blob: 8e5726ab0d85789ee553241f4172787d3c88d0d6 [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 Torvalds1da177e2005-04-16 15:20:36 -070057
58#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070059#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070060#include "internal.h"
61
Lee Schermerhorn72812012010-05-26 14:44:56 -070062#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
63DEFINE_PER_CPU(int, numa_node);
64EXPORT_PER_CPU_SYMBOL(numa_node);
65#endif
66
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070067#ifdef CONFIG_HAVE_MEMORYLESS_NODES
68/*
69 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
70 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
71 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
72 * defined in <linux/topology.h>.
73 */
74DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
75EXPORT_PER_CPU_SYMBOL(_numa_mem_);
76#endif
77
Linus Torvalds1da177e2005-04-16 15:20:36 -070078/*
Christoph Lameter13808912007-10-16 01:25:27 -070079 * Array of node states.
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 */
Christoph Lameter13808912007-10-16 01:25:27 -070081nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
82 [N_POSSIBLE] = NODE_MASK_ALL,
83 [N_ONLINE] = { { [0] = 1UL } },
84#ifndef CONFIG_NUMA
85 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
86#ifdef CONFIG_HIGHMEM
87 [N_HIGH_MEMORY] = { { [0] = 1UL } },
88#endif
89 [N_CPU] = { { [0] = 1UL } },
90#endif /* NUMA */
91};
92EXPORT_SYMBOL(node_states);
93
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070094unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -070095unsigned long totalreserve_pages __read_mostly;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -080096int percpu_pagelist_fraction;
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +100097gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -070098
Rafael J. Wysocki452aa692010-03-05 13:42:13 -080099#ifdef CONFIG_PM_SLEEP
100/*
101 * The following functions are used by the suspend/hibernate code to temporarily
102 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
103 * while devices are suspended. To avoid races with the suspend/hibernate code,
104 * they should always be called with pm_mutex held (gfp_allowed_mask also should
105 * only be modified with pm_mutex held, unless the suspend/hibernate code is
106 * guaranteed not to run in parallel with that modification).
107 */
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100108
109static gfp_t saved_gfp_mask;
110
111void pm_restore_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800112{
113 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100114 if (saved_gfp_mask) {
115 gfp_allowed_mask = saved_gfp_mask;
116 saved_gfp_mask = 0;
117 }
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800118}
119
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100120void pm_restrict_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800121{
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800122 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100123 WARN_ON(saved_gfp_mask);
124 saved_gfp_mask = gfp_allowed_mask;
125 gfp_allowed_mask &= ~GFP_IOFS;
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800126}
127#endif /* CONFIG_PM_SLEEP */
128
Mel Gormand9c23402007-10-16 01:26:01 -0700129#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
130int pageblock_order __read_mostly;
131#endif
132
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800133static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -0800134
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135/*
136 * results with 256, 32 in the lowmem_reserve sysctl:
137 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
138 * 1G machine -> (16M dma, 784M normal, 224M high)
139 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
140 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
141 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +0100142 *
143 * TBD: should special case ZONE_DMA32 machines here - in those we normally
144 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700146int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800147#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700148 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -0800149#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700150#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700151 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700152#endif
Christoph Lametere53ef382006-09-25 23:31:14 -0700153#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700154 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700155#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700156 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700157};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158
159EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160
Helge Deller15ad7cd2006-12-06 20:40:36 -0800161static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800162#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700163 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800164#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700165#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700166 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700167#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700168 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700169#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700170 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700171#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700172 "Movable",
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700173};
174
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175int min_free_kbytes = 1024;
176
Jan Beulich2c85f512009-09-21 17:03:07 -0700177static unsigned long __meminitdata nr_kernel_pages;
178static unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700179static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180
Mel Gormanc7132162006-09-27 01:49:43 -0700181#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
182 /*
Simon Arlott183ff222007-10-20 01:27:18 +0200183 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
Mel Gormanc7132162006-09-27 01:49:43 -0700184 * ranges of memory (RAM) that may be registered with add_active_range().
185 * Ranges passed to add_active_range() will be merged if possible
186 * so the number of times add_active_range() can be called is
187 * related to the number of nodes and the number of holes
188 */
189 #ifdef CONFIG_MAX_ACTIVE_REGIONS
190 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
191 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
192 #else
193 #if MAX_NUMNODES >= 32
194 /* If there can be many nodes, allow up to 50 holes per node */
195 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
196 #else
197 /* By default, allow up to 256 distinct regions */
198 #define MAX_ACTIVE_REGIONS 256
199 #endif
200 #endif
201
Jan Beulich98011f52007-07-15 23:38:17 -0700202 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
203 static int __meminitdata nr_nodemap_entries;
204 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
205 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
Adrian Bunkb69a7282008-07-23 21:28:12 -0700206 static unsigned long __initdata required_kernelcore;
Adrian Bunk484f51f2007-10-16 01:26:03 -0700207 static unsigned long __initdata required_movablecore;
Adrian Bunkb69a7282008-07-23 21:28:12 -0700208 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Mel Gorman2a1e2742007-07-17 04:03:12 -0700209
210 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
211 int movable_zone;
212 EXPORT_SYMBOL(movable_zone);
Mel Gormanc7132162006-09-27 01:49:43 -0700213#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
214
Miklos Szeredi418508c2007-05-23 13:57:55 -0700215#if MAX_NUMNODES > 1
216int nr_node_ids __read_mostly = MAX_NUMNODES;
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700217int nr_online_nodes __read_mostly = 1;
Miklos Szeredi418508c2007-05-23 13:57:55 -0700218EXPORT_SYMBOL(nr_node_ids);
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700219EXPORT_SYMBOL(nr_online_nodes);
Miklos Szeredi418508c2007-05-23 13:57:55 -0700220#endif
221
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700222int page_group_by_mobility_disabled __read_mostly;
223
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700224static void set_pageblock_migratetype(struct page *page, int migratetype)
225{
Mel Gorman49255c62009-06-16 15:31:58 -0700226
227 if (unlikely(page_group_by_mobility_disabled))
228 migratetype = MIGRATE_UNMOVABLE;
229
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700230 set_pageblock_flags_group(page, (unsigned long)migratetype,
231 PB_migrate, PB_migrate_end);
232}
233
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -0700234bool oom_killer_disabled __read_mostly;
235
Nick Piggin13e74442006-01-06 00:10:58 -0800236#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700237static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700239 int ret = 0;
240 unsigned seq;
241 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -0700242
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700243 do {
244 seq = zone_span_seqbegin(zone);
245 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
246 ret = 1;
247 else if (pfn < zone->zone_start_pfn)
248 ret = 1;
249 } while (zone_span_seqretry(zone, seq));
250
251 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700252}
253
254static int page_is_consistent(struct zone *zone, struct page *page)
255{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700256 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700257 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700259 return 0;
260
261 return 1;
262}
263/*
264 * Temporary debugging check for pages not lying within a given zone.
265 */
266static int bad_range(struct zone *zone, struct page *page)
267{
268 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700270 if (!page_is_consistent(zone, page))
271 return 1;
272
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 return 0;
274}
Nick Piggin13e74442006-01-06 00:10:58 -0800275#else
276static inline int bad_range(struct zone *zone, struct page *page)
277{
278 return 0;
279}
280#endif
281
Nick Piggin224abf92006-01-06 00:11:11 -0800282static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800284 static unsigned long resume;
285 static unsigned long nr_shown;
286 static unsigned long nr_unshown;
287
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200288 /* Don't complain about poisoned pages */
289 if (PageHWPoison(page)) {
Andrea Arcangelief2b4b92011-03-18 00:16:35 +0100290 reset_page_mapcount(page); /* remove PageBuddy */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200291 return;
292 }
293
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800294 /*
295 * Allow a burst of 60 reports, then keep quiet for that minute;
296 * or allow a steady drip of one report per second.
297 */
298 if (nr_shown == 60) {
299 if (time_before(jiffies, resume)) {
300 nr_unshown++;
301 goto out;
302 }
303 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800304 printk(KERN_ALERT
305 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800306 nr_unshown);
307 nr_unshown = 0;
308 }
309 nr_shown = 0;
310 }
311 if (nr_shown++ == 0)
312 resume = jiffies + 60 * HZ;
313
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800314 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800315 current->comm, page_to_pfn(page));
Wu Fengguang718a3822010-03-10 15:20:43 -0800316 dump_page(page);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700317
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800319out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800320 /* Leave bad fields for debug, except PageBuddy could make trouble */
Andrea Arcangelief2b4b92011-03-18 00:16:35 +0100321 reset_page_mapcount(page); /* remove PageBuddy */
Randy Dunlap9f158332005-09-13 01:25:16 -0700322 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323}
324
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325/*
326 * Higher-order pages are called "compound pages". They are structured thusly:
327 *
328 * The first PAGE_SIZE page is called the "head page".
329 *
330 * The remaining PAGE_SIZE pages are called "tail pages".
331 *
332 * All pages have PG_compound set. All pages have their ->private pointing at
333 * the head page (even the head page has this).
334 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800335 * The first tail page's ->lru.next holds the address of the compound page's
336 * put_page() function. Its ->lru.prev holds the order of allocation.
337 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800339
340static void free_compound_page(struct page *page)
341{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700342 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800343}
344
Andi Kleen01ad1c02008-07-23 21:27:46 -0700345void prep_compound_page(struct page *page, unsigned long order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346{
347 int i;
348 int nr_pages = 1 << order;
349
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800350 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700351 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700352 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800353 for (i = 1; i < nr_pages; i++) {
354 struct page *p = page + i;
355
Christoph Lameterd85f3382007-05-06 14:49:39 -0700356 __SetPageTail(p);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700357 p->first_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 }
359}
360
Andrea Arcangeli59ff4212011-01-13 15:46:38 -0800361/* update __split_huge_page_refcount if you change this function */
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800362static int destroy_compound_page(struct page *page, unsigned long order)
Andy Whitcroft18229df2008-11-06 12:53:27 -0800363{
364 int i;
365 int nr_pages = 1 << order;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800366 int bad = 0;
Andy Whitcroft18229df2008-11-06 12:53:27 -0800367
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800368 if (unlikely(compound_order(page) != order) ||
369 unlikely(!PageHead(page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800370 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800371 bad++;
372 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373
Christoph Lameter6d777952007-05-06 14:49:40 -0700374 __ClearPageHead(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800375
Andy Whitcroft18229df2008-11-06 12:53:27 -0800376 for (i = 1; i < nr_pages; i++) {
377 struct page *p = page + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378
Alexey Zaytseve713a212009-01-10 02:47:57 +0300379 if (unlikely(!PageTail(p) || (p->first_page != page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800380 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800381 bad++;
382 }
Christoph Lameterd85f3382007-05-06 14:49:39 -0700383 __ClearPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800385
386 return bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
Nick Piggin17cf4402006-03-22 00:08:41 -0800389static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
390{
391 int i;
392
Andrew Morton6626c5d2006-03-22 00:08:42 -0800393 /*
394 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
395 * and __GFP_HIGHMEM from hard or soft interrupt context.
396 */
Nick Piggin725d7042006-09-25 23:30:55 -0700397 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800398 for (i = 0; i < (1 << order); i++)
399 clear_highpage(page + i);
400}
401
Andrew Morton6aa30012006-04-18 22:20:52 -0700402static inline void set_page_order(struct page *page, int order)
403{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700404 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000405 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406}
407
408static inline void rmv_page_order(struct page *page)
409{
Nick Piggin676165a2006-04-10 11:21:48 +1000410 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700411 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412}
413
414/*
415 * Locate the struct page for both the matching buddy in our
416 * pair (buddy1) and the combined O(n+1) page they form (page).
417 *
418 * 1) Any buddy B1 will have an order O twin B2 which satisfies
419 * the following equation:
420 * B2 = B1 ^ (1 << O)
421 * For example, if the starting buddy (buddy2) is #8 its order
422 * 1 buddy is #10:
423 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
424 *
425 * 2) Any buddy B will have an order O+1 parent P which
426 * satisfies the following equation:
427 * P = B & ~(1 << O)
428 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200429 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431static inline unsigned long
KyongHo Cho43506fa2011-01-13 15:47:24 -0800432__find_buddy_index(unsigned long page_idx, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433{
KyongHo Cho43506fa2011-01-13 15:47:24 -0800434 return page_idx ^ (1 << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435}
436
437/*
438 * This function checks whether a page is free && is the buddy
439 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800440 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000441 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700442 * (c) a page and its buddy have the same order &&
443 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 *
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800445 * For recording whether a page is in the buddy system, we set ->_mapcount -2.
446 * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock.
Nick Piggin676165a2006-04-10 11:21:48 +1000447 *
448 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700450static inline int page_is_buddy(struct page *page, struct page *buddy,
451 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700453 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800454 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800455
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700456 if (page_zone_id(page) != page_zone_id(buddy))
457 return 0;
458
459 if (PageBuddy(buddy) && page_order(buddy) == order) {
Nick Piggina3af9c32009-06-16 15:32:10 -0700460 VM_BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa30012006-04-18 22:20:52 -0700461 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000462 }
Andrew Morton6aa30012006-04-18 22:20:52 -0700463 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464}
465
466/*
467 * Freeing function for a buddy system allocator.
468 *
469 * The concept of a buddy system is to maintain direct-mapped table
470 * (containing bit values) for memory blocks of various "orders".
471 * The bottom level table contains the map for the smallest allocatable
472 * units of memory (here, pages), and each level above it describes
473 * pairs of units from the levels below, hence, "buddies".
474 * At a high level, all that happens here is marking the table entry
475 * at the bottom level available, and propagating the changes upward
476 * as necessary, plus some accounting needed to play nicely with other
477 * parts of the VM system.
478 * At each level, we keep a list of pages, which are heads of continuous
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800479 * free pages of length of (1 << order) and marked with _mapcount -2. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700480 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 * So when we are allocating or freeing one, we can derive the state of the
482 * other. That is, if we allocate a small block, and both were
483 * free, the remainder of the region must be split into blocks.
484 * If a block is freed, and its buddy is also free, then this
485 * triggers coalescing into a block of larger size.
486 *
487 * -- wli
488 */
489
Nick Piggin48db57f2006-01-08 01:00:42 -0800490static inline void __free_one_page(struct page *page,
Mel Gormaned0ae212009-06-16 15:32:07 -0700491 struct zone *zone, unsigned int order,
492 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493{
494 unsigned long page_idx;
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700495 unsigned long combined_idx;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800496 unsigned long uninitialized_var(buddy_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700497 struct page *buddy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498
Nick Piggin224abf92006-01-06 00:11:11 -0800499 if (unlikely(PageCompound(page)))
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800500 if (unlikely(destroy_compound_page(page, order)))
501 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502
Mel Gormaned0ae212009-06-16 15:32:07 -0700503 VM_BUG_ON(migratetype == -1);
504
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
506
Mel Gormanf2260e62009-06-16 15:32:13 -0700507 VM_BUG_ON(page_idx & ((1 << order) - 1));
Nick Piggin725d7042006-09-25 23:30:55 -0700508 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 while (order < MAX_ORDER-1) {
KyongHo Cho43506fa2011-01-13 15:47:24 -0800511 buddy_idx = __find_buddy_index(page_idx, order);
512 buddy = page + (buddy_idx - page_idx);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700513 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700514 break;
Nick Piggin13e74442006-01-06 00:10:58 -0800515
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700516 /* Our buddy is free, merge with it and move up one order. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 list_del(&buddy->lru);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700518 zone->free_area[order].nr_free--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 rmv_page_order(buddy);
KyongHo Cho43506fa2011-01-13 15:47:24 -0800520 combined_idx = buddy_idx & page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 page = page + (combined_idx - page_idx);
522 page_idx = combined_idx;
523 order++;
524 }
525 set_page_order(page, order);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700526
527 /*
528 * If this is not the largest possible page, check if the buddy
529 * of the next-highest order is free. If it is, it's possible
530 * that pages are being freed that will coalesce soon. In case,
531 * that is happening, add the free page to the tail of the list
532 * so it's less likely to be used soon and more likely to be merged
533 * as a higher order page
534 */
Mel Gormanb7f50cf2010-10-26 14:21:11 -0700535 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700536 struct page *higher_page, *higher_buddy;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800537 combined_idx = buddy_idx & page_idx;
538 higher_page = page + (combined_idx - page_idx);
539 buddy_idx = __find_buddy_index(combined_idx, order + 1);
540 higher_buddy = page + (buddy_idx - combined_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700541 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
542 list_add_tail(&page->lru,
543 &zone->free_area[order].free_list[migratetype]);
544 goto out;
545 }
546 }
547
548 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
549out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 zone->free_area[order].nr_free++;
551}
552
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700553/*
554 * free_page_mlock() -- clean up attempts to free and mlocked() page.
555 * Page should not be on lru, so no need to fix that up.
556 * free_pages_check() will verify...
557 */
558static inline void free_page_mlock(struct page *page)
559{
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700560 __dec_zone_page_state(page, NR_MLOCK);
561 __count_vm_event(UNEVICTABLE_MLOCKFREED);
562}
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700563
Nick Piggin224abf92006-01-06 00:11:11 -0800564static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565{
Nick Piggin92be2e32006-01-06 00:10:57 -0800566 if (unlikely(page_mapcount(page) |
567 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700568 (atomic_read(&page->_count) != 0) |
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -0700569 (page->flags & PAGE_FLAGS_CHECK_AT_FREE) |
570 (mem_cgroup_bad_page_check(page)))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800571 bad_page(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800572 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800573 }
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800574 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
575 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
576 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577}
578
579/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700580 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700582 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 *
584 * If the zone was previously in an "all pages pinned" state then look to
585 * see if this freeing clears that state.
586 *
587 * And clear the zone's pages_scanned counter, to hold off the "all pages are
588 * pinned" detection logic.
589 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700590static void free_pcppages_bulk(struct zone *zone, int count,
591 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592{
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700593 int migratetype = 0;
Mel Gormana6f9edd2009-09-21 17:03:20 -0700594 int batch_free = 0;
Mel Gorman72853e22010-09-09 16:38:16 -0700595 int to_free = count;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700596
Nick Pigginc54ad302006-01-06 00:10:56 -0800597 spin_lock(&zone->lock);
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -0800598 zone->all_unreclaimable = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700600
Mel Gorman72853e22010-09-09 16:38:16 -0700601 while (to_free) {
Nick Piggin48db57f2006-01-08 01:00:42 -0800602 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700603 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -0800604
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700605 /*
Mel Gormana6f9edd2009-09-21 17:03:20 -0700606 * Remove pages from lists in a round-robin fashion. A
607 * batch_free count is maintained that is incremented when an
608 * empty list is encountered. This is so more pages are freed
609 * off fuller lists instead of spinning excessively around empty
610 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700611 */
612 do {
Mel Gormana6f9edd2009-09-21 17:03:20 -0700613 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700614 if (++migratetype == MIGRATE_PCPTYPES)
615 migratetype = 0;
616 list = &pcp->lists[migratetype];
617 } while (list_empty(list));
618
Namhyung Kim1d168712011-03-22 16:32:45 -0700619 /* This is the only non-empty list. Free them all. */
620 if (batch_free == MIGRATE_PCPTYPES)
621 batch_free = to_free;
622
Mel Gormana6f9edd2009-09-21 17:03:20 -0700623 do {
624 page = list_entry(list->prev, struct page, lru);
625 /* must delete as __free_one_page list manipulates */
626 list_del(&page->lru);
Hugh Dickinsa7016232010-01-29 17:46:34 +0000627 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
628 __free_one_page(page, zone, 0, page_private(page));
629 trace_mm_page_pcpu_drain(page, 0, page_private(page));
Mel Gorman72853e22010-09-09 16:38:16 -0700630 } while (--to_free && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 }
Mel Gorman72853e22010-09-09 16:38:16 -0700632 __mod_zone_page_state(zone, NR_FREE_PAGES, count);
Nick Pigginc54ad302006-01-06 00:10:56 -0800633 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634}
635
Mel Gormaned0ae212009-06-16 15:32:07 -0700636static void free_one_page(struct zone *zone, struct page *page, int order,
637 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800638{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700639 spin_lock(&zone->lock);
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -0800640 zone->all_unreclaimable = 0;
Christoph Lameter006d22d2006-09-25 23:31:48 -0700641 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700642
Mel Gormaned0ae212009-06-16 15:32:07 -0700643 __free_one_page(page, zone, order, migratetype);
Mel Gorman72853e22010-09-09 16:38:16 -0700644 __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700645 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800646}
647
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700648static bool free_pages_prepare(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 int i;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800651 int bad = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
Li Hongf6503162010-03-05 13:41:52 -0800653 trace_mm_page_free_direct(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100654 kmemcheck_free_shadow(page, order);
655
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -0800656 if (PageAnon(page))
657 page->mapping = NULL;
658 for (i = 0; i < (1 << order); i++)
659 bad += free_pages_check(page + i);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800660 if (bad)
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700661 return false;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800662
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700663 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -0700664 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700665 debug_check_no_obj_freed(page_address(page),
666 PAGE_SIZE << order);
667 }
Nick Piggindafb1362006-10-11 01:21:30 -0700668 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800669 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700670
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700671 return true;
672}
673
674static void __free_pages_ok(struct page *page, unsigned int order)
675{
676 unsigned long flags;
677 int wasMlocked = __TestClearPageMlocked(page);
678
679 if (!free_pages_prepare(page, order))
680 return;
681
Nick Pigginc54ad302006-01-06 00:10:56 -0800682 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +0200683 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -0700684 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700685 __count_vm_events(PGFREE, 1 << order);
Mel Gormaned0ae212009-06-16 15:32:07 -0700686 free_one_page(page_zone(page), page, order,
687 get_pageblock_migratetype(page));
Nick Pigginc54ad302006-01-06 00:10:56 -0800688 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689}
690
David Howellsa226f6c2006-01-06 00:11:08 -0800691/*
692 * permit the bootmem allocator to evade page validation on high-order frees
693 */
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700694void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800695{
696 if (order == 0) {
697 __ClearPageReserved(page);
698 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800699 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800700 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800701 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800702 int loop;
703
Nick Piggin545b1ea2006-03-22 00:08:07 -0800704 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800705 for (loop = 0; loop < BITS_PER_LONG; loop++) {
706 struct page *p = &page[loop];
707
Nick Piggin545b1ea2006-03-22 00:08:07 -0800708 if (loop + 1 < BITS_PER_LONG)
709 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800710 __ClearPageReserved(p);
711 set_page_count(p, 0);
712 }
713
Nick Piggin7835e982006-03-22 00:08:40 -0800714 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800715 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800716 }
717}
718
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719
720/*
721 * The order of subdivision here is critical for the IO subsystem.
722 * Please do not alter this order without good reasons and regression
723 * testing. Specifically, as large blocks of memory are subdivided,
724 * the order in which smaller blocks are delivered depends on the order
725 * they're subdivided in this function. This is the primary factor
726 * influencing the order in which pages are delivered to the IO
727 * subsystem according to empirical testing, and this is also justified
728 * by considering the behavior of a buddy system containing a single
729 * large block of memory acted on by a series of small allocations.
730 * This behavior is a critical factor in sglist merging's success.
