07001776 /* This is the last chance, in general, before the goto nopage. */
Mel Gorman19770b32008-04-28 02:12:18 -07001777 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
Peter Zijlstra341ce062009-06-16 15:32:02 -07001778 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
1779 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001780 if (page)
1781 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782
Kirill Korotaevb43a57b2006-12-06 20:32:27 -08001783rebalance:
Mel Gorman11e33f62009-06-16 15:31:57 -07001784 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07001785 if (alloc_flags & ALLOC_NO_WATERMARKS) {
1786 page = __alloc_pages_high_priority(gfp_mask, order,
1787 zonelist, high_zoneidx, nodemask,
1788 preferred_zone, migratetype);
1789 if (page)
1790 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 }
1792
1793 /* Atomic allocations - we can't balance anything */
1794 if (!wait)
1795 goto nopage;
1796
Peter Zijlstra341ce062009-06-16 15:32:02 -07001797 /* Avoid recursion of direct reclaim */
1798 if (p->flags & PF_MEMALLOC)
1799 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800
David Rientjes6583bb62009-07-29 15:02:06 -07001801 /* Avoid allocations with no watermarks from looping endlessly */
1802 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
1803 goto nopage;
1804
Mel Gorman11e33f62009-06-16 15:31:57 -07001805 /* Try direct reclaim and then allocating */
1806 page = __alloc_pages_direct_reclaim(gfp_mask, order,
1807 zonelist, high_zoneidx,
1808 nodemask,
Mel Gorman5117f452009-06-16 15:31:59 -07001809 alloc_flags, preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001810 migratetype, &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07001811 if (page)
1812 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 /*
Mel Gorman11e33f62009-06-16 15:31:57 -07001815 * If we failed to make any progress reclaiming, then we are
1816 * running out of options and have to consider going OOM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 */
Mel Gorman11e33f62009-06-16 15:31:57 -07001818 if (!did_some_progress) {
1819 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -07001820 if (oom_killer_disabled)
1821 goto nopage;
Mel Gorman11e33f62009-06-16 15:31:57 -07001822 page = __alloc_pages_may_oom(gfp_mask, order,
1823 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001824 nodemask, preferred_zone,
1825 migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001826 if (page)
1827 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828
Mel Gorman11e33f62009-06-16 15:31:57 -07001829 /*
David Rientjes82553a92009-06-16 15:32:58 -07001830 * The OOM killer does not trigger for high-order
1831 * ~__GFP_NOFAIL allocations so if no progress is being
1832 * made, there are no other options and retrying is
1833 * unlikely to help.
Mel Gorman11e33f62009-06-16 15:31:57 -07001834 */
David Rientjes82553a92009-06-16 15:32:58 -07001835 if (order > PAGE_ALLOC_COSTLY_ORDER &&
1836 !(gfp_mask & __GFP_NOFAIL))
Mel Gorman11e33f62009-06-16 15:31:57 -07001837 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 goto restart;
1840 }
1841 }
1842
Mel Gorman11e33f62009-06-16 15:31:57 -07001843 /* Check if we should retry the allocation */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07001844 pages_reclaimed += did_some_progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07001845 if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
1846 /* Wait for some write requests to complete then retry */
Jens Axboe8aa7e842009-07-09 14:52:32 +02001847 congestion_wait(BLK_RW_ASYNC, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 goto rebalance;
1849 }
1850
1851nopage:
1852 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1853 printk(KERN_WARNING "%s: page allocation failure."
1854 " order:%d, mode:0x%x\n",
1855 p->comm, order, gfp_mask);
1856 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001857 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 }
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001859 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860got_pg:
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001861 if (kmemcheck_enabled)
1862 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 return page;
Mel Gorman11e33f62009-06-16 15:31:57 -07001864
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865}
Mel Gorman11e33f62009-06-16 15:31:57 -07001866
1867/*
1868 * This is the 'heart' of the zoned buddy allocator.
1869 */
1870struct page *
1871__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
1872 struct zonelist *zonelist, nodemask_t *nodemask)
1873{
1874 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Mel Gorman5117f452009-06-16 15:31:59 -07001875 struct zone *preferred_zone;
Mel Gorman11e33f62009-06-16 15:31:57 -07001876 struct page *page;
Mel Gorman3dd28262009-06-16 15:32:00 -07001877 int migratetype = allocflags_to_migratetype(gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07001878
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10001879 gfp_mask &= gfp_allowed_mask;
1880
Mel Gorman11e33f62009-06-16 15:31:57 -07001881 lockdep_trace_alloc(gfp_mask);
1882
1883 might_sleep_if(gfp_mask & __GFP_WAIT);
1884
1885 if (should_fail_alloc_page(gfp_mask, order))
1886 return NULL;
1887
1888 /*
1889 * Check the zones suitable for the gfp_mask contain at least one
1890 * valid zone. It's possible to have an empty zonelist as a result
1891 * of GFP_THISNODE and a memoryless node
1892 */
1893 if (unlikely(!zonelist->_zonerefs->zone))
1894 return NULL;
1895
Mel Gorman5117f452009-06-16 15:31:59 -07001896 /* The preferred zone is used for statistics later */
1897 first_zones_zonelist(zonelist, high_zoneidx, nodemask, &preferred_zone);
1898 if (!preferred_zone)
1899 return NULL;
1900
1901 /* First allocation attempt */
Mel Gorman11e33f62009-06-16 15:31:57 -07001902 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001903 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07001904 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001905 if (unlikely(!page))
1906 page = __alloc_pages_slowpath(gfp_mask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001907 zonelist, high_zoneidx, nodemask,
Mel Gorman3dd28262009-06-16 15:32:00 -07001908 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001909
1910 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911}
Mel Gormand2391712009-06-16 15:31:52 -07001912EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913
1914/*
1915 * Common helper functions.
1916 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001917unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918{
1919 struct page * page;
1920 page = alloc_pages(gfp_mask, order);
1921 if (!page)
1922 return 0;
1923 return (unsigned long) page_address(page);
1924}
1925
1926EXPORT_SYMBOL(__get_free_pages);
1927
Harvey Harrison920c7a52008-02-04 22:29:26 -08001928unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929{
1930 struct page * page;
1931
1932 /*
1933 * get_zeroed_page() returns a 32-bit address, which cannot represent
1934 * a highmem page
1935 */
Nick Piggin725d7042006-09-25 23:30:55 -07001936 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937
1938 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1939 if (page)
1940 return (unsigned long) page_address(page);
1941 return 0;
1942}
1943
1944EXPORT_SYMBOL(get_zeroed_page);
1945
1946void __pagevec_free(struct pagevec *pvec)
1947{
1948 int i = pagevec_count(pvec);
1949
1950 while (--i >= 0)
1951 free_hot_cold_page(pvec->pages[i], pvec->cold);
1952}
1953
Harvey Harrison920c7a52008-02-04 22:29:26 -08001954void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955{
Nick Pigginb5810032005-10-29 18:16:12 -07001956 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 if (order == 0)
1958 free_hot_page(page);
1959 else
1960 __free_pages_ok(page, order);
1961 }
1962}
1963
1964EXPORT_SYMBOL(__free_pages);
1965
Harvey Harrison920c7a52008-02-04 22:29:26 -08001966void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967{
1968 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07001969 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 __free_pages(virt_to_page((void *)addr), order);
1971 }
1972}
1973
1974EXPORT_SYMBOL(free_pages);
1975
Timur Tabi2be0ffe2008-07-23 21:28:11 -07001976/**
1977 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1978 * @size: the number of bytes to allocate
1979 * @gfp_mask: GFP flags for the allocation
1980 *
1981 * This function is similar to alloc_pages(), except that it allocates the
1982 * minimum number of pages to satisfy the request. alloc_pages() can only
1983 * allocate memory in power-of-two pages.
1984 *
1985 * This function is also limited by MAX_ORDER.
1986 *
1987 * Memory allocated by this function must be released by free_pages_exact().
1988 */
1989void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
1990{
1991 unsigned int order = get_order(size);
1992 unsigned long addr;
1993
1994 addr = __get_free_pages(gfp_mask, order);
1995 if (addr) {
1996 unsigned long alloc_end = addr + (PAGE_SIZE << order);
1997 unsigned long used = addr + PAGE_ALIGN(size);
1998
Kevin Cernekee5bfd7562009-07-05 12:08:19 -07001999 split_page(virt_to_page((void *)addr), order);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002000 while (used < alloc_end) {
2001 free_page(used);
2002 used += PAGE_SIZE;
2003 }
2004 }
2005
2006 return (void *)addr;
2007}
2008EXPORT_SYMBOL(alloc_pages_exact);
2009
2010/**
2011 * free_pages_exact - release memory allocated via alloc_pages_exact()
2012 * @virt: the value returned by alloc_pages_exact.
2013 * @size: size of allocation, same value as passed to alloc_pages_exact().
2014 *
2015 * Release the memory allocated by a previous call to alloc_pages_exact.
2016 */
2017void free_pages_exact(void *virt, size_t size)
2018{
2019 unsigned long addr = (unsigned long)virt;
2020 unsigned long end = addr + PAGE_ALIGN(size);
2021
2022 while (addr < end) {
2023 free_page(addr);
2024 addr += PAGE_SIZE;
2025 }
2026}
2027EXPORT_SYMBOL(free_pages_exact);
2028
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029static unsigned int nr_free_zone_pages(int offset)
2030{
Mel Gormandd1a2392008-04-28 02:12:17 -07002031 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002032 struct zone *zone;
2033
Martin J. Blighe310fd42005-07-29 22:59:18 -07002034 /* Just pick one node, since fallback list is circular */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 unsigned int sum = 0;
2036
Mel Gorman0e884602008-04-28 02:12:14 -07002037 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038
Mel Gorman54a6eb52008-04-28 02:12:16 -07002039 for_each_zone_zonelist(zone, z, zonelist, offset) {
Martin J. Blighe310fd42005-07-29 22:59:18 -07002040 unsigned long size = zone->present_pages;
Mel Gorman41858962009-06-16 15:32:12 -07002041 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07002042 if (size > high)
2043 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 }
2045
2046 return sum;
2047}
2048
2049/*
2050 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2051 */
2052unsigned int nr_free_buffer_pages(void)
2053{
Al Viroaf4ca452005-10-21 02:55:38 -04002054 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055}
Meelap Shahc2f1a552007-07-17 04:04:39 -07002056EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057
2058/*
2059 * Amount of free RAM allocatable within all zones
2060 */
2061unsigned int nr_free_pagecache_pages(void)
2062{
Mel Gorman2a1e2742007-07-17 04:03:12 -07002063 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002065
2066static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067{
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002068 if (NUMA_BUILD)
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08002069 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072void si_meminfo(struct sysinfo *val)
2073{
2074 val->totalram = totalram_pages;
2075 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002076 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 val->totalhigh = totalhigh_pages;
2079 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 val->mem_unit = PAGE_SIZE;
2081}
2082
2083EXPORT_SYMBOL(si_meminfo);
2084
2085#ifdef CONFIG_NUMA
2086void si_meminfo_node(struct sysinfo *val, int nid)
2087{
2088 pg_data_t *pgdat = NODE_DATA(nid);
2089
2090 val->totalram = pgdat->node_present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002091 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002092#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002094 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2095 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002096#else
2097 val->totalhigh = 0;
2098 val->freehigh = 0;
2099#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 val->mem_unit = PAGE_SIZE;
2101}
2102#endif
2103
2104#define K(x) ((x) << (PAGE_SHIFT-10))
2105
2106/*
2107 * Show free area list (used inside shift_scroll-lock stuff)
2108 * We also calculate the percentage fragmentation. We do this by counting the
2109 * memory on each free list with the exception of the first item on the list.
