blob: c2b59dbda196dcaaffcd44f367143e3e83a15c0f [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>
Dave Hansena238ab52011-05-24 17:12:16 -070033#include <linux/ratelimit.h>
David Rientjes5a3135c22007-10-16 23:25:53 -070034#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <linux/notifier.h>
36#include <linux/topology.h>
37#include <linux/sysctl.h>
38#include <linux/cpu.h>
39#include <linux/cpuset.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070040#include <linux/memory_hotplug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include <linux/nodemask.h>
42#include <linux/vmalloc.h>
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -070043#include <linux/vmstat.h>
Christoph Lameter4be38e32006-01-06 00:11:17 -080044#include <linux/mempolicy.h>
Yasunori Goto68113782006-06-23 02:03:11 -070045#include <linux/stop_machine.h>
Mel Gormanc7132162006-09-27 01:49:43 -070046#include <linux/sort.h>
47#include <linux/pfn.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070048#include <linux/backing-dev.h>
Akinobu Mita933e3122006-12-08 02:39:45 -080049#include <linux/fault-inject.h>
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -070050#include <linux/page-isolation.h>
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070051#include <linux/page_cgroup.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070052#include <linux/debugobjects.h>
Catalin Marinasdbb1f812009-06-11 13:23:19 +010053#include <linux/kmemleak.h>
Mel Gorman56de7262010-05-24 14:32:30 -070054#include <linux/compaction.h>
Mel Gorman0d3d0622009-09-21 17:02:44 -070055#include <trace/events/kmem.h>
Wu Fengguang718a3822010-03-10 15:20:43 -080056#include <linux/ftrace_event.h>
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -070057#include <linux/memcontrol.h>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070058#include <linux/prefetch.h>
Michal Nazarewicz041d3a82011-12-29 13:09:50 +010059#include <linux/migrate.h>
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -080060#include <linux/page-debug-flags.h>
David Rientjes949f7ec2013-04-29 15:07:48 -070061#include <linux/hugetlb.h>
Clark Williams8bd75c72013-02-07 09:47:07 -060062#include <linux/sched/rt.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070063
Jiang Liu7ee3d4e2013-07-03 15:03:41 -070064#include <asm/sections.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070065#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070066#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070067#include "internal.h"
68
Cody P Schaferc8e251f2013-07-03 15:01:29 -070069/* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
70static DEFINE_MUTEX(pcp_batch_high_lock);
71
Lee Schermerhorn72812012010-05-26 14:44:56 -070072#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
73DEFINE_PER_CPU(int, numa_node);
74EXPORT_PER_CPU_SYMBOL(numa_node);
75#endif
76
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070077#ifdef CONFIG_HAVE_MEMORYLESS_NODES
78/*
79 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
80 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
81 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
82 * defined in <linux/topology.h>.
83 */
84DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
85EXPORT_PER_CPU_SYMBOL(_numa_mem_);
86#endif
87
Linus Torvalds1da177e2005-04-16 15:20:36 -070088/*
Christoph Lameter13808912007-10-16 01:25:27 -070089 * Array of node states.
Linus Torvalds1da177e2005-04-16 15:20:36 -070090 */
Christoph Lameter13808912007-10-16 01:25:27 -070091nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
92 [N_POSSIBLE] = NODE_MASK_ALL,
93 [N_ONLINE] = { { [0] = 1UL } },
94#ifndef CONFIG_NUMA
95 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
96#ifdef CONFIG_HIGHMEM
97 [N_HIGH_MEMORY] = { { [0] = 1UL } },
98#endif
Lai Jiangshan20b2f522012-12-12 13:52:00 -080099#ifdef CONFIG_MOVABLE_NODE
100 [N_MEMORY] = { { [0] = 1UL } },
101#endif
Christoph Lameter13808912007-10-16 01:25:27 -0700102 [N_CPU] = { { [0] = 1UL } },
103#endif /* NUMA */
104};
105EXPORT_SYMBOL(node_states);
106
Jiang Liuc3d5f5f2013-07-03 15:03:14 -0700107/* Protect totalram_pages and zone->managed_pages */
108static DEFINE_SPINLOCK(managed_page_count_lock);
109
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700110unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -0700111unsigned long totalreserve_pages __read_mostly;
Johannes Weinerab8fabd2012-01-10 15:07:42 -0800112/*
113 * When calculating the number of globally allowed dirty pages, there
114 * is a certain number of per-zone reserves that should not be
115 * considered dirtyable memory. This is the sum of those reserves
116 * over all existing zones that contribute dirtyable memory.
117 */
118unsigned long dirty_balance_reserve __read_mostly;
119
Hugh Dickins1b76b022012-05-11 01:00:07 -0700120int percpu_pagelist_fraction;
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000121gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800123#ifdef CONFIG_PM_SLEEP
124/*
125 * The following functions are used by the suspend/hibernate code to temporarily
126 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
127 * while devices are suspended. To avoid races with the suspend/hibernate code,
128 * they should always be called with pm_mutex held (gfp_allowed_mask also should
129 * only be modified with pm_mutex held, unless the suspend/hibernate code is
130 * guaranteed not to run in parallel with that modification).
131 */
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100132
133static gfp_t saved_gfp_mask;
134
135void pm_restore_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800136{
137 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100138 if (saved_gfp_mask) {
139 gfp_allowed_mask = saved_gfp_mask;
140 saved_gfp_mask = 0;
141 }
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800142}
143
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100144void pm_restrict_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800145{
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800146 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100147 WARN_ON(saved_gfp_mask);
148 saved_gfp_mask = gfp_allowed_mask;
149 gfp_allowed_mask &= ~GFP_IOFS;
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800150}
Mel Gormanf90ac392012-01-10 15:07:15 -0800151
152bool pm_suspended_storage(void)
153{
154 if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS)
155 return false;
156 return true;
157}
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800158#endif /* CONFIG_PM_SLEEP */
159
Mel Gormand9c23402007-10-16 01:26:01 -0700160#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
161int pageblock_order __read_mostly;
162#endif
163
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800164static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -0800165
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166/*
167 * results with 256, 32 in the lowmem_reserve sysctl:
168 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
169 * 1G machine -> (16M dma, 784M normal, 224M high)
170 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
171 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
172 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +0100173 *
174 * TBD: should special case ZONE_DMA32 machines here - in those we normally
175 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700177int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800178#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700179 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -0800180#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700181#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700182 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700183#endif
Christoph Lametere53ef382006-09-25 23:31:14 -0700184#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700185 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700186#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700187 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700188};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189
190EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191
Helge Deller15ad7cd2006-12-06 20:40:36 -0800192static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800193#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700194 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800195#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700196#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700197 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700198#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700199 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700200#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700201 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700202#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700203 "Movable",
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700204};
205
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206int min_free_kbytes = 1024;
Michal Hocko5f127332013-07-08 16:00:40 -0700207int user_min_free_kbytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208
Jan Beulich2c85f512009-09-21 17:03:07 -0700209static unsigned long __meminitdata nr_kernel_pages;
210static unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700211static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212
Tejun Heo0ee332c2011-12-08 10:22:09 -0800213#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
214static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
215static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
216static unsigned long __initdata required_kernelcore;
217static unsigned long __initdata required_movablecore;
218static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Mel Gormanc7132162006-09-27 01:49:43 -0700219
Tejun Heo0ee332c2011-12-08 10:22:09 -0800220/* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
221int movable_zone;
222EXPORT_SYMBOL(movable_zone);
223#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -0700224
Miklos Szeredi418508c2007-05-23 13:57:55 -0700225#if MAX_NUMNODES > 1
226int nr_node_ids __read_mostly = MAX_NUMNODES;
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700227int nr_online_nodes __read_mostly = 1;
Miklos Szeredi418508c2007-05-23 13:57:55 -0700228EXPORT_SYMBOL(nr_node_ids);
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700229EXPORT_SYMBOL(nr_online_nodes);
Miklos Szeredi418508c2007-05-23 13:57:55 -0700230#endif
231
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700232int page_group_by_mobility_disabled __read_mostly;
233
Minchan Kimee6f5092012-07-31 16:43:50 -0700234void set_pageblock_migratetype(struct page *page, int migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700235{
Mel Gorman49255c62009-06-16 15:31:58 -0700236
237 if (unlikely(page_group_by_mobility_disabled))
238 migratetype = MIGRATE_UNMOVABLE;
239
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700240 set_pageblock_flags_group(page, (unsigned long)migratetype,
241 PB_migrate, PB_migrate_end);
242}
243
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -0700244bool oom_killer_disabled __read_mostly;
245
Nick Piggin13e74442006-01-06 00:10:58 -0800246#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700247static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700249 int ret = 0;
250 unsigned seq;
251 unsigned long pfn = page_to_pfn(page);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800252 unsigned long sp, start_pfn;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700253
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700254 do {
255 seq = zone_span_seqbegin(zone);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800256 start_pfn = zone->zone_start_pfn;
257 sp = zone->spanned_pages;
Cody P Schafer108bcc92013-02-22 16:35:23 -0800258 if (!zone_spans_pfn(zone, pfn))
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700259 ret = 1;
260 } while (zone_span_seqretry(zone, seq));
261
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800262 if (ret)
263 pr_err("page %lu outside zone [ %lu - %lu ]\n",
264 pfn, start_pfn, start_pfn + sp);
265
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700266 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700267}
268
269static int page_is_consistent(struct zone *zone, struct page *page)
270{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700271 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700272 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700274 return 0;
275
276 return 1;
277}
278/*
279 * Temporary debugging check for pages not lying within a given zone.
