blob: 9d5498e2d0f5a73c527c32b4ffe14cbfa897a7be [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>
Yinghai Luedbe7d22010-08-25 13:39:16 -070024#include <linux/memblock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/compiler.h>
Randy Dunlap9f158332005-09-13 01:25:16 -070026#include <linux/kernel.h>
Vegard Nossumb1eeab62008-11-25 16:55:53 +010027#include <linux/kmemcheck.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/module.h>
29#include <linux/suspend.h>
30#include <linux/pagevec.h>
31#include <linux/blkdev.h>
32#include <linux/slab.h>
David Rientjes5a3135c2007-10-16 23:25:53 -070033#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <linux/notifier.h>
35#include <linux/topology.h>
36#include <linux/sysctl.h>
37#include <linux/cpu.h>
38#include <linux/cpuset.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070039#include <linux/memory_hotplug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include <linux/nodemask.h>
41#include <linux/vmalloc.h>
Christoph Lameter4be38e32006-01-06 00:11:17 -080042#include <linux/mempolicy.h>
Yasunori Goto68113782006-06-23 02:03:11 -070043#include <linux/stop_machine.h>
Mel Gormanc7132162006-09-27 01:49:43 -070044#include <linux/sort.h>
45#include <linux/pfn.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070046#include <linux/backing-dev.h>
Akinobu Mita933e3122006-12-08 02:39:45 -080047#include <linux/fault-inject.h>
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -070048#include <linux/page-isolation.h>
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070049#include <linux/page_cgroup.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070050#include <linux/debugobjects.h>
Catalin Marinasdbb1f812009-06-11 13:23:19 +010051#include <linux/kmemleak.h>
Robert Jennings925cc712009-12-17 14:44:38 +000052#include <linux/memory.h>
Mel Gorman56de7262010-05-24 14:32:30 -070053#include <linux/compaction.h>
Mel Gorman0d3d0622009-09-21 17:02:44 -070054#include <trace/events/kmem.h>
Wu Fengguang718a3822010-03-10 15:20:43 -080055#include <linux/ftrace_event.h>
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -070056#include <linux/memcontrol.h>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070057#include <linux/prefetch.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
59#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070060#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070061#include "internal.h"
62
Lee Schermerhorn72812012010-05-26 14:44:56 -070063#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
64DEFINE_PER_CPU(int, numa_node);
65EXPORT_PER_CPU_SYMBOL(numa_node);
66#endif
67
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070068#ifdef CONFIG_HAVE_MEMORYLESS_NODES
69/*
70 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
71 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
72 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
73 * defined in <linux/topology.h>.
74 */
75DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
76EXPORT_PER_CPU_SYMBOL(_numa_mem_);
77#endif
78
Linus Torvalds1da177e2005-04-16 15:20:36 -070079/*
Christoph Lameter13808912007-10-16 01:25:27 -070080 * Array of node states.
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 */
Christoph Lameter13808912007-10-16 01:25:27 -070082nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
83 [N_POSSIBLE] = NODE_MASK_ALL,
84 [N_ONLINE] = { { [0] = 1UL } },
85#ifndef CONFIG_NUMA
86 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
87#ifdef CONFIG_HIGHMEM
88 [N_HIGH_MEMORY] = { { [0] = 1UL } },
89#endif
90 [N_CPU] = { { [0] = 1UL } },
91#endif /* NUMA */
92};
93EXPORT_SYMBOL(node_states);
94
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070095unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -070096unsigned long totalreserve_pages __read_mostly;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -080097int percpu_pagelist_fraction;
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +100098gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800100#ifdef CONFIG_PM_SLEEP
101/*
102 * The following functions are used by the suspend/hibernate code to temporarily
103 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
104 * while devices are suspended. To avoid races with the suspend/hibernate code,
105 * they should always be called with pm_mutex held (gfp_allowed_mask also should
106 * only be modified with pm_mutex held, unless the suspend/hibernate code is
107 * guaranteed not to run in parallel with that modification).
108 */
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100109
110static gfp_t saved_gfp_mask;
111
112void pm_restore_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800113{
114 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100115 if (saved_gfp_mask) {
116 gfp_allowed_mask = saved_gfp_mask;
117 saved_gfp_mask = 0;
118 }
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800119}
120
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100121void pm_restrict_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800122{
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800123 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100124 WARN_ON(saved_gfp_mask);
125 saved_gfp_mask = gfp_allowed_mask;
126 gfp_allowed_mask &= ~GFP_IOFS;
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800127}
128#endif /* CONFIG_PM_SLEEP */
129
Mel Gormand9c23402007-10-16 01:26:01 -0700130#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
131int pageblock_order __read_mostly;
132#endif
133
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800134static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -0800135
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136/*
137 * results with 256, 32 in the lowmem_reserve sysctl:
138 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
139 * 1G machine -> (16M dma, 784M normal, 224M high)
140 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
141 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
142 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +0100143 *
144 * TBD: should special case ZONE_DMA32 machines here - in those we normally
145 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700147int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800148#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700149 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -0800150#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700151#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700152 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700153#endif
Christoph Lametere53ef382006-09-25 23:31:14 -0700154#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700155 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700156#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700157 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700158};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159
160EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
Helge Deller15ad7cd2006-12-06 20:40:36 -0800162static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800163#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700164 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800165#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700166#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700167 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700168#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700169 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700170#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700171 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700172#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700173 "Movable",
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700174};
175
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176int min_free_kbytes = 1024;
177
Jan Beulich2c85f512009-09-21 17:03:07 -0700178static unsigned long __meminitdata nr_kernel_pages;
179static unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700180static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181
Mel Gormanc7132162006-09-27 01:49:43 -0700182#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
183 /*
Simon Arlott183ff222007-10-20 01:27:18 +0200184 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
Mel Gormanc7132162006-09-27 01:49:43 -0700185 * ranges of memory (RAM) that may be registered with add_active_range().
186 * Ranges passed to add_active_range() will be merged if possible
187 * so the number of times add_active_range() can be called is
188 * related to the number of nodes and the number of holes
189 */
190 #ifdef CONFIG_MAX_ACTIVE_REGIONS
191 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
192 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
193 #else
194 #if MAX_NUMNODES >= 32
195 /* If there can be many nodes, allow up to 50 holes per node */
196 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
197 #else
198 /* By default, allow up to 256 distinct regions */
199 #define MAX_ACTIVE_REGIONS 256
200 #endif
201 #endif
202
Jan Beulich98011f52007-07-15 23:38:17 -0700203 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
204 static int __meminitdata nr_nodemap_entries;
205 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
206 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
Adrian Bunkb69a7282008-07-23 21:28:12 -0700207 static unsigned long __initdata required_kernelcore;
Adrian Bunk484f51f2007-10-16 01:26:03 -0700208 static unsigned long __initdata required_movablecore;
Adrian Bunkb69a7282008-07-23 21:28:12 -0700209 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Mel Gorman2a1e2742007-07-17 04:03:12 -0700210
211 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
212 int movable_zone;
213 EXPORT_SYMBOL(movable_zone);
Mel Gormanc7132162006-09-27 01:49:43 -0700214#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
215
Miklos Szeredi418508c2007-05-23 13:57:55 -0700216#if MAX_NUMNODES > 1
217int nr_node_ids __read_mostly = MAX_NUMNODES;
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700218int nr_online_nodes __read_mostly = 1;
Miklos Szeredi418508c2007-05-23 13:57:55 -0700219EXPORT_SYMBOL(nr_node_ids);
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700220EXPORT_SYMBOL(nr_online_nodes);
Miklos Szeredi418508c2007-05-23 13:57:55 -0700221#endif
222
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700223int page_group_by_mobility_disabled __read_mostly;
224
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700225static void set_pageblock_migratetype(struct page *page, int migratetype)
226{
Mel Gorman49255c62009-06-16 15:31:58 -0700227
228 if (unlikely(page_group_by_mobility_disabled))
229 migratetype = MIGRATE_UNMOVABLE;
230
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700231 set_pageblock_flags_group(page, (unsigned long)migratetype,
232 PB_migrate, PB_migrate_end);
233}
234
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -0700235bool oom_killer_disabled __read_mostly;
236
Nick Piggin13e74442006-01-06 00:10:58 -0800237#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700238static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700240 int ret = 0;
241 unsigned seq;
242 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -0700243
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700244 do {
245 seq = zone_span_seqbegin(zone);
246 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
247 ret = 1;
248 else if (pfn < zone->zone_start_pfn)
249 ret = 1;
250 } while (zone_span_seqretry(zone, seq));
251
252 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700253}
254
255static int page_is_consistent(struct zone *zone, struct page *page)
256{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700257 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700258 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700260 return 0;
261
262 return 1;
263}
264/*
265 * Temporary debugging check for pages not lying within a given zone.
266 */
267static int bad_range(struct zone *zone, struct page *page)
268{
269 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700271 if (!page_is_consistent(zone, page))
272 return 1;
273
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 return 0;
275}
Nick Piggin13e74442006-01-06 00:10:58 -0800276#else
277static inline int bad_range(struct zone *zone, struct page *page)
278{
279 return 0;
280}
281#endif
282
Nick Piggin224abf92006-01-06 00:11:11 -0800283static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800285 static unsigned long resume;
286 static unsigned long nr_shown;
287 static unsigned long nr_unshown;
288
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200289 /* Don't complain about poisoned pages */
290 if (PageHWPoison(page)) {
Andrea Arcangelief2b4b92011-03-18 00:16:35 +0100291 reset_page_mapcount(page); /* remove PageBuddy */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200292 return;
293 }
294
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800295 /*
296 * Allow a burst of 60 reports, then keep quiet for that minute;
297 * or allow a steady drip of one report per second.
298 */
299 if (nr_shown == 60) {
300 if (time_before(jiffies, resume)) {
301 nr_unshown++;
302 goto out;
303 }
304 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800305 printk(KERN_ALERT
306 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800307 nr_unshown);
308 nr_unshown = 0;
309 }
310 nr_shown = 0;
311 }
312 if (nr_shown++ == 0)
313 resume = jiffies + 60 * HZ;
314
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800315 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800316 current->comm, page_to_pfn(page));
Wu Fengguang718a3822010-03-10 15:20:43 -0800317 dump_page(page);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700318
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800320out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800321 /* Leave bad fields for debug, except PageBuddy could make trouble */
Andrea Arcangelief2b4b92011-03-18 00:16:35 +0100322 reset_page_mapcount(page); /* remove PageBuddy */
Randy Dunlap9f158332005-09-13 01:25:16 -0700323 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324}
325
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326/*
327 * Higher-order pages are called "compound pages". They are structured thusly:
328 *
329 * The first PAGE_SIZE page is called the "head page".
330 *
331 * The remaining PAGE_SIZE pages are called "tail pages".
332 *
333 * All pages have PG_compound set. All pages have their ->private pointing at
334 * the head page (even the head page has this).
335 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800336 * The first tail page's ->lru.next holds the address of the compound page's
337 * put_page() function. Its ->lru.prev holds the order of allocation.
338 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800340
341static void free_compound_page(struct page *page)
342{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700343 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800344}
345
Andi Kleen01ad1c02008-07-23 21:27:46 -0700346void prep_compound_page(struct page *page, unsigned long order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347{
348 int i;
349 int nr_pages = 1 << order;
350
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800351 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700352 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700353 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800354 for (i = 1; i < nr_pages; i++) {
355 struct page *p = page + i;
356
Christoph Lameterd85f3382007-05-06 14:49:39 -0700357 __SetPageTail(p);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700358 p->first_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 }
360}
361
Andrea Arcangeli59ff4212011-01-13 15:46:38 -0800362/* update __split_huge_page_refcount if you change this function */
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800363static int destroy_compound_page(struct page *page, unsigned long order)
Andy Whitcroft18229df2008-11-06 12:53:27 -0800364{
365 int i;
366 int nr_pages = 1 << order;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800367 int bad = 0;
Andy Whitcroft18229df2008-11-06 12:53:27 -0800368
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800369 if (unlikely(compound_order(page) != order) ||
370 unlikely(!PageHead(page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800371 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800372 bad++;
373 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374
Christoph Lameter6d777952007-05-06 14:49:40 -0700375 __ClearPageHead(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800376
Andy Whitcroft18229df2008-11-06 12:53:27 -0800377 for (i = 1; i < nr_pages; i++) {
378 struct page *p = page + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379
Alexey Zaytseve713a212009-01-10 02:47:57 +0300380 if (unlikely(!PageTail(p) || (p->first_page != page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800381 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800382 bad++;
383 }
Christoph Lameterd85f3382007-05-06 14:49:39 -0700384 __ClearPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800386
387 return bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389
Nick Piggin17cf4402006-03-22 00:08:41 -0800390static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
391{
392 int i;
393
Andrew Morton6626c5d2006-03-22 00:08:42 -0800394 /*
395 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
396 * and __GFP_HIGHMEM from hard or soft interrupt context.
397 */
Nick Piggin725d7042006-09-25 23:30:55 -0700398 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800399 for (i = 0; i < (1 << order); i++)
400 clear_highpage(page + i);
401}
402
Andrew Morton6aa30012006-04-18 22:20:52 -0700403static inline void set_page_order(struct page *page, int order)
404{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700405 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000406 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407}
408
409static inline void rmv_page_order(struct page *page)
410{
Nick Piggin676165a2006-04-10 11:21:48 +1000411 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700412 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413}
414
415/*
416 * Locate the struct page for both the matching buddy in our
417 * pair (buddy1) and the combined O(n+1) page they form (page).
418 *
419 * 1) Any buddy B1 will have an order O twin B2 which satisfies
420 * the following equation:
421 * B2 = B1 ^ (1 << O)
422 * For example, if the starting buddy (buddy2) is #8 its order
423 * 1 buddy is #10:
424 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
425 *
426 * 2) Any buddy B will have an order O+1 parent P which
427 * satisfies the following equation:
428 * P = B & ~(1 << O)
429 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200430 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432static inline unsigned long
KyongHo Cho43506fa2011-01-13 15:47:24 -0800433__find_buddy_index(unsigned long page_idx, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434{
KyongHo Cho43506fa2011-01-13 15:47:24 -0800435 return page_idx ^ (1 << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436}
437
438/*
439 * This function checks whether a page is free && is the buddy
440 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800441 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000442 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700443 * (c) a page and its buddy have the same order &&
444 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 *
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800446 * For recording whether a page is in the buddy system, we set ->_mapcount -2.
447 * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock.
Nick Piggin676165a2006-04-10 11:21:48 +1000448 *
449 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700451static inline int page_is_buddy(struct page *page, struct page *buddy,
452 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700454 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800455 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800456
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700457 if (page_zone_id(page) != page_zone_id(buddy))
458 return 0;
459
460 if (PageBuddy(buddy) && page_order(buddy) == order) {
Nick Piggina3af9c32009-06-16 15:32:10 -0700461 VM_BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa30012006-04-18 22:20:52 -0700462 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000463 }
Andrew Morton6aa30012006-04-18 22:20:52 -0700464 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465}
466
467/*
468 * Freeing function for a buddy system allocator.
469 *
470 * The concept of a buddy system is to maintain direct-mapped table
471 * (containing bit values) for memory blocks of various "orders".
472 * The bottom level table contains the map for the smallest allocatable
473 * units of memory (here, pages), and each level above it describes
474 * pairs of units from the levels below, hence, "buddies".
475 * At a high level, all that happens here is marking the table entry
476 * at the bottom level available, and propagating the changes upward
477 * as necessary, plus some accounting needed to play nicely with other
478 * parts of the VM system.
479 * At each level, we keep a list of pages, which are heads of continuous
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800480 * free pages of length of (1 << order) and marked with _mapcount -2. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700481 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 * So when we are allocating or freeing one, we can derive the state of the
483 * other. That is, if we allocate a small block, and both were
484 * free, the remainder of the region must be split into blocks.
485 * If a block is freed, and its buddy is also free, then this
486 * triggers coalescing into a block of larger size.
487 *
488 * -- wli
489 */
490
Nick Piggin48db57f2006-01-08 01:00:42 -0800491static inline void __free_one_page(struct page *page,
Mel Gormaned0ae212009-06-16 15:32:07 -0700492 struct zone *zone, unsigned int order,
493 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494{
495 unsigned long page_idx;
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700496 unsigned long combined_idx;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800497 unsigned long uninitialized_var(buddy_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700498 struct page *buddy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499
Nick Piggin224abf92006-01-06 00:11:11 -0800500 if (unlikely(PageCompound(page)))
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800501 if (unlikely(destroy_compound_page(page, order)))
502 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503
Mel Gormaned0ae212009-06-16 15:32:07 -0700504 VM_BUG_ON(migratetype == -1);
505
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
507
Mel Gormanf2260e62009-06-16 15:32:13 -0700508 VM_BUG_ON(page_idx & ((1 << order) - 1));
Nick Piggin725d7042006-09-25 23:30:55 -0700509 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 while (order < MAX_ORDER-1) {
KyongHo Cho43506fa2011-01-13 15:47:24 -0800512 buddy_idx = __find_buddy_index(page_idx, order);
513 buddy = page + (buddy_idx - page_idx);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700514 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700515 break;
Nick Piggin13e74442006-01-06 00:10:58 -0800516
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700517 /* Our buddy is free, merge with it and move up one order. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 list_del(&buddy->lru);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700519 zone->free_area[order].nr_free--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 rmv_page_order(buddy);
KyongHo Cho43506fa2011-01-13 15:47:24 -0800521 combined_idx = buddy_idx & page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 page = page + (combined_idx - page_idx);
523 page_idx = combined_idx;
524 order++;
525 }
526 set_page_order(page, order);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700527
528 /*
529 * If this is not the largest possible page, check if the buddy
530 * of the next-highest order is free. If it is, it's possible
531 * that pages are being freed that will coalesce soon. In case,
532 * that is happening, add the free page to the tail of the list
533 * so it's less likely to be used soon and more likely to be merged
534 * as a higher order page
535 */
Mel Gormanb7f50cf2010-10-26 14:21:11 -0700536 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700537 struct page *higher_page, *higher_buddy;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800538 combined_idx = buddy_idx & page_idx;
539 higher_page = page + (combined_idx - page_idx);
540 buddy_idx = __find_buddy_index(combined_idx, order + 1);
541 higher_buddy = page + (buddy_idx - combined_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700542 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
543 list_add_tail(&page->lru,
544 &zone->free_area[order].free_list[migratetype]);
545 goto out;
546 }
547 }
548
549 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
550out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 zone->free_area[order].nr_free++;
552}
553
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700554/*
555 * free_page_mlock() -- clean up attempts to free and mlocked() page.
556 * Page should not be on lru, so no need to fix that up.
557 * free_pages_check() will verify...
558 */
559static inline void free_page_mlock(struct page *page)
560{
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700561 __dec_zone_page_state(page, NR_MLOCK);
562 __count_vm_event(UNEVICTABLE_MLOCKFREED);
563}
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700564
Nick Piggin224abf92006-01-06 00:11:11 -0800565static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566{
Nick Piggin92be2e32006-01-06 00:10:57 -0800567 if (unlikely(page_mapcount(page) |
568 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700569 (atomic_read(&page->_count) != 0) |
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -0700570 (page->flags & PAGE_FLAGS_CHECK_AT_FREE) |
571 (mem_cgroup_bad_page_check(page)))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800572 bad_page(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800573 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800574 }
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800575 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
576 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
577 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578}
579
580/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700581 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700583 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 *
585 * If the zone was previously in an "all pages pinned" state then look to
586 * see if this freeing clears that state.
587 *
588 * And clear the zone's pages_scanned counter, to hold off the "all pages are
589 * pinned" detection logic.
590 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700591static void free_pcppages_bulk(struct zone *zone, int count,
592 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593{
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700594 int migratetype = 0;
Mel Gormana6f9edd2009-09-21 17:03:20 -0700595 int batch_free = 0;
Mel Gorman72853e22010-09-09 16:38:16 -0700596 int to_free = count;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700597
Nick Pigginc54ad302006-01-06 00:10:56 -0800598 spin_lock(&zone->lock);
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -0800599 zone->all_unreclaimable = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700601
Mel Gorman72853e22010-09-09 16:38:16 -0700602 while (to_free) {
Nick Piggin48db57f2006-01-08 01:00:42 -0800603 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700604 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -0800605
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700606 /*
Mel Gormana6f9edd2009-09-21 17:03:20 -0700607 * Remove pages from lists in a round-robin fashion. A
608 * batch_free count is maintained that is incremented when an
609 * empty list is encountered. This is so more pages are freed
610 * off fuller lists instead of spinning excessively around empty
611 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700612 */
613 do {
Mel Gormana6f9edd2009-09-21 17:03:20 -0700614 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700615 if (++migratetype == MIGRATE_PCPTYPES)
616 migratetype = 0;
617 list = &pcp->lists[migratetype];
618 } while (list_empty(list));
619
Namhyung Kim1d168712011-03-22 16:32:45 -0700620 /* This is the only non-empty list. Free them all. */
621 if (batch_free == MIGRATE_PCPTYPES)
622 batch_free = to_free;
623
Mel Gormana6f9edd2009-09-21 17:03:20 -0700624 do {
625 page = list_entry(list->prev, struct page, lru);
626 /* must delete as __free_one_page list manipulates */
627 list_del(&page->lru);
Hugh Dickinsa7016232010-01-29 17:46:34 +0000628 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
629 __free_one_page(page, zone, 0, page_private(page));
630 trace_mm_page_pcpu_drain(page, 0, page_private(page));
Mel Gorman72853e22010-09-09 16:38:16 -0700631 } while (--to_free && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 }
Mel Gorman72853e22010-09-09 16:38:16 -0700633 __mod_zone_page_state(zone, NR_FREE_PAGES, count);
Nick Pigginc54ad302006-01-06 00:10:56 -0800634 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635}
636
Mel Gormaned0ae212009-06-16 15:32:07 -0700637static void free_one_page(struct zone *zone, struct page *page, int order,
638 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800639{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700640 spin_lock(&zone->lock);
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -0800641 zone->all_unreclaimable = 0;
Christoph Lameter006d22d2006-09-25 23:31:48 -0700642 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700643
Mel Gormaned0ae212009-06-16 15:32:07 -0700644 __free_one_page(page, zone, order, migratetype);
Mel Gorman72853e22010-09-09 16:38:16 -0700645 __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700646 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800647}
648
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700649static bool free_pages_prepare(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 int i;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800652 int bad = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653
Li Hongf6503162010-03-05 13:41:52 -0800654 trace_mm_page_free_direct(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100655 kmemcheck_free_shadow(page, order);
656
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -0800657 if (PageAnon(page))
658 page->mapping = NULL;
659 for (i = 0; i < (1 << order); i++)
660 bad += free_pages_check(page + i);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800661 if (bad)
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700662 return false;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800663
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700664 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -0700665 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700666 debug_check_no_obj_freed(page_address(page),
667 PAGE_SIZE << order);
668 }
Nick Piggindafb1362006-10-11 01:21:30 -0700669 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800670 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700671
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700672 return true;
673}
674
675static void __free_pages_ok(struct page *page, unsigned int order)
676{
677 unsigned long flags;
678 int wasMlocked = __TestClearPageMlocked(page);
679
680 if (!free_pages_prepare(page, order))
681 return;
682
Nick Pigginc54ad302006-01-06 00:10:56 -0800683 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +0200684 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -0700685 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700686 __count_vm_events(PGFREE, 1 << order);
Mel Gormaned0ae212009-06-16 15:32:07 -0700687 free_one_page(page_zone(page), page, order,
688 get_pageblock_migratetype(page));
Nick Pigginc54ad302006-01-06 00:10:56 -0800689 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690}
691
David Howellsa226f6c2006-01-06 00:11:08 -0800692/*
693 * permit the bootmem allocator to evade page validation on high-order frees
694 */
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700695void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800696{
697 if (order == 0) {
698 __ClearPageReserved(page);
699 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800700 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800701 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800702 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800703 int loop;
704
Nick Piggin545b1ea2006-03-22 00:08:07 -0800705 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800706 for (loop = 0; loop < BITS_PER_LONG; loop++) {
707 struct page *p = &page[loop];
708
Nick Piggin545b1ea2006-03-22 00:08:07 -0800709 if (loop + 1 < BITS_PER_LONG)
710 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800711 __ClearPageReserved(p);
712 set_page_count(p, 0);
713 }
714
Nick Piggin7835e982006-03-22 00:08:40 -0800715 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800716 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800717 }
718}
719
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720
721/*
722 * The order of subdivision here is critical for the IO subsystem.