2110 */
2111void show_free_areas(void)
2112{
Jes Sorensenc7241912006-09-27 01:50:05 -07002113 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 struct zone *zone;
2115
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002116 for_each_populated_zone(zone) {
Jes Sorensenc7241912006-09-27 01:50:05 -07002117 show_node(zone);
2118 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119
Dave Jones6b482c62005-11-10 15:45:56 -05002120 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 struct per_cpu_pageset *pageset;
2122
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002123 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002125 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2126 cpu, pageset->pcp.high,
2127 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 }
2129 }
2130
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002131 printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n"
2132 " inactive_file:%lu"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002133 " unevictable:%lu"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002134 " dirty:%lu writeback:%lu unstable:%lu\n"
Christoph Lameterd23ad422007-02-10 01:43:02 -08002135 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002136 global_page_state(NR_ACTIVE_ANON),
2137 global_page_state(NR_ACTIVE_FILE),
2138 global_page_state(NR_INACTIVE_ANON),
2139 global_page_state(NR_INACTIVE_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002140 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07002141 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07002142 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07002143 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08002144 global_page_state(NR_FREE_PAGES),
Christoph Lameter972d1a72006-09-25 23:31:51 -07002145 global_page_state(NR_SLAB_RECLAIMABLE) +
2146 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07002147 global_page_state(NR_FILE_MAPPED),
Andrew Mortona25700a2007-02-08 14:20:40 -08002148 global_page_state(NR_PAGETABLE),
2149 global_page_state(NR_BOUNCE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002151 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 int i;
2153
2154 show_node(zone);
2155 printk("%s"
2156 " free:%lukB"
2157 " min:%lukB"
2158 " low:%lukB"
2159 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002160 " active_anon:%lukB"
2161 " inactive_anon:%lukB"
2162 " active_file:%lukB"
2163 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002164 " unevictable:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 " present:%lukB"
2166 " pages_scanned:%lu"
2167 " all_unreclaimable? %s"
2168 "\n",
2169 zone->name,
Christoph Lameterd23ad422007-02-10 01:43:02 -08002170 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07002171 K(min_wmark_pages(zone)),
2172 K(low_wmark_pages(zone)),
2173 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002174 K(zone_page_state(zone, NR_ACTIVE_ANON)),
2175 K(zone_page_state(zone, NR_INACTIVE_ANON)),
2176 K(zone_page_state(zone, NR_ACTIVE_FILE)),
2177 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002178 K(zone_page_state(zone, NR_UNEVICTABLE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 K(zone->present_pages),
2180 zone->pages_scanned,
David Rientjese815af92007-10-16 23:25:54 -07002181 (zone_is_all_unreclaimable(zone) ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 );
2183 printk("lowmem_reserve[]:");
2184 for (i = 0; i < MAX_NR_ZONES; i++)
2185 printk(" %lu", zone->lowmem_reserve[i]);
2186 printk("\n");
2187 }
2188
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002189 for_each_populated_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002190 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191
2192 show_node(zone);
2193 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194
2195 spin_lock_irqsave(&zone->lock, flags);
2196 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002197 nr[order] = zone->free_area[order].nr_free;
2198 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 }
2200 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002201 for (order = 0; order < MAX_ORDER; order++)
2202 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 printk("= %lukB\n", K(total));
2204 }
2205
Larry Woodmane6f36022008-02-04 22:29:30 -08002206 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2207
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 show_swap_cache_info();
2209}
2210
Mel Gorman19770b32008-04-28 02:12:18 -07002211static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2212{
2213 zoneref->zone = zone;
2214 zoneref->zone_idx = zone_idx(zone);
2215}
2216
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217/*
2218 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08002219 *
2220 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002222static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2223 int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224{
Christoph Lameter1a932052006-01-06 00:11:16 -08002225 struct zone *zone;
2226
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002227 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002228 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08002229
2230 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002231 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08002232 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08002233 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002234 zoneref_set_zone(zone,
2235 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08002236 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08002238
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002239 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08002240 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241}
2242
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002243
2244/*
2245 * zonelist_order:
2246 * 0 = automatic detection of better ordering.
2247 * 1 = order by ([node] distance, -zonetype)
2248 * 2 = order by (-zonetype, [node] distance)
2249 *
2250 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2251 * the same zonelist. So only NUMA can configure this param.
2252 */
2253#define ZONELIST_ORDER_DEFAULT 0
2254#define ZONELIST_ORDER_NODE 1
2255#define ZONELIST_ORDER_ZONE 2
2256
2257/* zonelist order in the kernel.
2258 * set_zonelist_order() will set this to NODE or ZONE.
2259 */
2260static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2261static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2262
2263
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002265/* The value user specified ....changed by config */
2266static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2267/* string for sysctl */
2268#define NUMA_ZONELIST_ORDER_LEN 16
2269char numa_zonelist_order[16] = "default";
2270
2271/*
2272 * interface for configure zonelist ordering.
2273 * command line option "numa_zonelist_order"
2274 * = "[dD]efault - default, automatic configuration.
2275 * = "[nN]ode - order by node locality, then by zone within node
2276 * = "[zZ]one - order by zone, then by locality within zone
2277 */
2278
2279static int __parse_numa_zonelist_order(char *s)
2280{
2281 if (*s == 'd' || *s == 'D') {
2282 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2283 } else if (*s == 'n' || *s == 'N') {
2284 user_zonelist_order = ZONELIST_ORDER_NODE;
2285 } else if (*s == 'z' || *s == 'Z') {
2286 user_zonelist_order = ZONELIST_ORDER_ZONE;
2287 } else {
2288 printk(KERN_WARNING
2289 "Ignoring invalid numa_zonelist_order value: "
2290 "%s\n", s);
2291 return -EINVAL;
2292 }
2293 return 0;
2294}
2295
2296static __init int setup_numa_zonelist_order(char *s)
2297{
2298 if (s)
2299 return __parse_numa_zonelist_order(s);
2300 return 0;
2301}
2302early_param("numa_zonelist_order", setup_numa_zonelist_order);
2303
2304/*
2305 * sysctl handler for numa_zonelist_order
2306 */
2307int numa_zonelist_order_handler(ctl_table *table, int write,
2308 struct file *file, void __user *buffer, size_t *length,
2309 loff_t *ppos)
2310{
2311 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2312 int ret;
2313
2314 if (write)
2315 strncpy(saved_string, (char*)table->data,
2316 NUMA_ZONELIST_ORDER_LEN);
2317 ret = proc_dostring(table, write, file, buffer, length, ppos);
2318 if (ret)
2319 return ret;
2320 if (write) {
2321 int oldval = user_zonelist_order;
2322 if (__parse_numa_zonelist_order((char*)table->data)) {
2323 /*
2324 * bogus value. restore saved string
2325 */
2326 strncpy((char*)table->data, saved_string,
2327 NUMA_ZONELIST_ORDER_LEN);
2328 user_zonelist_order = oldval;
2329 } else if (oldval != user_zonelist_order)
2330 build_all_zonelists();
2331 }
2332 return 0;
2333}
2334
2335
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002336#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002337static int node_load[MAX_NUMNODES];
2338
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07002340 * 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 -07002341 * @node: node whose fallback list we're appending
2342 * @used_node_mask: nodemask_t of already used nodes
2343 *
2344 * We use a number of factors to determine which is the next node that should
2345 * appear on a given node's fallback list. The node should not have appeared
2346 * already in @node's fallback list, and it should be the next closest node
2347 * according to the distance array (which contains arbitrary distance values
2348 * from each node to each node in the system), and should also prefer nodes
2349 * with no CPUs, since presumably they'll have very little allocation pressure
2350 * on them otherwise.
2351 * It returns -1 if no node is found.
2352 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002353static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002355 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356 int min_val = INT_MAX;
2357 int best_node = -1;
Rusty Russella70f7302009-03-13 14:49:46 +10302358 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002360 /* Use the local node if we haven't already */
2361 if (!node_isset(node, *used_node_mask)) {
2362 node_set(node, *used_node_mask);
2363 return node;
2364 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002366 for_each_node_state(n, N_HIGH_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367
2368 /* Don't want a node to appear more than once */
2369 if (node_isset(n, *used_node_mask))
2370 continue;
2371
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 /* Use the distance array to find the distance */
2373 val = node_distance(node, n);
2374
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002375 /* Penalize nodes under us ("prefer the next node") */
2376 val += (n < node);
2377
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10302379 tmp = cpumask_of_node(n);
2380 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 val += PENALTY_FOR_NODE_WITH_CPUS;
2382
2383 /* Slight preference for less loaded node */
2384 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2385 val += node_load[n];
2386
2387 if (val < min_val) {
2388 min_val = val;
2389 best_node = n;
2390 }
2391 }
2392
2393 if (best_node >= 0)
2394 node_set(best_node, *used_node_mask);
2395
2396 return best_node;
2397}
2398
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002399
2400/*
2401 * Build zonelists ordered by node and zones within node.
2402 * This results in maximum locality--normal zone overflows into local
2403 * DMA zone, if any--but risks exhausting DMA zone.
2404 */
2405static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002407 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002409
Mel Gorman54a6eb52008-04-28 02:12:16 -07002410 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07002411 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07002412 ;
2413 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2414 MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002415 zonelist->_zonerefs[j].zone = NULL;
2416 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002417}
2418
2419/*
Christoph Lameter523b9452007-10-16 01:25:37 -07002420 * Build gfp_thisnode zonelists
2421 */
2422static void build_thisnode_zonelists(pg_data_t *pgdat)
2423{
Christoph Lameter523b9452007-10-16 01:25:37 -07002424 int j;
2425 struct zonelist *zonelist;
2426
Mel Gorman54a6eb52008-04-28 02:12:16 -07002427 zonelist = &pgdat->node_zonelists[1];
2428 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002429 zonelist->_zonerefs[j].zone = NULL;
2430 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07002431}
2432
2433/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002434 * Build zonelists ordered by zone and nodes within zones.
2435 * This results in conserving DMA zone[s] until all Normal memory is
2436 * exhausted, but results in overflowing to remote node while memory
2437 * may still exist in local DMA zone.
2438 */
2439static int node_order[MAX_NUMNODES];
2440
2441static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2442{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002443 int pos, j, node;
2444 int zone_type; /* needs to be signed */
2445 struct zone *z;
2446 struct zonelist *zonelist;
2447
Mel Gorman54a6eb52008-04-28 02:12:16 -07002448 zonelist = &pgdat->node_zonelists[0];
2449 pos = 0;
2450 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2451 for (j = 0; j < nr_nodes; j++) {
2452 node = node_order[j];
2453 z = &NODE_DATA(node)->node_zones[zone_type];
2454 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002455 zoneref_set_zone(z,
2456 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002457 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002458 }
2459 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002460 }
Mel Gormandd1a2392008-04-28 02:12:17 -07002461 zonelist->_zonerefs[pos].zone = NULL;
2462 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002463}
2464
2465static int default_zonelist_order(void)
2466{
2467 int nid, zone_type;
2468 unsigned long low_kmem_size,total_size;
2469 struct zone *z;
2470 int average_size;
2471 /*
2472 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2473 * If they are really small and used heavily, the system can fall
2474 * into OOM very easily.
2475 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2476 */
2477 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2478 low_kmem_size = 0;
2479 total_size = 0;
2480 for_each_online_node(nid) {
2481 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2482 z = &NODE_DATA(nid)->node_zones[zone_type];
2483 if (populated_zone(z)) {
2484 if (zone_type < ZONE_NORMAL)
2485 low_kmem_size += z->present_pages;
2486 total_size += z->present_pages;
2487 }
2488 }
2489 }
2490 if (!low_kmem_size || /* there are no DMA area. */
2491 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2492 return ZONELIST_ORDER_NODE;
2493 /*
2494 * look into each node's config.
2495 * If there is a node whose DMA/DMA32 memory is very big area on
2496 * local memory, NODE_ORDER may be suitable.
2497 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002498 average_size = total_size /
2499 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002500 for_each_online_node(nid) {
2501 low_kmem_size = 0;
2502 total_size = 0;
2503 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2504 z = &NODE_DATA(nid)->node_zones[zone_type];
2505 if (populated_zone(z)) {
2506 if (zone_type < ZONE_NORMAL)
2507 low_kmem_size += z->present_pages;
2508 total_size += z->present_pages;
2509 }
2510 }
2511 if (low_kmem_size &&
2512 total_size > average_size && /* ignore small node */
2513 low_kmem_size > total_size * 70/100)
2514 return ZONELIST_ORDER_NODE;
2515 }
2516 return ZONELIST_ORDER_ZONE;
2517}
2518
2519static void set_zonelist_order(void)
2520{
2521 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2522 current_zonelist_order = default_zonelist_order();
2523 else
2524 current_zonelist_order = user_zonelist_order;
2525}
2526
2527static void build_zonelists(pg_data_t *pgdat)
2528{
2529 int j, node, load;
2530 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002532 int local_node, prev_node;
2533 struct zonelist *zonelist;
2534 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535
2536 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07002537 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07002539 zonelist->_zonerefs[0].zone = NULL;
2540 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002541 }
2542
2543 /* NUMA-aware ordering of nodes */
2544 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002545 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 prev_node = local_node;
2547 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002548
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002549 memset(node_order, 0, sizeof(node_order));
2550 j = 0;
2551
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08002553 int distance = node_distance(local_node, node);
2554
2555 /*
2556 * If another node is sufficiently far away then it is better
2557 * to reclaim pages in a zone before going off node.
2558 */
2559 if (distance > RECLAIM_DISTANCE)
2560 zone_reclaim_mode = 1;
2561
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562 /*
2563 * We don't want to pressure a particular node.
2564 * So adding penalty to the first node in same
2565 * distance group to make it round-robin.
2566 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08002567 if (distance != node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002568 node_load[node] = load;
2569
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 prev_node = node;
2571 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002572 if (order == ZONELIST_ORDER_NODE)
2573 build_zonelists_in_node_order(pgdat, node);
2574 else
2575 node_order[j++] = node; /* remember order */
2576 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002578 if (order == ZONELIST_ORDER_ZONE) {
2579 /* calculate node order -- i.e., DMA last! */
2580 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581 }
Christoph Lameter523b9452007-10-16 01:25:37 -07002582
2583 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584}
2585
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002586/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002587static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002588{
Mel Gorman54a6eb52008-04-28 02:12:16 -07002589 struct zonelist *zonelist;
2590 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07002591 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002592
Mel Gorman54a6eb52008-04-28 02:12:16 -07002593 zonelist = &pgdat->node_zonelists[0];
2594 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2595 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07002596 for (z = zonelist->_zonerefs; z->zone; z++)
2597 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002598}
2599
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002600
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601#else /* CONFIG_NUMA */
2602
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002603static void set_zonelist_order(void)
2604{
2605 current_zonelist_order = ZONELIST_ORDER_ZONE;
2606}
2607
2608static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609{
Christoph Lameter19655d32006-09-25 23:31:19 -07002610 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002611 enum zone_type j;
2612 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613
2614 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615
Mel Gorman54a6eb52008-04-28 02:12:16 -07002616 zonelist = &pgdat->node_zonelists[0];
2617 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618
Mel Gorman54a6eb52008-04-28 02:12:16 -07002619 /*
2620 * Now we build the zonelist so that it contains the zones
2621 * of all the other nodes.