280 */
281static int bad_range(struct zone *zone, struct page *page)
282{
283 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700285 if (!page_is_consistent(zone, page))
286 return 1;
287
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 return 0;
289}
Nick Piggin13e74442006-01-06 00:10:58 -0800290#else
291static inline int bad_range(struct zone *zone, struct page *page)
292{
293 return 0;
294}
295#endif
296
Nick Piggin224abf92006-01-06 00:11:11 -0800297static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800299 static unsigned long resume;
300 static unsigned long nr_shown;
301 static unsigned long nr_unshown;
302
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200303 /* Don't complain about poisoned pages */
304 if (PageHWPoison(page)) {
Mel Gorman22b751c2013-02-22 16:34:59 -0800305 page_mapcount_reset(page); /* remove PageBuddy */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200306 return;
307 }
308
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800309 /*
310 * Allow a burst of 60 reports, then keep quiet for that minute;
311 * or allow a steady drip of one report per second.
312 */
313 if (nr_shown == 60) {
314 if (time_before(jiffies, resume)) {
315 nr_unshown++;
316 goto out;
317 }
318 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800319 printk(KERN_ALERT
320 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800321 nr_unshown);
322 nr_unshown = 0;
323 }
324 nr_shown = 0;
325 }
326 if (nr_shown++ == 0)
327 resume = jiffies + 60 * HZ;
328
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800329 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800330 current->comm, page_to_pfn(page));
Wu Fengguang718a3822010-03-10 15:20:43 -0800331 dump_page(page);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700332
Dave Jones4f318882011-10-31 17:07:24 -0700333 print_modules();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800335out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800336 /* Leave bad fields for debug, except PageBuddy could make trouble */
Mel Gorman22b751c2013-02-22 16:34:59 -0800337 page_mapcount_reset(page); /* remove PageBuddy */
Rusty Russell373d4d02013-01-21 17:17:39 +1030338 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339}
340
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341/*
342 * Higher-order pages are called "compound pages". They are structured thusly:
343 *
344 * The first PAGE_SIZE page is called the "head page".
345 *
346 * The remaining PAGE_SIZE pages are called "tail pages".
347 *
Wang Sheng-Hui6416b9f2011-11-17 10:53:50 +0100348 * All pages have PG_compound set. All tail pages have their ->first_page
349 * pointing at the head page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800351 * The first tail page's ->lru.next holds the address of the compound page's
352 * put_page() function. Its ->lru.prev holds the order of allocation.
353 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800355
356static void free_compound_page(struct page *page)
357{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700358 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800359}
360
Andi Kleen01ad1c02008-07-23 21:27:46 -0700361void prep_compound_page(struct page *page, unsigned long order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362{
363 int i;
364 int nr_pages = 1 << order;
365
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800366 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700367 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700368 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800369 for (i = 1; i < nr_pages; i++) {
370 struct page *p = page + i;
Christoph Lameterd85f3382007-05-06 14:49:39 -0700371 __SetPageTail(p);
Youquan Song58a84aa2011-12-08 14:34:18 -0800372 set_page_count(p, 0);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700373 p->first_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 }
375}
376
Andrea Arcangeli59ff4212011-01-13 15:46:38 -0800377/* update __split_huge_page_refcount if you change this function */
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800378static int destroy_compound_page(struct page *page, unsigned long order)
Andy Whitcroft18229df2008-11-06 12:53:27 -0800379{
380 int i;
381 int nr_pages = 1 << order;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800382 int bad = 0;
Andy Whitcroft18229df2008-11-06 12:53:27 -0800383
Gavin Shan0bb2c762012-12-18 14:21:32 -0800384 if (unlikely(compound_order(page) != order)) {
Nick Piggin224abf92006-01-06 00:11:11 -0800385 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800386 bad++;
387 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
Christoph Lameter6d777952007-05-06 14:49:40 -0700389 __ClearPageHead(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800390
Andy Whitcroft18229df2008-11-06 12:53:27 -0800391 for (i = 1; i < nr_pages; i++) {
392 struct page *p = page + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393
Alexey Zaytseve713a212009-01-10 02:47:57 +0300394 if (unlikely(!PageTail(p) || (p->first_page != page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800395 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800396 bad++;
397 }
Christoph Lameterd85f3382007-05-06 14:49:39 -0700398 __ClearPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800400
401 return bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403
Nick Piggin17cf4402006-03-22 00:08:41 -0800404static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
405{
406 int i;
407
Andrew Morton6626c5d2006-03-22 00:08:42 -0800408 /*
409 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
410 * and __GFP_HIGHMEM from hard or soft interrupt context.
411 */
Nick Piggin725d7042006-09-25 23:30:55 -0700412 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800413 for (i = 0; i < (1 << order); i++)
414 clear_highpage(page + i);
415}
416
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800417#ifdef CONFIG_DEBUG_PAGEALLOC
418unsigned int _debug_guardpage_minorder;
419
420static int __init debug_guardpage_minorder_setup(char *buf)
421{
422 unsigned long res;
423
424 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
425 printk(KERN_ERR "Bad debug_guardpage_minorder value\n");
426 return 0;
427 }
428 _debug_guardpage_minorder = res;
429 printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res);
430 return 0;
431}
432__setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
433
434static inline void set_page_guard_flag(struct page *page)
435{
436 __set_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags);
437}
438
439static inline void clear_page_guard_flag(struct page *page)
440{
441 __clear_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags);
442}
443#else
444static inline void set_page_guard_flag(struct page *page) { }
445static inline void clear_page_guard_flag(struct page *page) { }
446#endif
447
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700448static inline void set_page_order(struct page *page, int order)
449{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700450 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000451 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452}
453
454static inline void rmv_page_order(struct page *page)
455{
Nick Piggin676165a2006-04-10 11:21:48 +1000456 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700457 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458}
459
460/*
461 * Locate the struct page for both the matching buddy in our
462 * pair (buddy1) and the combined O(n+1) page they form (page).
