blob: ff7e158723987d4355ec84a2c090716ec8accbe9 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
57#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070058#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070059#include "internal.h"
60
Lee Schermerhorn72812012010-05-26 14:44:56 -070061#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
62DEFINE_PER_CPU(int, numa_node);
63EXPORT_PER_CPU_SYMBOL(numa_node);
64#endif
65
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070066#ifdef CONFIG_HAVE_MEMORYLESS_NODES
67/*
68 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
69 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
70 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
71 * defined in <linux/topology.h>.
72 */
73DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
74EXPORT_PER_CPU_SYMBOL(_numa_mem_);
75#endif
76
Linus Torvalds1da177e2005-04-16 15:20:36 -070077/*
Christoph Lameter13808912007-10-16 01:25:27 -070078 * Array of node states.
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 */
Christoph Lameter13808912007-10-16 01:25:27 -070080nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
81 [N_POSSIBLE] = NODE_MASK_ALL,
82 [N_ONLINE] = { { [0] = 1UL } },
83#ifndef CONFIG_NUMA
84 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
85#ifdef CONFIG_HIGHMEM
86 [N_HIGH_MEMORY] = { { [0] = 1UL } },
87#endif
88 [N_CPU] = { { [0] = 1UL } },
89#endif /* NUMA */
90};
91EXPORT_SYMBOL(node_states);
92
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070093unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -070094unsigned long totalreserve_pages __read_mostly;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -080095int percpu_pagelist_fraction;
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +100096gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -070097
Rafael J. Wysocki452aa692010-03-05 13:42:13 -080098#ifdef CONFIG_PM_SLEEP
99/*
100 * The following functions are used by the suspend/hibernate code to temporarily
101 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
102 * while devices are suspended. To avoid races with the suspend/hibernate code,
103 * they should always be called with pm_mutex held (gfp_allowed_mask also should
104 * only be modified with pm_mutex held, unless the suspend/hibernate code is
105 * guaranteed not to run in parallel with that modification).
106 */
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100107
108static gfp_t saved_gfp_mask;
109
110void pm_restore_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800111{
112 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100113 if (saved_gfp_mask) {
114 gfp_allowed_mask = saved_gfp_mask;
115 saved_gfp_mask = 0;
116 }
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800117}
118
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100119void pm_restrict_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800120{
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800121 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100122 WARN_ON(saved_gfp_mask);
123 saved_gfp_mask = gfp_allowed_mask;
124 gfp_allowed_mask &= ~GFP_IOFS;
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800125}
126#endif /* CONFIG_PM_SLEEP */
127
Mel Gormand9c23402007-10-16 01:26:01 -0700128#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
129int pageblock_order __read_mostly;
130#endif
131
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800132static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -0800133
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134/*
135 * results with 256, 32 in the lowmem_reserve sysctl:
136 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
137 * 1G machine -> (16M dma, 784M normal, 224M high)
138 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
139 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
140 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +0100141 *
142 * TBD: should special case ZONE_DMA32 machines here - in those we normally
143 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700145int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800146#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700147 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -0800148#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700149#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700150 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700151#endif
Christoph Lametere53ef382006-09-25 23:31:14 -0700152#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700153 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700154#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700155 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700156};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157
158EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159
Helge Deller15ad7cd2006-12-06 20:40:36 -0800160static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800161#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700162 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800163#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700164#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700165 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700166#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700167 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700168#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700169 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700170#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700171 "Movable",
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700172};
173
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174int min_free_kbytes = 1024;
175
Jan Beulich2c85f512009-09-21 17:03:07 -0700176static unsigned long __meminitdata nr_kernel_pages;
177static unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700178static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179
Mel Gormanc7132162006-09-27 01:49:43 -0700180#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
181 /*
Simon Arlott183ff222007-10-20 01:27:18 +0200182 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
Mel Gormanc7132162006-09-27 01:49:43 -0700183 * ranges of memory (RAM) that may be registered with add_active_range().
184 * Ranges passed to add_active_range() will be merged if possible
185 * so the number of times add_active_range() can be called is
186 * related to the number of nodes and the number of holes
187 */
188 #ifdef CONFIG_MAX_ACTIVE_REGIONS
189 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
190 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
191 #else
192 #if MAX_NUMNODES >= 32
193 /* If there can be many nodes, allow up to 50 holes per node */
194 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
195 #else
196 /* By default, allow up to 256 distinct regions */
197 #define MAX_ACTIVE_REGIONS 256
198 #endif
199 #endif
200
Jan Beulich98011f52007-07-15 23:38:17 -0700201 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
202 static int __meminitdata nr_nodemap_entries;
203 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
204 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
Adrian Bunkb69a7282008-07-23 21:28:12 -0700205 static unsigned long __initdata required_kernelcore;
Adrian Bunk484f51f2007-10-16 01:26:03 -0700206 static unsigned long __initdata required_movablecore;
Adrian Bunkb69a7282008-07-23 21:28:12 -0700207 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Mel Gorman2a1e2742007-07-17 04:03:12 -0700208
209 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
210 int movable_zone;
211 EXPORT_SYMBOL(movable_zone);
Mel Gormanc7132162006-09-27 01:49:43 -0700212#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
213
Miklos Szeredi418508c2007-05-23 13:57:55 -0700214#if MAX_NUMNODES > 1
215int nr_node_ids __read_mostly = MAX_NUMNODES;
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700216int nr_online_nodes __read_mostly = 1;
Miklos Szeredi418508c2007-05-23 13:57:55 -0700217EXPORT_SYMBOL(nr_node_ids);
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700218EXPORT_SYMBOL(nr_online_nodes);
Miklos Szeredi418508c2007-05-23 13:57:55 -0700219#endif
220
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700221int page_group_by_mobility_disabled __read_mostly;
222
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700223static void set_pageblock_migratetype(struct page *page, int migratetype)
224{
Mel Gorman49255c62009-06-16 15:31:58 -0700225
226 if (unlikely(page_group_by_mobility_disabled))
227 migratetype = MIGRATE_UNMOVABLE;
228
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700229 set_pageblock_flags_group(page, (unsigned long)migratetype,
230 PB_migrate, PB_migrate_end);
231}
232
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -0700233bool oom_killer_disabled __read_mostly;
234
Nick Piggin13e74442006-01-06 00:10:58 -0800235#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700236static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700238 int ret = 0;
239 unsigned seq;
240 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -0700241
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700242 do {
243 seq = zone_span_seqbegin(zone);
244 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
245 ret = 1;
246 else if (pfn < zone->zone_start_pfn)
247 ret = 1;
248 } while (zone_span_seqretry(zone, seq));
249
250 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700251}
252
253static int page_is_consistent(struct zone *zone, struct page *page)
254{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700255 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700256 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700258 return 0;
259
260 return 1;
261}
262/*
263 * Temporary debugging check for pages not lying within a given zone.
264 */
265static int bad_range(struct zone *zone, struct page *page)
266{
267 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700269 if (!page_is_consistent(zone, page))
270 return 1;
271
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 return 0;
273}
Nick Piggin13e74442006-01-06 00:10:58 -0800274#else
275static inline int bad_range(struct zone *zone, struct page *page)
276{
277 return 0;
278}
279#endif
280
Nick Piggin224abf92006-01-06 00:11:11 -0800281static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800283 static unsigned long resume;
284 static unsigned long nr_shown;
285 static unsigned long nr_unshown;
286
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200287 /* Don't complain about poisoned pages */
288 if (PageHWPoison(page)) {
289 __ClearPageBuddy(page);
290 return;
291 }
292
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800293 /*
294 * Allow a burst of 60 reports, then keep quiet for that minute;
295 * or allow a steady drip of one report per second.
296 */
297 if (nr_shown == 60) {
298 if (time_before(jiffies, resume)) {
299 nr_unshown++;
300 goto out;
301 }
302 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800303 printk(KERN_ALERT
304 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800305 nr_unshown);
306 nr_unshown = 0;
307 }
308 nr_shown = 0;
309 }
310 if (nr_shown++ == 0)
311 resume = jiffies + 60 * HZ;
312
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800313 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800314 current->comm, page_to_pfn(page));
Wu Fengguang718a3822010-03-10 15:20:43 -0800315 dump_page(page);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700316
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800318out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800319 /* Leave bad fields for debug, except PageBuddy could make trouble */
320 __ClearPageBuddy(page);
Randy Dunlap9f158332005-09-13 01:25:16 -0700321 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322}
323
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324/*
325 * Higher-order pages are called "compound pages". They are structured thusly:
326 *
327 * The first PAGE_SIZE page is called the "head page".
328 *
329 * The remaining PAGE_SIZE pages are called "tail pages".
330 *
331 * All pages have PG_compound set. All pages have their ->private pointing at
332 * the head page (even the head page has this).
333 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800334 * The first tail page's ->lru.next holds the address of the compound page's
335 * put_page() function. Its ->lru.prev holds the order of allocation.
336 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800338
339static void free_compound_page(struct page *page)
340{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700341 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800342}
343
Andi Kleen01ad1c02008-07-23 21:27:46 -0700344void prep_compound_page(struct page *page, unsigned long order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345{
346 int i;
347 int nr_pages = 1 << order;
348
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800349 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700350 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700351 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800352 for (i = 1; i < nr_pages; i++) {
353 struct page *p = page + i;
354
Christoph Lameterd85f3382007-05-06 14:49:39 -0700355 __SetPageTail(p);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700356 p->first_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 }
358}
359
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800360static int destroy_compound_page(struct page *page, unsigned long order)
Andy Whitcroft18229df2008-11-06 12:53:27 -0800361{
362 int i;
363 int nr_pages = 1 << order;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800364 int bad = 0;
Andy Whitcroft18229df2008-11-06 12:53:27 -0800365
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800366 if (unlikely(compound_order(page) != order) ||
367 unlikely(!PageHead(page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800368 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800369 bad++;
370 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371
Christoph Lameter6d777952007-05-06 14:49:40 -0700372 __ClearPageHead(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800373
Andy Whitcroft18229df2008-11-06 12:53:27 -0800374 for (i = 1; i < nr_pages; i++) {
375 struct page *p = page + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376
Alexey Zaytseve713a212009-01-10 02:47:57 +0300377 if (unlikely(!PageTail(p) || (p->first_page != page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800378 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800379 bad++;
380 }
Christoph Lameterd85f3382007-05-06 14:49:39 -0700381 __ClearPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800383
384 return bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386
Nick Piggin17cf4402006-03-22 00:08:41 -0800387static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
388{
389 int i;
390
Andrew Morton6626c5d2006-03-22 00:08:42 -0800391 /*
392 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
393 * and __GFP_HIGHMEM from hard or soft interrupt context.
394 */
Nick Piggin725d7042006-09-25 23:30:55 -0700395 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800396 for (i = 0; i < (1 << order); i++)
397 clear_highpage(page + i);
398}
399
Andrew Morton6aa30012006-04-18 22:20:52 -0700400static inline void set_page_order(struct page *page, int order)
401{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700402 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000403 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404}
405
406static inline void rmv_page_order(struct page *page)
407{
Nick Piggin676165a2006-04-10 11:21:48 +1000408 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700409 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410}
411
412/*
413 * Locate the struct page for both the matching buddy in our
414 * pair (buddy1) and the combined O(n+1) page they form (page).
415 *
416 * 1) Any buddy B1 will have an order O twin B2 which satisfies
417 * the following equation:
418 * B2 = B1 ^ (1 << O)
419 * For example, if the starting buddy (buddy2) is #8 its order
420 * 1 buddy is #10:
421 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
422 *
423 * 2) Any buddy B will have an order O+1 parent P which
424 * satisfies the following equation:
425 * P = B & ~(1 << O)
426 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200427 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 */
429static inline struct page *
430__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
431{
432 unsigned long buddy_idx = page_idx ^ (1 << order);
433
434 return page + (buddy_idx - page_idx);
435}
436
437static inline unsigned long
438__find_combined_index(unsigned long page_idx, unsigned int order)
439{
440 return (page_idx & ~(1 << order));
441}
442
443/*
444 * This function checks whether a page is free && is the buddy
445 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800446 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000447 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700448 * (c) a page and its buddy have the same order &&
449 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 *
Nick Piggin676165a2006-04-10 11:21:48 +1000451 * For recording whether a page is in the buddy system, we use PG_buddy.
452 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
453 *
454 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700456static inline int page_is_buddy(struct page *page, struct page *buddy,
457 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700459 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800460 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800461
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700462 if (page_zone_id(page) != page_zone_id(buddy))
463 return 0;
464
465 if (PageBuddy(buddy) && page_order(buddy) == order) {
Nick Piggina3af9c32009-06-16 15:32:10 -0700466 VM_BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa30012006-04-18 22:20:52 -0700467 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000468 }
Andrew Morton6aa30012006-04-18 22:20:52 -0700469 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470}
471
472/*
473 * Freeing function for a buddy system allocator.
474 *
475 * The concept of a buddy system is to maintain direct-mapped table
476 * (containing bit values) for memory blocks of various "orders".
477 * The bottom level table contains the map for the smallest allocatable
478 * units of memory (here, pages), and each level above it describes
479 * pairs of units from the levels below, hence, "buddies".
480 * At a high level, all that happens here is marking the table entry
481 * at the bottom level available, and propagating the changes upward
482 * as necessary, plus some accounting needed to play nicely with other
483 * parts of the VM system.
484 * At each level, we keep a list of pages, which are heads of continuous
Nick Piggin676165a2006-04-10 11:21:48 +1000485 * free pages of length of (1 << order) and marked with PG_buddy. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700486 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 * So when we are allocating or freeing one, we can derive the state of the
488 * other. That is, if we allocate a small block, and both were
489 * free, the remainder of the region must be split into blocks.
490 * If a block is freed, and its buddy is also free, then this
491 * triggers coalescing into a block of larger size.
492 *
493 * -- wli
494 */
495
Nick Piggin48db57f2006-01-08 01:00:42 -0800496static inline void __free_one_page(struct page *page,
Mel Gormaned0ae212009-06-16 15:32:07 -0700497 struct zone *zone, unsigned int order,
498 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499{
500 unsigned long page_idx;
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700501 unsigned long combined_idx;
502 struct page *buddy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503
Nick Piggin224abf92006-01-06 00:11:11 -0800504 if (unlikely(PageCompound(page)))
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800505 if (unlikely(destroy_compound_page(page, order)))
506 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507
Mel Gormaned0ae212009-06-16 15:32:07 -0700508 VM_BUG_ON(migratetype == -1);
509
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
511
Mel Gormanf2260e62009-06-16 15:32:13 -0700512 VM_BUG_ON(page_idx & ((1 << order) - 1));
Nick Piggin725d7042006-09-25 23:30:55 -0700513 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 while (order < MAX_ORDER-1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 buddy = __page_find_buddy(page, page_idx, order);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700517 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700518 break;
Nick Piggin13e74442006-01-06 00:10:58 -0800519
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700520 /* Our buddy is free, merge with it and move up one order. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 list_del(&buddy->lru);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700522 zone->free_area[order].nr_free--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 rmv_page_order(buddy);
Nick Piggin13e74442006-01-06 00:10:58 -0800524 combined_idx = __find_combined_index(page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 page = page + (combined_idx - page_idx);
526 page_idx = combined_idx;
527 order++;
528 }
529 set_page_order(page, order);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700530
531 /*
532 * If this is not the largest possible page, check if the buddy
533 * of the next-highest order is free. If it is, it's possible
534 * that pages are being freed that will coalesce soon. In case,
535 * that is happening, add the free page to the tail of the list
536 * so it's less likely to be used soon and more likely to be merged
537 * as a higher order page
538 */
Mel Gormanb7f50cf2010-10-26 14:21:11 -0700539 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700540 struct page *higher_page, *higher_buddy;
541 combined_idx = __find_combined_index(page_idx, order);
542 higher_page = page + combined_idx - page_idx;
543 higher_buddy = __page_find_buddy(higher_page, combined_idx, order + 1);
544 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
545 list_add_tail(&page->lru,
546 &zone->free_area[order].free_list[migratetype]);
547 goto out;
548 }
549 }
550
551 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
552out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 zone->free_area[order].nr_free++;
554}
555
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700556/*
557 * free_page_mlock() -- clean up attempts to free and mlocked() page.
558 * Page should not be on lru, so no need to fix that up.
559 * free_pages_check() will verify...
560 */
561static inline void free_page_mlock(struct page *page)
562{
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700563 __dec_zone_page_state(page, NR_MLOCK);
564 __count_vm_event(UNEVICTABLE_MLOCKFREED);
565}
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700566
Nick Piggin224abf92006-01-06 00:11:11 -0800567static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568{
Nick Piggin92be2e32006-01-06 00:10:57 -0800569 if (unlikely(page_mapcount(page) |
570 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700571 (atomic_read(&page->_count) != 0) |
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800572 (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800573 bad_page(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800574 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800575 }
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800576 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
577 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
578 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579}
580
581/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700582 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700584 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 *
586 * If the zone was previously in an "all pages pinned" state then look to
587 * see if this freeing clears that state.
588 *
589 * And clear the zone's pages_scanned counter, to hold off the "all pages are
590 * pinned" detection logic.
591 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700592static void free_pcppages_bulk(struct zone *zone, int count,
593 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594{
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700595 int migratetype = 0;
Mel Gormana6f9edd2009-09-21 17:03:20 -0700596 int batch_free = 0;
Mel Gorman72853e22010-09-09 16:38:16 -0700597 int to_free = count;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700598
Nick Pigginc54ad302006-01-06 00:10:56 -0800599 spin_lock(&zone->lock);
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -0800600 zone->all_unreclaimable = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700602
Mel Gorman72853e22010-09-09 16:38:16 -0700603 while (to_free) {
Nick Piggin48db57f2006-01-08 01:00:42 -0800604 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700605 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -0800606
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700607 /*
Mel Gormana6f9edd2009-09-21 17:03:20 -0700608 * Remove pages from lists in a round-robin fashion. A
609 * batch_free count is maintained that is incremented when an
610 * empty list is encountered. This is so more pages are freed
611 * off fuller lists instead of spinning excessively around empty
612 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700613 */
614 do {
Mel Gormana6f9edd2009-09-21 17:03:20 -0700615 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700616 if (++migratetype == MIGRATE_PCPTYPES)
617 migratetype = 0;
618 list = &pcp->lists[migratetype];
619 } while (list_empty(list));
620
Mel Gormana6f9edd2009-09-21 17:03:20 -0700621 do {
622 page = list_entry(list->prev, struct page, lru);
623 /* must delete as __free_one_page list manipulates */
624 list_del(&page->lru);
Hugh Dickinsa7016232010-01-29 17:46:34 +0000625 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
626 __free_one_page(page, zone, 0, page_private(page));
627 trace_mm_page_pcpu_drain(page, 0, page_private(page));
Mel Gorman72853e22010-09-09 16:38:16 -0700628 } while (--to_free && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 }
Mel Gorman72853e22010-09-09 16:38:16 -0700630 __mod_zone_page_state(zone, NR_FREE_PAGES, count);
Nick Pigginc54ad302006-01-06 00:10:56 -0800631 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632}
633
Mel Gormaned0ae212009-06-16 15:32:07 -0700634static void free_one_page(struct zone *zone, struct page *page, int order,
635 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800636{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700637 spin_lock(&zone->lock);
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -0800638 zone->all_unreclaimable = 0;
Christoph Lameter006d22d2006-09-25 23:31:48 -0700639 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700640
Mel Gormaned0ae212009-06-16 15:32:07 -0700641 __free_one_page(page, zone, order, migratetype);
Mel Gorman72853e22010-09-09 16:38:16 -0700642 __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700643 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800644}
645
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700646static bool free_pages_prepare(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 int i;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800649 int bad = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650
Li Hongf6503162010-03-05 13:41:52 -0800651 trace_mm_page_free_direct(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100652 kmemcheck_free_shadow(page, order);
653
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700654 for (i = 0; i < (1 << order); i++) {
655 struct page *pg = page + i;
656
657 if (PageAnon(pg))
658 pg->mapping = NULL;
659 bad += free_pages_check(pg);
660 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800661 if (bad)
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700662 return false;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800663
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700664 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -0700665 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700666 debug_check_no_obj_freed(page_address(page),
667 PAGE_SIZE << order);
668 }
Nick Piggindafb1362006-10-11 01:21:30 -0700669 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800670 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700671
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700672 return true;
673}
674
675static void __free_pages_ok(struct page *page, unsigned int order)
676{
677 unsigned long flags;
678 int wasMlocked = __TestClearPageMlocked(page);
679
680 if (!free_pages_prepare(page, order))
681 return;
682
Nick Pigginc54ad302006-01-06 00:10:56 -0800683 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +0200684 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -0700685 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700686 __count_vm_events(PGFREE, 1 << order);
Mel Gormaned0ae212009-06-16 15:32:07 -0700687 free_one_page(page_zone(page), page, order,
688 get_pageblock_migratetype(page));
Nick Pigginc54ad302006-01-06 00:10:56 -0800689 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690}
691
David Howellsa226f6c2006-01-06 00:11:08 -0800692/*
693 * permit the bootmem allocator to evade page validation on high-order frees
694 */
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700695void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800696{
697 if (order == 0) {
698 __ClearPageReserved(page);
699 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800700 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800701 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800702 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800703 int loop;
704
Nick Piggin545b1ea2006-03-22 00:08:07 -0800705 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800706 for (loop = 0; loop < BITS_PER_LONG; loop++) {
707 struct page *p = &page[loop];
708
Nick Piggin545b1ea2006-03-22 00:08:07 -0800709 if (loop + 1 < BITS_PER_LONG)
710 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800711 __ClearPageReserved(p);
712 set_page_count(p, 0);
713 }
714
Nick Piggin7835e982006-03-22 00:08:40 -0800715 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800716 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800717 }
718}
719
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720
721/*
722 * The order of subdivision here is critical for the IO subsystem.
723 * Please do not alter this order without good reasons and regression
724 * testing. Specifically, as large blocks of memory are subdivided,
725 * the order in which smaller blocks are delivered depends on the order
726 * they're subdivided in this function. This is the primary factor
727 * influencing the order in which pages are delivered to the IO
728 * subsystem according to empirical testing, and this is also justified
729 * by considering the behavior of a buddy system containing a single
730 * large block of memory acted on by a series of small allocations.
731 * This behavior is a critical factor in sglist merging's success.