723 * Please do not alter this order without good reasons and regression
724 * testing. Specifically, as large blocks of memory are subdivided,
725 * the order in which smaller blocks are delivered depends on the order
726 * they're subdivided in this function. This is the primary factor
727 * influencing the order in which pages are delivered to the IO
728 * subsystem according to empirical testing, and this is also justified
729 * by considering the behavior of a buddy system containing a single
730 * large block of memory acted on by a series of small allocations.
731 * This behavior is a critical factor in sglist merging's success.
732 *
733 * -- wli
734 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800735static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700736 int low, int high, struct free_area *area,
737 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738{
739 unsigned long size = 1 << high;
740
741 while (high > low) {
742 area--;
743 high--;
744 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700745 VM_BUG_ON(bad_range(zone, &page[size]));
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700746 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 area->nr_free++;
748 set_page_order(&page[size], high);
749 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750}
751
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752/*
753 * This page is about to be returned from the page allocator
754 */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200755static inline int check_new_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756{
Nick Piggin92be2e32006-01-06 00:10:57 -0800757 if (unlikely(page_mapcount(page) |
758 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700759 (atomic_read(&page->_count) != 0) |
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -0700760 (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
761 (mem_cgroup_bad_page_check(page)))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800762 bad_page(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800763 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800764 }
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200765 return 0;
766}
767
768static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
769{
770 int i;
771
772 for (i = 0; i < (1 << order); i++) {
773 struct page *p = page + i;
774 if (unlikely(check_new_page(p)))
775 return 1;
776 }
Hugh Dickins689bceb2005-11-21 21:32:20 -0800777
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700778 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800779 set_page_refcounted(page);
Nick Piggincc102502006-12-06 20:32:00 -0800780
781 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800783
784 if (gfp_flags & __GFP_ZERO)
785 prep_zero_page(page, order, gfp_flags);
786
787 if (order && (gfp_flags & __GFP_COMP))
788 prep_compound_page(page, order);
789
Hugh Dickins689bceb2005-11-21 21:32:20 -0800790 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791}
792
Mel Gorman56fd56b2007-10-16 01:25:58 -0700793/*
794 * Go through the free lists for the given migratetype and remove
795 * the smallest available page from the freelists
796 */
Mel Gorman728ec982009-06-16 15:32:04 -0700797static inline
798struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -0700799 int migratetype)
800{
801 unsigned int current_order;
802 struct free_area * area;
803 struct page *page;
804
805 /* Find a page of the appropriate size in the preferred list */
806 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
807 area = &(zone->free_area[current_order]);
808 if (list_empty(&area->free_list[migratetype]))
809 continue;
810
811 page = list_entry(area->free_list[migratetype].next,
812 struct page, lru);
813 list_del(&page->lru);
814 rmv_page_order(page);
815 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -0700816 expand(zone, page, order, current_order, area, migratetype);
817 return page;
818 }
819
820 return NULL;
821}
822
823
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700824/*
825 * This array describes the order lists are fallen back to when
826 * the free lists for the desirable migrate type are depleted
827 */
828static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
Mel Gorman64c5e132007-10-16 01:25:59 -0700829 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
830 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
831 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
832 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700833};
834
Mel Gormanc361be52007-10-16 01:25:51 -0700835/*
836 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -0700837 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -0700838 * boundary. If alignment is required, use move_freepages_block()
839 */
Adrian Bunkb69a7282008-07-23 21:28:12 -0700840static int move_freepages(struct zone *zone,
841 struct page *start_page, struct page *end_page,
842 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700843{
844 struct page *page;
845 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -0700846 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -0700847
848#ifndef CONFIG_HOLES_IN_ZONE
849 /*
850 * page_zone is not safe to call in this context when
851 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
852 * anyway as we check zone boundaries in move_freepages_block().
853 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -0700854 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -0700855 */
856 BUG_ON(page_zone(start_page) != page_zone(end_page));
857#endif
858
859 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -0700860 /* Make sure we are not inadvertently changing nodes */
861 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
862
Mel Gormanc361be52007-10-16 01:25:51 -0700863 if (!pfn_valid_within(page_to_pfn(page))) {
864 page++;
865 continue;
866 }
867
868 if (!PageBuddy(page)) {
869 page++;
870 continue;
871 }
872
873 order = page_order(page);
Kirill A. Shutemov84be48d2011-03-22 16:33:41 -0700874 list_move(&page->lru,
875 &zone->free_area[order].free_list[migratetype]);
Mel Gormanc361be52007-10-16 01:25:51 -0700876 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -0700877 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -0700878 }
879
Mel Gormand1003132007-10-16 01:26:00 -0700880 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -0700881}
882
Adrian Bunkb69a7282008-07-23 21:28:12 -0700883static int move_freepages_block(struct zone *zone, struct page *page,
884 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700885{
886 unsigned long start_pfn, end_pfn;
887 struct page *start_page, *end_page;
888
889 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -0700890 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -0700891 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -0700892 end_page = start_page + pageblock_nr_pages - 1;
893 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -0700894
895 /* Do not cross zone boundaries */
896 if (start_pfn < zone->zone_start_pfn)
897 start_page = page;
898 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
899 return 0;
900
901 return move_freepages(zone, start_page, end_page, migratetype);
902}
903
Mel Gorman2f66a682009-09-21 17:02:31 -0700904static void change_pageblock_range(struct page *pageblock_page,
905 int start_order, int migratetype)
906{
907 int nr_pageblocks = 1 << (start_order - pageblock_order);
908
909 while (nr_pageblocks--) {
910 set_pageblock_migratetype(pageblock_page, migratetype);
911 pageblock_page += pageblock_nr_pages;
912 }
913}
914
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700915/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -0700916static inline struct page *
917__rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700918{
919 struct free_area * area;
920 int current_order;
921 struct page *page;
922 int migratetype, i;
923
924 /* Find the largest possible block of pages in the other list */
925 for (current_order = MAX_ORDER-1; current_order >= order;
926 --current_order) {
927 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
928 migratetype = fallbacks[start_migratetype][i];
929
Mel Gorman56fd56b2007-10-16 01:25:58 -0700930 /* MIGRATE_RESERVE handled later if necessary */
931 if (migratetype == MIGRATE_RESERVE)
932 continue;
Mel Gormane0104872007-10-16 01:25:53 -0700933
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700934 area = &(zone->free_area[current_order]);
935 if (list_empty(&area->free_list[migratetype]))
936 continue;
937
938 page = list_entry(area->free_list[migratetype].next,
939 struct page, lru);
940 area->nr_free--;
941
942 /*
Mel Gormanc361be52007-10-16 01:25:51 -0700943 * If breaking a large block of pages, move all free
Mel Gorman46dafbc2007-10-16 01:25:55 -0700944 * pages to the preferred allocation list. If falling
945 * back for a reclaimable kernel allocation, be more
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300946 * aggressive about taking ownership of free pages
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700947 */
Mel Gormand9c23402007-10-16 01:26:01 -0700948 if (unlikely(current_order >= (pageblock_order >> 1)) ||
Mel Gormandd5d2412009-09-05 11:17:11 -0700949 start_migratetype == MIGRATE_RECLAIMABLE ||
950 page_group_by_mobility_disabled) {
Mel Gorman46dafbc2007-10-16 01:25:55 -0700951 unsigned long pages;
952 pages = move_freepages_block(zone, page,
953 start_migratetype);
954
955 /* Claim the whole block if over half of it is free */
Mel Gormandd5d2412009-09-05 11:17:11 -0700956 if (pages >= (1 << (pageblock_order-1)) ||
957 page_group_by_mobility_disabled)
Mel Gorman46dafbc2007-10-16 01:25:55 -0700958 set_pageblock_migratetype(page,
959 start_migratetype);
960
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700961 migratetype = start_migratetype;
Mel Gormanc361be52007-10-16 01:25:51 -0700962 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700963
964 /* Remove the page from the freelists */
965 list_del(&page->lru);
966 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700967
Mel Gorman2f66a682009-09-21 17:02:31 -0700968 /* Take ownership for orders >= pageblock_order */
969 if (current_order >= pageblock_order)
970 change_pageblock_range(page, current_order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700971 start_migratetype);
972
973 expand(zone, page, order, current_order, area, migratetype);
Mel Gormane0fff1b2009-09-21 17:02:42 -0700974
975 trace_mm_page_alloc_extfrag(page, order, current_order,
976 start_migratetype, migratetype);
977
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700978 return page;
979 }
980 }
981
Mel Gorman728ec982009-06-16 15:32:04 -0700982 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700983}
984
Mel Gorman56fd56b2007-10-16 01:25:58 -0700985/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 * Do the hard work of removing an element from the buddy allocator.
987 * Call me with the zone->lock already held.
988 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700989static struct page *__rmqueue(struct zone *zone, unsigned int order,
990 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 struct page *page;
993
Mel Gorman728ec982009-06-16 15:32:04 -0700994retry_reserve:
Mel Gorman56fd56b2007-10-16 01:25:58 -0700995 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996
Mel Gorman728ec982009-06-16 15:32:04 -0700997 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Mel Gorman56fd56b2007-10-16 01:25:58 -0700998 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700999
Mel Gorman728ec982009-06-16 15:32:04 -07001000 /*
1001 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1002 * is used because __rmqueue_smallest is an inline function
1003 * and we want just one call site
1004 */
1005 if (!page) {
1006 migratetype = MIGRATE_RESERVE;
1007 goto retry_reserve;
1008 }
1009 }
1010
Mel Gorman0d3d0622009-09-21 17:02:44 -07001011 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001012 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013}
1014
1015/*
1016 * Obtain a specified number of elements from the buddy allocator, all under
1017 * a single hold of the lock, for efficiency. Add them to the supplied list.
1018 * Returns the number of new pages which were placed at *list.
1019 */
1020static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001021 unsigned long count, struct list_head *list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001022 int migratetype, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025
Nick Pigginc54ad302006-01-06 00:10:56 -08001026 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001028 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001029 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08001031
1032 /*
1033 * Split buddy pages returned by expand() are received here
1034 * in physical page order. The page is added to the callers and
1035 * list and the list head then moves forward. From the callers
1036 * perspective, the linked list is ordered by page number in
1037 * some conditions. This is useful for IO devices that can
1038 * merge IO requests if the physical pages are ordered
1039 * properly.
1040 */
Mel Gormane084b2d2009-07-29 15:02:04 -07001041 if (likely(cold == 0))
1042 list_add(&page->lru, list);
1043 else
1044 list_add_tail(&page->lru, list);
Mel Gorman535131e2007-10-16 01:25:49 -07001045 set_page_private(page, migratetype);
Mel Gorman81eabcb2007-12-17 16:20:05 -08001046 list = &page->lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 }
Mel Gormanf2260e62009-06-16 15:32:13 -07001048 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08001049 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001050 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051}
1052
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001053#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001054/*
Christoph Lameter4037d452007-05-09 02:35:14 -07001055 * Called from the vmstat counter updater to drain pagesets of this
1056 * currently executing processor on remote nodes after they have
1057 * expired.
1058 *
Christoph Lameter879336c2006-03-22 00:09:08 -08001059 * Note that this function must be called with the thread pinned to
1060 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001061 */
Christoph Lameter4037d452007-05-09 02:35:14 -07001062void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001063{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001064 unsigned long flags;
Christoph Lameter4037d452007-05-09 02:35:14 -07001065 int to_drain;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001066
Christoph Lameter4037d452007-05-09 02:35:14 -07001067 local_irq_save(flags);
1068 if (pcp->count >= pcp->batch)
1069 to_drain = pcp->batch;
1070 else
1071 to_drain = pcp->count;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001072 free_pcppages_bulk(zone, to_drain, pcp);
Christoph Lameter4037d452007-05-09 02:35:14 -07001073 pcp->count -= to_drain;
1074 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001075}
1076#endif
1077
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001078/*
1079 * Drain pages of the indicated processor.
1080 *
1081 * The processor must either be the current processor and the
1082 * thread pinned to the current processor or a processor that
1083 * is not online.
1084 */
1085static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086{
Nick Pigginc54ad302006-01-06 00:10:56 -08001087 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07001090 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 struct per_cpu_pageset *pset;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001092 struct per_cpu_pages *pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001094 local_irq_save(flags);
1095 pset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001097 pcp = &pset->pcp;
David Rientjes2ff754f2011-01-25 15:07:23 -08001098 if (pcp->count) {
1099 free_pcppages_bulk(zone, pcp->count, pcp);
1100 pcp->count = 0;
1101 }
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001102 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 }
1104}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001106/*
1107 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1108 */
1109void drain_local_pages(void *arg)
1110{
1111 drain_pages(smp_processor_id());
1112}
1113
1114/*
1115 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
1116 */
1117void drain_all_pages(void)
1118{
Jens Axboe15c8b6c2008-05-09 09:39:44 +02001119 on_each_cpu(drain_local_pages, NULL, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001120}
1121
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02001122#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123
1124void mark_free_pages(struct zone *zone)
1125{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001126 unsigned long pfn, max_zone_pfn;
1127 unsigned long flags;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001128 int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 struct list_head *curr;
1130
1131 if (!zone->spanned_pages)
1132 return;
1133
1134 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001135
1136 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1137 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1138 if (pfn_valid(pfn)) {
1139 struct page *page = pfn_to_page(pfn);
1140
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001141 if (!swsusp_page_is_forbidden(page))
1142 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001143 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001145 for_each_migratetype_order(order, t) {
1146 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001147 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001149 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1150 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001151 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001152 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001153 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 spin_unlock_irqrestore(&zone->lock, flags);
1155}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001156#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157
1158/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 * Free a 0-order page
Li Hongfc916682010-03-05 13:41:54 -08001160 * cold == 1 ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 */
Li Hongfc916682010-03-05 13:41:54 -08001162void free_hot_cold_page(struct page *page, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163{
1164 struct zone *zone = page_zone(page);
1165 struct per_cpu_pages *pcp;
1166 unsigned long flags;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001167 int migratetype;
Johannes Weiner451ea252009-09-21 17:01:48 -07001168 int wasMlocked = __TestClearPageMlocked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001170 if (!free_pages_prepare(page, 0))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001171 return;
1172
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001173 migratetype = get_pageblock_migratetype(page);
1174 set_page_private(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +02001176 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -07001177 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001178 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001179
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001180 /*
1181 * We only track unmovable, reclaimable and movable on pcp lists.
1182 * Free ISOLATE pages back to the allocator because they are being
1183 * offlined but treat RESERVE as movable pages so we can get those
1184 * areas back if necessary. Otherwise, we may have to free
1185 * excessively into the page allocator
1186 */
1187 if (migratetype >= MIGRATE_PCPTYPES) {
1188 if (unlikely(migratetype == MIGRATE_ISOLATE)) {
1189 free_one_page(zone, page, 0, migratetype);
1190 goto out;
1191 }
1192 migratetype = MIGRATE_MOVABLE;
1193 }
1194
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001195 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001196 if (cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001197 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001198 else
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001199 list_add(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001201 if (pcp->count >= pcp->high) {
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001202 free_pcppages_bulk(zone, pcp->batch, pcp);
Nick Piggin48db57f2006-01-08 01:00:42 -08001203 pcp->count -= pcp->batch;
1204 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001205
1206out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208}
1209
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001210/*
1211 * split_page takes a non-compound higher-order page, and splits it into
1212 * n (1<<order) sub-pages: page[0..n]
1213 * Each sub-page must be freed individually.
1214 *
1215 * Note: this is probably too low level an operation for use in drivers.
1216 * Please consult with lkml before using this in your driver.
1217 */
1218void split_page(struct page *page, unsigned int order)
1219{
1220 int i;
1221
Nick Piggin725d7042006-09-25 23:30:55 -07001222 VM_BUG_ON(PageCompound(page));
1223 VM_BUG_ON(!page_count(page));
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001224
1225#ifdef CONFIG_KMEMCHECK
1226 /*
1227 * Split shadow pages too, because free(page[0]) would
1228 * otherwise free the whole shadow.
1229 */
1230 if (kmemcheck_page_is_tracked(page))
1231 split_page(virt_to_page(page[0].shadow), order);
1232#endif
1233
Nick Piggin7835e982006-03-22 00:08:40 -08001234 for (i = 1; i < (1 << order); i++)
1235 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001236}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001237
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238/*
Mel Gorman748446b2010-05-24 14:32:27 -07001239 * Similar to split_page except the page is already free. As this is only
1240 * being used for migration, the migratetype of the block also changes.
1241 * As this is called with interrupts disabled, the caller is responsible
1242 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1243 * are enabled.
1244 *
1245 * Note: this is probably too low level an operation for use in drivers.
1246 * Please consult with lkml before using this in your driver.
1247 */
1248int split_free_page(struct page *page)
1249{
1250 unsigned int order;
1251 unsigned long watermark;
1252 struct zone *zone;
1253
1254 BUG_ON(!PageBuddy(page));
1255
1256 zone = page_zone(page);
1257 order = page_order(page);
1258
1259 /* Obey watermarks as if the page was being allocated */
1260 watermark = low_wmark_pages(zone) + (1 << order);
1261 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1262 return 0;
1263
1264 /* Remove page from free list */
1265 list_del(&page->lru);
1266 zone->free_area[order].nr_free--;
1267 rmv_page_order(page);
1268 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1UL << order));
1269
1270 /* Split into individual pages */
1271 set_page_refcounted(page);
1272 split_page(page, order);
1273
1274 if (order >= pageblock_order - 1) {
1275 struct page *endpage = page + (1 << order) - 1;
1276 for (; page < endpage; page += pageblock_nr_pages)
1277 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1278 }
1279
1280 return 1 << order;
1281}
1282
1283/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1285 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1286 * or two.
1287 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07001288static inline
1289struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001290 struct zone *zone, int order, gfp_t gfp_flags,
1291 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292{
1293 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001294 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 int cold = !!(gfp_flags & __GFP_COLD);
1296
Hugh Dickins689bceb2005-11-21 21:32:20 -08001297again:
Nick Piggin48db57f2006-01-08 01:00:42 -08001298 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001300 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 local_irq_save(flags);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001303 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1304 list = &pcp->lists[migratetype];
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001305 if (list_empty(list)) {
Mel Gorman535131e2007-10-16 01:25:49 -07001306 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001307 pcp->batch, list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001308 migratetype, cold);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001309 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07001310 goto failed;
Mel Gorman535131e2007-10-16 01:25:49 -07001311 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001312
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001313 if (cold)
1314 page = list_entry(list->prev, struct page, lru);
1315 else
1316 page = list_entry(list->next, struct page, lru);
1317
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001318 list_del(&page->lru);
1319 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001320 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07001321 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1322 /*
1323 * __GFP_NOFAIL is not to be used in new code.
1324 *
1325 * All __GFP_NOFAIL callers should be fixed so that they
1326 * properly detect and handle allocation failures.
1327 *
1328 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07001329 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07001330 * __GFP_NOFAIL.
1331 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07001332 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07001333 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001335 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001336 spin_unlock(&zone->lock);
1337 if (!page)
1338 goto failed;
KOSAKI Motohiro6ccf80e2010-01-15 17:01:18 -08001339 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 }
1341
Christoph Lameterf8891e52006-06-30 01:55:45 -07001342 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Andi Kleen78afd562011-03-22 16:33:12 -07001343 zone_statistics(preferred_zone, zone, gfp_flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001344 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345
Nick Piggin725d7042006-09-25 23:30:55 -07001346 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -08001347 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001348 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001350
1351failed:
1352 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001353 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354}
1355
Mel Gorman41858962009-06-16 15:32:12 -07001356/* The ALLOC_WMARK bits are used as an index to zone->watermark */
1357#define ALLOC_WMARK_MIN WMARK_MIN
1358#define ALLOC_WMARK_LOW WMARK_LOW
1359#define ALLOC_WMARK_HIGH WMARK_HIGH
1360#define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1361
1362/* Mask to get the watermark bits */
1363#define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1364
Nick Piggin31488902005-11-28 13:44:03 -08001365#define ALLOC_HARDER 0x10 /* try to alloc harder */
1366#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1367#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001368
Akinobu Mita933e3122006-12-08 02:39:45 -08001369#ifdef CONFIG_FAIL_PAGE_ALLOC
1370
1371static struct fail_page_alloc_attr {
1372 struct fault_attr attr;
1373
1374 u32 ignore_gfp_highmem;
1375 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001376 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001377
1378#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1379
1380 struct dentry *ignore_gfp_highmem_file;
1381 struct dentry *ignore_gfp_wait_file;
Akinobu Mita54114992007-07-15 23:40:23 -07001382 struct dentry *min_order_file;
Akinobu Mita933e3122006-12-08 02:39:45 -08001383
1384#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1385
1386} fail_page_alloc = {
1387 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001388 .ignore_gfp_wait = 1,
1389 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001390 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001391};
1392
1393static int __init setup_fail_page_alloc(char *str)
1394{
1395 return setup_fault_attr(&fail_page_alloc.attr, str);
1396}
1397__setup("fail_page_alloc=", setup_fail_page_alloc);
1398
1399static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1400{
Akinobu Mita54114992007-07-15 23:40:23 -07001401 if (order < fail_page_alloc.min_order)
1402 return 0;
Akinobu Mita933e3122006-12-08 02:39:45 -08001403 if (gfp_mask & __GFP_NOFAIL)
1404 return 0;
1405 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1406 return 0;
1407 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1408 return 0;
1409
1410 return should_fail(&fail_page_alloc.attr, 1 << order);
1411}
1412
1413#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1414
1415static int __init fail_page_alloc_debugfs(void)
1416{
1417 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1418 struct dentry *dir;
1419 int err;
1420
1421 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1422 "fail_page_alloc");
1423 if (err)
1424 return err;
1425 dir = fail_page_alloc.attr.dentries.dir;
1426
1427 fail_page_alloc.ignore_gfp_wait_file =
1428 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1429 &fail_page_alloc.ignore_gfp_wait);
1430
1431 fail_page_alloc.ignore_gfp_highmem_file =
1432 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1433 &fail_page_alloc.ignore_gfp_highmem);
Akinobu Mita54114992007-07-15 23:40:23 -07001434 fail_page_alloc.min_order_file =
1435 debugfs_create_u32("min-order", mode, dir,
1436 &fail_page_alloc.min_order);
Akinobu Mita933e3122006-12-08 02:39:45 -08001437
1438 if (!fail_page_alloc.ignore_gfp_wait_file ||
Akinobu Mita54114992007-07-15 23:40:23 -07001439 !fail_page_alloc.ignore_gfp_highmem_file ||
1440 !fail_page_alloc.min_order_file) {
Akinobu Mita933e3122006-12-08 02:39:45 -08001441 err = -ENOMEM;
1442 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1443 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
Akinobu Mita54114992007-07-15 23:40:23 -07001444 debugfs_remove(fail_page_alloc.min_order_file);
Akinobu Mita933e3122006-12-08 02:39:45 -08001445 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1446 }
1447
1448 return err;
1449}
1450
1451late_initcall(fail_page_alloc_debugfs);
1452
1453#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1454
1455#else /* CONFIG_FAIL_PAGE_ALLOC */
1456
1457static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1458{
1459 return 0;
1460}
1461
1462#endif /* CONFIG_FAIL_PAGE_ALLOC */
1463
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464/*
Mel Gorman88f5acf2011-01-13 15:45:41 -08001465 * Return true if free pages are above 'mark'. This takes into account the order
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 * of the allocation.
1467 */
Mel Gorman88f5acf2011-01-13 15:45:41 -08001468static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1469 int classzone_idx, int alloc_flags, long free_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470{
1471 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001472 long min = mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 int o;
1474
Mel Gorman88f5acf2011-01-13 15:45:41 -08001475 free_pages -= (1 << order) + 1;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001476 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001478 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 min -= min / 4;
1480
1481 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
Mel Gorman88f5acf2011-01-13 15:45:41 -08001482 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 for (o = 0; o < order; o++) {
1484 /* At the next order, this order's pages become unavailable */
1485 free_pages -= z->free_area[o].nr_free << o;
1486
1487 /* Require fewer higher order pages to be free */
1488 min >>= 1;
1489
1490 if (free_pages <= min)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001491 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 }
Mel Gorman88f5acf2011-01-13 15:45:41 -08001493 return true;
1494}
1495
1496bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1497 int classzone_idx, int alloc_flags)
1498{
1499 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1500 zone_page_state(z, NR_FREE_PAGES));
1501}
1502
1503bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
1504 int classzone_idx, int alloc_flags)
1505{
1506 long free_pages = zone_page_state(z, NR_FREE_PAGES);
1507
1508 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1509 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1510
1511 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1512 free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513}
1514
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001515#ifdef CONFIG_NUMA
1516/*
1517 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1518 * skip over zones that are not allowed by the cpuset, or that have
1519 * been recently (in last second) found to be nearly full. See further
1520 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001521 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001522 *
1523 * If the zonelist cache is present in the passed in zonelist, then
1524 * returns a pointer to the allowed node mask (either the current
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001525 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001526 *
1527 * If the zonelist cache is not available for this zonelist, does
1528 * nothing and returns NULL.
