blob: a873e61e312e6dd7795b4b734a0370bc844d9f29 [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 Rientjes5a3135c22007-10-16 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
Andrea Arcangeli59ff4212011-01-13 15:46:38 -0800360/* update __split_huge_page_refcount if you change this function */
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800361static int destroy_compound_page(struct page *page, unsigned long order)
Andy Whitcroft18229df2008-11-06 12:53:27 -0800362{
363 int i;
364 int nr_pages = 1 << order;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800365 int bad = 0;
Andy Whitcroft18229df2008-11-06 12:53:27 -0800366
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800367 if (unlikely(compound_order(page) != order) ||
368 unlikely(!PageHead(page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800369 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800370 bad++;
371 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372
Christoph Lameter6d777952007-05-06 14:49:40 -0700373 __ClearPageHead(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800374
Andy Whitcroft18229df2008-11-06 12:53:27 -0800375 for (i = 1; i < nr_pages; i++) {
376 struct page *p = page + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377
Alexey Zaytseve713a212009-01-10 02:47:57 +0300378 if (unlikely(!PageTail(p) || (p->first_page != page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800379 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800380 bad++;
381 }
Christoph Lameterd85f3382007-05-06 14:49:39 -0700382 __ClearPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800384
385 return bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387
Nick Piggin17cf4402006-03-22 00:08:41 -0800388static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
389{
390 int i;
391
Andrew Morton6626c5d2006-03-22 00:08:42 -0800392 /*
393 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
394 * and __GFP_HIGHMEM from hard or soft interrupt context.
395 */
Nick Piggin725d7042006-09-25 23:30:55 -0700396 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800397 for (i = 0; i < (1 << order); i++)
398 clear_highpage(page + i);
399}
400
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700401static inline void set_page_order(struct page *page, int order)
402{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700403 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000404 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405}
406
407static inline void rmv_page_order(struct page *page)
408{
Nick Piggin676165a2006-04-10 11:21:48 +1000409 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700410 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411}
412
413/*
414 * Locate the struct page for both the matching buddy in our
415 * pair (buddy1) and the combined O(n+1) page they form (page).
416 *
417 * 1) Any buddy B1 will have an order O twin B2 which satisfies
418 * the following equation:
419 * B2 = B1 ^ (1 << O)
420 * For example, if the starting buddy (buddy2) is #8 its order
421 * 1 buddy is #10:
422 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
423 *
424 * 2) Any buddy B will have an order O+1 parent P which
425 * satisfies the following equation:
426 * P = B & ~(1 << O)
427 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200428 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430static inline unsigned long
KyongHo Cho43506fa2011-01-13 15:47:24 -0800431__find_buddy_index(unsigned long page_idx, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432{
KyongHo Cho43506fa2011-01-13 15:47:24 -0800433 return page_idx ^ (1 << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434}
435
436/*
437 * This function checks whether a page is free && is the buddy
438 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800439 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000440 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700441 * (c) a page and its buddy have the same order &&
442 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 *
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800444 * For recording whether a page is in the buddy system, we set ->_mapcount -2.
445 * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock.
Nick Piggin676165a2006-04-10 11:21:48 +1000446 *
447 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700449static inline int page_is_buddy(struct page *page, struct page *buddy,
450 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700452 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800453 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800454
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700455 if (page_zone_id(page) != page_zone_id(buddy))
456 return 0;
457
458 if (PageBuddy(buddy) && page_order(buddy) == order) {
Nick Piggina3af9c32009-06-16 15:32:10 -0700459 VM_BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700460 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000461 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700462 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463}
464
465/*
466 * Freeing function for a buddy system allocator.
467 *
468 * The concept of a buddy system is to maintain direct-mapped table
469 * (containing bit values) for memory blocks of various "orders".
470 * The bottom level table contains the map for the smallest allocatable
471 * units of memory (here, pages), and each level above it describes
472 * pairs of units from the levels below, hence, "buddies".
473 * At a high level, all that happens here is marking the table entry
474 * at the bottom level available, and propagating the changes upward
475 * as necessary, plus some accounting needed to play nicely with other
476 * parts of the VM system.
477 * At each level, we keep a list of pages, which are heads of continuous
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800478 * free pages of length of (1 << order) and marked with _mapcount -2. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700479 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 * So when we are allocating or freeing one, we can derive the state of the
481 * other. That is, if we allocate a small block, and both were
482 * free, the remainder of the region must be split into blocks.
483 * If a block is freed, and its buddy is also free, then this
484 * triggers coalescing into a block of larger size.
485 *
486 * -- wli
487 */
488
Nick Piggin48db57f2006-01-08 01:00:42 -0800489static inline void __free_one_page(struct page *page,
Mel Gormaned0ae212009-06-16 15:32:07 -0700490 struct zone *zone, unsigned int order,
491 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492{
493 unsigned long page_idx;
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700494 unsigned long combined_idx;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800495 unsigned long uninitialized_var(buddy_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700496 struct page *buddy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497
Nick Piggin224abf92006-01-06 00:11:11 -0800498 if (unlikely(PageCompound(page)))
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800499 if (unlikely(destroy_compound_page(page, order)))
500 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501
Mel Gormaned0ae212009-06-16 15:32:07 -0700502 VM_BUG_ON(migratetype == -1);
503
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
505
Mel Gormanf2260e62009-06-16 15:32:13 -0700506 VM_BUG_ON(page_idx & ((1 << order) - 1));
Nick Piggin725d7042006-09-25 23:30:55 -0700507 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 while (order < MAX_ORDER-1) {
KyongHo Cho43506fa2011-01-13 15:47:24 -0800510 buddy_idx = __find_buddy_index(page_idx, order);
511 buddy = page + (buddy_idx - page_idx);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700512 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700513 break;
Nick Piggin13e74442006-01-06 00:10:58 -0800514
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700515 /* Our buddy is free, merge with it and move up one order. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 list_del(&buddy->lru);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700517 zone->free_area[order].nr_free--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 rmv_page_order(buddy);
KyongHo Cho43506fa2011-01-13 15:47:24 -0800519 combined_idx = buddy_idx & page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 page = page + (combined_idx - page_idx);
521 page_idx = combined_idx;
522 order++;
523 }
524 set_page_order(page, order);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700525
526 /*
527 * If this is not the largest possible page, check if the buddy
528 * of the next-highest order is free. If it is, it's possible
529 * that pages are being freed that will coalesce soon. In case,
530 * that is happening, add the free page to the tail of the list
531 * so it's less likely to be used soon and more likely to be merged
532 * as a higher order page
533 */
Mel Gormanb7f50cf2010-10-26 14:21:11 -0700534 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700535 struct page *higher_page, *higher_buddy;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800536 combined_idx = buddy_idx & page_idx;
537 higher_page = page + (combined_idx - page_idx);
538 buddy_idx = __find_buddy_index(combined_idx, order + 1);
539 higher_buddy = page + (buddy_idx - combined_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700540 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
541 list_add_tail(&page->lru,
542 &zone->free_area[order].free_list[migratetype]);
543 goto out;
544 }
545 }
546
547 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
548out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 zone->free_area[order].nr_free++;
550}
551
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700552/*
553 * free_page_mlock() -- clean up attempts to free and mlocked() page.
554 * Page should not be on lru, so no need to fix that up.
555 * free_pages_check() will verify...
556 */
557static inline void free_page_mlock(struct page *page)
558{
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700559 __dec_zone_page_state(page, NR_MLOCK);
560 __count_vm_event(UNEVICTABLE_MLOCKFREED);
561}
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700562
Nick Piggin224abf92006-01-06 00:11:11 -0800563static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564{
Nick Piggin92be2e332006-01-06 00:10:57 -0800565 if (unlikely(page_mapcount(page) |
566 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700567 (atomic_read(&page->_count) != 0) |
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800568 (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800569 bad_page(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800570 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800571 }
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800572 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
573 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
574 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575}
576
577/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700578 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700580 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 *
582 * If the zone was previously in an "all pages pinned" state then look to
583 * see if this freeing clears that state.
584 *
585 * And clear the zone's pages_scanned counter, to hold off the "all pages are
586 * pinned" detection logic.
587 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700588static void free_pcppages_bulk(struct zone *zone, int count,
589 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590{
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700591 int migratetype = 0;
Mel Gormana6f9edd2009-09-21 17:03:20 -0700592 int batch_free = 0;
Mel Gorman72853e22010-09-09 16:38:16 -0700593 int to_free = count;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700594
Nick Pigginc54ad302006-01-06 00:10:56 -0800595 spin_lock(&zone->lock);
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -0800596 zone->all_unreclaimable = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700598
Mel Gorman72853e22010-09-09 16:38:16 -0700599 while (to_free) {
Nick Piggin48db57f2006-01-08 01:00:42 -0800600 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700601 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -0800602
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700603 /*
Mel Gormana6f9edd2009-09-21 17:03:20 -0700604 * Remove pages from lists in a round-robin fashion. A
605 * batch_free count is maintained that is incremented when an
606 * empty list is encountered. This is so more pages are freed
607 * off fuller lists instead of spinning excessively around empty
608 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700609 */
610 do {
Mel Gormana6f9edd2009-09-21 17:03:20 -0700611 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700612 if (++migratetype == MIGRATE_PCPTYPES)
613 migratetype = 0;
614 list = &pcp->lists[migratetype];
615 } while (list_empty(list));
616
Mel Gormana6f9edd2009-09-21 17:03:20 -0700617 do {
618 page = list_entry(list->prev, struct page, lru);
619 /* must delete as __free_one_page list manipulates */
620 list_del(&page->lru);
Hugh Dickinsa7016232010-01-29 17:46:34 +0000621 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
622 __free_one_page(page, zone, 0, page_private(page));
623 trace_mm_page_pcpu_drain(page, 0, page_private(page));
Mel Gorman72853e22010-09-09 16:38:16 -0700624 } while (--to_free && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 }
Mel Gorman72853e22010-09-09 16:38:16 -0700626 __mod_zone_page_state(zone, NR_FREE_PAGES, count);
Nick Pigginc54ad302006-01-06 00:10:56 -0800627 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628}
629
Mel Gormaned0ae212009-06-16 15:32:07 -0700630static void free_one_page(struct zone *zone, struct page *page, int order,
631 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800632{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700633 spin_lock(&zone->lock);
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -0800634 zone->all_unreclaimable = 0;
Christoph Lameter006d22d2006-09-25 23:31:48 -0700635 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700636
Mel Gormaned0ae212009-06-16 15:32:07 -0700637 __free_one_page(page, zone, order, migratetype);
Mel Gorman72853e22010-09-09 16:38:16 -0700638 __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700639 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800640}
641
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700642static bool free_pages_prepare(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 int i;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800645 int bad = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646
Li Hongf6503162010-03-05 13:41:52 -0800647 trace_mm_page_free_direct(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100648 kmemcheck_free_shadow(page, order);
649
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -0800650 if (PageAnon(page))
651 page->mapping = NULL;
652 for (i = 0; i < (1 << order); i++)
653 bad += free_pages_check(page + i);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800654 if (bad)
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700655 return false;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800656
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700657 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -0700658 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700659 debug_check_no_obj_freed(page_address(page),
660 PAGE_SIZE << order);
661 }
Nick Piggindafb1362006-10-11 01:21:30 -0700662 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800663 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700664
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700665 return true;
666}
667
668static void __free_pages_ok(struct page *page, unsigned int order)
669{
670 unsigned long flags;
671 int wasMlocked = __TestClearPageMlocked(page);
672
673 if (!free_pages_prepare(page, order))
674 return;
675
Nick Pigginc54ad302006-01-06 00:10:56 -0800676 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +0200677 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -0700678 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700679 __count_vm_events(PGFREE, 1 << order);
Mel Gormaned0ae212009-06-16 15:32:07 -0700680 free_one_page(page_zone(page), page, order,
681 get_pageblock_migratetype(page));
Nick Pigginc54ad302006-01-06 00:10:56 -0800682 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683}
684
David Howellsa226f6c2006-01-06 00:11:08 -0800685/*
686 * permit the bootmem allocator to evade page validation on high-order frees
687 */
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700688void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800689{
690 if (order == 0) {
691 __ClearPageReserved(page);
692 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800693 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800694 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800695 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800696 int loop;
697
Nick Piggin545b1ea2006-03-22 00:08:07 -0800698 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800699 for (loop = 0; loop < BITS_PER_LONG; loop++) {
700 struct page *p = &page[loop];
701
Nick Piggin545b1ea2006-03-22 00:08:07 -0800702 if (loop + 1 < BITS_PER_LONG)
703 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800704 __ClearPageReserved(p);
705 set_page_count(p, 0);
706 }
707
Nick Piggin7835e982006-03-22 00:08:40 -0800708 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800709 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800710 }
711}
712
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713
714/*
715 * The order of subdivision here is critical for the IO subsystem.
716 * Please do not alter this order without good reasons and regression
717 * testing. Specifically, as large blocks of memory are subdivided,
718 * the order in which smaller blocks are delivered depends on the order
719 * they're subdivided in this function. This is the primary factor
720 * influencing the order in which pages are delivered to the IO
721 * subsystem according to empirical testing, and this is also justified
722 * by considering the behavior of a buddy system containing a single
723 * large block of memory acted on by a series of small allocations.
724 * This behavior is a critical factor in sglist merging's success.
725 *
726 * -- wli
727 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800728static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700729 int low, int high, struct free_area *area,
730 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731{
732 unsigned long size = 1 << high;
733
734 while (high > low) {
735 area--;
736 high--;
737 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700738 VM_BUG_ON(bad_range(zone, &page[size]));
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700739 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 area->nr_free++;
741 set_page_order(&page[size], high);
742 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743}
744
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745/*
746 * This page is about to be returned from the page allocator
747 */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200748static inline int check_new_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749{
Nick Piggin92be2e332006-01-06 00:10:57 -0800750 if (unlikely(page_mapcount(page) |
751 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700752 (atomic_read(&page->_count) != 0) |
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800753 (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800754 bad_page(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800755 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800756 }
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200757 return 0;
758}
759
760static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
761{
762 int i;
763
764 for (i = 0; i < (1 << order); i++) {
765 struct page *p = page + i;
766 if (unlikely(check_new_page(p)))
767 return 1;
768 }
Hugh Dickins689bceb2005-11-21 21:32:20 -0800769
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700770 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800771 set_page_refcounted(page);
Nick Piggincc1025092006-12-06 20:32:00 -0800772
773 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800775
776 if (gfp_flags & __GFP_ZERO)
777 prep_zero_page(page, order, gfp_flags);
778
779 if (order && (gfp_flags & __GFP_COMP))
780 prep_compound_page(page, order);
781
Hugh Dickins689bceb2005-11-21 21:32:20 -0800782 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783}
784
Mel Gorman56fd56b2007-10-16 01:25:58 -0700785/*
786 * Go through the free lists for the given migratetype and remove
787 * the smallest available page from the freelists
788 */
Mel Gorman728ec982009-06-16 15:32:04 -0700789static inline
790struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -0700791 int migratetype)
792{
793 unsigned int current_order;
794 struct free_area * area;
795 struct page *page;
796
797 /* Find a page of the appropriate size in the preferred list */
798 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
799 area = &(zone->free_area[current_order]);
800 if (list_empty(&area->free_list[migratetype]))
801 continue;
802
803 page = list_entry(area->free_list[migratetype].next,
804 struct page, lru);
805 list_del(&page->lru);
806 rmv_page_order(page);
807 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -0700808 expand(zone, page, order, current_order, area, migratetype);
809 return page;
810 }
811
812 return NULL;
813}
814
815
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700816/*
817 * This array describes the order lists are fallen back to when
818 * the free lists for the desirable migrate type are depleted
819 */
820static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
Mel Gorman64c5e132007-10-16 01:25:59 -0700821 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
822 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
823 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
824 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700825};
826
Mel Gormanc361be52007-10-16 01:25:51 -0700827/*
828 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -0700829 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -0700830 * boundary. If alignment is required, use move_freepages_block()
831 */
Adrian Bunkb69a7282008-07-23 21:28:12 -0700832static int move_freepages(struct zone *zone,
833 struct page *start_page, struct page *end_page,
834 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700835{
836 struct page *page;
837 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -0700838 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -0700839
840#ifndef CONFIG_HOLES_IN_ZONE
841 /*
842 * page_zone is not safe to call in this context when
843 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
844 * anyway as we check zone boundaries in move_freepages_block().
845 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -0700846 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -0700847 */
848 BUG_ON(page_zone(start_page) != page_zone(end_page));
849#endif
850
851 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -0700852 /* Make sure we are not inadvertently changing nodes */
853 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
854
Mel Gormanc361be52007-10-16 01:25:51 -0700855 if (!pfn_valid_within(page_to_pfn(page))) {
856 page++;
857 continue;
858 }
859
860 if (!PageBuddy(page)) {
861 page++;
862 continue;
863 }
864
865 order = page_order(page);
866 list_del(&page->lru);
867 list_add(&page->lru,
868 &zone->free_area[order].free_list[migratetype]);
869 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -0700870 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -0700871 }
872
Mel Gormand1003132007-10-16 01:26:00 -0700873 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -0700874}
875
Adrian Bunkb69a7282008-07-23 21:28:12 -0700876static int move_freepages_block(struct zone *zone, struct page *page,
877 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700878{
879 unsigned long start_pfn, end_pfn;
880 struct page *start_page, *end_page;
881
882 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -0700883 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -0700884 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -0700885 end_page = start_page + pageblock_nr_pages - 1;
886 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -0700887
888 /* Do not cross zone boundaries */
889 if (start_pfn < zone->zone_start_pfn)
890 start_page = page;
891 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
892 return 0;
893
894 return move_freepages(zone, start_page, end_page, migratetype);
895}
896
Mel Gorman2f66a682009-09-21 17:02:31 -0700897static void change_pageblock_range(struct page *pageblock_page,
898 int start_order, int migratetype)
899{
900 int nr_pageblocks = 1 << (start_order - pageblock_order);
901
902 while (nr_pageblocks--) {
903 set_pageblock_migratetype(pageblock_page, migratetype);
904 pageblock_page += pageblock_nr_pages;
905 }
906}
907
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700908/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -0700909static inline struct page *
910__rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700911{
912 struct free_area * area;
913 int current_order;
914 struct page *page;
915 int migratetype, i;
916
917 /* Find the largest possible block of pages in the other list */
918 for (current_order = MAX_ORDER-1; current_order >= order;
919 --current_order) {
920 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
921 migratetype = fallbacks[start_migratetype][i];
922
Mel Gorman56fd56b2007-10-16 01:25:58 -0700923 /* MIGRATE_RESERVE handled later if necessary */
924 if (migratetype == MIGRATE_RESERVE)
925 continue;
Mel Gormane0104872007-10-16 01:25:53 -0700926
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700927 area = &(zone->free_area[current_order]);
928 if (list_empty(&area->free_list[migratetype]))
929 continue;
930
931 page = list_entry(area->free_list[migratetype].next,
932 struct page, lru);
933 area->nr_free--;
934
935 /*
Mel Gormanc361be52007-10-16 01:25:51 -0700936 * If breaking a large block of pages, move all free
Mel Gorman46dafbc2007-10-16 01:25:55 -0700937 * pages to the preferred allocation list. If falling
938 * back for a reclaimable kernel allocation, be more
939 * agressive about taking ownership of free pages
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700940 */
Mel Gormand9c23402007-10-16 01:26:01 -0700941 if (unlikely(current_order >= (pageblock_order >> 1)) ||
Mel Gormandd5d2412009-09-05 11:17:11 -0700942 start_migratetype == MIGRATE_RECLAIMABLE ||
943 page_group_by_mobility_disabled) {
Mel Gorman46dafbc2007-10-16 01:25:55 -0700944 unsigned long pages;
945 pages = move_freepages_block(zone, page,
946 start_migratetype);
947
948 /* Claim the whole block if over half of it is free */
Mel Gormandd5d2412009-09-05 11:17:11 -0700949 if (pages >= (1 << (pageblock_order-1)) ||
950 page_group_by_mobility_disabled)
Mel Gorman46dafbc2007-10-16 01:25:55 -0700951 set_pageblock_migratetype(page,
952 start_migratetype);
953
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700954 migratetype = start_migratetype;
Mel Gormanc361be52007-10-16 01:25:51 -0700955 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700956
957 /* Remove the page from the freelists */
958 list_del(&page->lru);
959 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700960
Mel Gorman2f66a682009-09-21 17:02:31 -0700961 /* Take ownership for orders >= pageblock_order */
962 if (current_order >= pageblock_order)
963 change_pageblock_range(page, current_order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700964 start_migratetype);
965
966 expand(zone, page, order, current_order, area, migratetype);
Mel Gormane0fff1b2009-09-21 17:02:42 -0700967
968 trace_mm_page_alloc_extfrag(page, order, current_order,
969 start_migratetype, migratetype);
970
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700971 return page;
972 }
973 }
974
Mel Gorman728ec982009-06-16 15:32:04 -0700975 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700976}
977
Mel Gorman56fd56b2007-10-16 01:25:58 -0700978/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 * Do the hard work of removing an element from the buddy allocator.
980 * Call me with the zone->lock already held.
981 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700982static struct page *__rmqueue(struct zone *zone, unsigned int order,
983 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 struct page *page;
986
Mel Gorman728ec982009-06-16 15:32:04 -0700987retry_reserve:
Mel Gorman56fd56b2007-10-16 01:25:58 -0700988 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989
Mel Gorman728ec982009-06-16 15:32:04 -0700990 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Mel Gorman56fd56b2007-10-16 01:25:58 -0700991 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700992
Mel Gorman728ec982009-06-16 15:32:04 -0700993 /*
994 * Use MIGRATE_RESERVE rather than fail an allocation. goto
995 * is used because __rmqueue_smallest is an inline function
996 * and we want just one call site
997 */
998 if (!page) {
999 migratetype = MIGRATE_RESERVE;
1000 goto retry_reserve;
1001 }
1002 }
1003
Mel Gorman0d3d0622009-09-21 17:02:44 -07001004 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001005 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006}
1007
1008/*
1009 * Obtain a specified number of elements from the buddy allocator, all under
1010 * a single hold of the lock, for efficiency. Add them to the supplied list.
1011 * Returns the number of new pages which were placed at *list.
1012 */
1013static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001014 unsigned long count, struct list_head *list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001015 int migratetype, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018
Nick Pigginc54ad302006-01-06 00:10:56 -08001019 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001021 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001022 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08001024
1025 /*
1026 * Split buddy pages returned by expand() are received here
1027 * in physical page order. The page is added to the callers and
1028 * list and the list head then moves forward. From the callers
1029 * perspective, the linked list is ordered by page number in
1030 * some conditions. This is useful for IO devices that can
1031 * merge IO requests if the physical pages are ordered
1032 * properly.