463 *
464 * 1) Any buddy B1 will have an order O twin B2 which satisfies
465 * the following equation:
466 * B2 = B1 ^ (1 << O)
467 * For example, if the starting buddy (buddy2) is #8 its order
468 * 1 buddy is #10:
469 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
470 *
471 * 2) Any buddy B will have an order O+1 parent P which
472 * satisfies the following equation:
473 * P = B & ~(1 << O)
474 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200475 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477static inline unsigned long
KyongHo Cho43506fa2011-01-13 15:47:24 -0800478__find_buddy_index(unsigned long page_idx, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479{
KyongHo Cho43506fa2011-01-13 15:47:24 -0800480 return page_idx ^ (1 << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481}
482
483/*
484 * This function checks whether a page is free && is the buddy
485 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800486 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000487 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700488 * (c) a page and its buddy have the same order &&
489 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 *
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800491 * For recording whether a page is in the buddy system, we set ->_mapcount -2.
492 * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock.
Nick Piggin676165a2006-04-10 11:21:48 +1000493 *
494 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700496static inline int page_is_buddy(struct page *page, struct page *buddy,
497 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700499 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800500 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800501
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700502 if (page_zone_id(page) != page_zone_id(buddy))
503 return 0;
504
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800505 if (page_is_guard(buddy) && page_order(buddy) == order) {
506 VM_BUG_ON(page_count(buddy) != 0);
507 return 1;
508 }
509
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700510 if (PageBuddy(buddy) && page_order(buddy) == order) {
Nick Piggina3af9c32009-06-16 15:32:10 -0700511 VM_BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700512 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000513 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700514 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515}
516
517/*
518 * Freeing function for a buddy system allocator.
519 *
520 * The concept of a buddy system is to maintain direct-mapped table
521 * (containing bit values) for memory blocks of various "orders".
522 * The bottom level table contains the map for the smallest allocatable
523 * units of memory (here, pages), and each level above it describes
524 * pairs of units from the levels below, hence, "buddies".
525 * At a high level, all that happens here is marking the table entry
526 * at the bottom level available, and propagating the changes upward
527 * as necessary, plus some accounting needed to play nicely with other
528 * parts of the VM system.
529 * At each level, we keep a list of pages, which are heads of continuous
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800530 * free pages of length of (1 << order) and marked with _mapcount -2. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700531 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 * So when we are allocating or freeing one, we can derive the state of the
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100533 * other. That is, if we allocate a small block, and both were
534 * free, the remainder of the region must be split into blocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 * If a block is freed, and its buddy is also free, then this
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100536 * triggers coalescing into a block of larger size.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +0100538 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 */
540
Nick Piggin48db57f2006-01-08 01:00:42 -0800541static inline void __free_one_page(struct page *page,
Mel Gormaned0ae212009-06-16 15:32:07 -0700542 struct zone *zone, unsigned int order,
543 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544{
545 unsigned long page_idx;
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700546 unsigned long combined_idx;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800547 unsigned long uninitialized_var(buddy_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700548 struct page *buddy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549
Cody P Schaferd29bb972013-02-22 16:35:25 -0800550 VM_BUG_ON(!zone_is_initialized(zone));
551
Nick Piggin224abf92006-01-06 00:11:11 -0800552 if (unlikely(PageCompound(page)))
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800553 if (unlikely(destroy_compound_page(page, order)))
554 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555
Mel Gormaned0ae212009-06-16 15:32:07 -0700556 VM_BUG_ON(migratetype == -1);
557
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
559
Mel Gormanf2260e62009-06-16 15:32:13 -0700560 VM_BUG_ON(page_idx & ((1 << order) - 1));
Nick Piggin725d7042006-09-25 23:30:55 -0700561 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 while (order < MAX_ORDER-1) {
KyongHo Cho43506fa2011-01-13 15:47:24 -0800564 buddy_idx = __find_buddy_index(page_idx, order);
565 buddy = page + (buddy_idx - page_idx);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700566 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700567 break;
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800568 /*
569 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
570 * merge with it and move up one order.
571 */
572 if (page_is_guard(buddy)) {
573 clear_page_guard_flag(buddy);
574 set_page_private(page, 0);
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -0700575 __mod_zone_freepage_state(zone, 1 << order,
576 migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800577 } else {
578 list_del(&buddy->lru);
579 zone->free_area[order].nr_free--;
580 rmv_page_order(buddy);
581 }
KyongHo Cho43506fa2011-01-13 15:47:24 -0800582 combined_idx = buddy_idx & page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 page = page + (combined_idx - page_idx);
584 page_idx = combined_idx;
585 order++;
586 }
587 set_page_order(page, order);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700588
589 /*
590 * If this is not the largest possible page, check if the buddy
591 * of the next-highest order is free. If it is, it's possible
592 * that pages are being freed that will coalesce soon. In case,
593 * that is happening, add the free page to the tail of the list
594 * so it's less likely to be used soon and more likely to be merged
595 * as a higher order page
596 */
Mel Gormanb7f50cf2010-10-26 14:21:11 -0700597 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700598 struct page *higher_page, *higher_buddy;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800599 combined_idx = buddy_idx & page_idx;
600 higher_page = page + (combined_idx - page_idx);
601 buddy_idx = __find_buddy_index(combined_idx, order + 1);
Li Haifeng0ba8f2d2012-09-17 14:09:21 -0700602 higher_buddy = higher_page + (buddy_idx - combined_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700603 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
604 list_add_tail(&page->lru,
605 &zone->free_area[order].free_list[migratetype]);
606 goto out;
607 }
608 }
609
610 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
611out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 zone->free_area[order].nr_free++;
613}
614
Nick Piggin224abf92006-01-06 00:11:11 -0800615static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616{
Nick Piggin92be2e332006-01-06 00:10:57 -0800617 if (unlikely(page_mapcount(page) |
618 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700619 (atomic_read(&page->_count) != 0) |
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -0700620 (page->flags & PAGE_FLAGS_CHECK_AT_FREE) |
621 (mem_cgroup_bad_page_check(page)))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800622 bad_page(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800623 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800624 }
Mel Gorman22b751c2013-02-22 16:34:59 -0800625 page_nid_reset_last(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800626 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
627 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
628 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629}
630
631/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700632 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700634 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 *
636 * If the zone was previously in an "all pages pinned" state then look to
637 * see if this freeing clears that state.
638 *
639 * And clear the zone's pages_scanned counter, to hold off the "all pages are
640 * pinned" detection logic.