2622 * We don't want to pressure a particular node, so when
2623 * building the zones for node N, we make sure that the
2624 * zones coming right after the local ones are those from
2625 * node N+1 (modulo N)
2626 */
2627 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2628 if (!node_online(node))
2629 continue;
2630 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2631 MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07002633 for (node = 0; node < local_node; node++) {
2634 if (!node_online(node))
2635 continue;
2636 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2637 MAX_NR_ZONES - 1);
2638 }
2639
Mel Gormandd1a2392008-04-28 02:12:17 -07002640 zonelist->_zonerefs[j].zone = NULL;
2641 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642}
2643
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002644/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002645static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002646{
Mel Gorman54a6eb52008-04-28 02:12:16 -07002647 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002648}
2649
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650#endif /* CONFIG_NUMA */
2651
Rusty Russell9b1a4d32008-07-28 12:16:30 -05002652/* return values int ....just for stop_machine() */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002653static int __build_all_zonelists(void *dummy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654{
Yasunori Goto68113782006-06-23 02:03:11 -07002655 int nid;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002656
Bo Liu7f9cfb32009-08-18 14:11:19 -07002657#ifdef CONFIG_NUMA
2658 memset(node_load, 0, sizeof(node_load));
2659#endif
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002660 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07002661 pg_data_t *pgdat = NODE_DATA(nid);
2662
2663 build_zonelists(pgdat);
2664 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002665 }
Yasunori Goto68113782006-06-23 02:03:11 -07002666 return 0;
2667}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002669void build_all_zonelists(void)
Yasunori Goto68113782006-06-23 02:03:11 -07002670{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002671 set_zonelist_order();
2672
Yasunori Goto68113782006-06-23 02:03:11 -07002673 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07002674 __build_all_zonelists(NULL);
Mel Gorman68ad8df2008-07-23 21:26:52 -07002675 mminit_verify_zonelist();
Yasunori Goto68113782006-06-23 02:03:11 -07002676 cpuset_init_current_mems_allowed();
2677 } else {
Simon Arlott183ff222007-10-20 01:27:18 +02002678 /* we have to stop all cpus to guarantee there is no user
Yasunori Goto68113782006-06-23 02:03:11 -07002679 of zonelist */
Rusty Russell9b1a4d32008-07-28 12:16:30 -05002680 stop_machine(__build_all_zonelists, NULL, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07002681 /* cpuset refresh routine should be here */
2682 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07002683 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002684 /*
2685 * Disable grouping by mobility if the number of pages in the
2686 * system is too low to allow the mechanism to work. It would be
2687 * more accurate, but expensive to check per-zone. This check is
2688 * made on memory-hotadd so a system can start with mobility
2689 * disabled and enable it later
2690 */
Mel Gormand9c23402007-10-16 01:26:01 -07002691 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002692 page_group_by_mobility_disabled = 1;
2693 else
2694 page_group_by_mobility_disabled = 0;
2695
2696 printk("Built %i zonelists in %s order, mobility grouping %s. "
2697 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002698 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002699 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002700 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002701 vm_total_pages);
2702#ifdef CONFIG_NUMA
2703 printk("Policy zone: %s\n", zone_names[policy_zone]);
2704#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705}
2706
2707/*
2708 * Helper functions to size the waitqueue hash table.
2709 * Essentially these want to choose hash table sizes sufficiently
2710 * large so that collisions trying to wait on pages are rare.
2711 * But in fact, the number of active page waitqueues on typical
2712 * systems is ridiculously low, less than 200. So this is even
2713 * conservative, even though it seems large.
2714 *
2715 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2716 * waitqueues, i.e. the size of the waitq table given the number of pages.
2717 */
2718#define PAGES_PER_WAITQUEUE 256
2719
Yasunori Gotocca448f2006-06-23 02:03:10 -07002720#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07002721static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722{
2723 unsigned long size = 1;
2724
2725 pages /= PAGES_PER_WAITQUEUE;
2726
2727 while (size < pages)
2728 size <<= 1;
2729
2730 /*
2731 * Once we have dozens or even hundreds of threads sleeping
2732 * on IO we've got bigger problems than wait queue collision.
2733 * Limit the size of the wait table to a reasonable size.
2734 */
2735 size = min(size, 4096UL);
2736
2737 return max(size, 4UL);
2738}
Yasunori Gotocca448f2006-06-23 02:03:10 -07002739#else
2740/*
2741 * A zone's size might be changed by hot-add, so it is not possible to determine
2742 * a suitable size for its wait_table. So we use the maximum size now.
2743 *
2744 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2745 *
2746 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2747 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2748 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2749 *
2750 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2751 * or more by the traditional way. (See above). It equals:
2752 *
2753 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2754 * ia64(16K page size) : = ( 8G + 4M)byte.
2755 * powerpc (64K page size) : = (32G +16M)byte.
2756 */
2757static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2758{
2759 return 4096UL;
2760}
2761#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762
2763/*
2764 * This is an integer logarithm so that shifts can be used later
2765 * to extract the more random high bits from the multiplicative
2766 * hash function before the remainder is taken.
2767 */
2768static inline unsigned long wait_table_bits(unsigned long size)
2769{
2770 return ffz(~size);
2771}
2772
2773#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2774
Mel Gorman56fd56b2007-10-16 01:25:58 -07002775/*
Mel Gormand9c23402007-10-16 01:26:01 -07002776 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07002777 * of blocks reserved is based on min_wmark_pages(zone). The memory within
2778 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07002779 * higher will lead to a bigger reserve which will get freed as contiguous
2780 * blocks as reclaim kicks in
2781 */
2782static void setup_zone_migrate_reserve(struct zone *zone)
2783{
2784 unsigned long start_pfn, pfn, end_pfn;
2785 struct page *page;
2786 unsigned long reserve, block_migratetype;
2787
2788 /* Get the start pfn, end pfn and the number of blocks to reserve */
2789 start_pfn = zone->zone_start_pfn;
2790 end_pfn = start_pfn + zone->spanned_pages;
Mel Gorman41858962009-06-16 15:32:12 -07002791 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07002792 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07002793
Mel Gormand9c23402007-10-16 01:26:01 -07002794 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07002795 if (!pfn_valid(pfn))
2796 continue;
2797 page = pfn_to_page(pfn);
2798
Adam Litke344c7902008-09-02 14:35:38 -07002799 /* Watch out for overlapping nodes */
2800 if (page_to_nid(page) != zone_to_nid(zone))
2801 continue;
2802
Mel Gorman56fd56b2007-10-16 01:25:58 -07002803 /* Blocks with reserved pages will never free, skip them. */
2804 if (PageReserved(page))
2805 continue;
2806
2807 block_migratetype = get_pageblock_migratetype(page);
2808
2809 /* If this block is reserved, account for it */
2810 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
2811 reserve--;
2812 continue;
2813 }
2814
2815 /* Suitable for reserving if this block is movable */
2816 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
2817 set_pageblock_migratetype(page, MIGRATE_RESERVE);
2818 move_freepages_block(zone, page, MIGRATE_RESERVE);
2819 reserve--;
2820 continue;
2821 }
2822
2823 /*
2824 * If the reserve is met and this is a previous reserved block,
2825 * take it back
2826 */
2827 if (block_migratetype == MIGRATE_RESERVE) {
2828 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2829 move_freepages_block(zone, page, MIGRATE_MOVABLE);
2830 }
2831 }
2832}
Mel Gormanac0e5b72007-10-16 01:25:58 -07002833
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834/*
2835 * Initially all pages are reserved - free ones are freed
2836 * up by free_all_bootmem() once the early boot process is
2837 * done. Non-atomic initialization, single-pass.
2838 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002839void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa02007-01-10 23:15:30 -08002840 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002842 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07002843 unsigned long end_pfn = start_pfn + size;
2844 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002845 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846
Hugh Dickins22b31ee2009-01-06 14:40:09 -08002847 if (highest_memmap_pfn < end_pfn - 1)
2848 highest_memmap_pfn = end_pfn - 1;
2849
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002850 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08002851 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa02007-01-10 23:15:30 -08002852 /*
2853 * There can be holes in boot-time mem_map[]s
2854 * handed to this function. They do not
2855 * exist on hotplugged memory.
2856 */
2857 if (context == MEMMAP_EARLY) {
2858 if (!early_pfn_valid(pfn))
2859 continue;
2860 if (!early_pfn_in_nid(pfn, nid))
2861 continue;
2862 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002863 page = pfn_to_page(pfn);
2864 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07002865 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08002866 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 reset_page_mapcount(page);
2868 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002869 /*
2870 * Mark the block movable so that blocks are reserved for
2871 * movable at startup. This will force kernel allocations
2872 * to reserve their blocks rather than leaking throughout
2873 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07002874 * kernel allocations are made. Later some blocks near
2875 * the start are marked MIGRATE_RESERVE by
2876 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002877 *
2878 * bitmap is created for zone's valid pfn range. but memmap
2879 * can be created for invalid pages (for alignment)
2880 * check here not to call set_pageblock_migratetype() against
2881 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002882 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002883 if ((z->zone_start_pfn <= pfn)
2884 && (pfn < z->zone_start_pfn + z->spanned_pages)
2885 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07002886 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002887
Linus Torvalds1da177e2005-04-16 15:20:36 -07002888 INIT_LIST_HEAD(&page->lru);
2889#ifdef WANT_PAGE_VIRTUAL
2890 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2891 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07002892 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002893#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002894 }
2895}
2896
Andi Kleen1e548de2008-02-04 22:29:26 -08002897static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898{
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002899 int order, t;
2900 for_each_migratetype_order(order, t) {
2901 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902 zone->free_area[order].nr_free = 0;
2903 }
2904}
2905
2906#ifndef __HAVE_ARCH_MEMMAP_INIT
2907#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa02007-01-10 23:15:30 -08002908 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909#endif
2910
Sam Ravnborg1d6f4e62008-01-17 15:21:12 -08002911static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002912{
David Howells3a6be872009-05-06 16:03:03 -07002913#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002914 int batch;
2915
2916 /*
2917 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07002918 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002919 *
2920 * OK, so we don't know how big the cache is. So guess.
2921 */
2922 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07002923 if (batch * PAGE_SIZE > 512 * 1024)
2924 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002925 batch /= 4; /* We effectively *= 4 below */
2926 if (batch < 1)
2927 batch = 1;
2928
2929 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002930 * Clamp the batch to a 2^n - 1 value. Having a power
2931 * of 2 value was found to be more likely to have
2932 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002933 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002934 * For example if 2 tasks are alternately allocating
2935 * batches of pages, one task can end up with a lot
2936 * of pages of one half of the possible page colors
2937 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002938 */
David Howells91552032009-05-06 16:03:02 -07002939 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07002940
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002941 return batch;
David Howells3a6be872009-05-06 16:03:03 -07002942
2943#else
2944 /* The deferral and batching of frees should be suppressed under NOMMU
2945 * conditions.
2946 *
2947 * The problem is that NOMMU needs to be able to allocate large chunks
2948 * of contiguous memory as there's no hardware page translation to
2949 * assemble apparent contiguous memory from discontiguous pages.
2950 *
2951 * Queueing large contiguous runs of pages for batching, however,
2952 * causes the pages to actually be freed in smaller chunks. As there
2953 * can be a significant delay between the individual batches being
2954 * recycled, this leads to the once large chunks of space being
2955 * fragmented and becoming unavailable for high-order allocations.
2956 */
2957 return 0;
2958#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002959}
2960
Adrian Bunkb69a7282008-07-23 21:28:12 -07002961static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
Christoph Lameter2caaad42005-06-21 17:15:00 -07002962{
2963 struct per_cpu_pages *pcp;
2964
Magnus Damm1c6fe942005-10-26 01:58:59 -07002965 memset(p, 0, sizeof(*p));
2966
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002967 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07002968 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07002969 pcp->high = 6 * batch;
2970 pcp->batch = max(1UL, 1 * batch);
2971 INIT_LIST_HEAD(&pcp->list);
Christoph Lameter2caaad42005-06-21 17:15:00 -07002972}
2973
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002974/*
2975 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2976 * to the value high for the pageset p.
2977 */
2978
2979static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2980 unsigned long high)
2981{
2982 struct per_cpu_pages *pcp;
2983
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002984 pcp = &p->pcp;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002985 pcp->high = high;
2986 pcp->batch = max(1UL, high/4);
2987 if ((high/4) > (PAGE_SHIFT * 8))
2988 pcp->batch = PAGE_SHIFT * 8;
2989}
2990
2991
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002992#ifdef CONFIG_NUMA
2993/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07002994 * Boot pageset table. One per cpu which is going to be used for all
2995 * zones and all nodes. The parameters will be set in such a way
2996 * that an item put on a list will immediately be handed over to
2997 * the buddy list. This is safe since pageset manipulation is done
2998 * with interrupts disabled.
2999 *
3000 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07003001 *
3002 * The boot_pagesets must be kept even after bootup is complete for
3003 * unused processors and/or zones. They do play a role for bootstrapping
3004 * hotplugged processors.
3005 *
3006 * zoneinfo_show() and maybe other functions do
3007 * not check if the processor is online before following the pageset pointer.
3008 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07003009 */
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08003010static struct per_cpu_pageset boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07003011
3012/*
3013 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003014 * per cpu pageset array in struct zone.
