blob: caa92689aac951e5d6c08c89993526898bc695d1 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/page_alloc.c
3 *
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
6 *
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
15 */
16
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/stddef.h>
18#include <linux/mm.h>
19#include <linux/swap.h>
20#include <linux/interrupt.h>
21#include <linux/pagemap.h>
KOSAKI Motohiro10ed2732008-03-04 14:28:32 -080022#include <linux/jiffies.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/bootmem.h>
24#include <linux/compiler.h>
Randy Dunlap9f158332005-09-13 01:25:16 -070025#include <linux/kernel.h>
Vegard Nossumb1eeab62008-11-25 16:55:53 +010026#include <linux/kmemcheck.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/module.h>
28#include <linux/suspend.h>
29#include <linux/pagevec.h>
30#include <linux/blkdev.h>
31#include <linux/slab.h>
David Rientjes5a3135c22007-10-16 23:25:53 -070032#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include <linux/notifier.h>
34#include <linux/topology.h>
35#include <linux/sysctl.h>
36#include <linux/cpu.h>
37#include <linux/cpuset.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070038#include <linux/memory_hotplug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include <linux/nodemask.h>
40#include <linux/vmalloc.h>
Christoph Lameter4be38e32006-01-06 00:11:17 -080041#include <linux/mempolicy.h>
Yasunori Goto68113782006-06-23 02:03:11 -070042#include <linux/stop_machine.h>
Mel Gormanc7132162006-09-27 01:49:43 -070043#include <linux/sort.h>
44#include <linux/pfn.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070045#include <linux/backing-dev.h>
Akinobu Mita933e3122006-12-08 02:39:45 -080046#include <linux/fault-inject.h>
KAMEZAWA 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 Morton6aa3001b22006-04-18 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 Morton6aa3001b22006-04-18 22:20:52 -0700414 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000415 }
Andrew Morton6aa3001b22006-04-18 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 Piggin92be2e332006-01-06 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 Piggin92be2e332006-01-06 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 Piggincc1025092006-12-06 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 Gorman46dafbc2007-10-16 01:25:55 -0700820 start_migratetype == MIGRATE_RECLAIMABLE) {
821 unsigned long pages;
822 pages = move_freepages_block(zone, page,
823 start_migratetype);
824
825 /* Claim the whole block if over half of it is free */
Mel Gormand9c23402007-10-16 01:26:01 -0700826 if (pages >= (1 << (pageblock_order-1)))
Mel Gorman46dafbc2007-10-16 01:25:55 -0700827 set_pageblock_migratetype(page,
828 start_migratetype);
829
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700830 migratetype = start_migratetype;
Mel Gormanc361be52007-10-16 01:25:51 -0700831 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700832
833 /* Remove the page from the freelists */
834 list_del(&page->lru);
835 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700836
Mel Gormand9c23402007-10-16 01:26:01 -0700837 if (current_order == pageblock_order)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700838 set_pageblock_migratetype(page,
839 start_migratetype);
840
841 expand(zone, page, order, current_order, area, migratetype);
842 return page;
843 }
844 }
845
Mel Gorman728ec982009-06-16 15:32:04 -0700846 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700847}
848
Mel Gorman56fd56b2007-10-16 01:25:58 -0700849/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 * Do the hard work of removing an element from the buddy allocator.
851 * Call me with the zone->lock already held.
852 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700853static struct page *__rmqueue(struct zone *zone, unsigned int order,
854 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 struct page *page;
857
Mel Gorman728ec982009-06-16 15:32:04 -0700858retry_reserve:
Mel Gorman56fd56b2007-10-16 01:25:58 -0700859 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860
Mel Gorman728ec982009-06-16 15:32:04 -0700861 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Mel Gorman56fd56b2007-10-16 01:25:58 -0700862 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700863
Mel Gorman728ec982009-06-16 15:32:04 -0700864 /*
865 * Use MIGRATE_RESERVE rather than fail an allocation. goto
866 * is used because __rmqueue_smallest is an inline function
867 * and we want just one call site
868 */
869 if (!page) {
870 migratetype = MIGRATE_RESERVE;
871 goto retry_reserve;
872 }
873 }
874
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700875 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876}
877
878/*
879 * Obtain a specified number of elements from the buddy allocator, all under
880 * a single hold of the lock, for efficiency. Add them to the supplied list.
881 * Returns the number of new pages which were placed at *list.
882 */
883static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700884 unsigned long count, struct list_head *list,
885 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888
Nick Pigginc54ad302006-01-06 00:10:56 -0800889 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700891 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800892 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -0800894
895 /*
896 * Split buddy pages returned by expand() are received here
897 * in physical page order. The page is added to the callers and
898 * list and the list head then moves forward. From the callers
899 * perspective, the linked list is ordered by page number in
900 * some conditions. This is useful for IO devices that can
901 * merge IO requests if the physical pages are ordered
902 * properly.
903 */
Mel Gorman535131e62007-10-16 01:25:49 -0700904 list_add(&page->lru, list);
905 set_page_private(page, migratetype);
Mel Gorman81eabcb2007-12-17 16:20:05 -0800906 list = &page->lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 }
Mel Gormanf2260e62009-06-16 15:32:13 -0700908 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -0800909 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800910 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911}
912
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700913#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800914/*
Christoph Lameter4037d452007-05-09 02:35:14 -0700915 * Called from the vmstat counter updater to drain pagesets of this
916 * currently executing processor on remote nodes after they have
917 * expired.
918 *
Christoph Lameter879336c2006-03-22 00:09:08 -0800919 * Note that this function must be called with the thread pinned to
920 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800921 */
Christoph Lameter4037d452007-05-09 02:35:14 -0700922void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700923{
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700924 unsigned long flags;
Christoph Lameter4037d452007-05-09 02:35:14 -0700925 int to_drain;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700926
Christoph Lameter4037d452007-05-09 02:35:14 -0700927 local_irq_save(flags);
928 if (pcp->count >= pcp->batch)
929 to_drain = pcp->batch;
930 else
931 to_drain = pcp->count;
932 free_pages_bulk(zone, to_drain, &pcp->list, 0);
933 pcp->count -= to_drain;
934 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700935}
936#endif
937
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800938/*
939 * Drain pages of the indicated processor.
940 *
941 * The processor must either be the current processor and the
942 * thread pinned to the current processor or a processor that
943 * is not online.
944 */
945static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946{
Nick Pigginc54ad302006-01-06 00:10:56 -0800947 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949
KOSAKI Motohiroee99c712009-03-31 15:19:31 -0700950 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 struct per_cpu_pageset *pset;
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800952 struct per_cpu_pages *pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700954 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800956 pcp = &pset->pcp;
957 local_irq_save(flags);
958 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
959 pcp->count = 0;
960 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 }
962}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800964/*
965 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
966 */
967void drain_local_pages(void *arg)
968{
969 drain_pages(smp_processor_id());
970}
971
972/*
973 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
974 */
975void drain_all_pages(void)
976{
Jens Axboe15c8b6c2008-05-09 09:39:44 +0200977 on_each_cpu(drain_local_pages, NULL, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800978}
979
Rafael J. Wysocki296699d2007-07-29 23:27:18 +0200980#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981
982void mark_free_pages(struct zone *zone)
983{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700984 unsigned long pfn, max_zone_pfn;
985 unsigned long flags;
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700986 int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 struct list_head *curr;
988
989 if (!zone->spanned_pages)
990 return;
991
992 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700993
994 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
995 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
996 if (pfn_valid(pfn)) {
997 struct page *page = pfn_to_page(pfn);
998
Rafael J. Wysocki7be98232007-05-06 14:50:42 -0700999 if (!swsusp_page_is_forbidden(page))
1000 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001001 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001003 for_each_migratetype_order(order, t) {
1004 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001005 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001007 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1008 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001009 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001010 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001011 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 spin_unlock_irqrestore(&zone->lock, flags);
1013}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001014#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015
1016/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 * Free a 0-order page
1018 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001019static void free_hot_cold_page(struct page *page, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020{
1021 struct zone *zone = page_zone(page);
1022 struct per_cpu_pages *pcp;
1023 unsigned long flags;
Johannes Weinerc2773312009-06-19 19:30:56 +02001024 int wasMlocked = TestClearPageMlocked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001026 kmemcheck_free_shadow(page, 0);
1027
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 if (PageAnon(page))
1029 page->mapping = NULL;
Nick Piggin224abf92006-01-06 00:11:11 -08001030 if (free_pages_check(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001031 return;
1032
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001033 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -07001034 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001035 debug_check_no_obj_freed(page_address(page), PAGE_SIZE);
1036 }
Nick Piggindafb1362006-10-11 01:21:30 -07001037 arch_free_page(page, 0);
Hugh Dickins689bceb2005-11-21 21:32:20 -08001038 kernel_map_pages(page, 1, 0);
1039
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001040 pcp = &zone_pcp(zone, get_cpu())->pcp;
Mel Gorman974709b2009-06-16 15:32:14 -07001041 set_page_private(page, get_pageblock_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +02001043 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -07001044 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001045 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001046
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001047 if (cold)
1048 list_add_tail(&page->lru, &pcp->list);
1049 else
1050 list_add(&page->lru, &pcp->list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001052 if (pcp->count >= pcp->high) {
1053 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
1054 pcp->count -= pcp->batch;
1055 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 local_irq_restore(flags);
1057 put_cpu();
1058}
1059
Harvey Harrison920c7a52008-02-04 22:29:26 -08001060void free_hot_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061{
1062 free_hot_cold_page(page, 0);
1063}
1064
Harvey Harrison920c7a52008-02-04 22:29:26 -08001065void free_cold_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066{
1067 free_hot_cold_page(page, 1);
1068}
1069
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001070/*
1071 * split_page takes a non-compound higher-order page, and splits it into
1072 * n (1<<order) sub-pages: page[0..n]
1073 * Each sub-page must be freed individually.
1074 *
1075 * Note: this is probably too low level an operation for use in drivers.
1076 * Please consult with lkml before using this in your driver.
1077 */
1078void split_page(struct page *page, unsigned int order)
1079{
1080 int i;
1081
Nick Piggin725d7042006-09-25 23:30:55 -07001082 VM_BUG_ON(PageCompound(page));
1083 VM_BUG_ON(!page_count(page));
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001084
1085#ifdef CONFIG_KMEMCHECK
1086 /*
1087 * Split shadow pages too, because free(page[0]) would
1088 * otherwise free the whole shadow.
1089 */
1090 if (kmemcheck_page_is_tracked(page))
1091 split_page(virt_to_page(page[0].shadow), order);
1092#endif
1093
Nick Piggin7835e982006-03-22 00:08:40 -08001094 for (i = 1; i < (1 << order); i++)
1095 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001096}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001097
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098/*
1099 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1100 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1101 * or two.
1102 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07001103static inline
1104struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001105 struct zone *zone, int order, gfp_t gfp_flags,
1106 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107{
1108 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001109 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 int cold = !!(gfp_flags & __GFP_COLD);
Nick Piggina74609f2006-01-06 00:11:20 -08001111 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112
Hugh Dickins689bceb2005-11-21 21:32:20 -08001113again:
Nick Piggina74609f2006-01-06 00:11:20 -08001114 cpu = get_cpu();
Nick Piggin48db57f2006-01-08 01:00:42 -08001115 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 struct per_cpu_pages *pcp;
1117
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001118 pcp = &zone_pcp(zone, cpu)->pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001120 if (!pcp->count) {
nkalmala941c7102006-11-02 22:07:04 -08001121 pcp->count = rmqueue_bulk(zone, 0,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001122 pcp->batch, &pcp->list, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001123 if (unlikely(!pcp->count))
1124 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 }
Mel Gorman535131e62007-10-16 01:25:49 -07001126
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001127 /* Find a page of the appropriate migrate type */
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001128 if (cold) {
1129 list_for_each_entry_reverse(page, &pcp->list, lru)
1130 if (page_private(page) == migratetype)
1131 break;
1132 } else {
1133 list_for_each_entry(page, &pcp->list, lru)
1134 if (page_private(page) == migratetype)
1135 break;
1136 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001137
1138 /* Allocate more to the pcp list if necessary */
1139 if (unlikely(&page->lru == &pcp->list)) {
Mel Gorman535131e62007-10-16 01:25:49 -07001140 pcp->count += rmqueue_bulk(zone, 0,
1141 pcp->batch, &pcp->list, migratetype);
1142 page = list_entry(pcp->list.next, struct page, lru);
Mel Gorman535131e62007-10-16 01:25:49 -07001143 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001144
1145 list_del(&page->lru);
1146 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001147 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07001148 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1149 /*
1150 * __GFP_NOFAIL is not to be used in new code.
1151 *
1152 * All __GFP_NOFAIL callers should be fixed so that they
1153 * properly detect and handle allocation failures.
1154 *
1155 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07001156 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07001157 * __GFP_NOFAIL.
1158 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07001159 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07001160 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001162 page = __rmqueue(zone, order, migratetype);
Mel Gormanf2260e62009-06-16 15:32:13 -07001163 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order));
Nick Piggina74609f2006-01-06 00:11:20 -08001164 spin_unlock(&zone->lock);
1165 if (!page)
1166 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 }
1168
Christoph Lameterf8891e52006-06-30 01:55:45 -07001169 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Mel Gorman18ea7e72008-04-28 02:12:14 -07001170 zone_statistics(preferred_zone, zone);
Nick Piggina74609f2006-01-06 00:11:20 -08001171 local_irq_restore(flags);
1172 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173
Nick Piggin725d7042006-09-25 23:30:55 -07001174 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -08001175 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001176 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001178
1179failed:
1180 local_irq_restore(flags);
1181 put_cpu();
1182 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183}
1184
Mel Gorman41858962009-06-16 15:32:12 -07001185/* The ALLOC_WMARK bits are used as an index to zone->watermark */
1186#define ALLOC_WMARK_MIN WMARK_MIN
1187#define ALLOC_WMARK_LOW WMARK_LOW
1188#define ALLOC_WMARK_HIGH WMARK_HIGH
1189#define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1190
1191/* Mask to get the watermark bits */
1192#define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1193
Nick Piggin31488902005-11-28 13:44:03 -08001194#define ALLOC_HARDER 0x10 /* try to alloc harder */
1195#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1196#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001197
Akinobu Mita933e3122006-12-08 02:39:45 -08001198#ifdef CONFIG_FAIL_PAGE_ALLOC
1199
1200static struct fail_page_alloc_attr {
1201 struct fault_attr attr;
1202
1203 u32 ignore_gfp_highmem;
1204 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001205 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001206
1207#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1208
1209 struct dentry *ignore_gfp_highmem_file;
1210 struct dentry *ignore_gfp_wait_file;
Akinobu Mita54114992007-07-15 23:40:23 -07001211 struct dentry *min_order_file;
Akinobu Mita933e3122006-12-08 02:39:45 -08001212
1213#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1214
1215} fail_page_alloc = {
1216 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001217 .ignore_gfp_wait = 1,
1218 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001219 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001220};
1221
1222static int __init setup_fail_page_alloc(char *str)
1223{
1224 return setup_fault_attr(&fail_page_alloc.attr, str);
1225}
1226__setup("fail_page_alloc=", setup_fail_page_alloc);
1227
1228static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1229{
Akinobu Mita54114992007-07-15 23:40:23 -07001230 if (order < fail_page_alloc.min_order)
1231 return 0;
Akinobu Mita933e3122006-12-08 02:39:45 -08001232 if (gfp_mask & __GFP_NOFAIL)
1233 return 0;
1234 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1235 return 0;
1236 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1237 return 0;
1238
1239 return should_fail(&fail_page_alloc.attr, 1 << order);
1240}
1241
1242#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1243
1244static int __init fail_page_alloc_debugfs(void)
1245{
1246 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1247 struct dentry *dir;
1248 int err;
1249
1250 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1251 "fail_page_alloc");
1252 if (err)
1253 return err;
1254 dir = fail_page_alloc.attr.dentries.dir;
1255
1256 fail_page_alloc.ignore_gfp_wait_file =
1257 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1258 &fail_page_alloc.ignore_gfp_wait);
1259
1260 fail_page_alloc.ignore_gfp_highmem_file =
1261 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1262 &fail_page_alloc.ignore_gfp_highmem);
Akinobu Mita54114992007-07-15 23:40:23 -07001263 fail_page_alloc.min_order_file =
1264 debugfs_create_u32("min-order", mode, dir,
1265 &fail_page_alloc.min_order);
Akinobu Mita933e3122006-12-08 02:39:45 -08001266
1267 if (!fail_page_alloc.ignore_gfp_wait_file ||
Akinobu Mita54114992007-07-15 23:40:23 -07001268 !fail_page_alloc.ignore_gfp_highmem_file ||
1269 !fail_page_alloc.min_order_file) {
Akinobu Mita933e3122006-12-08 02:39:45 -08001270 err = -ENOMEM;
1271 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1272 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
Akinobu Mita54114992007-07-15 23:40:23 -07001273 debugfs_remove(fail_page_alloc.min_order_file);
Akinobu Mita933e3122006-12-08 02:39:45 -08001274 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1275 }
1276
1277 return err;
1278}
1279
1280late_initcall(fail_page_alloc_debugfs);
1281
1282#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1283
1284#else /* CONFIG_FAIL_PAGE_ALLOC */
1285
1286static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1287{
1288 return 0;
1289}
1290
1291#endif /* CONFIG_FAIL_PAGE_ALLOC */
1292
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293/*
1294 * Return 1 if free pages are above 'mark'. This takes into account the order
1295 * of the allocation.