1033 */
Mel Gormane084b2d2009-07-29 15:02:04 -07001034 if (likely(cold == 0))
1035 list_add(&page->lru, list);
1036 else
1037 list_add_tail(&page->lru, list);
Mel Gorman535131e62007-10-16 01:25:49 -07001038 set_page_private(page, migratetype);
Mel Gorman81eabcb2007-12-17 16:20:05 -08001039 list = &page->lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 }
Mel Gormanf2260e62009-06-16 15:32:13 -07001041 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08001042 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001043 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044}
1045
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001046#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001047/*
Christoph Lameter4037d452007-05-09 02:35:14 -07001048 * Called from the vmstat counter updater to drain pagesets of this
1049 * currently executing processor on remote nodes after they have
1050 * expired.
1051 *
Christoph Lameter879336c2006-03-22 00:09:08 -08001052 * Note that this function must be called with the thread pinned to
1053 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001054 */
Christoph Lameter4037d452007-05-09 02:35:14 -07001055void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001056{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001057 unsigned long flags;
Christoph Lameter4037d452007-05-09 02:35:14 -07001058 int to_drain;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001059
Christoph Lameter4037d452007-05-09 02:35:14 -07001060 local_irq_save(flags);
1061 if (pcp->count >= pcp->batch)
1062 to_drain = pcp->batch;
1063 else
1064 to_drain = pcp->count;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001065 free_pcppages_bulk(zone, to_drain, pcp);
Christoph Lameter4037d452007-05-09 02:35:14 -07001066 pcp->count -= to_drain;
1067 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001068}
1069#endif
1070
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001071/*
1072 * Drain pages of the indicated processor.
1073 *
1074 * The processor must either be the current processor and the
1075 * thread pinned to the current processor or a processor that
1076 * is not online.
1077 */
1078static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079{
Nick Pigginc54ad302006-01-06 00:10:56 -08001080 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07001083 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 struct per_cpu_pageset *pset;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001085 struct per_cpu_pages *pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001087 local_irq_save(flags);
1088 pset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001090 pcp = &pset->pcp;
David Rientjes2ff754f2011-01-25 15:07:23 -08001091 if (pcp->count) {
1092 free_pcppages_bulk(zone, pcp->count, pcp);
1093 pcp->count = 0;
1094 }
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001095 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 }
1097}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001099/*
1100 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1101 */
1102void drain_local_pages(void *arg)
1103{
1104 drain_pages(smp_processor_id());
1105}
1106
1107/*
1108 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
1109 */
1110void drain_all_pages(void)
1111{
Jens Axboe15c8b6c2008-05-09 09:39:44 +02001112 on_each_cpu(drain_local_pages, NULL, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001113}
1114
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02001115#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116
1117void mark_free_pages(struct zone *zone)
1118{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001119 unsigned long pfn, max_zone_pfn;
1120 unsigned long flags;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001121 int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 struct list_head *curr;
1123
1124 if (!zone->spanned_pages)
1125 return;
1126
1127 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001128
1129 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1130 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1131 if (pfn_valid(pfn)) {
1132 struct page *page = pfn_to_page(pfn);
1133
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001134 if (!swsusp_page_is_forbidden(page))
1135 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001136 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001138 for_each_migratetype_order(order, t) {
1139 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001140 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001142 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1143 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001144 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001145 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001146 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 spin_unlock_irqrestore(&zone->lock, flags);
1148}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001149#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150
1151/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 * Free a 0-order page
Li Hongfc916682010-03-05 13:41:54 -08001153 * cold == 1 ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 */
Li Hongfc916682010-03-05 13:41:54 -08001155void free_hot_cold_page(struct page *page, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156{
1157 struct zone *zone = page_zone(page);
1158 struct per_cpu_pages *pcp;
1159 unsigned long flags;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001160 int migratetype;
Johannes Weiner451ea252009-09-21 17:01:48 -07001161 int wasMlocked = __TestClearPageMlocked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001163 if (!free_pages_prepare(page, 0))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001164 return;
1165
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001166 migratetype = get_pageblock_migratetype(page);
1167 set_page_private(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +02001169 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -07001170 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001171 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001172
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001173 /*
1174 * We only track unmovable, reclaimable and movable on pcp lists.
1175 * Free ISOLATE pages back to the allocator because they are being
1176 * offlined but treat RESERVE as movable pages so we can get those
1177 * areas back if necessary. Otherwise, we may have to free
1178 * excessively into the page allocator
1179 */
1180 if (migratetype >= MIGRATE_PCPTYPES) {
1181 if (unlikely(migratetype == MIGRATE_ISOLATE)) {
1182 free_one_page(zone, page, 0, migratetype);
1183 goto out;
1184 }
1185 migratetype = MIGRATE_MOVABLE;
1186 }
1187
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001188 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001189 if (cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001190 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001191 else
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001192 list_add(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001194 if (pcp->count >= pcp->high) {
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001195 free_pcppages_bulk(zone, pcp->batch, pcp);
Nick Piggin48db57f2006-01-08 01:00:42 -08001196 pcp->count -= pcp->batch;
1197 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001198
1199out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201}
1202
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001203/*
1204 * split_page takes a non-compound higher-order page, and splits it into
1205 * n (1<<order) sub-pages: page[0..n]
1206 * Each sub-page must be freed individually.
1207 *
1208 * Note: this is probably too low level an operation for use in drivers.
1209 * Please consult with lkml before using this in your driver.
1210 */
1211void split_page(struct page *page, unsigned int order)
1212{
1213 int i;
1214
Nick Piggin725d7042006-09-25 23:30:55 -07001215 VM_BUG_ON(PageCompound(page));
1216 VM_BUG_ON(!page_count(page));
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001217
1218#ifdef CONFIG_KMEMCHECK
1219 /*
1220 * Split shadow pages too, because free(page[0]) would
1221 * otherwise free the whole shadow.
1222 */
1223 if (kmemcheck_page_is_tracked(page))
1224 split_page(virt_to_page(page[0].shadow), order);
1225#endif
1226
Nick Piggin7835e982006-03-22 00:08:40 -08001227 for (i = 1; i < (1 << order); i++)
1228 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001229}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001230
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231/*
Mel Gorman748446b2010-05-24 14:32:27 -07001232 * Similar to split_page except the page is already free. As this is only
1233 * being used for migration, the migratetype of the block also changes.
1234 * As this is called with interrupts disabled, the caller is responsible
1235 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1236 * are enabled.
1237 *
1238 * Note: this is probably too low level an operation for use in drivers.
1239 * Please consult with lkml before using this in your driver.
1240 */
1241int split_free_page(struct page *page)
1242{
1243 unsigned int order;
1244 unsigned long watermark;
1245 struct zone *zone;
1246
1247 BUG_ON(!PageBuddy(page));
1248
1249 zone = page_zone(page);
1250 order = page_order(page);
1251
1252 /* Obey watermarks as if the page was being allocated */
1253 watermark = low_wmark_pages(zone) + (1 << order);
1254 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1255 return 0;
1256
1257 /* Remove page from free list */
1258 list_del(&page->lru);
1259 zone->free_area[order].nr_free--;
1260 rmv_page_order(page);
1261 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1UL << order));
1262
1263 /* Split into individual pages */
1264 set_page_refcounted(page);
1265 split_page(page, order);
1266
1267 if (order >= pageblock_order - 1) {
1268 struct page *endpage = page + (1 << order) - 1;
1269 for (; page < endpage; page += pageblock_nr_pages)
1270 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1271 }
1272
1273 return 1 << order;
1274}
1275
1276/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1278 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1279 * or two.
1280 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07001281static inline
1282struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001283 struct zone *zone, int order, gfp_t gfp_flags,
1284 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285{
1286 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001287 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 int cold = !!(gfp_flags & __GFP_COLD);
1289
Hugh Dickins689bceb2005-11-21 21:32:20 -08001290again:
Nick Piggin48db57f2006-01-08 01:00:42 -08001291 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001293 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 local_irq_save(flags);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001296 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1297 list = &pcp->lists[migratetype];
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001298 if (list_empty(list)) {
Mel Gorman535131e62007-10-16 01:25:49 -07001299 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001300 pcp->batch, list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001301 migratetype, cold);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001302 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07001303 goto failed;
Mel Gorman535131e62007-10-16 01:25:49 -07001304 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001305
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001306 if (cold)
1307 page = list_entry(list->prev, struct page, lru);
1308 else
1309 page = list_entry(list->next, struct page, lru);
1310
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001311 list_del(&page->lru);
1312 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001313 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07001314 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1315 /*
1316 * __GFP_NOFAIL is not to be used in new code.
1317 *
1318 * All __GFP_NOFAIL callers should be fixed so that they
1319 * properly detect and handle allocation failures.
1320 *
1321 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07001322 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07001323 * __GFP_NOFAIL.
1324 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07001325 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07001326 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001328 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001329 spin_unlock(&zone->lock);
1330 if (!page)
1331 goto failed;
KOSAKI Motohiro6ccf80e2010-01-15 17:01:18 -08001332 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 }
1334
Christoph Lameterf8891e52006-06-30 01:55:45 -07001335 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Mel Gorman18ea7e72008-04-28 02:12:14 -07001336 zone_statistics(preferred_zone, zone);
Nick Piggina74609f2006-01-06 00:11:20 -08001337 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338
Nick Piggin725d7042006-09-25 23:30:55 -07001339 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -08001340 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001341 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001343
1344failed:
1345 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001346 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347}
1348
Mel Gorman41858962009-06-16 15:32:12 -07001349/* The ALLOC_WMARK bits are used as an index to zone->watermark */
1350#define ALLOC_WMARK_MIN WMARK_MIN
1351#define ALLOC_WMARK_LOW WMARK_LOW
1352#define ALLOC_WMARK_HIGH WMARK_HIGH
1353#define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1354
1355/* Mask to get the watermark bits */
1356#define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1357
Nick Piggin31488902005-11-28 13:44:03 -08001358#define ALLOC_HARDER 0x10 /* try to alloc harder */
1359#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1360#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001361
Akinobu Mita933e3122006-12-08 02:39:45 -08001362#ifdef CONFIG_FAIL_PAGE_ALLOC
1363
1364static struct fail_page_alloc_attr {
1365 struct fault_attr attr;
1366
1367 u32 ignore_gfp_highmem;
1368 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001369 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001370
1371#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1372
1373 struct dentry *ignore_gfp_highmem_file;
1374 struct dentry *ignore_gfp_wait_file;
Akinobu Mita54114992007-07-15 23:40:23 -07001375 struct dentry *min_order_file;
Akinobu Mita933e3122006-12-08 02:39:45 -08001376
1377#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1378
1379} fail_page_alloc = {
1380 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001381 .ignore_gfp_wait = 1,
1382 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001383 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001384};
1385
1386static int __init setup_fail_page_alloc(char *str)
1387{
1388 return setup_fault_attr(&fail_page_alloc.attr, str);
1389}
1390__setup("fail_page_alloc=", setup_fail_page_alloc);
1391
1392static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1393{
Akinobu Mita54114992007-07-15 23:40:23 -07001394 if (order < fail_page_alloc.min_order)
1395 return 0;
Akinobu Mita933e3122006-12-08 02:39:45 -08001396 if (gfp_mask & __GFP_NOFAIL)
1397 return 0;
1398 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1399 return 0;
1400 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1401 return 0;
1402
1403 return should_fail(&fail_page_alloc.attr, 1 << order);
1404}
1405
1406#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1407
1408static int __init fail_page_alloc_debugfs(void)
1409{
1410 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1411 struct dentry *dir;
1412 int err;
1413
1414 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1415 "fail_page_alloc");
1416 if (err)
1417 return err;
1418 dir = fail_page_alloc.attr.dentries.dir;
1419
1420 fail_page_alloc.ignore_gfp_wait_file =
1421 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1422 &fail_page_alloc.ignore_gfp_wait);
1423
1424 fail_page_alloc.ignore_gfp_highmem_file =
1425 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1426 &fail_page_alloc.ignore_gfp_highmem);
Akinobu Mita54114992007-07-15 23:40:23 -07001427 fail_page_alloc.min_order_file =
1428 debugfs_create_u32("min-order", mode, dir,
1429 &fail_page_alloc.min_order);
Akinobu Mita933e3122006-12-08 02:39:45 -08001430
1431 if (!fail_page_alloc.ignore_gfp_wait_file ||
Akinobu Mita54114992007-07-15 23:40:23 -07001432 !fail_page_alloc.ignore_gfp_highmem_file ||
1433 !fail_page_alloc.min_order_file) {
Akinobu Mita933e3122006-12-08 02:39:45 -08001434 err = -ENOMEM;
1435 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1436 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
Akinobu Mita54114992007-07-15 23:40:23 -07001437 debugfs_remove(fail_page_alloc.min_order_file);
Akinobu Mita933e3122006-12-08 02:39:45 -08001438 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1439 }
1440
1441 return err;
1442}
1443
1444late_initcall(fail_page_alloc_debugfs);
1445
1446#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1447
1448#else /* CONFIG_FAIL_PAGE_ALLOC */
1449
1450static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1451{
1452 return 0;
1453}
1454
1455#endif /* CONFIG_FAIL_PAGE_ALLOC */
1456
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457/*
Mel Gorman88f5acf2011-01-13 15:45:41 -08001458 * Return true if free pages are above 'mark'. This takes into account the order
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 * of the allocation.
1460 */
Mel Gorman88f5acf2011-01-13 15:45:41 -08001461static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1462 int classzone_idx, int alloc_flags, long free_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463{
1464 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001465 long min = mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 int o;
1467
Mel Gorman88f5acf2011-01-13 15:45:41 -08001468 free_pages -= (1 << order) + 1;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001469 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001471 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 min -= min / 4;
1473
1474 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
Mel Gorman88f5acf2011-01-13 15:45:41 -08001475 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 for (o = 0; o < order; o++) {
1477 /* At the next order, this order's pages become unavailable */
1478 free_pages -= z->free_area[o].nr_free << o;
1479
1480 /* Require fewer higher order pages to be free */
1481 min >>= 1;
1482
1483 if (free_pages <= min)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001484 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 }
Mel Gorman88f5acf2011-01-13 15:45:41 -08001486 return true;
1487}
1488
1489bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1490 int classzone_idx, int alloc_flags)
1491{
1492 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1493 zone_page_state(z, NR_FREE_PAGES));
1494}
1495
1496bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
1497 int classzone_idx, int alloc_flags)
1498{
1499 long free_pages = zone_page_state(z, NR_FREE_PAGES);
1500
1501 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1502 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1503
1504 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1505 free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506}
1507
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001508#ifdef CONFIG_NUMA
1509/*
1510 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1511 * skip over zones that are not allowed by the cpuset, or that have
1512 * been recently (in last second) found to be nearly full. See further
1513 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001514 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001515 *
1516 * If the zonelist cache is present in the passed in zonelist, then
1517 * returns a pointer to the allowed node mask (either the current
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001518 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001519 *
1520 * If the zonelist cache is not available for this zonelist, does
1521 * nothing and returns NULL.
1522 *
1523 * If the fullzones BITMAP in the zonelist cache is stale (more than
1524 * a second since last zap'd) then we zap it out (clear its bits.)
1525 *
1526 * We hold off even calling zlc_setup, until after we've checked the
1527 * first zone in the zonelist, on the theory that most allocations will
1528 * be satisfied from that first zone, so best to examine that zone as
1529 * quickly as we can.
1530 */
1531static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1532{
1533 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1534 nodemask_t *allowednodes; /* zonelist_cache approximation */
1535
1536 zlc = zonelist->zlcache_ptr;
1537 if (!zlc)
1538 return NULL;
1539
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001540 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001541 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1542 zlc->last_full_zap = jiffies;
1543 }
1544
1545 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1546 &cpuset_current_mems_allowed :
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001547 &node_states[N_HIGH_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001548 return allowednodes;
1549}
1550
1551/*
1552 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1553 * if it is worth looking at further for free memory:
1554 * 1) Check that the zone isn't thought to be full (doesn't have its
1555 * bit set in the zonelist_cache fullzones BITMAP).
1556 * 2) Check that the zones node (obtained from the zonelist_cache
1557 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1558 * Return true (non-zero) if zone is worth looking at further, or
1559 * else return false (zero) if it is not.
1560 *
1561 * This check -ignores- the distinction between various watermarks,
1562 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1563 * found to be full for any variation of these watermarks, it will
1564 * be considered full for up to one second by all requests, unless
1565 * we are so low on memory on all allowed nodes that we are forced
1566 * into the second scan of the zonelist.
1567 *
1568 * In the second scan we ignore this zonelist cache and exactly
1569 * apply the watermarks to all zones, even it is slower to do so.
1570 * We are low on memory in the second scan, and should leave no stone
1571 * unturned looking for a free page.
1572 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001573static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001574 nodemask_t *allowednodes)
1575{
1576 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1577 int i; /* index of *z in zonelist zones */
1578 int n; /* node that zone *z is on */
1579
1580 zlc = zonelist->zlcache_ptr;
1581 if (!zlc)
1582 return 1;
1583
Mel Gormandd1a2392008-04-28 02:12:17 -07001584 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001585 n = zlc->z_to_n[i];
1586
1587 /* This zone is worth trying if it is allowed but not full */
1588 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1589}
1590
1591/*
1592 * Given 'z' scanning a zonelist, set the corresponding bit in
1593 * zlc->fullzones, so that subsequent attempts to allocate a page
1594 * from that zone don't waste time re-examining it.
1595 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001596static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001597{
1598 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1599 int i; /* index of *z in zonelist zones */
1600
1601 zlc = zonelist->zlcache_ptr;
1602 if (!zlc)
1603 return;
1604
Mel Gormandd1a2392008-04-28 02:12:17 -07001605 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001606
1607 set_bit(i, zlc->fullzones);
1608}
1609
1610#else /* CONFIG_NUMA */
1611
1612static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1613{
1614 return NULL;
1615}
1616
Mel Gormandd1a2392008-04-28 02:12:17 -07001617static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001618 nodemask_t *allowednodes)
1619{
1620 return 1;
1621}
1622
Mel Gormandd1a2392008-04-28 02:12:17 -07001623static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001624{
1625}
1626#endif /* CONFIG_NUMA */
1627
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001628/*
Paul Jackson0798e512006-12-06 20:31:38 -08001629 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001630 * a page.
1631 */
1632static struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001633get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
Mel Gorman5117f452009-06-16 15:31:59 -07001634 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
Mel Gorman3dd28262009-06-16 15:32:00 -07001635 struct zone *preferred_zone, int migratetype)
Martin Hicks753ee722005-06-21 17:14:41 -07001636{
Mel Gormandd1a2392008-04-28 02:12:17 -07001637 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001638 struct page *page = NULL;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001639 int classzone_idx;
Mel Gorman5117f452009-06-16 15:31:59 -07001640 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001641 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1642 int zlc_active = 0; /* set if using zonelist_cache */
1643 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gorman54a6eb52008-04-28 02:12:16 -07001644
Mel Gorman19770b32008-04-28 02:12:18 -07001645 classzone_idx = zone_idx(preferred_zone);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001646zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001647 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001648 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001649 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1650 */
Mel Gorman19770b32008-04-28 02:12:18 -07001651 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1652 high_zoneidx, nodemask) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001653 if (NUMA_BUILD && zlc_active &&
1654 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1655 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001656 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001657 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001658 goto try_next_zone;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001659
Mel Gorman41858962009-06-16 15:32:12 -07001660 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001661 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001662 unsigned long mark;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001663 int ret;
1664
Mel Gorman41858962009-06-16 15:32:12 -07001665 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
Mel Gormanfa5e0842009-06-16 15:33:22 -07001666 if (zone_watermark_ok(zone, order, mark,
1667 classzone_idx, alloc_flags))
1668 goto try_this_zone;
1669
1670 if (zone_reclaim_mode == 0)
1671 goto this_zone_full;
1672
1673 ret = zone_reclaim(zone, gfp_mask, order);
1674 switch (ret) {
1675 case ZONE_RECLAIM_NOSCAN:
1676 /* did not scan */
1677 goto try_next_zone;
1678 case ZONE_RECLAIM_FULL:
1679 /* scanned but unreclaimable */
1680 goto this_zone_full;
1681 default:
1682 /* did we reclaim enough */
1683 if (!zone_watermark_ok(zone, order, mark,
1684 classzone_idx, alloc_flags))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001685 goto this_zone_full;
Paul Jackson0798e512006-12-06 20:31:38 -08001686 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001687 }
1688
Mel Gormanfa5e0842009-06-16 15:33:22 -07001689try_this_zone:
Mel Gorman3dd28262009-06-16 15:32:00 -07001690 page = buffered_rmqueue(preferred_zone, zone, order,
1691 gfp_mask, migratetype);
Paul Jackson0798e512006-12-06 20:31:38 -08001692 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001693 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001694this_zone_full:
1695 if (NUMA_BUILD)
1696 zlc_mark_zone_full(zonelist, z);
1697try_next_zone:
Christoph Lameter62bc62a2009-06-16 15:32:15 -07001698 if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) {
Mel Gormand395b732009-06-16 15:32:09 -07001699 /*
1700 * we do zlc_setup after the first zone is tried but only
1701 * if there are multiple nodes make it worthwhile
1702 */
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001703 allowednodes = zlc_setup(zonelist, alloc_flags);
1704 zlc_active = 1;
1705 did_zlc_setup = 1;
1706 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07001707 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001708
1709 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1710 /* Disable zlc cache for second zonelist scan */
1711 zlc_active = 0;
1712 goto zonelist_scan;
1713 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001714 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07001715}
1716
Mel Gorman11e33f62009-06-16 15:31:57 -07001717static inline int
1718should_alloc_retry(gfp_t gfp_mask, unsigned int order,
1719 unsigned long pages_reclaimed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720{
Mel Gorman11e33f62009-06-16 15:31:57 -07001721 /* Do not loop if specifically requested */
1722 if (gfp_mask & __GFP_NORETRY)
1723 return 0;
1724
1725 /*
1726 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1727 * means __GFP_NOFAIL, but that may not be true in other
1728 * implementations.
1729 */
1730 if (order <= PAGE_ALLOC_COSTLY_ORDER)
1731 return 1;
1732
1733 /*
1734 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1735 * specified, then we retry until we no longer reclaim any pages
1736 * (above), or we've reclaimed an order of pages at least as
1737 * large as the allocation's order. In both cases, if the
1738 * allocation still fails, we stop retrying.