1529 *
1530 * If the fullzones BITMAP in the zonelist cache is stale (more than
1531 * a second since last zap'd) then we zap it out (clear its bits.)
1532 *
1533 * We hold off even calling zlc_setup, until after we've checked the
1534 * first zone in the zonelist, on the theory that most allocations will
1535 * be satisfied from that first zone, so best to examine that zone as
1536 * quickly as we can.
1537 */
1538static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1539{
1540 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1541 nodemask_t *allowednodes; /* zonelist_cache approximation */
1542
1543 zlc = zonelist->zlcache_ptr;
1544 if (!zlc)
1545 return NULL;
1546
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001547 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001548 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1549 zlc->last_full_zap = jiffies;
1550 }
1551
1552 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1553 &cpuset_current_mems_allowed :
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001554 &node_states[N_HIGH_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001555 return allowednodes;
1556}
1557
1558/*
1559 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1560 * if it is worth looking at further for free memory:
1561 * 1) Check that the zone isn't thought to be full (doesn't have its
1562 * bit set in the zonelist_cache fullzones BITMAP).
1563 * 2) Check that the zones node (obtained from the zonelist_cache
1564 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1565 * Return true (non-zero) if zone is worth looking at further, or
1566 * else return false (zero) if it is not.
1567 *
1568 * This check -ignores- the distinction between various watermarks,
1569 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1570 * found to be full for any variation of these watermarks, it will
1571 * be considered full for up to one second by all requests, unless
1572 * we are so low on memory on all allowed nodes that we are forced
1573 * into the second scan of the zonelist.
1574 *
1575 * In the second scan we ignore this zonelist cache and exactly
1576 * apply the watermarks to all zones, even it is slower to do so.
1577 * We are low on memory in the second scan, and should leave no stone
1578 * unturned looking for a free page.
1579 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001580static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001581 nodemask_t *allowednodes)
1582{
1583 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1584 int i; /* index of *z in zonelist zones */
1585 int n; /* node that zone *z is on */
1586
1587 zlc = zonelist->zlcache_ptr;
1588 if (!zlc)
1589 return 1;
1590
Mel Gormandd1a2392008-04-28 02:12:17 -07001591 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001592 n = zlc->z_to_n[i];
1593
1594 /* This zone is worth trying if it is allowed but not full */
1595 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1596}
1597
1598/*
1599 * Given 'z' scanning a zonelist, set the corresponding bit in
1600 * zlc->fullzones, so that subsequent attempts to allocate a page
1601 * from that zone don't waste time re-examining it.
1602 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001603static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001604{
1605 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1606 int i; /* index of *z in zonelist zones */
1607
1608 zlc = zonelist->zlcache_ptr;
1609 if (!zlc)
1610 return;
1611
Mel Gormandd1a2392008-04-28 02:12:17 -07001612 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001613
1614 set_bit(i, zlc->fullzones);
1615}
1616
1617#else /* CONFIG_NUMA */
1618
1619static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1620{
1621 return NULL;
1622}
1623
Mel Gormandd1a2392008-04-28 02:12:17 -07001624static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001625 nodemask_t *allowednodes)
1626{
1627 return 1;
1628}
1629
Mel Gormandd1a2392008-04-28 02:12:17 -07001630static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001631{
1632}
1633#endif /* CONFIG_NUMA */
1634
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001635/*
Paul Jackson0798e512006-12-06 20:31:38 -08001636 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001637 * a page.
1638 */
1639static struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001640get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
Mel Gorman5117f452009-06-16 15:31:59 -07001641 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
Mel Gorman3dd28262009-06-16 15:32:00 -07001642 struct zone *preferred_zone, int migratetype)
Martin Hicks753ee722005-06-21 17:14:41 -07001643{
Mel Gormandd1a2392008-04-28 02:12:17 -07001644 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001645 struct page *page = NULL;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001646 int classzone_idx;
Mel Gorman5117f452009-06-16 15:31:59 -07001647 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001648 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1649 int zlc_active = 0; /* set if using zonelist_cache */
1650 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gorman54a6eb52008-04-28 02:12:16 -07001651
Mel Gorman19770b32008-04-28 02:12:18 -07001652 classzone_idx = zone_idx(preferred_zone);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001653zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001654 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001655 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001656 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1657 */
Mel Gorman19770b32008-04-28 02:12:18 -07001658 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1659 high_zoneidx, nodemask) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001660 if (NUMA_BUILD && zlc_active &&
1661 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1662 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001663 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001664 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001665 goto try_next_zone;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001666
Mel Gorman41858962009-06-16 15:32:12 -07001667 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001668 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001669 unsigned long mark;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001670 int ret;
1671
Mel Gorman41858962009-06-16 15:32:12 -07001672 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
Mel Gormanfa5e0842009-06-16 15:33:22 -07001673 if (zone_watermark_ok(zone, order, mark,
1674 classzone_idx, alloc_flags))
1675 goto try_this_zone;
1676
1677 if (zone_reclaim_mode == 0)
1678 goto this_zone_full;
1679
1680 ret = zone_reclaim(zone, gfp_mask, order);
1681 switch (ret) {
1682 case ZONE_RECLAIM_NOSCAN:
1683 /* did not scan */
1684 goto try_next_zone;
1685 case ZONE_RECLAIM_FULL:
1686 /* scanned but unreclaimable */
1687 goto this_zone_full;
1688 default:
1689 /* did we reclaim enough */
1690 if (!zone_watermark_ok(zone, order, mark,
1691 classzone_idx, alloc_flags))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001692 goto this_zone_full;
Paul Jackson0798e512006-12-06 20:31:38 -08001693 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001694 }
1695
Mel Gormanfa5e0842009-06-16 15:33:22 -07001696try_this_zone:
Mel Gorman3dd28262009-06-16 15:32:00 -07001697 page = buffered_rmqueue(preferred_zone, zone, order,
1698 gfp_mask, migratetype);
Paul Jackson0798e512006-12-06 20:31:38 -08001699 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001700 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001701this_zone_full:
1702 if (NUMA_BUILD)
1703 zlc_mark_zone_full(zonelist, z);
1704try_next_zone:
Christoph Lameter62bc62a2009-06-16 15:32:15 -07001705 if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) {
Mel Gormand395b732009-06-16 15:32:09 -07001706 /*
1707 * we do zlc_setup after the first zone is tried but only
1708 * if there are multiple nodes make it worthwhile
1709 */
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001710 allowednodes = zlc_setup(zonelist, alloc_flags);
1711 zlc_active = 1;
1712 did_zlc_setup = 1;
1713 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07001714 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001715
1716 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1717 /* Disable zlc cache for second zonelist scan */
1718 zlc_active = 0;
1719 goto zonelist_scan;
1720 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001721 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07001722}
1723
David Rientjes29423e72011-03-22 16:30:47 -07001724/*
1725 * Large machines with many possible nodes should not always dump per-node
1726 * meminfo in irq context.
1727 */
1728static inline bool should_suppress_show_mem(void)
1729{
1730 bool ret = false;
1731
1732#if NODES_SHIFT > 8
1733 ret = in_interrupt();
1734#endif
1735 return ret;
1736}
1737
Mel Gorman11e33f62009-06-16 15:31:57 -07001738static inline int
1739should_alloc_retry(gfp_t gfp_mask, unsigned int order,
1740 unsigned long pages_reclaimed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741{
Mel Gorman11e33f62009-06-16 15:31:57 -07001742 /* Do not loop if specifically requested */
1743 if (gfp_mask & __GFP_NORETRY)
1744 return 0;
1745
1746 /*
1747 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1748 * means __GFP_NOFAIL, but that may not be true in other
1749 * implementations.
1750 */
1751 if (order <= PAGE_ALLOC_COSTLY_ORDER)
1752 return 1;
1753
1754 /*
1755 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1756 * specified, then we retry until we no longer reclaim any pages
1757 * (above), or we've reclaimed an order of pages at least as
1758 * large as the allocation's order. In both cases, if the
1759 * allocation still fails, we stop retrying.
1760 */
1761 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
1762 return 1;
1763
1764 /*
1765 * Don't let big-order allocations loop unless the caller
1766 * explicitly requests that.
1767 */
1768 if (gfp_mask & __GFP_NOFAIL)
1769 return 1;
1770
1771 return 0;
1772}
1773
1774static inline struct page *
1775__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1776 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001777 nodemask_t *nodemask, struct zone *preferred_zone,
1778 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001779{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781
Mel Gorman11e33f62009-06-16 15:31:57 -07001782 /* Acquire the OOM killer lock for the zones in zonelist */
Minchan Kimff321fe2010-08-09 17:18:57 -07001783 if (!try_set_zonelist_oom(zonelist, gfp_mask)) {
Mel Gorman11e33f62009-06-16 15:31:57 -07001784 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 return NULL;
1786 }
Jens Axboe6b1de912005-11-17 21:35:02 +01001787
Mel Gorman11e33f62009-06-16 15:31:57 -07001788 /*
1789 * Go through the zonelist yet one more time, keep very high watermark
1790 * here, this is only to catch a parallel oom killing, we must fail if
1791 * we're still under heavy pressure.
1792 */
1793 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1794 order, zonelist, high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001795 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07001796 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001797 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07001798 goto out;
1799
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001800 if (!(gfp_mask & __GFP_NOFAIL)) {
1801 /* The OOM killer will not help higher order allocs */
1802 if (order > PAGE_ALLOC_COSTLY_ORDER)
1803 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07001804 /* The OOM killer does not needlessly kill tasks for lowmem */
1805 if (high_zoneidx < ZONE_NORMAL)
1806 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001807 /*
1808 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
1809 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
1810 * The caller should handle page allocation failure by itself if
1811 * it specifies __GFP_THISNODE.
1812 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
1813 */
1814 if (gfp_mask & __GFP_THISNODE)
1815 goto out;
1816 }
Mel Gorman11e33f62009-06-16 15:31:57 -07001817 /* Exhausted what can be done so it's blamo time */
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001818 out_of_memory(zonelist, gfp_mask, order, nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07001819
1820out:
1821 clear_zonelist_oom(zonelist, gfp_mask);
1822 return page;
1823}
1824
Mel Gorman56de7262010-05-24 14:32:30 -07001825#ifdef CONFIG_COMPACTION
1826/* Try memory compaction for high-order allocations before reclaim */
1827static struct page *
1828__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
1829 struct zonelist *zonelist, enum zone_type high_zoneidx,
1830 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman77f1fe62011-01-13 15:45:57 -08001831 int migratetype, unsigned long *did_some_progress,
1832 bool sync_migration)
Mel Gorman56de7262010-05-24 14:32:30 -07001833{
1834 struct page *page;
1835
Mel Gorman4f92e252010-05-24 14:32:32 -07001836 if (!order || compaction_deferred(preferred_zone))
Mel Gorman56de7262010-05-24 14:32:30 -07001837 return NULL;
1838
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001839 current->flags |= PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07001840 *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
Mel Gorman77f1fe62011-01-13 15:45:57 -08001841 nodemask, sync_migration);
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001842 current->flags &= ~PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07001843 if (*did_some_progress != COMPACT_SKIPPED) {
1844
1845 /* Page migration frees to the PCP lists but we want merging */
1846 drain_pages(get_cpu());
1847 put_cpu();
1848
1849 page = get_page_from_freelist(gfp_mask, nodemask,
1850 order, zonelist, high_zoneidx,
1851 alloc_flags, preferred_zone,
1852 migratetype);
1853 if (page) {
Mel Gorman4f92e252010-05-24 14:32:32 -07001854 preferred_zone->compact_considered = 0;
1855 preferred_zone->compact_defer_shift = 0;
Mel Gorman56de7262010-05-24 14:32:30 -07001856 count_vm_event(COMPACTSUCCESS);
1857 return page;
1858 }
1859
1860 /*
1861 * It's bad if compaction run occurs and fails.
1862 * The most likely reason is that pages exist,
1863 * but not enough to satisfy watermarks.
1864 */
1865 count_vm_event(COMPACTFAIL);
Mel Gorman4f92e252010-05-24 14:32:32 -07001866 defer_compaction(preferred_zone);
Mel Gorman56de7262010-05-24 14:32:30 -07001867
1868 cond_resched();
1869 }
1870
1871 return NULL;
1872}
1873#else
1874static inline struct page *
1875__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
1876 struct zonelist *zonelist, enum zone_type high_zoneidx,
1877 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman77f1fe62011-01-13 15:45:57 -08001878 int migratetype, unsigned long *did_some_progress,
1879 bool sync_migration)
Mel Gorman56de7262010-05-24 14:32:30 -07001880{
1881 return NULL;
1882}
1883#endif /* CONFIG_COMPACTION */
1884
Mel Gorman11e33f62009-06-16 15:31:57 -07001885/* The really slow allocator path where we enter direct reclaim */
1886static inline struct page *
1887__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
1888 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001889 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001890 int migratetype, unsigned long *did_some_progress)
Mel Gorman11e33f62009-06-16 15:31:57 -07001891{
1892 struct page *page = NULL;
1893 struct reclaim_state reclaim_state;
Mel Gorman9ee493c2010-09-09 16:38:18 -07001894 bool drained = false;
Mel Gorman11e33f62009-06-16 15:31:57 -07001895
1896 cond_resched();
1897
1898 /* We now go into synchronous reclaim */
1899 cpuset_memory_pressure_bump();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001900 current->flags |= PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07001901 lockdep_set_current_reclaim_state(gfp_mask);
1902 reclaim_state.reclaimed_slab = 0;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001903 current->reclaim_state = &reclaim_state;
Mel Gorman11e33f62009-06-16 15:31:57 -07001904
1905 *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
1906
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001907 current->reclaim_state = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07001908 lockdep_clear_current_reclaim_state();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001909 current->flags &= ~PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07001910
1911 cond_resched();
1912
Mel Gorman9ee493c2010-09-09 16:38:18 -07001913 if (unlikely(!(*did_some_progress)))
1914 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07001915
Mel Gorman9ee493c2010-09-09 16:38:18 -07001916retry:
1917 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001918 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001919 alloc_flags, preferred_zone,
1920 migratetype);
Mel Gorman9ee493c2010-09-09 16:38:18 -07001921
1922 /*
1923 * If an allocation failed after direct reclaim, it could be because
1924 * pages are pinned on the per-cpu lists. Drain them and try again
1925 */
1926 if (!page && !drained) {
1927 drain_all_pages();
1928 drained = true;
1929 goto retry;
1930 }
1931
Mel Gorman11e33f62009-06-16 15:31:57 -07001932 return page;
1933}
1934
Mel Gorman11e33f62009-06-16 15:31:57 -07001935/*
1936 * This is called in the allocator slow-path if the allocation request is of
1937 * sufficient urgency to ignore watermarks and take other desperate measures
1938 */
1939static inline struct page *
1940__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
1941 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001942 nodemask_t *nodemask, struct zone *preferred_zone,
1943 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001944{
1945 struct page *page;
1946
1947 do {
1948 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001949 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
Mel Gorman3dd28262009-06-16 15:32:00 -07001950 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001951
1952 if (!page && gfp_mask & __GFP_NOFAIL)
Mel Gorman0e093d992010-10-26 14:21:45 -07001953 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07001954 } while (!page && (gfp_mask & __GFP_NOFAIL));
1955
1956 return page;
1957}
1958
1959static inline
1960void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
Mel Gorman99504742011-01-13 15:46:20 -08001961 enum zone_type high_zoneidx,
1962 enum zone_type classzone_idx)
Mel Gorman11e33f62009-06-16 15:31:57 -07001963{
1964 struct zoneref *z;
1965 struct zone *zone;
1966
1967 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
Mel Gorman99504742011-01-13 15:46:20 -08001968 wakeup_kswapd(zone, order, classzone_idx);
Mel Gorman11e33f62009-06-16 15:31:57 -07001969}
1970
Peter Zijlstra341ce062009-06-16 15:32:02 -07001971static inline int
1972gfp_to_alloc_flags(gfp_t gfp_mask)
1973{
Peter Zijlstra341ce062009-06-16 15:32:02 -07001974 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
1975 const gfp_t wait = gfp_mask & __GFP_WAIT;
1976
Mel Gormana56f57f2009-06-16 15:32:02 -07001977 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07001978 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07001979
Peter Zijlstra341ce062009-06-16 15:32:02 -07001980 /*
1981 * The caller may dip into page reserves a bit more if the caller
1982 * cannot run direct reclaim, or if the caller has realtime scheduling
1983 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1984 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1985 */
Namhyung Kime6223a32010-10-26 14:21:59 -07001986 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07001987
1988 if (!wait) {
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08001989 /*
1990 * Not worth trying to allocate harder for
1991 * __GFP_NOMEMALLOC even if it can't schedule.
1992 */
1993 if (!(gfp_mask & __GFP_NOMEMALLOC))
1994 alloc_flags |= ALLOC_HARDER;
Peter Zijlstra341ce062009-06-16 15:32:02 -07001995 /*
1996 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1997 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1998 */
1999 alloc_flags &= ~ALLOC_CPUSET;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002000 } else if (unlikely(rt_task(current)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07002001 alloc_flags |= ALLOC_HARDER;
2002
2003 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2004 if (!in_interrupt() &&
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002005 ((current->flags & PF_MEMALLOC) ||
Peter Zijlstra341ce062009-06-16 15:32:02 -07002006 unlikely(test_thread_flag(TIF_MEMDIE))))
2007 alloc_flags |= ALLOC_NO_WATERMARKS;
2008 }
2009
2010 return alloc_flags;
2011}
2012
Mel Gorman11e33f62009-06-16 15:31:57 -07002013static inline struct page *
2014__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
2015 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002016 nodemask_t *nodemask, struct zone *preferred_zone,
2017 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002018{
2019 const gfp_t wait = gfp_mask & __GFP_WAIT;
2020 struct page *page = NULL;
2021 int alloc_flags;
2022 unsigned long pages_reclaimed = 0;
2023 unsigned long did_some_progress;
Mel Gorman77f1fe62011-01-13 15:45:57 -08002024 bool sync_migration = false;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002025
Christoph Lameter952f3b52006-12-06 20:33:26 -08002026 /*
Mel Gorman72807a72009-06-16 15:32:18 -07002027 * In the slowpath, we sanity check order to avoid ever trying to
2028 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2029 * be using allocators in order of preference for an area that is
2030 * too large.
2031 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07002032 if (order >= MAX_ORDER) {
2033 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07002034 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07002035 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002036
Christoph Lameter952f3b52006-12-06 20:33:26 -08002037 /*
2038 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2039 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2040 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2041 * using a larger set of nodes after it has established that the
2042 * allowed per node queues are empty and that nodes are
2043 * over allocated.
2044 */
2045 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
2046 goto nopage;
2047
Mel Gormancc4a6852009-11-11 14:26:14 -08002048restart:
Andrea Arcangeli32dba982011-01-13 15:46:49 -08002049 if (!(gfp_mask & __GFP_NO_KSWAPD))
2050 wake_all_kswapd(order, zonelist, high_zoneidx,
Mel Gorman99504742011-01-13 15:46:20 -08002051 zone_idx(preferred_zone));
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002052
Paul Jackson9bf22292005-09-06 15:18:12 -07002053 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002054 * OK, we're below the kswapd watermark and have kicked background
2055 * reclaim. Now things get more complex, so set up alloc_flags according
2056 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07002057 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002058 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059
David Rientjesf33261d2011-01-25 15:07:20 -08002060 /*
2061 * Find the true preferred zone if the allocation is unconstrained by
2062 * cpusets.
2063 */
2064 if (!(alloc_flags & ALLOC_CPUSET) && !nodemask)
2065 first_zones_zonelist(zonelist, high_zoneidx, NULL,
2066 &preferred_zone);
2067
Peter Zijlstra341ce062009-06-16 15:32:02 -07002068 /* This is the last chance, in general, before the goto nopage. */
Mel Gorman19770b32008-04-28 02:12:18 -07002069 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
Peter Zijlstra341ce062009-06-16 15:32:02 -07002070 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
2071 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002072 if (page)
2073 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074
Kirill Korotaevb43a57b2006-12-06 20:32:27 -08002075rebalance:
Mel Gorman11e33f62009-06-16 15:31:57 -07002076 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002077 if (alloc_flags & ALLOC_NO_WATERMARKS) {
2078 page = __alloc_pages_high_priority(gfp_mask, order,
2079 zonelist, high_zoneidx, nodemask,
2080 preferred_zone, migratetype);
2081 if (page)
2082 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 }
2084
2085 /* Atomic allocations - we can't balance anything */
2086 if (!wait)
2087 goto nopage;
2088
Peter Zijlstra341ce062009-06-16 15:32:02 -07002089 /* Avoid recursion of direct reclaim */
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002090 if (current->flags & PF_MEMALLOC)
Peter Zijlstra341ce062009-06-16 15:32:02 -07002091 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092
David Rientjes6583bb62009-07-29 15:02:06 -07002093 /* Avoid allocations with no watermarks from looping endlessly */
2094 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2095 goto nopage;
2096
Mel Gorman77f1fe62011-01-13 15:45:57 -08002097 /*
2098 * Try direct compaction. The first pass is asynchronous. Subsequent
2099 * attempts after direct reclaim are synchronous
2100 */
Mel Gorman56de7262010-05-24 14:32:30 -07002101 page = __alloc_pages_direct_compact(gfp_mask, order,
2102 zonelist, high_zoneidx,
2103 nodemask,
2104 alloc_flags, preferred_zone,
Mel Gorman77f1fe62011-01-13 15:45:57 -08002105 migratetype, &did_some_progress,
2106 sync_migration);
Mel Gorman56de7262010-05-24 14:32:30 -07002107 if (page)
2108 goto got_pg;
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -07002109 sync_migration = !(gfp_mask & __GFP_NO_KSWAPD);
Mel Gorman56de7262010-05-24 14:32:30 -07002110
Mel Gorman11e33f62009-06-16 15:31:57 -07002111 /* Try direct reclaim and then allocating */
2112 page = __alloc_pages_direct_reclaim(gfp_mask, order,
2113 zonelist, high_zoneidx,
2114 nodemask,
Mel Gorman5117f452009-06-16 15:31:59 -07002115 alloc_flags, preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07002116 migratetype, &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07002117 if (page)
2118 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120 /*
Mel Gorman11e33f62009-06-16 15:31:57 -07002121 * If we failed to make any progress reclaiming, then we are
2122 * running out of options and have to consider going OOM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 */
Mel Gorman11e33f62009-06-16 15:31:57 -07002124 if (!did_some_progress) {
2125 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -07002126 if (oom_killer_disabled)
2127 goto nopage;
Mel Gorman11e33f62009-06-16 15:31:57 -07002128 page = __alloc_pages_may_oom(gfp_mask, order,
2129 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002130 nodemask, preferred_zone,
2131 migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002132 if (page)
2133 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134
David Rientjes03668b32010-08-09 17:18:54 -07002135 if (!(gfp_mask & __GFP_NOFAIL)) {
2136 /*
2137 * The oom killer is not called for high-order
2138 * allocations that may fail, so if no progress
2139 * is being made, there are no other options and
2140 * retrying is unlikely to help.
2141 */
2142 if (order > PAGE_ALLOC_COSTLY_ORDER)
2143 goto nopage;
2144 /*
2145 * The oom killer is not called for lowmem
2146 * allocations to prevent needlessly killing
2147 * innocent tasks.