3015 */
Ashok Raj6292d9a2006-02-01 03:04:44 -08003016static int __cpuinit process_zones(int cpu)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003017{
3018 struct zone *zone, *dzone;
Christoph Lameter37c07082007-10-16 01:25:36 -07003019 int node = cpu_to_node(cpu);
3020
3021 node_set_state(node, N_CPU); /* this node has a cpu */
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003022
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003023 for_each_populated_zone(zone) {
Nick Piggin23316bc2006-01-08 01:00:41 -08003024 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lameter37c07082007-10-16 01:25:36 -07003025 GFP_KERNEL, node);
Nick Piggin23316bc2006-01-08 01:00:41 -08003026 if (!zone_pcp(zone, cpu))
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003027 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003028
Nick Piggin23316bc2006-01-08 01:00:41 -08003029 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003030
3031 if (percpu_pagelist_fraction)
3032 setup_pagelist_highmark(zone_pcp(zone, cpu),
3033 (zone->present_pages / percpu_pagelist_fraction));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003034 }
3035
3036 return 0;
3037bad:
3038 for_each_zone(dzone) {
Andrew Morton64191682007-08-30 23:56:17 -07003039 if (!populated_zone(dzone))
3040 continue;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003041 if (dzone == zone)
3042 break;
Nick Piggin23316bc2006-01-08 01:00:41 -08003043 kfree(zone_pcp(dzone, cpu));
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07003044 zone_pcp(dzone, cpu) = &boot_pageset[cpu];
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003045 }
3046 return -ENOMEM;
3047}
3048
3049static inline void free_zone_pagesets(int cpu)
3050{
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003051 struct zone *zone;
3052
3053 for_each_zone(zone) {
3054 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
3055
David Rientjesf3ef9ea2006-09-25 16:24:57 -07003056 /* Free per_cpu_pageset if it is slab allocated */
3057 if (pset != &boot_pageset[cpu])
3058 kfree(pset);
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07003059 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003060 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003061}
3062
Chandra Seetharaman9c7b2162006-06-27 02:54:07 -07003063static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003064 unsigned long action,
3065 void *hcpu)
3066{
3067 int cpu = (long)hcpu;
3068 int ret = NOTIFY_OK;
3069
3070 switch (action) {
Andy Whitcroftce421c72006-12-06 20:33:08 -08003071 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003072 case CPU_UP_PREPARE_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08003073 if (process_zones(cpu))
3074 ret = NOTIFY_BAD;
3075 break;
3076 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003077 case CPU_UP_CANCELED_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08003078 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003079 case CPU_DEAD_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08003080 free_zone_pagesets(cpu);
3081 break;
3082 default:
3083 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003084 }
3085 return ret;
3086}
3087
Chandra Seetharaman74b85f32006-06-27 02:54:09 -07003088static struct notifier_block __cpuinitdata pageset_notifier =
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003089 { &pageset_cpuup_callback, NULL, 0 };
3090
Al Viro78d99552005-12-15 09:18:25 +00003091void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003092{
3093 int err;
3094
3095 /* Initialize per_cpu_pageset for cpu 0.
3096 * A cpuup callback will do this for every cpu
3097 * as it comes online
3098 */
3099 err = process_zones(smp_processor_id());
3100 BUG_ON(err);
3101 register_cpu_notifier(&pageset_notifier);
3102}
3103
3104#endif
3105
Sam Ravnborg577a32f2007-05-17 23:29:25 +02003106static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07003107int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07003108{
3109 int i;
3110 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003111 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07003112
3113 /*
3114 * The per-page waitqueue mechanism uses hashed waitqueues
3115 * per zone.
3116 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07003117 zone->wait_table_hash_nr_entries =
3118 wait_table_hash_nr_entries(zone_size_pages);
3119 zone->wait_table_bits =
3120 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003121 alloc_size = zone->wait_table_hash_nr_entries
3122 * sizeof(wait_queue_head_t);
3123
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07003124 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07003125 zone->wait_table = (wait_queue_head_t *)
3126 alloc_bootmem_node(pgdat, alloc_size);
3127 } else {
3128 /*
3129 * This case means that a zone whose size was 0 gets new memory
3130 * via memory hot-add.
3131 * But it may be the case that a new node was hot-added. In
3132 * this case vmalloc() will not be able to use this new node's
3133 * memory - this wait_table must be initialized to use this new
3134 * node itself as well.
3135 * To use this new node's memory, further consideration will be
3136 * necessary.
3137 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07003138 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003139 }
3140 if (!zone->wait_table)
3141 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07003142
Yasunori Goto02b694d2006-06-23 02:03:08 -07003143 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07003144 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003145
3146 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003147}
3148
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003149static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07003150{
3151 int cpu;
3152 unsigned long batch = zone_batchsize(zone);
3153
3154 for (cpu = 0; cpu < NR_CPUS; cpu++) {
3155#ifdef CONFIG_NUMA
3156 /* Early boot. Slab allocator not functional yet */
Nick Piggin23316bc2006-01-08 01:00:41 -08003157 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Dave Hansened8ece22005-10-29 18:16:50 -07003158 setup_pageset(&boot_pageset[cpu],0);
3159#else
3160 setup_pageset(zone_pcp(zone,cpu), batch);
3161#endif
3162 }
Anton Blanchardf5335c02006-03-25 03:06:49 -08003163 if (zone->present_pages)
3164 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
3165 zone->name, zone->present_pages, batch);
Dave Hansened8ece22005-10-29 18:16:50 -07003166}
3167
Yasunori Goto718127c2006-06-23 02:03:10 -07003168__meminit int init_currently_empty_zone(struct zone *zone,
3169 unsigned long zone_start_pfn,
Dave Hansena2f3aa02007-01-10 23:15:30 -08003170 unsigned long size,
3171 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07003172{
3173 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003174 int ret;
3175 ret = zone_wait_table_init(zone, size);
3176 if (ret)
3177 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07003178 pgdat->nr_zones = zone_idx(zone) + 1;
3179
Dave Hansened8ece22005-10-29 18:16:50 -07003180 zone->zone_start_pfn = zone_start_pfn;
3181
Mel Gorman708614e2008-07-23 21:26:51 -07003182 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3183 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3184 pgdat->node_id,
3185 (unsigned long)zone_idx(zone),
3186 zone_start_pfn, (zone_start_pfn + size));
3187
Andi Kleen1e548de2008-02-04 22:29:26 -08003188 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07003189
3190 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003191}
3192
Mel Gormanc7132162006-09-27 01:49:43 -07003193#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3194/*
3195 * Basic iterator support. Return the first range of PFNs for a node
3196 * Note: nid == MAX_NUMNODES returns first region regardless of node
3197 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003198static int __meminit first_active_region_index_in_nid(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003199{
3200 int i;
3201
3202 for (i = 0; i < nr_nodemap_entries; i++)
3203 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3204 return i;
3205
3206 return -1;
3207}
3208
3209/*
3210 * Basic iterator support. Return the next active range of PFNs for a node
Simon Arlott183ff222007-10-20 01:27:18 +02003211 * Note: nid == MAX_NUMNODES returns next region regardless of node
Mel Gormanc7132162006-09-27 01:49:43 -07003212 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003213static int __meminit next_active_region_index_in_nid(int index, int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003214{
3215 for (index = index + 1; index < nr_nodemap_entries; index++)
3216 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3217 return index;
3218
3219 return -1;
3220}
3221
3222#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3223/*
3224 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3225 * Architectures may implement their own version but if add_active_range()
3226 * was used and there are no special requirements, this is a convenient
3227 * alternative
3228 */
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003229int __meminit __early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07003230{
3231 int i;
3232
3233 for (i = 0; i < nr_nodemap_entries; i++) {
3234 unsigned long start_pfn = early_node_map[i].start_pfn;
3235 unsigned long end_pfn = early_node_map[i].end_pfn;
3236
3237 if (start_pfn <= pfn && pfn < end_pfn)
3238 return early_node_map[i].nid;
3239 }
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003240 /* This is a memory hole */
3241 return -1;
Mel Gormanc7132162006-09-27 01:49:43 -07003242}
3243#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3244
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003245int __meminit early_pfn_to_nid(unsigned long pfn)
3246{
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003247 int nid;
3248
3249 nid = __early_pfn_to_nid(pfn);
3250 if (nid >= 0)
3251 return nid;
3252 /* just returns 0 */
3253 return 0;
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003254}
3255
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003256#ifdef CONFIG_NODES_SPAN_OTHER_NODES
3257bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3258{
3259 int nid;
3260
3261 nid = __early_pfn_to_nid(pfn);
3262 if (nid >= 0 && nid != node)
3263 return false;
3264 return true;
3265}
3266#endif
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003267
Mel Gormanc7132162006-09-27 01:49:43 -07003268/* Basic iterator support to walk early_node_map[] */
3269#define for_each_active_range_index_in_nid(i, nid) \
3270 for (i = first_active_region_index_in_nid(nid); i != -1; \
3271 i = next_active_region_index_in_nid(i, nid))
3272
3273/**
3274 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003275 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3276 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07003277 *
3278 * If an architecture guarantees that all ranges registered with
3279 * add_active_ranges() contain no holes and may be freed, this
3280 * this function may be used instead of calling free_bootmem() manually.
3281 */
3282void __init free_bootmem_with_active_regions(int nid,
3283 unsigned long max_low_pfn)
3284{
3285 int i;
3286
3287 for_each_active_range_index_in_nid(i, nid) {
3288 unsigned long size_pages = 0;
3289 unsigned long end_pfn = early_node_map[i].end_pfn;
3290
3291 if (early_node_map[i].start_pfn >= max_low_pfn)
3292 continue;
3293
3294 if (end_pfn > max_low_pfn)
3295 end_pfn = max_low_pfn;
3296
3297 size_pages = end_pfn - early_node_map[i].start_pfn;
3298 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3299 PFN_PHYS(early_node_map[i].start_pfn),
3300 size_pages << PAGE_SHIFT);
3301 }
3302}
3303
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003304void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3305{
3306 int i;
Yinghai Lud52d53b2008-06-16 20:10:55 -07003307 int ret;
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003308
Yinghai Lud52d53b2008-06-16 20:10:55 -07003309 for_each_active_range_index_in_nid(i, nid) {
3310 ret = work_fn(early_node_map[i].start_pfn,
3311 early_node_map[i].end_pfn, data);
3312 if (ret)
3313 break;
3314 }
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003315}
Mel Gormanc7132162006-09-27 01:49:43 -07003316/**
3317 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003318 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07003319 *
3320 * If an architecture guarantees that all ranges registered with
3321 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003322 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07003323 */
3324void __init sparse_memory_present_with_active_regions(int nid)
3325{
3326 int i;
3327
3328 for_each_active_range_index_in_nid(i, nid)
3329 memory_present(early_node_map[i].nid,
3330 early_node_map[i].start_pfn,
3331 early_node_map[i].end_pfn);
3332}
3333
3334/**
3335 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003336 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3337 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3338 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07003339 *
3340 * It returns the start and end page frame of a node based on information
3341 * provided by an arch calling add_active_range(). If called for a node
3342 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003343 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07003344 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003345void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003346 unsigned long *start_pfn, unsigned long *end_pfn)
3347{
3348 int i;
3349 *start_pfn = -1UL;
3350 *end_pfn = 0;
3351
3352 for_each_active_range_index_in_nid(i, nid) {
3353 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3354 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3355 }
3356
Christoph Lameter633c0662007-10-16 01:25:37 -07003357 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07003358 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07003359}
3360
3361/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07003362 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3363 * assumption is made that zones within a node are ordered in monotonic
3364 * increasing memory addresses so that the "highest" populated zone is used
3365 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003366static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07003367{
3368 int zone_index;
3369 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3370 if (zone_index == ZONE_MOVABLE)
3371 continue;
3372
3373 if (arch_zone_highest_possible_pfn[zone_index] >
3374 arch_zone_lowest_possible_pfn[zone_index])
3375 break;
3376 }
3377
3378 VM_BUG_ON(zone_index == -1);
3379 movable_zone = zone_index;
3380}
3381
3382/*
3383 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3384 * because it is sized independant of architecture. Unlike the other zones,
3385 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3386 * in each node depending on the size of each node and how evenly kernelcore
3387 * is distributed. This helper function adjusts the zone ranges
3388 * provided by the architecture for a given node by using the end of the
3389 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3390 * zones within a node are in order of monotonic increases memory addresses
3391 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003392static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07003393 unsigned long zone_type,
3394 unsigned long node_start_pfn,
3395 unsigned long node_end_pfn,
3396 unsigned long *zone_start_pfn,
3397 unsigned long *zone_end_pfn)
3398{
