blob: 5675b30738546451c47771e5703f06830a63898a [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/module.h>
27#include <linux/suspend.h>
28#include <linux/pagevec.h>
29#include <linux/blkdev.h>
30#include <linux/slab.h>
David Rientjes5a3135c22007-10-16 23:25:53 -070031#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <linux/notifier.h>
33#include <linux/topology.h>
34#include <linux/sysctl.h>
35#include <linux/cpu.h>
36#include <linux/cpuset.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070037#include <linux/memory_hotplug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <linux/nodemask.h>
39#include <linux/vmalloc.h>
Christoph Lameter4be38e32006-01-06 00:11:17 -080040#include <linux/mempolicy.h>
Yasunori Goto68113782006-06-23 02:03:11 -070041#include <linux/stop_machine.h>
Mel Gormanc7132162006-09-27 01:49:43 -070042#include <linux/sort.h>
43#include <linux/pfn.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070044#include <linux/backing-dev.h>
Akinobu Mita933e3122006-12-08 02:39:45 -080045#include <linux/fault-inject.h>
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -070046#include <linux/page-isolation.h>
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070047#include <linux/page_cgroup.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070048#include <linux/debugobjects.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
50#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070051#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include "internal.h"
53
54/*
Christoph Lameter13808912007-10-16 01:25:27 -070055 * Array of node states.
Linus Torvalds1da177e2005-04-16 15:20:36 -070056 */
Christoph Lameter13808912007-10-16 01:25:27 -070057nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
58 [N_POSSIBLE] = NODE_MASK_ALL,
59 [N_ONLINE] = { { [0] = 1UL } },
60#ifndef CONFIG_NUMA
61 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
62#ifdef CONFIG_HIGHMEM
63 [N_HIGH_MEMORY] = { { [0] = 1UL } },
64#endif
65 [N_CPU] = { { [0] = 1UL } },
66#endif /* NUMA */
67};
68EXPORT_SYMBOL(node_states);
69
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070070unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -070071unsigned long totalreserve_pages __read_mostly;
Hugh Dickins22b31ee2009-01-06 14:40:09 -080072unsigned long highest_memmap_pfn __read_mostly;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -080073int percpu_pagelist_fraction;
Linus Torvalds1da177e2005-04-16 15:20:36 -070074
Mel Gormand9c23402007-10-16 01:26:01 -070075#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
76int pageblock_order __read_mostly;
77#endif
78
Hugh Dickinsd98c7a02006-02-14 13:52:59 -080079static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -080080
Linus Torvalds1da177e2005-04-16 15:20:36 -070081/*
82 * results with 256, 32 in the lowmem_reserve sysctl:
83 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
84 * 1G machine -> (16M dma, 784M normal, 224M high)
85 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
86 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
87 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +010088 *
89 * TBD: should special case ZONE_DMA32 machines here - in those we normally
90 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -070092int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -080093#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -070094 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -080095#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -070096#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -070097 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -070098#endif
Christoph Lametere53ef382006-09-25 23:31:14 -070099#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700100 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700101#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700102 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700103};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104
105EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106
Helge Deller15ad7cd2006-12-06 20:40:36 -0800107static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800108#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700109 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800110#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700111#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700112 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700113#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700114 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700115#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700116 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700117#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700118 "Movable",
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700119};
120
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121int min_free_kbytes = 1024;
122
Yasunori Goto86356ab2006-06-23 02:03:09 -0700123unsigned long __meminitdata nr_kernel_pages;
124unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700125static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
Mel Gormanc7132162006-09-27 01:49:43 -0700127#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
128 /*
Simon Arlott183ff222007-10-20 01:27:18 +0200129 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
Mel Gormanc7132162006-09-27 01:49:43 -0700130 * ranges of memory (RAM) that may be registered with add_active_range().
131 * Ranges passed to add_active_range() will be merged if possible
132 * so the number of times add_active_range() can be called is
133 * related to the number of nodes and the number of holes
134 */
135 #ifdef CONFIG_MAX_ACTIVE_REGIONS
136 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
137 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
138 #else
139 #if MAX_NUMNODES >= 32
140 /* If there can be many nodes, allow up to 50 holes per node */
141 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
142 #else
143 /* By default, allow up to 256 distinct regions */
144 #define MAX_ACTIVE_REGIONS 256
145 #endif
146 #endif
147
Jan Beulich98011f52007-07-15 23:38:17 -0700148 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
149 static int __meminitdata nr_nodemap_entries;
150 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
151 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
Mel Gormanfb014392006-09-27 01:49:59 -0700152#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
Jan Beulich98011f52007-07-15 23:38:17 -0700153 static unsigned long __meminitdata node_boundary_start_pfn[MAX_NUMNODES];
154 static unsigned long __meminitdata node_boundary_end_pfn[MAX_NUMNODES];
Mel Gormanfb014392006-09-27 01:49:59 -0700155#endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
Adrian Bunkb69a7282008-07-23 21:28:12 -0700156 static unsigned long __initdata required_kernelcore;
Adrian Bunk484f51f2007-10-16 01:26:03 -0700157 static unsigned long __initdata required_movablecore;
Adrian Bunkb69a7282008-07-23 21:28:12 -0700158 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Mel Gorman2a1e2742007-07-17 04:03:12 -0700159
160 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
161 int movable_zone;
162 EXPORT_SYMBOL(movable_zone);
Mel Gormanc7132162006-09-27 01:49:43 -0700163#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
164
Miklos Szeredi418508c2007-05-23 13:57:55 -0700165#if MAX_NUMNODES > 1
166int nr_node_ids __read_mostly = MAX_NUMNODES;
167EXPORT_SYMBOL(nr_node_ids);
168#endif
169
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700170int page_group_by_mobility_disabled __read_mostly;
171
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700172static void set_pageblock_migratetype(struct page *page, int migratetype)
173{
174 set_pageblock_flags_group(page, (unsigned long)migratetype,
175 PB_migrate, PB_migrate_end);
176}
177
Nick Piggin13e74442006-01-06 00:10:58 -0800178#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700179static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700181 int ret = 0;
182 unsigned seq;
183 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -0700184
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700185 do {
186 seq = zone_span_seqbegin(zone);
187 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
188 ret = 1;
189 else if (pfn < zone->zone_start_pfn)
190 ret = 1;
191 } while (zone_span_seqretry(zone, seq));
192
193 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700194}
195
196static int page_is_consistent(struct zone *zone, struct page *page)
197{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700198 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700199 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700201 return 0;
202
203 return 1;
204}
205/*
206 * Temporary debugging check for pages not lying within a given zone.
207 */
208static int bad_range(struct zone *zone, struct page *page)
209{
210 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700212 if (!page_is_consistent(zone, page))
213 return 1;
214
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 return 0;
216}
Nick Piggin13e74442006-01-06 00:10:58 -0800217#else
218static inline int bad_range(struct zone *zone, struct page *page)
219{
220 return 0;
221}
222#endif
223
Nick Piggin224abf92006-01-06 00:11:11 -0800224static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800226 static unsigned long resume;
227 static unsigned long nr_shown;
228 static unsigned long nr_unshown;
229
230 /*
231 * Allow a burst of 60 reports, then keep quiet for that minute;
232 * or allow a steady drip of one report per second.
233 */
234 if (nr_shown == 60) {
235 if (time_before(jiffies, resume)) {
236 nr_unshown++;
237 goto out;
238 }
239 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800240 printk(KERN_ALERT
241 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800242 nr_unshown);
243 nr_unshown = 0;
244 }
245 nr_shown = 0;
246 }
247 if (nr_shown++ == 0)
248 resume = jiffies + 60 * HZ;
249
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800250 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800251 current->comm, page_to_pfn(page));
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800252 printk(KERN_ALERT
Hugh Dickins3dc14742009-01-06 14:40:08 -0800253 "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n",
254 page, (void *)page->flags, page_count(page),
255 page_mapcount(page), page->mapping, page->index);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700256
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800258out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800259 /* Leave bad fields for debug, except PageBuddy could make trouble */
260 __ClearPageBuddy(page);
Randy Dunlap9f158332005-09-13 01:25:16 -0700261 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262}
263
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264/*
265 * Higher-order pages are called "compound pages". They are structured thusly:
266 *
267 * The first PAGE_SIZE page is called the "head page".
268 *
269 * The remaining PAGE_SIZE pages are called "tail pages".
270 *
271 * All pages have PG_compound set. All pages have their ->private pointing at
272 * the head page (even the head page has this).
273 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800274 * The first tail page's ->lru.next holds the address of the compound page's
275 * put_page() function. Its ->lru.prev holds the order of allocation.
276 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800278
279static void free_compound_page(struct page *page)
280{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700281 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800282}
283
Andi Kleen01ad1c02008-07-23 21:27:46 -0700284void prep_compound_page(struct page *page, unsigned long order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285{
286 int i;
287 int nr_pages = 1 << order;
288
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800289 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700290 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700291 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800292 for (i = 1; i < nr_pages; i++) {
293 struct page *p = page + i;
294
Christoph Lameterd85f3382007-05-06 14:49:39 -0700295 __SetPageTail(p);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700296 p->first_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 }
298}
299
Andy Whitcroft18229df2008-11-06 12:53:27 -0800300#ifdef CONFIG_HUGETLBFS
301void prep_compound_gigantic_page(struct page *page, unsigned long order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302{
303 int i;
304 int nr_pages = 1 << order;
Andy Whitcroft6babc322008-10-02 14:50:18 -0700305 struct page *p = page + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306
Andy Whitcroft18229df2008-11-06 12:53:27 -0800307 set_compound_page_dtor(page, free_compound_page);
308 set_compound_order(page, order);
309 __SetPageHead(page);
310 for (i = 1; i < nr_pages; i++, p = mem_map_next(p, page, i)) {
311 __SetPageTail(p);
312 p->first_page = page;
313 }
314}
315#endif
316
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800317static int destroy_compound_page(struct page *page, unsigned long order)
Andy Whitcroft18229df2008-11-06 12:53:27 -0800318{
319 int i;
320 int nr_pages = 1 << order;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800321 int bad = 0;
Andy Whitcroft18229df2008-11-06 12:53:27 -0800322
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800323 if (unlikely(compound_order(page) != order) ||
324 unlikely(!PageHead(page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800325 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800326 bad++;
327 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328
Christoph Lameter6d777952007-05-06 14:49:40 -0700329 __ClearPageHead(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800330
Andy Whitcroft18229df2008-11-06 12:53:27 -0800331 for (i = 1; i < nr_pages; i++) {
332 struct page *p = page + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800334 if (unlikely(!PageTail(p) | (p->first_page != page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800335 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800336 bad++;
337 }
Christoph Lameterd85f3382007-05-06 14:49:39 -0700338 __ClearPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800340
341 return bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343
Nick Piggin17cf4402006-03-22 00:08:41 -0800344static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
345{
346 int i;
347
Andrew Morton6626c5d2006-03-22 00:08:42 -0800348 /*
349 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
350 * and __GFP_HIGHMEM from hard or soft interrupt context.
351 */
Nick Piggin725d7042006-09-25 23:30:55 -0700352 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800353 for (i = 0; i < (1 << order); i++)
354 clear_highpage(page + i);
355}
356
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700357static inline void set_page_order(struct page *page, int order)
358{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700359 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000360 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361}
362
363static inline void rmv_page_order(struct page *page)
364{
Nick Piggin676165a2006-04-10 11:21:48 +1000365 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700366 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367}
368
369/*
370 * Locate the struct page for both the matching buddy in our
371 * pair (buddy1) and the combined O(n+1) page they form (page).
372 *
373 * 1) Any buddy B1 will have an order O twin B2 which satisfies
374 * the following equation:
375 * B2 = B1 ^ (1 << O)
376 * For example, if the starting buddy (buddy2) is #8 its order
377 * 1 buddy is #10:
378 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
379 *
380 * 2) Any buddy B will have an order O+1 parent P which
381 * satisfies the following equation:
382 * P = B & ~(1 << O)
383 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200384 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 */
386static inline struct page *
387__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
388{
389 unsigned long buddy_idx = page_idx ^ (1 << order);
390
391 return page + (buddy_idx - page_idx);
392}
393
394static inline unsigned long
395__find_combined_index(unsigned long page_idx, unsigned int order)
396{
397 return (page_idx & ~(1 << order));
398}
399
400/*
401 * This function checks whether a page is free && is the buddy
402 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800403 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000404 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700405 * (c) a page and its buddy have the same order &&
406 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 *
Nick Piggin676165a2006-04-10 11:21:48 +1000408 * For recording whether a page is in the buddy system, we use PG_buddy.
409 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
410 *
411 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700413static inline int page_is_buddy(struct page *page, struct page *buddy,
414 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700416 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800417 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800418
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700419 if (page_zone_id(page) != page_zone_id(buddy))
420 return 0;
421
422 if (PageBuddy(buddy) && page_order(buddy) == order) {
423 BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700424 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000425 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700426 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427}
428
429/*
430 * Freeing function for a buddy system allocator.
431 *
432 * The concept of a buddy system is to maintain direct-mapped table
433 * (containing bit values) for memory blocks of various "orders".
434 * The bottom level table contains the map for the smallest allocatable
435 * units of memory (here, pages), and each level above it describes
436 * pairs of units from the levels below, hence, "buddies".
437 * At a high level, all that happens here is marking the table entry
438 * at the bottom level available, and propagating the changes upward
439 * as necessary, plus some accounting needed to play nicely with other
440 * parts of the VM system.
441 * At each level, we keep a list of pages, which are heads of continuous
Nick Piggin676165a2006-04-10 11:21:48 +1000442 * free pages of length of (1 << order) and marked with PG_buddy. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700443 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 * So when we are allocating or freeing one, we can derive the state of the
445 * other. That is, if we allocate a small block, and both were
446 * free, the remainder of the region must be split into blocks.
447 * If a block is freed, and its buddy is also free, then this
448 * triggers coalescing into a block of larger size.
449 *
450 * -- wli
451 */
452
Nick Piggin48db57f2006-01-08 01:00:42 -0800453static inline void __free_one_page(struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 struct zone *zone, unsigned int order)
455{
456 unsigned long page_idx;
457 int order_size = 1 << order;
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700458 int migratetype = get_pageblock_migratetype(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459
Nick Piggin224abf92006-01-06 00:11:11 -0800460 if (unlikely(PageCompound(page)))
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800461 if (unlikely(destroy_compound_page(page, order)))
462 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463
464 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
465
Nick Piggin725d7042006-09-25 23:30:55 -0700466 VM_BUG_ON(page_idx & (order_size - 1));
467 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468
Christoph Lameterd23ad422007-02-10 01:43:02 -0800469 __mod_zone_page_state(zone, NR_FREE_PAGES, order_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 while (order < MAX_ORDER-1) {
471 unsigned long combined_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 struct page *buddy;
473
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 buddy = __page_find_buddy(page, page_idx, order);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700475 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700476 break;
Nick Piggin13e74442006-01-06 00:10:58 -0800477
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700478 /* Our buddy is free, merge with it and move up one order. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 list_del(&buddy->lru);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700480 zone->free_area[order].nr_free--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 rmv_page_order(buddy);
Nick Piggin13e74442006-01-06 00:10:58 -0800482 combined_idx = __find_combined_index(page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 page = page + (combined_idx - page_idx);
484 page_idx = combined_idx;
485 order++;
486 }
487 set_page_order(page, order);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700488 list_add(&page->lru,
489 &zone->free_area[order].free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 zone->free_area[order].nr_free++;
491}
492
Nick Piggin224abf92006-01-06 00:11:11 -0800493static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494{
Lee Schermerhorn985737c2008-10-18 20:26:53 -0700495 free_page_mlock(page);
Nick Piggin92be2e332006-01-06 00:10:57 -0800496 if (unlikely(page_mapcount(page) |
497 (page->mapping != NULL) |
498 (page_count(page) != 0) |
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800499 (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800500 bad_page(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800501 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800502 }
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800503 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
504 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
505 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506}
507
508/*
509 * Frees a list of pages.
510 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700511 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 *
513 * If the zone was previously in an "all pages pinned" state then look to
514 * see if this freeing clears that state.
515 *
516 * And clear the zone's pages_scanned counter, to hold off the "all pages are
517 * pinned" detection logic.
518 */
Nick Piggin48db57f2006-01-08 01:00:42 -0800519static void free_pages_bulk(struct zone *zone, int count,
520 struct list_head *list, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521{
Nick Pigginc54ad302006-01-06 00:10:56 -0800522 spin_lock(&zone->lock);
David Rientjese815af92007-10-16 23:25:54 -0700523 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 zone->pages_scanned = 0;
Nick Piggin48db57f2006-01-08 01:00:42 -0800525 while (count--) {
526 struct page *page;
527
Nick Piggin725d7042006-09-25 23:30:55 -0700528 VM_BUG_ON(list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 page = list_entry(list->prev, struct page, lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800530 /* have to delete it as __free_one_page list manipulates */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 list_del(&page->lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800532 __free_one_page(page, zone, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800534 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535}
536
Nick Piggin48db57f2006-01-08 01:00:42 -0800537static void free_one_page(struct zone *zone, struct page *page, int order)
538{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700539 spin_lock(&zone->lock);
David Rientjese815af92007-10-16 23:25:54 -0700540 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700541 zone->pages_scanned = 0;
Paul Jackson0798e512006-12-06 20:31:38 -0800542 __free_one_page(page, zone, order);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700543 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800544}
545
546static void __free_pages_ok(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547{
Nick Pigginc54ad302006-01-06 00:10:56 -0800548 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 int i;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800550 int bad = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 for (i = 0 ; i < (1 << order) ; ++i)
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800553 bad += free_pages_check(page + i);
554 if (bad)
Hugh Dickins689bceb2005-11-21 21:32:20 -0800555 return;
556
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700557 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -0700558 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700559 debug_check_no_obj_freed(page_address(page),
560 PAGE_SIZE << order);
561 }
Nick Piggindafb1362006-10-11 01:21:30 -0700562 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800563 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700564
Nick Pigginc54ad302006-01-06 00:10:56 -0800565 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700566 __count_vm_events(PGFREE, 1 << order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800567 free_one_page(page_zone(page), page, order);
Nick Pigginc54ad302006-01-06 00:10:56 -0800568 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569}
570
David Howellsa226f6c2006-01-06 00:11:08 -0800571/*
572 * permit the bootmem allocator to evade page validation on high-order frees
573 */
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700574void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800575{
576 if (order == 0) {
577 __ClearPageReserved(page);
578 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800579 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800580 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800581 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800582 int loop;
583
Nick Piggin545b1ea2006-03-22 00:08:07 -0800584 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800585 for (loop = 0; loop < BITS_PER_LONG; loop++) {
586 struct page *p = &page[loop];
587
Nick Piggin545b1ea2006-03-22 00:08:07 -0800588 if (loop + 1 < BITS_PER_LONG)
589 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800590 __ClearPageReserved(p);
591 set_page_count(p, 0);
592 }
593
Nick Piggin7835e982006-03-22 00:08:40 -0800594 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800595 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800596 }
597}
598
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599
600/*
601 * The order of subdivision here is critical for the IO subsystem.
602 * Please do not alter this order without good reasons and regression
603 * testing. Specifically, as large blocks of memory are subdivided,
604 * the order in which smaller blocks are delivered depends on the order
605 * they're subdivided in this function. This is the primary factor
606 * influencing the order in which pages are delivered to the IO
607 * subsystem according to empirical testing, and this is also justified
608 * by considering the behavior of a buddy system containing a single
609 * large block of memory acted on by a series of small allocations.
610 * This behavior is a critical factor in sglist merging's success.