641 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700642static void free_pcppages_bulk(struct zone *zone, int count,
643 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644{
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700645 int migratetype = 0;
Mel Gormana6f9edd62009-09-21 17:03:20 -0700646 int batch_free = 0;
Mel Gorman72853e22010-09-09 16:38:16 -0700647 int to_free = count;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700648
Nick Pigginc54ad302006-01-06 00:10:56 -0800649 spin_lock(&zone->lock);
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -0800650 zone->all_unreclaimable = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700652
Mel Gorman72853e22010-09-09 16:38:16 -0700653 while (to_free) {
Nick Piggin48db57f2006-01-08 01:00:42 -0800654 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700655 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -0800656
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700657 /*
Mel Gormana6f9edd62009-09-21 17:03:20 -0700658 * Remove pages from lists in a round-robin fashion. A
659 * batch_free count is maintained that is incremented when an
660 * empty list is encountered. This is so more pages are freed
661 * off fuller lists instead of spinning excessively around empty
662 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700663 */
664 do {
Mel Gormana6f9edd62009-09-21 17:03:20 -0700665 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700666 if (++migratetype == MIGRATE_PCPTYPES)
667 migratetype = 0;
668 list = &pcp->lists[migratetype];
669 } while (list_empty(list));
670
Namhyung Kim1d168712011-03-22 16:32:45 -0700671 /* This is the only non-empty list. Free them all. */
672 if (batch_free == MIGRATE_PCPTYPES)
673 batch_free = to_free;
674
Mel Gormana6f9edd62009-09-21 17:03:20 -0700675 do {
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -0700676 int mt; /* migratetype of the to-be-freed page */
677
Mel Gormana6f9edd62009-09-21 17:03:20 -0700678 page = list_entry(list->prev, struct page, lru);
679 /* must delete as __free_one_page list manipulates */
680 list_del(&page->lru);
Minchan Kimb12c4ad2012-10-08 16:32:08 -0700681 mt = get_freepage_migratetype(page);
Hugh Dickinsa7016232010-01-29 17:46:34 +0000682 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -0700683 __free_one_page(page, zone, 0, mt);
684 trace_mm_page_pcpu_drain(page, 0, mt);
Minchan Kim194159f2013-02-22 16:33:58 -0800685 if (likely(!is_migrate_isolate_page(page))) {
Wen Congyang97d0da22012-12-11 16:00:52 -0800686 __mod_zone_page_state(zone, NR_FREE_PAGES, 1);
687 if (is_migrate_cma(mt))
688 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, 1);
689 }
Mel Gorman72853e22010-09-09 16:38:16 -0700690 } while (--to_free && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800692 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693}
694
Mel Gormaned0ae212009-06-16 15:32:07 -0700695static void free_one_page(struct zone *zone, struct page *page, int order,
696 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800697{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700698 spin_lock(&zone->lock);
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -0800699 zone->all_unreclaimable = 0;
Christoph Lameter006d22d2006-09-25 23:31:48 -0700700 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700701
Mel Gormaned0ae212009-06-16 15:32:07 -0700702 __free_one_page(page, zone, order, migratetype);
Minchan Kim194159f2013-02-22 16:33:58 -0800703 if (unlikely(!is_migrate_isolate(migratetype)))
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -0700704 __mod_zone_freepage_state(zone, 1 << order, migratetype);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700705 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800706}
707
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700708static bool free_pages_prepare(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 int i;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800711 int bad = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -0800713 trace_mm_page_free(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100714 kmemcheck_free_shadow(page, order);
715
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -0800716 if (PageAnon(page))
717 page->mapping = NULL;
718 for (i = 0; i < (1 << order); i++)
719 bad += free_pages_check(page + i);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800720 if (bad)
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700721 return false;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800722
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700723 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -0700724 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700725 debug_check_no_obj_freed(page_address(page),
726 PAGE_SIZE << order);
727 }
Nick Piggindafb1362006-10-11 01:21:30 -0700728 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800729 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700730
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700731 return true;
732}
733
734static void __free_pages_ok(struct page *page, unsigned int order)
735{
736 unsigned long flags;
Minchan Kim95e34412012-10-08 16:32:11 -0700737 int migratetype;
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700738
739 if (!free_pages_prepare(page, order))
740 return;
741
Nick Pigginc54ad302006-01-06 00:10:56 -0800742 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700743 __count_vm_events(PGFREE, 1 << order);
Minchan Kim95e34412012-10-08 16:32:11 -0700744 migratetype = get_pageblock_migratetype(page);
745 set_freepage_migratetype(page, migratetype);
746 free_one_page(page_zone(page), page, order, migratetype);
Nick Pigginc54ad302006-01-06 00:10:56 -0800747 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748}
749
Jiang Liu170a5a72013-07-03 15:03:17 -0700750void __init __free_pages_bootmem(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800751{
Johannes Weinerc3993072012-01-10 15:08:10 -0800752 unsigned int nr_pages = 1 << order;
753 unsigned int loop;
David Howellsa226f6c2006-01-06 00:11:08 -0800754
Johannes Weinerc3993072012-01-10 15:08:10 -0800755 prefetchw(page);
756 for (loop = 0; loop < nr_pages; loop++) {
757 struct page *p = &page[loop];
David Howellsa226f6c2006-01-06 00:11:08 -0800758
Johannes Weinerc3993072012-01-10 15:08:10 -0800759 if (loop + 1 < nr_pages)
760 prefetchw(p + 1);
761 __ClearPageReserved(p);
762 set_page_count(p, 0);
David Howellsa226f6c2006-01-06 00:11:08 -0800763 }
Johannes Weinerc3993072012-01-10 15:08:10 -0800764
Jiang Liu9feedc92012-12-12 13:52:12 -0800765 page_zone(page)->managed_pages += 1 << order;
Johannes Weinerc3993072012-01-10 15:08:10 -0800766 set_page_refcounted(page);
767 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800768}
769
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100770#ifdef CONFIG_CMA
Li Zhong9cf510a2013-08-23 13:52:52 +0800771/* Free whole pageblock and set its migration type to MIGRATE_CMA. */
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100772void __init init_cma_reserved_pageblock(struct page *page)
773{
774 unsigned i = pageblock_nr_pages;
775 struct page *p = page;
776
777 do {
778 __ClearPageReserved(p);
779 set_page_count(p, 0);
780 } while (++p, --i);
781
782 set_page_refcounted(page);
783 set_pageblock_migratetype(page, MIGRATE_CMA);
784 __free_pages(page, pageblock_order);
Jiang Liu3dcc0572013-07-03 15:03:21 -0700785 adjust_managed_page_count(page, pageblock_nr_pages);
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100786}
787#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788
789/*
790 * The order of subdivision here is critical for the IO subsystem.
791 * Please do not alter this order without good reasons and regression
792 * testing. Specifically, as large blocks of memory are subdivided,
793 * the order in which smaller blocks are delivered depends on the order
794 * they're subdivided in this function. This is the primary factor
795 * influencing the order in which pages are delivered to the IO
796 * subsystem according to empirical testing, and this is also justified
797 * by considering the behavior of a buddy system containing a single
798 * large block of memory acted on by a series of small allocations.
799 * This behavior is a critical factor in sglist merging's success.
800 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +0100801 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800803static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700804 int low, int high, struct free_area *area,
805 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806{
807 unsigned long size = 1 << high;
808
809 while (high > low) {
810 area--;
811 high--;
812 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700813 VM_BUG_ON(bad_range(zone, &page[size]));
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800814
815#ifdef CONFIG_DEBUG_PAGEALLOC
816 if (high < debug_guardpage_minorder()) {
817 /*
818 * Mark as guard pages (or page), that will allow to
819 * merge back to allocator when buddy will be freed.
