blob: 5cc986eb9f6f605c0b88246a700d4ac46afd4b7f [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/page_alloc.c
3 *
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
6 *
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
15 */
16
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/stddef.h>
18#include <linux/mm.h>
19#include <linux/swap.h>
20#include <linux/interrupt.h>
21#include <linux/pagemap.h>
KOSAKI Motohiro10ed2732008-03-04 14:28:32 -080022#include <linux/jiffies.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/bootmem.h>
24#include <linux/compiler.h>
Randy Dunlap9f158332005-09-13 01:25:16 -070025#include <linux/kernel.h>
Vegard Nossumb1eeab62008-11-25 16:55:53 +010026#include <linux/kmemcheck.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/module.h>
28#include <linux/suspend.h>
29#include <linux/pagevec.h>
30#include <linux/blkdev.h>
31#include <linux/slab.h>
David Rientjes5a3135c2007-10-16 23:25:53 -070032#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include <linux/notifier.h>
34#include <linux/topology.h>
35#include <linux/sysctl.h>
36#include <linux/cpu.h>
37#include <linux/cpuset.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070038#include <linux/memory_hotplug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include <linux/nodemask.h>
40#include <linux/vmalloc.h>
Christoph Lameter4be38e32006-01-06 00:11:17 -080041#include <linux/mempolicy.h>
Yasunori Goto68113782006-06-23 02:03:11 -070042#include <linux/stop_machine.h>
Mel Gormanc7132162006-09-27 01:49:43 -070043#include <linux/sort.h>
44#include <linux/pfn.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070045#include <linux/backing-dev.h>
Akinobu Mita933e3122006-12-08 02:39:45 -080046#include <linux/fault-inject.h>
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -070047#include <linux/page-isolation.h>
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070048#include <linux/page_cgroup.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070049#include <linux/debugobjects.h>
Catalin Marinasdbb1f812009-06-11 13:23:19 +010050#include <linux/kmemleak.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
52#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070053#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070054#include "internal.h"
55
56/*
Christoph Lameter13808912007-10-16 01:25:27 -070057 * Array of node states.
Linus Torvalds1da177e2005-04-16 15:20:36 -070058 */
Christoph Lameter13808912007-10-16 01:25:27 -070059nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
60 [N_POSSIBLE] = NODE_MASK_ALL,
61 [N_ONLINE] = { { [0] = 1UL } },
62#ifndef CONFIG_NUMA
63 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
64#ifdef CONFIG_HIGHMEM
65 [N_HIGH_MEMORY] = { { [0] = 1UL } },
66#endif
67 [N_CPU] = { { [0] = 1UL } },
68#endif /* NUMA */
69};
70EXPORT_SYMBOL(node_states);
71
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070072unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -070073unsigned long totalreserve_pages __read_mostly;
Hugh Dickins22b31ee2009-01-06 14:40:09 -080074unsigned long highest_memmap_pfn __read_mostly;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -080075int percpu_pagelist_fraction;
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +100076gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -070077
Mel Gormand9c23402007-10-16 01:26:01 -070078#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
79int pageblock_order __read_mostly;
80#endif
81
Hugh Dickinsd98c7a02006-02-14 13:52:59 -080082static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -080083
Linus Torvalds1da177e2005-04-16 15:20:36 -070084/*
85 * results with 256, 32 in the lowmem_reserve sysctl:
86 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
87 * 1G machine -> (16M dma, 784M normal, 224M high)
88 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
89 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
90 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +010091 *
92 * TBD: should special case ZONE_DMA32 machines here - in those we normally
93 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -070095int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -080096#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -070097 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -080098#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -070099#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700100 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700101#endif
Christoph Lametere53ef382006-09-25 23:31:14 -0700102#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700103 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700104#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700105 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700106};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107
108EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109
Helge Deller15ad7cd2006-12-06 20:40:36 -0800110static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800111#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700112 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800113#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700114#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700115 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700116#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700117 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700118#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700119 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700120#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700121 "Movable",
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700122};
123
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124int min_free_kbytes = 1024;
125
Yasunori Goto86356ab2006-06-23 02:03:09 -0700126unsigned long __meminitdata nr_kernel_pages;
127unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700128static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129
Mel Gormanc7132162006-09-27 01:49:43 -0700130#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
131 /*
Simon Arlott183ff222007-10-20 01:27:18 +0200132 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
Mel Gormanc7132162006-09-27 01:49:43 -0700133 * ranges of memory (RAM) that may be registered with add_active_range().
134 * Ranges passed to add_active_range() will be merged if possible
135 * so the number of times add_active_range() can be called is
136 * related to the number of nodes and the number of holes
137 */
138 #ifdef CONFIG_MAX_ACTIVE_REGIONS
139 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
140 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
141 #else
142 #if MAX_NUMNODES >= 32
143 /* If there can be many nodes, allow up to 50 holes per node */
144 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
145 #else
146 /* By default, allow up to 256 distinct regions */
147 #define MAX_ACTIVE_REGIONS 256
148 #endif
149 #endif
150
Jan Beulich98011f52007-07-15 23:38:17 -0700151 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
152 static int __meminitdata nr_nodemap_entries;
153 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
154 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
Adrian Bunkb69a7282008-07-23 21:28:12 -0700155 static unsigned long __initdata required_kernelcore;
Adrian Bunk484f51f2007-10-16 01:26:03 -0700156 static unsigned long __initdata required_movablecore;
Adrian Bunkb69a7282008-07-23 21:28:12 -0700157 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Mel Gorman2a1e2742007-07-17 04:03:12 -0700158
159 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
160 int movable_zone;
161 EXPORT_SYMBOL(movable_zone);
Mel Gormanc7132162006-09-27 01:49:43 -0700162#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
163
Miklos Szeredi418508c2007-05-23 13:57:55 -0700164#if MAX_NUMNODES > 1
165int nr_node_ids __read_mostly = MAX_NUMNODES;
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700166int nr_online_nodes __read_mostly = 1;
Miklos Szeredi418508c2007-05-23 13:57:55 -0700167EXPORT_SYMBOL(nr_node_ids);
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700168EXPORT_SYMBOL(nr_online_nodes);
Miklos Szeredi418508c2007-05-23 13:57:55 -0700169#endif
170
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700171int page_group_by_mobility_disabled __read_mostly;
172
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700173static void set_pageblock_migratetype(struct page *page, int migratetype)
174{
Mel Gorman49255c62009-06-16 15:31:58 -0700175
176 if (unlikely(page_group_by_mobility_disabled))
177 migratetype = MIGRATE_UNMOVABLE;
178
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700179 set_pageblock_flags_group(page, (unsigned long)migratetype,
180 PB_migrate, PB_migrate_end);
181}
182
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -0700183bool oom_killer_disabled __read_mostly;
184
Nick Piggin13e74442006-01-06 00:10:58 -0800185#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700186static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700188 int ret = 0;
189 unsigned seq;
190 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -0700191
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700192 do {
193 seq = zone_span_seqbegin(zone);
194 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
195 ret = 1;
196 else if (pfn < zone->zone_start_pfn)
197 ret = 1;
198 } while (zone_span_seqretry(zone, seq));
199
200 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700201}
202
203static int page_is_consistent(struct zone *zone, struct page *page)
204{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700205 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700206 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700208 return 0;
209
210 return 1;
211}
212/*
213 * Temporary debugging check for pages not lying within a given zone.
214 */
215static int bad_range(struct zone *zone, struct page *page)
216{
217 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700219 if (!page_is_consistent(zone, page))
220 return 1;
221
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 return 0;
223}
Nick Piggin13e74442006-01-06 00:10:58 -0800224#else
225static inline int bad_range(struct zone *zone, struct page *page)
226{
227 return 0;
228}
229#endif
230
Nick Piggin224abf92006-01-06 00:11:11 -0800231static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800233 static unsigned long resume;
234 static unsigned long nr_shown;
235 static unsigned long nr_unshown;
236
237 /*
238 * Allow a burst of 60 reports, then keep quiet for that minute;
239 * or allow a steady drip of one report per second.
240 */
241 if (nr_shown == 60) {
242 if (time_before(jiffies, resume)) {
243 nr_unshown++;
244 goto out;
245 }
246 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800247 printk(KERN_ALERT
248 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800249 nr_unshown);
250 nr_unshown = 0;
251 }
252 nr_shown = 0;
253 }
254 if (nr_shown++ == 0)
255 resume = jiffies + 60 * HZ;
256
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800257 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800258 current->comm, page_to_pfn(page));
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800259 printk(KERN_ALERT
Hugh Dickins3dc14742009-01-06 14:40:08 -0800260 "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n",
261 page, (void *)page->flags, page_count(page),
262 page_mapcount(page), page->mapping, page->index);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700263
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800265out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800266 /* Leave bad fields for debug, except PageBuddy could make trouble */
267 __ClearPageBuddy(page);
Randy Dunlap9f158332005-09-13 01:25:16 -0700268 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269}
270
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271/*
272 * Higher-order pages are called "compound pages". They are structured thusly:
273 *
274 * The first PAGE_SIZE page is called the "head page".
275 *
276 * The remaining PAGE_SIZE pages are called "tail pages".
277 *
278 * All pages have PG_compound set. All pages have their ->private pointing at
279 * the head page (even the head page has this).
280 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800281 * The first tail page's ->lru.next holds the address of the compound page's
282 * put_page() function. Its ->lru.prev holds the order of allocation.
283 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800285
286static void free_compound_page(struct page *page)
287{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700288 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800289}
290
Andi Kleen01ad1c02008-07-23 21:27:46 -0700291void prep_compound_page(struct page *page, unsigned long order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292{
293 int i;
294 int nr_pages = 1 << order;
295
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800296 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700297 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700298 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800299 for (i = 1; i < nr_pages; i++) {
300 struct page *p = page + i;
301
Christoph Lameterd85f3382007-05-06 14:49:39 -0700302 __SetPageTail(p);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700303 p->first_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 }
305}
306
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800307static int destroy_compound_page(struct page *page, unsigned long order)
Andy Whitcroft18229df2008-11-06 12:53:27 -0800308{
309 int i;
310 int nr_pages = 1 << order;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800311 int bad = 0;
Andy Whitcroft18229df2008-11-06 12:53:27 -0800312
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800313 if (unlikely(compound_order(page) != order) ||
314 unlikely(!PageHead(page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800315 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800316 bad++;
317 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318
Christoph Lameter6d777952007-05-06 14:49:40 -0700319 __ClearPageHead(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800320
Andy Whitcroft18229df2008-11-06 12:53:27 -0800321 for (i = 1; i < nr_pages; i++) {
322 struct page *p = page + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323
Alexey Zaytseve713a212009-01-10 02:47:57 +0300324 if (unlikely(!PageTail(p) || (p->first_page != page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800325 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800326 bad++;
327 }
Christoph Lameterd85f3382007-05-06 14:49:39 -0700328 __ClearPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800330
331 return bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333
Nick Piggin17cf4402006-03-22 00:08:41 -0800334static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
335{
336 int i;
337
Andrew Morton6626c5d2006-03-22 00:08:42 -0800338 /*
339 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
340 * and __GFP_HIGHMEM from hard or soft interrupt context.
341 */
Nick Piggin725d7042006-09-25 23:30:55 -0700342 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800343 for (i = 0; i < (1 << order); i++)
344 clear_highpage(page + i);
345}
346
Andrew Morton6aa30012006-04-18 22:20:52 -0700347static inline void set_page_order(struct page *page, int order)
348{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700349 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000350 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351}
352
353static inline void rmv_page_order(struct page *page)
354{
Nick Piggin676165a2006-04-10 11:21:48 +1000355 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700356 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357}
358
359/*
360 * Locate the struct page for both the matching buddy in our
361 * pair (buddy1) and the combined O(n+1) page they form (page).
362 *
363 * 1) Any buddy B1 will have an order O twin B2 which satisfies
364 * the following equation:
365 * B2 = B1 ^ (1 << O)
366 * For example, if the starting buddy (buddy2) is #8 its order
367 * 1 buddy is #10:
368 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
369 *
370 * 2) Any buddy B will have an order O+1 parent P which
371 * satisfies the following equation:
372 * P = B & ~(1 << O)
373 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200374 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 */
376static inline struct page *
377__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
378{
379 unsigned long buddy_idx = page_idx ^ (1 << order);
380
381 return page + (buddy_idx - page_idx);
382}
383
384static inline unsigned long
385__find_combined_index(unsigned long page_idx, unsigned int order)
386{
387 return (page_idx & ~(1 << order));
388}
389
390/*
391 * This function checks whether a page is free && is the buddy
392 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800393 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000394 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700395 * (c) a page and its buddy have the same order &&
396 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 *
Nick Piggin676165a2006-04-10 11:21:48 +1000398 * For recording whether a page is in the buddy system, we use PG_buddy.
399 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
400 *
401 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700403static inline int page_is_buddy(struct page *page, struct page *buddy,
404 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700406 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800407 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800408
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700409 if (page_zone_id(page) != page_zone_id(buddy))
410 return 0;
411
412 if (PageBuddy(buddy) && page_order(buddy) == order) {
Nick Piggina3af9c32009-06-16 15:32:10 -0700413 VM_BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa30012006-04-18 22:20:52 -0700414 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000415 }
Andrew Morton6aa30012006-04-18 22:20:52 -0700416 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417}
418
419/*
420 * Freeing function for a buddy system allocator.
421 *
422 * The concept of a buddy system is to maintain direct-mapped table
423 * (containing bit values) for memory blocks of various "orders".
424 * The bottom level table contains the map for the smallest allocatable
425 * units of memory (here, pages), and each level above it describes
426 * pairs of units from the levels below, hence, "buddies".
427 * At a high level, all that happens here is marking the table entry
428 * at the bottom level available, and propagating the changes upward
429 * as necessary, plus some accounting needed to play nicely with other
430 * parts of the VM system.
431 * At each level, we keep a list of pages, which are heads of continuous
Nick Piggin676165a2006-04-10 11:21:48 +1000432 * free pages of length of (1 << order) and marked with PG_buddy. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700433 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 * So when we are allocating or freeing one, we can derive the state of the
435 * other. That is, if we allocate a small block, and both were
436 * free, the remainder of the region must be split into blocks.
437 * If a block is freed, and its buddy is also free, then this
438 * triggers coalescing into a block of larger size.
439 *
440 * -- wli
441 */
442
Nick Piggin48db57f2006-01-08 01:00:42 -0800443static inline void __free_one_page(struct page *page,
Mel Gormaned0ae212009-06-16 15:32:07 -0700444 struct zone *zone, unsigned int order,
445 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446{
447 unsigned long page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448
Nick Piggin224abf92006-01-06 00:11:11 -0800449 if (unlikely(PageCompound(page)))
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800450 if (unlikely(destroy_compound_page(page, order)))
451 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452
Mel Gormaned0ae212009-06-16 15:32:07 -0700453 VM_BUG_ON(migratetype == -1);
454
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
456
Mel Gormanf2260e62009-06-16 15:32:13 -0700457 VM_BUG_ON(page_idx & ((1 << order) - 1));
Nick Piggin725d7042006-09-25 23:30:55 -0700458 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 while (order < MAX_ORDER-1) {
461 unsigned long combined_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 struct page *buddy;
463
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 buddy = __page_find_buddy(page, page_idx, order);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700465 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700466 break;
Nick Piggin13e74442006-01-06 00:10:58 -0800467
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700468 /* Our buddy is free, merge with it and move up one order. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 list_del(&buddy->lru);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700470 zone->free_area[order].nr_free--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 rmv_page_order(buddy);
Nick Piggin13e74442006-01-06 00:10:58 -0800472 combined_idx = __find_combined_index(page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 page = page + (combined_idx - page_idx);
474 page_idx = combined_idx;
475 order++;
476 }
477 set_page_order(page, order);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700478 list_add(&page->lru,
479 &zone->free_area[order].free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 zone->free_area[order].nr_free++;
481}
482
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700483#ifdef CONFIG_HAVE_MLOCKED_PAGE_BIT
484/*
485 * free_page_mlock() -- clean up attempts to free and mlocked() page.
486 * Page should not be on lru, so no need to fix that up.
487 * free_pages_check() will verify...
488 */
489static inline void free_page_mlock(struct page *page)
490{
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700491 __dec_zone_page_state(page, NR_MLOCK);
492 __count_vm_event(UNEVICTABLE_MLOCKFREED);
493}
494#else
495static void free_page_mlock(struct page *page) { }
496#endif
497
Nick Piggin224abf92006-01-06 00:11:11 -0800498static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499{
Nick Piggin92be2e32006-01-06 00:10:57 -0800500 if (unlikely(page_mapcount(page) |
501 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700502 (atomic_read(&page->_count) != 0) |
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800503 (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800504 bad_page(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800505 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800506 }
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800507 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
508 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
509 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510}
511
512/*
513 * Frees a list of pages.
514 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700515 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 *
517 * If the zone was previously in an "all pages pinned" state then look to
518 * see if this freeing clears that state.
519 *
520 * And clear the zone's pages_scanned counter, to hold off the "all pages are
521 * pinned" detection logic.
522 */
Nick Piggin48db57f2006-01-08 01:00:42 -0800523static void free_pages_bulk(struct zone *zone, int count,
524 struct list_head *list, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525{
Nick Pigginc54ad302006-01-06 00:10:56 -0800526 spin_lock(&zone->lock);
David Rientjese815af92007-10-16 23:25:54 -0700527 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700529
530 __mod_zone_page_state(zone, NR_FREE_PAGES, count << order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800531 while (count--) {
532 struct page *page;
533
Nick Piggin725d7042006-09-25 23:30:55 -0700534 VM_BUG_ON(list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 page = list_entry(list->prev, struct page, lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800536 /* have to delete it as __free_one_page list manipulates */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 list_del(&page->lru);
Mel Gormaned0ae212009-06-16 15:32:07 -0700538 __free_one_page(page, zone, order, page_private(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800540 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541}
542
Mel Gormaned0ae212009-06-16 15:32:07 -0700543static void free_one_page(struct zone *zone, struct page *page, int order,
544 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800545{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700546 spin_lock(&zone->lock);
David Rientjese815af92007-10-16 23:25:54 -0700547 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700548 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700549
550 __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
Mel Gormaned0ae212009-06-16 15:32:07 -0700551 __free_one_page(page, zone, order, migratetype);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700552 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800553}
554
555static void __free_pages_ok(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556{
Nick Pigginc54ad302006-01-06 00:10:56 -0800557 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 int i;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800559 int bad = 0;
Johannes Weinerc2773312009-06-19 19:30:56 +0200560 int wasMlocked = TestClearPageMlocked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100562 kmemcheck_free_shadow(page, order);
563
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 for (i = 0 ; i < (1 << order) ; ++i)
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800565 bad += free_pages_check(page + i);
566 if (bad)
Hugh Dickins689bceb2005-11-21 21:32:20 -0800567 return;
568
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700569 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -0700570 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700571 debug_check_no_obj_freed(page_address(page),
572 PAGE_SIZE << order);
573 }
Nick Piggindafb1362006-10-11 01:21:30 -0700574 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800575 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700576
Nick Pigginc54ad302006-01-06 00:10:56 -0800577 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +0200578 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -0700579 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700580 __count_vm_events(PGFREE, 1 << order);
Mel Gormaned0ae212009-06-16 15:32:07 -0700581 free_one_page(page_zone(page), page, order,
582 get_pageblock_migratetype(page));
Nick Pigginc54ad302006-01-06 00:10:56 -0800583 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584}
585
David Howellsa226f6c2006-01-06 00:11:08 -0800586/*
587 * permit the bootmem allocator to evade page validation on high-order frees
588 */
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700589void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800590{
591 if (order == 0) {
592 __ClearPageReserved(page);
593 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800594 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800595 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800596 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800597 int loop;
598
Nick Piggin545b1ea2006-03-22 00:08:07 -0800599 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800600 for (loop = 0; loop < BITS_PER_LONG; loop++) {
601 struct page *p = &page[loop];
602
Nick Piggin545b1ea2006-03-22 00:08:07 -0800603 if (loop + 1 < BITS_PER_LONG)
604 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800605 __ClearPageReserved(p);
606 set_page_count(p, 0);
607 }
608
Nick Piggin7835e982006-03-22 00:08:40 -0800609 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800610 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800611 }
612}
613
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614
615/*
616 * The order of subdivision here is critical for the IO subsystem.