732 *
733 * -- wli
734 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800735static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700736 int low, int high, struct free_area *area,
737 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738{
739 unsigned long size = 1 << high;
740
741 while (high > low) {
742 area--;
743 high--;
744 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700745 VM_BUG_ON(bad_range(zone, &page[size]));
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700746 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 area->nr_free++;
748 set_page_order(&page[size], high);
749 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750}
751
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752/*
753 * This page is about to be returned from the page allocator
754 */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200755static inline int check_new_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756{
Nick Piggin92be2e32006-01-06 00:10:57 -0800757 if (unlikely(page_mapcount(page) |
758 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700759 (atomic_read(&page->_count) != 0) |
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800760 (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) {
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);
873 list_del(&page->lru);
874 list_add(&page->lru,
875 &zone->free_area[order].free_list[migratetype]);
876 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -0700877 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -0700878 }
879
Mel Gormand1003132007-10-16 01:26:00 -0700880 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -0700881}
882
Adrian Bunkb69a7282008-07-23 21:28:12 -0700883static int move_freepages_block(struct zone *zone, struct page *page,
884 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700885{
886 unsigned long start_pfn, end_pfn;
887 struct page *start_page, *end_page;
888
889 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -0700890 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -0700891 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -0700892 end_page = start_page + pageblock_nr_pages - 1;
893 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -0700894
895 /* Do not cross zone boundaries */
896 if (start_pfn < zone->zone_start_pfn)
897 start_page = page;
898 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
899 return 0;
900
901 return move_freepages(zone, start_page, end_page, migratetype);
902}
903
Mel Gorman2f66a682009-09-21 17:02:31 -0700904static void change_pageblock_range(struct page *pageblock_page,
905 int start_order, int migratetype)
906{
907 int nr_pageblocks = 1 << (start_order - pageblock_order);
908
909 while (nr_pageblocks--) {
910 set_pageblock_migratetype(pageblock_page, migratetype);
911 pageblock_page += pageblock_nr_pages;
912 }
913}
914
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700915/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -0700916static inline struct page *
917__rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700918{
919 struct free_area * area;
920 int current_order;
921 struct page *page;
922 int migratetype, i;
923
924 /* Find the largest possible block of pages in the other list */
925 for (current_order = MAX_ORDER-1; current_order >= order;
926 --current_order) {
927 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
928 migratetype = fallbacks[start_migratetype][i];
929
Mel Gorman56fd56b2007-10-16 01:25:58 -0700930 /* MIGRATE_RESERVE handled later if necessary */
931 if (migratetype == MIGRATE_RESERVE)
932 continue;
Mel Gormane0104872007-10-16 01:25:53 -0700933
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700934 area = &(zone->free_area[current_order]);
935 if (list_empty(&area->free_list[migratetype]))
936 continue;
937
938 page = list_entry(area->free_list[migratetype].next,
939 struct page, lru);
940 area->nr_free--;
941
942 /*
Mel Gormanc361be52007-10-16 01:25:51 -0700943 * If breaking a large block of pages, move all free
Mel Gorman46dafbc2007-10-16 01:25:55 -0700944 * pages to the preferred allocation list. If falling
945 * back for a reclaimable kernel allocation, be more
946 * agressive about taking ownership of free pages
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700947 */
Mel Gormand9c23402007-10-16 01:26:01 -0700948 if (unlikely(current_order >= (pageblock_order >> 1)) ||
Mel Gormandd5d2412009-09-05 11:17:11 -0700949 start_migratetype == MIGRATE_RECLAIMABLE ||
950 page_group_by_mobility_disabled) {
Mel Gorman46dafbc2007-10-16 01:25:55 -0700951 unsigned long pages;
952 pages = move_freepages_block(zone, page,
953 start_migratetype);
954
955 /* Claim the whole block if over half of it is free */
Mel Gormandd5d2412009-09-05 11:17:11 -0700956 if (pages >= (1 << (pageblock_order-1)) ||
957 page_group_by_mobility_disabled)
Mel Gorman46dafbc2007-10-16 01:25:55 -0700958 set_pageblock_migratetype(page,
959 start_migratetype);
960
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700961 migratetype = start_migratetype;
Mel Gormanc361be52007-10-16 01:25:51 -0700962 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700963
964 /* Remove the page from the freelists */
965 list_del(&page->lru);
966 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700967
Mel Gorman2f66a682009-09-21 17:02:31 -0700968 /* Take ownership for orders >= pageblock_order */
969 if (current_order >= pageblock_order)
970 change_pageblock_range(page, current_order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700971 start_migratetype);
972
973 expand(zone, page, order, current_order, area, migratetype);
Mel Gormane0fff1b2009-09-21 17:02:42 -0700974
975 trace_mm_page_alloc_extfrag(page, order, current_order,
976 start_migratetype, migratetype);
977
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700978 return page;
979 }
980 }
981
Mel Gorman728ec982009-06-16 15:32:04 -0700982 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700983}
984
Mel Gorman56fd56b2007-10-16 01:25:58 -0700985/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 * Do the hard work of removing an element from the buddy allocator.
987 * Call me with the zone->lock already held.
988 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700989static struct page *__rmqueue(struct zone *zone, unsigned int order,
990 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 struct page *page;
993
Mel Gorman728ec982009-06-16 15:32:04 -0700994retry_reserve:
Mel Gorman56fd56b2007-10-16 01:25:58 -0700995 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996
Mel Gorman728ec982009-06-16 15:32:04 -0700997 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Mel Gorman56fd56b2007-10-16 01:25:58 -0700998 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700999
Mel Gorman728ec982009-06-16 15:32:04 -07001000 /*
1001 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1002 * is used because __rmqueue_smallest is an inline function
1003 * and we want just one call site
1004 */
1005 if (!page) {
1006 migratetype = MIGRATE_RESERVE;
1007 goto retry_reserve;
1008 }
1009 }
1010
Mel Gorman0d3d0622009-09-21 17:02:44 -07001011 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001012 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013}
1014
1015/*
1016 * Obtain a specified number of elements from the buddy allocator, all under
1017 * a single hold of the lock, for efficiency. Add them to the supplied list.
1018 * Returns the number of new pages which were placed at *list.
1019 */
1020static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001021 unsigned long count, struct list_head *list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001022 int migratetype, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025
Nick Pigginc54ad302006-01-06 00:10:56 -08001026 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001028 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001029 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08001031
1032 /*
1033 * Split buddy pages returned by expand() are received here
1034 * in physical page order. The page is added to the callers and
1035 * list and the list head then moves forward. From the callers
1036 * perspective, the linked list is ordered by page number in
1037 * some conditions. This is useful for IO devices that can
1038 * merge IO requests if the physical pages are ordered
1039 * properly.
1040 */
Mel Gormane084b2d2009-07-29 15:02:04 -07001041 if (likely(cold == 0))
1042 list_add(&page->lru, list);
1043 else
1044 list_add_tail(&page->lru, list);
Mel Gorman535131e2007-10-16 01:25:49 -07001045 set_page_private(page, migratetype);
Mel Gorman81eabcb2007-12-17 16:20:05 -08001046 list = &page->lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 }
Mel Gormanf2260e62009-06-16 15:32:13 -07001048 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08001049 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001050 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051}
1052
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001053#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001054/*
Christoph Lameter4037d452007-05-09 02:35:14 -07001055 * Called from the vmstat counter updater to drain pagesets of this
1056 * currently executing processor on remote nodes after they have
1057 * expired.
1058 *
Christoph Lameter879336c2006-03-22 00:09:08 -08001059 * Note that this function must be called with the thread pinned to
1060 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001061 */
Christoph Lameter4037d452007-05-09 02:35:14 -07001062void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001063{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001064 unsigned long flags;
Christoph Lameter4037d452007-05-09 02:35:14 -07001065 int to_drain;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001066
Christoph Lameter4037d452007-05-09 02:35:14 -07001067 local_irq_save(flags);
1068 if (pcp->count >= pcp->batch)
1069 to_drain = pcp->batch;
1070 else
1071 to_drain = pcp->count;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001072 free_pcppages_bulk(zone, to_drain, pcp);
Christoph Lameter4037d452007-05-09 02:35:14 -07001073 pcp->count -= to_drain;
1074 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001075}
1076#endif
1077
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001078/*
1079 * Drain pages of the indicated processor.
1080 *
1081 * The processor must either be the current processor and the
1082 * thread pinned to the current processor or a processor that
1083 * is not online.
1084 */
1085static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086{
Nick Pigginc54ad302006-01-06 00:10:56 -08001087 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07001090 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 struct per_cpu_pageset *pset;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001092 struct per_cpu_pages *pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001094 local_irq_save(flags);
1095 pset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001097 pcp = &pset->pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001098 free_pcppages_bulk(zone, pcp->count, pcp);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001099 pcp->count = 0;
1100 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 }
1102}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001104/*
1105 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1106 */
1107void drain_local_pages(void *arg)
1108{
1109 drain_pages(smp_processor_id());
1110}
1111
1112/*
1113 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
1114 */
1115void drain_all_pages(void)
1116{
Jens Axboe15c8b6c2008-05-09 09:39:44 +02001117 on_each_cpu(drain_local_pages, NULL, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001118}
1119
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02001120#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121
1122void mark_free_pages(struct zone *zone)
1123{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001124 unsigned long pfn, max_zone_pfn;
1125 unsigned long flags;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001126 int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 struct list_head *curr;
1128
1129 if (!zone->spanned_pages)
1130 return;
1131
1132 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001133
1134 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1135 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1136 if (pfn_valid(pfn)) {
1137 struct page *page = pfn_to_page(pfn);
1138
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001139 if (!swsusp_page_is_forbidden(page))
1140 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001141 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001143 for_each_migratetype_order(order, t) {
1144 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001145 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001147 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1148 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001149 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001150 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001151 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 spin_unlock_irqrestore(&zone->lock, flags);
1153}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001154#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155
1156/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 * Free a 0-order page
Li Hongfc916682010-03-05 13:41:54 -08001158 * cold == 1 ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 */
Li Hongfc916682010-03-05 13:41:54 -08001160void free_hot_cold_page(struct page *page, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161{
1162 struct zone *zone = page_zone(page);
1163 struct per_cpu_pages *pcp;
1164 unsigned long flags;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001165 int migratetype;
Johannes Weiner451ea252009-09-21 17:01:48 -07001166 int wasMlocked = __TestClearPageMlocked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001168 if (!free_pages_prepare(page, 0))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001169 return;
1170
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001171 migratetype = get_pageblock_migratetype(page);
1172 set_page_private(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +02001174 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -07001175 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001176 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001177
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001178 /*
1179 * We only track unmovable, reclaimable and movable on pcp lists.
1180 * Free ISOLATE pages back to the allocator because they are being
1181 * offlined but treat RESERVE as movable pages so we can get those
1182 * areas back if necessary. Otherwise, we may have to free
1183 * excessively into the page allocator
1184 */
1185 if (migratetype >= MIGRATE_PCPTYPES) {
1186 if (unlikely(migratetype == MIGRATE_ISOLATE)) {
1187 free_one_page(zone, page, 0, migratetype);
1188 goto out;
1189 }
1190 migratetype = MIGRATE_MOVABLE;
1191 }
1192
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001193 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001194 if (cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001195 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001196 else
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001197 list_add(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001199 if (pcp->count >= pcp->high) {
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001200 free_pcppages_bulk(zone, pcp->batch, pcp);
Nick Piggin48db57f2006-01-08 01:00:42 -08001201 pcp->count -= pcp->batch;
1202 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001203
1204out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206}
1207
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001208/*
1209 * split_page takes a non-compound higher-order page, and splits it into
1210 * n (1<<order) sub-pages: page[0..n]
1211 * Each sub-page must be freed individually.
1212 *
1213 * Note: this is probably too low level an operation for use in drivers.
1214 * Please consult with lkml before using this in your driver.
1215 */
1216void split_page(struct page *page, unsigned int order)
1217{
1218 int i;
1219
Nick Piggin725d7042006-09-25 23:30:55 -07001220 VM_BUG_ON(PageCompound(page));
1221 VM_BUG_ON(!page_count(page));
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001222
1223#ifdef CONFIG_KMEMCHECK
1224 /*
1225 * Split shadow pages too, because free(page[0]) would
1226 * otherwise free the whole shadow.
1227 */
1228 if (kmemcheck_page_is_tracked(page))
1229 split_page(virt_to_page(page[0].shadow), order);
1230#endif
1231
Nick Piggin7835e982006-03-22 00:08:40 -08001232 for (i = 1; i < (1 << order); i++)
1233 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001234}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001235
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236/*
Mel Gorman748446b2010-05-24 14:32:27 -07001237 * Similar to split_page except the page is already free. As this is only
1238 * being used for migration, the migratetype of the block also changes.
1239 * As this is called with interrupts disabled, the caller is responsible
1240 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1241 * are enabled.
1242 *
1243 * Note: this is probably too low level an operation for use in drivers.
1244 * Please consult with lkml before using this in your driver.
1245 */
1246int split_free_page(struct page *page)
1247{
1248 unsigned int order;
1249 unsigned long watermark;
1250 struct zone *zone;
1251
1252 BUG_ON(!PageBuddy(page));
1253
1254 zone = page_zone(page);
1255 order = page_order(page);
1256
1257 /* Obey watermarks as if the page was being allocated */
1258 watermark = low_wmark_pages(zone) + (1 << order);
1259 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1260 return 0;
1261
1262 /* Remove page from free list */
1263 list_del(&page->lru);
1264 zone->free_area[order].nr_free--;
1265 rmv_page_order(page);
1266 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1UL << order));
1267
1268 /* Split into individual pages */
1269 set_page_refcounted(page);
1270 split_page(page, order);
1271
1272 if (order >= pageblock_order - 1) {
1273 struct page *endpage = page + (1 << order) - 1;
1274 for (; page < endpage; page += pageblock_nr_pages)
1275 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1276 }
1277
1278 return 1 << order;
1279}
1280
1281/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1283 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1284 * or two.
1285 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07001286static inline
1287struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001288 struct zone *zone, int order, gfp_t gfp_flags,
1289 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290{
1291 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001292 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 int cold = !!(gfp_flags & __GFP_COLD);
1294
Hugh Dickins689bceb2005-11-21 21:32:20 -08001295again:
Nick Piggin48db57f2006-01-08 01:00:42 -08001296 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001298 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 local_irq_save(flags);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001301 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1302 list = &pcp->lists[migratetype];
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001303 if (list_empty(list)) {
Mel Gorman535131e2007-10-16 01:25:49 -07001304 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001305 pcp->batch, list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001306 migratetype, cold);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001307 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07001308 goto failed;
Mel Gorman535131e2007-10-16 01:25:49 -07001309 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001310
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001311 if (cold)
1312 page = list_entry(list->prev, struct page, lru);
1313 else
1314 page = list_entry(list->next, struct page, lru);
1315
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001316 list_del(&page->lru);
1317 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001318 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07001319 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1320 /*
1321 * __GFP_NOFAIL is not to be used in new code.
1322 *
1323 * All __GFP_NOFAIL callers should be fixed so that they
1324 * properly detect and handle allocation failures.
1325 *
1326 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07001327 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07001328 * __GFP_NOFAIL.
1329 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07001330 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07001331 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001333 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001334 spin_unlock(&zone->lock);
1335 if (!page)
1336 goto failed;
KOSAKI Motohiro6ccf80e2010-01-15 17:01:18 -08001337 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 }
1339
Christoph Lameterf8891e52006-06-30 01:55:45 -07001340 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Mel Gorman18ea7e72008-04-28 02:12:14 -07001341 zone_statistics(preferred_zone, zone);
Nick Piggina74609f2006-01-06 00:11:20 -08001342 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343
Nick Piggin725d7042006-09-25 23:30:55 -07001344 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -08001345 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001346 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001348
1349failed:
1350 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001351 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352}
1353
Mel Gorman41858962009-06-16 15:32:12 -07001354/* The ALLOC_WMARK bits are used as an index to zone->watermark */
1355#define ALLOC_WMARK_MIN WMARK_MIN
1356#define ALLOC_WMARK_LOW WMARK_LOW
1357#define ALLOC_WMARK_HIGH WMARK_HIGH
1358#define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1359
1360/* Mask to get the watermark bits */
1361#define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1362
Nick Piggin31488902005-11-28 13:44:03 -08001363#define ALLOC_HARDER 0x10 /* try to alloc harder */
1364#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1365#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001366
Akinobu Mita933e3122006-12-08 02:39:45 -08001367#ifdef CONFIG_FAIL_PAGE_ALLOC
1368
1369static struct fail_page_alloc_attr {
1370 struct fault_attr attr;
1371
1372 u32 ignore_gfp_highmem;
1373 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001374 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001375
1376#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1377
1378 struct dentry *ignore_gfp_highmem_file;
1379 struct dentry *ignore_gfp_wait_file;
Akinobu Mita54114992007-07-15 23:40:23 -07001380 struct dentry *min_order_file;
Akinobu Mita933e3122006-12-08 02:39:45 -08001381
1382#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1383
1384} fail_page_alloc = {
1385 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001386 .ignore_gfp_wait = 1,
1387 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001388 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001389};
1390
1391static int __init setup_fail_page_alloc(char *str)
1392{
1393 return setup_fault_attr(&fail_page_alloc.attr, str);
1394}
1395__setup("fail_page_alloc=", setup_fail_page_alloc);
1396
1397static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1398{
Akinobu Mita54114992007-07-15 23:40:23 -07001399 if (order < fail_page_alloc.min_order)
1400 return 0;
Akinobu Mita933e3122006-12-08 02:39:45 -08001401 if (gfp_mask & __GFP_NOFAIL)
1402 return 0;
1403 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1404 return 0;
1405 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1406 return 0;
1407
1408 return should_fail(&fail_page_alloc.attr, 1 << order);
1409}
1410
1411#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1412
1413static int __init fail_page_alloc_debugfs(void)
1414{
1415 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1416 struct dentry *dir;
1417 int err;
1418
1419 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1420 "fail_page_alloc");
1421 if (err)
1422 return err;
1423 dir = fail_page_alloc.attr.dentries.dir;
1424
1425 fail_page_alloc.ignore_gfp_wait_file =
1426 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1427 &fail_page_alloc.ignore_gfp_wait);
1428
1429 fail_page_alloc.ignore_gfp_highmem_file =
1430 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1431 &fail_page_alloc.ignore_gfp_highmem);
Akinobu Mita54114992007-07-15 23:40:23 -07001432 fail_page_alloc.min_order_file =
1433 debugfs_create_u32("min-order", mode, dir,
1434 &fail_page_alloc.min_order);
Akinobu Mita933e3122006-12-08 02:39:45 -08001435
1436 if (!fail_page_alloc.ignore_gfp_wait_file ||
Akinobu Mita54114992007-07-15 23:40:23 -07001437 !fail_page_alloc.ignore_gfp_highmem_file ||
1438 !fail_page_alloc.min_order_file) {
Akinobu Mita933e3122006-12-08 02:39:45 -08001439 err = -ENOMEM;
1440 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1441 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
Akinobu Mita54114992007-07-15 23:40:23 -07001442 debugfs_remove(fail_page_alloc.min_order_file);
Akinobu Mita933e3122006-12-08 02:39:45 -08001443 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1444 }
1445
1446 return err;
1447}
1448
1449late_initcall(fail_page_alloc_debugfs);
1450
1451#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1452
1453#else /* CONFIG_FAIL_PAGE_ALLOC */
1454
1455static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1456{
1457 return 0;
1458}
1459
1460#endif /* CONFIG_FAIL_PAGE_ALLOC */
1461
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462/*
1463 * Return 1 if free pages are above 'mark'. This takes into account the order
1464 * of the allocation.
1465 */
1466int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001467 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468{
1469 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001470 long min = mark;
Christoph Lameteraa454842010-09-09 16:38:17 -07001471 long free_pages = zone_nr_free_pages(z) - (1 << order) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 int o;
1473
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001474 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001476 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 min -= min / 4;
1478
1479 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1480 return 0;
1481 for (o = 0; o < order; o++) {
1482 /* At the next order, this order's pages become unavailable */
1483 free_pages -= z->free_area[o].nr_free << o;
1484
1485 /* Require fewer higher order pages to be free */
1486 min >>= 1;
1487
1488 if (free_pages <= min)
1489 return 0;
1490 }
1491 return 1;
1492}
1493
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001494#ifdef CONFIG_NUMA
1495/*
1496 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1497 * skip over zones that are not allowed by the cpuset, or that have
1498 * been recently (in last second) found to be nearly full. See further
1499 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001500 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001501 *
1502 * If the zonelist cache is present in the passed in zonelist, then
1503 * returns a pointer to the allowed node mask (either the current
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001504 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001505 *
1506 * If the zonelist cache is not available for this zonelist, does
1507 * nothing and returns NULL.
1508 *
1509 * If the fullzones BITMAP in the zonelist cache is stale (more than
1510 * a second since last zap'd) then we zap it out (clear its bits.)
1511 *
1512 * We hold off even calling zlc_setup, until after we've checked the
1513 * first zone in the zonelist, on the theory that most allocations will
1514 * be satisfied from that first zone, so best to examine that zone as
1515 * quickly as we can.
1516 */
1517static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1518{
1519 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1520 nodemask_t *allowednodes; /* zonelist_cache approximation */
1521
1522 zlc = zonelist->zlcache_ptr;
1523 if (!zlc)
1524 return NULL;
1525
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001526 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001527 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1528 zlc->last_full_zap = jiffies;
1529 }
1530
1531 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1532 &cpuset_current_mems_allowed :
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001533 &node_states[N_HIGH_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001534 return allowednodes;
1535}
1536
1537/*
1538 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1539 * if it is worth looking at further for free memory:
1540 * 1) Check that the zone isn't thought to be full (doesn't have its
1541 * bit set in the zonelist_cache fullzones BITMAP).
1542 * 2) Check that the zones node (obtained from the zonelist_cache
1543 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1544 * Return true (non-zero) if zone is worth looking at further, or
1545 * else return false (zero) if it is not.
1546 *
1547 * This check -ignores- the distinction between various watermarks,
1548 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1549 * found to be full for any variation of these watermarks, it will
1550 * be considered full for up to one second by all requests, unless
1551 * we are so low on memory on all allowed nodes that we are forced
1552 * into the second scan of the zonelist.
1553 *
1554 * In the second scan we ignore this zonelist cache and exactly
1555 * apply the watermarks to all zones, even it is slower to do so.
1556 * We are low on memory in the second scan, and should leave no stone
1557 * unturned looking for a free page.
1558 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001559static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001560 nodemask_t *allowednodes)
1561{
1562 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1563 int i; /* index of *z in zonelist zones */
1564 int n; /* node that zone *z is on */
1565
1566 zlc = zonelist->zlcache_ptr;
1567 if (!zlc)
1568 return 1;
1569
Mel Gormandd1a2392008-04-28 02:12:17 -07001570 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001571 n = zlc->z_to_n[i];
1572
1573 /* This zone is worth trying if it is allowed but not full */
1574 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1575}
1576
1577/*
1578 * Given 'z' scanning a zonelist, set the corresponding bit in
1579 * zlc->fullzones, so that subsequent attempts to allocate a page
1580 * from that zone don't waste time re-examining it.