1739 */
1740 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
1741 return 1;
1742
1743 /*
1744 * Don't let big-order allocations loop unless the caller
1745 * explicitly requests that.
1746 */
1747 if (gfp_mask & __GFP_NOFAIL)
1748 return 1;
1749
1750 return 0;
1751}
1752
1753static inline struct page *
1754__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1755 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001756 nodemask_t *nodemask, struct zone *preferred_zone,
1757 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001758{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760
Mel Gorman11e33f62009-06-16 15:31:57 -07001761 /* Acquire the OOM killer lock for the zones in zonelist */
Minchan Kimff321fe2010-08-09 17:18:57 -07001762 if (!try_set_zonelist_oom(zonelist, gfp_mask)) {
Mel Gorman11e33f62009-06-16 15:31:57 -07001763 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 return NULL;
1765 }
Jens Axboe6b1de912005-11-17 21:35:02 +01001766
Mel Gorman11e33f62009-06-16 15:31:57 -07001767 /*
1768 * Go through the zonelist yet one more time, keep very high watermark
1769 * here, this is only to catch a parallel oom killing, we must fail if
1770 * we're still under heavy pressure.
1771 */
1772 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1773 order, zonelist, high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001774 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07001775 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001776 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07001777 goto out;
1778
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001779 if (!(gfp_mask & __GFP_NOFAIL)) {
1780 /* The OOM killer will not help higher order allocs */
1781 if (order > PAGE_ALLOC_COSTLY_ORDER)
1782 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07001783 /* The OOM killer does not needlessly kill tasks for lowmem */
1784 if (high_zoneidx < ZONE_NORMAL)
1785 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001786 /*
1787 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
1788 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
1789 * The caller should handle page allocation failure by itself if
1790 * it specifies __GFP_THISNODE.
1791 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
1792 */
1793 if (gfp_mask & __GFP_THISNODE)
1794 goto out;
1795 }
Mel Gorman11e33f62009-06-16 15:31:57 -07001796 /* Exhausted what can be done so it's blamo time */
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001797 out_of_memory(zonelist, gfp_mask, order, nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07001798
1799out:
1800 clear_zonelist_oom(zonelist, gfp_mask);
1801 return page;
1802}
1803
Mel Gorman56de7262010-05-24 14:32:30 -07001804#ifdef CONFIG_COMPACTION
1805/* Try memory compaction for high-order allocations before reclaim */
1806static struct page *
1807__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
1808 struct zonelist *zonelist, enum zone_type high_zoneidx,
1809 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman77f1fe62011-01-13 15:45:57 -08001810 int migratetype, unsigned long *did_some_progress,
1811 bool sync_migration)
Mel Gorman56de7262010-05-24 14:32:30 -07001812{
1813 struct page *page;
1814
Mel Gorman4f92e252010-05-24 14:32:32 -07001815 if (!order || compaction_deferred(preferred_zone))
Mel Gorman56de7262010-05-24 14:32:30 -07001816 return NULL;
1817
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001818 current->flags |= PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07001819 *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
Mel Gorman77f1fe62011-01-13 15:45:57 -08001820 nodemask, sync_migration);
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001821 current->flags &= ~PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07001822 if (*did_some_progress != COMPACT_SKIPPED) {
1823
1824 /* Page migration frees to the PCP lists but we want merging */
1825 drain_pages(get_cpu());
1826 put_cpu();
1827
1828 page = get_page_from_freelist(gfp_mask, nodemask,
1829 order, zonelist, high_zoneidx,
1830 alloc_flags, preferred_zone,
1831 migratetype);
1832 if (page) {
Mel Gorman4f92e252010-05-24 14:32:32 -07001833 preferred_zone->compact_considered = 0;
1834 preferred_zone->compact_defer_shift = 0;
Mel Gorman56de7262010-05-24 14:32:30 -07001835 count_vm_event(COMPACTSUCCESS);
1836 return page;
1837 }
1838
1839 /*
1840 * It's bad if compaction run occurs and fails.
1841 * The most likely reason is that pages exist,
1842 * but not enough to satisfy watermarks.
1843 */
1844 count_vm_event(COMPACTFAIL);
Mel Gorman4f92e252010-05-24 14:32:32 -07001845 defer_compaction(preferred_zone);
Mel Gorman56de7262010-05-24 14:32:30 -07001846
1847 cond_resched();
1848 }
1849
1850 return NULL;
1851}
1852#else
1853static inline struct page *
1854__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
1855 struct zonelist *zonelist, enum zone_type high_zoneidx,
1856 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman77f1fe62011-01-13 15:45:57 -08001857 int migratetype, unsigned long *did_some_progress,
1858 bool sync_migration)
Mel Gorman56de7262010-05-24 14:32:30 -07001859{
1860 return NULL;
1861}
1862#endif /* CONFIG_COMPACTION */
1863
Mel Gorman11e33f62009-06-16 15:31:57 -07001864/* The really slow allocator path where we enter direct reclaim */
1865static inline struct page *
1866__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
1867 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001868 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001869 int migratetype, unsigned long *did_some_progress)
Mel Gorman11e33f62009-06-16 15:31:57 -07001870{
1871 struct page *page = NULL;
1872 struct reclaim_state reclaim_state;
Mel Gorman9ee493c2010-09-09 16:38:18 -07001873 bool drained = false;
Mel Gorman11e33f62009-06-16 15:31:57 -07001874
1875 cond_resched();
1876
1877 /* We now go into synchronous reclaim */
1878 cpuset_memory_pressure_bump();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001879 current->flags |= PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07001880 lockdep_set_current_reclaim_state(gfp_mask);
1881 reclaim_state.reclaimed_slab = 0;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001882 current->reclaim_state = &reclaim_state;
Mel Gorman11e33f62009-06-16 15:31:57 -07001883
1884 *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
1885
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001886 current->reclaim_state = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07001887 lockdep_clear_current_reclaim_state();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001888 current->flags &= ~PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07001889
1890 cond_resched();
1891
Mel Gorman9ee493c2010-09-09 16:38:18 -07001892 if (unlikely(!(*did_some_progress)))
1893 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07001894
Mel Gorman9ee493c2010-09-09 16:38:18 -07001895retry:
1896 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001897 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001898 alloc_flags, preferred_zone,
1899 migratetype);
Mel Gorman9ee493c2010-09-09 16:38:18 -07001900
1901 /*
1902 * If an allocation failed after direct reclaim, it could be because
1903 * pages are pinned on the per-cpu lists. Drain them and try again
1904 */
1905 if (!page && !drained) {
1906 drain_all_pages();
1907 drained = true;
1908 goto retry;
1909 }
1910
Mel Gorman11e33f62009-06-16 15:31:57 -07001911 return page;
1912}
1913
Mel Gorman11e33f62009-06-16 15:31:57 -07001914/*
1915 * This is called in the allocator slow-path if the allocation request is of
1916 * sufficient urgency to ignore watermarks and take other desperate measures
1917 */
1918static inline struct page *
1919__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
1920 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001921 nodemask_t *nodemask, struct zone *preferred_zone,
1922 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001923{
1924 struct page *page;
1925
1926 do {
1927 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001928 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
Mel Gorman3dd28262009-06-16 15:32:00 -07001929 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001930
1931 if (!page && gfp_mask & __GFP_NOFAIL)
Mel Gorman0e093d992010-10-26 14:21:45 -07001932 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07001933 } while (!page && (gfp_mask & __GFP_NOFAIL));
1934
1935 return page;
1936}
1937
1938static inline
1939void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
Mel Gorman99504742011-01-13 15:46:20 -08001940 enum zone_type high_zoneidx,
1941 enum zone_type classzone_idx)
Mel Gorman11e33f62009-06-16 15:31:57 -07001942{
1943 struct zoneref *z;
1944 struct zone *zone;
1945
1946 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
Mel Gorman99504742011-01-13 15:46:20 -08001947 wakeup_kswapd(zone, order, classzone_idx);
Mel Gorman11e33f62009-06-16 15:31:57 -07001948}
1949
Peter Zijlstra341ce062009-06-16 15:32:02 -07001950static inline int
1951gfp_to_alloc_flags(gfp_t gfp_mask)
1952{
Peter Zijlstra341ce062009-06-16 15:32:02 -07001953 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
1954 const gfp_t wait = gfp_mask & __GFP_WAIT;
1955
Mel Gormana56f57f2009-06-16 15:32:02 -07001956 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07001957 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07001958
Peter Zijlstra341ce062009-06-16 15:32:02 -07001959 /*
1960 * The caller may dip into page reserves a bit more if the caller
1961 * cannot run direct reclaim, or if the caller has realtime scheduling
1962 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1963 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1964 */
Namhyung Kime6223a32010-10-26 14:21:59 -07001965 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07001966
1967 if (!wait) {
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08001968 /*
1969 * Not worth trying to allocate harder for
1970 * __GFP_NOMEMALLOC even if it can't schedule.
1971 */
1972 if (!(gfp_mask & __GFP_NOMEMALLOC))
1973 alloc_flags |= ALLOC_HARDER;
Peter Zijlstra341ce062009-06-16 15:32:02 -07001974 /*
1975 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1976 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1977 */
1978 alloc_flags &= ~ALLOC_CPUSET;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001979 } else if (unlikely(rt_task(current)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07001980 alloc_flags |= ALLOC_HARDER;
1981
1982 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
1983 if (!in_interrupt() &&
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001984 ((current->flags & PF_MEMALLOC) ||
Peter Zijlstra341ce062009-06-16 15:32:02 -07001985 unlikely(test_thread_flag(TIF_MEMDIE))))
1986 alloc_flags |= ALLOC_NO_WATERMARKS;
1987 }
1988
1989 return alloc_flags;
1990}
1991
Mel Gorman11e33f62009-06-16 15:31:57 -07001992static inline struct page *
1993__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
1994 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001995 nodemask_t *nodemask, struct zone *preferred_zone,
1996 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001997{
1998 const gfp_t wait = gfp_mask & __GFP_WAIT;
1999 struct page *page = NULL;
2000 int alloc_flags;
2001 unsigned long pages_reclaimed = 0;
2002 unsigned long did_some_progress;
Mel Gorman77f1fe62011-01-13 15:45:57 -08002003 bool sync_migration = false;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002004
Christoph Lameter952f3b52006-12-06 20:33:26 -08002005 /*
Mel Gorman72807a72009-06-16 15:32:18 -07002006 * In the slowpath, we sanity check order to avoid ever trying to
2007 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2008 * be using allocators in order of preference for an area that is
2009 * too large.
2010 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07002011 if (order >= MAX_ORDER) {
2012 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07002013 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07002014 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002015
Christoph Lameter952f3b52006-12-06 20:33:26 -08002016 /*
2017 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2018 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2019 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2020 * using a larger set of nodes after it has established that the
2021 * allowed per node queues are empty and that nodes are
2022 * over allocated.
2023 */
2024 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
2025 goto nopage;
2026
Mel Gormancc4a6852009-11-11 14:26:14 -08002027restart:
Andrea Arcangeli32dba982011-01-13 15:46:49 -08002028 if (!(gfp_mask & __GFP_NO_KSWAPD))
2029 wake_all_kswapd(order, zonelist, high_zoneidx,
Mel Gorman99504742011-01-13 15:46:20 -08002030 zone_idx(preferred_zone));
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002031
Paul Jackson9bf22292005-09-06 15:18:12 -07002032 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002033 * OK, we're below the kswapd watermark and have kicked background
2034 * reclaim. Now things get more complex, so set up alloc_flags according
2035 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07002036 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002037 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038
David Rientjesf33261d2011-01-25 15:07:20 -08002039 /*
2040 * Find the true preferred zone if the allocation is unconstrained by
2041 * cpusets.
2042 */
2043 if (!(alloc_flags & ALLOC_CPUSET) && !nodemask)
2044 first_zones_zonelist(zonelist, high_zoneidx, NULL,
2045 &preferred_zone);
2046
Peter Zijlstra341ce062009-06-16 15:32:02 -07002047 /* This is the last chance, in general, before the goto nopage. */
Mel Gorman19770b32008-04-28 02:12:18 -07002048 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
Peter Zijlstra341ce062009-06-16 15:32:02 -07002049 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
2050 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002051 if (page)
2052 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053
Kirill Korotaevb43a57b2006-12-06 20:32:27 -08002054rebalance:
Mel Gorman11e33f62009-06-16 15:31:57 -07002055 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002056 if (alloc_flags & ALLOC_NO_WATERMARKS) {
2057 page = __alloc_pages_high_priority(gfp_mask, order,
2058 zonelist, high_zoneidx, nodemask,
2059 preferred_zone, migratetype);
2060 if (page)
2061 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 }
2063
2064 /* Atomic allocations - we can't balance anything */
2065 if (!wait)
2066 goto nopage;
2067
Peter Zijlstra341ce062009-06-16 15:32:02 -07002068 /* Avoid recursion of direct reclaim */
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002069 if (current->flags & PF_MEMALLOC)
Peter Zijlstra341ce062009-06-16 15:32:02 -07002070 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071
David Rientjes6583bb62009-07-29 15:02:06 -07002072 /* Avoid allocations with no watermarks from looping endlessly */
2073 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2074 goto nopage;
2075
Mel Gorman77f1fe62011-01-13 15:45:57 -08002076 /*
2077 * Try direct compaction. The first pass is asynchronous. Subsequent
2078 * attempts after direct reclaim are synchronous
2079 */
Mel Gorman56de7262010-05-24 14:32:30 -07002080 page = __alloc_pages_direct_compact(gfp_mask, order,
2081 zonelist, high_zoneidx,
2082 nodemask,
2083 alloc_flags, preferred_zone,
Mel Gorman77f1fe62011-01-13 15:45:57 -08002084 migratetype, &did_some_progress,
2085 sync_migration);
Mel Gorman56de7262010-05-24 14:32:30 -07002086 if (page)
2087 goto got_pg;
Mel Gorman77f1fe62011-01-13 15:45:57 -08002088 sync_migration = true;
Mel Gorman56de7262010-05-24 14:32:30 -07002089
Mel Gorman11e33f62009-06-16 15:31:57 -07002090 /* Try direct reclaim and then allocating */
2091 page = __alloc_pages_direct_reclaim(gfp_mask, order,
2092 zonelist, high_zoneidx,
2093 nodemask,
Mel Gorman5117f452009-06-16 15:31:59 -07002094 alloc_flags, preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07002095 migratetype, &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07002096 if (page)
2097 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 /*
Mel Gorman11e33f62009-06-16 15:31:57 -07002100 * If we failed to make any progress reclaiming, then we are
2101 * running out of options and have to consider going OOM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102 */
Mel Gorman11e33f62009-06-16 15:31:57 -07002103 if (!did_some_progress) {
2104 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -07002105 if (oom_killer_disabled)
2106 goto nopage;
Mel Gorman11e33f62009-06-16 15:31:57 -07002107 page = __alloc_pages_may_oom(gfp_mask, order,
2108 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002109 nodemask, preferred_zone,
2110 migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002111 if (page)
2112 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113
David Rientjes03668b32010-08-09 17:18:54 -07002114 if (!(gfp_mask & __GFP_NOFAIL)) {
2115 /*
2116 * The oom killer is not called for high-order
2117 * allocations that may fail, so if no progress
2118 * is being made, there are no other options and
2119 * retrying is unlikely to help.
2120 */
2121 if (order > PAGE_ALLOC_COSTLY_ORDER)
2122 goto nopage;
2123 /*
2124 * The oom killer is not called for lowmem
2125 * allocations to prevent needlessly killing
2126 * innocent tasks.
2127 */
2128 if (high_zoneidx < ZONE_NORMAL)
2129 goto nopage;
2130 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 goto restart;
2133 }
2134 }
2135
Mel Gorman11e33f62009-06-16 15:31:57 -07002136 /* Check if we should retry the allocation */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07002137 pages_reclaimed += did_some_progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07002138 if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
2139 /* Wait for some write requests to complete then retry */
Mel Gorman0e093d992010-10-26 14:21:45 -07002140 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 goto rebalance;
Mel Gorman3e7d3442011-01-13 15:45:56 -08002142 } else {
2143 /*
2144 * High-order allocations do not necessarily loop after
2145 * direct reclaim and reclaim/compaction depends on compaction
2146 * being called after reclaim so call directly if necessary
2147 */
2148 page = __alloc_pages_direct_compact(gfp_mask, order,
2149 zonelist, high_zoneidx,
2150 nodemask,
2151 alloc_flags, preferred_zone,
Mel Gorman77f1fe62011-01-13 15:45:57 -08002152 migratetype, &did_some_progress,
2153 sync_migration);
Mel Gorman3e7d3442011-01-13 15:45:56 -08002154 if (page)
2155 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156 }
2157
2158nopage:
2159 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
2160 printk(KERN_WARNING "%s: page allocation failure."
2161 " order:%d, mode:0x%x\n",
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002162 current->comm, order, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07002164 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 }
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002166 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167got_pg:
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002168 if (kmemcheck_enabled)
2169 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 return page;
Mel Gorman11e33f62009-06-16 15:31:57 -07002171
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172}
Mel Gorman11e33f62009-06-16 15:31:57 -07002173
2174/*
2175 * This is the 'heart' of the zoned buddy allocator.
2176 */
2177struct page *
2178__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2179 struct zonelist *zonelist, nodemask_t *nodemask)
2180{
2181 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Mel Gorman5117f452009-06-16 15:31:59 -07002182 struct zone *preferred_zone;
Mel Gorman11e33f62009-06-16 15:31:57 -07002183 struct page *page;
Mel Gorman3dd28262009-06-16 15:32:00 -07002184 int migratetype = allocflags_to_migratetype(gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002185
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10002186 gfp_mask &= gfp_allowed_mask;
2187
Mel Gorman11e33f62009-06-16 15:31:57 -07002188 lockdep_trace_alloc(gfp_mask);
2189
2190 might_sleep_if(gfp_mask & __GFP_WAIT);
2191
2192 if (should_fail_alloc_page(gfp_mask, order))
2193 return NULL;
2194
2195 /*
2196 * Check the zones suitable for the gfp_mask contain at least one
2197 * valid zone. It's possible to have an empty zonelist as a result
2198 * of GFP_THISNODE and a memoryless node
2199 */
2200 if (unlikely(!zonelist->_zonerefs->zone))
2201 return NULL;
2202
Miao Xiec0ff7452010-05-24 14:32:08 -07002203 get_mems_allowed();
Mel Gorman5117f452009-06-16 15:31:59 -07002204 /* The preferred zone is used for statistics later */
David Rientjesf33261d2011-01-25 15:07:20 -08002205 first_zones_zonelist(zonelist, high_zoneidx,
2206 nodemask ? : &cpuset_current_mems_allowed,
2207 &preferred_zone);
Miao Xiec0ff7452010-05-24 14:32:08 -07002208 if (!preferred_zone) {
2209 put_mems_allowed();
Mel Gorman5117f452009-06-16 15:31:59 -07002210 return NULL;
Miao Xiec0ff7452010-05-24 14:32:08 -07002211 }
Mel Gorman5117f452009-06-16 15:31:59 -07002212
2213 /* First allocation attempt */
Mel Gorman11e33f62009-06-16 15:31:57 -07002214 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002215 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07002216 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002217 if (unlikely(!page))
2218 page = __alloc_pages_slowpath(gfp_mask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002219 zonelist, high_zoneidx, nodemask,
Mel Gorman3dd28262009-06-16 15:32:00 -07002220 preferred_zone, migratetype);
Miao Xiec0ff7452010-05-24 14:32:08 -07002221 put_mems_allowed();
Mel Gorman11e33f62009-06-16 15:31:57 -07002222
Mel Gorman4b4f2782009-09-21 17:02:41 -07002223 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002224 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225}
Mel Gormand2391712009-06-16 15:31:52 -07002226EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227
2228/*
2229 * Common helper functions.
2230 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08002231unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232{
Akinobu Mita945a1112009-09-21 17:01:47 -07002233 struct page *page;
2234
2235 /*
2236 * __get_free_pages() returns a 32-bit address, which cannot represent
2237 * a highmem page
2238 */
2239 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2240
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241 page = alloc_pages(gfp_mask, order);
2242 if (!page)
2243 return 0;
2244 return (unsigned long) page_address(page);
2245}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246EXPORT_SYMBOL(__get_free_pages);
2247
Harvey Harrison920c7a52008-02-04 22:29:26 -08002248unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249{
Akinobu Mita945a1112009-09-21 17:01:47 -07002250 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252EXPORT_SYMBOL(get_zeroed_page);
2253
2254void __pagevec_free(struct pagevec *pvec)
2255{
2256 int i = pagevec_count(pvec);
2257
Mel Gorman4b4f2782009-09-21 17:02:41 -07002258 while (--i >= 0) {
2259 trace_mm_pagevec_free(pvec->pages[i], pvec->cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 free_hot_cold_page(pvec->pages[i], pvec->cold);
Mel Gorman4b4f2782009-09-21 17:02:41 -07002261 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262}
2263
Harvey Harrison920c7a52008-02-04 22:29:26 -08002264void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265{
Nick Pigginb5810032005-10-29 18:16:12 -07002266 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 if (order == 0)
Li Hongfc916682010-03-05 13:41:54 -08002268 free_hot_cold_page(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 else
2270 __free_pages_ok(page, order);
2271 }
2272}
2273
2274EXPORT_SYMBOL(__free_pages);
2275
Harvey Harrison920c7a52008-02-04 22:29:26 -08002276void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277{
2278 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07002279 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280 __free_pages(virt_to_page((void *)addr), order);
2281 }
2282}
2283
2284EXPORT_SYMBOL(free_pages);
2285
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002286/**
2287 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2288 * @size: the number of bytes to allocate
2289 * @gfp_mask: GFP flags for the allocation
2290 *
2291 * This function is similar to alloc_pages(), except that it allocates the
2292 * minimum number of pages to satisfy the request. alloc_pages() can only
2293 * allocate memory in power-of-two pages.
2294 *
2295 * This function is also limited by MAX_ORDER.
2296 *
2297 * Memory allocated by this function must be released by free_pages_exact().
2298 */
2299void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2300{
2301 unsigned int order = get_order(size);
2302 unsigned long addr;
2303
2304 addr = __get_free_pages(gfp_mask, order);
2305 if (addr) {
2306 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2307 unsigned long used = addr + PAGE_ALIGN(size);
2308
Kevin Cernekee5bfd7562009-07-05 12:08:19 -07002309 split_page(virt_to_page((void *)addr), order);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002310 while (used < alloc_end) {
2311 free_page(used);
2312 used += PAGE_SIZE;
2313 }
2314 }
2315
2316 return (void *)addr;
2317}
2318EXPORT_SYMBOL(alloc_pages_exact);
2319
2320/**
2321 * free_pages_exact - release memory allocated via alloc_pages_exact()
2322 * @virt: the value returned by alloc_pages_exact.