820 * Corresponding page table entries will not be touched,
821 * pages will stay not present in virtual address space
822 */
823 INIT_LIST_HEAD(&page[size].lru);
824 set_page_guard_flag(&page[size]);
825 set_page_private(&page[size], high);
826 /* Guard pages are not available for any usage */
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -0700827 __mod_zone_freepage_state(zone, -(1 << high),
828 migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800829 continue;
830 }
831#endif
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700832 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 area->nr_free++;
834 set_page_order(&page[size], high);
835 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836}
837
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838/*
839 * This page is about to be returned from the page allocator
840 */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200841static inline int check_new_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842{
Nick Piggin92be2e332006-01-06 00:10:57 -0800843 if (unlikely(page_mapcount(page) |
844 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700845 (atomic_read(&page->_count) != 0) |
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -0700846 (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
847 (mem_cgroup_bad_page_check(page)))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800848 bad_page(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800849 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800850 }
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200851 return 0;
852}
853
854static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
855{
856 int i;
857
858 for (i = 0; i < (1 << order); i++) {
859 struct page *p = page + i;
860 if (unlikely(check_new_page(p)))
861 return 1;
862 }
Hugh Dickins689bceb2005-11-21 21:32:20 -0800863
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700864 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800865 set_page_refcounted(page);
Nick Piggincc1025092006-12-06 20:32:00 -0800866
867 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800869
870 if (gfp_flags & __GFP_ZERO)
871 prep_zero_page(page, order, gfp_flags);
872
873 if (order && (gfp_flags & __GFP_COMP))
874 prep_compound_page(page, order);
875
Hugh Dickins689bceb2005-11-21 21:32:20 -0800876 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877}
878
Mel Gorman56fd56b2007-10-16 01:25:58 -0700879/*
880 * Go through the free lists for the given migratetype and remove
881 * the smallest available page from the freelists
882 */
Mel Gorman728ec982009-06-16 15:32:04 -0700883static inline
884struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -0700885 int migratetype)
886{
887 unsigned int current_order;
888 struct free_area * area;
889 struct page *page;
890
891 /* Find a page of the appropriate size in the preferred list */
892 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
893 area = &(zone->free_area[current_order]);
894 if (list_empty(&area->free_list[migratetype]))
895 continue;
896
897 page = list_entry(area->free_list[migratetype].next,
898 struct page, lru);
899 list_del(&page->lru);
900 rmv_page_order(page);
901 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -0700902 expand(zone, page, order, current_order, area, migratetype);
903 return page;
904 }
905
906 return NULL;
907}
908
909
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700910/*
911 * This array describes the order lists are fallen back to when
912 * the free lists for the desirable migrate type are depleted
913 */
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100914static int fallbacks[MIGRATE_TYPES][4] = {
915 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
916 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
917#ifdef CONFIG_CMA
918 [MIGRATE_MOVABLE] = { MIGRATE_CMA, MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
919 [MIGRATE_CMA] = { MIGRATE_RESERVE }, /* Never used */
920#else
921 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
922#endif
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +0100923 [MIGRATE_RESERVE] = { MIGRATE_RESERVE }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -0800924#ifdef CONFIG_MEMORY_ISOLATION
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +0100925 [MIGRATE_ISOLATE] = { MIGRATE_RESERVE }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -0800926#endif
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700927};
928
Mel Gormanc361be52007-10-16 01:25:51 -0700929/*
930 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -0700931 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -0700932 * boundary. If alignment is required, use move_freepages_block()
933 */
Minchan Kim435b4052012-10-08 16:32:16 -0700934int move_freepages(struct zone *zone,
Adrian Bunkb69a7282008-07-23 21:28:12 -0700935 struct page *start_page, struct page *end_page,
936 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700937{
938 struct page *page;
939 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -0700940 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -0700941
942#ifndef CONFIG_HOLES_IN_ZONE
943 /*
944 * page_zone is not safe to call in this context when
945 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
946 * anyway as we check zone boundaries in move_freepages_block().
947 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -0700948 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -0700949 */
950 BUG_ON(page_zone(start_page) != page_zone(end_page));
951#endif
952
953 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -0700954 /* Make sure we are not inadvertently changing nodes */
955 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
956
Mel Gormanc361be52007-10-16 01:25:51 -0700957 if (!pfn_valid_within(page_to_pfn(page))) {
958 page++;
959 continue;
960 }
961
962 if (!PageBuddy(page)) {
963 page++;
964 continue;
965 }
966
967 order = page_order(page);
Kirill A. Shutemov84be48d2011-03-22 16:33:41 -0700968 list_move(&page->lru,
969 &zone->free_area[order].free_list[migratetype]);
Minchan Kim95e34412012-10-08 16:32:11 -0700970 set_freepage_migratetype(page, migratetype);
Mel Gormanc361be52007-10-16 01:25:51 -0700971 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -0700972 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -0700973 }
974
Mel Gormand1003132007-10-16 01:26:00 -0700975 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -0700976}
977
Minchan Kimee6f5092012-07-31 16:43:50 -0700978int move_freepages_block(struct zone *zone, struct page *page,
Linus Torvalds68e3e922012-06-03 20:05:57 -0700979 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700980{
981 unsigned long start_pfn, end_pfn;
982 struct page *start_page, *end_page;
983
984 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -0700985 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -0700986 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -0700987 end_page = start_page + pageblock_nr_pages - 1;
988 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -0700989
990 /* Do not cross zone boundaries */
Cody P Schafer108bcc92013-02-22 16:35:23 -0800991 if (!zone_spans_pfn(zone, start_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -0700992 start_page = page;
Cody P Schafer108bcc92013-02-22 16:35:23 -0800993 if (!zone_spans_pfn(zone, end_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -0700994 return 0;
995
996 return move_freepages(zone, start_page, end_page, migratetype);
997}
998
Mel Gorman2f66a682009-09-21 17:02:31 -0700999static void change_pageblock_range(struct page *pageblock_page,
1000 int start_order, int migratetype)
1001{
1002 int nr_pageblocks = 1 << (start_order - pageblock_order);
1003
1004 while (nr_pageblocks--) {
1005 set_pageblock_migratetype(pageblock_page, migratetype);
1006 pageblock_page += pageblock_nr_pages;
1007 }
1008}
1009
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001010/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -07001011static inline struct page *
1012__rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001013{
1014 struct free_area * area;
1015 int current_order;
1016 struct page *page;
1017 int migratetype, i;
1018
1019 /* Find the largest possible block of pages in the other list */
1020 for (current_order = MAX_ORDER-1; current_order >= order;
1021 --current_order) {
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001022 for (i = 0;; i++) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001023 migratetype = fallbacks[start_migratetype][i];
1024
Mel Gorman56fd56b2007-10-16 01:25:58 -07001025 /* MIGRATE_RESERVE handled later if necessary */
1026 if (migratetype == MIGRATE_RESERVE)
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001027 break;
Mel Gormane0104872007-10-16 01:25:53 -07001028
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001029 area = &(zone->free_area[current_order]);
1030 if (list_empty(&area->free_list[migratetype]))
1031 continue;
1032
1033 page = list_entry(area->free_list[migratetype].next,
1034 struct page, lru);
1035 area->nr_free--;
1036
1037 /*
Mel Gormanc361be52007-10-16 01:25:51 -07001038 * If breaking a large block of pages, move all free
Mel Gorman46dafbc2007-10-16 01:25:55 -07001039 * pages to the preferred allocation list. If falling
1040 * back for a reclaimable kernel allocation, be more
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001041 * aggressive about taking ownership of free pages
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001042 *
1043 * On the other hand, never change migration
1044 * type of MIGRATE_CMA pageblocks nor move CMA
1045 * pages on different free lists. We don't
1046 * want unmovable pages to be allocated from
1047 * MIGRATE_CMA areas.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001048 */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001049 if (!is_migrate_cma(migratetype) &&
Zhang Yanfei345606d2013-07-08 16:00:08 -07001050 (current_order >= pageblock_order / 2 ||
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001051 start_migratetype == MIGRATE_RECLAIMABLE ||
1052 page_group_by_mobility_disabled)) {
1053 int pages;
Mel Gorman46dafbc2007-10-16 01:25:55 -07001054 pages = move_freepages_block(zone, page,
1055 start_migratetype);
1056
1057 /* Claim the whole block if over half of it is free */
Mel Gormandd5d2412009-09-05 11:17:11 -07001058 if (pages >= (1 << (pageblock_order-1)) ||
1059 page_group_by_mobility_disabled)
Mel Gorman46dafbc2007-10-16 01:25:55 -07001060 set_pageblock_migratetype(page,
1061 start_migratetype);
1062
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001063 migratetype = start_migratetype;
Mel Gormanc361be52007-10-16 01:25:51 -07001064 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001065
1066 /* Remove the page from the freelists */
1067 list_del(&page->lru);
1068 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001069
Mel Gorman2f66a682009-09-21 17:02:31 -07001070 /* Take ownership for orders >= pageblock_order */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001071 if (current_order >= pageblock_order &&
1072 !is_migrate_cma(migratetype))
Mel Gorman2f66a682009-09-21 17:02:31 -07001073 change_pageblock_range(page, current_order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001074 start_migratetype);
1075
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001076 expand(zone, page, order, current_order, area,
1077 is_migrate_cma(migratetype)
1078 ? migratetype : start_migratetype);
Mel Gormane0fff1b2009-09-21 17:02:42 -07001079
1080 trace_mm_page_alloc_extfrag(page, order, current_order,
1081 start_migratetype, migratetype);
1082
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001083 return page;
1084 }
1085 }
1086
Mel Gorman728ec982009-06-16 15:32:04 -07001087 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001088}
1089
Mel Gorman56fd56b2007-10-16 01:25:58 -07001090/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 * Do the hard work of removing an element from the buddy allocator.