617 * Please do not alter this order without good reasons and regression
618 * testing. Specifically, as large blocks of memory are subdivided,
619 * the order in which smaller blocks are delivered depends on the order
620 * they're subdivided in this function. This is the primary factor
621 * influencing the order in which pages are delivered to the IO
622 * subsystem according to empirical testing, and this is also justified
623 * by considering the behavior of a buddy system containing a single
624 * large block of memory acted on by a series of small allocations.
625 * This behavior is a critical factor in sglist merging's success.
626 *
627 * -- wli
628 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800629static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700630 int low, int high, struct free_area *area,
631 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632{
633 unsigned long size = 1 << high;
634
635 while (high > low) {
636 area--;
637 high--;
638 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700639 VM_BUG_ON(bad_range(zone, &page[size]));
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700640 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 area->nr_free++;
642 set_page_order(&page[size], high);
643 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644}
645
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646/*
647 * This page is about to be returned from the page allocator
648 */
Nick Piggin17cf4402006-03-22 00:08:41 -0800649static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650{
Nick Piggin92be2e32006-01-06 00:10:57 -0800651 if (unlikely(page_mapcount(page) |
652 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700653 (atomic_read(&page->_count) != 0) |
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800654 (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800655 bad_page(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800656 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800657 }
Hugh Dickins689bceb2005-11-21 21:32:20 -0800658
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700659 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800660 set_page_refcounted(page);
Nick Piggincc102502006-12-06 20:32:00 -0800661
662 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800664
665 if (gfp_flags & __GFP_ZERO)
666 prep_zero_page(page, order, gfp_flags);
667
668 if (order && (gfp_flags & __GFP_COMP))
669 prep_compound_page(page, order);
670
Hugh Dickins689bceb2005-11-21 21:32:20 -0800671 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672}
673
Mel Gorman56fd56b2007-10-16 01:25:58 -0700674/*
675 * Go through the free lists for the given migratetype and remove
676 * the smallest available page from the freelists
677 */
Mel Gorman728ec982009-06-16 15:32:04 -0700678static inline
679struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -0700680 int migratetype)
681{
682 unsigned int current_order;
683 struct free_area * area;
684 struct page *page;
685
686 /* Find a page of the appropriate size in the preferred list */
687 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
688 area = &(zone->free_area[current_order]);
689 if (list_empty(&area->free_list[migratetype]))
690 continue;
691
692 page = list_entry(area->free_list[migratetype].next,
693 struct page, lru);
694 list_del(&page->lru);
695 rmv_page_order(page);
696 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -0700697 expand(zone, page, order, current_order, area, migratetype);
698 return page;
699 }
700
701 return NULL;
702}
703
704
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700705/*
706 * This array describes the order lists are fallen back to when
707 * the free lists for the desirable migrate type are depleted
708 */
709static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
Mel Gorman64c5e132007-10-16 01:25:59 -0700710 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
711 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
712 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
713 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700714};
715
Mel Gormanc361be52007-10-16 01:25:51 -0700716/*
717 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -0700718 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -0700719 * boundary. If alignment is required, use move_freepages_block()
720 */
Adrian Bunkb69a7282008-07-23 21:28:12 -0700721static int move_freepages(struct zone *zone,
722 struct page *start_page, struct page *end_page,
723 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700724{
725 struct page *page;
726 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -0700727 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -0700728
729#ifndef CONFIG_HOLES_IN_ZONE
730 /*
731 * page_zone is not safe to call in this context when
732 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
733 * anyway as we check zone boundaries in move_freepages_block().
734 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -0700735 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -0700736 */
737 BUG_ON(page_zone(start_page) != page_zone(end_page));
738#endif
739
740 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -0700741 /* Make sure we are not inadvertently changing nodes */
742 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
743
Mel Gormanc361be52007-10-16 01:25:51 -0700744 if (!pfn_valid_within(page_to_pfn(page))) {
745 page++;
746 continue;
747 }
748
749 if (!PageBuddy(page)) {
750 page++;
751 continue;
752 }
753
754 order = page_order(page);
755 list_del(&page->lru);
756 list_add(&page->lru,
757 &zone->free_area[order].free_list[migratetype]);
758 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -0700759 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -0700760 }
761
Mel Gormand1003132007-10-16 01:26:00 -0700762 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -0700763}
764
Adrian Bunkb69a7282008-07-23 21:28:12 -0700765static int move_freepages_block(struct zone *zone, struct page *page,
766 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700767{
768 unsigned long start_pfn, end_pfn;
769 struct page *start_page, *end_page;
770
771 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -0700772 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -0700773 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -0700774 end_page = start_page + pageblock_nr_pages - 1;
775 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -0700776
777 /* Do not cross zone boundaries */
778 if (start_pfn < zone->zone_start_pfn)
779 start_page = page;
780 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
781 return 0;
782
783 return move_freepages(zone, start_page, end_page, migratetype);
784}
785
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700786/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -0700787static inline struct page *
788__rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700789{
790 struct free_area * area;
791 int current_order;
792 struct page *page;
793 int migratetype, i;
794
795 /* Find the largest possible block of pages in the other list */
796 for (current_order = MAX_ORDER-1; current_order >= order;
797 --current_order) {
798 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
799 migratetype = fallbacks[start_migratetype][i];
800
Mel Gorman56fd56b2007-10-16 01:25:58 -0700801 /* MIGRATE_RESERVE handled later if necessary */
802 if (migratetype == MIGRATE_RESERVE)
803 continue;
Mel Gormane0104872007-10-16 01:25:53 -0700804
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700805 area = &(zone->free_area[current_order]);
806 if (list_empty(&area->free_list[migratetype]))
807 continue;
808
809 page = list_entry(area->free_list[migratetype].next,
810 struct page, lru);
811 area->nr_free--;
812
813 /*
Mel Gormanc361be52007-10-16 01:25:51 -0700814 * If breaking a large block of pages, move all free
Mel Gorman46dafbc2007-10-16 01:25:55 -0700815 * pages to the preferred allocation list. If falling
816 * back for a reclaimable kernel allocation, be more
817 * agressive about taking ownership of free pages
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700818 */
Mel Gormand9c23402007-10-16 01:26:01 -0700819 if (unlikely(current_order >= (pageblock_order >> 1)) ||
Mel 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,
Mel Gormane084b2d2009-07-29 15:02:04 -0700885 int migratetype, int cold)
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 Gormane084b2d2009-07-29 15:02:04 -0700904 if (likely(cold == 0))
905 list_add(&page->lru, list);
906 else
907 list_add_tail(&page->lru, list);
Mel Gorman535131e2007-10-16 01:25:49 -0700908 set_page_private(page, migratetype);
Mel Gorman81eabcb2007-12-17 16:20:05 -0800909 list = &page->lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 }
Mel Gormanf2260e62009-06-16 15:32:13 -0700911 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -0800912 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800913 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914}
915
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700916#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800917/*
Christoph Lameter4037d452007-05-09 02:35:14 -0700918 * Called from the vmstat counter updater to drain pagesets of this
919 * currently executing processor on remote nodes after they have
920 * expired.
921 *
Christoph Lameter879336c2006-03-22 00:09:08 -0800922 * Note that this function must be called with the thread pinned to
923 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800924 */
Christoph Lameter4037d452007-05-09 02:35:14 -0700925void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700926{
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700927 unsigned long flags;
Christoph Lameter4037d452007-05-09 02:35:14 -0700928 int to_drain;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700929
Christoph Lameter4037d452007-05-09 02:35:14 -0700930 local_irq_save(flags);
931 if (pcp->count >= pcp->batch)
932 to_drain = pcp->batch;
933 else
934 to_drain = pcp->count;
935 free_pages_bulk(zone, to_drain, &pcp->list, 0);
936 pcp->count -= to_drain;
937 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700938}
939#endif
940
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800941/*
942 * Drain pages of the indicated processor.
943 *
944 * The processor must either be the current processor and the
945 * thread pinned to the current processor or a processor that
946 * is not online.
947 */
948static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949{
Nick Pigginc54ad302006-01-06 00:10:56 -0800950 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952
KOSAKI Motohiroee99c712009-03-31 15:19:31 -0700953 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 struct per_cpu_pageset *pset;
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800955 struct per_cpu_pages *pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700957 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800959 pcp = &pset->pcp;
960 local_irq_save(flags);
961 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
962 pcp->count = 0;
963 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 }
965}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800967/*
968 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
969 */
970void drain_local_pages(void *arg)
971{
972 drain_pages(smp_processor_id());
973}
974
975/*
976 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
977 */
978void drain_all_pages(void)
979{
Jens Axboe15c8b6c2008-05-09 09:39:44 +0200980 on_each_cpu(drain_local_pages, NULL, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800981}
982
Rafael J. Wysocki296699d2007-07-29 23:27:18 +0200983#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984
985void mark_free_pages(struct zone *zone)
986{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700987 unsigned long pfn, max_zone_pfn;
988 unsigned long flags;
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700989 int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 struct list_head *curr;
991
992 if (!zone->spanned_pages)
993 return;
994
995 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700996
997 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
998 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
999 if (pfn_valid(pfn)) {
1000 struct page *page = pfn_to_page(pfn);
1001
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001002 if (!swsusp_page_is_forbidden(page))
1003 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001004 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001006 for_each_migratetype_order(order, t) {
1007 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001008 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001010 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1011 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001012 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001013 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001014 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 spin_unlock_irqrestore(&zone->lock, flags);
1016}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001017#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018
1019/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 * Free a 0-order page
1021 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001022static void free_hot_cold_page(struct page *page, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023{
1024 struct zone *zone = page_zone(page);
1025 struct per_cpu_pages *pcp;
1026 unsigned long flags;
Johannes Weinerc2773312009-06-19 19:30:56 +02001027 int wasMlocked = TestClearPageMlocked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001029 kmemcheck_free_shadow(page, 0);
1030
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 if (PageAnon(page))
1032 page->mapping = NULL;
Nick Piggin224abf92006-01-06 00:11:11 -08001033 if (free_pages_check(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001034 return;
1035
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001036 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -07001037 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001038 debug_check_no_obj_freed(page_address(page), PAGE_SIZE);
1039 }
Nick Piggindafb1362006-10-11 01:21:30 -07001040 arch_free_page(page, 0);
Hugh Dickins689bceb2005-11-21 21:32:20 -08001041 kernel_map_pages(page, 1, 0);
1042
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001043 pcp = &zone_pcp(zone, get_cpu())->pcp;
Mel Gorman974709b2009-06-16 15:32:14 -07001044 set_page_private(page, get_pageblock_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +02001046 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -07001047 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001048 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001049
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001050 if (cold)
1051 list_add_tail(&page->lru, &pcp->list);
1052 else
1053 list_add(&page->lru, &pcp->list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001055 if (pcp->count >= pcp->high) {
1056 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
1057 pcp->count -= pcp->batch;
1058 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 local_irq_restore(flags);
1060 put_cpu();
1061}
1062
Harvey Harrison920c7a52008-02-04 22:29:26 -08001063void free_hot_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064{
1065 free_hot_cold_page(page, 0);
1066}
1067
Harvey Harrison920c7a52008-02-04 22:29:26 -08001068void free_cold_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069{
1070 free_hot_cold_page(page, 1);
1071}
1072
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001073/*
1074 * split_page takes a non-compound higher-order page, and splits it into
1075 * n (1<<order) sub-pages: page[0..n]
1076 * Each sub-page must be freed individually.
1077 *
1078 * Note: this is probably too low level an operation for use in drivers.
1079 * Please consult with lkml before using this in your driver.
1080 */
1081void split_page(struct page *page, unsigned int order)
1082{
1083 int i;
1084
Nick Piggin725d7042006-09-25 23:30:55 -07001085 VM_BUG_ON(PageCompound(page));
1086 VM_BUG_ON(!page_count(page));
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001087
1088#ifdef CONFIG_KMEMCHECK
1089 /*
1090 * Split shadow pages too, because free(page[0]) would
1091 * otherwise free the whole shadow.
1092 */
1093 if (kmemcheck_page_is_tracked(page))
1094 split_page(virt_to_page(page[0].shadow), order);
1095#endif
1096
Nick Piggin7835e982006-03-22 00:08:40 -08001097 for (i = 1; i < (1 << order); i++)
1098 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001099}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001100
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101/*
1102 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1103 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1104 * or two.
1105 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07001106static inline
1107struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001108 struct zone *zone, int order, gfp_t gfp_flags,
1109 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110{
1111 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001112 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 int cold = !!(gfp_flags & __GFP_COLD);
Nick Piggina74609f2006-01-06 00:11:20 -08001114 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115
Hugh Dickins689bceb2005-11-21 21:32:20 -08001116again:
Nick Piggina74609f2006-01-06 00:11:20 -08001117 cpu = get_cpu();
Nick Piggin48db57f2006-01-08 01:00:42 -08001118 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 struct per_cpu_pages *pcp;
1120
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001121 pcp = &zone_pcp(zone, cpu)->pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001123 if (!pcp->count) {
nkalmala941c7102006-11-02 22:07:04 -08001124 pcp->count = rmqueue_bulk(zone, 0,
Mel Gormane084b2d2009-07-29 15:02:04 -07001125 pcp->batch, &pcp->list,
1126 migratetype, cold);
Nick Piggina74609f2006-01-06 00:11:20 -08001127 if (unlikely(!pcp->count))
1128 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 }
Mel Gorman535131e2007-10-16 01:25:49 -07001130
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001131 /* Find a page of the appropriate migrate type */
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001132 if (cold) {
1133 list_for_each_entry_reverse(page, &pcp->list, lru)
1134 if (page_private(page) == migratetype)
1135 break;
1136 } else {
1137 list_for_each_entry(page, &pcp->list, lru)
1138 if (page_private(page) == migratetype)
1139 break;
1140 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001141
1142 /* Allocate more to the pcp list if necessary */
1143 if (unlikely(&page->lru == &pcp->list)) {
Mel Gorman535131e2007-10-16 01:25:49 -07001144 pcp->count += rmqueue_bulk(zone, 0,
Mel Gormane084b2d2009-07-29 15:02:04 -07001145 pcp->batch, &pcp->list,
1146 migratetype, cold);
Mel Gorman535131e2007-10-16 01:25:49 -07001147 page = list_entry(pcp->list.next, struct page, lru);
Mel Gorman535131e2007-10-16 01:25:49 -07001148 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001149
1150 list_del(&page->lru);
1151 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001152 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07001153 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1154 /*
1155 * __GFP_NOFAIL is not to be used in new code.
1156 *
1157 * All __GFP_NOFAIL callers should be fixed so that they
1158 * properly detect and handle allocation failures.
1159 *
1160 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07001161 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07001162 * __GFP_NOFAIL.
1163 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07001164 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07001165 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001167 page = __rmqueue(zone, order, migratetype);
Mel Gormanf2260e62009-06-16 15:32:13 -07001168 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order));
Nick Piggina74609f2006-01-06 00:11:20 -08001169 spin_unlock(&zone->lock);
1170 if (!page)
1171 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 }
1173
Christoph Lameterf8891e52006-06-30 01:55:45 -07001174 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Mel Gorman18ea7e72008-04-28 02:12:14 -07001175 zone_statistics(preferred_zone, zone);
Nick Piggina74609f2006-01-06 00:11:20 -08001176 local_irq_restore(flags);
1177 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178
Nick Piggin725d7042006-09-25 23:30:55 -07001179 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -08001180 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001181 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001183
1184failed:
1185 local_irq_restore(flags);
1186 put_cpu();
1187 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188}
1189
Mel Gorman41858962009-06-16 15:32:12 -07001190/* The ALLOC_WMARK bits are used as an index to zone->watermark */
1191#define ALLOC_WMARK_MIN WMARK_MIN
1192#define ALLOC_WMARK_LOW WMARK_LOW
1193#define ALLOC_WMARK_HIGH WMARK_HIGH
1194#define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1195
1196/* Mask to get the watermark bits */
1197#define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1198
Nick Piggin31488902005-11-28 13:44:03 -08001199#define ALLOC_HARDER 0x10 /* try to alloc harder */
1200#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1201#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001202
Akinobu Mita933e3122006-12-08 02:39:45 -08001203#ifdef CONFIG_FAIL_PAGE_ALLOC
1204
1205static struct fail_page_alloc_attr {
1206 struct fault_attr attr;
1207
1208 u32 ignore_gfp_highmem;
1209 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001210 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001211
1212#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1213
1214 struct dentry *ignore_gfp_highmem_file;
1215 struct dentry *ignore_gfp_wait_file;
Akinobu Mita54114992007-07-15 23:40:23 -07001216 struct dentry *min_order_file;
Akinobu Mita933e3122006-12-08 02:39:45 -08001217
1218#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1219
1220} fail_page_alloc = {
1221 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001222 .ignore_gfp_wait = 1,
1223 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001224 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001225};
1226
1227static int __init setup_fail_page_alloc(char *str)
1228{
1229 return setup_fault_attr(&fail_page_alloc.attr, str);
1230}
1231__setup("fail_page_alloc=", setup_fail_page_alloc);
1232
1233static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1234{
Akinobu Mita54114992007-07-15 23:40:23 -07001235 if (order < fail_page_alloc.min_order)
1236 return 0;
Akinobu Mita933e3122006-12-08 02:39:45 -08001237 if (gfp_mask & __GFP_NOFAIL)
1238 return 0;
1239 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1240 return 0;
1241 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1242 return 0;
1243
1244 return should_fail(&fail_page_alloc.attr, 1 << order);
1245}
1246
1247#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1248
1249static int __init fail_page_alloc_debugfs(void)
1250{
1251 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1252 struct dentry *dir;
1253 int err;
1254
1255 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1256 "fail_page_alloc");
1257 if (err)
1258 return err;
1259 dir = fail_page_alloc.attr.dentries.dir;
1260
1261 fail_page_alloc.ignore_gfp_wait_file =
1262 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1263 &fail_page_alloc.ignore_gfp_wait);
1264
1265 fail_page_alloc.ignore_gfp_highmem_file =
1266 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1267 &fail_page_alloc.ignore_gfp_highmem);
Akinobu Mita54114992007-07-15 23:40:23 -07001268 fail_page_alloc.min_order_file =
1269 debugfs_create_u32("min-order", mode, dir,
1270 &fail_page_alloc.min_order);
Akinobu Mita933e3122006-12-08 02:39:45 -08001271
1272 if (!fail_page_alloc.ignore_gfp_wait_file ||
Akinobu Mita54114992007-07-15 23:40:23 -07001273 !fail_page_alloc.ignore_gfp_highmem_file ||
1274 !fail_page_alloc.min_order_file) {
Akinobu Mita933e3122006-12-08 02:39:45 -08001275 err = -ENOMEM;
1276 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1277 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
Akinobu Mita54114992007-07-15 23:40:23 -07001278 debugfs_remove(fail_page_alloc.min_order_file);
Akinobu Mita933e3122006-12-08 02:39:45 -08001279 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1280 }
1281
1282 return err;
1283}
1284
1285late_initcall(fail_page_alloc_debugfs);
1286
1287#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1288
1289#else /* CONFIG_FAIL_PAGE_ALLOC */
1290
1291static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1292{
1293 return 0;
1294}
1295
1296#endif /* CONFIG_FAIL_PAGE_ALLOC */
1297
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298/*
1299 * Return 1 if free pages are above 'mark'. This takes into account the order
1300 * of the allocation.
1301 */
1302int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001303 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304{
1305 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001306 long min = mark;
1307 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 int o;
1309
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001310 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001312 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 min -= min / 4;
1314
1315 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1316 return 0;
1317 for (o = 0; o < order; o++) {
1318 /* At the next order, this order's pages become unavailable */
1319 free_pages -= z->free_area[o].nr_free << o;
1320
1321 /* Require fewer higher order pages to be free */
1322 min >>= 1;
1323
1324 if (free_pages <= min)
1325 return 0;
1326 }
1327 return 1;
1328}
1329
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001330#ifdef CONFIG_NUMA
1331/*
1332 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1333 * skip over zones that are not allowed by the cpuset, or that have
1334 * been recently (in last second) found to be nearly full. See further
1335 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001336 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001337 *
1338 * If the zonelist cache is present in the passed in zonelist, then
1339 * returns a pointer to the allowed node mask (either the current
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001340 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001341 *
1342 * If the zonelist cache is not available for this zonelist, does
1343 * nothing and returns NULL.