1296 */
1297int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001298 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299{
1300 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001301 long min = mark;
1302 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 int o;
1304
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001305 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001307 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 min -= min / 4;
1309
1310 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1311 return 0;
1312 for (o = 0; o < order; o++) {
1313 /* At the next order, this order's pages become unavailable */
1314 free_pages -= z->free_area[o].nr_free << o;
1315
1316 /* Require fewer higher order pages to be free */
1317 min >>= 1;
1318
1319 if (free_pages <= min)
1320 return 0;
1321 }
1322 return 1;
1323}
1324
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001325#ifdef CONFIG_NUMA
1326/*
1327 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1328 * skip over zones that are not allowed by the cpuset, or that have
1329 * been recently (in last second) found to be nearly full. See further
1330 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001331 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001332 *
1333 * If the zonelist cache is present in the passed in zonelist, then
1334 * returns a pointer to the allowed node mask (either the current
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001335 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001336 *
1337 * If the zonelist cache is not available for this zonelist, does
1338 * nothing and returns NULL.
1339 *
1340 * If the fullzones BITMAP in the zonelist cache is stale (more than
1341 * a second since last zap'd) then we zap it out (clear its bits.)
1342 *
1343 * We hold off even calling zlc_setup, until after we've checked the
1344 * first zone in the zonelist, on the theory that most allocations will
1345 * be satisfied from that first zone, so best to examine that zone as
1346 * quickly as we can.
1347 */
1348static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1349{
1350 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1351 nodemask_t *allowednodes; /* zonelist_cache approximation */
1352
1353 zlc = zonelist->zlcache_ptr;
1354 if (!zlc)
1355 return NULL;
1356
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001357 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001358 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1359 zlc->last_full_zap = jiffies;
1360 }
1361
1362 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1363 &cpuset_current_mems_allowed :
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001364 &node_states[N_HIGH_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001365 return allowednodes;
1366}
1367
1368/*
1369 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1370 * if it is worth looking at further for free memory:
1371 * 1) Check that the zone isn't thought to be full (doesn't have its
1372 * bit set in the zonelist_cache fullzones BITMAP).
1373 * 2) Check that the zones node (obtained from the zonelist_cache
1374 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1375 * Return true (non-zero) if zone is worth looking at further, or
1376 * else return false (zero) if it is not.
1377 *
1378 * This check -ignores- the distinction between various watermarks,
1379 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1380 * found to be full for any variation of these watermarks, it will
1381 * be considered full for up to one second by all requests, unless
1382 * we are so low on memory on all allowed nodes that we are forced
1383 * into the second scan of the zonelist.
1384 *
1385 * In the second scan we ignore this zonelist cache and exactly
1386 * apply the watermarks to all zones, even it is slower to do so.
1387 * We are low on memory in the second scan, and should leave no stone
1388 * unturned looking for a free page.
1389 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001390static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001391 nodemask_t *allowednodes)
1392{
1393 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1394 int i; /* index of *z in zonelist zones */
1395 int n; /* node that zone *z is on */
1396
1397 zlc = zonelist->zlcache_ptr;
1398 if (!zlc)
1399 return 1;
1400
Mel Gormandd1a2392008-04-28 02:12:17 -07001401 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001402 n = zlc->z_to_n[i];
1403
1404 /* This zone is worth trying if it is allowed but not full */
1405 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1406}
1407
1408/*
1409 * Given 'z' scanning a zonelist, set the corresponding bit in
1410 * zlc->fullzones, so that subsequent attempts to allocate a page
1411 * from that zone don't waste time re-examining it.
1412 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001413static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001414{
1415 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1416 int i; /* index of *z in zonelist zones */
1417
1418 zlc = zonelist->zlcache_ptr;
1419 if (!zlc)
1420 return;
1421
Mel Gormandd1a2392008-04-28 02:12:17 -07001422 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001423
1424 set_bit(i, zlc->fullzones);
1425}
1426
1427#else /* CONFIG_NUMA */
1428
1429static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1430{
1431 return NULL;
1432}
1433
Mel Gormandd1a2392008-04-28 02:12:17 -07001434static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001435 nodemask_t *allowednodes)
1436{
1437 return 1;
1438}
1439
Mel Gormandd1a2392008-04-28 02:12:17 -07001440static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001441{
1442}
1443#endif /* CONFIG_NUMA */
1444
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001445/*
Paul Jackson0798e512006-12-06 20:31:38 -08001446 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001447 * a page.
1448 */
1449static struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001450get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
Mel Gorman5117f452009-06-16 15:31:59 -07001451 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
Mel Gorman3dd28262009-06-16 15:32:00 -07001452 struct zone *preferred_zone, int migratetype)
Martin Hicks753ee722005-06-21 17:14:41 -07001453{
Mel Gormandd1a2392008-04-28 02:12:17 -07001454 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001455 struct page *page = NULL;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001456 int classzone_idx;
Mel Gorman5117f452009-06-16 15:31:59 -07001457 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001458 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1459 int zlc_active = 0; /* set if using zonelist_cache */
1460 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gorman54a6eb52008-04-28 02:12:16 -07001461
Mel Gorman19770b32008-04-28 02:12:18 -07001462 classzone_idx = zone_idx(preferred_zone);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001463zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001464 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001465 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001466 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1467 */
Mel Gorman19770b32008-04-28 02:12:18 -07001468 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1469 high_zoneidx, nodemask) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001470 if (NUMA_BUILD && zlc_active &&
1471 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1472 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001473 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001474 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001475 goto try_next_zone;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001476
Mel Gorman41858962009-06-16 15:32:12 -07001477 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001478 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001479 unsigned long mark;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001480 int ret;
1481
Mel Gorman41858962009-06-16 15:32:12 -07001482 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
Mel Gormanfa5e0842009-06-16 15:33:22 -07001483 if (zone_watermark_ok(zone, order, mark,
1484 classzone_idx, alloc_flags))
1485 goto try_this_zone;
1486
1487 if (zone_reclaim_mode == 0)
1488 goto this_zone_full;
1489
1490 ret = zone_reclaim(zone, gfp_mask, order);
1491 switch (ret) {
1492 case ZONE_RECLAIM_NOSCAN:
1493 /* did not scan */
1494 goto try_next_zone;
1495 case ZONE_RECLAIM_FULL:
1496 /* scanned but unreclaimable */
1497 goto this_zone_full;
1498 default:
1499 /* did we reclaim enough */
1500 if (!zone_watermark_ok(zone, order, mark,
1501 classzone_idx, alloc_flags))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001502 goto this_zone_full;
Paul Jackson0798e512006-12-06 20:31:38 -08001503 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001504 }
1505
Mel Gormanfa5e0842009-06-16 15:33:22 -07001506try_this_zone:
Mel Gorman3dd28262009-06-16 15:32:00 -07001507 page = buffered_rmqueue(preferred_zone, zone, order,
1508 gfp_mask, migratetype);
Paul Jackson0798e512006-12-06 20:31:38 -08001509 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001510 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001511this_zone_full:
1512 if (NUMA_BUILD)
1513 zlc_mark_zone_full(zonelist, z);
1514try_next_zone:
Christoph Lameter62bc62a2009-06-16 15:32:15 -07001515 if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) {
Mel Gormand395b732009-06-16 15:32:09 -07001516 /*
1517 * we do zlc_setup after the first zone is tried but only
1518 * if there are multiple nodes make it worthwhile
1519 */
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001520 allowednodes = zlc_setup(zonelist, alloc_flags);
1521 zlc_active = 1;
1522 did_zlc_setup = 1;
1523 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07001524 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001525
1526 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1527 /* Disable zlc cache for second zonelist scan */
1528 zlc_active = 0;
1529 goto zonelist_scan;
1530 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001531 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07001532}
1533
Mel Gorman11e33f62009-06-16 15:31:57 -07001534static inline int
1535should_alloc_retry(gfp_t gfp_mask, unsigned int order,
1536 unsigned long pages_reclaimed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537{
Mel Gorman11e33f62009-06-16 15:31:57 -07001538 /* Do not loop if specifically requested */
1539 if (gfp_mask & __GFP_NORETRY)
1540 return 0;
1541
1542 /*
1543 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1544 * means __GFP_NOFAIL, but that may not be true in other
1545 * implementations.
1546 */
1547 if (order <= PAGE_ALLOC_COSTLY_ORDER)
1548 return 1;
1549
1550 /*
1551 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1552 * specified, then we retry until we no longer reclaim any pages
1553 * (above), or we've reclaimed an order of pages at least as
1554 * large as the allocation's order. In both cases, if the
1555 * allocation still fails, we stop retrying.
1556 */
1557 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
1558 return 1;
1559
1560 /*
1561 * Don't let big-order allocations loop unless the caller
1562 * explicitly requests that.
1563 */
1564 if (gfp_mask & __GFP_NOFAIL)
1565 return 1;
1566
1567 return 0;
1568}
1569
1570static inline struct page *
1571__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1572 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001573 nodemask_t *nodemask, struct zone *preferred_zone,
1574 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001575{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577
Mel Gorman11e33f62009-06-16 15:31:57 -07001578 /* Acquire the OOM killer lock for the zones in zonelist */
1579 if (!try_set_zone_oom(zonelist, gfp_mask)) {
1580 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 return NULL;
1582 }
Jens Axboe6b1de912005-11-17 21:35:02 +01001583
Mel Gorman11e33f62009-06-16 15:31:57 -07001584 /*
1585 * Go through the zonelist yet one more time, keep very high watermark
1586 * here, this is only to catch a parallel oom killing, we must fail if
1587 * we're still under heavy pressure.
1588 */
1589 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1590 order, zonelist, high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001591 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07001592 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001593 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07001594 goto out;
1595
1596 /* The OOM killer will not help higher order allocs */
David Rientjes82553a92009-06-16 15:32:58 -07001597 if (order > PAGE_ALLOC_COSTLY_ORDER && !(gfp_mask & __GFP_NOFAIL))
Mel Gorman11e33f62009-06-16 15:31:57 -07001598 goto out;
1599
1600 /* Exhausted what can be done so it's blamo time */
1601 out_of_memory(zonelist, gfp_mask, order);
1602
1603out:
1604 clear_zonelist_oom(zonelist, gfp_mask);
1605 return page;
1606}
1607
1608/* The really slow allocator path where we enter direct reclaim */
1609static inline struct page *
1610__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
1611 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001612 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001613 int migratetype, unsigned long *did_some_progress)
Mel Gorman11e33f62009-06-16 15:31:57 -07001614{
1615 struct page *page = NULL;
1616 struct reclaim_state reclaim_state;
1617 struct task_struct *p = current;
1618
1619 cond_resched();
1620
1621 /* We now go into synchronous reclaim */
1622 cpuset_memory_pressure_bump();
1623
1624 /*
1625 * The task's cpuset might have expanded its set of allowable nodes
1626 */
1627 p->flags |= PF_MEMALLOC;
1628 lockdep_set_current_reclaim_state(gfp_mask);
1629 reclaim_state.reclaimed_slab = 0;
1630 p->reclaim_state = &reclaim_state;
1631
1632 *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
1633
1634 p->reclaim_state = NULL;
1635 lockdep_clear_current_reclaim_state();
1636 p->flags &= ~PF_MEMALLOC;
1637
1638 cond_resched();
1639
1640 if (order != 0)
1641 drain_all_pages();
1642
1643 if (likely(*did_some_progress))
1644 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001645 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001646 alloc_flags, preferred_zone,
1647 migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001648 return page;
1649}
1650
Mel Gorman11e33f62009-06-16 15:31:57 -07001651/*
1652 * This is called in the allocator slow-path if the allocation request is of
1653 * sufficient urgency to ignore watermarks and take other desperate measures
1654 */
1655static inline struct page *
1656__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
1657 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001658 nodemask_t *nodemask, struct zone *preferred_zone,
1659 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001660{
1661 struct page *page;
1662
1663 do {
1664 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001665 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
Mel Gorman3dd28262009-06-16 15:32:00 -07001666 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001667
1668 if (!page && gfp_mask & __GFP_NOFAIL)
Jens Axboe8aa7e842009-07-09 14:52:32 +02001669 congestion_wait(BLK_RW_ASYNC, HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07001670 } while (!page && (gfp_mask & __GFP_NOFAIL));
1671
1672 return page;
1673}
1674
1675static inline
1676void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
1677 enum zone_type high_zoneidx)
1678{
1679 struct zoneref *z;
1680 struct zone *zone;
1681
1682 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
1683 wakeup_kswapd(zone, order);
1684}
1685
Peter Zijlstra341ce062009-06-16 15:32:02 -07001686static inline int
1687gfp_to_alloc_flags(gfp_t gfp_mask)
1688{
1689 struct task_struct *p = current;
1690 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
1691 const gfp_t wait = gfp_mask & __GFP_WAIT;
1692
Mel Gormana56f57f2009-06-16 15:32:02 -07001693 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
1694 BUILD_BUG_ON(__GFP_HIGH != ALLOC_HIGH);
1695
Peter Zijlstra341ce062009-06-16 15:32:02 -07001696 /*
1697 * The caller may dip into page reserves a bit more if the caller
1698 * cannot run direct reclaim, or if the caller has realtime scheduling
1699 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1700 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1701 */
Mel Gormana56f57f2009-06-16 15:32:02 -07001702 alloc_flags |= (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07001703
1704 if (!wait) {
1705 alloc_flags |= ALLOC_HARDER;
1706 /*
1707 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1708 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1709 */
1710 alloc_flags &= ~ALLOC_CPUSET;
1711 } else if (unlikely(rt_task(p)))
1712 alloc_flags |= ALLOC_HARDER;
1713
1714 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
1715 if (!in_interrupt() &&
1716 ((p->flags & PF_MEMALLOC) ||
1717 unlikely(test_thread_flag(TIF_MEMDIE))))
1718 alloc_flags |= ALLOC_NO_WATERMARKS;
1719 }
1720
1721 return alloc_flags;
1722}
1723
Mel Gorman11e33f62009-06-16 15:31:57 -07001724static inline struct page *
1725__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
1726 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001727 nodemask_t *nodemask, struct zone *preferred_zone,
1728 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001729{
1730 const gfp_t wait = gfp_mask & __GFP_WAIT;
1731 struct page *page = NULL;
1732 int alloc_flags;
1733 unsigned long pages_reclaimed = 0;
1734 unsigned long did_some_progress;
1735 struct task_struct *p = current;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001736
Christoph Lameter952f3b52006-12-06 20:33:26 -08001737 /*
Mel Gorman72807a72009-06-16 15:32:18 -07001738 * In the slowpath, we sanity check order to avoid ever trying to
1739 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
1740 * be using allocators in order of preference for an area that is
1741 * too large.
1742 */
1743 if (WARN_ON_ONCE(order >= MAX_ORDER))
1744 return NULL;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001745
Christoph Lameter952f3b52006-12-06 20:33:26 -08001746 /*
1747 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1748 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1749 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1750 * using a larger set of nodes after it has established that the
1751 * allowed per node queues are empty and that nodes are
1752 * over allocated.
1753 */
1754 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1755 goto nopage;
1756
Mel Gorman11e33f62009-06-16 15:31:57 -07001757 wake_all_kswapd(order, zonelist, high_zoneidx);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001758
Paul Jackson9bf22292005-09-06 15:18:12 -07001759 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001760 * OK, we're below the kswapd watermark and have kicked background
1761 * reclaim. Now things get more complex, so set up alloc_flags according
1762 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07001763 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07001764 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765
Mel Gorman11e33f62009-06-16 15:31:57 -07001766restart:
Peter Zijlstra341ce062009-06-16 15:32:02 -07001767 /* This is the last chance, in general, before the goto nopage. */
Mel Gorman19770b32008-04-28 02:12:18 -07001768 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
Peter Zijlstra341ce062009-06-16 15:32:02 -07001769 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
1770 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001771 if (page)
1772 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773
Kirill Korotaevb43a57b2006-12-06 20:32:27 -08001774rebalance:
Mel Gorman11e33f62009-06-16 15:31:57 -07001775 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07001776 if (alloc_flags & ALLOC_NO_WATERMARKS) {
1777 page = __alloc_pages_high_priority(gfp_mask, order,
1778 zonelist, high_zoneidx, nodemask,
1779 preferred_zone, migratetype);
1780 if (page)
1781 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 }
1783
1784 /* Atomic allocations - we can't balance anything */
1785 if (!wait)
1786 goto nopage;
1787
Peter Zijlstra341ce062009-06-16 15:32:02 -07001788 /* Avoid recursion of direct reclaim */
1789 if (p->flags & PF_MEMALLOC)
1790 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791
Mel Gorman11e33f62009-06-16 15:31:57 -07001792 /* Try direct reclaim and then allocating */
1793 page = __alloc_pages_direct_reclaim(gfp_mask, order,
1794 zonelist, high_zoneidx,
1795 nodemask,
Mel Gorman5117f452009-06-16 15:31:59 -07001796 alloc_flags, preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001797 migratetype, &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07001798 if (page)
1799 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 /*
Mel Gorman11e33f62009-06-16 15:31:57 -07001802 * If we failed to make any progress reclaiming, then we are
1803 * running out of options and have to consider going OOM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 */
Mel Gorman11e33f62009-06-16 15:31:57 -07001805 if (!did_some_progress) {
1806 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -07001807 if (oom_killer_disabled)
1808 goto nopage;
Mel Gorman11e33f62009-06-16 15:31:57 -07001809 page = __alloc_pages_may_oom(gfp_mask, order,
1810 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001811 nodemask, preferred_zone,
1812 migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001813 if (page)
1814 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815
Mel Gorman11e33f62009-06-16 15:31:57 -07001816 /*
David Rientjes82553a92009-06-16 15:32:58 -07001817 * The OOM killer does not trigger for high-order
1818 * ~__GFP_NOFAIL allocations so if no progress is being
1819 * made, there are no other options and retrying is
1820 * unlikely to help.