2148 */
2149 if (high_zoneidx < ZONE_NORMAL)
2150 goto nopage;
2151 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 goto restart;
2154 }
2155 }
2156
Mel Gorman11e33f62009-06-16 15:31:57 -07002157 /* Check if we should retry the allocation */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07002158 pages_reclaimed += did_some_progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07002159 if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
2160 /* Wait for some write requests to complete then retry */
Mel Gorman0e093d992010-10-26 14:21:45 -07002161 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 goto rebalance;
Mel Gorman3e7d3442011-01-13 15:45:56 -08002163 } else {
2164 /*
2165 * High-order allocations do not necessarily loop after
2166 * direct reclaim and reclaim/compaction depends on compaction
2167 * being called after reclaim so call directly if necessary
2168 */
2169 page = __alloc_pages_direct_compact(gfp_mask, order,
2170 zonelist, high_zoneidx,
2171 nodemask,
2172 alloc_flags, preferred_zone,
Mel Gorman77f1fe62011-01-13 15:45:57 -08002173 migratetype, &did_some_progress,
2174 sync_migration);
Mel Gorman3e7d3442011-01-13 15:45:56 -08002175 if (page)
2176 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 }
2178
2179nopage:
2180 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
David Rientjescbf978b2011-03-22 16:30:48 -07002181 unsigned int filter = SHOW_MEM_FILTER_NODES;
2182
2183 /*
2184 * This documents exceptions given to allocations in certain
2185 * contexts that are allowed to allocate outside current's set
2186 * of allowed nodes.
2187 */
2188 if (!(gfp_mask & __GFP_NOMEMALLOC))
2189 if (test_thread_flag(TIF_MEMDIE) ||
2190 (current->flags & (PF_MEMALLOC | PF_EXITING)))
2191 filter &= ~SHOW_MEM_FILTER_NODES;
2192 if (in_interrupt() || !wait)
2193 filter &= ~SHOW_MEM_FILTER_NODES;
2194
2195 pr_warning("%s: page allocation failure. order:%d, mode:0x%x\n",
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002196 current->comm, order, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197 dump_stack();
David Rientjes29423e72011-03-22 16:30:47 -07002198 if (!should_suppress_show_mem())
David Rientjesb2b755b2011-03-24 15:18:15 -07002199 show_mem(filter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 }
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002201 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202got_pg:
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002203 if (kmemcheck_enabled)
2204 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 return page;
Mel Gorman11e33f62009-06-16 15:31:57 -07002206
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207}
Mel Gorman11e33f62009-06-16 15:31:57 -07002208
2209/*
2210 * This is the 'heart' of the zoned buddy allocator.
2211 */
2212struct page *
2213__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2214 struct zonelist *zonelist, nodemask_t *nodemask)
2215{
2216 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Mel Gorman5117f452009-06-16 15:31:59 -07002217 struct zone *preferred_zone;
Mel Gorman11e33f62009-06-16 15:31:57 -07002218 struct page *page;
Mel Gorman3dd28262009-06-16 15:32:00 -07002219 int migratetype = allocflags_to_migratetype(gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002220
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10002221 gfp_mask &= gfp_allowed_mask;
2222
Mel Gorman11e33f62009-06-16 15:31:57 -07002223 lockdep_trace_alloc(gfp_mask);
2224
2225 might_sleep_if(gfp_mask & __GFP_WAIT);
2226
2227 if (should_fail_alloc_page(gfp_mask, order))
2228 return NULL;
2229
2230 /*
2231 * Check the zones suitable for the gfp_mask contain at least one
2232 * valid zone. It's possible to have an empty zonelist as a result
2233 * of GFP_THISNODE and a memoryless node
2234 */
2235 if (unlikely(!zonelist->_zonerefs->zone))
2236 return NULL;
2237
Miao Xiec0ff7452010-05-24 14:32:08 -07002238 get_mems_allowed();
Mel Gorman5117f452009-06-16 15:31:59 -07002239 /* The preferred zone is used for statistics later */
David Rientjesf33261d2011-01-25 15:07:20 -08002240 first_zones_zonelist(zonelist, high_zoneidx,
2241 nodemask ? : &cpuset_current_mems_allowed,
2242 &preferred_zone);
Miao Xiec0ff7452010-05-24 14:32:08 -07002243 if (!preferred_zone) {
2244 put_mems_allowed();
Mel Gorman5117f452009-06-16 15:31:59 -07002245 return NULL;
Miao Xiec0ff7452010-05-24 14:32:08 -07002246 }
Mel Gorman5117f452009-06-16 15:31:59 -07002247
2248 /* First allocation attempt */
Mel Gorman11e33f62009-06-16 15:31:57 -07002249 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002250 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07002251 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002252 if (unlikely(!page))
2253 page = __alloc_pages_slowpath(gfp_mask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002254 zonelist, high_zoneidx, nodemask,
Mel Gorman3dd28262009-06-16 15:32:00 -07002255 preferred_zone, migratetype);
Miao Xiec0ff7452010-05-24 14:32:08 -07002256 put_mems_allowed();
Mel Gorman11e33f62009-06-16 15:31:57 -07002257
Mel Gorman4b4f2782009-09-21 17:02:41 -07002258 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002259 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260}
Mel Gormand2391712009-06-16 15:31:52 -07002261EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262
2263/*
2264 * Common helper functions.
2265 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08002266unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267{
Akinobu Mita945a1112009-09-21 17:01:47 -07002268 struct page *page;
2269
2270 /*
2271 * __get_free_pages() returns a 32-bit address, which cannot represent
2272 * a highmem page
2273 */
2274 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2275
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276 page = alloc_pages(gfp_mask, order);
2277 if (!page)
2278 return 0;
2279 return (unsigned long) page_address(page);
2280}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281EXPORT_SYMBOL(__get_free_pages);
2282
Harvey Harrison920c7a52008-02-04 22:29:26 -08002283unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284{
Akinobu Mita945a1112009-09-21 17:01:47 -07002285 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287EXPORT_SYMBOL(get_zeroed_page);
2288
2289void __pagevec_free(struct pagevec *pvec)
2290{
2291 int i = pagevec_count(pvec);
2292
Mel Gorman4b4f2782009-09-21 17:02:41 -07002293 while (--i >= 0) {
2294 trace_mm_pagevec_free(pvec->pages[i], pvec->cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 free_hot_cold_page(pvec->pages[i], pvec->cold);
Mel Gorman4b4f2782009-09-21 17:02:41 -07002296 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297}
2298
Harvey Harrison920c7a52008-02-04 22:29:26 -08002299void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300{
Nick Pigginb5810032005-10-29 18:16:12 -07002301 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302 if (order == 0)
Li Hongfc916682010-03-05 13:41:54 -08002303 free_hot_cold_page(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 else
2305 __free_pages_ok(page, order);
2306 }
2307}
2308
2309EXPORT_SYMBOL(__free_pages);
2310
Harvey Harrison920c7a52008-02-04 22:29:26 -08002311void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312{
2313 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07002314 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 __free_pages(virt_to_page((void *)addr), order);
2316 }
2317}
2318
2319EXPORT_SYMBOL(free_pages);
2320
Andi Kleenee85c2e2011-05-11 15:13:34 -07002321static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
2322{
2323 if (addr) {
2324 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2325 unsigned long used = addr + PAGE_ALIGN(size);
2326
2327 split_page(virt_to_page((void *)addr), order);
2328 while (used < alloc_end) {
2329 free_page(used);
2330 used += PAGE_SIZE;
2331 }
2332 }
2333 return (void *)addr;
2334}
2335
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002336/**
2337 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2338 * @size: the number of bytes to allocate
2339 * @gfp_mask: GFP flags for the allocation
2340 *
2341 * This function is similar to alloc_pages(), except that it allocates the
2342 * minimum number of pages to satisfy the request. alloc_pages() can only
2343 * allocate memory in power-of-two pages.
2344 *
2345 * This function is also limited by MAX_ORDER.
2346 *
2347 * Memory allocated by this function must be released by free_pages_exact().
2348 */
2349void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2350{
2351 unsigned int order = get_order(size);
2352 unsigned long addr;
2353
2354 addr = __get_free_pages(gfp_mask, order);
Andi Kleenee85c2e2011-05-11 15:13:34 -07002355 return make_alloc_exact(addr, order, size);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002356}
2357EXPORT_SYMBOL(alloc_pages_exact);
2358
2359/**
Andi Kleenee85c2e2011-05-11 15:13:34 -07002360 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2361 * pages on a node.
Randy Dunlapb5e6ab52011-05-16 13:16:54 -07002362 * @nid: the preferred node ID where memory should be allocated
Andi Kleenee85c2e2011-05-11 15:13:34 -07002363 * @size: the number of bytes to allocate
2364 * @gfp_mask: GFP flags for the allocation
2365 *
2366 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2367 * back.
2368 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2369 * but is not exact.
2370 */
2371void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
2372{
2373 unsigned order = get_order(size);
2374 struct page *p = alloc_pages_node(nid, gfp_mask, order);
2375 if (!p)
2376 return NULL;
2377 return make_alloc_exact((unsigned long)page_address(p), order, size);
2378}
2379EXPORT_SYMBOL(alloc_pages_exact_nid);
2380
2381/**
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002382 * free_pages_exact - release memory allocated via alloc_pages_exact()
2383 * @virt: the value returned by alloc_pages_exact.
2384 * @size: size of allocation, same value as passed to alloc_pages_exact().
2385 *
2386 * Release the memory allocated by a previous call to alloc_pages_exact.
2387 */
2388void free_pages_exact(void *virt, size_t size)
2389{
2390 unsigned long addr = (unsigned long)virt;
2391 unsigned long end = addr + PAGE_ALIGN(size);
2392
2393 while (addr < end) {
2394 free_page(addr);
2395 addr += PAGE_SIZE;
2396 }
2397}
2398EXPORT_SYMBOL(free_pages_exact);
2399
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400static unsigned int nr_free_zone_pages(int offset)
2401{
Mel Gormandd1a2392008-04-28 02:12:17 -07002402 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002403 struct zone *zone;
2404
Martin J. Blighe310fd42005-07-29 22:59:18 -07002405 /* Just pick one node, since fallback list is circular */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406 unsigned int sum = 0;
2407
Mel Gorman0e884602008-04-28 02:12:14 -07002408 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409
Mel Gorman54a6eb52008-04-28 02:12:16 -07002410 for_each_zone_zonelist(zone, z, zonelist, offset) {
Martin J. Blighe310fd42005-07-29 22:59:18 -07002411 unsigned long size = zone->present_pages;
Mel Gorman41858962009-06-16 15:32:12 -07002412 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07002413 if (size > high)
2414 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415 }
2416
2417 return sum;
2418}
2419
2420/*
2421 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2422 */
2423unsigned int nr_free_buffer_pages(void)
2424{
Al Viroaf4ca452005-10-21 02:55:38 -04002425 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426}
Meelap Shahc2f1a552007-07-17 04:04:39 -07002427EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428
2429/*
2430 * Amount of free RAM allocatable within all zones
2431 */
2432unsigned int nr_free_pagecache_pages(void)
2433{
Mel Gorman2a1e2742007-07-17 04:03:12 -07002434 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002436
2437static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438{
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002439 if (NUMA_BUILD)
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08002440 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443void si_meminfo(struct sysinfo *val)
2444{
2445 val->totalram = totalram_pages;
2446 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002447 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449 val->totalhigh = totalhigh_pages;
2450 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451 val->mem_unit = PAGE_SIZE;
2452}
2453
2454EXPORT_SYMBOL(si_meminfo);
2455
2456#ifdef CONFIG_NUMA
2457void si_meminfo_node(struct sysinfo *val, int nid)
2458{
2459 pg_data_t *pgdat = NODE_DATA(nid);
2460
2461 val->totalram = pgdat->node_present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002462 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002463#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002465 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2466 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002467#else
2468 val->totalhigh = 0;
2469 val->freehigh = 0;
2470#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471 val->mem_unit = PAGE_SIZE;
2472}
2473#endif
2474
David Rientjesddd588b2011-03-22 16:30:46 -07002475/*
2476 * Determine whether the zone's node should be displayed or not, depending on
2477 * whether SHOW_MEM_FILTER_NODES was passed to __show_free_areas().
2478 */
2479static bool skip_free_areas_zone(unsigned int flags, const struct zone *zone)
2480{
2481 bool ret = false;
2482
2483 if (!(flags & SHOW_MEM_FILTER_NODES))
2484 goto out;
2485
2486 get_mems_allowed();
2487 ret = !node_isset(zone->zone_pgdat->node_id,
2488 cpuset_current_mems_allowed);
2489 put_mems_allowed();
2490out:
2491 return ret;
2492}
2493
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494#define K(x) ((x) << (PAGE_SHIFT-10))
2495
2496/*
2497 * Show free area list (used inside shift_scroll-lock stuff)
2498 * We also calculate the percentage fragmentation. We do this by counting the
2499 * memory on each free list with the exception of the first item on the list.
David Rientjesddd588b2011-03-22 16:30:46 -07002500 * Suppresses nodes that are not allowed by current's cpuset if
2501 * SHOW_MEM_FILTER_NODES is passed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502 */
David Rientjesddd588b2011-03-22 16:30:46 -07002503void __show_free_areas(unsigned int filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504{
Jes Sorensenc7241912006-09-27 01:50:05 -07002505 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506 struct zone *zone;
2507
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002508 for_each_populated_zone(zone) {
David Rientjesddd588b2011-03-22 16:30:46 -07002509 if (skip_free_areas_zone(filter, zone))
2510 continue;
Jes Sorensenc7241912006-09-27 01:50:05 -07002511 show_node(zone);
2512 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513
Dave Jones6b482c62005-11-10 15:45:56 -05002514 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515 struct per_cpu_pageset *pageset;
2516
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002517 pageset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002519 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2520 cpu, pageset->pcp.high,
2521 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 }
2523 }
2524
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002525 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2526 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002527 " unevictable:%lu"
Andrew Mortonb76146e2009-10-26 16:49:52 -07002528 " dirty:%lu writeback:%lu unstable:%lu\n"
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07002529 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002530 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002531 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002532 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002533 global_page_state(NR_ISOLATED_ANON),
2534 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002535 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002536 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002537 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07002538 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07002539 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07002540 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08002541 global_page_state(NR_FREE_PAGES),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07002542 global_page_state(NR_SLAB_RECLAIMABLE),
2543 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07002544 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002545 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08002546 global_page_state(NR_PAGETABLE),
2547 global_page_state(NR_BOUNCE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002549 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550 int i;
2551
David Rientjesddd588b2011-03-22 16:30:46 -07002552 if (skip_free_areas_zone(filter, zone))
2553 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554 show_node(zone);
2555 printk("%s"
2556 " free:%lukB"
2557 " min:%lukB"
2558 " low:%lukB"
2559 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002560 " active_anon:%lukB"
2561 " inactive_anon:%lukB"
2562 " active_file:%lukB"
2563 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002564 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002565 " isolated(anon):%lukB"
2566 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567 " present:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002568 " mlocked:%lukB"
2569 " dirty:%lukB"
2570 " writeback:%lukB"
2571 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002572 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002573 " slab_reclaimable:%lukB"
2574 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07002575 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002576 " pagetables:%lukB"
2577 " unstable:%lukB"
2578 " bounce:%lukB"
2579 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580 " pages_scanned:%lu"
2581 " all_unreclaimable? %s"
2582 "\n",
2583 zone->name,
Mel Gorman88f5acf2011-01-13 15:45:41 -08002584 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07002585 K(min_wmark_pages(zone)),
2586 K(low_wmark_pages(zone)),
2587 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002588 K(zone_page_state(zone, NR_ACTIVE_ANON)),
2589 K(zone_page_state(zone, NR_INACTIVE_ANON)),
2590 K(zone_page_state(zone, NR_ACTIVE_FILE)),
2591 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002592 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002593 K(zone_page_state(zone, NR_ISOLATED_ANON)),
2594 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595 K(zone->present_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002596 K(zone_page_state(zone, NR_MLOCK)),
2597 K(zone_page_state(zone, NR_FILE_DIRTY)),
2598 K(zone_page_state(zone, NR_WRITEBACK)),
2599 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002600 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002601 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
2602 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07002603 zone_page_state(zone, NR_KERNEL_STACK) *
2604 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002605 K(zone_page_state(zone, NR_PAGETABLE)),
2606 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
2607 K(zone_page_state(zone, NR_BOUNCE)),
2608 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609 zone->pages_scanned,
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -08002610 (zone->all_unreclaimable ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611 );
2612 printk("lowmem_reserve[]:");
2613 for (i = 0; i < MAX_NR_ZONES; i++)
2614 printk(" %lu", zone->lowmem_reserve[i]);
2615 printk("\n");
2616 }
2617
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002618 for_each_populated_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002619 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620
David Rientjesddd588b2011-03-22 16:30:46 -07002621 if (skip_free_areas_zone(filter, zone))
2622 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 show_node(zone);
2624 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625
2626 spin_lock_irqsave(&zone->lock, flags);
2627 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002628 nr[order] = zone->free_area[order].nr_free;
2629 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630 }
2631 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002632 for (order = 0; order < MAX_ORDER; order++)
2633 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634 printk("= %lukB\n", K(total));
2635 }
2636
Larry Woodmane6f36022008-02-04 22:29:30 -08002637 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2638
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 show_swap_cache_info();
2640}
2641
David Rientjesddd588b2011-03-22 16:30:46 -07002642void show_free_areas(void)
2643{
2644 __show_free_areas(0);
2645}
2646
Mel Gorman19770b32008-04-28 02:12:18 -07002647static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2648{
2649 zoneref->zone = zone;
2650 zoneref->zone_idx = zone_idx(zone);
2651}
2652
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653/*
2654 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08002655 *
2656 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002658static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2659 int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660{
Christoph Lameter1a932052006-01-06 00:11:16 -08002661 struct zone *zone;
2662
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002663 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002664 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08002665
2666 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002667 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08002668 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08002669 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002670 zoneref_set_zone(zone,
2671 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08002672 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08002674
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002675 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08002676 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677}
2678
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002679
2680/*
2681 * zonelist_order:
2682 * 0 = automatic detection of better ordering.
2683 * 1 = order by ([node] distance, -zonetype)
2684 * 2 = order by (-zonetype, [node] distance)
2685 *
2686 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2687 * the same zonelist. So only NUMA can configure this param.
2688 */
2689#define ZONELIST_ORDER_DEFAULT 0
2690#define ZONELIST_ORDER_NODE 1
2691#define ZONELIST_ORDER_ZONE 2
2692
2693/* zonelist order in the kernel.
2694 * set_zonelist_order() will set this to NODE or ZONE.
2695 */
2696static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2697static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2698
2699
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002701/* The value user specified ....changed by config */
2702static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2703/* string for sysctl */
2704#define NUMA_ZONELIST_ORDER_LEN 16
2705char numa_zonelist_order[16] = "default";
2706
2707/*
2708 * interface for configure zonelist ordering.
2709 * command line option "numa_zonelist_order"
2710 * = "[dD]efault - default, automatic configuration.
2711 * = "[nN]ode - order by node locality, then by zone within node
2712 * = "[zZ]one - order by zone, then by locality within zone
2713 */
2714
2715static int __parse_numa_zonelist_order(char *s)
2716{
2717 if (*s == 'd' || *s == 'D') {
2718 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2719 } else if (*s == 'n' || *s == 'N') {
2720 user_zonelist_order = ZONELIST_ORDER_NODE;
2721 } else if (*s == 'z' || *s == 'Z') {
2722 user_zonelist_order = ZONELIST_ORDER_ZONE;
2723 } else {
2724 printk(KERN_WARNING
2725 "Ignoring invalid numa_zonelist_order value: "
2726 "%s\n", s);
2727 return -EINVAL;
2728 }
2729 return 0;
2730}
2731
2732static __init int setup_numa_zonelist_order(char *s)
2733{
Volodymyr G. Lukiianykecb256f2011-01-13 15:46:26 -08002734 int ret;
2735
2736 if (!s)
2737 return 0;
2738
2739 ret = __parse_numa_zonelist_order(s);
2740 if (ret == 0)
2741 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
2742
2743 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002744}
2745early_param("numa_zonelist_order", setup_numa_zonelist_order);
2746
2747/*
2748 * sysctl handler for numa_zonelist_order
2749 */
2750int numa_zonelist_order_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07002751 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002752 loff_t *ppos)
2753{
2754 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2755 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01002756 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002757
Andi Kleen443c6f12009-12-23 21:00:47 +01002758 mutex_lock(&zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002759 if (write)
Andi Kleen443c6f12009-12-23 21:00:47 +01002760 strcpy(saved_string, (char*)table->data);
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07002761 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002762 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01002763 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002764 if (write) {
2765 int oldval = user_zonelist_order;
2766 if (__parse_numa_zonelist_order((char*)table->data)) {
2767 /*
2768 * bogus value. restore saved string
2769 */
2770 strncpy((char*)table->data, saved_string,
2771 NUMA_ZONELIST_ORDER_LEN);
2772 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07002773 } else if (oldval != user_zonelist_order) {
2774 mutex_lock(&zonelists_mutex);
Haicheng Li1f522502010-05-24 14:32:51 -07002775 build_all_zonelists(NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07002776 mutex_unlock(&zonelists_mutex);
2777 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002778 }
Andi Kleen443c6f12009-12-23 21:00:47 +01002779out:
2780 mutex_unlock(&zl_order_mutex);
2781 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002782}
2783
2784
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002785#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002786static int node_load[MAX_NUMNODES];
2787
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07002789 * 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 -07002790 * @node: node whose fallback list we're appending
2791 * @used_node_mask: nodemask_t of already used nodes
2792 *
2793 * We use a number of factors to determine which is the next node that should
2794 * appear on a given node's fallback list. The node should not have appeared
2795 * already in @node's fallback list, and it should be the next closest node
2796 * according to the distance array (which contains arbitrary distance values
2797 * from each node to each node in the system), and should also prefer nodes
2798 * with no CPUs, since presumably they'll have very little allocation pressure
2799 * on them otherwise.
2800 * It returns -1 if no node is found.
2801 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002802static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002804 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805 int min_val = INT_MAX;
2806 int best_node = -1;
Rusty Russella70f7302009-03-13 14:49:46 +10302807 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002809 /* Use the local node if we haven't already */
2810 if (!node_isset(node, *used_node_mask)) {
2811 node_set(node, *used_node_mask);
2812 return node;
2813 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002815 for_each_node_state(n, N_HIGH_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816
2817 /* Don't want a node to appear more than once */
2818 if (node_isset(n, *used_node_mask))
2819 continue;
2820
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 /* Use the distance array to find the distance */
2822 val = node_distance(node, n);
2823
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002824 /* Penalize nodes under us ("prefer the next node") */
2825 val += (n < node);
2826
Linus Torvalds1da177e2005-04-16 15:20:36 -07002827 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10302828 tmp = cpumask_of_node(n);
2829 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830 val += PENALTY_FOR_NODE_WITH_CPUS;
2831
2832 /* Slight preference for less loaded node */
2833 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2834 val += node_load[n];
2835
2836 if (val < min_val) {
2837 min_val = val;
2838 best_node = n;
2839 }
2840 }
2841
2842 if (best_node >= 0)
2843 node_set(best_node, *used_node_mask);
2844
2845 return best_node;
2846}
2847
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002848
2849/*
2850 * Build zonelists ordered by node and zones within node.
2851 * This results in maximum locality--normal zone overflows into local
2852 * DMA zone, if any--but risks exhausting DMA zone.
2853 */
2854static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002856 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002858
Mel Gorman54a6eb52008-04-28 02:12:16 -07002859 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07002860 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07002861 ;
2862 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2863 MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002864 zonelist->_zonerefs[j].zone = NULL;
2865 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002866}
2867
2868/*
Christoph Lameter523b9452007-10-16 01:25:37 -07002869 * Build gfp_thisnode zonelists
2870 */
2871static void build_thisnode_zonelists(pg_data_t *pgdat)
2872{
Christoph Lameter523b9452007-10-16 01:25:37 -07002873 int j;
2874 struct zonelist *zonelist;
2875
Mel Gorman54a6eb52008-04-28 02:12:16 -07002876 zonelist = &pgdat->node_zonelists[1];
2877 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002878 zonelist->_zonerefs[j].zone = NULL;
2879 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07002880}
2881
2882/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002883 * Build zonelists ordered by zone and nodes within zones.
2884 * This results in conserving DMA zone[s] until all Normal memory is
2885 * exhausted, but results in overflowing to remote node while memory
2886 * may still exist in local DMA zone.