1344 *
1345 * If the fullzones BITMAP in the zonelist cache is stale (more than
1346 * a second since last zap'd) then we zap it out (clear its bits.)
1347 *
1348 * We hold off even calling zlc_setup, until after we've checked the
1349 * first zone in the zonelist, on the theory that most allocations will
1350 * be satisfied from that first zone, so best to examine that zone as
1351 * quickly as we can.
1352 */
1353static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1354{
1355 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1356 nodemask_t *allowednodes; /* zonelist_cache approximation */
1357
1358 zlc = zonelist->zlcache_ptr;
1359 if (!zlc)
1360 return NULL;
1361
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001362 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001363 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1364 zlc->last_full_zap = jiffies;
1365 }
1366
1367 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1368 &cpuset_current_mems_allowed :
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001369 &node_states[N_HIGH_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001370 return allowednodes;
1371}
1372
1373/*
1374 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1375 * if it is worth looking at further for free memory:
1376 * 1) Check that the zone isn't thought to be full (doesn't have its
1377 * bit set in the zonelist_cache fullzones BITMAP).
1378 * 2) Check that the zones node (obtained from the zonelist_cache
1379 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1380 * Return true (non-zero) if zone is worth looking at further, or
1381 * else return false (zero) if it is not.
1382 *
1383 * This check -ignores- the distinction between various watermarks,
1384 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1385 * found to be full for any variation of these watermarks, it will
1386 * be considered full for up to one second by all requests, unless
1387 * we are so low on memory on all allowed nodes that we are forced
1388 * into the second scan of the zonelist.
1389 *
1390 * In the second scan we ignore this zonelist cache and exactly
1391 * apply the watermarks to all zones, even it is slower to do so.
1392 * We are low on memory in the second scan, and should leave no stone
1393 * unturned looking for a free page.
1394 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001395static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001396 nodemask_t *allowednodes)
1397{
1398 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1399 int i; /* index of *z in zonelist zones */
1400 int n; /* node that zone *z is on */
1401
1402 zlc = zonelist->zlcache_ptr;
1403 if (!zlc)
1404 return 1;
1405
Mel Gormandd1a2392008-04-28 02:12:17 -07001406 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001407 n = zlc->z_to_n[i];
1408
1409 /* This zone is worth trying if it is allowed but not full */
1410 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1411}
1412
1413/*
1414 * Given 'z' scanning a zonelist, set the corresponding bit in
1415 * zlc->fullzones, so that subsequent attempts to allocate a page
1416 * from that zone don't waste time re-examining it.
1417 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001418static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001419{
1420 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1421 int i; /* index of *z in zonelist zones */
1422
1423 zlc = zonelist->zlcache_ptr;
1424 if (!zlc)
1425 return;
1426
Mel Gormandd1a2392008-04-28 02:12:17 -07001427 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001428
1429 set_bit(i, zlc->fullzones);
1430}
1431
1432#else /* CONFIG_NUMA */
1433
1434static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1435{
1436 return NULL;
1437}
1438
Mel Gormandd1a2392008-04-28 02:12:17 -07001439static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001440 nodemask_t *allowednodes)
1441{
1442 return 1;
1443}
1444
Mel Gormandd1a2392008-04-28 02:12:17 -07001445static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001446{
1447}
1448#endif /* CONFIG_NUMA */
1449
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001450/*
Paul Jackson0798e512006-12-06 20:31:38 -08001451 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001452 * a page.
1453 */
1454static struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001455get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
Mel Gorman5117f452009-06-16 15:31:59 -07001456 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
Mel Gorman3dd28262009-06-16 15:32:00 -07001457 struct zone *preferred_zone, int migratetype)
Martin Hicks753ee722005-06-21 17:14:41 -07001458{
Mel Gormandd1a2392008-04-28 02:12:17 -07001459 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001460 struct page *page = NULL;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001461 int classzone_idx;
Mel Gorman5117f452009-06-16 15:31:59 -07001462 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001463 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1464 int zlc_active = 0; /* set if using zonelist_cache */
1465 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gorman54a6eb52008-04-28 02:12:16 -07001466
Mel Gorman19770b32008-04-28 02:12:18 -07001467 classzone_idx = zone_idx(preferred_zone);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001468zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001469 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001470 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001471 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1472 */
Mel Gorman19770b32008-04-28 02:12:18 -07001473 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1474 high_zoneidx, nodemask) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001475 if (NUMA_BUILD && zlc_active &&
1476 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1477 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001478 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001479 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001480 goto try_next_zone;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001481
Mel Gorman41858962009-06-16 15:32:12 -07001482 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001483 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001484 unsigned long mark;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001485 int ret;
1486
Mel Gorman41858962009-06-16 15:32:12 -07001487 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
Mel Gormanfa5e0842009-06-16 15:33:22 -07001488 if (zone_watermark_ok(zone, order, mark,
1489 classzone_idx, alloc_flags))
1490 goto try_this_zone;
1491
1492 if (zone_reclaim_mode == 0)
1493 goto this_zone_full;
1494
1495 ret = zone_reclaim(zone, gfp_mask, order);
1496 switch (ret) {
1497 case ZONE_RECLAIM_NOSCAN:
1498 /* did not scan */
1499 goto try_next_zone;
1500 case ZONE_RECLAIM_FULL:
1501 /* scanned but unreclaimable */
1502 goto this_zone_full;
1503 default:
1504 /* did we reclaim enough */
1505 if (!zone_watermark_ok(zone, order, mark,
1506 classzone_idx, alloc_flags))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001507 goto this_zone_full;
Paul Jackson0798e512006-12-06 20:31:38 -08001508 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001509 }
1510
Mel Gormanfa5e0842009-06-16 15:33:22 -07001511try_this_zone:
Mel Gorman3dd28262009-06-16 15:32:00 -07001512 page = buffered_rmqueue(preferred_zone, zone, order,
1513 gfp_mask, migratetype);
Paul Jackson0798e512006-12-06 20:31:38 -08001514 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001515 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001516this_zone_full:
1517 if (NUMA_BUILD)
1518 zlc_mark_zone_full(zonelist, z);
1519try_next_zone:
Christoph Lameter62bc62a2009-06-16 15:32:15 -07001520 if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) {
Mel Gormand395b732009-06-16 15:32:09 -07001521 /*
1522 * we do zlc_setup after the first zone is tried but only
1523 * if there are multiple nodes make it worthwhile
1524 */
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001525 allowednodes = zlc_setup(zonelist, alloc_flags);
1526 zlc_active = 1;
1527 did_zlc_setup = 1;
1528 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07001529 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001530
1531 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1532 /* Disable zlc cache for second zonelist scan */
1533 zlc_active = 0;
1534 goto zonelist_scan;
1535 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001536 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07001537}
1538
Mel Gorman11e33f62009-06-16 15:31:57 -07001539static inline int
1540should_alloc_retry(gfp_t gfp_mask, unsigned int order,
1541 unsigned long pages_reclaimed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542{
Mel Gorman11e33f62009-06-16 15:31:57 -07001543 /* Do not loop if specifically requested */
1544 if (gfp_mask & __GFP_NORETRY)
1545 return 0;
1546
1547 /*
1548 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1549 * means __GFP_NOFAIL, but that may not be true in other
1550 * implementations.
1551 */
1552 if (order <= PAGE_ALLOC_COSTLY_ORDER)
1553 return 1;
1554
1555 /*
1556 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1557 * specified, then we retry until we no longer reclaim any pages
1558 * (above), or we've reclaimed an order of pages at least as
1559 * large as the allocation's order. In both cases, if the
1560 * allocation still fails, we stop retrying.
1561 */
1562 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
1563 return 1;
1564
1565 /*
1566 * Don't let big-order allocations loop unless the caller
1567 * explicitly requests that.
1568 */
1569 if (gfp_mask & __GFP_NOFAIL)
1570 return 1;
1571
1572 return 0;
1573}
1574
1575static inline struct page *
1576__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1577 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001578 nodemask_t *nodemask, struct zone *preferred_zone,
1579 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001580{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582
Mel Gorman11e33f62009-06-16 15:31:57 -07001583 /* Acquire the OOM killer lock for the zones in zonelist */
1584 if (!try_set_zone_oom(zonelist, gfp_mask)) {
1585 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 return NULL;
1587 }
Jens Axboe6b1de912005-11-17 21:35:02 +01001588
Mel Gorman11e33f62009-06-16 15:31:57 -07001589 /*
1590 * Go through the zonelist yet one more time, keep very high watermark
1591 * here, this is only to catch a parallel oom killing, we must fail if
1592 * we're still under heavy pressure.
1593 */
1594 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1595 order, zonelist, high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001596 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07001597 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001598 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07001599 goto out;
1600
1601 /* The OOM killer will not help higher order allocs */
David Rientjes82553a92009-06-16 15:32:58 -07001602 if (order > PAGE_ALLOC_COSTLY_ORDER && !(gfp_mask & __GFP_NOFAIL))
Mel Gorman11e33f62009-06-16 15:31:57 -07001603 goto out;
1604
1605 /* Exhausted what can be done so it's blamo time */
1606 out_of_memory(zonelist, gfp_mask, order);
1607
1608out:
1609 clear_zonelist_oom(zonelist, gfp_mask);
1610 return page;
1611}
1612
1613/* The really slow allocator path where we enter direct reclaim */
1614static inline struct page *
1615__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
1616 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001617 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001618 int migratetype, unsigned long *did_some_progress)
Mel Gorman11e33f62009-06-16 15:31:57 -07001619{
1620 struct page *page = NULL;
1621 struct reclaim_state reclaim_state;
1622 struct task_struct *p = current;
1623
1624 cond_resched();
1625
1626 /* We now go into synchronous reclaim */
1627 cpuset_memory_pressure_bump();
1628
1629 /*
1630 * The task's cpuset might have expanded its set of allowable nodes
1631 */
1632 p->flags |= PF_MEMALLOC;
1633 lockdep_set_current_reclaim_state(gfp_mask);
1634 reclaim_state.reclaimed_slab = 0;
1635 p->reclaim_state = &reclaim_state;
1636
1637 *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
1638
1639 p->reclaim_state = NULL;
1640 lockdep_clear_current_reclaim_state();
1641 p->flags &= ~PF_MEMALLOC;
1642
1643 cond_resched();
1644
1645 if (order != 0)
1646 drain_all_pages();
1647
1648 if (likely(*did_some_progress))
1649 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001650 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001651 alloc_flags, preferred_zone,
1652 migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001653 return page;
1654}
1655
Mel Gorman11e33f62009-06-16 15:31:57 -07001656/*
1657 * This is called in the allocator slow-path if the allocation request is of
1658 * sufficient urgency to ignore watermarks and take other desperate measures
1659 */
1660static inline struct page *
1661__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
1662 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001663 nodemask_t *nodemask, struct zone *preferred_zone,
1664 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001665{
1666 struct page *page;
1667
1668 do {
1669 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001670 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
Mel Gorman3dd28262009-06-16 15:32:00 -07001671 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001672
1673 if (!page && gfp_mask & __GFP_NOFAIL)
Jens Axboe8aa7e842009-07-09 14:52:32 +02001674 congestion_wait(BLK_RW_ASYNC, HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07001675 } while (!page && (gfp_mask & __GFP_NOFAIL));
1676
1677 return page;
1678}
1679
1680static inline
1681void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
1682 enum zone_type high_zoneidx)
1683{
1684 struct zoneref *z;
1685 struct zone *zone;
1686
1687 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
1688 wakeup_kswapd(zone, order);
1689}
1690
Peter Zijlstra341ce062009-06-16 15:32:02 -07001691static inline int
1692gfp_to_alloc_flags(gfp_t gfp_mask)
1693{
1694 struct task_struct *p = current;
1695 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
1696 const gfp_t wait = gfp_mask & __GFP_WAIT;
1697
Mel Gormana56f57f2009-06-16 15:32:02 -07001698 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
1699 BUILD_BUG_ON(__GFP_HIGH != ALLOC_HIGH);
1700
Peter Zijlstra341ce062009-06-16 15:32:02 -07001701 /*
1702 * The caller may dip into page reserves a bit more if the caller
1703 * cannot run direct reclaim, or if the caller has realtime scheduling
1704 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1705 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1706 */
Mel Gormana56f57f2009-06-16 15:32:02 -07001707 alloc_flags |= (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07001708
1709 if (!wait) {
1710 alloc_flags |= ALLOC_HARDER;
1711 /*
1712 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1713 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1714 */
1715 alloc_flags &= ~ALLOC_CPUSET;
1716 } else if (unlikely(rt_task(p)))
1717 alloc_flags |= ALLOC_HARDER;
1718
1719 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
1720 if (!in_interrupt() &&
1721 ((p->flags & PF_MEMALLOC) ||
1722 unlikely(test_thread_flag(TIF_MEMDIE))))
1723 alloc_flags |= ALLOC_NO_WATERMARKS;
1724 }
1725
1726 return alloc_flags;
1727}
1728
Mel Gorman11e33f62009-06-16 15:31:57 -07001729static inline struct page *
1730__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
1731 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001732 nodemask_t *nodemask, struct zone *preferred_zone,
1733 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001734{
1735 const gfp_t wait = gfp_mask & __GFP_WAIT;
1736 struct page *page = NULL;
1737 int alloc_flags;
1738 unsigned long pages_reclaimed = 0;
1739 unsigned long did_some_progress;
1740 struct task_struct *p = current;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001741
Christoph Lameter952f3b52006-12-06 20:33:26 -08001742 /*
Mel Gorman72807a72009-06-16 15:32:18 -07001743 * In the slowpath, we sanity check order to avoid ever trying to
1744 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
1745 * be using allocators in order of preference for an area that is
1746 * too large.
1747 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07001748 if (order >= MAX_ORDER) {
1749 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07001750 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07001751 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001752
Christoph Lameter952f3b52006-12-06 20:33:26 -08001753 /*
1754 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1755 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1756 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1757 * using a larger set of nodes after it has established that the
1758 * allowed per node queues are empty and that nodes are
1759 * over allocated.
1760 */
1761 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1762 goto nopage;
1763
Mel Gorman11e33f62009-06-16 15:31:57 -07001764 wake_all_kswapd(order, zonelist, high_zoneidx);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001765
Paul Jackson9bf22292005-09-06 15:18:12 -07001766 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001767 * OK, we're below the kswapd watermark and have kicked background
1768 * reclaim. Now things get more complex, so set up alloc_flags according
1769 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07001770 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07001771 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772
Mel Gorman11e33f62009-06-16 15:31:57 -07001773restart:
Peter Zijlstra341ce062009-06-16 15:32:02 -07001774 /* This is the last chance, in general, before the goto nopage. */
Mel Gorman19770b32008-04-28 02:12:18 -07001775 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
Peter Zijlstra341ce062009-06-16 15:32:02 -07001776 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
1777 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001778 if (page)
1779 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780
Kirill Korotaevb43a57b2006-12-06 20:32:27 -08001781rebalance:
Mel Gorman11e33f62009-06-16 15:31:57 -07001782 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07001783 if (alloc_flags & ALLOC_NO_WATERMARKS) {
1784 page = __alloc_pages_high_priority(gfp_mask, order,
1785 zonelist, high_zoneidx, nodemask,
1786 preferred_zone, migratetype);
1787 if (page)
1788 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 }
1790
1791 /* Atomic allocations - we can't balance anything */
1792 if (!wait)
1793 goto nopage;
1794
Peter Zijlstra341ce062009-06-16 15:32:02 -07001795 /* Avoid recursion of direct reclaim */
1796 if (p->flags & PF_MEMALLOC)
1797 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798
David Rientjes6583bb62009-07-29 15:02:06 -07001799 /* Avoid allocations with no watermarks from looping endlessly */
1800 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
1801 goto nopage;
1802
Mel Gorman11e33f62009-06-16 15:31:57 -07001803 /* Try direct reclaim and then allocating */
1804 page = __alloc_pages_direct_reclaim(gfp_mask, order,
1805 zonelist, high_zoneidx,
1806 nodemask,
Mel Gorman5117f452009-06-16 15:31:59 -07001807 alloc_flags, preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001808 migratetype, &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07001809 if (page)
1810 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 /*
Mel Gorman11e33f62009-06-16 15:31:57 -07001813 * If we failed to make any progress reclaiming, then we are
1814 * running out of options and have to consider going OOM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 */
Mel Gorman11e33f62009-06-16 15:31:57 -07001816 if (!did_some_progress) {
1817 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -07001818 if (oom_killer_disabled)
1819 goto nopage;
Mel Gorman11e33f62009-06-16 15:31:57 -07001820 page = __alloc_pages_may_oom(gfp_mask, order,
1821 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001822 nodemask, preferred_zone,
1823 migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001824 if (page)
1825 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826
Mel Gorman11e33f62009-06-16 15:31:57 -07001827 /*
David Rientjes82553a92009-06-16 15:32:58 -07001828 * The OOM killer does not trigger for high-order
1829 * ~__GFP_NOFAIL allocations so if no progress is being
1830 * made, there are no other options and retrying is
1831 * unlikely to help.
Mel Gorman11e33f62009-06-16 15:31:57 -07001832 */
David Rientjes82553a92009-06-16 15:32:58 -07001833 if (order > PAGE_ALLOC_COSTLY_ORDER &&
1834 !(gfp_mask & __GFP_NOFAIL))
Mel Gorman11e33f62009-06-16 15:31:57 -07001835 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 goto restart;
1838 }
1839 }
1840
Mel Gorman11e33f62009-06-16 15:31:57 -07001841 /* Check if we should retry the allocation */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07001842 pages_reclaimed += did_some_progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07001843 if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
1844 /* Wait for some write requests to complete then retry */
Jens Axboe8aa7e842009-07-09 14:52:32 +02001845 congestion_wait(BLK_RW_ASYNC, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 goto rebalance;
1847 }
1848
1849nopage:
1850 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1851 printk(KERN_WARNING "%s: page allocation failure."
1852 " order:%d, mode:0x%x\n",
1853 p->comm, order, gfp_mask);
1854 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001855 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 }
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001857 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858got_pg:
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001859 if (kmemcheck_enabled)
1860 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 return page;
Mel Gorman11e33f62009-06-16 15:31:57 -07001862
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863}
Mel Gorman11e33f62009-06-16 15:31:57 -07001864
1865/*
1866 * This is the 'heart' of the zoned buddy allocator.
1867 */
1868struct page *
1869__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
1870 struct zonelist *zonelist, nodemask_t *nodemask)
1871{
1872 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Mel Gorman5117f452009-06-16 15:31:59 -07001873 struct zone *preferred_zone;
Mel Gorman11e33f62009-06-16 15:31:57 -07001874 struct page *page;
Mel Gorman3dd28262009-06-16 15:32:00 -07001875 int migratetype = allocflags_to_migratetype(gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07001876
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10001877 gfp_mask &= gfp_allowed_mask;
1878
Mel Gorman11e33f62009-06-16 15:31:57 -07001879 lockdep_trace_alloc(gfp_mask);
1880
1881 might_sleep_if(gfp_mask & __GFP_WAIT);
1882
1883 if (should_fail_alloc_page(gfp_mask, order))
1884 return NULL;
1885
1886 /*
1887 * Check the zones suitable for the gfp_mask contain at least one
1888 * valid zone. It's possible to have an empty zonelist as a result
1889 * of GFP_THISNODE and a memoryless node
1890 */
1891 if (unlikely(!zonelist->_zonerefs->zone))
1892 return NULL;
1893
Mel Gorman5117f452009-06-16 15:31:59 -07001894 /* The preferred zone is used for statistics later */
1895 first_zones_zonelist(zonelist, high_zoneidx, nodemask, &preferred_zone);
1896 if (!preferred_zone)
1897 return NULL;
1898
1899 /* First allocation attempt */
Mel Gorman11e33f62009-06-16 15:31:57 -07001900 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001901 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07001902 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001903 if (unlikely(!page))
1904 page = __alloc_pages_slowpath(gfp_mask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001905 zonelist, high_zoneidx, nodemask,
Mel Gorman3dd28262009-06-16 15:32:00 -07001906 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001907
1908 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909}
Mel Gormand2391712009-06-16 15:31:52 -07001910EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911
1912/*
1913 * Common helper functions.