Mel Gorman11e33f62009-06-16 15:31:57 -07001821 */
David Rientjes82553a92009-06-16 15:32:58 -07001822 if (order > PAGE_ALLOC_COSTLY_ORDER &&
1823 !(gfp_mask & __GFP_NOFAIL))
Mel Gorman11e33f62009-06-16 15:31:57 -07001824 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 goto restart;
1827 }
1828 }
1829
Mel Gorman11e33f62009-06-16 15:31:57 -07001830 /* Check if we should retry the allocation */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07001831 pages_reclaimed += did_some_progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07001832 if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
1833 /* Wait for some write requests to complete then retry */
Jens Axboe8aa7e842009-07-09 14:52:32 +02001834 congestion_wait(BLK_RW_ASYNC, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 goto rebalance;
1836 }
1837
1838nopage:
1839 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1840 printk(KERN_WARNING "%s: page allocation failure."
1841 " order:%d, mode:0x%x\n",
1842 p->comm, order, gfp_mask);
1843 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001844 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 }
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001846 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847got_pg:
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001848 if (kmemcheck_enabled)
1849 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 return page;
Mel Gorman11e33f62009-06-16 15:31:57 -07001851
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852}
Mel Gorman11e33f62009-06-16 15:31:57 -07001853
1854/*
1855 * This is the 'heart' of the zoned buddy allocator.
1856 */
1857struct page *
1858__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
1859 struct zonelist *zonelist, nodemask_t *nodemask)
1860{
1861 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Mel Gorman5117f452009-06-16 15:31:59 -07001862 struct zone *preferred_zone;
Mel Gorman11e33f62009-06-16 15:31:57 -07001863 struct page *page;
Mel Gorman3dd28262009-06-16 15:32:00 -07001864 int migratetype = allocflags_to_migratetype(gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07001865
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10001866 gfp_mask &= gfp_allowed_mask;
1867
Mel Gorman11e33f62009-06-16 15:31:57 -07001868 lockdep_trace_alloc(gfp_mask);
1869
1870 might_sleep_if(gfp_mask & __GFP_WAIT);
1871
1872 if (should_fail_alloc_page(gfp_mask, order))
1873 return NULL;
1874
1875 /*
1876 * Check the zones suitable for the gfp_mask contain at least one
1877 * valid zone. It's possible to have an empty zonelist as a result
1878 * of GFP_THISNODE and a memoryless node
1879 */
1880 if (unlikely(!zonelist->_zonerefs->zone))
1881 return NULL;
1882
Mel Gorman5117f452009-06-16 15:31:59 -07001883 /* The preferred zone is used for statistics later */
1884 first_zones_zonelist(zonelist, high_zoneidx, nodemask, &preferred_zone);
1885 if (!preferred_zone)
1886 return NULL;
1887
1888 /* First allocation attempt */
Mel Gorman11e33f62009-06-16 15:31:57 -07001889 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001890 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07001891 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001892 if (unlikely(!page))
1893 page = __alloc_pages_slowpath(gfp_mask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001894 zonelist, high_zoneidx, nodemask,
Mel Gorman3dd28262009-06-16 15:32:00 -07001895 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001896
1897 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898}
Mel Gormand2391712009-06-16 15:31:52 -07001899EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900
1901/*
1902 * Common helper functions.
1903 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001904unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905{
1906 struct page * page;
1907 page = alloc_pages(gfp_mask, order);
1908 if (!page)
1909 return 0;
1910 return (unsigned long) page_address(page);
1911}
1912
1913EXPORT_SYMBOL(__get_free_pages);
1914
Harvey Harrison920c7a52008-02-04 22:29:26 -08001915unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916{
1917 struct page * page;
1918
1919 /*
1920 * get_zeroed_page() returns a 32-bit address, which cannot represent
1921 * a highmem page
1922 */
Nick Piggin725d7042006-09-25 23:30:55 -07001923 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924
1925 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1926 if (page)
1927 return (unsigned long) page_address(page);
1928 return 0;
1929}
1930
1931EXPORT_SYMBOL(get_zeroed_page);
1932
1933void __pagevec_free(struct pagevec *pvec)
1934{
1935 int i = pagevec_count(pvec);
1936
1937 while (--i >= 0)
1938 free_hot_cold_page(pvec->pages[i], pvec->cold);
1939}
1940
Harvey Harrison920c7a52008-02-04 22:29:26 -08001941void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942{
Nick Pigginb5810032005-10-29 18:16:12 -07001943 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 if (order == 0)
1945 free_hot_page(page);
1946 else
1947 __free_pages_ok(page, order);
1948 }
1949}
1950
1951EXPORT_SYMBOL(__free_pages);
1952
Harvey Harrison920c7a52008-02-04 22:29:26 -08001953void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954{
1955 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07001956 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 __free_pages(virt_to_page((void *)addr), order);
1958 }
1959}
1960
1961EXPORT_SYMBOL(free_pages);
1962
Timur Tabi2be0ffe2008-07-23 21:28:11 -07001963/**
1964 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1965 * @size: the number of bytes to allocate
1966 * @gfp_mask: GFP flags for the allocation
1967 *
1968 * This function is similar to alloc_pages(), except that it allocates the
1969 * minimum number of pages to satisfy the request. alloc_pages() can only
1970 * allocate memory in power-of-two pages.
1971 *
1972 * This function is also limited by MAX_ORDER.
1973 *
1974 * Memory allocated by this function must be released by free_pages_exact().
1975 */
1976void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
1977{
1978 unsigned int order = get_order(size);
1979 unsigned long addr;
1980
1981 addr = __get_free_pages(gfp_mask, order);
1982 if (addr) {
1983 unsigned long alloc_end = addr + (PAGE_SIZE << order);
1984 unsigned long used = addr + PAGE_ALIGN(size);
1985
Kevin Cernekee5bfd7562009-07-05 12:08:19 -07001986 split_page(virt_to_page((void *)addr), order);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07001987 while (used < alloc_end) {
1988 free_page(used);
1989 used += PAGE_SIZE;
1990 }
1991 }
1992
1993 return (void *)addr;
1994}
1995EXPORT_SYMBOL(alloc_pages_exact);
1996
1997/**
1998 * free_pages_exact - release memory allocated via alloc_pages_exact()
1999 * @virt: the value returned by alloc_pages_exact.
2000 * @size: size of allocation, same value as passed to alloc_pages_exact().
2001 *
2002 * Release the memory allocated by a previous call to alloc_pages_exact.
2003 */
2004void free_pages_exact(void *virt, size_t size)
2005{
2006 unsigned long addr = (unsigned long)virt;
2007 unsigned long end = addr + PAGE_ALIGN(size);
2008
2009 while (addr < end) {
2010 free_page(addr);
2011 addr += PAGE_SIZE;
2012 }
2013}
2014EXPORT_SYMBOL(free_pages_exact);
2015
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016static unsigned int nr_free_zone_pages(int offset)
2017{
Mel Gormandd1a2392008-04-28 02:12:17 -07002018 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002019 struct zone *zone;
2020
Martin J. Blighe310fd42005-07-29 22:59:18 -07002021 /* Just pick one node, since fallback list is circular */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 unsigned int sum = 0;
2023
Mel Gorman0e884602008-04-28 02:12:14 -07002024 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025
Mel Gorman54a6eb52008-04-28 02:12:16 -07002026 for_each_zone_zonelist(zone, z, zonelist, offset) {
Martin J. Blighe310fd42005-07-29 22:59:18 -07002027 unsigned long size = zone->present_pages;
Mel Gorman41858962009-06-16 15:32:12 -07002028 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07002029 if (size > high)
2030 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 }
2032
2033 return sum;
2034}
2035
2036/*
2037 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2038 */
2039unsigned int nr_free_buffer_pages(void)
2040{
Al Viroaf4ca452005-10-21 02:55:38 -04002041 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042}
Meelap Shahc2f1a552007-07-17 04:04:39 -07002043EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044
2045/*
2046 * Amount of free RAM allocatable within all zones
2047 */
2048unsigned int nr_free_pagecache_pages(void)
2049{
Mel Gorman2a1e2742007-07-17 04:03:12 -07002050 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002052
2053static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054{
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002055 if (NUMA_BUILD)
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08002056 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059void si_meminfo(struct sysinfo *val)
2060{
2061 val->totalram = totalram_pages;
2062 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002063 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 val->totalhigh = totalhigh_pages;
2066 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 val->mem_unit = PAGE_SIZE;
2068}
2069
2070EXPORT_SYMBOL(si_meminfo);
2071
2072#ifdef CONFIG_NUMA
2073void si_meminfo_node(struct sysinfo *val, int nid)
2074{
2075 pg_data_t *pgdat = NODE_DATA(nid);
2076
2077 val->totalram = pgdat->node_present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002078 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002079#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002081 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2082 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002083#else
2084 val->totalhigh = 0;
2085 val->freehigh = 0;
2086#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 val->mem_unit = PAGE_SIZE;
2088}
2089#endif
2090
2091#define K(x) ((x) << (PAGE_SHIFT-10))
2092
2093/*
2094 * Show free area list (used inside shift_scroll-lock stuff)
2095 * We also calculate the percentage fragmentation. We do this by counting the
2096 * memory on each free list with the exception of the first item on the list.
2097 */
2098void show_free_areas(void)
2099{
Jes Sorensenc7241912006-09-27 01:50:05 -07002100 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101 struct zone *zone;
2102
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002103 for_each_populated_zone(zone) {
Jes Sorensenc7241912006-09-27 01:50:05 -07002104 show_node(zone);
2105 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106
Dave Jones6b482c62005-11-10 15:45:56 -05002107 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 struct per_cpu_pageset *pageset;
2109
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002110 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002112 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2113 cpu, pageset->pcp.high,
2114 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 }
2116 }
2117
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002118 printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n"
2119 " inactive_file:%lu"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002120 " unevictable:%lu"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002121 " dirty:%lu writeback:%lu unstable:%lu\n"
Christoph Lameterd23ad422007-02-10 01:43:02 -08002122 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002123 global_page_state(NR_ACTIVE_ANON),
2124 global_page_state(NR_ACTIVE_FILE),
2125 global_page_state(NR_INACTIVE_ANON),
2126 global_page_state(NR_INACTIVE_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002127 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07002128 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07002129 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07002130 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08002131 global_page_state(NR_FREE_PAGES),
Christoph Lameter972d1a72006-09-25 23:31:51 -07002132 global_page_state(NR_SLAB_RECLAIMABLE) +
2133 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07002134 global_page_state(NR_FILE_MAPPED),
Andrew Mortona25700a2007-02-08 14:20:40 -08002135 global_page_state(NR_PAGETABLE),
2136 global_page_state(NR_BOUNCE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002138 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139 int i;
2140
2141 show_node(zone);
2142 printk("%s"
2143 " free:%lukB"
2144 " min:%lukB"
2145 " low:%lukB"
2146 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002147 " active_anon:%lukB"
2148 " inactive_anon:%lukB"
2149 " active_file:%lukB"
2150 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002151 " unevictable:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 " present:%lukB"
2153 " pages_scanned:%lu"
2154 " all_unreclaimable? %s"
2155 "\n",
2156 zone->name,
Christoph Lameterd23ad422007-02-10 01:43:02 -08002157 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07002158 K(min_wmark_pages(zone)),
2159 K(low_wmark_pages(zone)),
2160 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002161 K(zone_page_state(zone, NR_ACTIVE_ANON)),
2162 K(zone_page_state(zone, NR_INACTIVE_ANON)),
2163 K(zone_page_state(zone, NR_ACTIVE_FILE)),
2164 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002165 K(zone_page_state(zone, NR_UNEVICTABLE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166 K(zone->present_pages),
2167 zone->pages_scanned,
David Rientjese815af92007-10-16 23:25:54 -07002168 (zone_is_all_unreclaimable(zone) ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 );
2170 printk("lowmem_reserve[]:");
2171 for (i = 0; i < MAX_NR_ZONES; i++)
2172 printk(" %lu", zone->lowmem_reserve[i]);
2173 printk("\n");
2174 }
2175
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002176 for_each_populated_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002177 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178
2179 show_node(zone);
2180 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181
2182 spin_lock_irqsave(&zone->lock, flags);
2183 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002184 nr[order] = zone->free_area[order].nr_free;
2185 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186 }
2187 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002188 for (order = 0; order < MAX_ORDER; order++)
2189 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190 printk("= %lukB\n", K(total));
2191 }
2192
Larry Woodmane6f36022008-02-04 22:29:30 -08002193 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2194
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 show_swap_cache_info();
2196}
2197
Mel Gorman19770b32008-04-28 02:12:18 -07002198static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2199{
2200 zoneref->zone = zone;
2201 zoneref->zone_idx = zone_idx(zone);
2202}
2203
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204/*
2205 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08002206 *
2207 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002209static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2210 int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211{
Christoph Lameter1a932052006-01-06 00:11:16 -08002212 struct zone *zone;
2213
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002214 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002215 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08002216
2217 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002218 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08002219 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08002220 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002221 zoneref_set_zone(zone,
2222 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08002223 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08002225
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002226 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08002227 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228}
2229
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002230
2231/*
2232 * zonelist_order:
2233 * 0 = automatic detection of better ordering.
2234 * 1 = order by ([node] distance, -zonetype)
2235 * 2 = order by (-zonetype, [node] distance)
2236 *
2237 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2238 * the same zonelist. So only NUMA can configure this param.
2239 */
2240#define ZONELIST_ORDER_DEFAULT 0
2241#define ZONELIST_ORDER_NODE 1
2242#define ZONELIST_ORDER_ZONE 2
2243
2244/* zonelist order in the kernel.
2245 * set_zonelist_order() will set this to NODE or ZONE.
2246 */
2247static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2248static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2249
2250
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002252/* The value user specified ....changed by config */
2253static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2254/* string for sysctl */
2255#define NUMA_ZONELIST_ORDER_LEN 16
2256char numa_zonelist_order[16] = "default";
2257
2258/*
2259 * interface for configure zonelist ordering.
2260 * command line option "numa_zonelist_order"
2261 * = "[dD]efault - default, automatic configuration.
2262 * = "[nN]ode - order by node locality, then by zone within node
2263 * = "[zZ]one - order by zone, then by locality within zone
2264 */
2265
2266static int __parse_numa_zonelist_order(char *s)
2267{
2268 if (*s == 'd' || *s == 'D') {
2269 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2270 } else if (*s == 'n' || *s == 'N') {
2271 user_zonelist_order = ZONELIST_ORDER_NODE;
2272 } else if (*s == 'z' || *s == 'Z') {
2273 user_zonelist_order = ZONELIST_ORDER_ZONE;
2274 } else {
2275 printk(KERN_WARNING
2276 "Ignoring invalid numa_zonelist_order value: "
2277 "%s\n", s);
2278 return -EINVAL;
2279 }
2280 return 0;
2281}
2282
2283static __init int setup_numa_zonelist_order(char *s)
2284{
2285 if (s)
2286 return __parse_numa_zonelist_order(s);
2287 return 0;
2288}
2289early_param("numa_zonelist_order", setup_numa_zonelist_order);
2290
2291/*
2292 * sysctl handler for numa_zonelist_order
2293 */
2294int numa_zonelist_order_handler(ctl_table *table, int write,
2295 struct file *file, void __user *buffer, size_t *length,
2296 loff_t *ppos)
2297{
2298 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2299 int ret;
2300
2301 if (write)
2302 strncpy(saved_string, (char*)table->data,
2303 NUMA_ZONELIST_ORDER_LEN);
2304 ret = proc_dostring(table, write, file, buffer, length, ppos);
2305 if (ret)
2306 return ret;
2307 if (write) {
2308 int oldval = user_zonelist_order;
2309 if (__parse_numa_zonelist_order((char*)table->data)) {
2310 /*
2311 * bogus value. restore saved string
2312 */
2313 strncpy((char*)table->data, saved_string,
2314 NUMA_ZONELIST_ORDER_LEN);
2315 user_zonelist_order = oldval;
2316 } else if (oldval != user_zonelist_order)
2317 build_all_zonelists();
2318 }
2319 return 0;
2320}
2321
2322
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002323#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002324static int node_load[MAX_NUMNODES];
2325
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07002327 * 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 -07002328 * @node: node whose fallback list we're appending
2329 * @used_node_mask: nodemask_t of already used nodes
2330 *
2331 * We use a number of factors to determine which is the next node that should
2332 * appear on a given node's fallback list. The node should not have appeared
2333 * already in @node's fallback list, and it should be the next closest node
2334 * according to the distance array (which contains arbitrary distance values
2335 * from each node to each node in the system), and should also prefer nodes
2336 * with no CPUs, since presumably they'll have very little allocation pressure
2337 * on them otherwise.
2338 * It returns -1 if no node is found.