2323 * @size: size of allocation, same value as passed to alloc_pages_exact().
2324 *
2325 * Release the memory allocated by a previous call to alloc_pages_exact.
2326 */
2327void free_pages_exact(void *virt, size_t size)
2328{
2329 unsigned long addr = (unsigned long)virt;
2330 unsigned long end = addr + PAGE_ALIGN(size);
2331
2332 while (addr < end) {
2333 free_page(addr);
2334 addr += PAGE_SIZE;
2335 }
2336}
2337EXPORT_SYMBOL(free_pages_exact);
2338
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339static unsigned int nr_free_zone_pages(int offset)
2340{
Mel Gormandd1a2392008-04-28 02:12:17 -07002341 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002342 struct zone *zone;
2343
Martin J. Blighe310fd42005-07-29 22:59:18 -07002344 /* Just pick one node, since fallback list is circular */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 unsigned int sum = 0;
2346
Mel Gorman0e884602008-04-28 02:12:14 -07002347 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348
Mel Gorman54a6eb52008-04-28 02:12:16 -07002349 for_each_zone_zonelist(zone, z, zonelist, offset) {
Martin J. Blighe310fd42005-07-29 22:59:18 -07002350 unsigned long size = zone->present_pages;
Mel Gorman41858962009-06-16 15:32:12 -07002351 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07002352 if (size > high)
2353 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354 }
2355
2356 return sum;
2357}
2358
2359/*
2360 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2361 */
2362unsigned int nr_free_buffer_pages(void)
2363{
Al Viroaf4ca452005-10-21 02:55:38 -04002364 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365}
Meelap Shahc2f1a552007-07-17 04:04:39 -07002366EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367
2368/*
2369 * Amount of free RAM allocatable within all zones
2370 */
2371unsigned int nr_free_pagecache_pages(void)
2372{
Mel Gorman2a1e2742007-07-17 04:03:12 -07002373 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002375
2376static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377{
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002378 if (NUMA_BUILD)
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08002379 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382void si_meminfo(struct sysinfo *val)
2383{
2384 val->totalram = totalram_pages;
2385 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002386 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 val->totalhigh = totalhigh_pages;
2389 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390 val->mem_unit = PAGE_SIZE;
2391}
2392
2393EXPORT_SYMBOL(si_meminfo);
2394
2395#ifdef CONFIG_NUMA
2396void si_meminfo_node(struct sysinfo *val, int nid)
2397{
2398 pg_data_t *pgdat = NODE_DATA(nid);
2399
2400 val->totalram = pgdat->node_present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002401 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002402#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002404 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2405 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002406#else
2407 val->totalhigh = 0;
2408 val->freehigh = 0;
2409#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410 val->mem_unit = PAGE_SIZE;
2411}
2412#endif
2413
2414#define K(x) ((x) << (PAGE_SHIFT-10))
2415
2416/*
2417 * Show free area list (used inside shift_scroll-lock stuff)
2418 * We also calculate the percentage fragmentation. We do this by counting the
2419 * memory on each free list with the exception of the first item on the list.
2420 */
2421void show_free_areas(void)
2422{
Jes Sorensenc7241912006-09-27 01:50:05 -07002423 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 struct zone *zone;
2425
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002426 for_each_populated_zone(zone) {
Jes Sorensenc7241912006-09-27 01:50:05 -07002427 show_node(zone);
2428 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429
Dave Jones6b482c62005-11-10 15:45:56 -05002430 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 struct per_cpu_pageset *pageset;
2432
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002433 pageset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002435 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2436 cpu, pageset->pcp.high,
2437 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 }
2439 }
2440
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002441 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2442 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002443 " unevictable:%lu"
Andrew Mortonb76146e2009-10-26 16:49:52 -07002444 " dirty:%lu writeback:%lu unstable:%lu\n"
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07002445 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002446 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002447 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002448 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002449 global_page_state(NR_ISOLATED_ANON),
2450 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002451 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002452 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002453 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07002454 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07002455 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07002456 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08002457 global_page_state(NR_FREE_PAGES),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07002458 global_page_state(NR_SLAB_RECLAIMABLE),
2459 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07002460 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002461 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08002462 global_page_state(NR_PAGETABLE),
2463 global_page_state(NR_BOUNCE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002465 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466 int i;
2467
2468 show_node(zone);
2469 printk("%s"
2470 " free:%lukB"
2471 " min:%lukB"
2472 " low:%lukB"
2473 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002474 " active_anon:%lukB"
2475 " inactive_anon:%lukB"
2476 " active_file:%lukB"
2477 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002478 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002479 " isolated(anon):%lukB"
2480 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 " present:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002482 " mlocked:%lukB"
2483 " dirty:%lukB"
2484 " writeback:%lukB"
2485 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002486 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002487 " slab_reclaimable:%lukB"
2488 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07002489 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002490 " pagetables:%lukB"
2491 " unstable:%lukB"
2492 " bounce:%lukB"
2493 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 " pages_scanned:%lu"
2495 " all_unreclaimable? %s"
2496 "\n",
2497 zone->name,
Mel Gorman88f5acf2011-01-13 15:45:41 -08002498 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07002499 K(min_wmark_pages(zone)),
2500 K(low_wmark_pages(zone)),
2501 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002502 K(zone_page_state(zone, NR_ACTIVE_ANON)),
2503 K(zone_page_state(zone, NR_INACTIVE_ANON)),
2504 K(zone_page_state(zone, NR_ACTIVE_FILE)),
2505 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002506 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002507 K(zone_page_state(zone, NR_ISOLATED_ANON)),
2508 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509 K(zone->present_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002510 K(zone_page_state(zone, NR_MLOCK)),
2511 K(zone_page_state(zone, NR_FILE_DIRTY)),
2512 K(zone_page_state(zone, NR_WRITEBACK)),
2513 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002514 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002515 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
2516 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07002517 zone_page_state(zone, NR_KERNEL_STACK) *
2518 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002519 K(zone_page_state(zone, NR_PAGETABLE)),
2520 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
2521 K(zone_page_state(zone, NR_BOUNCE)),
2522 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523 zone->pages_scanned,
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -08002524 (zone->all_unreclaimable ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525 );
2526 printk("lowmem_reserve[]:");
2527 for (i = 0; i < MAX_NR_ZONES; i++)
2528 printk(" %lu", zone->lowmem_reserve[i]);
2529 printk("\n");
2530 }
2531
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002532 for_each_populated_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002533 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534
2535 show_node(zone);
2536 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537
2538 spin_lock_irqsave(&zone->lock, flags);
2539 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002540 nr[order] = zone->free_area[order].nr_free;
2541 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 }
2543 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002544 for (order = 0; order < MAX_ORDER; order++)
2545 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 printk("= %lukB\n", K(total));
2547 }
2548
Larry Woodmane6f36022008-02-04 22:29:30 -08002549 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2550
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 show_swap_cache_info();
2552}
2553
Mel Gorman19770b32008-04-28 02:12:18 -07002554static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2555{
2556 zoneref->zone = zone;
2557 zoneref->zone_idx = zone_idx(zone);
2558}
2559
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560/*
2561 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08002562 *
2563 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002565static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2566 int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567{
Christoph Lameter1a932052006-01-06 00:11:16 -08002568 struct zone *zone;
2569
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002570 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002571 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08002572
2573 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002574 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08002575 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08002576 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002577 zoneref_set_zone(zone,
2578 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08002579 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08002581
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002582 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08002583 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584}
2585
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002586
2587/*
2588 * zonelist_order:
2589 * 0 = automatic detection of better ordering.
2590 * 1 = order by ([node] distance, -zonetype)
2591 * 2 = order by (-zonetype, [node] distance)
2592 *
2593 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2594 * the same zonelist. So only NUMA can configure this param.
2595 */
2596#define ZONELIST_ORDER_DEFAULT 0
2597#define ZONELIST_ORDER_NODE 1
2598#define ZONELIST_ORDER_ZONE 2
2599
2600/* zonelist order in the kernel.
2601 * set_zonelist_order() will set this to NODE or ZONE.
2602 */
2603static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2604static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2605
2606
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002608/* The value user specified ....changed by config */
2609static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2610/* string for sysctl */
2611#define NUMA_ZONELIST_ORDER_LEN 16
2612char numa_zonelist_order[16] = "default";
2613
2614/*
2615 * interface for configure zonelist ordering.
2616 * command line option "numa_zonelist_order"
2617 * = "[dD]efault - default, automatic configuration.
2618 * = "[nN]ode - order by node locality, then by zone within node
2619 * = "[zZ]one - order by zone, then by locality within zone
2620 */
2621
2622static int __parse_numa_zonelist_order(char *s)
2623{
2624 if (*s == 'd' || *s == 'D') {
2625 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2626 } else if (*s == 'n' || *s == 'N') {
2627 user_zonelist_order = ZONELIST_ORDER_NODE;
2628 } else if (*s == 'z' || *s == 'Z') {
2629 user_zonelist_order = ZONELIST_ORDER_ZONE;
2630 } else {
2631 printk(KERN_WARNING
2632 "Ignoring invalid numa_zonelist_order value: "
2633 "%s\n", s);
2634 return -EINVAL;
2635 }
2636 return 0;
2637}
2638
2639static __init int setup_numa_zonelist_order(char *s)
2640{
Volodymyr G. Lukiianykecb256f2011-01-13 15:46:26 -08002641 int ret;
2642
2643 if (!s)
2644 return 0;
2645
2646 ret = __parse_numa_zonelist_order(s);
2647 if (ret == 0)
2648 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
2649
2650 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002651}
2652early_param("numa_zonelist_order", setup_numa_zonelist_order);
2653
2654/*
2655 * sysctl handler for numa_zonelist_order
2656 */
2657int numa_zonelist_order_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07002658 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002659 loff_t *ppos)
2660{
2661 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2662 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01002663 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002664
Andi Kleen443c6f12009-12-23 21:00:47 +01002665 mutex_lock(&zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002666 if (write)
Andi Kleen443c6f12009-12-23 21:00:47 +01002667 strcpy(saved_string, (char*)table->data);
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07002668 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002669 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01002670 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002671 if (write) {
2672 int oldval = user_zonelist_order;
2673 if (__parse_numa_zonelist_order((char*)table->data)) {
2674 /*
2675 * bogus value. restore saved string
2676 */
2677 strncpy((char*)table->data, saved_string,
2678 NUMA_ZONELIST_ORDER_LEN);
2679 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07002680 } else if (oldval != user_zonelist_order) {
2681 mutex_lock(&zonelists_mutex);
Haicheng Li1f522502010-05-24 14:32:51 -07002682 build_all_zonelists(NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07002683 mutex_unlock(&zonelists_mutex);
2684 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002685 }
Andi Kleen443c6f12009-12-23 21:00:47 +01002686out:
2687 mutex_unlock(&zl_order_mutex);
2688 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002689}
2690
2691
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002692#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002693static int node_load[MAX_NUMNODES];
2694
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07002696 * 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 -07002697 * @node: node whose fallback list we're appending
2698 * @used_node_mask: nodemask_t of already used nodes
2699 *
2700 * We use a number of factors to determine which is the next node that should
2701 * appear on a given node's fallback list. The node should not have appeared
2702 * already in @node's fallback list, and it should be the next closest node
2703 * according to the distance array (which contains arbitrary distance values
2704 * from each node to each node in the system), and should also prefer nodes
2705 * with no CPUs, since presumably they'll have very little allocation pressure
2706 * on them otherwise.
2707 * It returns -1 if no node is found.
2708 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002709static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002711 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712 int min_val = INT_MAX;
2713 int best_node = -1;
Rusty Russella70f7302009-03-13 14:49:46 +10302714 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002716 /* Use the local node if we haven't already */
2717 if (!node_isset(node, *used_node_mask)) {
2718 node_set(node, *used_node_mask);
2719 return node;
2720 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002721
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002722 for_each_node_state(n, N_HIGH_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723
2724 /* Don't want a node to appear more than once */
2725 if (node_isset(n, *used_node_mask))
2726 continue;
2727
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728 /* Use the distance array to find the distance */
2729 val = node_distance(node, n);
2730
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002731 /* Penalize nodes under us ("prefer the next node") */
2732 val += (n < node);
2733
Linus Torvalds1da177e2005-04-16 15:20:36 -07002734 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10302735 tmp = cpumask_of_node(n);
2736 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737 val += PENALTY_FOR_NODE_WITH_CPUS;
2738
2739 /* Slight preference for less loaded node */
2740 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2741 val += node_load[n];
2742
2743 if (val < min_val) {
2744 min_val = val;
2745 best_node = n;
2746 }
2747 }
2748
2749 if (best_node >= 0)
2750 node_set(best_node, *used_node_mask);
2751
2752 return best_node;
2753}
2754
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002755
2756/*
2757 * Build zonelists ordered by node and zones within node.
2758 * This results in maximum locality--normal zone overflows into local
2759 * DMA zone, if any--but risks exhausting DMA zone.
2760 */
2761static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002763 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002764 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002765
Mel Gorman54a6eb52008-04-28 02:12:16 -07002766 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07002767 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07002768 ;
2769 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2770 MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002771 zonelist->_zonerefs[j].zone = NULL;
2772 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002773}
2774
2775/*
Christoph Lameter523b9452007-10-16 01:25:37 -07002776 * Build gfp_thisnode zonelists
2777 */
2778static void build_thisnode_zonelists(pg_data_t *pgdat)
2779{
Christoph Lameter523b9452007-10-16 01:25:37 -07002780 int j;
2781 struct zonelist *zonelist;
2782
Mel Gorman54a6eb52008-04-28 02:12:16 -07002783 zonelist = &pgdat->node_zonelists[1];
2784 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002785 zonelist->_zonerefs[j].zone = NULL;
2786 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07002787}
2788
2789/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002790 * Build zonelists ordered by zone and nodes within zones.
2791 * This results in conserving DMA zone[s] until all Normal memory is
2792 * exhausted, but results in overflowing to remote node while memory
2793 * may still exist in local DMA zone.
2794 */
2795static int node_order[MAX_NUMNODES];
2796
2797static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2798{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002799 int pos, j, node;
2800 int zone_type; /* needs to be signed */
2801 struct zone *z;
2802 struct zonelist *zonelist;
2803
Mel Gorman54a6eb52008-04-28 02:12:16 -07002804 zonelist = &pgdat->node_zonelists[0];
2805 pos = 0;
2806 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2807 for (j = 0; j < nr_nodes; j++) {
2808 node = node_order[j];
2809 z = &NODE_DATA(node)->node_zones[zone_type];
2810 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002811 zoneref_set_zone(z,
2812 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002813 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002814 }
2815 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002816 }
Mel Gormandd1a2392008-04-28 02:12:17 -07002817 zonelist->_zonerefs[pos].zone = NULL;
2818 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002819}
2820
2821static int default_zonelist_order(void)
2822{
2823 int nid, zone_type;
2824 unsigned long low_kmem_size,total_size;
2825 struct zone *z;
2826 int average_size;
2827 /*
Thomas Weber88393162010-03-16 11:47:56 +01002828 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002829 * If they are really small and used heavily, the system can fall
2830 * into OOM very easily.
David Rientjese325c902010-05-24 14:32:13 -07002831 * This function detect ZONE_DMA/DMA32 size and configures zone order.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002832 */
2833 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2834 low_kmem_size = 0;
2835 total_size = 0;
2836 for_each_online_node(nid) {
2837 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2838 z = &NODE_DATA(nid)->node_zones[zone_type];
2839 if (populated_zone(z)) {
2840 if (zone_type < ZONE_NORMAL)
2841 low_kmem_size += z->present_pages;
2842 total_size += z->present_pages;
David Rientjese325c902010-05-24 14:32:13 -07002843 } else if (zone_type == ZONE_NORMAL) {
2844 /*
2845 * If any node has only lowmem, then node order
2846 * is preferred to allow kernel allocations
2847 * locally; otherwise, they can easily infringe
2848 * on other nodes when there is an abundance of
2849 * lowmem available to allocate from.
2850 */
2851 return ZONELIST_ORDER_NODE;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002852 }
2853 }
2854 }
2855 if (!low_kmem_size || /* there are no DMA area. */
2856 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2857 return ZONELIST_ORDER_NODE;
2858 /*
2859 * look into each node's config.
2860 * If there is a node whose DMA/DMA32 memory is very big area on
2861 * local memory, NODE_ORDER may be suitable.
2862 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002863 average_size = total_size /
2864 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002865 for_each_online_node(nid) {
2866 low_kmem_size = 0;
2867 total_size = 0;
2868 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2869 z = &NODE_DATA(nid)->node_zones[zone_type];
2870 if (populated_zone(z)) {
2871 if (zone_type < ZONE_NORMAL)
2872 low_kmem_size += z->present_pages;
2873 total_size += z->present_pages;
2874 }
2875 }
2876 if (low_kmem_size &&
2877 total_size > average_size && /* ignore small node */
2878 low_kmem_size > total_size * 70/100)
2879 return ZONELIST_ORDER_NODE;
2880 }
2881 return ZONELIST_ORDER_ZONE;
2882}
2883
2884static void set_zonelist_order(void)
2885{
2886 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2887 current_zonelist_order = default_zonelist_order();
2888 else
2889 current_zonelist_order = user_zonelist_order;
2890}
2891
2892static void build_zonelists(pg_data_t *pgdat)
2893{
2894 int j, node, load;
2895 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002897 int local_node, prev_node;
2898 struct zonelist *zonelist;
2899 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002900
2901 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07002902 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002903 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07002904 zonelist->_zonerefs[0].zone = NULL;
2905 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906 }
2907
2908 /* NUMA-aware ordering of nodes */
2909 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002910 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002911 prev_node = local_node;
2912 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002913
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002914 memset(node_order, 0, sizeof(node_order));
2915 j = 0;
2916
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08002918 int distance = node_distance(local_node, node);
2919
2920 /*
2921 * If another node is sufficiently far away then it is better
2922 * to reclaim pages in a zone before going off node.
2923 */
2924 if (distance > RECLAIM_DISTANCE)
2925 zone_reclaim_mode = 1;
2926
Linus Torvalds1da177e2005-04-16 15:20:36 -07002927 /*
2928 * We don't want to pressure a particular node.
2929 * So adding penalty to the first node in same
2930 * distance group to make it round-robin.
2931 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08002932 if (distance != node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002933 node_load[node] = load;
2934
Linus Torvalds1da177e2005-04-16 15:20:36 -07002935 prev_node = node;
2936 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002937 if (order == ZONELIST_ORDER_NODE)
2938 build_zonelists_in_node_order(pgdat, node);
2939 else
2940 node_order[j++] = node; /* remember order */
2941 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002942
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002943 if (order == ZONELIST_ORDER_ZONE) {
2944 /* calculate node order -- i.e., DMA last! */
2945 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002946 }
Christoph Lameter523b9452007-10-16 01:25:37 -07002947
2948 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949}
2950
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002951/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002952static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002953{
Mel Gorman54a6eb52008-04-28 02:12:16 -07002954 struct zonelist *zonelist;
2955 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07002956 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002957
Mel Gorman54a6eb52008-04-28 02:12:16 -07002958 zonelist = &pgdat->node_zonelists[0];
2959 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2960 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07002961 for (z = zonelist->_zonerefs; z->zone; z++)
2962 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002963}
2964
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07002965#ifdef CONFIG_HAVE_MEMORYLESS_NODES
2966/*
2967 * Return node id of node used for "local" allocations.
2968 * I.e., first node id of first zone in arg node's generic zonelist.
2969 * Used for initializing percpu 'numa_mem', which is used primarily
2970 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
2971 */
2972int local_memory_node(int node)
2973{
2974 struct zone *zone;
2975
2976 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
2977 gfp_zone(GFP_KERNEL),
2978 NULL,
2979 &zone);
2980 return zone->node;
2981}
2982#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002983
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984#else /* CONFIG_NUMA */
2985
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002986static void set_zonelist_order(void)
2987{
2988 current_zonelist_order = ZONELIST_ORDER_ZONE;
2989}
2990
2991static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992{
Christoph Lameter19655d32006-09-25 23:31:19 -07002993 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002994 enum zone_type j;
2995 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002996
2997 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998
Mel Gorman54a6eb52008-04-28 02:12:16 -07002999 zonelist = &pgdat->node_zonelists[0];
3000 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001
Mel Gorman54a6eb52008-04-28 02:12:16 -07003002 /*
3003 * Now we build the zonelist so that it contains the zones
3004 * of all the other nodes.
3005 * We don't want to pressure a particular node, so when
3006 * building the zones for node N, we make sure that the
3007 * zones coming right after the local ones are those from
3008 * node N+1 (modulo N)
3009 */
3010 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3011 if (!node_online(node))
3012 continue;
3013 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3014 MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003015 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07003016 for (node = 0; node < local_node; node++) {
3017 if (!node_online(node))
3018 continue;
3019 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3020 MAX_NR_ZONES - 1);
3021 }
3022
Mel Gormandd1a2392008-04-28 02:12:17 -07003023 zonelist->_zonerefs[j].zone = NULL;
3024 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003025}
3026
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003027/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003028static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003029{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003030 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003031}
3032
Linus Torvalds1da177e2005-04-16 15:20:36 -07003033#endif /* CONFIG_NUMA */
3034
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003035/*
3036 * Boot pageset table. One per cpu which is going to be used for all
3037 * zones and all nodes. The parameters will be set in such a way
3038 * that an item put on a list will immediately be handed over to
3039 * the buddy list. This is safe since pageset manipulation is done
3040 * with interrupts disabled.
3041 *
3042 * The boot_pagesets must be kept even after bootup is complete for
3043 * unused processors and/or zones. They do play a role for bootstrapping
3044 * hotplugged processors.
3045 *
3046 * zoneinfo_show() and maybe other functions do
3047 * not check if the processor is online before following the pageset pointer.
3048 * Other parts of the kernel may not check if the zone is available.
3049 */
3050static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3051static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07003052static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003053
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003054/*
3055 * Global mutex to protect against size modification of zonelists
3056 * as well as to serialize pageset setup for the new populated zone.