611 *
612 * -- wli
613 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800614static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700615 int low, int high, struct free_area *area,
616 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617{
618 unsigned long size = 1 << high;
619
620 while (high > low) {
621 area--;
622 high--;
623 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700624 VM_BUG_ON(bad_range(zone, &page[size]));
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700625 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 area->nr_free++;
627 set_page_order(&page[size], high);
628 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629}
630
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631/*
632 * This page is about to be returned from the page allocator
633 */
Nick Piggin17cf4402006-03-22 00:08:41 -0800634static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635{
Nick Piggin92be2e332006-01-06 00:10:57 -0800636 if (unlikely(page_mapcount(page) |
637 (page->mapping != NULL) |
638 (page_count(page) != 0) |
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800639 (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800640 bad_page(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800641 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800642 }
Hugh Dickins689bceb2005-11-21 21:32:20 -0800643
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700644 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800645 set_page_refcounted(page);
Nick Piggincc1025092006-12-06 20:32:00 -0800646
647 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800649
650 if (gfp_flags & __GFP_ZERO)
651 prep_zero_page(page, order, gfp_flags);
652
653 if (order && (gfp_flags & __GFP_COMP))
654 prep_compound_page(page, order);
655
Hugh Dickins689bceb2005-11-21 21:32:20 -0800656 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657}
658
Mel Gorman56fd56b2007-10-16 01:25:58 -0700659/*
660 * Go through the free lists for the given migratetype and remove
661 * the smallest available page from the freelists
662 */
663static struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
664 int migratetype)
665{
666 unsigned int current_order;
667 struct free_area * area;
668 struct page *page;
669
670 /* Find a page of the appropriate size in the preferred list */
671 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
672 area = &(zone->free_area[current_order]);
673 if (list_empty(&area->free_list[migratetype]))
674 continue;
675
676 page = list_entry(area->free_list[migratetype].next,
677 struct page, lru);
678 list_del(&page->lru);
679 rmv_page_order(page);
680 area->nr_free--;
681 __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order));
682 expand(zone, page, order, current_order, area, migratetype);
683 return page;
684 }
685
686 return NULL;
687}
688
689
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700690/*
691 * This array describes the order lists are fallen back to when
692 * the free lists for the desirable migrate type are depleted
693 */
694static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
Mel Gorman64c5e132007-10-16 01:25:59 -0700695 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
696 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
697 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
698 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700699};
700
Mel Gormanc361be52007-10-16 01:25:51 -0700701/*
702 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -0700703 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -0700704 * boundary. If alignment is required, use move_freepages_block()
705 */
Adrian Bunkb69a7282008-07-23 21:28:12 -0700706static int move_freepages(struct zone *zone,
707 struct page *start_page, struct page *end_page,
708 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700709{
710 struct page *page;
711 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -0700712 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -0700713
714#ifndef CONFIG_HOLES_IN_ZONE
715 /*
716 * page_zone is not safe to call in this context when
717 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
718 * anyway as we check zone boundaries in move_freepages_block().
719 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -0700720 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -0700721 */
722 BUG_ON(page_zone(start_page) != page_zone(end_page));
723#endif
724
725 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -0700726 /* Make sure we are not inadvertently changing nodes */
727 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
728
Mel Gormanc361be52007-10-16 01:25:51 -0700729 if (!pfn_valid_within(page_to_pfn(page))) {
730 page++;
731 continue;
732 }
733
734 if (!PageBuddy(page)) {
735 page++;
736 continue;
737 }
738
739 order = page_order(page);
740 list_del(&page->lru);
741 list_add(&page->lru,
742 &zone->free_area[order].free_list[migratetype]);
743 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -0700744 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -0700745 }
746
Mel Gormand1003132007-10-16 01:26:00 -0700747 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -0700748}
749
Adrian Bunkb69a7282008-07-23 21:28:12 -0700750static int move_freepages_block(struct zone *zone, struct page *page,
751 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700752{
753 unsigned long start_pfn, end_pfn;
754 struct page *start_page, *end_page;
755
756 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -0700757 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -0700758 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -0700759 end_page = start_page + pageblock_nr_pages - 1;
760 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -0700761
762 /* Do not cross zone boundaries */
763 if (start_pfn < zone->zone_start_pfn)
764 start_page = page;
765 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
766 return 0;
767
768 return move_freepages(zone, start_page, end_page, migratetype);
769}
770
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700771/* Remove an element from the buddy allocator from the fallback list */
772static struct page *__rmqueue_fallback(struct zone *zone, int order,
773 int start_migratetype)
774{
775 struct free_area * area;
776 int current_order;
777 struct page *page;
778 int migratetype, i;
779
780 /* Find the largest possible block of pages in the other list */
781 for (current_order = MAX_ORDER-1; current_order >= order;
782 --current_order) {
783 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
784 migratetype = fallbacks[start_migratetype][i];
785
Mel Gorman56fd56b2007-10-16 01:25:58 -0700786 /* MIGRATE_RESERVE handled later if necessary */
787 if (migratetype == MIGRATE_RESERVE)
788 continue;
Mel Gormane0104872007-10-16 01:25:53 -0700789
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700790 area = &(zone->free_area[current_order]);
791 if (list_empty(&area->free_list[migratetype]))
792 continue;
793
794 page = list_entry(area->free_list[migratetype].next,
795 struct page, lru);
796 area->nr_free--;
797
798 /*
Mel Gormanc361be52007-10-16 01:25:51 -0700799 * If breaking a large block of pages, move all free
Mel Gorman46dafbc2007-10-16 01:25:55 -0700800 * pages to the preferred allocation list. If falling
801 * back for a reclaimable kernel allocation, be more
802 * agressive about taking ownership of free pages
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700803 */
Mel Gormand9c23402007-10-16 01:26:01 -0700804 if (unlikely(current_order >= (pageblock_order >> 1)) ||
Mel Gorman46dafbc2007-10-16 01:25:55 -0700805 start_migratetype == MIGRATE_RECLAIMABLE) {
806 unsigned long pages;
807 pages = move_freepages_block(zone, page,
808 start_migratetype);
809
810 /* Claim the whole block if over half of it is free */
Mel Gormand9c23402007-10-16 01:26:01 -0700811 if (pages >= (1 << (pageblock_order-1)))
Mel Gorman46dafbc2007-10-16 01:25:55 -0700812 set_pageblock_migratetype(page,
813 start_migratetype);
814
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700815 migratetype = start_migratetype;
Mel Gormanc361be52007-10-16 01:25:51 -0700816 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700817
818 /* Remove the page from the freelists */
819 list_del(&page->lru);
820 rmv_page_order(page);
821 __mod_zone_page_state(zone, NR_FREE_PAGES,
822 -(1UL << order));
823
Mel Gormand9c23402007-10-16 01:26:01 -0700824 if (current_order == pageblock_order)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700825 set_pageblock_migratetype(page,
826 start_migratetype);
827
828 expand(zone, page, order, current_order, area, migratetype);
829 return page;
830 }
831 }
832
Mel Gorman56fd56b2007-10-16 01:25:58 -0700833 /* Use MIGRATE_RESERVE rather than fail an allocation */
834 return __rmqueue_smallest(zone, order, MIGRATE_RESERVE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700835}
836
Mel Gorman56fd56b2007-10-16 01:25:58 -0700837/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 * Do the hard work of removing an element from the buddy allocator.
839 * Call me with the zone->lock already held.
840 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700841static struct page *__rmqueue(struct zone *zone, unsigned int order,
842 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 struct page *page;
845
Mel Gorman56fd56b2007-10-16 01:25:58 -0700846 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847
Mel Gorman56fd56b2007-10-16 01:25:58 -0700848 if (unlikely(!page))
849 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700850
851 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852}
853
854/*
855 * Obtain a specified number of elements from the buddy allocator, all under
856 * a single hold of the lock, for efficiency. Add them to the supplied list.
857 * Returns the number of new pages which were placed at *list.
858 */
859static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700860 unsigned long count, struct list_head *list,
861 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864
Nick Pigginc54ad302006-01-06 00:10:56 -0800865 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700867 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800868 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -0800870
871 /*
872 * Split buddy pages returned by expand() are received here
873 * in physical page order. The page is added to the callers and
874 * list and the list head then moves forward. From the callers
875 * perspective, the linked list is ordered by page number in
876 * some conditions. This is useful for IO devices that can
877 * merge IO requests if the physical pages are ordered
878 * properly.
879 */
Mel Gorman535131e62007-10-16 01:25:49 -0700880 list_add(&page->lru, list);
881 set_page_private(page, migratetype);
Mel Gorman81eabcb2007-12-17 16:20:05 -0800882 list = &page->lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800884 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800885 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886}
887
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700888#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800889/*
Christoph Lameter4037d452007-05-09 02:35:14 -0700890 * Called from the vmstat counter updater to drain pagesets of this
891 * currently executing processor on remote nodes after they have
892 * expired.
893 *
Christoph Lameter879336c2006-03-22 00:09:08 -0800894 * Note that this function must be called with the thread pinned to
895 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800896 */
Christoph Lameter4037d452007-05-09 02:35:14 -0700897void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700898{
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700899 unsigned long flags;
Christoph Lameter4037d452007-05-09 02:35:14 -0700900 int to_drain;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700901
Christoph Lameter4037d452007-05-09 02:35:14 -0700902 local_irq_save(flags);
903 if (pcp->count >= pcp->batch)
904 to_drain = pcp->batch;
905 else
906 to_drain = pcp->count;
907 free_pages_bulk(zone, to_drain, &pcp->list, 0);
908 pcp->count -= to_drain;
909 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700910}
911#endif
912
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800913/*
914 * Drain pages of the indicated processor.
915 *
916 * The processor must either be the current processor and the
917 * thread pinned to the current processor or a processor that
918 * is not online.
919 */
920static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921{
Nick Pigginc54ad302006-01-06 00:10:56 -0800922 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924
925 for_each_zone(zone) {
926 struct per_cpu_pageset *pset;
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800927 struct per_cpu_pages *pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928
Christoph Lameterf2e12bb2007-01-05 16:37:02 -0800929 if (!populated_zone(zone))
930 continue;
931
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700932 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800934 pcp = &pset->pcp;
935 local_irq_save(flags);
936 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
937 pcp->count = 0;
938 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 }
940}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800942/*
943 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
944 */
945void drain_local_pages(void *arg)
946{
947 drain_pages(smp_processor_id());
948}
949
950/*
951 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
952 */
953void drain_all_pages(void)
954{
Jens Axboe15c8b6c2008-05-09 09:39:44 +0200955 on_each_cpu(drain_local_pages, NULL, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800956}
957
Rafael J. Wysocki296699d2007-07-29 23:27:18 +0200958#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959
960void mark_free_pages(struct zone *zone)
961{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700962 unsigned long pfn, max_zone_pfn;
963 unsigned long flags;
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700964 int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 struct list_head *curr;
966
967 if (!zone->spanned_pages)
968 return;
969
970 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700971
972 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
973 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
974 if (pfn_valid(pfn)) {
975 struct page *page = pfn_to_page(pfn);
976
Rafael J. Wysocki7be98232007-05-06 14:50:42 -0700977 if (!swsusp_page_is_forbidden(page))
978 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700979 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700981 for_each_migratetype_order(order, t) {
982 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700983 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700985 pfn = page_to_pfn(list_entry(curr, struct page, lru));
986 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -0700987 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700988 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700989 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 spin_unlock_irqrestore(&zone->lock, flags);
991}
Mel Gormane2c55dc2007-10-16 01:25:50 -0700992#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993
994/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 * Free a 0-order page
996 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800997static void free_hot_cold_page(struct page *page, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998{
999 struct zone *zone = page_zone(page);
1000 struct per_cpu_pages *pcp;
1001 unsigned long flags;
1002
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 if (PageAnon(page))
1004 page->mapping = NULL;
Nick Piggin224abf92006-01-06 00:11:11 -08001005 if (free_pages_check(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001006 return;
1007
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001008 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -07001009 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001010 debug_check_no_obj_freed(page_address(page), PAGE_SIZE);
1011 }
Nick Piggindafb1362006-10-11 01:21:30 -07001012 arch_free_page(page, 0);
Hugh Dickins689bceb2005-11-21 21:32:20 -08001013 kernel_map_pages(page, 1, 0);
1014
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001015 pcp = &zone_pcp(zone, get_cpu())->pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001017 __count_vm_event(PGFREE);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001018 if (cold)
1019 list_add_tail(&page->lru, &pcp->list);
1020 else
1021 list_add(&page->lru, &pcp->list);
Mel Gorman535131e62007-10-16 01:25:49 -07001022 set_page_private(page, get_pageblock_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001024 if (pcp->count >= pcp->high) {
1025 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
1026 pcp->count -= pcp->batch;
1027 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 local_irq_restore(flags);
1029 put_cpu();
1030}
1031
Harvey Harrison920c7a52008-02-04 22:29:26 -08001032void free_hot_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033{
1034 free_hot_cold_page(page, 0);
1035}
1036
Harvey Harrison920c7a52008-02-04 22:29:26 -08001037void free_cold_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038{
1039 free_hot_cold_page(page, 1);
1040}
1041
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001042/*
1043 * split_page takes a non-compound higher-order page, and splits it into
1044 * n (1<<order) sub-pages: page[0..n]
1045 * Each sub-page must be freed individually.
1046 *
1047 * Note: this is probably too low level an operation for use in drivers.
1048 * Please consult with lkml before using this in your driver.
1049 */
1050void split_page(struct page *page, unsigned int order)
1051{
1052 int i;
1053
Nick Piggin725d7042006-09-25 23:30:55 -07001054 VM_BUG_ON(PageCompound(page));
1055 VM_BUG_ON(!page_count(page));
Nick Piggin7835e982006-03-22 00:08:40 -08001056 for (i = 1; i < (1 << order); i++)
1057 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001058}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001059
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060/*
1061 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1062 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1063 * or two.
1064 */
Mel Gorman18ea7e72008-04-28 02:12:14 -07001065static struct page *buffered_rmqueue(struct zone *preferred_zone,
Nick Piggina74609f2006-01-06 00:11:20 -08001066 struct zone *zone, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067{
1068 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001069 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 int cold = !!(gfp_flags & __GFP_COLD);
Nick Piggina74609f2006-01-06 00:11:20 -08001071 int cpu;
Mel Gorman64c5e132007-10-16 01:25:59 -07001072 int migratetype = allocflags_to_migratetype(gfp_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073
Hugh Dickins689bceb2005-11-21 21:32:20 -08001074again:
Nick Piggina74609f2006-01-06 00:11:20 -08001075 cpu = get_cpu();
Nick Piggin48db57f2006-01-08 01:00:42 -08001076 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 struct per_cpu_pages *pcp;
1078
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001079 pcp = &zone_pcp(zone, cpu)->pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001081 if (!pcp->count) {
nkalmala941c7102006-11-02 22:07:04 -08001082 pcp->count = rmqueue_bulk(zone, 0,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001083 pcp->batch, &pcp->list, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001084 if (unlikely(!pcp->count))
1085 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 }
Mel Gorman535131e62007-10-16 01:25:49 -07001087
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001088 /* Find a page of the appropriate migrate type */
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001089 if (cold) {
1090 list_for_each_entry_reverse(page, &pcp->list, lru)
1091 if (page_private(page) == migratetype)
1092 break;
1093 } else {
1094 list_for_each_entry(page, &pcp->list, lru)
1095 if (page_private(page) == migratetype)
1096 break;
1097 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001098
1099 /* Allocate more to the pcp list if necessary */
1100 if (unlikely(&page->lru == &pcp->list)) {
Mel Gorman535131e62007-10-16 01:25:49 -07001101 pcp->count += rmqueue_bulk(zone, 0,
1102 pcp->batch, &pcp->list, migratetype);
1103 page = list_entry(pcp->list.next, struct page, lru);
Mel Gorman535131e62007-10-16 01:25:49 -07001104 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001105
1106 list_del(&page->lru);
1107 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001108 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001110 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001111 spin_unlock(&zone->lock);
1112 if (!page)
1113 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 }
1115
Christoph Lameterf8891e52006-06-30 01:55:45 -07001116 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Mel Gorman18ea7e72008-04-28 02:12:14 -07001117 zone_statistics(preferred_zone, zone);
Nick Piggina74609f2006-01-06 00:11:20 -08001118 local_irq_restore(flags);
1119 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120
Nick Piggin725d7042006-09-25 23:30:55 -07001121 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -08001122 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001123 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001125
1126failed:
1127 local_irq_restore(flags);
1128 put_cpu();
1129 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130}
1131
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001132#define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
Nick Piggin31488902005-11-28 13:44:03 -08001133#define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1134#define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1135#define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1136#define ALLOC_HARDER 0x10 /* try to alloc harder */
1137#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1138#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001139
Akinobu Mita933e3122006-12-08 02:39:45 -08001140#ifdef CONFIG_FAIL_PAGE_ALLOC
1141
1142static struct fail_page_alloc_attr {
1143 struct fault_attr attr;
1144
1145 u32 ignore_gfp_highmem;
1146 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001147 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001148
1149#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1150
1151 struct dentry *ignore_gfp_highmem_file;
1152 struct dentry *ignore_gfp_wait_file;
Akinobu Mita54114992007-07-15 23:40:23 -07001153 struct dentry *min_order_file;
Akinobu Mita933e3122006-12-08 02:39:45 -08001154
1155#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1156
1157} fail_page_alloc = {
1158 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001159 .ignore_gfp_wait = 1,
1160 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001161 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001162};
1163
1164static int __init setup_fail_page_alloc(char *str)
1165{
1166 return setup_fault_attr(&fail_page_alloc.attr, str);
1167}
1168__setup("fail_page_alloc=", setup_fail_page_alloc);
1169
1170static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1171{
Akinobu Mita54114992007-07-15 23:40:23 -07001172 if (order < fail_page_alloc.min_order)
1173 return 0;
Akinobu Mita933e3122006-12-08 02:39:45 -08001174 if (gfp_mask & __GFP_NOFAIL)
1175 return 0;
1176 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1177 return 0;
1178 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1179 return 0;
1180
1181 return should_fail(&fail_page_alloc.attr, 1 << order);
1182}
1183
1184#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1185
1186static int __init fail_page_alloc_debugfs(void)
1187{
1188 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1189 struct dentry *dir;
1190 int err;
1191
1192 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1193 "fail_page_alloc");
1194 if (err)
1195 return err;
1196 dir = fail_page_alloc.attr.dentries.dir;
1197
1198 fail_page_alloc.ignore_gfp_wait_file =
1199 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1200 &fail_page_alloc.ignore_gfp_wait);
1201
1202 fail_page_alloc.ignore_gfp_highmem_file =
1203 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1204 &fail_page_alloc.ignore_gfp_highmem);
Akinobu Mita54114992007-07-15 23:40:23 -07001205 fail_page_alloc.min_order_file =
1206 debugfs_create_u32("min-order", mode, dir,
1207 &fail_page_alloc.min_order);
Akinobu Mita933e3122006-12-08 02:39:45 -08001208
1209 if (!fail_page_alloc.ignore_gfp_wait_file ||
Akinobu Mita54114992007-07-15 23:40:23 -07001210 !fail_page_alloc.ignore_gfp_highmem_file ||
1211 !fail_page_alloc.min_order_file) {
Akinobu Mita933e3122006-12-08 02:39:45 -08001212 err = -ENOMEM;
1213 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1214 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
Akinobu Mita54114992007-07-15 23:40:23 -07001215 debugfs_remove(fail_page_alloc.min_order_file);
Akinobu Mita933e3122006-12-08 02:39:45 -08001216 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1217 }
1218
1219 return err;
1220}
1221
1222late_initcall(fail_page_alloc_debugfs);
1223
1224#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1225
1226#else /* CONFIG_FAIL_PAGE_ALLOC */
1227
1228static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1229{
1230 return 0;
1231}
1232
1233#endif /* CONFIG_FAIL_PAGE_ALLOC */
1234
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235/*
1236 * Return 1 if free pages are above 'mark'. This takes into account the order
1237 * of the allocation.
1238 */
1239int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001240 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241{
1242 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001243 long min = mark;
1244 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 int o;
1246
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001247 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001249 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 min -= min / 4;
1251
1252 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1253 return 0;
1254 for (o = 0; o < order; o++) {
1255 /* At the next order, this order's pages become unavailable */
1256 free_pages -= z->free_area[o].nr_free << o;
1257
1258 /* Require fewer higher order pages to be free */
1259 min >>= 1;
1260
1261 if (free_pages <= min)
1262 return 0;
1263 }
1264 return 1;
1265}
1266
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001267#ifdef CONFIG_NUMA
1268/*
1269 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1270 * skip over zones that are not allowed by the cpuset, or that have
1271 * been recently (in last second) found to be nearly full. See further
1272 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001273 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001274 *
1275 * If the zonelist cache is present in the passed in zonelist, then
1276 * returns a pointer to the allowed node mask (either the current
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001277 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001278 *
1279 * If the zonelist cache is not available for this zonelist, does
1280 * nothing and returns NULL.