2339 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002340static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002342 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 int min_val = INT_MAX;
2344 int best_node = -1;
Rusty Russella70f7302009-03-13 14:49:46 +10302345 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002347 /* Use the local node if we haven't already */
2348 if (!node_isset(node, *used_node_mask)) {
2349 node_set(node, *used_node_mask);
2350 return node;
2351 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002353 for_each_node_state(n, N_HIGH_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354
2355 /* Don't want a node to appear more than once */
2356 if (node_isset(n, *used_node_mask))
2357 continue;
2358
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 /* Use the distance array to find the distance */
2360 val = node_distance(node, n);
2361
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002362 /* Penalize nodes under us ("prefer the next node") */
2363 val += (n < node);
2364
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10302366 tmp = cpumask_of_node(n);
2367 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368 val += PENALTY_FOR_NODE_WITH_CPUS;
2369
2370 /* Slight preference for less loaded node */
2371 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2372 val += node_load[n];
2373
2374 if (val < min_val) {
2375 min_val = val;
2376 best_node = n;
2377 }
2378 }
2379
2380 if (best_node >= 0)
2381 node_set(best_node, *used_node_mask);
2382
2383 return best_node;
2384}
2385
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002386
2387/*
2388 * Build zonelists ordered by node and zones within node.
2389 * This results in maximum locality--normal zone overflows into local
2390 * DMA zone, if any--but risks exhausting DMA zone.
2391 */
2392static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002394 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002396
Mel Gorman54a6eb52008-04-28 02:12:16 -07002397 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07002398 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07002399 ;
2400 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2401 MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002402 zonelist->_zonerefs[j].zone = NULL;
2403 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002404}
2405
2406/*
Christoph Lameter523b9452007-10-16 01:25:37 -07002407 * Build gfp_thisnode zonelists
2408 */
2409static void build_thisnode_zonelists(pg_data_t *pgdat)
2410{
Christoph Lameter523b9452007-10-16 01:25:37 -07002411 int j;
2412 struct zonelist *zonelist;
2413
Mel Gorman54a6eb52008-04-28 02:12:16 -07002414 zonelist = &pgdat->node_zonelists[1];
2415 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002416 zonelist->_zonerefs[j].zone = NULL;
2417 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07002418}
2419
2420/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002421 * Build zonelists ordered by zone and nodes within zones.
2422 * This results in conserving DMA zone[s] until all Normal memory is
2423 * exhausted, but results in overflowing to remote node while memory
2424 * may still exist in local DMA zone.
2425 */
2426static int node_order[MAX_NUMNODES];
2427
2428static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2429{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002430 int pos, j, node;
2431 int zone_type; /* needs to be signed */
2432 struct zone *z;
2433 struct zonelist *zonelist;
2434
Mel Gorman54a6eb52008-04-28 02:12:16 -07002435 zonelist = &pgdat->node_zonelists[0];
2436 pos = 0;
2437 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2438 for (j = 0; j < nr_nodes; j++) {
2439 node = node_order[j];
2440 z = &NODE_DATA(node)->node_zones[zone_type];
2441 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002442 zoneref_set_zone(z,
2443 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002444 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002445 }
2446 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002447 }
Mel Gormandd1a2392008-04-28 02:12:17 -07002448 zonelist->_zonerefs[pos].zone = NULL;
2449 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002450}
2451
2452static int default_zonelist_order(void)
2453{
2454 int nid, zone_type;
2455 unsigned long low_kmem_size,total_size;
2456 struct zone *z;
2457 int average_size;
2458 /*
2459 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2460 * If they are really small and used heavily, the system can fall
2461 * into OOM very easily.
2462 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2463 */
2464 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2465 low_kmem_size = 0;
2466 total_size = 0;
2467 for_each_online_node(nid) {
2468 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2469 z = &NODE_DATA(nid)->node_zones[zone_type];
2470 if (populated_zone(z)) {
2471 if (zone_type < ZONE_NORMAL)
2472 low_kmem_size += z->present_pages;
2473 total_size += z->present_pages;
2474 }
2475 }
2476 }
2477 if (!low_kmem_size || /* there are no DMA area. */
2478 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2479 return ZONELIST_ORDER_NODE;
2480 /*
2481 * look into each node's config.
2482 * If there is a node whose DMA/DMA32 memory is very big area on
2483 * local memory, NODE_ORDER may be suitable.
2484 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002485 average_size = total_size /
2486 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002487 for_each_online_node(nid) {
2488 low_kmem_size = 0;
2489 total_size = 0;
2490 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2491 z = &NODE_DATA(nid)->node_zones[zone_type];
2492 if (populated_zone(z)) {
2493 if (zone_type < ZONE_NORMAL)
2494 low_kmem_size += z->present_pages;
2495 total_size += z->present_pages;
2496 }
2497 }
2498 if (low_kmem_size &&
2499 total_size > average_size && /* ignore small node */
2500 low_kmem_size > total_size * 70/100)
2501 return ZONELIST_ORDER_NODE;
2502 }
2503 return ZONELIST_ORDER_ZONE;
2504}
2505
2506static void set_zonelist_order(void)
2507{
2508 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2509 current_zonelist_order = default_zonelist_order();
2510 else
2511 current_zonelist_order = user_zonelist_order;
2512}
2513
2514static void build_zonelists(pg_data_t *pgdat)
2515{
2516 int j, node, load;
2517 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002519 int local_node, prev_node;
2520 struct zonelist *zonelist;
2521 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522
2523 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07002524 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07002526 zonelist->_zonerefs[0].zone = NULL;
2527 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528 }
2529
2530 /* NUMA-aware ordering of nodes */
2531 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002532 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533 prev_node = local_node;
2534 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002535
2536 memset(node_load, 0, sizeof(node_load));
2537 memset(node_order, 0, sizeof(node_order));
2538 j = 0;
2539
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08002541 int distance = node_distance(local_node, node);
2542
2543 /*
2544 * If another node is sufficiently far away then it is better
2545 * to reclaim pages in a zone before going off node.
2546 */
2547 if (distance > RECLAIM_DISTANCE)
2548 zone_reclaim_mode = 1;
2549
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550 /*
2551 * We don't want to pressure a particular node.
2552 * So adding penalty to the first node in same
2553 * distance group to make it round-robin.
2554 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08002555 if (distance != node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002556 node_load[node] = load;
2557
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 prev_node = node;
2559 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002560 if (order == ZONELIST_ORDER_NODE)
2561 build_zonelists_in_node_order(pgdat, node);
2562 else
2563 node_order[j++] = node; /* remember order */
2564 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002566 if (order == ZONELIST_ORDER_ZONE) {
2567 /* calculate node order -- i.e., DMA last! */
2568 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569 }
Christoph Lameter523b9452007-10-16 01:25:37 -07002570
2571 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572}
2573
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002574/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002575static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002576{
Mel Gorman54a6eb52008-04-28 02:12:16 -07002577 struct zonelist *zonelist;
2578 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07002579 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002580
Mel Gorman54a6eb52008-04-28 02:12:16 -07002581 zonelist = &pgdat->node_zonelists[0];
2582 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2583 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07002584 for (z = zonelist->_zonerefs; z->zone; z++)
2585 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002586}
2587
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002588
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589#else /* CONFIG_NUMA */
2590
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002591static void set_zonelist_order(void)
2592{
2593 current_zonelist_order = ZONELIST_ORDER_ZONE;
2594}
2595
2596static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597{
Christoph Lameter19655d32006-09-25 23:31:19 -07002598 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002599 enum zone_type j;
2600 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601
2602 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603
Mel Gorman54a6eb52008-04-28 02:12:16 -07002604 zonelist = &pgdat->node_zonelists[0];
2605 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606
Mel Gorman54a6eb52008-04-28 02:12:16 -07002607 /*
2608 * Now we build the zonelist so that it contains the zones
2609 * of all the other nodes.
2610 * We don't want to pressure a particular node, so when
2611 * building the zones for node N, we make sure that the
2612 * zones coming right after the local ones are those from
2613 * node N+1 (modulo N)
2614 */
2615 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2616 if (!node_online(node))
2617 continue;
2618 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2619 MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07002621 for (node = 0; node < local_node; node++) {
2622 if (!node_online(node))
2623 continue;
2624 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2625 MAX_NR_ZONES - 1);
2626 }
2627
Mel Gormandd1a2392008-04-28 02:12:17 -07002628 zonelist->_zonerefs[j].zone = NULL;
2629 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630}
2631
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002632/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002633static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002634{
Mel Gorman54a6eb52008-04-28 02:12:16 -07002635 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002636}
2637
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638#endif /* CONFIG_NUMA */
2639
Rusty Russell9b1a4d32008-07-28 12:16:30 -05002640/* return values int ....just for stop_machine() */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002641static int __build_all_zonelists(void *dummy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642{
Yasunori Goto68113782006-06-23 02:03:11 -07002643 int nid;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002644
2645 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07002646 pg_data_t *pgdat = NODE_DATA(nid);
2647
2648 build_zonelists(pgdat);
2649 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002650 }
Yasunori Goto68113782006-06-23 02:03:11 -07002651 return 0;
2652}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002654void build_all_zonelists(void)
Yasunori Goto68113782006-06-23 02:03:11 -07002655{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002656 set_zonelist_order();
2657
Yasunori Goto68113782006-06-23 02:03:11 -07002658 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07002659 __build_all_zonelists(NULL);
Mel Gorman68ad8df2008-07-23 21:26:52 -07002660 mminit_verify_zonelist();
Yasunori Goto68113782006-06-23 02:03:11 -07002661 cpuset_init_current_mems_allowed();
2662 } else {
Simon Arlott183ff222007-10-20 01:27:18 +02002663 /* we have to stop all cpus to guarantee there is no user
Yasunori Goto68113782006-06-23 02:03:11 -07002664 of zonelist */
Rusty Russell9b1a4d32008-07-28 12:16:30 -05002665 stop_machine(__build_all_zonelists, NULL, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07002666 /* cpuset refresh routine should be here */
2667 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07002668 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002669 /*
2670 * Disable grouping by mobility if the number of pages in the
2671 * system is too low to allow the mechanism to work. It would be
2672 * more accurate, but expensive to check per-zone. This check is
2673 * made on memory-hotadd so a system can start with mobility
2674 * disabled and enable it later
2675 */
Mel Gormand9c23402007-10-16 01:26:01 -07002676 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002677 page_group_by_mobility_disabled = 1;
2678 else
2679 page_group_by_mobility_disabled = 0;
2680
2681 printk("Built %i zonelists in %s order, mobility grouping %s. "
2682 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002683 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002684 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002685 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002686 vm_total_pages);
2687#ifdef CONFIG_NUMA
2688 printk("Policy zone: %s\n", zone_names[policy_zone]);
2689#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690}
2691
2692/*
2693 * Helper functions to size the waitqueue hash table.
2694 * Essentially these want to choose hash table sizes sufficiently
2695 * large so that collisions trying to wait on pages are rare.
2696 * But in fact, the number of active page waitqueues on typical
2697 * systems is ridiculously low, less than 200. So this is even
2698 * conservative, even though it seems large.
2699 *
2700 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2701 * waitqueues, i.e. the size of the waitq table given the number of pages.
2702 */
2703#define PAGES_PER_WAITQUEUE 256
2704
Yasunori Gotocca448f2006-06-23 02:03:10 -07002705#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07002706static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707{
2708 unsigned long size = 1;
2709
2710 pages /= PAGES_PER_WAITQUEUE;
2711
2712 while (size < pages)
2713 size <<= 1;
2714
2715 /*
2716 * Once we have dozens or even hundreds of threads sleeping
2717 * on IO we've got bigger problems than wait queue collision.
2718 * Limit the size of the wait table to a reasonable size.
2719 */
2720 size = min(size, 4096UL);
2721
2722 return max(size, 4UL);
2723}
Yasunori Gotocca448f2006-06-23 02:03:10 -07002724#else
2725/*
2726 * A zone's size might be changed by hot-add, so it is not possible to determine
2727 * a suitable size for its wait_table. So we use the maximum size now.
2728 *
2729 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2730 *
2731 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2732 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2733 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2734 *
2735 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2736 * or more by the traditional way. (See above). It equals:
2737 *
2738 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2739 * ia64(16K page size) : = ( 8G + 4M)byte.
2740 * powerpc (64K page size) : = (32G +16M)byte.
2741 */
2742static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2743{
2744 return 4096UL;
2745}
2746#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747
2748/*
2749 * This is an integer logarithm so that shifts can be used later
2750 * to extract the more random high bits from the multiplicative
2751 * hash function before the remainder is taken.
2752 */
2753static inline unsigned long wait_table_bits(unsigned long size)
2754{
2755 return ffz(~size);
2756}
2757
2758#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2759
Mel Gorman56fd56b2007-10-16 01:25:58 -07002760/*
Mel Gormand9c23402007-10-16 01:26:01 -07002761 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07002762 * of blocks reserved is based on min_wmark_pages(zone). The memory within
2763 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07002764 * higher will lead to a bigger reserve which will get freed as contiguous
2765 * blocks as reclaim kicks in
2766 */
2767static void setup_zone_migrate_reserve(struct zone *zone)
2768{
2769 unsigned long start_pfn, pfn, end_pfn;
2770 struct page *page;
2771 unsigned long reserve, block_migratetype;
2772
2773 /* Get the start pfn, end pfn and the number of blocks to reserve */
2774 start_pfn = zone->zone_start_pfn;
2775 end_pfn = start_pfn + zone->spanned_pages;
Mel Gorman41858962009-06-16 15:32:12 -07002776 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07002777 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07002778
Mel Gormand9c23402007-10-16 01:26:01 -07002779 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07002780 if (!pfn_valid(pfn))
2781 continue;
2782 page = pfn_to_page(pfn);
2783
Adam Litke344c7902008-09-02 14:35:38 -07002784 /* Watch out for overlapping nodes */
2785 if (page_to_nid(page) != zone_to_nid(zone))
2786 continue;
2787
Mel Gorman56fd56b2007-10-16 01:25:58 -07002788 /* Blocks with reserved pages will never free, skip them. */
2789 if (PageReserved(page))
2790 continue;
2791
2792 block_migratetype = get_pageblock_migratetype(page);
2793
2794 /* If this block is reserved, account for it */
2795 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
2796 reserve--;
2797 continue;
2798 }
2799
2800 /* Suitable for reserving if this block is movable */
2801 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
2802 set_pageblock_migratetype(page, MIGRATE_RESERVE);
2803 move_freepages_block(zone, page, MIGRATE_RESERVE);
2804 reserve--;
2805 continue;
2806 }
2807
2808 /*
2809 * If the reserve is met and this is a previous reserved block,
2810 * take it back
2811 */
2812 if (block_migratetype == MIGRATE_RESERVE) {
2813 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2814 move_freepages_block(zone, page, MIGRATE_MOVABLE);
2815 }
2816 }
2817}
Mel Gormanac0e5b72007-10-16 01:25:58 -07002818
Linus Torvalds1da177e2005-04-16 15:20:36 -07002819/*
2820 * Initially all pages are reserved - free ones are freed
2821 * up by free_all_bootmem() once the early boot process is
2822 * done. Non-atomic initialization, single-pass.
2823 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002824void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08002825 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002827 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07002828 unsigned long end_pfn = start_pfn + size;
2829 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002830 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831
Hugh Dickins22b31ee2009-01-06 14:40:09 -08002832 if (highest_memmap_pfn < end_pfn - 1)
2833 highest_memmap_pfn = end_pfn - 1;
2834
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002835 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08002836 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08002837 /*
2838 * There can be holes in boot-time mem_map[]s
2839 * handed to this function. They do not
2840 * exist on hotplugged memory.
2841 */
2842 if (context == MEMMAP_EARLY) {
2843 if (!early_pfn_valid(pfn))
2844 continue;
2845 if (!early_pfn_in_nid(pfn, nid))
2846 continue;
2847 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002848 page = pfn_to_page(pfn);
2849 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07002850 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08002851 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852 reset_page_mapcount(page);
2853 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002854 /*
2855 * Mark the block movable so that blocks are reserved for
2856 * movable at startup. This will force kernel allocations
2857 * to reserve their blocks rather than leaking throughout
2858 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07002859 * kernel allocations are made. Later some blocks near
2860 * the start are marked MIGRATE_RESERVE by
2861 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002862 *
2863 * bitmap is created for zone's valid pfn range. but memmap
2864 * can be created for invalid pages (for alignment)
2865 * check here not to call set_pageblock_migratetype() against
2866 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002867 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002868 if ((z->zone_start_pfn <= pfn)
2869 && (pfn < z->zone_start_pfn + z->spanned_pages)
2870 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07002871 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002872
Linus Torvalds1da177e2005-04-16 15:20:36 -07002873 INIT_LIST_HEAD(&page->lru);
2874#ifdef WANT_PAGE_VIRTUAL
2875 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2876 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07002877 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879 }
2880}
2881
Andi Kleen1e548de2008-02-04 22:29:26 -08002882static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883{
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002884 int order, t;
2885 for_each_migratetype_order(order, t) {
2886 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002887 zone->free_area[order].nr_free = 0;
2888 }
2889}
2890
2891#ifndef __HAVE_ARCH_MEMMAP_INIT
2892#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08002893 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002894#endif
2895
Sam Ravnborg1d6f4e62008-01-17 15:21:12 -08002896static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002897{
David Howells3a6be872009-05-06 16:03:03 -07002898#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002899 int batch;
2900
2901 /*
2902 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07002903 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002904 *
2905 * OK, so we don't know how big the cache is. So guess.