3057 */
3058DEFINE_MUTEX(zonelists_mutex);
3059
Rusty Russell9b1a4d32008-07-28 12:16:30 -05003060/* return values int ....just for stop_machine() */
Haicheng Li1f522502010-05-24 14:32:51 -07003061static __init_refok int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003062{
Yasunori Goto68113782006-06-23 02:03:11 -07003063 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003064 int cpu;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003065
Bo Liu7f9cfb32009-08-18 14:11:19 -07003066#ifdef CONFIG_NUMA
3067 memset(node_load, 0, sizeof(node_load));
3068#endif
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003069 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07003070 pg_data_t *pgdat = NODE_DATA(nid);
3071
3072 build_zonelists(pgdat);
3073 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003074 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003075
3076 /*
3077 * Initialize the boot_pagesets that are going to be used
3078 * for bootstrapping processors. The real pagesets for
3079 * each zone will be allocated later when the per cpu
3080 * allocator is available.
3081 *
3082 * boot_pagesets are used also for bootstrapping offline
3083 * cpus if the system is already booted because the pagesets
3084 * are needed to initialize allocators on a specific cpu too.
3085 * F.e. the percpu allocator needs the page allocator which
3086 * needs the percpu allocator in order to allocate its pagesets
3087 * (a chicken-egg dilemma).
3088 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003089 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003090 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3091
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003092#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3093 /*
3094 * We now know the "local memory node" for each node--
3095 * i.e., the node of the first zone in the generic zonelist.
3096 * Set up numa_mem percpu variable for on-line cpus. During
3097 * boot, only the boot cpu should be on-line; we'll init the
3098 * secondary cpus' numa_mem as they come on-line. During
3099 * node/memory hotplug, we'll fixup all on-line cpus.
3100 */
3101 if (cpu_online(cpu))
3102 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3103#endif
3104 }
3105
Yasunori Goto68113782006-06-23 02:03:11 -07003106 return 0;
3107}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003109/*
3110 * Called with zonelists_mutex held always
3111 * unless system_state == SYSTEM_BOOTING.
3112 */
Haicheng Li1f522502010-05-24 14:32:51 -07003113void build_all_zonelists(void *data)
Yasunori Goto68113782006-06-23 02:03:11 -07003114{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003115 set_zonelist_order();
3116
Yasunori Goto68113782006-06-23 02:03:11 -07003117 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07003118 __build_all_zonelists(NULL);
Mel Gorman68ad8df2008-07-23 21:26:52 -07003119 mminit_verify_zonelist();
Yasunori Goto68113782006-06-23 02:03:11 -07003120 cpuset_init_current_mems_allowed();
3121 } else {
Simon Arlott183ff222007-10-20 01:27:18 +02003122 /* we have to stop all cpus to guarantee there is no user
Yasunori Goto68113782006-06-23 02:03:11 -07003123 of zonelist */
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003124#ifdef CONFIG_MEMORY_HOTPLUG
3125 if (data)
3126 setup_zone_pageset((struct zone *)data);
3127#endif
3128 stop_machine(__build_all_zonelists, NULL, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07003129 /* cpuset refresh routine should be here */
3130 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07003131 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003132 /*
3133 * Disable grouping by mobility if the number of pages in the
3134 * system is too low to allow the mechanism to work. It would be
3135 * more accurate, but expensive to check per-zone. This check is
3136 * made on memory-hotadd so a system can start with mobility
3137 * disabled and enable it later
3138 */
Mel Gormand9c23402007-10-16 01:26:01 -07003139 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003140 page_group_by_mobility_disabled = 1;
3141 else
3142 page_group_by_mobility_disabled = 0;
3143
3144 printk("Built %i zonelists in %s order, mobility grouping %s. "
3145 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003146 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003147 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003148 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003149 vm_total_pages);
3150#ifdef CONFIG_NUMA
3151 printk("Policy zone: %s\n", zone_names[policy_zone]);
3152#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003153}
3154
3155/*
3156 * Helper functions to size the waitqueue hash table.
3157 * Essentially these want to choose hash table sizes sufficiently
3158 * large so that collisions trying to wait on pages are rare.
3159 * But in fact, the number of active page waitqueues on typical
3160 * systems is ridiculously low, less than 200. So this is even
3161 * conservative, even though it seems large.
3162 *
3163 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3164 * waitqueues, i.e. the size of the waitq table given the number of pages.
3165 */
3166#define PAGES_PER_WAITQUEUE 256
3167
Yasunori Gotocca448f2006-06-23 02:03:10 -07003168#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07003169static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003170{
3171 unsigned long size = 1;
3172
3173 pages /= PAGES_PER_WAITQUEUE;
3174
3175 while (size < pages)
3176 size <<= 1;
3177
3178 /*
3179 * Once we have dozens or even hundreds of threads sleeping
3180 * on IO we've got bigger problems than wait queue collision.
3181 * Limit the size of the wait table to a reasonable size.
3182 */
3183 size = min(size, 4096UL);
3184
3185 return max(size, 4UL);
3186}
Yasunori Gotocca448f2006-06-23 02:03:10 -07003187#else
3188/*
3189 * A zone's size might be changed by hot-add, so it is not possible to determine
3190 * a suitable size for its wait_table. So we use the maximum size now.
3191 *
3192 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3193 *
3194 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3195 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3196 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3197 *
3198 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3199 * or more by the traditional way. (See above). It equals:
3200 *
3201 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3202 * ia64(16K page size) : = ( 8G + 4M)byte.
3203 * powerpc (64K page size) : = (32G +16M)byte.
3204 */
3205static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3206{
3207 return 4096UL;
3208}
3209#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003210
3211/*
3212 * This is an integer logarithm so that shifts can be used later
3213 * to extract the more random high bits from the multiplicative
3214 * hash function before the remainder is taken.
3215 */
3216static inline unsigned long wait_table_bits(unsigned long size)
3217{
3218 return ffz(~size);
3219}
3220
3221#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3222
Mel Gorman56fd56b2007-10-16 01:25:58 -07003223/*
Mel Gormand9c23402007-10-16 01:26:01 -07003224 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07003225 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3226 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07003227 * higher will lead to a bigger reserve which will get freed as contiguous
3228 * blocks as reclaim kicks in
3229 */
3230static void setup_zone_migrate_reserve(struct zone *zone)
3231{
3232 unsigned long start_pfn, pfn, end_pfn;
3233 struct page *page;
Mel Gorman78986a62009-09-21 17:03:02 -07003234 unsigned long block_migratetype;
3235 int reserve;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003236
3237 /* Get the start pfn, end pfn and the number of blocks to reserve */
3238 start_pfn = zone->zone_start_pfn;
3239 end_pfn = start_pfn + zone->spanned_pages;
Mel Gorman41858962009-06-16 15:32:12 -07003240 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07003241 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003242
Mel Gorman78986a62009-09-21 17:03:02 -07003243 /*
3244 * Reserve blocks are generally in place to help high-order atomic
3245 * allocations that are short-lived. A min_free_kbytes value that
3246 * would result in more than 2 reserve blocks for atomic allocations
3247 * is assumed to be in place to help anti-fragmentation for the
3248 * future allocation of hugepages at runtime.
3249 */
3250 reserve = min(2, reserve);
3251
Mel Gormand9c23402007-10-16 01:26:01 -07003252 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07003253 if (!pfn_valid(pfn))
3254 continue;
3255 page = pfn_to_page(pfn);
3256
Adam Litke344c7902008-09-02 14:35:38 -07003257 /* Watch out for overlapping nodes */
3258 if (page_to_nid(page) != zone_to_nid(zone))
3259 continue;
3260
Mel Gorman56fd56b2007-10-16 01:25:58 -07003261 /* Blocks with reserved pages will never free, skip them. */
3262 if (PageReserved(page))
3263 continue;
3264
3265 block_migratetype = get_pageblock_migratetype(page);
3266
3267 /* If this block is reserved, account for it */
3268 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
3269 reserve--;
3270 continue;
3271 }
3272
3273 /* Suitable for reserving if this block is movable */
3274 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
3275 set_pageblock_migratetype(page, MIGRATE_RESERVE);
3276 move_freepages_block(zone, page, MIGRATE_RESERVE);
3277 reserve--;
3278 continue;
3279 }
3280
3281 /*
3282 * If the reserve is met and this is a previous reserved block,
3283 * take it back
3284 */
3285 if (block_migratetype == MIGRATE_RESERVE) {
3286 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
3287 move_freepages_block(zone, page, MIGRATE_MOVABLE);
3288 }
3289 }
3290}
Mel Gormanac0e5b72007-10-16 01:25:58 -07003291
Linus Torvalds1da177e2005-04-16 15:20:36 -07003292/*
3293 * Initially all pages are reserved - free ones are freed
3294 * up by free_all_bootmem() once the early boot process is
3295 * done. Non-atomic initialization, single-pass.
3296 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003297void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08003298 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003299{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003300 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07003301 unsigned long end_pfn = start_pfn + size;
3302 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003303 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003304
Hugh Dickins22b31ee2009-01-06 14:40:09 -08003305 if (highest_memmap_pfn < end_pfn - 1)
3306 highest_memmap_pfn = end_pfn - 1;
3307
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003308 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08003309 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08003310 /*
3311 * There can be holes in boot-time mem_map[]s
3312 * handed to this function. They do not
3313 * exist on hotplugged memory.
3314 */
3315 if (context == MEMMAP_EARLY) {
3316 if (!early_pfn_valid(pfn))
3317 continue;
3318 if (!early_pfn_in_nid(pfn, nid))
3319 continue;
3320 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003321 page = pfn_to_page(pfn);
3322 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07003323 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08003324 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003325 reset_page_mapcount(page);
3326 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003327 /*
3328 * Mark the block movable so that blocks are reserved for
3329 * movable at startup. This will force kernel allocations
3330 * to reserve their blocks rather than leaking throughout
3331 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07003332 * kernel allocations are made. Later some blocks near
3333 * the start are marked MIGRATE_RESERVE by
3334 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003335 *
3336 * bitmap is created for zone's valid pfn range. but memmap
3337 * can be created for invalid pages (for alignment)
3338 * check here not to call set_pageblock_migratetype() against
3339 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003340 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003341 if ((z->zone_start_pfn <= pfn)
3342 && (pfn < z->zone_start_pfn + z->spanned_pages)
3343 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07003344 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003345
Linus Torvalds1da177e2005-04-16 15:20:36 -07003346 INIT_LIST_HEAD(&page->lru);
3347#ifdef WANT_PAGE_VIRTUAL
3348 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3349 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07003350 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003351#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003352 }
3353}
3354
Andi Kleen1e548de2008-02-04 22:29:26 -08003355static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003356{
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003357 int order, t;
3358 for_each_migratetype_order(order, t) {
3359 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003360 zone->free_area[order].nr_free = 0;
3361 }
3362}
3363
3364#ifndef __HAVE_ARCH_MEMMAP_INIT
3365#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08003366 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003367#endif
3368
Sam Ravnborg1d6f4e62008-01-17 15:21:12 -08003369static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003370{
David Howells3a6be872009-05-06 16:03:03 -07003371#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003372 int batch;
3373
3374 /*
3375 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07003376 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003377 *
3378 * OK, so we don't know how big the cache is. So guess.
3379 */
3380 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07003381 if (batch * PAGE_SIZE > 512 * 1024)
3382 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003383 batch /= 4; /* We effectively *= 4 below */
3384 if (batch < 1)
3385 batch = 1;
3386
3387 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11003388 * Clamp the batch to a 2^n - 1 value. Having a power
3389 * of 2 value was found to be more likely to have
3390 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003391 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11003392 * For example if 2 tasks are alternately allocating
3393 * batches of pages, one task can end up with a lot
3394 * of pages of one half of the possible page colors
3395 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003396 */
David Howells91552032009-05-06 16:03:02 -07003397 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07003398
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003399 return batch;
David Howells3a6be872009-05-06 16:03:03 -07003400
3401#else
3402 /* The deferral and batching of frees should be suppressed under NOMMU
3403 * conditions.
3404 *
3405 * The problem is that NOMMU needs to be able to allocate large chunks
3406 * of contiguous memory as there's no hardware page translation to
3407 * assemble apparent contiguous memory from discontiguous pages.
3408 *
3409 * Queueing large contiguous runs of pages for batching, however,
3410 * causes the pages to actually be freed in smaller chunks. As there
3411 * can be a significant delay between the individual batches being
3412 * recycled, this leads to the once large chunks of space being
3413 * fragmented and becoming unavailable for high-order allocations.
3414 */
3415 return 0;
3416#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003417}
3418
Adrian Bunkb69a7282008-07-23 21:28:12 -07003419static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
Christoph Lameter2caaad42005-06-21 17:15:00 -07003420{
3421 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003422 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003423
Magnus Damm1c6fe942005-10-26 01:58:59 -07003424 memset(p, 0, sizeof(*p));
3425
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003426 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003427 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003428 pcp->high = 6 * batch;
3429 pcp->batch = max(1UL, 1 * batch);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003430 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
3431 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07003432}
3433
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003434/*
3435 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3436 * to the value high for the pageset p.
3437 */
3438
3439static void setup_pagelist_highmark(struct per_cpu_pageset *p,
3440 unsigned long high)
3441{
3442 struct per_cpu_pages *pcp;
3443
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003444 pcp = &p->pcp;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003445 pcp->high = high;
3446 pcp->batch = max(1UL, high/4);
3447 if ((high/4) > (PAGE_SHIFT * 8))
3448 pcp->batch = PAGE_SHIFT * 8;
3449}
3450
Wu Fengguang319774e2010-05-24 14:32:49 -07003451static __meminit void setup_zone_pageset(struct zone *zone)
3452{
3453 int cpu;
3454
3455 zone->pageset = alloc_percpu(struct per_cpu_pageset);
3456
3457 for_each_possible_cpu(cpu) {
3458 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
3459
3460 setup_pageset(pcp, zone_batchsize(zone));
3461
3462 if (percpu_pagelist_fraction)
3463 setup_pagelist_highmark(pcp,
3464 (zone->present_pages /
3465 percpu_pagelist_fraction));
3466 }
3467}
3468
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003469/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003470 * Allocate per cpu pagesets and initialize them.
3471 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07003472 */
Al Viro78d99552005-12-15 09:18:25 +00003473void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003474{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003475 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003476
Wu Fengguang319774e2010-05-24 14:32:49 -07003477 for_each_populated_zone(zone)
3478 setup_zone_pageset(zone);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003479}
3480
Sam Ravnborg577a32f2007-05-17 23:29:25 +02003481static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07003482int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07003483{
3484 int i;
3485 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003486 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07003487
3488 /*
3489 * The per-page waitqueue mechanism uses hashed waitqueues
3490 * per zone.
3491 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07003492 zone->wait_table_hash_nr_entries =
3493 wait_table_hash_nr_entries(zone_size_pages);
3494 zone->wait_table_bits =
3495 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003496 alloc_size = zone->wait_table_hash_nr_entries
3497 * sizeof(wait_queue_head_t);
3498
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07003499 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07003500 zone->wait_table = (wait_queue_head_t *)
3501 alloc_bootmem_node(pgdat, alloc_size);
3502 } else {
3503 /*
3504 * This case means that a zone whose size was 0 gets new memory
3505 * via memory hot-add.
3506 * But it may be the case that a new node was hot-added. In
3507 * this case vmalloc() will not be able to use this new node's
3508 * memory - this wait_table must be initialized to use this new
3509 * node itself as well.
3510 * To use this new node's memory, further consideration will be
3511 * necessary.
3512 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07003513 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003514 }
3515 if (!zone->wait_table)
3516 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07003517
Yasunori Goto02b694d2006-06-23 02:03:08 -07003518 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07003519 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003520
3521 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003522}
3523
Shaohua Li112067f2009-09-21 17:01:16 -07003524static int __zone_pcp_update(void *data)
3525{
3526 struct zone *zone = data;
3527 int cpu;
3528 unsigned long batch = zone_batchsize(zone), flags;
3529
Thomas Gleixner2d30a1f2010-03-10 15:20:40 -08003530 for_each_possible_cpu(cpu) {
Shaohua Li112067f2009-09-21 17:01:16 -07003531 struct per_cpu_pageset *pset;
3532 struct per_cpu_pages *pcp;
3533
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003534 pset = per_cpu_ptr(zone->pageset, cpu);
Shaohua Li112067f2009-09-21 17:01:16 -07003535 pcp = &pset->pcp;
3536
3537 local_irq_save(flags);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003538 free_pcppages_bulk(zone, pcp->count, pcp);
Shaohua Li112067f2009-09-21 17:01:16 -07003539 setup_pageset(pset, batch);
3540 local_irq_restore(flags);
3541 }
3542 return 0;
3543}
3544
3545void zone_pcp_update(struct zone *zone)
3546{
3547 stop_machine(__zone_pcp_update, zone, NULL);
3548}
3549
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003550static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07003551{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003552 /*
3553 * per cpu subsystem is not up at this point. The following code
3554 * relies on the ability of the linker to provide the
3555 * offset of a (static) per cpu variable into the per cpu area.
3556 */
3557 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07003558
Anton Blanchardf5335c02006-03-25 03:06:49 -08003559 if (zone->present_pages)
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003560 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
3561 zone->name, zone->present_pages,
3562 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07003563}
3564
Yasunori Goto718127c2006-06-23 02:03:10 -07003565__meminit int init_currently_empty_zone(struct zone *zone,
3566 unsigned long zone_start_pfn,
Dave Hansena2f3aa022007-01-10 23:15:30 -08003567 unsigned long size,
3568 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07003569{
3570 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003571 int ret;
3572 ret = zone_wait_table_init(zone, size);
3573 if (ret)
3574 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07003575 pgdat->nr_zones = zone_idx(zone) + 1;
3576
Dave Hansened8ece22005-10-29 18:16:50 -07003577 zone->zone_start_pfn = zone_start_pfn;
3578
Mel Gorman708614e2008-07-23 21:26:51 -07003579 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3580 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3581 pgdat->node_id,
3582 (unsigned long)zone_idx(zone),
3583 zone_start_pfn, (zone_start_pfn + size));
3584
Andi Kleen1e548de2008-02-04 22:29:26 -08003585 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07003586
3587 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003588}
3589
Mel Gormanc7132162006-09-27 01:49:43 -07003590#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3591/*
3592 * Basic iterator support. Return the first range of PFNs for a node
3593 * Note: nid == MAX_NUMNODES returns first region regardless of node
3594 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003595static int __meminit first_active_region_index_in_nid(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003596{
3597 int i;
3598
3599 for (i = 0; i < nr_nodemap_entries; i++)
3600 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3601 return i;
3602
3603 return -1;
3604}
3605
3606/*
3607 * Basic iterator support. Return the next active range of PFNs for a node
Simon Arlott183ff222007-10-20 01:27:18 +02003608 * Note: nid == MAX_NUMNODES returns next region regardless of node
Mel Gormanc7132162006-09-27 01:49:43 -07003609 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003610static int __meminit next_active_region_index_in_nid(int index, int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003611{
3612 for (index = index + 1; index < nr_nodemap_entries; index++)
3613 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3614 return index;
3615
3616 return -1;
3617}
3618
3619#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3620/*
3621 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3622 * Architectures may implement their own version but if add_active_range()
3623 * was used and there are no special requirements, this is a convenient
3624 * alternative
3625 */
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003626int __meminit __early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07003627{
3628 int i;
3629
3630 for (i = 0; i < nr_nodemap_entries; i++) {
3631 unsigned long start_pfn = early_node_map[i].start_pfn;
3632 unsigned long end_pfn = early_node_map[i].end_pfn;
3633
3634 if (start_pfn <= pfn && pfn < end_pfn)
3635 return early_node_map[i].nid;
3636 }
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003637 /* This is a memory hole */
3638 return -1;
Mel Gormanc7132162006-09-27 01:49:43 -07003639}
3640#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3641
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003642int __meminit early_pfn_to_nid(unsigned long pfn)
3643{
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003644 int nid;
3645
3646 nid = __early_pfn_to_nid(pfn);
3647 if (nid >= 0)
3648 return nid;
3649 /* just returns 0 */
3650 return 0;
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003651}
3652
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003653#ifdef CONFIG_NODES_SPAN_OTHER_NODES
3654bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3655{
3656 int nid;
3657
3658 nid = __early_pfn_to_nid(pfn);
3659 if (nid >= 0 && nid != node)
3660 return false;
3661 return true;
3662}
3663#endif
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003664
Mel Gormanc7132162006-09-27 01:49:43 -07003665/* Basic iterator support to walk early_node_map[] */
3666#define for_each_active_range_index_in_nid(i, nid) \
3667 for (i = first_active_region_index_in_nid(nid); i != -1; \
3668 i = next_active_region_index_in_nid(i, nid))
3669
3670/**
3671 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003672 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3673 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07003674 *
3675 * If an architecture guarantees that all ranges registered with
3676 * add_active_ranges() contain no holes and may be freed, this
3677 * this function may be used instead of calling free_bootmem() manually.