1914 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001915unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916{
1917 struct page * page;
1918 page = alloc_pages(gfp_mask, order);
1919 if (!page)
1920 return 0;
1921 return (unsigned long) page_address(page);
1922}
1923
1924EXPORT_SYMBOL(__get_free_pages);
1925
Harvey Harrison920c7a52008-02-04 22:29:26 -08001926unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927{
1928 struct page * page;
1929
1930 /*
1931 * get_zeroed_page() returns a 32-bit address, which cannot represent
1932 * a highmem page
1933 */
Nick Piggin725d7042006-09-25 23:30:55 -07001934 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935
1936 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1937 if (page)
1938 return (unsigned long) page_address(page);
1939 return 0;
1940}
1941
1942EXPORT_SYMBOL(get_zeroed_page);
1943
1944void __pagevec_free(struct pagevec *pvec)
1945{
1946 int i = pagevec_count(pvec);
1947
1948 while (--i >= 0)
1949 free_hot_cold_page(pvec->pages[i], pvec->cold);
1950}
1951
Harvey Harrison920c7a52008-02-04 22:29:26 -08001952void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953{
Nick Pigginb5810032005-10-29 18:16:12 -07001954 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 if (order == 0)
1956 free_hot_page(page);
1957 else
1958 __free_pages_ok(page, order);
1959 }
1960}
1961
1962EXPORT_SYMBOL(__free_pages);
1963
Harvey Harrison920c7a52008-02-04 22:29:26 -08001964void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965{
1966 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07001967 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 __free_pages(virt_to_page((void *)addr), order);
1969 }
1970}
1971
1972EXPORT_SYMBOL(free_pages);
1973
Timur Tabi2be0ffe2008-07-23 21:28:11 -07001974/**
1975 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1976 * @size: the number of bytes to allocate
1977 * @gfp_mask: GFP flags for the allocation
1978 *
1979 * This function is similar to alloc_pages(), except that it allocates the
1980 * minimum number of pages to satisfy the request. alloc_pages() can only
1981 * allocate memory in power-of-two pages.
1982 *
1983 * This function is also limited by MAX_ORDER.
1984 *
1985 * Memory allocated by this function must be released by free_pages_exact().
1986 */
1987void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
1988{
1989 unsigned int order = get_order(size);
1990 unsigned long addr;
1991
1992 addr = __get_free_pages(gfp_mask, order);
1993 if (addr) {
1994 unsigned long alloc_end = addr + (PAGE_SIZE << order);
1995 unsigned long used = addr + PAGE_ALIGN(size);
1996
Kevin Cernekee5bfd7562009-07-05 12:08:19 -07001997 split_page(virt_to_page((void *)addr), order);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07001998 while (used < alloc_end) {
1999 free_page(used);
2000 used += PAGE_SIZE;
2001 }
2002 }
2003
2004 return (void *)addr;
2005}
2006EXPORT_SYMBOL(alloc_pages_exact);
2007
2008/**
2009 * free_pages_exact - release memory allocated via alloc_pages_exact()
2010 * @virt: the value returned by alloc_pages_exact.
2011 * @size: size of allocation, same value as passed to alloc_pages_exact().
2012 *
2013 * Release the memory allocated by a previous call to alloc_pages_exact.
2014 */
2015void free_pages_exact(void *virt, size_t size)
2016{
2017 unsigned long addr = (unsigned long)virt;
2018 unsigned long end = addr + PAGE_ALIGN(size);
2019
2020 while (addr < end) {
2021 free_page(addr);
2022 addr += PAGE_SIZE;
2023 }
2024}
2025EXPORT_SYMBOL(free_pages_exact);
2026
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027static unsigned int nr_free_zone_pages(int offset)
2028{
Mel Gormandd1a2392008-04-28 02:12:17 -07002029 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002030 struct zone *zone;
2031
Martin J. Blighe310fd42005-07-29 22:59:18 -07002032 /* Just pick one node, since fallback list is circular */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 unsigned int sum = 0;
2034
Mel Gorman0e884602008-04-28 02:12:14 -07002035 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036
Mel Gorman54a6eb52008-04-28 02:12:16 -07002037 for_each_zone_zonelist(zone, z, zonelist, offset) {
Martin J. Blighe310fd42005-07-29 22:59:18 -07002038 unsigned long size = zone->present_pages;
Mel Gorman41858962009-06-16 15:32:12 -07002039 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07002040 if (size > high)
2041 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 }
2043
2044 return sum;
2045}
2046
2047/*
2048 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2049 */
2050unsigned int nr_free_buffer_pages(void)
2051{
Al Viroaf4ca452005-10-21 02:55:38 -04002052 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053}
Meelap Shahc2f1a552007-07-17 04:04:39 -07002054EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055
2056/*
2057 * Amount of free RAM allocatable within all zones
2058 */
2059unsigned int nr_free_pagecache_pages(void)
2060{
Mel Gorman2a1e2742007-07-17 04:03:12 -07002061 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002063
2064static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065{
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002066 if (NUMA_BUILD)
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08002067 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070void si_meminfo(struct sysinfo *val)
2071{
2072 val->totalram = totalram_pages;
2073 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002074 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 val->totalhigh = totalhigh_pages;
2077 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 val->mem_unit = PAGE_SIZE;
2079}
2080
2081EXPORT_SYMBOL(si_meminfo);
2082
2083#ifdef CONFIG_NUMA
2084void si_meminfo_node(struct sysinfo *val, int nid)
2085{
2086 pg_data_t *pgdat = NODE_DATA(nid);
2087
2088 val->totalram = pgdat->node_present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002089 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002090#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002092 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2093 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002094#else
2095 val->totalhigh = 0;
2096 val->freehigh = 0;
2097#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098 val->mem_unit = PAGE_SIZE;
2099}
2100#endif
2101
2102#define K(x) ((x) << (PAGE_SHIFT-10))
2103
2104/*
2105 * Show free area list (used inside shift_scroll-lock stuff)
2106 * We also calculate the percentage fragmentation. We do this by counting the
2107 * memory on each free list with the exception of the first item on the list.
2108 */
2109void show_free_areas(void)
2110{
Jes Sorensenc7241912006-09-27 01:50:05 -07002111 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 struct zone *zone;
2113
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002114 for_each_populated_zone(zone) {
Jes Sorensenc7241912006-09-27 01:50:05 -07002115 show_node(zone);
2116 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117
Dave Jones6b482c62005-11-10 15:45:56 -05002118 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 struct per_cpu_pageset *pageset;
2120
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002121 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002123 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2124 cpu, pageset->pcp.high,
2125 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 }
2127 }
2128
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002129 printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n"
2130 " inactive_file:%lu"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002131 " unevictable:%lu"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002132 " dirty:%lu writeback:%lu unstable:%lu\n"
Christoph Lameterd23ad422007-02-10 01:43:02 -08002133 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002134 global_page_state(NR_ACTIVE_ANON),
2135 global_page_state(NR_ACTIVE_FILE),
2136 global_page_state(NR_INACTIVE_ANON),
2137 global_page_state(NR_INACTIVE_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002138 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07002139 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07002140 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07002141 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08002142 global_page_state(NR_FREE_PAGES),
Christoph Lameter972d1a72006-09-25 23:31:51 -07002143 global_page_state(NR_SLAB_RECLAIMABLE) +
2144 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07002145 global_page_state(NR_FILE_MAPPED),
Andrew Mortona25700a2007-02-08 14:20:40 -08002146 global_page_state(NR_PAGETABLE),
2147 global_page_state(NR_BOUNCE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002149 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150 int i;
2151
2152 show_node(zone);
2153 printk("%s"
2154 " free:%lukB"
2155 " min:%lukB"
2156 " low:%lukB"
2157 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002158 " active_anon:%lukB"
2159 " inactive_anon:%lukB"
2160 " active_file:%lukB"
2161 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002162 " unevictable:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 " present:%lukB"
2164 " pages_scanned:%lu"
2165 " all_unreclaimable? %s"
2166 "\n",
2167 zone->name,
Christoph Lameterd23ad422007-02-10 01:43:02 -08002168 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07002169 K(min_wmark_pages(zone)),
2170 K(low_wmark_pages(zone)),
2171 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002172 K(zone_page_state(zone, NR_ACTIVE_ANON)),
2173 K(zone_page_state(zone, NR_INACTIVE_ANON)),
2174 K(zone_page_state(zone, NR_ACTIVE_FILE)),
2175 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002176 K(zone_page_state(zone, NR_UNEVICTABLE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 K(zone->present_pages),
2178 zone->pages_scanned,
David Rientjese815af92007-10-16 23:25:54 -07002179 (zone_is_all_unreclaimable(zone) ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 );
2181 printk("lowmem_reserve[]:");
2182 for (i = 0; i < MAX_NR_ZONES; i++)
2183 printk(" %lu", zone->lowmem_reserve[i]);
2184 printk("\n");
2185 }
2186
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002187 for_each_populated_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002188 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189
2190 show_node(zone);
2191 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192
2193 spin_lock_irqsave(&zone->lock, flags);
2194 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002195 nr[order] = zone->free_area[order].nr_free;
2196 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197 }
2198 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002199 for (order = 0; order < MAX_ORDER; order++)
2200 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 printk("= %lukB\n", K(total));
2202 }
2203
Larry Woodmane6f36022008-02-04 22:29:30 -08002204 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2205
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 show_swap_cache_info();
2207}
2208
Mel Gorman19770b32008-04-28 02:12:18 -07002209static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2210{
2211 zoneref->zone = zone;
2212 zoneref->zone_idx = zone_idx(zone);
2213}
2214
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215/*
2216 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08002217 *
2218 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002220static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2221 int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222{
Christoph Lameter1a932052006-01-06 00:11:16 -08002223 struct zone *zone;
2224
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002225 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002226 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08002227
2228 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002229 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08002230 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08002231 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002232 zoneref_set_zone(zone,
2233 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08002234 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08002236
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002237 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08002238 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239}
2240
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002241
2242/*
2243 * zonelist_order:
2244 * 0 = automatic detection of better ordering.
2245 * 1 = order by ([node] distance, -zonetype)
2246 * 2 = order by (-zonetype, [node] distance)
2247 *
2248 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2249 * the same zonelist. So only NUMA can configure this param.
2250 */
2251#define ZONELIST_ORDER_DEFAULT 0
2252#define ZONELIST_ORDER_NODE 1
2253#define ZONELIST_ORDER_ZONE 2
2254
2255/* zonelist order in the kernel.
2256 * set_zonelist_order() will set this to NODE or ZONE.
2257 */
2258static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2259static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2260
2261
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002263/* The value user specified ....changed by config */
2264static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2265/* string for sysctl */
2266#define NUMA_ZONELIST_ORDER_LEN 16
2267char numa_zonelist_order[16] = "default";
2268
2269/*
2270 * interface for configure zonelist ordering.
2271 * command line option "numa_zonelist_order"
2272 * = "[dD]efault - default, automatic configuration.
2273 * = "[nN]ode - order by node locality, then by zone within node
2274 * = "[zZ]one - order by zone, then by locality within zone
2275 */
2276
2277static int __parse_numa_zonelist_order(char *s)
2278{
2279 if (*s == 'd' || *s == 'D') {
2280 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2281 } else if (*s == 'n' || *s == 'N') {
2282 user_zonelist_order = ZONELIST_ORDER_NODE;
2283 } else if (*s == 'z' || *s == 'Z') {
2284 user_zonelist_order = ZONELIST_ORDER_ZONE;
2285 } else {
2286 printk(KERN_WARNING
2287 "Ignoring invalid numa_zonelist_order value: "
2288 "%s\n", s);
2289 return -EINVAL;
2290 }
2291 return 0;
2292}
2293
2294static __init int setup_numa_zonelist_order(char *s)
2295{
2296 if (s)
2297 return __parse_numa_zonelist_order(s);
2298 return 0;
2299}
2300early_param("numa_zonelist_order", setup_numa_zonelist_order);
2301
2302/*
2303 * sysctl handler for numa_zonelist_order
2304 */
2305int numa_zonelist_order_handler(ctl_table *table, int write,
2306 struct file *file, void __user *buffer, size_t *length,
2307 loff_t *ppos)
2308{
2309 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2310 int ret;
2311
2312 if (write)
2313 strncpy(saved_string, (char*)table->data,
2314 NUMA_ZONELIST_ORDER_LEN);
2315 ret = proc_dostring(table, write, file, buffer, length, ppos);
2316 if (ret)
2317 return ret;
2318 if (write) {
2319 int oldval = user_zonelist_order;
2320 if (__parse_numa_zonelist_order((char*)table->data)) {
2321 /*
2322 * bogus value. restore saved string
2323 */
2324 strncpy((char*)table->data, saved_string,
2325 NUMA_ZONELIST_ORDER_LEN);
2326 user_zonelist_order = oldval;
2327 } else if (oldval != user_zonelist_order)
2328 build_all_zonelists();
2329 }
2330 return 0;
2331}
2332
2333
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002334#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002335static int node_load[MAX_NUMNODES];
2336
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07002338 * 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 -07002339 * @node: node whose fallback list we're appending
2340 * @used_node_mask: nodemask_t of already used nodes
2341 *
2342 * We use a number of factors to determine which is the next node that should
2343 * appear on a given node's fallback list. The node should not have appeared
2344 * already in @node's fallback list, and it should be the next closest node
2345 * according to the distance array (which contains arbitrary distance values
2346 * from each node to each node in the system), and should also prefer nodes
2347 * with no CPUs, since presumably they'll have very little allocation pressure
2348 * on them otherwise.
2349 * It returns -1 if no node is found.
2350 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002351static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002353 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354 int min_val = INT_MAX;
2355 int best_node = -1;
Rusty Russella70f7302009-03-13 14:49:46 +10302356 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002358 /* Use the local node if we haven't already */
2359 if (!node_isset(node, *used_node_mask)) {
2360 node_set(node, *used_node_mask);
2361 return node;
2362 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002364 for_each_node_state(n, N_HIGH_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365
2366 /* Don't want a node to appear more than once */
2367 if (node_isset(n, *used_node_mask))
2368 continue;
2369
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 /* Use the distance array to find the distance */
2371 val = node_distance(node, n);
2372
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002373 /* Penalize nodes under us ("prefer the next node") */
2374 val += (n < node);
2375
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10302377 tmp = cpumask_of_node(n);
2378 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 val += PENALTY_FOR_NODE_WITH_CPUS;
2380
2381 /* Slight preference for less loaded node */
2382 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2383 val += node_load[n];
2384
2385 if (val < min_val) {
2386 min_val = val;
2387 best_node = n;
2388 }
2389 }
2390
2391 if (best_node >= 0)
2392 node_set(best_node, *used_node_mask);
2393
2394 return best_node;
2395}
2396
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002397
2398/*
2399 * Build zonelists ordered by node and zones within node.
2400 * This results in maximum locality--normal zone overflows into local
2401 * DMA zone, if any--but risks exhausting DMA zone.
2402 */
2403static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002405 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002407
Mel Gorman54a6eb52008-04-28 02:12:16 -07002408 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07002409 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07002410 ;
2411 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2412 MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002413 zonelist->_zonerefs[j].zone = NULL;
2414 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002415}
2416
2417/*
Christoph Lameter523b9452007-10-16 01:25:37 -07002418 * Build gfp_thisnode zonelists
2419 */
2420static void build_thisnode_zonelists(pg_data_t *pgdat)
2421{
Christoph Lameter523b9452007-10-16 01:25:37 -07002422 int j;
2423 struct zonelist *zonelist;
2424
Mel Gorman54a6eb52008-04-28 02:12:16 -07002425 zonelist = &pgdat->node_zonelists[1];
2426 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002427 zonelist->_zonerefs[j].zone = NULL;
2428 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07002429}
2430
2431/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002432 * Build zonelists ordered by zone and nodes within zones.
2433 * This results in conserving DMA zone[s] until all Normal memory is
2434 * exhausted, but results in overflowing to remote node while memory
2435 * may still exist in local DMA zone.
2436 */
2437static int node_order[MAX_NUMNODES];
2438
2439static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2440{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002441 int pos, j, node;
2442 int zone_type; /* needs to be signed */
2443 struct zone *z;
2444 struct zonelist *zonelist;
2445
Mel Gorman54a6eb52008-04-28 02:12:16 -07002446 zonelist = &pgdat->node_zonelists[0];
2447 pos = 0;
2448 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2449 for (j = 0; j < nr_nodes; j++) {
2450 node = node_order[j];
2451 z = &NODE_DATA(node)->node_zones[zone_type];
2452 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002453 zoneref_set_zone(z,
2454 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002455 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002456 }
2457 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002458 }
Mel Gormandd1a2392008-04-28 02:12:17 -07002459 zonelist->_zonerefs[pos].zone = NULL;
2460 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002461}
2462
2463static int default_zonelist_order(void)
2464{
2465 int nid, zone_type;
2466 unsigned long low_kmem_size,total_size;
2467 struct zone *z;
2468 int average_size;
2469 /*
2470 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2471 * If they are really small and used heavily, the system can fall
2472 * into OOM very easily.
2473 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2474 */
2475 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2476 low_kmem_size = 0;
2477 total_size = 0;
2478 for_each_online_node(nid) {
2479 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2480 z = &NODE_DATA(nid)->node_zones[zone_type];
2481 if (populated_zone(z)) {
2482 if (zone_type < ZONE_NORMAL)
2483 low_kmem_size += z->present_pages;
2484 total_size += z->present_pages;
2485 }
2486 }
2487 }
2488 if (!low_kmem_size || /* there are no DMA area. */
2489 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2490 return ZONELIST_ORDER_NODE;
2491 /*
2492 * look into each node's config.
2493 * If there is a node whose DMA/DMA32 memory is very big area on
2494 * local memory, NODE_ORDER may be suitable.
2495 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002496 average_size = total_size /
2497 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002498 for_each_online_node(nid) {
2499 low_kmem_size = 0;
2500 total_size = 0;
2501 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2502 z = &NODE_DATA(nid)->node_zones[zone_type];
2503 if (populated_zone(z)) {
2504 if (zone_type < ZONE_NORMAL)
2505 low_kmem_size += z->present_pages;
2506 total_size += z->present_pages;
2507 }
2508 }
2509 if (low_kmem_size &&
2510 total_size > average_size && /* ignore small node */
2511 low_kmem_size > total_size * 70/100)
2512 return ZONELIST_ORDER_NODE;
2513 }
2514 return ZONELIST_ORDER_ZONE;
2515}
2516
2517static void set_zonelist_order(void)
2518{
2519 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2520 current_zonelist_order = default_zonelist_order();
2521 else
2522 current_zonelist_order = user_zonelist_order;
2523}
2524
2525static void build_zonelists(pg_data_t *pgdat)
2526{
2527 int j, node, load;
2528 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002530 int local_node, prev_node;
2531 struct zonelist *zonelist;
2532 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533
2534 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07002535 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07002537 zonelist->_zonerefs[0].zone = NULL;
2538 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539 }
2540
2541 /* NUMA-aware ordering of nodes */
2542 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002543 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544 prev_node = local_node;
2545 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002546
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002547 memset(node_order, 0, sizeof(node_order));
2548 j = 0;
2549
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08002551 int distance = node_distance(local_node, node);
2552
2553 /*
2554 * If another node is sufficiently far away then it is better
2555 * to reclaim pages in a zone before going off node.
2556 */
2557 if (distance > RECLAIM_DISTANCE)
2558 zone_reclaim_mode = 1;
2559
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560 /*
2561 * We don't want to pressure a particular node.
2562 * So adding penalty to the first node in same
2563 * distance group to make it round-robin.