2906 */
2907 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07002908 if (batch * PAGE_SIZE > 512 * 1024)
2909 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002910 batch /= 4; /* We effectively *= 4 below */
2911 if (batch < 1)
2912 batch = 1;
2913
2914 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002915 * Clamp the batch to a 2^n - 1 value. Having a power
2916 * of 2 value was found to be more likely to have
2917 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002918 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002919 * For example if 2 tasks are alternately allocating
2920 * batches of pages, one task can end up with a lot
2921 * of pages of one half of the possible page colors
2922 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002923 */
David Howells91552032009-05-06 16:03:02 -07002924 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07002925
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002926 return batch;
David Howells3a6be872009-05-06 16:03:03 -07002927
2928#else
2929 /* The deferral and batching of frees should be suppressed under NOMMU
2930 * conditions.
2931 *
2932 * The problem is that NOMMU needs to be able to allocate large chunks
2933 * of contiguous memory as there's no hardware page translation to
2934 * assemble apparent contiguous memory from discontiguous pages.
2935 *
2936 * Queueing large contiguous runs of pages for batching, however,
2937 * causes the pages to actually be freed in smaller chunks. As there
2938 * can be a significant delay between the individual batches being
2939 * recycled, this leads to the once large chunks of space being
2940 * fragmented and becoming unavailable for high-order allocations.
2941 */
2942 return 0;
2943#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002944}
2945
Adrian Bunkb69a7282008-07-23 21:28:12 -07002946static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
Christoph Lameter2caaad42005-06-21 17:15:00 -07002947{
2948 struct per_cpu_pages *pcp;
2949
Magnus Damm1c6fe942005-10-26 01:58:59 -07002950 memset(p, 0, sizeof(*p));
2951
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002952 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07002953 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07002954 pcp->high = 6 * batch;
2955 pcp->batch = max(1UL, 1 * batch);
2956 INIT_LIST_HEAD(&pcp->list);
Christoph Lameter2caaad42005-06-21 17:15:00 -07002957}
2958
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002959/*
2960 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2961 * to the value high for the pageset p.
2962 */
2963
2964static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2965 unsigned long high)
2966{
2967 struct per_cpu_pages *pcp;
2968
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002969 pcp = &p->pcp;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002970 pcp->high = high;
2971 pcp->batch = max(1UL, high/4);
2972 if ((high/4) > (PAGE_SHIFT * 8))
2973 pcp->batch = PAGE_SHIFT * 8;
2974}
2975
2976
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002977#ifdef CONFIG_NUMA
2978/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07002979 * Boot pageset table. One per cpu which is going to be used for all
2980 * zones and all nodes. The parameters will be set in such a way
2981 * that an item put on a list will immediately be handed over to
2982 * the buddy list. This is safe since pageset manipulation is done
2983 * with interrupts disabled.
2984 *
2985 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07002986 *
2987 * The boot_pagesets must be kept even after bootup is complete for
2988 * unused processors and/or zones. They do play a role for bootstrapping
2989 * hotplugged processors.
2990 *
2991 * zoneinfo_show() and maybe other functions do
2992 * not check if the processor is online before following the pageset pointer.
2993 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07002994 */
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08002995static struct per_cpu_pageset boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07002996
2997/*
2998 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002999 * per cpu pageset array in struct zone.
3000 */
Ashok Raj6292d9a2006-02-01 03:04:44 -08003001static int __cpuinit process_zones(int cpu)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003002{
3003 struct zone *zone, *dzone;
Christoph Lameter37c07082007-10-16 01:25:36 -07003004 int node = cpu_to_node(cpu);
3005
3006 node_set_state(node, N_CPU); /* this node has a cpu */
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003007
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003008 for_each_populated_zone(zone) {
Nick Piggin23316bc2006-01-08 01:00:41 -08003009 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lameter37c07082007-10-16 01:25:36 -07003010 GFP_KERNEL, node);
Nick Piggin23316bc2006-01-08 01:00:41 -08003011 if (!zone_pcp(zone, cpu))
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003012 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003013
Nick Piggin23316bc2006-01-08 01:00:41 -08003014 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003015
3016 if (percpu_pagelist_fraction)
3017 setup_pagelist_highmark(zone_pcp(zone, cpu),
3018 (zone->present_pages / percpu_pagelist_fraction));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003019 }
3020
3021 return 0;
3022bad:
3023 for_each_zone(dzone) {
Andrew Morton64191682007-08-30 23:56:17 -07003024 if (!populated_zone(dzone))
3025 continue;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003026 if (dzone == zone)
3027 break;
Nick Piggin23316bc2006-01-08 01:00:41 -08003028 kfree(zone_pcp(dzone, cpu));
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07003029 zone_pcp(dzone, cpu) = &boot_pageset[cpu];
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003030 }
3031 return -ENOMEM;
3032}
3033
3034static inline void free_zone_pagesets(int cpu)
3035{
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003036 struct zone *zone;
3037
3038 for_each_zone(zone) {
3039 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
3040
David Rientjesf3ef9ea2006-09-25 16:24:57 -07003041 /* Free per_cpu_pageset if it is slab allocated */
3042 if (pset != &boot_pageset[cpu])
3043 kfree(pset);
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07003044 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003045 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003046}
3047
Chandra Seetharaman9c7b2162006-06-27 02:54:07 -07003048static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003049 unsigned long action,
3050 void *hcpu)
3051{
3052 int cpu = (long)hcpu;
3053 int ret = NOTIFY_OK;
3054
3055 switch (action) {
Andy Whitcroftce421c72006-12-06 20:33:08 -08003056 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003057 case CPU_UP_PREPARE_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08003058 if (process_zones(cpu))
3059 ret = NOTIFY_BAD;
3060 break;
3061 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003062 case CPU_UP_CANCELED_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08003063 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003064 case CPU_DEAD_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08003065 free_zone_pagesets(cpu);
3066 break;
3067 default:
3068 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003069 }
3070 return ret;
3071}
3072
Chandra Seetharaman74b85f32006-06-27 02:54:09 -07003073static struct notifier_block __cpuinitdata pageset_notifier =
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003074 { &pageset_cpuup_callback, NULL, 0 };
3075
Al Viro78d99552005-12-15 09:18:25 +00003076void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003077{
3078 int err;
3079
3080 /* Initialize per_cpu_pageset for cpu 0.
3081 * A cpuup callback will do this for every cpu
3082 * as it comes online
3083 */
3084 err = process_zones(smp_processor_id());
3085 BUG_ON(err);
3086 register_cpu_notifier(&pageset_notifier);
3087}
3088
3089#endif
3090
Sam Ravnborg577a32f2007-05-17 23:29:25 +02003091static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07003092int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07003093{
3094 int i;
3095 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003096 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07003097
3098 /*
3099 * The per-page waitqueue mechanism uses hashed waitqueues
3100 * per zone.
3101 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07003102 zone->wait_table_hash_nr_entries =
3103 wait_table_hash_nr_entries(zone_size_pages);
3104 zone->wait_table_bits =
3105 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003106 alloc_size = zone->wait_table_hash_nr_entries
3107 * sizeof(wait_queue_head_t);
3108
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07003109 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07003110 zone->wait_table = (wait_queue_head_t *)
3111 alloc_bootmem_node(pgdat, alloc_size);
3112 } else {
3113 /*
3114 * This case means that a zone whose size was 0 gets new memory
3115 * via memory hot-add.
3116 * But it may be the case that a new node was hot-added. In
3117 * this case vmalloc() will not be able to use this new node's
3118 * memory - this wait_table must be initialized to use this new
3119 * node itself as well.
3120 * To use this new node's memory, further consideration will be
3121 * necessary.
3122 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07003123 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003124 }
3125 if (!zone->wait_table)
3126 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07003127
Yasunori Goto02b694d2006-06-23 02:03:08 -07003128 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07003129 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003130
3131 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003132}
3133
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003134static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07003135{
3136 int cpu;
3137 unsigned long batch = zone_batchsize(zone);
3138
3139 for (cpu = 0; cpu < NR_CPUS; cpu++) {
3140#ifdef CONFIG_NUMA
3141 /* Early boot. Slab allocator not functional yet */
Nick Piggin23316bc2006-01-08 01:00:41 -08003142 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Dave Hansened8ece22005-10-29 18:16:50 -07003143 setup_pageset(&boot_pageset[cpu],0);
3144#else
3145 setup_pageset(zone_pcp(zone,cpu), batch);
3146#endif
3147 }
Anton Blanchardf5335c02006-03-25 03:06:49 -08003148 if (zone->present_pages)
3149 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
3150 zone->name, zone->present_pages, batch);
Dave Hansened8ece22005-10-29 18:16:50 -07003151}
3152
Yasunori Goto718127c2006-06-23 02:03:10 -07003153__meminit int init_currently_empty_zone(struct zone *zone,
3154 unsigned long zone_start_pfn,
Dave Hansena2f3aa022007-01-10 23:15:30 -08003155 unsigned long size,
3156 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07003157{
3158 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003159 int ret;
3160 ret = zone_wait_table_init(zone, size);
3161 if (ret)
3162 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07003163 pgdat->nr_zones = zone_idx(zone) + 1;
3164
Dave Hansened8ece22005-10-29 18:16:50 -07003165 zone->zone_start_pfn = zone_start_pfn;
3166
Mel Gorman708614e2008-07-23 21:26:51 -07003167 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3168 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3169 pgdat->node_id,
3170 (unsigned long)zone_idx(zone),
3171 zone_start_pfn, (zone_start_pfn + size));
3172
Andi Kleen1e548de2008-02-04 22:29:26 -08003173 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07003174
3175 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003176}
3177
Mel Gormanc7132162006-09-27 01:49:43 -07003178#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3179/*
3180 * Basic iterator support. Return the first range of PFNs for a node
3181 * Note: nid == MAX_NUMNODES returns first region regardless of node
3182 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003183static int __meminit first_active_region_index_in_nid(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003184{
3185 int i;
3186
3187 for (i = 0; i < nr_nodemap_entries; i++)
3188 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3189 return i;
3190
3191 return -1;
3192}
3193
3194/*
3195 * Basic iterator support. Return the next active range of PFNs for a node
Simon Arlott183ff222007-10-20 01:27:18 +02003196 * Note: nid == MAX_NUMNODES returns next region regardless of node
Mel Gormanc7132162006-09-27 01:49:43 -07003197 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003198static int __meminit next_active_region_index_in_nid(int index, int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003199{
3200 for (index = index + 1; index < nr_nodemap_entries; index++)
3201 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3202 return index;
3203
3204 return -1;
3205}
3206
3207#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3208/*
3209 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3210 * Architectures may implement their own version but if add_active_range()
3211 * was used and there are no special requirements, this is a convenient
3212 * alternative
3213 */
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003214int __meminit __early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07003215{
3216 int i;
3217
3218 for (i = 0; i < nr_nodemap_entries; i++) {
3219 unsigned long start_pfn = early_node_map[i].start_pfn;
3220 unsigned long end_pfn = early_node_map[i].end_pfn;
3221
3222 if (start_pfn <= pfn && pfn < end_pfn)
3223 return early_node_map[i].nid;
3224 }
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003225 /* This is a memory hole */
3226 return -1;
Mel Gormanc7132162006-09-27 01:49:43 -07003227}
3228#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3229
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003230int __meminit early_pfn_to_nid(unsigned long pfn)
3231{
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003232 int nid;
3233
3234 nid = __early_pfn_to_nid(pfn);
3235 if (nid >= 0)
3236 return nid;
3237 /* just returns 0 */
3238 return 0;
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003239}
3240
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003241#ifdef CONFIG_NODES_SPAN_OTHER_NODES
3242bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3243{
3244 int nid;
3245
3246 nid = __early_pfn_to_nid(pfn);
3247 if (nid >= 0 && nid != node)
3248 return false;
3249 return true;
3250}
3251#endif
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003252
Mel Gormanc7132162006-09-27 01:49:43 -07003253/* Basic iterator support to walk early_node_map[] */
3254#define for_each_active_range_index_in_nid(i, nid) \
3255 for (i = first_active_region_index_in_nid(nid); i != -1; \
3256 i = next_active_region_index_in_nid(i, nid))
3257
3258/**
3259 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003260 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3261 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07003262 *
3263 * If an architecture guarantees that all ranges registered with
3264 * add_active_ranges() contain no holes and may be freed, this
3265 * this function may be used instead of calling free_bootmem() manually.
3266 */
3267void __init free_bootmem_with_active_regions(int nid,
3268 unsigned long max_low_pfn)
3269{
3270 int i;
3271
3272 for_each_active_range_index_in_nid(i, nid) {
3273 unsigned long size_pages = 0;
3274 unsigned long end_pfn = early_node_map[i].end_pfn;
3275
3276 if (early_node_map[i].start_pfn >= max_low_pfn)
3277 continue;
3278
3279 if (end_pfn > max_low_pfn)
3280 end_pfn = max_low_pfn;
3281
3282 size_pages = end_pfn - early_node_map[i].start_pfn;
3283 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3284 PFN_PHYS(early_node_map[i].start_pfn),
3285 size_pages << PAGE_SHIFT);
3286 }
3287}
3288
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003289void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3290{
3291 int i;
Yinghai Lud52d53b2008-06-16 20:10:55 -07003292 int ret;
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003293
Yinghai Lud52d53b2008-06-16 20:10:55 -07003294 for_each_active_range_index_in_nid(i, nid) {
3295 ret = work_fn(early_node_map[i].start_pfn,
3296 early_node_map[i].end_pfn, data);
3297 if (ret)
3298 break;
3299 }
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003300}
Mel Gormanc7132162006-09-27 01:49:43 -07003301/**
3302 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003303 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07003304 *
3305 * If an architecture guarantees that all ranges registered with
3306 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003307 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07003308 */
3309void __init sparse_memory_present_with_active_regions(int nid)
3310{
3311 int i;
3312
3313 for_each_active_range_index_in_nid(i, nid)
3314 memory_present(early_node_map[i].nid,
3315 early_node_map[i].start_pfn,
3316 early_node_map[i].end_pfn);
3317}
3318
3319/**
3320 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003321 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3322 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3323 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07003324 *
3325 * It returns the start and end page frame of a node based on information
3326 * provided by an arch calling add_active_range(). If called for a node
3327 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003328 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07003329 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003330void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003331 unsigned long *start_pfn, unsigned long *end_pfn)
3332{
3333 int i;
3334 *start_pfn = -1UL;
3335 *end_pfn = 0;
3336
3337 for_each_active_range_index_in_nid(i, nid) {
3338 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3339 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3340 }
3341
Christoph Lameter633c0662007-10-16 01:25:37 -07003342 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07003343 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07003344}
3345
3346/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07003347 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3348 * assumption is made that zones within a node are ordered in monotonic
3349 * increasing memory addresses so that the "highest" populated zone is used
3350 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003351static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07003352{
3353 int zone_index;
3354 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3355 if (zone_index == ZONE_MOVABLE)
3356 continue;
3357
3358 if (arch_zone_highest_possible_pfn[zone_index] >
3359 arch_zone_lowest_possible_pfn[zone_index])
3360 break;
3361 }
3362
3363 VM_BUG_ON(zone_index == -1);
3364 movable_zone = zone_index;
3365}
3366
3367/*
3368 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3369 * because it is sized independant of architecture. Unlike the other zones,
3370 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3371 * in each node depending on the size of each node and how evenly kernelcore
3372 * is distributed. This helper function adjusts the zone ranges
3373 * provided by the architecture for a given node by using the end of the
3374 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3375 * zones within a node are in order of monotonic increases memory addresses
3376 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003377static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07003378 unsigned long zone_type,
3379 unsigned long node_start_pfn,
3380 unsigned long node_end_pfn,
3381 unsigned long *zone_start_pfn,
3382 unsigned long *zone_end_pfn)
3383{
3384 /* Only adjust if ZONE_MOVABLE is on this node */
3385 if (zone_movable_pfn[nid]) {
3386 /* Size ZONE_MOVABLE */
3387 if (zone_type == ZONE_MOVABLE) {
3388 *zone_start_pfn = zone_movable_pfn[nid];
3389 *zone_end_pfn = min(node_end_pfn,
3390 arch_zone_highest_possible_pfn[movable_zone]);
3391
3392 /* Adjust for ZONE_MOVABLE starting within this range */
3393 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3394 *zone_end_pfn > zone_movable_pfn[nid]) {
3395 *zone_end_pfn = zone_movable_pfn[nid];
3396
3397 /* Check if this whole range is within ZONE_MOVABLE */
3398 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3399 *zone_start_pfn = *zone_end_pfn;
3400 }
3401}
3402
3403/*
Mel Gormanc7132162006-09-27 01:49:43 -07003404 * Return the number of pages a zone spans in a node, including holes
3405 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3406 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003407static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003408 unsigned long zone_type,
3409 unsigned long *ignored)
3410{
3411 unsigned long node_start_pfn, node_end_pfn;
3412 unsigned long zone_start_pfn, zone_end_pfn;
3413
3414 /* Get the start and end of the node and zone */
3415 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3416 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3417 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07003418 adjust_zone_range_for_zone_movable(nid, zone_type,
3419 node_start_pfn, node_end_pfn,
3420 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003421
3422 /* Check that this node has pages within the zone's required range */
3423 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3424 return 0;
3425
3426 /* Move the zone boundaries inside the node if necessary */
3427 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3428 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3429
3430 /* Return the spanned pages */
3431 return zone_end_pfn - zone_start_pfn;
3432}
3433
3434/*
3435 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003436 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07003437 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003438static unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003439 unsigned long range_start_pfn,
3440 unsigned long range_end_pfn)
3441{
3442 int i = 0;
3443 unsigned long prev_end_pfn = 0, hole_pages = 0;
3444 unsigned long start_pfn;
3445
3446 /* Find the end_pfn of the first active range of pfns in the node */
3447 i = first_active_region_index_in_nid(nid);
3448 if (i == -1)
3449 return 0;
3450
Mel Gormanb5445f92007-07-26 10:41:18 -07003451 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3452
Mel Gorman9c7cd682006-09-27 01:49:58 -07003453 /* Account for ranges before physical memory on this node */
3454 if (early_node_map[i].start_pfn > range_start_pfn)
Mel Gormanb5445f92007-07-26 10:41:18 -07003455 hole_pages = prev_end_pfn - range_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003456
3457 /* Find all holes for the zone within the node */
3458 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3459
3460 /* No need to continue if prev_end_pfn is outside the zone */
3461 if (prev_end_pfn >= range_end_pfn)
3462 break;
3463
3464 /* Make sure the end of the zone is not within the hole */
3465 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3466 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3467
3468 /* Update the hole size cound and move on */
3469 if (start_pfn > range_start_pfn) {
3470 BUG_ON(prev_end_pfn > start_pfn);
3471 hole_pages += start_pfn - prev_end_pfn;
3472 }
3473 prev_end_pfn = early_node_map[i].end_pfn;
3474 }
3475
Mel Gorman9c7cd682006-09-27 01:49:58 -07003476 /* Account for ranges past physical memory on this node */
3477 if (range_end_pfn > prev_end_pfn)
Mel Gorman0c6cb972006-10-28 10:38:59 -07003478 hole_pages += range_end_pfn -
Mel Gorman9c7cd682006-09-27 01:49:58 -07003479 max(range_start_pfn, prev_end_pfn);
3480
Mel Gormanc7132162006-09-27 01:49:43 -07003481 return hole_pages;
3482}
3483
3484/**
3485 * absent_pages_in_range - Return number of page frames in holes within a range
3486 * @start_pfn: The start PFN to start searching for holes
3487 * @end_pfn: The end PFN to stop searching for holes
3488 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003489 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07003490 */
3491unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3492 unsigned long end_pfn)
3493{
3494 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3495}
3496
3497/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003498static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003499 unsigned long zone_type,
3500 unsigned long *ignored)
3501{
Mel Gorman9c7cd682006-09-27 01:49:58 -07003502 unsigned long node_start_pfn, node_end_pfn;
3503 unsigned long zone_start_pfn, zone_end_pfn;
3504
3505 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3506 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3507 node_start_pfn);
3508 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3509 node_end_pfn);
3510
Mel Gorman2a1e2742007-07-17 04:03:12 -07003511 adjust_zone_range_for_zone_movable(nid, zone_type,
3512 node_start_pfn, node_end_pfn,
3513 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07003514 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003515}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003516
Mel Gormanc7132162006-09-27 01:49:43 -07003517#else
Paul Mundt6ea6e682007-07-15 23:38:20 -07003518static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003519 unsigned long zone_type,
3520 unsigned long *zones_size)
3521{
3522 return zones_size[zone_type];
3523}
3524
Paul Mundt6ea6e682007-07-15 23:38:20 -07003525static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003526 unsigned long zone_type,
3527 unsigned long *zholes_size)
3528{
3529 if (!zholes_size)
3530 return 0;
3531
3532 return zholes_size[zone_type];
3533}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003534
Mel Gormanc7132162006-09-27 01:49:43 -07003535#endif
3536
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003537static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Mel Gormanc7132162006-09-27 01:49:43 -07003538 unsigned long *zones_size, unsigned long *zholes_size)
3539{
3540 unsigned long realtotalpages, totalpages = 0;
3541 enum zone_type i;
3542
3543 for (i = 0; i < MAX_NR_ZONES; i++)
3544 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3545 zones_size);
3546 pgdat->node_spanned_pages = totalpages;
3547
3548 realtotalpages = totalpages;
3549 for (i = 0; i < MAX_NR_ZONES; i++)
3550 realtotalpages -=
3551 zone_absent_pages_in_node(pgdat->node_id, i,
3552 zholes_size);
3553 pgdat->node_present_pages = realtotalpages;
3554 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3555 realtotalpages);
3556}
3557
Mel Gorman835c1342007-10-16 01:25:47 -07003558#ifndef CONFIG_SPARSEMEM
3559/*
3560 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07003561 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3562 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07003563 * round what is now in bits to nearest long in bits, then return it in
3564 * bytes.