731 *
732 * -- wli
733 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800734static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700735 int low, int high, struct free_area *area,
736 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737{
738 unsigned long size = 1 << high;
739
740 while (high > low) {
741 area--;
742 high--;
743 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700744 VM_BUG_ON(bad_range(zone, &page[size]));
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700745 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 area->nr_free++;
747 set_page_order(&page[size], high);
748 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749}
750
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751/*
752 * This page is about to be returned from the page allocator
753 */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200754static inline int check_new_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755{
Nick Piggin92be2e32006-01-06 00:10:57 -0800756 if (unlikely(page_mapcount(page) |
757 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700758 (atomic_read(&page->_count) != 0) |
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -0700759 (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
760 (mem_cgroup_bad_page_check(page)))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800761 bad_page(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800762 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800763 }
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200764 return 0;
765}
766
767static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
768{
769 int i;
770
771 for (i = 0; i < (1 << order); i++) {
772 struct page *p = page + i;
773 if (unlikely(check_new_page(p)))
774 return 1;
775 }
Hugh Dickins689bceb2005-11-21 21:32:20 -0800776
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700777 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800778 set_page_refcounted(page);
Nick Piggincc102502006-12-06 20:32:00 -0800779
780 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800782
783 if (gfp_flags & __GFP_ZERO)
784 prep_zero_page(page, order, gfp_flags);
785
786 if (order && (gfp_flags & __GFP_COMP))
787 prep_compound_page(page, order);
788
Hugh Dickins689bceb2005-11-21 21:32:20 -0800789 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790}
791
Mel Gorman56fd56b2007-10-16 01:25:58 -0700792/*
793 * Go through the free lists for the given migratetype and remove
794 * the smallest available page from the freelists
795 */
Mel Gorman728ec982009-06-16 15:32:04 -0700796static inline
797struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -0700798 int migratetype)
799{
800 unsigned int current_order;
801 struct free_area * area;
802 struct page *page;
803
804 /* Find a page of the appropriate size in the preferred list */
805 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
806 area = &(zone->free_area[current_order]);
807 if (list_empty(&area->free_list[migratetype]))
808 continue;
809
810 page = list_entry(area->free_list[migratetype].next,
811 struct page, lru);
812 list_del(&page->lru);
813 rmv_page_order(page);
814 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -0700815 expand(zone, page, order, current_order, area, migratetype);
816 return page;
817 }
818
819 return NULL;
820}
821
822
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700823/*
824 * This array describes the order lists are fallen back to when
825 * the free lists for the desirable migrate type are depleted
826 */
827static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
Mel Gorman64c5e132007-10-16 01:25:59 -0700828 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
829 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
830 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
831 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700832};
833
Mel Gormanc361be52007-10-16 01:25:51 -0700834/*
835 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -0700836 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -0700837 * boundary. If alignment is required, use move_freepages_block()
838 */
Adrian Bunkb69a7282008-07-23 21:28:12 -0700839static int move_freepages(struct zone *zone,
840 struct page *start_page, struct page *end_page,
841 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700842{
843 struct page *page;
844 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -0700845 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -0700846
847#ifndef CONFIG_HOLES_IN_ZONE
848 /*
849 * page_zone is not safe to call in this context when
850 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
851 * anyway as we check zone boundaries in move_freepages_block().
852 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -0700853 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -0700854 */
855 BUG_ON(page_zone(start_page) != page_zone(end_page));
856#endif
857
858 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -0700859 /* Make sure we are not inadvertently changing nodes */
860 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
861
Mel Gormanc361be52007-10-16 01:25:51 -0700862 if (!pfn_valid_within(page_to_pfn(page))) {
863 page++;
864 continue;
865 }
866
867 if (!PageBuddy(page)) {
868 page++;
869 continue;
870 }
871
872 order = page_order(page);
Kirill A. Shutemov84be48d2011-03-22 16:33:41 -0700873 list_move(&page->lru,
874 &zone->free_area[order].free_list[migratetype]);
Mel Gormanc361be52007-10-16 01:25:51 -0700875 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -0700876 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -0700877 }
878
Mel Gormand1003132007-10-16 01:26:00 -0700879 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -0700880}
881
Adrian Bunkb69a7282008-07-23 21:28:12 -0700882static int move_freepages_block(struct zone *zone, struct page *page,
883 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700884{
885 unsigned long start_pfn, end_pfn;
886 struct page *start_page, *end_page;
887
888 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -0700889 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -0700890 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -0700891 end_page = start_page + pageblock_nr_pages - 1;
892 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -0700893
894 /* Do not cross zone boundaries */
895 if (start_pfn < zone->zone_start_pfn)
896 start_page = page;
897 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
898 return 0;
899
900 return move_freepages(zone, start_page, end_page, migratetype);
901}
902
Mel Gorman2f66a682009-09-21 17:02:31 -0700903static void change_pageblock_range(struct page *pageblock_page,
904 int start_order, int migratetype)
905{
906 int nr_pageblocks = 1 << (start_order - pageblock_order);
907
908 while (nr_pageblocks--) {
909 set_pageblock_migratetype(pageblock_page, migratetype);
910 pageblock_page += pageblock_nr_pages;
911 }
912}
913
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700914/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -0700915static inline struct page *
916__rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700917{
918 struct free_area * area;
919 int current_order;
920 struct page *page;
921 int migratetype, i;
922
923 /* Find the largest possible block of pages in the other list */
924 for (current_order = MAX_ORDER-1; current_order >= order;
925 --current_order) {
926 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
927 migratetype = fallbacks[start_migratetype][i];
928
Mel Gorman56fd56b2007-10-16 01:25:58 -0700929 /* MIGRATE_RESERVE handled later if necessary */
930 if (migratetype == MIGRATE_RESERVE)
931 continue;
Mel Gormane0104872007-10-16 01:25:53 -0700932
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700933 area = &(zone->free_area[current_order]);
934 if (list_empty(&area->free_list[migratetype]))
935 continue;
936
937 page = list_entry(area->free_list[migratetype].next,
938 struct page, lru);
939 area->nr_free--;
940
941 /*
Mel Gormanc361be52007-10-16 01:25:51 -0700942 * If breaking a large block of pages, move all free
Mel Gorman46dafbc2007-10-16 01:25:55 -0700943 * pages to the preferred allocation list. If falling
944 * back for a reclaimable kernel allocation, be more
945 * agressive about taking ownership of free pages
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700946 */
Mel Gormand9c23402007-10-16 01:26:01 -0700947 if (unlikely(current_order >= (pageblock_order >> 1)) ||
Mel Gormandd5d2412009-09-05 11:17:11 -0700948 start_migratetype == MIGRATE_RECLAIMABLE ||
949 page_group_by_mobility_disabled) {
Mel Gorman46dafbc2007-10-16 01:25:55 -0700950 unsigned long pages;
951 pages = move_freepages_block(zone, page,
952 start_migratetype);
953
954 /* Claim the whole block if over half of it is free */
Mel Gormandd5d2412009-09-05 11:17:11 -0700955 if (pages >= (1 << (pageblock_order-1)) ||
956 page_group_by_mobility_disabled)
Mel Gorman46dafbc2007-10-16 01:25:55 -0700957 set_pageblock_migratetype(page,
958 start_migratetype);
959
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700960 migratetype = start_migratetype;
Mel Gormanc361be52007-10-16 01:25:51 -0700961 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700962
963 /* Remove the page from the freelists */
964 list_del(&page->lru);
965 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700966
Mel Gorman2f66a682009-09-21 17:02:31 -0700967 /* Take ownership for orders >= pageblock_order */
968 if (current_order >= pageblock_order)
969 change_pageblock_range(page, current_order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700970 start_migratetype);
971
972 expand(zone, page, order, current_order, area, migratetype);
Mel Gormane0fff1b2009-09-21 17:02:42 -0700973
974 trace_mm_page_alloc_extfrag(page, order, current_order,
975 start_migratetype, migratetype);
976
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700977 return page;
978 }
979 }
980
Mel Gorman728ec982009-06-16 15:32:04 -0700981 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700982}
983
Mel Gorman56fd56b2007-10-16 01:25:58 -0700984/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 * Do the hard work of removing an element from the buddy allocator.
986 * Call me with the zone->lock already held.
987 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700988static struct page *__rmqueue(struct zone *zone, unsigned int order,
989 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 struct page *page;
992
Mel Gorman728ec982009-06-16 15:32:04 -0700993retry_reserve:
Mel Gorman56fd56b2007-10-16 01:25:58 -0700994 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995
Mel Gorman728ec982009-06-16 15:32:04 -0700996 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Mel Gorman56fd56b2007-10-16 01:25:58 -0700997 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700998
Mel Gorman728ec982009-06-16 15:32:04 -0700999 /*
1000 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1001 * is used because __rmqueue_smallest is an inline function
1002 * and we want just one call site
1003 */
1004 if (!page) {
1005 migratetype = MIGRATE_RESERVE;
1006 goto retry_reserve;
1007 }
1008 }
1009
Mel Gorman0d3d0622009-09-21 17:02:44 -07001010 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001011 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012}
1013
1014/*
1015 * Obtain a specified number of elements from the buddy allocator, all under
1016 * a single hold of the lock, for efficiency. Add them to the supplied list.
1017 * Returns the number of new pages which were placed at *list.
1018 */
1019static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001020 unsigned long count, struct list_head *list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001021 int migratetype, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024
Nick Pigginc54ad302006-01-06 00:10:56 -08001025 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001027 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001028 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08001030
1031 /*
1032 * Split buddy pages returned by expand() are received here
1033 * in physical page order. The page is added to the callers and
1034 * list and the list head then moves forward. From the callers
1035 * perspective, the linked list is ordered by page number in
1036 * some conditions. This is useful for IO devices that can
1037 * merge IO requests if the physical pages are ordered
1038 * properly.
1039 */
Mel Gormane084b2d2009-07-29 15:02:04 -07001040 if (likely(cold == 0))
1041 list_add(&page->lru, list);
1042 else
1043 list_add_tail(&page->lru, list);
Mel Gorman535131e2007-10-16 01:25:49 -07001044 set_page_private(page, migratetype);
Mel Gorman81eabcb2007-12-17 16:20:05 -08001045 list = &page->lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 }
Mel Gormanf2260e62009-06-16 15:32:13 -07001047 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08001048 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001049 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050}
1051
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001052#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001053/*
Christoph Lameter4037d452007-05-09 02:35:14 -07001054 * Called from the vmstat counter updater to drain pagesets of this
1055 * currently executing processor on remote nodes after they have
1056 * expired.
1057 *
Christoph Lameter879336c2006-03-22 00:09:08 -08001058 * Note that this function must be called with the thread pinned to
1059 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001060 */
Christoph Lameter4037d452007-05-09 02:35:14 -07001061void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001062{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001063 unsigned long flags;
Christoph Lameter4037d452007-05-09 02:35:14 -07001064 int to_drain;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001065
Christoph Lameter4037d452007-05-09 02:35:14 -07001066 local_irq_save(flags);
1067 if (pcp->count >= pcp->batch)
1068 to_drain = pcp->batch;
1069 else
1070 to_drain = pcp->count;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001071 free_pcppages_bulk(zone, to_drain, pcp);
Christoph Lameter4037d452007-05-09 02:35:14 -07001072 pcp->count -= to_drain;
1073 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001074}
1075#endif
1076
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001077/*
1078 * Drain pages of the indicated processor.
1079 *
1080 * The processor must either be the current processor and the
1081 * thread pinned to the current processor or a processor that
1082 * is not online.
1083 */
1084static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085{
Nick Pigginc54ad302006-01-06 00:10:56 -08001086 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07001089 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 struct per_cpu_pageset *pset;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001091 struct per_cpu_pages *pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001093 local_irq_save(flags);
1094 pset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001096 pcp = &pset->pcp;
David Rientjes2ff754f2011-01-25 15:07:23 -08001097 if (pcp->count) {
1098 free_pcppages_bulk(zone, pcp->count, pcp);
1099 pcp->count = 0;
1100 }
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001101 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 }
1103}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001105/*
1106 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1107 */
1108void drain_local_pages(void *arg)
1109{
1110 drain_pages(smp_processor_id());
1111}
1112
1113/*
1114 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
1115 */
1116void drain_all_pages(void)
1117{
Jens Axboe15c8b6c2008-05-09 09:39:44 +02001118 on_each_cpu(drain_local_pages, NULL, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001119}
1120
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02001121#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122
1123void mark_free_pages(struct zone *zone)
1124{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001125 unsigned long pfn, max_zone_pfn;
1126 unsigned long flags;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001127 int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 struct list_head *curr;
1129
1130 if (!zone->spanned_pages)
1131 return;
1132
1133 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001134
1135 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1136 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1137 if (pfn_valid(pfn)) {
1138 struct page *page = pfn_to_page(pfn);
1139
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001140 if (!swsusp_page_is_forbidden(page))
1141 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001142 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001144 for_each_migratetype_order(order, t) {
1145 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001146 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001148 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1149 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001150 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001151 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001152 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 spin_unlock_irqrestore(&zone->lock, flags);
1154}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001155#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156
1157/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 * Free a 0-order page
Li Hongfc916682010-03-05 13:41:54 -08001159 * cold == 1 ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 */
Li Hongfc916682010-03-05 13:41:54 -08001161void free_hot_cold_page(struct page *page, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162{
1163 struct zone *zone = page_zone(page);
1164 struct per_cpu_pages *pcp;
1165 unsigned long flags;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001166 int migratetype;
Johannes Weiner451ea252009-09-21 17:01:48 -07001167 int wasMlocked = __TestClearPageMlocked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001169 if (!free_pages_prepare(page, 0))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001170 return;
1171
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001172 migratetype = get_pageblock_migratetype(page);
1173 set_page_private(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +02001175 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -07001176 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001177 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001178
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001179 /*
1180 * We only track unmovable, reclaimable and movable on pcp lists.
1181 * Free ISOLATE pages back to the allocator because they are being
1182 * offlined but treat RESERVE as movable pages so we can get those
1183 * areas back if necessary. Otherwise, we may have to free
1184 * excessively into the page allocator
1185 */
1186 if (migratetype >= MIGRATE_PCPTYPES) {
1187 if (unlikely(migratetype == MIGRATE_ISOLATE)) {
1188 free_one_page(zone, page, 0, migratetype);
1189 goto out;
1190 }
1191 migratetype = MIGRATE_MOVABLE;
1192 }
1193
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001194 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001195 if (cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001196 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001197 else
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001198 list_add(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001200 if (pcp->count >= pcp->high) {
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001201 free_pcppages_bulk(zone, pcp->batch, pcp);
Nick Piggin48db57f2006-01-08 01:00:42 -08001202 pcp->count -= pcp->batch;
1203 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001204
1205out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207}
1208
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001209/*
1210 * split_page takes a non-compound higher-order page, and splits it into
1211 * n (1<<order) sub-pages: page[0..n]
1212 * Each sub-page must be freed individually.
1213 *
1214 * Note: this is probably too low level an operation for use in drivers.
1215 * Please consult with lkml before using this in your driver.
1216 */
1217void split_page(struct page *page, unsigned int order)
1218{
1219 int i;
1220
Nick Piggin725d7042006-09-25 23:30:55 -07001221 VM_BUG_ON(PageCompound(page));
1222 VM_BUG_ON(!page_count(page));
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001223
1224#ifdef CONFIG_KMEMCHECK
1225 /*
1226 * Split shadow pages too, because free(page[0]) would
1227 * otherwise free the whole shadow.
1228 */
1229 if (kmemcheck_page_is_tracked(page))
1230 split_page(virt_to_page(page[0].shadow), order);
1231#endif
1232
Nick Piggin7835e982006-03-22 00:08:40 -08001233 for (i = 1; i < (1 << order); i++)
1234 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001235}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001236
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237/*
Mel Gorman748446b2010-05-24 14:32:27 -07001238 * Similar to split_page except the page is already free. As this is only
1239 * being used for migration, the migratetype of the block also changes.
1240 * As this is called with interrupts disabled, the caller is responsible
1241 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1242 * are enabled.
1243 *
1244 * Note: this is probably too low level an operation for use in drivers.
1245 * Please consult with lkml before using this in your driver.
1246 */
1247int split_free_page(struct page *page)
1248{
1249 unsigned int order;
1250 unsigned long watermark;
1251 struct zone *zone;
1252
1253 BUG_ON(!PageBuddy(page));
1254
1255 zone = page_zone(page);
1256 order = page_order(page);
1257
1258 /* Obey watermarks as if the page was being allocated */
1259 watermark = low_wmark_pages(zone) + (1 << order);
1260 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1261 return 0;
1262
1263 /* Remove page from free list */
1264 list_del(&page->lru);
1265 zone->free_area[order].nr_free--;
1266 rmv_page_order(page);
1267 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1UL << order));
1268
1269 /* Split into individual pages */
1270 set_page_refcounted(page);
1271 split_page(page, order);
1272
1273 if (order >= pageblock_order - 1) {
1274 struct page *endpage = page + (1 << order) - 1;
1275 for (; page < endpage; page += pageblock_nr_pages)
1276 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1277 }
1278
1279 return 1 << order;
1280}
1281
1282/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1284 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1285 * or two.
1286 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07001287static inline
1288struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001289 struct zone *zone, int order, gfp_t gfp_flags,
1290 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291{
1292 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001293 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 int cold = !!(gfp_flags & __GFP_COLD);
1295
Hugh Dickins689bceb2005-11-21 21:32:20 -08001296again:
Nick Piggin48db57f2006-01-08 01:00:42 -08001297 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001299 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 local_irq_save(flags);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001302 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1303 list = &pcp->lists[migratetype];
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001304 if (list_empty(list)) {
Mel Gorman535131e2007-10-16 01:25:49 -07001305 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001306 pcp->batch, list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001307 migratetype, cold);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001308 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07001309 goto failed;
Mel Gorman535131e2007-10-16 01:25:49 -07001310 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001311
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001312 if (cold)
1313 page = list_entry(list->prev, struct page, lru);
1314 else
1315 page = list_entry(list->next, struct page, lru);
1316
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001317 list_del(&page->lru);
1318 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001319 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07001320 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1321 /*
1322 * __GFP_NOFAIL is not to be used in new code.
1323 *
1324 * All __GFP_NOFAIL callers should be fixed so that they
1325 * properly detect and handle allocation failures.
1326 *
1327 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07001328 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07001329 * __GFP_NOFAIL.
1330 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07001331 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07001332 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001334 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001335 spin_unlock(&zone->lock);
1336 if (!page)
1337 goto failed;
KOSAKI Motohiro6ccf80e2010-01-15 17:01:18 -08001338 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 }
1340
Christoph Lameterf8891e52006-06-30 01:55:45 -07001341 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Andi Kleen78afd562011-03-22 16:33:12 -07001342 zone_statistics(preferred_zone, zone, gfp_flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001343 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344
Nick Piggin725d7042006-09-25 23:30:55 -07001345 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -08001346 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001347 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001349
1350failed:
1351 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001352 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353}
1354
Mel Gorman41858962009-06-16 15:32:12 -07001355/* The ALLOC_WMARK bits are used as an index to zone->watermark */
1356#define ALLOC_WMARK_MIN WMARK_MIN
1357#define ALLOC_WMARK_LOW WMARK_LOW
1358#define ALLOC_WMARK_HIGH WMARK_HIGH
1359#define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1360
1361/* Mask to get the watermark bits */
1362#define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1363
Nick Piggin31488902005-11-28 13:44:03 -08001364#define ALLOC_HARDER 0x10 /* try to alloc harder */
1365#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1366#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001367
Akinobu Mita933e3122006-12-08 02:39:45 -08001368#ifdef CONFIG_FAIL_PAGE_ALLOC
1369
1370static struct fail_page_alloc_attr {
1371 struct fault_attr attr;
1372
1373 u32 ignore_gfp_highmem;
1374 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001375 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001376
1377#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1378
1379 struct dentry *ignore_gfp_highmem_file;
1380 struct dentry *ignore_gfp_wait_file;
Akinobu Mita54114992007-07-15 23:40:23 -07001381 struct dentry *min_order_file;
Akinobu Mita933e3122006-12-08 02:39:45 -08001382
1383#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1384
1385} fail_page_alloc = {
1386 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001387 .ignore_gfp_wait = 1,
1388 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001389 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001390};
1391
1392static int __init setup_fail_page_alloc(char *str)
1393{
1394 return setup_fault_attr(&fail_page_alloc.attr, str);
1395}
1396__setup("fail_page_alloc=", setup_fail_page_alloc);
1397
1398static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1399{
Akinobu Mita54114992007-07-15 23:40:23 -07001400 if (order < fail_page_alloc.min_order)
1401 return 0;
Akinobu Mita933e3122006-12-08 02:39:45 -08001402 if (gfp_mask & __GFP_NOFAIL)
1403 return 0;
1404 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1405 return 0;
1406 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1407 return 0;
1408
1409 return should_fail(&fail_page_alloc.attr, 1 << order);
1410}
1411
1412#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1413
1414static int __init fail_page_alloc_debugfs(void)
1415{
1416 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1417 struct dentry *dir;
1418 int err;
1419
1420 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1421 "fail_page_alloc");
1422 if (err)
1423 return err;
1424 dir = fail_page_alloc.attr.dentries.dir;
1425
1426 fail_page_alloc.ignore_gfp_wait_file =
1427 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1428 &fail_page_alloc.ignore_gfp_wait);
1429
1430 fail_page_alloc.ignore_gfp_highmem_file =
1431 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1432 &fail_page_alloc.ignore_gfp_highmem);
Akinobu Mita54114992007-07-15 23:40:23 -07001433 fail_page_alloc.min_order_file =
1434 debugfs_create_u32("min-order", mode, dir,
1435 &fail_page_alloc.min_order);
Akinobu Mita933e3122006-12-08 02:39:45 -08001436
1437 if (!fail_page_alloc.ignore_gfp_wait_file ||
Akinobu Mita54114992007-07-15 23:40:23 -07001438 !fail_page_alloc.ignore_gfp_highmem_file ||
1439 !fail_page_alloc.min_order_file) {
Akinobu Mita933e3122006-12-08 02:39:45 -08001440 err = -ENOMEM;
1441 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1442 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
Akinobu Mita54114992007-07-15 23:40:23 -07001443 debugfs_remove(fail_page_alloc.min_order_file);
Akinobu Mita933e3122006-12-08 02:39:45 -08001444 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1445 }
1446
1447 return err;
1448}
1449
1450late_initcall(fail_page_alloc_debugfs);
1451
1452#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1453
1454#else /* CONFIG_FAIL_PAGE_ALLOC */
1455
1456static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1457{
1458 return 0;
1459}
1460
1461#endif /* CONFIG_FAIL_PAGE_ALLOC */
1462
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463/*
Mel Gorman88f5acf2011-01-13 15:45:41 -08001464 * Return true if free pages are above 'mark'. This takes into account the order
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 * of the allocation.
1466 */
Mel Gorman88f5acf2011-01-13 15:45:41 -08001467static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1468 int classzone_idx, int alloc_flags, long free_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469{
1470 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001471 long min = mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 int o;
1473
Mel Gorman88f5acf2011-01-13 15:45:41 -08001474 free_pages -= (1 << order) + 1;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001475 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001477 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 min -= min / 4;
1479
1480 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
Mel Gorman88f5acf2011-01-13 15:45:41 -08001481 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 for (o = 0; o < order; o++) {
1483 /* At the next order, this order's pages become unavailable */
1484 free_pages -= z->free_area[o].nr_free << o;
1485
1486 /* Require fewer higher order pages to be free */
1487 min >>= 1;
1488
1489 if (free_pages <= min)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001490 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 }
Mel Gorman88f5acf2011-01-13 15:45:41 -08001492 return true;
1493}
1494
1495bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1496 int classzone_idx, int alloc_flags)
1497{
1498 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1499 zone_page_state(z, NR_FREE_PAGES));
1500}
1501
1502bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
1503 int classzone_idx, int alloc_flags)
1504{
1505 long free_pages = zone_page_state(z, NR_FREE_PAGES);
1506
1507 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1508 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1509
1510 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1511 free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512}
1513
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001514#ifdef CONFIG_NUMA
1515/*
1516 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1517 * skip over zones that are not allowed by the cpuset, or that have
1518 * been recently (in last second) found to be nearly full. See further
1519 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001520 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001521 *
1522 * If the zonelist cache is present in the passed in zonelist, then
1523 * returns a pointer to the allowed node mask (either the current
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001524 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001525 *
1526 * If the zonelist cache is not available for this zonelist, does
1527 * nothing and returns NULL.
1528 *
1529 * If the fullzones BITMAP in the zonelist cache is stale (more than
1530 * a second since last zap'd) then we zap it out (clear its bits.)
1531 *
1532 * We hold off even calling zlc_setup, until after we've checked the
1533 * first zone in the zonelist, on the theory that most allocations will
1534 * be satisfied from that first zone, so best to examine that zone as
1535 * quickly as we can.