2887 */
2888static int node_order[MAX_NUMNODES];
2889
2890static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2891{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002892 int pos, j, node;
2893 int zone_type; /* needs to be signed */
2894 struct zone *z;
2895 struct zonelist *zonelist;
2896
Mel Gorman54a6eb52008-04-28 02:12:16 -07002897 zonelist = &pgdat->node_zonelists[0];
2898 pos = 0;
2899 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2900 for (j = 0; j < nr_nodes; j++) {
2901 node = node_order[j];
2902 z = &NODE_DATA(node)->node_zones[zone_type];
2903 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002904 zoneref_set_zone(z,
2905 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002906 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002907 }
2908 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002909 }
Mel Gormandd1a2392008-04-28 02:12:17 -07002910 zonelist->_zonerefs[pos].zone = NULL;
2911 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002912}
2913
2914static int default_zonelist_order(void)
2915{
2916 int nid, zone_type;
2917 unsigned long low_kmem_size,total_size;
2918 struct zone *z;
2919 int average_size;
2920 /*
Thomas Weber88393162010-03-16 11:47:56 +01002921 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002922 * If they are really small and used heavily, the system can fall
2923 * into OOM very easily.
David Rientjese325c902010-05-24 14:32:13 -07002924 * This function detect ZONE_DMA/DMA32 size and configures zone order.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002925 */
2926 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2927 low_kmem_size = 0;
2928 total_size = 0;
2929 for_each_online_node(nid) {
2930 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2931 z = &NODE_DATA(nid)->node_zones[zone_type];
2932 if (populated_zone(z)) {
2933 if (zone_type < ZONE_NORMAL)
2934 low_kmem_size += z->present_pages;
2935 total_size += z->present_pages;
David Rientjese325c902010-05-24 14:32:13 -07002936 } else if (zone_type == ZONE_NORMAL) {
2937 /*
2938 * If any node has only lowmem, then node order
2939 * is preferred to allow kernel allocations
2940 * locally; otherwise, they can easily infringe
2941 * on other nodes when there is an abundance of
2942 * lowmem available to allocate from.
2943 */
2944 return ZONELIST_ORDER_NODE;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002945 }
2946 }
2947 }
2948 if (!low_kmem_size || /* there are no DMA area. */
2949 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2950 return ZONELIST_ORDER_NODE;
2951 /*
2952 * look into each node's config.
2953 * If there is a node whose DMA/DMA32 memory is very big area on
2954 * local memory, NODE_ORDER may be suitable.
2955 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002956 average_size = total_size /
2957 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002958 for_each_online_node(nid) {
2959 low_kmem_size = 0;
2960 total_size = 0;
2961 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2962 z = &NODE_DATA(nid)->node_zones[zone_type];
2963 if (populated_zone(z)) {
2964 if (zone_type < ZONE_NORMAL)
2965 low_kmem_size += z->present_pages;
2966 total_size += z->present_pages;
2967 }
2968 }
2969 if (low_kmem_size &&
2970 total_size > average_size && /* ignore small node */
2971 low_kmem_size > total_size * 70/100)
2972 return ZONELIST_ORDER_NODE;
2973 }
2974 return ZONELIST_ORDER_ZONE;
2975}
2976
2977static void set_zonelist_order(void)
2978{
2979 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2980 current_zonelist_order = default_zonelist_order();
2981 else
2982 current_zonelist_order = user_zonelist_order;
2983}
2984
2985static void build_zonelists(pg_data_t *pgdat)
2986{
2987 int j, node, load;
2988 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002990 int local_node, prev_node;
2991 struct zonelist *zonelist;
2992 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002993
2994 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07002995 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002996 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07002997 zonelist->_zonerefs[0].zone = NULL;
2998 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002999 }
3000
3001 /* NUMA-aware ordering of nodes */
3002 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003003 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004 prev_node = local_node;
3005 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003006
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003007 memset(node_order, 0, sizeof(node_order));
3008 j = 0;
3009
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08003011 int distance = node_distance(local_node, node);
3012
3013 /*
3014 * If another node is sufficiently far away then it is better
3015 * to reclaim pages in a zone before going off node.
3016 */
3017 if (distance > RECLAIM_DISTANCE)
3018 zone_reclaim_mode = 1;
3019
Linus Torvalds1da177e2005-04-16 15:20:36 -07003020 /*
3021 * We don't want to pressure a particular node.
3022 * So adding penalty to the first node in same
3023 * distance group to make it round-robin.
3024 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08003025 if (distance != node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003026 node_load[node] = load;
3027
Linus Torvalds1da177e2005-04-16 15:20:36 -07003028 prev_node = node;
3029 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003030 if (order == ZONELIST_ORDER_NODE)
3031 build_zonelists_in_node_order(pgdat, node);
3032 else
3033 node_order[j++] = node; /* remember order */
3034 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003035
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003036 if (order == ZONELIST_ORDER_ZONE) {
3037 /* calculate node order -- i.e., DMA last! */
3038 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003039 }
Christoph Lameter523b9452007-10-16 01:25:37 -07003040
3041 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042}
3043
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003044/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003045static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003046{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003047 struct zonelist *zonelist;
3048 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07003049 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003050
Mel Gorman54a6eb52008-04-28 02:12:16 -07003051 zonelist = &pgdat->node_zonelists[0];
3052 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3053 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07003054 for (z = zonelist->_zonerefs; z->zone; z++)
3055 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003056}
3057
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003058#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3059/*
3060 * Return node id of node used for "local" allocations.
3061 * I.e., first node id of first zone in arg node's generic zonelist.
3062 * Used for initializing percpu 'numa_mem', which is used primarily
3063 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3064 */
3065int local_memory_node(int node)
3066{
3067 struct zone *zone;
3068
3069 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3070 gfp_zone(GFP_KERNEL),
3071 NULL,
3072 &zone);
3073 return zone->node;
3074}
3075#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003076
Linus Torvalds1da177e2005-04-16 15:20:36 -07003077#else /* CONFIG_NUMA */
3078
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003079static void set_zonelist_order(void)
3080{
3081 current_zonelist_order = ZONELIST_ORDER_ZONE;
3082}
3083
3084static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003085{
Christoph Lameter19655d32006-09-25 23:31:19 -07003086 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003087 enum zone_type j;
3088 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003089
3090 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091
Mel Gorman54a6eb52008-04-28 02:12:16 -07003092 zonelist = &pgdat->node_zonelists[0];
3093 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003094
Mel Gorman54a6eb52008-04-28 02:12:16 -07003095 /*
3096 * Now we build the zonelist so that it contains the zones
3097 * of all the other nodes.
3098 * We don't want to pressure a particular node, so when
3099 * building the zones for node N, we make sure that the
3100 * zones coming right after the local ones are those from
3101 * node N+1 (modulo N)
3102 */
3103 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3104 if (!node_online(node))
3105 continue;
3106 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3107 MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07003109 for (node = 0; node < local_node; node++) {
3110 if (!node_online(node))
3111 continue;
3112 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3113 MAX_NR_ZONES - 1);
3114 }
3115
Mel Gormandd1a2392008-04-28 02:12:17 -07003116 zonelist->_zonerefs[j].zone = NULL;
3117 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003118}
3119
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003120/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003121static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003122{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003123 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003124}
3125
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126#endif /* CONFIG_NUMA */
3127
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003128/*
3129 * Boot pageset table. One per cpu which is going to be used for all
3130 * zones and all nodes. The parameters will be set in such a way
3131 * that an item put on a list will immediately be handed over to
3132 * the buddy list. This is safe since pageset manipulation is done
3133 * with interrupts disabled.
3134 *
3135 * The boot_pagesets must be kept even after bootup is complete for
3136 * unused processors and/or zones. They do play a role for bootstrapping
3137 * hotplugged processors.
3138 *
3139 * zoneinfo_show() and maybe other functions do
3140 * not check if the processor is online before following the pageset pointer.
3141 * Other parts of the kernel may not check if the zone is available.
3142 */
3143static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3144static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07003145static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003146
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003147/*
3148 * Global mutex to protect against size modification of zonelists
3149 * as well as to serialize pageset setup for the new populated zone.
3150 */
3151DEFINE_MUTEX(zonelists_mutex);
3152
Rusty Russell9b1a4d32008-07-28 12:16:30 -05003153/* return values int ....just for stop_machine() */
Haicheng Li1f522502010-05-24 14:32:51 -07003154static __init_refok int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003155{
Yasunori Goto68113782006-06-23 02:03:11 -07003156 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003157 int cpu;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003158
Bo Liu7f9cfb32009-08-18 14:11:19 -07003159#ifdef CONFIG_NUMA
3160 memset(node_load, 0, sizeof(node_load));
3161#endif
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003162 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07003163 pg_data_t *pgdat = NODE_DATA(nid);
3164
3165 build_zonelists(pgdat);
3166 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003167 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003168
3169 /*
3170 * Initialize the boot_pagesets that are going to be used
3171 * for bootstrapping processors. The real pagesets for
3172 * each zone will be allocated later when the per cpu
3173 * allocator is available.
3174 *
3175 * boot_pagesets are used also for bootstrapping offline
3176 * cpus if the system is already booted because the pagesets
3177 * are needed to initialize allocators on a specific cpu too.
3178 * F.e. the percpu allocator needs the page allocator which
3179 * needs the percpu allocator in order to allocate its pagesets
3180 * (a chicken-egg dilemma).
3181 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003182 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003183 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3184
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003185#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3186 /*
3187 * We now know the "local memory node" for each node--
3188 * i.e., the node of the first zone in the generic zonelist.
3189 * Set up numa_mem percpu variable for on-line cpus. During
3190 * boot, only the boot cpu should be on-line; we'll init the
3191 * secondary cpus' numa_mem as they come on-line. During
3192 * node/memory hotplug, we'll fixup all on-line cpus.
3193 */
3194 if (cpu_online(cpu))
3195 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3196#endif
3197 }
3198
Yasunori Goto68113782006-06-23 02:03:11 -07003199 return 0;
3200}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003201
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003202/*
3203 * Called with zonelists_mutex held always
3204 * unless system_state == SYSTEM_BOOTING.
3205 */
Paul Mundt9f6ae442011-04-14 15:21:57 -07003206void __ref build_all_zonelists(void *data)
Yasunori Goto68113782006-06-23 02:03:11 -07003207{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003208 set_zonelist_order();
3209
Yasunori Goto68113782006-06-23 02:03:11 -07003210 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07003211 __build_all_zonelists(NULL);
Mel Gorman68ad8df2008-07-23 21:26:52 -07003212 mminit_verify_zonelist();
Yasunori Goto68113782006-06-23 02:03:11 -07003213 cpuset_init_current_mems_allowed();
3214 } else {
Simon Arlott183ff222007-10-20 01:27:18 +02003215 /* we have to stop all cpus to guarantee there is no user
Yasunori Goto68113782006-06-23 02:03:11 -07003216 of zonelist */
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003217#ifdef CONFIG_MEMORY_HOTPLUG
3218 if (data)
3219 setup_zone_pageset((struct zone *)data);
3220#endif
3221 stop_machine(__build_all_zonelists, NULL, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07003222 /* cpuset refresh routine should be here */
3223 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07003224 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003225 /*
3226 * Disable grouping by mobility if the number of pages in the
3227 * system is too low to allow the mechanism to work. It would be
3228 * more accurate, but expensive to check per-zone. This check is
3229 * made on memory-hotadd so a system can start with mobility
3230 * disabled and enable it later
3231 */
Mel Gormand9c23402007-10-16 01:26:01 -07003232 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003233 page_group_by_mobility_disabled = 1;
3234 else
3235 page_group_by_mobility_disabled = 0;
3236
3237 printk("Built %i zonelists in %s order, mobility grouping %s. "
3238 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003239 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003240 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003241 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003242 vm_total_pages);
3243#ifdef CONFIG_NUMA
3244 printk("Policy zone: %s\n", zone_names[policy_zone]);
3245#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003246}
3247
3248/*
3249 * Helper functions to size the waitqueue hash table.
3250 * Essentially these want to choose hash table sizes sufficiently
3251 * large so that collisions trying to wait on pages are rare.
3252 * But in fact, the number of active page waitqueues on typical
3253 * systems is ridiculously low, less than 200. So this is even
3254 * conservative, even though it seems large.
3255 *
3256 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3257 * waitqueues, i.e. the size of the waitq table given the number of pages.
3258 */
3259#define PAGES_PER_WAITQUEUE 256
3260
Yasunori Gotocca448f2006-06-23 02:03:10 -07003261#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07003262static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003263{
3264 unsigned long size = 1;
3265
3266 pages /= PAGES_PER_WAITQUEUE;
3267
3268 while (size < pages)
3269 size <<= 1;
3270
3271 /*
3272 * Once we have dozens or even hundreds of threads sleeping
3273 * on IO we've got bigger problems than wait queue collision.
3274 * Limit the size of the wait table to a reasonable size.
3275 */
3276 size = min(size, 4096UL);
3277
3278 return max(size, 4UL);
3279}
Yasunori Gotocca448f2006-06-23 02:03:10 -07003280#else
3281/*
3282 * A zone's size might be changed by hot-add, so it is not possible to determine
3283 * a suitable size for its wait_table. So we use the maximum size now.
3284 *
3285 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3286 *
3287 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3288 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3289 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3290 *
3291 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3292 * or more by the traditional way. (See above). It equals:
3293 *
3294 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3295 * ia64(16K page size) : = ( 8G + 4M)byte.
3296 * powerpc (64K page size) : = (32G +16M)byte.
3297 */
3298static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3299{
3300 return 4096UL;
3301}
3302#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003303
3304/*
3305 * This is an integer logarithm so that shifts can be used later
3306 * to extract the more random high bits from the multiplicative
3307 * hash function before the remainder is taken.
3308 */
3309static inline unsigned long wait_table_bits(unsigned long size)
3310{
3311 return ffz(~size);
3312}
3313
3314#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3315
Mel Gorman56fd56b2007-10-16 01:25:58 -07003316/*
Mel Gormand9c23402007-10-16 01:26:01 -07003317 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07003318 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3319 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07003320 * higher will lead to a bigger reserve which will get freed as contiguous
3321 * blocks as reclaim kicks in
3322 */
3323static void setup_zone_migrate_reserve(struct zone *zone)
3324{
3325 unsigned long start_pfn, pfn, end_pfn;
3326 struct page *page;
Mel Gorman78986a62009-09-21 17:03:02 -07003327 unsigned long block_migratetype;
3328 int reserve;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003329
3330 /* Get the start pfn, end pfn and the number of blocks to reserve */
3331 start_pfn = zone->zone_start_pfn;
3332 end_pfn = start_pfn + zone->spanned_pages;
Mel Gorman41858962009-06-16 15:32:12 -07003333 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07003334 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003335
Mel Gorman78986a62009-09-21 17:03:02 -07003336 /*
3337 * Reserve blocks are generally in place to help high-order atomic
3338 * allocations that are short-lived. A min_free_kbytes value that
3339 * would result in more than 2 reserve blocks for atomic allocations
3340 * is assumed to be in place to help anti-fragmentation for the
3341 * future allocation of hugepages at runtime.
3342 */
3343 reserve = min(2, reserve);
3344
Mel Gormand9c23402007-10-16 01:26:01 -07003345 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07003346 if (!pfn_valid(pfn))
3347 continue;
3348 page = pfn_to_page(pfn);
3349
Adam Litke344c7902008-09-02 14:35:38 -07003350 /* Watch out for overlapping nodes */
3351 if (page_to_nid(page) != zone_to_nid(zone))
3352 continue;
3353
Mel Gorman56fd56b2007-10-16 01:25:58 -07003354 /* Blocks with reserved pages will never free, skip them. */
3355 if (PageReserved(page))
3356 continue;
3357
3358 block_migratetype = get_pageblock_migratetype(page);
3359
3360 /* If this block is reserved, account for it */
3361 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
3362 reserve--;
3363 continue;
3364 }
3365
3366 /* Suitable for reserving if this block is movable */
3367 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
3368 set_pageblock_migratetype(page, MIGRATE_RESERVE);
3369 move_freepages_block(zone, page, MIGRATE_RESERVE);
3370 reserve--;
3371 continue;
3372 }
3373
3374 /*
3375 * If the reserve is met and this is a previous reserved block,
3376 * take it back
3377 */
3378 if (block_migratetype == MIGRATE_RESERVE) {
3379 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
3380 move_freepages_block(zone, page, MIGRATE_MOVABLE);
3381 }
3382 }
3383}
Mel Gormanac0e5b72007-10-16 01:25:58 -07003384
Linus Torvalds1da177e2005-04-16 15:20:36 -07003385/*
3386 * Initially all pages are reserved - free ones are freed
3387 * up by free_all_bootmem() once the early boot process is
3388 * done. Non-atomic initialization, single-pass.
3389 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003390void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa02007-01-10 23:15:30 -08003391 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003392{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003393 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07003394 unsigned long end_pfn = start_pfn + size;
3395 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003396 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003397
Hugh Dickins22b31ee2009-01-06 14:40:09 -08003398 if (highest_memmap_pfn < end_pfn - 1)
3399 highest_memmap_pfn = end_pfn - 1;
3400
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003401 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08003402 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa02007-01-10 23:15:30 -08003403 /*
3404 * There can be holes in boot-time mem_map[]s
3405 * handed to this function. They do not
3406 * exist on hotplugged memory.
3407 */
3408 if (context == MEMMAP_EARLY) {
3409 if (!early_pfn_valid(pfn))
3410 continue;
3411 if (!early_pfn_in_nid(pfn, nid))
3412 continue;
3413 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003414 page = pfn_to_page(pfn);
3415 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07003416 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08003417 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003418 reset_page_mapcount(page);
3419 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003420 /*
3421 * Mark the block movable so that blocks are reserved for
3422 * movable at startup. This will force kernel allocations
3423 * to reserve their blocks rather than leaking throughout
3424 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07003425 * kernel allocations are made. Later some blocks near
3426 * the start are marked MIGRATE_RESERVE by
3427 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003428 *
3429 * bitmap is created for zone's valid pfn range. but memmap
3430 * can be created for invalid pages (for alignment)
3431 * check here not to call set_pageblock_migratetype() against
3432 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003433 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003434 if ((z->zone_start_pfn <= pfn)
3435 && (pfn < z->zone_start_pfn + z->spanned_pages)
3436 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07003437 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003438
Linus Torvalds1da177e2005-04-16 15:20:36 -07003439 INIT_LIST_HEAD(&page->lru);
3440#ifdef WANT_PAGE_VIRTUAL
3441 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3442 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07003443 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003444#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003445 }
3446}
3447
Andi Kleen1e548de2008-02-04 22:29:26 -08003448static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003449{
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003450 int order, t;
3451 for_each_migratetype_order(order, t) {
3452 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003453 zone->free_area[order].nr_free = 0;
3454 }
3455}
3456
3457#ifndef __HAVE_ARCH_MEMMAP_INIT
3458#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa02007-01-10 23:15:30 -08003459 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003460#endif
3461
Sam Ravnborg1d6f4e62008-01-17 15:21:12 -08003462static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003463{
David Howells3a6be872009-05-06 16:03:03 -07003464#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003465 int batch;
3466
3467 /*
3468 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07003469 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003470 *
3471 * OK, so we don't know how big the cache is. So guess.
3472 */
3473 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07003474 if (batch * PAGE_SIZE > 512 * 1024)
3475 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003476 batch /= 4; /* We effectively *= 4 below */
3477 if (batch < 1)
3478 batch = 1;
3479
3480 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11003481 * Clamp the batch to a 2^n - 1 value. Having a power
3482 * of 2 value was found to be more likely to have
3483 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003484 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11003485 * For example if 2 tasks are alternately allocating
3486 * batches of pages, one task can end up with a lot
3487 * of pages of one half of the possible page colors
3488 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003489 */
David Howells91552032009-05-06 16:03:02 -07003490 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07003491
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003492 return batch;
David Howells3a6be872009-05-06 16:03:03 -07003493
3494#else
3495 /* The deferral and batching of frees should be suppressed under NOMMU
3496 * conditions.
3497 *
3498 * The problem is that NOMMU needs to be able to allocate large chunks
3499 * of contiguous memory as there's no hardware page translation to
3500 * assemble apparent contiguous memory from discontiguous pages.
3501 *
3502 * Queueing large contiguous runs of pages for batching, however,
3503 * causes the pages to actually be freed in smaller chunks. As there
3504 * can be a significant delay between the individual batches being
3505 * recycled, this leads to the once large chunks of space being
3506 * fragmented and becoming unavailable for high-order allocations.
3507 */
3508 return 0;
3509#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003510}
3511
Adrian Bunkb69a7282008-07-23 21:28:12 -07003512static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
Christoph Lameter2caaad42005-06-21 17:15:00 -07003513{
3514 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003515 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003516
Magnus Damm1c6fe942005-10-26 01:58:59 -07003517 memset(p, 0, sizeof(*p));
3518
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003519 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003520 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003521 pcp->high = 6 * batch;
3522 pcp->batch = max(1UL, 1 * batch);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003523 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
3524 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07003525}
3526
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003527/*
3528 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3529 * to the value high for the pageset p.
3530 */
3531
3532static void setup_pagelist_highmark(struct per_cpu_pageset *p,
3533 unsigned long high)
3534{
3535 struct per_cpu_pages *pcp;
3536
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003537 pcp = &p->pcp;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003538 pcp->high = high;
3539 pcp->batch = max(1UL, high/4);
3540 if ((high/4) > (PAGE_SHIFT * 8))
3541 pcp->batch = PAGE_SHIFT * 8;
3542}
3543
Nikanth Karthikesan58c2ee42011-03-15 10:59:02 +05303544static void setup_zone_pageset(struct zone *zone)
Wu Fengguang319774e2010-05-24 14:32:49 -07003545{
3546 int cpu;
3547
3548 zone->pageset = alloc_percpu(struct per_cpu_pageset);
3549
3550 for_each_possible_cpu(cpu) {
3551 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
3552
3553 setup_pageset(pcp, zone_batchsize(zone));
3554
3555 if (percpu_pagelist_fraction)
3556 setup_pagelist_highmark(pcp,
3557 (zone->present_pages /
3558 percpu_pagelist_fraction));
3559 }
3560}
3561
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003562/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003563 * Allocate per cpu pagesets and initialize them.
3564 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07003565 */
Al Viro78d99552005-12-15 09:18:25 +00003566void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003567{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003568 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003569
Wu Fengguang319774e2010-05-24 14:32:49 -07003570 for_each_populated_zone(zone)
3571 setup_zone_pageset(zone);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003572}
3573
Sam Ravnborg577a32f2007-05-17 23:29:25 +02003574static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07003575int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07003576{
3577 int i;
3578 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003579 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07003580
3581 /*
3582 * The per-page waitqueue mechanism uses hashed waitqueues
3583 * per zone.
3584 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07003585 zone->wait_table_hash_nr_entries =
3586 wait_table_hash_nr_entries(zone_size_pages);
3587 zone->wait_table_bits =
3588 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003589 alloc_size = zone->wait_table_hash_nr_entries
3590 * sizeof(wait_queue_head_t);
3591
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07003592 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07003593 zone->wait_table = (wait_queue_head_t *)
Yinghai Lu8f389a992011-05-11 15:13:32 -07003594 alloc_bootmem_node_nopanic(pgdat, alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003595 } else {
3596 /*
3597 * This case means that a zone whose size was 0 gets new memory
3598 * via memory hot-add.
3599 * But it may be the case that a new node was hot-added. In
3600 * this case vmalloc() will not be able to use this new node's
3601 * memory - this wait_table must be initialized to use this new
3602 * node itself as well.
3603 * To use this new node's memory, further consideration will be
3604 * necessary.
3605 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07003606 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003607 }
3608 if (!zone->wait_table)
3609 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07003610
Yasunori Goto02b694d2006-06-23 02:03:08 -07003611 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07003612 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003613
3614 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003615}
3616
Shaohua Li112067f2009-09-21 17:01:16 -07003617static int __zone_pcp_update(void *data)
3618{
3619 struct zone *zone = data;
3620 int cpu;
3621 unsigned long batch = zone_batchsize(zone), flags;
3622
Thomas Gleixner2d30a1f2010-03-10 15:20:40 -08003623 for_each_possible_cpu(cpu) {
Shaohua Li112067f2009-09-21 17:01:16 -07003624 struct per_cpu_pageset *pset;
3625 struct per_cpu_pages *pcp;
3626
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003627 pset = per_cpu_ptr(zone->pageset, cpu);
Shaohua Li112067f2009-09-21 17:01:16 -07003628 pcp = &pset->pcp;
3629
3630 local_irq_save(flags);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003631 free_pcppages_bulk(zone, pcp->count, pcp);
Shaohua Li112067f2009-09-21 17:01:16 -07003632 setup_pageset(pset, batch);
3633 local_irq_restore(flags);
3634 }
3635 return 0;
3636}
3637
3638void zone_pcp_update(struct zone *zone)
3639{
3640 stop_machine(__zone_pcp_update, zone, NULL);
3641}
3642
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003643static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07003644{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003645 /*
3646 * per cpu subsystem is not up at this point. The following code
3647 * relies on the ability of the linker to provide the
3648 * offset of a (static) per cpu variable into the per cpu area.