3565 */
3566static unsigned long __init usemap_size(unsigned long zonesize)
3567{
3568 unsigned long usemapsize;
3569
Mel Gormand9c23402007-10-16 01:26:01 -07003570 usemapsize = roundup(zonesize, pageblock_nr_pages);
3571 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07003572 usemapsize *= NR_PAGEBLOCK_BITS;
3573 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3574
3575 return usemapsize / 8;
3576}
3577
3578static void __init setup_usemap(struct pglist_data *pgdat,
3579 struct zone *zone, unsigned long zonesize)
3580{
3581 unsigned long usemapsize = usemap_size(zonesize);
3582 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08003583 if (usemapsize)
Mel Gorman835c1342007-10-16 01:25:47 -07003584 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
Mel Gorman835c1342007-10-16 01:25:47 -07003585}
3586#else
3587static void inline setup_usemap(struct pglist_data *pgdat,
3588 struct zone *zone, unsigned long zonesize) {}
3589#endif /* CONFIG_SPARSEMEM */
3590
Mel Gormand9c23402007-10-16 01:26:01 -07003591#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08003592
3593/* Return a sensible default order for the pageblock size. */
3594static inline int pageblock_default_order(void)
3595{
3596 if (HPAGE_SHIFT > PAGE_SHIFT)
3597 return HUGETLB_PAGE_ORDER;
3598
3599 return MAX_ORDER-1;
3600}
3601
Mel Gormand9c23402007-10-16 01:26:01 -07003602/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3603static inline void __init set_pageblock_order(unsigned int order)
3604{
3605 /* Check that pageblock_nr_pages has not already been setup */
3606 if (pageblock_order)
3607 return;
3608
3609 /*
3610 * Assume the largest contiguous order of interest is a huge page.
3611 * This value may be variable depending on boot parameters on IA64
3612 */
3613 pageblock_order = order;
3614}
3615#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3616
Mel Gormanba72cb82007-11-28 16:21:13 -08003617/*
3618 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3619 * and pageblock_default_order() are unused as pageblock_order is set
3620 * at compile-time. See include/linux/pageblock-flags.h for the values of
3621 * pageblock_order based on the kernel config
3622 */
3623static inline int pageblock_default_order(unsigned int order)
3624{
3625 return MAX_ORDER-1;
3626}
Mel Gormand9c23402007-10-16 01:26:01 -07003627#define set_pageblock_order(x) do {} while (0)
3628
3629#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3630
Linus Torvalds1da177e2005-04-16 15:20:36 -07003631/*
3632 * Set up the zone data structures:
3633 * - mark all pages reserved
3634 * - mark all memory queues empty
3635 * - clear the memory bitmaps
3636 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08003637static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003638 unsigned long *zones_size, unsigned long *zholes_size)
3639{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07003640 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07003641 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003642 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07003643 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003644
Dave Hansen208d54e2005-10-29 18:16:52 -07003645 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003646 pgdat->nr_zones = 0;
3647 init_waitqueue_head(&pgdat->kswapd_wait);
3648 pgdat->kswapd_max_order = 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003649 pgdat_page_cgroup_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003650
3651 for (j = 0; j < MAX_NR_ZONES; j++) {
3652 struct zone *zone = pgdat->node_zones + j;
Mel Gorman0e0b8642006-09-27 01:49:56 -07003653 unsigned long size, realsize, memmap_pages;
Christoph Lameterb69408e2008-10-18 20:26:14 -07003654 enum lru_list l;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003655
Mel Gormanc7132162006-09-27 01:49:43 -07003656 size = zone_spanned_pages_in_node(nid, j, zones_size);
3657 realsize = size - zone_absent_pages_in_node(nid, j,
3658 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003659
Mel Gorman0e0b8642006-09-27 01:49:56 -07003660 /*
3661 * Adjust realsize so that it accounts for how much memory
3662 * is used by this zone for memmap. This affects the watermark
3663 * and per-cpu initialisations
3664 */
Johannes Weinerf7232152008-05-23 13:04:21 -07003665 memmap_pages =
3666 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
Mel Gorman0e0b8642006-09-27 01:49:56 -07003667 if (realsize >= memmap_pages) {
3668 realsize -= memmap_pages;
Yinghai Lu5594c8c2009-01-06 14:39:14 -08003669 if (memmap_pages)
3670 printk(KERN_DEBUG
3671 " %s zone: %lu pages used for memmap\n",
3672 zone_names[j], memmap_pages);
Mel Gorman0e0b8642006-09-27 01:49:56 -07003673 } else
3674 printk(KERN_WARNING
3675 " %s zone: %lu pages exceeds realsize %lu\n",
3676 zone_names[j], memmap_pages, realsize);
3677
Christoph Lameter62672762007-02-10 01:43:07 -08003678 /* Account for reserved pages */
3679 if (j == 0 && realsize > dma_reserve) {
Mel Gorman0e0b8642006-09-27 01:49:56 -07003680 realsize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07003681 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08003682 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07003683 }
3684
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003685 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003686 nr_kernel_pages += realsize;
3687 nr_all_pages += realsize;
3688
3689 zone->spanned_pages = size;
3690 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07003691#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07003692 zone->node = nid;
Christoph Lameter8417bba2006-09-25 23:31:51 -07003693 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07003694 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07003695 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07003696#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003697 zone->name = zone_names[j];
3698 spin_lock_init(&zone->lock);
3699 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07003700 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003701 zone->zone_pgdat = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003702
Martin Bligh3bb1a8522006-10-28 10:38:24 -07003703 zone->prev_priority = DEF_PRIORITY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003704
Dave Hansened8ece22005-10-29 18:16:50 -07003705 zone_pcp_init(zone);
Christoph Lameterb69408e2008-10-18 20:26:14 -07003706 for_each_lru(l) {
3707 INIT_LIST_HEAD(&zone->lru[l].list);
Wu Fengguang6e08a362009-06-16 15:32:29 -07003708 zone->lru[l].nr_saved_scan = 0;
Christoph Lameterb69408e2008-10-18 20:26:14 -07003709 }
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08003710 zone->reclaim_stat.recent_rotated[0] = 0;
3711 zone->reclaim_stat.recent_rotated[1] = 0;
3712 zone->reclaim_stat.recent_scanned[0] = 0;
3713 zone->reclaim_stat.recent_scanned[1] = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07003714 zap_zone_vm_stats(zone);
David Rientjese815af92007-10-16 23:25:54 -07003715 zone->flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003716 if (!size)
3717 continue;
3718
Mel Gormanba72cb82007-11-28 16:21:13 -08003719 set_pageblock_order(pageblock_default_order());
Mel Gorman835c1342007-10-16 01:25:47 -07003720 setup_usemap(pgdat, zone, size);
Dave Hansena2f3aa022007-01-10 23:15:30 -08003721 ret = init_currently_empty_zone(zone, zone_start_pfn,
3722 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07003723 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07003724 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003725 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003726 }
3727}
3728
Sam Ravnborg577a32f2007-05-17 23:29:25 +02003729static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003730{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003731 /* Skip empty nodes */
3732 if (!pgdat->node_spanned_pages)
3733 return;
3734
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003735#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07003736 /* ia64 gets its own node_mem_map, before this, without bootmem */
3737 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07003738 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003739 struct page *map;
3740
Bob Piccoe984bb42006-05-20 15:00:31 -07003741 /*
3742 * The zone's endpoints aren't required to be MAX_ORDER
3743 * aligned but the node_mem_map endpoints must be in order
3744 * for the buddy allocator to function correctly.
3745 */
3746 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3747 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3748 end = ALIGN(end, MAX_ORDER_NR_PAGES);
3749 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07003750 map = alloc_remap(pgdat->node_id, size);
3751 if (!map)
3752 map = alloc_bootmem_node(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07003753 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003754 }
Roman Zippel12d810c2007-05-31 00:40:54 -07003755#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07003756 /*
3757 * With no DISCONTIG, the global mem_map is just set as node 0's
3758 */
Mel Gormanc7132162006-09-27 01:49:43 -07003759 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003760 mem_map = NODE_DATA(0)->node_mem_map;
Mel Gormanc7132162006-09-27 01:49:43 -07003761#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3762 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08003763 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Mel Gormanc7132162006-09-27 01:49:43 -07003764#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3765 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003766#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003767#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003768}
3769
Johannes Weiner9109fb72008-07-23 21:27:20 -07003770void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
3771 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003772{
Johannes Weiner9109fb72008-07-23 21:27:20 -07003773 pg_data_t *pgdat = NODE_DATA(nid);
3774
Linus Torvalds1da177e2005-04-16 15:20:36 -07003775 pgdat->node_id = nid;
3776 pgdat->node_start_pfn = node_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003777 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003778
3779 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07003780#ifdef CONFIG_FLAT_NODE_MEM_MAP
3781 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3782 nid, (unsigned long)pgdat,
3783 (unsigned long)pgdat->node_mem_map);
3784#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003785
3786 free_area_init_core(pgdat, zones_size, zholes_size);
3787}
3788
Mel Gormanc7132162006-09-27 01:49:43 -07003789#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07003790
3791#if MAX_NUMNODES > 1
3792/*
3793 * Figure out the number of possible node ids.
3794 */
3795static void __init setup_nr_node_ids(void)
3796{
3797 unsigned int node;
3798 unsigned int highest = 0;
3799
3800 for_each_node_mask(node, node_possible_map)
3801 highest = node;
3802 nr_node_ids = highest + 1;
3803}
3804#else
3805static inline void setup_nr_node_ids(void)
3806{
3807}
3808#endif
3809
Mel Gormanc7132162006-09-27 01:49:43 -07003810/**
3811 * add_active_range - Register a range of PFNs backed by physical memory
3812 * @nid: The node ID the range resides on
3813 * @start_pfn: The start PFN of the available physical memory
3814 * @end_pfn: The end PFN of the available physical memory
3815 *
3816 * These ranges are stored in an early_node_map[] and later used by
3817 * free_area_init_nodes() to calculate zone sizes and holes. If the
3818 * range spans a memory hole, it is up to the architecture to ensure
3819 * the memory is not freed by the bootmem allocator. If possible
3820 * the range being registered will be merged with existing ranges.
3821 */
3822void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3823 unsigned long end_pfn)
3824{
3825 int i;
3826
Mel Gorman6b74ab92008-07-23 21:26:49 -07003827 mminit_dprintk(MMINIT_TRACE, "memory_register",
3828 "Entering add_active_range(%d, %#lx, %#lx) "
3829 "%d entries of %d used\n",
3830 nid, start_pfn, end_pfn,
3831 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
Mel Gormanc7132162006-09-27 01:49:43 -07003832
Mel Gorman2dbb51c2008-07-23 21:26:52 -07003833 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
3834
Mel Gormanc7132162006-09-27 01:49:43 -07003835 /* Merge with existing active regions if possible */
3836 for (i = 0; i < nr_nodemap_entries; i++) {
3837 if (early_node_map[i].nid != nid)
3838 continue;
3839
3840 /* Skip if an existing region covers this new one */
3841 if (start_pfn >= early_node_map[i].start_pfn &&
3842 end_pfn <= early_node_map[i].end_pfn)
3843 return;
3844
3845 /* Merge forward if suitable */
3846 if (start_pfn <= early_node_map[i].end_pfn &&
3847 end_pfn > early_node_map[i].end_pfn) {
3848 early_node_map[i].end_pfn = end_pfn;
3849 return;
3850 }
3851
3852 /* Merge backward if suitable */
3853 if (start_pfn < early_node_map[i].end_pfn &&
3854 end_pfn >= early_node_map[i].start_pfn) {
3855 early_node_map[i].start_pfn = start_pfn;
3856 return;
3857 }
3858 }
3859
3860 /* Check that early_node_map is large enough */
3861 if (i >= MAX_ACTIVE_REGIONS) {
3862 printk(KERN_CRIT "More than %d memory regions, truncating\n",
3863 MAX_ACTIVE_REGIONS);
3864 return;
3865 }
3866
3867 early_node_map[i].nid = nid;
3868 early_node_map[i].start_pfn = start_pfn;
3869 early_node_map[i].end_pfn = end_pfn;
3870 nr_nodemap_entries = i + 1;
3871}
3872
3873/**
Yinghai Lucc1050b2008-06-13 19:08:52 -07003874 * remove_active_range - Shrink an existing registered range of PFNs
Mel Gormanc7132162006-09-27 01:49:43 -07003875 * @nid: The node id the range is on that should be shrunk
Yinghai Lucc1050b2008-06-13 19:08:52 -07003876 * @start_pfn: The new PFN of the range
3877 * @end_pfn: The new PFN of the range
Mel Gormanc7132162006-09-27 01:49:43 -07003878 *
3879 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003880 * The map is kept near the end physical page range that has already been
3881 * registered. This function allows an arch to shrink an existing registered
3882 * range.