1581 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001582static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001583{
1584 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1585 int i; /* index of *z in zonelist zones */
1586
1587 zlc = zonelist->zlcache_ptr;
1588 if (!zlc)
1589 return;
1590
Mel Gormandd1a2392008-04-28 02:12:17 -07001591 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001592
1593 set_bit(i, zlc->fullzones);
1594}
1595
1596#else /* CONFIG_NUMA */
1597
1598static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1599{
1600 return NULL;
1601}
1602
Mel Gormandd1a2392008-04-28 02:12:17 -07001603static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001604 nodemask_t *allowednodes)
1605{
1606 return 1;
1607}
1608
Mel Gormandd1a2392008-04-28 02:12:17 -07001609static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001610{
1611}
1612#endif /* CONFIG_NUMA */
1613
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001614/*
Paul Jackson0798e512006-12-06 20:31:38 -08001615 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001616 * a page.
1617 */
1618static struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001619get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
Mel Gorman5117f452009-06-16 15:31:59 -07001620 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
Mel Gorman3dd28262009-06-16 15:32:00 -07001621 struct zone *preferred_zone, int migratetype)
Martin Hicks753ee722005-06-21 17:14:41 -07001622{
Mel Gormandd1a2392008-04-28 02:12:17 -07001623 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001624 struct page *page = NULL;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001625 int classzone_idx;
Mel Gorman5117f452009-06-16 15:31:59 -07001626 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001627 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1628 int zlc_active = 0; /* set if using zonelist_cache */
1629 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gorman54a6eb52008-04-28 02:12:16 -07001630
Mel Gorman19770b32008-04-28 02:12:18 -07001631 classzone_idx = zone_idx(preferred_zone);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001632zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001633 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001634 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001635 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1636 */
Mel Gorman19770b32008-04-28 02:12:18 -07001637 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1638 high_zoneidx, nodemask) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001639 if (NUMA_BUILD && zlc_active &&
1640 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1641 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001642 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001643 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001644 goto try_next_zone;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001645
Mel Gorman41858962009-06-16 15:32:12 -07001646 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001647 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001648 unsigned long mark;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001649 int ret;
1650
Mel Gorman41858962009-06-16 15:32:12 -07001651 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
Mel Gormanfa5e0842009-06-16 15:33:22 -07001652 if (zone_watermark_ok(zone, order, mark,
1653 classzone_idx, alloc_flags))
1654 goto try_this_zone;
1655
1656 if (zone_reclaim_mode == 0)
1657 goto this_zone_full;
1658
1659 ret = zone_reclaim(zone, gfp_mask, order);
1660 switch (ret) {
1661 case ZONE_RECLAIM_NOSCAN:
1662 /* did not scan */
1663 goto try_next_zone;
1664 case ZONE_RECLAIM_FULL:
1665 /* scanned but unreclaimable */
1666 goto this_zone_full;
1667 default:
1668 /* did we reclaim enough */
1669 if (!zone_watermark_ok(zone, order, mark,
1670 classzone_idx, alloc_flags))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001671 goto this_zone_full;
Paul Jackson0798e512006-12-06 20:31:38 -08001672 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001673 }
1674
Mel Gormanfa5e0842009-06-16 15:33:22 -07001675try_this_zone:
Mel Gorman3dd28262009-06-16 15:32:00 -07001676 page = buffered_rmqueue(preferred_zone, zone, order,
1677 gfp_mask, migratetype);
Paul Jackson0798e512006-12-06 20:31:38 -08001678 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001679 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001680this_zone_full:
1681 if (NUMA_BUILD)
1682 zlc_mark_zone_full(zonelist, z);
1683try_next_zone:
Christoph Lameter62bc62a2009-06-16 15:32:15 -07001684 if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) {
Mel Gormand395b732009-06-16 15:32:09 -07001685 /*
1686 * we do zlc_setup after the first zone is tried but only
1687 * if there are multiple nodes make it worthwhile
1688 */
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001689 allowednodes = zlc_setup(zonelist, alloc_flags);
1690 zlc_active = 1;
1691 did_zlc_setup = 1;
1692 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07001693 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001694
1695 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1696 /* Disable zlc cache for second zonelist scan */
1697 zlc_active = 0;
1698 goto zonelist_scan;
1699 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001700 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07001701}
1702
Mel Gorman11e33f62009-06-16 15:31:57 -07001703static inline int
1704should_alloc_retry(gfp_t gfp_mask, unsigned int order,
1705 unsigned long pages_reclaimed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706{
Mel Gorman11e33f62009-06-16 15:31:57 -07001707 /* Do not loop if specifically requested */
1708 if (gfp_mask & __GFP_NORETRY)
1709 return 0;
1710
1711 /*
1712 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1713 * means __GFP_NOFAIL, but that may not be true in other
1714 * implementations.
1715 */
1716 if (order <= PAGE_ALLOC_COSTLY_ORDER)
1717 return 1;
1718
1719 /*
1720 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1721 * specified, then we retry until we no longer reclaim any pages
1722 * (above), or we've reclaimed an order of pages at least as
1723 * large as the allocation's order. In both cases, if the
1724 * allocation still fails, we stop retrying.
1725 */
1726 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
1727 return 1;
1728
1729 /*
1730 * Don't let big-order allocations loop unless the caller
1731 * explicitly requests that.
1732 */
1733 if (gfp_mask & __GFP_NOFAIL)
1734 return 1;
1735
1736 return 0;
1737}
1738
1739static inline struct page *
1740__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1741 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001742 nodemask_t *nodemask, struct zone *preferred_zone,
1743 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001744{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746
Mel Gorman11e33f62009-06-16 15:31:57 -07001747 /* Acquire the OOM killer lock for the zones in zonelist */
Minchan Kimff321fe2010-08-09 17:18:57 -07001748 if (!try_set_zonelist_oom(zonelist, gfp_mask)) {
Mel Gorman11e33f62009-06-16 15:31:57 -07001749 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 return NULL;
1751 }
Jens Axboe6b1de912005-11-17 21:35:02 +01001752
Mel Gorman11e33f62009-06-16 15:31:57 -07001753 /*
1754 * Go through the zonelist yet one more time, keep very high watermark
1755 * here, this is only to catch a parallel oom killing, we must fail if
1756 * we're still under heavy pressure.
1757 */
1758 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1759 order, zonelist, high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001760 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07001761 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001762 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07001763 goto out;
1764
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001765 if (!(gfp_mask & __GFP_NOFAIL)) {
1766 /* The OOM killer will not help higher order allocs */
1767 if (order > PAGE_ALLOC_COSTLY_ORDER)
1768 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07001769 /* The OOM killer does not needlessly kill tasks for lowmem */
1770 if (high_zoneidx < ZONE_NORMAL)
1771 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001772 /*
1773 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
1774 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
1775 * The caller should handle page allocation failure by itself if
1776 * it specifies __GFP_THISNODE.
1777 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
1778 */
1779 if (gfp_mask & __GFP_THISNODE)
1780 goto out;
1781 }
Mel Gorman11e33f62009-06-16 15:31:57 -07001782 /* Exhausted what can be done so it's blamo time */
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001783 out_of_memory(zonelist, gfp_mask, order, nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07001784
1785out:
1786 clear_zonelist_oom(zonelist, gfp_mask);
1787 return page;
1788}
1789
Mel Gorman56de7262010-05-24 14:32:30 -07001790#ifdef CONFIG_COMPACTION
1791/* Try memory compaction for high-order allocations before reclaim */
1792static struct page *
1793__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
1794 struct zonelist *zonelist, enum zone_type high_zoneidx,
1795 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
1796 int migratetype, unsigned long *did_some_progress)
1797{
1798 struct page *page;
1799
Mel Gorman4f92e252010-05-24 14:32:32 -07001800 if (!order || compaction_deferred(preferred_zone))
Mel Gorman56de7262010-05-24 14:32:30 -07001801 return NULL;
1802
1803 *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
1804 nodemask);
1805 if (*did_some_progress != COMPACT_SKIPPED) {
1806
1807 /* Page migration frees to the PCP lists but we want merging */
1808 drain_pages(get_cpu());
1809 put_cpu();
1810
1811 page = get_page_from_freelist(gfp_mask, nodemask,
1812 order, zonelist, high_zoneidx,
1813 alloc_flags, preferred_zone,
1814 migratetype);
1815 if (page) {
Mel Gorman4f92e252010-05-24 14:32:32 -07001816 preferred_zone->compact_considered = 0;
1817 preferred_zone->compact_defer_shift = 0;
Mel Gorman56de7262010-05-24 14:32:30 -07001818 count_vm_event(COMPACTSUCCESS);
1819 return page;
1820 }
1821
1822 /*
1823 * It's bad if compaction run occurs and fails.
1824 * The most likely reason is that pages exist,
1825 * but not enough to satisfy watermarks.
1826 */
1827 count_vm_event(COMPACTFAIL);
Mel Gorman4f92e252010-05-24 14:32:32 -07001828 defer_compaction(preferred_zone);
Mel Gorman56de7262010-05-24 14:32:30 -07001829
1830 cond_resched();
1831 }
1832
1833 return NULL;
1834}
1835#else
1836static inline struct page *
1837__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
1838 struct zonelist *zonelist, enum zone_type high_zoneidx,
1839 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
1840 int migratetype, unsigned long *did_some_progress)
1841{
1842 return NULL;
1843}
1844#endif /* CONFIG_COMPACTION */
1845
Mel Gorman11e33f62009-06-16 15:31:57 -07001846/* The really slow allocator path where we enter direct reclaim */
1847static inline struct page *
1848__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
1849 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001850 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001851 int migratetype, unsigned long *did_some_progress)
Mel Gorman11e33f62009-06-16 15:31:57 -07001852{
1853 struct page *page = NULL;
1854 struct reclaim_state reclaim_state;
1855 struct task_struct *p = current;
Mel Gorman9ee493c2010-09-09 16:38:18 -07001856 bool drained = false;
Mel Gorman11e33f62009-06-16 15:31:57 -07001857
1858 cond_resched();
1859
1860 /* We now go into synchronous reclaim */
1861 cpuset_memory_pressure_bump();
Mel Gorman11e33f62009-06-16 15:31:57 -07001862 p->flags |= PF_MEMALLOC;
1863 lockdep_set_current_reclaim_state(gfp_mask);
1864 reclaim_state.reclaimed_slab = 0;
1865 p->reclaim_state = &reclaim_state;
1866
1867 *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
1868
1869 p->reclaim_state = NULL;
1870 lockdep_clear_current_reclaim_state();
1871 p->flags &= ~PF_MEMALLOC;
1872
1873 cond_resched();
1874
Mel Gorman9ee493c2010-09-09 16:38:18 -07001875 if (unlikely(!(*did_some_progress)))
1876 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07001877
Mel Gorman9ee493c2010-09-09 16:38:18 -07001878retry:
1879 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001880 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001881 alloc_flags, preferred_zone,
1882 migratetype);
Mel Gorman9ee493c2010-09-09 16:38:18 -07001883
1884 /*
1885 * If an allocation failed after direct reclaim, it could be because
1886 * pages are pinned on the per-cpu lists. Drain them and try again
1887 */
1888 if (!page && !drained) {
1889 drain_all_pages();
1890 drained = true;
1891 goto retry;
1892 }
1893
Mel Gorman11e33f62009-06-16 15:31:57 -07001894 return page;
1895}
1896
Mel Gorman11e33f62009-06-16 15:31:57 -07001897/*
1898 * This is called in the allocator slow-path if the allocation request is of
1899 * sufficient urgency to ignore watermarks and take other desperate measures
1900 */
1901static inline struct page *
1902__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
1903 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001904 nodemask_t *nodemask, struct zone *preferred_zone,
1905 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001906{
1907 struct page *page;
1908
1909 do {
1910 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001911 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
Mel Gorman3dd28262009-06-16 15:32:00 -07001912 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001913
1914 if (!page && gfp_mask & __GFP_NOFAIL)
Mel Gorman0e093d992010-10-26 14:21:45 -07001915 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07001916 } while (!page && (gfp_mask & __GFP_NOFAIL));
1917
1918 return page;
1919}
1920
1921static inline
1922void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
1923 enum zone_type high_zoneidx)
1924{
1925 struct zoneref *z;
1926 struct zone *zone;
1927
1928 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
1929 wakeup_kswapd(zone, order);
1930}
1931
Peter Zijlstra341ce062009-06-16 15:32:02 -07001932static inline int
1933gfp_to_alloc_flags(gfp_t gfp_mask)
1934{
1935 struct task_struct *p = current;
1936 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
1937 const gfp_t wait = gfp_mask & __GFP_WAIT;
1938
Mel Gormana56f57f2009-06-16 15:32:02 -07001939 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07001940 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07001941
Peter Zijlstra341ce062009-06-16 15:32:02 -07001942 /*
1943 * The caller may dip into page reserves a bit more if the caller
1944 * cannot run direct reclaim, or if the caller has realtime scheduling
1945 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1946 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1947 */
Namhyung Kime6223a32010-10-26 14:21:59 -07001948 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07001949
1950 if (!wait) {
1951 alloc_flags |= ALLOC_HARDER;
1952 /*
1953 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1954 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1955 */
1956 alloc_flags &= ~ALLOC_CPUSET;
Mel Gorman9d0ed602009-11-11 14:26:17 -08001957 } else if (unlikely(rt_task(p)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07001958 alloc_flags |= ALLOC_HARDER;
1959
1960 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
1961 if (!in_interrupt() &&
1962 ((p->flags & PF_MEMALLOC) ||
1963 unlikely(test_thread_flag(TIF_MEMDIE))))
1964 alloc_flags |= ALLOC_NO_WATERMARKS;
1965 }
1966
1967 return alloc_flags;
1968}
1969
Mel Gorman11e33f62009-06-16 15:31:57 -07001970static inline struct page *
1971__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
1972 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001973 nodemask_t *nodemask, struct zone *preferred_zone,
1974 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001975{
1976 const gfp_t wait = gfp_mask & __GFP_WAIT;
1977 struct page *page = NULL;
1978 int alloc_flags;
1979 unsigned long pages_reclaimed = 0;
1980 unsigned long did_some_progress;
1981 struct task_struct *p = current;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001982
Christoph Lameter952f3b52006-12-06 20:33:26 -08001983 /*
Mel Gorman72807a72009-06-16 15:32:18 -07001984 * In the slowpath, we sanity check order to avoid ever trying to
1985 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
1986 * be using allocators in order of preference for an area that is
1987 * too large.
1988 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07001989 if (order >= MAX_ORDER) {
1990 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07001991 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07001992 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001993
Christoph Lameter952f3b52006-12-06 20:33:26 -08001994 /*
1995 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1996 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1997 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1998 * using a larger set of nodes after it has established that the
1999 * allowed per node queues are empty and that nodes are
2000 * over allocated.
2001 */
2002 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
2003 goto nopage;
2004
Mel Gormancc4a6852009-11-11 14:26:14 -08002005restart:
Mel Gorman11e33f62009-06-16 15:31:57 -07002006 wake_all_kswapd(order, zonelist, high_zoneidx);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002007
Paul Jackson9bf22292005-09-06 15:18:12 -07002008 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002009 * OK, we're below the kswapd watermark and have kicked background
2010 * reclaim. Now things get more complex, so set up alloc_flags according
2011 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07002012 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002013 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014
Peter Zijlstra341ce062009-06-16 15:32:02 -07002015 /* This is the last chance, in general, before the goto nopage. */
Mel Gorman19770b32008-04-28 02:12:18 -07002016 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
Peter Zijlstra341ce062009-06-16 15:32:02 -07002017 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
2018 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002019 if (page)
2020 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021
Kirill Korotaevb43a57b2006-12-06 20:32:27 -08002022rebalance:
Mel Gorman11e33f62009-06-16 15:31:57 -07002023 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002024 if (alloc_flags & ALLOC_NO_WATERMARKS) {
2025 page = __alloc_pages_high_priority(gfp_mask, order,
2026 zonelist, high_zoneidx, nodemask,
2027 preferred_zone, migratetype);
2028 if (page)
2029 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 }
2031
2032 /* Atomic allocations - we can't balance anything */
2033 if (!wait)
2034 goto nopage;
2035
Peter Zijlstra341ce062009-06-16 15:32:02 -07002036 /* Avoid recursion of direct reclaim */
2037 if (p->flags & PF_MEMALLOC)
2038 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039
David Rientjes6583bb62009-07-29 15:02:06 -07002040 /* Avoid allocations with no watermarks from looping endlessly */
2041 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2042 goto nopage;
2043
Mel Gorman56de7262010-05-24 14:32:30 -07002044 /* Try direct compaction */
2045 page = __alloc_pages_direct_compact(gfp_mask, order,
2046 zonelist, high_zoneidx,
2047 nodemask,
2048 alloc_flags, preferred_zone,
2049 migratetype, &did_some_progress);
2050 if (page)
2051 goto got_pg;
2052
Mel Gorman11e33f62009-06-16 15:31:57 -07002053 /* Try direct reclaim and then allocating */
2054 page = __alloc_pages_direct_reclaim(gfp_mask, order,
2055 zonelist, high_zoneidx,
2056 nodemask,
Mel Gorman5117f452009-06-16 15:31:59 -07002057 alloc_flags, preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07002058 migratetype, &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07002059 if (page)
2060 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 /*
Mel Gorman11e33f62009-06-16 15:31:57 -07002063 * If we failed to make any progress reclaiming, then we are
2064 * running out of options and have to consider going OOM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 */
Mel Gorman11e33f62009-06-16 15:31:57 -07002066 if (!did_some_progress) {
2067 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -07002068 if (oom_killer_disabled)
2069 goto nopage;
Mel Gorman11e33f62009-06-16 15:31:57 -07002070 page = __alloc_pages_may_oom(gfp_mask, order,
2071 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002072 nodemask, preferred_zone,
2073 migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002074 if (page)
2075 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076
David Rientjes03668b32010-08-09 17:18:54 -07002077 if (!(gfp_mask & __GFP_NOFAIL)) {
2078 /*
2079 * The oom killer is not called for high-order
2080 * allocations that may fail, so if no progress
2081 * is being made, there are no other options and
2082 * retrying is unlikely to help.
2083 */
2084 if (order > PAGE_ALLOC_COSTLY_ORDER)
2085 goto nopage;
2086 /*
2087 * The oom killer is not called for lowmem
2088 * allocations to prevent needlessly killing
2089 * innocent tasks.
2090 */
2091 if (high_zoneidx < ZONE_NORMAL)
2092 goto nopage;
2093 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 goto restart;
2096 }
2097 }
2098
Mel Gorman11e33f62009-06-16 15:31:57 -07002099 /* Check if we should retry the allocation */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07002100 pages_reclaimed += did_some_progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07002101 if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
2102 /* Wait for some write requests to complete then retry */
Mel Gorman0e093d992010-10-26 14:21:45 -07002103 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104 goto rebalance;
2105 }
2106
2107nopage:
2108 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
2109 printk(KERN_WARNING "%s: page allocation failure."
2110 " order:%d, mode:0x%x\n",
2111 p->comm, order, gfp_mask);
2112 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07002113 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 }
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002115 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116got_pg:
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002117 if (kmemcheck_enabled)
2118 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 return page;
Mel Gorman11e33f62009-06-16 15:31:57 -07002120
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121}
Mel Gorman11e33f62009-06-16 15:31:57 -07002122
2123/*
2124 * This is the 'heart' of the zoned buddy allocator.
2125 */
2126struct page *
2127__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2128 struct zonelist *zonelist, nodemask_t *nodemask)
2129{
2130 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Mel Gorman5117f452009-06-16 15:31:59 -07002131 struct zone *preferred_zone;
Mel Gorman11e33f62009-06-16 15:31:57 -07002132 struct page *page;
Mel Gorman3dd28262009-06-16 15:32:00 -07002133 int migratetype = allocflags_to_migratetype(gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002134
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10002135 gfp_mask &= gfp_allowed_mask;
2136
Mel Gorman11e33f62009-06-16 15:31:57 -07002137 lockdep_trace_alloc(gfp_mask);
2138
2139 might_sleep_if(gfp_mask & __GFP_WAIT);
2140
2141 if (should_fail_alloc_page(gfp_mask, order))
2142 return NULL;
2143
2144 /*
2145 * Check the zones suitable for the gfp_mask contain at least one
2146 * valid zone. It's possible to have an empty zonelist as a result
2147 * of GFP_THISNODE and a memoryless node
2148 */
2149 if (unlikely(!zonelist->_zonerefs->zone))
2150 return NULL;
2151
Miao Xiec0ff7452010-05-24 14:32:08 -07002152 get_mems_allowed();
Mel Gorman5117f452009-06-16 15:31:59 -07002153 /* The preferred zone is used for statistics later */
2154 first_zones_zonelist(zonelist, high_zoneidx, nodemask, &preferred_zone);
Miao Xiec0ff7452010-05-24 14:32:08 -07002155 if (!preferred_zone) {
2156 put_mems_allowed();
Mel Gorman5117f452009-06-16 15:31:59 -07002157 return NULL;
Miao Xiec0ff7452010-05-24 14:32:08 -07002158 }
Mel Gorman5117f452009-06-16 15:31:59 -07002159
2160 /* First allocation attempt */
Mel Gorman11e33f62009-06-16 15:31:57 -07002161 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002162 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07002163 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002164 if (unlikely(!page))
2165 page = __alloc_pages_slowpath(gfp_mask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002166 zonelist, high_zoneidx, nodemask,
Mel Gorman3dd28262009-06-16 15:32:00 -07002167 preferred_zone, migratetype);
Miao Xiec0ff7452010-05-24 14:32:08 -07002168 put_mems_allowed();
Mel Gorman11e33f62009-06-16 15:31:57 -07002169
Mel Gorman4b4f2782009-09-21 17:02:41 -07002170 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002171 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172}
Mel Gormand2391712009-06-16 15:31:52 -07002173EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174
2175/*
2176 * Common helper functions.