3399 /* Only adjust if ZONE_MOVABLE is on this node */
3400 if (zone_movable_pfn[nid]) {
3401 /* Size ZONE_MOVABLE */
3402 if (zone_type == ZONE_MOVABLE) {
3403 *zone_start_pfn = zone_movable_pfn[nid];
3404 *zone_end_pfn = min(node_end_pfn,
3405 arch_zone_highest_possible_pfn[movable_zone]);
3406
3407 /* Adjust for ZONE_MOVABLE starting within this range */
3408 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3409 *zone_end_pfn > zone_movable_pfn[nid]) {
3410 *zone_end_pfn = zone_movable_pfn[nid];
3411
3412 /* Check if this whole range is within ZONE_MOVABLE */
3413 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3414 *zone_start_pfn = *zone_end_pfn;
3415 }
3416}
3417
3418/*
Mel Gormanc7132162006-09-27 01:49:43 -07003419 * Return the number of pages a zone spans in a node, including holes
3420 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3421 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003422static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003423 unsigned long zone_type,
3424 unsigned long *ignored)
3425{
3426 unsigned long node_start_pfn, node_end_pfn;
3427 unsigned long zone_start_pfn, zone_end_pfn;
3428
3429 /* Get the start and end of the node and zone */
3430 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3431 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3432 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07003433 adjust_zone_range_for_zone_movable(nid, zone_type,
3434 node_start_pfn, node_end_pfn,
3435 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003436
3437 /* Check that this node has pages within the zone's required range */
3438 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3439 return 0;
3440
3441 /* Move the zone boundaries inside the node if necessary */
3442 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3443 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3444
3445 /* Return the spanned pages */
3446 return zone_end_pfn - zone_start_pfn;
3447}
3448
3449/*
3450 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003451 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07003452 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003453static unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003454 unsigned long range_start_pfn,
3455 unsigned long range_end_pfn)
3456{
3457 int i = 0;
3458 unsigned long prev_end_pfn = 0, hole_pages = 0;
3459 unsigned long start_pfn;
3460
3461 /* Find the end_pfn of the first active range of pfns in the node */
3462 i = first_active_region_index_in_nid(nid);
3463 if (i == -1)
3464 return 0;
3465
Mel Gormanb5445f92007-07-26 10:41:18 -07003466 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3467
Mel Gorman9c7cd682006-09-27 01:49:58 -07003468 /* Account for ranges before physical memory on this node */
3469 if (early_node_map[i].start_pfn > range_start_pfn)
Mel Gormanb5445f92007-07-26 10:41:18 -07003470 hole_pages = prev_end_pfn - range_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003471
3472 /* Find all holes for the zone within the node */
3473 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3474
3475 /* No need to continue if prev_end_pfn is outside the zone */
3476 if (prev_end_pfn >= range_end_pfn)
3477 break;
3478
3479 /* Make sure the end of the zone is not within the hole */
3480 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3481 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3482
3483 /* Update the hole size cound and move on */
3484 if (start_pfn > range_start_pfn) {
3485 BUG_ON(prev_end_pfn > start_pfn);
3486 hole_pages += start_pfn - prev_end_pfn;
3487 }
3488 prev_end_pfn = early_node_map[i].end_pfn;
3489 }
3490
Mel Gorman9c7cd682006-09-27 01:49:58 -07003491 /* Account for ranges past physical memory on this node */
3492 if (range_end_pfn > prev_end_pfn)
Mel Gorman0c6cb972006-10-28 10:38:59 -07003493 hole_pages += range_end_pfn -
Mel Gorman9c7cd682006-09-27 01:49:58 -07003494 max(range_start_pfn, prev_end_pfn);
3495
Mel Gormanc7132162006-09-27 01:49:43 -07003496 return hole_pages;
3497}
3498
3499/**
3500 * absent_pages_in_range - Return number of page frames in holes within a range
3501 * @start_pfn: The start PFN to start searching for holes
3502 * @end_pfn: The end PFN to stop searching for holes
3503 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003504 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07003505 */
3506unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3507 unsigned long end_pfn)
3508{
3509 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3510}
3511
3512/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003513static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003514 unsigned long zone_type,
3515 unsigned long *ignored)
3516{
Mel Gorman9c7cd682006-09-27 01:49:58 -07003517 unsigned long node_start_pfn, node_end_pfn;
3518 unsigned long zone_start_pfn, zone_end_pfn;
3519
3520 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3521 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3522 node_start_pfn);
3523 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3524 node_end_pfn);
3525
Mel Gorman2a1e2742007-07-17 04:03:12 -07003526 adjust_zone_range_for_zone_movable(nid, zone_type,
3527 node_start_pfn, node_end_pfn,
3528 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07003529 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003530}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003531
Mel Gormanc7132162006-09-27 01:49:43 -07003532#else
Paul Mundt6ea6e682007-07-15 23:38:20 -07003533static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003534 unsigned long zone_type,
3535 unsigned long *zones_size)
3536{
3537 return zones_size[zone_type];
3538}
3539
Paul Mundt6ea6e682007-07-15 23:38:20 -07003540static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003541 unsigned long zone_type,
3542 unsigned long *zholes_size)
3543{
3544 if (!zholes_size)
3545 return 0;
3546
3547 return zholes_size[zone_type];
3548}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003549
Mel Gormanc7132162006-09-27 01:49:43 -07003550#endif
3551
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003552static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Mel Gormanc7132162006-09-27 01:49:43 -07003553 unsigned long *zones_size, unsigned long *zholes_size)
3554{
3555 unsigned long realtotalpages, totalpages = 0;
3556 enum zone_type i;
3557
3558 for (i = 0; i < MAX_NR_ZONES; i++)
3559 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3560 zones_size);
3561 pgdat->node_spanned_pages = totalpages;
3562
3563 realtotalpages = totalpages;
3564 for (i = 0; i < MAX_NR_ZONES; i++)
3565 realtotalpages -=
3566 zone_absent_pages_in_node(pgdat->node_id, i,
3567 zholes_size);
3568 pgdat->node_present_pages = realtotalpages;
3569 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3570 realtotalpages);
3571}
3572
Mel Gorman835c1342007-10-16 01:25:47 -07003573#ifndef CONFIG_SPARSEMEM
3574/*
3575 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07003576 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3577 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07003578 * round what is now in bits to nearest long in bits, then return it in
3579 * bytes.
3580 */
3581static unsigned long __init usemap_size(unsigned long zonesize)
3582{
3583 unsigned long usemapsize;
3584
Mel Gormand9c23402007-10-16 01:26:01 -07003585 usemapsize = roundup(zonesize, pageblock_nr_pages);
3586 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07003587 usemapsize *= NR_PAGEBLOCK_BITS;
3588 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3589
3590 return usemapsize / 8;
3591}
3592
3593static void __init setup_usemap(struct pglist_data *pgdat,
3594 struct zone *zone, unsigned long zonesize)
3595{
3596 unsigned long usemapsize = usemap_size(zonesize);
3597 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08003598 if (usemapsize)
Mel Gorman835c1342007-10-16 01:25:47 -07003599 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
Mel Gorman835c1342007-10-16 01:25:47 -07003600}
3601#else
3602static void inline setup_usemap(struct pglist_data *pgdat,
3603 struct zone *zone, unsigned long zonesize) {}
3604#endif /* CONFIG_SPARSEMEM */
3605
Mel Gormand9c23402007-10-16 01:26:01 -07003606#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08003607
3608/* Return a sensible default order for the pageblock size. */
3609static inline int pageblock_default_order(void)
3610{
3611 if (HPAGE_SHIFT > PAGE_SHIFT)
3612 return HUGETLB_PAGE_ORDER;
3613
3614 return MAX_ORDER-1;
3615}
3616
Mel Gormand9c23402007-10-16 01:26:01 -07003617/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3618static inline void __init set_pageblock_order(unsigned int order)
3619{
3620 /* Check that pageblock_nr_pages has not already been setup */
3621 if (pageblock_order)
3622 return;
3623
3624 /*
3625 * Assume the largest contiguous order of interest is a huge page.
3626 * This value may be variable depending on boot parameters on IA64
3627 */
3628 pageblock_order = order;
3629}
3630#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3631
Mel Gormanba72cb82007-11-28 16:21:13 -08003632/*
3633 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3634 * and pageblock_default_order() are unused as pageblock_order is set
3635 * at compile-time. See include/linux/pageblock-flags.h for the values of
3636 * pageblock_order based on the kernel config
3637 */
3638static inline int pageblock_default_order(unsigned int order)
3639{
3640 return MAX_ORDER-1;
3641}
Mel Gormand9c23402007-10-16 01:26:01 -07003642#define set_pageblock_order(x) do {} while (0)
3643
3644#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3645
Linus Torvalds1da177e2005-04-16 15:20:36 -07003646/*
3647 * Set up the zone data structures:
3648 * - mark all pages reserved
3649 * - mark all memory queues empty
3650 * - clear the memory bitmaps
3651 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08003652static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003653 unsigned long *zones_size, unsigned long *zholes_size)
3654{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07003655 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07003656 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003657 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07003658 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003659
Dave Hansen208d54e2005-10-29 18:16:52 -07003660 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003661 pgdat->nr_zones = 0;
3662 init_waitqueue_head(&pgdat->kswapd_wait);
3663 pgdat->kswapd_max_order = 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003664 pgdat_page_cgroup_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003665
3666 for (j = 0; j < MAX_NR_ZONES; j++) {
3667 struct zone *zone = pgdat->node_zones + j;
Mel Gorman0e0b8642006-09-27 01:49:56 -07003668 unsigned long size, realsize, memmap_pages;
Christoph Lameterb69408e2008-10-18 20:26:14 -07003669 enum lru_list l;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003670
Mel Gormanc7132162006-09-27 01:49:43 -07003671 size = zone_spanned_pages_in_node(nid, j, zones_size);
3672 realsize = size - zone_absent_pages_in_node(nid, j,
3673 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003674
Mel Gorman0e0b8642006-09-27 01:49:56 -07003675 /*
3676 * Adjust realsize so that it accounts for how much memory
3677 * is used by this zone for memmap. This affects the watermark
3678 * and per-cpu initialisations
3679 */
Johannes Weinerf7232152008-05-23 13:04:21 -07003680 memmap_pages =
3681 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
Mel Gorman0e0b8642006-09-27 01:49:56 -07003682 if (realsize >= memmap_pages) {
3683 realsize -= memmap_pages;
Yinghai Lu5594c8c2009-01-06 14:39:14 -08003684 if (memmap_pages)
3685 printk(KERN_DEBUG
3686 " %s zone: %lu pages used for memmap\n",
3687 zone_names[j], memmap_pages);
Mel Gorman0e0b8642006-09-27 01:49:56 -07003688 } else
3689 printk(KERN_WARNING
3690 " %s zone: %lu pages exceeds realsize %lu\n",
3691 zone_names[j], memmap_pages, realsize);
3692
Christoph Lameter62672762007-02-10 01:43:07 -08003693 /* Account for reserved pages */
3694 if (j == 0 && realsize > dma_reserve) {
Mel Gorman0e0b8642006-09-27 01:49:56 -07003695 realsize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07003696 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08003697 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07003698 }
3699
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003700 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003701 nr_kernel_pages += realsize;
3702 nr_all_pages += realsize;
3703
3704 zone->spanned_pages = size;
3705 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07003706#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07003707 zone->node = nid;
Christoph Lameter8417bba2006-09-25 23:31:51 -07003708 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07003709 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07003710 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07003711#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003712 zone->name = zone_names[j];
3713 spin_lock_init(&zone->lock);
3714 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07003715 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003716 zone->zone_pgdat = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003717
Martin Bligh3bb1a852006-10-28 10:38:24 -07003718 zone->prev_priority = DEF_PRIORITY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003719
Dave Hansened8ece22005-10-29 18:16:50 -07003720 zone_pcp_init(zone);
Christoph Lameterb69408e2008-10-18 20:26:14 -07003721 for_each_lru(l) {
3722 INIT_LIST_HEAD(&zone->lru[l].list);
Wu Fengguang6e08a362009-06-16 15:32:29 -07003723 zone->lru[l].nr_saved_scan = 0;
Christoph Lameterb69408e2008-10-18 20:26:14 -07003724 }
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08003725 zone->reclaim_stat.recent_rotated[0] = 0;
3726 zone->reclaim_stat.recent_rotated[1] = 0;
3727 zone->reclaim_stat.recent_scanned[0] = 0;
3728 zone->reclaim_stat.recent_scanned[1] = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07003729 zap_zone_vm_stats(zone);
David Rientjese815af92007-10-16 23:25:54 -07003730 zone->flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003731 if (!size)
3732 continue;
3733
Mel Gormanba72cb82007-11-28 16:21:13 -08003734 set_pageblock_order(pageblock_default_order());
Mel Gorman835c1342007-10-16 01:25:47 -07003735 setup_usemap(pgdat, zone, size);
Dave Hansena2f3aa02007-01-10 23:15:30 -08003736 ret = init_currently_empty_zone(zone, zone_start_pfn,
3737 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07003738 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07003739 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003740 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003741 }
3742}
3743
Sam Ravnborg577a32f2007-05-17 23:29:25 +02003744static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003745{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003746 /* Skip empty nodes */
3747 if (!pgdat->node_spanned_pages)
3748 return;
3749
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003750#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07003751 /* ia64 gets its own node_mem_map, before this, without bootmem */
3752 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07003753 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003754 struct page *map;
3755
Bob Piccoe984bb42006-05-20 15:00:31 -07003756 /*
3757 * The zone's endpoints aren't required to be MAX_ORDER
3758 * aligned but the node_mem_map endpoints must be in order
3759 * for the buddy allocator to function correctly.