2564 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08002565 if (distance != node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002566 node_load[node] = load;
2567
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 prev_node = node;
2569 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002570 if (order == ZONELIST_ORDER_NODE)
2571 build_zonelists_in_node_order(pgdat, node);
2572 else
2573 node_order[j++] = node; /* remember order */
2574 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002576 if (order == ZONELIST_ORDER_ZONE) {
2577 /* calculate node order -- i.e., DMA last! */
2578 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 }
Christoph Lameter523b9452007-10-16 01:25:37 -07002580
2581 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582}
2583
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002584/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002585static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002586{
Mel Gorman54a6eb52008-04-28 02:12:16 -07002587 struct zonelist *zonelist;
2588 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07002589 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002590
Mel Gorman54a6eb52008-04-28 02:12:16 -07002591 zonelist = &pgdat->node_zonelists[0];
2592 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2593 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07002594 for (z = zonelist->_zonerefs; z->zone; z++)
2595 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002596}
2597
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002598
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599#else /* CONFIG_NUMA */
2600
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002601static void set_zonelist_order(void)
2602{
2603 current_zonelist_order = ZONELIST_ORDER_ZONE;
2604}
2605
2606static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607{
Christoph Lameter19655d32006-09-25 23:31:19 -07002608 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002609 enum zone_type j;
2610 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611
2612 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613
Mel Gorman54a6eb52008-04-28 02:12:16 -07002614 zonelist = &pgdat->node_zonelists[0];
2615 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616
Mel Gorman54a6eb52008-04-28 02:12:16 -07002617 /*
2618 * Now we build the zonelist so that it contains the zones
2619 * of all the other nodes.
2620 * We don't want to pressure a particular node, so when
2621 * building the zones for node N, we make sure that the
2622 * zones coming right after the local ones are those from
2623 * node N+1 (modulo N)
2624 */
2625 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2626 if (!node_online(node))
2627 continue;
2628 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2629 MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07002631 for (node = 0; node < local_node; node++) {
2632 if (!node_online(node))
2633 continue;
2634 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2635 MAX_NR_ZONES - 1);
2636 }
2637
Mel Gormandd1a2392008-04-28 02:12:17 -07002638 zonelist->_zonerefs[j].zone = NULL;
2639 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640}
2641
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002642/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002643static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002644{
Mel Gorman54a6eb52008-04-28 02:12:16 -07002645 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002646}
2647
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648#endif /* CONFIG_NUMA */
2649
Rusty Russell9b1a4d32008-07-28 12:16:30 -05002650/* return values int ....just for stop_machine() */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002651static int __build_all_zonelists(void *dummy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652{
Yasunori Goto68113782006-06-23 02:03:11 -07002653 int nid;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002654
Bo Liu7f9cfb32009-08-18 14:11:19 -07002655#ifdef CONFIG_NUMA
2656 memset(node_load, 0, sizeof(node_load));
2657#endif
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002658 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07002659 pg_data_t *pgdat = NODE_DATA(nid);
2660
2661 build_zonelists(pgdat);
2662 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002663 }
Yasunori Goto68113782006-06-23 02:03:11 -07002664 return 0;
2665}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002667void build_all_zonelists(void)
Yasunori Goto68113782006-06-23 02:03:11 -07002668{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002669 set_zonelist_order();
2670
Yasunori Goto68113782006-06-23 02:03:11 -07002671 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07002672 __build_all_zonelists(NULL);
Mel Gorman68ad8df2008-07-23 21:26:52 -07002673 mminit_verify_zonelist();
Yasunori Goto68113782006-06-23 02:03:11 -07002674 cpuset_init_current_mems_allowed();
2675 } else {
Simon Arlott183ff222007-10-20 01:27:18 +02002676 /* we have to stop all cpus to guarantee there is no user
Yasunori Goto68113782006-06-23 02:03:11 -07002677 of zonelist */
Rusty Russell9b1a4d32008-07-28 12:16:30 -05002678 stop_machine(__build_all_zonelists, NULL, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07002679 /* cpuset refresh routine should be here */
2680 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07002681 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002682 /*
2683 * Disable grouping by mobility if the number of pages in the
2684 * system is too low to allow the mechanism to work. It would be
2685 * more accurate, but expensive to check per-zone. This check is
2686 * made on memory-hotadd so a system can start with mobility
2687 * disabled and enable it later
2688 */
Mel Gormand9c23402007-10-16 01:26:01 -07002689 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002690 page_group_by_mobility_disabled = 1;
2691 else
2692 page_group_by_mobility_disabled = 0;
2693
2694 printk("Built %i zonelists in %s order, mobility grouping %s. "
2695 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002696 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002697 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002698 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002699 vm_total_pages);
2700#ifdef CONFIG_NUMA
2701 printk("Policy zone: %s\n", zone_names[policy_zone]);
2702#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703}
2704
2705/*
2706 * Helper functions to size the waitqueue hash table.
2707 * Essentially these want to choose hash table sizes sufficiently
2708 * large so that collisions trying to wait on pages are rare.
2709 * But in fact, the number of active page waitqueues on typical
2710 * systems is ridiculously low, less than 200. So this is even
2711 * conservative, even though it seems large.
2712 *
2713 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2714 * waitqueues, i.e. the size of the waitq table given the number of pages.
2715 */
2716#define PAGES_PER_WAITQUEUE 256
2717
Yasunori Gotocca448f2006-06-23 02:03:10 -07002718#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07002719static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720{
2721 unsigned long size = 1;
2722
2723 pages /= PAGES_PER_WAITQUEUE;
2724
2725 while (size < pages)
2726 size <<= 1;
2727
2728 /*
2729 * Once we have dozens or even hundreds of threads sleeping
2730 * on IO we've got bigger problems than wait queue collision.
2731 * Limit the size of the wait table to a reasonable size.
2732 */
2733 size = min(size, 4096UL);
2734
2735 return max(size, 4UL);
2736}
Yasunori Gotocca448f2006-06-23 02:03:10 -07002737#else
2738/*
2739 * A zone's size might be changed by hot-add, so it is not possible to determine
2740 * a suitable size for its wait_table. So we use the maximum size now.
2741 *
2742 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2743 *
2744 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2745 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2746 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2747 *
2748 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2749 * or more by the traditional way. (See above). It equals:
2750 *
2751 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2752 * ia64(16K page size) : = ( 8G + 4M)byte.
2753 * powerpc (64K page size) : = (32G +16M)byte.
2754 */
2755static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2756{
2757 return 4096UL;
2758}
2759#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760
2761/*
2762 * This is an integer logarithm so that shifts can be used later
2763 * to extract the more random high bits from the multiplicative
2764 * hash function before the remainder is taken.
2765 */
2766static inline unsigned long wait_table_bits(unsigned long size)
2767{
2768 return ffz(~size);
2769}
2770
2771#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2772
Mel Gorman56fd56b2007-10-16 01:25:58 -07002773/*
Mel Gormand9c23402007-10-16 01:26:01 -07002774 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07002775 * of blocks reserved is based on min_wmark_pages(zone). The memory within
2776 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07002777 * higher will lead to a bigger reserve which will get freed as contiguous
2778 * blocks as reclaim kicks in
2779 */
2780static void setup_zone_migrate_reserve(struct zone *zone)
2781{
2782 unsigned long start_pfn, pfn, end_pfn;
2783 struct page *page;
2784 unsigned long reserve, block_migratetype;
2785
2786 /* Get the start pfn, end pfn and the number of blocks to reserve */
2787 start_pfn = zone->zone_start_pfn;
2788 end_pfn = start_pfn + zone->spanned_pages;
Mel Gorman41858962009-06-16 15:32:12 -07002789 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07002790 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07002791
Mel Gormand9c23402007-10-16 01:26:01 -07002792 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07002793 if (!pfn_valid(pfn))
2794 continue;
2795 page = pfn_to_page(pfn);
2796
Adam Litke344c7902008-09-02 14:35:38 -07002797 /* Watch out for overlapping nodes */
2798 if (page_to_nid(page) != zone_to_nid(zone))
2799 continue;
2800
Mel Gorman56fd56b2007-10-16 01:25:58 -07002801 /* Blocks with reserved pages will never free, skip them. */
2802 if (PageReserved(page))
2803 continue;
2804
2805 block_migratetype = get_pageblock_migratetype(page);
2806
2807 /* If this block is reserved, account for it */
2808 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
2809 reserve--;
2810 continue;
2811 }
2812
2813 /* Suitable for reserving if this block is movable */
2814 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
2815 set_pageblock_migratetype(page, MIGRATE_RESERVE);
2816 move_freepages_block(zone, page, MIGRATE_RESERVE);
2817 reserve--;
2818 continue;
2819 }
2820
2821 /*
2822 * If the reserve is met and this is a previous reserved block,
2823 * take it back
2824 */
2825 if (block_migratetype == MIGRATE_RESERVE) {
2826 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2827 move_freepages_block(zone, page, MIGRATE_MOVABLE);
2828 }
2829 }
2830}
Mel Gormanac0e5b72007-10-16 01:25:58 -07002831
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832/*
2833 * Initially all pages are reserved - free ones are freed
2834 * up by free_all_bootmem() once the early boot process is
2835 * done. Non-atomic initialization, single-pass.
2836 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002837void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa02007-01-10 23:15:30 -08002838 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002840 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07002841 unsigned long end_pfn = start_pfn + size;
2842 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002843 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844
Hugh Dickins22b31ee2009-01-06 14:40:09 -08002845 if (highest_memmap_pfn < end_pfn - 1)
2846 highest_memmap_pfn = end_pfn - 1;
2847
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002848 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08002849 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa02007-01-10 23:15:30 -08002850 /*
2851 * There can be holes in boot-time mem_map[]s
2852 * handed to this function. They do not
2853 * exist on hotplugged memory.
2854 */
2855 if (context == MEMMAP_EARLY) {
2856 if (!early_pfn_valid(pfn))
2857 continue;
2858 if (!early_pfn_in_nid(pfn, nid))
2859 continue;
2860 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002861 page = pfn_to_page(pfn);
2862 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07002863 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08002864 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002865 reset_page_mapcount(page);
2866 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002867 /*
2868 * Mark the block movable so that blocks are reserved for
2869 * movable at startup. This will force kernel allocations
2870 * to reserve their blocks rather than leaking throughout
2871 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07002872 * kernel allocations are made. Later some blocks near
2873 * the start are marked MIGRATE_RESERVE by
2874 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002875 *
2876 * bitmap is created for zone's valid pfn range. but memmap
2877 * can be created for invalid pages (for alignment)
2878 * check here not to call set_pageblock_migratetype() against
2879 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002880 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002881 if ((z->zone_start_pfn <= pfn)
2882 && (pfn < z->zone_start_pfn + z->spanned_pages)
2883 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07002884 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002885
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886 INIT_LIST_HEAD(&page->lru);
2887#ifdef WANT_PAGE_VIRTUAL
2888 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2889 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07002890 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002892 }
2893}
2894
Andi Kleen1e548de2008-02-04 22:29:26 -08002895static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896{
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002897 int order, t;
2898 for_each_migratetype_order(order, t) {
2899 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002900 zone->free_area[order].nr_free = 0;
2901 }
2902}
2903
2904#ifndef __HAVE_ARCH_MEMMAP_INIT
2905#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa02007-01-10 23:15:30 -08002906 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907#endif
2908
Sam Ravnborg1d6f4e62008-01-17 15:21:12 -08002909static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002910{
David Howells3a6be872009-05-06 16:03:03 -07002911#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002912 int batch;
2913
2914 /*
2915 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07002916 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002917 *
2918 * OK, so we don't know how big the cache is. So guess.
2919 */
2920 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07002921 if (batch * PAGE_SIZE > 512 * 1024)
2922 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002923 batch /= 4; /* We effectively *= 4 below */
2924 if (batch < 1)
2925 batch = 1;
2926
2927 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002928 * Clamp the batch to a 2^n - 1 value. Having a power
2929 * of 2 value was found to be more likely to have
2930 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002931 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002932 * For example if 2 tasks are alternately allocating
2933 * batches of pages, one task can end up with a lot
2934 * of pages of one half of the possible page colors
2935 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002936 */
David Howells91552032009-05-06 16:03:02 -07002937 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07002938
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002939 return batch;
David Howells3a6be872009-05-06 16:03:03 -07002940
2941#else
2942 /* The deferral and batching of frees should be suppressed under NOMMU
2943 * conditions.
2944 *
2945 * The problem is that NOMMU needs to be able to allocate large chunks
2946 * of contiguous memory as there's no hardware page translation to
2947 * assemble apparent contiguous memory from discontiguous pages.
2948 *
2949 * Queueing large contiguous runs of pages for batching, however,
2950 * causes the pages to actually be freed in smaller chunks. As there
2951 * can be a significant delay between the individual batches being
2952 * recycled, this leads to the once large chunks of space being
2953 * fragmented and becoming unavailable for high-order allocations.
2954 */
2955 return 0;
2956#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002957}
2958
Adrian Bunkb69a7282008-07-23 21:28:12 -07002959static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
Christoph Lameter2caaad42005-06-21 17:15:00 -07002960{
2961 struct per_cpu_pages *pcp;
2962
Magnus Damm1c6fe942005-10-26 01:58:59 -07002963 memset(p, 0, sizeof(*p));
2964
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002965 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07002966 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07002967 pcp->high = 6 * batch;
2968 pcp->batch = max(1UL, 1 * batch);
2969 INIT_LIST_HEAD(&pcp->list);
Christoph Lameter2caaad42005-06-21 17:15:00 -07002970}
2971
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002972/*
2973 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2974 * to the value high for the pageset p.
2975 */
2976
2977static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2978 unsigned long high)
2979{
2980 struct per_cpu_pages *pcp;
2981
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002982 pcp = &p->pcp;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002983 pcp->high = high;
2984 pcp->batch = max(1UL, high/4);
2985 if ((high/4) > (PAGE_SHIFT * 8))
2986 pcp->batch = PAGE_SHIFT * 8;
2987}
2988
2989
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002990#ifdef CONFIG_NUMA
2991/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07002992 * Boot pageset table. One per cpu which is going to be used for all
2993 * zones and all nodes. The parameters will be set in such a way
2994 * that an item put on a list will immediately be handed over to
2995 * the buddy list. This is safe since pageset manipulation is done
2996 * with interrupts disabled.
2997 *
2998 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07002999 *
3000 * The boot_pagesets must be kept even after bootup is complete for
3001 * unused processors and/or zones. They do play a role for bootstrapping
3002 * hotplugged processors.
3003 *
3004 * zoneinfo_show() and maybe other functions do
3005 * not check if the processor is online before following the pageset pointer.
3006 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07003007 */
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08003008static struct per_cpu_pageset boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07003009
3010/*
3011 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003012 * per cpu pageset array in struct zone.
3013 */
Ashok Raj6292d9a2006-02-01 03:04:44 -08003014static int __cpuinit process_zones(int cpu)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003015{
3016 struct zone *zone, *dzone;
Christoph Lameter37c07082007-10-16 01:25:36 -07003017 int node = cpu_to_node(cpu);
3018
3019 node_set_state(node, N_CPU); /* this node has a cpu */
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003020
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003021 for_each_populated_zone(zone) {
Nick Piggin23316bc2006-01-08 01:00:41 -08003022 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lameter37c07082007-10-16 01:25:36 -07003023 GFP_KERNEL, node);
Nick Piggin23316bc2006-01-08 01:00:41 -08003024 if (!zone_pcp(zone, cpu))
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003025 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003026
Nick Piggin23316bc2006-01-08 01:00:41 -08003027 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003028
3029 if (percpu_pagelist_fraction)
3030 setup_pagelist_highmark(zone_pcp(zone, cpu),
3031 (zone->present_pages / percpu_pagelist_fraction));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003032 }
3033
3034 return 0;
3035bad:
3036 for_each_zone(dzone) {
Andrew Morton64191682007-08-30 23:56:17 -07003037 if (!populated_zone(dzone))
3038 continue;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003039 if (dzone == zone)
3040 break;
Nick Piggin23316bc2006-01-08 01:00:41 -08003041 kfree(zone_pcp(dzone, cpu));
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07003042 zone_pcp(dzone, cpu) = &boot_pageset[cpu];
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003043 }
3044 return -ENOMEM;
3045}
3046
3047static inline void free_zone_pagesets(int cpu)
3048{
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003049 struct zone *zone;
3050
3051 for_each_zone(zone) {
3052 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
3053
David Rientjesf3ef9ea2006-09-25 16:24:57 -07003054 /* Free per_cpu_pageset if it is slab allocated */
3055 if (pset != &boot_pageset[cpu])
3056 kfree(pset);
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07003057 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003058 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003059}
3060
Chandra Seetharaman9c7b2162006-06-27 02:54:07 -07003061static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003062 unsigned long action,
3063 void *hcpu)
3064{
3065 int cpu = (long)hcpu;
3066 int ret = NOTIFY_OK;
3067
3068 switch (action) {
Andy Whitcroftce421c72006-12-06 20:33:08 -08003069 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003070 case CPU_UP_PREPARE_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08003071 if (process_zones(cpu))
3072 ret = NOTIFY_BAD;
3073 break;
3074 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003075 case CPU_UP_CANCELED_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08003076 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003077 case CPU_DEAD_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08003078 free_zone_pagesets(cpu);
3079 break;
3080 default:
3081 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003082 }
3083 return ret;
3084}
3085
Chandra Seetharaman74b85f32006-06-27 02:54:09 -07003086static struct notifier_block __cpuinitdata pageset_notifier =
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003087 { &pageset_cpuup_callback, NULL, 0 };
3088
Al Viro78d99552005-12-15 09:18:25 +00003089void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003090{
3091 int err;
3092
3093 /* Initialize per_cpu_pageset for cpu 0.
3094 * A cpuup callback will do this for every cpu
3095 * as it comes online
3096 */
3097 err = process_zones(smp_processor_id());
3098 BUG_ON(err);
3099 register_cpu_notifier(&pageset_notifier);
3100}
3101
3102#endif
3103
Sam Ravnborg577a32f2007-05-17 23:29:25 +02003104static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07003105int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07003106{
3107 int i;
3108 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003109 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07003110
3111 /*
3112 * The per-page waitqueue mechanism uses hashed waitqueues
3113 * per zone.
3114 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07003115 zone->wait_table_hash_nr_entries =
3116 wait_table_hash_nr_entries(zone_size_pages);
3117 zone->wait_table_bits =
3118 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003119 alloc_size = zone->wait_table_hash_nr_entries
3120 * sizeof(wait_queue_head_t);
3121
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07003122 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07003123 zone->wait_table = (wait_queue_head_t *)
3124 alloc_bootmem_node(pgdat, alloc_size);
3125 } else {
3126 /*
3127 * This case means that a zone whose size was 0 gets new memory
3128 * via memory hot-add.
3129 * But it may be the case that a new node was hot-added. In
3130 * this case vmalloc() will not be able to use this new node's
3131 * memory - this wait_table must be initialized to use this new
3132 * node itself as well.
3133 * To use this new node's memory, further consideration will be
3134 * necessary.