1281 *
1282 * If the fullzones BITMAP in the zonelist cache is stale (more than
1283 * a second since last zap'd) then we zap it out (clear its bits.)
1284 *
1285 * We hold off even calling zlc_setup, until after we've checked the
1286 * first zone in the zonelist, on the theory that most allocations will
1287 * be satisfied from that first zone, so best to examine that zone as
1288 * quickly as we can.
1289 */
1290static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1291{
1292 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1293 nodemask_t *allowednodes; /* zonelist_cache approximation */
1294
1295 zlc = zonelist->zlcache_ptr;
1296 if (!zlc)
1297 return NULL;
1298
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001299 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001300 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1301 zlc->last_full_zap = jiffies;
1302 }
1303
1304 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1305 &cpuset_current_mems_allowed :
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001306 &node_states[N_HIGH_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001307 return allowednodes;
1308}
1309
1310/*
1311 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1312 * if it is worth looking at further for free memory:
1313 * 1) Check that the zone isn't thought to be full (doesn't have its
1314 * bit set in the zonelist_cache fullzones BITMAP).
1315 * 2) Check that the zones node (obtained from the zonelist_cache
1316 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1317 * Return true (non-zero) if zone is worth looking at further, or
1318 * else return false (zero) if it is not.
1319 *
1320 * This check -ignores- the distinction between various watermarks,
1321 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1322 * found to be full for any variation of these watermarks, it will
1323 * be considered full for up to one second by all requests, unless
1324 * we are so low on memory on all allowed nodes that we are forced
1325 * into the second scan of the zonelist.
1326 *
1327 * In the second scan we ignore this zonelist cache and exactly
1328 * apply the watermarks to all zones, even it is slower to do so.
1329 * We are low on memory in the second scan, and should leave no stone
1330 * unturned looking for a free page.
1331 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001332static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001333 nodemask_t *allowednodes)
1334{
1335 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1336 int i; /* index of *z in zonelist zones */
1337 int n; /* node that zone *z is on */
1338
1339 zlc = zonelist->zlcache_ptr;
1340 if (!zlc)
1341 return 1;
1342
Mel Gormandd1a2392008-04-28 02:12:17 -07001343 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001344 n = zlc->z_to_n[i];
1345
1346 /* This zone is worth trying if it is allowed but not full */
1347 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1348}
1349
1350/*
1351 * Given 'z' scanning a zonelist, set the corresponding bit in
1352 * zlc->fullzones, so that subsequent attempts to allocate a page
1353 * from that zone don't waste time re-examining it.
1354 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001355static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001356{
1357 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1358 int i; /* index of *z in zonelist zones */
1359
1360 zlc = zonelist->zlcache_ptr;
1361 if (!zlc)
1362 return;
1363
Mel Gormandd1a2392008-04-28 02:12:17 -07001364 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001365
1366 set_bit(i, zlc->fullzones);
1367}
1368
1369#else /* CONFIG_NUMA */
1370
1371static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1372{
1373 return NULL;
1374}
1375
Mel Gormandd1a2392008-04-28 02:12:17 -07001376static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001377 nodemask_t *allowednodes)
1378{
1379 return 1;
1380}
1381
Mel Gormandd1a2392008-04-28 02:12:17 -07001382static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001383{
1384}
1385#endif /* CONFIG_NUMA */
1386
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001387/*
Paul Jackson0798e512006-12-06 20:31:38 -08001388 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001389 * a page.
1390 */
1391static struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001392get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
Mel Gorman54a6eb52008-04-28 02:12:16 -07001393 struct zonelist *zonelist, int high_zoneidx, int alloc_flags)
Martin Hicks753ee722005-06-21 17:14:41 -07001394{
Mel Gormandd1a2392008-04-28 02:12:17 -07001395 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001396 struct page *page = NULL;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001397 int classzone_idx;
Mel Gorman18ea7e72008-04-28 02:12:14 -07001398 struct zone *zone, *preferred_zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001399 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1400 int zlc_active = 0; /* set if using zonelist_cache */
1401 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gorman54a6eb52008-04-28 02:12:16 -07001402
Mel Gorman19770b32008-04-28 02:12:18 -07001403 (void)first_zones_zonelist(zonelist, high_zoneidx, nodemask,
1404 &preferred_zone);
Andy Whitcroft7eb54822008-05-23 13:04:50 -07001405 if (!preferred_zone)
1406 return NULL;
1407
Mel Gorman19770b32008-04-28 02:12:18 -07001408 classzone_idx = zone_idx(preferred_zone);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001409
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001410zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001411 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001412 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001413 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1414 */
Mel Gorman19770b32008-04-28 02:12:18 -07001415 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1416 high_zoneidx, nodemask) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001417 if (NUMA_BUILD && zlc_active &&
1418 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1419 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001420 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001421 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001422 goto try_next_zone;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001423
1424 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001425 unsigned long mark;
1426 if (alloc_flags & ALLOC_WMARK_MIN)
Christoph Lameter1192d522006-09-25 23:31:45 -07001427 mark = zone->pages_min;
Nick Piggin31488902005-11-28 13:44:03 -08001428 else if (alloc_flags & ALLOC_WMARK_LOW)
Christoph Lameter1192d522006-09-25 23:31:45 -07001429 mark = zone->pages_low;
Nick Piggin31488902005-11-28 13:44:03 -08001430 else
Christoph Lameter1192d522006-09-25 23:31:45 -07001431 mark = zone->pages_high;
Paul Jackson0798e512006-12-06 20:31:38 -08001432 if (!zone_watermark_ok(zone, order, mark,
1433 classzone_idx, alloc_flags)) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08001434 if (!zone_reclaim_mode ||
Christoph Lameter1192d522006-09-25 23:31:45 -07001435 !zone_reclaim(zone, gfp_mask, order))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001436 goto this_zone_full;
Paul Jackson0798e512006-12-06 20:31:38 -08001437 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001438 }
1439
Mel Gorman18ea7e72008-04-28 02:12:14 -07001440 page = buffered_rmqueue(preferred_zone, zone, order, gfp_mask);
Paul Jackson0798e512006-12-06 20:31:38 -08001441 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001442 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001443this_zone_full:
1444 if (NUMA_BUILD)
1445 zlc_mark_zone_full(zonelist, z);
1446try_next_zone:
1447 if (NUMA_BUILD && !did_zlc_setup) {
1448 /* we do zlc_setup after the first zone is tried */
1449 allowednodes = zlc_setup(zonelist, alloc_flags);
1450 zlc_active = 1;
1451 did_zlc_setup = 1;
1452 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07001453 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001454
1455 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1456 /* Disable zlc cache for second zonelist scan */
1457 zlc_active = 0;
1458 goto zonelist_scan;
1459 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001460 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07001461}
1462
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463/*
1464 * This is the 'heart' of the zoned buddy allocator.
1465 */
KOSAKI Motohiroe4048e52008-07-23 21:27:01 -07001466struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001467__alloc_pages_internal(gfp_t gfp_mask, unsigned int order,
1468 struct zonelist *zonelist, nodemask_t *nodemask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469{
Al Viro260b2362005-10-21 03:22:44 -04001470 const gfp_t wait = gfp_mask & __GFP_WAIT;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001471 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Mel Gormandd1a2392008-04-28 02:12:17 -07001472 struct zoneref *z;
1473 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 struct page *page;
1475 struct reclaim_state reclaim_state;
1476 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 int do_retry;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001478 int alloc_flags;
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07001479 unsigned long did_some_progress;
1480 unsigned long pages_reclaimed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481
1482 might_sleep_if(wait);
1483
Akinobu Mita933e3122006-12-08 02:39:45 -08001484 if (should_fail_alloc_page(gfp_mask, order))
1485 return NULL;
1486
Jens Axboe6b1de912005-11-17 21:35:02 +01001487restart:
Mel Gormandd1a2392008-04-28 02:12:17 -07001488 z = zonelist->_zonerefs; /* the list of zones suitable for gfp_mask */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489
Mel Gormandd1a2392008-04-28 02:12:17 -07001490 if (unlikely(!z->zone)) {
Christoph Lameter523b9452007-10-16 01:25:37 -07001491 /*
1492 * Happens if we have an empty zonelist as a result of
1493 * GFP_THISNODE being used on a memoryless node
1494 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 return NULL;
1496 }
Jens Axboe6b1de912005-11-17 21:35:02 +01001497
Mel Gorman19770b32008-04-28 02:12:18 -07001498 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
Mel Gorman54a6eb52008-04-28 02:12:16 -07001499 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001500 if (page)
1501 goto got_pg;
1502
Christoph Lameter952f3b52006-12-06 20:33:26 -08001503 /*
1504 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1505 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1506 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1507 * using a larger set of nodes after it has established that the
1508 * allowed per node queues are empty and that nodes are
1509 * over allocated.
1510 */
1511 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1512 goto nopage;
1513
Mel Gormandd1a2392008-04-28 02:12:17 -07001514 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
1515 wakeup_kswapd(zone, order);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001516
Paul Jackson9bf22292005-09-06 15:18:12 -07001517 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001518 * OK, we're below the kswapd watermark and have kicked background
1519 * reclaim. Now things get more complex, so set up alloc_flags according
1520 * to how we want to proceed.
1521 *
1522 * The caller may dip into page reserves a bit more if the caller
1523 * cannot run direct reclaim, or if the caller has realtime scheduling
Paul Jackson4eac9152006-01-11 12:17:19 -08001524 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1525 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
Paul Jackson9bf22292005-09-06 15:18:12 -07001526 */
Nick Piggin31488902005-11-28 13:44:03 -08001527 alloc_flags = ALLOC_WMARK_MIN;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001528 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
1529 alloc_flags |= ALLOC_HARDER;
1530 if (gfp_mask & __GFP_HIGH)
1531 alloc_flags |= ALLOC_HIGH;
Paul Jacksonbdd804f2006-05-20 15:00:09 -07001532 if (wait)
1533 alloc_flags |= ALLOC_CPUSET;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534
1535 /*
1536 * Go through the zonelist again. Let __GFP_HIGH and allocations
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001537 * coming from realtime tasks go deeper into reserves.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 *
1539 * This is the last chance, in general, before the goto nopage.
1540 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
Paul Jackson9bf22292005-09-06 15:18:12 -07001541 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 */
Mel Gorman19770b32008-04-28 02:12:18 -07001543 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
Mel Gorman54a6eb52008-04-28 02:12:16 -07001544 high_zoneidx, alloc_flags);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001545 if (page)
1546 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547
1548 /* This allocation should allow future memory freeing. */
Nick Pigginb84a35b2005-05-01 08:58:36 -07001549
Kirill Korotaevb43a57b2006-12-06 20:32:27 -08001550rebalance:
Nick Pigginb84a35b2005-05-01 08:58:36 -07001551 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1552 && !in_interrupt()) {
1553 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
Kirill Korotaev885036d2005-11-13 16:06:41 -08001554nofail_alloc:
Nick Pigginb84a35b2005-05-01 08:58:36 -07001555 /* go through the zonelist yet again, ignoring mins */
Mel Gorman19770b32008-04-28 02:12:18 -07001556 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman54a6eb52008-04-28 02:12:16 -07001557 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001558 if (page)
1559 goto got_pg;
Kirill Korotaev885036d2005-11-13 16:06:41 -08001560 if (gfp_mask & __GFP_NOFAIL) {
Andrew Morton3fcfab12006-10-19 23:28:16 -07001561 congestion_wait(WRITE, HZ/50);
Kirill Korotaev885036d2005-11-13 16:06:41 -08001562 goto nofail_alloc;
1563 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 }
1565 goto nopage;
1566 }
1567
1568 /* Atomic allocations - we can't balance anything */
1569 if (!wait)
1570 goto nopage;
1571
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 cond_resched();
1573
1574 /* We now go into synchronous reclaim */
Paul Jackson3e0d98b2006-01-08 01:01:49 -08001575 cpuset_memory_pressure_bump();
David Rientjese33c3b52008-11-12 13:25:37 -08001576 /*
1577 * The task's cpuset might have expanded its set of allowable nodes
1578 */
1579 cpuset_update_task_memory_state();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 p->flags |= PF_MEMALLOC;
1581 reclaim_state.reclaimed_slab = 0;
1582 p->reclaim_state = &reclaim_state;
1583
Mel Gormandac1d272008-04-28 02:12:12 -07001584 did_some_progress = try_to_free_pages(zonelist, order, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585
1586 p->reclaim_state = NULL;
1587 p->flags &= ~PF_MEMALLOC;
1588
1589 cond_resched();
1590
Mel Gormane2c55dc2007-10-16 01:25:50 -07001591 if (order != 0)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001592 drain_all_pages();
Mel Gormane2c55dc2007-10-16 01:25:50 -07001593
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 if (likely(did_some_progress)) {
Mel Gorman19770b32008-04-28 02:12:18 -07001595 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman54a6eb52008-04-28 02:12:16 -07001596 zonelist, high_zoneidx, alloc_flags);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001597 if (page)
1598 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07001600 if (!try_set_zone_oom(zonelist, gfp_mask)) {
David Rientjesff0ceb92007-10-16 23:25:56 -07001601 schedule_timeout_uninterruptible(1);
1602 goto restart;
1603 }
1604
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 /*
1606 * Go through the zonelist yet one more time, keep
1607 * very high watermark here, this is only to catch
1608 * a parallel oom killing, we must fail if we're still
1609 * under heavy pressure.
1610 */
Mel Gorman19770b32008-04-28 02:12:18 -07001611 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1612 order, zonelist, high_zoneidx,
1613 ALLOC_WMARK_HIGH|ALLOC_CPUSET);
David Rientjesff0ceb92007-10-16 23:25:56 -07001614 if (page) {
Mel Gormandd1a2392008-04-28 02:12:17 -07001615 clear_zonelist_oom(zonelist, gfp_mask);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001616 goto got_pg;
David Rientjesff0ceb92007-10-16 23:25:56 -07001617 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618
Mel Gormana8bbf722007-07-31 00:37:30 -07001619 /* The OOM killer will not help higher order allocs so fail */
David Rientjesff0ceb92007-10-16 23:25:56 -07001620 if (order > PAGE_ALLOC_COSTLY_ORDER) {
Mel Gormandd1a2392008-04-28 02:12:17 -07001621 clear_zonelist_oom(zonelist, gfp_mask);
Mel Gormana8bbf722007-07-31 00:37:30 -07001622 goto nopage;
David Rientjesff0ceb92007-10-16 23:25:56 -07001623 }
Mel Gormana8bbf722007-07-31 00:37:30 -07001624
Christoph Lameter9b0f8b02006-02-20 18:27:52 -08001625 out_of_memory(zonelist, gfp_mask, order);
Mel Gormandd1a2392008-04-28 02:12:17 -07001626 clear_zonelist_oom(zonelist, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 goto restart;
1628 }
1629
1630 /*
1631 * Don't let big-order allocations loop unless the caller explicitly
1632 * requests that. Wait for some write requests to complete then retry.
1633 *
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07001634 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1635 * means __GFP_NOFAIL, but that may not be true in other
Nishanth Aravamudanab857d02008-04-29 00:58:23 -07001636 * implementations.
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07001637 *
1638 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1639 * specified, then we retry until we no longer reclaim any pages
1640 * (above), or we've reclaimed an order of pages at least as
1641 * large as the allocation's order. In both cases, if the
1642 * allocation still fails, we stop retrying.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07001644 pages_reclaimed += did_some_progress;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 do_retry = 0;
1646 if (!(gfp_mask & __GFP_NORETRY)) {
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07001647 if (order <= PAGE_ALLOC_COSTLY_ORDER) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 do_retry = 1;
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07001649 } else {
1650 if (gfp_mask & __GFP_REPEAT &&
1651 pages_reclaimed < (1 << order))
1652 do_retry = 1;
1653 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 if (gfp_mask & __GFP_NOFAIL)
1655 do_retry = 1;
1656 }
1657 if (do_retry) {
Andrew Morton3fcfab12006-10-19 23:28:16 -07001658 congestion_wait(WRITE, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 goto rebalance;
1660 }
1661
1662nopage:
1663 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1664 printk(KERN_WARNING "%s: page allocation failure."
1665 " order:%d, mode:0x%x\n",
1666 p->comm, order, gfp_mask);
1667 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001668 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670got_pg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 return page;
1672}
KOSAKI Motohiroe4048e52008-07-23 21:27:01 -07001673EXPORT_SYMBOL(__alloc_pages_internal);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674
1675/*
1676 * Common helper functions.
1677 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001678unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679{
1680 struct page * page;
1681 page = alloc_pages(gfp_mask, order);
1682 if (!page)
1683 return 0;
1684 return (unsigned long) page_address(page);
1685}
1686
1687EXPORT_SYMBOL(__get_free_pages);
1688
Harvey Harrison920c7a52008-02-04 22:29:26 -08001689unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690{
1691 struct page * page;
1692
1693 /*
1694 * get_zeroed_page() returns a 32-bit address, which cannot represent
1695 * a highmem page
1696 */
Nick Piggin725d7042006-09-25 23:30:55 -07001697 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698
1699 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1700 if (page)
1701 return (unsigned long) page_address(page);
1702 return 0;
1703}
1704
1705EXPORT_SYMBOL(get_zeroed_page);
1706
1707void __pagevec_free(struct pagevec *pvec)
1708{
1709 int i = pagevec_count(pvec);
1710
1711 while (--i >= 0)
1712 free_hot_cold_page(pvec->pages[i], pvec->cold);
1713}
1714
Harvey Harrison920c7a52008-02-04 22:29:26 -08001715void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716{
Nick Pigginb5810032005-10-29 18:16:12 -07001717 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 if (order == 0)
1719 free_hot_page(page);
1720 else
1721 __free_pages_ok(page, order);
1722 }
1723}
1724
1725EXPORT_SYMBOL(__free_pages);
1726
Harvey Harrison920c7a52008-02-04 22:29:26 -08001727void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728{
1729 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07001730 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 __free_pages(virt_to_page((void *)addr), order);
1732 }
1733}
1734
1735EXPORT_SYMBOL(free_pages);
1736
Timur Tabi2be0ffe2008-07-23 21:28:11 -07001737/**
1738 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1739 * @size: the number of bytes to allocate
1740 * @gfp_mask: GFP flags for the allocation
1741 *
1742 * This function is similar to alloc_pages(), except that it allocates the
1743 * minimum number of pages to satisfy the request. alloc_pages() can only
1744 * allocate memory in power-of-two pages.
1745 *
1746 * This function is also limited by MAX_ORDER.
1747 *
1748 * Memory allocated by this function must be released by free_pages_exact().
1749 */
1750void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
1751{
1752 unsigned int order = get_order(size);
1753 unsigned long addr;
1754
1755 addr = __get_free_pages(gfp_mask, order);
1756 if (addr) {
1757 unsigned long alloc_end = addr + (PAGE_SIZE << order);
1758 unsigned long used = addr + PAGE_ALIGN(size);
1759
1760 split_page(virt_to_page(addr), order);
1761 while (used < alloc_end) {
1762 free_page(used);
1763 used += PAGE_SIZE;
1764 }
1765 }
1766
1767 return (void *)addr;
1768}
1769EXPORT_SYMBOL(alloc_pages_exact);
1770
1771/**
1772 * free_pages_exact - release memory allocated via alloc_pages_exact()
1773 * @virt: the value returned by alloc_pages_exact.
1774 * @size: size of allocation, same value as passed to alloc_pages_exact().
1775 *
1776 * Release the memory allocated by a previous call to alloc_pages_exact.