2177 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08002178unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179{
Akinobu Mita945a1112009-09-21 17:01:47 -07002180 struct page *page;
2181
2182 /*
2183 * __get_free_pages() returns a 32-bit address, which cannot represent
2184 * a highmem page
2185 */
2186 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2187
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 page = alloc_pages(gfp_mask, order);
2189 if (!page)
2190 return 0;
2191 return (unsigned long) page_address(page);
2192}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193EXPORT_SYMBOL(__get_free_pages);
2194
Harvey Harrison920c7a52008-02-04 22:29:26 -08002195unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196{
Akinobu Mita945a1112009-09-21 17:01:47 -07002197 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199EXPORT_SYMBOL(get_zeroed_page);
2200
2201void __pagevec_free(struct pagevec *pvec)
2202{
2203 int i = pagevec_count(pvec);
2204
Mel Gorman4b4f2782009-09-21 17:02:41 -07002205 while (--i >= 0) {
2206 trace_mm_pagevec_free(pvec->pages[i], pvec->cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 free_hot_cold_page(pvec->pages[i], pvec->cold);
Mel Gorman4b4f2782009-09-21 17:02:41 -07002208 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209}
2210
Harvey Harrison920c7a52008-02-04 22:29:26 -08002211void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212{
Nick Pigginb5810032005-10-29 18:16:12 -07002213 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 if (order == 0)
Li Hongfc916682010-03-05 13:41:54 -08002215 free_hot_cold_page(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216 else
2217 __free_pages_ok(page, order);
2218 }
2219}
2220
2221EXPORT_SYMBOL(__free_pages);
2222
Harvey Harrison920c7a52008-02-04 22:29:26 -08002223void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224{
2225 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07002226 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227 __free_pages(virt_to_page((void *)addr), order);
2228 }
2229}
2230
2231EXPORT_SYMBOL(free_pages);
2232
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002233/**
2234 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2235 * @size: the number of bytes to allocate
2236 * @gfp_mask: GFP flags for the allocation
2237 *
2238 * This function is similar to alloc_pages(), except that it allocates the
2239 * minimum number of pages to satisfy the request. alloc_pages() can only
2240 * allocate memory in power-of-two pages.
2241 *
2242 * This function is also limited by MAX_ORDER.
2243 *
2244 * Memory allocated by this function must be released by free_pages_exact().
2245 */
2246void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2247{
2248 unsigned int order = get_order(size);
2249 unsigned long addr;
2250
2251 addr = __get_free_pages(gfp_mask, order);
2252 if (addr) {
2253 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2254 unsigned long used = addr + PAGE_ALIGN(size);
2255
Kevin Cernekee5bfd7562009-07-05 12:08:19 -07002256 split_page(virt_to_page((void *)addr), order);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002257 while (used < alloc_end) {
2258 free_page(used);
2259 used += PAGE_SIZE;
2260 }
2261 }
2262
2263 return (void *)addr;
2264}
2265EXPORT_SYMBOL(alloc_pages_exact);
2266
2267/**
2268 * free_pages_exact - release memory allocated via alloc_pages_exact()
2269 * @virt: the value returned by alloc_pages_exact.
2270 * @size: size of allocation, same value as passed to alloc_pages_exact().
2271 *
2272 * Release the memory allocated by a previous call to alloc_pages_exact.
2273 */
2274void free_pages_exact(void *virt, size_t size)
2275{
2276 unsigned long addr = (unsigned long)virt;
2277 unsigned long end = addr + PAGE_ALIGN(size);
2278
2279 while (addr < end) {
2280 free_page(addr);
2281 addr += PAGE_SIZE;
2282 }
2283}
2284EXPORT_SYMBOL(free_pages_exact);
2285
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286static unsigned int nr_free_zone_pages(int offset)
2287{
Mel Gormandd1a2392008-04-28 02:12:17 -07002288 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002289 struct zone *zone;
2290
Martin J. Blighe310fd42005-07-29 22:59:18 -07002291 /* Just pick one node, since fallback list is circular */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 unsigned int sum = 0;
2293
Mel Gorman0e884602008-04-28 02:12:14 -07002294 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295
Mel Gorman54a6eb52008-04-28 02:12:16 -07002296 for_each_zone_zonelist(zone, z, zonelist, offset) {
Martin J. Blighe310fd42005-07-29 22:59:18 -07002297 unsigned long size = zone->present_pages;
Mel Gorman41858962009-06-16 15:32:12 -07002298 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07002299 if (size > high)
2300 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301 }
2302
2303 return sum;
2304}
2305
2306/*
2307 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2308 */
2309unsigned int nr_free_buffer_pages(void)
2310{
Al Viroaf4ca452005-10-21 02:55:38 -04002311 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312}
Meelap Shahc2f1a552007-07-17 04:04:39 -07002313EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314
2315/*
2316 * Amount of free RAM allocatable within all zones
2317 */
2318unsigned int nr_free_pagecache_pages(void)
2319{
Mel Gorman2a1e2742007-07-17 04:03:12 -07002320 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002322
2323static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324{
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002325 if (NUMA_BUILD)
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08002326 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329void si_meminfo(struct sysinfo *val)
2330{
2331 val->totalram = totalram_pages;
2332 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002333 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 val->totalhigh = totalhigh_pages;
2336 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337 val->mem_unit = PAGE_SIZE;
2338}
2339
2340EXPORT_SYMBOL(si_meminfo);
2341
2342#ifdef CONFIG_NUMA
2343void si_meminfo_node(struct sysinfo *val, int nid)
2344{
2345 pg_data_t *pgdat = NODE_DATA(nid);
2346
2347 val->totalram = pgdat->node_present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002348 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002349#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002351 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2352 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002353#else
2354 val->totalhigh = 0;
2355 val->freehigh = 0;
2356#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357 val->mem_unit = PAGE_SIZE;
2358}
2359#endif
2360
2361#define K(x) ((x) << (PAGE_SHIFT-10))
2362
2363/*
2364 * Show free area list (used inside shift_scroll-lock stuff)
2365 * We also calculate the percentage fragmentation. We do this by counting the
2366 * memory on each free list with the exception of the first item on the list.
2367 */
2368void show_free_areas(void)
2369{
Jes Sorensenc7241912006-09-27 01:50:05 -07002370 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371 struct zone *zone;
2372
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002373 for_each_populated_zone(zone) {
Jes Sorensenc7241912006-09-27 01:50:05 -07002374 show_node(zone);
2375 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376
Dave Jones6b482c62005-11-10 15:45:56 -05002377 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378 struct per_cpu_pageset *pageset;
2379
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002380 pageset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002382 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2383 cpu, pageset->pcp.high,
2384 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 }
2386 }
2387
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002388 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2389 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002390 " unevictable:%lu"
Andrew Mortonb76146e2009-10-26 16:49:52 -07002391 " dirty:%lu writeback:%lu unstable:%lu\n"
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07002392 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002393 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002394 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002395 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002396 global_page_state(NR_ISOLATED_ANON),
2397 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002398 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002399 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002400 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07002401 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07002402 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07002403 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08002404 global_page_state(NR_FREE_PAGES),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07002405 global_page_state(NR_SLAB_RECLAIMABLE),
2406 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07002407 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002408 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08002409 global_page_state(NR_PAGETABLE),
2410 global_page_state(NR_BOUNCE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002412 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 int i;
2414
2415 show_node(zone);
2416 printk("%s"
2417 " free:%lukB"
2418 " min:%lukB"
2419 " low:%lukB"
2420 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002421 " active_anon:%lukB"
2422 " inactive_anon:%lukB"
2423 " active_file:%lukB"
2424 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002425 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002426 " isolated(anon):%lukB"
2427 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428 " present:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002429 " mlocked:%lukB"
2430 " dirty:%lukB"
2431 " writeback:%lukB"
2432 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002433 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002434 " slab_reclaimable:%lukB"
2435 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07002436 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002437 " pagetables:%lukB"
2438 " unstable:%lukB"
2439 " bounce:%lukB"
2440 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441 " pages_scanned:%lu"
2442 " all_unreclaimable? %s"
2443 "\n",
2444 zone->name,
Christoph Lameteraa454842010-09-09 16:38:17 -07002445 K(zone_nr_free_pages(zone)),
Mel Gorman41858962009-06-16 15:32:12 -07002446 K(min_wmark_pages(zone)),
2447 K(low_wmark_pages(zone)),
2448 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002449 K(zone_page_state(zone, NR_ACTIVE_ANON)),
2450 K(zone_page_state(zone, NR_INACTIVE_ANON)),
2451 K(zone_page_state(zone, NR_ACTIVE_FILE)),
2452 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002453 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002454 K(zone_page_state(zone, NR_ISOLATED_ANON)),
2455 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456 K(zone->present_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002457 K(zone_page_state(zone, NR_MLOCK)),
2458 K(zone_page_state(zone, NR_FILE_DIRTY)),
2459 K(zone_page_state(zone, NR_WRITEBACK)),
2460 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002461 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002462 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
2463 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07002464 zone_page_state(zone, NR_KERNEL_STACK) *
2465 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002466 K(zone_page_state(zone, NR_PAGETABLE)),
2467 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
2468 K(zone_page_state(zone, NR_BOUNCE)),
2469 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470 zone->pages_scanned,
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -08002471 (zone->all_unreclaimable ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 );
2473 printk("lowmem_reserve[]:");
2474 for (i = 0; i < MAX_NR_ZONES; i++)
2475 printk(" %lu", zone->lowmem_reserve[i]);
2476 printk("\n");
2477 }
2478
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002479 for_each_populated_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002480 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481
2482 show_node(zone);
2483 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484
2485 spin_lock_irqsave(&zone->lock, flags);
2486 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002487 nr[order] = zone->free_area[order].nr_free;
2488 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489 }
2490 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002491 for (order = 0; order < MAX_ORDER; order++)
2492 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493 printk("= %lukB\n", K(total));
2494 }
2495
Larry Woodmane6f36022008-02-04 22:29:30 -08002496 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2497
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498 show_swap_cache_info();
2499}
2500
Mel Gorman19770b32008-04-28 02:12:18 -07002501static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2502{
2503 zoneref->zone = zone;
2504 zoneref->zone_idx = zone_idx(zone);
2505}
2506
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507/*
2508 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08002509 *
2510 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002512static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2513 int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514{
Christoph Lameter1a932052006-01-06 00:11:16 -08002515 struct zone *zone;
2516
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002517 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002518 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08002519
2520 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002521 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08002522 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08002523 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002524 zoneref_set_zone(zone,
2525 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08002526 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08002528
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002529 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08002530 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531}
2532
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002533
2534/*
2535 * zonelist_order:
2536 * 0 = automatic detection of better ordering.
2537 * 1 = order by ([node] distance, -zonetype)
2538 * 2 = order by (-zonetype, [node] distance)
2539 *
2540 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2541 * the same zonelist. So only NUMA can configure this param.
2542 */
2543#define ZONELIST_ORDER_DEFAULT 0
2544#define ZONELIST_ORDER_NODE 1
2545#define ZONELIST_ORDER_ZONE 2
2546
2547/* zonelist order in the kernel.
2548 * set_zonelist_order() will set this to NODE or ZONE.
2549 */
2550static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2551static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2552
2553
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002555/* The value user specified ....changed by config */
2556static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2557/* string for sysctl */
2558#define NUMA_ZONELIST_ORDER_LEN 16
2559char numa_zonelist_order[16] = "default";
2560
2561/*
2562 * interface for configure zonelist ordering.
2563 * command line option "numa_zonelist_order"
2564 * = "[dD]efault - default, automatic configuration.
2565 * = "[nN]ode - order by node locality, then by zone within node
2566 * = "[zZ]one - order by zone, then by locality within zone
2567 */
2568
2569static int __parse_numa_zonelist_order(char *s)
2570{
2571 if (*s == 'd' || *s == 'D') {
2572 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2573 } else if (*s == 'n' || *s == 'N') {
2574 user_zonelist_order = ZONELIST_ORDER_NODE;
2575 } else if (*s == 'z' || *s == 'Z') {
2576 user_zonelist_order = ZONELIST_ORDER_ZONE;
2577 } else {
2578 printk(KERN_WARNING
2579 "Ignoring invalid numa_zonelist_order value: "
2580 "%s\n", s);
2581 return -EINVAL;
2582 }
2583 return 0;
2584}
2585
2586static __init int setup_numa_zonelist_order(char *s)
2587{
2588 if (s)
2589 return __parse_numa_zonelist_order(s);
2590 return 0;
2591}
2592early_param("numa_zonelist_order", setup_numa_zonelist_order);
2593
2594/*
2595 * sysctl handler for numa_zonelist_order
2596 */
2597int numa_zonelist_order_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07002598 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002599 loff_t *ppos)
2600{
2601 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2602 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01002603 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002604
Andi Kleen443c6f12009-12-23 21:00:47 +01002605 mutex_lock(&zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002606 if (write)
Andi Kleen443c6f12009-12-23 21:00:47 +01002607 strcpy(saved_string, (char*)table->data);
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07002608 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002609 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01002610 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002611 if (write) {
2612 int oldval = user_zonelist_order;
2613 if (__parse_numa_zonelist_order((char*)table->data)) {
2614 /*
2615 * bogus value. restore saved string
2616 */
2617 strncpy((char*)table->data, saved_string,
2618 NUMA_ZONELIST_ORDER_LEN);
2619 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07002620 } else if (oldval != user_zonelist_order) {
2621 mutex_lock(&zonelists_mutex);
Haicheng Li1f522502010-05-24 14:32:51 -07002622 build_all_zonelists(NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07002623 mutex_unlock(&zonelists_mutex);
2624 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002625 }
Andi Kleen443c6f12009-12-23 21:00:47 +01002626out:
2627 mutex_unlock(&zl_order_mutex);
2628 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002629}
2630
2631
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002632#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002633static int node_load[MAX_NUMNODES];
2634
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07002636 * 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 -07002637 * @node: node whose fallback list we're appending
2638 * @used_node_mask: nodemask_t of already used nodes
2639 *
2640 * We use a number of factors to determine which is the next node that should
2641 * appear on a given node's fallback list. The node should not have appeared
2642 * already in @node's fallback list, and it should be the next closest node
2643 * according to the distance array (which contains arbitrary distance values
2644 * from each node to each node in the system), and should also prefer nodes
2645 * with no CPUs, since presumably they'll have very little allocation pressure
2646 * on them otherwise.
2647 * It returns -1 if no node is found.
2648 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002649static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002651 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652 int min_val = INT_MAX;
2653 int best_node = -1;
Rusty Russella70f7302009-03-13 14:49:46 +10302654 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002656 /* Use the local node if we haven't already */
2657 if (!node_isset(node, *used_node_mask)) {
2658 node_set(node, *used_node_mask);
2659 return node;
2660 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002662 for_each_node_state(n, N_HIGH_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663
2664 /* Don't want a node to appear more than once */
2665 if (node_isset(n, *used_node_mask))
2666 continue;
2667
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668 /* Use the distance array to find the distance */
2669 val = node_distance(node, n);
2670
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002671 /* Penalize nodes under us ("prefer the next node") */
2672 val += (n < node);
2673
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10302675 tmp = cpumask_of_node(n);
2676 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 val += PENALTY_FOR_NODE_WITH_CPUS;
2678
2679 /* Slight preference for less loaded node */
2680 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2681 val += node_load[n];
2682
2683 if (val < min_val) {
2684 min_val = val;
2685 best_node = n;
2686 }
2687 }
2688
2689 if (best_node >= 0)
2690 node_set(best_node, *used_node_mask);
2691
2692 return best_node;
2693}
2694
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002695
2696/*
2697 * Build zonelists ordered by node and zones within node.
2698 * This results in maximum locality--normal zone overflows into local
2699 * DMA zone, if any--but risks exhausting DMA zone.
2700 */
2701static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002703 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002705
Mel Gorman54a6eb52008-04-28 02:12:16 -07002706 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07002707 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07002708 ;
2709 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2710 MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002711 zonelist->_zonerefs[j].zone = NULL;
2712 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002713}
2714
2715/*
Christoph Lameter523b9452007-10-16 01:25:37 -07002716 * Build gfp_thisnode zonelists
2717 */
2718static void build_thisnode_zonelists(pg_data_t *pgdat)
2719{
Christoph Lameter523b9452007-10-16 01:25:37 -07002720 int j;
2721 struct zonelist *zonelist;
2722
Mel Gorman54a6eb52008-04-28 02:12:16 -07002723 zonelist = &pgdat->node_zonelists[1];
2724 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002725 zonelist->_zonerefs[j].zone = NULL;
2726 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07002727}
2728
2729/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002730 * Build zonelists ordered by zone and nodes within zones.
2731 * This results in conserving DMA zone[s] until all Normal memory is
2732 * exhausted, but results in overflowing to remote node while memory
2733 * may still exist in local DMA zone.
2734 */
2735static int node_order[MAX_NUMNODES];
2736
2737static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2738{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002739 int pos, j, node;
2740 int zone_type; /* needs to be signed */
2741 struct zone *z;
2742 struct zonelist *zonelist;
2743
Mel Gorman54a6eb52008-04-28 02:12:16 -07002744 zonelist = &pgdat->node_zonelists[0];
2745 pos = 0;
2746 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2747 for (j = 0; j < nr_nodes; j++) {
2748 node = node_order[j];
2749 z = &NODE_DATA(node)->node_zones[zone_type];
2750 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002751 zoneref_set_zone(z,
2752 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002753 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002754 }
2755 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002756 }
Mel Gormandd1a2392008-04-28 02:12:17 -07002757 zonelist->_zonerefs[pos].zone = NULL;
2758 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002759}
2760
2761static int default_zonelist_order(void)
2762{
2763 int nid, zone_type;
2764 unsigned long low_kmem_size,total_size;
2765 struct zone *z;
2766 int average_size;
2767 /*
Thomas Weber88393162010-03-16 11:47:56 +01002768 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002769 * If they are really small and used heavily, the system can fall
2770 * into OOM very easily.
David Rientjese325c902010-05-24 14:32:13 -07002771 * This function detect ZONE_DMA/DMA32 size and configures zone order.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002772 */
2773 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2774 low_kmem_size = 0;
2775 total_size = 0;
2776 for_each_online_node(nid) {
2777 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2778 z = &NODE_DATA(nid)->node_zones[zone_type];
2779 if (populated_zone(z)) {
2780 if (zone_type < ZONE_NORMAL)
2781 low_kmem_size += z->present_pages;
2782 total_size += z->present_pages;
David Rientjese325c902010-05-24 14:32:13 -07002783 } else if (zone_type == ZONE_NORMAL) {
2784 /*
2785 * If any node has only lowmem, then node order
2786 * is preferred to allow kernel allocations
2787 * locally; otherwise, they can easily infringe
2788 * on other nodes when there is an abundance of
2789 * lowmem available to allocate from.
2790 */
2791 return ZONELIST_ORDER_NODE;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002792 }
2793 }
2794 }
2795 if (!low_kmem_size || /* there are no DMA area. */
2796 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2797 return ZONELIST_ORDER_NODE;
2798 /*
2799 * look into each node's config.
2800 * If there is a node whose DMA/DMA32 memory is very big area on
2801 * local memory, NODE_ORDER may be suitable.
2802 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002803 average_size = total_size /
2804 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002805 for_each_online_node(nid) {
2806 low_kmem_size = 0;
2807 total_size = 0;
2808 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2809 z = &NODE_DATA(nid)->node_zones[zone_type];
2810 if (populated_zone(z)) {
2811 if (zone_type < ZONE_NORMAL)
2812 low_kmem_size += z->present_pages;
2813 total_size += z->present_pages;
2814 }
2815 }
2816 if (low_kmem_size &&
2817 total_size > average_size && /* ignore small node */
2818 low_kmem_size > total_size * 70/100)
2819 return ZONELIST_ORDER_NODE;
2820 }
2821 return ZONELIST_ORDER_ZONE;
2822}
2823
2824static void set_zonelist_order(void)
2825{
2826 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2827 current_zonelist_order = default_zonelist_order();
2828 else
2829 current_zonelist_order = user_zonelist_order;
2830}
2831
2832static void build_zonelists(pg_data_t *pgdat)
2833{
2834 int j, node, load;
2835 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002837 int local_node, prev_node;
2838 struct zonelist *zonelist;
2839 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002840
2841 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07002842 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07002844 zonelist->_zonerefs[0].zone = NULL;
2845 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846 }
2847
2848 /* NUMA-aware ordering of nodes */
2849 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002850 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851 prev_node = local_node;
2852 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002853
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002854 memset(node_order, 0, sizeof(node_order));
2855 j = 0;
2856
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08002858 int distance = node_distance(local_node, node);
2859
2860 /*
2861 * If another node is sufficiently far away then it is better
2862 * to reclaim pages in a zone before going off node.
2863 */
2864 if (distance > RECLAIM_DISTANCE)
2865 zone_reclaim_mode = 1;
2866
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 /*
2868 * We don't want to pressure a particular node.
2869 * So adding penalty to the first node in same
2870 * distance group to make it round-robin.
2871 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08002872 if (distance != node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002873 node_load[node] = load;
2874
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875 prev_node = node;
2876 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002877 if (order == ZONELIST_ORDER_NODE)
2878 build_zonelists_in_node_order(pgdat, node);
2879 else
2880 node_order[j++] = node; /* remember order */
2881 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002883 if (order == ZONELIST_ORDER_ZONE) {
2884 /* calculate node order -- i.e., DMA last! */
2885 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886 }
Christoph Lameter523b9452007-10-16 01:25:37 -07002887
2888 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889}
2890
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002891/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002892static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002893{
Mel Gorman54a6eb52008-04-28 02:12:16 -07002894 struct zonelist *zonelist;
2895 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07002896 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002897
Mel Gorman54a6eb52008-04-28 02:12:16 -07002898 zonelist = &pgdat->node_zonelists[0];
2899 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2900 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07002901 for (z = zonelist->_zonerefs; z->zone; z++)
2902 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002903}
2904
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07002905#ifdef CONFIG_HAVE_MEMORYLESS_NODES
2906/*
2907 * Return node id of node used for "local" allocations.
2908 * I.e., first node id of first zone in arg node's generic zonelist.
2909 * Used for initializing percpu 'numa_mem', which is used primarily
2910 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
2911 */
2912int local_memory_node(int node)
2913{
2914 struct zone *zone;
2915
2916 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
2917 gfp_zone(GFP_KERNEL),
2918 NULL,
2919 &zone);
2920 return zone->node;
2921}
2922#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002923
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924#else /* CONFIG_NUMA */
2925
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002926static void set_zonelist_order(void)
2927{
2928 current_zonelist_order = ZONELIST_ORDER_ZONE;
2929}
2930
2931static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002932{
Christoph Lameter19655d32006-09-25 23:31:19 -07002933 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002934 enum zone_type j;
2935 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002936
2937 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002938
Mel Gorman54a6eb52008-04-28 02:12:16 -07002939 zonelist = &pgdat->node_zonelists[0];
2940 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002941
Mel Gorman54a6eb52008-04-28 02:12:16 -07002942 /*
2943 * Now we build the zonelist so that it contains the zones
2944 * of all the other nodes.
2945 * We don't want to pressure a particular node, so when
2946 * building the zones for node N, we make sure that the
2947 * zones coming right after the local ones are those from
2948 * node N+1 (modulo N)
2949 */
2950 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2951 if (!node_online(node))
2952 continue;
2953 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2954 MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07002956 for (node = 0; node < local_node; node++) {
2957 if (!node_online(node))
2958 continue;
2959 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2960 MAX_NR_ZONES - 1);
2961 }
2962
Mel Gormandd1a2392008-04-28 02:12:17 -07002963 zonelist->_zonerefs[j].zone = NULL;
2964 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002965}
2966
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002967/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002968static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002969{
Mel Gorman54a6eb52008-04-28 02:12:16 -07002970 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002971}
2972
Linus Torvalds1da177e2005-04-16 15:20:36 -07002973#endif /* CONFIG_NUMA */
2974
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002975/*
2976 * Boot pageset table. One per cpu which is going to be used for all
2977 * zones and all nodes. The parameters will be set in such a way
2978 * that an item put on a list will immediately be handed over to
2979 * the buddy list. This is safe since pageset manipulation is done
2980 * with interrupts disabled.