1536 */
1537static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1538{
1539 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1540 nodemask_t *allowednodes; /* zonelist_cache approximation */
1541
1542 zlc = zonelist->zlcache_ptr;
1543 if (!zlc)
1544 return NULL;
1545
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001546 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001547 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1548 zlc->last_full_zap = jiffies;
1549 }
1550
1551 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1552 &cpuset_current_mems_allowed :
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001553 &node_states[N_HIGH_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001554 return allowednodes;
1555}
1556
1557/*
1558 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1559 * if it is worth looking at further for free memory:
1560 * 1) Check that the zone isn't thought to be full (doesn't have its
1561 * bit set in the zonelist_cache fullzones BITMAP).
1562 * 2) Check that the zones node (obtained from the zonelist_cache
1563 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1564 * Return true (non-zero) if zone is worth looking at further, or
1565 * else return false (zero) if it is not.
1566 *
1567 * This check -ignores- the distinction between various watermarks,
1568 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1569 * found to be full for any variation of these watermarks, it will
1570 * be considered full for up to one second by all requests, unless
1571 * we are so low on memory on all allowed nodes that we are forced
1572 * into the second scan of the zonelist.
1573 *
1574 * In the second scan we ignore this zonelist cache and exactly
1575 * apply the watermarks to all zones, even it is slower to do so.
1576 * We are low on memory in the second scan, and should leave no stone
1577 * unturned looking for a free page.
1578 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001579static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001580 nodemask_t *allowednodes)
1581{
1582 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1583 int i; /* index of *z in zonelist zones */
1584 int n; /* node that zone *z is on */
1585
1586 zlc = zonelist->zlcache_ptr;
1587 if (!zlc)
1588 return 1;
1589
Mel Gormandd1a2392008-04-28 02:12:17 -07001590 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001591 n = zlc->z_to_n[i];
1592
1593 /* This zone is worth trying if it is allowed but not full */
1594 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1595}
1596
1597/*
1598 * Given 'z' scanning a zonelist, set the corresponding bit in
1599 * zlc->fullzones, so that subsequent attempts to allocate a page
1600 * from that zone don't waste time re-examining it.
1601 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001602static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001603{
1604 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1605 int i; /* index of *z in zonelist zones */
1606
1607 zlc = zonelist->zlcache_ptr;
1608 if (!zlc)
1609 return;
1610
Mel Gormandd1a2392008-04-28 02:12:17 -07001611 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001612
1613 set_bit(i, zlc->fullzones);
1614}
1615
1616#else /* CONFIG_NUMA */
1617
1618static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1619{
1620 return NULL;
1621}
1622
Mel Gormandd1a2392008-04-28 02:12:17 -07001623static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001624 nodemask_t *allowednodes)
1625{
1626 return 1;
1627}
1628
Mel Gormandd1a2392008-04-28 02:12:17 -07001629static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001630{
1631}
1632#endif /* CONFIG_NUMA */
1633
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001634/*
Paul Jackson0798e512006-12-06 20:31:38 -08001635 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001636 * a page.
1637 */
1638static struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001639get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
Mel Gorman5117f452009-06-16 15:31:59 -07001640 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
Mel Gorman3dd28262009-06-16 15:32:00 -07001641 struct zone *preferred_zone, int migratetype)
Martin Hicks753ee722005-06-21 17:14:41 -07001642{
Mel Gormandd1a2392008-04-28 02:12:17 -07001643 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001644 struct page *page = NULL;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001645 int classzone_idx;
Mel Gorman5117f452009-06-16 15:31:59 -07001646 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001647 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1648 int zlc_active = 0; /* set if using zonelist_cache */
1649 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gorman54a6eb52008-04-28 02:12:16 -07001650
Mel Gorman19770b32008-04-28 02:12:18 -07001651 classzone_idx = zone_idx(preferred_zone);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001652zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001653 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001654 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001655 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1656 */
Mel Gorman19770b32008-04-28 02:12:18 -07001657 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1658 high_zoneidx, nodemask) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001659 if (NUMA_BUILD && zlc_active &&
1660 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1661 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001662 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001663 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001664 goto try_next_zone;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001665
Mel Gorman41858962009-06-16 15:32:12 -07001666 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001667 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001668 unsigned long mark;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001669 int ret;
1670
Mel Gorman41858962009-06-16 15:32:12 -07001671 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
Mel Gormanfa5e0842009-06-16 15:33:22 -07001672 if (zone_watermark_ok(zone, order, mark,
1673 classzone_idx, alloc_flags))
1674 goto try_this_zone;
1675
1676 if (zone_reclaim_mode == 0)
1677 goto this_zone_full;
1678
1679 ret = zone_reclaim(zone, gfp_mask, order);
1680 switch (ret) {
1681 case ZONE_RECLAIM_NOSCAN:
1682 /* did not scan */
1683 goto try_next_zone;
1684 case ZONE_RECLAIM_FULL:
1685 /* scanned but unreclaimable */
1686 goto this_zone_full;
1687 default:
1688 /* did we reclaim enough */
1689 if (!zone_watermark_ok(zone, order, mark,
1690 classzone_idx, alloc_flags))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001691 goto this_zone_full;
Paul Jackson0798e512006-12-06 20:31:38 -08001692 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001693 }
1694
Mel Gormanfa5e0842009-06-16 15:33:22 -07001695try_this_zone:
Mel Gorman3dd28262009-06-16 15:32:00 -07001696 page = buffered_rmqueue(preferred_zone, zone, order,
1697 gfp_mask, migratetype);
Paul Jackson0798e512006-12-06 20:31:38 -08001698 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001699 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001700this_zone_full:
1701 if (NUMA_BUILD)
1702 zlc_mark_zone_full(zonelist, z);
1703try_next_zone:
Christoph Lameter62bc62a2009-06-16 15:32:15 -07001704 if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) {
Mel Gormand395b732009-06-16 15:32:09 -07001705 /*
1706 * we do zlc_setup after the first zone is tried but only
1707 * if there are multiple nodes make it worthwhile
1708 */
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001709 allowednodes = zlc_setup(zonelist, alloc_flags);
1710 zlc_active = 1;
1711 did_zlc_setup = 1;
1712 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07001713 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001714
1715 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1716 /* Disable zlc cache for second zonelist scan */
1717 zlc_active = 0;
1718 goto zonelist_scan;
1719 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001720 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07001721}
1722
David Rientjes29423e72011-03-22 16:30:47 -07001723/*
1724 * Large machines with many possible nodes should not always dump per-node
1725 * meminfo in irq context.
1726 */
1727static inline bool should_suppress_show_mem(void)
1728{
1729 bool ret = false;
1730
1731#if NODES_SHIFT > 8
1732 ret = in_interrupt();
1733#endif
1734 return ret;
1735}
1736
Mel Gorman11e33f62009-06-16 15:31:57 -07001737static inline int
1738should_alloc_retry(gfp_t gfp_mask, unsigned int order,
1739 unsigned long pages_reclaimed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740{
Mel Gorman11e33f62009-06-16 15:31:57 -07001741 /* Do not loop if specifically requested */
1742 if (gfp_mask & __GFP_NORETRY)
1743 return 0;
1744
1745 /*
1746 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1747 * means __GFP_NOFAIL, but that may not be true in other
1748 * implementations.
1749 */
1750 if (order <= PAGE_ALLOC_COSTLY_ORDER)
1751 return 1;
1752
1753 /*
1754 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1755 * specified, then we retry until we no longer reclaim any pages
1756 * (above), or we've reclaimed an order of pages at least as
1757 * large as the allocation's order. In both cases, if the
1758 * allocation still fails, we stop retrying.
1759 */
1760 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
1761 return 1;
1762
1763 /*
1764 * Don't let big-order allocations loop unless the caller
1765 * explicitly requests that.
1766 */
1767 if (gfp_mask & __GFP_NOFAIL)
1768 return 1;
1769
1770 return 0;
1771}
1772
1773static inline struct page *
1774__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1775 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001776 nodemask_t *nodemask, struct zone *preferred_zone,
1777 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001778{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780
Mel Gorman11e33f62009-06-16 15:31:57 -07001781 /* Acquire the OOM killer lock for the zones in zonelist */
Minchan Kimff321fe2010-08-09 17:18:57 -07001782 if (!try_set_zonelist_oom(zonelist, gfp_mask)) {
Mel Gorman11e33f62009-06-16 15:31:57 -07001783 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 return NULL;
1785 }
Jens Axboe6b1de912005-11-17 21:35:02 +01001786
Mel Gorman11e33f62009-06-16 15:31:57 -07001787 /*
1788 * Go through the zonelist yet one more time, keep very high watermark
1789 * here, this is only to catch a parallel oom killing, we must fail if
1790 * we're still under heavy pressure.
1791 */
1792 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1793 order, zonelist, high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001794 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07001795 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001796 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07001797 goto out;
1798
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001799 if (!(gfp_mask & __GFP_NOFAIL)) {
1800 /* The OOM killer will not help higher order allocs */
1801 if (order > PAGE_ALLOC_COSTLY_ORDER)
1802 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07001803 /* The OOM killer does not needlessly kill tasks for lowmem */
1804 if (high_zoneidx < ZONE_NORMAL)
1805 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001806 /*
1807 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
1808 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
1809 * The caller should handle page allocation failure by itself if
1810 * it specifies __GFP_THISNODE.
1811 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
1812 */
1813 if (gfp_mask & __GFP_THISNODE)
1814 goto out;
1815 }
Mel Gorman11e33f62009-06-16 15:31:57 -07001816 /* Exhausted what can be done so it's blamo time */
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001817 out_of_memory(zonelist, gfp_mask, order, nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07001818
1819out:
1820 clear_zonelist_oom(zonelist, gfp_mask);
1821 return page;
1822}
1823
Mel Gorman56de7262010-05-24 14:32:30 -07001824#ifdef CONFIG_COMPACTION
1825/* Try memory compaction for high-order allocations before reclaim */
1826static struct page *
1827__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
1828 struct zonelist *zonelist, enum zone_type high_zoneidx,
1829 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman77f1fe62011-01-13 15:45:57 -08001830 int migratetype, unsigned long *did_some_progress,
1831 bool sync_migration)
Mel Gorman56de7262010-05-24 14:32:30 -07001832{
1833 struct page *page;
1834
Mel Gorman4f92e252010-05-24 14:32:32 -07001835 if (!order || compaction_deferred(preferred_zone))
Mel Gorman56de7262010-05-24 14:32:30 -07001836 return NULL;
1837
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001838 current->flags |= PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07001839 *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
Mel Gorman77f1fe62011-01-13 15:45:57 -08001840 nodemask, sync_migration);
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001841 current->flags &= ~PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07001842 if (*did_some_progress != COMPACT_SKIPPED) {
1843
1844 /* Page migration frees to the PCP lists but we want merging */
1845 drain_pages(get_cpu());
1846 put_cpu();
1847
1848 page = get_page_from_freelist(gfp_mask, nodemask,
1849 order, zonelist, high_zoneidx,
1850 alloc_flags, preferred_zone,
1851 migratetype);
1852 if (page) {
Mel Gorman4f92e252010-05-24 14:32:32 -07001853 preferred_zone->compact_considered = 0;
1854 preferred_zone->compact_defer_shift = 0;
Mel Gorman56de7262010-05-24 14:32:30 -07001855 count_vm_event(COMPACTSUCCESS);
1856 return page;
1857 }
1858
1859 /*
1860 * It's bad if compaction run occurs and fails.
1861 * The most likely reason is that pages exist,
1862 * but not enough to satisfy watermarks.
1863 */
1864 count_vm_event(COMPACTFAIL);
Mel Gorman4f92e252010-05-24 14:32:32 -07001865 defer_compaction(preferred_zone);
Mel Gorman56de7262010-05-24 14:32:30 -07001866
1867 cond_resched();
1868 }
1869
1870 return NULL;
1871}
1872#else
1873static inline struct page *
1874__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
1875 struct zonelist *zonelist, enum zone_type high_zoneidx,
1876 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman77f1fe62011-01-13 15:45:57 -08001877 int migratetype, unsigned long *did_some_progress,
1878 bool sync_migration)
Mel Gorman56de7262010-05-24 14:32:30 -07001879{
1880 return NULL;
1881}
1882#endif /* CONFIG_COMPACTION */
1883
Mel Gorman11e33f62009-06-16 15:31:57 -07001884/* The really slow allocator path where we enter direct reclaim */
1885static inline struct page *
1886__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
1887 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001888 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001889 int migratetype, unsigned long *did_some_progress)
Mel Gorman11e33f62009-06-16 15:31:57 -07001890{
1891 struct page *page = NULL;
1892 struct reclaim_state reclaim_state;
Mel Gorman9ee493c2010-09-09 16:38:18 -07001893 bool drained = false;
Mel Gorman11e33f62009-06-16 15:31:57 -07001894
1895 cond_resched();
1896
1897 /* We now go into synchronous reclaim */
1898 cpuset_memory_pressure_bump();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001899 current->flags |= PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07001900 lockdep_set_current_reclaim_state(gfp_mask);
1901 reclaim_state.reclaimed_slab = 0;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001902 current->reclaim_state = &reclaim_state;
Mel Gorman11e33f62009-06-16 15:31:57 -07001903
1904 *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
1905
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001906 current->reclaim_state = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07001907 lockdep_clear_current_reclaim_state();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001908 current->flags &= ~PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07001909
1910 cond_resched();
1911
Mel Gorman9ee493c2010-09-09 16:38:18 -07001912 if (unlikely(!(*did_some_progress)))
1913 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07001914
Mel Gorman9ee493c2010-09-09 16:38:18 -07001915retry:
1916 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001917 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001918 alloc_flags, preferred_zone,
1919 migratetype);
Mel Gorman9ee493c2010-09-09 16:38:18 -07001920
1921 /*
1922 * If an allocation failed after direct reclaim, it could be because
1923 * pages are pinned on the per-cpu lists. Drain them and try again
1924 */
1925 if (!page && !drained) {
1926 drain_all_pages();
1927 drained = true;
1928 goto retry;
1929 }
1930
Mel Gorman11e33f62009-06-16 15:31:57 -07001931 return page;
1932}
1933
Mel Gorman11e33f62009-06-16 15:31:57 -07001934/*
1935 * This is called in the allocator slow-path if the allocation request is of
1936 * sufficient urgency to ignore watermarks and take other desperate measures
1937 */
1938static inline struct page *
1939__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
1940 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001941 nodemask_t *nodemask, struct zone *preferred_zone,
1942 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001943{
1944 struct page *page;
1945
1946 do {
1947 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001948 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
Mel Gorman3dd28262009-06-16 15:32:00 -07001949 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001950
1951 if (!page && gfp_mask & __GFP_NOFAIL)
Mel Gorman0e093d992010-10-26 14:21:45 -07001952 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07001953 } while (!page && (gfp_mask & __GFP_NOFAIL));
1954
1955 return page;
1956}
1957
1958static inline
1959void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
Mel Gorman99504742011-01-13 15:46:20 -08001960 enum zone_type high_zoneidx,
1961 enum zone_type classzone_idx)
Mel Gorman11e33f62009-06-16 15:31:57 -07001962{
1963 struct zoneref *z;
1964 struct zone *zone;
1965
1966 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
Mel Gorman99504742011-01-13 15:46:20 -08001967 wakeup_kswapd(zone, order, classzone_idx);
Mel Gorman11e33f62009-06-16 15:31:57 -07001968}
1969
Peter Zijlstra341ce062009-06-16 15:32:02 -07001970static inline int
1971gfp_to_alloc_flags(gfp_t gfp_mask)
1972{
Peter Zijlstra341ce062009-06-16 15:32:02 -07001973 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
1974 const gfp_t wait = gfp_mask & __GFP_WAIT;
1975
Mel Gormana56f57f2009-06-16 15:32:02 -07001976 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07001977 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07001978
Peter Zijlstra341ce062009-06-16 15:32:02 -07001979 /*
1980 * The caller may dip into page reserves a bit more if the caller
1981 * cannot run direct reclaim, or if the caller has realtime scheduling
1982 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1983 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1984 */
Namhyung Kime6223a32010-10-26 14:21:59 -07001985 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07001986
1987 if (!wait) {
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08001988 /*
1989 * Not worth trying to allocate harder for
1990 * __GFP_NOMEMALLOC even if it can't schedule.
1991 */
1992 if (!(gfp_mask & __GFP_NOMEMALLOC))
1993 alloc_flags |= ALLOC_HARDER;
Peter Zijlstra341ce062009-06-16 15:32:02 -07001994 /*
1995 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1996 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1997 */
1998 alloc_flags &= ~ALLOC_CPUSET;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001999 } else if (unlikely(rt_task(current)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07002000 alloc_flags |= ALLOC_HARDER;
2001
2002 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2003 if (!in_interrupt() &&
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002004 ((current->flags & PF_MEMALLOC) ||
Peter Zijlstra341ce062009-06-16 15:32:02 -07002005 unlikely(test_thread_flag(TIF_MEMDIE))))
2006 alloc_flags |= ALLOC_NO_WATERMARKS;
2007 }
2008
2009 return alloc_flags;
2010}
2011
Mel Gorman11e33f62009-06-16 15:31:57 -07002012static inline struct page *
2013__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
2014 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002015 nodemask_t *nodemask, struct zone *preferred_zone,
2016 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002017{
2018 const gfp_t wait = gfp_mask & __GFP_WAIT;
2019 struct page *page = NULL;
2020 int alloc_flags;
2021 unsigned long pages_reclaimed = 0;
2022 unsigned long did_some_progress;
Mel Gorman77f1fe62011-01-13 15:45:57 -08002023 bool sync_migration = false;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002024
Christoph Lameter952f3b52006-12-06 20:33:26 -08002025 /*
Mel Gorman72807a72009-06-16 15:32:18 -07002026 * In the slowpath, we sanity check order to avoid ever trying to
2027 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2028 * be using allocators in order of preference for an area that is
2029 * too large.
2030 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07002031 if (order >= MAX_ORDER) {
2032 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07002033 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07002034 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002035
Christoph Lameter952f3b52006-12-06 20:33:26 -08002036 /*
2037 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2038 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2039 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2040 * using a larger set of nodes after it has established that the
2041 * allowed per node queues are empty and that nodes are
2042 * over allocated.
2043 */
2044 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
2045 goto nopage;
2046
Mel Gormancc4a6852009-11-11 14:26:14 -08002047restart:
Andrea Arcangeli32dba982011-01-13 15:46:49 -08002048 if (!(gfp_mask & __GFP_NO_KSWAPD))
2049 wake_all_kswapd(order, zonelist, high_zoneidx,
Mel Gorman99504742011-01-13 15:46:20 -08002050 zone_idx(preferred_zone));
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002051
Paul Jackson9bf22292005-09-06 15:18:12 -07002052 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002053 * OK, we're below the kswapd watermark and have kicked background
2054 * reclaim. Now things get more complex, so set up alloc_flags according
2055 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07002056 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002057 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058
David Rientjesf33261d2011-01-25 15:07:20 -08002059 /*
2060 * Find the true preferred zone if the allocation is unconstrained by
2061 * cpusets.
2062 */
2063 if (!(alloc_flags & ALLOC_CPUSET) && !nodemask)
2064 first_zones_zonelist(zonelist, high_zoneidx, NULL,
2065 &preferred_zone);
2066
Peter Zijlstra341ce062009-06-16 15:32:02 -07002067 /* This is the last chance, in general, before the goto nopage. */
Mel Gorman19770b32008-04-28 02:12:18 -07002068 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
Peter Zijlstra341ce062009-06-16 15:32:02 -07002069 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
2070 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002071 if (page)
2072 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073
Kirill Korotaevb43a57b2006-12-06 20:32:27 -08002074rebalance:
Mel Gorman11e33f62009-06-16 15:31:57 -07002075 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002076 if (alloc_flags & ALLOC_NO_WATERMARKS) {
2077 page = __alloc_pages_high_priority(gfp_mask, order,
2078 zonelist, high_zoneidx, nodemask,
2079 preferred_zone, migratetype);
2080 if (page)
2081 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 }
2083
2084 /* Atomic allocations - we can't balance anything */
2085 if (!wait)
2086 goto nopage;
2087
Peter Zijlstra341ce062009-06-16 15:32:02 -07002088 /* Avoid recursion of direct reclaim */
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002089 if (current->flags & PF_MEMALLOC)
Peter Zijlstra341ce062009-06-16 15:32:02 -07002090 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091
David Rientjes6583bb62009-07-29 15:02:06 -07002092 /* Avoid allocations with no watermarks from looping endlessly */
2093 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2094 goto nopage;
2095
Mel Gorman77f1fe62011-01-13 15:45:57 -08002096 /*
2097 * Try direct compaction. The first pass is asynchronous. Subsequent
2098 * attempts after direct reclaim are synchronous
2099 */
Mel Gorman56de7262010-05-24 14:32:30 -07002100 page = __alloc_pages_direct_compact(gfp_mask, order,
2101 zonelist, high_zoneidx,
2102 nodemask,
2103 alloc_flags, preferred_zone,
Mel Gorman77f1fe62011-01-13 15:45:57 -08002104 migratetype, &did_some_progress,
2105 sync_migration);
Mel Gorman56de7262010-05-24 14:32:30 -07002106 if (page)
2107 goto got_pg;
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -07002108 sync_migration = !(gfp_mask & __GFP_NO_KSWAPD);
Mel Gorman56de7262010-05-24 14:32:30 -07002109
Mel Gorman11e33f62009-06-16 15:31:57 -07002110 /* Try direct reclaim and then allocating */
2111 page = __alloc_pages_direct_reclaim(gfp_mask, order,
2112 zonelist, high_zoneidx,
2113 nodemask,
Mel Gorman5117f452009-06-16 15:31:59 -07002114 alloc_flags, preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07002115 migratetype, &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07002116 if (page)
2117 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 /*
Mel Gorman11e33f62009-06-16 15:31:57 -07002120 * If we failed to make any progress reclaiming, then we are
2121 * running out of options and have to consider going OOM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122 */
Mel Gorman11e33f62009-06-16 15:31:57 -07002123 if (!did_some_progress) {
2124 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -07002125 if (oom_killer_disabled)
2126 goto nopage;
Mel Gorman11e33f62009-06-16 15:31:57 -07002127 page = __alloc_pages_may_oom(gfp_mask, order,
2128 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002129 nodemask, preferred_zone,
2130 migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002131 if (page)
2132 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133
David Rientjes03668b32010-08-09 17:18:54 -07002134 if (!(gfp_mask & __GFP_NOFAIL)) {
2135 /*
2136 * The oom killer is not called for high-order
2137 * allocations that may fail, so if no progress
2138 * is being made, there are no other options and
2139 * retrying is unlikely to help.
2140 */
2141 if (order > PAGE_ALLOC_COSTLY_ORDER)
2142 goto nopage;
2143 /*
2144 * The oom killer is not called for lowmem
2145 * allocations to prevent needlessly killing
2146 * innocent tasks.
2147 */
2148 if (high_zoneidx < ZONE_NORMAL)
2149 goto nopage;
2150 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 goto restart;
2153 }
2154 }
2155
Mel Gorman11e33f62009-06-16 15:31:57 -07002156 /* Check if we should retry the allocation */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07002157 pages_reclaimed += did_some_progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07002158 if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
2159 /* Wait for some write requests to complete then retry */
Mel Gorman0e093d992010-10-26 14:21:45 -07002160 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 goto rebalance;
Mel Gorman3e7d3442011-01-13 15:45:56 -08002162 } else {
2163 /*
2164 * High-order allocations do not necessarily loop after
2165 * direct reclaim and reclaim/compaction depends on compaction
2166 * being called after reclaim so call directly if necessary
2167 */
2168 page = __alloc_pages_direct_compact(gfp_mask, order,
2169 zonelist, high_zoneidx,
2170 nodemask,
2171 alloc_flags, preferred_zone,
Mel Gorman77f1fe62011-01-13 15:45:57 -08002172 migratetype, &did_some_progress,
2173 sync_migration);
Mel Gorman3e7d3442011-01-13 15:45:56 -08002174 if (page)
2175 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 }
2177
2178nopage:
2179 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
David Rientjescbf978b2011-03-22 16:30:48 -07002180 unsigned int filter = SHOW_MEM_FILTER_NODES;
2181
2182 /*
2183 * This documents exceptions given to allocations in certain
2184 * contexts that are allowed to allocate outside current's set
2185 * of allowed nodes.