3678 */
3679void __init free_bootmem_with_active_regions(int nid,
3680 unsigned long max_low_pfn)
3681{
3682 int i;
3683
3684 for_each_active_range_index_in_nid(i, nid) {
3685 unsigned long size_pages = 0;
3686 unsigned long end_pfn = early_node_map[i].end_pfn;
3687
3688 if (early_node_map[i].start_pfn >= max_low_pfn)
3689 continue;
3690
3691 if (end_pfn > max_low_pfn)
3692 end_pfn = max_low_pfn;
3693
3694 size_pages = end_pfn - early_node_map[i].start_pfn;
3695 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3696 PFN_PHYS(early_node_map[i].start_pfn),
3697 size_pages << PAGE_SHIFT);
3698 }
3699}
3700
Yinghai Luedbe7d22010-08-25 13:39:16 -07003701#ifdef CONFIG_HAVE_MEMBLOCK
3702u64 __init find_memory_core_early(int nid, u64 size, u64 align,
3703 u64 goal, u64 limit)
3704{
3705 int i;
3706
3707 /* Need to go over early_node_map to find out good range for node */
3708 for_each_active_range_index_in_nid(i, nid) {
3709 u64 addr;
3710 u64 ei_start, ei_last;
3711 u64 final_start, final_end;
3712
3713 ei_last = early_node_map[i].end_pfn;
3714 ei_last <<= PAGE_SHIFT;
3715 ei_start = early_node_map[i].start_pfn;
3716 ei_start <<= PAGE_SHIFT;
3717
3718 final_start = max(ei_start, goal);
3719 final_end = min(ei_last, limit);
3720
3721 if (final_start >= final_end)
3722 continue;
3723
3724 addr = memblock_find_in_range(final_start, final_end, size, align);
3725
3726 if (addr == MEMBLOCK_ERROR)
3727 continue;
3728
3729 return addr;
3730 }
3731
3732 return MEMBLOCK_ERROR;
3733}
3734#endif
3735
Yinghai Lu08677212010-02-10 01:20:20 -08003736int __init add_from_early_node_map(struct range *range, int az,
3737 int nr_range, int nid)
3738{
3739 int i;
3740 u64 start, end;
3741
3742 /* need to go over early_node_map to find out good range for node */
3743 for_each_active_range_index_in_nid(i, nid) {
3744 start = early_node_map[i].start_pfn;
3745 end = early_node_map[i].end_pfn;
3746 nr_range = add_range(range, az, nr_range, start, end);
3747 }
3748 return nr_range;
3749}
3750
Yinghai Lu2ee78f72010-02-17 11:29:49 -08003751#ifdef CONFIG_NO_BOOTMEM
Yinghai Lu08677212010-02-10 01:20:20 -08003752void * __init __alloc_memory_core_early(int nid, u64 size, u64 align,
3753 u64 goal, u64 limit)
3754{
Yinghai Lu08677212010-02-10 01:20:20 -08003755 void *ptr;
Yinghai Lu72d7c3b2010-08-25 13:39:17 -07003756 u64 addr;
Yinghai Lu08677212010-02-10 01:20:20 -08003757
Yinghai Lu72d7c3b2010-08-25 13:39:17 -07003758 if (limit > memblock.current_limit)
3759 limit = memblock.current_limit;
Yinghai Lub8ab9f82010-07-20 13:24:31 -07003760
Yinghai Lu72d7c3b2010-08-25 13:39:17 -07003761 addr = find_memory_core_early(nid, size, align, goal, limit);
Yinghai Lu08677212010-02-10 01:20:20 -08003762
Yinghai Lu72d7c3b2010-08-25 13:39:17 -07003763 if (addr == MEMBLOCK_ERROR)
3764 return NULL;
Yinghai Lu08677212010-02-10 01:20:20 -08003765
Yinghai Lu72d7c3b2010-08-25 13:39:17 -07003766 ptr = phys_to_virt(addr);
3767 memset(ptr, 0, size);
3768 memblock_x86_reserve_range(addr, addr + size, "BOOTMEM");
3769 /*
3770 * The min_count is set to 0 so that bootmem allocated blocks
3771 * are never reported as leaks.
3772 */
3773 kmemleak_alloc(ptr, size, 0, 0);
3774 return ptr;
Yinghai Lu08677212010-02-10 01:20:20 -08003775}
Yinghai Lu2ee78f72010-02-17 11:29:49 -08003776#endif
Yinghai Lu08677212010-02-10 01:20:20 -08003777
3778
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003779void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3780{
3781 int i;
Yinghai Lud52d53b2008-06-16 20:10:55 -07003782 int ret;
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003783
Yinghai Lud52d53b2008-06-16 20:10:55 -07003784 for_each_active_range_index_in_nid(i, nid) {
3785 ret = work_fn(early_node_map[i].start_pfn,
3786 early_node_map[i].end_pfn, data);
3787 if (ret)
3788 break;
3789 }
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003790}
Mel Gormanc7132162006-09-27 01:49:43 -07003791/**
3792 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003793 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07003794 *
3795 * If an architecture guarantees that all ranges registered with
3796 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003797 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07003798 */
3799void __init sparse_memory_present_with_active_regions(int nid)
3800{
3801 int i;
3802
3803 for_each_active_range_index_in_nid(i, nid)
3804 memory_present(early_node_map[i].nid,
3805 early_node_map[i].start_pfn,
3806 early_node_map[i].end_pfn);
3807}
3808
3809/**
3810 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003811 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3812 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3813 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07003814 *
3815 * It returns the start and end page frame of a node based on information
3816 * provided by an arch calling add_active_range(). If called for a node
3817 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003818 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07003819 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003820void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003821 unsigned long *start_pfn, unsigned long *end_pfn)
3822{
3823 int i;
3824 *start_pfn = -1UL;
3825 *end_pfn = 0;
3826
3827 for_each_active_range_index_in_nid(i, nid) {
3828 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3829 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3830 }
3831
Christoph Lameter633c0662007-10-16 01:25:37 -07003832 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07003833 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07003834}
3835
3836/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07003837 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3838 * assumption is made that zones within a node are ordered in monotonic
3839 * increasing memory addresses so that the "highest" populated zone is used
3840 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003841static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07003842{
3843 int zone_index;
3844 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3845 if (zone_index == ZONE_MOVABLE)
3846 continue;
3847
3848 if (arch_zone_highest_possible_pfn[zone_index] >
3849 arch_zone_lowest_possible_pfn[zone_index])
3850 break;
3851 }
3852
3853 VM_BUG_ON(zone_index == -1);
3854 movable_zone = zone_index;
3855}
3856
3857/*
3858 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3859 * because it is sized independant of architecture. Unlike the other zones,
3860 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3861 * in each node depending on the size of each node and how evenly kernelcore
3862 * is distributed. This helper function adjusts the zone ranges
3863 * provided by the architecture for a given node by using the end of the
3864 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3865 * zones within a node are in order of monotonic increases memory addresses
3866 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003867static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07003868 unsigned long zone_type,
3869 unsigned long node_start_pfn,
3870 unsigned long node_end_pfn,
3871 unsigned long *zone_start_pfn,
3872 unsigned long *zone_end_pfn)
3873{
3874 /* Only adjust if ZONE_MOVABLE is on this node */
3875 if (zone_movable_pfn[nid]) {
3876 /* Size ZONE_MOVABLE */
3877 if (zone_type == ZONE_MOVABLE) {
3878 *zone_start_pfn = zone_movable_pfn[nid];
3879 *zone_end_pfn = min(node_end_pfn,
3880 arch_zone_highest_possible_pfn[movable_zone]);
3881
3882 /* Adjust for ZONE_MOVABLE starting within this range */
3883 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3884 *zone_end_pfn > zone_movable_pfn[nid]) {
3885 *zone_end_pfn = zone_movable_pfn[nid];
3886
3887 /* Check if this whole range is within ZONE_MOVABLE */
3888 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3889 *zone_start_pfn = *zone_end_pfn;
3890 }
3891}
3892
3893/*
Mel Gormanc7132162006-09-27 01:49:43 -07003894 * Return the number of pages a zone spans in a node, including holes
3895 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3896 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003897static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003898 unsigned long zone_type,
3899 unsigned long *ignored)
3900{
3901 unsigned long node_start_pfn, node_end_pfn;
3902 unsigned long zone_start_pfn, zone_end_pfn;
3903
3904 /* Get the start and end of the node and zone */
3905 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3906 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3907 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07003908 adjust_zone_range_for_zone_movable(nid, zone_type,
3909 node_start_pfn, node_end_pfn,
3910 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003911
3912 /* Check that this node has pages within the zone's required range */
3913 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3914 return 0;
3915
3916 /* Move the zone boundaries inside the node if necessary */
3917 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3918 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3919
3920 /* Return the spanned pages */
3921 return zone_end_pfn - zone_start_pfn;
3922}
3923
3924/*
3925 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003926 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07003927 */
Yinghai Lu32996252009-12-15 17:59:02 -08003928unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003929 unsigned long range_start_pfn,
3930 unsigned long range_end_pfn)
3931{
3932 int i = 0;
3933 unsigned long prev_end_pfn = 0, hole_pages = 0;
3934 unsigned long start_pfn;
3935
3936 /* Find the end_pfn of the first active range of pfns in the node */
3937 i = first_active_region_index_in_nid(nid);
3938 if (i == -1)
3939 return 0;
3940
Mel Gormanb5445f92007-07-26 10:41:18 -07003941 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3942
Mel Gorman9c7cd682006-09-27 01:49:58 -07003943 /* Account for ranges before physical memory on this node */
3944 if (early_node_map[i].start_pfn > range_start_pfn)
Mel Gormanb5445f92007-07-26 10:41:18 -07003945 hole_pages = prev_end_pfn - range_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003946
3947 /* Find all holes for the zone within the node */
3948 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3949
3950 /* No need to continue if prev_end_pfn is outside the zone */
3951 if (prev_end_pfn >= range_end_pfn)
3952 break;
3953
3954 /* Make sure the end of the zone is not within the hole */
3955 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3956 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3957
3958 /* Update the hole size cound and move on */
3959 if (start_pfn > range_start_pfn) {
3960 BUG_ON(prev_end_pfn > start_pfn);
3961 hole_pages += start_pfn - prev_end_pfn;
3962 }
3963 prev_end_pfn = early_node_map[i].end_pfn;
3964 }
3965
Mel Gorman9c7cd682006-09-27 01:49:58 -07003966 /* Account for ranges past physical memory on this node */
3967 if (range_end_pfn > prev_end_pfn)
Mel Gorman0c6cb972006-10-28 10:38:59 -07003968 hole_pages += range_end_pfn -
Mel Gorman9c7cd682006-09-27 01:49:58 -07003969 max(range_start_pfn, prev_end_pfn);
3970
Mel Gormanc7132162006-09-27 01:49:43 -07003971 return hole_pages;
3972}
3973
3974/**
3975 * absent_pages_in_range - Return number of page frames in holes within a range
3976 * @start_pfn: The start PFN to start searching for holes
3977 * @end_pfn: The end PFN to stop searching for holes
3978 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003979 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07003980 */
3981unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3982 unsigned long end_pfn)
3983{
3984 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3985}
3986
3987/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003988static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003989 unsigned long zone_type,
3990 unsigned long *ignored)
3991{
Mel Gorman9c7cd682006-09-27 01:49:58 -07003992 unsigned long node_start_pfn, node_end_pfn;
3993 unsigned long zone_start_pfn, zone_end_pfn;
3994
3995 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3996 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3997 node_start_pfn);
3998 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3999 node_end_pfn);
4000
Mel Gorman2a1e2742007-07-17 04:03:12 -07004001 adjust_zone_range_for_zone_movable(nid, zone_type,
4002 node_start_pfn, node_end_pfn,
4003 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004004 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004005}
Mel Gorman0e0b8642006-09-27 01:49:56 -07004006
Mel Gormanc7132162006-09-27 01:49:43 -07004007#else
Paul Mundt6ea6e682007-07-15 23:38:20 -07004008static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004009 unsigned long zone_type,
4010 unsigned long *zones_size)
4011{
4012 return zones_size[zone_type];
4013}
4014
Paul Mundt6ea6e682007-07-15 23:38:20 -07004015static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004016 unsigned long zone_type,
4017 unsigned long *zholes_size)
4018{
4019 if (!zholes_size)
4020 return 0;
4021
4022 return zholes_size[zone_type];
4023}
Mel Gorman0e0b8642006-09-27 01:49:56 -07004024
Mel Gormanc7132162006-09-27 01:49:43 -07004025#endif
4026
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004027static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Mel Gormanc7132162006-09-27 01:49:43 -07004028 unsigned long *zones_size, unsigned long *zholes_size)
4029{
4030 unsigned long realtotalpages, totalpages = 0;
4031 enum zone_type i;
4032
4033 for (i = 0; i < MAX_NR_ZONES; i++)
4034 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
4035 zones_size);
4036 pgdat->node_spanned_pages = totalpages;
4037
4038 realtotalpages = totalpages;
4039 for (i = 0; i < MAX_NR_ZONES; i++)
4040 realtotalpages -=
4041 zone_absent_pages_in_node(pgdat->node_id, i,
4042 zholes_size);
4043 pgdat->node_present_pages = realtotalpages;
4044 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4045 realtotalpages);
4046}
4047
Mel Gorman835c1342007-10-16 01:25:47 -07004048#ifndef CONFIG_SPARSEMEM
4049/*
4050 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07004051 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4052 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07004053 * round what is now in bits to nearest long in bits, then return it in
4054 * bytes.
4055 */
4056static unsigned long __init usemap_size(unsigned long zonesize)
4057{
4058 unsigned long usemapsize;
4059
Mel Gormand9c23402007-10-16 01:26:01 -07004060 usemapsize = roundup(zonesize, pageblock_nr_pages);
4061 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07004062 usemapsize *= NR_PAGEBLOCK_BITS;
4063 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4064
4065 return usemapsize / 8;
4066}
4067
4068static void __init setup_usemap(struct pglist_data *pgdat,
4069 struct zone *zone, unsigned long zonesize)
4070{
4071 unsigned long usemapsize = usemap_size(zonesize);
4072 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08004073 if (usemapsize)
Mel Gorman835c1342007-10-16 01:25:47 -07004074 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
Mel Gorman835c1342007-10-16 01:25:47 -07004075}
4076#else
Jesper Juhlfa9f90b2010-11-28 21:39:34 +01004077static inline void setup_usemap(struct pglist_data *pgdat,
Mel Gorman835c1342007-10-16 01:25:47 -07004078 struct zone *zone, unsigned long zonesize) {}
4079#endif /* CONFIG_SPARSEMEM */
4080
Mel Gormand9c23402007-10-16 01:26:01 -07004081#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08004082
4083/* Return a sensible default order for the pageblock size. */
4084static inline int pageblock_default_order(void)
4085{
4086 if (HPAGE_SHIFT > PAGE_SHIFT)
4087 return HUGETLB_PAGE_ORDER;
4088
4089 return MAX_ORDER-1;
4090}
4091
Mel Gormand9c23402007-10-16 01:26:01 -07004092/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4093static inline void __init set_pageblock_order(unsigned int order)
4094{
4095 /* Check that pageblock_nr_pages has not already been setup */
4096 if (pageblock_order)
4097 return;
4098
4099 /*
4100 * Assume the largest contiguous order of interest is a huge page.
4101 * This value may be variable depending on boot parameters on IA64
4102 */
4103 pageblock_order = order;
4104}
4105#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4106
Mel Gormanba72cb82007-11-28 16:21:13 -08004107/*
4108 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4109 * and pageblock_default_order() are unused as pageblock_order is set
4110 * at compile-time. See include/linux/pageblock-flags.h for the values of
4111 * pageblock_order based on the kernel config
4112 */
4113static inline int pageblock_default_order(unsigned int order)
4114{
4115 return MAX_ORDER-1;
4116}
Mel Gormand9c23402007-10-16 01:26:01 -07004117#define set_pageblock_order(x) do {} while (0)
4118
4119#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4120
Linus Torvalds1da177e2005-04-16 15:20:36 -07004121/*
4122 * Set up the zone data structures:
4123 * - mark all pages reserved
4124 * - mark all memory queues empty
4125 * - clear the memory bitmaps
4126 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08004127static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004128 unsigned long *zones_size, unsigned long *zholes_size)
4129{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07004130 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07004131 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004132 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07004133 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004134
Dave Hansen208d54e2005-10-29 18:16:52 -07004135 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004136 pgdat->nr_zones = 0;
4137 init_waitqueue_head(&pgdat->kswapd_wait);
4138 pgdat->kswapd_max_order = 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004139 pgdat_page_cgroup_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004140
4141 for (j = 0; j < MAX_NR_ZONES; j++) {
4142 struct zone *zone = pgdat->node_zones + j;
Mel Gorman0e0b8642006-09-27 01:49:56 -07004143 unsigned long size, realsize, memmap_pages;
Christoph Lameterb69408e2008-10-18 20:26:14 -07004144 enum lru_list l;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004145
Mel Gormanc7132162006-09-27 01:49:43 -07004146 size = zone_spanned_pages_in_node(nid, j, zones_size);
4147 realsize = size - zone_absent_pages_in_node(nid, j,
4148 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004149
Mel Gorman0e0b8642006-09-27 01:49:56 -07004150 /*
4151 * Adjust realsize so that it accounts for how much memory
4152 * is used by this zone for memmap. This affects the watermark
4153 * and per-cpu initialisations
4154 */
Johannes Weinerf7232152008-05-23 13:04:21 -07004155 memmap_pages =
4156 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
Mel Gorman0e0b8642006-09-27 01:49:56 -07004157 if (realsize >= memmap_pages) {
4158 realsize -= memmap_pages;
Yinghai Lu5594c8c2009-01-06 14:39:14 -08004159 if (memmap_pages)
4160 printk(KERN_DEBUG
4161 " %s zone: %lu pages used for memmap\n",
4162 zone_names[j], memmap_pages);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004163 } else
4164 printk(KERN_WARNING
4165 " %s zone: %lu pages exceeds realsize %lu\n",
4166 zone_names[j], memmap_pages, realsize);
4167
Christoph Lameter62672762007-02-10 01:43:07 -08004168 /* Account for reserved pages */
4169 if (j == 0 && realsize > dma_reserve) {
Mel Gorman0e0b8642006-09-27 01:49:56 -07004170 realsize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07004171 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08004172 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004173 }
4174
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004175 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004176 nr_kernel_pages += realsize;
4177 nr_all_pages += realsize;
4178
4179 zone->spanned_pages = size;
4180 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07004181#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07004182 zone->node = nid;
Christoph Lameter8417bba2006-09-25 23:31:51 -07004183 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07004184 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07004185 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07004186#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004187 zone->name = zone_names[j];
4188 spin_lock_init(&zone->lock);
4189 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07004190 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004191 zone->zone_pgdat = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004192
Dave Hansened8ece22005-10-29 18:16:50 -07004193 zone_pcp_init(zone);
Christoph Lameterb69408e2008-10-18 20:26:14 -07004194 for_each_lru(l) {
4195 INIT_LIST_HEAD(&zone->lru[l].list);
Wu Fengguangf8629632009-09-21 17:03:11 -07004196 zone->reclaim_stat.nr_saved_scan[l] = 0;
Christoph Lameterb69408e2008-10-18 20:26:14 -07004197 }
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08004198 zone->reclaim_stat.recent_rotated[0] = 0;
4199 zone->reclaim_stat.recent_rotated[1] = 0;
4200 zone->reclaim_stat.recent_scanned[0] = 0;
4201 zone->reclaim_stat.recent_scanned[1] = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07004202 zap_zone_vm_stats(zone);
David Rientjese815af92007-10-16 23:25:54 -07004203 zone->flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004204 if (!size)
4205 continue;
4206
Mel Gormanba72cb82007-11-28 16:21:13 -08004207 set_pageblock_order(pageblock_default_order());
Mel Gorman835c1342007-10-16 01:25:47 -07004208 setup_usemap(pgdat, zone, size);
Dave Hansena2f3aa022007-01-10 23:15:30 -08004209 ret = init_currently_empty_zone(zone, zone_start_pfn,
4210 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07004211 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07004212 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004213 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004214 }
4215}
4216
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004217static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004218{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004219 /* Skip empty nodes */
4220 if (!pgdat->node_spanned_pages)
4221 return;
4222
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004223#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07004224 /* ia64 gets its own node_mem_map, before this, without bootmem */
4225 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07004226 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004227 struct page *map;
4228
Bob Piccoe984bb42006-05-20 15:00:31 -07004229 /*
4230 * The zone's endpoints aren't required to be MAX_ORDER
4231 * aligned but the node_mem_map endpoints must be in order
4232 * for the buddy allocator to function correctly.
4233 */
4234 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
4235 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
4236 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4237 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07004238 map = alloc_remap(pgdat->node_id, size);
4239 if (!map)
4240 map = alloc_bootmem_node(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07004241 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004242 }
Roman Zippel12d810c2007-05-31 00:40:54 -07004243#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07004244 /*
4245 * With no DISCONTIG, the global mem_map is just set as node 0's
4246 */
Mel Gormanc7132162006-09-27 01:49:43 -07004247 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004248 mem_map = NODE_DATA(0)->node_mem_map;
Mel Gormanc7132162006-09-27 01:49:43 -07004249#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
4250 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08004251 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Mel Gormanc7132162006-09-27 01:49:43 -07004252#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4253 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004254#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004255#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004256}
4257
Johannes Weiner9109fb72008-07-23 21:27:20 -07004258void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4259 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004260{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004261 pg_data_t *pgdat = NODE_DATA(nid);
4262
Linus Torvalds1da177e2005-04-16 15:20:36 -07004263 pgdat->node_id = nid;
4264 pgdat->node_start_pfn = node_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004265 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004266
4267 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07004268#ifdef CONFIG_FLAT_NODE_MEM_MAP
4269 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4270 nid, (unsigned long)pgdat,
4271 (unsigned long)pgdat->node_mem_map);
4272#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004273
4274 free_area_init_core(pgdat, zones_size, zholes_size);
4275}
4276
Mel Gormanc7132162006-09-27 01:49:43 -07004277#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07004278
4279#if MAX_NUMNODES > 1
4280/*
4281 * Figure out the number of possible node ids.
4282 */
4283static void __init setup_nr_node_ids(void)
4284{
4285 unsigned int node;
4286 unsigned int highest = 0;
4287
4288 for_each_node_mask(node, node_possible_map)
4289 highest = node;
4290 nr_node_ids = highest + 1;
4291}
4292#else
4293static inline void setup_nr_node_ids(void)
4294{
4295}
4296#endif
4297
Mel Gormanc7132162006-09-27 01:49:43 -07004298/**
4299 * add_active_range - Register a range of PFNs backed by physical memory
4300 * @nid: The node ID the range resides on
4301 * @start_pfn: The start PFN of the available physical memory
4302 * @end_pfn: The end PFN of the available physical memory
4303 *
4304 * These ranges are stored in an early_node_map[] and later used by
4305 * free_area_init_nodes() to calculate zone sizes and holes. If the
4306 * range spans a memory hole, it is up to the architecture to ensure
4307 * the memory is not freed by the bootmem allocator. If possible
4308 * the range being registered will be merged with existing ranges.