2981 *
2982 * The boot_pagesets must be kept even after bootup is complete for
2983 * unused processors and/or zones. They do play a role for bootstrapping
2984 * hotplugged processors.
2985 *
2986 * zoneinfo_show() and maybe other functions do
2987 * not check if the processor is online before following the pageset pointer.
2988 * Other parts of the kernel may not check if the zone is available.
2989 */
2990static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
2991static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07002992static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002993
Haicheng Li4eaf3f62010-05-24 14:32:52 -07002994/*
2995 * Global mutex to protect against size modification of zonelists
2996 * as well as to serialize pageset setup for the new populated zone.
2997 */
2998DEFINE_MUTEX(zonelists_mutex);
2999
Rusty Russell9b1a4d32008-07-28 12:16:30 -05003000/* return values int ....just for stop_machine() */
Haicheng Li1f522502010-05-24 14:32:51 -07003001static __init_refok int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002{
Yasunori Goto68113782006-06-23 02:03:11 -07003003 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003004 int cpu;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003005
Bo Liu7f9cfb32009-08-18 14:11:19 -07003006#ifdef CONFIG_NUMA
3007 memset(node_load, 0, sizeof(node_load));
3008#endif
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003009 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07003010 pg_data_t *pgdat = NODE_DATA(nid);
3011
3012 build_zonelists(pgdat);
3013 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003014 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003015
3016 /*
3017 * Initialize the boot_pagesets that are going to be used
3018 * for bootstrapping processors. The real pagesets for
3019 * each zone will be allocated later when the per cpu
3020 * allocator is available.
3021 *
3022 * boot_pagesets are used also for bootstrapping offline
3023 * cpus if the system is already booted because the pagesets
3024 * are needed to initialize allocators on a specific cpu too.
3025 * F.e. the percpu allocator needs the page allocator which
3026 * needs the percpu allocator in order to allocate its pagesets
3027 * (a chicken-egg dilemma).
3028 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003029 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003030 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3031
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003032#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3033 /*
3034 * We now know the "local memory node" for each node--
3035 * i.e., the node of the first zone in the generic zonelist.
3036 * Set up numa_mem percpu variable for on-line cpus. During
3037 * boot, only the boot cpu should be on-line; we'll init the
3038 * secondary cpus' numa_mem as they come on-line. During
3039 * node/memory hotplug, we'll fixup all on-line cpus.
3040 */
3041 if (cpu_online(cpu))
3042 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3043#endif
3044 }
3045
Yasunori Goto68113782006-06-23 02:03:11 -07003046 return 0;
3047}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003048
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003049/*
3050 * Called with zonelists_mutex held always
3051 * unless system_state == SYSTEM_BOOTING.
3052 */
Haicheng Li1f522502010-05-24 14:32:51 -07003053void build_all_zonelists(void *data)
Yasunori Goto68113782006-06-23 02:03:11 -07003054{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003055 set_zonelist_order();
3056
Yasunori Goto68113782006-06-23 02:03:11 -07003057 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07003058 __build_all_zonelists(NULL);
Mel Gorman68ad8df2008-07-23 21:26:52 -07003059 mminit_verify_zonelist();
Yasunori Goto68113782006-06-23 02:03:11 -07003060 cpuset_init_current_mems_allowed();
3061 } else {
Simon Arlott183ff222007-10-20 01:27:18 +02003062 /* we have to stop all cpus to guarantee there is no user
Yasunori Goto68113782006-06-23 02:03:11 -07003063 of zonelist */
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003064#ifdef CONFIG_MEMORY_HOTPLUG
3065 if (data)
3066 setup_zone_pageset((struct zone *)data);
3067#endif
3068 stop_machine(__build_all_zonelists, NULL, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07003069 /* cpuset refresh routine should be here */
3070 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07003071 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003072 /*
3073 * Disable grouping by mobility if the number of pages in the
3074 * system is too low to allow the mechanism to work. It would be
3075 * more accurate, but expensive to check per-zone. This check is
3076 * made on memory-hotadd so a system can start with mobility
3077 * disabled and enable it later
3078 */
Mel Gormand9c23402007-10-16 01:26:01 -07003079 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003080 page_group_by_mobility_disabled = 1;
3081 else
3082 page_group_by_mobility_disabled = 0;
3083
3084 printk("Built %i zonelists in %s order, mobility grouping %s. "
3085 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003086 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003087 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003088 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003089 vm_total_pages);
3090#ifdef CONFIG_NUMA
3091 printk("Policy zone: %s\n", zone_names[policy_zone]);
3092#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093}
3094
3095/*
3096 * Helper functions to size the waitqueue hash table.
3097 * Essentially these want to choose hash table sizes sufficiently
3098 * large so that collisions trying to wait on pages are rare.
3099 * But in fact, the number of active page waitqueues on typical
3100 * systems is ridiculously low, less than 200. So this is even
3101 * conservative, even though it seems large.
3102 *
3103 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3104 * waitqueues, i.e. the size of the waitq table given the number of pages.
3105 */
3106#define PAGES_PER_WAITQUEUE 256
3107
Yasunori Gotocca448f2006-06-23 02:03:10 -07003108#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07003109static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003110{
3111 unsigned long size = 1;
3112
3113 pages /= PAGES_PER_WAITQUEUE;
3114
3115 while (size < pages)
3116 size <<= 1;
3117
3118 /*
3119 * Once we have dozens or even hundreds of threads sleeping
3120 * on IO we've got bigger problems than wait queue collision.
3121 * Limit the size of the wait table to a reasonable size.
3122 */
3123 size = min(size, 4096UL);
3124
3125 return max(size, 4UL);
3126}
Yasunori Gotocca448f2006-06-23 02:03:10 -07003127#else
3128/*
3129 * A zone's size might be changed by hot-add, so it is not possible to determine
3130 * a suitable size for its wait_table. So we use the maximum size now.
3131 *
3132 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3133 *
3134 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3135 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3136 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3137 *
3138 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3139 * or more by the traditional way. (See above). It equals:
3140 *
3141 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3142 * ia64(16K page size) : = ( 8G + 4M)byte.
3143 * powerpc (64K page size) : = (32G +16M)byte.
3144 */
3145static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3146{
3147 return 4096UL;
3148}
3149#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003150
3151/*
3152 * This is an integer logarithm so that shifts can be used later
3153 * to extract the more random high bits from the multiplicative
3154 * hash function before the remainder is taken.
3155 */
3156static inline unsigned long wait_table_bits(unsigned long size)
3157{
3158 return ffz(~size);
3159}
3160
3161#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3162
Mel Gorman56fd56b2007-10-16 01:25:58 -07003163/*
Mel Gormand9c23402007-10-16 01:26:01 -07003164 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07003165 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3166 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07003167 * higher will lead to a bigger reserve which will get freed as contiguous
3168 * blocks as reclaim kicks in
3169 */
3170static void setup_zone_migrate_reserve(struct zone *zone)
3171{
3172 unsigned long start_pfn, pfn, end_pfn;
3173 struct page *page;
Mel Gorman78986a62009-09-21 17:03:02 -07003174 unsigned long block_migratetype;
3175 int reserve;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003176
3177 /* Get the start pfn, end pfn and the number of blocks to reserve */
3178 start_pfn = zone->zone_start_pfn;
3179 end_pfn = start_pfn + zone->spanned_pages;
Mel Gorman41858962009-06-16 15:32:12 -07003180 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07003181 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003182
Mel Gorman78986a62009-09-21 17:03:02 -07003183 /*
3184 * Reserve blocks are generally in place to help high-order atomic
3185 * allocations that are short-lived. A min_free_kbytes value that
3186 * would result in more than 2 reserve blocks for atomic allocations
3187 * is assumed to be in place to help anti-fragmentation for the
3188 * future allocation of hugepages at runtime.
3189 */
3190 reserve = min(2, reserve);
3191
Mel Gormand9c23402007-10-16 01:26:01 -07003192 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07003193 if (!pfn_valid(pfn))
3194 continue;
3195 page = pfn_to_page(pfn);
3196
Adam Litke344c7902008-09-02 14:35:38 -07003197 /* Watch out for overlapping nodes */
3198 if (page_to_nid(page) != zone_to_nid(zone))
3199 continue;
3200
Mel Gorman56fd56b2007-10-16 01:25:58 -07003201 /* Blocks with reserved pages will never free, skip them. */
3202 if (PageReserved(page))
3203 continue;
3204
3205 block_migratetype = get_pageblock_migratetype(page);
3206
3207 /* If this block is reserved, account for it */
3208 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
3209 reserve--;
3210 continue;
3211 }
3212
3213 /* Suitable for reserving if this block is movable */
3214 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
3215 set_pageblock_migratetype(page, MIGRATE_RESERVE);
3216 move_freepages_block(zone, page, MIGRATE_RESERVE);
3217 reserve--;
3218 continue;
3219 }
3220
3221 /*
3222 * If the reserve is met and this is a previous reserved block,
3223 * take it back
3224 */
3225 if (block_migratetype == MIGRATE_RESERVE) {
3226 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
3227 move_freepages_block(zone, page, MIGRATE_MOVABLE);
3228 }
3229 }
3230}
Mel Gormanac0e5b72007-10-16 01:25:58 -07003231
Linus Torvalds1da177e2005-04-16 15:20:36 -07003232/*
3233 * Initially all pages are reserved - free ones are freed
3234 * up by free_all_bootmem() once the early boot process is
3235 * done. Non-atomic initialization, single-pass.
3236 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003237void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa02007-01-10 23:15:30 -08003238 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003239{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003240 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07003241 unsigned long end_pfn = start_pfn + size;
3242 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003243 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003244
Hugh Dickins22b31ee2009-01-06 14:40:09 -08003245 if (highest_memmap_pfn < end_pfn - 1)
3246 highest_memmap_pfn = end_pfn - 1;
3247
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003248 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08003249 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa02007-01-10 23:15:30 -08003250 /*
3251 * There can be holes in boot-time mem_map[]s
3252 * handed to this function. They do not
3253 * exist on hotplugged memory.
3254 */
3255 if (context == MEMMAP_EARLY) {
3256 if (!early_pfn_valid(pfn))
3257 continue;
3258 if (!early_pfn_in_nid(pfn, nid))
3259 continue;
3260 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003261 page = pfn_to_page(pfn);
3262 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07003263 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08003264 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003265 reset_page_mapcount(page);
3266 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003267 /*
3268 * Mark the block movable so that blocks are reserved for
3269 * movable at startup. This will force kernel allocations
3270 * to reserve their blocks rather than leaking throughout
3271 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07003272 * kernel allocations are made. Later some blocks near
3273 * the start are marked MIGRATE_RESERVE by
3274 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003275 *
3276 * bitmap is created for zone's valid pfn range. but memmap
3277 * can be created for invalid pages (for alignment)
3278 * check here not to call set_pageblock_migratetype() against
3279 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003280 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003281 if ((z->zone_start_pfn <= pfn)
3282 && (pfn < z->zone_start_pfn + z->spanned_pages)
3283 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07003284 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003285
Linus Torvalds1da177e2005-04-16 15:20:36 -07003286 INIT_LIST_HEAD(&page->lru);
3287#ifdef WANT_PAGE_VIRTUAL
3288 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3289 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07003290 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003291#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003292 }
3293}
3294
Andi Kleen1e548de2008-02-04 22:29:26 -08003295static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003296{
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003297 int order, t;
3298 for_each_migratetype_order(order, t) {
3299 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003300 zone->free_area[order].nr_free = 0;
3301 }
3302}
3303
3304#ifndef __HAVE_ARCH_MEMMAP_INIT
3305#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa02007-01-10 23:15:30 -08003306 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003307#endif
3308
Sam Ravnborg1d6f4e62008-01-17 15:21:12 -08003309static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003310{
David Howells3a6be872009-05-06 16:03:03 -07003311#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003312 int batch;
3313
3314 /*
3315 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07003316 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003317 *
3318 * OK, so we don't know how big the cache is. So guess.
3319 */
3320 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07003321 if (batch * PAGE_SIZE > 512 * 1024)
3322 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003323 batch /= 4; /* We effectively *= 4 below */
3324 if (batch < 1)
3325 batch = 1;
3326
3327 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11003328 * Clamp the batch to a 2^n - 1 value. Having a power
3329 * of 2 value was found to be more likely to have
3330 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003331 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11003332 * For example if 2 tasks are alternately allocating
3333 * batches of pages, one task can end up with a lot
3334 * of pages of one half of the possible page colors
3335 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003336 */
David Howells91552032009-05-06 16:03:02 -07003337 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07003338
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003339 return batch;
David Howells3a6be872009-05-06 16:03:03 -07003340
3341#else
3342 /* The deferral and batching of frees should be suppressed under NOMMU
3343 * conditions.
3344 *
3345 * The problem is that NOMMU needs to be able to allocate large chunks
3346 * of contiguous memory as there's no hardware page translation to
3347 * assemble apparent contiguous memory from discontiguous pages.
3348 *
3349 * Queueing large contiguous runs of pages for batching, however,
3350 * causes the pages to actually be freed in smaller chunks. As there
3351 * can be a significant delay between the individual batches being
3352 * recycled, this leads to the once large chunks of space being
3353 * fragmented and becoming unavailable for high-order allocations.
3354 */
3355 return 0;
3356#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003357}
3358
Adrian Bunkb69a7282008-07-23 21:28:12 -07003359static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
Christoph Lameter2caaad42005-06-21 17:15:00 -07003360{
3361 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003362 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003363
Magnus Damm1c6fe942005-10-26 01:58:59 -07003364 memset(p, 0, sizeof(*p));
3365
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003366 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003367 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003368 pcp->high = 6 * batch;
3369 pcp->batch = max(1UL, 1 * batch);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003370 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
3371 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07003372}
3373
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003374/*
3375 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3376 * to the value high for the pageset p.
3377 */
3378
3379static void setup_pagelist_highmark(struct per_cpu_pageset *p,
3380 unsigned long high)
3381{
3382 struct per_cpu_pages *pcp;
3383
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003384 pcp = &p->pcp;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003385 pcp->high = high;
3386 pcp->batch = max(1UL, high/4);
3387 if ((high/4) > (PAGE_SHIFT * 8))
3388 pcp->batch = PAGE_SHIFT * 8;
3389}
3390
Wu Fengguang319774e2010-05-24 14:32:49 -07003391static __meminit void setup_zone_pageset(struct zone *zone)
3392{
3393 int cpu;
3394
3395 zone->pageset = alloc_percpu(struct per_cpu_pageset);
3396
3397 for_each_possible_cpu(cpu) {
3398 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
3399
3400 setup_pageset(pcp, zone_batchsize(zone));
3401
3402 if (percpu_pagelist_fraction)
3403 setup_pagelist_highmark(pcp,
3404 (zone->present_pages /
3405 percpu_pagelist_fraction));
3406 }
3407}
3408
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003409/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003410 * Allocate per cpu pagesets and initialize them.
3411 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07003412 */
Al Viro78d99552005-12-15 09:18:25 +00003413void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003414{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003415 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003416
Wu Fengguang319774e2010-05-24 14:32:49 -07003417 for_each_populated_zone(zone)
3418 setup_zone_pageset(zone);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003419}
3420
Sam Ravnborg577a32f2007-05-17 23:29:25 +02003421static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07003422int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07003423{
3424 int i;
3425 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003426 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07003427
3428 /*
3429 * The per-page waitqueue mechanism uses hashed waitqueues
3430 * per zone.
3431 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07003432 zone->wait_table_hash_nr_entries =
3433 wait_table_hash_nr_entries(zone_size_pages);
3434 zone->wait_table_bits =
3435 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003436 alloc_size = zone->wait_table_hash_nr_entries
3437 * sizeof(wait_queue_head_t);
3438
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07003439 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07003440 zone->wait_table = (wait_queue_head_t *)
3441 alloc_bootmem_node(pgdat, alloc_size);
3442 } else {
3443 /*
3444 * This case means that a zone whose size was 0 gets new memory
3445 * via memory hot-add.
3446 * But it may be the case that a new node was hot-added. In
3447 * this case vmalloc() will not be able to use this new node's
3448 * memory - this wait_table must be initialized to use this new
3449 * node itself as well.
3450 * To use this new node's memory, further consideration will be
3451 * necessary.
3452 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07003453 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003454 }
3455 if (!zone->wait_table)
3456 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07003457
Yasunori Goto02b694d2006-06-23 02:03:08 -07003458 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07003459 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003460
3461 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003462}
3463
Shaohua Li112067f2009-09-21 17:01:16 -07003464static int __zone_pcp_update(void *data)
3465{
3466 struct zone *zone = data;
3467 int cpu;
3468 unsigned long batch = zone_batchsize(zone), flags;
3469
Thomas Gleixner2d30a1f2010-03-10 15:20:40 -08003470 for_each_possible_cpu(cpu) {
Shaohua Li112067f2009-09-21 17:01:16 -07003471 struct per_cpu_pageset *pset;
3472 struct per_cpu_pages *pcp;
3473
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003474 pset = per_cpu_ptr(zone->pageset, cpu);
Shaohua Li112067f2009-09-21 17:01:16 -07003475 pcp = &pset->pcp;
3476
3477 local_irq_save(flags);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003478 free_pcppages_bulk(zone, pcp->count, pcp);
Shaohua Li112067f2009-09-21 17:01:16 -07003479 setup_pageset(pset, batch);
3480 local_irq_restore(flags);
3481 }
3482 return 0;
3483}
3484
3485void zone_pcp_update(struct zone *zone)
3486{
3487 stop_machine(__zone_pcp_update, zone, NULL);
3488}
3489
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003490static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07003491{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003492 /*
3493 * per cpu subsystem is not up at this point. The following code
3494 * relies on the ability of the linker to provide the
3495 * offset of a (static) per cpu variable into the per cpu area.
3496 */
3497 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07003498
Anton Blanchardf5335c02006-03-25 03:06:49 -08003499 if (zone->present_pages)
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003500 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
3501 zone->name, zone->present_pages,
3502 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07003503}
3504
Yasunori Goto718127c2006-06-23 02:03:10 -07003505__meminit int init_currently_empty_zone(struct zone *zone,
3506 unsigned long zone_start_pfn,
Dave Hansena2f3aa02007-01-10 23:15:30 -08003507 unsigned long size,
3508 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07003509{
3510 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003511 int ret;
3512 ret = zone_wait_table_init(zone, size);
3513 if (ret)
3514 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07003515 pgdat->nr_zones = zone_idx(zone) + 1;
3516
Dave Hansened8ece22005-10-29 18:16:50 -07003517 zone->zone_start_pfn = zone_start_pfn;
3518
Mel Gorman708614e2008-07-23 21:26:51 -07003519 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3520 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3521 pgdat->node_id,
3522 (unsigned long)zone_idx(zone),
3523 zone_start_pfn, (zone_start_pfn + size));
3524
Andi Kleen1e548de2008-02-04 22:29:26 -08003525 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07003526
3527 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003528}
3529
Mel Gormanc7132162006-09-27 01:49:43 -07003530#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3531/*
3532 * Basic iterator support. Return the first range of PFNs for a node
3533 * Note: nid == MAX_NUMNODES returns first region regardless of node
3534 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003535static int __meminit first_active_region_index_in_nid(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003536{
3537 int i;
3538
3539 for (i = 0; i < nr_nodemap_entries; i++)
3540 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3541 return i;
3542
3543 return -1;
3544}
3545
3546/*
3547 * Basic iterator support. Return the next active range of PFNs for a node
Simon Arlott183ff222007-10-20 01:27:18 +02003548 * Note: nid == MAX_NUMNODES returns next region regardless of node
Mel Gormanc7132162006-09-27 01:49:43 -07003549 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003550static int __meminit next_active_region_index_in_nid(int index, int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003551{
3552 for (index = index + 1; index < nr_nodemap_entries; index++)
3553 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3554 return index;
3555
3556 return -1;
3557}
3558
3559#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3560/*
3561 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3562 * Architectures may implement their own version but if add_active_range()
3563 * was used and there are no special requirements, this is a convenient
3564 * alternative
3565 */
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003566int __meminit __early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07003567{
3568 int i;
3569
3570 for (i = 0; i < nr_nodemap_entries; i++) {
3571 unsigned long start_pfn = early_node_map[i].start_pfn;
3572 unsigned long end_pfn = early_node_map[i].end_pfn;
3573
3574 if (start_pfn <= pfn && pfn < end_pfn)
3575 return early_node_map[i].nid;
3576 }
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003577 /* This is a memory hole */
3578 return -1;
Mel Gormanc7132162006-09-27 01:49:43 -07003579}
3580#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3581
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003582int __meminit early_pfn_to_nid(unsigned long pfn)
3583{
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003584 int nid;
3585
3586 nid = __early_pfn_to_nid(pfn);
3587 if (nid >= 0)
3588 return nid;
3589 /* just returns 0 */
3590 return 0;
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003591}
3592
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003593#ifdef CONFIG_NODES_SPAN_OTHER_NODES
3594bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3595{
3596 int nid;
3597
3598 nid = __early_pfn_to_nid(pfn);
3599 if (nid >= 0 && nid != node)
3600 return false;
3601 return true;
3602}
3603#endif
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003604
Mel Gormanc7132162006-09-27 01:49:43 -07003605/* Basic iterator support to walk early_node_map[] */
3606#define for_each_active_range_index_in_nid(i, nid) \
3607 for (i = first_active_region_index_in_nid(nid); i != -1; \
3608 i = next_active_region_index_in_nid(i, nid))
3609
3610/**
3611 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003612 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3613 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07003614 *
3615 * If an architecture guarantees that all ranges registered with
3616 * add_active_ranges() contain no holes and may be freed, this
3617 * this function may be used instead of calling free_bootmem() manually.