2186 */
2187 if (!(gfp_mask & __GFP_NOMEMALLOC))
2188 if (test_thread_flag(TIF_MEMDIE) ||
2189 (current->flags & (PF_MEMALLOC | PF_EXITING)))
2190 filter &= ~SHOW_MEM_FILTER_NODES;
2191 if (in_interrupt() || !wait)
2192 filter &= ~SHOW_MEM_FILTER_NODES;
2193
2194 pr_warning("%s: page allocation failure. order:%d, mode:0x%x\n",
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002195 current->comm, order, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 dump_stack();
David Rientjes29423e72011-03-22 16:30:47 -07002197 if (!should_suppress_show_mem())
David Rientjescbf978b2011-03-22 16:30:48 -07002198 __show_mem(filter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 }
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002200 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201got_pg:
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002202 if (kmemcheck_enabled)
2203 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 return page;
Mel Gorman11e33f62009-06-16 15:31:57 -07002205
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206}
Mel Gorman11e33f62009-06-16 15:31:57 -07002207
2208/*
2209 * This is the 'heart' of the zoned buddy allocator.
2210 */
2211struct page *
2212__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2213 struct zonelist *zonelist, nodemask_t *nodemask)
2214{
2215 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Mel Gorman5117f452009-06-16 15:31:59 -07002216 struct zone *preferred_zone;
Mel Gorman11e33f62009-06-16 15:31:57 -07002217 struct page *page;
Mel Gorman3dd28262009-06-16 15:32:00 -07002218 int migratetype = allocflags_to_migratetype(gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002219
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10002220 gfp_mask &= gfp_allowed_mask;
2221
Mel Gorman11e33f62009-06-16 15:31:57 -07002222 lockdep_trace_alloc(gfp_mask);
2223
2224 might_sleep_if(gfp_mask & __GFP_WAIT);
2225
2226 if (should_fail_alloc_page(gfp_mask, order))
2227 return NULL;
2228
2229 /*
2230 * Check the zones suitable for the gfp_mask contain at least one
2231 * valid zone. It's possible to have an empty zonelist as a result
2232 * of GFP_THISNODE and a memoryless node
2233 */
2234 if (unlikely(!zonelist->_zonerefs->zone))
2235 return NULL;
2236
Miao Xiec0ff7452010-05-24 14:32:08 -07002237 get_mems_allowed();
Mel Gorman5117f452009-06-16 15:31:59 -07002238 /* The preferred zone is used for statistics later */
David Rientjesf33261d2011-01-25 15:07:20 -08002239 first_zones_zonelist(zonelist, high_zoneidx,
2240 nodemask ? : &cpuset_current_mems_allowed,
2241 &preferred_zone);
Miao Xiec0ff7452010-05-24 14:32:08 -07002242 if (!preferred_zone) {
2243 put_mems_allowed();
Mel Gorman5117f452009-06-16 15:31:59 -07002244 return NULL;
Miao Xiec0ff7452010-05-24 14:32:08 -07002245 }
Mel Gorman5117f452009-06-16 15:31:59 -07002246
2247 /* First allocation attempt */
Mel Gorman11e33f62009-06-16 15:31:57 -07002248 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002249 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07002250 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002251 if (unlikely(!page))
2252 page = __alloc_pages_slowpath(gfp_mask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002253 zonelist, high_zoneidx, nodemask,
Mel Gorman3dd28262009-06-16 15:32:00 -07002254 preferred_zone, migratetype);
Miao Xiec0ff7452010-05-24 14:32:08 -07002255 put_mems_allowed();
Mel Gorman11e33f62009-06-16 15:31:57 -07002256
Mel Gorman4b4f2782009-09-21 17:02:41 -07002257 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002258 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259}
Mel Gormand2391712009-06-16 15:31:52 -07002260EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261
2262/*
2263 * Common helper functions.
2264 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08002265unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266{
Akinobu Mita945a1112009-09-21 17:01:47 -07002267 struct page *page;
2268
2269 /*
2270 * __get_free_pages() returns a 32-bit address, which cannot represent
2271 * a highmem page
2272 */
2273 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2274
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 page = alloc_pages(gfp_mask, order);
2276 if (!page)
2277 return 0;
2278 return (unsigned long) page_address(page);
2279}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280EXPORT_SYMBOL(__get_free_pages);
2281
Harvey Harrison920c7a52008-02-04 22:29:26 -08002282unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283{
Akinobu Mita945a1112009-09-21 17:01:47 -07002284 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286EXPORT_SYMBOL(get_zeroed_page);
2287
2288void __pagevec_free(struct pagevec *pvec)
2289{
2290 int i = pagevec_count(pvec);
2291
Mel Gorman4b4f2782009-09-21 17:02:41 -07002292 while (--i >= 0) {
2293 trace_mm_pagevec_free(pvec->pages[i], pvec->cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 free_hot_cold_page(pvec->pages[i], pvec->cold);
Mel Gorman4b4f2782009-09-21 17:02:41 -07002295 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296}
2297
Harvey Harrison920c7a52008-02-04 22:29:26 -08002298void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299{
Nick Pigginb5810032005-10-29 18:16:12 -07002300 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301 if (order == 0)
Li Hongfc916682010-03-05 13:41:54 -08002302 free_hot_cold_page(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303 else
2304 __free_pages_ok(page, order);
2305 }
2306}
2307
2308EXPORT_SYMBOL(__free_pages);
2309
Harvey Harrison920c7a52008-02-04 22:29:26 -08002310void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311{
2312 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07002313 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314 __free_pages(virt_to_page((void *)addr), order);
2315 }
2316}
2317
2318EXPORT_SYMBOL(free_pages);
2319
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002320/**
2321 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2322 * @size: the number of bytes to allocate
2323 * @gfp_mask: GFP flags for the allocation
2324 *
2325 * This function is similar to alloc_pages(), except that it allocates the
2326 * minimum number of pages to satisfy the request. alloc_pages() can only
2327 * allocate memory in power-of-two pages.
2328 *
2329 * This function is also limited by MAX_ORDER.
2330 *
2331 * Memory allocated by this function must be released by free_pages_exact().
2332 */
2333void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2334{
2335 unsigned int order = get_order(size);
2336 unsigned long addr;
2337
2338 addr = __get_free_pages(gfp_mask, order);
2339 if (addr) {
2340 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2341 unsigned long used = addr + PAGE_ALIGN(size);
2342
Kevin Cernekee5bfd7562009-07-05 12:08:19 -07002343 split_page(virt_to_page((void *)addr), order);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002344 while (used < alloc_end) {
2345 free_page(used);
2346 used += PAGE_SIZE;
2347 }
2348 }
2349
2350 return (void *)addr;
2351}
2352EXPORT_SYMBOL(alloc_pages_exact);
2353
2354/**
2355 * free_pages_exact - release memory allocated via alloc_pages_exact()
2356 * @virt: the value returned by alloc_pages_exact.
2357 * @size: size of allocation, same value as passed to alloc_pages_exact().
2358 *
2359 * Release the memory allocated by a previous call to alloc_pages_exact.
2360 */
2361void free_pages_exact(void *virt, size_t size)
2362{
2363 unsigned long addr = (unsigned long)virt;
2364 unsigned long end = addr + PAGE_ALIGN(size);
2365
2366 while (addr < end) {
2367 free_page(addr);
2368 addr += PAGE_SIZE;
2369 }
2370}
2371EXPORT_SYMBOL(free_pages_exact);
2372
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373static unsigned int nr_free_zone_pages(int offset)
2374{
Mel Gormandd1a2392008-04-28 02:12:17 -07002375 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002376 struct zone *zone;
2377
Martin J. Blighe310fd42005-07-29 22:59:18 -07002378 /* Just pick one node, since fallback list is circular */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 unsigned int sum = 0;
2380
Mel Gorman0e884602008-04-28 02:12:14 -07002381 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382
Mel Gorman54a6eb52008-04-28 02:12:16 -07002383 for_each_zone_zonelist(zone, z, zonelist, offset) {
Martin J. Blighe310fd42005-07-29 22:59:18 -07002384 unsigned long size = zone->present_pages;
Mel Gorman41858962009-06-16 15:32:12 -07002385 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07002386 if (size > high)
2387 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 }
2389
2390 return sum;
2391}
2392
2393/*
2394 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2395 */
2396unsigned int nr_free_buffer_pages(void)
2397{
Al Viroaf4ca452005-10-21 02:55:38 -04002398 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399}
Meelap Shahc2f1a552007-07-17 04:04:39 -07002400EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401
2402/*
2403 * Amount of free RAM allocatable within all zones
2404 */
2405unsigned int nr_free_pagecache_pages(void)
2406{
Mel Gorman2a1e2742007-07-17 04:03:12 -07002407 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002409
2410static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411{
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002412 if (NUMA_BUILD)
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08002413 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416void si_meminfo(struct sysinfo *val)
2417{
2418 val->totalram = totalram_pages;
2419 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002420 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422 val->totalhigh = totalhigh_pages;
2423 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 val->mem_unit = PAGE_SIZE;
2425}
2426
2427EXPORT_SYMBOL(si_meminfo);
2428
2429#ifdef CONFIG_NUMA
2430void si_meminfo_node(struct sysinfo *val, int nid)
2431{
2432 pg_data_t *pgdat = NODE_DATA(nid);
2433
2434 val->totalram = pgdat->node_present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002435 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002436#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002438 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2439 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002440#else
2441 val->totalhigh = 0;
2442 val->freehigh = 0;
2443#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444 val->mem_unit = PAGE_SIZE;
2445}
2446#endif
2447
David Rientjesddd588b2011-03-22 16:30:46 -07002448/*
2449 * Determine whether the zone's node should be displayed or not, depending on
2450 * whether SHOW_MEM_FILTER_NODES was passed to __show_free_areas().
2451 */
2452static bool skip_free_areas_zone(unsigned int flags, const struct zone *zone)
2453{
2454 bool ret = false;
2455
2456 if (!(flags & SHOW_MEM_FILTER_NODES))
2457 goto out;
2458
2459 get_mems_allowed();
2460 ret = !node_isset(zone->zone_pgdat->node_id,
2461 cpuset_current_mems_allowed);
2462 put_mems_allowed();
2463out:
2464 return ret;
2465}
2466
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467#define K(x) ((x) << (PAGE_SHIFT-10))
2468
2469/*
2470 * Show free area list (used inside shift_scroll-lock stuff)
2471 * We also calculate the percentage fragmentation. We do this by counting the
2472 * memory on each free list with the exception of the first item on the list.
David Rientjesddd588b2011-03-22 16:30:46 -07002473 * Suppresses nodes that are not allowed by current's cpuset if
2474 * SHOW_MEM_FILTER_NODES is passed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475 */
David Rientjesddd588b2011-03-22 16:30:46 -07002476void __show_free_areas(unsigned int filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477{
Jes Sorensenc7241912006-09-27 01:50:05 -07002478 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479 struct zone *zone;
2480
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002481 for_each_populated_zone(zone) {
David Rientjesddd588b2011-03-22 16:30:46 -07002482 if (skip_free_areas_zone(filter, zone))
2483 continue;
Jes Sorensenc7241912006-09-27 01:50:05 -07002484 show_node(zone);
2485 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002486
Dave Jones6b482c62005-11-10 15:45:56 -05002487 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488 struct per_cpu_pageset *pageset;
2489
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002490 pageset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002492 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2493 cpu, pageset->pcp.high,
2494 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495 }
2496 }
2497
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002498 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2499 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002500 " unevictable:%lu"
Andrew Mortonb76146e2009-10-26 16:49:52 -07002501 " dirty:%lu writeback:%lu unstable:%lu\n"
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07002502 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002503 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002504 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002505 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002506 global_page_state(NR_ISOLATED_ANON),
2507 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002508 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002509 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002510 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07002511 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07002512 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07002513 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08002514 global_page_state(NR_FREE_PAGES),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07002515 global_page_state(NR_SLAB_RECLAIMABLE),
2516 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07002517 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002518 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08002519 global_page_state(NR_PAGETABLE),
2520 global_page_state(NR_BOUNCE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002522 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523 int i;
2524
David Rientjesddd588b2011-03-22 16:30:46 -07002525 if (skip_free_areas_zone(filter, zone))
2526 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527 show_node(zone);
2528 printk("%s"
2529 " free:%lukB"
2530 " min:%lukB"
2531 " low:%lukB"
2532 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002533 " active_anon:%lukB"
2534 " inactive_anon:%lukB"
2535 " active_file:%lukB"
2536 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002537 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002538 " isolated(anon):%lukB"
2539 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 " present:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002541 " mlocked:%lukB"
2542 " dirty:%lukB"
2543 " writeback:%lukB"
2544 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002545 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002546 " slab_reclaimable:%lukB"
2547 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07002548 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002549 " pagetables:%lukB"
2550 " unstable:%lukB"
2551 " bounce:%lukB"
2552 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553 " pages_scanned:%lu"
2554 " all_unreclaimable? %s"
2555 "\n",
2556 zone->name,
Mel Gorman88f5acf2011-01-13 15:45:41 -08002557 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07002558 K(min_wmark_pages(zone)),
2559 K(low_wmark_pages(zone)),
2560 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002561 K(zone_page_state(zone, NR_ACTIVE_ANON)),
2562 K(zone_page_state(zone, NR_INACTIVE_ANON)),
2563 K(zone_page_state(zone, NR_ACTIVE_FILE)),
2564 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002565 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002566 K(zone_page_state(zone, NR_ISOLATED_ANON)),
2567 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 K(zone->present_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002569 K(zone_page_state(zone, NR_MLOCK)),
2570 K(zone_page_state(zone, NR_FILE_DIRTY)),
2571 K(zone_page_state(zone, NR_WRITEBACK)),
2572 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002573 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002574 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
2575 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07002576 zone_page_state(zone, NR_KERNEL_STACK) *
2577 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002578 K(zone_page_state(zone, NR_PAGETABLE)),
2579 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
2580 K(zone_page_state(zone, NR_BOUNCE)),
2581 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582 zone->pages_scanned,
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -08002583 (zone->all_unreclaimable ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 );
2585 printk("lowmem_reserve[]:");
2586 for (i = 0; i < MAX_NR_ZONES; i++)
2587 printk(" %lu", zone->lowmem_reserve[i]);
2588 printk("\n");
2589 }
2590
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002591 for_each_populated_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002592 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593
David Rientjesddd588b2011-03-22 16:30:46 -07002594 if (skip_free_areas_zone(filter, zone))
2595 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596 show_node(zone);
2597 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598
2599 spin_lock_irqsave(&zone->lock, flags);
2600 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002601 nr[order] = zone->free_area[order].nr_free;
2602 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 }
2604 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002605 for (order = 0; order < MAX_ORDER; order++)
2606 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607 printk("= %lukB\n", K(total));
2608 }
2609
Larry Woodmane6f36022008-02-04 22:29:30 -08002610 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2611
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612 show_swap_cache_info();
2613}
2614
David Rientjesddd588b2011-03-22 16:30:46 -07002615void show_free_areas(void)
2616{
2617 __show_free_areas(0);
2618}
2619
Mel Gorman19770b32008-04-28 02:12:18 -07002620static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2621{
2622 zoneref->zone = zone;
2623 zoneref->zone_idx = zone_idx(zone);
2624}
2625
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626/*
2627 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08002628 *
2629 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002631static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2632 int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633{
Christoph Lameter1a932052006-01-06 00:11:16 -08002634 struct zone *zone;
2635
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002636 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002637 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08002638
2639 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002640 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08002641 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08002642 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002643 zoneref_set_zone(zone,
2644 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08002645 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08002647
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002648 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08002649 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650}
2651
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002652
2653/*
2654 * zonelist_order:
2655 * 0 = automatic detection of better ordering.
2656 * 1 = order by ([node] distance, -zonetype)
2657 * 2 = order by (-zonetype, [node] distance)
2658 *
2659 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2660 * the same zonelist. So only NUMA can configure this param.
2661 */
2662#define ZONELIST_ORDER_DEFAULT 0
2663#define ZONELIST_ORDER_NODE 1
2664#define ZONELIST_ORDER_ZONE 2
2665
2666/* zonelist order in the kernel.
2667 * set_zonelist_order() will set this to NODE or ZONE.
2668 */
2669static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2670static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2671
2672
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002674/* The value user specified ....changed by config */
2675static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2676/* string for sysctl */
2677#define NUMA_ZONELIST_ORDER_LEN 16
2678char numa_zonelist_order[16] = "default";
2679
2680/*
2681 * interface for configure zonelist ordering.
2682 * command line option "numa_zonelist_order"
2683 * = "[dD]efault - default, automatic configuration.
2684 * = "[nN]ode - order by node locality, then by zone within node
2685 * = "[zZ]one - order by zone, then by locality within zone
2686 */
2687
2688static int __parse_numa_zonelist_order(char *s)
2689{
2690 if (*s == 'd' || *s == 'D') {
2691 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2692 } else if (*s == 'n' || *s == 'N') {
2693 user_zonelist_order = ZONELIST_ORDER_NODE;
2694 } else if (*s == 'z' || *s == 'Z') {
2695 user_zonelist_order = ZONELIST_ORDER_ZONE;
2696 } else {
2697 printk(KERN_WARNING
2698 "Ignoring invalid numa_zonelist_order value: "
2699 "%s\n", s);
2700 return -EINVAL;
2701 }
2702 return 0;
2703}
2704
2705static __init int setup_numa_zonelist_order(char *s)
2706{
Volodymyr G. Lukiianykecb256f2011-01-13 15:46:26 -08002707 int ret;
2708
2709 if (!s)
2710 return 0;
2711
2712 ret = __parse_numa_zonelist_order(s);
2713 if (ret == 0)
2714 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
2715
2716 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002717}
2718early_param("numa_zonelist_order", setup_numa_zonelist_order);
2719
2720/*
2721 * sysctl handler for numa_zonelist_order
2722 */
2723int numa_zonelist_order_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07002724 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002725 loff_t *ppos)
2726{
2727 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2728 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01002729 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002730
Andi Kleen443c6f12009-12-23 21:00:47 +01002731 mutex_lock(&zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002732 if (write)
Andi Kleen443c6f12009-12-23 21:00:47 +01002733 strcpy(saved_string, (char*)table->data);
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07002734 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002735 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01002736 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002737 if (write) {
2738 int oldval = user_zonelist_order;
2739 if (__parse_numa_zonelist_order((char*)table->data)) {
2740 /*
2741 * bogus value. restore saved string
2742 */
2743 strncpy((char*)table->data, saved_string,
2744 NUMA_ZONELIST_ORDER_LEN);
2745 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07002746 } else if (oldval != user_zonelist_order) {
2747 mutex_lock(&zonelists_mutex);
Haicheng Li1f522502010-05-24 14:32:51 -07002748 build_all_zonelists(NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07002749 mutex_unlock(&zonelists_mutex);
2750 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002751 }
Andi Kleen443c6f12009-12-23 21:00:47 +01002752out:
2753 mutex_unlock(&zl_order_mutex);
2754 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002755}
2756
2757
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002758#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002759static int node_load[MAX_NUMNODES];
2760
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07002762 * 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 -07002763 * @node: node whose fallback list we're appending
2764 * @used_node_mask: nodemask_t of already used nodes
2765 *
2766 * We use a number of factors to determine which is the next node that should
2767 * appear on a given node's fallback list. The node should not have appeared
2768 * already in @node's fallback list, and it should be the next closest node
2769 * according to the distance array (which contains arbitrary distance values
2770 * from each node to each node in the system), and should also prefer nodes
2771 * with no CPUs, since presumably they'll have very little allocation pressure
2772 * on them otherwise.
2773 * It returns -1 if no node is found.
2774 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002775static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002777 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778 int min_val = INT_MAX;
2779 int best_node = -1;
Rusty Russella70f7302009-03-13 14:49:46 +10302780 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002782 /* Use the local node if we haven't already */
2783 if (!node_isset(node, *used_node_mask)) {
2784 node_set(node, *used_node_mask);
2785 return node;
2786 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002788 for_each_node_state(n, N_HIGH_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789
2790 /* Don't want a node to appear more than once */
2791 if (node_isset(n, *used_node_mask))
2792 continue;
2793
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794 /* Use the distance array to find the distance */
2795 val = node_distance(node, n);
2796
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002797 /* Penalize nodes under us ("prefer the next node") */
2798 val += (n < node);
2799
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10302801 tmp = cpumask_of_node(n);
2802 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803 val += PENALTY_FOR_NODE_WITH_CPUS;
2804
2805 /* Slight preference for less loaded node */
2806 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2807 val += node_load[n];
2808
2809 if (val < min_val) {
2810 min_val = val;
2811 best_node = n;
2812 }
2813 }
2814
2815 if (best_node >= 0)
2816 node_set(best_node, *used_node_mask);
2817
2818 return best_node;
2819}
2820
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002821
2822/*
2823 * Build zonelists ordered by node and zones within node.
2824 * This results in maximum locality--normal zone overflows into local
2825 * DMA zone, if any--but risks exhausting DMA zone.
2826 */
2827static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002829 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002831
Mel Gorman54a6eb52008-04-28 02:12:16 -07002832 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07002833 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07002834 ;
2835 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2836 MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002837 zonelist->_zonerefs[j].zone = NULL;
2838 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002839}
2840
2841/*
Christoph Lameter523b9452007-10-16 01:25:37 -07002842 * Build gfp_thisnode zonelists
2843 */
2844static void build_thisnode_zonelists(pg_data_t *pgdat)
2845{
Christoph Lameter523b9452007-10-16 01:25:37 -07002846 int j;
2847 struct zonelist *zonelist;
2848
Mel Gorman54a6eb52008-04-28 02:12:16 -07002849 zonelist = &pgdat->node_zonelists[1];
2850 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002851 zonelist->_zonerefs[j].zone = NULL;
2852 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07002853}
2854
2855/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002856 * Build zonelists ordered by zone and nodes within zones.
2857 * This results in conserving DMA zone[s] until all Normal memory is
2858 * exhausted, but results in overflowing to remote node while memory
2859 * may still exist in local DMA zone.
2860 */
2861static int node_order[MAX_NUMNODES];
2862
2863static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2864{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002865 int pos, j, node;
2866 int zone_type; /* needs to be signed */
2867 struct zone *z;
2868 struct zonelist *zonelist;
2869
Mel Gorman54a6eb52008-04-28 02:12:16 -07002870 zonelist = &pgdat->node_zonelists[0];
2871 pos = 0;
2872 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2873 for (j = 0; j < nr_nodes; j++) {
2874 node = node_order[j];
2875 z = &NODE_DATA(node)->node_zones[zone_type];
2876 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002877 zoneref_set_zone(z,
2878 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002879 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002880 }
2881 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002882 }
Mel Gormandd1a2392008-04-28 02:12:17 -07002883 zonelist->_zonerefs[pos].zone = NULL;
2884 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002885}
2886
2887static int default_zonelist_order(void)
2888{
2889 int nid, zone_type;
2890 unsigned long low_kmem_size,total_size;
2891 struct zone *z;
2892 int average_size;
2893 /*
Thomas Weber88393162010-03-16 11:47:56 +01002894 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002895 * If they are really small and used heavily, the system can fall
2896 * into OOM very easily.
David Rientjese325c902010-05-24 14:32:13 -07002897 * This function detect ZONE_DMA/DMA32 size and configures zone order.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002898 */
2899 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2900 low_kmem_size = 0;
2901 total_size = 0;
2902 for_each_online_node(nid) {
2903 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2904 z = &NODE_DATA(nid)->node_zones[zone_type];
2905 if (populated_zone(z)) {
2906 if (zone_type < ZONE_NORMAL)
2907 low_kmem_size += z->present_pages;
2908 total_size += z->present_pages;
David Rientjese325c902010-05-24 14:32:13 -07002909 } else if (zone_type == ZONE_NORMAL) {
2910 /*
2911 * If any node has only lowmem, then node order
2912 * is preferred to allow kernel allocations
2913 * locally; otherwise, they can easily infringe
2914 * on other nodes when there is an abundance of
2915 * lowmem available to allocate from.