1777 */
1778void free_pages_exact(void *virt, size_t size)
1779{
1780 unsigned long addr = (unsigned long)virt;
1781 unsigned long end = addr + PAGE_ALIGN(size);
1782
1783 while (addr < end) {
1784 free_page(addr);
1785 addr += PAGE_SIZE;
1786 }
1787}
1788EXPORT_SYMBOL(free_pages_exact);
1789
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790static unsigned int nr_free_zone_pages(int offset)
1791{
Mel Gormandd1a2392008-04-28 02:12:17 -07001792 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001793 struct zone *zone;
1794
Martin J. Blighe310fd42005-07-29 22:59:18 -07001795 /* Just pick one node, since fallback list is circular */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 unsigned int sum = 0;
1797
Mel Gorman0e884602008-04-28 02:12:14 -07001798 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799
Mel Gorman54a6eb52008-04-28 02:12:16 -07001800 for_each_zone_zonelist(zone, z, zonelist, offset) {
Martin J. Blighe310fd42005-07-29 22:59:18 -07001801 unsigned long size = zone->present_pages;
1802 unsigned long high = zone->pages_high;
1803 if (size > high)
1804 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 }
1806
1807 return sum;
1808}
1809
1810/*
1811 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1812 */
1813unsigned int nr_free_buffer_pages(void)
1814{
Al Viroaf4ca452005-10-21 02:55:38 -04001815 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816}
Meelap Shahc2f1a552007-07-17 04:04:39 -07001817EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818
1819/*
1820 * Amount of free RAM allocatable within all zones
1821 */
1822unsigned int nr_free_pagecache_pages(void)
1823{
Mel Gorman2a1e2742007-07-17 04:03:12 -07001824 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07001826
1827static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828{
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07001829 if (NUMA_BUILD)
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08001830 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833void si_meminfo(struct sysinfo *val)
1834{
1835 val->totalram = totalram_pages;
1836 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08001837 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 val->totalhigh = totalhigh_pages;
1840 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 val->mem_unit = PAGE_SIZE;
1842}
1843
1844EXPORT_SYMBOL(si_meminfo);
1845
1846#ifdef CONFIG_NUMA
1847void si_meminfo_node(struct sysinfo *val, int nid)
1848{
1849 pg_data_t *pgdat = NODE_DATA(nid);
1850
1851 val->totalram = pgdat->node_present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08001852 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001853#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08001855 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
1856 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001857#else
1858 val->totalhigh = 0;
1859 val->freehigh = 0;
1860#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 val->mem_unit = PAGE_SIZE;
1862}
1863#endif
1864
1865#define K(x) ((x) << (PAGE_SHIFT-10))
1866
1867/*
1868 * Show free area list (used inside shift_scroll-lock stuff)
1869 * We also calculate the percentage fragmentation. We do this by counting the
1870 * memory on each free list with the exception of the first item on the list.
1871 */
1872void show_free_areas(void)
1873{
Jes Sorensenc7241912006-09-27 01:50:05 -07001874 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 struct zone *zone;
1876
1877 for_each_zone(zone) {
Jes Sorensenc7241912006-09-27 01:50:05 -07001878 if (!populated_zone(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 continue;
Jes Sorensenc7241912006-09-27 01:50:05 -07001880
1881 show_node(zone);
1882 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883
Dave Jones6b482c62005-11-10 15:45:56 -05001884 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 struct per_cpu_pageset *pageset;
1886
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001887 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001889 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
1890 cpu, pageset->pcp.high,
1891 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 }
1893 }
1894
Lee Schermerhorn7b854122008-10-18 20:26:40 -07001895 printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n"
1896 " inactive_file:%lu"
1897//TODO: check/adjust line lengths
1898#ifdef CONFIG_UNEVICTABLE_LRU
1899 " unevictable:%lu"
1900#endif
1901 " dirty:%lu writeback:%lu unstable:%lu\n"
Christoph Lameterd23ad422007-02-10 01:43:02 -08001902 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001903 global_page_state(NR_ACTIVE_ANON),
1904 global_page_state(NR_ACTIVE_FILE),
1905 global_page_state(NR_INACTIVE_ANON),
1906 global_page_state(NR_INACTIVE_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07001907#ifdef CONFIG_UNEVICTABLE_LRU
1908 global_page_state(NR_UNEVICTABLE),
1909#endif
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07001910 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07001911 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07001912 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08001913 global_page_state(NR_FREE_PAGES),
Christoph Lameter972d1a72006-09-25 23:31:51 -07001914 global_page_state(NR_SLAB_RECLAIMABLE) +
1915 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07001916 global_page_state(NR_FILE_MAPPED),
Andrew Mortona25700a2007-02-08 14:20:40 -08001917 global_page_state(NR_PAGETABLE),
1918 global_page_state(NR_BOUNCE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919
1920 for_each_zone(zone) {
1921 int i;
1922
Jes Sorensenc7241912006-09-27 01:50:05 -07001923 if (!populated_zone(zone))
1924 continue;
1925
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 show_node(zone);
1927 printk("%s"
1928 " free:%lukB"
1929 " min:%lukB"
1930 " low:%lukB"
1931 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001932 " active_anon:%lukB"
1933 " inactive_anon:%lukB"
1934 " active_file:%lukB"
1935 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07001936#ifdef CONFIG_UNEVICTABLE_LRU
1937 " unevictable:%lukB"
1938#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 " present:%lukB"
1940 " pages_scanned:%lu"
1941 " all_unreclaimable? %s"
1942 "\n",
1943 zone->name,
Christoph Lameterd23ad422007-02-10 01:43:02 -08001944 K(zone_page_state(zone, NR_FREE_PAGES)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 K(zone->pages_min),
1946 K(zone->pages_low),
1947 K(zone->pages_high),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001948 K(zone_page_state(zone, NR_ACTIVE_ANON)),
1949 K(zone_page_state(zone, NR_INACTIVE_ANON)),
1950 K(zone_page_state(zone, NR_ACTIVE_FILE)),
1951 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07001952#ifdef CONFIG_UNEVICTABLE_LRU
1953 K(zone_page_state(zone, NR_UNEVICTABLE)),
1954#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 K(zone->present_pages),
1956 zone->pages_scanned,
David Rientjese815af92007-10-16 23:25:54 -07001957 (zone_is_all_unreclaimable(zone) ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 );
1959 printk("lowmem_reserve[]:");
1960 for (i = 0; i < MAX_NR_ZONES; i++)
1961 printk(" %lu", zone->lowmem_reserve[i]);
1962 printk("\n");
1963 }
1964
1965 for_each_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001966 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967
Jes Sorensenc7241912006-09-27 01:50:05 -07001968 if (!populated_zone(zone))
1969 continue;
1970
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 show_node(zone);
1972 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973
1974 spin_lock_irqsave(&zone->lock, flags);
1975 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001976 nr[order] = zone->free_area[order].nr_free;
1977 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978 }
1979 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001980 for (order = 0; order < MAX_ORDER; order++)
1981 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 printk("= %lukB\n", K(total));
1983 }
1984
Larry Woodmane6f36022008-02-04 22:29:30 -08001985 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
1986
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 show_swap_cache_info();
1988}
1989
Mel Gorman19770b32008-04-28 02:12:18 -07001990static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
1991{
1992 zoneref->zone = zone;
1993 zoneref->zone_idx = zone_idx(zone);
1994}
1995
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996/*
1997 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08001998 *
1999 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002001static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2002 int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003{
Christoph Lameter1a932052006-01-06 00:11:16 -08002004 struct zone *zone;
2005
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002006 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002007 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08002008
2009 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002010 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08002011 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08002012 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002013 zoneref_set_zone(zone,
2014 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08002015 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08002017
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002018 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08002019 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020}
2021
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002022
2023/*
2024 * zonelist_order:
2025 * 0 = automatic detection of better ordering.
2026 * 1 = order by ([node] distance, -zonetype)
2027 * 2 = order by (-zonetype, [node] distance)
2028 *
2029 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2030 * the same zonelist. So only NUMA can configure this param.
2031 */
2032#define ZONELIST_ORDER_DEFAULT 0
2033#define ZONELIST_ORDER_NODE 1
2034#define ZONELIST_ORDER_ZONE 2
2035
2036/* zonelist order in the kernel.
2037 * set_zonelist_order() will set this to NODE or ZONE.
2038 */
2039static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2040static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2041
2042
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002044/* The value user specified ....changed by config */
2045static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2046/* string for sysctl */
2047#define NUMA_ZONELIST_ORDER_LEN 16
2048char numa_zonelist_order[16] = "default";
2049
2050/*
2051 * interface for configure zonelist ordering.
2052 * command line option "numa_zonelist_order"
2053 * = "[dD]efault - default, automatic configuration.
2054 * = "[nN]ode - order by node locality, then by zone within node
2055 * = "[zZ]one - order by zone, then by locality within zone
2056 */
2057
2058static int __parse_numa_zonelist_order(char *s)
2059{
2060 if (*s == 'd' || *s == 'D') {
2061 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2062 } else if (*s == 'n' || *s == 'N') {
2063 user_zonelist_order = ZONELIST_ORDER_NODE;
2064 } else if (*s == 'z' || *s == 'Z') {
2065 user_zonelist_order = ZONELIST_ORDER_ZONE;
2066 } else {
2067 printk(KERN_WARNING
2068 "Ignoring invalid numa_zonelist_order value: "
2069 "%s\n", s);
2070 return -EINVAL;
2071 }
2072 return 0;
2073}
2074
2075static __init int setup_numa_zonelist_order(char *s)
2076{
2077 if (s)
2078 return __parse_numa_zonelist_order(s);
2079 return 0;
2080}
2081early_param("numa_zonelist_order", setup_numa_zonelist_order);
2082
2083/*
2084 * sysctl handler for numa_zonelist_order
2085 */
2086int numa_zonelist_order_handler(ctl_table *table, int write,
2087 struct file *file, void __user *buffer, size_t *length,
2088 loff_t *ppos)
2089{
2090 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2091 int ret;
2092
2093 if (write)
2094 strncpy(saved_string, (char*)table->data,
2095 NUMA_ZONELIST_ORDER_LEN);
2096 ret = proc_dostring(table, write, file, buffer, length, ppos);
2097 if (ret)
2098 return ret;
2099 if (write) {
2100 int oldval = user_zonelist_order;
2101 if (__parse_numa_zonelist_order((char*)table->data)) {
2102 /*
2103 * bogus value. restore saved string
2104 */
2105 strncpy((char*)table->data, saved_string,
2106 NUMA_ZONELIST_ORDER_LEN);
2107 user_zonelist_order = oldval;
2108 } else if (oldval != user_zonelist_order)
2109 build_all_zonelists();
2110 }
2111 return 0;
2112}
2113
2114
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115#define MAX_NODE_LOAD (num_online_nodes())
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002116static int node_load[MAX_NUMNODES];
2117
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07002119 * 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 -07002120 * @node: node whose fallback list we're appending
2121 * @used_node_mask: nodemask_t of already used nodes
2122 *
2123 * We use a number of factors to determine which is the next node that should
2124 * appear on a given node's fallback list. The node should not have appeared
2125 * already in @node's fallback list, and it should be the next closest node
2126 * according to the distance array (which contains arbitrary distance values
2127 * from each node to each node in the system), and should also prefer nodes
2128 * with no CPUs, since presumably they'll have very little allocation pressure
2129 * on them otherwise.
2130 * It returns -1 if no node is found.
2131 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002132static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002134 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 int min_val = INT_MAX;
2136 int best_node = -1;
Mike Travisc5f59f02008-04-04 18:11:10 -07002137 node_to_cpumask_ptr(tmp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002139 /* Use the local node if we haven't already */
2140 if (!node_isset(node, *used_node_mask)) {
2141 node_set(node, *used_node_mask);
2142 return node;
2143 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002145 for_each_node_state(n, N_HIGH_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146
2147 /* Don't want a node to appear more than once */
2148 if (node_isset(n, *used_node_mask))
2149 continue;
2150
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 /* Use the distance array to find the distance */
2152 val = node_distance(node, n);
2153
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002154 /* Penalize nodes under us ("prefer the next node") */
2155 val += (n < node);
2156
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157 /* Give preference to headless and unused nodes */
Mike Travisc5f59f02008-04-04 18:11:10 -07002158 node_to_cpumask_ptr_next(tmp, n);
2159 if (!cpus_empty(*tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 val += PENALTY_FOR_NODE_WITH_CPUS;
2161
2162 /* Slight preference for less loaded node */
2163 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2164 val += node_load[n];
2165
2166 if (val < min_val) {
2167 min_val = val;
2168 best_node = n;
2169 }
2170 }
2171
2172 if (best_node >= 0)
2173 node_set(best_node, *used_node_mask);
2174
2175 return best_node;
2176}
2177
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002178
2179/*
2180 * Build zonelists ordered by node and zones within node.
2181 * This results in maximum locality--normal zone overflows into local
2182 * DMA zone, if any--but risks exhausting DMA zone.
2183 */
2184static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002186 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002188
Mel Gorman54a6eb52008-04-28 02:12:16 -07002189 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07002190 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07002191 ;
2192 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2193 MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002194 zonelist->_zonerefs[j].zone = NULL;
2195 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002196}
2197
2198/*
Christoph Lameter523b9452007-10-16 01:25:37 -07002199 * Build gfp_thisnode zonelists
2200 */
2201static void build_thisnode_zonelists(pg_data_t *pgdat)
2202{
Christoph Lameter523b9452007-10-16 01:25:37 -07002203 int j;
2204 struct zonelist *zonelist;
2205
Mel Gorman54a6eb52008-04-28 02:12:16 -07002206 zonelist = &pgdat->node_zonelists[1];
2207 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002208 zonelist->_zonerefs[j].zone = NULL;
2209 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07002210}
2211
2212/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002213 * Build zonelists ordered by zone and nodes within zones.
2214 * This results in conserving DMA zone[s] until all Normal memory is
2215 * exhausted, but results in overflowing to remote node while memory
2216 * may still exist in local DMA zone.
2217 */
2218static int node_order[MAX_NUMNODES];
2219
2220static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2221{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002222 int pos, j, node;
2223 int zone_type; /* needs to be signed */
2224 struct zone *z;
2225 struct zonelist *zonelist;
2226
Mel Gorman54a6eb52008-04-28 02:12:16 -07002227 zonelist = &pgdat->node_zonelists[0];
2228 pos = 0;
2229 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2230 for (j = 0; j < nr_nodes; j++) {
2231 node = node_order[j];
2232 z = &NODE_DATA(node)->node_zones[zone_type];
2233 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002234 zoneref_set_zone(z,
2235 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002236 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002237 }
2238 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002239 }
Mel Gormandd1a2392008-04-28 02:12:17 -07002240 zonelist->_zonerefs[pos].zone = NULL;
2241 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002242}
2243
2244static int default_zonelist_order(void)
2245{
2246 int nid, zone_type;
2247 unsigned long low_kmem_size,total_size;
2248 struct zone *z;
2249 int average_size;
2250 /*
2251 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2252 * If they are really small and used heavily, the system can fall
2253 * into OOM very easily.
2254 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2255 */
2256 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2257 low_kmem_size = 0;
2258 total_size = 0;
2259 for_each_online_node(nid) {
2260 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2261 z = &NODE_DATA(nid)->node_zones[zone_type];
2262 if (populated_zone(z)) {
2263 if (zone_type < ZONE_NORMAL)
2264 low_kmem_size += z->present_pages;
2265 total_size += z->present_pages;
2266 }
2267 }
2268 }
2269 if (!low_kmem_size || /* there are no DMA area. */
2270 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2271 return ZONELIST_ORDER_NODE;
2272 /*
2273 * look into each node's config.
2274 * If there is a node whose DMA/DMA32 memory is very big area on
2275 * local memory, NODE_ORDER may be suitable.
2276 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002277 average_size = total_size /
2278 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002279 for_each_online_node(nid) {
2280 low_kmem_size = 0;
2281 total_size = 0;
2282 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2283 z = &NODE_DATA(nid)->node_zones[zone_type];
2284 if (populated_zone(z)) {
2285 if (zone_type < ZONE_NORMAL)
2286 low_kmem_size += z->present_pages;
2287 total_size += z->present_pages;
2288 }
2289 }
2290 if (low_kmem_size &&
2291 total_size > average_size && /* ignore small node */
2292 low_kmem_size > total_size * 70/100)
2293 return ZONELIST_ORDER_NODE;
2294 }
2295 return ZONELIST_ORDER_ZONE;
2296}
2297
2298static void set_zonelist_order(void)
2299{
2300 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2301 current_zonelist_order = default_zonelist_order();
2302 else
2303 current_zonelist_order = user_zonelist_order;
2304}
2305
2306static void build_zonelists(pg_data_t *pgdat)
2307{
2308 int j, node, load;
2309 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002311 int local_node, prev_node;
2312 struct zonelist *zonelist;
2313 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314
2315 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07002316 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07002318 zonelist->_zonerefs[0].zone = NULL;
2319 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 }
2321
2322 /* NUMA-aware ordering of nodes */
2323 local_node = pgdat->node_id;
2324 load = num_online_nodes();
2325 prev_node = local_node;
2326 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002327
2328 memset(node_load, 0, sizeof(node_load));
2329 memset(node_order, 0, sizeof(node_order));
2330 j = 0;
2331
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08002333 int distance = node_distance(local_node, node);
2334
2335 /*
2336 * If another node is sufficiently far away then it is better
2337 * to reclaim pages in a zone before going off node.
2338 */
2339 if (distance > RECLAIM_DISTANCE)
2340 zone_reclaim_mode = 1;
2341
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342 /*
2343 * We don't want to pressure a particular node.
2344 * So adding penalty to the first node in same
2345 * distance group to make it round-robin.
2346 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08002347 if (distance != node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002348 node_load[node] = load;
2349
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350 prev_node = node;
2351 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002352 if (order == ZONELIST_ORDER_NODE)
2353 build_zonelists_in_node_order(pgdat, node);
2354 else
2355 node_order[j++] = node; /* remember order */
2356 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002358 if (order == ZONELIST_ORDER_ZONE) {
2359 /* calculate node order -- i.e., DMA last! */
2360 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 }
Christoph Lameter523b9452007-10-16 01:25:37 -07002362
2363 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364}
2365
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002366/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002367static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002368{
Mel Gorman54a6eb52008-04-28 02:12:16 -07002369 struct zonelist *zonelist;
2370 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07002371 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002372
Mel Gorman54a6eb52008-04-28 02:12:16 -07002373 zonelist = &pgdat->node_zonelists[0];
2374 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2375 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07002376 for (z = zonelist->_zonerefs; z->zone; z++)
2377 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002378}
2379
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002380
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381#else /* CONFIG_NUMA */
2382
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002383static void set_zonelist_order(void)
2384{
2385 current_zonelist_order = ZONELIST_ORDER_ZONE;
2386}
2387
2388static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389{
Christoph Lameter19655d32006-09-25 23:31:19 -07002390 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002391 enum zone_type j;
2392 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393
2394 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395
Mel Gorman54a6eb52008-04-28 02:12:16 -07002396 zonelist = &pgdat->node_zonelists[0];
2397 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398
Mel Gorman54a6eb52008-04-28 02:12:16 -07002399 /*
2400 * Now we build the zonelist so that it contains the zones
2401 * of all the other nodes.