3649 */
3650 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07003651
Anton Blanchardf5335c02006-03-25 03:06:49 -08003652 if (zone->present_pages)
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003653 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
3654 zone->name, zone->present_pages,
3655 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07003656}
3657
Yasunori Goto718127c2006-06-23 02:03:10 -07003658__meminit int init_currently_empty_zone(struct zone *zone,
3659 unsigned long zone_start_pfn,
Dave Hansena2f3aa02007-01-10 23:15:30 -08003660 unsigned long size,
3661 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07003662{
3663 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003664 int ret;
3665 ret = zone_wait_table_init(zone, size);
3666 if (ret)
3667 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07003668 pgdat->nr_zones = zone_idx(zone) + 1;
3669
Dave Hansened8ece22005-10-29 18:16:50 -07003670 zone->zone_start_pfn = zone_start_pfn;
3671
Mel Gorman708614e2008-07-23 21:26:51 -07003672 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3673 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3674 pgdat->node_id,
3675 (unsigned long)zone_idx(zone),
3676 zone_start_pfn, (zone_start_pfn + size));
3677
Andi Kleen1e548de2008-02-04 22:29:26 -08003678 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07003679
3680 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003681}
3682
Mel Gormanc7132162006-09-27 01:49:43 -07003683#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3684/*
3685 * Basic iterator support. Return the first range of PFNs for a node
3686 * Note: nid == MAX_NUMNODES returns first region regardless of node
3687 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003688static int __meminit first_active_region_index_in_nid(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003689{
3690 int i;
3691
3692 for (i = 0; i < nr_nodemap_entries; i++)
3693 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3694 return i;
3695
3696 return -1;
3697}
3698
3699/*
3700 * Basic iterator support. Return the next active range of PFNs for a node
Simon Arlott183ff222007-10-20 01:27:18 +02003701 * Note: nid == MAX_NUMNODES returns next region regardless of node
Mel Gormanc7132162006-09-27 01:49:43 -07003702 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003703static int __meminit next_active_region_index_in_nid(int index, int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003704{
3705 for (index = index + 1; index < nr_nodemap_entries; index++)
3706 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3707 return index;
3708
3709 return -1;
3710}
3711
3712#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3713/*
3714 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3715 * Architectures may implement their own version but if add_active_range()
3716 * was used and there are no special requirements, this is a convenient
3717 * alternative
3718 */
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003719int __meminit __early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07003720{
3721 int i;
3722
3723 for (i = 0; i < nr_nodemap_entries; i++) {
3724 unsigned long start_pfn = early_node_map[i].start_pfn;
3725 unsigned long end_pfn = early_node_map[i].end_pfn;
3726
3727 if (start_pfn <= pfn && pfn < end_pfn)
3728 return early_node_map[i].nid;
3729 }
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003730 /* This is a memory hole */
3731 return -1;
Mel Gormanc7132162006-09-27 01:49:43 -07003732}
3733#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3734
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003735int __meminit early_pfn_to_nid(unsigned long pfn)
3736{
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003737 int nid;
3738
3739 nid = __early_pfn_to_nid(pfn);
3740 if (nid >= 0)
3741 return nid;
3742 /* just returns 0 */
3743 return 0;
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003744}
3745
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003746#ifdef CONFIG_NODES_SPAN_OTHER_NODES
3747bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3748{
3749 int nid;
3750
3751 nid = __early_pfn_to_nid(pfn);
3752 if (nid >= 0 && nid != node)
3753 return false;
3754 return true;
3755}
3756#endif
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003757
Mel Gormanc7132162006-09-27 01:49:43 -07003758/* Basic iterator support to walk early_node_map[] */
3759#define for_each_active_range_index_in_nid(i, nid) \
3760 for (i = first_active_region_index_in_nid(nid); i != -1; \
3761 i = next_active_region_index_in_nid(i, nid))
3762
3763/**
3764 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003765 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3766 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07003767 *
3768 * If an architecture guarantees that all ranges registered with
3769 * add_active_ranges() contain no holes and may be freed, this
3770 * this function may be used instead of calling free_bootmem() manually.
3771 */
3772void __init free_bootmem_with_active_regions(int nid,
3773 unsigned long max_low_pfn)
3774{
3775 int i;
3776
3777 for_each_active_range_index_in_nid(i, nid) {
3778 unsigned long size_pages = 0;
3779 unsigned long end_pfn = early_node_map[i].end_pfn;
3780
3781 if (early_node_map[i].start_pfn >= max_low_pfn)
3782 continue;
3783
3784 if (end_pfn > max_low_pfn)
3785 end_pfn = max_low_pfn;
3786
3787 size_pages = end_pfn - early_node_map[i].start_pfn;
3788 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3789 PFN_PHYS(early_node_map[i].start_pfn),
3790 size_pages << PAGE_SHIFT);
3791 }
3792}
3793
Yinghai Luedbe7d22010-08-25 13:39:16 -07003794#ifdef CONFIG_HAVE_MEMBLOCK
Yinghai Lucc289892011-02-26 13:05:43 +01003795/*
3796 * Basic iterator support. Return the last range of PFNs for a node
3797 * Note: nid == MAX_NUMNODES returns last region regardless of node
3798 */
3799static int __meminit last_active_region_index_in_nid(int nid)
3800{
3801 int i;
3802
3803 for (i = nr_nodemap_entries - 1; i >= 0; i--)
3804 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3805 return i;
3806
3807 return -1;
3808}
3809
3810/*
3811 * Basic iterator support. Return the previous active range of PFNs for a node
3812 * Note: nid == MAX_NUMNODES returns next region regardless of node
3813 */
3814static int __meminit previous_active_region_index_in_nid(int index, int nid)
3815{
3816 for (index = index - 1; index >= 0; index--)
3817 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3818 return index;
3819
3820 return -1;
3821}
3822
3823#define for_each_active_range_index_in_nid_reverse(i, nid) \
3824 for (i = last_active_region_index_in_nid(nid); i != -1; \
3825 i = previous_active_region_index_in_nid(i, nid))
3826
Yinghai Luedbe7d22010-08-25 13:39:16 -07003827u64 __init find_memory_core_early(int nid, u64 size, u64 align,
3828 u64 goal, u64 limit)
3829{
3830 int i;
3831
3832 /* Need to go over early_node_map to find out good range for node */
Yinghai Lu1a4a6782010-12-17 16:59:07 -08003833 for_each_active_range_index_in_nid_reverse(i, nid) {
Yinghai Luedbe7d22010-08-25 13:39:16 -07003834 u64 addr;
3835 u64 ei_start, ei_last;
3836 u64 final_start, final_end;
3837
3838 ei_last = early_node_map[i].end_pfn;
3839 ei_last <<= PAGE_SHIFT;
3840 ei_start = early_node_map[i].start_pfn;
3841 ei_start <<= PAGE_SHIFT;
3842
3843 final_start = max(ei_start, goal);
3844 final_end = min(ei_last, limit);
3845
3846 if (final_start >= final_end)
3847 continue;
3848
3849 addr = memblock_find_in_range(final_start, final_end, size, align);
3850
3851 if (addr == MEMBLOCK_ERROR)
3852 continue;
3853
3854 return addr;
3855 }
3856
3857 return MEMBLOCK_ERROR;
3858}
3859#endif
3860
Yinghai Lu08677212010-02-10 01:20:20 -08003861int __init add_from_early_node_map(struct range *range, int az,
3862 int nr_range, int nid)
3863{
3864 int i;
3865 u64 start, end;
3866
3867 /* need to go over early_node_map to find out good range for node */
3868 for_each_active_range_index_in_nid(i, nid) {
3869 start = early_node_map[i].start_pfn;
3870 end = early_node_map[i].end_pfn;
3871 nr_range = add_range(range, az, nr_range, start, end);
3872 }
3873 return nr_range;
3874}
3875
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003876void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3877{
3878 int i;
Yinghai Lud52d53b2008-06-16 20:10:55 -07003879 int ret;
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003880
Yinghai Lud52d53b2008-06-16 20:10:55 -07003881 for_each_active_range_index_in_nid(i, nid) {
3882 ret = work_fn(early_node_map[i].start_pfn,
3883 early_node_map[i].end_pfn, data);
3884 if (ret)
3885 break;
3886 }
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003887}
Mel Gormanc7132162006-09-27 01:49:43 -07003888/**
3889 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003890 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07003891 *
3892 * If an architecture guarantees that all ranges registered with
3893 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003894 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07003895 */
3896void __init sparse_memory_present_with_active_regions(int nid)
3897{
3898 int i;
3899
3900 for_each_active_range_index_in_nid(i, nid)
3901 memory_present(early_node_map[i].nid,
3902 early_node_map[i].start_pfn,
3903 early_node_map[i].end_pfn);
3904}
3905
3906/**
3907 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003908 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3909 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3910 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07003911 *
3912 * It returns the start and end page frame of a node based on information
3913 * provided by an arch calling add_active_range(). If called for a node
3914 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003915 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07003916 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003917void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003918 unsigned long *start_pfn, unsigned long *end_pfn)
3919{
3920 int i;
3921 *start_pfn = -1UL;
3922 *end_pfn = 0;
3923
3924 for_each_active_range_index_in_nid(i, nid) {
3925 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3926 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3927 }
3928
Christoph Lameter633c0662007-10-16 01:25:37 -07003929 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07003930 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07003931}
3932
3933/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07003934 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3935 * assumption is made that zones within a node are ordered in monotonic
3936 * increasing memory addresses so that the "highest" populated zone is used
3937 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003938static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07003939{
3940 int zone_index;
3941 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3942 if (zone_index == ZONE_MOVABLE)
3943 continue;
3944
3945 if (arch_zone_highest_possible_pfn[zone_index] >
3946 arch_zone_lowest_possible_pfn[zone_index])
3947 break;
3948 }
3949
3950 VM_BUG_ON(zone_index == -1);
3951 movable_zone = zone_index;
3952}
3953
3954/*
3955 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003956 * because it is sized independent of architecture. Unlike the other zones,
Mel Gorman2a1e2742007-07-17 04:03:12 -07003957 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3958 * in each node depending on the size of each node and how evenly kernelcore
3959 * is distributed. This helper function adjusts the zone ranges
3960 * provided by the architecture for a given node by using the end of the
3961 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3962 * zones within a node are in order of monotonic increases memory addresses
3963 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003964static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07003965 unsigned long zone_type,
3966 unsigned long node_start_pfn,
3967 unsigned long node_end_pfn,
3968 unsigned long *zone_start_pfn,
3969 unsigned long *zone_end_pfn)
3970{
3971 /* Only adjust if ZONE_MOVABLE is on this node */
3972 if (zone_movable_pfn[nid]) {
3973 /* Size ZONE_MOVABLE */
3974 if (zone_type == ZONE_MOVABLE) {
3975 *zone_start_pfn = zone_movable_pfn[nid];
3976 *zone_end_pfn = min(node_end_pfn,
3977 arch_zone_highest_possible_pfn[movable_zone]);
3978
3979 /* Adjust for ZONE_MOVABLE starting within this range */
3980 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3981 *zone_end_pfn > zone_movable_pfn[nid]) {
3982 *zone_end_pfn = zone_movable_pfn[nid];
3983
3984 /* Check if this whole range is within ZONE_MOVABLE */
3985 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3986 *zone_start_pfn = *zone_end_pfn;
3987 }
3988}
3989
3990/*
Mel Gormanc7132162006-09-27 01:49:43 -07003991 * Return the number of pages a zone spans in a node, including holes
3992 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3993 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003994static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003995 unsigned long zone_type,
3996 unsigned long *ignored)
3997{
3998 unsigned long node_start_pfn, node_end_pfn;
3999 unsigned long zone_start_pfn, zone_end_pfn;
4000
4001 /* Get the start and end of the node and zone */
4002 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
4003 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4004 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07004005 adjust_zone_range_for_zone_movable(nid, zone_type,
4006 node_start_pfn, node_end_pfn,
4007 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004008
4009 /* Check that this node has pages within the zone's required range */
4010 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4011 return 0;
4012
4013 /* Move the zone boundaries inside the node if necessary */
4014 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4015 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4016
4017 /* Return the spanned pages */
4018 return zone_end_pfn - zone_start_pfn;
4019}
4020
4021/*
4022 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004023 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07004024 */
Yinghai Lu32996252009-12-15 17:59:02 -08004025unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004026 unsigned long range_start_pfn,
4027 unsigned long range_end_pfn)
4028{
4029 int i = 0;
4030 unsigned long prev_end_pfn = 0, hole_pages = 0;
4031 unsigned long start_pfn;
4032
4033 /* Find the end_pfn of the first active range of pfns in the node */
4034 i = first_active_region_index_in_nid(nid);
4035 if (i == -1)
4036 return 0;
4037
Mel Gormanb5445f92007-07-26 10:41:18 -07004038 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
4039
Mel Gorman9c7cd682006-09-27 01:49:58 -07004040 /* Account for ranges before physical memory on this node */
4041 if (early_node_map[i].start_pfn > range_start_pfn)
Mel Gormanb5445f92007-07-26 10:41:18 -07004042 hole_pages = prev_end_pfn - range_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004043
4044 /* Find all holes for the zone within the node */
4045 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
4046
4047 /* No need to continue if prev_end_pfn is outside the zone */
4048 if (prev_end_pfn >= range_end_pfn)
4049 break;
4050
4051 /* Make sure the end of the zone is not within the hole */
4052 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
4053 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
4054
4055 /* Update the hole size cound and move on */
4056 if (start_pfn > range_start_pfn) {
4057 BUG_ON(prev_end_pfn > start_pfn);
4058 hole_pages += start_pfn - prev_end_pfn;
4059 }
4060 prev_end_pfn = early_node_map[i].end_pfn;
4061 }
4062
Mel Gorman9c7cd682006-09-27 01:49:58 -07004063 /* Account for ranges past physical memory on this node */
4064 if (range_end_pfn > prev_end_pfn)
Mel Gorman0c6cb972006-10-28 10:38:59 -07004065 hole_pages += range_end_pfn -
Mel Gorman9c7cd682006-09-27 01:49:58 -07004066 max(range_start_pfn, prev_end_pfn);
4067
Mel Gormanc7132162006-09-27 01:49:43 -07004068 return hole_pages;
4069}
4070
4071/**
4072 * absent_pages_in_range - Return number of page frames in holes within a range
4073 * @start_pfn: The start PFN to start searching for holes
4074 * @end_pfn: The end PFN to stop searching for holes
4075 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004076 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07004077 */
4078unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4079 unsigned long end_pfn)
4080{
4081 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4082}
4083
4084/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004085static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004086 unsigned long zone_type,
4087 unsigned long *ignored)
4088{
Mel Gorman9c7cd682006-09-27 01:49:58 -07004089 unsigned long node_start_pfn, node_end_pfn;
4090 unsigned long zone_start_pfn, zone_end_pfn;
4091
4092 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
4093 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
4094 node_start_pfn);
4095 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
4096 node_end_pfn);
4097
Mel Gorman2a1e2742007-07-17 04:03:12 -07004098 adjust_zone_range_for_zone_movable(nid, zone_type,
4099 node_start_pfn, node_end_pfn,
4100 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004101 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004102}
Mel Gorman0e0b8642006-09-27 01:49:56 -07004103
Mel Gormanc7132162006-09-27 01:49:43 -07004104#else
Paul Mundt6ea6e682007-07-15 23:38:20 -07004105static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004106 unsigned long zone_type,
4107 unsigned long *zones_size)
4108{
4109 return zones_size[zone_type];
4110}
4111
Paul Mundt6ea6e682007-07-15 23:38:20 -07004112static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004113 unsigned long zone_type,
4114 unsigned long *zholes_size)
4115{
4116 if (!zholes_size)
4117 return 0;
4118
4119 return zholes_size[zone_type];
4120}
Mel Gorman0e0b8642006-09-27 01:49:56 -07004121
Mel Gormanc7132162006-09-27 01:49:43 -07004122#endif
4123
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004124static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Mel Gormanc7132162006-09-27 01:49:43 -07004125 unsigned long *zones_size, unsigned long *zholes_size)
4126{
4127 unsigned long realtotalpages, totalpages = 0;
4128 enum zone_type i;
4129
4130 for (i = 0; i < MAX_NR_ZONES; i++)
4131 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
4132 zones_size);
4133 pgdat->node_spanned_pages = totalpages;
4134
4135 realtotalpages = totalpages;
4136 for (i = 0; i < MAX_NR_ZONES; i++)
4137 realtotalpages -=
4138 zone_absent_pages_in_node(pgdat->node_id, i,
4139 zholes_size);
4140 pgdat->node_present_pages = realtotalpages;
4141 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4142 realtotalpages);
4143}
4144
Mel Gorman835c1342007-10-16 01:25:47 -07004145#ifndef CONFIG_SPARSEMEM
4146/*
4147 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07004148 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4149 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07004150 * round what is now in bits to nearest long in bits, then return it in
4151 * bytes.
4152 */
4153static unsigned long __init usemap_size(unsigned long zonesize)
4154{
4155 unsigned long usemapsize;
4156
Mel Gormand9c23402007-10-16 01:26:01 -07004157 usemapsize = roundup(zonesize, pageblock_nr_pages);
4158 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07004159 usemapsize *= NR_PAGEBLOCK_BITS;
4160 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4161
4162 return usemapsize / 8;
4163}
4164
4165static void __init setup_usemap(struct pglist_data *pgdat,
4166 struct zone *zone, unsigned long zonesize)
4167{
4168 unsigned long usemapsize = usemap_size(zonesize);
4169 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08004170 if (usemapsize)
Yinghai Lu8f389a992011-05-11 15:13:32 -07004171 zone->pageblock_flags = alloc_bootmem_node_nopanic(pgdat,
4172 usemapsize);
Mel Gorman835c1342007-10-16 01:25:47 -07004173}
4174#else
Jesper Juhlfa9f90b2010-11-28 21:39:34 +01004175static inline void setup_usemap(struct pglist_data *pgdat,
Mel Gorman835c1342007-10-16 01:25:47 -07004176 struct zone *zone, unsigned long zonesize) {}
4177#endif /* CONFIG_SPARSEMEM */
4178
Mel Gormand9c23402007-10-16 01:26:01 -07004179#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08004180
4181/* Return a sensible default order for the pageblock size. */
4182static inline int pageblock_default_order(void)
4183{
4184 if (HPAGE_SHIFT > PAGE_SHIFT)
4185 return HUGETLB_PAGE_ORDER;
4186
4187 return MAX_ORDER-1;
4188}
4189
Mel Gormand9c23402007-10-16 01:26:01 -07004190/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4191static inline void __init set_pageblock_order(unsigned int order)
4192{
4193 /* Check that pageblock_nr_pages has not already been setup */
4194 if (pageblock_order)
4195 return;
4196
4197 /*
4198 * Assume the largest contiguous order of interest is a huge page.
4199 * This value may be variable depending on boot parameters on IA64
4200 */
4201 pageblock_order = order;
4202}
4203#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4204
Mel Gormanba72cb82007-11-28 16:21:13 -08004205/*
4206 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4207 * and pageblock_default_order() are unused as pageblock_order is set
4208 * at compile-time. See include/linux/pageblock-flags.h for the values of
4209 * pageblock_order based on the kernel config
4210 */
4211static inline int pageblock_default_order(unsigned int order)
4212{
4213 return MAX_ORDER-1;
4214}
Mel Gormand9c23402007-10-16 01:26:01 -07004215#define set_pageblock_order(x) do {} while (0)
4216
4217#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4218
Linus Torvalds1da177e2005-04-16 15:20:36 -07004219/*
4220 * Set up the zone data structures:
4221 * - mark all pages reserved
4222 * - mark all memory queues empty
4223 * - clear the memory bitmaps
4224 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08004225static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004226 unsigned long *zones_size, unsigned long *zholes_size)
4227{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07004228 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07004229 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004230 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07004231 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004232
Dave Hansen208d54e2005-10-29 18:16:52 -07004233 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004234 pgdat->nr_zones = 0;
4235 init_waitqueue_head(&pgdat->kswapd_wait);
4236 pgdat->kswapd_max_order = 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004237 pgdat_page_cgroup_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004238
4239 for (j = 0; j < MAX_NR_ZONES; j++) {
4240 struct zone *zone = pgdat->node_zones + j;
Mel Gorman0e0b8642006-09-27 01:49:56 -07004241 unsigned long size, realsize, memmap_pages;
Christoph Lameterb69408e2008-10-18 20:26:14 -07004242 enum lru_list l;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004243
Mel Gormanc7132162006-09-27 01:49:43 -07004244 size = zone_spanned_pages_in_node(nid, j, zones_size);
4245 realsize = size - zone_absent_pages_in_node(nid, j,
4246 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004247
Mel Gorman0e0b8642006-09-27 01:49:56 -07004248 /*
4249 * Adjust realsize so that it accounts for how much memory
4250 * is used by this zone for memmap. This affects the watermark
4251 * and per-cpu initialisations
4252 */
Johannes Weinerf7232152008-05-23 13:04:21 -07004253 memmap_pages =
4254 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
Mel Gorman0e0b8642006-09-27 01:49:56 -07004255 if (realsize >= memmap_pages) {
4256 realsize -= memmap_pages;
Yinghai Lu5594c8c2009-01-06 14:39:14 -08004257 if (memmap_pages)
4258 printk(KERN_DEBUG
4259 " %s zone: %lu pages used for memmap\n",
4260 zone_names[j], memmap_pages);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004261 } else
4262 printk(KERN_WARNING
4263 " %s zone: %lu pages exceeds realsize %lu\n",
4264 zone_names[j], memmap_pages, realsize);
4265
Christoph Lameter62672762007-02-10 01:43:07 -08004266 /* Account for reserved pages */
4267 if (j == 0 && realsize > dma_reserve) {
Mel Gorman0e0b8642006-09-27 01:49:56 -07004268 realsize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07004269 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08004270 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004271 }
4272
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004273 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004274 nr_kernel_pages += realsize;
4275 nr_all_pages += realsize;
4276
4277 zone->spanned_pages = size;
4278 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07004279#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07004280 zone->node = nid;
Christoph Lameter8417bba2006-09-25 23:31:51 -07004281 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07004282 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07004283 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07004284#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004285 zone->name = zone_names[j];
4286 spin_lock_init(&zone->lock);
4287 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07004288 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004289 zone->zone_pgdat = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004290
Dave Hansened8ece22005-10-29 18:16:50 -07004291 zone_pcp_init(zone);
Christoph Lameterb69408e2008-10-18 20:26:14 -07004292 for_each_lru(l) {
4293 INIT_LIST_HEAD(&zone->lru[l].list);
Wu Fengguangf8629632009-09-21 17:03:11 -07004294 zone->reclaim_stat.nr_saved_scan[l] = 0;
Christoph Lameterb69408e2008-10-18 20:26:14 -07004295 }
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08004296 zone->reclaim_stat.recent_rotated[0] = 0;
4297 zone->reclaim_stat.recent_rotated[1] = 0;
4298 zone->reclaim_stat.recent_scanned[0] = 0;
4299 zone->reclaim_stat.recent_scanned[1] = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07004300 zap_zone_vm_stats(zone);
David Rientjese815af92007-10-16 23:25:54 -07004301 zone->flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004302 if (!size)
4303 continue;
4304
Mel Gormanba72cb82007-11-28 16:21:13 -08004305 set_pageblock_order(pageblock_default_order());
Mel Gorman835c1342007-10-16 01:25:47 -07004306 setup_usemap(pgdat, zone, size);
Dave Hansena2f3aa02007-01-10 23:15:30 -08004307 ret = init_currently_empty_zone(zone, zone_start_pfn,
4308 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07004309 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07004310 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004311 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004312 }
4313}
4314
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004315static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004316{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004317 /* Skip empty nodes */
4318 if (!pgdat->node_spanned_pages)
4319 return;
4320
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004321#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07004322 /* ia64 gets its own node_mem_map, before this, without bootmem */
4323 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07004324 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004325 struct page *map;
4326
Bob Piccoe984bb42006-05-20 15:00:31 -07004327 /*
4328 * The zone's endpoints aren't required to be MAX_ORDER
4329 * aligned but the node_mem_map endpoints must be in order
4330 * for the buddy allocator to function correctly.