3760 */
3761 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3762 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3763 end = ALIGN(end, MAX_ORDER_NR_PAGES);
3764 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07003765 map = alloc_remap(pgdat->node_id, size);
3766 if (!map)
3767 map = alloc_bootmem_node(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07003768 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003769 }
Roman Zippel12d810c2007-05-31 00:40:54 -07003770#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07003771 /*
3772 * With no DISCONTIG, the global mem_map is just set as node 0's
3773 */
Mel Gormanc7132162006-09-27 01:49:43 -07003774 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003775 mem_map = NODE_DATA(0)->node_mem_map;
Mel Gormanc7132162006-09-27 01:49:43 -07003776#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3777 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08003778 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Mel Gormanc7132162006-09-27 01:49:43 -07003779#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3780 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003781#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003782#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003783}
3784
Johannes Weiner9109fb72008-07-23 21:27:20 -07003785void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
3786 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003787{
Johannes Weiner9109fb72008-07-23 21:27:20 -07003788 pg_data_t *pgdat = NODE_DATA(nid);
3789
Linus Torvalds1da177e2005-04-16 15:20:36 -07003790 pgdat->node_id = nid;
3791 pgdat->node_start_pfn = node_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003792 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003793
3794 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07003795#ifdef CONFIG_FLAT_NODE_MEM_MAP
3796 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3797 nid, (unsigned long)pgdat,
3798 (unsigned long)pgdat->node_mem_map);
3799#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003800
3801 free_area_init_core(pgdat, zones_size, zholes_size);
3802}
3803
Mel Gormanc7132162006-09-27 01:49:43 -07003804#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07003805
3806#if MAX_NUMNODES > 1
3807/*
3808 * Figure out the number of possible node ids.
3809 */
3810static void __init setup_nr_node_ids(void)
3811{
3812 unsigned int node;
3813 unsigned int highest = 0;
3814
3815 for_each_node_mask(node, node_possible_map)
3816 highest = node;
3817 nr_node_ids = highest + 1;
3818}
3819#else
3820static inline void setup_nr_node_ids(void)
3821{
3822}
3823#endif
3824
Mel Gormanc7132162006-09-27 01:49:43 -07003825/**
3826 * add_active_range - Register a range of PFNs backed by physical memory
3827 * @nid: The node ID the range resides on
3828 * @start_pfn: The start PFN of the available physical memory
3829 * @end_pfn: The end PFN of the available physical memory
3830 *
3831 * These ranges are stored in an early_node_map[] and later used by
3832 * free_area_init_nodes() to calculate zone sizes and holes. If the
3833 * range spans a memory hole, it is up to the architecture to ensure
3834 * the memory is not freed by the bootmem allocator. If possible
3835 * the range being registered will be merged with existing ranges.
3836 */
3837void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3838 unsigned long end_pfn)
3839{
3840 int i;
3841
Mel Gorman6b74ab92008-07-23 21:26:49 -07003842 mminit_dprintk(MMINIT_TRACE, "memory_register",
3843 "Entering add_active_range(%d, %#lx, %#lx) "
3844 "%d entries of %d used\n",
3845 nid, start_pfn, end_pfn,
3846 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
Mel Gormanc7132162006-09-27 01:49:43 -07003847
Mel Gorman2dbb51c2008-07-23 21:26:52 -07003848 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
3849
Mel Gormanc7132162006-09-27 01:49:43 -07003850 /* Merge with existing active regions if possible */
3851 for (i = 0; i < nr_nodemap_entries; i++) {
3852 if (early_node_map[i].nid != nid)
3853 continue;
3854
3855 /* Skip if an existing region covers this new one */
3856 if (start_pfn >= early_node_map[i].start_pfn &&
3857 end_pfn <= early_node_map[i].end_pfn)
3858 return;
3859
3860 /* Merge forward if suitable */
3861 if (start_pfn <= early_node_map[i].end_pfn &&
3862 end_pfn > early_node_map[i].end_pfn) {
3863 early_node_map[i].end_pfn = end_pfn;
3864 return;
3865 }
3866
3867 /* Merge backward if suitable */
3868 if (start_pfn < early_node_map[i].end_pfn &&
3869 end_pfn >= early_node_map[i].start_pfn) {
3870 early_node_map[i].start_pfn = start_pfn;
3871 return;
3872 }
3873 }
3874
3875 /* Check that early_node_map is large enough */
3876 if (i >= MAX_ACTIVE_REGIONS) {
3877 printk(KERN_CRIT "More than %d memory regions, truncating\n",
3878 MAX_ACTIVE_REGIONS);
3879 return;
3880 }
3881
3882 early_node_map[i].nid = nid;
3883 early_node_map[i].start_pfn = start_pfn;
3884 early_node_map[i].end_pfn = end_pfn;
3885 nr_nodemap_entries = i + 1;
3886}
3887
3888/**
Yinghai Lucc1050b2008-06-13 19:08:52 -07003889 * remove_active_range - Shrink an existing registered range of PFNs
Mel Gormanc7132162006-09-27 01:49:43 -07003890 * @nid: The node id the range is on that should be shrunk
Yinghai Lucc1050b2008-06-13 19:08:52 -07003891 * @start_pfn: The new PFN of the range
3892 * @end_pfn: The new PFN of the range
Mel Gormanc7132162006-09-27 01:49:43 -07003893 *
3894 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003895 * The map is kept near the end physical page range that has already been
3896 * registered. This function allows an arch to shrink an existing registered
3897 * range.
Mel Gormanc7132162006-09-27 01:49:43 -07003898 */
Yinghai Lucc1050b2008-06-13 19:08:52 -07003899void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
3900 unsigned long end_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07003901{
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003902 int i, j;
3903 int removed = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07003904
Yinghai Lucc1050b2008-06-13 19:08:52 -07003905 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
3906 nid, start_pfn, end_pfn);
3907
Mel Gormanc7132162006-09-27 01:49:43 -07003908 /* Find the old active region end and shrink */
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003909 for_each_active_range_index_in_nid(i, nid) {
Yinghai Lucc1050b2008-06-13 19:08:52 -07003910 if (early_node_map[i].start_pfn >= start_pfn &&
3911 early_node_map[i].end_pfn <= end_pfn) {
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003912 /* clear it */
Yinghai Lucc1050b2008-06-13 19:08:52 -07003913 early_node_map[i].start_pfn = 0;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003914 early_node_map[i].end_pfn = 0;
3915 removed = 1;
3916 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07003917 }
Yinghai Lucc1050b2008-06-13 19:08:52 -07003918 if (early_node_map[i].start_pfn < start_pfn &&
3919 early_node_map[i].end_pfn > start_pfn) {
3920 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
3921 early_node_map[i].end_pfn = start_pfn;
3922 if (temp_end_pfn > end_pfn)
3923 add_active_range(nid, end_pfn, temp_end_pfn);
3924 continue;
3925 }
3926 if (early_node_map[i].start_pfn >= start_pfn &&
3927 early_node_map[i].end_pfn > end_pfn &&
3928 early_node_map[i].start_pfn < end_pfn) {
3929 early_node_map[i].start_pfn = end_pfn;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003930 continue;
3931 }
3932 }
3933
3934 if (!removed)
3935 return;
3936
3937 /* remove the blank ones */
3938 for (i = nr_nodemap_entries - 1; i > 0; i--) {
3939 if (early_node_map[i].nid != nid)
3940 continue;
3941 if (early_node_map[i].end_pfn)
3942 continue;
3943 /* we found it, get rid of it */
3944 for (j = i; j < nr_nodemap_entries - 1; j++)
3945 memcpy(&early_node_map[j], &early_node_map[j+1],
3946 sizeof(early_node_map[j]));
3947 j = nr_nodemap_entries - 1;
3948 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
3949 nr_nodemap_entries--;
3950 }
Mel Gormanc7132162006-09-27 01:49:43 -07003951}
3952
3953/**
3954 * remove_all_active_ranges - Remove all currently registered regions
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003955 *
Mel Gormanc7132162006-09-27 01:49:43 -07003956 * During discovery, it may be found that a table like SRAT is invalid
3957 * and an alternative discovery method must be used. This function removes
3958 * all currently registered regions.
3959 */
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003960void __init remove_all_active_ranges(void)
Mel Gormanc7132162006-09-27 01:49:43 -07003961{
3962 memset(early_node_map, 0, sizeof(early_node_map));
3963 nr_nodemap_entries = 0;
3964}
3965
3966/* Compare two active node_active_regions */
3967static int __init cmp_node_active_region(const void *a, const void *b)
3968{
3969 struct node_active_region *arange = (struct node_active_region *)a;
3970 struct node_active_region *brange = (struct node_active_region *)b;
3971
3972 /* Done this way to avoid overflows */
3973 if (arange->start_pfn > brange->start_pfn)
3974 return 1;
3975 if (arange->start_pfn < brange->start_pfn)
3976 return -1;
3977
3978 return 0;
3979}
3980
3981/* sort the node_map by start_pfn */
3982static void __init sort_node_map(void)
3983{
3984 sort(early_node_map, (size_t)nr_nodemap_entries,
3985 sizeof(struct node_active_region),
3986 cmp_node_active_region, NULL);
3987}
3988
Mel Gormana6af2bc2007-02-10 01:42:57 -08003989/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003990static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003991{
3992 int i;
Mel Gormana6af2bc2007-02-10 01:42:57 -08003993 unsigned long min_pfn = ULONG_MAX;
Mel Gorman1abbfb42006-11-23 12:01:41 +00003994
Mel Gormanc7132162006-09-27 01:49:43 -07003995 /* Assuming a sorted map, the first range found has the starting pfn */
3996 for_each_active_range_index_in_nid(i, nid)
Mel Gormana6af2bc2007-02-10 01:42:57 -08003997 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003998
Mel Gormana6af2bc2007-02-10 01:42:57 -08003999 if (min_pfn == ULONG_MAX) {
4000 printk(KERN_WARNING
Paul Jackson2bc0d262008-06-22 07:22:17 -07004001 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08004002 return 0;
4003 }
4004
4005 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004006}
4007
4008/**
4009 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4010 *
4011 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004012 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07004013 */
4014unsigned long __init find_min_pfn_with_active_regions(void)
4015{
4016 return find_min_pfn_for_node(MAX_NUMNODES);
4017}
4018
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004019/*
4020 * early_calculate_totalpages()
4021 * Sum pages in active regions for movable zone.
4022 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4023 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07004024static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07004025{
4026 int i;
4027 unsigned long totalpages = 0;
4028
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004029 for (i = 0; i < nr_nodemap_entries; i++) {
4030 unsigned long pages = early_node_map[i].end_pfn -
Mel Gorman7e63efe2007-07-17 04:03:15 -07004031 early_node_map[i].start_pfn;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004032 totalpages += pages;
4033 if (pages)
4034 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
4035 }
4036 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07004037}
4038
Mel Gorman2a1e2742007-07-17 04:03:12 -07004039/*
4040 * Find the PFN the Movable zone begins in each node. Kernel memory
4041 * is spread evenly between nodes as long as the nodes have enough
4042 * memory. When they don't, some nodes will have more kernelcore than
4043 * others
4044 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004045static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004046{
4047 int i, nid;
4048 unsigned long usable_startpfn;
4049 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07004050 /* save the state before borrow the nodemask */
4051 nodemask_t saved_node_state = node_states[N_HIGH_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004052 unsigned long totalpages = early_calculate_totalpages();
4053 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004054
Mel Gorman7e63efe2007-07-17 04:03:15 -07004055 /*
4056 * If movablecore was specified, calculate what size of
4057 * kernelcore that corresponds so that memory usable for
4058 * any allocation type is evenly spread. If both kernelcore
4059 * and movablecore are specified, then the value of kernelcore
4060 * will be used for required_kernelcore if it's greater than
4061 * what movablecore would have allowed.