2402 * We don't want to pressure a particular node, so when
2403 * building the zones for node N, we make sure that the
2404 * zones coming right after the local ones are those from
2405 * node N+1 (modulo N)
2406 */
2407 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2408 if (!node_online(node))
2409 continue;
2410 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2411 MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07002413 for (node = 0; node < local_node; node++) {
2414 if (!node_online(node))
2415 continue;
2416 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2417 MAX_NR_ZONES - 1);
2418 }
2419
Mel Gormandd1a2392008-04-28 02:12:17 -07002420 zonelist->_zonerefs[j].zone = NULL;
2421 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422}
2423
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002424/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002425static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002426{
Mel Gorman54a6eb52008-04-28 02:12:16 -07002427 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002428}
2429
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430#endif /* CONFIG_NUMA */
2431
Rusty Russell9b1a4d32008-07-28 12:16:30 -05002432/* return values int ....just for stop_machine() */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002433static int __build_all_zonelists(void *dummy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434{
Yasunori Goto68113782006-06-23 02:03:11 -07002435 int nid;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002436
2437 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07002438 pg_data_t *pgdat = NODE_DATA(nid);
2439
2440 build_zonelists(pgdat);
2441 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002442 }
Yasunori Goto68113782006-06-23 02:03:11 -07002443 return 0;
2444}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002446void build_all_zonelists(void)
Yasunori Goto68113782006-06-23 02:03:11 -07002447{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002448 set_zonelist_order();
2449
Yasunori Goto68113782006-06-23 02:03:11 -07002450 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07002451 __build_all_zonelists(NULL);
Mel Gorman68ad8df2008-07-23 21:26:52 -07002452 mminit_verify_zonelist();
Yasunori Goto68113782006-06-23 02:03:11 -07002453 cpuset_init_current_mems_allowed();
2454 } else {
Simon Arlott183ff222007-10-20 01:27:18 +02002455 /* we have to stop all cpus to guarantee there is no user
Yasunori Goto68113782006-06-23 02:03:11 -07002456 of zonelist */
Rusty Russell9b1a4d32008-07-28 12:16:30 -05002457 stop_machine(__build_all_zonelists, NULL, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07002458 /* cpuset refresh routine should be here */
2459 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07002460 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002461 /*
2462 * Disable grouping by mobility if the number of pages in the
2463 * system is too low to allow the mechanism to work. It would be
2464 * more accurate, but expensive to check per-zone. This check is
2465 * made on memory-hotadd so a system can start with mobility
2466 * disabled and enable it later
2467 */
Mel Gormand9c23402007-10-16 01:26:01 -07002468 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002469 page_group_by_mobility_disabled = 1;
2470 else
2471 page_group_by_mobility_disabled = 0;
2472
2473 printk("Built %i zonelists in %s order, mobility grouping %s. "
2474 "Total pages: %ld\n",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002475 num_online_nodes(),
2476 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002477 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002478 vm_total_pages);
2479#ifdef CONFIG_NUMA
2480 printk("Policy zone: %s\n", zone_names[policy_zone]);
2481#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482}
2483
2484/*
2485 * Helper functions to size the waitqueue hash table.
2486 * Essentially these want to choose hash table sizes sufficiently
2487 * large so that collisions trying to wait on pages are rare.
2488 * But in fact, the number of active page waitqueues on typical
2489 * systems is ridiculously low, less than 200. So this is even
2490 * conservative, even though it seems large.
2491 *
2492 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2493 * waitqueues, i.e. the size of the waitq table given the number of pages.
2494 */
2495#define PAGES_PER_WAITQUEUE 256
2496
Yasunori Gotocca448f2006-06-23 02:03:10 -07002497#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07002498static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499{
2500 unsigned long size = 1;
2501
2502 pages /= PAGES_PER_WAITQUEUE;
2503
2504 while (size < pages)
2505 size <<= 1;
2506
2507 /*
2508 * Once we have dozens or even hundreds of threads sleeping
2509 * on IO we've got bigger problems than wait queue collision.
2510 * Limit the size of the wait table to a reasonable size.
2511 */
2512 size = min(size, 4096UL);
2513
2514 return max(size, 4UL);
2515}
Yasunori Gotocca448f2006-06-23 02:03:10 -07002516#else
2517/*
2518 * A zone's size might be changed by hot-add, so it is not possible to determine
2519 * a suitable size for its wait_table. So we use the maximum size now.
2520 *
2521 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2522 *
2523 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2524 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2525 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2526 *
2527 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2528 * or more by the traditional way. (See above). It equals:
2529 *
2530 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2531 * ia64(16K page size) : = ( 8G + 4M)byte.
2532 * powerpc (64K page size) : = (32G +16M)byte.
2533 */
2534static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2535{
2536 return 4096UL;
2537}
2538#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539
2540/*
2541 * This is an integer logarithm so that shifts can be used later
2542 * to extract the more random high bits from the multiplicative
2543 * hash function before the remainder is taken.
2544 */
2545static inline unsigned long wait_table_bits(unsigned long size)
2546{
2547 return ffz(~size);
2548}
2549
2550#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2551
Mel Gorman56fd56b2007-10-16 01:25:58 -07002552/*
Mel Gormand9c23402007-10-16 01:26:01 -07002553 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman56fd56b2007-10-16 01:25:58 -07002554 * of blocks reserved is based on zone->pages_min. The memory within the
2555 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2556 * higher will lead to a bigger reserve which will get freed as contiguous
2557 * blocks as reclaim kicks in
2558 */
2559static void setup_zone_migrate_reserve(struct zone *zone)
2560{
2561 unsigned long start_pfn, pfn, end_pfn;
2562 struct page *page;
2563 unsigned long reserve, block_migratetype;
2564
2565 /* Get the start pfn, end pfn and the number of blocks to reserve */
2566 start_pfn = zone->zone_start_pfn;
2567 end_pfn = start_pfn + zone->spanned_pages;
Mel Gormand9c23402007-10-16 01:26:01 -07002568 reserve = roundup(zone->pages_min, pageblock_nr_pages) >>
2569 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07002570
Mel Gormand9c23402007-10-16 01:26:01 -07002571 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07002572 if (!pfn_valid(pfn))
2573 continue;
2574 page = pfn_to_page(pfn);
2575
Adam Litke344c7902008-09-02 14:35:38 -07002576 /* Watch out for overlapping nodes */
2577 if (page_to_nid(page) != zone_to_nid(zone))
2578 continue;
2579
Mel Gorman56fd56b2007-10-16 01:25:58 -07002580 /* Blocks with reserved pages will never free, skip them. */
2581 if (PageReserved(page))
2582 continue;
2583
2584 block_migratetype = get_pageblock_migratetype(page);
2585
2586 /* If this block is reserved, account for it */
2587 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
2588 reserve--;
2589 continue;
2590 }
2591
2592 /* Suitable for reserving if this block is movable */
2593 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
2594 set_pageblock_migratetype(page, MIGRATE_RESERVE);
2595 move_freepages_block(zone, page, MIGRATE_RESERVE);
2596 reserve--;
2597 continue;
2598 }
2599
2600 /*
2601 * If the reserve is met and this is a previous reserved block,
2602 * take it back
2603 */
2604 if (block_migratetype == MIGRATE_RESERVE) {
2605 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2606 move_freepages_block(zone, page, MIGRATE_MOVABLE);
2607 }
2608 }
2609}
Mel Gormanac0e5b72007-10-16 01:25:58 -07002610
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611/*
2612 * Initially all pages are reserved - free ones are freed
2613 * up by free_all_bootmem() once the early boot process is
2614 * done. Non-atomic initialization, single-pass.
2615 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002616void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08002617 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07002620 unsigned long end_pfn = start_pfn + size;
2621 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002622 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623
Hugh Dickins22b31ee2009-01-06 14:40:09 -08002624 if (highest_memmap_pfn < end_pfn - 1)
2625 highest_memmap_pfn = end_pfn - 1;
2626
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002627 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08002628 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08002629 /*
2630 * There can be holes in boot-time mem_map[]s
2631 * handed to this function. They do not
2632 * exist on hotplugged memory.
2633 */
2634 if (context == MEMMAP_EARLY) {
2635 if (!early_pfn_valid(pfn))
2636 continue;
2637 if (!early_pfn_in_nid(pfn, nid))
2638 continue;
2639 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002640 page = pfn_to_page(pfn);
2641 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07002642 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08002643 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644 reset_page_mapcount(page);
2645 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002646 /*
2647 * Mark the block movable so that blocks are reserved for
2648 * movable at startup. This will force kernel allocations
2649 * to reserve their blocks rather than leaking throughout
2650 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07002651 * kernel allocations are made. Later some blocks near
2652 * the start are marked MIGRATE_RESERVE by
2653 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002654 *
2655 * bitmap is created for zone's valid pfn range. but memmap
2656 * can be created for invalid pages (for alignment)
2657 * check here not to call set_pageblock_migratetype() against
2658 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002659 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002660 if ((z->zone_start_pfn <= pfn)
2661 && (pfn < z->zone_start_pfn + z->spanned_pages)
2662 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07002663 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002664
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665 INIT_LIST_HEAD(&page->lru);
2666#ifdef WANT_PAGE_VIRTUAL
2667 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2668 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07002669 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671 }
2672}
2673
Andi Kleen1e548de2008-02-04 22:29:26 -08002674static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675{
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002676 int order, t;
2677 for_each_migratetype_order(order, t) {
2678 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679 zone->free_area[order].nr_free = 0;
2680 }
2681}
2682
2683#ifndef __HAVE_ARCH_MEMMAP_INIT
2684#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08002685 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686#endif
2687
Sam Ravnborg1d6f4e62008-01-17 15:21:12 -08002688static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002689{
2690 int batch;
2691
2692 /*
2693 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07002694 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002695 *
2696 * OK, so we don't know how big the cache is. So guess.
2697 */
2698 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07002699 if (batch * PAGE_SIZE > 512 * 1024)
2700 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002701 batch /= 4; /* We effectively *= 4 below */
2702 if (batch < 1)
2703 batch = 1;
2704
2705 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002706 * Clamp the batch to a 2^n - 1 value. Having a power
2707 * of 2 value was found to be more likely to have
2708 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002709 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002710 * For example if 2 tasks are alternately allocating
2711 * batches of pages, one task can end up with a lot
2712 * of pages of one half of the possible page colors
2713 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002714 */
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002715 batch = (1 << (fls(batch + batch/2)-1)) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07002716
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002717 return batch;
2718}
2719
Adrian Bunkb69a7282008-07-23 21:28:12 -07002720static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
Christoph Lameter2caaad42005-06-21 17:15:00 -07002721{
2722 struct per_cpu_pages *pcp;
2723
Magnus Damm1c6fe942005-10-26 01:58:59 -07002724 memset(p, 0, sizeof(*p));
2725
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002726 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07002727 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07002728 pcp->high = 6 * batch;
2729 pcp->batch = max(1UL, 1 * batch);
2730 INIT_LIST_HEAD(&pcp->list);
Christoph Lameter2caaad42005-06-21 17:15:00 -07002731}
2732
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002733/*
2734 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2735 * to the value high for the pageset p.
2736 */
2737
2738static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2739 unsigned long high)
2740{
2741 struct per_cpu_pages *pcp;
2742
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002743 pcp = &p->pcp;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002744 pcp->high = high;
2745 pcp->batch = max(1UL, high/4);
2746 if ((high/4) > (PAGE_SHIFT * 8))
2747 pcp->batch = PAGE_SHIFT * 8;
2748}
2749
2750
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002751#ifdef CONFIG_NUMA
2752/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07002753 * Boot pageset table. One per cpu which is going to be used for all
2754 * zones and all nodes. The parameters will be set in such a way
2755 * that an item put on a list will immediately be handed over to
2756 * the buddy list. This is safe since pageset manipulation is done
2757 * with interrupts disabled.
2758 *
2759 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07002760 *
2761 * The boot_pagesets must be kept even after bootup is complete for
2762 * unused processors and/or zones. They do play a role for bootstrapping
2763 * hotplugged processors.
2764 *
2765 * zoneinfo_show() and maybe other functions do
2766 * not check if the processor is online before following the pageset pointer.
2767 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07002768 */
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08002769static struct per_cpu_pageset boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07002770
2771/*
2772 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002773 * per cpu pageset array in struct zone.
2774 */
Ashok Raj6292d9a2006-02-01 03:04:44 -08002775static int __cpuinit process_zones(int cpu)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002776{
2777 struct zone *zone, *dzone;
Christoph Lameter37c07082007-10-16 01:25:36 -07002778 int node = cpu_to_node(cpu);
2779
2780 node_set_state(node, N_CPU); /* this node has a cpu */
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002781
2782 for_each_zone(zone) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002783
Christoph Lameter66a55032006-09-27 01:50:09 -07002784 if (!populated_zone(zone))
2785 continue;
2786
Nick Piggin23316bc2006-01-08 01:00:41 -08002787 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lameter37c07082007-10-16 01:25:36 -07002788 GFP_KERNEL, node);
Nick Piggin23316bc2006-01-08 01:00:41 -08002789 if (!zone_pcp(zone, cpu))
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002790 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002791
Nick Piggin23316bc2006-01-08 01:00:41 -08002792 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002793
2794 if (percpu_pagelist_fraction)
2795 setup_pagelist_highmark(zone_pcp(zone, cpu),
2796 (zone->present_pages / percpu_pagelist_fraction));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002797 }
2798
2799 return 0;
2800bad:
2801 for_each_zone(dzone) {
Andrew Morton64191682007-08-30 23:56:17 -07002802 if (!populated_zone(dzone))
2803 continue;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002804 if (dzone == zone)
2805 break;
Nick Piggin23316bc2006-01-08 01:00:41 -08002806 kfree(zone_pcp(dzone, cpu));
2807 zone_pcp(dzone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002808 }
2809 return -ENOMEM;
2810}
2811
2812static inline void free_zone_pagesets(int cpu)
2813{
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002814 struct zone *zone;
2815
2816 for_each_zone(zone) {
2817 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2818
David Rientjesf3ef9ea2006-09-25 16:24:57 -07002819 /* Free per_cpu_pageset if it is slab allocated */
2820 if (pset != &boot_pageset[cpu])
2821 kfree(pset);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002822 zone_pcp(zone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002823 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002824}
2825
Chandra Seetharaman9c7b2162006-06-27 02:54:07 -07002826static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002827 unsigned long action,
2828 void *hcpu)
2829{
2830 int cpu = (long)hcpu;
2831 int ret = NOTIFY_OK;
2832
2833 switch (action) {
Andy Whitcroftce421c72006-12-06 20:33:08 -08002834 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002835 case CPU_UP_PREPARE_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08002836 if (process_zones(cpu))
2837 ret = NOTIFY_BAD;
2838 break;
2839 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002840 case CPU_UP_CANCELED_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08002841 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002842 case CPU_DEAD_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08002843 free_zone_pagesets(cpu);
2844 break;
2845 default:
2846 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002847 }
2848 return ret;
2849}
2850
Chandra Seetharaman74b85f32006-06-27 02:54:09 -07002851static struct notifier_block __cpuinitdata pageset_notifier =
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002852 { &pageset_cpuup_callback, NULL, 0 };
2853
Al Viro78d99552005-12-15 09:18:25 +00002854void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002855{
2856 int err;
2857
2858 /* Initialize per_cpu_pageset for cpu 0.
2859 * A cpuup callback will do this for every cpu
2860 * as it comes online
2861 */
2862 err = process_zones(smp_processor_id());
2863 BUG_ON(err);
2864 register_cpu_notifier(&pageset_notifier);
2865}
2866
2867#endif
2868
Sam Ravnborg577a32f2007-05-17 23:29:25 +02002869static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07002870int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07002871{
2872 int i;
2873 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07002874 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07002875
2876 /*
2877 * The per-page waitqueue mechanism uses hashed waitqueues
2878 * per zone.
2879 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07002880 zone->wait_table_hash_nr_entries =
2881 wait_table_hash_nr_entries(zone_size_pages);
2882 zone->wait_table_bits =
2883 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07002884 alloc_size = zone->wait_table_hash_nr_entries
2885 * sizeof(wait_queue_head_t);
2886
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07002887 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07002888 zone->wait_table = (wait_queue_head_t *)
2889 alloc_bootmem_node(pgdat, alloc_size);
2890 } else {
2891 /*
2892 * This case means that a zone whose size was 0 gets new memory
2893 * via memory hot-add.
2894 * But it may be the case that a new node was hot-added. In
2895 * this case vmalloc() will not be able to use this new node's
2896 * memory - this wait_table must be initialized to use this new
2897 * node itself as well.
2898 * To use this new node's memory, further consideration will be
2899 * necessary.
2900 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07002901 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07002902 }
2903 if (!zone->wait_table)
2904 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07002905
Yasunori Goto02b694d2006-06-23 02:03:08 -07002906 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07002907 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07002908
2909 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07002910}
2911
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002912static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07002913{
2914 int cpu;
2915 unsigned long batch = zone_batchsize(zone);
2916
2917 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2918#ifdef CONFIG_NUMA
2919 /* Early boot. Slab allocator not functional yet */
Nick Piggin23316bc2006-01-08 01:00:41 -08002920 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Dave Hansened8ece22005-10-29 18:16:50 -07002921 setup_pageset(&boot_pageset[cpu],0);
2922#else
2923 setup_pageset(zone_pcp(zone,cpu), batch);
2924#endif
2925 }
Anton Blanchardf5335c02006-03-25 03:06:49 -08002926 if (zone->present_pages)
2927 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2928 zone->name, zone->present_pages, batch);
Dave Hansened8ece22005-10-29 18:16:50 -07002929}
2930
Yasunori Goto718127c2006-06-23 02:03:10 -07002931__meminit int init_currently_empty_zone(struct zone *zone,
2932 unsigned long zone_start_pfn,
Dave Hansena2f3aa022007-01-10 23:15:30 -08002933 unsigned long size,
2934 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07002935{
2936 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07002937 int ret;
2938 ret = zone_wait_table_init(zone, size);
2939 if (ret)
2940 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07002941 pgdat->nr_zones = zone_idx(zone) + 1;
2942
Dave Hansened8ece22005-10-29 18:16:50 -07002943 zone->zone_start_pfn = zone_start_pfn;
2944
Mel Gorman708614e2008-07-23 21:26:51 -07002945 mminit_dprintk(MMINIT_TRACE, "memmap_init",
2946 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
2947 pgdat->node_id,
2948 (unsigned long)zone_idx(zone),
2949 zone_start_pfn, (zone_start_pfn + size));
2950
Andi Kleen1e548de2008-02-04 22:29:26 -08002951 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07002952
2953 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07002954}
2955
Mel Gormanc7132162006-09-27 01:49:43 -07002956#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2957/*
2958 * Basic iterator support. Return the first range of PFNs for a node
2959 * Note: nid == MAX_NUMNODES returns first region regardless of node
2960 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07002961static int __meminit first_active_region_index_in_nid(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07002962{
2963 int i;
2964
2965 for (i = 0; i < nr_nodemap_entries; i++)
2966 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
2967 return i;
2968
2969 return -1;
2970}
2971
2972/*
2973 * Basic iterator support. Return the next active range of PFNs for a node
Simon Arlott183ff222007-10-20 01:27:18 +02002974 * Note: nid == MAX_NUMNODES returns next region regardless of node
Mel Gormanc7132162006-09-27 01:49:43 -07002975 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07002976static int __meminit next_active_region_index_in_nid(int index, int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07002977{
2978 for (index = index + 1; index < nr_nodemap_entries; index++)
2979 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
2980 return index;
2981
2982 return -1;
2983}
2984
2985#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2986/*
2987 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2988 * Architectures may implement their own version but if add_active_range()
2989 * was used and there are no special requirements, this is a convenient
2990 * alternative
2991 */
Stephen Rothwell6f076f52007-05-10 03:15:27 -07002992int __meminit early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07002993{
2994 int i;
2995
2996 for (i = 0; i < nr_nodemap_entries; i++) {
2997 unsigned long start_pfn = early_node_map[i].start_pfn;
2998 unsigned long end_pfn = early_node_map[i].end_pfn;
2999
3000 if (start_pfn <= pfn && pfn < end_pfn)
3001 return early_node_map[i].nid;
3002 }
3003
3004 return 0;
3005}
3006#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3007
3008/* Basic iterator support to walk early_node_map[] */
3009#define for_each_active_range_index_in_nid(i, nid) \
3010 for (i = first_active_region_index_in_nid(nid); i != -1; \
3011 i = next_active_region_index_in_nid(i, nid))
3012
3013/**
3014 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003015 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3016 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07003017 *
3018 * If an architecture guarantees that all ranges registered with
3019 * add_active_ranges() contain no holes and may be freed, this
3020 * this function may be used instead of calling free_bootmem() manually.