4331 */
4332 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
4333 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
4334 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4335 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07004336 map = alloc_remap(pgdat->node_id, size);
4337 if (!map)
Yinghai Lu8f389a992011-05-11 15:13:32 -07004338 map = alloc_bootmem_node_nopanic(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07004339 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004340 }
Roman Zippel12d810c2007-05-31 00:40:54 -07004341#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07004342 /*
4343 * With no DISCONTIG, the global mem_map is just set as node 0's
4344 */
Mel Gormanc7132162006-09-27 01:49:43 -07004345 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004346 mem_map = NODE_DATA(0)->node_mem_map;
Mel Gormanc7132162006-09-27 01:49:43 -07004347#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
4348 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08004349 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Mel Gormanc7132162006-09-27 01:49:43 -07004350#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4351 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004352#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004353#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004354}
4355
Johannes Weiner9109fb72008-07-23 21:27:20 -07004356void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4357 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004358{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004359 pg_data_t *pgdat = NODE_DATA(nid);
4360
Linus Torvalds1da177e2005-04-16 15:20:36 -07004361 pgdat->node_id = nid;
4362 pgdat->node_start_pfn = node_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004363 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004364
4365 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07004366#ifdef CONFIG_FLAT_NODE_MEM_MAP
4367 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4368 nid, (unsigned long)pgdat,
4369 (unsigned long)pgdat->node_mem_map);
4370#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004371
4372 free_area_init_core(pgdat, zones_size, zholes_size);
4373}
4374
Mel Gormanc7132162006-09-27 01:49:43 -07004375#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07004376
4377#if MAX_NUMNODES > 1
4378/*
4379 * Figure out the number of possible node ids.
4380 */
4381static void __init setup_nr_node_ids(void)
4382{
4383 unsigned int node;
4384 unsigned int highest = 0;
4385
4386 for_each_node_mask(node, node_possible_map)
4387 highest = node;
4388 nr_node_ids = highest + 1;
4389}
4390#else
4391static inline void setup_nr_node_ids(void)
4392{
4393}
4394#endif
4395
Mel Gormanc7132162006-09-27 01:49:43 -07004396/**
4397 * add_active_range - Register a range of PFNs backed by physical memory
4398 * @nid: The node ID the range resides on
4399 * @start_pfn: The start PFN of the available physical memory
4400 * @end_pfn: The end PFN of the available physical memory
4401 *
4402 * These ranges are stored in an early_node_map[] and later used by
4403 * free_area_init_nodes() to calculate zone sizes and holes. If the
4404 * range spans a memory hole, it is up to the architecture to ensure
4405 * the memory is not freed by the bootmem allocator. If possible
4406 * the range being registered will be merged with existing ranges.
4407 */
4408void __init add_active_range(unsigned int nid, unsigned long start_pfn,
4409 unsigned long end_pfn)
4410{
4411 int i;
4412
Mel Gorman6b74ab92008-07-23 21:26:49 -07004413 mminit_dprintk(MMINIT_TRACE, "memory_register",
4414 "Entering add_active_range(%d, %#lx, %#lx) "
4415 "%d entries of %d used\n",
4416 nid, start_pfn, end_pfn,
4417 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
Mel Gormanc7132162006-09-27 01:49:43 -07004418
Mel Gorman2dbb51c2008-07-23 21:26:52 -07004419 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
4420
Mel Gormanc7132162006-09-27 01:49:43 -07004421 /* Merge with existing active regions if possible */
4422 for (i = 0; i < nr_nodemap_entries; i++) {
4423 if (early_node_map[i].nid != nid)
4424 continue;
4425
4426 /* Skip if an existing region covers this new one */
4427 if (start_pfn >= early_node_map[i].start_pfn &&
4428 end_pfn <= early_node_map[i].end_pfn)
4429 return;
4430
4431 /* Merge forward if suitable */
4432 if (start_pfn <= early_node_map[i].end_pfn &&
4433 end_pfn > early_node_map[i].end_pfn) {
4434 early_node_map[i].end_pfn = end_pfn;
4435 return;
4436 }
4437
4438 /* Merge backward if suitable */
Kazuhisa Ichikawad2dbe082010-01-15 17:01:20 -08004439 if (start_pfn < early_node_map[i].start_pfn &&
Mel Gormanc7132162006-09-27 01:49:43 -07004440 end_pfn >= early_node_map[i].start_pfn) {
4441 early_node_map[i].start_pfn = start_pfn;
4442 return;
4443 }
4444 }
4445
4446 /* Check that early_node_map is large enough */
4447 if (i >= MAX_ACTIVE_REGIONS) {
4448 printk(KERN_CRIT "More than %d memory regions, truncating\n",
4449 MAX_ACTIVE_REGIONS);
4450 return;
4451 }
4452
4453 early_node_map[i].nid = nid;
4454 early_node_map[i].start_pfn = start_pfn;
4455 early_node_map[i].end_pfn = end_pfn;
4456 nr_nodemap_entries = i + 1;
4457}
4458
4459/**
Yinghai Lucc1050b2008-06-13 19:08:52 -07004460 * remove_active_range - Shrink an existing registered range of PFNs
Mel Gormanc7132162006-09-27 01:49:43 -07004461 * @nid: The node id the range is on that should be shrunk
Yinghai Lucc1050b2008-06-13 19:08:52 -07004462 * @start_pfn: The new PFN of the range
4463 * @end_pfn: The new PFN of the range
Mel Gormanc7132162006-09-27 01:49:43 -07004464 *
4465 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004466 * The map is kept near the end physical page range that has already been
4467 * registered. This function allows an arch to shrink an existing registered
4468 * range.
Mel Gormanc7132162006-09-27 01:49:43 -07004469 */
Yinghai Lucc1050b2008-06-13 19:08:52 -07004470void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
4471 unsigned long end_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004472{
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004473 int i, j;
4474 int removed = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07004475
Yinghai Lucc1050b2008-06-13 19:08:52 -07004476 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
4477 nid, start_pfn, end_pfn);
4478
Mel Gormanc7132162006-09-27 01:49:43 -07004479 /* Find the old active region end and shrink */
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004480 for_each_active_range_index_in_nid(i, nid) {
Yinghai Lucc1050b2008-06-13 19:08:52 -07004481 if (early_node_map[i].start_pfn >= start_pfn &&
4482 early_node_map[i].end_pfn <= end_pfn) {
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004483 /* clear it */
Yinghai Lucc1050b2008-06-13 19:08:52 -07004484 early_node_map[i].start_pfn = 0;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004485 early_node_map[i].end_pfn = 0;
4486 removed = 1;
4487 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004488 }
Yinghai Lucc1050b2008-06-13 19:08:52 -07004489 if (early_node_map[i].start_pfn < start_pfn &&
4490 early_node_map[i].end_pfn > start_pfn) {
4491 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
4492 early_node_map[i].end_pfn = start_pfn;
4493 if (temp_end_pfn > end_pfn)
4494 add_active_range(nid, end_pfn, temp_end_pfn);
4495 continue;
4496 }
4497 if (early_node_map[i].start_pfn >= start_pfn &&
4498 early_node_map[i].end_pfn > end_pfn &&
4499 early_node_map[i].start_pfn < end_pfn) {
4500 early_node_map[i].start_pfn = end_pfn;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004501 continue;
4502 }
4503 }
4504
4505 if (!removed)
4506 return;
4507
4508 /* remove the blank ones */
4509 for (i = nr_nodemap_entries - 1; i > 0; i--) {
4510 if (early_node_map[i].nid != nid)
4511 continue;
4512 if (early_node_map[i].end_pfn)
4513 continue;
4514 /* we found it, get rid of it */
4515 for (j = i; j < nr_nodemap_entries - 1; j++)
4516 memcpy(&early_node_map[j], &early_node_map[j+1],
4517 sizeof(early_node_map[j]));
4518 j = nr_nodemap_entries - 1;
4519 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
4520 nr_nodemap_entries--;
4521 }
Mel Gormanc7132162006-09-27 01:49:43 -07004522}
4523
4524/**
4525 * remove_all_active_ranges - Remove all currently registered regions
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004526 *
Mel Gormanc7132162006-09-27 01:49:43 -07004527 * During discovery, it may be found that a table like SRAT is invalid
4528 * and an alternative discovery method must be used. This function removes
4529 * all currently registered regions.
4530 */
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004531void __init remove_all_active_ranges(void)
Mel Gormanc7132162006-09-27 01:49:43 -07004532{
4533 memset(early_node_map, 0, sizeof(early_node_map));
4534 nr_nodemap_entries = 0;
4535}
4536
4537/* Compare two active node_active_regions */
4538static int __init cmp_node_active_region(const void *a, const void *b)
4539{
4540 struct node_active_region *arange = (struct node_active_region *)a;
4541 struct node_active_region *brange = (struct node_active_region *)b;
4542
4543 /* Done this way to avoid overflows */
4544 if (arange->start_pfn > brange->start_pfn)
4545 return 1;
4546 if (arange->start_pfn < brange->start_pfn)
4547 return -1;
4548
4549 return 0;
4550}
4551
4552/* sort the node_map by start_pfn */
Yinghai Lu32996252009-12-15 17:59:02 -08004553void __init sort_node_map(void)
Mel Gormanc7132162006-09-27 01:49:43 -07004554{
4555 sort(early_node_map, (size_t)nr_nodemap_entries,
4556 sizeof(struct node_active_region),
4557 cmp_node_active_region, NULL);
4558}
4559
Mel Gormana6af2bc2007-02-10 01:42:57 -08004560/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004561static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07004562{
4563 int i;
Mel Gormana6af2bc2007-02-10 01:42:57 -08004564 unsigned long min_pfn = ULONG_MAX;
Mel Gorman1abbfb42006-11-23 12:01:41 +00004565
Mel Gormanc7132162006-09-27 01:49:43 -07004566 /* Assuming a sorted map, the first range found has the starting pfn */
4567 for_each_active_range_index_in_nid(i, nid)
Mel Gormana6af2bc2007-02-10 01:42:57 -08004568 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004569
Mel Gormana6af2bc2007-02-10 01:42:57 -08004570 if (min_pfn == ULONG_MAX) {
4571 printk(KERN_WARNING
Paul Jackson2bc0d262008-06-22 07:22:17 -07004572 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08004573 return 0;
4574 }
4575
4576 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004577}
4578
4579/**
4580 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4581 *
4582 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004583 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07004584 */
4585unsigned long __init find_min_pfn_with_active_regions(void)
4586{
4587 return find_min_pfn_for_node(MAX_NUMNODES);
4588}
4589
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004590/*
4591 * early_calculate_totalpages()
4592 * Sum pages in active regions for movable zone.
4593 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4594 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07004595static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07004596{
4597 int i;
4598 unsigned long totalpages = 0;
4599
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004600 for (i = 0; i < nr_nodemap_entries; i++) {
4601 unsigned long pages = early_node_map[i].end_pfn -
Mel Gorman7e63efe2007-07-17 04:03:15 -07004602 early_node_map[i].start_pfn;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004603 totalpages += pages;
4604 if (pages)
4605 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
4606 }
4607 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07004608}
4609
Mel Gorman2a1e2742007-07-17 04:03:12 -07004610/*
4611 * Find the PFN the Movable zone begins in each node. Kernel memory
4612 * is spread evenly between nodes as long as the nodes have enough
4613 * memory. When they don't, some nodes will have more kernelcore than
4614 * others
4615 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004616static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004617{
4618 int i, nid;
4619 unsigned long usable_startpfn;
4620 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07004621 /* save the state before borrow the nodemask */
4622 nodemask_t saved_node_state = node_states[N_HIGH_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004623 unsigned long totalpages = early_calculate_totalpages();
4624 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004625
Mel Gorman7e63efe2007-07-17 04:03:15 -07004626 /*
4627 * If movablecore was specified, calculate what size of
4628 * kernelcore that corresponds so that memory usable for
4629 * any allocation type is evenly spread. If both kernelcore
4630 * and movablecore are specified, then the value of kernelcore
4631 * will be used for required_kernelcore if it's greater than
4632 * what movablecore would have allowed.
4633 */
4634 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07004635 unsigned long corepages;
4636
4637 /*
4638 * Round-up so that ZONE_MOVABLE is at least as large as what
4639 * was requested by the user
4640 */
4641 required_movablecore =
4642 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
4643 corepages = totalpages - required_movablecore;
4644
4645 required_kernelcore = max(required_kernelcore, corepages);
4646 }
4647
Mel Gorman2a1e2742007-07-17 04:03:12 -07004648 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4649 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07004650 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004651
4652 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4653 find_usable_zone_for_movable();
4654 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4655
4656restart:
4657 /* Spread kernelcore memory as evenly as possible throughout nodes */
4658 kernelcore_node = required_kernelcore / usable_nodes;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004659 for_each_node_state(nid, N_HIGH_MEMORY) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004660 /*
4661 * Recalculate kernelcore_node if the division per node
4662 * now exceeds what is necessary to satisfy the requested
4663 * amount of memory for the kernel
4664 */
4665 if (required_kernelcore < kernelcore_node)
4666 kernelcore_node = required_kernelcore / usable_nodes;
4667
4668 /*
4669 * As the map is walked, we track how much memory is usable
4670 * by the kernel using kernelcore_remaining. When it is
4671 * 0, the rest of the node is usable by ZONE_MOVABLE
4672 */
4673 kernelcore_remaining = kernelcore_node;
4674
4675 /* Go through each range of PFNs within this node */
4676 for_each_active_range_index_in_nid(i, nid) {
4677 unsigned long start_pfn, end_pfn;
4678 unsigned long size_pages;
4679
4680 start_pfn = max(early_node_map[i].start_pfn,
4681 zone_movable_pfn[nid]);
4682 end_pfn = early_node_map[i].end_pfn;
4683 if (start_pfn >= end_pfn)
4684 continue;
4685
4686 /* Account for what is only usable for kernelcore */
4687 if (start_pfn < usable_startpfn) {
4688 unsigned long kernel_pages;
4689 kernel_pages = min(end_pfn, usable_startpfn)
4690 - start_pfn;
4691
4692 kernelcore_remaining -= min(kernel_pages,
4693 kernelcore_remaining);
4694 required_kernelcore -= min(kernel_pages,
4695 required_kernelcore);
4696
4697 /* Continue if range is now fully accounted */
4698 if (end_pfn <= usable_startpfn) {
4699
4700 /*
4701 * Push zone_movable_pfn to the end so
4702 * that if we have to rebalance
4703 * kernelcore across nodes, we will
4704 * not double account here
4705 */
4706 zone_movable_pfn[nid] = end_pfn;
4707 continue;
4708 }
4709 start_pfn = usable_startpfn;
4710 }
4711
4712 /*
4713 * The usable PFN range for ZONE_MOVABLE is from
4714 * start_pfn->end_pfn. Calculate size_pages as the
4715 * number of pages used as kernelcore
4716 */
4717 size_pages = end_pfn - start_pfn;
4718 if (size_pages > kernelcore_remaining)
4719 size_pages = kernelcore_remaining;
4720 zone_movable_pfn[nid] = start_pfn + size_pages;
4721
4722 /*
4723 * Some kernelcore has been met, update counts and
4724 * break if the kernelcore for this node has been
4725 * satisified
4726 */
4727 required_kernelcore -= min(required_kernelcore,
4728 size_pages);
4729 kernelcore_remaining -= size_pages;
4730 if (!kernelcore_remaining)
4731 break;
4732 }
4733 }
4734
4735 /*
4736 * If there is still required_kernelcore, we do another pass with one
4737 * less node in the count. This will push zone_movable_pfn[nid] further
4738 * along on the nodes that still have memory until kernelcore is
4739 * satisified
4740 */
4741 usable_nodes--;
4742 if (usable_nodes && required_kernelcore > usable_nodes)
4743 goto restart;
4744
4745 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4746 for (nid = 0; nid < MAX_NUMNODES; nid++)
4747 zone_movable_pfn[nid] =
4748 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07004749
4750out:
4751 /* restore the node_state */
4752 node_states[N_HIGH_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004753}
4754
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004755/* Any regular memory on that node ? */
4756static void check_for_regular_memory(pg_data_t *pgdat)
4757{
4758#ifdef CONFIG_HIGHMEM
4759 enum zone_type zone_type;
4760
4761 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4762 struct zone *zone = &pgdat->node_zones[zone_type];
4763 if (zone->present_pages)
4764 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4765 }
4766#endif
4767}
4768
Mel Gormanc7132162006-09-27 01:49:43 -07004769/**
4770 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004771 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07004772 *
4773 * This will call free_area_init_node() for each active node in the system.
4774 * Using the page ranges provided by add_active_range(), the size of each
4775 * zone in each node and their holes is calculated. If the maximum PFN
4776 * between two adjacent zones match, it is assumed that the zone is empty.
4777 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4778 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4779 * starts where the previous one ended. For example, ZONE_DMA32 starts
4780 * at arch_max_dma_pfn.
4781 */
4782void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4783{
4784 unsigned long nid;
Andrew Mortondb991002008-10-15 22:01:07 -07004785 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004786
Mel Gormana6af2bc2007-02-10 01:42:57 -08004787 /* Sort early_node_map as initialisation assumes it is sorted */
4788 sort_node_map();
4789
Mel Gormanc7132162006-09-27 01:49:43 -07004790 /* Record where the zone boundaries are */
4791 memset(arch_zone_lowest_possible_pfn, 0,
4792 sizeof(arch_zone_lowest_possible_pfn));
4793 memset(arch_zone_highest_possible_pfn, 0,
4794 sizeof(arch_zone_highest_possible_pfn));
4795 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4796 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4797 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004798 if (i == ZONE_MOVABLE)
4799 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004800 arch_zone_lowest_possible_pfn[i] =
4801 arch_zone_highest_possible_pfn[i-1];
4802 arch_zone_highest_possible_pfn[i] =
4803 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4804 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07004805 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4806 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4807
4808 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4809 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4810 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004811
Mel Gormanc7132162006-09-27 01:49:43 -07004812 /* Print out the zone ranges */
4813 printk("Zone PFN ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07004814 for (i = 0; i < MAX_NR_ZONES; i++) {
4815 if (i == ZONE_MOVABLE)
4816 continue;
David Rientjes72f0ba02010-03-05 13:42:14 -08004817 printk(" %-8s ", zone_names[i]);
4818 if (arch_zone_lowest_possible_pfn[i] ==
4819 arch_zone_highest_possible_pfn[i])
4820 printk("empty\n");
4821 else
4822 printk("%0#10lx -> %0#10lx\n",
Mel Gormanc7132162006-09-27 01:49:43 -07004823 arch_zone_lowest_possible_pfn[i],
4824 arch_zone_highest_possible_pfn[i]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004825 }
4826
4827 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4828 printk("Movable zone start PFN for each node\n");
4829 for (i = 0; i < MAX_NUMNODES; i++) {
4830 if (zone_movable_pfn[i])
4831 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4832 }
Mel Gormanc7132162006-09-27 01:49:43 -07004833
4834 /* Print out the early_node_map[] */
4835 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4836 for (i = 0; i < nr_nodemap_entries; i++)
Paul Jackson5dab8ec2008-06-25 05:44:40 -07004837 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004838 early_node_map[i].start_pfn,
4839 early_node_map[i].end_pfn);
4840
4841 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07004842 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08004843 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07004844 for_each_online_node(nid) {
4845 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07004846 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07004847 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004848
4849 /* Any memory on that node */
4850 if (pgdat->node_present_pages)
4851 node_set_state(nid, N_HIGH_MEMORY);
4852 check_for_regular_memory(pgdat);
Mel Gormanc7132162006-09-27 01:49:43 -07004853 }
4854}
Mel Gorman2a1e2742007-07-17 04:03:12 -07004855
Mel Gorman7e63efe2007-07-17 04:03:15 -07004856static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004857{
4858 unsigned long long coremem;
4859 if (!p)
4860 return -EINVAL;
4861
4862 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004863 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004864
Mel Gorman7e63efe2007-07-17 04:03:15 -07004865 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07004866 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4867
4868 return 0;
4869}
Mel Gormaned7ed362007-07-17 04:03:14 -07004870
Mel Gorman7e63efe2007-07-17 04:03:15 -07004871/*
4872 * kernelcore=size sets the amount of memory for use for allocations that
4873 * cannot be reclaimed or migrated.
4874 */
4875static int __init cmdline_parse_kernelcore(char *p)
4876{
4877 return cmdline_parse_core(p, &required_kernelcore);
4878}
4879
4880/*
4881 * movablecore=size sets the amount of memory for use for allocations that
4882 * can be reclaimed or migrated.
4883 */
4884static int __init cmdline_parse_movablecore(char *p)
4885{
4886 return cmdline_parse_core(p, &required_movablecore);
4887}
4888
Mel Gormaned7ed362007-07-17 04:03:14 -07004889early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004890early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07004891
Mel Gormanc7132162006-09-27 01:49:43 -07004892#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4893
Mel Gorman0e0b8642006-09-27 01:49:56 -07004894/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004895 * set_dma_reserve - set the specified number of pages reserved in the first zone
4896 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07004897 *
4898 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4899 * In the DMA zone, a significant percentage may be consumed by kernel image
4900 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004901 * function may optionally be used to account for unfreeable pages in the
4902 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4903 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07004904 */
4905void __init set_dma_reserve(unsigned long new_dma_reserve)
4906{
4907 dma_reserve = new_dma_reserve;
4908}
4909
Linus Torvalds1da177e2005-04-16 15:20:36 -07004910void __init free_area_init(unsigned long *zones_size)
4911{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004912 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004913 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4914}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004915
Linus Torvalds1da177e2005-04-16 15:20:36 -07004916static int page_alloc_cpu_notify(struct notifier_block *self,
4917 unsigned long action, void *hcpu)
4918{
4919 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004920
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07004921 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004922 drain_pages(cpu);
4923
4924 /*
4925 * Spill the event counters of the dead processor
4926 * into the current processors event counters.
4927 * This artificially elevates the count of the current
4928 * processor.
4929 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07004930 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004931
4932 /*
4933 * Zero the differential counters of the dead processor
4934 * so that the vm statistics are consistent.
4935 *
4936 * This is only okay since the processor is dead and cannot
4937 * race with what we are doing.
4938 */
Christoph Lameter2244b952006-06-30 01:55:33 -07004939 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004940 }
4941 return NOTIFY_OK;
4942}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004943
4944void __init page_alloc_init(void)
4945{
4946 hotcpu_notifier(page_alloc_cpu_notify, 0);
4947}
4948
4949/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004950 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4951 * or min_free_kbytes changes.
4952 */
4953static void calculate_totalreserve_pages(void)
4954{
4955 struct pglist_data *pgdat;
4956 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004957 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004958
4959 for_each_online_pgdat(pgdat) {
4960 for (i = 0; i < MAX_NR_ZONES; i++) {
4961 struct zone *zone = pgdat->node_zones + i;
4962 unsigned long max = 0;
4963
4964 /* Find valid and maximum lowmem_reserve in the zone */
4965 for (j = i; j < MAX_NR_ZONES; j++) {
4966 if (zone->lowmem_reserve[j] > max)
4967 max = zone->lowmem_reserve[j];
4968 }
4969
Mel Gorman41858962009-06-16 15:32:12 -07004970 /* we treat the high watermark as reserved pages. */
4971 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004972
4973 if (max > zone->present_pages)
4974 max = zone->present_pages;
4975 reserve_pages += max;
4976 }
4977 }
4978 totalreserve_pages = reserve_pages;
4979}
4980
4981/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07004982 * setup_per_zone_lowmem_reserve - called whenever
4983 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4984 * has a correct pages reserved value, so an adequate number of
4985 * pages are left in the zone after a successful __alloc_pages().