2916 */
2917 return ZONELIST_ORDER_NODE;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002918 }
2919 }
2920 }
2921 if (!low_kmem_size || /* there are no DMA area. */
2922 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2923 return ZONELIST_ORDER_NODE;
2924 /*
2925 * look into each node's config.
2926 * If there is a node whose DMA/DMA32 memory is very big area on
2927 * local memory, NODE_ORDER may be suitable.
2928 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002929 average_size = total_size /
2930 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002931 for_each_online_node(nid) {
2932 low_kmem_size = 0;
2933 total_size = 0;
2934 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2935 z = &NODE_DATA(nid)->node_zones[zone_type];
2936 if (populated_zone(z)) {
2937 if (zone_type < ZONE_NORMAL)
2938 low_kmem_size += z->present_pages;
2939 total_size += z->present_pages;
2940 }
2941 }
2942 if (low_kmem_size &&
2943 total_size > average_size && /* ignore small node */
2944 low_kmem_size > total_size * 70/100)
2945 return ZONELIST_ORDER_NODE;
2946 }
2947 return ZONELIST_ORDER_ZONE;
2948}
2949
2950static void set_zonelist_order(void)
2951{
2952 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2953 current_zonelist_order = default_zonelist_order();
2954 else
2955 current_zonelist_order = user_zonelist_order;
2956}
2957
2958static void build_zonelists(pg_data_t *pgdat)
2959{
2960 int j, node, load;
2961 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002962 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002963 int local_node, prev_node;
2964 struct zonelist *zonelist;
2965 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002966
2967 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07002968 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002969 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07002970 zonelist->_zonerefs[0].zone = NULL;
2971 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002972 }
2973
2974 /* NUMA-aware ordering of nodes */
2975 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002976 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002977 prev_node = local_node;
2978 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002979
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002980 memset(node_order, 0, sizeof(node_order));
2981 j = 0;
2982
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08002984 int distance = node_distance(local_node, node);
2985
2986 /*
2987 * If another node is sufficiently far away then it is better
2988 * to reclaim pages in a zone before going off node.
2989 */
2990 if (distance > RECLAIM_DISTANCE)
2991 zone_reclaim_mode = 1;
2992
Linus Torvalds1da177e2005-04-16 15:20:36 -07002993 /*
2994 * We don't want to pressure a particular node.
2995 * So adding penalty to the first node in same
2996 * distance group to make it round-robin.
2997 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08002998 if (distance != node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002999 node_load[node] = load;
3000
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001 prev_node = node;
3002 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003003 if (order == ZONELIST_ORDER_NODE)
3004 build_zonelists_in_node_order(pgdat, node);
3005 else
3006 node_order[j++] = node; /* remember order */
3007 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003008
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003009 if (order == ZONELIST_ORDER_ZONE) {
3010 /* calculate node order -- i.e., DMA last! */
3011 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003012 }
Christoph Lameter523b9452007-10-16 01:25:37 -07003013
3014 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003015}
3016
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003017/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003018static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003019{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003020 struct zonelist *zonelist;
3021 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07003022 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003023
Mel Gorman54a6eb52008-04-28 02:12:16 -07003024 zonelist = &pgdat->node_zonelists[0];
3025 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3026 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07003027 for (z = zonelist->_zonerefs; z->zone; z++)
3028 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003029}
3030
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003031#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3032/*
3033 * Return node id of node used for "local" allocations.
3034 * I.e., first node id of first zone in arg node's generic zonelist.
3035 * Used for initializing percpu 'numa_mem', which is used primarily
3036 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3037 */
3038int local_memory_node(int node)
3039{
3040 struct zone *zone;
3041
3042 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3043 gfp_zone(GFP_KERNEL),
3044 NULL,
3045 &zone);
3046 return zone->node;
3047}
3048#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003049
Linus Torvalds1da177e2005-04-16 15:20:36 -07003050#else /* CONFIG_NUMA */
3051
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003052static void set_zonelist_order(void)
3053{
3054 current_zonelist_order = ZONELIST_ORDER_ZONE;
3055}
3056
3057static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058{
Christoph Lameter19655d32006-09-25 23:31:19 -07003059 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003060 enum zone_type j;
3061 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003062
3063 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064
Mel Gorman54a6eb52008-04-28 02:12:16 -07003065 zonelist = &pgdat->node_zonelists[0];
3066 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067
Mel Gorman54a6eb52008-04-28 02:12:16 -07003068 /*
3069 * Now we build the zonelist so that it contains the zones
3070 * of all the other nodes.
3071 * We don't want to pressure a particular node, so when
3072 * building the zones for node N, we make sure that the
3073 * zones coming right after the local ones are those from
3074 * node N+1 (modulo N)
3075 */
3076 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3077 if (!node_online(node))
3078 continue;
3079 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3080 MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003081 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07003082 for (node = 0; node < local_node; node++) {
3083 if (!node_online(node))
3084 continue;
3085 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3086 MAX_NR_ZONES - 1);
3087 }
3088
Mel Gormandd1a2392008-04-28 02:12:17 -07003089 zonelist->_zonerefs[j].zone = NULL;
3090 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091}
3092
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003093/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003094static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003095{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003096 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003097}
3098
Linus Torvalds1da177e2005-04-16 15:20:36 -07003099#endif /* CONFIG_NUMA */
3100
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003101/*
3102 * Boot pageset table. One per cpu which is going to be used for all
3103 * zones and all nodes. The parameters will be set in such a way
3104 * that an item put on a list will immediately be handed over to
3105 * the buddy list. This is safe since pageset manipulation is done
3106 * with interrupts disabled.
3107 *
3108 * The boot_pagesets must be kept even after bootup is complete for
3109 * unused processors and/or zones. They do play a role for bootstrapping
3110 * hotplugged processors.
3111 *
3112 * zoneinfo_show() and maybe other functions do
3113 * not check if the processor is online before following the pageset pointer.
3114 * Other parts of the kernel may not check if the zone is available.
3115 */
3116static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3117static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07003118static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003119
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003120/*
3121 * Global mutex to protect against size modification of zonelists
3122 * as well as to serialize pageset setup for the new populated zone.
3123 */
3124DEFINE_MUTEX(zonelists_mutex);
3125
Rusty Russell9b1a4d32008-07-28 12:16:30 -05003126/* return values int ....just for stop_machine() */
Haicheng Li1f522502010-05-24 14:32:51 -07003127static __init_refok int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003128{
Yasunori Goto68113782006-06-23 02:03:11 -07003129 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003130 int cpu;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003131
Bo Liu7f9cfb32009-08-18 14:11:19 -07003132#ifdef CONFIG_NUMA
3133 memset(node_load, 0, sizeof(node_load));
3134#endif
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003135 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07003136 pg_data_t *pgdat = NODE_DATA(nid);
3137
3138 build_zonelists(pgdat);
3139 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003140 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003141
3142 /*
3143 * Initialize the boot_pagesets that are going to be used
3144 * for bootstrapping processors. The real pagesets for
3145 * each zone will be allocated later when the per cpu
3146 * allocator is available.
3147 *
3148 * boot_pagesets are used also for bootstrapping offline
3149 * cpus if the system is already booted because the pagesets
3150 * are needed to initialize allocators on a specific cpu too.
3151 * F.e. the percpu allocator needs the page allocator which
3152 * needs the percpu allocator in order to allocate its pagesets
3153 * (a chicken-egg dilemma).
3154 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003155 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003156 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3157
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003158#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3159 /*
3160 * We now know the "local memory node" for each node--
3161 * i.e., the node of the first zone in the generic zonelist.
3162 * Set up numa_mem percpu variable for on-line cpus. During
3163 * boot, only the boot cpu should be on-line; we'll init the
3164 * secondary cpus' numa_mem as they come on-line. During
3165 * node/memory hotplug, we'll fixup all on-line cpus.
3166 */
3167 if (cpu_online(cpu))
3168 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3169#endif
3170 }
3171
Yasunori Goto68113782006-06-23 02:03:11 -07003172 return 0;
3173}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003174
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003175/*
3176 * Called with zonelists_mutex held always
3177 * unless system_state == SYSTEM_BOOTING.
3178 */
Haicheng Li1f522502010-05-24 14:32:51 -07003179void build_all_zonelists(void *data)
Yasunori Goto68113782006-06-23 02:03:11 -07003180{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003181 set_zonelist_order();
3182
Yasunori Goto68113782006-06-23 02:03:11 -07003183 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07003184 __build_all_zonelists(NULL);
Mel Gorman68ad8df2008-07-23 21:26:52 -07003185 mminit_verify_zonelist();
Yasunori Goto68113782006-06-23 02:03:11 -07003186 cpuset_init_current_mems_allowed();
3187 } else {
Simon Arlott183ff222007-10-20 01:27:18 +02003188 /* we have to stop all cpus to guarantee there is no user
Yasunori Goto68113782006-06-23 02:03:11 -07003189 of zonelist */
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003190#ifdef CONFIG_MEMORY_HOTPLUG
3191 if (data)
3192 setup_zone_pageset((struct zone *)data);
3193#endif
3194 stop_machine(__build_all_zonelists, NULL, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07003195 /* cpuset refresh routine should be here */
3196 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07003197 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003198 /*
3199 * Disable grouping by mobility if the number of pages in the
3200 * system is too low to allow the mechanism to work. It would be
3201 * more accurate, but expensive to check per-zone. This check is
3202 * made on memory-hotadd so a system can start with mobility
3203 * disabled and enable it later
3204 */
Mel Gormand9c23402007-10-16 01:26:01 -07003205 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003206 page_group_by_mobility_disabled = 1;
3207 else
3208 page_group_by_mobility_disabled = 0;
3209
3210 printk("Built %i zonelists in %s order, mobility grouping %s. "
3211 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003212 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003213 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003214 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003215 vm_total_pages);
3216#ifdef CONFIG_NUMA
3217 printk("Policy zone: %s\n", zone_names[policy_zone]);
3218#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003219}
3220
3221/*
3222 * Helper functions to size the waitqueue hash table.
3223 * Essentially these want to choose hash table sizes sufficiently
3224 * large so that collisions trying to wait on pages are rare.
3225 * But in fact, the number of active page waitqueues on typical
3226 * systems is ridiculously low, less than 200. So this is even
3227 * conservative, even though it seems large.
3228 *
3229 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3230 * waitqueues, i.e. the size of the waitq table given the number of pages.
3231 */
3232#define PAGES_PER_WAITQUEUE 256
3233
Yasunori Gotocca448f2006-06-23 02:03:10 -07003234#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07003235static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003236{
3237 unsigned long size = 1;
3238
3239 pages /= PAGES_PER_WAITQUEUE;
3240
3241 while (size < pages)
3242 size <<= 1;
3243
3244 /*
3245 * Once we have dozens or even hundreds of threads sleeping
3246 * on IO we've got bigger problems than wait queue collision.
3247 * Limit the size of the wait table to a reasonable size.
3248 */
3249 size = min(size, 4096UL);
3250
3251 return max(size, 4UL);
3252}
Yasunori Gotocca448f2006-06-23 02:03:10 -07003253#else
3254/*
3255 * A zone's size might be changed by hot-add, so it is not possible to determine
3256 * a suitable size for its wait_table. So we use the maximum size now.
3257 *
3258 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3259 *
3260 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3261 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3262 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3263 *
3264 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3265 * or more by the traditional way. (See above). It equals:
3266 *
3267 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3268 * ia64(16K page size) : = ( 8G + 4M)byte.
3269 * powerpc (64K page size) : = (32G +16M)byte.
3270 */
3271static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3272{
3273 return 4096UL;
3274}
3275#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003276
3277/*
3278 * This is an integer logarithm so that shifts can be used later
3279 * to extract the more random high bits from the multiplicative
3280 * hash function before the remainder is taken.
3281 */
3282static inline unsigned long wait_table_bits(unsigned long size)
3283{
3284 return ffz(~size);
3285}
3286
3287#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3288
Mel Gorman56fd56b2007-10-16 01:25:58 -07003289/*
Mel Gormand9c23402007-10-16 01:26:01 -07003290 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07003291 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3292 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07003293 * higher will lead to a bigger reserve which will get freed as contiguous
3294 * blocks as reclaim kicks in
3295 */
3296static void setup_zone_migrate_reserve(struct zone *zone)
3297{
3298 unsigned long start_pfn, pfn, end_pfn;
3299 struct page *page;
Mel Gorman78986a62009-09-21 17:03:02 -07003300 unsigned long block_migratetype;
3301 int reserve;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003302
3303 /* Get the start pfn, end pfn and the number of blocks to reserve */
3304 start_pfn = zone->zone_start_pfn;
3305 end_pfn = start_pfn + zone->spanned_pages;
Mel Gorman41858962009-06-16 15:32:12 -07003306 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07003307 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003308
Mel Gorman78986a62009-09-21 17:03:02 -07003309 /*
3310 * Reserve blocks are generally in place to help high-order atomic
3311 * allocations that are short-lived. A min_free_kbytes value that
3312 * would result in more than 2 reserve blocks for atomic allocations
3313 * is assumed to be in place to help anti-fragmentation for the
3314 * future allocation of hugepages at runtime.
3315 */
3316 reserve = min(2, reserve);
3317
Mel Gormand9c23402007-10-16 01:26:01 -07003318 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07003319 if (!pfn_valid(pfn))
3320 continue;
3321 page = pfn_to_page(pfn);
3322
Adam Litke344c7902008-09-02 14:35:38 -07003323 /* Watch out for overlapping nodes */
3324 if (page_to_nid(page) != zone_to_nid(zone))
3325 continue;
3326
Mel Gorman56fd56b2007-10-16 01:25:58 -07003327 /* Blocks with reserved pages will never free, skip them. */
3328 if (PageReserved(page))
3329 continue;
3330
3331 block_migratetype = get_pageblock_migratetype(page);
3332
3333 /* If this block is reserved, account for it */
3334 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
3335 reserve--;
3336 continue;
3337 }
3338
3339 /* Suitable for reserving if this block is movable */
3340 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
3341 set_pageblock_migratetype(page, MIGRATE_RESERVE);
3342 move_freepages_block(zone, page, MIGRATE_RESERVE);
3343 reserve--;
3344 continue;
3345 }
3346
3347 /*
3348 * If the reserve is met and this is a previous reserved block,
3349 * take it back
3350 */
3351 if (block_migratetype == MIGRATE_RESERVE) {
3352 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
3353 move_freepages_block(zone, page, MIGRATE_MOVABLE);
3354 }
3355 }
3356}
Mel Gormanac0e5b72007-10-16 01:25:58 -07003357
Linus Torvalds1da177e2005-04-16 15:20:36 -07003358/*
3359 * Initially all pages are reserved - free ones are freed
3360 * up by free_all_bootmem() once the early boot process is
3361 * done. Non-atomic initialization, single-pass.
3362 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003363void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa02007-01-10 23:15:30 -08003364 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003365{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003366 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07003367 unsigned long end_pfn = start_pfn + size;
3368 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003369 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003370
Hugh Dickins22b31ee2009-01-06 14:40:09 -08003371 if (highest_memmap_pfn < end_pfn - 1)
3372 highest_memmap_pfn = end_pfn - 1;
3373
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003374 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08003375 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa02007-01-10 23:15:30 -08003376 /*
3377 * There can be holes in boot-time mem_map[]s
3378 * handed to this function. They do not
3379 * exist on hotplugged memory.
3380 */
3381 if (context == MEMMAP_EARLY) {
3382 if (!early_pfn_valid(pfn))
3383 continue;
3384 if (!early_pfn_in_nid(pfn, nid))
3385 continue;
3386 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003387 page = pfn_to_page(pfn);
3388 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07003389 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08003390 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003391 reset_page_mapcount(page);
3392 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003393 /*
3394 * Mark the block movable so that blocks are reserved for
3395 * movable at startup. This will force kernel allocations
3396 * to reserve their blocks rather than leaking throughout
3397 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07003398 * kernel allocations are made. Later some blocks near
3399 * the start are marked MIGRATE_RESERVE by
3400 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003401 *
3402 * bitmap is created for zone's valid pfn range. but memmap
3403 * can be created for invalid pages (for alignment)
3404 * check here not to call set_pageblock_migratetype() against
3405 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003406 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003407 if ((z->zone_start_pfn <= pfn)
3408 && (pfn < z->zone_start_pfn + z->spanned_pages)
3409 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07003410 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003411
Linus Torvalds1da177e2005-04-16 15:20:36 -07003412 INIT_LIST_HEAD(&page->lru);
3413#ifdef WANT_PAGE_VIRTUAL
3414 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3415 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07003416 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003417#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003418 }
3419}
3420
Andi Kleen1e548de2008-02-04 22:29:26 -08003421static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003422{
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003423 int order, t;
3424 for_each_migratetype_order(order, t) {
3425 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003426 zone->free_area[order].nr_free = 0;
3427 }
3428}
3429
3430#ifndef __HAVE_ARCH_MEMMAP_INIT
3431#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa02007-01-10 23:15:30 -08003432 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003433#endif
3434
Sam Ravnborg1d6f4e62008-01-17 15:21:12 -08003435static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003436{
David Howells3a6be872009-05-06 16:03:03 -07003437#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003438 int batch;
3439
3440 /*
3441 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07003442 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003443 *
3444 * OK, so we don't know how big the cache is. So guess.
3445 */
3446 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07003447 if (batch * PAGE_SIZE > 512 * 1024)
3448 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003449 batch /= 4; /* We effectively *= 4 below */
3450 if (batch < 1)
3451 batch = 1;
3452
3453 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11003454 * Clamp the batch to a 2^n - 1 value. Having a power
3455 * of 2 value was found to be more likely to have
3456 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003457 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11003458 * For example if 2 tasks are alternately allocating
3459 * batches of pages, one task can end up with a lot
3460 * of pages of one half of the possible page colors
3461 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003462 */
David Howells91552032009-05-06 16:03:02 -07003463 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07003464
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003465 return batch;
David Howells3a6be872009-05-06 16:03:03 -07003466
3467#else
3468 /* The deferral and batching of frees should be suppressed under NOMMU
3469 * conditions.
3470 *
3471 * The problem is that NOMMU needs to be able to allocate large chunks
3472 * of contiguous memory as there's no hardware page translation to
3473 * assemble apparent contiguous memory from discontiguous pages.
3474 *
3475 * Queueing large contiguous runs of pages for batching, however,
3476 * causes the pages to actually be freed in smaller chunks. As there
3477 * can be a significant delay between the individual batches being
3478 * recycled, this leads to the once large chunks of space being
3479 * fragmented and becoming unavailable for high-order allocations.
3480 */
3481 return 0;
3482#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003483}
3484
Adrian Bunkb69a7282008-07-23 21:28:12 -07003485static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
Christoph Lameter2caaad42005-06-21 17:15:00 -07003486{
3487 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003488 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003489
Magnus Damm1c6fe942005-10-26 01:58:59 -07003490 memset(p, 0, sizeof(*p));
3491
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003492 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003493 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003494 pcp->high = 6 * batch;
3495 pcp->batch = max(1UL, 1 * batch);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003496 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
3497 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07003498}
3499
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003500/*
3501 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3502 * to the value high for the pageset p.
3503 */
3504
3505static void setup_pagelist_highmark(struct per_cpu_pageset *p,
3506 unsigned long high)
3507{
3508 struct per_cpu_pages *pcp;
3509
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003510 pcp = &p->pcp;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003511 pcp->high = high;
3512 pcp->batch = max(1UL, high/4);
3513 if ((high/4) > (PAGE_SHIFT * 8))
3514 pcp->batch = PAGE_SHIFT * 8;
3515}
3516
Wu Fengguang319774e2010-05-24 14:32:49 -07003517static __meminit void setup_zone_pageset(struct zone *zone)
3518{
3519 int cpu;
3520
3521 zone->pageset = alloc_percpu(struct per_cpu_pageset);
3522
3523 for_each_possible_cpu(cpu) {
3524 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
3525
3526 setup_pageset(pcp, zone_batchsize(zone));
3527
3528 if (percpu_pagelist_fraction)
3529 setup_pagelist_highmark(pcp,
3530 (zone->present_pages /
3531 percpu_pagelist_fraction));
3532 }
3533}
3534
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003535/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003536 * Allocate per cpu pagesets and initialize them.
3537 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07003538 */
Al Viro78d99552005-12-15 09:18:25 +00003539void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003540{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003541 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003542
Wu Fengguang319774e2010-05-24 14:32:49 -07003543 for_each_populated_zone(zone)
3544 setup_zone_pageset(zone);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003545}
3546
Sam Ravnborg577a32f2007-05-17 23:29:25 +02003547static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07003548int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07003549{
3550 int i;
3551 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003552 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07003553
3554 /*
3555 * The per-page waitqueue mechanism uses hashed waitqueues
3556 * per zone.
3557 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07003558 zone->wait_table_hash_nr_entries =
3559 wait_table_hash_nr_entries(zone_size_pages);
3560 zone->wait_table_bits =
3561 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003562 alloc_size = zone->wait_table_hash_nr_entries
3563 * sizeof(wait_queue_head_t);
3564
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07003565 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07003566 zone->wait_table = (wait_queue_head_t *)
3567 alloc_bootmem_node(pgdat, alloc_size);
3568 } else {
3569 /*
3570 * This case means that a zone whose size was 0 gets new memory
3571 * via memory hot-add.
3572 * But it may be the case that a new node was hot-added. In
3573 * this case vmalloc() will not be able to use this new node's
3574 * memory - this wait_table must be initialized to use this new
3575 * node itself as well.
3576 * To use this new node's memory, further consideration will be
3577 * necessary.
3578 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07003579 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003580 }
3581 if (!zone->wait_table)
3582 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07003583
Yasunori Goto02b694d2006-06-23 02:03:08 -07003584 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07003585 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003586
3587 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003588}
3589
Shaohua Li112067f2009-09-21 17:01:16 -07003590static int __zone_pcp_update(void *data)
3591{
3592 struct zone *zone = data;
3593 int cpu;
3594 unsigned long batch = zone_batchsize(zone), flags;
3595
Thomas Gleixner2d30a1f2010-03-10 15:20:40 -08003596 for_each_possible_cpu(cpu) {
Shaohua Li112067f2009-09-21 17:01:16 -07003597 struct per_cpu_pageset *pset;
3598 struct per_cpu_pages *pcp;
3599
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003600 pset = per_cpu_ptr(zone->pageset, cpu);
Shaohua Li112067f2009-09-21 17:01:16 -07003601 pcp = &pset->pcp;
3602
3603 local_irq_save(flags);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003604 free_pcppages_bulk(zone, pcp->count, pcp);
Shaohua Li112067f2009-09-21 17:01:16 -07003605 setup_pageset(pset, batch);
3606 local_irq_restore(flags);
3607 }
3608 return 0;
3609}
3610
3611void zone_pcp_update(struct zone *zone)
3612{
3613 stop_machine(__zone_pcp_update, zone, NULL);
3614}
3615
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003616static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07003617{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003618 /*
3619 * per cpu subsystem is not up at this point. The following code
3620 * relies on the ability of the linker to provide the
3621 * offset of a (static) per cpu variable into the per cpu area.