4062 */
4063 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07004064 unsigned long corepages;
4065
4066 /*
4067 * Round-up so that ZONE_MOVABLE is at least as large as what
4068 * was requested by the user
4069 */
4070 required_movablecore =
4071 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
4072 corepages = totalpages - required_movablecore;
4073
4074 required_kernelcore = max(required_kernelcore, corepages);
4075 }
4076
Mel Gorman2a1e2742007-07-17 04:03:12 -07004077 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4078 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07004079 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004080
4081 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4082 find_usable_zone_for_movable();
4083 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4084
4085restart:
4086 /* Spread kernelcore memory as evenly as possible throughout nodes */
4087 kernelcore_node = required_kernelcore / usable_nodes;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004088 for_each_node_state(nid, N_HIGH_MEMORY) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004089 /*
4090 * Recalculate kernelcore_node if the division per node
4091 * now exceeds what is necessary to satisfy the requested
4092 * amount of memory for the kernel
4093 */
4094 if (required_kernelcore < kernelcore_node)
4095 kernelcore_node = required_kernelcore / usable_nodes;
4096
4097 /*
4098 * As the map is walked, we track how much memory is usable
4099 * by the kernel using kernelcore_remaining. When it is
4100 * 0, the rest of the node is usable by ZONE_MOVABLE
4101 */
4102 kernelcore_remaining = kernelcore_node;
4103
4104 /* Go through each range of PFNs within this node */
4105 for_each_active_range_index_in_nid(i, nid) {
4106 unsigned long start_pfn, end_pfn;
4107 unsigned long size_pages;
4108
4109 start_pfn = max(early_node_map[i].start_pfn,
4110 zone_movable_pfn[nid]);
4111 end_pfn = early_node_map[i].end_pfn;
4112 if (start_pfn >= end_pfn)
4113 continue;
4114
4115 /* Account for what is only usable for kernelcore */
4116 if (start_pfn < usable_startpfn) {
4117 unsigned long kernel_pages;
4118 kernel_pages = min(end_pfn, usable_startpfn)
4119 - start_pfn;
4120
4121 kernelcore_remaining -= min(kernel_pages,
4122 kernelcore_remaining);
4123 required_kernelcore -= min(kernel_pages,
4124 required_kernelcore);
4125
4126 /* Continue if range is now fully accounted */
4127 if (end_pfn <= usable_startpfn) {
4128
4129 /*
4130 * Push zone_movable_pfn to the end so
4131 * that if we have to rebalance
4132 * kernelcore across nodes, we will
4133 * not double account here
4134 */
4135 zone_movable_pfn[nid] = end_pfn;
4136 continue;
4137 }
4138 start_pfn = usable_startpfn;
4139 }
4140
4141 /*
4142 * The usable PFN range for ZONE_MOVABLE is from
4143 * start_pfn->end_pfn. Calculate size_pages as the
4144 * number of pages used as kernelcore
4145 */
4146 size_pages = end_pfn - start_pfn;
4147 if (size_pages > kernelcore_remaining)
4148 size_pages = kernelcore_remaining;
4149 zone_movable_pfn[nid] = start_pfn + size_pages;
4150
4151 /*
4152 * Some kernelcore has been met, update counts and
4153 * break if the kernelcore for this node has been
4154 * satisified
4155 */
4156 required_kernelcore -= min(required_kernelcore,
4157 size_pages);
4158 kernelcore_remaining -= size_pages;
4159 if (!kernelcore_remaining)
4160 break;
4161 }
4162 }
4163
4164 /*
4165 * If there is still required_kernelcore, we do another pass with one
4166 * less node in the count. This will push zone_movable_pfn[nid] further
4167 * along on the nodes that still have memory until kernelcore is
4168 * satisified
4169 */
4170 usable_nodes--;
4171 if (usable_nodes && required_kernelcore > usable_nodes)
4172 goto restart;
4173
4174 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4175 for (nid = 0; nid < MAX_NUMNODES; nid++)
4176 zone_movable_pfn[nid] =
4177 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07004178
4179out:
4180 /* restore the node_state */
4181 node_states[N_HIGH_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004182}
4183
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004184/* Any regular memory on that node ? */
4185static void check_for_regular_memory(pg_data_t *pgdat)
4186{
4187#ifdef CONFIG_HIGHMEM
4188 enum zone_type zone_type;
4189
4190 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4191 struct zone *zone = &pgdat->node_zones[zone_type];
4192 if (zone->present_pages)
4193 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4194 }
4195#endif
4196}
4197
Mel Gormanc7132162006-09-27 01:49:43 -07004198/**
4199 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004200 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07004201 *
4202 * This will call free_area_init_node() for each active node in the system.
4203 * Using the page ranges provided by add_active_range(), the size of each
4204 * zone in each node and their holes is calculated. If the maximum PFN
4205 * between two adjacent zones match, it is assumed that the zone is empty.
4206 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4207 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4208 * starts where the previous one ended. For example, ZONE_DMA32 starts
4209 * at arch_max_dma_pfn.
4210 */
4211void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4212{
4213 unsigned long nid;
Andrew Mortondb991002008-10-15 22:01:07 -07004214 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004215
Mel Gormana6af2bc2007-02-10 01:42:57 -08004216 /* Sort early_node_map as initialisation assumes it is sorted */
4217 sort_node_map();
4218
Mel Gormanc7132162006-09-27 01:49:43 -07004219 /* Record where the zone boundaries are */
4220 memset(arch_zone_lowest_possible_pfn, 0,
4221 sizeof(arch_zone_lowest_possible_pfn));
4222 memset(arch_zone_highest_possible_pfn, 0,
4223 sizeof(arch_zone_highest_possible_pfn));
4224 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4225 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4226 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004227 if (i == ZONE_MOVABLE)
4228 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004229 arch_zone_lowest_possible_pfn[i] =
4230 arch_zone_highest_possible_pfn[i-1];
4231 arch_zone_highest_possible_pfn[i] =
4232 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4233 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07004234 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4235 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4236
4237 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4238 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4239 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004240
Mel Gormanc7132162006-09-27 01:49:43 -07004241 /* Print out the zone ranges */
4242 printk("Zone PFN ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07004243 for (i = 0; i < MAX_NR_ZONES; i++) {
4244 if (i == ZONE_MOVABLE)
4245 continue;
Paul Jackson5dab8ec2008-06-25 05:44:40 -07004246 printk(" %-8s %0#10lx -> %0#10lx\n",
Mel Gormanc7132162006-09-27 01:49:43 -07004247 zone_names[i],
4248 arch_zone_lowest_possible_pfn[i],
4249 arch_zone_highest_possible_pfn[i]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004250 }
4251
4252 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4253 printk("Movable zone start PFN for each node\n");
4254 for (i = 0; i < MAX_NUMNODES; i++) {
4255 if (zone_movable_pfn[i])
4256 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4257 }
Mel Gormanc7132162006-09-27 01:49:43 -07004258
4259 /* Print out the early_node_map[] */
4260 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4261 for (i = 0; i < nr_nodemap_entries; i++)
Paul Jackson5dab8ec2008-06-25 05:44:40 -07004262 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004263 early_node_map[i].start_pfn,
4264 early_node_map[i].end_pfn);
4265
4266 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07004267 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08004268 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07004269 for_each_online_node(nid) {
4270 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07004271 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07004272 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004273
4274 /* Any memory on that node */
4275 if (pgdat->node_present_pages)
4276 node_set_state(nid, N_HIGH_MEMORY);
4277 check_for_regular_memory(pgdat);
Mel Gormanc7132162006-09-27 01:49:43 -07004278 }
4279}
Mel Gorman2a1e2742007-07-17 04:03:12 -07004280
Mel Gorman7e63efe2007-07-17 04:03:15 -07004281static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004282{
4283 unsigned long long coremem;
4284 if (!p)
4285 return -EINVAL;
4286
4287 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004288 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004289
Mel Gorman7e63efe2007-07-17 04:03:15 -07004290 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07004291 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4292
4293 return 0;
4294}
Mel Gormaned7ed362007-07-17 04:03:14 -07004295
Mel Gorman7e63efe2007-07-17 04:03:15 -07004296/*
4297 * kernelcore=size sets the amount of memory for use for allocations that
4298 * cannot be reclaimed or migrated.
4299 */
4300static int __init cmdline_parse_kernelcore(char *p)
4301{
4302 return cmdline_parse_core(p, &required_kernelcore);
4303}
4304
4305/*
4306 * movablecore=size sets the amount of memory for use for allocations that
4307 * can be reclaimed or migrated.
4308 */
4309static int __init cmdline_parse_movablecore(char *p)
4310{
4311 return cmdline_parse_core(p, &required_movablecore);
4312}
4313
Mel Gormaned7ed362007-07-17 04:03:14 -07004314early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004315early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07004316
Mel Gormanc7132162006-09-27 01:49:43 -07004317#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4318
Mel Gorman0e0b8642006-09-27 01:49:56 -07004319/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004320 * set_dma_reserve - set the specified number of pages reserved in the first zone
4321 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07004322 *
4323 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4324 * In the DMA zone, a significant percentage may be consumed by kernel image
4325 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004326 * function may optionally be used to account for unfreeable pages in the
4327 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4328 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07004329 */
4330void __init set_dma_reserve(unsigned long new_dma_reserve)
4331{
4332 dma_reserve = new_dma_reserve;
4333}
4334
Dave Hansen93b75042005-06-23 00:07:47 -07004335#ifndef CONFIG_NEED_MULTIPLE_NODES
Marcin Slusarz52765582008-09-02 14:35:41 -07004336struct pglist_data __refdata contig_page_data = { .bdata = &bootmem_node_data[0] };
Linus Torvalds1da177e2005-04-16 15:20:36 -07004337EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07004338#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004339
4340void __init free_area_init(unsigned long *zones_size)
4341{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004342 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004343 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4344}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004345
Linus Torvalds1da177e2005-04-16 15:20:36 -07004346static int page_alloc_cpu_notify(struct notifier_block *self,
4347 unsigned long action, void *hcpu)
4348{
4349 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004350
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07004351 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004352 drain_pages(cpu);
4353
4354 /*
4355 * Spill the event counters of the dead processor
4356 * into the current processors event counters.
4357 * This artificially elevates the count of the current
4358 * processor.
4359 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07004360 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004361
4362 /*
4363 * Zero the differential counters of the dead processor
4364 * so that the vm statistics are consistent.
4365 *
4366 * This is only okay since the processor is dead and cannot
4367 * race with what we are doing.
4368 */
Christoph Lameter2244b952006-06-30 01:55:33 -07004369 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004370 }
4371 return NOTIFY_OK;
4372}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004373
4374void __init page_alloc_init(void)
4375{
4376 hotcpu_notifier(page_alloc_cpu_notify, 0);
4377}
4378
4379/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004380 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4381 * or min_free_kbytes changes.
4382 */
4383static void calculate_totalreserve_pages(void)
4384{
4385 struct pglist_data *pgdat;
4386 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004387 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004388
4389 for_each_online_pgdat(pgdat) {
4390 for (i = 0; i < MAX_NR_ZONES; i++) {
4391 struct zone *zone = pgdat->node_zones + i;
4392 unsigned long max = 0;
4393
4394 /* Find valid and maximum lowmem_reserve in the zone */
4395 for (j = i; j < MAX_NR_ZONES; j++) {
4396 if (zone->lowmem_reserve[j] > max)
4397 max = zone->lowmem_reserve[j];
4398 }
4399
Mel Gorman41858962009-06-16 15:32:12 -07004400 /* we treat the high watermark as reserved pages. */
4401 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004402
4403 if (max > zone->present_pages)
4404 max = zone->present_pages;
4405 reserve_pages += max;
4406 }
4407 }
4408 totalreserve_pages = reserve_pages;
4409}
4410
4411/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07004412 * setup_per_zone_lowmem_reserve - called whenever
4413 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4414 * has a correct pages reserved value, so an adequate number of
4415 * pages are left in the zone after a successful __alloc_pages().
4416 */
4417static void setup_per_zone_lowmem_reserve(void)
4418{
4419 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004420 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004421
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08004422 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004423 for (j = 0; j < MAX_NR_ZONES; j++) {
4424 struct zone *zone = pgdat->node_zones + j;
4425 unsigned long present_pages = zone->present_pages;
4426
4427 zone->lowmem_reserve[j] = 0;
4428
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004429 idx = j;
4430 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004431 struct zone *lower_zone;
4432
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004433 idx--;
4434
Linus Torvalds1da177e2005-04-16 15:20:36 -07004435 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4436 sysctl_lowmem_reserve_ratio[idx] = 1;
4437
4438 lower_zone = pgdat->node_zones + idx;
4439 lower_zone->lowmem_reserve[j] = present_pages /
4440 sysctl_lowmem_reserve_ratio[idx];
4441 present_pages += lower_zone->present_pages;
4442 }
4443 }
4444 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004445
4446 /* update totalreserve_pages */
4447 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004448}
4449
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004450/**
Minchan Kimbc75d332009-06-16 15:32:48 -07004451 * setup_per_zone_wmarks - called when min_free_kbytes changes
Minchan Kimbce73942009-06-16 15:32:50 -07004452 * or when memory is hot-{added|removed}
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004453 *
Minchan Kimbc75d332009-06-16 15:32:48 -07004454 * Ensures that the watermark[min,low,high] values for each zone are set
4455 * correctly with respect to min_free_kbytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004456 */
Minchan Kimbc75d332009-06-16 15:32:48 -07004457void setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004458{
4459 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4460 unsigned long lowmem_pages = 0;
4461 struct zone *zone;
4462 unsigned long flags;
4463
4464 /* Calculate total number of !ZONE_HIGHMEM pages */
4465 for_each_zone(zone) {
4466 if (!is_highmem(zone))
4467 lowmem_pages += zone->present_pages;
4468 }
4469
4470 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07004471 u64 tmp;
4472
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07004473 spin_lock_irqsave(&zone->lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07004474 tmp = (u64)pages_min * zone->present_pages;
4475 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004476 if (is_highmem(zone)) {
4477 /*
Nick Piggin669ed172005-11-13 16:06:45 -08004478 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4479 * need highmem pages, so cap pages_min to a small
4480 * value here.
4481 *
Mel Gorman41858962009-06-16 15:32:12 -07004482 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Nick Piggin669ed172005-11-13 16:06:45 -08004483 * deltas controls asynch page reclaim, and so should
4484 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004485 */
4486 int min_pages;
4487
4488 min_pages = zone->present_pages / 1024;
4489 if (min_pages < SWAP_CLUSTER_MAX)
4490 min_pages = SWAP_CLUSTER_MAX;
4491 if (min_pages > 128)
4492 min_pages = 128;
Mel Gorman41858962009-06-16 15:32:12 -07004493 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004494 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08004495 /*
4496 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07004497 * proportionate to the zone's size.
4498 */
Mel Gorman41858962009-06-16 15:32:12 -07004499 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004500 }
4501
Mel Gorman41858962009-06-16 15:32:12 -07004502 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
4503 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Mel Gorman56fd56b2007-10-16 01:25:58 -07004504 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07004505 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004506 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004507
4508 /* update totalreserve_pages */
4509 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004510}
4511
Rik van Riel556adec2008-10-18 20:26:34 -07004512/**
Rik van Riel556adec2008-10-18 20:26:34 -07004513 * The inactive anon list should be small enough that the VM never has to
4514 * do too much work, but large enough that each inactive page has a chance
4515 * to be referenced again before it is swapped out.
4516 *
4517 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4518 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4519 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4520 * the anonymous pages are kept on the inactive list.