1092 * Call me with the zone->lock already held.
1093 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001094static struct page *__rmqueue(struct zone *zone, unsigned int order,
1095 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 struct page *page;
1098
Mel Gorman728ec982009-06-16 15:32:04 -07001099retry_reserve:
Mel Gorman56fd56b2007-10-16 01:25:58 -07001100 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101
Mel Gorman728ec982009-06-16 15:32:04 -07001102 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07001103 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001104
Mel Gorman728ec982009-06-16 15:32:04 -07001105 /*
1106 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1107 * is used because __rmqueue_smallest is an inline function
1108 * and we want just one call site
1109 */
1110 if (!page) {
1111 migratetype = MIGRATE_RESERVE;
1112 goto retry_reserve;
1113 }
1114 }
1115
Mel Gorman0d3d0622009-09-21 17:02:44 -07001116 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001117 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118}
1119
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001120/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 * Obtain a specified number of elements from the buddy allocator, all under
1122 * a single hold of the lock, for efficiency. Add them to the supplied list.
1123 * Returns the number of new pages which were placed at *list.
1124 */
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001125static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001126 unsigned long count, struct list_head *list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001127 int migratetype, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128{
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001129 int mt = migratetype, i;
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001130
Nick Pigginc54ad302006-01-06 00:10:56 -08001131 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001133 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001134 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08001136
1137 /*
1138 * Split buddy pages returned by expand() are received here
1139 * in physical page order. The page is added to the callers and
1140 * list and the list head then moves forward. From the callers
1141 * perspective, the linked list is ordered by page number in
1142 * some conditions. This is useful for IO devices that can
1143 * merge IO requests if the physical pages are ordered
1144 * properly.
1145 */
Mel Gormane084b2d2009-07-29 15:02:04 -07001146 if (likely(cold == 0))
1147 list_add(&page->lru, list);
1148 else
1149 list_add_tail(&page->lru, list);
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001150 if (IS_ENABLED(CONFIG_CMA)) {
1151 mt = get_pageblock_migratetype(page);
Minchan Kim194159f2013-02-22 16:33:58 -08001152 if (!is_migrate_cma(mt) && !is_migrate_isolate(mt))
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001153 mt = migratetype;
1154 }
Minchan Kimb12c4ad2012-10-08 16:32:08 -07001155 set_freepage_migratetype(page, mt);
Mel Gorman81eabcb2007-12-17 16:20:05 -08001156 list = &page->lru;
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07001157 if (is_migrate_cma(mt))
1158 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
1159 -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 }
Mel Gormanf2260e62009-06-16 15:32:13 -07001161 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08001162 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001163 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164}
1165
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001166#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001167/*
Christoph Lameter4037d452007-05-09 02:35:14 -07001168 * Called from the vmstat counter updater to drain pagesets of this
1169 * currently executing processor on remote nodes after they have
1170 * expired.
1171 *
Christoph Lameter879336c2006-03-22 00:09:08 -08001172 * Note that this function must be called with the thread pinned to
1173 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001174 */
Christoph Lameter4037d452007-05-09 02:35:14 -07001175void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001176{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001177 unsigned long flags;
Christoph Lameter4037d452007-05-09 02:35:14 -07001178 int to_drain;
Cody P Schafer998d39c2013-07-03 15:01:32 -07001179 unsigned long batch;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001180
Christoph Lameter4037d452007-05-09 02:35:14 -07001181 local_irq_save(flags);
Cody P Schafer998d39c2013-07-03 15:01:32 -07001182 batch = ACCESS_ONCE(pcp->batch);
1183 if (pcp->count >= batch)
1184 to_drain = batch;
Christoph Lameter4037d452007-05-09 02:35:14 -07001185 else
1186 to_drain = pcp->count;
KOSAKI Motohiro2a135152012-07-31 16:42:53 -07001187 if (to_drain > 0) {
1188 free_pcppages_bulk(zone, to_drain, pcp);
1189 pcp->count -= to_drain;
1190 }
Christoph Lameter4037d452007-05-09 02:35:14 -07001191 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001192}
1193#endif
1194
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001195/*
1196 * Drain pages of the indicated processor.
1197 *
1198 * The processor must either be the current processor and the
1199 * thread pinned to the current processor or a processor that
1200 * is not online.
1201 */
1202static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203{
Nick Pigginc54ad302006-01-06 00:10:56 -08001204 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07001207 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 struct per_cpu_pageset *pset;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001209 struct per_cpu_pages *pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001211 local_irq_save(flags);
1212 pset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001214 pcp = &pset->pcp;
David Rientjes2ff754f2011-01-25 15:07:23 -08001215 if (pcp->count) {
1216 free_pcppages_bulk(zone, pcp->count, pcp);
1217 pcp->count = 0;
1218 }
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001219 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 }
1221}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001223/*
1224 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1225 */
1226void drain_local_pages(void *arg)
1227{
1228 drain_pages(smp_processor_id());
1229}
1230
1231/*
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001232 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1233 *
1234 * Note that this code is protected against sending an IPI to an offline
1235 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1236 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1237 * nothing keeps CPUs from showing up after we populated the cpumask and
1238 * before the call to on_each_cpu_mask().