Mel Gormanc7132162006-09-27 01:49:43 -07003883 */
Yinghai Lucc1050b2008-06-13 19:08:52 -07003884void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
3885 unsigned long end_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07003886{
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003887 int i, j;
3888 int removed = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07003889
Yinghai Lucc1050b2008-06-13 19:08:52 -07003890 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
3891 nid, start_pfn, end_pfn);
3892
Mel Gormanc7132162006-09-27 01:49:43 -07003893 /* Find the old active region end and shrink */
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003894 for_each_active_range_index_in_nid(i, nid) {
Yinghai Lucc1050b2008-06-13 19:08:52 -07003895 if (early_node_map[i].start_pfn >= start_pfn &&
3896 early_node_map[i].end_pfn <= end_pfn) {
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003897 /* clear it */
Yinghai Lucc1050b2008-06-13 19:08:52 -07003898 early_node_map[i].start_pfn = 0;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003899 early_node_map[i].end_pfn = 0;
3900 removed = 1;
3901 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07003902 }
Yinghai Lucc1050b2008-06-13 19:08:52 -07003903 if (early_node_map[i].start_pfn < start_pfn &&
3904 early_node_map[i].end_pfn > start_pfn) {
3905 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
3906 early_node_map[i].end_pfn = start_pfn;
3907 if (temp_end_pfn > end_pfn)
3908 add_active_range(nid, end_pfn, temp_end_pfn);
3909 continue;
3910 }
3911 if (early_node_map[i].start_pfn >= start_pfn &&
3912 early_node_map[i].end_pfn > end_pfn &&
3913 early_node_map[i].start_pfn < end_pfn) {
3914 early_node_map[i].start_pfn = end_pfn;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003915 continue;
3916 }
3917 }
3918
3919 if (!removed)
3920 return;
3921
3922 /* remove the blank ones */
3923 for (i = nr_nodemap_entries - 1; i > 0; i--) {
3924 if (early_node_map[i].nid != nid)
3925 continue;
3926 if (early_node_map[i].end_pfn)
3927 continue;
3928 /* we found it, get rid of it */
3929 for (j = i; j < nr_nodemap_entries - 1; j++)
3930 memcpy(&early_node_map[j], &early_node_map[j+1],
3931 sizeof(early_node_map[j]));
3932 j = nr_nodemap_entries - 1;
3933 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
3934 nr_nodemap_entries--;
3935 }
Mel Gormanc7132162006-09-27 01:49:43 -07003936}
3937
3938/**
3939 * remove_all_active_ranges - Remove all currently registered regions
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003940 *
Mel Gormanc7132162006-09-27 01:49:43 -07003941 * During discovery, it may be found that a table like SRAT is invalid
3942 * and an alternative discovery method must be used. This function removes
3943 * all currently registered regions.
3944 */
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003945void __init remove_all_active_ranges(void)
Mel Gormanc7132162006-09-27 01:49:43 -07003946{
3947 memset(early_node_map, 0, sizeof(early_node_map));
3948 nr_nodemap_entries = 0;
3949}
3950
3951/* Compare two active node_active_regions */
3952static int __init cmp_node_active_region(const void *a, const void *b)
3953{
3954 struct node_active_region *arange = (struct node_active_region *)a;
3955 struct node_active_region *brange = (struct node_active_region *)b;
3956
3957 /* Done this way to avoid overflows */
3958 if (arange->start_pfn > brange->start_pfn)
3959 return 1;
3960 if (arange->start_pfn < brange->start_pfn)
3961 return -1;
3962
3963 return 0;
3964}
3965
3966/* sort the node_map by start_pfn */
3967static void __init sort_node_map(void)
3968{
3969 sort(early_node_map, (size_t)nr_nodemap_entries,
3970 sizeof(struct node_active_region),
3971 cmp_node_active_region, NULL);
3972}
3973
Mel Gormana6af2bc2007-02-10 01:42:57 -08003974/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003975static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003976{
3977 int i;
Mel Gormana6af2bc2007-02-10 01:42:57 -08003978 unsigned long min_pfn = ULONG_MAX;
Mel Gorman1abbfb42006-11-23 12:01:41 +00003979
Mel Gormanc7132162006-09-27 01:49:43 -07003980 /* Assuming a sorted map, the first range found has the starting pfn */
3981 for_each_active_range_index_in_nid(i, nid)
Mel Gormana6af2bc2007-02-10 01:42:57 -08003982 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003983
Mel Gormana6af2bc2007-02-10 01:42:57 -08003984 if (min_pfn == ULONG_MAX) {
3985 printk(KERN_WARNING
Paul Jackson2bc0d262008-06-22 07:22:17 -07003986 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08003987 return 0;
3988 }
3989
3990 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003991}
3992
3993/**
3994 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3995 *
3996 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003997 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07003998 */
3999unsigned long __init find_min_pfn_with_active_regions(void)
4000{
4001 return find_min_pfn_for_node(MAX_NUMNODES);
4002}
4003
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004004/*
4005 * early_calculate_totalpages()
4006 * Sum pages in active regions for movable zone.
4007 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4008 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07004009static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07004010{
4011 int i;
4012 unsigned long totalpages = 0;
4013
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004014 for (i = 0; i < nr_nodemap_entries; i++) {
4015 unsigned long pages = early_node_map[i].end_pfn -
Mel Gorman7e63efe2007-07-17 04:03:15 -07004016 early_node_map[i].start_pfn;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004017 totalpages += pages;
4018 if (pages)
4019 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
4020 }
4021 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07004022}
4023
Mel Gorman2a1e2742007-07-17 04:03:12 -07004024/*
4025 * Find the PFN the Movable zone begins in each node. Kernel memory
4026 * is spread evenly between nodes as long as the nodes have enough
4027 * memory. When they don't, some nodes will have more kernelcore than
4028 * others
4029 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004030static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004031{
4032 int i, nid;
4033 unsigned long usable_startpfn;
4034 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07004035 /* save the state before borrow the nodemask */
4036 nodemask_t saved_node_state = node_states[N_HIGH_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004037 unsigned long totalpages = early_calculate_totalpages();
4038 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004039
Mel Gorman7e63efe2007-07-17 04:03:15 -07004040 /*
4041 * If movablecore was specified, calculate what size of
4042 * kernelcore that corresponds so that memory usable for
4043 * any allocation type is evenly spread. If both kernelcore
4044 * and movablecore are specified, then the value of kernelcore
4045 * will be used for required_kernelcore if it's greater than
4046 * what movablecore would have allowed.
4047 */
4048 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07004049 unsigned long corepages;
4050
4051 /*
4052 * Round-up so that ZONE_MOVABLE is at least as large as what
4053 * was requested by the user
4054 */
4055 required_movablecore =
4056 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
4057 corepages = totalpages - required_movablecore;
4058
4059 required_kernelcore = max(required_kernelcore, corepages);
4060 }
4061
Mel Gorman2a1e2742007-07-17 04:03:12 -07004062 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4063 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07004064 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004065
4066 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4067 find_usable_zone_for_movable();
4068 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4069
4070restart:
4071 /* Spread kernelcore memory as evenly as possible throughout nodes */
4072 kernelcore_node = required_kernelcore / usable_nodes;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004073 for_each_node_state(nid, N_HIGH_MEMORY) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004074 /*
4075 * Recalculate kernelcore_node if the division per node
4076 * now exceeds what is necessary to satisfy the requested
4077 * amount of memory for the kernel
4078 */
4079 if (required_kernelcore < kernelcore_node)
4080 kernelcore_node = required_kernelcore / usable_nodes;
4081
4082 /*
4083 * As the map is walked, we track how much memory is usable
4084 * by the kernel using kernelcore_remaining. When it is
4085 * 0, the rest of the node is usable by ZONE_MOVABLE
4086 */
4087 kernelcore_remaining = kernelcore_node;
4088
4089 /* Go through each range of PFNs within this node */
4090 for_each_active_range_index_in_nid(i, nid) {
4091 unsigned long start_pfn, end_pfn;
4092 unsigned long size_pages;
4093
4094 start_pfn = max(early_node_map[i].start_pfn,
4095 zone_movable_pfn[nid]);
4096 end_pfn = early_node_map[i].end_pfn;
4097 if (start_pfn >= end_pfn)
4098 continue;
4099
4100 /* Account for what is only usable for kernelcore */
4101 if (start_pfn < usable_startpfn) {
4102 unsigned long kernel_pages;
4103 kernel_pages = min(end_pfn, usable_startpfn)
4104 - start_pfn;
4105
4106 kernelcore_remaining -= min(kernel_pages,
4107 kernelcore_remaining);
4108 required_kernelcore -= min(kernel_pages,
4109 required_kernelcore);
4110
4111 /* Continue if range is now fully accounted */
4112 if (end_pfn <= usable_startpfn) {
4113
4114 /*
4115 * Push zone_movable_pfn to the end so
4116 * that if we have to rebalance
4117 * kernelcore across nodes, we will
4118 * not double account here
4119 */
4120 zone_movable_pfn[nid] = end_pfn;
4121 continue;
4122 }
4123 start_pfn = usable_startpfn;
4124 }
4125
4126 /*
4127 * The usable PFN range for ZONE_MOVABLE is from
4128 * start_pfn->end_pfn. Calculate size_pages as the
4129 * number of pages used as kernelcore
4130 */
4131 size_pages = end_pfn - start_pfn;
4132 if (size_pages > kernelcore_remaining)
4133 size_pages = kernelcore_remaining;
4134 zone_movable_pfn[nid] = start_pfn + size_pages;
4135
4136 /*
4137 * Some kernelcore has been met, update counts and
4138 * break if the kernelcore for this node has been
4139 * satisified
4140 */
4141 required_kernelcore -= min(required_kernelcore,
4142 size_pages);
4143 kernelcore_remaining -= size_pages;
4144 if (!kernelcore_remaining)
4145 break;
4146 }
4147 }
4148
4149 /*
4150 * If there is still required_kernelcore, we do another pass with one
4151 * less node in the count. This will push zone_movable_pfn[nid] further
4152 * along on the nodes that still have memory until kernelcore is
4153 * satisified
4154 */
4155 usable_nodes--;
4156 if (usable_nodes && required_kernelcore > usable_nodes)
4157 goto restart;
4158
4159 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4160 for (nid = 0; nid < MAX_NUMNODES; nid++)
4161 zone_movable_pfn[nid] =
4162 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07004163
4164out:
4165 /* restore the node_state */
4166 node_states[N_HIGH_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004167}
4168
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004169/* Any regular memory on that node ? */
4170static void check_for_regular_memory(pg_data_t *pgdat)
4171{
4172#ifdef CONFIG_HIGHMEM
4173 enum zone_type zone_type;
4174
4175 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4176 struct zone *zone = &pgdat->node_zones[zone_type];
4177 if (zone->present_pages)
4178 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4179 }
4180#endif
4181}
4182
Mel Gormanc7132162006-09-27 01:49:43 -07004183/**
4184 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004185 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07004186 *
4187 * This will call free_area_init_node() for each active node in the system.
4188 * Using the page ranges provided by add_active_range(), the size of each
4189 * zone in each node and their holes is calculated. If the maximum PFN
4190 * between two adjacent zones match, it is assumed that the zone is empty.
4191 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4192 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4193 * starts where the previous one ended. For example, ZONE_DMA32 starts
4194 * at arch_max_dma_pfn.
4195 */
4196void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4197{
4198 unsigned long nid;
Andrew Mortondb991002008-10-15 22:01:07 -07004199 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004200
Mel Gormana6af2bc2007-02-10 01:42:57 -08004201 /* Sort early_node_map as initialisation assumes it is sorted */
4202 sort_node_map();
4203
Mel Gormanc7132162006-09-27 01:49:43 -07004204 /* Record where the zone boundaries are */
4205 memset(arch_zone_lowest_possible_pfn, 0,
4206 sizeof(arch_zone_lowest_possible_pfn));
4207 memset(arch_zone_highest_possible_pfn, 0,
4208 sizeof(arch_zone_highest_possible_pfn));
4209 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4210 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4211 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004212 if (i == ZONE_MOVABLE)
4213 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004214 arch_zone_lowest_possible_pfn[i] =
4215 arch_zone_highest_possible_pfn[i-1];
4216 arch_zone_highest_possible_pfn[i] =
4217 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4218 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07004219 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4220 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4221
4222 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4223 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4224 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004225
Mel Gormanc7132162006-09-27 01:49:43 -07004226 /* Print out the zone ranges */
4227 printk("Zone PFN ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07004228 for (i = 0; i < MAX_NR_ZONES; i++) {
4229 if (i == ZONE_MOVABLE)
4230 continue;
Paul Jackson5dab8ec2008-06-25 05:44:40 -07004231 printk(" %-8s %0#10lx -> %0#10lx\n",
Mel Gormanc7132162006-09-27 01:49:43 -07004232 zone_names[i],
4233 arch_zone_lowest_possible_pfn[i],
4234 arch_zone_highest_possible_pfn[i]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004235 }
4236
4237 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4238 printk("Movable zone start PFN for each node\n");
4239 for (i = 0; i < MAX_NUMNODES; i++) {
4240 if (zone_movable_pfn[i])
4241 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4242 }
Mel Gormanc7132162006-09-27 01:49:43 -07004243
4244 /* Print out the early_node_map[] */
4245 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4246 for (i = 0; i < nr_nodemap_entries; i++)
Paul Jackson5dab8ec2008-06-25 05:44:40 -07004247 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004248 early_node_map[i].start_pfn,
4249 early_node_map[i].end_pfn);
4250
4251 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07004252 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08004253 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07004254 for_each_online_node(nid) {
4255 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07004256 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07004257 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004258
4259 /* Any memory on that node */
4260 if (pgdat->node_present_pages)
4261 node_set_state(nid, N_HIGH_MEMORY);
4262 check_for_regular_memory(pgdat);
Mel Gormanc7132162006-09-27 01:49:43 -07004263 }
4264}
Mel Gorman2a1e2742007-07-17 04:03:12 -07004265
Mel Gorman7e63efe2007-07-17 04:03:15 -07004266static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004267{
4268 unsigned long long coremem;
4269 if (!p)
4270 return -EINVAL;
4271
4272 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004273 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004274
Mel Gorman7e63efe2007-07-17 04:03:15 -07004275 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07004276 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4277
4278 return 0;
4279}
Mel Gormaned7ed362007-07-17 04:03:14 -07004280
Mel Gorman7e63efe2007-07-17 04:03:15 -07004281/*
4282 * kernelcore=size sets the amount of memory for use for allocations that
4283 * cannot be reclaimed or migrated.
4284 */
4285static int __init cmdline_parse_kernelcore(char *p)
4286{
4287 return cmdline_parse_core(p, &required_kernelcore);
4288}
4289
4290/*
4291 * movablecore=size sets the amount of memory for use for allocations that
4292 * can be reclaimed or migrated.
4293 */
4294static int __init cmdline_parse_movablecore(char *p)
4295{
4296 return cmdline_parse_core(p, &required_movablecore);
4297}
4298
Mel Gormaned7ed362007-07-17 04:03:14 -07004299early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004300early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07004301
Mel Gormanc7132162006-09-27 01:49:43 -07004302#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4303
Mel Gorman0e0b8642006-09-27 01:49:56 -07004304/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004305 * set_dma_reserve - set the specified number of pages reserved in the first zone
4306 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07004307 *
4308 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4309 * In the DMA zone, a significant percentage may be consumed by kernel image
4310 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004311 * function may optionally be used to account for unfreeable pages in the
4312 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4313 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07004314 */
4315void __init set_dma_reserve(unsigned long new_dma_reserve)
4316{
4317 dma_reserve = new_dma_reserve;
4318}
4319
Dave Hansen93b75042005-06-23 00:07:47 -07004320#ifndef CONFIG_NEED_MULTIPLE_NODES
Marcin Slusarz52765582008-09-02 14:35:41 -07004321struct pglist_data __refdata contig_page_data = { .bdata = &bootmem_node_data[0] };
Linus Torvalds1da177e2005-04-16 15:20:36 -07004322EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07004323#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004324
4325void __init free_area_init(unsigned long *zones_size)
4326{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004327 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004328 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4329}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004330
Linus Torvalds1da177e2005-04-16 15:20:36 -07004331static int page_alloc_cpu_notify(struct notifier_block *self,
4332 unsigned long action, void *hcpu)
4333{
4334 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004335
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07004336 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004337 drain_pages(cpu);
4338
4339 /*
4340 * Spill the event counters of the dead processor
4341 * into the current processors event counters.
4342 * This artificially elevates the count of the current
4343 * processor.
4344 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07004345 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004346
4347 /*
4348 * Zero the differential counters of the dead processor
4349 * so that the vm statistics are consistent.
4350 *
4351 * This is only okay since the processor is dead and cannot
4352 * race with what we are doing.
4353 */
Christoph Lameter2244b952006-06-30 01:55:33 -07004354 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004355 }
4356 return NOTIFY_OK;
4357}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004358
4359void __init page_alloc_init(void)
4360{
4361 hotcpu_notifier(page_alloc_cpu_notify, 0);
4362}
4363
4364/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004365 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4366 * or min_free_kbytes changes.
4367 */
4368static void calculate_totalreserve_pages(void)
4369{
4370 struct pglist_data *pgdat;
4371 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004372 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004373
4374 for_each_online_pgdat(pgdat) {
4375 for (i = 0; i < MAX_NR_ZONES; i++) {
4376 struct zone *zone = pgdat->node_zones + i;
4377 unsigned long max = 0;
4378
4379 /* Find valid and maximum lowmem_reserve in the zone */
4380 for (j = i; j < MAX_NR_ZONES; j++) {
4381 if (zone->lowmem_reserve[j] > max)
4382 max = zone->lowmem_reserve[j];
4383 }
4384
Mel Gorman41858962009-06-16 15:32:12 -07004385 /* we treat the high watermark as reserved pages. */
4386 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004387
4388 if (max > zone->present_pages)
4389 max = zone->present_pages;
4390 reserve_pages += max;
4391 }
4392 }
4393 totalreserve_pages = reserve_pages;
4394}
4395
4396/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07004397 * setup_per_zone_lowmem_reserve - called whenever
4398 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4399 * has a correct pages reserved value, so an adequate number of
4400 * pages are left in the zone after a successful __alloc_pages().