4309 */
4310void __init add_active_range(unsigned int nid, unsigned long start_pfn,
4311 unsigned long end_pfn)
4312{
4313 int i;
4314
Mel Gorman6b74ab92008-07-23 21:26:49 -07004315 mminit_dprintk(MMINIT_TRACE, "memory_register",
4316 "Entering add_active_range(%d, %#lx, %#lx) "
4317 "%d entries of %d used\n",
4318 nid, start_pfn, end_pfn,
4319 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
Mel Gormanc7132162006-09-27 01:49:43 -07004320
Mel Gorman2dbb51c2008-07-23 21:26:52 -07004321 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
4322
Mel Gormanc7132162006-09-27 01:49:43 -07004323 /* Merge with existing active regions if possible */
4324 for (i = 0; i < nr_nodemap_entries; i++) {
4325 if (early_node_map[i].nid != nid)
4326 continue;
4327
4328 /* Skip if an existing region covers this new one */
4329 if (start_pfn >= early_node_map[i].start_pfn &&
4330 end_pfn <= early_node_map[i].end_pfn)
4331 return;
4332
4333 /* Merge forward if suitable */
4334 if (start_pfn <= early_node_map[i].end_pfn &&
4335 end_pfn > early_node_map[i].end_pfn) {
4336 early_node_map[i].end_pfn = end_pfn;
4337 return;
4338 }
4339
4340 /* Merge backward if suitable */
Kazuhisa Ichikawad2dbe082010-01-15 17:01:20 -08004341 if (start_pfn < early_node_map[i].start_pfn &&
Mel Gormanc7132162006-09-27 01:49:43 -07004342 end_pfn >= early_node_map[i].start_pfn) {
4343 early_node_map[i].start_pfn = start_pfn;
4344 return;
4345 }
4346 }
4347
4348 /* Check that early_node_map is large enough */
4349 if (i >= MAX_ACTIVE_REGIONS) {
4350 printk(KERN_CRIT "More than %d memory regions, truncating\n",
4351 MAX_ACTIVE_REGIONS);
4352 return;
4353 }
4354
4355 early_node_map[i].nid = nid;
4356 early_node_map[i].start_pfn = start_pfn;
4357 early_node_map[i].end_pfn = end_pfn;
4358 nr_nodemap_entries = i + 1;
4359}
4360
4361/**
Yinghai Lucc1050b2008-06-13 19:08:52 -07004362 * remove_active_range - Shrink an existing registered range of PFNs
Mel Gormanc7132162006-09-27 01:49:43 -07004363 * @nid: The node id the range is on that should be shrunk
Yinghai Lucc1050b2008-06-13 19:08:52 -07004364 * @start_pfn: The new PFN of the range
4365 * @end_pfn: The new PFN of the range
Mel Gormanc7132162006-09-27 01:49:43 -07004366 *
4367 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004368 * The map is kept near the end physical page range that has already been
4369 * registered. This function allows an arch to shrink an existing registered
4370 * range.
Mel Gormanc7132162006-09-27 01:49:43 -07004371 */
Yinghai Lucc1050b2008-06-13 19:08:52 -07004372void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
4373 unsigned long end_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004374{
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004375 int i, j;
4376 int removed = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07004377
Yinghai Lucc1050b2008-06-13 19:08:52 -07004378 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
4379 nid, start_pfn, end_pfn);
4380
Mel Gormanc7132162006-09-27 01:49:43 -07004381 /* Find the old active region end and shrink */
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004382 for_each_active_range_index_in_nid(i, nid) {
Yinghai Lucc1050b2008-06-13 19:08:52 -07004383 if (early_node_map[i].start_pfn >= start_pfn &&
4384 early_node_map[i].end_pfn <= end_pfn) {
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004385 /* clear it */
Yinghai Lucc1050b2008-06-13 19:08:52 -07004386 early_node_map[i].start_pfn = 0;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004387 early_node_map[i].end_pfn = 0;
4388 removed = 1;
4389 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004390 }
Yinghai Lucc1050b2008-06-13 19:08:52 -07004391 if (early_node_map[i].start_pfn < start_pfn &&
4392 early_node_map[i].end_pfn > start_pfn) {
4393 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
4394 early_node_map[i].end_pfn = start_pfn;
4395 if (temp_end_pfn > end_pfn)
4396 add_active_range(nid, end_pfn, temp_end_pfn);
4397 continue;
4398 }
4399 if (early_node_map[i].start_pfn >= start_pfn &&
4400 early_node_map[i].end_pfn > end_pfn &&
4401 early_node_map[i].start_pfn < end_pfn) {
4402 early_node_map[i].start_pfn = end_pfn;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004403 continue;
4404 }
4405 }
4406
4407 if (!removed)
4408 return;
4409
4410 /* remove the blank ones */
4411 for (i = nr_nodemap_entries - 1; i > 0; i--) {
4412 if (early_node_map[i].nid != nid)
4413 continue;
4414 if (early_node_map[i].end_pfn)
4415 continue;
4416 /* we found it, get rid of it */
4417 for (j = i; j < nr_nodemap_entries - 1; j++)
4418 memcpy(&early_node_map[j], &early_node_map[j+1],
4419 sizeof(early_node_map[j]));
4420 j = nr_nodemap_entries - 1;
4421 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
4422 nr_nodemap_entries--;
4423 }
Mel Gormanc7132162006-09-27 01:49:43 -07004424}
4425
4426/**
4427 * remove_all_active_ranges - Remove all currently registered regions
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004428 *
Mel Gormanc7132162006-09-27 01:49:43 -07004429 * During discovery, it may be found that a table like SRAT is invalid
4430 * and an alternative discovery method must be used. This function removes
4431 * all currently registered regions.
4432 */
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004433void __init remove_all_active_ranges(void)
Mel Gormanc7132162006-09-27 01:49:43 -07004434{
4435 memset(early_node_map, 0, sizeof(early_node_map));
4436 nr_nodemap_entries = 0;
4437}
4438
4439/* Compare two active node_active_regions */
4440static int __init cmp_node_active_region(const void *a, const void *b)
4441{
4442 struct node_active_region *arange = (struct node_active_region *)a;
4443 struct node_active_region *brange = (struct node_active_region *)b;
4444
4445 /* Done this way to avoid overflows */
4446 if (arange->start_pfn > brange->start_pfn)
4447 return 1;
4448 if (arange->start_pfn < brange->start_pfn)
4449 return -1;
4450
4451 return 0;
4452}
4453
4454/* sort the node_map by start_pfn */
Yinghai Lu32996252009-12-15 17:59:02 -08004455void __init sort_node_map(void)
Mel Gormanc7132162006-09-27 01:49:43 -07004456{
4457 sort(early_node_map, (size_t)nr_nodemap_entries,
4458 sizeof(struct node_active_region),
4459 cmp_node_active_region, NULL);
4460}
4461
Mel Gormana6af2bc2007-02-10 01:42:57 -08004462/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004463static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07004464{
4465 int i;
Mel Gormana6af2bc2007-02-10 01:42:57 -08004466 unsigned long min_pfn = ULONG_MAX;
Mel Gorman1abbfb42006-11-23 12:01:41 +00004467
Mel Gormanc7132162006-09-27 01:49:43 -07004468 /* Assuming a sorted map, the first range found has the starting pfn */
4469 for_each_active_range_index_in_nid(i, nid)
Mel Gormana6af2bc2007-02-10 01:42:57 -08004470 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004471
Mel Gormana6af2bc2007-02-10 01:42:57 -08004472 if (min_pfn == ULONG_MAX) {
4473 printk(KERN_WARNING
Paul Jackson2bc0d262008-06-22 07:22:17 -07004474 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08004475 return 0;
4476 }
4477
4478 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004479}
4480
4481/**
4482 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4483 *
4484 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004485 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07004486 */
4487unsigned long __init find_min_pfn_with_active_regions(void)
4488{
4489 return find_min_pfn_for_node(MAX_NUMNODES);
4490}
4491
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004492/*
4493 * early_calculate_totalpages()
4494 * Sum pages in active regions for movable zone.
4495 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4496 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07004497static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07004498{
4499 int i;
4500 unsigned long totalpages = 0;
4501
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004502 for (i = 0; i < nr_nodemap_entries; i++) {
4503 unsigned long pages = early_node_map[i].end_pfn -
Mel Gorman7e63efe2007-07-17 04:03:15 -07004504 early_node_map[i].start_pfn;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004505 totalpages += pages;
4506 if (pages)
4507 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
4508 }
4509 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07004510}
4511
Mel Gorman2a1e2742007-07-17 04:03:12 -07004512/*
4513 * Find the PFN the Movable zone begins in each node. Kernel memory
4514 * is spread evenly between nodes as long as the nodes have enough
4515 * memory. When they don't, some nodes will have more kernelcore than
4516 * others
4517 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004518static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004519{
4520 int i, nid;
4521 unsigned long usable_startpfn;
4522 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07004523 /* save the state before borrow the nodemask */
4524 nodemask_t saved_node_state = node_states[N_HIGH_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004525 unsigned long totalpages = early_calculate_totalpages();
4526 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004527
Mel Gorman7e63efe2007-07-17 04:03:15 -07004528 /*
4529 * If movablecore was specified, calculate what size of
4530 * kernelcore that corresponds so that memory usable for
4531 * any allocation type is evenly spread. If both kernelcore
4532 * and movablecore are specified, then the value of kernelcore
4533 * will be used for required_kernelcore if it's greater than
4534 * what movablecore would have allowed.
4535 */
4536 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07004537 unsigned long corepages;
4538
4539 /*
4540 * Round-up so that ZONE_MOVABLE is at least as large as what
4541 * was requested by the user
4542 */
4543 required_movablecore =
4544 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
4545 corepages = totalpages - required_movablecore;
4546
4547 required_kernelcore = max(required_kernelcore, corepages);
4548 }
4549
Mel Gorman2a1e2742007-07-17 04:03:12 -07004550 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4551 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07004552 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004553
4554 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4555 find_usable_zone_for_movable();
4556 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4557
4558restart:
4559 /* Spread kernelcore memory as evenly as possible throughout nodes */
4560 kernelcore_node = required_kernelcore / usable_nodes;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004561 for_each_node_state(nid, N_HIGH_MEMORY) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004562 /*
4563 * Recalculate kernelcore_node if the division per node
4564 * now exceeds what is necessary to satisfy the requested
4565 * amount of memory for the kernel
4566 */
4567 if (required_kernelcore < kernelcore_node)
4568 kernelcore_node = required_kernelcore / usable_nodes;
4569
4570 /*
4571 * As the map is walked, we track how much memory is usable
4572 * by the kernel using kernelcore_remaining. When it is
4573 * 0, the rest of the node is usable by ZONE_MOVABLE
4574 */
4575 kernelcore_remaining = kernelcore_node;
4576
4577 /* Go through each range of PFNs within this node */
4578 for_each_active_range_index_in_nid(i, nid) {
4579 unsigned long start_pfn, end_pfn;
4580 unsigned long size_pages;
4581
4582 start_pfn = max(early_node_map[i].start_pfn,
4583 zone_movable_pfn[nid]);
4584 end_pfn = early_node_map[i].end_pfn;
4585 if (start_pfn >= end_pfn)
4586 continue;
4587
4588 /* Account for what is only usable for kernelcore */
4589 if (start_pfn < usable_startpfn) {
4590 unsigned long kernel_pages;
4591 kernel_pages = min(end_pfn, usable_startpfn)
4592 - start_pfn;
4593
4594 kernelcore_remaining -= min(kernel_pages,
4595 kernelcore_remaining);
4596 required_kernelcore -= min(kernel_pages,
4597 required_kernelcore);
4598
4599 /* Continue if range is now fully accounted */
4600 if (end_pfn <= usable_startpfn) {
4601
4602 /*
4603 * Push zone_movable_pfn to the end so
4604 * that if we have to rebalance
4605 * kernelcore across nodes, we will
4606 * not double account here
4607 */
4608 zone_movable_pfn[nid] = end_pfn;
4609 continue;
4610 }
4611 start_pfn = usable_startpfn;
4612 }
4613
4614 /*
4615 * The usable PFN range for ZONE_MOVABLE is from
4616 * start_pfn->end_pfn. Calculate size_pages as the
4617 * number of pages used as kernelcore
4618 */
4619 size_pages = end_pfn - start_pfn;
4620 if (size_pages > kernelcore_remaining)
4621 size_pages = kernelcore_remaining;
4622 zone_movable_pfn[nid] = start_pfn + size_pages;
4623
4624 /*
4625 * Some kernelcore has been met, update counts and
4626 * break if the kernelcore for this node has been
4627 * satisified
4628 */
4629 required_kernelcore -= min(required_kernelcore,
4630 size_pages);
4631 kernelcore_remaining -= size_pages;
4632 if (!kernelcore_remaining)
4633 break;
4634 }
4635 }
4636
4637 /*
4638 * If there is still required_kernelcore, we do another pass with one
4639 * less node in the count. This will push zone_movable_pfn[nid] further
4640 * along on the nodes that still have memory until kernelcore is
4641 * satisified
4642 */
4643 usable_nodes--;
4644 if (usable_nodes && required_kernelcore > usable_nodes)
4645 goto restart;
4646
4647 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4648 for (nid = 0; nid < MAX_NUMNODES; nid++)
4649 zone_movable_pfn[nid] =
4650 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07004651
4652out:
4653 /* restore the node_state */
4654 node_states[N_HIGH_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004655}
4656
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004657/* Any regular memory on that node ? */
4658static void check_for_regular_memory(pg_data_t *pgdat)
4659{
4660#ifdef CONFIG_HIGHMEM
4661 enum zone_type zone_type;
4662
4663 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4664 struct zone *zone = &pgdat->node_zones[zone_type];
4665 if (zone->present_pages)
4666 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4667 }
4668#endif
4669}
4670
Mel Gormanc7132162006-09-27 01:49:43 -07004671/**
4672 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004673 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07004674 *
4675 * This will call free_area_init_node() for each active node in the system.
4676 * Using the page ranges provided by add_active_range(), the size of each
4677 * zone in each node and their holes is calculated. If the maximum PFN
4678 * between two adjacent zones match, it is assumed that the zone is empty.
4679 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4680 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4681 * starts where the previous one ended. For example, ZONE_DMA32 starts
4682 * at arch_max_dma_pfn.
4683 */
4684void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4685{
4686 unsigned long nid;
Andrew Mortondb991002008-10-15 22:01:07 -07004687 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004688
Mel Gormana6af2bc2007-02-10 01:42:57 -08004689 /* Sort early_node_map as initialisation assumes it is sorted */
4690 sort_node_map();
4691
Mel Gormanc7132162006-09-27 01:49:43 -07004692 /* Record where the zone boundaries are */
4693 memset(arch_zone_lowest_possible_pfn, 0,
4694 sizeof(arch_zone_lowest_possible_pfn));
4695 memset(arch_zone_highest_possible_pfn, 0,
4696 sizeof(arch_zone_highest_possible_pfn));
4697 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4698 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4699 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004700 if (i == ZONE_MOVABLE)
4701 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004702 arch_zone_lowest_possible_pfn[i] =
4703 arch_zone_highest_possible_pfn[i-1];
4704 arch_zone_highest_possible_pfn[i] =
4705 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4706 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07004707 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4708 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4709
4710 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4711 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4712 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004713
Mel Gormanc7132162006-09-27 01:49:43 -07004714 /* Print out the zone ranges */
4715 printk("Zone PFN ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07004716 for (i = 0; i < MAX_NR_ZONES; i++) {
4717 if (i == ZONE_MOVABLE)
4718 continue;
David Rientjes72f0ba02010-03-05 13:42:14 -08004719 printk(" %-8s ", zone_names[i]);
4720 if (arch_zone_lowest_possible_pfn[i] ==
4721 arch_zone_highest_possible_pfn[i])
4722 printk("empty\n");
4723 else
4724 printk("%0#10lx -> %0#10lx\n",
Mel Gormanc7132162006-09-27 01:49:43 -07004725 arch_zone_lowest_possible_pfn[i],
4726 arch_zone_highest_possible_pfn[i]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004727 }
4728
4729 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4730 printk("Movable zone start PFN for each node\n");
4731 for (i = 0; i < MAX_NUMNODES; i++) {
4732 if (zone_movable_pfn[i])
4733 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4734 }
Mel Gormanc7132162006-09-27 01:49:43 -07004735
4736 /* Print out the early_node_map[] */
4737 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4738 for (i = 0; i < nr_nodemap_entries; i++)
Paul Jackson5dab8ec2008-06-25 05:44:40 -07004739 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004740 early_node_map[i].start_pfn,
4741 early_node_map[i].end_pfn);
4742
4743 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07004744 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08004745 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07004746 for_each_online_node(nid) {
4747 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07004748 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07004749 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004750
4751 /* Any memory on that node */
4752 if (pgdat->node_present_pages)
4753 node_set_state(nid, N_HIGH_MEMORY);
4754 check_for_regular_memory(pgdat);
Mel Gormanc7132162006-09-27 01:49:43 -07004755 }
4756}
Mel Gorman2a1e2742007-07-17 04:03:12 -07004757
Mel Gorman7e63efe2007-07-17 04:03:15 -07004758static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004759{
4760 unsigned long long coremem;
4761 if (!p)
4762 return -EINVAL;
4763
4764 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004765 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004766
Mel Gorman7e63efe2007-07-17 04:03:15 -07004767 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07004768 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4769
4770 return 0;
4771}
Mel Gormaned7ed362007-07-17 04:03:14 -07004772
Mel Gorman7e63efe2007-07-17 04:03:15 -07004773/*
4774 * kernelcore=size sets the amount of memory for use for allocations that
4775 * cannot be reclaimed or migrated.
4776 */
4777static int __init cmdline_parse_kernelcore(char *p)
4778{
4779 return cmdline_parse_core(p, &required_kernelcore);
4780}
4781
4782/*
4783 * movablecore=size sets the amount of memory for use for allocations that
4784 * can be reclaimed or migrated.
4785 */
4786static int __init cmdline_parse_movablecore(char *p)
4787{
4788 return cmdline_parse_core(p, &required_movablecore);
4789}
4790
Mel Gormaned7ed362007-07-17 04:03:14 -07004791early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004792early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07004793
Mel Gormanc7132162006-09-27 01:49:43 -07004794#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4795
Mel Gorman0e0b8642006-09-27 01:49:56 -07004796/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004797 * set_dma_reserve - set the specified number of pages reserved in the first zone
4798 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07004799 *
4800 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4801 * In the DMA zone, a significant percentage may be consumed by kernel image
4802 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004803 * function may optionally be used to account for unfreeable pages in the
4804 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4805 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07004806 */
4807void __init set_dma_reserve(unsigned long new_dma_reserve)
4808{
4809 dma_reserve = new_dma_reserve;
4810}
4811
Dave Hansen93b75042005-06-23 00:07:47 -07004812#ifndef CONFIG_NEED_MULTIPLE_NODES
Yinghai Lu08677212010-02-10 01:20:20 -08004813struct pglist_data __refdata contig_page_data = {
4814#ifndef CONFIG_NO_BOOTMEM
4815 .bdata = &bootmem_node_data[0]
4816#endif
4817 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07004818EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07004819#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004820
4821void __init free_area_init(unsigned long *zones_size)
4822{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004823 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004824 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4825}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004826
Linus Torvalds1da177e2005-04-16 15:20:36 -07004827static int page_alloc_cpu_notify(struct notifier_block *self,
4828 unsigned long action, void *hcpu)
4829{
4830 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004831
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07004832 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004833 drain_pages(cpu);
4834
4835 /*
4836 * Spill the event counters of the dead processor
4837 * into the current processors event counters.
4838 * This artificially elevates the count of the current
4839 * processor.
4840 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07004841 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004842
4843 /*
4844 * Zero the differential counters of the dead processor
4845 * so that the vm statistics are consistent.
4846 *
4847 * This is only okay since the processor is dead and cannot
4848 * race with what we are doing.
4849 */
Christoph Lameter2244b952006-06-30 01:55:33 -07004850 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004851 }
4852 return NOTIFY_OK;
4853}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004854
4855void __init page_alloc_init(void)
4856{
4857 hotcpu_notifier(page_alloc_cpu_notify, 0);
4858}
4859
4860/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004861 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4862 * or min_free_kbytes changes.
4863 */
4864static void calculate_totalreserve_pages(void)
4865{
4866 struct pglist_data *pgdat;
4867 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004868 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004869
4870 for_each_online_pgdat(pgdat) {
4871 for (i = 0; i < MAX_NR_ZONES; i++) {
4872 struct zone *zone = pgdat->node_zones + i;
4873 unsigned long max = 0;
4874
4875 /* Find valid and maximum lowmem_reserve in the zone */
4876 for (j = i; j < MAX_NR_ZONES; j++) {
4877 if (zone->lowmem_reserve[j] > max)
4878 max = zone->lowmem_reserve[j];
4879 }
4880
Mel Gorman41858962009-06-16 15:32:12 -07004881 /* we treat the high watermark as reserved pages. */
4882 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004883
4884 if (max > zone->present_pages)
4885 max = zone->present_pages;
4886 reserve_pages += max;
4887 }
4888 }
4889 totalreserve_pages = reserve_pages;
4890}
4891
4892/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07004893 * setup_per_zone_lowmem_reserve - called whenever
4894 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4895 * has a correct pages reserved value, so an adequate number of
4896 * pages are left in the zone after a successful __alloc_pages().
4897 */
4898static void setup_per_zone_lowmem_reserve(void)
4899{
4900 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004901 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004902
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08004903 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004904 for (j = 0; j < MAX_NR_ZONES; j++) {
4905 struct zone *zone = pgdat->node_zones + j;
4906 unsigned long present_pages = zone->present_pages;
4907
4908 zone->lowmem_reserve[j] = 0;
4909
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004910 idx = j;
4911 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004912 struct zone *lower_zone;
4913
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004914 idx--;
4915
Linus Torvalds1da177e2005-04-16 15:20:36 -07004916 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4917 sysctl_lowmem_reserve_ratio[idx] = 1;
4918
4919 lower_zone = pgdat->node_zones + idx;
4920 lower_zone->lowmem_reserve[j] = present_pages /
4921 sysctl_lowmem_reserve_ratio[idx];
4922 present_pages += lower_zone->present_pages;
4923 }
4924 }
4925 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004926
4927 /* update totalreserve_pages */
4928 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004929}
4930
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004931/**
Minchan Kimbc75d332009-06-16 15:32:48 -07004932 * setup_per_zone_wmarks - called when min_free_kbytes changes
Minchan Kimbce73942009-06-16 15:32:50 -07004933 * or when memory is hot-{added|removed}
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004934 *
Minchan Kimbc75d332009-06-16 15:32:48 -07004935 * Ensures that the watermark[min,low,high] values for each zone are set
4936 * correctly with respect to min_free_kbytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004937 */
Minchan Kimbc75d332009-06-16 15:32:48 -07004938void setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004939{
4940 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4941 unsigned long lowmem_pages = 0;
4942 struct zone *zone;
4943 unsigned long flags;
4944
4945 /* Calculate total number of !ZONE_HIGHMEM pages */
4946 for_each_zone(zone) {
4947 if (!is_highmem(zone))
4948 lowmem_pages += zone->present_pages;
4949 }
4950
4951 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07004952 u64 tmp;
4953
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07004954 spin_lock_irqsave(&zone->lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07004955 tmp = (u64)pages_min * zone->present_pages;
4956 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004957 if (is_highmem(zone)) {
4958 /*
Nick Piggin669ed172005-11-13 16:06:45 -08004959 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4960 * need highmem pages, so cap pages_min to a small
4961 * value here.