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001239 */
1240void drain_all_pages(void)
1241{
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001242 int cpu;
1243 struct per_cpu_pageset *pcp;
1244 struct zone *zone;
1245
1246 /*
1247 * Allocate in the BSS so we wont require allocation in
1248 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1249 */
1250 static cpumask_t cpus_with_pcps;
1251
1252 /*
1253 * We don't care about racing with CPU hotplug event
1254 * as offline notification will cause the notified
1255 * cpu to drain that CPU pcps and on_each_cpu_mask
1256 * disables preemption as part of its processing
1257 */
1258 for_each_online_cpu(cpu) {
1259 bool has_pcps = false;
1260 for_each_populated_zone(zone) {
1261 pcp = per_cpu_ptr(zone->pageset, cpu);
1262 if (pcp->pcp.count) {
1263 has_pcps = true;
1264 break;
1265 }
1266 }
1267 if (has_pcps)
1268 cpumask_set_cpu(cpu, &cpus_with_pcps);
1269 else
1270 cpumask_clear_cpu(cpu, &cpus_with_pcps);
1271 }
1272 on_each_cpu_mask(&cpus_with_pcps, drain_local_pages, NULL, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001273}
1274
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02001275#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276
1277void mark_free_pages(struct zone *zone)
1278{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001279 unsigned long pfn, max_zone_pfn;
1280 unsigned long flags;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001281 int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 struct list_head *curr;
1283
1284 if (!zone->spanned_pages)
1285 return;
1286
1287 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001288
Cody P Schafer108bcc92013-02-22 16:35:23 -08001289 max_zone_pfn = zone_end_pfn(zone);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001290 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1291 if (pfn_valid(pfn)) {
1292 struct page *page = pfn_to_page(pfn);
1293
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001294 if (!swsusp_page_is_forbidden(page))
1295 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001296 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001298 for_each_migratetype_order(order, t) {
1299 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001300 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001302 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1303 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001304 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001305 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001306 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 spin_unlock_irqrestore(&zone->lock, flags);
1308}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001309#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310
1311/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 * Free a 0-order page
Li Hongfc916682010-03-05 13:41:54 -08001313 * cold == 1 ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 */
Li Hongfc916682010-03-05 13:41:54 -08001315void free_hot_cold_page(struct page *page, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316{
1317 struct zone *zone = page_zone(page);
1318 struct per_cpu_pages *pcp;
1319 unsigned long flags;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001320 int migratetype;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001322 if (!free_pages_prepare(page, 0))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001323 return;
1324
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001325 migratetype = get_pageblock_migratetype(page);
Minchan Kimb12c4ad2012-10-08 16:32:08 -07001326 set_freepage_migratetype(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001328 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001329
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001330 /*
1331 * We only track unmovable, reclaimable and movable on pcp lists.
1332 * Free ISOLATE pages back to the allocator because they are being
1333 * offlined but treat RESERVE as movable pages so we can get those
1334 * areas back if necessary. Otherwise, we may have to free
1335 * excessively into the page allocator
1336 */
1337 if (migratetype >= MIGRATE_PCPTYPES) {
Minchan Kim194159f2013-02-22 16:33:58 -08001338 if (unlikely(is_migrate_isolate(migratetype))) {
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001339 free_one_page(zone, page, 0, migratetype);
1340 goto out;
1341 }
1342 migratetype = MIGRATE_MOVABLE;
1343 }
1344
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001345 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001346 if (cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001347 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001348 else
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001349 list_add(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001351 if (pcp->count >= pcp->high) {
Cody P Schafer998d39c2013-07-03 15:01:32 -07001352 unsigned long batch = ACCESS_ONCE(pcp->batch);
1353 free_pcppages_bulk(zone, batch, pcp);
1354 pcp->count -= batch;
Nick Piggin48db57f2006-01-08 01:00:42 -08001355 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001356
1357out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359}
1360
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001361/*
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001362 * Free a list of 0-order pages
1363 */
1364void free_hot_cold_page_list(struct list_head *list, int cold)
1365{
1366 struct page *page, *next;
1367
1368 list_for_each_entry_safe(page, next, list, lru) {
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -08001369 trace_mm_page_free_batched(page, cold);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001370 free_hot_cold_page(page, cold);
1371 }
1372}
1373
1374/*
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001375 * split_page takes a non-compound higher-order page, and splits it into
1376 * n (1<<order) sub-pages: page[0..n]
1377 * Each sub-page must be freed individually.
1378 *
1379 * Note: this is probably too low level an operation for use in drivers.
1380 * Please consult with lkml before using this in your driver.
1381 */
1382void split_page(struct page *page, unsigned int order)
1383{
1384 int i;
1385
Nick Piggin725d7042006-09-25 23:30:55 -07001386 VM_BUG_ON(PageCompound(page));
1387 VM_BUG_ON(!page_count(page));
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001388
1389#ifdef CONFIG_KMEMCHECK
1390 /*
1391 * Split shadow pages too, because free(page[0]) would
1392 * otherwise free the whole shadow.
1393 */
1394 if (kmemcheck_page_is_tracked(page))
1395 split_page(virt_to_page(page[0].shadow), order);
1396#endif
1397
Nick Piggin7835e982006-03-22 00:08:40 -08001398 for (i = 1; i < (1 << order); i++)
1399 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001400}
K. Y. Srinivasan5853ff22013-03-25 15:47:38 -07001401EXPORT_SYMBOL_GPL(split_page);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001402
Mel Gorman8fb74b92013-01-11 14:32:16 -08001403static int __isolate_free_page(struct page *page, unsigned int order)
Mel Gorman748446b2010-05-24 14:32:27 -07001404{
Mel Gorman748446b2010-05-24 14:32:27 -07001405 unsigned long watermark;
1406 struct zone *zone;
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07001407 int mt;
Mel Gorman748446b2010-05-24 14:32:27 -07001408
1409 BUG_ON(!PageBuddy(page));
1410
1411 zone = page_zone(page);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001412 mt = get_pageblock_migratetype(page);
Mel Gorman748446b2010-05-24 14:32:27 -07001413
Minchan Kim194159f2013-02-22 16:33:58 -08001414 if (!is_migrate_isolate(mt)) {
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001415 /* Obey watermarks as if the page was being allocated */
1416 watermark = low_wmark_pages(zone) + (1 << order);
1417 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1418 return 0;
1419
Mel Gorman8fb74b92013-01-11 14:32:16 -08001420 __mod_zone_freepage_state(zone, -(1UL << order), mt);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001421 }
Mel Gorman748446b2010-05-24 14:32:27 -07001422
1423 /* Remove page from free list */
1424 list_del(&page->lru);
1425 zone->free_area[order].nr_free--;
1426 rmv_page_order(page);
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07001427
Mel Gorman8fb74b92013-01-11 14:32:16 -08001428 /* Set the pageblock if the isolated page is at least a pageblock */
Mel Gorman748446b2010-05-24 14:32:27 -07001429 if (order >= pageblock_order - 1) {
1430 struct page *endpage = page + (1 << order) - 1;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001431 for (; page < endpage; page += pageblock_nr_pages) {
1432 int mt = get_pageblock_migratetype(page);
Minchan Kim194159f2013-02-22 16:33:58 -08001433 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001434 set_pageblock_migratetype(page,
1435 MIGRATE_MOVABLE);
1436 }
Mel Gorman748446b2010-05-24 14:32:27 -07001437 }
1438
Mel Gorman8fb74b92013-01-11 14:32:16 -08001439 return 1UL << order;
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001440}
1441
1442/*
1443 * Similar to split_page except the page is already free. As this is only
1444 * being used for migration, the migratetype of the block also changes.
1445 * As this is called with interrupts disabled, the caller is responsible
1446 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1447 * are enabled.
1448 *
1449 * Note: this is probably too low level an operation for use in drivers.
1450 * Please consult with lkml before using this in your driver.