3618 */
3619void __init free_bootmem_with_active_regions(int nid,
3620 unsigned long max_low_pfn)
3621{
3622 int i;
3623
3624 for_each_active_range_index_in_nid(i, nid) {
3625 unsigned long size_pages = 0;
3626 unsigned long end_pfn = early_node_map[i].end_pfn;
3627
3628 if (early_node_map[i].start_pfn >= max_low_pfn)
3629 continue;
3630
3631 if (end_pfn > max_low_pfn)
3632 end_pfn = max_low_pfn;
3633
3634 size_pages = end_pfn - early_node_map[i].start_pfn;
3635 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3636 PFN_PHYS(early_node_map[i].start_pfn),
3637 size_pages << PAGE_SHIFT);
3638 }
3639}
3640
Yinghai Luedbe7d22010-08-25 13:39:16 -07003641#ifdef CONFIG_HAVE_MEMBLOCK
3642u64 __init find_memory_core_early(int nid, u64 size, u64 align,
3643 u64 goal, u64 limit)
3644{
3645 int i;
3646
3647 /* Need to go over early_node_map to find out good range for node */
3648 for_each_active_range_index_in_nid(i, nid) {
3649 u64 addr;
3650 u64 ei_start, ei_last;
3651 u64 final_start, final_end;
3652
3653 ei_last = early_node_map[i].end_pfn;
3654 ei_last <<= PAGE_SHIFT;
3655 ei_start = early_node_map[i].start_pfn;
3656 ei_start <<= PAGE_SHIFT;
3657
3658 final_start = max(ei_start, goal);
3659 final_end = min(ei_last, limit);
3660
3661 if (final_start >= final_end)
3662 continue;
3663
3664 addr = memblock_find_in_range(final_start, final_end, size, align);
3665
3666 if (addr == MEMBLOCK_ERROR)
3667 continue;
3668
3669 return addr;
3670 }
3671
3672 return MEMBLOCK_ERROR;
3673}
3674#endif
3675
Yinghai Lu08677212010-02-10 01:20:20 -08003676int __init add_from_early_node_map(struct range *range, int az,
3677 int nr_range, int nid)
3678{
3679 int i;
3680 u64 start, end;
3681
3682 /* need to go over early_node_map to find out good range for node */
3683 for_each_active_range_index_in_nid(i, nid) {
3684 start = early_node_map[i].start_pfn;
3685 end = early_node_map[i].end_pfn;
3686 nr_range = add_range(range, az, nr_range, start, end);
3687 }
3688 return nr_range;
3689}
3690
Yinghai Lu2ee78f72010-02-17 11:29:49 -08003691#ifdef CONFIG_NO_BOOTMEM
Yinghai Lu08677212010-02-10 01:20:20 -08003692void * __init __alloc_memory_core_early(int nid, u64 size, u64 align,
3693 u64 goal, u64 limit)
3694{
Yinghai Lu08677212010-02-10 01:20:20 -08003695 void *ptr;
Yinghai Lu72d7c3b2010-08-25 13:39:17 -07003696 u64 addr;
Yinghai Lu08677212010-02-10 01:20:20 -08003697
Yinghai Lu72d7c3b2010-08-25 13:39:17 -07003698 if (limit > memblock.current_limit)
3699 limit = memblock.current_limit;
Yinghai Lub8ab9f82010-07-20 13:24:31 -07003700
Yinghai Lu72d7c3b2010-08-25 13:39:17 -07003701 addr = find_memory_core_early(nid, size, align, goal, limit);
Yinghai Lu08677212010-02-10 01:20:20 -08003702
Yinghai Lu72d7c3b2010-08-25 13:39:17 -07003703 if (addr == MEMBLOCK_ERROR)
3704 return NULL;
Yinghai Lu08677212010-02-10 01:20:20 -08003705
Yinghai Lu72d7c3b2010-08-25 13:39:17 -07003706 ptr = phys_to_virt(addr);
3707 memset(ptr, 0, size);
3708 memblock_x86_reserve_range(addr, addr + size, "BOOTMEM");
3709 /*
3710 * The min_count is set to 0 so that bootmem allocated blocks
3711 * are never reported as leaks.
3712 */
3713 kmemleak_alloc(ptr, size, 0, 0);
3714 return ptr;
Yinghai Lu08677212010-02-10 01:20:20 -08003715}
Yinghai Lu2ee78f72010-02-17 11:29:49 -08003716#endif
Yinghai Lu08677212010-02-10 01:20:20 -08003717
3718
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003719void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3720{
3721 int i;
Yinghai Lud52d53b2008-06-16 20:10:55 -07003722 int ret;
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003723
Yinghai Lud52d53b2008-06-16 20:10:55 -07003724 for_each_active_range_index_in_nid(i, nid) {
3725 ret = work_fn(early_node_map[i].start_pfn,
3726 early_node_map[i].end_pfn, data);
3727 if (ret)
3728 break;
3729 }
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003730}
Mel Gormanc7132162006-09-27 01:49:43 -07003731/**
3732 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003733 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07003734 *
3735 * If an architecture guarantees that all ranges registered with
3736 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003737 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07003738 */
3739void __init sparse_memory_present_with_active_regions(int nid)
3740{
3741 int i;
3742
3743 for_each_active_range_index_in_nid(i, nid)
3744 memory_present(early_node_map[i].nid,
3745 early_node_map[i].start_pfn,
3746 early_node_map[i].end_pfn);
3747}
3748
3749/**
3750 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003751 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3752 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3753 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07003754 *
3755 * It returns the start and end page frame of a node based on information
3756 * provided by an arch calling add_active_range(). If called for a node
3757 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003758 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07003759 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003760void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003761 unsigned long *start_pfn, unsigned long *end_pfn)
3762{
3763 int i;
3764 *start_pfn = -1UL;
3765 *end_pfn = 0;
3766
3767 for_each_active_range_index_in_nid(i, nid) {
3768 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3769 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3770 }
3771
Christoph Lameter633c0662007-10-16 01:25:37 -07003772 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07003773 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07003774}
3775
3776/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07003777 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3778 * assumption is made that zones within a node are ordered in monotonic
3779 * increasing memory addresses so that the "highest" populated zone is used
3780 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003781static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07003782{
3783 int zone_index;
3784 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3785 if (zone_index == ZONE_MOVABLE)
3786 continue;
3787
3788 if (arch_zone_highest_possible_pfn[zone_index] >
3789 arch_zone_lowest_possible_pfn[zone_index])
3790 break;
3791 }
3792
3793 VM_BUG_ON(zone_index == -1);
3794 movable_zone = zone_index;
3795}
3796
3797/*
3798 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3799 * because it is sized independant of architecture. Unlike the other zones,
3800 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3801 * in each node depending on the size of each node and how evenly kernelcore
3802 * is distributed. This helper function adjusts the zone ranges
3803 * provided by the architecture for a given node by using the end of the
3804 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3805 * zones within a node are in order of monotonic increases memory addresses
3806 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003807static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07003808 unsigned long zone_type,
3809 unsigned long node_start_pfn,
3810 unsigned long node_end_pfn,
3811 unsigned long *zone_start_pfn,
3812 unsigned long *zone_end_pfn)
3813{
3814 /* Only adjust if ZONE_MOVABLE is on this node */
3815 if (zone_movable_pfn[nid]) {
3816 /* Size ZONE_MOVABLE */
3817 if (zone_type == ZONE_MOVABLE) {
3818 *zone_start_pfn = zone_movable_pfn[nid];
3819 *zone_end_pfn = min(node_end_pfn,
3820 arch_zone_highest_possible_pfn[movable_zone]);
3821
3822 /* Adjust for ZONE_MOVABLE starting within this range */
3823 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3824 *zone_end_pfn > zone_movable_pfn[nid]) {
3825 *zone_end_pfn = zone_movable_pfn[nid];
3826
3827 /* Check if this whole range is within ZONE_MOVABLE */
3828 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3829 *zone_start_pfn = *zone_end_pfn;
3830 }
3831}
3832
3833/*
Mel Gormanc7132162006-09-27 01:49:43 -07003834 * Return the number of pages a zone spans in a node, including holes
3835 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3836 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003837static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003838 unsigned long zone_type,
3839 unsigned long *ignored)
3840{
3841 unsigned long node_start_pfn, node_end_pfn;
3842 unsigned long zone_start_pfn, zone_end_pfn;
3843
3844 /* Get the start and end of the node and zone */
3845 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3846 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3847 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07003848 adjust_zone_range_for_zone_movable(nid, zone_type,
3849 node_start_pfn, node_end_pfn,
3850 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003851
3852 /* Check that this node has pages within the zone's required range */
3853 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3854 return 0;
3855
3856 /* Move the zone boundaries inside the node if necessary */
3857 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3858 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3859
3860 /* Return the spanned pages */
3861 return zone_end_pfn - zone_start_pfn;
3862}
3863
3864/*
3865 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003866 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07003867 */
Yinghai Lu32996252009-12-15 17:59:02 -08003868unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003869 unsigned long range_start_pfn,
3870 unsigned long range_end_pfn)
3871{
3872 int i = 0;
3873 unsigned long prev_end_pfn = 0, hole_pages = 0;
3874 unsigned long start_pfn;
3875
3876 /* Find the end_pfn of the first active range of pfns in the node */
3877 i = first_active_region_index_in_nid(nid);
3878 if (i == -1)
3879 return 0;
3880
Mel Gormanb5445f92007-07-26 10:41:18 -07003881 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3882
Mel Gorman9c7cd682006-09-27 01:49:58 -07003883 /* Account for ranges before physical memory on this node */
3884 if (early_node_map[i].start_pfn > range_start_pfn)
Mel Gormanb5445f92007-07-26 10:41:18 -07003885 hole_pages = prev_end_pfn - range_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003886
3887 /* Find all holes for the zone within the node */
3888 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3889
3890 /* No need to continue if prev_end_pfn is outside the zone */
3891 if (prev_end_pfn >= range_end_pfn)
3892 break;
3893
3894 /* Make sure the end of the zone is not within the hole */
3895 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3896 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3897
3898 /* Update the hole size cound and move on */
3899 if (start_pfn > range_start_pfn) {
3900 BUG_ON(prev_end_pfn > start_pfn);
3901 hole_pages += start_pfn - prev_end_pfn;
3902 }
3903 prev_end_pfn = early_node_map[i].end_pfn;
3904 }
3905
Mel Gorman9c7cd682006-09-27 01:49:58 -07003906 /* Account for ranges past physical memory on this node */
3907 if (range_end_pfn > prev_end_pfn)
Mel Gorman0c6cb972006-10-28 10:38:59 -07003908 hole_pages += range_end_pfn -
Mel Gorman9c7cd682006-09-27 01:49:58 -07003909 max(range_start_pfn, prev_end_pfn);
3910
Mel Gormanc7132162006-09-27 01:49:43 -07003911 return hole_pages;
3912}
3913
3914/**
3915 * absent_pages_in_range - Return number of page frames in holes within a range
3916 * @start_pfn: The start PFN to start searching for holes
3917 * @end_pfn: The end PFN to stop searching for holes
3918 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003919 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07003920 */
3921unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3922 unsigned long end_pfn)
3923{
3924 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3925}
3926
3927/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003928static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003929 unsigned long zone_type,
3930 unsigned long *ignored)
3931{
Mel Gorman9c7cd682006-09-27 01:49:58 -07003932 unsigned long node_start_pfn, node_end_pfn;
3933 unsigned long zone_start_pfn, zone_end_pfn;
3934
3935 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3936 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3937 node_start_pfn);
3938 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3939 node_end_pfn);
3940
Mel Gorman2a1e2742007-07-17 04:03:12 -07003941 adjust_zone_range_for_zone_movable(nid, zone_type,
3942 node_start_pfn, node_end_pfn,
3943 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07003944 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003945}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003946
Mel Gormanc7132162006-09-27 01:49:43 -07003947#else
Paul Mundt6ea6e682007-07-15 23:38:20 -07003948static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003949 unsigned long zone_type,
3950 unsigned long *zones_size)
3951{
3952 return zones_size[zone_type];
3953}
3954
Paul Mundt6ea6e682007-07-15 23:38:20 -07003955static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003956 unsigned long zone_type,
3957 unsigned long *zholes_size)
3958{
3959 if (!zholes_size)
3960 return 0;
3961
3962 return zholes_size[zone_type];
3963}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003964
Mel Gormanc7132162006-09-27 01:49:43 -07003965#endif
3966
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003967static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Mel Gormanc7132162006-09-27 01:49:43 -07003968 unsigned long *zones_size, unsigned long *zholes_size)
3969{
3970 unsigned long realtotalpages, totalpages = 0;
3971 enum zone_type i;
3972
3973 for (i = 0; i < MAX_NR_ZONES; i++)
3974 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3975 zones_size);
3976 pgdat->node_spanned_pages = totalpages;
3977
3978 realtotalpages = totalpages;
3979 for (i = 0; i < MAX_NR_ZONES; i++)
3980 realtotalpages -=
3981 zone_absent_pages_in_node(pgdat->node_id, i,
3982 zholes_size);
3983 pgdat->node_present_pages = realtotalpages;
3984 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3985 realtotalpages);
3986}
3987
Mel Gorman835c1342007-10-16 01:25:47 -07003988#ifndef CONFIG_SPARSEMEM
3989/*
3990 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07003991 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3992 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07003993 * round what is now in bits to nearest long in bits, then return it in
3994 * bytes.
3995 */
3996static unsigned long __init usemap_size(unsigned long zonesize)
3997{
3998 unsigned long usemapsize;
3999
Mel Gormand9c23402007-10-16 01:26:01 -07004000 usemapsize = roundup(zonesize, pageblock_nr_pages);
4001 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07004002 usemapsize *= NR_PAGEBLOCK_BITS;
4003 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4004
4005 return usemapsize / 8;
4006}
4007
4008static void __init setup_usemap(struct pglist_data *pgdat,
4009 struct zone *zone, unsigned long zonesize)
4010{
4011 unsigned long usemapsize = usemap_size(zonesize);
4012 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08004013 if (usemapsize)
Mel Gorman835c1342007-10-16 01:25:47 -07004014 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
Mel Gorman835c1342007-10-16 01:25:47 -07004015}
4016#else
4017static void inline setup_usemap(struct pglist_data *pgdat,
4018 struct zone *zone, unsigned long zonesize) {}
4019#endif /* CONFIG_SPARSEMEM */
4020
Mel Gormand9c23402007-10-16 01:26:01 -07004021#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08004022
4023/* Return a sensible default order for the pageblock size. */
4024static inline int pageblock_default_order(void)
4025{
4026 if (HPAGE_SHIFT > PAGE_SHIFT)
4027 return HUGETLB_PAGE_ORDER;
4028
4029 return MAX_ORDER-1;
4030}
4031
Mel Gormand9c23402007-10-16 01:26:01 -07004032/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4033static inline void __init set_pageblock_order(unsigned int order)
4034{
4035 /* Check that pageblock_nr_pages has not already been setup */
4036 if (pageblock_order)
4037 return;
4038
4039 /*
4040 * Assume the largest contiguous order of interest is a huge page.
4041 * This value may be variable depending on boot parameters on IA64
4042 */
4043 pageblock_order = order;
4044}
4045#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4046
Mel Gormanba72cb82007-11-28 16:21:13 -08004047/*
4048 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4049 * and pageblock_default_order() are unused as pageblock_order is set
4050 * at compile-time. See include/linux/pageblock-flags.h for the values of
4051 * pageblock_order based on the kernel config
4052 */
4053static inline int pageblock_default_order(unsigned int order)
4054{
4055 return MAX_ORDER-1;
4056}
Mel Gormand9c23402007-10-16 01:26:01 -07004057#define set_pageblock_order(x) do {} while (0)
4058
4059#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4060
Linus Torvalds1da177e2005-04-16 15:20:36 -07004061/*
4062 * Set up the zone data structures:
4063 * - mark all pages reserved
4064 * - mark all memory queues empty
4065 * - clear the memory bitmaps
4066 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08004067static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004068 unsigned long *zones_size, unsigned long *zholes_size)
4069{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07004070 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07004071 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004072 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07004073 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004074
Dave Hansen208d54e2005-10-29 18:16:52 -07004075 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004076 pgdat->nr_zones = 0;
4077 init_waitqueue_head(&pgdat->kswapd_wait);
4078 pgdat->kswapd_max_order = 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004079 pgdat_page_cgroup_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004080
4081 for (j = 0; j < MAX_NR_ZONES; j++) {
4082 struct zone *zone = pgdat->node_zones + j;
Mel Gorman0e0b8642006-09-27 01:49:56 -07004083 unsigned long size, realsize, memmap_pages;
Christoph Lameterb69408e2008-10-18 20:26:14 -07004084 enum lru_list l;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004085
Mel Gormanc7132162006-09-27 01:49:43 -07004086 size = zone_spanned_pages_in_node(nid, j, zones_size);
4087 realsize = size - zone_absent_pages_in_node(nid, j,
4088 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004089
Mel Gorman0e0b8642006-09-27 01:49:56 -07004090 /*
4091 * Adjust realsize so that it accounts for how much memory
4092 * is used by this zone for memmap. This affects the watermark
4093 * and per-cpu initialisations
4094 */
Johannes Weinerf7232152008-05-23 13:04:21 -07004095 memmap_pages =
4096 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
Mel Gorman0e0b8642006-09-27 01:49:56 -07004097 if (realsize >= memmap_pages) {
4098 realsize -= memmap_pages;
Yinghai Lu5594c8c2009-01-06 14:39:14 -08004099 if (memmap_pages)
4100 printk(KERN_DEBUG
4101 " %s zone: %lu pages used for memmap\n",
4102 zone_names[j], memmap_pages);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004103 } else
4104 printk(KERN_WARNING
4105 " %s zone: %lu pages exceeds realsize %lu\n",
4106 zone_names[j], memmap_pages, realsize);
4107
Christoph Lameter62672762007-02-10 01:43:07 -08004108 /* Account for reserved pages */
4109 if (j == 0 && realsize > dma_reserve) {
Mel Gorman0e0b8642006-09-27 01:49:56 -07004110 realsize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07004111 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08004112 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004113 }
4114
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004115 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004116 nr_kernel_pages += realsize;
4117 nr_all_pages += realsize;
4118
4119 zone->spanned_pages = size;
4120 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07004121#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07004122 zone->node = nid;
Christoph Lameter8417bba2006-09-25 23:31:51 -07004123 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07004124 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07004125 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07004126#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004127 zone->name = zone_names[j];
4128 spin_lock_init(&zone->lock);
4129 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07004130 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004131 zone->zone_pgdat = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004132
Dave Hansened8ece22005-10-29 18:16:50 -07004133 zone_pcp_init(zone);
Christoph Lameterb69408e2008-10-18 20:26:14 -07004134 for_each_lru(l) {
4135 INIT_LIST_HEAD(&zone->lru[l].list);
Wu Fengguangf8629632009-09-21 17:03:11 -07004136 zone->reclaim_stat.nr_saved_scan[l] = 0;
Christoph Lameterb69408e2008-10-18 20:26:14 -07004137 }
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08004138 zone->reclaim_stat.recent_rotated[0] = 0;
4139 zone->reclaim_stat.recent_rotated[1] = 0;
4140 zone->reclaim_stat.recent_scanned[0] = 0;
4141 zone->reclaim_stat.recent_scanned[1] = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07004142 zap_zone_vm_stats(zone);
David Rientjese815af92007-10-16 23:25:54 -07004143 zone->flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004144 if (!size)
4145 continue;
4146
Mel Gormanba72cb82007-11-28 16:21:13 -08004147 set_pageblock_order(pageblock_default_order());
Mel Gorman835c1342007-10-16 01:25:47 -07004148 setup_usemap(pgdat, zone, size);
Dave Hansena2f3aa02007-01-10 23:15:30 -08004149 ret = init_currently_empty_zone(zone, zone_start_pfn,
4150 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07004151 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07004152 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004153 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004154 }
4155}
4156
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004157static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004158{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004159 /* Skip empty nodes */
4160 if (!pgdat->node_spanned_pages)
4161 return;
4162
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004163#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07004164 /* ia64 gets its own node_mem_map, before this, without bootmem */
4165 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07004166 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004167 struct page *map;
4168
Bob Piccoe984bb42006-05-20 15:00:31 -07004169 /*
4170 * The zone's endpoints aren't required to be MAX_ORDER
4171 * aligned but the node_mem_map endpoints must be in order
4172 * for the buddy allocator to function correctly.
4173 */
4174 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
4175 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
4176 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4177 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07004178 map = alloc_remap(pgdat->node_id, size);
4179 if (!map)
4180 map = alloc_bootmem_node(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07004181 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004182 }
Roman Zippel12d810c2007-05-31 00:40:54 -07004183#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07004184 /*
4185 * With no DISCONTIG, the global mem_map is just set as node 0's
4186 */
Mel Gormanc7132162006-09-27 01:49:43 -07004187 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004188 mem_map = NODE_DATA(0)->node_mem_map;
Mel Gormanc7132162006-09-27 01:49:43 -07004189#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
4190 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08004191 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Mel Gormanc7132162006-09-27 01:49:43 -07004192#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4193 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004194#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004195#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004196}
4197
Johannes Weiner9109fb72008-07-23 21:27:20 -07004198void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4199 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004200{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004201 pg_data_t *pgdat = NODE_DATA(nid);
4202
Linus Torvalds1da177e2005-04-16 15:20:36 -07004203 pgdat->node_id = nid;
4204 pgdat->node_start_pfn = node_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004205 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004206
4207 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07004208#ifdef CONFIG_FLAT_NODE_MEM_MAP
4209 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4210 nid, (unsigned long)pgdat,
4211 (unsigned long)pgdat->node_mem_map);
4212#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004213
4214 free_area_init_core(pgdat, zones_size, zholes_size);
4215}
4216
Mel Gormanc7132162006-09-27 01:49:43 -07004217#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07004218
4219#if MAX_NUMNODES > 1
4220/*
4221 * Figure out the number of possible node ids.
4222 */
4223static void __init setup_nr_node_ids(void)
4224{
4225 unsigned int node;
4226 unsigned int highest = 0;
4227
4228 for_each_node_mask(node, node_possible_map)
4229 highest = node;
4230 nr_node_ids = highest + 1;
4231}
4232#else
4233static inline void setup_nr_node_ids(void)
4234{
4235}
4236#endif
4237
Mel Gormanc7132162006-09-27 01:49:43 -07004238/**
4239 * add_active_range - Register a range of PFNs backed by physical memory
4240 * @nid: The node ID the range resides on
4241 * @start_pfn: The start PFN of the available physical memory
4242 * @end_pfn: The end PFN of the available physical memory
4243 *
4244 * These ranges are stored in an early_node_map[] and later used by
4245 * free_area_init_nodes() to calculate zone sizes and holes. If the
4246 * range spans a memory hole, it is up to the architecture to ensure
4247 * the memory is not freed by the bootmem allocator. If possible
4248 * the range being registered will be merged with existing ranges.