3021 */
3022void __init free_bootmem_with_active_regions(int nid,
3023 unsigned long max_low_pfn)
3024{
3025 int i;
3026
3027 for_each_active_range_index_in_nid(i, nid) {
3028 unsigned long size_pages = 0;
3029 unsigned long end_pfn = early_node_map[i].end_pfn;
3030
3031 if (early_node_map[i].start_pfn >= max_low_pfn)
3032 continue;
3033
3034 if (end_pfn > max_low_pfn)
3035 end_pfn = max_low_pfn;
3036
3037 size_pages = end_pfn - early_node_map[i].start_pfn;
3038 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3039 PFN_PHYS(early_node_map[i].start_pfn),
3040 size_pages << PAGE_SHIFT);
3041 }
3042}
3043
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003044void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3045{
3046 int i;
Yinghai Lud52d53b2008-06-16 20:10:55 -07003047 int ret;
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003048
Yinghai Lud52d53b2008-06-16 20:10:55 -07003049 for_each_active_range_index_in_nid(i, nid) {
3050 ret = work_fn(early_node_map[i].start_pfn,
3051 early_node_map[i].end_pfn, data);
3052 if (ret)
3053 break;
3054 }
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003055}
Mel Gormanc7132162006-09-27 01:49:43 -07003056/**
3057 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003058 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07003059 *
3060 * If an architecture guarantees that all ranges registered with
3061 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003062 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07003063 */
3064void __init sparse_memory_present_with_active_regions(int nid)
3065{
3066 int i;
3067
3068 for_each_active_range_index_in_nid(i, nid)
3069 memory_present(early_node_map[i].nid,
3070 early_node_map[i].start_pfn,
3071 early_node_map[i].end_pfn);
3072}
3073
3074/**
Mel Gormanfb014392006-09-27 01:49:59 -07003075 * push_node_boundaries - Push node boundaries to at least the requested boundary
3076 * @nid: The nid of the node to push the boundary for
3077 * @start_pfn: The start pfn of the node
3078 * @end_pfn: The end pfn of the node
3079 *
3080 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
3081 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
3082 * be hotplugged even though no physical memory exists. This function allows
3083 * an arch to push out the node boundaries so mem_map is allocated that can
3084 * be used later.
3085 */
3086#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3087void __init push_node_boundaries(unsigned int nid,
3088 unsigned long start_pfn, unsigned long end_pfn)
3089{
Mel Gorman6b74ab92008-07-23 21:26:49 -07003090 mminit_dprintk(MMINIT_TRACE, "zoneboundary",
3091 "Entering push_node_boundaries(%u, %lu, %lu)\n",
Mel Gormanfb014392006-09-27 01:49:59 -07003092 nid, start_pfn, end_pfn);
3093
3094 /* Initialise the boundary for this node if necessary */
3095 if (node_boundary_end_pfn[nid] == 0)
3096 node_boundary_start_pfn[nid] = -1UL;
3097
3098 /* Update the boundaries */
3099 if (node_boundary_start_pfn[nid] > start_pfn)
3100 node_boundary_start_pfn[nid] = start_pfn;
3101 if (node_boundary_end_pfn[nid] < end_pfn)
3102 node_boundary_end_pfn[nid] = end_pfn;
3103}
3104
3105/* If necessary, push the node boundary out for reserve hotadd */
Jan Beulich98011f52007-07-15 23:38:17 -07003106static void __meminit account_node_boundary(unsigned int nid,
Mel Gormanfb014392006-09-27 01:49:59 -07003107 unsigned long *start_pfn, unsigned long *end_pfn)
3108{
Mel Gorman6b74ab92008-07-23 21:26:49 -07003109 mminit_dprintk(MMINIT_TRACE, "zoneboundary",
3110 "Entering account_node_boundary(%u, %lu, %lu)\n",
Mel Gormanfb014392006-09-27 01:49:59 -07003111 nid, *start_pfn, *end_pfn);
3112
3113 /* Return if boundary information has not been provided */
3114 if (node_boundary_end_pfn[nid] == 0)
3115 return;
3116
3117 /* Check the boundaries and update if necessary */
3118 if (node_boundary_start_pfn[nid] < *start_pfn)
3119 *start_pfn = node_boundary_start_pfn[nid];
3120 if (node_boundary_end_pfn[nid] > *end_pfn)
3121 *end_pfn = node_boundary_end_pfn[nid];
3122}
3123#else
3124void __init push_node_boundaries(unsigned int nid,
3125 unsigned long start_pfn, unsigned long end_pfn) {}
3126
Jan Beulich98011f52007-07-15 23:38:17 -07003127static void __meminit account_node_boundary(unsigned int nid,
Mel Gormanfb014392006-09-27 01:49:59 -07003128 unsigned long *start_pfn, unsigned long *end_pfn) {}
3129#endif
3130
3131
3132/**
Mel Gormanc7132162006-09-27 01:49:43 -07003133 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003134 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3135 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3136 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07003137 *
3138 * It returns the start and end page frame of a node based on information
3139 * provided by an arch calling add_active_range(). If called for a node
3140 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003141 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07003142 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003143void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003144 unsigned long *start_pfn, unsigned long *end_pfn)
3145{
3146 int i;
3147 *start_pfn = -1UL;
3148 *end_pfn = 0;
3149
3150 for_each_active_range_index_in_nid(i, nid) {
3151 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3152 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3153 }
3154
Christoph Lameter633c0662007-10-16 01:25:37 -07003155 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07003156 *start_pfn = 0;
Mel Gormanfb014392006-09-27 01:49:59 -07003157
3158 /* Push the node boundaries out if requested */
3159 account_node_boundary(nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003160}
3161
3162/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07003163 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3164 * assumption is made that zones within a node are ordered in monotonic
3165 * increasing memory addresses so that the "highest" populated zone is used
3166 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003167static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07003168{
3169 int zone_index;
3170 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3171 if (zone_index == ZONE_MOVABLE)
3172 continue;
3173
3174 if (arch_zone_highest_possible_pfn[zone_index] >
3175 arch_zone_lowest_possible_pfn[zone_index])
3176 break;
3177 }
3178
3179 VM_BUG_ON(zone_index == -1);
3180 movable_zone = zone_index;
3181}
3182
3183/*
3184 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3185 * because it is sized independant of architecture. Unlike the other zones,
3186 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3187 * in each node depending on the size of each node and how evenly kernelcore
3188 * is distributed. This helper function adjusts the zone ranges
3189 * provided by the architecture for a given node by using the end of the
3190 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3191 * zones within a node are in order of monotonic increases memory addresses
3192 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003193static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07003194 unsigned long zone_type,
3195 unsigned long node_start_pfn,
3196 unsigned long node_end_pfn,
3197 unsigned long *zone_start_pfn,
3198 unsigned long *zone_end_pfn)
3199{
3200 /* Only adjust if ZONE_MOVABLE is on this node */
3201 if (zone_movable_pfn[nid]) {
3202 /* Size ZONE_MOVABLE */
3203 if (zone_type == ZONE_MOVABLE) {
3204 *zone_start_pfn = zone_movable_pfn[nid];
3205 *zone_end_pfn = min(node_end_pfn,
3206 arch_zone_highest_possible_pfn[movable_zone]);
3207
3208 /* Adjust for ZONE_MOVABLE starting within this range */
3209 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3210 *zone_end_pfn > zone_movable_pfn[nid]) {
3211 *zone_end_pfn = zone_movable_pfn[nid];
3212
3213 /* Check if this whole range is within ZONE_MOVABLE */
3214 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3215 *zone_start_pfn = *zone_end_pfn;
3216 }
3217}
3218
3219/*
Mel Gormanc7132162006-09-27 01:49:43 -07003220 * Return the number of pages a zone spans in a node, including holes
3221 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3222 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003223static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003224 unsigned long zone_type,
3225 unsigned long *ignored)
3226{
3227 unsigned long node_start_pfn, node_end_pfn;
3228 unsigned long zone_start_pfn, zone_end_pfn;
3229
3230 /* Get the start and end of the node and zone */
3231 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3232 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3233 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07003234 adjust_zone_range_for_zone_movable(nid, zone_type,
3235 node_start_pfn, node_end_pfn,
3236 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003237
3238 /* Check that this node has pages within the zone's required range */
3239 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3240 return 0;
3241
3242 /* Move the zone boundaries inside the node if necessary */
3243 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3244 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3245
3246 /* Return the spanned pages */
3247 return zone_end_pfn - zone_start_pfn;
3248}
3249
3250/*
3251 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003252 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07003253 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003254static unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003255 unsigned long range_start_pfn,
3256 unsigned long range_end_pfn)
3257{
3258 int i = 0;
3259 unsigned long prev_end_pfn = 0, hole_pages = 0;
3260 unsigned long start_pfn;
3261
3262 /* Find the end_pfn of the first active range of pfns in the node */
3263 i = first_active_region_index_in_nid(nid);
3264 if (i == -1)
3265 return 0;
3266
Mel Gormanb5445f92007-07-26 10:41:18 -07003267 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3268
Mel Gorman9c7cd682006-09-27 01:49:58 -07003269 /* Account for ranges before physical memory on this node */
3270 if (early_node_map[i].start_pfn > range_start_pfn)
Mel Gormanb5445f92007-07-26 10:41:18 -07003271 hole_pages = prev_end_pfn - range_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003272
3273 /* Find all holes for the zone within the node */
3274 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3275
3276 /* No need to continue if prev_end_pfn is outside the zone */
3277 if (prev_end_pfn >= range_end_pfn)
3278 break;
3279
3280 /* Make sure the end of the zone is not within the hole */
3281 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3282 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3283
3284 /* Update the hole size cound and move on */
3285 if (start_pfn > range_start_pfn) {
3286 BUG_ON(prev_end_pfn > start_pfn);
3287 hole_pages += start_pfn - prev_end_pfn;
3288 }
3289 prev_end_pfn = early_node_map[i].end_pfn;
3290 }
3291
Mel Gorman9c7cd682006-09-27 01:49:58 -07003292 /* Account for ranges past physical memory on this node */
3293 if (range_end_pfn > prev_end_pfn)
Mel Gorman0c6cb972006-10-28 10:38:59 -07003294 hole_pages += range_end_pfn -
Mel Gorman9c7cd682006-09-27 01:49:58 -07003295 max(range_start_pfn, prev_end_pfn);
3296
Mel Gormanc7132162006-09-27 01:49:43 -07003297 return hole_pages;
3298}
3299
3300/**
3301 * absent_pages_in_range - Return number of page frames in holes within a range
3302 * @start_pfn: The start PFN to start searching for holes
3303 * @end_pfn: The end PFN to stop searching for holes
3304 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003305 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07003306 */
3307unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3308 unsigned long end_pfn)
3309{
3310 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3311}
3312
3313/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003314static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003315 unsigned long zone_type,
3316 unsigned long *ignored)
3317{
Mel Gorman9c7cd682006-09-27 01:49:58 -07003318 unsigned long node_start_pfn, node_end_pfn;
3319 unsigned long zone_start_pfn, zone_end_pfn;
3320
3321 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3322 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3323 node_start_pfn);
3324 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3325 node_end_pfn);
3326
Mel Gorman2a1e2742007-07-17 04:03:12 -07003327 adjust_zone_range_for_zone_movable(nid, zone_type,
3328 node_start_pfn, node_end_pfn,
3329 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07003330 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003331}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003332
Mel Gormanc7132162006-09-27 01:49:43 -07003333#else
Paul Mundt6ea6e682007-07-15 23:38:20 -07003334static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003335 unsigned long zone_type,
3336 unsigned long *zones_size)
3337{
3338 return zones_size[zone_type];
3339}
3340
Paul Mundt6ea6e682007-07-15 23:38:20 -07003341static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003342 unsigned long zone_type,
3343 unsigned long *zholes_size)
3344{
3345 if (!zholes_size)
3346 return 0;
3347
3348 return zholes_size[zone_type];
3349}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003350
Mel Gormanc7132162006-09-27 01:49:43 -07003351#endif
3352
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003353static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Mel Gormanc7132162006-09-27 01:49:43 -07003354 unsigned long *zones_size, unsigned long *zholes_size)
3355{
3356 unsigned long realtotalpages, totalpages = 0;
3357 enum zone_type i;
3358
3359 for (i = 0; i < MAX_NR_ZONES; i++)
3360 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3361 zones_size);
3362 pgdat->node_spanned_pages = totalpages;
3363
3364 realtotalpages = totalpages;
3365 for (i = 0; i < MAX_NR_ZONES; i++)
3366 realtotalpages -=
3367 zone_absent_pages_in_node(pgdat->node_id, i,
3368 zholes_size);
3369 pgdat->node_present_pages = realtotalpages;
3370 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3371 realtotalpages);
3372}
3373
Mel Gorman835c1342007-10-16 01:25:47 -07003374#ifndef CONFIG_SPARSEMEM
3375/*
3376 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07003377 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3378 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07003379 * round what is now in bits to nearest long in bits, then return it in
3380 * bytes.
3381 */
3382static unsigned long __init usemap_size(unsigned long zonesize)
3383{
3384 unsigned long usemapsize;
3385
Mel Gormand9c23402007-10-16 01:26:01 -07003386 usemapsize = roundup(zonesize, pageblock_nr_pages);
3387 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07003388 usemapsize *= NR_PAGEBLOCK_BITS;
3389 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3390
3391 return usemapsize / 8;
3392}
3393
3394static void __init setup_usemap(struct pglist_data *pgdat,
3395 struct zone *zone, unsigned long zonesize)
3396{
3397 unsigned long usemapsize = usemap_size(zonesize);
3398 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08003399 if (usemapsize)
Mel Gorman835c1342007-10-16 01:25:47 -07003400 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
Mel Gorman835c1342007-10-16 01:25:47 -07003401}
3402#else
3403static void inline setup_usemap(struct pglist_data *pgdat,
3404 struct zone *zone, unsigned long zonesize) {}
3405#endif /* CONFIG_SPARSEMEM */
3406
Mel Gormand9c23402007-10-16 01:26:01 -07003407#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08003408
3409/* Return a sensible default order for the pageblock size. */
3410static inline int pageblock_default_order(void)
3411{
3412 if (HPAGE_SHIFT > PAGE_SHIFT)
3413 return HUGETLB_PAGE_ORDER;
3414
3415 return MAX_ORDER-1;
3416}
3417
Mel Gormand9c23402007-10-16 01:26:01 -07003418/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3419static inline void __init set_pageblock_order(unsigned int order)
3420{
3421 /* Check that pageblock_nr_pages has not already been setup */
3422 if (pageblock_order)
3423 return;
3424
3425 /*
3426 * Assume the largest contiguous order of interest is a huge page.
3427 * This value may be variable depending on boot parameters on IA64
3428 */
3429 pageblock_order = order;
3430}
3431#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3432
Mel Gormanba72cb82007-11-28 16:21:13 -08003433/*
3434 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3435 * and pageblock_default_order() are unused as pageblock_order is set
3436 * at compile-time. See include/linux/pageblock-flags.h for the values of
3437 * pageblock_order based on the kernel config
3438 */
3439static inline int pageblock_default_order(unsigned int order)
3440{
3441 return MAX_ORDER-1;
3442}
Mel Gormand9c23402007-10-16 01:26:01 -07003443#define set_pageblock_order(x) do {} while (0)
3444
3445#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3446
Linus Torvalds1da177e2005-04-16 15:20:36 -07003447/*
3448 * Set up the zone data structures:
3449 * - mark all pages reserved
3450 * - mark all memory queues empty
3451 * - clear the memory bitmaps
3452 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08003453static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003454 unsigned long *zones_size, unsigned long *zholes_size)
3455{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07003456 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07003457 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003458 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07003459 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003460
Dave Hansen208d54e2005-10-29 18:16:52 -07003461 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003462 pgdat->nr_zones = 0;
3463 init_waitqueue_head(&pgdat->kswapd_wait);
3464 pgdat->kswapd_max_order = 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003465 pgdat_page_cgroup_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003466
3467 for (j = 0; j < MAX_NR_ZONES; j++) {
3468 struct zone *zone = pgdat->node_zones + j;
Mel Gorman0e0b8642006-09-27 01:49:56 -07003469 unsigned long size, realsize, memmap_pages;
Christoph Lameterb69408e2008-10-18 20:26:14 -07003470 enum lru_list l;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003471
Mel Gormanc7132162006-09-27 01:49:43 -07003472 size = zone_spanned_pages_in_node(nid, j, zones_size);
3473 realsize = size - zone_absent_pages_in_node(nid, j,
3474 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003475
Mel Gorman0e0b8642006-09-27 01:49:56 -07003476 /*
3477 * Adjust realsize so that it accounts for how much memory
3478 * is used by this zone for memmap. This affects the watermark
3479 * and per-cpu initialisations
3480 */
Johannes Weinerf7232152008-05-23 13:04:21 -07003481 memmap_pages =
3482 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
Mel Gorman0e0b8642006-09-27 01:49:56 -07003483 if (realsize >= memmap_pages) {
3484 realsize -= memmap_pages;
Yinghai Lu5594c8c2009-01-06 14:39:14 -08003485 if (memmap_pages)
3486 printk(KERN_DEBUG
3487 " %s zone: %lu pages used for memmap\n",
3488 zone_names[j], memmap_pages);
Mel Gorman0e0b8642006-09-27 01:49:56 -07003489 } else
3490 printk(KERN_WARNING
3491 " %s zone: %lu pages exceeds realsize %lu\n",
3492 zone_names[j], memmap_pages, realsize);
3493
Christoph Lameter62672762007-02-10 01:43:07 -08003494 /* Account for reserved pages */
3495 if (j == 0 && realsize > dma_reserve) {
Mel Gorman0e0b8642006-09-27 01:49:56 -07003496 realsize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07003497 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08003498 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07003499 }
3500
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003501 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003502 nr_kernel_pages += realsize;
3503 nr_all_pages += realsize;
3504
3505 zone->spanned_pages = size;
3506 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07003507#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07003508 zone->node = nid;
Christoph Lameter8417bba2006-09-25 23:31:51 -07003509 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07003510 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07003511 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07003512#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003513 zone->name = zone_names[j];
3514 spin_lock_init(&zone->lock);
3515 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07003516 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003517 zone->zone_pgdat = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003518
Martin Bligh3bb1a8522006-10-28 10:38:24 -07003519 zone->prev_priority = DEF_PRIORITY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003520
Dave Hansened8ece22005-10-29 18:16:50 -07003521 zone_pcp_init(zone);
Christoph Lameterb69408e2008-10-18 20:26:14 -07003522 for_each_lru(l) {
3523 INIT_LIST_HEAD(&zone->lru[l].list);
3524 zone->lru[l].nr_scan = 0;
3525 }
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08003526 zone->reclaim_stat.recent_rotated[0] = 0;
3527 zone->reclaim_stat.recent_rotated[1] = 0;
3528 zone->reclaim_stat.recent_scanned[0] = 0;
3529 zone->reclaim_stat.recent_scanned[1] = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07003530 zap_zone_vm_stats(zone);
David Rientjese815af92007-10-16 23:25:54 -07003531 zone->flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003532 if (!size)
3533 continue;
3534
Mel Gormanba72cb82007-11-28 16:21:13 -08003535 set_pageblock_order(pageblock_default_order());
Mel Gorman835c1342007-10-16 01:25:47 -07003536 setup_usemap(pgdat, zone, size);
Dave Hansena2f3aa022007-01-10 23:15:30 -08003537 ret = init_currently_empty_zone(zone, zone_start_pfn,
3538 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07003539 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07003540 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003541 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003542 }
3543}
3544
Sam Ravnborg577a32f2007-05-17 23:29:25 +02003545static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003546{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003547 /* Skip empty nodes */
3548 if (!pgdat->node_spanned_pages)
3549 return;
3550
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003551#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07003552 /* ia64 gets its own node_mem_map, before this, without bootmem */
3553 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07003554 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003555 struct page *map;
3556
Bob Piccoe984bb42006-05-20 15:00:31 -07003557 /*
3558 * The zone's endpoints aren't required to be MAX_ORDER
3559 * aligned but the node_mem_map endpoints must be in order
3560 * for the buddy allocator to function correctly.