3622 */
3623 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07003624
Anton Blanchardf5335c02006-03-25 03:06:49 -08003625 if (zone->present_pages)
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003626 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
3627 zone->name, zone->present_pages,
3628 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07003629}
3630
Yasunori Goto718127c2006-06-23 02:03:10 -07003631__meminit int init_currently_empty_zone(struct zone *zone,
3632 unsigned long zone_start_pfn,
Dave Hansena2f3aa02007-01-10 23:15:30 -08003633 unsigned long size,
3634 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07003635{
3636 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003637 int ret;
3638 ret = zone_wait_table_init(zone, size);
3639 if (ret)
3640 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07003641 pgdat->nr_zones = zone_idx(zone) + 1;
3642
Dave Hansened8ece22005-10-29 18:16:50 -07003643 zone->zone_start_pfn = zone_start_pfn;
3644
Mel Gorman708614e2008-07-23 21:26:51 -07003645 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3646 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3647 pgdat->node_id,
3648 (unsigned long)zone_idx(zone),
3649 zone_start_pfn, (zone_start_pfn + size));
3650
Andi Kleen1e548de2008-02-04 22:29:26 -08003651 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07003652
3653 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003654}
3655
Mel Gormanc7132162006-09-27 01:49:43 -07003656#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3657/*
3658 * Basic iterator support. Return the first range of PFNs for a node
3659 * Note: nid == MAX_NUMNODES returns first region regardless of node
3660 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003661static int __meminit first_active_region_index_in_nid(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003662{
3663 int i;
3664
3665 for (i = 0; i < nr_nodemap_entries; i++)
3666 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3667 return i;
3668
3669 return -1;
3670}
3671
3672/*
3673 * Basic iterator support. Return the next active range of PFNs for a node
Simon Arlott183ff222007-10-20 01:27:18 +02003674 * Note: nid == MAX_NUMNODES returns next region regardless of node
Mel Gormanc7132162006-09-27 01:49:43 -07003675 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003676static int __meminit next_active_region_index_in_nid(int index, int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003677{
3678 for (index = index + 1; index < nr_nodemap_entries; index++)
3679 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3680 return index;
3681
3682 return -1;
3683}
3684
3685#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3686/*
3687 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3688 * Architectures may implement their own version but if add_active_range()
3689 * was used and there are no special requirements, this is a convenient
3690 * alternative
3691 */
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003692int __meminit __early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07003693{
3694 int i;
3695
3696 for (i = 0; i < nr_nodemap_entries; i++) {
3697 unsigned long start_pfn = early_node_map[i].start_pfn;
3698 unsigned long end_pfn = early_node_map[i].end_pfn;
3699
3700 if (start_pfn <= pfn && pfn < end_pfn)
3701 return early_node_map[i].nid;
3702 }
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003703 /* This is a memory hole */
3704 return -1;
Mel Gormanc7132162006-09-27 01:49:43 -07003705}
3706#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3707
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003708int __meminit early_pfn_to_nid(unsigned long pfn)
3709{
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003710 int nid;
3711
3712 nid = __early_pfn_to_nid(pfn);
3713 if (nid >= 0)
3714 return nid;
3715 /* just returns 0 */
3716 return 0;
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003717}
3718
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003719#ifdef CONFIG_NODES_SPAN_OTHER_NODES
3720bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3721{
3722 int nid;
3723
3724 nid = __early_pfn_to_nid(pfn);
3725 if (nid >= 0 && nid != node)
3726 return false;
3727 return true;
3728}
3729#endif
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003730
Mel Gormanc7132162006-09-27 01:49:43 -07003731/* Basic iterator support to walk early_node_map[] */
3732#define for_each_active_range_index_in_nid(i, nid) \
3733 for (i = first_active_region_index_in_nid(nid); i != -1; \
3734 i = next_active_region_index_in_nid(i, nid))
3735
3736/**
3737 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003738 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3739 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07003740 *
3741 * If an architecture guarantees that all ranges registered with
3742 * add_active_ranges() contain no holes and may be freed, this
3743 * this function may be used instead of calling free_bootmem() manually.
3744 */
3745void __init free_bootmem_with_active_regions(int nid,
3746 unsigned long max_low_pfn)
3747{
3748 int i;
3749
3750 for_each_active_range_index_in_nid(i, nid) {
3751 unsigned long size_pages = 0;
3752 unsigned long end_pfn = early_node_map[i].end_pfn;
3753
3754 if (early_node_map[i].start_pfn >= max_low_pfn)
3755 continue;
3756
3757 if (end_pfn > max_low_pfn)
3758 end_pfn = max_low_pfn;
3759
3760 size_pages = end_pfn - early_node_map[i].start_pfn;
3761 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3762 PFN_PHYS(early_node_map[i].start_pfn),
3763 size_pages << PAGE_SHIFT);
3764 }
3765}
3766
Yinghai Luedbe7d22010-08-25 13:39:16 -07003767#ifdef CONFIG_HAVE_MEMBLOCK
Yinghai Lucc289892011-02-26 13:05:43 +01003768/*
3769 * Basic iterator support. Return the last range of PFNs for a node
3770 * Note: nid == MAX_NUMNODES returns last region regardless of node
3771 */
3772static int __meminit last_active_region_index_in_nid(int nid)
3773{
3774 int i;
3775
3776 for (i = nr_nodemap_entries - 1; i >= 0; i--)
3777 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3778 return i;
3779
3780 return -1;
3781}
3782
3783/*
3784 * Basic iterator support. Return the previous active range of PFNs for a node
3785 * Note: nid == MAX_NUMNODES returns next region regardless of node
3786 */
3787static int __meminit previous_active_region_index_in_nid(int index, int nid)
3788{
3789 for (index = index - 1; index >= 0; index--)
3790 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3791 return index;
3792
3793 return -1;
3794}
3795
3796#define for_each_active_range_index_in_nid_reverse(i, nid) \
3797 for (i = last_active_region_index_in_nid(nid); i != -1; \
3798 i = previous_active_region_index_in_nid(i, nid))
3799
Yinghai Luedbe7d22010-08-25 13:39:16 -07003800u64 __init find_memory_core_early(int nid, u64 size, u64 align,
3801 u64 goal, u64 limit)
3802{
3803 int i;
3804
3805 /* Need to go over early_node_map to find out good range for node */
Yinghai Lu1a4a6782010-12-17 16:59:07 -08003806 for_each_active_range_index_in_nid_reverse(i, nid) {
Yinghai Luedbe7d22010-08-25 13:39:16 -07003807 u64 addr;
3808 u64 ei_start, ei_last;
3809 u64 final_start, final_end;
3810
3811 ei_last = early_node_map[i].end_pfn;
3812 ei_last <<= PAGE_SHIFT;
3813 ei_start = early_node_map[i].start_pfn;
3814 ei_start <<= PAGE_SHIFT;
3815
3816 final_start = max(ei_start, goal);
3817 final_end = min(ei_last, limit);
3818
3819 if (final_start >= final_end)
3820 continue;
3821
3822 addr = memblock_find_in_range(final_start, final_end, size, align);
3823
3824 if (addr == MEMBLOCK_ERROR)
3825 continue;
3826
3827 return addr;
3828 }
3829
3830 return MEMBLOCK_ERROR;
3831}
3832#endif
3833
Yinghai Lu08677212010-02-10 01:20:20 -08003834int __init add_from_early_node_map(struct range *range, int az,
3835 int nr_range, int nid)
3836{
3837 int i;
3838 u64 start, end;
3839
3840 /* need to go over early_node_map to find out good range for node */
3841 for_each_active_range_index_in_nid(i, nid) {
3842 start = early_node_map[i].start_pfn;
3843 end = early_node_map[i].end_pfn;
3844 nr_range = add_range(range, az, nr_range, start, end);
3845 }
3846 return nr_range;
3847}
3848
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003849void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3850{
3851 int i;
Yinghai Lud52d53b2008-06-16 20:10:55 -07003852 int ret;
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003853
Yinghai Lud52d53b2008-06-16 20:10:55 -07003854 for_each_active_range_index_in_nid(i, nid) {
3855 ret = work_fn(early_node_map[i].start_pfn,
3856 early_node_map[i].end_pfn, data);
3857 if (ret)
3858 break;
3859 }
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003860}
Mel Gormanc7132162006-09-27 01:49:43 -07003861/**
3862 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003863 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07003864 *
3865 * If an architecture guarantees that all ranges registered with
3866 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003867 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07003868 */
3869void __init sparse_memory_present_with_active_regions(int nid)
3870{
3871 int i;
3872
3873 for_each_active_range_index_in_nid(i, nid)
3874 memory_present(early_node_map[i].nid,
3875 early_node_map[i].start_pfn,
3876 early_node_map[i].end_pfn);
3877}
3878
3879/**
3880 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003881 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3882 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3883 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07003884 *
3885 * It returns the start and end page frame of a node based on information
3886 * provided by an arch calling add_active_range(). If called for a node
3887 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003888 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07003889 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003890void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003891 unsigned long *start_pfn, unsigned long *end_pfn)
3892{
3893 int i;
3894 *start_pfn = -1UL;
3895 *end_pfn = 0;
3896
3897 for_each_active_range_index_in_nid(i, nid) {
3898 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3899 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3900 }
3901
Christoph Lameter633c0662007-10-16 01:25:37 -07003902 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07003903 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07003904}
3905
3906/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07003907 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3908 * assumption is made that zones within a node are ordered in monotonic
3909 * increasing memory addresses so that the "highest" populated zone is used
3910 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003911static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07003912{
3913 int zone_index;
3914 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3915 if (zone_index == ZONE_MOVABLE)
3916 continue;
3917
3918 if (arch_zone_highest_possible_pfn[zone_index] >
3919 arch_zone_lowest_possible_pfn[zone_index])
3920 break;
3921 }
3922
3923 VM_BUG_ON(zone_index == -1);
3924 movable_zone = zone_index;
3925}
3926
3927/*
3928 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3929 * because it is sized independant of architecture. Unlike the other zones,
3930 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3931 * in each node depending on the size of each node and how evenly kernelcore
3932 * is distributed. This helper function adjusts the zone ranges
3933 * provided by the architecture for a given node by using the end of the
3934 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3935 * zones within a node are in order of monotonic increases memory addresses
3936 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003937static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07003938 unsigned long zone_type,
3939 unsigned long node_start_pfn,
3940 unsigned long node_end_pfn,
3941 unsigned long *zone_start_pfn,
3942 unsigned long *zone_end_pfn)
3943{
3944 /* Only adjust if ZONE_MOVABLE is on this node */
3945 if (zone_movable_pfn[nid]) {
3946 /* Size ZONE_MOVABLE */
3947 if (zone_type == ZONE_MOVABLE) {
3948 *zone_start_pfn = zone_movable_pfn[nid];
3949 *zone_end_pfn = min(node_end_pfn,
3950 arch_zone_highest_possible_pfn[movable_zone]);
3951
3952 /* Adjust for ZONE_MOVABLE starting within this range */
3953 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3954 *zone_end_pfn > zone_movable_pfn[nid]) {
3955 *zone_end_pfn = zone_movable_pfn[nid];
3956
3957 /* Check if this whole range is within ZONE_MOVABLE */
3958 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3959 *zone_start_pfn = *zone_end_pfn;
3960 }
3961}
3962
3963/*
Mel Gormanc7132162006-09-27 01:49:43 -07003964 * Return the number of pages a zone spans in a node, including holes
3965 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3966 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003967static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003968 unsigned long zone_type,
3969 unsigned long *ignored)
3970{
3971 unsigned long node_start_pfn, node_end_pfn;
3972 unsigned long zone_start_pfn, zone_end_pfn;
3973
3974 /* Get the start and end of the node and zone */
3975 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3976 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3977 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07003978 adjust_zone_range_for_zone_movable(nid, zone_type,
3979 node_start_pfn, node_end_pfn,
3980 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003981
3982 /* Check that this node has pages within the zone's required range */
3983 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3984 return 0;
3985
3986 /* Move the zone boundaries inside the node if necessary */
3987 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3988 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3989
3990 /* Return the spanned pages */
3991 return zone_end_pfn - zone_start_pfn;
3992}
3993
3994/*
3995 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003996 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07003997 */
Yinghai Lu32996252009-12-15 17:59:02 -08003998unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003999 unsigned long range_start_pfn,
4000 unsigned long range_end_pfn)
4001{
4002 int i = 0;
4003 unsigned long prev_end_pfn = 0, hole_pages = 0;
4004 unsigned long start_pfn;
4005
4006 /* Find the end_pfn of the first active range of pfns in the node */
4007 i = first_active_region_index_in_nid(nid);
4008 if (i == -1)
4009 return 0;
4010
Mel Gormanb5445f92007-07-26 10:41:18 -07004011 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
4012
Mel Gorman9c7cd682006-09-27 01:49:58 -07004013 /* Account for ranges before physical memory on this node */
4014 if (early_node_map[i].start_pfn > range_start_pfn)
Mel Gormanb5445f92007-07-26 10:41:18 -07004015 hole_pages = prev_end_pfn - range_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004016
4017 /* Find all holes for the zone within the node */
4018 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
4019
4020 /* No need to continue if prev_end_pfn is outside the zone */
4021 if (prev_end_pfn >= range_end_pfn)
4022 break;
4023
4024 /* Make sure the end of the zone is not within the hole */
4025 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
4026 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
4027
4028 /* Update the hole size cound and move on */
4029 if (start_pfn > range_start_pfn) {
4030 BUG_ON(prev_end_pfn > start_pfn);
4031 hole_pages += start_pfn - prev_end_pfn;
4032 }
4033 prev_end_pfn = early_node_map[i].end_pfn;
4034 }
4035
Mel Gorman9c7cd682006-09-27 01:49:58 -07004036 /* Account for ranges past physical memory on this node */
4037 if (range_end_pfn > prev_end_pfn)
Mel Gorman0c6cb972006-10-28 10:38:59 -07004038 hole_pages += range_end_pfn -
Mel Gorman9c7cd682006-09-27 01:49:58 -07004039 max(range_start_pfn, prev_end_pfn);
4040
Mel Gormanc7132162006-09-27 01:49:43 -07004041 return hole_pages;
4042}
4043
4044/**
4045 * absent_pages_in_range - Return number of page frames in holes within a range
4046 * @start_pfn: The start PFN to start searching for holes
4047 * @end_pfn: The end PFN to stop searching for holes
4048 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004049 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07004050 */
4051unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4052 unsigned long end_pfn)
4053{
4054 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4055}
4056
4057/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004058static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004059 unsigned long zone_type,
4060 unsigned long *ignored)
4061{
Mel Gorman9c7cd682006-09-27 01:49:58 -07004062 unsigned long node_start_pfn, node_end_pfn;
4063 unsigned long zone_start_pfn, zone_end_pfn;
4064
4065 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
4066 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
4067 node_start_pfn);
4068 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
4069 node_end_pfn);
4070
Mel Gorman2a1e2742007-07-17 04:03:12 -07004071 adjust_zone_range_for_zone_movable(nid, zone_type,
4072 node_start_pfn, node_end_pfn,
4073 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004074 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004075}
Mel Gorman0e0b8642006-09-27 01:49:56 -07004076
Mel Gormanc7132162006-09-27 01:49:43 -07004077#else
Paul Mundt6ea6e682007-07-15 23:38:20 -07004078static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004079 unsigned long zone_type,
4080 unsigned long *zones_size)
4081{
4082 return zones_size[zone_type];
4083}
4084
Paul Mundt6ea6e682007-07-15 23:38:20 -07004085static inline 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 *zholes_size)
4088{
4089 if (!zholes_size)
4090 return 0;
4091
4092 return zholes_size[zone_type];
4093}
Mel Gorman0e0b8642006-09-27 01:49:56 -07004094
Mel Gormanc7132162006-09-27 01:49:43 -07004095#endif
4096
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004097static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Mel Gormanc7132162006-09-27 01:49:43 -07004098 unsigned long *zones_size, unsigned long *zholes_size)
4099{
4100 unsigned long realtotalpages, totalpages = 0;
4101 enum zone_type i;
4102
4103 for (i = 0; i < MAX_NR_ZONES; i++)
4104 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
4105 zones_size);
4106 pgdat->node_spanned_pages = totalpages;
4107
4108 realtotalpages = totalpages;
4109 for (i = 0; i < MAX_NR_ZONES; i++)
4110 realtotalpages -=
4111 zone_absent_pages_in_node(pgdat->node_id, i,
4112 zholes_size);
4113 pgdat->node_present_pages = realtotalpages;
4114 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4115 realtotalpages);
4116}
4117
Mel Gorman835c1342007-10-16 01:25:47 -07004118#ifndef CONFIG_SPARSEMEM
4119/*
4120 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07004121 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4122 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07004123 * round what is now in bits to nearest long in bits, then return it in
4124 * bytes.
4125 */
4126static unsigned long __init usemap_size(unsigned long zonesize)
4127{
4128 unsigned long usemapsize;
4129
Mel Gormand9c23402007-10-16 01:26:01 -07004130 usemapsize = roundup(zonesize, pageblock_nr_pages);
4131 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07004132 usemapsize *= NR_PAGEBLOCK_BITS;
4133 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4134
4135 return usemapsize / 8;
4136}
4137
4138static void __init setup_usemap(struct pglist_data *pgdat,
4139 struct zone *zone, unsigned long zonesize)
4140{
4141 unsigned long usemapsize = usemap_size(zonesize);
4142 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08004143 if (usemapsize)
Mel Gorman835c1342007-10-16 01:25:47 -07004144 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
Mel Gorman835c1342007-10-16 01:25:47 -07004145}
4146#else
Jesper Juhlfa9f90b2010-11-28 21:39:34 +01004147static inline void setup_usemap(struct pglist_data *pgdat,
Mel Gorman835c1342007-10-16 01:25:47 -07004148 struct zone *zone, unsigned long zonesize) {}
4149#endif /* CONFIG_SPARSEMEM */
4150
Mel Gormand9c23402007-10-16 01:26:01 -07004151#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08004152
4153/* Return a sensible default order for the pageblock size. */
4154static inline int pageblock_default_order(void)
4155{
4156 if (HPAGE_SHIFT > PAGE_SHIFT)
4157 return HUGETLB_PAGE_ORDER;
4158
4159 return MAX_ORDER-1;
4160}
4161
Mel Gormand9c23402007-10-16 01:26:01 -07004162/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4163static inline void __init set_pageblock_order(unsigned int order)
4164{
4165 /* Check that pageblock_nr_pages has not already been setup */
4166 if (pageblock_order)
4167 return;
4168
4169 /*
4170 * Assume the largest contiguous order of interest is a huge page.
4171 * This value may be variable depending on boot parameters on IA64
4172 */
4173 pageblock_order = order;
4174}
4175#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4176
Mel Gormanba72cb82007-11-28 16:21:13 -08004177/*
4178 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4179 * and pageblock_default_order() are unused as pageblock_order is set
4180 * at compile-time. See include/linux/pageblock-flags.h for the values of
4181 * pageblock_order based on the kernel config
4182 */
4183static inline int pageblock_default_order(unsigned int order)
4184{
4185 return MAX_ORDER-1;
4186}
Mel Gormand9c23402007-10-16 01:26:01 -07004187#define set_pageblock_order(x) do {} while (0)
4188
4189#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4190
Linus Torvalds1da177e2005-04-16 15:20:36 -07004191/*
4192 * Set up the zone data structures:
4193 * - mark all pages reserved
4194 * - mark all memory queues empty
4195 * - clear the memory bitmaps
4196 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08004197static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004198 unsigned long *zones_size, unsigned long *zholes_size)
4199{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07004200 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07004201 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004202 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07004203 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004204
Dave Hansen208d54e2005-10-29 18:16:52 -07004205 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004206 pgdat->nr_zones = 0;
4207 init_waitqueue_head(&pgdat->kswapd_wait);
4208 pgdat->kswapd_max_order = 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004209 pgdat_page_cgroup_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004210
4211 for (j = 0; j < MAX_NR_ZONES; j++) {
4212 struct zone *zone = pgdat->node_zones + j;
Mel Gorman0e0b8642006-09-27 01:49:56 -07004213 unsigned long size, realsize, memmap_pages;
Christoph Lameterb69408e2008-10-18 20:26:14 -07004214 enum lru_list l;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004215
Mel Gormanc7132162006-09-27 01:49:43 -07004216 size = zone_spanned_pages_in_node(nid, j, zones_size);
4217 realsize = size - zone_absent_pages_in_node(nid, j,
4218 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004219
Mel Gorman0e0b8642006-09-27 01:49:56 -07004220 /*
4221 * Adjust realsize so that it accounts for how much memory
4222 * is used by this zone for memmap. This affects the watermark
4223 * and per-cpu initialisations
4224 */
Johannes Weinerf7232152008-05-23 13:04:21 -07004225 memmap_pages =
4226 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
Mel Gorman0e0b8642006-09-27 01:49:56 -07004227 if (realsize >= memmap_pages) {
4228 realsize -= memmap_pages;
Yinghai Lu5594c8c2009-01-06 14:39:14 -08004229 if (memmap_pages)
4230 printk(KERN_DEBUG
4231 " %s zone: %lu pages used for memmap\n",
4232 zone_names[j], memmap_pages);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004233 } else
4234 printk(KERN_WARNING
4235 " %s zone: %lu pages exceeds realsize %lu\n",
4236 zone_names[j], memmap_pages, realsize);
4237
Christoph Lameter62672762007-02-10 01:43:07 -08004238 /* Account for reserved pages */
4239 if (j == 0 && realsize > dma_reserve) {
Mel Gorman0e0b8642006-09-27 01:49:56 -07004240 realsize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07004241 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08004242 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004243 }
4244
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004245 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004246 nr_kernel_pages += realsize;
4247 nr_all_pages += realsize;
4248
4249 zone->spanned_pages = size;
4250 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07004251#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07004252 zone->node = nid;
Christoph Lameter8417bba2006-09-25 23:31:51 -07004253 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07004254 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07004255 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07004256#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004257 zone->name = zone_names[j];
4258 spin_lock_init(&zone->lock);
4259 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07004260 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004261 zone->zone_pgdat = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004262
Dave Hansened8ece22005-10-29 18:16:50 -07004263 zone_pcp_init(zone);
Christoph Lameterb69408e2008-10-18 20:26:14 -07004264 for_each_lru(l) {
4265 INIT_LIST_HEAD(&zone->lru[l].list);
Wu Fengguangf8629632009-09-21 17:03:11 -07004266 zone->reclaim_stat.nr_saved_scan[l] = 0;
Christoph Lameterb69408e2008-10-18 20:26:14 -07004267 }
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08004268 zone->reclaim_stat.recent_rotated[0] = 0;
4269 zone->reclaim_stat.recent_rotated[1] = 0;
4270 zone->reclaim_stat.recent_scanned[0] = 0;
4271 zone->reclaim_stat.recent_scanned[1] = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07004272 zap_zone_vm_stats(zone);
David Rientjese815af92007-10-16 23:25:54 -07004273 zone->flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004274 if (!size)
4275 continue;
4276
Mel Gormanba72cb82007-11-28 16:21:13 -08004277 set_pageblock_order(pageblock_default_order());
Mel Gorman835c1342007-10-16 01:25:47 -07004278 setup_usemap(pgdat, zone, size);
Dave Hansena2f3aa02007-01-10 23:15:30 -08004279 ret = init_currently_empty_zone(zone, zone_start_pfn,
4280 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07004281 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07004282 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004283 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004284 }
4285}
4286
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004287static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004288{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004289 /* Skip empty nodes */
4290 if (!pgdat->node_spanned_pages)
4291 return;
4292
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004293#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07004294 /* ia64 gets its own node_mem_map, before this, without bootmem */
4295 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07004296 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004297 struct page *map;
4298
Bob Piccoe984bb42006-05-20 15:00:31 -07004299 /*
4300 * The zone's endpoints aren't required to be MAX_ORDER
4301 * aligned but the node_mem_map endpoints must be in order
4302 * for the buddy allocator to function correctly.