3135 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07003136 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003137 }
3138 if (!zone->wait_table)
3139 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07003140
Yasunori Goto02b694d2006-06-23 02:03:08 -07003141 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07003142 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003143
3144 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003145}
3146
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003147static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07003148{
3149 int cpu;
3150 unsigned long batch = zone_batchsize(zone);
3151
3152 for (cpu = 0; cpu < NR_CPUS; cpu++) {
3153#ifdef CONFIG_NUMA
3154 /* Early boot. Slab allocator not functional yet */
Nick Piggin23316bc2006-01-08 01:00:41 -08003155 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Dave Hansened8ece22005-10-29 18:16:50 -07003156 setup_pageset(&boot_pageset[cpu],0);
3157#else
3158 setup_pageset(zone_pcp(zone,cpu), batch);
3159#endif
3160 }
Anton Blanchardf5335c02006-03-25 03:06:49 -08003161 if (zone->present_pages)
3162 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
3163 zone->name, zone->present_pages, batch);
Dave Hansened8ece22005-10-29 18:16:50 -07003164}
3165
Yasunori Goto718127c2006-06-23 02:03:10 -07003166__meminit int init_currently_empty_zone(struct zone *zone,
3167 unsigned long zone_start_pfn,
Dave Hansena2f3aa02007-01-10 23:15:30 -08003168 unsigned long size,
3169 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07003170{
3171 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003172 int ret;
3173 ret = zone_wait_table_init(zone, size);
3174 if (ret)
3175 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07003176 pgdat->nr_zones = zone_idx(zone) + 1;
3177
Dave Hansened8ece22005-10-29 18:16:50 -07003178 zone->zone_start_pfn = zone_start_pfn;
3179
Mel Gorman708614e2008-07-23 21:26:51 -07003180 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3181 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3182 pgdat->node_id,
3183 (unsigned long)zone_idx(zone),
3184 zone_start_pfn, (zone_start_pfn + size));
3185
Andi Kleen1e548de2008-02-04 22:29:26 -08003186 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07003187
3188 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003189}
3190
Mel Gormanc7132162006-09-27 01:49:43 -07003191#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3192/*
3193 * Basic iterator support. Return the first range of PFNs for a node
3194 * Note: nid == MAX_NUMNODES returns first region regardless of node
3195 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003196static int __meminit first_active_region_index_in_nid(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003197{
3198 int i;
3199
3200 for (i = 0; i < nr_nodemap_entries; i++)
3201 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3202 return i;
3203
3204 return -1;
3205}
3206
3207/*
3208 * Basic iterator support. Return the next active range of PFNs for a node
Simon Arlott183ff222007-10-20 01:27:18 +02003209 * Note: nid == MAX_NUMNODES returns next region regardless of node
Mel Gormanc7132162006-09-27 01:49:43 -07003210 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003211static int __meminit next_active_region_index_in_nid(int index, int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003212{
3213 for (index = index + 1; index < nr_nodemap_entries; index++)
3214 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3215 return index;
3216
3217 return -1;
3218}
3219
3220#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3221/*
3222 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3223 * Architectures may implement their own version but if add_active_range()
3224 * was used and there are no special requirements, this is a convenient
3225 * alternative
3226 */
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003227int __meminit __early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07003228{
3229 int i;
3230
3231 for (i = 0; i < nr_nodemap_entries; i++) {
3232 unsigned long start_pfn = early_node_map[i].start_pfn;
3233 unsigned long end_pfn = early_node_map[i].end_pfn;
3234
3235 if (start_pfn <= pfn && pfn < end_pfn)
3236 return early_node_map[i].nid;
3237 }
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003238 /* This is a memory hole */
3239 return -1;
Mel Gormanc7132162006-09-27 01:49:43 -07003240}
3241#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3242
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003243int __meminit early_pfn_to_nid(unsigned long pfn)
3244{
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003245 int nid;
3246
3247 nid = __early_pfn_to_nid(pfn);
3248 if (nid >= 0)
3249 return nid;
3250 /* just returns 0 */
3251 return 0;
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003252}
3253
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003254#ifdef CONFIG_NODES_SPAN_OTHER_NODES
3255bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3256{
3257 int nid;
3258
3259 nid = __early_pfn_to_nid(pfn);
3260 if (nid >= 0 && nid != node)
3261 return false;
3262 return true;
3263}
3264#endif
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003265
Mel Gormanc7132162006-09-27 01:49:43 -07003266/* Basic iterator support to walk early_node_map[] */
3267#define for_each_active_range_index_in_nid(i, nid) \
3268 for (i = first_active_region_index_in_nid(nid); i != -1; \
3269 i = next_active_region_index_in_nid(i, nid))
3270
3271/**
3272 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003273 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3274 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07003275 *
3276 * If an architecture guarantees that all ranges registered with
3277 * add_active_ranges() contain no holes and may be freed, this
3278 * this function may be used instead of calling free_bootmem() manually.
3279 */
3280void __init free_bootmem_with_active_regions(int nid,
3281 unsigned long max_low_pfn)
3282{
3283 int i;
3284
3285 for_each_active_range_index_in_nid(i, nid) {
3286 unsigned long size_pages = 0;
3287 unsigned long end_pfn = early_node_map[i].end_pfn;
3288
3289 if (early_node_map[i].start_pfn >= max_low_pfn)
3290 continue;
3291
3292 if (end_pfn > max_low_pfn)
3293 end_pfn = max_low_pfn;
3294
3295 size_pages = end_pfn - early_node_map[i].start_pfn;
3296 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3297 PFN_PHYS(early_node_map[i].start_pfn),
3298 size_pages << PAGE_SHIFT);
3299 }
3300}
3301
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003302void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3303{
3304 int i;
Yinghai Lud52d53b2008-06-16 20:10:55 -07003305 int ret;
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003306
Yinghai Lud52d53b2008-06-16 20:10:55 -07003307 for_each_active_range_index_in_nid(i, nid) {
3308 ret = work_fn(early_node_map[i].start_pfn,
3309 early_node_map[i].end_pfn, data);
3310 if (ret)
3311 break;
3312 }
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003313}
Mel Gormanc7132162006-09-27 01:49:43 -07003314/**
3315 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003316 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07003317 *
3318 * If an architecture guarantees that all ranges registered with
3319 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003320 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07003321 */
3322void __init sparse_memory_present_with_active_regions(int nid)
3323{
3324 int i;
3325
3326 for_each_active_range_index_in_nid(i, nid)
3327 memory_present(early_node_map[i].nid,
3328 early_node_map[i].start_pfn,
3329 early_node_map[i].end_pfn);
3330}
3331
3332/**
3333 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003334 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3335 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3336 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07003337 *
3338 * It returns the start and end page frame of a node based on information
3339 * provided by an arch calling add_active_range(). If called for a node
3340 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003341 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07003342 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003343void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003344 unsigned long *start_pfn, unsigned long *end_pfn)
3345{
3346 int i;
3347 *start_pfn = -1UL;
3348 *end_pfn = 0;
3349
3350 for_each_active_range_index_in_nid(i, nid) {
3351 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3352 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3353 }
3354
Christoph Lameter633c0662007-10-16 01:25:37 -07003355 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07003356 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07003357}
3358
3359/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07003360 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3361 * assumption is made that zones within a node are ordered in monotonic
3362 * increasing memory addresses so that the "highest" populated zone is used
3363 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003364static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07003365{
3366 int zone_index;
3367 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3368 if (zone_index == ZONE_MOVABLE)
3369 continue;
3370
3371 if (arch_zone_highest_possible_pfn[zone_index] >
3372 arch_zone_lowest_possible_pfn[zone_index])
3373 break;
3374 }
3375
3376 VM_BUG_ON(zone_index == -1);
3377 movable_zone = zone_index;
3378}
3379
3380/*
3381 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3382 * because it is sized independant of architecture. Unlike the other zones,
3383 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3384 * in each node depending on the size of each node and how evenly kernelcore
3385 * is distributed. This helper function adjusts the zone ranges
3386 * provided by the architecture for a given node by using the end of the
3387 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3388 * zones within a node are in order of monotonic increases memory addresses
3389 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003390static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07003391 unsigned long zone_type,
3392 unsigned long node_start_pfn,
3393 unsigned long node_end_pfn,
3394 unsigned long *zone_start_pfn,
3395 unsigned long *zone_end_pfn)
3396{
3397 /* Only adjust if ZONE_MOVABLE is on this node */
3398 if (zone_movable_pfn[nid]) {
3399 /* Size ZONE_MOVABLE */
3400 if (zone_type == ZONE_MOVABLE) {
3401 *zone_start_pfn = zone_movable_pfn[nid];
3402 *zone_end_pfn = min(node_end_pfn,
3403 arch_zone_highest_possible_pfn[movable_zone]);
3404
3405 /* Adjust for ZONE_MOVABLE starting within this range */
3406 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3407 *zone_end_pfn > zone_movable_pfn[nid]) {
3408 *zone_end_pfn = zone_movable_pfn[nid];
3409
3410 /* Check if this whole range is within ZONE_MOVABLE */
3411 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3412 *zone_start_pfn = *zone_end_pfn;
3413 }
3414}
3415
3416/*
Mel Gormanc7132162006-09-27 01:49:43 -07003417 * Return the number of pages a zone spans in a node, including holes
3418 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3419 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003420static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003421 unsigned long zone_type,
3422 unsigned long *ignored)
3423{
3424 unsigned long node_start_pfn, node_end_pfn;
3425 unsigned long zone_start_pfn, zone_end_pfn;
3426
3427 /* Get the start and end of the node and zone */
3428 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3429 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3430 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07003431 adjust_zone_range_for_zone_movable(nid, zone_type,
3432 node_start_pfn, node_end_pfn,
3433 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003434
3435 /* Check that this node has pages within the zone's required range */
3436 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3437 return 0;
3438
3439 /* Move the zone boundaries inside the node if necessary */
3440 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3441 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3442
3443 /* Return the spanned pages */
3444 return zone_end_pfn - zone_start_pfn;
3445}
3446
3447/*
3448 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003449 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07003450 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003451static unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003452 unsigned long range_start_pfn,
3453 unsigned long range_end_pfn)
3454{
3455 int i = 0;
3456 unsigned long prev_end_pfn = 0, hole_pages = 0;
3457 unsigned long start_pfn;
3458
3459 /* Find the end_pfn of the first active range of pfns in the node */
3460 i = first_active_region_index_in_nid(nid);
3461 if (i == -1)
3462 return 0;
3463
Mel Gormanb5445f92007-07-26 10:41:18 -07003464 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3465
Mel Gorman9c7cd682006-09-27 01:49:58 -07003466 /* Account for ranges before physical memory on this node */
3467 if (early_node_map[i].start_pfn > range_start_pfn)
Mel Gormanb5445f92007-07-26 10:41:18 -07003468 hole_pages = prev_end_pfn - range_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003469
3470 /* Find all holes for the zone within the node */
3471 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3472
3473 /* No need to continue if prev_end_pfn is outside the zone */
3474 if (prev_end_pfn >= range_end_pfn)
3475 break;
3476
3477 /* Make sure the end of the zone is not within the hole */
3478 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3479 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3480
3481 /* Update the hole size cound and move on */
3482 if (start_pfn > range_start_pfn) {
3483 BUG_ON(prev_end_pfn > start_pfn);
3484 hole_pages += start_pfn - prev_end_pfn;
3485 }
3486 prev_end_pfn = early_node_map[i].end_pfn;
3487 }
3488
Mel Gorman9c7cd682006-09-27 01:49:58 -07003489 /* Account for ranges past physical memory on this node */
3490 if (range_end_pfn > prev_end_pfn)
Mel Gorman0c6cb972006-10-28 10:38:59 -07003491 hole_pages += range_end_pfn -
Mel Gorman9c7cd682006-09-27 01:49:58 -07003492 max(range_start_pfn, prev_end_pfn);
3493
Mel Gormanc7132162006-09-27 01:49:43 -07003494 return hole_pages;
3495}
3496
3497/**
3498 * absent_pages_in_range - Return number of page frames in holes within a range
3499 * @start_pfn: The start PFN to start searching for holes
3500 * @end_pfn: The end PFN to stop searching for holes
3501 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003502 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07003503 */
3504unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3505 unsigned long end_pfn)
3506{
3507 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3508}
3509
3510/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003511static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003512 unsigned long zone_type,
3513 unsigned long *ignored)
3514{
Mel Gorman9c7cd682006-09-27 01:49:58 -07003515 unsigned long node_start_pfn, node_end_pfn;
3516 unsigned long zone_start_pfn, zone_end_pfn;
3517
3518 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3519 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3520 node_start_pfn);
3521 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3522 node_end_pfn);
3523
Mel Gorman2a1e2742007-07-17 04:03:12 -07003524 adjust_zone_range_for_zone_movable(nid, zone_type,
3525 node_start_pfn, node_end_pfn,
3526 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07003527 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003528}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003529
Mel Gormanc7132162006-09-27 01:49:43 -07003530#else
Paul Mundt6ea6e682007-07-15 23:38:20 -07003531static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003532 unsigned long zone_type,
3533 unsigned long *zones_size)
3534{
3535 return zones_size[zone_type];
3536}
3537
Paul Mundt6ea6e682007-07-15 23:38:20 -07003538static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003539 unsigned long zone_type,
3540 unsigned long *zholes_size)
3541{
3542 if (!zholes_size)
3543 return 0;
3544
3545 return zholes_size[zone_type];
3546}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003547
Mel Gormanc7132162006-09-27 01:49:43 -07003548#endif
3549
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003550static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Mel Gormanc7132162006-09-27 01:49:43 -07003551 unsigned long *zones_size, unsigned long *zholes_size)
3552{
3553 unsigned long realtotalpages, totalpages = 0;
3554 enum zone_type i;
3555
3556 for (i = 0; i < MAX_NR_ZONES; i++)
3557 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3558 zones_size);
3559 pgdat->node_spanned_pages = totalpages;
3560
3561 realtotalpages = totalpages;
3562 for (i = 0; i < MAX_NR_ZONES; i++)
3563 realtotalpages -=
3564 zone_absent_pages_in_node(pgdat->node_id, i,
3565 zholes_size);
3566 pgdat->node_present_pages = realtotalpages;
3567 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3568 realtotalpages);
3569}
3570
Mel Gorman835c1342007-10-16 01:25:47 -07003571#ifndef CONFIG_SPARSEMEM
3572/*
3573 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07003574 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3575 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07003576 * round what is now in bits to nearest long in bits, then return it in
3577 * bytes.
3578 */
3579static unsigned long __init usemap_size(unsigned long zonesize)
3580{
3581 unsigned long usemapsize;
3582
Mel Gormand9c23402007-10-16 01:26:01 -07003583 usemapsize = roundup(zonesize, pageblock_nr_pages);
3584 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07003585 usemapsize *= NR_PAGEBLOCK_BITS;
3586 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3587
3588 return usemapsize / 8;
3589}
3590
3591static void __init setup_usemap(struct pglist_data *pgdat,
3592 struct zone *zone, unsigned long zonesize)
3593{
3594 unsigned long usemapsize = usemap_size(zonesize);
3595 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08003596 if (usemapsize)
Mel Gorman835c1342007-10-16 01:25:47 -07003597 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
Mel Gorman835c1342007-10-16 01:25:47 -07003598}
3599#else
3600static void inline setup_usemap(struct pglist_data *pgdat,
3601 struct zone *zone, unsigned long zonesize) {}
3602#endif /* CONFIG_SPARSEMEM */
3603
Mel Gormand9c23402007-10-16 01:26:01 -07003604#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08003605
3606/* Return a sensible default order for the pageblock size. */
3607static inline int pageblock_default_order(void)
3608{
3609 if (HPAGE_SHIFT > PAGE_SHIFT)
3610 return HUGETLB_PAGE_ORDER;
3611
3612 return MAX_ORDER-1;
3613}
3614
Mel Gormand9c23402007-10-16 01:26:01 -07003615/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3616static inline void __init set_pageblock_order(unsigned int order)
3617{
3618 /* Check that pageblock_nr_pages has not already been setup */
3619 if (pageblock_order)
3620 return;
3621
3622 /*
3623 * Assume the largest contiguous order of interest is a huge page.
3624 * This value may be variable depending on boot parameters on IA64
3625 */
3626 pageblock_order = order;
3627}
3628#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3629
Mel Gormanba72cb82007-11-28 16:21:13 -08003630/*
3631 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3632 * and pageblock_default_order() are unused as pageblock_order is set
3633 * at compile-time. See include/linux/pageblock-flags.h for the values of
3634 * pageblock_order based on the kernel config
3635 */
3636static inline int pageblock_default_order(unsigned int order)
3637{
3638 return MAX_ORDER-1;
3639}
Mel Gormand9c23402007-10-16 01:26:01 -07003640#define set_pageblock_order(x) do {} while (0)
3641
3642#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3643
Linus Torvalds1da177e2005-04-16 15:20:36 -07003644/*
3645 * Set up the zone data structures:
3646 * - mark all pages reserved
3647 * - mark all memory queues empty
3648 * - clear the memory bitmaps
3649 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08003650static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003651 unsigned long *zones_size, unsigned long *zholes_size)
3652{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07003653 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07003654 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003655 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07003656 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003657
Dave Hansen208d54e2005-10-29 18:16:52 -07003658 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003659 pgdat->nr_zones = 0;
3660 init_waitqueue_head(&pgdat->kswapd_wait);
3661 pgdat->kswapd_max_order = 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003662 pgdat_page_cgroup_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003663
3664 for (j = 0; j < MAX_NR_ZONES; j++) {
3665 struct zone *zone = pgdat->node_zones + j;
Mel Gorman0e0b8642006-09-27 01:49:56 -07003666 unsigned long size, realsize, memmap_pages;
Christoph Lameterb69408e2008-10-18 20:26:14 -07003667 enum lru_list l;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003668
Mel Gormanc7132162006-09-27 01:49:43 -07003669 size = zone_spanned_pages_in_node(nid, j, zones_size);
3670 realsize = size - zone_absent_pages_in_node(nid, j,
3671 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003672
Mel Gorman0e0b8642006-09-27 01:49:56 -07003673 /*
3674 * Adjust realsize so that it accounts for how much memory
3675 * is used by this zone for memmap. This affects the watermark
3676 * and per-cpu initialisations
3677 */
Johannes Weinerf7232152008-05-23 13:04:21 -07003678 memmap_pages =
3679 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
Mel Gorman0e0b8642006-09-27 01:49:56 -07003680 if (realsize >= memmap_pages) {
3681 realsize -= memmap_pages;
Yinghai Lu5594c8c2009-01-06 14:39:14 -08003682 if (memmap_pages)
3683 printk(KERN_DEBUG
3684 " %s zone: %lu pages used for memmap\n",
3685 zone_names[j], memmap_pages);
Mel Gorman0e0b8642006-09-27 01:49:56 -07003686 } else
3687 printk(KERN_WARNING
3688 " %s zone: %lu pages exceeds realsize %lu\n",
3689 zone_names[j], memmap_pages, realsize);
3690
Christoph Lameter62672762007-02-10 01:43:07 -08003691 /* Account for reserved pages */
3692 if (j == 0 && realsize > dma_reserve) {
Mel Gorman0e0b8642006-09-27 01:49:56 -07003693 realsize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07003694 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08003695 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07003696 }
3697
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003698 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003699 nr_kernel_pages += realsize;
3700 nr_all_pages += realsize;
3701
3702 zone->spanned_pages = size;
3703 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07003704#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07003705 zone->node = nid;
Christoph Lameter8417bba2006-09-25 23:31:51 -07003706 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07003707 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07003708 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07003709#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003710 zone->name = zone_names[j];
3711 spin_lock_init(&zone->lock);
3712 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07003713 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003714 zone->zone_pgdat = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003715
Martin Bligh3bb1a852006-10-28 10:38:24 -07003716 zone->prev_priority = DEF_PRIORITY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003717
Dave Hansened8ece22005-10-29 18:16:50 -07003718 zone_pcp_init(zone);
Christoph Lameterb69408e2008-10-18 20:26:14 -07003719 for_each_lru(l) {
3720 INIT_LIST_HEAD(&zone->lru[l].list);
Wu Fengguang6e08a362009-06-16 15:32:29 -07003721 zone->lru[l].nr_saved_scan = 0;
Christoph Lameterb69408e2008-10-18 20:26:14 -07003722 }
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08003723 zone->reclaim_stat.recent_rotated[0] = 0;
3724 zone->reclaim_stat.recent_rotated[1] = 0;
3725 zone->reclaim_stat.recent_scanned[0] = 0;
3726 zone->reclaim_stat.recent_scanned[1] = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07003727 zap_zone_vm_stats(zone);
David Rientjese815af92007-10-16 23:25:54 -07003728 zone->flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003729 if (!size)
3730 continue;
3731
Mel Gormanba72cb82007-11-28 16:21:13 -08003732 set_pageblock_order(pageblock_default_order());
Mel Gorman835c1342007-10-16 01:25:47 -07003733 setup_usemap(pgdat, zone, size);
Dave Hansena2f3aa02007-01-10 23:15:30 -08003734 ret = init_currently_empty_zone(zone, zone_start_pfn,
3735 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07003736 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07003737 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003738 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003739 }
3740}
3741
Sam Ravnborg577a32f2007-05-17 23:29:25 +02003742static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003743{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003744 /* Skip empty nodes */
3745 if (!pgdat->node_spanned_pages)
3746 return;
3747
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003748#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07003749 /* ia64 gets its own node_mem_map, before this, without bootmem */
3750 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07003751 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003752 struct page *map;
3753
Bob Piccoe984bb42006-05-20 15:00:31 -07003754 /*
3755 * The zone's endpoints aren't required to be MAX_ORDER
3756 * aligned but the node_mem_map endpoints must be in order
3757 * for the buddy allocator to function correctly.