3561 */
3562 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3563 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3564 end = ALIGN(end, MAX_ORDER_NR_PAGES);
3565 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07003566 map = alloc_remap(pgdat->node_id, size);
3567 if (!map)
3568 map = alloc_bootmem_node(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07003569 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003570 }
Roman Zippel12d810c2007-05-31 00:40:54 -07003571#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07003572 /*
3573 * With no DISCONTIG, the global mem_map is just set as node 0's
3574 */
Mel Gormanc7132162006-09-27 01:49:43 -07003575 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003576 mem_map = NODE_DATA(0)->node_mem_map;
Mel Gormanc7132162006-09-27 01:49:43 -07003577#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3578 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08003579 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Mel Gormanc7132162006-09-27 01:49:43 -07003580#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3581 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003582#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003583#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003584}
3585
Johannes Weiner9109fb72008-07-23 21:27:20 -07003586void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
3587 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003588{
Johannes Weiner9109fb72008-07-23 21:27:20 -07003589 pg_data_t *pgdat = NODE_DATA(nid);
3590
Linus Torvalds1da177e2005-04-16 15:20:36 -07003591 pgdat->node_id = nid;
3592 pgdat->node_start_pfn = node_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003593 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003594
3595 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07003596#ifdef CONFIG_FLAT_NODE_MEM_MAP
3597 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3598 nid, (unsigned long)pgdat,
3599 (unsigned long)pgdat->node_mem_map);
3600#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003601
3602 free_area_init_core(pgdat, zones_size, zholes_size);
3603}
3604
Mel Gormanc7132162006-09-27 01:49:43 -07003605#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07003606
3607#if MAX_NUMNODES > 1
3608/*
3609 * Figure out the number of possible node ids.
3610 */
3611static void __init setup_nr_node_ids(void)
3612{
3613 unsigned int node;
3614 unsigned int highest = 0;
3615
3616 for_each_node_mask(node, node_possible_map)
3617 highest = node;
3618 nr_node_ids = highest + 1;
3619}
3620#else
3621static inline void setup_nr_node_ids(void)
3622{
3623}
3624#endif
3625
Mel Gormanc7132162006-09-27 01:49:43 -07003626/**
3627 * add_active_range - Register a range of PFNs backed by physical memory
3628 * @nid: The node ID the range resides on
3629 * @start_pfn: The start PFN of the available physical memory
3630 * @end_pfn: The end PFN of the available physical memory
3631 *
3632 * These ranges are stored in an early_node_map[] and later used by
3633 * free_area_init_nodes() to calculate zone sizes and holes. If the
3634 * range spans a memory hole, it is up to the architecture to ensure
3635 * the memory is not freed by the bootmem allocator. If possible
3636 * the range being registered will be merged with existing ranges.
3637 */
3638void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3639 unsigned long end_pfn)
3640{
3641 int i;
3642
Mel Gorman6b74ab92008-07-23 21:26:49 -07003643 mminit_dprintk(MMINIT_TRACE, "memory_register",
3644 "Entering add_active_range(%d, %#lx, %#lx) "
3645 "%d entries of %d used\n",
3646 nid, start_pfn, end_pfn,
3647 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
Mel Gormanc7132162006-09-27 01:49:43 -07003648
Mel Gorman2dbb51c2008-07-23 21:26:52 -07003649 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
3650
Mel Gormanc7132162006-09-27 01:49:43 -07003651 /* Merge with existing active regions if possible */
3652 for (i = 0; i < nr_nodemap_entries; i++) {
3653 if (early_node_map[i].nid != nid)
3654 continue;
3655
3656 /* Skip if an existing region covers this new one */
3657 if (start_pfn >= early_node_map[i].start_pfn &&
3658 end_pfn <= early_node_map[i].end_pfn)
3659 return;
3660
3661 /* Merge forward if suitable */
3662 if (start_pfn <= early_node_map[i].end_pfn &&
3663 end_pfn > early_node_map[i].end_pfn) {
3664 early_node_map[i].end_pfn = end_pfn;
3665 return;
3666 }
3667
3668 /* Merge backward if suitable */
3669 if (start_pfn < early_node_map[i].end_pfn &&
3670 end_pfn >= early_node_map[i].start_pfn) {
3671 early_node_map[i].start_pfn = start_pfn;
3672 return;
3673 }
3674 }
3675
3676 /* Check that early_node_map is large enough */
3677 if (i >= MAX_ACTIVE_REGIONS) {
3678 printk(KERN_CRIT "More than %d memory regions, truncating\n",
3679 MAX_ACTIVE_REGIONS);
3680 return;
3681 }
3682
3683 early_node_map[i].nid = nid;
3684 early_node_map[i].start_pfn = start_pfn;
3685 early_node_map[i].end_pfn = end_pfn;
3686 nr_nodemap_entries = i + 1;
3687}
3688
3689/**
Yinghai Lucc1050b2008-06-13 19:08:52 -07003690 * remove_active_range - Shrink an existing registered range of PFNs
Mel Gormanc7132162006-09-27 01:49:43 -07003691 * @nid: The node id the range is on that should be shrunk
Yinghai Lucc1050b2008-06-13 19:08:52 -07003692 * @start_pfn: The new PFN of the range
3693 * @end_pfn: The new PFN of the range
Mel Gormanc7132162006-09-27 01:49:43 -07003694 *
3695 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003696 * The map is kept near the end physical page range that has already been
3697 * registered. This function allows an arch to shrink an existing registered
3698 * range.
Mel Gormanc7132162006-09-27 01:49:43 -07003699 */
Yinghai Lucc1050b2008-06-13 19:08:52 -07003700void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
3701 unsigned long end_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07003702{
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003703 int i, j;
3704 int removed = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07003705
Yinghai Lucc1050b2008-06-13 19:08:52 -07003706 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
3707 nid, start_pfn, end_pfn);
3708
Mel Gormanc7132162006-09-27 01:49:43 -07003709 /* Find the old active region end and shrink */
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003710 for_each_active_range_index_in_nid(i, nid) {
Yinghai Lucc1050b2008-06-13 19:08:52 -07003711 if (early_node_map[i].start_pfn >= start_pfn &&
3712 early_node_map[i].end_pfn <= end_pfn) {
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003713 /* clear it */
Yinghai Lucc1050b2008-06-13 19:08:52 -07003714 early_node_map[i].start_pfn = 0;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003715 early_node_map[i].end_pfn = 0;
3716 removed = 1;
3717 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07003718 }
Yinghai Lucc1050b2008-06-13 19:08:52 -07003719 if (early_node_map[i].start_pfn < start_pfn &&
3720 early_node_map[i].end_pfn > start_pfn) {
3721 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
3722 early_node_map[i].end_pfn = start_pfn;
3723 if (temp_end_pfn > end_pfn)
3724 add_active_range(nid, end_pfn, temp_end_pfn);
3725 continue;
3726 }
3727 if (early_node_map[i].start_pfn >= start_pfn &&
3728 early_node_map[i].end_pfn > end_pfn &&
3729 early_node_map[i].start_pfn < end_pfn) {
3730 early_node_map[i].start_pfn = end_pfn;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003731 continue;
3732 }
3733 }
3734
3735 if (!removed)
3736 return;
3737
3738 /* remove the blank ones */
3739 for (i = nr_nodemap_entries - 1; i > 0; i--) {
3740 if (early_node_map[i].nid != nid)
3741 continue;
3742 if (early_node_map[i].end_pfn)
3743 continue;
3744 /* we found it, get rid of it */
3745 for (j = i; j < nr_nodemap_entries - 1; j++)
3746 memcpy(&early_node_map[j], &early_node_map[j+1],
3747 sizeof(early_node_map[j]));
3748 j = nr_nodemap_entries - 1;
3749 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
3750 nr_nodemap_entries--;
3751 }
Mel Gormanc7132162006-09-27 01:49:43 -07003752}
3753
3754/**
3755 * remove_all_active_ranges - Remove all currently registered regions
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003756 *
Mel Gormanc7132162006-09-27 01:49:43 -07003757 * During discovery, it may be found that a table like SRAT is invalid
3758 * and an alternative discovery method must be used. This function removes
3759 * all currently registered regions.
3760 */
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003761void __init remove_all_active_ranges(void)
Mel Gormanc7132162006-09-27 01:49:43 -07003762{
3763 memset(early_node_map, 0, sizeof(early_node_map));
3764 nr_nodemap_entries = 0;
Mel Gormanfb014392006-09-27 01:49:59 -07003765#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3766 memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
3767 memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
3768#endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
Mel Gormanc7132162006-09-27 01:49:43 -07003769}
3770
3771/* Compare two active node_active_regions */
3772static int __init cmp_node_active_region(const void *a, const void *b)
3773{
3774 struct node_active_region *arange = (struct node_active_region *)a;
3775 struct node_active_region *brange = (struct node_active_region *)b;
3776
3777 /* Done this way to avoid overflows */
3778 if (arange->start_pfn > brange->start_pfn)
3779 return 1;
3780 if (arange->start_pfn < brange->start_pfn)
3781 return -1;
3782
3783 return 0;
3784}
3785
3786/* sort the node_map by start_pfn */
3787static void __init sort_node_map(void)
3788{
3789 sort(early_node_map, (size_t)nr_nodemap_entries,
3790 sizeof(struct node_active_region),
3791 cmp_node_active_region, NULL);
3792}
3793
Mel Gormana6af2bc2007-02-10 01:42:57 -08003794/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003795static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003796{
3797 int i;
Mel Gormana6af2bc2007-02-10 01:42:57 -08003798 unsigned long min_pfn = ULONG_MAX;
Mel Gorman1abbfb42006-11-23 12:01:41 +00003799
Mel Gormanc7132162006-09-27 01:49:43 -07003800 /* Assuming a sorted map, the first range found has the starting pfn */
3801 for_each_active_range_index_in_nid(i, nid)
Mel Gormana6af2bc2007-02-10 01:42:57 -08003802 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003803
Mel Gormana6af2bc2007-02-10 01:42:57 -08003804 if (min_pfn == ULONG_MAX) {
3805 printk(KERN_WARNING
Paul Jackson2bc0d262008-06-22 07:22:17 -07003806 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08003807 return 0;
3808 }
3809
3810 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003811}
3812
3813/**
3814 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3815 *
3816 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003817 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07003818 */
3819unsigned long __init find_min_pfn_with_active_regions(void)
3820{
3821 return find_min_pfn_for_node(MAX_NUMNODES);
3822}
3823
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003824/*
3825 * early_calculate_totalpages()
3826 * Sum pages in active regions for movable zone.
3827 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3828 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07003829static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07003830{
3831 int i;
3832 unsigned long totalpages = 0;
3833
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003834 for (i = 0; i < nr_nodemap_entries; i++) {
3835 unsigned long pages = early_node_map[i].end_pfn -
Mel Gorman7e63efe2007-07-17 04:03:15 -07003836 early_node_map[i].start_pfn;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003837 totalpages += pages;
3838 if (pages)
3839 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
3840 }
3841 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07003842}
3843
Mel Gorman2a1e2742007-07-17 04:03:12 -07003844/*
3845 * Find the PFN the Movable zone begins in each node. Kernel memory
3846 * is spread evenly between nodes as long as the nodes have enough
3847 * memory. When they don't, some nodes will have more kernelcore than
3848 * others
3849 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003850static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
Mel Gorman2a1e2742007-07-17 04:03:12 -07003851{
3852 int i, nid;
3853 unsigned long usable_startpfn;
3854 unsigned long kernelcore_node, kernelcore_remaining;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003855 unsigned long totalpages = early_calculate_totalpages();
3856 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07003857
Mel Gorman7e63efe2007-07-17 04:03:15 -07003858 /*
3859 * If movablecore was specified, calculate what size of
3860 * kernelcore that corresponds so that memory usable for
3861 * any allocation type is evenly spread. If both kernelcore
3862 * and movablecore are specified, then the value of kernelcore
3863 * will be used for required_kernelcore if it's greater than
3864 * what movablecore would have allowed.
3865 */
3866 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07003867 unsigned long corepages;
3868
3869 /*
3870 * Round-up so that ZONE_MOVABLE is at least as large as what
3871 * was requested by the user
3872 */
3873 required_movablecore =
3874 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
3875 corepages = totalpages - required_movablecore;
3876
3877 required_kernelcore = max(required_kernelcore, corepages);
3878 }
3879
Mel Gorman2a1e2742007-07-17 04:03:12 -07003880 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3881 if (!required_kernelcore)
3882 return;
3883
3884 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3885 find_usable_zone_for_movable();
3886 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
3887
3888restart:
3889 /* Spread kernelcore memory as evenly as possible throughout nodes */
3890 kernelcore_node = required_kernelcore / usable_nodes;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003891 for_each_node_state(nid, N_HIGH_MEMORY) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07003892 /*
3893 * Recalculate kernelcore_node if the division per node
3894 * now exceeds what is necessary to satisfy the requested
3895 * amount of memory for the kernel
3896 */
3897 if (required_kernelcore < kernelcore_node)
3898 kernelcore_node = required_kernelcore / usable_nodes;
3899
3900 /*
3901 * As the map is walked, we track how much memory is usable
3902 * by the kernel using kernelcore_remaining. When it is
3903 * 0, the rest of the node is usable by ZONE_MOVABLE
3904 */
3905 kernelcore_remaining = kernelcore_node;
3906
3907 /* Go through each range of PFNs within this node */
3908 for_each_active_range_index_in_nid(i, nid) {
3909 unsigned long start_pfn, end_pfn;
3910 unsigned long size_pages;
3911
3912 start_pfn = max(early_node_map[i].start_pfn,
3913 zone_movable_pfn[nid]);
3914 end_pfn = early_node_map[i].end_pfn;
3915 if (start_pfn >= end_pfn)
3916 continue;
3917
3918 /* Account for what is only usable for kernelcore */
3919 if (start_pfn < usable_startpfn) {
3920 unsigned long kernel_pages;
3921 kernel_pages = min(end_pfn, usable_startpfn)
3922 - start_pfn;
3923
3924 kernelcore_remaining -= min(kernel_pages,
3925 kernelcore_remaining);
3926 required_kernelcore -= min(kernel_pages,
3927 required_kernelcore);
3928
3929 /* Continue if range is now fully accounted */
3930 if (end_pfn <= usable_startpfn) {
3931
3932 /*
3933 * Push zone_movable_pfn to the end so
3934 * that if we have to rebalance
3935 * kernelcore across nodes, we will
3936 * not double account here
3937 */
3938 zone_movable_pfn[nid] = end_pfn;
3939 continue;
3940 }
3941 start_pfn = usable_startpfn;
3942 }
3943
3944 /*
3945 * The usable PFN range for ZONE_MOVABLE is from
3946 * start_pfn->end_pfn. Calculate size_pages as the
3947 * number of pages used as kernelcore
3948 */
3949 size_pages = end_pfn - start_pfn;
3950 if (size_pages > kernelcore_remaining)
3951 size_pages = kernelcore_remaining;
3952 zone_movable_pfn[nid] = start_pfn + size_pages;
3953
3954 /*
3955 * Some kernelcore has been met, update counts and
3956 * break if the kernelcore for this node has been
3957 * satisified
3958 */
3959 required_kernelcore -= min(required_kernelcore,
3960 size_pages);
3961 kernelcore_remaining -= size_pages;
3962 if (!kernelcore_remaining)
3963 break;
3964 }
3965 }
3966
3967 /*
3968 * If there is still required_kernelcore, we do another pass with one
3969 * less node in the count. This will push zone_movable_pfn[nid] further
3970 * along on the nodes that still have memory until kernelcore is
3971 * satisified
3972 */
3973 usable_nodes--;
3974 if (usable_nodes && required_kernelcore > usable_nodes)
3975 goto restart;
3976
3977 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3978 for (nid = 0; nid < MAX_NUMNODES; nid++)
3979 zone_movable_pfn[nid] =
3980 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
3981}
3982
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003983/* Any regular memory on that node ? */
3984static void check_for_regular_memory(pg_data_t *pgdat)
3985{
3986#ifdef CONFIG_HIGHMEM
3987 enum zone_type zone_type;
3988
3989 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
3990 struct zone *zone = &pgdat->node_zones[zone_type];
3991 if (zone->present_pages)
3992 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
3993 }
3994#endif
3995}
3996
Mel Gormanc7132162006-09-27 01:49:43 -07003997/**
3998 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003999 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07004000 *
4001 * This will call free_area_init_node() for each active node in the system.
4002 * Using the page ranges provided by add_active_range(), the size of each
4003 * zone in each node and their holes is calculated. If the maximum PFN
4004 * between two adjacent zones match, it is assumed that the zone is empty.
4005 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4006 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4007 * starts where the previous one ended. For example, ZONE_DMA32 starts
4008 * at arch_max_dma_pfn.
4009 */
4010void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4011{
4012 unsigned long nid;
Andrew Mortondb991002008-10-15 22:01:07 -07004013 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004014
Mel Gormana6af2bc2007-02-10 01:42:57 -08004015 /* Sort early_node_map as initialisation assumes it is sorted */
4016 sort_node_map();
4017
Mel Gormanc7132162006-09-27 01:49:43 -07004018 /* Record where the zone boundaries are */
4019 memset(arch_zone_lowest_possible_pfn, 0,
4020 sizeof(arch_zone_lowest_possible_pfn));
4021 memset(arch_zone_highest_possible_pfn, 0,
4022 sizeof(arch_zone_highest_possible_pfn));
4023 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4024 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4025 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004026 if (i == ZONE_MOVABLE)
4027 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004028 arch_zone_lowest_possible_pfn[i] =
4029 arch_zone_highest_possible_pfn[i-1];
4030 arch_zone_highest_possible_pfn[i] =
4031 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4032 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07004033 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4034 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4035
4036 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4037 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4038 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004039
Mel Gormanc7132162006-09-27 01:49:43 -07004040 /* Print out the zone ranges */
4041 printk("Zone PFN ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07004042 for (i = 0; i < MAX_NR_ZONES; i++) {
4043 if (i == ZONE_MOVABLE)
4044 continue;
Paul Jackson5dab8ec2008-06-25 05:44:40 -07004045 printk(" %-8s %0#10lx -> %0#10lx\n",
Mel Gormanc7132162006-09-27 01:49:43 -07004046 zone_names[i],
4047 arch_zone_lowest_possible_pfn[i],
4048 arch_zone_highest_possible_pfn[i]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004049 }
4050
4051 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4052 printk("Movable zone start PFN for each node\n");
4053 for (i = 0; i < MAX_NUMNODES; i++) {
4054 if (zone_movable_pfn[i])
4055 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4056 }
Mel Gormanc7132162006-09-27 01:49:43 -07004057
4058 /* Print out the early_node_map[] */
4059 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4060 for (i = 0; i < nr_nodemap_entries; i++)
Paul Jackson5dab8ec2008-06-25 05:44:40 -07004061 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004062 early_node_map[i].start_pfn,
4063 early_node_map[i].end_pfn);
4064
4065 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07004066 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08004067 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07004068 for_each_online_node(nid) {
4069 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07004070 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07004071 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004072
4073 /* Any memory on that node */
4074 if (pgdat->node_present_pages)
4075 node_set_state(nid, N_HIGH_MEMORY);
4076 check_for_regular_memory(pgdat);
Mel Gormanc7132162006-09-27 01:49:43 -07004077 }
4078}
Mel Gorman2a1e2742007-07-17 04:03:12 -07004079
Mel Gorman7e63efe2007-07-17 04:03:15 -07004080static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004081{
4082 unsigned long long coremem;
4083 if (!p)
4084 return -EINVAL;
4085
4086 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004087 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004088
Mel Gorman7e63efe2007-07-17 04:03:15 -07004089 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07004090 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4091
4092 return 0;
4093}
Mel Gormaned7ed362007-07-17 04:03:14 -07004094
Mel Gorman7e63efe2007-07-17 04:03:15 -07004095/*
4096 * kernelcore=size sets the amount of memory for use for allocations that
4097 * cannot be reclaimed or migrated.
4098 */
4099static int __init cmdline_parse_kernelcore(char *p)
4100{
4101 return cmdline_parse_core(p, &required_kernelcore);
4102}
4103
4104/*
4105 * movablecore=size sets the amount of memory for use for allocations that
4106 * can be reclaimed or migrated.