4521 *
4522 * total target max
4523 * memory ratio inactive anon
4524 * -------------------------------------
4525 * 10MB 1 5MB
4526 * 100MB 1 50MB
4527 * 1GB 3 250MB
4528 * 10GB 10 0.9GB
4529 * 100GB 31 3GB
4530 * 1TB 101 10GB
4531 * 10TB 320 32GB
4532 */
Minchan Kim96cb4df2009-06-16 15:32:49 -07004533void calculate_zone_inactive_ratio(struct zone *zone)
4534{
4535 unsigned int gb, ratio;
4536
4537 /* Zone size in gigabytes */
4538 gb = zone->present_pages >> (30 - PAGE_SHIFT);
4539 if (gb)
4540 ratio = int_sqrt(10 * gb);
4541 else
4542 ratio = 1;
4543
4544 zone->inactive_ratio = ratio;
4545}
4546
Cyrill Gorcunove9bb35d2009-06-16 15:32:32 -07004547static void __init setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07004548{
4549 struct zone *zone;
4550
Minchan Kim96cb4df2009-06-16 15:32:49 -07004551 for_each_zone(zone)
4552 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07004553}
4554
Linus Torvalds1da177e2005-04-16 15:20:36 -07004555/*
4556 * Initialise min_free_kbytes.
4557 *
4558 * For small machines we want it small (128k min). For large machines
4559 * we want it large (64MB max). But it is not linear, because network
4560 * bandwidth does not increase linearly with machine size. We use
4561 *
4562 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4563 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4564 *
4565 * which yields
4566 *
4567 * 16MB: 512k
4568 * 32MB: 724k
4569 * 64MB: 1024k
4570 * 128MB: 1448k
4571 * 256MB: 2048k
4572 * 512MB: 2896k
4573 * 1024MB: 4096k
4574 * 2048MB: 5792k
4575 * 4096MB: 8192k
4576 * 8192MB: 11584k
4577 * 16384MB: 16384k
4578 */
Minchan Kimbc75d332009-06-16 15:32:48 -07004579static int __init init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004580{
4581 unsigned long lowmem_kbytes;
4582
4583 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4584
4585 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4586 if (min_free_kbytes < 128)
4587 min_free_kbytes = 128;
4588 if (min_free_kbytes > 65536)
4589 min_free_kbytes = 65536;
Minchan Kimbc75d332009-06-16 15:32:48 -07004590 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004591 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07004592 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004593 return 0;
4594}
Minchan Kimbc75d332009-06-16 15:32:48 -07004595module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004596
4597/*
4598 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4599 * that we can call two helper functions whenever min_free_kbytes
4600 * changes.
4601 */
4602int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
4603 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4604{
4605 proc_dointvec(table, write, file, buffer, length, ppos);
Mel Gorman3b1d92c2007-05-06 14:49:30 -07004606 if (write)
Minchan Kimbc75d332009-06-16 15:32:48 -07004607 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004608 return 0;
4609}
4610
Christoph Lameter96146342006-07-03 00:24:13 -07004611#ifdef CONFIG_NUMA
4612int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
4613 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4614{
4615 struct zone *zone;
4616 int rc;
4617
4618 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4619 if (rc)
4620 return rc;
4621
4622 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07004623 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07004624 sysctl_min_unmapped_ratio) / 100;
4625 return 0;
4626}
Christoph Lameter0ff38492006-09-25 23:31:52 -07004627
4628int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
4629 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4630{
4631 struct zone *zone;
4632 int rc;
4633
4634 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4635 if (rc)
4636 return rc;
4637
4638 for_each_zone(zone)
4639 zone->min_slab_pages = (zone->present_pages *
4640 sysctl_min_slab_ratio) / 100;
4641 return 0;
4642}
Christoph Lameter96146342006-07-03 00:24:13 -07004643#endif
4644
Linus Torvalds1da177e2005-04-16 15:20:36 -07004645/*
4646 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4647 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4648 * whenever sysctl_lowmem_reserve_ratio changes.
4649 *
4650 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07004651 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07004652 * if in function of the boot time zone sizes.
4653 */
4654int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
4655 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4656{
4657 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4658 setup_per_zone_lowmem_reserve();
4659 return 0;
4660}
4661
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004662/*
4663 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4664 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4665 * can have before it gets flushed back to buddy allocator.
4666 */
4667
4668int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
4669 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4670{
4671 struct zone *zone;
4672 unsigned int cpu;
4673 int ret;
4674
4675 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4676 if (!write || (ret == -EINVAL))
4677 return ret;
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07004678 for_each_populated_zone(zone) {
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004679 for_each_online_cpu(cpu) {
4680 unsigned long high;
4681 high = zone->present_pages / percpu_pagelist_fraction;
4682 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4683 }
4684 }
4685 return 0;
4686}
4687
David S. Millerf034b5d2006-08-24 03:08:07 -07004688int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004689
4690#ifdef CONFIG_NUMA
4691static int __init set_hashdist(char *str)
4692{
4693 if (!str)
4694 return 0;
4695 hashdist = simple_strtoul(str, &str, 0);
4696 return 1;
4697}
4698__setup("hashdist=", set_hashdist);
4699#endif
4700
4701/*
4702 * allocate a large system hash table from bootmem
4703 * - it is assumed that the hash table must contain an exact power-of-2
4704 * quantity of entries
4705 * - limit is the number of hash buckets, not the total allocation size
4706 */
4707void *__init alloc_large_system_hash(const char *tablename,
4708 unsigned long bucketsize,
4709 unsigned long numentries,
4710 int scale,
4711 int flags,
4712 unsigned int *_hash_shift,
4713 unsigned int *_hash_mask,
4714 unsigned long limit)
4715{
4716 unsigned long long max = limit;
4717 unsigned long log2qty, size;
4718 void *table = NULL;
4719
4720 /* allow the kernel cmdline to have a say */
4721 if (!numentries) {
4722 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08004723 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004724 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4725 numentries >>= 20 - PAGE_SHIFT;
4726 numentries <<= 20 - PAGE_SHIFT;
4727
4728 /* limit to 1 bucket per 2^scale bytes of low memory */
4729 if (scale > PAGE_SHIFT)
4730 numentries >>= (scale - PAGE_SHIFT);
4731 else
4732 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08004733
4734 /* Make sure we've got at least a 0-order allocation.. */
4735 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
4736 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004737 }
John Hawkes6e692ed2006-03-25 03:08:02 -08004738 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004739
4740 /* limit allocation size to 1/16 total memory by default */
4741 if (max == 0) {
4742 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4743 do_div(max, bucketsize);
4744 }
4745
4746 if (numentries > max)
4747 numentries = max;
4748
David Howellsf0d1b0b2006-12-08 02:37:49 -08004749 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004750
4751 do {
4752 size = bucketsize << log2qty;
4753 if (flags & HASH_EARLY)
Jan Beulich74768ed2008-08-12 15:08:39 -07004754 table = alloc_bootmem_nopanic(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004755 else if (hashdist)
4756 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4757 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07004758 /*
4759 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07004760 * some pages at the end of hash table which
4761 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07004762 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01004763 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07004764 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01004765 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
4766 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004767 }
4768 } while (!table && size > PAGE_SIZE && --log2qty);
4769
4770 if (!table)
4771 panic("Failed to allocate %s hash table\n", tablename);
4772
Dan Alonib49ad482007-07-15 23:38:23 -07004773 printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07004774 tablename,
4775 (1U << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08004776 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004777 size);
4778
4779 if (_hash_shift)
4780 *_hash_shift = log2qty;
4781 if (_hash_mask)
4782 *_hash_mask = (1 << log2qty) - 1;
4783
4784 return table;
4785}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004786
Mel Gorman835c1342007-10-16 01:25:47 -07004787/* Return a pointer to the bitmap storing bits affecting a block of pages */
4788static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4789 unsigned long pfn)
4790{
4791#ifdef CONFIG_SPARSEMEM
4792 return __pfn_to_section(pfn)->pageblock_flags;
4793#else
4794 return zone->pageblock_flags;
4795#endif /* CONFIG_SPARSEMEM */
4796}
Andrew Morton6220ec72006-10-19 23:29:05 -07004797
Mel Gorman835c1342007-10-16 01:25:47 -07004798static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4799{
4800#ifdef CONFIG_SPARSEMEM
4801 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07004802 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07004803#else
4804 pfn = pfn - zone->zone_start_pfn;
Mel Gormand9c23402007-10-16 01:26:01 -07004805 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07004806#endif /* CONFIG_SPARSEMEM */
4807}
4808
4809/**
Mel Gormand9c23402007-10-16 01:26:01 -07004810 * 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 -07004811 * @page: The page within the block of interest
4812 * @start_bitidx: The first bit of interest to retrieve
4813 * @end_bitidx: The last bit of interest
4814 * returns pageblock_bits flags
4815 */
4816unsigned long get_pageblock_flags_group(struct page *page,
4817 int start_bitidx, int end_bitidx)
4818{
4819 struct zone *zone;
4820 unsigned long *bitmap;
4821 unsigned long pfn, bitidx;
4822 unsigned long flags = 0;
4823 unsigned long value = 1;
4824
4825 zone = page_zone(page);
4826 pfn = page_to_pfn(page);
4827 bitmap = get_pageblock_bitmap(zone, pfn);
4828 bitidx = pfn_to_bitidx(zone, pfn);
4829
4830 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4831 if (test_bit(bitidx + start_bitidx, bitmap))
4832 flags |= value;
4833
4834 return flags;
4835}
4836
4837/**
Mel Gormand9c23402007-10-16 01:26:01 -07004838 * 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 -07004839 * @page: The page within the block of interest
4840 * @start_bitidx: The first bit of interest
4841 * @end_bitidx: The last bit of interest
4842 * @flags: The flags to set
4843 */
4844void set_pageblock_flags_group(struct page *page, unsigned long flags,
4845 int start_bitidx, int end_bitidx)
4846{
4847 struct zone *zone;
4848 unsigned long *bitmap;
4849 unsigned long pfn, bitidx;
4850 unsigned long value = 1;
4851
4852 zone = page_zone(page);
4853 pfn = page_to_pfn(page);
4854 bitmap = get_pageblock_bitmap(zone, pfn);
4855 bitidx = pfn_to_bitidx(zone, pfn);
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004856 VM_BUG_ON(pfn < zone->zone_start_pfn);
4857 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
Mel Gorman835c1342007-10-16 01:25:47 -07004858
4859 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4860 if (flags & value)
4861 __set_bit(bitidx + start_bitidx, bitmap);
4862 else
4863 __clear_bit(bitidx + start_bitidx, bitmap);
4864}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07004865
4866/*
4867 * This is designed as sub function...plz see page_isolation.c also.
4868 * set/clear page block's type to be ISOLATE.
4869 * page allocater never alloc memory from ISOLATE block.
4870 */
4871
4872int set_migratetype_isolate(struct page *page)
4873{
4874 struct zone *zone;
4875 unsigned long flags;
4876 int ret = -EBUSY;
4877
4878 zone = page_zone(page);
4879 spin_lock_irqsave(&zone->lock, flags);
4880 /*
4881 * In future, more migrate types will be able to be isolation target.
4882 */
4883 if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE)
4884 goto out;
4885 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
4886 move_freepages_block(zone, page, MIGRATE_ISOLATE);
4887 ret = 0;
4888out:
4889 spin_unlock_irqrestore(&zone->lock, flags);
4890 if (!ret)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004891 drain_all_pages();
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07004892 return ret;
4893}
4894
4895void unset_migratetype_isolate(struct page *page)
4896{
4897 struct zone *zone;
4898 unsigned long flags;
4899 zone = page_zone(page);
4900 spin_lock_irqsave(&zone->lock, flags);
4901 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
4902 goto out;
4903 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4904 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4905out:
4906 spin_unlock_irqrestore(&zone->lock, flags);
4907}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07004908
4909#ifdef CONFIG_MEMORY_HOTREMOVE
4910/*
4911 * All pages in the range must be isolated before calling this.
4912 */
4913void
4914__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
4915{
4916 struct page *page;
4917 struct zone *zone;
4918 int order, i;
4919 unsigned long pfn;
4920 unsigned long flags;
4921 /* find the first valid pfn */
4922 for (pfn = start_pfn; pfn < end_pfn; pfn++)
4923 if (pfn_valid(pfn))
4924 break;
4925 if (pfn == end_pfn)
4926 return;
4927 zone = page_zone(pfn_to_page(pfn));
4928 spin_lock_irqsave(&zone->lock, flags);
4929 pfn = start_pfn;
4930 while (pfn < end_pfn) {
4931 if (!pfn_valid(pfn)) {
4932 pfn++;
4933 continue;
4934 }
4935 page = pfn_to_page(pfn);
4936 BUG_ON(page_count(page));
4937 BUG_ON(!PageBuddy(page));
4938 order = page_order(page);
4939#ifdef CONFIG_DEBUG_VM
4940 printk(KERN_INFO "remove from free list %lx %d %lx\n",
4941 pfn, 1 << order, end_pfn);
4942#endif
4943 list_del(&page->lru);
4944 rmv_page_order(page);
4945 zone->free_area[order].nr_free--;
4946 __mod_zone_page_state(zone, NR_FREE_PAGES,
4947 - (1UL << order));
4948 for (i = 0; i < (1 << order); i++)
4949 SetPageReserved((page+i));
4950 pfn += (1 << order);
4951 }
4952 spin_unlock_irqrestore(&zone->lock, flags);
4953}
4954#endif