4401 */
4402static void setup_per_zone_lowmem_reserve(void)
4403{
4404 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004405 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004406
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08004407 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004408 for (j = 0; j < MAX_NR_ZONES; j++) {
4409 struct zone *zone = pgdat->node_zones + j;
4410 unsigned long present_pages = zone->present_pages;
4411
4412 zone->lowmem_reserve[j] = 0;
4413
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004414 idx = j;
4415 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004416 struct zone *lower_zone;
4417
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004418 idx--;
4419
Linus Torvalds1da177e2005-04-16 15:20:36 -07004420 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4421 sysctl_lowmem_reserve_ratio[idx] = 1;
4422
4423 lower_zone = pgdat->node_zones + idx;
4424 lower_zone->lowmem_reserve[j] = present_pages /
4425 sysctl_lowmem_reserve_ratio[idx];
4426 present_pages += lower_zone->present_pages;
4427 }
4428 }
4429 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004430
4431 /* update totalreserve_pages */
4432 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004433}
4434
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004435/**
Minchan Kimbc75d332009-06-16 15:32:48 -07004436 * setup_per_zone_wmarks - called when min_free_kbytes changes
Minchan Kimbce73942009-06-16 15:32:50 -07004437 * or when memory is hot-{added|removed}
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004438 *
Minchan Kimbc75d332009-06-16 15:32:48 -07004439 * Ensures that the watermark[min,low,high] values for each zone are set
4440 * correctly with respect to min_free_kbytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004441 */
Minchan Kimbc75d332009-06-16 15:32:48 -07004442void setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004443{
4444 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4445 unsigned long lowmem_pages = 0;
4446 struct zone *zone;
4447 unsigned long flags;
4448
4449 /* Calculate total number of !ZONE_HIGHMEM pages */
4450 for_each_zone(zone) {
4451 if (!is_highmem(zone))
4452 lowmem_pages += zone->present_pages;
4453 }
4454
4455 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07004456 u64 tmp;
4457
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07004458 spin_lock_irqsave(&zone->lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07004459 tmp = (u64)pages_min * zone->present_pages;
4460 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004461 if (is_highmem(zone)) {
4462 /*
Nick Piggin669ed172005-11-13 16:06:45 -08004463 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4464 * need highmem pages, so cap pages_min to a small
4465 * value here.
4466 *
Mel Gorman41858962009-06-16 15:32:12 -07004467 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Nick Piggin669ed172005-11-13 16:06:45 -08004468 * deltas controls asynch page reclaim, and so should
4469 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004470 */
4471 int min_pages;
4472
4473 min_pages = zone->present_pages / 1024;
4474 if (min_pages < SWAP_CLUSTER_MAX)
4475 min_pages = SWAP_CLUSTER_MAX;
4476 if (min_pages > 128)
4477 min_pages = 128;
Mel Gorman41858962009-06-16 15:32:12 -07004478 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004479 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08004480 /*
4481 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07004482 * proportionate to the zone's size.
4483 */
Mel Gorman41858962009-06-16 15:32:12 -07004484 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004485 }
4486
Mel Gorman41858962009-06-16 15:32:12 -07004487 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
4488 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Mel Gorman56fd56b2007-10-16 01:25:58 -07004489 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07004490 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004491 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004492
4493 /* update totalreserve_pages */
4494 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004495}
4496
Rik van Riel556adec2008-10-18 20:26:34 -07004497/**
Rik van Riel556adec2008-10-18 20:26:34 -07004498 * The inactive anon list should be small enough that the VM never has to
4499 * do too much work, but large enough that each inactive page has a chance
4500 * to be referenced again before it is swapped out.
4501 *
4502 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4503 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4504 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4505 * the anonymous pages are kept on the inactive list.
4506 *
4507 * total target max
4508 * memory ratio inactive anon
4509 * -------------------------------------
4510 * 10MB 1 5MB
4511 * 100MB 1 50MB
4512 * 1GB 3 250MB
4513 * 10GB 10 0.9GB
4514 * 100GB 31 3GB
4515 * 1TB 101 10GB
4516 * 10TB 320 32GB
4517 */
Minchan Kim96cb4df2009-06-16 15:32:49 -07004518void calculate_zone_inactive_ratio(struct zone *zone)
4519{
4520 unsigned int gb, ratio;
4521
4522 /* Zone size in gigabytes */
4523 gb = zone->present_pages >> (30 - PAGE_SHIFT);
4524 if (gb)
4525 ratio = int_sqrt(10 * gb);
4526 else
4527 ratio = 1;
4528
4529 zone->inactive_ratio = ratio;
4530}
4531
Cyrill Gorcunove9bb35d2009-06-16 15:32:32 -07004532static void __init setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07004533{
4534 struct zone *zone;
4535
Minchan Kim96cb4df2009-06-16 15:32:49 -07004536 for_each_zone(zone)
4537 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07004538}
4539
Linus Torvalds1da177e2005-04-16 15:20:36 -07004540/*
4541 * Initialise min_free_kbytes.
4542 *
4543 * For small machines we want it small (128k min). For large machines
4544 * we want it large (64MB max). But it is not linear, because network
4545 * bandwidth does not increase linearly with machine size. We use
4546 *
4547 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4548 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4549 *
4550 * which yields
4551 *
4552 * 16MB: 512k
4553 * 32MB: 724k
4554 * 64MB: 1024k
4555 * 128MB: 1448k
4556 * 256MB: 2048k
4557 * 512MB: 2896k
4558 * 1024MB: 4096k
4559 * 2048MB: 5792k
4560 * 4096MB: 8192k
4561 * 8192MB: 11584k
4562 * 16384MB: 16384k
4563 */
Minchan Kimbc75d332009-06-16 15:32:48 -07004564static int __init init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004565{
4566 unsigned long lowmem_kbytes;
4567
4568 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4569
4570 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4571 if (min_free_kbytes < 128)
4572 min_free_kbytes = 128;
4573 if (min_free_kbytes > 65536)
4574 min_free_kbytes = 65536;
Minchan Kimbc75d332009-06-16 15:32:48 -07004575 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004576 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07004577 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004578 return 0;
4579}
Minchan Kimbc75d332009-06-16 15:32:48 -07004580module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004581
4582/*
4583 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4584 * that we can call two helper functions whenever min_free_kbytes
4585 * changes.
4586 */
4587int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
4588 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4589{
4590 proc_dointvec(table, write, file, buffer, length, ppos);
Mel Gorman3b1d92c2007-05-06 14:49:30 -07004591 if (write)
Minchan Kimbc75d332009-06-16 15:32:48 -07004592 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004593 return 0;
4594}
4595
Christoph Lameter96146342006-07-03 00:24:13 -07004596#ifdef CONFIG_NUMA
4597int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
4598 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4599{
4600 struct zone *zone;
4601 int rc;
4602
4603 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4604 if (rc)
4605 return rc;
4606
4607 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07004608 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07004609 sysctl_min_unmapped_ratio) / 100;
4610 return 0;
4611}
Christoph Lameter0ff38492006-09-25 23:31:52 -07004612
4613int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
4614 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4615{
4616 struct zone *zone;
4617 int rc;
4618
4619 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4620 if (rc)
4621 return rc;
4622
4623 for_each_zone(zone)
4624 zone->min_slab_pages = (zone->present_pages *
4625 sysctl_min_slab_ratio) / 100;
4626 return 0;
4627}
Christoph Lameter96146342006-07-03 00:24:13 -07004628#endif
4629
Linus Torvalds1da177e2005-04-16 15:20:36 -07004630/*
4631 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4632 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4633 * whenever sysctl_lowmem_reserve_ratio changes.
4634 *
4635 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07004636 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07004637 * if in function of the boot time zone sizes.
4638 */
4639int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
4640 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4641{
4642 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4643 setup_per_zone_lowmem_reserve();
4644 return 0;
4645}
4646
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004647/*
4648 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4649 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4650 * can have before it gets flushed back to buddy allocator.
4651 */
4652
4653int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
4654 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4655{
4656 struct zone *zone;
4657 unsigned int cpu;
4658 int ret;
4659
4660 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4661 if (!write || (ret == -EINVAL))
4662 return ret;
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07004663 for_each_populated_zone(zone) {
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004664 for_each_online_cpu(cpu) {
4665 unsigned long high;
4666 high = zone->present_pages / percpu_pagelist_fraction;
4667 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4668 }
4669 }
4670 return 0;
4671}
4672
David S. Millerf034b5d2006-08-24 03:08:07 -07004673int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004674
4675#ifdef CONFIG_NUMA
4676static int __init set_hashdist(char *str)
4677{
4678 if (!str)
4679 return 0;
4680 hashdist = simple_strtoul(str, &str, 0);
4681 return 1;
4682}
4683__setup("hashdist=", set_hashdist);
4684#endif
4685
4686/*
4687 * allocate a large system hash table from bootmem
4688 * - it is assumed that the hash table must contain an exact power-of-2
4689 * quantity of entries
4690 * - limit is the number of hash buckets, not the total allocation size
4691 */
4692void *__init alloc_large_system_hash(const char *tablename,
4693 unsigned long bucketsize,
4694 unsigned long numentries,
4695 int scale,
4696 int flags,
4697 unsigned int *_hash_shift,
4698 unsigned int *_hash_mask,
4699 unsigned long limit)
4700{
4701 unsigned long long max = limit;
4702 unsigned long log2qty, size;
4703 void *table = NULL;
4704
4705 /* allow the kernel cmdline to have a say */
4706 if (!numentries) {
4707 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08004708 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004709 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4710 numentries >>= 20 - PAGE_SHIFT;
4711 numentries <<= 20 - PAGE_SHIFT;
4712
4713 /* limit to 1 bucket per 2^scale bytes of low memory */
4714 if (scale > PAGE_SHIFT)
4715 numentries >>= (scale - PAGE_SHIFT);
4716 else
4717 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08004718
4719 /* Make sure we've got at least a 0-order allocation.. */
4720 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
4721 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004722 }
John Hawkes6e692ed2006-03-25 03:08:02 -08004723 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004724
4725 /* limit allocation size to 1/16 total memory by default */
4726 if (max == 0) {
4727 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4728 do_div(max, bucketsize);
4729 }
4730
4731 if (numentries > max)
4732 numentries = max;
4733
David Howellsf0d1b0b2006-12-08 02:37:49 -08004734 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004735
4736 do {
4737 size = bucketsize << log2qty;
4738 if (flags & HASH_EARLY)
Jan Beulich74768ed2008-08-12 15:08:39 -07004739 table = alloc_bootmem_nopanic(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004740 else if (hashdist)
4741 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4742 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07004743 /*
4744 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07004745 * some pages at the end of hash table which
4746 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07004747 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01004748 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07004749 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01004750 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
4751 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004752 }
4753 } while (!table && size > PAGE_SIZE && --log2qty);
4754
4755 if (!table)
4756 panic("Failed to allocate %s hash table\n", tablename);
4757
Dan Alonib49ad482007-07-15 23:38:23 -07004758 printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07004759 tablename,
4760 (1U << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08004761 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004762 size);
4763
4764 if (_hash_shift)
4765 *_hash_shift = log2qty;
4766 if (_hash_mask)
4767 *_hash_mask = (1 << log2qty) - 1;
4768
4769 return table;
4770}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004771
Mel Gorman835c1342007-10-16 01:25:47 -07004772/* Return a pointer to the bitmap storing bits affecting a block of pages */
4773static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4774 unsigned long pfn)
4775{
4776#ifdef CONFIG_SPARSEMEM
4777 return __pfn_to_section(pfn)->pageblock_flags;
4778#else
4779 return zone->pageblock_flags;
4780#endif /* CONFIG_SPARSEMEM */
4781}
Andrew Morton6220ec72006-10-19 23:29:05 -07004782
Mel Gorman835c1342007-10-16 01:25:47 -07004783static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4784{
4785#ifdef CONFIG_SPARSEMEM
4786 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07004787 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07004788#else
4789 pfn = pfn - zone->zone_start_pfn;
Mel Gormand9c23402007-10-16 01:26:01 -07004790 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07004791#endif /* CONFIG_SPARSEMEM */
4792}
4793
4794/**
Mel Gormand9c23402007-10-16 01:26:01 -07004795 * 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 -07004796 * @page: The page within the block of interest
4797 * @start_bitidx: The first bit of interest to retrieve
4798 * @end_bitidx: The last bit of interest
4799 * returns pageblock_bits flags
4800 */
4801unsigned long get_pageblock_flags_group(struct page *page,
4802 int start_bitidx, int end_bitidx)
4803{
4804 struct zone *zone;
4805 unsigned long *bitmap;
4806 unsigned long pfn, bitidx;
4807 unsigned long flags = 0;
4808 unsigned long value = 1;
4809
4810 zone = page_zone(page);
4811 pfn = page_to_pfn(page);
4812 bitmap = get_pageblock_bitmap(zone, pfn);
4813 bitidx = pfn_to_bitidx(zone, pfn);
4814
4815 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4816 if (test_bit(bitidx + start_bitidx, bitmap))
4817 flags |= value;
4818
4819 return flags;
4820}
4821
4822/**
Mel Gormand9c23402007-10-16 01:26:01 -07004823 * 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 -07004824 * @page: The page within the block of interest
4825 * @start_bitidx: The first bit of interest
4826 * @end_bitidx: The last bit of interest
4827 * @flags: The flags to set
4828 */
4829void set_pageblock_flags_group(struct page *page, unsigned long flags,
4830 int start_bitidx, int end_bitidx)
4831{
4832 struct zone *zone;
4833 unsigned long *bitmap;
4834 unsigned long pfn, bitidx;
4835 unsigned long value = 1;
4836
4837 zone = page_zone(page);
4838 pfn = page_to_pfn(page);
4839 bitmap = get_pageblock_bitmap(zone, pfn);
4840 bitidx = pfn_to_bitidx(zone, pfn);
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004841 VM_BUG_ON(pfn < zone->zone_start_pfn);
4842 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
Mel Gorman835c1342007-10-16 01:25:47 -07004843
4844 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4845 if (flags & value)
4846 __set_bit(bitidx + start_bitidx, bitmap);
4847 else
4848 __clear_bit(bitidx + start_bitidx, bitmap);
4849}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07004850
4851/*
4852 * This is designed as sub function...plz see page_isolation.c also.
4853 * set/clear page block's type to be ISOLATE.
4854 * page allocater never alloc memory from ISOLATE block.
4855 */
4856
4857int set_migratetype_isolate(struct page *page)
4858{
4859 struct zone *zone;
4860 unsigned long flags;
4861 int ret = -EBUSY;
4862
4863 zone = page_zone(page);
4864 spin_lock_irqsave(&zone->lock, flags);
4865 /*
4866 * In future, more migrate types will be able to be isolation target.
4867 */
4868 if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE)
4869 goto out;
4870 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
4871 move_freepages_block(zone, page, MIGRATE_ISOLATE);
4872 ret = 0;
4873out:
4874 spin_unlock_irqrestore(&zone->lock, flags);
4875 if (!ret)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004876 drain_all_pages();
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07004877 return ret;
4878}
4879
4880void unset_migratetype_isolate(struct page *page)
4881{
4882 struct zone *zone;
4883 unsigned long flags;
4884 zone = page_zone(page);
4885 spin_lock_irqsave(&zone->lock, flags);
4886 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
4887 goto out;
4888 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4889 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4890out:
4891 spin_unlock_irqrestore(&zone->lock, flags);
4892}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07004893
4894#ifdef CONFIG_MEMORY_HOTREMOVE
4895/*
4896 * All pages in the range must be isolated before calling this.
4897 */
4898void
4899__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
4900{
4901 struct page *page;
4902 struct zone *zone;
4903 int order, i;
4904 unsigned long pfn;
4905 unsigned long flags;
4906 /* find the first valid pfn */
4907 for (pfn = start_pfn; pfn < end_pfn; pfn++)
4908 if (pfn_valid(pfn))
4909 break;
4910 if (pfn == end_pfn)
4911 return;
4912 zone = page_zone(pfn_to_page(pfn));
4913 spin_lock_irqsave(&zone->lock, flags);
4914 pfn = start_pfn;
4915 while (pfn < end_pfn) {
4916 if (!pfn_valid(pfn)) {
4917 pfn++;
4918 continue;
4919 }
4920 page = pfn_to_page(pfn);
4921 BUG_ON(page_count(page));
4922 BUG_ON(!PageBuddy(page));
4923 order = page_order(page);
4924#ifdef CONFIG_DEBUG_VM
4925 printk(KERN_INFO "remove from free list %lx %d %lx\n",
4926 pfn, 1 << order, end_pfn);
4927#endif
4928 list_del(&page->lru);
4929 rmv_page_order(page);
4930 zone->free_area[order].nr_free--;
4931 __mod_zone_page_state(zone, NR_FREE_PAGES,
4932 - (1UL << order));
4933 for (i = 0; i < (1 << order); i++)
4934 SetPageReserved((page+i));
4935 pfn += (1 << order);
4936 }
4937 spin_unlock_irqrestore(&zone->lock, flags);
4938}
4939#endif