4303 */
4304 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
4305 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
4306 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4307 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07004308 map = alloc_remap(pgdat->node_id, size);
4309 if (!map)
4310 map = alloc_bootmem_node(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07004311 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004312 }
Roman Zippel12d810c2007-05-31 00:40:54 -07004313#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07004314 /*
4315 * With no DISCONTIG, the global mem_map is just set as node 0's
4316 */
Mel Gormanc7132162006-09-27 01:49:43 -07004317 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004318 mem_map = NODE_DATA(0)->node_mem_map;
Mel Gormanc7132162006-09-27 01:49:43 -07004319#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
4320 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08004321 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Mel Gormanc7132162006-09-27 01:49:43 -07004322#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4323 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004324#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004325#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004326}
4327
Johannes Weiner9109fb72008-07-23 21:27:20 -07004328void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4329 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004330{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004331 pg_data_t *pgdat = NODE_DATA(nid);
4332
Linus Torvalds1da177e2005-04-16 15:20:36 -07004333 pgdat->node_id = nid;
4334 pgdat->node_start_pfn = node_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004335 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004336
4337 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07004338#ifdef CONFIG_FLAT_NODE_MEM_MAP
4339 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4340 nid, (unsigned long)pgdat,
4341 (unsigned long)pgdat->node_mem_map);
4342#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004343
4344 free_area_init_core(pgdat, zones_size, zholes_size);
4345}
4346
Mel Gormanc7132162006-09-27 01:49:43 -07004347#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07004348
4349#if MAX_NUMNODES > 1
4350/*
4351 * Figure out the number of possible node ids.
4352 */
4353static void __init setup_nr_node_ids(void)
4354{
4355 unsigned int node;
4356 unsigned int highest = 0;
4357
4358 for_each_node_mask(node, node_possible_map)
4359 highest = node;
4360 nr_node_ids = highest + 1;
4361}
4362#else
4363static inline void setup_nr_node_ids(void)
4364{
4365}
4366#endif
4367
Mel Gormanc7132162006-09-27 01:49:43 -07004368/**
4369 * add_active_range - Register a range of PFNs backed by physical memory
4370 * @nid: The node ID the range resides on
4371 * @start_pfn: The start PFN of the available physical memory
4372 * @end_pfn: The end PFN of the available physical memory
4373 *
4374 * These ranges are stored in an early_node_map[] and later used by
4375 * free_area_init_nodes() to calculate zone sizes and holes. If the
4376 * range spans a memory hole, it is up to the architecture to ensure
4377 * the memory is not freed by the bootmem allocator. If possible
4378 * the range being registered will be merged with existing ranges.
4379 */
4380void __init add_active_range(unsigned int nid, unsigned long start_pfn,
4381 unsigned long end_pfn)
4382{
4383 int i;
4384
Mel Gorman6b74ab92008-07-23 21:26:49 -07004385 mminit_dprintk(MMINIT_TRACE, "memory_register",
4386 "Entering add_active_range(%d, %#lx, %#lx) "
4387 "%d entries of %d used\n",
4388 nid, start_pfn, end_pfn,
4389 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
Mel Gormanc7132162006-09-27 01:49:43 -07004390
Mel Gorman2dbb51c2008-07-23 21:26:52 -07004391 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
4392
Mel Gormanc7132162006-09-27 01:49:43 -07004393 /* Merge with existing active regions if possible */
4394 for (i = 0; i < nr_nodemap_entries; i++) {
4395 if (early_node_map[i].nid != nid)
4396 continue;
4397
4398 /* Skip if an existing region covers this new one */
4399 if (start_pfn >= early_node_map[i].start_pfn &&
4400 end_pfn <= early_node_map[i].end_pfn)
4401 return;
4402
4403 /* Merge forward if suitable */
4404 if (start_pfn <= early_node_map[i].end_pfn &&
4405 end_pfn > early_node_map[i].end_pfn) {
4406 early_node_map[i].end_pfn = end_pfn;
4407 return;
4408 }
4409
4410 /* Merge backward if suitable */
Kazuhisa Ichikawad2dbe082010-01-15 17:01:20 -08004411 if (start_pfn < early_node_map[i].start_pfn &&
Mel Gormanc7132162006-09-27 01:49:43 -07004412 end_pfn >= early_node_map[i].start_pfn) {
4413 early_node_map[i].start_pfn = start_pfn;
4414 return;
4415 }
4416 }
4417
4418 /* Check that early_node_map is large enough */
4419 if (i >= MAX_ACTIVE_REGIONS) {
4420 printk(KERN_CRIT "More than %d memory regions, truncating\n",
4421 MAX_ACTIVE_REGIONS);
4422 return;
4423 }
4424
4425 early_node_map[i].nid = nid;
4426 early_node_map[i].start_pfn = start_pfn;
4427 early_node_map[i].end_pfn = end_pfn;
4428 nr_nodemap_entries = i + 1;
4429}
4430
4431/**
Yinghai Lucc1050b2008-06-13 19:08:52 -07004432 * remove_active_range - Shrink an existing registered range of PFNs
Mel Gormanc7132162006-09-27 01:49:43 -07004433 * @nid: The node id the range is on that should be shrunk
Yinghai Lucc1050b2008-06-13 19:08:52 -07004434 * @start_pfn: The new PFN of the range
4435 * @end_pfn: The new PFN of the range
Mel Gormanc7132162006-09-27 01:49:43 -07004436 *
4437 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004438 * The map is kept near the end physical page range that has already been
4439 * registered. This function allows an arch to shrink an existing registered
4440 * range.
Mel Gormanc7132162006-09-27 01:49:43 -07004441 */
Yinghai Lucc1050b2008-06-13 19:08:52 -07004442void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
4443 unsigned long end_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004444{
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004445 int i, j;
4446 int removed = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07004447
Yinghai Lucc1050b2008-06-13 19:08:52 -07004448 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
4449 nid, start_pfn, end_pfn);
4450
Mel Gormanc7132162006-09-27 01:49:43 -07004451 /* Find the old active region end and shrink */
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004452 for_each_active_range_index_in_nid(i, nid) {
Yinghai Lucc1050b2008-06-13 19:08:52 -07004453 if (early_node_map[i].start_pfn >= start_pfn &&
4454 early_node_map[i].end_pfn <= end_pfn) {
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004455 /* clear it */
Yinghai Lucc1050b2008-06-13 19:08:52 -07004456 early_node_map[i].start_pfn = 0;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004457 early_node_map[i].end_pfn = 0;
4458 removed = 1;
4459 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004460 }
Yinghai Lucc1050b2008-06-13 19:08:52 -07004461 if (early_node_map[i].start_pfn < start_pfn &&
4462 early_node_map[i].end_pfn > start_pfn) {
4463 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
4464 early_node_map[i].end_pfn = start_pfn;
4465 if (temp_end_pfn > end_pfn)
4466 add_active_range(nid, end_pfn, temp_end_pfn);
4467 continue;
4468 }
4469 if (early_node_map[i].start_pfn >= start_pfn &&
4470 early_node_map[i].end_pfn > end_pfn &&
4471 early_node_map[i].start_pfn < end_pfn) {
4472 early_node_map[i].start_pfn = end_pfn;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004473 continue;
4474 }
4475 }
4476
4477 if (!removed)
4478 return;
4479
4480 /* remove the blank ones */
4481 for (i = nr_nodemap_entries - 1; i > 0; i--) {
4482 if (early_node_map[i].nid != nid)
4483 continue;
4484 if (early_node_map[i].end_pfn)
4485 continue;
4486 /* we found it, get rid of it */
4487 for (j = i; j < nr_nodemap_entries - 1; j++)
4488 memcpy(&early_node_map[j], &early_node_map[j+1],
4489 sizeof(early_node_map[j]));
4490 j = nr_nodemap_entries - 1;
4491 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
4492 nr_nodemap_entries--;
4493 }
Mel Gormanc7132162006-09-27 01:49:43 -07004494}
4495
4496/**
4497 * remove_all_active_ranges - Remove all currently registered regions
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004498 *
Mel Gormanc7132162006-09-27 01:49:43 -07004499 * During discovery, it may be found that a table like SRAT is invalid
4500 * and an alternative discovery method must be used. This function removes
4501 * all currently registered regions.
4502 */
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004503void __init remove_all_active_ranges(void)
Mel Gormanc7132162006-09-27 01:49:43 -07004504{
4505 memset(early_node_map, 0, sizeof(early_node_map));
4506 nr_nodemap_entries = 0;
4507}
4508
4509/* Compare two active node_active_regions */
4510static int __init cmp_node_active_region(const void *a, const void *b)
4511{
4512 struct node_active_region *arange = (struct node_active_region *)a;
4513 struct node_active_region *brange = (struct node_active_region *)b;
4514
4515 /* Done this way to avoid overflows */
4516 if (arange->start_pfn > brange->start_pfn)
4517 return 1;
4518 if (arange->start_pfn < brange->start_pfn)
4519 return -1;
4520
4521 return 0;
4522}
4523
4524/* sort the node_map by start_pfn */
Yinghai Lu32996252009-12-15 17:59:02 -08004525void __init sort_node_map(void)
Mel Gormanc7132162006-09-27 01:49:43 -07004526{
4527 sort(early_node_map, (size_t)nr_nodemap_entries,
4528 sizeof(struct node_active_region),
4529 cmp_node_active_region, NULL);
4530}
4531
Mel Gormana6af2bc2007-02-10 01:42:57 -08004532/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004533static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07004534{
4535 int i;
Mel Gormana6af2bc2007-02-10 01:42:57 -08004536 unsigned long min_pfn = ULONG_MAX;
Mel Gorman1abbfb42006-11-23 12:01:41 +00004537
Mel Gormanc7132162006-09-27 01:49:43 -07004538 /* Assuming a sorted map, the first range found has the starting pfn */
4539 for_each_active_range_index_in_nid(i, nid)
Mel Gormana6af2bc2007-02-10 01:42:57 -08004540 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004541
Mel Gormana6af2bc2007-02-10 01:42:57 -08004542 if (min_pfn == ULONG_MAX) {
4543 printk(KERN_WARNING
Paul Jackson2bc0d262008-06-22 07:22:17 -07004544 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08004545 return 0;
4546 }
4547
4548 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004549}
4550
4551/**
4552 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4553 *
4554 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004555 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07004556 */
4557unsigned long __init find_min_pfn_with_active_regions(void)
4558{
4559 return find_min_pfn_for_node(MAX_NUMNODES);
4560}
4561
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004562/*
4563 * early_calculate_totalpages()
4564 * Sum pages in active regions for movable zone.
4565 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4566 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07004567static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07004568{
4569 int i;
4570 unsigned long totalpages = 0;
4571
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004572 for (i = 0; i < nr_nodemap_entries; i++) {
4573 unsigned long pages = early_node_map[i].end_pfn -
Mel Gorman7e63efe2007-07-17 04:03:15 -07004574 early_node_map[i].start_pfn;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004575 totalpages += pages;
4576 if (pages)
4577 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
4578 }
4579 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07004580}
4581
Mel Gorman2a1e2742007-07-17 04:03:12 -07004582/*
4583 * Find the PFN the Movable zone begins in each node. Kernel memory
4584 * is spread evenly between nodes as long as the nodes have enough
4585 * memory. When they don't, some nodes will have more kernelcore than
4586 * others
4587 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004588static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004589{
4590 int i, nid;
4591 unsigned long usable_startpfn;
4592 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07004593 /* save the state before borrow the nodemask */
4594 nodemask_t saved_node_state = node_states[N_HIGH_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004595 unsigned long totalpages = early_calculate_totalpages();
4596 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004597
Mel Gorman7e63efe2007-07-17 04:03:15 -07004598 /*
4599 * If movablecore was specified, calculate what size of
4600 * kernelcore that corresponds so that memory usable for
4601 * any allocation type is evenly spread. If both kernelcore
4602 * and movablecore are specified, then the value of kernelcore
4603 * will be used for required_kernelcore if it's greater than
4604 * what movablecore would have allowed.
4605 */
4606 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07004607 unsigned long corepages;
4608
4609 /*
4610 * Round-up so that ZONE_MOVABLE is at least as large as what
4611 * was requested by the user
4612 */
4613 required_movablecore =
4614 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
4615 corepages = totalpages - required_movablecore;
4616
4617 required_kernelcore = max(required_kernelcore, corepages);
4618 }
4619
Mel Gorman2a1e2742007-07-17 04:03:12 -07004620 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4621 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07004622 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004623
4624 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4625 find_usable_zone_for_movable();
4626 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4627
4628restart:
4629 /* Spread kernelcore memory as evenly as possible throughout nodes */
4630 kernelcore_node = required_kernelcore / usable_nodes;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004631 for_each_node_state(nid, N_HIGH_MEMORY) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004632 /*
4633 * Recalculate kernelcore_node if the division per node
4634 * now exceeds what is necessary to satisfy the requested
4635 * amount of memory for the kernel
4636 */
4637 if (required_kernelcore < kernelcore_node)
4638 kernelcore_node = required_kernelcore / usable_nodes;
4639
4640 /*
4641 * As the map is walked, we track how much memory is usable
4642 * by the kernel using kernelcore_remaining. When it is
4643 * 0, the rest of the node is usable by ZONE_MOVABLE
4644 */
4645 kernelcore_remaining = kernelcore_node;
4646
4647 /* Go through each range of PFNs within this node */
4648 for_each_active_range_index_in_nid(i, nid) {
4649 unsigned long start_pfn, end_pfn;
4650 unsigned long size_pages;
4651
4652 start_pfn = max(early_node_map[i].start_pfn,
4653 zone_movable_pfn[nid]);
4654 end_pfn = early_node_map[i].end_pfn;
4655 if (start_pfn >= end_pfn)
4656 continue;
4657
4658 /* Account for what is only usable for kernelcore */
4659 if (start_pfn < usable_startpfn) {
4660 unsigned long kernel_pages;
4661 kernel_pages = min(end_pfn, usable_startpfn)
4662 - start_pfn;
4663
4664 kernelcore_remaining -= min(kernel_pages,
4665 kernelcore_remaining);
4666 required_kernelcore -= min(kernel_pages,
4667 required_kernelcore);
4668
4669 /* Continue if range is now fully accounted */
4670 if (end_pfn <= usable_startpfn) {
4671
4672 /*
4673 * Push zone_movable_pfn to the end so
4674 * that if we have to rebalance
4675 * kernelcore across nodes, we will
4676 * not double account here
4677 */
4678 zone_movable_pfn[nid] = end_pfn;
4679 continue;
4680 }
4681 start_pfn = usable_startpfn;
4682 }
4683
4684 /*
4685 * The usable PFN range for ZONE_MOVABLE is from
4686 * start_pfn->end_pfn. Calculate size_pages as the
4687 * number of pages used as kernelcore
4688 */
4689 size_pages = end_pfn - start_pfn;
4690 if (size_pages > kernelcore_remaining)
4691 size_pages = kernelcore_remaining;
4692 zone_movable_pfn[nid] = start_pfn + size_pages;
4693
4694 /*
4695 * Some kernelcore has been met, update counts and
4696 * break if the kernelcore for this node has been
4697 * satisified
4698 */
4699 required_kernelcore -= min(required_kernelcore,
4700 size_pages);
4701 kernelcore_remaining -= size_pages;
4702 if (!kernelcore_remaining)
4703 break;
4704 }
4705 }
4706
4707 /*
4708 * If there is still required_kernelcore, we do another pass with one
4709 * less node in the count. This will push zone_movable_pfn[nid] further
4710 * along on the nodes that still have memory until kernelcore is
4711 * satisified
4712 */
4713 usable_nodes--;
4714 if (usable_nodes && required_kernelcore > usable_nodes)
4715 goto restart;
4716
4717 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4718 for (nid = 0; nid < MAX_NUMNODES; nid++)
4719 zone_movable_pfn[nid] =
4720 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07004721
4722out:
4723 /* restore the node_state */
4724 node_states[N_HIGH_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004725}
4726
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004727/* Any regular memory on that node ? */
4728static void check_for_regular_memory(pg_data_t *pgdat)
4729{
4730#ifdef CONFIG_HIGHMEM
4731 enum zone_type zone_type;
4732
4733 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4734 struct zone *zone = &pgdat->node_zones[zone_type];
4735 if (zone->present_pages)
4736 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4737 }
4738#endif
4739}
4740
Mel Gormanc7132162006-09-27 01:49:43 -07004741/**
4742 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004743 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07004744 *
4745 * This will call free_area_init_node() for each active node in the system.
4746 * Using the page ranges provided by add_active_range(), the size of each
4747 * zone in each node and their holes is calculated. If the maximum PFN
4748 * between two adjacent zones match, it is assumed that the zone is empty.
4749 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4750 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4751 * starts where the previous one ended. For example, ZONE_DMA32 starts
4752 * at arch_max_dma_pfn.
4753 */
4754void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4755{
4756 unsigned long nid;
Andrew Mortondb991002008-10-15 22:01:07 -07004757 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004758
Mel Gormana6af2bc2007-02-10 01:42:57 -08004759 /* Sort early_node_map as initialisation assumes it is sorted */
4760 sort_node_map();
4761
Mel Gormanc7132162006-09-27 01:49:43 -07004762 /* Record where the zone boundaries are */
4763 memset(arch_zone_lowest_possible_pfn, 0,
4764 sizeof(arch_zone_lowest_possible_pfn));
4765 memset(arch_zone_highest_possible_pfn, 0,
4766 sizeof(arch_zone_highest_possible_pfn));
4767 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4768 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4769 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004770 if (i == ZONE_MOVABLE)
4771 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004772 arch_zone_lowest_possible_pfn[i] =
4773 arch_zone_highest_possible_pfn[i-1];
4774 arch_zone_highest_possible_pfn[i] =
4775 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4776 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07004777 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4778 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4779
4780 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4781 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4782 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004783
Mel Gormanc7132162006-09-27 01:49:43 -07004784 /* Print out the zone ranges */
4785 printk("Zone PFN ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07004786 for (i = 0; i < MAX_NR_ZONES; i++) {
4787 if (i == ZONE_MOVABLE)
4788 continue;
David Rientjes72f0ba02010-03-05 13:42:14 -08004789 printk(" %-8s ", zone_names[i]);
4790 if (arch_zone_lowest_possible_pfn[i] ==
4791 arch_zone_highest_possible_pfn[i])
4792 printk("empty\n");
4793 else
4794 printk("%0#10lx -> %0#10lx\n",
Mel Gormanc7132162006-09-27 01:49:43 -07004795 arch_zone_lowest_possible_pfn[i],
4796 arch_zone_highest_possible_pfn[i]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004797 }
4798
4799 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4800 printk("Movable zone start PFN for each node\n");
4801 for (i = 0; i < MAX_NUMNODES; i++) {
4802 if (zone_movable_pfn[i])
4803 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4804 }
Mel Gormanc7132162006-09-27 01:49:43 -07004805
4806 /* Print out the early_node_map[] */
4807 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4808 for (i = 0; i < nr_nodemap_entries; i++)
Paul Jackson5dab8ec2008-06-25 05:44:40 -07004809 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004810 early_node_map[i].start_pfn,
4811 early_node_map[i].end_pfn);
4812
4813 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07004814 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08004815 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07004816 for_each_online_node(nid) {
4817 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07004818 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07004819 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004820
4821 /* Any memory on that node */
4822 if (pgdat->node_present_pages)
4823 node_set_state(nid, N_HIGH_MEMORY);
4824 check_for_regular_memory(pgdat);
Mel Gormanc7132162006-09-27 01:49:43 -07004825 }
4826}
Mel Gorman2a1e2742007-07-17 04:03:12 -07004827
Mel Gorman7e63efe2007-07-17 04:03:15 -07004828static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004829{
4830 unsigned long long coremem;
4831 if (!p)
4832 return -EINVAL;
4833
4834 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004835 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004836
Mel Gorman7e63efe2007-07-17 04:03:15 -07004837 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07004838 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4839
4840 return 0;
4841}
Mel Gormaned7ed362007-07-17 04:03:14 -07004842
Mel Gorman7e63efe2007-07-17 04:03:15 -07004843/*
4844 * kernelcore=size sets the amount of memory for use for allocations that
4845 * cannot be reclaimed or migrated.
4846 */
4847static int __init cmdline_parse_kernelcore(char *p)
4848{
4849 return cmdline_parse_core(p, &required_kernelcore);
4850}
4851
4852/*
4853 * movablecore=size sets the amount of memory for use for allocations that
4854 * can be reclaimed or migrated.
4855 */
4856static int __init cmdline_parse_movablecore(char *p)
4857{
4858 return cmdline_parse_core(p, &required_movablecore);
4859}
4860
Mel Gormaned7ed362007-07-17 04:03:14 -07004861early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004862early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07004863
Mel Gormanc7132162006-09-27 01:49:43 -07004864#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4865
Mel Gorman0e0b8642006-09-27 01:49:56 -07004866/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004867 * set_dma_reserve - set the specified number of pages reserved in the first zone
4868 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07004869 *
4870 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4871 * In the DMA zone, a significant percentage may be consumed by kernel image
4872 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004873 * function may optionally be used to account for unfreeable pages in the
4874 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4875 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07004876 */
4877void __init set_dma_reserve(unsigned long new_dma_reserve)
4878{
4879 dma_reserve = new_dma_reserve;
4880}
4881
Linus Torvalds1da177e2005-04-16 15:20:36 -07004882void __init free_area_init(unsigned long *zones_size)
4883{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004884 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004885 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4886}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004887
Linus Torvalds1da177e2005-04-16 15:20:36 -07004888static int page_alloc_cpu_notify(struct notifier_block *self,
4889 unsigned long action, void *hcpu)
4890{
4891 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004892
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07004893 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004894 drain_pages(cpu);
4895
4896 /*
4897 * Spill the event counters of the dead processor
4898 * into the current processors event counters.
4899 * This artificially elevates the count of the current
4900 * processor.
4901 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07004902 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004903
4904 /*
4905 * Zero the differential counters of the dead processor
4906 * so that the vm statistics are consistent.
4907 *
4908 * This is only okay since the processor is dead and cannot
4909 * race with what we are doing.
4910 */
Christoph Lameter2244b952006-06-30 01:55:33 -07004911 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004912 }
4913 return NOTIFY_OK;
4914}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004915
4916void __init page_alloc_init(void)
4917{
4918 hotcpu_notifier(page_alloc_cpu_notify, 0);
4919}
4920
4921/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004922 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4923 * or min_free_kbytes changes.
4924 */
4925static void calculate_totalreserve_pages(void)
4926{
4927 struct pglist_data *pgdat;
4928 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004929 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004930
4931 for_each_online_pgdat(pgdat) {
4932 for (i = 0; i < MAX_NR_ZONES; i++) {
4933 struct zone *zone = pgdat->node_zones + i;
4934 unsigned long max = 0;
4935
4936 /* Find valid and maximum lowmem_reserve in the zone */
4937 for (j = i; j < MAX_NR_ZONES; j++) {
4938 if (zone->lowmem_reserve[j] > max)
4939 max = zone->lowmem_reserve[j];
4940 }
4941
Mel Gorman41858962009-06-16 15:32:12 -07004942 /* we treat the high watermark as reserved pages. */
4943 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004944
4945 if (max > zone->present_pages)
4946 max = zone->present_pages;
4947 reserve_pages += max;
4948 }
4949 }
4950 totalreserve_pages = reserve_pages;
4951}
4952
4953/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07004954 * setup_per_zone_lowmem_reserve - called whenever
4955 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4956 * has a correct pages reserved value, so an adequate number of
4957 * pages are left in the zone after a successful __alloc_pages().
4958 */
4959static void setup_per_zone_lowmem_reserve(void)
4960{
4961 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004962 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004963
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08004964 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004965 for (j = 0; j < MAX_NR_ZONES; j++) {
4966 struct zone *zone = pgdat->node_zones + j;
4967 unsigned long present_pages = zone->present_pages;
4968
4969 zone->lowmem_reserve[j] = 0;
4970
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004971 idx = j;
4972 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004973 struct zone *lower_zone;
4974
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004975 idx--;
4976
Linus Torvalds1da177e2005-04-16 15:20:36 -07004977 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4978 sysctl_lowmem_reserve_ratio[idx] = 1;
4979
4980 lower_zone = pgdat->node_zones + idx;
4981 lower_zone->lowmem_reserve[j] = present_pages /
4982 sysctl_lowmem_reserve_ratio[idx];
4983 present_pages += lower_zone->present_pages;
4984 }
4985 }
4986 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004987
4988 /* update totalreserve_pages */
4989 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004990}
4991
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004992/**
Minchan Kimbc75d332009-06-16 15:32:48 -07004993 * setup_per_zone_wmarks - called when min_free_kbytes changes
Minchan Kimbce73942009-06-16 15:32:50 -07004994 * or when memory is hot-{added|removed}
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004995 *
Minchan Kimbc75d332009-06-16 15:32:48 -07004996 * Ensures that the watermark[min,low,high] values for each zone are set
4997 * correctly with respect to min_free_kbytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004998 */
Minchan Kimbc75d332009-06-16 15:32:48 -07004999void setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005000{
5001 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
5002 unsigned long lowmem_pages = 0;
5003 struct zone *zone;
5004 unsigned long flags;
5005
5006 /* Calculate total number of !ZONE_HIGHMEM pages */
5007 for_each_zone(zone) {
5008 if (!is_highmem(zone))
5009 lowmem_pages += zone->present_pages;
5010 }
5011
5012 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07005013 u64 tmp;
5014
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005015 spin_lock_irqsave(&zone->lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07005016 tmp = (u64)pages_min * zone->present_pages;
5017 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005018 if (is_highmem(zone)) {
5019 /*
Nick Piggin669ed172005-11-13 16:06:45 -08005020 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5021 * need highmem pages, so cap pages_min to a small
5022 * value here.