4249 */
4250void __init add_active_range(unsigned int nid, unsigned long start_pfn,
4251 unsigned long end_pfn)
4252{
4253 int i;
4254
Mel Gorman6b74ab92008-07-23 21:26:49 -07004255 mminit_dprintk(MMINIT_TRACE, "memory_register",
4256 "Entering add_active_range(%d, %#lx, %#lx) "
4257 "%d entries of %d used\n",
4258 nid, start_pfn, end_pfn,
4259 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
Mel Gormanc7132162006-09-27 01:49:43 -07004260
Mel Gorman2dbb51c2008-07-23 21:26:52 -07004261 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
4262
Mel Gormanc7132162006-09-27 01:49:43 -07004263 /* Merge with existing active regions if possible */
4264 for (i = 0; i < nr_nodemap_entries; i++) {
4265 if (early_node_map[i].nid != nid)
4266 continue;
4267
4268 /* Skip if an existing region covers this new one */
4269 if (start_pfn >= early_node_map[i].start_pfn &&
4270 end_pfn <= early_node_map[i].end_pfn)
4271 return;
4272
4273 /* Merge forward if suitable */
4274 if (start_pfn <= early_node_map[i].end_pfn &&
4275 end_pfn > early_node_map[i].end_pfn) {
4276 early_node_map[i].end_pfn = end_pfn;
4277 return;
4278 }
4279
4280 /* Merge backward if suitable */
Kazuhisa Ichikawad2dbe082010-01-15 17:01:20 -08004281 if (start_pfn < early_node_map[i].start_pfn &&
Mel Gormanc7132162006-09-27 01:49:43 -07004282 end_pfn >= early_node_map[i].start_pfn) {
4283 early_node_map[i].start_pfn = start_pfn;
4284 return;
4285 }
4286 }
4287
4288 /* Check that early_node_map is large enough */
4289 if (i >= MAX_ACTIVE_REGIONS) {
4290 printk(KERN_CRIT "More than %d memory regions, truncating\n",
4291 MAX_ACTIVE_REGIONS);
4292 return;
4293 }
4294
4295 early_node_map[i].nid = nid;
4296 early_node_map[i].start_pfn = start_pfn;
4297 early_node_map[i].end_pfn = end_pfn;
4298 nr_nodemap_entries = i + 1;
4299}
4300
4301/**
Yinghai Lucc1050b2008-06-13 19:08:52 -07004302 * remove_active_range - Shrink an existing registered range of PFNs
Mel Gormanc7132162006-09-27 01:49:43 -07004303 * @nid: The node id the range is on that should be shrunk
Yinghai Lucc1050b2008-06-13 19:08:52 -07004304 * @start_pfn: The new PFN of the range
4305 * @end_pfn: The new PFN of the range
Mel Gormanc7132162006-09-27 01:49:43 -07004306 *
4307 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004308 * The map is kept near the end physical page range that has already been
4309 * registered. This function allows an arch to shrink an existing registered
4310 * range.
Mel Gormanc7132162006-09-27 01:49:43 -07004311 */
Yinghai Lucc1050b2008-06-13 19:08:52 -07004312void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
4313 unsigned long end_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004314{
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004315 int i, j;
4316 int removed = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07004317
Yinghai Lucc1050b2008-06-13 19:08:52 -07004318 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
4319 nid, start_pfn, end_pfn);
4320
Mel Gormanc7132162006-09-27 01:49:43 -07004321 /* Find the old active region end and shrink */
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004322 for_each_active_range_index_in_nid(i, nid) {
Yinghai Lucc1050b2008-06-13 19:08:52 -07004323 if (early_node_map[i].start_pfn >= start_pfn &&
4324 early_node_map[i].end_pfn <= end_pfn) {
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004325 /* clear it */
Yinghai Lucc1050b2008-06-13 19:08:52 -07004326 early_node_map[i].start_pfn = 0;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004327 early_node_map[i].end_pfn = 0;
4328 removed = 1;
4329 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004330 }
Yinghai Lucc1050b2008-06-13 19:08:52 -07004331 if (early_node_map[i].start_pfn < start_pfn &&
4332 early_node_map[i].end_pfn > start_pfn) {
4333 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
4334 early_node_map[i].end_pfn = start_pfn;
4335 if (temp_end_pfn > end_pfn)
4336 add_active_range(nid, end_pfn, temp_end_pfn);
4337 continue;
4338 }
4339 if (early_node_map[i].start_pfn >= start_pfn &&
4340 early_node_map[i].end_pfn > end_pfn &&
4341 early_node_map[i].start_pfn < end_pfn) {
4342 early_node_map[i].start_pfn = end_pfn;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004343 continue;
4344 }
4345 }
4346
4347 if (!removed)
4348 return;
4349
4350 /* remove the blank ones */
4351 for (i = nr_nodemap_entries - 1; i > 0; i--) {
4352 if (early_node_map[i].nid != nid)
4353 continue;
4354 if (early_node_map[i].end_pfn)
4355 continue;
4356 /* we found it, get rid of it */
4357 for (j = i; j < nr_nodemap_entries - 1; j++)
4358 memcpy(&early_node_map[j], &early_node_map[j+1],
4359 sizeof(early_node_map[j]));
4360 j = nr_nodemap_entries - 1;
4361 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
4362 nr_nodemap_entries--;
4363 }
Mel Gormanc7132162006-09-27 01:49:43 -07004364}
4365
4366/**
4367 * remove_all_active_ranges - Remove all currently registered regions
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004368 *
Mel Gormanc7132162006-09-27 01:49:43 -07004369 * During discovery, it may be found that a table like SRAT is invalid
4370 * and an alternative discovery method must be used. This function removes
4371 * all currently registered regions.
4372 */
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004373void __init remove_all_active_ranges(void)
Mel Gormanc7132162006-09-27 01:49:43 -07004374{
4375 memset(early_node_map, 0, sizeof(early_node_map));
4376 nr_nodemap_entries = 0;
4377}
4378
4379/* Compare two active node_active_regions */
4380static int __init cmp_node_active_region(const void *a, const void *b)
4381{
4382 struct node_active_region *arange = (struct node_active_region *)a;
4383 struct node_active_region *brange = (struct node_active_region *)b;
4384
4385 /* Done this way to avoid overflows */
4386 if (arange->start_pfn > brange->start_pfn)
4387 return 1;
4388 if (arange->start_pfn < brange->start_pfn)
4389 return -1;
4390
4391 return 0;
4392}
4393
4394/* sort the node_map by start_pfn */
Yinghai Lu32996252009-12-15 17:59:02 -08004395void __init sort_node_map(void)
Mel Gormanc7132162006-09-27 01:49:43 -07004396{
4397 sort(early_node_map, (size_t)nr_nodemap_entries,
4398 sizeof(struct node_active_region),
4399 cmp_node_active_region, NULL);
4400}
4401
Mel Gormana6af2bc2007-02-10 01:42:57 -08004402/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004403static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07004404{
4405 int i;
Mel Gormana6af2bc2007-02-10 01:42:57 -08004406 unsigned long min_pfn = ULONG_MAX;
Mel Gorman1abbfb42006-11-23 12:01:41 +00004407
Mel Gormanc7132162006-09-27 01:49:43 -07004408 /* Assuming a sorted map, the first range found has the starting pfn */
4409 for_each_active_range_index_in_nid(i, nid)
Mel Gormana6af2bc2007-02-10 01:42:57 -08004410 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004411
Mel Gormana6af2bc2007-02-10 01:42:57 -08004412 if (min_pfn == ULONG_MAX) {
4413 printk(KERN_WARNING
Paul Jackson2bc0d262008-06-22 07:22:17 -07004414 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08004415 return 0;
4416 }
4417
4418 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004419}
4420
4421/**
4422 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4423 *
4424 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004425 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07004426 */
4427unsigned long __init find_min_pfn_with_active_regions(void)
4428{
4429 return find_min_pfn_for_node(MAX_NUMNODES);
4430}
4431
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004432/*
4433 * early_calculate_totalpages()
4434 * Sum pages in active regions for movable zone.
4435 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4436 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07004437static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07004438{
4439 int i;
4440 unsigned long totalpages = 0;
4441
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004442 for (i = 0; i < nr_nodemap_entries; i++) {
4443 unsigned long pages = early_node_map[i].end_pfn -
Mel Gorman7e63efe2007-07-17 04:03:15 -07004444 early_node_map[i].start_pfn;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004445 totalpages += pages;
4446 if (pages)
4447 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
4448 }
4449 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07004450}
4451
Mel Gorman2a1e2742007-07-17 04:03:12 -07004452/*
4453 * Find the PFN the Movable zone begins in each node. Kernel memory
4454 * is spread evenly between nodes as long as the nodes have enough
4455 * memory. When they don't, some nodes will have more kernelcore than
4456 * others
4457 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004458static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004459{
4460 int i, nid;
4461 unsigned long usable_startpfn;
4462 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07004463 /* save the state before borrow the nodemask */
4464 nodemask_t saved_node_state = node_states[N_HIGH_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004465 unsigned long totalpages = early_calculate_totalpages();
4466 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004467
Mel Gorman7e63efe2007-07-17 04:03:15 -07004468 /*
4469 * If movablecore was specified, calculate what size of
4470 * kernelcore that corresponds so that memory usable for
4471 * any allocation type is evenly spread. If both kernelcore
4472 * and movablecore are specified, then the value of kernelcore
4473 * will be used for required_kernelcore if it's greater than
4474 * what movablecore would have allowed.
4475 */
4476 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07004477 unsigned long corepages;
4478
4479 /*
4480 * Round-up so that ZONE_MOVABLE is at least as large as what
4481 * was requested by the user
4482 */
4483 required_movablecore =
4484 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
4485 corepages = totalpages - required_movablecore;
4486
4487 required_kernelcore = max(required_kernelcore, corepages);
4488 }
4489
Mel Gorman2a1e2742007-07-17 04:03:12 -07004490 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4491 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07004492 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004493
4494 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4495 find_usable_zone_for_movable();
4496 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4497
4498restart:
4499 /* Spread kernelcore memory as evenly as possible throughout nodes */
4500 kernelcore_node = required_kernelcore / usable_nodes;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004501 for_each_node_state(nid, N_HIGH_MEMORY) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004502 /*
4503 * Recalculate kernelcore_node if the division per node
4504 * now exceeds what is necessary to satisfy the requested
4505 * amount of memory for the kernel
4506 */
4507 if (required_kernelcore < kernelcore_node)
4508 kernelcore_node = required_kernelcore / usable_nodes;
4509
4510 /*
4511 * As the map is walked, we track how much memory is usable
4512 * by the kernel using kernelcore_remaining. When it is
4513 * 0, the rest of the node is usable by ZONE_MOVABLE
4514 */
4515 kernelcore_remaining = kernelcore_node;
4516
4517 /* Go through each range of PFNs within this node */
4518 for_each_active_range_index_in_nid(i, nid) {
4519 unsigned long start_pfn, end_pfn;
4520 unsigned long size_pages;
4521
4522 start_pfn = max(early_node_map[i].start_pfn,
4523 zone_movable_pfn[nid]);
4524 end_pfn = early_node_map[i].end_pfn;
4525 if (start_pfn >= end_pfn)
4526 continue;
4527
4528 /* Account for what is only usable for kernelcore */
4529 if (start_pfn < usable_startpfn) {
4530 unsigned long kernel_pages;
4531 kernel_pages = min(end_pfn, usable_startpfn)
4532 - start_pfn;
4533
4534 kernelcore_remaining -= min(kernel_pages,
4535 kernelcore_remaining);
4536 required_kernelcore -= min(kernel_pages,
4537 required_kernelcore);
4538
4539 /* Continue if range is now fully accounted */
4540 if (end_pfn <= usable_startpfn) {
4541
4542 /*
4543 * Push zone_movable_pfn to the end so
4544 * that if we have to rebalance
4545 * kernelcore across nodes, we will
4546 * not double account here
4547 */
4548 zone_movable_pfn[nid] = end_pfn;
4549 continue;
4550 }
4551 start_pfn = usable_startpfn;
4552 }
4553
4554 /*
4555 * The usable PFN range for ZONE_MOVABLE is from
4556 * start_pfn->end_pfn. Calculate size_pages as the
4557 * number of pages used as kernelcore
4558 */
4559 size_pages = end_pfn - start_pfn;
4560 if (size_pages > kernelcore_remaining)
4561 size_pages = kernelcore_remaining;
4562 zone_movable_pfn[nid] = start_pfn + size_pages;
4563
4564 /*
4565 * Some kernelcore has been met, update counts and
4566 * break if the kernelcore for this node has been
4567 * satisified
4568 */
4569 required_kernelcore -= min(required_kernelcore,
4570 size_pages);
4571 kernelcore_remaining -= size_pages;
4572 if (!kernelcore_remaining)
4573 break;
4574 }
4575 }
4576
4577 /*
4578 * If there is still required_kernelcore, we do another pass with one
4579 * less node in the count. This will push zone_movable_pfn[nid] further
4580 * along on the nodes that still have memory until kernelcore is
4581 * satisified
4582 */
4583 usable_nodes--;
4584 if (usable_nodes && required_kernelcore > usable_nodes)
4585 goto restart;
4586
4587 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4588 for (nid = 0; nid < MAX_NUMNODES; nid++)
4589 zone_movable_pfn[nid] =
4590 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07004591
4592out:
4593 /* restore the node_state */
4594 node_states[N_HIGH_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004595}
4596
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004597/* Any regular memory on that node ? */
4598static void check_for_regular_memory(pg_data_t *pgdat)
4599{
4600#ifdef CONFIG_HIGHMEM
4601 enum zone_type zone_type;
4602
4603 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4604 struct zone *zone = &pgdat->node_zones[zone_type];
4605 if (zone->present_pages)
4606 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4607 }
4608#endif
4609}
4610
Mel Gormanc7132162006-09-27 01:49:43 -07004611/**
4612 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004613 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07004614 *
4615 * This will call free_area_init_node() for each active node in the system.
4616 * Using the page ranges provided by add_active_range(), the size of each
4617 * zone in each node and their holes is calculated. If the maximum PFN
4618 * between two adjacent zones match, it is assumed that the zone is empty.
4619 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4620 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4621 * starts where the previous one ended. For example, ZONE_DMA32 starts
4622 * at arch_max_dma_pfn.
4623 */
4624void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4625{
4626 unsigned long nid;
Andrew Mortondb991002008-10-15 22:01:07 -07004627 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004628
Mel Gormana6af2bc2007-02-10 01:42:57 -08004629 /* Sort early_node_map as initialisation assumes it is sorted */
4630 sort_node_map();
4631
Mel Gormanc7132162006-09-27 01:49:43 -07004632 /* Record where the zone boundaries are */
4633 memset(arch_zone_lowest_possible_pfn, 0,
4634 sizeof(arch_zone_lowest_possible_pfn));
4635 memset(arch_zone_highest_possible_pfn, 0,
4636 sizeof(arch_zone_highest_possible_pfn));
4637 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4638 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4639 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004640 if (i == ZONE_MOVABLE)
4641 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004642 arch_zone_lowest_possible_pfn[i] =
4643 arch_zone_highest_possible_pfn[i-1];
4644 arch_zone_highest_possible_pfn[i] =
4645 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4646 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07004647 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4648 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4649
4650 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4651 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4652 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004653
Mel Gormanc7132162006-09-27 01:49:43 -07004654 /* Print out the zone ranges */
4655 printk("Zone PFN ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07004656 for (i = 0; i < MAX_NR_ZONES; i++) {
4657 if (i == ZONE_MOVABLE)
4658 continue;
David Rientjes72f0ba02010-03-05 13:42:14 -08004659 printk(" %-8s ", zone_names[i]);
4660 if (arch_zone_lowest_possible_pfn[i] ==
4661 arch_zone_highest_possible_pfn[i])
4662 printk("empty\n");
4663 else
4664 printk("%0#10lx -> %0#10lx\n",
Mel Gormanc7132162006-09-27 01:49:43 -07004665 arch_zone_lowest_possible_pfn[i],
4666 arch_zone_highest_possible_pfn[i]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004667 }
4668
4669 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4670 printk("Movable zone start PFN for each node\n");
4671 for (i = 0; i < MAX_NUMNODES; i++) {
4672 if (zone_movable_pfn[i])
4673 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4674 }
Mel Gormanc7132162006-09-27 01:49:43 -07004675
4676 /* Print out the early_node_map[] */
4677 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4678 for (i = 0; i < nr_nodemap_entries; i++)
Paul Jackson5dab8ec2008-06-25 05:44:40 -07004679 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004680 early_node_map[i].start_pfn,
4681 early_node_map[i].end_pfn);
4682
4683 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07004684 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08004685 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07004686 for_each_online_node(nid) {
4687 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07004688 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07004689 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004690
4691 /* Any memory on that node */
4692 if (pgdat->node_present_pages)
4693 node_set_state(nid, N_HIGH_MEMORY);
4694 check_for_regular_memory(pgdat);
Mel Gormanc7132162006-09-27 01:49:43 -07004695 }
4696}
Mel Gorman2a1e2742007-07-17 04:03:12 -07004697
Mel Gorman7e63efe2007-07-17 04:03:15 -07004698static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004699{
4700 unsigned long long coremem;
4701 if (!p)
4702 return -EINVAL;
4703
4704 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004705 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004706
Mel Gorman7e63efe2007-07-17 04:03:15 -07004707 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07004708 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4709
4710 return 0;
4711}
Mel Gormaned7ed362007-07-17 04:03:14 -07004712
Mel Gorman7e63efe2007-07-17 04:03:15 -07004713/*
4714 * kernelcore=size sets the amount of memory for use for allocations that
4715 * cannot be reclaimed or migrated.
4716 */
4717static int __init cmdline_parse_kernelcore(char *p)
4718{
4719 return cmdline_parse_core(p, &required_kernelcore);
4720}
4721
4722/*
4723 * movablecore=size sets the amount of memory for use for allocations that
4724 * can be reclaimed or migrated.
4725 */
4726static int __init cmdline_parse_movablecore(char *p)
4727{
4728 return cmdline_parse_core(p, &required_movablecore);
4729}
4730
Mel Gormaned7ed362007-07-17 04:03:14 -07004731early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004732early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07004733
Mel Gormanc7132162006-09-27 01:49:43 -07004734#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4735
Mel Gorman0e0b8642006-09-27 01:49:56 -07004736/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004737 * set_dma_reserve - set the specified number of pages reserved in the first zone
4738 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07004739 *
4740 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4741 * In the DMA zone, a significant percentage may be consumed by kernel image
4742 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004743 * function may optionally be used to account for unfreeable pages in the
4744 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4745 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07004746 */
4747void __init set_dma_reserve(unsigned long new_dma_reserve)
4748{
4749 dma_reserve = new_dma_reserve;
4750}
4751
Dave Hansen93b75042005-06-23 00:07:47 -07004752#ifndef CONFIG_NEED_MULTIPLE_NODES
Yinghai Lu08677212010-02-10 01:20:20 -08004753struct pglist_data __refdata contig_page_data = {
4754#ifndef CONFIG_NO_BOOTMEM
4755 .bdata = &bootmem_node_data[0]
4756#endif
4757 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07004758EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07004759#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004760
4761void __init free_area_init(unsigned long *zones_size)
4762{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004763 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004764 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4765}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004766
Linus Torvalds1da177e2005-04-16 15:20:36 -07004767static int page_alloc_cpu_notify(struct notifier_block *self,
4768 unsigned long action, void *hcpu)
4769{
4770 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004771
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07004772 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004773 drain_pages(cpu);
4774
4775 /*
4776 * Spill the event counters of the dead processor
4777 * into the current processors event counters.
4778 * This artificially elevates the count of the current
4779 * processor.
4780 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07004781 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004782
4783 /*
4784 * Zero the differential counters of the dead processor
4785 * so that the vm statistics are consistent.
4786 *
4787 * This is only okay since the processor is dead and cannot
4788 * race with what we are doing.
4789 */
Christoph Lameter2244b952006-06-30 01:55:33 -07004790 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004791 }
4792 return NOTIFY_OK;
4793}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004794
4795void __init page_alloc_init(void)
4796{
4797 hotcpu_notifier(page_alloc_cpu_notify, 0);
4798}
4799
4800/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004801 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4802 * or min_free_kbytes changes.
4803 */
4804static void calculate_totalreserve_pages(void)
4805{
4806 struct pglist_data *pgdat;
4807 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004808 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004809
4810 for_each_online_pgdat(pgdat) {
4811 for (i = 0; i < MAX_NR_ZONES; i++) {
4812 struct zone *zone = pgdat->node_zones + i;
4813 unsigned long max = 0;
4814
4815 /* Find valid and maximum lowmem_reserve in the zone */
4816 for (j = i; j < MAX_NR_ZONES; j++) {
4817 if (zone->lowmem_reserve[j] > max)
4818 max = zone->lowmem_reserve[j];
4819 }
4820
Mel Gorman41858962009-06-16 15:32:12 -07004821 /* we treat the high watermark as reserved pages. */
4822 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004823
4824 if (max > zone->present_pages)
4825 max = zone->present_pages;
4826 reserve_pages += max;
4827 }
4828 }
4829 totalreserve_pages = reserve_pages;
4830}
4831
4832/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07004833 * setup_per_zone_lowmem_reserve - called whenever
4834 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4835 * has a correct pages reserved value, so an adequate number of
4836 * pages are left in the zone after a successful __alloc_pages().
4837 */
4838static void setup_per_zone_lowmem_reserve(void)
4839{
4840 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004841 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004842
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08004843 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004844 for (j = 0; j < MAX_NR_ZONES; j++) {
4845 struct zone *zone = pgdat->node_zones + j;
4846 unsigned long present_pages = zone->present_pages;
4847
4848 zone->lowmem_reserve[j] = 0;
4849
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004850 idx = j;
4851 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004852 struct zone *lower_zone;
4853
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004854 idx--;
4855
Linus Torvalds1da177e2005-04-16 15:20:36 -07004856 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4857 sysctl_lowmem_reserve_ratio[idx] = 1;
4858
4859 lower_zone = pgdat->node_zones + idx;
4860 lower_zone->lowmem_reserve[j] = present_pages /
4861 sysctl_lowmem_reserve_ratio[idx];
4862 present_pages += lower_zone->present_pages;
4863 }
4864 }
4865 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004866
4867 /* update totalreserve_pages */
4868 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004869}
4870
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004871/**
Minchan Kimbc75d332009-06-16 15:32:48 -07004872 * setup_per_zone_wmarks - called when min_free_kbytes changes
Minchan Kimbce73942009-06-16 15:32:50 -07004873 * or when memory is hot-{added|removed}
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004874 *
Minchan Kimbc75d332009-06-16 15:32:48 -07004875 * Ensures that the watermark[min,low,high] values for each zone are set
4876 * correctly with respect to min_free_kbytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004877 */
Minchan Kimbc75d332009-06-16 15:32:48 -07004878void setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004879{
4880 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4881 unsigned long lowmem_pages = 0;
4882 struct zone *zone;
4883 unsigned long flags;
4884
4885 /* Calculate total number of !ZONE_HIGHMEM pages */
4886 for_each_zone(zone) {
4887 if (!is_highmem(zone))
4888 lowmem_pages += zone->present_pages;
4889 }
4890
4891 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07004892 u64 tmp;
4893
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07004894 spin_lock_irqsave(&zone->lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07004895 tmp = (u64)pages_min * zone->present_pages;
4896 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004897 if (is_highmem(zone)) {
4898 /*
Nick Piggin669ed172005-11-13 16:06:45 -08004899 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4900 * need highmem pages, so cap pages_min to a small
4901 * value here.