3758 */
3759 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3760 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3761 end = ALIGN(end, MAX_ORDER_NR_PAGES);
3762 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07003763 map = alloc_remap(pgdat->node_id, size);
3764 if (!map)
3765 map = alloc_bootmem_node(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07003766 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003767 }
Roman Zippel12d810c2007-05-31 00:40:54 -07003768#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07003769 /*
3770 * With no DISCONTIG, the global mem_map is just set as node 0's
3771 */
Mel Gormanc7132162006-09-27 01:49:43 -07003772 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003773 mem_map = NODE_DATA(0)->node_mem_map;
Mel Gormanc7132162006-09-27 01:49:43 -07003774#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3775 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08003776 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Mel Gormanc7132162006-09-27 01:49:43 -07003777#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3778 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003779#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003780#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003781}
3782
Johannes Weiner9109fb72008-07-23 21:27:20 -07003783void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
3784 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003785{
Johannes Weiner9109fb72008-07-23 21:27:20 -07003786 pg_data_t *pgdat = NODE_DATA(nid);
3787
Linus Torvalds1da177e2005-04-16 15:20:36 -07003788 pgdat->node_id = nid;
3789 pgdat->node_start_pfn = node_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003790 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003791
3792 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07003793#ifdef CONFIG_FLAT_NODE_MEM_MAP
3794 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3795 nid, (unsigned long)pgdat,
3796 (unsigned long)pgdat->node_mem_map);
3797#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003798
3799 free_area_init_core(pgdat, zones_size, zholes_size);
3800}
3801
Mel Gormanc7132162006-09-27 01:49:43 -07003802#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07003803
3804#if MAX_NUMNODES > 1
3805/*
3806 * Figure out the number of possible node ids.
3807 */
3808static void __init setup_nr_node_ids(void)
3809{
3810 unsigned int node;
3811 unsigned int highest = 0;
3812
3813 for_each_node_mask(node, node_possible_map)
3814 highest = node;
3815 nr_node_ids = highest + 1;
3816}
3817#else
3818static inline void setup_nr_node_ids(void)
3819{
3820}
3821#endif
3822
Mel Gormanc7132162006-09-27 01:49:43 -07003823/**
3824 * add_active_range - Register a range of PFNs backed by physical memory
3825 * @nid: The node ID the range resides on
3826 * @start_pfn: The start PFN of the available physical memory
3827 * @end_pfn: The end PFN of the available physical memory
3828 *
3829 * These ranges are stored in an early_node_map[] and later used by
3830 * free_area_init_nodes() to calculate zone sizes and holes. If the
3831 * range spans a memory hole, it is up to the architecture to ensure
3832 * the memory is not freed by the bootmem allocator. If possible
3833 * the range being registered will be merged with existing ranges.
3834 */
3835void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3836 unsigned long end_pfn)
3837{
3838 int i;
3839
Mel Gorman6b74ab92008-07-23 21:26:49 -07003840 mminit_dprintk(MMINIT_TRACE, "memory_register",
3841 "Entering add_active_range(%d, %#lx, %#lx) "
3842 "%d entries of %d used\n",
3843 nid, start_pfn, end_pfn,
3844 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
Mel Gormanc7132162006-09-27 01:49:43 -07003845
Mel Gorman2dbb51c2008-07-23 21:26:52 -07003846 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
3847
Mel Gormanc7132162006-09-27 01:49:43 -07003848 /* Merge with existing active regions if possible */
3849 for (i = 0; i < nr_nodemap_entries; i++) {
3850 if (early_node_map[i].nid != nid)
3851 continue;
3852
3853 /* Skip if an existing region covers this new one */
3854 if (start_pfn >= early_node_map[i].start_pfn &&
3855 end_pfn <= early_node_map[i].end_pfn)
3856 return;
3857
3858 /* Merge forward if suitable */
3859 if (start_pfn <= early_node_map[i].end_pfn &&
3860 end_pfn > early_node_map[i].end_pfn) {
3861 early_node_map[i].end_pfn = end_pfn;
3862 return;
3863 }
3864
3865 /* Merge backward if suitable */
3866 if (start_pfn < early_node_map[i].end_pfn &&
3867 end_pfn >= early_node_map[i].start_pfn) {
3868 early_node_map[i].start_pfn = start_pfn;
3869 return;
3870 }
3871 }
3872
3873 /* Check that early_node_map is large enough */
3874 if (i >= MAX_ACTIVE_REGIONS) {
3875 printk(KERN_CRIT "More than %d memory regions, truncating\n",
3876 MAX_ACTIVE_REGIONS);
3877 return;
3878 }
3879
3880 early_node_map[i].nid = nid;
3881 early_node_map[i].start_pfn = start_pfn;
3882 early_node_map[i].end_pfn = end_pfn;
3883 nr_nodemap_entries = i + 1;
3884}
3885
3886/**
Yinghai Lucc1050b2008-06-13 19:08:52 -07003887 * remove_active_range - Shrink an existing registered range of PFNs
Mel Gormanc7132162006-09-27 01:49:43 -07003888 * @nid: The node id the range is on that should be shrunk
Yinghai Lucc1050b2008-06-13 19:08:52 -07003889 * @start_pfn: The new PFN of the range
3890 * @end_pfn: The new PFN of the range
Mel Gormanc7132162006-09-27 01:49:43 -07003891 *
3892 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003893 * The map is kept near the end physical page range that has already been
3894 * registered. This function allows an arch to shrink an existing registered
3895 * range.
Mel Gormanc7132162006-09-27 01:49:43 -07003896 */
Yinghai Lucc1050b2008-06-13 19:08:52 -07003897void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
3898 unsigned long end_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07003899{
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003900 int i, j;
3901 int removed = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07003902
Yinghai Lucc1050b2008-06-13 19:08:52 -07003903 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
3904 nid, start_pfn, end_pfn);
3905
Mel Gormanc7132162006-09-27 01:49:43 -07003906 /* Find the old active region end and shrink */
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003907 for_each_active_range_index_in_nid(i, nid) {
Yinghai Lucc1050b2008-06-13 19:08:52 -07003908 if (early_node_map[i].start_pfn >= start_pfn &&
3909 early_node_map[i].end_pfn <= end_pfn) {
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003910 /* clear it */
Yinghai Lucc1050b2008-06-13 19:08:52 -07003911 early_node_map[i].start_pfn = 0;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003912 early_node_map[i].end_pfn = 0;
3913 removed = 1;
3914 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07003915 }
Yinghai Lucc1050b2008-06-13 19:08:52 -07003916 if (early_node_map[i].start_pfn < start_pfn &&
3917 early_node_map[i].end_pfn > start_pfn) {
3918 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
3919 early_node_map[i].end_pfn = start_pfn;
3920 if (temp_end_pfn > end_pfn)
3921 add_active_range(nid, end_pfn, temp_end_pfn);
3922 continue;
3923 }
3924 if (early_node_map[i].start_pfn >= start_pfn &&
3925 early_node_map[i].end_pfn > end_pfn &&
3926 early_node_map[i].start_pfn < end_pfn) {
3927 early_node_map[i].start_pfn = end_pfn;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003928 continue;
3929 }
3930 }
3931
3932 if (!removed)
3933 return;
3934
3935 /* remove the blank ones */
3936 for (i = nr_nodemap_entries - 1; i > 0; i--) {
3937 if (early_node_map[i].nid != nid)
3938 continue;
3939 if (early_node_map[i].end_pfn)
3940 continue;
3941 /* we found it, get rid of it */
3942 for (j = i; j < nr_nodemap_entries - 1; j++)
3943 memcpy(&early_node_map[j], &early_node_map[j+1],
3944 sizeof(early_node_map[j]));
3945 j = nr_nodemap_entries - 1;
3946 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
3947 nr_nodemap_entries--;
3948 }
Mel Gormanc7132162006-09-27 01:49:43 -07003949}
3950
3951/**
3952 * remove_all_active_ranges - Remove all currently registered regions
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003953 *
Mel Gormanc7132162006-09-27 01:49:43 -07003954 * During discovery, it may be found that a table like SRAT is invalid
3955 * and an alternative discovery method must be used. This function removes
3956 * all currently registered regions.
3957 */
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003958void __init remove_all_active_ranges(void)
Mel Gormanc7132162006-09-27 01:49:43 -07003959{
3960 memset(early_node_map, 0, sizeof(early_node_map));
3961 nr_nodemap_entries = 0;
3962}
3963
3964/* Compare two active node_active_regions */
3965static int __init cmp_node_active_region(const void *a, const void *b)
3966{
3967 struct node_active_region *arange = (struct node_active_region *)a;
3968 struct node_active_region *brange = (struct node_active_region *)b;
3969
3970 /* Done this way to avoid overflows */
3971 if (arange->start_pfn > brange->start_pfn)
3972 return 1;
3973 if (arange->start_pfn < brange->start_pfn)
3974 return -1;
3975
3976 return 0;
3977}
3978
3979/* sort the node_map by start_pfn */
3980static void __init sort_node_map(void)
3981{
3982 sort(early_node_map, (size_t)nr_nodemap_entries,
3983 sizeof(struct node_active_region),
3984 cmp_node_active_region, NULL);
3985}
3986
Mel Gormana6af2bc2007-02-10 01:42:57 -08003987/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003988static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003989{
3990 int i;
Mel Gormana6af2bc2007-02-10 01:42:57 -08003991 unsigned long min_pfn = ULONG_MAX;
Mel Gorman1abbfb42006-11-23 12:01:41 +00003992
Mel Gormanc7132162006-09-27 01:49:43 -07003993 /* Assuming a sorted map, the first range found has the starting pfn */
3994 for_each_active_range_index_in_nid(i, nid)
Mel Gormana6af2bc2007-02-10 01:42:57 -08003995 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003996
Mel Gormana6af2bc2007-02-10 01:42:57 -08003997 if (min_pfn == ULONG_MAX) {
3998 printk(KERN_WARNING
Paul Jackson2bc0d262008-06-22 07:22:17 -07003999 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08004000 return 0;
4001 }
4002
4003 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004004}
4005
4006/**
4007 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4008 *
4009 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004010 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07004011 */
4012unsigned long __init find_min_pfn_with_active_regions(void)
4013{
4014 return find_min_pfn_for_node(MAX_NUMNODES);
4015}
4016
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004017/*
4018 * early_calculate_totalpages()
4019 * Sum pages in active regions for movable zone.
4020 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4021 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07004022static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07004023{
4024 int i;
4025 unsigned long totalpages = 0;
4026
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004027 for (i = 0; i < nr_nodemap_entries; i++) {
4028 unsigned long pages = early_node_map[i].end_pfn -
Mel Gorman7e63efe2007-07-17 04:03:15 -07004029 early_node_map[i].start_pfn;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004030 totalpages += pages;
4031 if (pages)
4032 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
4033 }
4034 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07004035}
4036
Mel Gorman2a1e2742007-07-17 04:03:12 -07004037/*
4038 * Find the PFN the Movable zone begins in each node. Kernel memory
4039 * is spread evenly between nodes as long as the nodes have enough
4040 * memory. When they don't, some nodes will have more kernelcore than
4041 * others
4042 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004043static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004044{
4045 int i, nid;
4046 unsigned long usable_startpfn;
4047 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07004048 /* save the state before borrow the nodemask */
4049 nodemask_t saved_node_state = node_states[N_HIGH_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004050 unsigned long totalpages = early_calculate_totalpages();
4051 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004052
Mel Gorman7e63efe2007-07-17 04:03:15 -07004053 /*
4054 * If movablecore was specified, calculate what size of
4055 * kernelcore that corresponds so that memory usable for
4056 * any allocation type is evenly spread. If both kernelcore
4057 * and movablecore are specified, then the value of kernelcore
4058 * will be used for required_kernelcore if it's greater than
4059 * what movablecore would have allowed.
4060 */
4061 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07004062 unsigned long corepages;
4063
4064 /*
4065 * Round-up so that ZONE_MOVABLE is at least as large as what
4066 * was requested by the user
4067 */
4068 required_movablecore =
4069 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
4070 corepages = totalpages - required_movablecore;
4071
4072 required_kernelcore = max(required_kernelcore, corepages);
4073 }
4074
Mel Gorman2a1e2742007-07-17 04:03:12 -07004075 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4076 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07004077 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004078
4079 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4080 find_usable_zone_for_movable();
4081 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4082
4083restart:
4084 /* Spread kernelcore memory as evenly as possible throughout nodes */
4085 kernelcore_node = required_kernelcore / usable_nodes;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004086 for_each_node_state(nid, N_HIGH_MEMORY) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004087 /*
4088 * Recalculate kernelcore_node if the division per node
4089 * now exceeds what is necessary to satisfy the requested
4090 * amount of memory for the kernel
4091 */
4092 if (required_kernelcore < kernelcore_node)
4093 kernelcore_node = required_kernelcore / usable_nodes;
4094
4095 /*
4096 * As the map is walked, we track how much memory is usable
4097 * by the kernel using kernelcore_remaining. When it is
4098 * 0, the rest of the node is usable by ZONE_MOVABLE
4099 */
4100 kernelcore_remaining = kernelcore_node;
4101
4102 /* Go through each range of PFNs within this node */
4103 for_each_active_range_index_in_nid(i, nid) {
4104 unsigned long start_pfn, end_pfn;
4105 unsigned long size_pages;
4106
4107 start_pfn = max(early_node_map[i].start_pfn,
4108 zone_movable_pfn[nid]);
4109 end_pfn = early_node_map[i].end_pfn;
4110 if (start_pfn >= end_pfn)
4111 continue;
4112
4113 /* Account for what is only usable for kernelcore */
4114 if (start_pfn < usable_startpfn) {
4115 unsigned long kernel_pages;
4116 kernel_pages = min(end_pfn, usable_startpfn)
4117 - start_pfn;
4118
4119 kernelcore_remaining -= min(kernel_pages,
4120 kernelcore_remaining);
4121 required_kernelcore -= min(kernel_pages,
4122 required_kernelcore);
4123
4124 /* Continue if range is now fully accounted */
4125 if (end_pfn <= usable_startpfn) {
4126
4127 /*
4128 * Push zone_movable_pfn to the end so
4129 * that if we have to rebalance
4130 * kernelcore across nodes, we will
4131 * not double account here
4132 */
4133 zone_movable_pfn[nid] = end_pfn;
4134 continue;
4135 }
4136 start_pfn = usable_startpfn;
4137 }
4138
4139 /*
4140 * The usable PFN range for ZONE_MOVABLE is from
4141 * start_pfn->end_pfn. Calculate size_pages as the
4142 * number of pages used as kernelcore
4143 */
4144 size_pages = end_pfn - start_pfn;
4145 if (size_pages > kernelcore_remaining)
4146 size_pages = kernelcore_remaining;
4147 zone_movable_pfn[nid] = start_pfn + size_pages;
4148
4149 /*
4150 * Some kernelcore has been met, update counts and
4151 * break if the kernelcore for this node has been
4152 * satisified
4153 */
4154 required_kernelcore -= min(required_kernelcore,
4155 size_pages);
4156 kernelcore_remaining -= size_pages;
4157 if (!kernelcore_remaining)
4158 break;
4159 }
4160 }
4161
4162 /*
4163 * If there is still required_kernelcore, we do another pass with one
4164 * less node in the count. This will push zone_movable_pfn[nid] further
4165 * along on the nodes that still have memory until kernelcore is
4166 * satisified
4167 */
4168 usable_nodes--;
4169 if (usable_nodes && required_kernelcore > usable_nodes)
4170 goto restart;
4171
4172 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4173 for (nid = 0; nid < MAX_NUMNODES; nid++)
4174 zone_movable_pfn[nid] =
4175 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07004176
4177out:
4178 /* restore the node_state */
4179 node_states[N_HIGH_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004180}
4181
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004182/* Any regular memory on that node ? */
4183static void check_for_regular_memory(pg_data_t *pgdat)
4184{
4185#ifdef CONFIG_HIGHMEM
4186 enum zone_type zone_type;
4187
4188 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4189 struct zone *zone = &pgdat->node_zones[zone_type];
4190 if (zone->present_pages)
4191 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4192 }
4193#endif
4194}
4195
Mel Gormanc7132162006-09-27 01:49:43 -07004196/**
4197 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004198 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07004199 *
4200 * This will call free_area_init_node() for each active node in the system.
4201 * Using the page ranges provided by add_active_range(), the size of each
4202 * zone in each node and their holes is calculated. If the maximum PFN
4203 * between two adjacent zones match, it is assumed that the zone is empty.
4204 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4205 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4206 * starts where the previous one ended. For example, ZONE_DMA32 starts
4207 * at arch_max_dma_pfn.
4208 */
4209void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4210{
4211 unsigned long nid;
Andrew Mortondb991002008-10-15 22:01:07 -07004212 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004213
Mel Gormana6af2bc2007-02-10 01:42:57 -08004214 /* Sort early_node_map as initialisation assumes it is sorted */
4215 sort_node_map();
4216
Mel Gormanc7132162006-09-27 01:49:43 -07004217 /* Record where the zone boundaries are */
4218 memset(arch_zone_lowest_possible_pfn, 0,
4219 sizeof(arch_zone_lowest_possible_pfn));
4220 memset(arch_zone_highest_possible_pfn, 0,
4221 sizeof(arch_zone_highest_possible_pfn));
4222 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4223 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4224 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004225 if (i == ZONE_MOVABLE)
4226 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004227 arch_zone_lowest_possible_pfn[i] =
4228 arch_zone_highest_possible_pfn[i-1];
4229 arch_zone_highest_possible_pfn[i] =
4230 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4231 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07004232 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4233 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4234
4235 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4236 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4237 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004238
Mel Gormanc7132162006-09-27 01:49:43 -07004239 /* Print out the zone ranges */
4240 printk("Zone PFN ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07004241 for (i = 0; i < MAX_NR_ZONES; i++) {
4242 if (i == ZONE_MOVABLE)
4243 continue;
Paul Jackson5dab8ec2008-06-25 05:44:40 -07004244 printk(" %-8s %0#10lx -> %0#10lx\n",
Mel Gormanc7132162006-09-27 01:49:43 -07004245 zone_names[i],
4246 arch_zone_lowest_possible_pfn[i],
4247 arch_zone_highest_possible_pfn[i]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004248 }
4249
4250 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4251 printk("Movable zone start PFN for each node\n");
4252 for (i = 0; i < MAX_NUMNODES; i++) {
4253 if (zone_movable_pfn[i])
4254 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4255 }
Mel Gormanc7132162006-09-27 01:49:43 -07004256
4257 /* Print out the early_node_map[] */
4258 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4259 for (i = 0; i < nr_nodemap_entries; i++)
Paul Jackson5dab8ec2008-06-25 05:44:40 -07004260 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004261 early_node_map[i].start_pfn,
4262 early_node_map[i].end_pfn);
4263
4264 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07004265 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08004266 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07004267 for_each_online_node(nid) {
4268 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07004269 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07004270 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004271
4272 /* Any memory on that node */
4273 if (pgdat->node_present_pages)
4274 node_set_state(nid, N_HIGH_MEMORY);
4275 check_for_regular_memory(pgdat);
Mel Gormanc7132162006-09-27 01:49:43 -07004276 }
4277}
Mel Gorman2a1e2742007-07-17 04:03:12 -07004278
Mel Gorman7e63efe2007-07-17 04:03:15 -07004279static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004280{
4281 unsigned long long coremem;
4282 if (!p)
4283 return -EINVAL;
4284
4285 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004286 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004287
Mel Gorman7e63efe2007-07-17 04:03:15 -07004288 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07004289 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4290
4291 return 0;
4292}
Mel Gormaned7ed362007-07-17 04:03:14 -07004293
Mel Gorman7e63efe2007-07-17 04:03:15 -07004294/*
4295 * kernelcore=size sets the amount of memory for use for allocations that
4296 * cannot be reclaimed or migrated.
4297 */
4298static int __init cmdline_parse_kernelcore(char *p)
4299{
4300 return cmdline_parse_core(p, &required_kernelcore);
4301}
4302
4303/*
4304 * movablecore=size sets the amount of memory for use for allocations that
4305 * can be reclaimed or migrated.