4986 */
4987static void setup_per_zone_lowmem_reserve(void)
4988{
4989 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004990 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004991
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08004992 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004993 for (j = 0; j < MAX_NR_ZONES; j++) {
4994 struct zone *zone = pgdat->node_zones + j;
4995 unsigned long present_pages = zone->present_pages;
4996
4997 zone->lowmem_reserve[j] = 0;
4998
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004999 idx = j;
5000 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005001 struct zone *lower_zone;
5002
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005003 idx--;
5004
Linus Torvalds1da177e2005-04-16 15:20:36 -07005005 if (sysctl_lowmem_reserve_ratio[idx] < 1)
5006 sysctl_lowmem_reserve_ratio[idx] = 1;
5007
5008 lower_zone = pgdat->node_zones + idx;
5009 lower_zone->lowmem_reserve[j] = present_pages /
5010 sysctl_lowmem_reserve_ratio[idx];
5011 present_pages += lower_zone->present_pages;
5012 }
5013 }
5014 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005015
5016 /* update totalreserve_pages */
5017 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005018}
5019
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005020/**
Minchan Kimbc75d332009-06-16 15:32:48 -07005021 * setup_per_zone_wmarks - called when min_free_kbytes changes
Minchan Kimbce73942009-06-16 15:32:50 -07005022 * or when memory is hot-{added|removed}
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005023 *
Minchan Kimbc75d332009-06-16 15:32:48 -07005024 * Ensures that the watermark[min,low,high] values for each zone are set
5025 * correctly with respect to min_free_kbytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005026 */
Minchan Kimbc75d332009-06-16 15:32:48 -07005027void setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005028{
5029 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
5030 unsigned long lowmem_pages = 0;
5031 struct zone *zone;
5032 unsigned long flags;
5033
5034 /* Calculate total number of !ZONE_HIGHMEM pages */
5035 for_each_zone(zone) {
5036 if (!is_highmem(zone))
5037 lowmem_pages += zone->present_pages;
5038 }
5039
5040 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07005041 u64 tmp;
5042
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005043 spin_lock_irqsave(&zone->lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07005044 tmp = (u64)pages_min * zone->present_pages;
5045 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005046 if (is_highmem(zone)) {
5047 /*
Nick Piggin669ed172005-11-13 16:06:45 -08005048 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5049 * need highmem pages, so cap pages_min to a small
5050 * value here.
5051 *
Mel Gorman41858962009-06-16 15:32:12 -07005052 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Nick Piggin669ed172005-11-13 16:06:45 -08005053 * deltas controls asynch page reclaim, and so should
5054 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005055 */
5056 int min_pages;
5057
5058 min_pages = zone->present_pages / 1024;
5059 if (min_pages < SWAP_CLUSTER_MAX)
5060 min_pages = SWAP_CLUSTER_MAX;
5061 if (min_pages > 128)
5062 min_pages = 128;
Mel Gorman41858962009-06-16 15:32:12 -07005063 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005064 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08005065 /*
5066 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07005067 * proportionate to the zone's size.
5068 */
Mel Gorman41858962009-06-16 15:32:12 -07005069 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005070 }
5071
Mel Gorman41858962009-06-16 15:32:12 -07005072 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
5073 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Mel Gorman56fd56b2007-10-16 01:25:58 -07005074 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005075 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005076 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005077
5078 /* update totalreserve_pages */
5079 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005080}
5081
Randy Dunlap55a44622009-09-21 17:01:20 -07005082/*
Rik van Riel556adec2008-10-18 20:26:34 -07005083 * The inactive anon list should be small enough that the VM never has to
5084 * do too much work, but large enough that each inactive page has a chance
5085 * to be referenced again before it is swapped out.
5086 *
5087 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5088 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5089 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5090 * the anonymous pages are kept on the inactive list.
5091 *
5092 * total target max
5093 * memory ratio inactive anon
5094 * -------------------------------------
5095 * 10MB 1 5MB
5096 * 100MB 1 50MB
5097 * 1GB 3 250MB
5098 * 10GB 10 0.9GB
5099 * 100GB 31 3GB
5100 * 1TB 101 10GB
5101 * 10TB 320 32GB
5102 */
Minchan Kim96cb4df2009-06-16 15:32:49 -07005103void calculate_zone_inactive_ratio(struct zone *zone)
5104{
5105 unsigned int gb, ratio;
5106
5107 /* Zone size in gigabytes */
5108 gb = zone->present_pages >> (30 - PAGE_SHIFT);
5109 if (gb)
5110 ratio = int_sqrt(10 * gb);
5111 else
5112 ratio = 1;
5113
5114 zone->inactive_ratio = ratio;
5115}
5116
Cyrill Gorcunove9bb35d2009-06-16 15:32:32 -07005117static void __init setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07005118{
5119 struct zone *zone;
5120
Minchan Kim96cb4df2009-06-16 15:32:49 -07005121 for_each_zone(zone)
5122 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07005123}
5124
Linus Torvalds1da177e2005-04-16 15:20:36 -07005125/*
5126 * Initialise min_free_kbytes.
5127 *
5128 * For small machines we want it small (128k min). For large machines
5129 * we want it large (64MB max). But it is not linear, because network
5130 * bandwidth does not increase linearly with machine size. We use
5131 *
5132 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5133 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5134 *
5135 * which yields
5136 *
5137 * 16MB: 512k
5138 * 32MB: 724k
5139 * 64MB: 1024k
5140 * 128MB: 1448k
5141 * 256MB: 2048k
5142 * 512MB: 2896k
5143 * 1024MB: 4096k
5144 * 2048MB: 5792k
5145 * 4096MB: 8192k
5146 * 8192MB: 11584k
5147 * 16384MB: 16384k
5148 */
Minchan Kimbc75d332009-06-16 15:32:48 -07005149static int __init init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005150{
5151 unsigned long lowmem_kbytes;
5152
5153 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
5154
5155 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
5156 if (min_free_kbytes < 128)
5157 min_free_kbytes = 128;
5158 if (min_free_kbytes > 65536)
5159 min_free_kbytes = 65536;
Minchan Kimbc75d332009-06-16 15:32:48 -07005160 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005161 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07005162 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005163 return 0;
5164}
Minchan Kimbc75d332009-06-16 15:32:48 -07005165module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005166
5167/*
5168 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5169 * that we can call two helper functions whenever min_free_kbytes
5170 * changes.
5171 */
5172int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005173 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005174{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005175 proc_dointvec(table, write, buffer, length, ppos);
Mel Gorman3b1d92c2007-05-06 14:49:30 -07005176 if (write)
Minchan Kimbc75d332009-06-16 15:32:48 -07005177 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005178 return 0;
5179}
5180
Christoph Lameter96146342006-07-03 00:24:13 -07005181#ifdef CONFIG_NUMA
5182int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005183 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07005184{
5185 struct zone *zone;
5186 int rc;
5187
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005188 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07005189 if (rc)
5190 return rc;
5191
5192 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07005193 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07005194 sysctl_min_unmapped_ratio) / 100;
5195 return 0;
5196}
Christoph Lameter0ff38492006-09-25 23:31:52 -07005197
5198int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005199 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07005200{
5201 struct zone *zone;
5202 int rc;
5203
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005204 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07005205 if (rc)
5206 return rc;
5207
5208 for_each_zone(zone)
5209 zone->min_slab_pages = (zone->present_pages *
5210 sysctl_min_slab_ratio) / 100;
5211 return 0;
5212}
Christoph Lameter96146342006-07-03 00:24:13 -07005213#endif
5214
Linus Torvalds1da177e2005-04-16 15:20:36 -07005215/*
5216 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5217 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5218 * whenever sysctl_lowmem_reserve_ratio changes.
5219 *
5220 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07005221 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07005222 * if in function of the boot time zone sizes.
5223 */
5224int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005225 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005226{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005227 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005228 setup_per_zone_lowmem_reserve();
5229 return 0;
5230}
5231
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005232/*
5233 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5234 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5235 * can have before it gets flushed back to buddy allocator.
5236 */
5237
5238int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005239 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005240{
5241 struct zone *zone;
5242 unsigned int cpu;
5243 int ret;
5244
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005245 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005246 if (!write || (ret == -EINVAL))
5247 return ret;
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07005248 for_each_populated_zone(zone) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005249 for_each_possible_cpu(cpu) {
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005250 unsigned long high;
5251 high = zone->present_pages / percpu_pagelist_fraction;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005252 setup_pagelist_highmark(
5253 per_cpu_ptr(zone->pageset, cpu), high);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005254 }
5255 }
5256 return 0;
5257}
5258
David S. Millerf034b5d2006-08-24 03:08:07 -07005259int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005260
5261#ifdef CONFIG_NUMA
5262static int __init set_hashdist(char *str)
5263{
5264 if (!str)
5265 return 0;
5266 hashdist = simple_strtoul(str, &str, 0);
5267 return 1;
5268}
5269__setup("hashdist=", set_hashdist);
5270#endif
5271
5272/*
5273 * allocate a large system hash table from bootmem
5274 * - it is assumed that the hash table must contain an exact power-of-2
5275 * quantity of entries
5276 * - limit is the number of hash buckets, not the total allocation size
5277 */
5278void *__init alloc_large_system_hash(const char *tablename,
5279 unsigned long bucketsize,
5280 unsigned long numentries,
5281 int scale,
5282 int flags,
5283 unsigned int *_hash_shift,
5284 unsigned int *_hash_mask,
5285 unsigned long limit)
5286{
5287 unsigned long long max = limit;
5288 unsigned long log2qty, size;
5289 void *table = NULL;
5290
5291 /* allow the kernel cmdline to have a say */
5292 if (!numentries) {
5293 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08005294 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005295 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
5296 numentries >>= 20 - PAGE_SHIFT;
5297 numentries <<= 20 - PAGE_SHIFT;
5298
5299 /* limit to 1 bucket per 2^scale bytes of low memory */
5300 if (scale > PAGE_SHIFT)
5301 numentries >>= (scale - PAGE_SHIFT);
5302 else
5303 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08005304
5305 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07005306 if (unlikely(flags & HASH_SMALL)) {
5307 /* Makes no sense without HASH_EARLY */
5308 WARN_ON(!(flags & HASH_EARLY));
5309 if (!(numentries >> *_hash_shift)) {
5310 numentries = 1UL << *_hash_shift;
5311 BUG_ON(!numentries);
5312 }
5313 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08005314 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005315 }
John Hawkes6e692ed2006-03-25 03:08:02 -08005316 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005317
5318 /* limit allocation size to 1/16 total memory by default */
5319 if (max == 0) {
5320 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
5321 do_div(max, bucketsize);
5322 }
5323
5324 if (numentries > max)
5325 numentries = max;
5326
David Howellsf0d1b0b2006-12-08 02:37:49 -08005327 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005328
5329 do {
5330 size = bucketsize << log2qty;
5331 if (flags & HASH_EARLY)
Jan Beulich74768ed2008-08-12 15:08:39 -07005332 table = alloc_bootmem_nopanic(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005333 else if (hashdist)
5334 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
5335 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07005336 /*
5337 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07005338 * some pages at the end of hash table which
5339 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07005340 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005341 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07005342 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005343 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
5344 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005345 }
5346 } while (!table && size > PAGE_SIZE && --log2qty);
5347
5348 if (!table)
5349 panic("Failed to allocate %s hash table\n", tablename);
5350
Robin Holtf241e662010-10-07 12:59:26 -07005351 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07005352 tablename,
Robin Holtf241e662010-10-07 12:59:26 -07005353 (1UL << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08005354 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005355 size);
5356
5357 if (_hash_shift)
5358 *_hash_shift = log2qty;
5359 if (_hash_mask)
5360 *_hash_mask = (1 << log2qty) - 1;
5361
5362 return table;
5363}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08005364
Mel Gorman835c1342007-10-16 01:25:47 -07005365/* Return a pointer to the bitmap storing bits affecting a block of pages */
5366static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
5367 unsigned long pfn)
5368{
5369#ifdef CONFIG_SPARSEMEM
5370 return __pfn_to_section(pfn)->pageblock_flags;
5371#else
5372 return zone->pageblock_flags;
5373#endif /* CONFIG_SPARSEMEM */
5374}
Andrew Morton6220ec72006-10-19 23:29:05 -07005375
Mel Gorman835c1342007-10-16 01:25:47 -07005376static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
5377{
5378#ifdef CONFIG_SPARSEMEM
5379 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07005380 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005381#else
5382 pfn = pfn - zone->zone_start_pfn;
Mel Gormand9c23402007-10-16 01:26:01 -07005383 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005384#endif /* CONFIG_SPARSEMEM */
5385}
5386
5387/**
Mel Gormand9c23402007-10-16 01:26:01 -07005388 * 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 -07005389 * @page: The page within the block of interest
5390 * @start_bitidx: The first bit of interest to retrieve
5391 * @end_bitidx: The last bit of interest
5392 * returns pageblock_bits flags
5393 */
5394unsigned long get_pageblock_flags_group(struct page *page,
5395 int start_bitidx, int end_bitidx)
5396{
5397 struct zone *zone;
5398 unsigned long *bitmap;
5399 unsigned long pfn, bitidx;
5400 unsigned long flags = 0;
5401 unsigned long value = 1;
5402
5403 zone = page_zone(page);
5404 pfn = page_to_pfn(page);
5405 bitmap = get_pageblock_bitmap(zone, pfn);
5406 bitidx = pfn_to_bitidx(zone, pfn);
5407
5408 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5409 if (test_bit(bitidx + start_bitidx, bitmap))
5410 flags |= value;
5411
5412 return flags;
5413}
5414
5415/**
Mel Gormand9c23402007-10-16 01:26:01 -07005416 * 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 -07005417 * @page: The page within the block of interest
5418 * @start_bitidx: The first bit of interest
5419 * @end_bitidx: The last bit of interest
5420 * @flags: The flags to set
5421 */
5422void set_pageblock_flags_group(struct page *page, unsigned long flags,
5423 int start_bitidx, int end_bitidx)
5424{
5425 struct zone *zone;
5426 unsigned long *bitmap;
5427 unsigned long pfn, bitidx;
5428 unsigned long value = 1;
5429
5430 zone = page_zone(page);
5431 pfn = page_to_pfn(page);
5432 bitmap = get_pageblock_bitmap(zone, pfn);
5433 bitidx = pfn_to_bitidx(zone, pfn);
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07005434 VM_BUG_ON(pfn < zone->zone_start_pfn);
5435 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
Mel Gorman835c1342007-10-16 01:25:47 -07005436
5437 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5438 if (flags & value)
5439 __set_bit(bitidx + start_bitidx, bitmap);
5440 else
5441 __clear_bit(bitidx + start_bitidx, bitmap);
5442}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005443
5444/*
5445 * This is designed as sub function...plz see page_isolation.c also.
5446 * set/clear page block's type to be ISOLATE.
5447 * page allocater never alloc memory from ISOLATE block.
5448 */
5449
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005450static int
5451__count_immobile_pages(struct zone *zone, struct page *page, int count)
5452{
5453 unsigned long pfn, iter, found;
5454 /*
5455 * For avoiding noise data, lru_add_drain_all() should be called
5456 * If ZONE_MOVABLE, the zone never contains immobile pages
5457 */
5458 if (zone_idx(zone) == ZONE_MOVABLE)
5459 return true;
5460
5461 if (get_pageblock_migratetype(page) == MIGRATE_MOVABLE)
5462 return true;
5463
5464 pfn = page_to_pfn(page);
5465 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
5466 unsigned long check = pfn + iter;
5467
Namhyung Kim29723fc2011-02-25 14:44:25 -08005468 if (!pfn_valid_within(check))
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005469 continue;
Namhyung Kim29723fc2011-02-25 14:44:25 -08005470
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005471 page = pfn_to_page(check);
5472 if (!page_count(page)) {
5473 if (PageBuddy(page))
5474 iter += (1 << page_order(page)) - 1;
5475 continue;
5476 }
5477 if (!PageLRU(page))
5478 found++;
5479 /*
5480 * If there are RECLAIMABLE pages, we need to check it.
5481 * But now, memory offline itself doesn't call shrink_slab()
5482 * and it still to be fixed.
5483 */
5484 /*
5485 * If the page is not RAM, page_count()should be 0.
5486 * we don't need more check. This is an _used_ not-movable page.
5487 *
5488 * The problematic thing here is PG_reserved pages. PG_reserved
5489 * is set to both of a memory hole page and a _used_ kernel
5490 * page at boot.
5491 */
5492 if (found > count)
5493 return false;
5494 }
5495 return true;
5496}
5497
5498bool is_pageblock_removable_nolock(struct page *page)
5499{
5500 struct zone *zone = page_zone(page);
5501 return __count_immobile_pages(zone, page, 0);
5502}
5503
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005504int set_migratetype_isolate(struct page *page)
5505{
5506 struct zone *zone;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005507 unsigned long flags, pfn;
Robert Jennings925cc712009-12-17 14:44:38 +00005508 struct memory_isolate_notify arg;
5509 int notifier_ret;
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005510 int ret = -EBUSY;
Shaohua Li8e7e40d2009-09-21 17:01:18 -07005511 int zone_idx;
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005512
5513 zone = page_zone(page);
Shaohua Li8e7e40d2009-09-21 17:01:18 -07005514 zone_idx = zone_idx(zone);
Robert Jennings925cc712009-12-17 14:44:38 +00005515
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005516 spin_lock_irqsave(&zone->lock, flags);
Robert Jennings925cc712009-12-17 14:44:38 +00005517
5518 pfn = page_to_pfn(page);
5519 arg.start_pfn = pfn;
5520 arg.nr_pages = pageblock_nr_pages;
5521 arg.pages_found = 0;
5522
5523 /*
5524 * It may be possible to isolate a pageblock even if the
5525 * migratetype is not MIGRATE_MOVABLE. The memory isolation
5526 * notifier chain is used by balloon drivers to return the
5527 * number of pages in a range that are held by the balloon
5528 * driver to shrink memory. If all the pages are accounted for
5529 * by balloons, are free, or on the LRU, isolation can continue.
5530 * Later, for example, when memory hotplug notifier runs, these
5531 * pages reported as "can be isolated" should be isolated(freed)
5532 * by the balloon driver through the memory notifier chain.
5533 */
5534 notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg);
5535 notifier_ret = notifier_to_errno(notifier_ret);
KAMEZAWA Hiroyuki4b204772010-10-26 14:21:29 -07005536 if (notifier_ret)
Robert Jennings925cc712009-12-17 14:44:38 +00005537 goto out;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005538 /*
5539 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
5540 * We just check MOVABLE pages.
5541 */
5542 if (__count_immobile_pages(zone, page, arg.pages_found))
Robert Jennings925cc712009-12-17 14:44:38 +00005543 ret = 0;
5544
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005545 /*
5546 * immobile means "not-on-lru" paes. If immobile is larger than
5547 * removable-by-driver pages reported by notifier, we'll fail.
5548 */
5549
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005550out:
Robert Jennings925cc712009-12-17 14:44:38 +00005551 if (!ret) {
5552 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
5553 move_freepages_block(zone, page, MIGRATE_ISOLATE);
5554 }
5555
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005556 spin_unlock_irqrestore(&zone->lock, flags);
5557 if (!ret)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005558 drain_all_pages();
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005559 return ret;
5560}
5561
5562void unset_migratetype_isolate(struct page *page)
5563{
5564 struct zone *zone;
5565 unsigned long flags;
5566 zone = page_zone(page);
5567 spin_lock_irqsave(&zone->lock, flags);
5568 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
5569 goto out;
5570 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5571 move_freepages_block(zone, page, MIGRATE_MOVABLE);
5572out:
5573 spin_unlock_irqrestore(&zone->lock, flags);
5574}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07005575
5576#ifdef CONFIG_MEMORY_HOTREMOVE
5577/*
5578 * All pages in the range must be isolated before calling this.
5579 */
5580void
5581__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
5582{
5583 struct page *page;
5584 struct zone *zone;
5585 int order, i;
5586 unsigned long pfn;
5587 unsigned long flags;
5588 /* find the first valid pfn */
5589 for (pfn = start_pfn; pfn < end_pfn; pfn++)
5590 if (pfn_valid(pfn))
5591 break;
5592 if (pfn == end_pfn)
5593 return;
5594 zone = page_zone(pfn_to_page(pfn));
5595 spin_lock_irqsave(&zone->lock, flags);
5596 pfn = start_pfn;
5597 while (pfn < end_pfn) {
5598 if (!pfn_valid(pfn)) {
5599 pfn++;
5600 continue;
5601 }
5602 page = pfn_to_page(pfn);
5603 BUG_ON(page_count(page));
5604 BUG_ON(!PageBuddy(page));
5605 order = page_order(page);
5606#ifdef CONFIG_DEBUG_VM
5607 printk(KERN_INFO "remove from free list %lx %d %lx\n",
5608 pfn, 1 << order, end_pfn);
5609#endif
5610 list_del(&page->lru);
5611 rmv_page_order(page);
5612 zone->free_area[order].nr_free--;
5613 __mod_zone_page_state(zone, NR_FREE_PAGES,
5614 - (1UL << order));
5615 for (i = 0; i < (1 << order); i++)
5616 SetPageReserved((page+i));
5617 pfn += (1 << order);
5618 }
5619 spin_unlock_irqrestore(&zone->lock, flags);
5620}
5621#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01005622
5623#ifdef CONFIG_MEMORY_FAILURE
5624bool is_free_buddy_page(struct page *page)
5625{
5626 struct zone *zone = page_zone(page);
5627 unsigned long pfn = page_to_pfn(page);
5628 unsigned long flags;
5629 int order;
5630
5631 spin_lock_irqsave(&zone->lock, flags);
5632 for (order = 0; order < MAX_ORDER; order++) {
5633 struct page *page_head = page - (pfn & ((1 << order) - 1));
5634
5635 if (PageBuddy(page_head) && page_order(page_head) >= order)
5636 break;
5637 }
5638 spin_unlock_irqrestore(&zone->lock, flags);
5639
5640 return order < MAX_ORDER;
5641}
5642#endif
Wu Fengguang718a3822010-03-10 15:20:43 -08005643
5644static struct trace_print_flags pageflag_names[] = {
5645 {1UL << PG_locked, "locked" },
5646 {1UL << PG_error, "error" },
5647 {1UL << PG_referenced, "referenced" },
5648 {1UL << PG_uptodate, "uptodate" },
5649 {1UL << PG_dirty, "dirty" },
5650 {1UL << PG_lru, "lru" },
5651 {1UL << PG_active, "active" },
5652 {1UL << PG_slab, "slab" },
5653 {1UL << PG_owner_priv_1, "owner_priv_1" },
5654 {1UL << PG_arch_1, "arch_1" },
5655 {1UL << PG_reserved, "reserved" },
5656 {1UL << PG_private, "private" },
5657 {1UL << PG_private_2, "private_2" },
5658 {1UL << PG_writeback, "writeback" },
5659#ifdef CONFIG_PAGEFLAGS_EXTENDED
5660 {1UL << PG_head, "head" },
5661 {1UL << PG_tail, "tail" },
5662#else
5663 {1UL << PG_compound, "compound" },
5664#endif
5665 {1UL << PG_swapcache, "swapcache" },
5666 {1UL << PG_mappedtodisk, "mappedtodisk" },
5667 {1UL << PG_reclaim, "reclaim" },
Wu Fengguang718a3822010-03-10 15:20:43 -08005668 {1UL << PG_swapbacked, "swapbacked" },
5669 {1UL << PG_unevictable, "unevictable" },
5670#ifdef CONFIG_MMU
5671 {1UL << PG_mlocked, "mlocked" },
5672#endif
5673#ifdef CONFIG_ARCH_USES_PG_UNCACHED
5674 {1UL << PG_uncached, "uncached" },
5675#endif
5676#ifdef CONFIG_MEMORY_FAILURE
5677 {1UL << PG_hwpoison, "hwpoison" },
5678#endif
5679 {-1UL, NULL },
5680};
5681
5682static void dump_page_flags(unsigned long flags)
5683{
5684 const char *delim = "";
5685 unsigned long mask;
5686 int i;
5687
5688 printk(KERN_ALERT "page flags: %#lx(", flags);
5689
5690 /* remove zone id */
5691 flags &= (1UL << NR_PAGEFLAGS) - 1;
5692
5693 for (i = 0; pageflag_names[i].name && flags; i++) {
5694
5695 mask = pageflag_names[i].mask;
5696 if ((flags & mask) != mask)
5697 continue;
5698
5699 flags &= ~mask;
5700 printk("%s%s", delim, pageflag_names[i].name);
5701 delim = "|";
5702 }
5703
5704 /* check for left over flags */
5705 if (flags)
5706 printk("%s%#lx", delim, flags);
5707
5708 printk(")\n");
5709}
5710
5711void dump_page(struct page *page)
5712{
5713 printk(KERN_ALERT
5714 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
Andrea Arcangeli4e9f64c2011-01-13 15:46:29 -08005715 page, atomic_read(&page->_count), page_mapcount(page),
Wu Fengguang718a3822010-03-10 15:20:43 -08005716 page->mapping, page->index);
5717 dump_page_flags(page->flags);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07005718 mem_cgroup_print_bad_page(page);
Wu Fengguang718a3822010-03-10 15:20:43 -08005719}