4902 *
Mel Gorman41858962009-06-16 15:32:12 -07004903 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Nick Piggin669ed172005-11-13 16:06:45 -08004904 * deltas controls asynch page reclaim, and so should
4905 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004906 */
4907 int min_pages;
4908
4909 min_pages = zone->present_pages / 1024;
4910 if (min_pages < SWAP_CLUSTER_MAX)
4911 min_pages = SWAP_CLUSTER_MAX;
4912 if (min_pages > 128)
4913 min_pages = 128;
Mel Gorman41858962009-06-16 15:32:12 -07004914 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004915 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08004916 /*
4917 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07004918 * proportionate to the zone's size.
4919 */
Mel Gorman41858962009-06-16 15:32:12 -07004920 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004921 }
4922
Mel Gorman41858962009-06-16 15:32:12 -07004923 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
4924 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Mel Gorman56fd56b2007-10-16 01:25:58 -07004925 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07004926 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004927 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004928
4929 /* update totalreserve_pages */
4930 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004931}
4932
Randy Dunlap55a44622009-09-21 17:01:20 -07004933/*
Rik van Riel556adec2008-10-18 20:26:34 -07004934 * The inactive anon list should be small enough that the VM never has to
4935 * do too much work, but large enough that each inactive page has a chance
4936 * to be referenced again before it is swapped out.
4937 *
4938 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4939 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4940 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4941 * the anonymous pages are kept on the inactive list.
4942 *
4943 * total target max
4944 * memory ratio inactive anon
4945 * -------------------------------------
4946 * 10MB 1 5MB
4947 * 100MB 1 50MB
4948 * 1GB 3 250MB
4949 * 10GB 10 0.9GB
4950 * 100GB 31 3GB
4951 * 1TB 101 10GB
4952 * 10TB 320 32GB
4953 */
Minchan Kim96cb4df2009-06-16 15:32:49 -07004954void calculate_zone_inactive_ratio(struct zone *zone)
4955{
4956 unsigned int gb, ratio;
4957
4958 /* Zone size in gigabytes */
4959 gb = zone->present_pages >> (30 - PAGE_SHIFT);
4960 if (gb)
4961 ratio = int_sqrt(10 * gb);
4962 else
4963 ratio = 1;
4964
4965 zone->inactive_ratio = ratio;
4966}
4967
Cyrill Gorcunove9bb35d2009-06-16 15:32:32 -07004968static void __init setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07004969{
4970 struct zone *zone;
4971
Minchan Kim96cb4df2009-06-16 15:32:49 -07004972 for_each_zone(zone)
4973 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07004974}
4975
Linus Torvalds1da177e2005-04-16 15:20:36 -07004976/*
4977 * Initialise min_free_kbytes.
4978 *
4979 * For small machines we want it small (128k min). For large machines
4980 * we want it large (64MB max). But it is not linear, because network
4981 * bandwidth does not increase linearly with machine size. We use
4982 *
4983 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4984 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4985 *
4986 * which yields
4987 *
4988 * 16MB: 512k
4989 * 32MB: 724k
4990 * 64MB: 1024k
4991 * 128MB: 1448k
4992 * 256MB: 2048k
4993 * 512MB: 2896k
4994 * 1024MB: 4096k
4995 * 2048MB: 5792k
4996 * 4096MB: 8192k
4997 * 8192MB: 11584k
4998 * 16384MB: 16384k
4999 */
Minchan Kimbc75d332009-06-16 15:32:48 -07005000static int __init init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005001{
5002 unsigned long lowmem_kbytes;
5003
5004 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
5005
5006 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
5007 if (min_free_kbytes < 128)
5008 min_free_kbytes = 128;
5009 if (min_free_kbytes > 65536)
5010 min_free_kbytes = 65536;
Minchan Kimbc75d332009-06-16 15:32:48 -07005011 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005012 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07005013 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005014 return 0;
5015}
Minchan Kimbc75d332009-06-16 15:32:48 -07005016module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005017
5018/*
5019 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5020 * that we can call two helper functions whenever min_free_kbytes
5021 * changes.
5022 */
5023int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005024 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005025{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005026 proc_dointvec(table, write, buffer, length, ppos);
Mel Gorman3b1d92c2007-05-06 14:49:30 -07005027 if (write)
Minchan Kimbc75d332009-06-16 15:32:48 -07005028 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005029 return 0;
5030}
5031
Christoph Lameter96146342006-07-03 00:24:13 -07005032#ifdef CONFIG_NUMA
5033int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005034 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07005035{
5036 struct zone *zone;
5037 int rc;
5038
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005039 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07005040 if (rc)
5041 return rc;
5042
5043 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07005044 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07005045 sysctl_min_unmapped_ratio) / 100;
5046 return 0;
5047}
Christoph Lameter0ff38492006-09-25 23:31:52 -07005048
5049int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005050 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07005051{
5052 struct zone *zone;
5053 int rc;
5054
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005055 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07005056 if (rc)
5057 return rc;
5058
5059 for_each_zone(zone)
5060 zone->min_slab_pages = (zone->present_pages *
5061 sysctl_min_slab_ratio) / 100;
5062 return 0;
5063}
Christoph Lameter96146342006-07-03 00:24:13 -07005064#endif
5065
Linus Torvalds1da177e2005-04-16 15:20:36 -07005066/*
5067 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5068 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5069 * whenever sysctl_lowmem_reserve_ratio changes.
5070 *
5071 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07005072 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07005073 * if in function of the boot time zone sizes.
5074 */
5075int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005076 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005077{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005078 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005079 setup_per_zone_lowmem_reserve();
5080 return 0;
5081}
5082
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005083/*
5084 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5085 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5086 * can have before it gets flushed back to buddy allocator.
5087 */
5088
5089int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005090 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005091{
5092 struct zone *zone;
5093 unsigned int cpu;
5094 int ret;
5095
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005096 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005097 if (!write || (ret == -EINVAL))
5098 return ret;
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07005099 for_each_populated_zone(zone) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005100 for_each_possible_cpu(cpu) {
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005101 unsigned long high;
5102 high = zone->present_pages / percpu_pagelist_fraction;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005103 setup_pagelist_highmark(
5104 per_cpu_ptr(zone->pageset, cpu), high);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005105 }
5106 }
5107 return 0;
5108}
5109
David S. Millerf034b5d2006-08-24 03:08:07 -07005110int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005111
5112#ifdef CONFIG_NUMA
5113static int __init set_hashdist(char *str)
5114{
5115 if (!str)
5116 return 0;
5117 hashdist = simple_strtoul(str, &str, 0);
5118 return 1;
5119}
5120__setup("hashdist=", set_hashdist);
5121#endif
5122
5123/*
5124 * allocate a large system hash table from bootmem
5125 * - it is assumed that the hash table must contain an exact power-of-2
5126 * quantity of entries
5127 * - limit is the number of hash buckets, not the total allocation size
5128 */
5129void *__init alloc_large_system_hash(const char *tablename,
5130 unsigned long bucketsize,
5131 unsigned long numentries,
5132 int scale,
5133 int flags,
5134 unsigned int *_hash_shift,
5135 unsigned int *_hash_mask,
5136 unsigned long limit)
5137{
5138 unsigned long long max = limit;
5139 unsigned long log2qty, size;
5140 void *table = NULL;
5141
5142 /* allow the kernel cmdline to have a say */
5143 if (!numentries) {
5144 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08005145 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005146 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
5147 numentries >>= 20 - PAGE_SHIFT;
5148 numentries <<= 20 - PAGE_SHIFT;
5149
5150 /* limit to 1 bucket per 2^scale bytes of low memory */
5151 if (scale > PAGE_SHIFT)
5152 numentries >>= (scale - PAGE_SHIFT);
5153 else
5154 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08005155
5156 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07005157 if (unlikely(flags & HASH_SMALL)) {
5158 /* Makes no sense without HASH_EARLY */
5159 WARN_ON(!(flags & HASH_EARLY));
5160 if (!(numentries >> *_hash_shift)) {
5161 numentries = 1UL << *_hash_shift;
5162 BUG_ON(!numentries);
5163 }
5164 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08005165 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005166 }
John Hawkes6e692ed2006-03-25 03:08:02 -08005167 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005168
5169 /* limit allocation size to 1/16 total memory by default */
5170 if (max == 0) {
5171 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
5172 do_div(max, bucketsize);
5173 }
5174
5175 if (numentries > max)
5176 numentries = max;
5177
David Howellsf0d1b0b2006-12-08 02:37:49 -08005178 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005179
5180 do {
5181 size = bucketsize << log2qty;
5182 if (flags & HASH_EARLY)
Jan Beulich74768ed2008-08-12 15:08:39 -07005183 table = alloc_bootmem_nopanic(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005184 else if (hashdist)
5185 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
5186 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07005187 /*
5188 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07005189 * some pages at the end of hash table which
5190 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07005191 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005192 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07005193 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005194 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
5195 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005196 }
5197 } while (!table && size > PAGE_SIZE && --log2qty);
5198
5199 if (!table)
5200 panic("Failed to allocate %s hash table\n", tablename);
5201
Robin Holtf241e662010-10-07 12:59:26 -07005202 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07005203 tablename,
Robin Holtf241e662010-10-07 12:59:26 -07005204 (1UL << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08005205 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005206 size);
5207
5208 if (_hash_shift)
5209 *_hash_shift = log2qty;
5210 if (_hash_mask)
5211 *_hash_mask = (1 << log2qty) - 1;
5212
5213 return table;
5214}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08005215
Mel Gorman835c1342007-10-16 01:25:47 -07005216/* Return a pointer to the bitmap storing bits affecting a block of pages */
5217static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
5218 unsigned long pfn)
5219{
5220#ifdef CONFIG_SPARSEMEM
5221 return __pfn_to_section(pfn)->pageblock_flags;
5222#else
5223 return zone->pageblock_flags;
5224#endif /* CONFIG_SPARSEMEM */
5225}
Andrew Morton6220ec72006-10-19 23:29:05 -07005226
Mel Gorman835c1342007-10-16 01:25:47 -07005227static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
5228{
5229#ifdef CONFIG_SPARSEMEM
5230 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07005231 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005232#else
5233 pfn = pfn - zone->zone_start_pfn;
Mel Gormand9c23402007-10-16 01:26:01 -07005234 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005235#endif /* CONFIG_SPARSEMEM */
5236}
5237
5238/**
Mel Gormand9c23402007-10-16 01:26:01 -07005239 * 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 -07005240 * @page: The page within the block of interest
5241 * @start_bitidx: The first bit of interest to retrieve
5242 * @end_bitidx: The last bit of interest
5243 * returns pageblock_bits flags
5244 */
5245unsigned long get_pageblock_flags_group(struct page *page,
5246 int start_bitidx, int end_bitidx)
5247{
5248 struct zone *zone;
5249 unsigned long *bitmap;
5250 unsigned long pfn, bitidx;
5251 unsigned long flags = 0;
5252 unsigned long value = 1;
5253
5254 zone = page_zone(page);
5255 pfn = page_to_pfn(page);
5256 bitmap = get_pageblock_bitmap(zone, pfn);
5257 bitidx = pfn_to_bitidx(zone, pfn);
5258
5259 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5260 if (test_bit(bitidx + start_bitidx, bitmap))
5261 flags |= value;
5262
5263 return flags;
5264}
5265
5266/**
Mel Gormand9c23402007-10-16 01:26:01 -07005267 * 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 -07005268 * @page: The page within the block of interest
5269 * @start_bitidx: The first bit of interest
5270 * @end_bitidx: The last bit of interest
5271 * @flags: The flags to set
5272 */
5273void set_pageblock_flags_group(struct page *page, unsigned long flags,
5274 int start_bitidx, int end_bitidx)
5275{
5276 struct zone *zone;
5277 unsigned long *bitmap;
5278 unsigned long pfn, bitidx;
5279 unsigned long value = 1;
5280
5281 zone = page_zone(page);
5282 pfn = page_to_pfn(page);
5283 bitmap = get_pageblock_bitmap(zone, pfn);
5284 bitidx = pfn_to_bitidx(zone, pfn);
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07005285 VM_BUG_ON(pfn < zone->zone_start_pfn);
5286 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
Mel Gorman835c1342007-10-16 01:25:47 -07005287
5288 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5289 if (flags & value)
5290 __set_bit(bitidx + start_bitidx, bitmap);
5291 else
5292 __clear_bit(bitidx + start_bitidx, bitmap);
5293}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005294
5295/*
5296 * This is designed as sub function...plz see page_isolation.c also.
5297 * set/clear page block's type to be ISOLATE.
5298 * page allocater never alloc memory from ISOLATE block.
5299 */
5300
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005301static int
5302__count_immobile_pages(struct zone *zone, struct page *page, int count)
5303{
5304 unsigned long pfn, iter, found;
5305 /*
5306 * For avoiding noise data, lru_add_drain_all() should be called
5307 * If ZONE_MOVABLE, the zone never contains immobile pages
5308 */
5309 if (zone_idx(zone) == ZONE_MOVABLE)
5310 return true;
5311
5312 if (get_pageblock_migratetype(page) == MIGRATE_MOVABLE)
5313 return true;
5314
5315 pfn = page_to_pfn(page);
5316 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
5317 unsigned long check = pfn + iter;
5318
5319 if (!pfn_valid_within(check)) {
5320 iter++;
5321 continue;
5322 }
5323 page = pfn_to_page(check);
5324 if (!page_count(page)) {
5325 if (PageBuddy(page))
5326 iter += (1 << page_order(page)) - 1;
5327 continue;
5328 }
5329 if (!PageLRU(page))
5330 found++;
5331 /*
5332 * If there are RECLAIMABLE pages, we need to check it.
5333 * But now, memory offline itself doesn't call shrink_slab()
5334 * and it still to be fixed.
5335 */
5336 /*
5337 * If the page is not RAM, page_count()should be 0.
5338 * we don't need more check. This is an _used_ not-movable page.
5339 *
5340 * The problematic thing here is PG_reserved pages. PG_reserved
5341 * is set to both of a memory hole page and a _used_ kernel
5342 * page at boot.
5343 */
5344 if (found > count)
5345 return false;
5346 }
5347 return true;
5348}
5349
5350bool is_pageblock_removable_nolock(struct page *page)
5351{
5352 struct zone *zone = page_zone(page);
5353 return __count_immobile_pages(zone, page, 0);
5354}
5355
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005356int set_migratetype_isolate(struct page *page)
5357{
5358 struct zone *zone;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005359 unsigned long flags, pfn;
Robert Jennings925cc712009-12-17 14:44:38 +00005360 struct memory_isolate_notify arg;
5361 int notifier_ret;
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005362 int ret = -EBUSY;
Shaohua Li8e7e40d2009-09-21 17:01:18 -07005363 int zone_idx;
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005364
5365 zone = page_zone(page);
Shaohua Li8e7e40d2009-09-21 17:01:18 -07005366 zone_idx = zone_idx(zone);
Robert Jennings925cc712009-12-17 14:44:38 +00005367
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005368 spin_lock_irqsave(&zone->lock, flags);
Robert Jennings925cc712009-12-17 14:44:38 +00005369
5370 pfn = page_to_pfn(page);
5371 arg.start_pfn = pfn;
5372 arg.nr_pages = pageblock_nr_pages;
5373 arg.pages_found = 0;
5374
5375 /*
5376 * It may be possible to isolate a pageblock even if the
5377 * migratetype is not MIGRATE_MOVABLE. The memory isolation
5378 * notifier chain is used by balloon drivers to return the
5379 * number of pages in a range that are held by the balloon
5380 * driver to shrink memory. If all the pages are accounted for
5381 * by balloons, are free, or on the LRU, isolation can continue.
5382 * Later, for example, when memory hotplug notifier runs, these
5383 * pages reported as "can be isolated" should be isolated(freed)
5384 * by the balloon driver through the memory notifier chain.
5385 */
5386 notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg);
5387 notifier_ret = notifier_to_errno(notifier_ret);
KAMEZAWA Hiroyuki4b204772010-10-26 14:21:29 -07005388 if (notifier_ret)
Robert Jennings925cc712009-12-17 14:44:38 +00005389 goto out;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005390 /*
5391 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
5392 * We just check MOVABLE pages.
5393 */
5394 if (__count_immobile_pages(zone, page, arg.pages_found))
Robert Jennings925cc712009-12-17 14:44:38 +00005395 ret = 0;
5396
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005397 /*
5398 * immobile means "not-on-lru" paes. If immobile is larger than
5399 * removable-by-driver pages reported by notifier, we'll fail.
5400 */
5401
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005402out:
Robert Jennings925cc712009-12-17 14:44:38 +00005403 if (!ret) {
5404 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
5405 move_freepages_block(zone, page, MIGRATE_ISOLATE);
5406 }
5407
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005408 spin_unlock_irqrestore(&zone->lock, flags);
5409 if (!ret)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005410 drain_all_pages();
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005411 return ret;
5412}
5413
5414void unset_migratetype_isolate(struct page *page)
5415{
5416 struct zone *zone;
5417 unsigned long flags;
5418 zone = page_zone(page);
5419 spin_lock_irqsave(&zone->lock, flags);
5420 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
5421 goto out;
5422 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5423 move_freepages_block(zone, page, MIGRATE_MOVABLE);
5424out:
5425 spin_unlock_irqrestore(&zone->lock, flags);
5426}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07005427
5428#ifdef CONFIG_MEMORY_HOTREMOVE
5429/*
5430 * All pages in the range must be isolated before calling this.
5431 */
5432void
5433__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
5434{
5435 struct page *page;
5436 struct zone *zone;
5437 int order, i;
5438 unsigned long pfn;
5439 unsigned long flags;
5440 /* find the first valid pfn */
5441 for (pfn = start_pfn; pfn < end_pfn; pfn++)
5442 if (pfn_valid(pfn))
5443 break;
5444 if (pfn == end_pfn)
5445 return;
5446 zone = page_zone(pfn_to_page(pfn));
5447 spin_lock_irqsave(&zone->lock, flags);
5448 pfn = start_pfn;
5449 while (pfn < end_pfn) {
5450 if (!pfn_valid(pfn)) {
5451 pfn++;
5452 continue;
5453 }
5454 page = pfn_to_page(pfn);
5455 BUG_ON(page_count(page));
5456 BUG_ON(!PageBuddy(page));
5457 order = page_order(page);
5458#ifdef CONFIG_DEBUG_VM
5459 printk(KERN_INFO "remove from free list %lx %d %lx\n",
5460 pfn, 1 << order, end_pfn);
5461#endif
5462 list_del(&page->lru);
5463 rmv_page_order(page);
5464 zone->free_area[order].nr_free--;
5465 __mod_zone_page_state(zone, NR_FREE_PAGES,
5466 - (1UL << order));
5467 for (i = 0; i < (1 << order); i++)
5468 SetPageReserved((page+i));
5469 pfn += (1 << order);
5470 }
5471 spin_unlock_irqrestore(&zone->lock, flags);
5472}
5473#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01005474
5475#ifdef CONFIG_MEMORY_FAILURE
5476bool is_free_buddy_page(struct page *page)
5477{
5478 struct zone *zone = page_zone(page);
5479 unsigned long pfn = page_to_pfn(page);
5480 unsigned long flags;
5481 int order;
5482
5483 spin_lock_irqsave(&zone->lock, flags);
5484 for (order = 0; order < MAX_ORDER; order++) {
5485 struct page *page_head = page - (pfn & ((1 << order) - 1));
5486
5487 if (PageBuddy(page_head) && page_order(page_head) >= order)
5488 break;
5489 }
5490 spin_unlock_irqrestore(&zone->lock, flags);
5491
5492 return order < MAX_ORDER;
5493}
5494#endif
Wu Fengguang718a3822010-03-10 15:20:43 -08005495
5496static struct trace_print_flags pageflag_names[] = {
5497 {1UL << PG_locked, "locked" },
5498 {1UL << PG_error, "error" },
5499 {1UL << PG_referenced, "referenced" },
5500 {1UL << PG_uptodate, "uptodate" },
5501 {1UL << PG_dirty, "dirty" },
5502 {1UL << PG_lru, "lru" },
5503 {1UL << PG_active, "active" },
5504 {1UL << PG_slab, "slab" },
5505 {1UL << PG_owner_priv_1, "owner_priv_1" },
5506 {1UL << PG_arch_1, "arch_1" },
5507 {1UL << PG_reserved, "reserved" },
5508 {1UL << PG_private, "private" },
5509 {1UL << PG_private_2, "private_2" },
5510 {1UL << PG_writeback, "writeback" },
5511#ifdef CONFIG_PAGEFLAGS_EXTENDED
5512 {1UL << PG_head, "head" },
5513 {1UL << PG_tail, "tail" },
5514#else
5515 {1UL << PG_compound, "compound" },
5516#endif
5517 {1UL << PG_swapcache, "swapcache" },
5518 {1UL << PG_mappedtodisk, "mappedtodisk" },
5519 {1UL << PG_reclaim, "reclaim" },
5520 {1UL << PG_buddy, "buddy" },
5521 {1UL << PG_swapbacked, "swapbacked" },
5522 {1UL << PG_unevictable, "unevictable" },
5523#ifdef CONFIG_MMU
5524 {1UL << PG_mlocked, "mlocked" },
5525#endif
5526#ifdef CONFIG_ARCH_USES_PG_UNCACHED
5527 {1UL << PG_uncached, "uncached" },
5528#endif
5529#ifdef CONFIG_MEMORY_FAILURE
5530 {1UL << PG_hwpoison, "hwpoison" },
5531#endif
5532 {-1UL, NULL },
5533};
5534
5535static void dump_page_flags(unsigned long flags)
5536{
5537 const char *delim = "";
5538 unsigned long mask;
5539 int i;
5540
5541 printk(KERN_ALERT "page flags: %#lx(", flags);
5542
5543 /* remove zone id */
5544 flags &= (1UL << NR_PAGEFLAGS) - 1;
5545
5546 for (i = 0; pageflag_names[i].name && flags; i++) {
5547
5548 mask = pageflag_names[i].mask;
5549 if ((flags & mask) != mask)
5550 continue;
5551
5552 flags &= ~mask;
5553 printk("%s%s", delim, pageflag_names[i].name);
5554 delim = "|";
5555 }
5556
5557 /* check for left over flags */
5558 if (flags)
5559 printk("%s%#lx", delim, flags);
5560
5561 printk(")\n");
5562}
5563
5564void dump_page(struct page *page)
5565{
5566 printk(KERN_ALERT
5567 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
5568 page, page_count(page), page_mapcount(page),
5569 page->mapping, page->index);
5570 dump_page_flags(page->flags);
5571}