4962 *
Mel Gorman41858962009-06-16 15:32:12 -07004963 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Nick Piggin669ed172005-11-13 16:06:45 -08004964 * deltas controls asynch page reclaim, and so should
4965 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004966 */
4967 int min_pages;
4968
4969 min_pages = zone->present_pages / 1024;
4970 if (min_pages < SWAP_CLUSTER_MAX)
4971 min_pages = SWAP_CLUSTER_MAX;
4972 if (min_pages > 128)
4973 min_pages = 128;
Mel Gorman41858962009-06-16 15:32:12 -07004974 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004975 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08004976 /*
4977 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07004978 * proportionate to the zone's size.
4979 */
Mel Gorman41858962009-06-16 15:32:12 -07004980 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004981 }
4982
Mel Gorman41858962009-06-16 15:32:12 -07004983 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
4984 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Mel Gorman56fd56b2007-10-16 01:25:58 -07004985 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07004986 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004987 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004988
4989 /* update totalreserve_pages */
4990 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004991}
4992
Randy Dunlap55a44622009-09-21 17:01:20 -07004993/*
Rik van Riel556adec2008-10-18 20:26:34 -07004994 * The inactive anon list should be small enough that the VM never has to
4995 * do too much work, but large enough that each inactive page has a chance
4996 * to be referenced again before it is swapped out.
4997 *
4998 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4999 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5000 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5001 * the anonymous pages are kept on the inactive list.
5002 *
5003 * total target max
5004 * memory ratio inactive anon
5005 * -------------------------------------
5006 * 10MB 1 5MB
5007 * 100MB 1 50MB
5008 * 1GB 3 250MB
5009 * 10GB 10 0.9GB
5010 * 100GB 31 3GB
5011 * 1TB 101 10GB
5012 * 10TB 320 32GB
5013 */
Minchan Kim96cb4df2009-06-16 15:32:49 -07005014void calculate_zone_inactive_ratio(struct zone *zone)
5015{
5016 unsigned int gb, ratio;
5017
5018 /* Zone size in gigabytes */
5019 gb = zone->present_pages >> (30 - PAGE_SHIFT);
5020 if (gb)
5021 ratio = int_sqrt(10 * gb);
5022 else
5023 ratio = 1;
5024
5025 zone->inactive_ratio = ratio;
5026}
5027
Cyrill Gorcunove9bb35d2009-06-16 15:32:32 -07005028static void __init setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07005029{
5030 struct zone *zone;
5031
Minchan Kim96cb4df2009-06-16 15:32:49 -07005032 for_each_zone(zone)
5033 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07005034}
5035
Linus Torvalds1da177e2005-04-16 15:20:36 -07005036/*
5037 * Initialise min_free_kbytes.
5038 *
5039 * For small machines we want it small (128k min). For large machines
5040 * we want it large (64MB max). But it is not linear, because network
5041 * bandwidth does not increase linearly with machine size. We use
5042 *
5043 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5044 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5045 *
5046 * which yields
5047 *
5048 * 16MB: 512k
5049 * 32MB: 724k
5050 * 64MB: 1024k
5051 * 128MB: 1448k
5052 * 256MB: 2048k
5053 * 512MB: 2896k
5054 * 1024MB: 4096k
5055 * 2048MB: 5792k
5056 * 4096MB: 8192k
5057 * 8192MB: 11584k
5058 * 16384MB: 16384k
5059 */
Minchan Kimbc75d332009-06-16 15:32:48 -07005060static int __init init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005061{
5062 unsigned long lowmem_kbytes;
5063
5064 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
5065
5066 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
5067 if (min_free_kbytes < 128)
5068 min_free_kbytes = 128;
5069 if (min_free_kbytes > 65536)
5070 min_free_kbytes = 65536;
Minchan Kimbc75d332009-06-16 15:32:48 -07005071 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005072 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07005073 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005074 return 0;
5075}
Minchan Kimbc75d332009-06-16 15:32:48 -07005076module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005077
5078/*
5079 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5080 * that we can call two helper functions whenever min_free_kbytes
5081 * changes.
5082 */
5083int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005084 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005085{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005086 proc_dointvec(table, write, buffer, length, ppos);
Mel Gorman3b1d92c2007-05-06 14:49:30 -07005087 if (write)
Minchan Kimbc75d332009-06-16 15:32:48 -07005088 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005089 return 0;
5090}
5091
Christoph Lameter96146342006-07-03 00:24:13 -07005092#ifdef CONFIG_NUMA
5093int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005094 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07005095{
5096 struct zone *zone;
5097 int rc;
5098
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005099 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07005100 if (rc)
5101 return rc;
5102
5103 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07005104 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07005105 sysctl_min_unmapped_ratio) / 100;
5106 return 0;
5107}
Christoph Lameter0ff38492006-09-25 23:31:52 -07005108
5109int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005110 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07005111{
5112 struct zone *zone;
5113 int rc;
5114
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005115 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07005116 if (rc)
5117 return rc;
5118
5119 for_each_zone(zone)
5120 zone->min_slab_pages = (zone->present_pages *
5121 sysctl_min_slab_ratio) / 100;
5122 return 0;
5123}
Christoph Lameter96146342006-07-03 00:24:13 -07005124#endif
5125
Linus Torvalds1da177e2005-04-16 15:20:36 -07005126/*
5127 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5128 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5129 * whenever sysctl_lowmem_reserve_ratio changes.
5130 *
5131 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07005132 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07005133 * if in function of the boot time zone sizes.
5134 */
5135int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005136 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005137{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005138 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005139 setup_per_zone_lowmem_reserve();
5140 return 0;
5141}
5142
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005143/*
5144 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5145 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5146 * can have before it gets flushed back to buddy allocator.
5147 */
5148
5149int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005150 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005151{
5152 struct zone *zone;
5153 unsigned int cpu;
5154 int ret;
5155
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005156 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005157 if (!write || (ret == -EINVAL))
5158 return ret;
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07005159 for_each_populated_zone(zone) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005160 for_each_possible_cpu(cpu) {
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005161 unsigned long high;
5162 high = zone->present_pages / percpu_pagelist_fraction;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005163 setup_pagelist_highmark(
5164 per_cpu_ptr(zone->pageset, cpu), high);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005165 }
5166 }
5167 return 0;
5168}
5169
David S. Millerf034b5d2006-08-24 03:08:07 -07005170int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005171
5172#ifdef CONFIG_NUMA
5173static int __init set_hashdist(char *str)
5174{
5175 if (!str)
5176 return 0;
5177 hashdist = simple_strtoul(str, &str, 0);
5178 return 1;
5179}
5180__setup("hashdist=", set_hashdist);
5181#endif
5182
5183/*
5184 * allocate a large system hash table from bootmem
5185 * - it is assumed that the hash table must contain an exact power-of-2
5186 * quantity of entries
5187 * - limit is the number of hash buckets, not the total allocation size
5188 */
5189void *__init alloc_large_system_hash(const char *tablename,
5190 unsigned long bucketsize,
5191 unsigned long numentries,
5192 int scale,
5193 int flags,
5194 unsigned int *_hash_shift,
5195 unsigned int *_hash_mask,
5196 unsigned long limit)
5197{
5198 unsigned long long max = limit;
5199 unsigned long log2qty, size;
5200 void *table = NULL;
5201
5202 /* allow the kernel cmdline to have a say */
5203 if (!numentries) {
5204 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08005205 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005206 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
5207 numentries >>= 20 - PAGE_SHIFT;
5208 numentries <<= 20 - PAGE_SHIFT;
5209
5210 /* limit to 1 bucket per 2^scale bytes of low memory */
5211 if (scale > PAGE_SHIFT)
5212 numentries >>= (scale - PAGE_SHIFT);
5213 else
5214 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08005215
5216 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07005217 if (unlikely(flags & HASH_SMALL)) {
5218 /* Makes no sense without HASH_EARLY */
5219 WARN_ON(!(flags & HASH_EARLY));
5220 if (!(numentries >> *_hash_shift)) {
5221 numentries = 1UL << *_hash_shift;
5222 BUG_ON(!numentries);
5223 }
5224 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08005225 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005226 }
John Hawkes6e692ed2006-03-25 03:08:02 -08005227 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005228
5229 /* limit allocation size to 1/16 total memory by default */
5230 if (max == 0) {
5231 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
5232 do_div(max, bucketsize);
5233 }
5234
5235 if (numentries > max)
5236 numentries = max;
5237
David Howellsf0d1b0b2006-12-08 02:37:49 -08005238 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005239
5240 do {
5241 size = bucketsize << log2qty;
5242 if (flags & HASH_EARLY)
Jan Beulich74768ed2008-08-12 15:08:39 -07005243 table = alloc_bootmem_nopanic(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005244 else if (hashdist)
5245 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
5246 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07005247 /*
5248 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07005249 * some pages at the end of hash table which
5250 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07005251 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005252 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07005253 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005254 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
5255 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005256 }
5257 } while (!table && size > PAGE_SIZE && --log2qty);
5258
5259 if (!table)
5260 panic("Failed to allocate %s hash table\n", tablename);
5261
Robin Holtf241e6602010-10-07 12:59:26 -07005262 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07005263 tablename,
Robin Holtf241e6602010-10-07 12:59:26 -07005264 (1UL << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08005265 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005266 size);
5267
5268 if (_hash_shift)
5269 *_hash_shift = log2qty;
5270 if (_hash_mask)
5271 *_hash_mask = (1 << log2qty) - 1;
5272
5273 return table;
5274}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08005275
Mel Gorman835c1342007-10-16 01:25:47 -07005276/* Return a pointer to the bitmap storing bits affecting a block of pages */
5277static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
5278 unsigned long pfn)
5279{
5280#ifdef CONFIG_SPARSEMEM
5281 return __pfn_to_section(pfn)->pageblock_flags;
5282#else
5283 return zone->pageblock_flags;
5284#endif /* CONFIG_SPARSEMEM */
5285}
Andrew Morton6220ec72006-10-19 23:29:05 -07005286
Mel Gorman835c1342007-10-16 01:25:47 -07005287static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
5288{
5289#ifdef CONFIG_SPARSEMEM
5290 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07005291 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005292#else
5293 pfn = pfn - zone->zone_start_pfn;
Mel Gormand9c23402007-10-16 01:26:01 -07005294 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005295#endif /* CONFIG_SPARSEMEM */
5296}
5297
5298/**
Mel Gormand9c23402007-10-16 01:26:01 -07005299 * 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 -07005300 * @page: The page within the block of interest
5301 * @start_bitidx: The first bit of interest to retrieve
5302 * @end_bitidx: The last bit of interest
5303 * returns pageblock_bits flags
5304 */
5305unsigned long get_pageblock_flags_group(struct page *page,
5306 int start_bitidx, int end_bitidx)
5307{
5308 struct zone *zone;
5309 unsigned long *bitmap;
5310 unsigned long pfn, bitidx;
5311 unsigned long flags = 0;
5312 unsigned long value = 1;
5313
5314 zone = page_zone(page);
5315 pfn = page_to_pfn(page);
5316 bitmap = get_pageblock_bitmap(zone, pfn);
5317 bitidx = pfn_to_bitidx(zone, pfn);
5318
5319 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5320 if (test_bit(bitidx + start_bitidx, bitmap))
5321 flags |= value;
5322
5323 return flags;
5324}
5325
5326/**
Mel Gormand9c23402007-10-16 01:26:01 -07005327 * 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 -07005328 * @page: The page within the block of interest
5329 * @start_bitidx: The first bit of interest
5330 * @end_bitidx: The last bit of interest
5331 * @flags: The flags to set
5332 */
5333void set_pageblock_flags_group(struct page *page, unsigned long flags,
5334 int start_bitidx, int end_bitidx)
5335{
5336 struct zone *zone;
5337 unsigned long *bitmap;
5338 unsigned long pfn, bitidx;
5339 unsigned long value = 1;
5340
5341 zone = page_zone(page);
5342 pfn = page_to_pfn(page);
5343 bitmap = get_pageblock_bitmap(zone, pfn);
5344 bitidx = pfn_to_bitidx(zone, pfn);
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07005345 VM_BUG_ON(pfn < zone->zone_start_pfn);
5346 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
Mel Gorman835c1342007-10-16 01:25:47 -07005347
5348 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5349 if (flags & value)
5350 __set_bit(bitidx + start_bitidx, bitmap);
5351 else
5352 __clear_bit(bitidx + start_bitidx, bitmap);
5353}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005354
5355/*
5356 * This is designed as sub function...plz see page_isolation.c also.
5357 * set/clear page block's type to be ISOLATE.
5358 * page allocater never alloc memory from ISOLATE block.
5359 */
5360
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005361static int
5362__count_immobile_pages(struct zone *zone, struct page *page, int count)
5363{
5364 unsigned long pfn, iter, found;
5365 /*
5366 * For avoiding noise data, lru_add_drain_all() should be called
5367 * If ZONE_MOVABLE, the zone never contains immobile pages
5368 */
5369 if (zone_idx(zone) == ZONE_MOVABLE)
5370 return true;
5371
5372 if (get_pageblock_migratetype(page) == MIGRATE_MOVABLE)
5373 return true;
5374
5375 pfn = page_to_pfn(page);
5376 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
5377 unsigned long check = pfn + iter;
5378
5379 if (!pfn_valid_within(check)) {
5380 iter++;
5381 continue;
5382 }
5383 page = pfn_to_page(check);
5384 if (!page_count(page)) {
5385 if (PageBuddy(page))
5386 iter += (1 << page_order(page)) - 1;
5387 continue;
5388 }
5389 if (!PageLRU(page))
5390 found++;
5391 /*
5392 * If there are RECLAIMABLE pages, we need to check it.
5393 * But now, memory offline itself doesn't call shrink_slab()
5394 * and it still to be fixed.
5395 */
5396 /*
5397 * If the page is not RAM, page_count()should be 0.
5398 * we don't need more check. This is an _used_ not-movable page.
5399 *
5400 * The problematic thing here is PG_reserved pages. PG_reserved
5401 * is set to both of a memory hole page and a _used_ kernel
5402 * page at boot.
5403 */
5404 if (found > count)
5405 return false;
5406 }
5407 return true;
5408}
5409
5410bool is_pageblock_removable_nolock(struct page *page)
5411{
5412 struct zone *zone = page_zone(page);
5413 return __count_immobile_pages(zone, page, 0);
5414}
5415
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005416int set_migratetype_isolate(struct page *page)
5417{
5418 struct zone *zone;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005419 unsigned long flags, pfn;
Robert Jennings925cc712009-12-17 14:44:38 +00005420 struct memory_isolate_notify arg;
5421 int notifier_ret;
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005422 int ret = -EBUSY;
Shaohua Li8e7e40d2009-09-21 17:01:18 -07005423 int zone_idx;
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005424
5425 zone = page_zone(page);
Shaohua Li8e7e40d2009-09-21 17:01:18 -07005426 zone_idx = zone_idx(zone);
Robert Jennings925cc712009-12-17 14:44:38 +00005427
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005428 spin_lock_irqsave(&zone->lock, flags);
Robert Jennings925cc712009-12-17 14:44:38 +00005429
5430 pfn = page_to_pfn(page);
5431 arg.start_pfn = pfn;
5432 arg.nr_pages = pageblock_nr_pages;
5433 arg.pages_found = 0;
5434
5435 /*
5436 * It may be possible to isolate a pageblock even if the
5437 * migratetype is not MIGRATE_MOVABLE. The memory isolation
5438 * notifier chain is used by balloon drivers to return the
5439 * number of pages in a range that are held by the balloon
5440 * driver to shrink memory. If all the pages are accounted for
5441 * by balloons, are free, or on the LRU, isolation can continue.
5442 * Later, for example, when memory hotplug notifier runs, these
5443 * pages reported as "can be isolated" should be isolated(freed)
5444 * by the balloon driver through the memory notifier chain.
5445 */
5446 notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg);
5447 notifier_ret = notifier_to_errno(notifier_ret);
KAMEZAWA Hiroyuki4b204772010-10-26 14:21:29 -07005448 if (notifier_ret)
Robert Jennings925cc712009-12-17 14:44:38 +00005449 goto out;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005450 /*
5451 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
5452 * We just check MOVABLE pages.
5453 */
5454 if (__count_immobile_pages(zone, page, arg.pages_found))
Robert Jennings925cc712009-12-17 14:44:38 +00005455 ret = 0;
5456
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005457 /*
5458 * immobile means "not-on-lru" paes. If immobile is larger than
5459 * removable-by-driver pages reported by notifier, we'll fail.
5460 */
5461
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005462out:
Robert Jennings925cc712009-12-17 14:44:38 +00005463 if (!ret) {
5464 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
5465 move_freepages_block(zone, page, MIGRATE_ISOLATE);
5466 }
5467
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005468 spin_unlock_irqrestore(&zone->lock, flags);
5469 if (!ret)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005470 drain_all_pages();
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005471 return ret;
5472}
5473
5474void unset_migratetype_isolate(struct page *page)
5475{
5476 struct zone *zone;
5477 unsigned long flags;
5478 zone = page_zone(page);
5479 spin_lock_irqsave(&zone->lock, flags);
5480 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
5481 goto out;
5482 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5483 move_freepages_block(zone, page, MIGRATE_MOVABLE);
5484out:
5485 spin_unlock_irqrestore(&zone->lock, flags);
5486}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07005487
5488#ifdef CONFIG_MEMORY_HOTREMOVE
5489/*
5490 * All pages in the range must be isolated before calling this.
5491 */
5492void
5493__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
5494{
5495 struct page *page;
5496 struct zone *zone;
5497 int order, i;
5498 unsigned long pfn;
5499 unsigned long flags;
5500 /* find the first valid pfn */
5501 for (pfn = start_pfn; pfn < end_pfn; pfn++)
5502 if (pfn_valid(pfn))
5503 break;
5504 if (pfn == end_pfn)
5505 return;
5506 zone = page_zone(pfn_to_page(pfn));
5507 spin_lock_irqsave(&zone->lock, flags);
5508 pfn = start_pfn;
5509 while (pfn < end_pfn) {
5510 if (!pfn_valid(pfn)) {
5511 pfn++;
5512 continue;
5513 }
5514 page = pfn_to_page(pfn);
5515 BUG_ON(page_count(page));
5516 BUG_ON(!PageBuddy(page));
5517 order = page_order(page);
5518#ifdef CONFIG_DEBUG_VM
5519 printk(KERN_INFO "remove from free list %lx %d %lx\n",
5520 pfn, 1 << order, end_pfn);
5521#endif
5522 list_del(&page->lru);
5523 rmv_page_order(page);
5524 zone->free_area[order].nr_free--;
5525 __mod_zone_page_state(zone, NR_FREE_PAGES,
5526 - (1UL << order));
5527 for (i = 0; i < (1 << order); i++)
5528 SetPageReserved((page+i));
5529 pfn += (1 << order);
5530 }
5531 spin_unlock_irqrestore(&zone->lock, flags);
5532}
5533#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01005534
5535#ifdef CONFIG_MEMORY_FAILURE
5536bool is_free_buddy_page(struct page *page)
5537{
5538 struct zone *zone = page_zone(page);
5539 unsigned long pfn = page_to_pfn(page);
5540 unsigned long flags;
5541 int order;
5542
5543 spin_lock_irqsave(&zone->lock, flags);
5544 for (order = 0; order < MAX_ORDER; order++) {
5545 struct page *page_head = page - (pfn & ((1 << order) - 1));
5546
5547 if (PageBuddy(page_head) && page_order(page_head) >= order)
5548 break;
5549 }
5550 spin_unlock_irqrestore(&zone->lock, flags);
5551
5552 return order < MAX_ORDER;
5553}
5554#endif
Wu Fengguang718a3822010-03-10 15:20:43 -08005555
5556static struct trace_print_flags pageflag_names[] = {
5557 {1UL << PG_locked, "locked" },
5558 {1UL << PG_error, "error" },
5559 {1UL << PG_referenced, "referenced" },
5560 {1UL << PG_uptodate, "uptodate" },
5561 {1UL << PG_dirty, "dirty" },
5562 {1UL << PG_lru, "lru" },
5563 {1UL << PG_active, "active" },
5564 {1UL << PG_slab, "slab" },
5565 {1UL << PG_owner_priv_1, "owner_priv_1" },
5566 {1UL << PG_arch_1, "arch_1" },
5567 {1UL << PG_reserved, "reserved" },
5568 {1UL << PG_private, "private" },
5569 {1UL << PG_private_2, "private_2" },
5570 {1UL << PG_writeback, "writeback" },
5571#ifdef CONFIG_PAGEFLAGS_EXTENDED
5572 {1UL << PG_head, "head" },
5573 {1UL << PG_tail, "tail" },
5574#else
5575 {1UL << PG_compound, "compound" },
5576#endif
5577 {1UL << PG_swapcache, "swapcache" },
5578 {1UL << PG_mappedtodisk, "mappedtodisk" },
5579 {1UL << PG_reclaim, "reclaim" },
Wu Fengguang718a3822010-03-10 15:20:43 -08005580 {1UL << PG_swapbacked, "swapbacked" },
5581 {1UL << PG_unevictable, "unevictable" },
5582#ifdef CONFIG_MMU
5583 {1UL << PG_mlocked, "mlocked" },
5584#endif
5585#ifdef CONFIG_ARCH_USES_PG_UNCACHED
5586 {1UL << PG_uncached, "uncached" },
5587#endif
5588#ifdef CONFIG_MEMORY_FAILURE
5589 {1UL << PG_hwpoison, "hwpoison" },
5590#endif
5591 {-1UL, NULL },
5592};
5593
5594static void dump_page_flags(unsigned long flags)
5595{
5596 const char *delim = "";
5597 unsigned long mask;
5598 int i;
5599
5600 printk(KERN_ALERT "page flags: %#lx(", flags);
5601
5602 /* remove zone id */
5603 flags &= (1UL << NR_PAGEFLAGS) - 1;
5604
5605 for (i = 0; pageflag_names[i].name && flags; i++) {
5606
5607 mask = pageflag_names[i].mask;
5608 if ((flags & mask) != mask)
5609 continue;
5610
5611 flags &= ~mask;
5612 printk("%s%s", delim, pageflag_names[i].name);
5613 delim = "|";
5614 }
5615
5616 /* check for left over flags */
5617 if (flags)
5618 printk("%s%#lx", delim, flags);
5619
5620 printk(")\n");
5621}
5622
5623void dump_page(struct page *page)
5624{
5625 printk(KERN_ALERT
5626 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
Andrea Arcangeli4e9f64c2011-01-13 15:46:29 -08005627 page, atomic_read(&page->_count), page_mapcount(page),
Wu Fengguang718a3822010-03-10 15:20:43 -08005628 page->mapping, page->index);
5629 dump_page_flags(page->flags);
5630}