5023 *
Mel Gorman41858962009-06-16 15:32:12 -07005024 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Nick Piggin669ed172005-11-13 16:06:45 -08005025 * deltas controls asynch page reclaim, and so should
5026 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005027 */
5028 int min_pages;
5029
5030 min_pages = zone->present_pages / 1024;
5031 if (min_pages < SWAP_CLUSTER_MAX)
5032 min_pages = SWAP_CLUSTER_MAX;
5033 if (min_pages > 128)
5034 min_pages = 128;
Mel Gorman41858962009-06-16 15:32:12 -07005035 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005036 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08005037 /*
5038 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07005039 * proportionate to the zone's size.
5040 */
Mel Gorman41858962009-06-16 15:32:12 -07005041 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005042 }
5043
Mel Gorman41858962009-06-16 15:32:12 -07005044 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
5045 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Mel Gorman56fd56b2007-10-16 01:25:58 -07005046 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005047 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005048 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005049
5050 /* update totalreserve_pages */
5051 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005052}
5053
Randy Dunlap55a44622009-09-21 17:01:20 -07005054/*
Rik van Riel556adec2008-10-18 20:26:34 -07005055 * The inactive anon list should be small enough that the VM never has to
5056 * do too much work, but large enough that each inactive page has a chance
5057 * to be referenced again before it is swapped out.
5058 *
5059 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5060 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5061 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5062 * the anonymous pages are kept on the inactive list.
5063 *
5064 * total target max
5065 * memory ratio inactive anon
5066 * -------------------------------------
5067 * 10MB 1 5MB
5068 * 100MB 1 50MB
5069 * 1GB 3 250MB
5070 * 10GB 10 0.9GB
5071 * 100GB 31 3GB
5072 * 1TB 101 10GB
5073 * 10TB 320 32GB
5074 */
Minchan Kim96cb4df2009-06-16 15:32:49 -07005075void calculate_zone_inactive_ratio(struct zone *zone)
5076{
5077 unsigned int gb, ratio;
5078
5079 /* Zone size in gigabytes */
5080 gb = zone->present_pages >> (30 - PAGE_SHIFT);
5081 if (gb)
5082 ratio = int_sqrt(10 * gb);
5083 else
5084 ratio = 1;
5085
5086 zone->inactive_ratio = ratio;
5087}
5088
Cyrill Gorcunove9bb35d2009-06-16 15:32:32 -07005089static void __init setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07005090{
5091 struct zone *zone;
5092
Minchan Kim96cb4df2009-06-16 15:32:49 -07005093 for_each_zone(zone)
5094 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07005095}
5096
Linus Torvalds1da177e2005-04-16 15:20:36 -07005097/*
5098 * Initialise min_free_kbytes.
5099 *
5100 * For small machines we want it small (128k min). For large machines
5101 * we want it large (64MB max). But it is not linear, because network
5102 * bandwidth does not increase linearly with machine size. We use
5103 *
5104 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5105 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5106 *
5107 * which yields
5108 *
5109 * 16MB: 512k
5110 * 32MB: 724k
5111 * 64MB: 1024k
5112 * 128MB: 1448k
5113 * 256MB: 2048k
5114 * 512MB: 2896k
5115 * 1024MB: 4096k
5116 * 2048MB: 5792k
5117 * 4096MB: 8192k
5118 * 8192MB: 11584k
5119 * 16384MB: 16384k
5120 */
Minchan Kimbc75d332009-06-16 15:32:48 -07005121static int __init init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005122{
5123 unsigned long lowmem_kbytes;
5124
5125 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
5126
5127 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
5128 if (min_free_kbytes < 128)
5129 min_free_kbytes = 128;
5130 if (min_free_kbytes > 65536)
5131 min_free_kbytes = 65536;
Minchan Kimbc75d332009-06-16 15:32:48 -07005132 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005133 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07005134 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005135 return 0;
5136}
Minchan Kimbc75d332009-06-16 15:32:48 -07005137module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005138
5139/*
5140 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5141 * that we can call two helper functions whenever min_free_kbytes
5142 * changes.
5143 */
5144int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005145 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005146{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005147 proc_dointvec(table, write, buffer, length, ppos);
Mel Gorman3b1d92c2007-05-06 14:49:30 -07005148 if (write)
Minchan Kimbc75d332009-06-16 15:32:48 -07005149 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005150 return 0;
5151}
5152
Christoph Lameter96146342006-07-03 00:24:13 -07005153#ifdef CONFIG_NUMA
5154int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005155 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07005156{
5157 struct zone *zone;
5158 int rc;
5159
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005160 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07005161 if (rc)
5162 return rc;
5163
5164 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07005165 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07005166 sysctl_min_unmapped_ratio) / 100;
5167 return 0;
5168}
Christoph Lameter0ff38492006-09-25 23:31:52 -07005169
5170int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005171 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07005172{
5173 struct zone *zone;
5174 int rc;
5175
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005176 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07005177 if (rc)
5178 return rc;
5179
5180 for_each_zone(zone)
5181 zone->min_slab_pages = (zone->present_pages *
5182 sysctl_min_slab_ratio) / 100;
5183 return 0;
5184}
Christoph Lameter96146342006-07-03 00:24:13 -07005185#endif
5186
Linus Torvalds1da177e2005-04-16 15:20:36 -07005187/*
5188 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5189 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5190 * whenever sysctl_lowmem_reserve_ratio changes.
5191 *
5192 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07005193 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07005194 * if in function of the boot time zone sizes.
5195 */
5196int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005197 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005198{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005199 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005200 setup_per_zone_lowmem_reserve();
5201 return 0;
5202}
5203
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005204/*
5205 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5206 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5207 * can have before it gets flushed back to buddy allocator.
5208 */
5209
5210int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005211 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005212{
5213 struct zone *zone;
5214 unsigned int cpu;
5215 int ret;
5216
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005217 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005218 if (!write || (ret == -EINVAL))
5219 return ret;
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07005220 for_each_populated_zone(zone) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005221 for_each_possible_cpu(cpu) {
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005222 unsigned long high;
5223 high = zone->present_pages / percpu_pagelist_fraction;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005224 setup_pagelist_highmark(
5225 per_cpu_ptr(zone->pageset, cpu), high);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005226 }
5227 }
5228 return 0;
5229}
5230
David S. Millerf034b5d2006-08-24 03:08:07 -07005231int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005232
5233#ifdef CONFIG_NUMA
5234static int __init set_hashdist(char *str)
5235{
5236 if (!str)
5237 return 0;
5238 hashdist = simple_strtoul(str, &str, 0);
5239 return 1;
5240}
5241__setup("hashdist=", set_hashdist);
5242#endif
5243
5244/*
5245 * allocate a large system hash table from bootmem
5246 * - it is assumed that the hash table must contain an exact power-of-2
5247 * quantity of entries
5248 * - limit is the number of hash buckets, not the total allocation size
5249 */
5250void *__init alloc_large_system_hash(const char *tablename,
5251 unsigned long bucketsize,
5252 unsigned long numentries,
5253 int scale,
5254 int flags,
5255 unsigned int *_hash_shift,
5256 unsigned int *_hash_mask,
5257 unsigned long limit)
5258{
5259 unsigned long long max = limit;
5260 unsigned long log2qty, size;
5261 void *table = NULL;
5262
5263 /* allow the kernel cmdline to have a say */
5264 if (!numentries) {
5265 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08005266 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005267 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
5268 numentries >>= 20 - PAGE_SHIFT;
5269 numentries <<= 20 - PAGE_SHIFT;
5270
5271 /* limit to 1 bucket per 2^scale bytes of low memory */
5272 if (scale > PAGE_SHIFT)
5273 numentries >>= (scale - PAGE_SHIFT);
5274 else
5275 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08005276
5277 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07005278 if (unlikely(flags & HASH_SMALL)) {
5279 /* Makes no sense without HASH_EARLY */
5280 WARN_ON(!(flags & HASH_EARLY));
5281 if (!(numentries >> *_hash_shift)) {
5282 numentries = 1UL << *_hash_shift;
5283 BUG_ON(!numentries);
5284 }
5285 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08005286 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005287 }
John Hawkes6e692ed2006-03-25 03:08:02 -08005288 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005289
5290 /* limit allocation size to 1/16 total memory by default */
5291 if (max == 0) {
5292 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
5293 do_div(max, bucketsize);
5294 }
5295
5296 if (numentries > max)
5297 numentries = max;
5298
David Howellsf0d1b0b2006-12-08 02:37:49 -08005299 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005300
5301 do {
5302 size = bucketsize << log2qty;
5303 if (flags & HASH_EARLY)
Jan Beulich74768ed2008-08-12 15:08:39 -07005304 table = alloc_bootmem_nopanic(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005305 else if (hashdist)
5306 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
5307 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07005308 /*
5309 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07005310 * some pages at the end of hash table which
5311 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07005312 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005313 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07005314 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005315 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
5316 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005317 }
5318 } while (!table && size > PAGE_SIZE && --log2qty);
5319
5320 if (!table)
5321 panic("Failed to allocate %s hash table\n", tablename);
5322
Robin Holtf241e662010-10-07 12:59:26 -07005323 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07005324 tablename,
Robin Holtf241e662010-10-07 12:59:26 -07005325 (1UL << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08005326 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005327 size);
5328
5329 if (_hash_shift)
5330 *_hash_shift = log2qty;
5331 if (_hash_mask)
5332 *_hash_mask = (1 << log2qty) - 1;
5333
5334 return table;
5335}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08005336
Mel Gorman835c1342007-10-16 01:25:47 -07005337/* Return a pointer to the bitmap storing bits affecting a block of pages */
5338static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
5339 unsigned long pfn)
5340{
5341#ifdef CONFIG_SPARSEMEM
5342 return __pfn_to_section(pfn)->pageblock_flags;
5343#else
5344 return zone->pageblock_flags;
5345#endif /* CONFIG_SPARSEMEM */
5346}
Andrew Morton6220ec72006-10-19 23:29:05 -07005347
Mel Gorman835c1342007-10-16 01:25:47 -07005348static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
5349{
5350#ifdef CONFIG_SPARSEMEM
5351 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07005352 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005353#else
5354 pfn = pfn - zone->zone_start_pfn;
Mel Gormand9c23402007-10-16 01:26:01 -07005355 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005356#endif /* CONFIG_SPARSEMEM */
5357}
5358
5359/**
Mel Gormand9c23402007-10-16 01:26:01 -07005360 * 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 -07005361 * @page: The page within the block of interest
5362 * @start_bitidx: The first bit of interest to retrieve
5363 * @end_bitidx: The last bit of interest
5364 * returns pageblock_bits flags
5365 */
5366unsigned long get_pageblock_flags_group(struct page *page,
5367 int start_bitidx, int end_bitidx)
5368{
5369 struct zone *zone;
5370 unsigned long *bitmap;
5371 unsigned long pfn, bitidx;
5372 unsigned long flags = 0;
5373 unsigned long value = 1;
5374
5375 zone = page_zone(page);
5376 pfn = page_to_pfn(page);
5377 bitmap = get_pageblock_bitmap(zone, pfn);
5378 bitidx = pfn_to_bitidx(zone, pfn);
5379
5380 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5381 if (test_bit(bitidx + start_bitidx, bitmap))
5382 flags |= value;
5383
5384 return flags;
5385}
5386
5387/**
Mel Gormand9c23402007-10-16 01:26:01 -07005388 * 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 -07005389 * @page: The page within the block of interest
5390 * @start_bitidx: The first bit of interest
5391 * @end_bitidx: The last bit of interest
5392 * @flags: The flags to set
5393 */
5394void set_pageblock_flags_group(struct page *page, unsigned long flags,
5395 int start_bitidx, int end_bitidx)
5396{
5397 struct zone *zone;
5398 unsigned long *bitmap;
5399 unsigned long pfn, bitidx;
5400 unsigned long value = 1;
5401
5402 zone = page_zone(page);
5403 pfn = page_to_pfn(page);
5404 bitmap = get_pageblock_bitmap(zone, pfn);
5405 bitidx = pfn_to_bitidx(zone, pfn);
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07005406 VM_BUG_ON(pfn < zone->zone_start_pfn);
5407 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
Mel Gorman835c1342007-10-16 01:25:47 -07005408
5409 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5410 if (flags & value)
5411 __set_bit(bitidx + start_bitidx, bitmap);
5412 else
5413 __clear_bit(bitidx + start_bitidx, bitmap);
5414}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005415
5416/*
5417 * This is designed as sub function...plz see page_isolation.c also.
5418 * set/clear page block's type to be ISOLATE.
5419 * page allocater never alloc memory from ISOLATE block.
5420 */
5421
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005422static int
5423__count_immobile_pages(struct zone *zone, struct page *page, int count)
5424{
5425 unsigned long pfn, iter, found;
5426 /*
5427 * For avoiding noise data, lru_add_drain_all() should be called
5428 * If ZONE_MOVABLE, the zone never contains immobile pages
5429 */
5430 if (zone_idx(zone) == ZONE_MOVABLE)
5431 return true;
5432
5433 if (get_pageblock_migratetype(page) == MIGRATE_MOVABLE)
5434 return true;
5435
5436 pfn = page_to_pfn(page);
5437 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
5438 unsigned long check = pfn + iter;
5439
Namhyung Kim29723fc2011-02-25 14:44:25 -08005440 if (!pfn_valid_within(check))
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005441 continue;
Namhyung Kim29723fc2011-02-25 14:44:25 -08005442
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005443 page = pfn_to_page(check);
5444 if (!page_count(page)) {
5445 if (PageBuddy(page))
5446 iter += (1 << page_order(page)) - 1;
5447 continue;
5448 }
5449 if (!PageLRU(page))
5450 found++;
5451 /*
5452 * If there are RECLAIMABLE pages, we need to check it.
5453 * But now, memory offline itself doesn't call shrink_slab()
5454 * and it still to be fixed.
5455 */
5456 /*
5457 * If the page is not RAM, page_count()should be 0.
5458 * we don't need more check. This is an _used_ not-movable page.
5459 *
5460 * The problematic thing here is PG_reserved pages. PG_reserved
5461 * is set to both of a memory hole page and a _used_ kernel
5462 * page at boot.
5463 */
5464 if (found > count)
5465 return false;
5466 }
5467 return true;
5468}
5469
5470bool is_pageblock_removable_nolock(struct page *page)
5471{
5472 struct zone *zone = page_zone(page);
5473 return __count_immobile_pages(zone, page, 0);
5474}
5475
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005476int set_migratetype_isolate(struct page *page)
5477{
5478 struct zone *zone;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005479 unsigned long flags, pfn;
Robert Jennings925cc712009-12-17 14:44:38 +00005480 struct memory_isolate_notify arg;
5481 int notifier_ret;
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005482 int ret = -EBUSY;
Shaohua Li8e7e40d2009-09-21 17:01:18 -07005483 int zone_idx;
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005484
5485 zone = page_zone(page);
Shaohua Li8e7e40d2009-09-21 17:01:18 -07005486 zone_idx = zone_idx(zone);
Robert Jennings925cc712009-12-17 14:44:38 +00005487
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005488 spin_lock_irqsave(&zone->lock, flags);
Robert Jennings925cc712009-12-17 14:44:38 +00005489
5490 pfn = page_to_pfn(page);
5491 arg.start_pfn = pfn;
5492 arg.nr_pages = pageblock_nr_pages;
5493 arg.pages_found = 0;
5494
5495 /*
5496 * It may be possible to isolate a pageblock even if the
5497 * migratetype is not MIGRATE_MOVABLE. The memory isolation
5498 * notifier chain is used by balloon drivers to return the
5499 * number of pages in a range that are held by the balloon
5500 * driver to shrink memory. If all the pages are accounted for
5501 * by balloons, are free, or on the LRU, isolation can continue.
5502 * Later, for example, when memory hotplug notifier runs, these
5503 * pages reported as "can be isolated" should be isolated(freed)
5504 * by the balloon driver through the memory notifier chain.
5505 */
5506 notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg);
5507 notifier_ret = notifier_to_errno(notifier_ret);
KAMEZAWA Hiroyuki4b204772010-10-26 14:21:29 -07005508 if (notifier_ret)
Robert Jennings925cc712009-12-17 14:44:38 +00005509 goto out;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005510 /*
5511 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
5512 * We just check MOVABLE pages.
5513 */
5514 if (__count_immobile_pages(zone, page, arg.pages_found))
Robert Jennings925cc712009-12-17 14:44:38 +00005515 ret = 0;
5516
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005517 /*
5518 * immobile means "not-on-lru" paes. If immobile is larger than
5519 * removable-by-driver pages reported by notifier, we'll fail.
5520 */
5521
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005522out:
Robert Jennings925cc712009-12-17 14:44:38 +00005523 if (!ret) {
5524 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
5525 move_freepages_block(zone, page, MIGRATE_ISOLATE);
5526 }
5527
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005528 spin_unlock_irqrestore(&zone->lock, flags);
5529 if (!ret)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005530 drain_all_pages();
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005531 return ret;
5532}
5533
5534void unset_migratetype_isolate(struct page *page)
5535{
5536 struct zone *zone;
5537 unsigned long flags;
5538 zone = page_zone(page);
5539 spin_lock_irqsave(&zone->lock, flags);
5540 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
5541 goto out;
5542 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5543 move_freepages_block(zone, page, MIGRATE_MOVABLE);
5544out:
5545 spin_unlock_irqrestore(&zone->lock, flags);
5546}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07005547
5548#ifdef CONFIG_MEMORY_HOTREMOVE
5549/*
5550 * All pages in the range must be isolated before calling this.
5551 */
5552void
5553__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
5554{
5555 struct page *page;
5556 struct zone *zone;
5557 int order, i;
5558 unsigned long pfn;
5559 unsigned long flags;
5560 /* find the first valid pfn */
5561 for (pfn = start_pfn; pfn < end_pfn; pfn++)
5562 if (pfn_valid(pfn))
5563 break;
5564 if (pfn == end_pfn)
5565 return;
5566 zone = page_zone(pfn_to_page(pfn));
5567 spin_lock_irqsave(&zone->lock, flags);
5568 pfn = start_pfn;
5569 while (pfn < end_pfn) {
5570 if (!pfn_valid(pfn)) {
5571 pfn++;
5572 continue;
5573 }
5574 page = pfn_to_page(pfn);
5575 BUG_ON(page_count(page));
5576 BUG_ON(!PageBuddy(page));
5577 order = page_order(page);
5578#ifdef CONFIG_DEBUG_VM
5579 printk(KERN_INFO "remove from free list %lx %d %lx\n",
5580 pfn, 1 << order, end_pfn);
5581#endif
5582 list_del(&page->lru);
5583 rmv_page_order(page);
5584 zone->free_area[order].nr_free--;
5585 __mod_zone_page_state(zone, NR_FREE_PAGES,
5586 - (1UL << order));
5587 for (i = 0; i < (1 << order); i++)
5588 SetPageReserved((page+i));
5589 pfn += (1 << order);
5590 }
5591 spin_unlock_irqrestore(&zone->lock, flags);
5592}
5593#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01005594
5595#ifdef CONFIG_MEMORY_FAILURE
5596bool is_free_buddy_page(struct page *page)
5597{
5598 struct zone *zone = page_zone(page);
5599 unsigned long pfn = page_to_pfn(page);
5600 unsigned long flags;
5601 int order;
5602
5603 spin_lock_irqsave(&zone->lock, flags);
5604 for (order = 0; order < MAX_ORDER; order++) {
5605 struct page *page_head = page - (pfn & ((1 << order) - 1));
5606
5607 if (PageBuddy(page_head) && page_order(page_head) >= order)
5608 break;
5609 }
5610 spin_unlock_irqrestore(&zone->lock, flags);
5611
5612 return order < MAX_ORDER;
5613}
5614#endif
Wu Fengguang718a3822010-03-10 15:20:43 -08005615
5616static struct trace_print_flags pageflag_names[] = {
5617 {1UL << PG_locked, "locked" },
5618 {1UL << PG_error, "error" },
5619 {1UL << PG_referenced, "referenced" },
5620 {1UL << PG_uptodate, "uptodate" },
5621 {1UL << PG_dirty, "dirty" },
5622 {1UL << PG_lru, "lru" },
5623 {1UL << PG_active, "active" },
5624 {1UL << PG_slab, "slab" },
5625 {1UL << PG_owner_priv_1, "owner_priv_1" },
5626 {1UL << PG_arch_1, "arch_1" },
5627 {1UL << PG_reserved, "reserved" },
5628 {1UL << PG_private, "private" },
5629 {1UL << PG_private_2, "private_2" },
5630 {1UL << PG_writeback, "writeback" },
5631#ifdef CONFIG_PAGEFLAGS_EXTENDED
5632 {1UL << PG_head, "head" },
5633 {1UL << PG_tail, "tail" },
5634#else
5635 {1UL << PG_compound, "compound" },
5636#endif
5637 {1UL << PG_swapcache, "swapcache" },
5638 {1UL << PG_mappedtodisk, "mappedtodisk" },
5639 {1UL << PG_reclaim, "reclaim" },
Wu Fengguang718a3822010-03-10 15:20:43 -08005640 {1UL << PG_swapbacked, "swapbacked" },
5641 {1UL << PG_unevictable, "unevictable" },
5642#ifdef CONFIG_MMU
5643 {1UL << PG_mlocked, "mlocked" },
5644#endif
5645#ifdef CONFIG_ARCH_USES_PG_UNCACHED
5646 {1UL << PG_uncached, "uncached" },
5647#endif
5648#ifdef CONFIG_MEMORY_FAILURE
5649 {1UL << PG_hwpoison, "hwpoison" },
5650#endif
5651 {-1UL, NULL },
5652};
5653
5654static void dump_page_flags(unsigned long flags)
5655{
5656 const char *delim = "";
5657 unsigned long mask;
5658 int i;
5659
5660 printk(KERN_ALERT "page flags: %#lx(", flags);
5661
5662 /* remove zone id */
5663 flags &= (1UL << NR_PAGEFLAGS) - 1;
5664
5665 for (i = 0; pageflag_names[i].name && flags; i++) {
5666
5667 mask = pageflag_names[i].mask;
5668 if ((flags & mask) != mask)
5669 continue;
5670
5671 flags &= ~mask;
5672 printk("%s%s", delim, pageflag_names[i].name);
5673 delim = "|";
5674 }
5675
5676 /* check for left over flags */
5677 if (flags)
5678 printk("%s%#lx", delim, flags);
5679
5680 printk(")\n");
5681}
5682
5683void dump_page(struct page *page)
5684{
5685 printk(KERN_ALERT
5686 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
Andrea Arcangeli4e9f64c2011-01-13 15:46:29 -08005687 page, atomic_read(&page->_count), page_mapcount(page),
Wu Fengguang718a3822010-03-10 15:20:43 -08005688 page->mapping, page->index);
5689 dump_page_flags(page->flags);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07005690 mem_cgroup_print_bad_page(page);
Wu Fengguang718a3822010-03-10 15:20:43 -08005691}