4107 */
4108static int __init cmdline_parse_movablecore(char *p)
4109{
4110 return cmdline_parse_core(p, &required_movablecore);
4111}
4112
Mel Gormaned7ed362007-07-17 04:03:14 -07004113early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004114early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07004115
Mel Gormanc7132162006-09-27 01:49:43 -07004116#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4117
Mel Gorman0e0b8642006-09-27 01:49:56 -07004118/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004119 * set_dma_reserve - set the specified number of pages reserved in the first zone
4120 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07004121 *
4122 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4123 * In the DMA zone, a significant percentage may be consumed by kernel image
4124 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004125 * function may optionally be used to account for unfreeable pages in the
4126 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4127 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07004128 */
4129void __init set_dma_reserve(unsigned long new_dma_reserve)
4130{
4131 dma_reserve = new_dma_reserve;
4132}
4133
Dave Hansen93b75042005-06-23 00:07:47 -07004134#ifndef CONFIG_NEED_MULTIPLE_NODES
Marcin Slusarz52765582008-09-02 14:35:41 -07004135struct pglist_data __refdata contig_page_data = { .bdata = &bootmem_node_data[0] };
Linus Torvalds1da177e2005-04-16 15:20:36 -07004136EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07004137#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004138
4139void __init free_area_init(unsigned long *zones_size)
4140{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004141 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004142 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4143}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004144
Linus Torvalds1da177e2005-04-16 15:20:36 -07004145static int page_alloc_cpu_notify(struct notifier_block *self,
4146 unsigned long action, void *hcpu)
4147{
4148 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004149
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07004150 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004151 drain_pages(cpu);
4152
4153 /*
4154 * Spill the event counters of the dead processor
4155 * into the current processors event counters.
4156 * This artificially elevates the count of the current
4157 * processor.
4158 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07004159 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004160
4161 /*
4162 * Zero the differential counters of the dead processor
4163 * so that the vm statistics are consistent.
4164 *
4165 * This is only okay since the processor is dead and cannot
4166 * race with what we are doing.
4167 */
Christoph Lameter2244b952006-06-30 01:55:33 -07004168 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004169 }
4170 return NOTIFY_OK;
4171}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004172
4173void __init page_alloc_init(void)
4174{
4175 hotcpu_notifier(page_alloc_cpu_notify, 0);
4176}
4177
4178/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004179 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4180 * or min_free_kbytes changes.
4181 */
4182static void calculate_totalreserve_pages(void)
4183{
4184 struct pglist_data *pgdat;
4185 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004186 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004187
4188 for_each_online_pgdat(pgdat) {
4189 for (i = 0; i < MAX_NR_ZONES; i++) {
4190 struct zone *zone = pgdat->node_zones + i;
4191 unsigned long max = 0;
4192
4193 /* Find valid and maximum lowmem_reserve in the zone */
4194 for (j = i; j < MAX_NR_ZONES; j++) {
4195 if (zone->lowmem_reserve[j] > max)
4196 max = zone->lowmem_reserve[j];
4197 }
4198
4199 /* we treat pages_high as reserved pages. */
4200 max += zone->pages_high;
4201
4202 if (max > zone->present_pages)
4203 max = zone->present_pages;
4204 reserve_pages += max;
4205 }
4206 }
4207 totalreserve_pages = reserve_pages;
4208}
4209
4210/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07004211 * setup_per_zone_lowmem_reserve - called whenever
4212 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4213 * has a correct pages reserved value, so an adequate number of
4214 * pages are left in the zone after a successful __alloc_pages().
4215 */
4216static void setup_per_zone_lowmem_reserve(void)
4217{
4218 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004219 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004220
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08004221 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004222 for (j = 0; j < MAX_NR_ZONES; j++) {
4223 struct zone *zone = pgdat->node_zones + j;
4224 unsigned long present_pages = zone->present_pages;
4225
4226 zone->lowmem_reserve[j] = 0;
4227
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004228 idx = j;
4229 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004230 struct zone *lower_zone;
4231
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004232 idx--;
4233
Linus Torvalds1da177e2005-04-16 15:20:36 -07004234 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4235 sysctl_lowmem_reserve_ratio[idx] = 1;
4236
4237 lower_zone = pgdat->node_zones + idx;
4238 lower_zone->lowmem_reserve[j] = present_pages /
4239 sysctl_lowmem_reserve_ratio[idx];
4240 present_pages += lower_zone->present_pages;
4241 }
4242 }
4243 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004244
4245 /* update totalreserve_pages */
4246 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004247}
4248
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004249/**
4250 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4251 *
4252 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4253 * with respect to min_free_kbytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004254 */
Dave Hansen3947be12005-10-29 18:16:54 -07004255void setup_per_zone_pages_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004256{
4257 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4258 unsigned long lowmem_pages = 0;
4259 struct zone *zone;
4260 unsigned long flags;
4261
4262 /* Calculate total number of !ZONE_HIGHMEM pages */
4263 for_each_zone(zone) {
4264 if (!is_highmem(zone))
4265 lowmem_pages += zone->present_pages;
4266 }
4267
4268 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07004269 u64 tmp;
4270
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07004271 spin_lock_irqsave(&zone->lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07004272 tmp = (u64)pages_min * zone->present_pages;
4273 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004274 if (is_highmem(zone)) {
4275 /*
Nick Piggin669ed172005-11-13 16:06:45 -08004276 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4277 * need highmem pages, so cap pages_min to a small
4278 * value here.
4279 *
4280 * The (pages_high-pages_low) and (pages_low-pages_min)
4281 * deltas controls asynch page reclaim, and so should
4282 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004283 */
4284 int min_pages;
4285
4286 min_pages = zone->present_pages / 1024;
4287 if (min_pages < SWAP_CLUSTER_MAX)
4288 min_pages = SWAP_CLUSTER_MAX;
4289 if (min_pages > 128)
4290 min_pages = 128;
4291 zone->pages_min = min_pages;
4292 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08004293 /*
4294 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07004295 * proportionate to the zone's size.
4296 */
Nick Piggin669ed172005-11-13 16:06:45 -08004297 zone->pages_min = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004298 }
4299
Andrew Mortonac924c62006-05-15 09:43:59 -07004300 zone->pages_low = zone->pages_min + (tmp >> 2);
4301 zone->pages_high = zone->pages_min + (tmp >> 1);
Mel Gorman56fd56b2007-10-16 01:25:58 -07004302 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07004303 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004304 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004305
4306 /* update totalreserve_pages */
4307 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004308}
4309
Rik van Riel556adec2008-10-18 20:26:34 -07004310/**
4311 * setup_per_zone_inactive_ratio - called when min_free_kbytes changes.
4312 *
4313 * The inactive anon list should be small enough that the VM never has to
4314 * do too much work, but large enough that each inactive page has a chance
4315 * to be referenced again before it is swapped out.
4316 *
4317 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4318 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4319 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4320 * the anonymous pages are kept on the inactive list.
4321 *
4322 * total target max
4323 * memory ratio inactive anon
4324 * -------------------------------------
4325 * 10MB 1 5MB
4326 * 100MB 1 50MB
4327 * 1GB 3 250MB
4328 * 10GB 10 0.9GB
4329 * 100GB 31 3GB
4330 * 1TB 101 10GB
4331 * 10TB 320 32GB
4332 */
KOSAKI Motohiroefab8182009-01-06 14:39:46 -08004333static void setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07004334{
4335 struct zone *zone;
4336
4337 for_each_zone(zone) {
4338 unsigned int gb, ratio;
4339
4340 /* Zone size in gigabytes */
4341 gb = zone->present_pages >> (30 - PAGE_SHIFT);
4342 ratio = int_sqrt(10 * gb);
4343 if (!ratio)
4344 ratio = 1;
4345
4346 zone->inactive_ratio = ratio;
4347 }
4348}
4349
Linus Torvalds1da177e2005-04-16 15:20:36 -07004350/*
4351 * Initialise min_free_kbytes.
4352 *
4353 * For small machines we want it small (128k min). For large machines
4354 * we want it large (64MB max). But it is not linear, because network
4355 * bandwidth does not increase linearly with machine size. We use
4356 *
4357 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4358 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4359 *
4360 * which yields
4361 *
4362 * 16MB: 512k
4363 * 32MB: 724k
4364 * 64MB: 1024k
4365 * 128MB: 1448k
4366 * 256MB: 2048k
4367 * 512MB: 2896k
4368 * 1024MB: 4096k
4369 * 2048MB: 5792k
4370 * 4096MB: 8192k
4371 * 8192MB: 11584k
4372 * 16384MB: 16384k
4373 */
4374static int __init init_per_zone_pages_min(void)
4375{
4376 unsigned long lowmem_kbytes;
4377
4378 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4379
4380 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4381 if (min_free_kbytes < 128)
4382 min_free_kbytes = 128;
4383 if (min_free_kbytes > 65536)
4384 min_free_kbytes = 65536;
4385 setup_per_zone_pages_min();
4386 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07004387 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004388 return 0;
4389}
4390module_init(init_per_zone_pages_min)
4391
4392/*
4393 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4394 * that we can call two helper functions whenever min_free_kbytes
4395 * changes.
4396 */
4397int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
4398 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4399{
4400 proc_dointvec(table, write, file, buffer, length, ppos);
Mel Gorman3b1d92c2007-05-06 14:49:30 -07004401 if (write)
4402 setup_per_zone_pages_min();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004403 return 0;
4404}
4405
Christoph Lameter96146342006-07-03 00:24:13 -07004406#ifdef CONFIG_NUMA
4407int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
4408 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4409{
4410 struct zone *zone;
4411 int rc;
4412
4413 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4414 if (rc)
4415 return rc;
4416
4417 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07004418 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07004419 sysctl_min_unmapped_ratio) / 100;
4420 return 0;
4421}
Christoph Lameter0ff38492006-09-25 23:31:52 -07004422
4423int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
4424 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4425{
4426 struct zone *zone;
4427 int rc;
4428
4429 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4430 if (rc)
4431 return rc;
4432
4433 for_each_zone(zone)
4434 zone->min_slab_pages = (zone->present_pages *
4435 sysctl_min_slab_ratio) / 100;
4436 return 0;
4437}
Christoph Lameter96146342006-07-03 00:24:13 -07004438#endif
4439
Linus Torvalds1da177e2005-04-16 15:20:36 -07004440/*
4441 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4442 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4443 * whenever sysctl_lowmem_reserve_ratio changes.
4444 *
4445 * The reserve ratio obviously has absolutely no relation with the
4446 * pages_min watermarks. The lowmem reserve ratio can only make sense
4447 * if in function of the boot time zone sizes.
4448 */
4449int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
4450 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4451{
4452 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4453 setup_per_zone_lowmem_reserve();
4454 return 0;
4455}
4456
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004457/*
4458 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4459 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4460 * can have before it gets flushed back to buddy allocator.
4461 */
4462
4463int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
4464 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4465{
4466 struct zone *zone;
4467 unsigned int cpu;
4468 int ret;
4469
4470 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4471 if (!write || (ret == -EINVAL))
4472 return ret;
4473 for_each_zone(zone) {
4474 for_each_online_cpu(cpu) {
4475 unsigned long high;
4476 high = zone->present_pages / percpu_pagelist_fraction;
4477 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4478 }
4479 }
4480 return 0;
4481}
4482
David S. Millerf034b5d2006-08-24 03:08:07 -07004483int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004484
4485#ifdef CONFIG_NUMA
4486static int __init set_hashdist(char *str)
4487{
4488 if (!str)
4489 return 0;
4490 hashdist = simple_strtoul(str, &str, 0);
4491 return 1;
4492}
4493__setup("hashdist=", set_hashdist);
4494#endif
4495
4496/*
4497 * allocate a large system hash table from bootmem
4498 * - it is assumed that the hash table must contain an exact power-of-2
4499 * quantity of entries
4500 * - limit is the number of hash buckets, not the total allocation size
4501 */
4502void *__init alloc_large_system_hash(const char *tablename,
4503 unsigned long bucketsize,
4504 unsigned long numentries,
4505 int scale,
4506 int flags,
4507 unsigned int *_hash_shift,
4508 unsigned int *_hash_mask,
4509 unsigned long limit)
4510{
4511 unsigned long long max = limit;
4512 unsigned long log2qty, size;
4513 void *table = NULL;
4514
4515 /* allow the kernel cmdline to have a say */
4516 if (!numentries) {
4517 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08004518 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004519 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4520 numentries >>= 20 - PAGE_SHIFT;
4521 numentries <<= 20 - PAGE_SHIFT;
4522
4523 /* limit to 1 bucket per 2^scale bytes of low memory */
4524 if (scale > PAGE_SHIFT)
4525 numentries >>= (scale - PAGE_SHIFT);
4526 else
4527 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08004528
4529 /* Make sure we've got at least a 0-order allocation.. */
4530 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
4531 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004532 }
John Hawkes6e692ed2006-03-25 03:08:02 -08004533 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004534
4535 /* limit allocation size to 1/16 total memory by default */
4536 if (max == 0) {
4537 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4538 do_div(max, bucketsize);
4539 }
4540
4541 if (numentries > max)
4542 numentries = max;
4543
David Howellsf0d1b0b2006-12-08 02:37:49 -08004544 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004545
4546 do {
4547 size = bucketsize << log2qty;
4548 if (flags & HASH_EARLY)
Jan Beulich74768ed2008-08-12 15:08:39 -07004549 table = alloc_bootmem_nopanic(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004550 else if (hashdist)
4551 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4552 else {
Pavel Machek2309f9e2008-04-28 02:13:35 -07004553 unsigned long order = get_order(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004554 table = (void*) __get_free_pages(GFP_ATOMIC, order);
Eric Dumazet1037b832007-07-15 23:38:05 -07004555 /*
4556 * If bucketsize is not a power-of-two, we may free
4557 * some pages at the end of hash table.
4558 */
4559 if (table) {
4560 unsigned long alloc_end = (unsigned long)table +
4561 (PAGE_SIZE << order);
4562 unsigned long used = (unsigned long)table +
4563 PAGE_ALIGN(size);
4564 split_page(virt_to_page(table), order);
4565 while (used < alloc_end) {
4566 free_page(used);
4567 used += PAGE_SIZE;
4568 }
4569 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004570 }
4571 } while (!table && size > PAGE_SIZE && --log2qty);
4572
4573 if (!table)
4574 panic("Failed to allocate %s hash table\n", tablename);
4575
Dan Alonib49ad482007-07-15 23:38:23 -07004576 printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07004577 tablename,
4578 (1U << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08004579 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004580 size);
4581
4582 if (_hash_shift)
4583 *_hash_shift = log2qty;
4584 if (_hash_mask)
4585 *_hash_mask = (1 << log2qty) - 1;
4586
4587 return table;
4588}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004589
Mel Gorman835c1342007-10-16 01:25:47 -07004590/* Return a pointer to the bitmap storing bits affecting a block of pages */
4591static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4592 unsigned long pfn)
4593{
4594#ifdef CONFIG_SPARSEMEM
4595 return __pfn_to_section(pfn)->pageblock_flags;
4596#else
4597 return zone->pageblock_flags;
4598#endif /* CONFIG_SPARSEMEM */
4599}
Andrew Morton6220ec72006-10-19 23:29:05 -07004600
Mel Gorman835c1342007-10-16 01:25:47 -07004601static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4602{
4603#ifdef CONFIG_SPARSEMEM
4604 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07004605 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07004606#else
4607 pfn = pfn - zone->zone_start_pfn;
Mel Gormand9c23402007-10-16 01:26:01 -07004608 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07004609#endif /* CONFIG_SPARSEMEM */
4610}
4611
4612/**
Mel Gormand9c23402007-10-16 01:26:01 -07004613 * 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 -07004614 * @page: The page within the block of interest
4615 * @start_bitidx: The first bit of interest to retrieve
4616 * @end_bitidx: The last bit of interest
4617 * returns pageblock_bits flags
4618 */
4619unsigned long get_pageblock_flags_group(struct page *page,
4620 int start_bitidx, int end_bitidx)
4621{
4622 struct zone *zone;
4623 unsigned long *bitmap;
4624 unsigned long pfn, bitidx;
4625 unsigned long flags = 0;
4626 unsigned long value = 1;
4627
4628 zone = page_zone(page);
4629 pfn = page_to_pfn(page);
4630 bitmap = get_pageblock_bitmap(zone, pfn);
4631 bitidx = pfn_to_bitidx(zone, pfn);
4632
4633 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4634 if (test_bit(bitidx + start_bitidx, bitmap))
4635 flags |= value;
4636
4637 return flags;
4638}
4639
4640/**
Mel Gormand9c23402007-10-16 01:26:01 -07004641 * 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 -07004642 * @page: The page within the block of interest
4643 * @start_bitidx: The first bit of interest
4644 * @end_bitidx: The last bit of interest
4645 * @flags: The flags to set
4646 */
4647void set_pageblock_flags_group(struct page *page, unsigned long flags,
4648 int start_bitidx, int end_bitidx)
4649{
4650 struct zone *zone;
4651 unsigned long *bitmap;
4652 unsigned long pfn, bitidx;
4653 unsigned long value = 1;
4654
4655 zone = page_zone(page);
4656 pfn = page_to_pfn(page);
4657 bitmap = get_pageblock_bitmap(zone, pfn);
4658 bitidx = pfn_to_bitidx(zone, pfn);
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004659 VM_BUG_ON(pfn < zone->zone_start_pfn);
4660 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
Mel Gorman835c1342007-10-16 01:25:47 -07004661
4662 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4663 if (flags & value)
4664 __set_bit(bitidx + start_bitidx, bitmap);
4665 else
4666 __clear_bit(bitidx + start_bitidx, bitmap);
4667}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07004668
4669/*
4670 * This is designed as sub function...plz see page_isolation.c also.
4671 * set/clear page block's type to be ISOLATE.
4672 * page allocater never alloc memory from ISOLATE block.
4673 */
4674
4675int set_migratetype_isolate(struct page *page)
4676{
4677 struct zone *zone;
4678 unsigned long flags;
4679 int ret = -EBUSY;
4680
4681 zone = page_zone(page);
4682 spin_lock_irqsave(&zone->lock, flags);
4683 /*
4684 * In future, more migrate types will be able to be isolation target.
4685 */
4686 if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE)
4687 goto out;
4688 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
4689 move_freepages_block(zone, page, MIGRATE_ISOLATE);
4690 ret = 0;
4691out:
4692 spin_unlock_irqrestore(&zone->lock, flags);
4693 if (!ret)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004694 drain_all_pages();
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07004695 return ret;
4696}
4697
4698void unset_migratetype_isolate(struct page *page)
4699{
4700 struct zone *zone;
4701 unsigned long flags;
4702 zone = page_zone(page);
4703 spin_lock_irqsave(&zone->lock, flags);
4704 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
4705 goto out;
4706 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4707 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4708out:
4709 spin_unlock_irqrestore(&zone->lock, flags);
4710}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07004711
4712#ifdef CONFIG_MEMORY_HOTREMOVE
4713/*
4714 * All pages in the range must be isolated before calling this.
4715 */
4716void
4717__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
4718{
4719 struct page *page;
4720 struct zone *zone;
4721 int order, i;
4722 unsigned long pfn;
4723 unsigned long flags;
4724 /* find the first valid pfn */
4725 for (pfn = start_pfn; pfn < end_pfn; pfn++)
4726 if (pfn_valid(pfn))
4727 break;
4728 if (pfn == end_pfn)
4729 return;
4730 zone = page_zone(pfn_to_page(pfn));
4731 spin_lock_irqsave(&zone->lock, flags);
4732 pfn = start_pfn;
4733 while (pfn < end_pfn) {
4734 if (!pfn_valid(pfn)) {
4735 pfn++;
4736 continue;
4737 }
4738 page = pfn_to_page(pfn);
4739 BUG_ON(page_count(page));
4740 BUG_ON(!PageBuddy(page));
4741 order = page_order(page);
4742#ifdef CONFIG_DEBUG_VM
4743 printk(KERN_INFO "remove from free list %lx %d %lx\n",
4744 pfn, 1 << order, end_pfn);
4745#endif
4746 list_del(&page->lru);
4747 rmv_page_order(page);
4748 zone->free_area[order].nr_free--;
4749 __mod_zone_page_state(zone, NR_FREE_PAGES,
4750 - (1UL << order));
4751 for (i = 0; i < (1 << order); i++)
4752 SetPageReserved((page+i));
4753 pfn += (1 << order);
4754 }
4755 spin_unlock_irqrestore(&zone->lock, flags);
4756}
4757#endif