4306 */
4307static int __init cmdline_parse_movablecore(char *p)
4308{
4309 return cmdline_parse_core(p, &required_movablecore);
4310}
4311
Mel Gormaned7ed362007-07-17 04:03:14 -07004312early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004313early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07004314
Mel Gormanc7132162006-09-27 01:49:43 -07004315#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4316
Mel Gorman0e0b8642006-09-27 01:49:56 -07004317/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004318 * set_dma_reserve - set the specified number of pages reserved in the first zone
4319 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07004320 *
4321 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4322 * In the DMA zone, a significant percentage may be consumed by kernel image
4323 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004324 * function may optionally be used to account for unfreeable pages in the
4325 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4326 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07004327 */
4328void __init set_dma_reserve(unsigned long new_dma_reserve)
4329{
4330 dma_reserve = new_dma_reserve;
4331}
4332
Dave Hansen93b75042005-06-23 00:07:47 -07004333#ifndef CONFIG_NEED_MULTIPLE_NODES
Marcin Slusarz52765582008-09-02 14:35:41 -07004334struct pglist_data __refdata contig_page_data = { .bdata = &bootmem_node_data[0] };
Linus Torvalds1da177e2005-04-16 15:20:36 -07004335EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07004336#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004337
4338void __init free_area_init(unsigned long *zones_size)
4339{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004340 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004341 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4342}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004343
Linus Torvalds1da177e2005-04-16 15:20:36 -07004344static int page_alloc_cpu_notify(struct notifier_block *self,
4345 unsigned long action, void *hcpu)
4346{
4347 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004348
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07004349 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004350 drain_pages(cpu);
4351
4352 /*
4353 * Spill the event counters of the dead processor
4354 * into the current processors event counters.
4355 * This artificially elevates the count of the current
4356 * processor.
4357 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07004358 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004359
4360 /*
4361 * Zero the differential counters of the dead processor
4362 * so that the vm statistics are consistent.
4363 *
4364 * This is only okay since the processor is dead and cannot
4365 * race with what we are doing.
4366 */
Christoph Lameter2244b952006-06-30 01:55:33 -07004367 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004368 }
4369 return NOTIFY_OK;
4370}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004371
4372void __init page_alloc_init(void)
4373{
4374 hotcpu_notifier(page_alloc_cpu_notify, 0);
4375}
4376
4377/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004378 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4379 * or min_free_kbytes changes.
4380 */
4381static void calculate_totalreserve_pages(void)
4382{
4383 struct pglist_data *pgdat;
4384 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004385 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004386
4387 for_each_online_pgdat(pgdat) {
4388 for (i = 0; i < MAX_NR_ZONES; i++) {
4389 struct zone *zone = pgdat->node_zones + i;
4390 unsigned long max = 0;
4391
4392 /* Find valid and maximum lowmem_reserve in the zone */
4393 for (j = i; j < MAX_NR_ZONES; j++) {
4394 if (zone->lowmem_reserve[j] > max)
4395 max = zone->lowmem_reserve[j];
4396 }
4397
Mel Gorman41858962009-06-16 15:32:12 -07004398 /* we treat the high watermark as reserved pages. */
4399 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004400
4401 if (max > zone->present_pages)
4402 max = zone->present_pages;
4403 reserve_pages += max;
4404 }
4405 }
4406 totalreserve_pages = reserve_pages;
4407}
4408
4409/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07004410 * setup_per_zone_lowmem_reserve - called whenever
4411 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4412 * has a correct pages reserved value, so an adequate number of
4413 * pages are left in the zone after a successful __alloc_pages().
4414 */
4415static void setup_per_zone_lowmem_reserve(void)
4416{
4417 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004418 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004419
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08004420 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004421 for (j = 0; j < MAX_NR_ZONES; j++) {
4422 struct zone *zone = pgdat->node_zones + j;
4423 unsigned long present_pages = zone->present_pages;
4424
4425 zone->lowmem_reserve[j] = 0;
4426
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004427 idx = j;
4428 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004429 struct zone *lower_zone;
4430
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004431 idx--;
4432
Linus Torvalds1da177e2005-04-16 15:20:36 -07004433 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4434 sysctl_lowmem_reserve_ratio[idx] = 1;
4435
4436 lower_zone = pgdat->node_zones + idx;
4437 lower_zone->lowmem_reserve[j] = present_pages /
4438 sysctl_lowmem_reserve_ratio[idx];
4439 present_pages += lower_zone->present_pages;
4440 }
4441 }
4442 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004443
4444 /* update totalreserve_pages */
4445 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004446}
4447
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004448/**
Minchan Kimbc75d332009-06-16 15:32:48 -07004449 * setup_per_zone_wmarks - called when min_free_kbytes changes
Minchan Kimbce73942009-06-16 15:32:50 -07004450 * or when memory is hot-{added|removed}
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004451 *
Minchan Kimbc75d332009-06-16 15:32:48 -07004452 * Ensures that the watermark[min,low,high] values for each zone are set
4453 * correctly with respect to min_free_kbytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004454 */
Minchan Kimbc75d332009-06-16 15:32:48 -07004455void setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004456{
4457 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4458 unsigned long lowmem_pages = 0;
4459 struct zone *zone;
4460 unsigned long flags;
4461
4462 /* Calculate total number of !ZONE_HIGHMEM pages */
4463 for_each_zone(zone) {
4464 if (!is_highmem(zone))
4465 lowmem_pages += zone->present_pages;
4466 }
4467
4468 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07004469 u64 tmp;
4470
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07004471 spin_lock_irqsave(&zone->lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07004472 tmp = (u64)pages_min * zone->present_pages;
4473 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004474 if (is_highmem(zone)) {
4475 /*
Nick Piggin669ed172005-11-13 16:06:45 -08004476 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4477 * need highmem pages, so cap pages_min to a small
4478 * value here.
4479 *
Mel Gorman41858962009-06-16 15:32:12 -07004480 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Nick Piggin669ed172005-11-13 16:06:45 -08004481 * deltas controls asynch page reclaim, and so should
4482 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004483 */
4484 int min_pages;
4485
4486 min_pages = zone->present_pages / 1024;
4487 if (min_pages < SWAP_CLUSTER_MAX)
4488 min_pages = SWAP_CLUSTER_MAX;
4489 if (min_pages > 128)
4490 min_pages = 128;
Mel Gorman41858962009-06-16 15:32:12 -07004491 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004492 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08004493 /*
4494 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07004495 * proportionate to the zone's size.
4496 */
Mel Gorman41858962009-06-16 15:32:12 -07004497 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004498 }
4499
Mel Gorman41858962009-06-16 15:32:12 -07004500 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
4501 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Mel Gorman56fd56b2007-10-16 01:25:58 -07004502 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07004503 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004504 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004505
4506 /* update totalreserve_pages */
4507 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004508}
4509
Rik van Riel556adec2008-10-18 20:26:34 -07004510/**
Rik van Riel556adec2008-10-18 20:26:34 -07004511 * The inactive anon list should be small enough that the VM never has to
4512 * do too much work, but large enough that each inactive page has a chance
4513 * to be referenced again before it is swapped out.
4514 *
4515 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4516 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4517 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4518 * the anonymous pages are kept on the inactive list.
4519 *
4520 * total target max
4521 * memory ratio inactive anon
4522 * -------------------------------------
4523 * 10MB 1 5MB
4524 * 100MB 1 50MB
4525 * 1GB 3 250MB
4526 * 10GB 10 0.9GB
4527 * 100GB 31 3GB
4528 * 1TB 101 10GB
4529 * 10TB 320 32GB
4530 */
Minchan Kim96cb4df2009-06-16 15:32:49 -07004531void calculate_zone_inactive_ratio(struct zone *zone)
4532{
4533 unsigned int gb, ratio;
4534
4535 /* Zone size in gigabytes */
4536 gb = zone->present_pages >> (30 - PAGE_SHIFT);
4537 if (gb)
4538 ratio = int_sqrt(10 * gb);
4539 else
4540 ratio = 1;
4541
4542 zone->inactive_ratio = ratio;
4543}
4544
Cyrill Gorcunove9bb35d2009-06-16 15:32:32 -07004545static void __init setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07004546{
4547 struct zone *zone;
4548
Minchan Kim96cb4df2009-06-16 15:32:49 -07004549 for_each_zone(zone)
4550 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07004551}
4552
Linus Torvalds1da177e2005-04-16 15:20:36 -07004553/*
4554 * Initialise min_free_kbytes.
4555 *
4556 * For small machines we want it small (128k min). For large machines
4557 * we want it large (64MB max). But it is not linear, because network
4558 * bandwidth does not increase linearly with machine size. We use
4559 *
4560 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4561 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4562 *
4563 * which yields
4564 *
4565 * 16MB: 512k
4566 * 32MB: 724k
4567 * 64MB: 1024k
4568 * 128MB: 1448k
4569 * 256MB: 2048k
4570 * 512MB: 2896k
4571 * 1024MB: 4096k
4572 * 2048MB: 5792k
4573 * 4096MB: 8192k
4574 * 8192MB: 11584k
4575 * 16384MB: 16384k
4576 */
Minchan Kimbc75d332009-06-16 15:32:48 -07004577static int __init init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004578{
4579 unsigned long lowmem_kbytes;
4580
4581 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4582
4583 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4584 if (min_free_kbytes < 128)
4585 min_free_kbytes = 128;
4586 if (min_free_kbytes > 65536)
4587 min_free_kbytes = 65536;
Minchan Kimbc75d332009-06-16 15:32:48 -07004588 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004589 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07004590 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004591 return 0;
4592}
Minchan Kimbc75d332009-06-16 15:32:48 -07004593module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004594
4595/*
4596 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4597 * that we can call two helper functions whenever min_free_kbytes
4598 * changes.
4599 */
4600int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
4601 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4602{
4603 proc_dointvec(table, write, file, buffer, length, ppos);
Mel Gorman3b1d92c2007-05-06 14:49:30 -07004604 if (write)
Minchan Kimbc75d332009-06-16 15:32:48 -07004605 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004606 return 0;
4607}
4608
Christoph Lameter96146342006-07-03 00:24:13 -07004609#ifdef CONFIG_NUMA
4610int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
4611 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4612{
4613 struct zone *zone;
4614 int rc;
4615
4616 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4617 if (rc)
4618 return rc;
4619
4620 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07004621 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07004622 sysctl_min_unmapped_ratio) / 100;
4623 return 0;
4624}
Christoph Lameter0ff38492006-09-25 23:31:52 -07004625
4626int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
4627 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4628{
4629 struct zone *zone;
4630 int rc;
4631
4632 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4633 if (rc)
4634 return rc;
4635
4636 for_each_zone(zone)
4637 zone->min_slab_pages = (zone->present_pages *
4638 sysctl_min_slab_ratio) / 100;
4639 return 0;
4640}
Christoph Lameter96146342006-07-03 00:24:13 -07004641#endif
4642
Linus Torvalds1da177e2005-04-16 15:20:36 -07004643/*
4644 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4645 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4646 * whenever sysctl_lowmem_reserve_ratio changes.
4647 *
4648 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07004649 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07004650 * if in function of the boot time zone sizes.
4651 */
4652int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
4653 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4654{
4655 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4656 setup_per_zone_lowmem_reserve();
4657 return 0;
4658}
4659
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004660/*
4661 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4662 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4663 * can have before it gets flushed back to buddy allocator.
4664 */
4665
4666int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
4667 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4668{
4669 struct zone *zone;
4670 unsigned int cpu;
4671 int ret;
4672
4673 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4674 if (!write || (ret == -EINVAL))
4675 return ret;
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07004676 for_each_populated_zone(zone) {
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004677 for_each_online_cpu(cpu) {
4678 unsigned long high;
4679 high = zone->present_pages / percpu_pagelist_fraction;
4680 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4681 }
4682 }
4683 return 0;
4684}
4685
David S. Millerf034b5d2006-08-24 03:08:07 -07004686int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004687
4688#ifdef CONFIG_NUMA
4689static int __init set_hashdist(char *str)
4690{
4691 if (!str)
4692 return 0;
4693 hashdist = simple_strtoul(str, &str, 0);
4694 return 1;
4695}
4696__setup("hashdist=", set_hashdist);
4697#endif
4698
4699/*
4700 * allocate a large system hash table from bootmem
4701 * - it is assumed that the hash table must contain an exact power-of-2
4702 * quantity of entries
4703 * - limit is the number of hash buckets, not the total allocation size
4704 */
4705void *__init alloc_large_system_hash(const char *tablename,
4706 unsigned long bucketsize,
4707 unsigned long numentries,
4708 int scale,
4709 int flags,
4710 unsigned int *_hash_shift,
4711 unsigned int *_hash_mask,
4712 unsigned long limit)
4713{
4714 unsigned long long max = limit;
4715 unsigned long log2qty, size;
4716 void *table = NULL;
4717
4718 /* allow the kernel cmdline to have a say */
4719 if (!numentries) {
4720 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08004721 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004722 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4723 numentries >>= 20 - PAGE_SHIFT;
4724 numentries <<= 20 - PAGE_SHIFT;
4725
4726 /* limit to 1 bucket per 2^scale bytes of low memory */
4727 if (scale > PAGE_SHIFT)
4728 numentries >>= (scale - PAGE_SHIFT);
4729 else
4730 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08004731
4732 /* Make sure we've got at least a 0-order allocation.. */
4733 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
4734 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004735 }
John Hawkes6e692ed2006-03-25 03:08:02 -08004736 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004737
4738 /* limit allocation size to 1/16 total memory by default */
4739 if (max == 0) {
4740 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4741 do_div(max, bucketsize);
4742 }
4743
4744 if (numentries > max)
4745 numentries = max;
4746
David Howellsf0d1b0b2006-12-08 02:37:49 -08004747 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004748
4749 do {
4750 size = bucketsize << log2qty;
4751 if (flags & HASH_EARLY)
Jan Beulich74768ed2008-08-12 15:08:39 -07004752 table = alloc_bootmem_nopanic(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004753 else if (hashdist)
4754 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4755 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07004756 /*
4757 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07004758 * some pages at the end of hash table which
4759 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07004760 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01004761 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07004762 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01004763 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
4764 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004765 }
4766 } while (!table && size > PAGE_SIZE && --log2qty);
4767
4768 if (!table)
4769 panic("Failed to allocate %s hash table\n", tablename);
4770
Dan Alonib49ad482007-07-15 23:38:23 -07004771 printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07004772 tablename,
4773 (1U << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08004774 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004775 size);
4776
4777 if (_hash_shift)
4778 *_hash_shift = log2qty;
4779 if (_hash_mask)
4780 *_hash_mask = (1 << log2qty) - 1;
4781
4782 return table;
4783}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004784
Mel Gorman835c1342007-10-16 01:25:47 -07004785/* Return a pointer to the bitmap storing bits affecting a block of pages */
4786static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4787 unsigned long pfn)
4788{
4789#ifdef CONFIG_SPARSEMEM
4790 return __pfn_to_section(pfn)->pageblock_flags;
4791#else
4792 return zone->pageblock_flags;
4793#endif /* CONFIG_SPARSEMEM */
4794}
Andrew Morton6220ec72006-10-19 23:29:05 -07004795
Mel Gorman835c1342007-10-16 01:25:47 -07004796static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4797{
4798#ifdef CONFIG_SPARSEMEM
4799 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07004800 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07004801#else
4802 pfn = pfn - zone->zone_start_pfn;
Mel Gormand9c23402007-10-16 01:26:01 -07004803 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07004804#endif /* CONFIG_SPARSEMEM */
4805}
4806
4807/**
Mel Gormand9c23402007-10-16 01:26:01 -07004808 * 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 -07004809 * @page: The page within the block of interest
4810 * @start_bitidx: The first bit of interest to retrieve
4811 * @end_bitidx: The last bit of interest
4812 * returns pageblock_bits flags
4813 */
4814unsigned long get_pageblock_flags_group(struct page *page,
4815 int start_bitidx, int end_bitidx)
4816{
4817 struct zone *zone;
4818 unsigned long *bitmap;
4819 unsigned long pfn, bitidx;
4820 unsigned long flags = 0;
4821 unsigned long value = 1;
4822
4823 zone = page_zone(page);
4824 pfn = page_to_pfn(page);
4825 bitmap = get_pageblock_bitmap(zone, pfn);
4826 bitidx = pfn_to_bitidx(zone, pfn);
4827
4828 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4829 if (test_bit(bitidx + start_bitidx, bitmap))
4830 flags |= value;
4831
4832 return flags;
4833}
4834
4835/**
Mel Gormand9c23402007-10-16 01:26:01 -07004836 * 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 -07004837 * @page: The page within the block of interest
4838 * @start_bitidx: The first bit of interest
4839 * @end_bitidx: The last bit of interest
4840 * @flags: The flags to set
4841 */
4842void set_pageblock_flags_group(struct page *page, unsigned long flags,
4843 int start_bitidx, int end_bitidx)
4844{
4845 struct zone *zone;
4846 unsigned long *bitmap;
4847 unsigned long pfn, bitidx;
4848 unsigned long value = 1;
4849
4850 zone = page_zone(page);
4851 pfn = page_to_pfn(page);
4852 bitmap = get_pageblock_bitmap(zone, pfn);
4853 bitidx = pfn_to_bitidx(zone, pfn);
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004854 VM_BUG_ON(pfn < zone->zone_start_pfn);
4855 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
Mel Gorman835c1342007-10-16 01:25:47 -07004856
4857 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4858 if (flags & value)
4859 __set_bit(bitidx + start_bitidx, bitmap);
4860 else
4861 __clear_bit(bitidx + start_bitidx, bitmap);
4862}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07004863
4864/*
4865 * This is designed as sub function...plz see page_isolation.c also.
4866 * set/clear page block's type to be ISOLATE.
4867 * page allocater never alloc memory from ISOLATE block.
4868 */
4869
4870int set_migratetype_isolate(struct page *page)
4871{
4872 struct zone *zone;
4873 unsigned long flags;
4874 int ret = -EBUSY;
4875
4876 zone = page_zone(page);
4877 spin_lock_irqsave(&zone->lock, flags);
4878 /*
4879 * In future, more migrate types will be able to be isolation target.
4880 */
4881 if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE)
4882 goto out;
4883 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
4884 move_freepages_block(zone, page, MIGRATE_ISOLATE);
4885 ret = 0;
4886out:
4887 spin_unlock_irqrestore(&zone->lock, flags);
4888 if (!ret)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004889 drain_all_pages();
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07004890 return ret;
4891}
4892
4893void unset_migratetype_isolate(struct page *page)
4894{
4895 struct zone *zone;
4896 unsigned long flags;
4897 zone = page_zone(page);
4898 spin_lock_irqsave(&zone->lock, flags);
4899 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
4900 goto out;
4901 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4902 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4903out:
4904 spin_unlock_irqrestore(&zone->lock, flags);
4905}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07004906
4907#ifdef CONFIG_MEMORY_HOTREMOVE
4908/*
4909 * All pages in the range must be isolated before calling this.
4910 */
4911void
4912__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
4913{
4914 struct page *page;
4915 struct zone *zone;
4916 int order, i;
4917 unsigned long pfn;
4918 unsigned long flags;
4919 /* find the first valid pfn */
4920 for (pfn = start_pfn; pfn < end_pfn; pfn++)
4921 if (pfn_valid(pfn))
4922 break;
4923 if (pfn == end_pfn)
4924 return;
4925 zone = page_zone(pfn_to_page(pfn));
4926 spin_lock_irqsave(&zone->lock, flags);
4927 pfn = start_pfn;
4928 while (pfn < end_pfn) {
4929 if (!pfn_valid(pfn)) {
4930 pfn++;
4931 continue;
4932 }
4933 page = pfn_to_page(pfn);
4934 BUG_ON(page_count(page));
4935 BUG_ON(!PageBuddy(page));
4936 order = page_order(page);
4937#ifdef CONFIG_DEBUG_VM
4938 printk(KERN_INFO "remove from free list %lx %d %lx\n",
4939 pfn, 1 << order, end_pfn);
4940#endif
4941 list_del(&page->lru);
4942 rmv_page_order(page);
4943 zone->free_area[order].nr_free--;
4944 __mod_zone_page_state(zone, NR_FREE_PAGES,
4945 - (1UL << order));
4946 for (i = 0; i < (1 << order); i++)
4947 SetPageReserved((page+i));
4948 pfn += (1 << order);
4949 }
4950 spin_unlock_irqrestore(&zone->lock, flags);
4951}
4952#endif