1451 */
1452int split_free_page(struct page *page)
1453{
1454 unsigned int order;
1455 int nr_pages;
1456
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001457 order = page_order(page);
1458
Mel Gorman8fb74b92013-01-11 14:32:16 -08001459 nr_pages = __isolate_free_page(page, order);
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001460 if (!nr_pages)
1461 return 0;
1462
1463 /* Split into individual pages */
1464 set_page_refcounted(page);
1465 split_page(page, order);
1466 return nr_pages;
Mel Gorman748446b2010-05-24 14:32:27 -07001467}
1468
1469/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1471 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1472 * or two.
1473 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07001474static inline
1475struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001476 struct zone *zone, int order, gfp_t gfp_flags,
1477 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478{
1479 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001480 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 int cold = !!(gfp_flags & __GFP_COLD);
1482
Hugh Dickins689bceb2005-11-21 21:32:20 -08001483again:
Nick Piggin48db57f2006-01-08 01:00:42 -08001484 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001486 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 local_irq_save(flags);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001489 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1490 list = &pcp->lists[migratetype];
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001491 if (list_empty(list)) {
Mel Gorman535131e62007-10-16 01:25:49 -07001492 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001493 pcp->batch, list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001494 migratetype, cold);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001495 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07001496 goto failed;
Mel Gorman535131e62007-10-16 01:25:49 -07001497 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001498
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001499 if (cold)
1500 page = list_entry(list->prev, struct page, lru);
1501 else
1502 page = list_entry(list->next, struct page, lru);
1503
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001504 list_del(&page->lru);
1505 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001506 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07001507 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1508 /*
1509 * __GFP_NOFAIL is not to be used in new code.
1510 *
1511 * All __GFP_NOFAIL callers should be fixed so that they
1512 * properly detect and handle allocation failures.
1513 *
1514 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07001515 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07001516 * __GFP_NOFAIL.
1517 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07001518 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07001519 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001521 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001522 spin_unlock(&zone->lock);
1523 if (!page)
1524 goto failed;
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07001525 __mod_zone_freepage_state(zone, -(1 << order),
1526 get_pageblock_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 }
1528
Christoph Lameterf8891e52006-06-30 01:55:45 -07001529 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Andi Kleen78afd562011-03-22 16:33:12 -07001530 zone_statistics(preferred_zone, zone, gfp_flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001531 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532
Nick Piggin725d7042006-09-25 23:30:55 -07001533 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -08001534 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001535 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001537
1538failed:
1539 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001540 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541}
1542
Akinobu Mita933e3122006-12-08 02:39:45 -08001543#ifdef CONFIG_FAIL_PAGE_ALLOC
1544
Akinobu Mitab2588c42011-07-26 16:09:03 -07001545static struct {
Akinobu Mita933e3122006-12-08 02:39:45 -08001546 struct fault_attr attr;
1547
1548 u32 ignore_gfp_highmem;
1549 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001550 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001551} fail_page_alloc = {
1552 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001553 .ignore_gfp_wait = 1,
1554 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001555 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001556};
1557
1558static int __init setup_fail_page_alloc(char *str)
1559{
1560 return setup_fault_attr(&fail_page_alloc.attr, str);
1561}
1562__setup("fail_page_alloc=", setup_fail_page_alloc);
1563
Gavin Shandeaf3862012-07-31 16:41:51 -07001564static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08001565{
Akinobu Mita54114992007-07-15 23:40:23 -07001566 if (order < fail_page_alloc.min_order)
Gavin Shandeaf3862012-07-31 16:41:51 -07001567 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001568 if (gfp_mask & __GFP_NOFAIL)
Gavin Shandeaf3862012-07-31 16:41:51 -07001569 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001570 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
Gavin Shandeaf3862012-07-31 16:41:51 -07001571 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001572 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
Gavin Shandeaf3862012-07-31 16:41:51 -07001573 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001574
1575 return should_fail(&fail_page_alloc.attr, 1 << order);
1576}
1577
1578#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1579
1580static int __init fail_page_alloc_debugfs(void)
1581{
Al Virof4ae40a2011-07-24 04:33:43 -04001582 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
Akinobu Mita933e3122006-12-08 02:39:45 -08001583 struct dentry *dir;
Akinobu Mita933e3122006-12-08 02:39:45 -08001584
Akinobu Mitadd48c082011-08-03 16:21:01 -07001585 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
1586 &fail_page_alloc.attr);
1587 if (IS_ERR(dir))
1588 return PTR_ERR(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08001589
Akinobu Mitab2588c42011-07-26 16:09:03 -07001590 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
1591 &fail_page_alloc.ignore_gfp_wait))
1592 goto fail;
1593 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1594 &fail_page_alloc.ignore_gfp_highmem))
1595 goto fail;
1596 if (!debugfs_create_u32("min-order", mode, dir,
1597 &fail_page_alloc.min_order))
1598 goto fail;
Akinobu Mita933e3122006-12-08 02:39:45 -08001599
Akinobu Mitab2588c42011-07-26 16:09:03 -07001600 return 0;
1601fail:
Akinobu Mitadd48c082011-08-03 16:21:01 -07001602 debugfs_remove_recursive(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08001603
Akinobu Mitab2588c42011-07-26 16:09:03 -07001604 return -ENOMEM;
Akinobu Mita933e3122006-12-08 02:39:45 -08001605}
1606
1607late_initcall(fail_page_alloc_debugfs);
1608
1609#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1610
1611#else /* CONFIG_FAIL_PAGE_ALLOC */
1612
Gavin Shandeaf3862012-07-31 16:41:51 -07001613static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08001614{
Gavin Shandeaf3862012-07-31 16:41:51 -07001615 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001616}
1617
1618#endif /* CONFIG_FAIL_PAGE_ALLOC */
1619
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620/*
Mel Gorman88f5acf2011-01-13 15:45:41 -08001621 * Return true if free pages are above 'mark'. This takes into account the order
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 * of the allocation.
1623 */
Mel Gorman88f5acf2011-01-13 15:45:41 -08001624static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1625 int classzone_idx, int alloc_flags, long free_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626{
1627 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001628 long min = mark;
Minchan Kim2cfed072012-07-31 16:43:53 -07001629 long lowmem_reserve = z->lowmem_reserve[classzone_idx];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 int o;
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07001631 long free_cma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632
Michal Hockodf0a6da2012-01-10 15:08:02 -08001633 free_pages -= (1 << order) - 1;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001634 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001636 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 min -= min / 4;
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07001638#ifdef CONFIG_CMA
1639 /* If allocation can't use CMA areas don't use free CMA pages */
1640 if (!(alloc_flags & ALLOC_CMA))
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07001641 free_cma = zone_page_state(z, NR_FREE_CMA_PAGES);
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07001642#endif
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07001643
1644 if (free_pages - free_cma <= min + lowmem_reserve)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001645 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 for (o = 0; o < order; o++) {
1647 /* At the next order, this order's pages become unavailable */
1648 free_pages -= z->free_area[o].nr_free << o;
1649
1650 /* Require fewer higher order pages to be free */
1651 min >>= 1;
1652
1653 if (free_pages <= min)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001654 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 }
Mel Gorman88f5acf2011-01-13 15:45:41 -08001656 return true;
1657}
1658
1659bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1660 int classzone_idx, int alloc_flags)
1661{
1662 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1663 zone_page_state(z, NR_FREE_PAGES));
1664}
1665
1666bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
1667 int classzone_idx, int alloc_flags)
1668{
1669 long free_pages = zone_page_state(z, NR_FREE_PAGES);
1670
1671 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1672 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1673
1674 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1675 free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676}
1677
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001678#ifdef CONFIG_NUMA
1679/*
1680 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1681 * skip over zones that are not allowed by the cpuset, or that have
1682 * been recently (in last second) found to be nearly full. See further
1683 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001684 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001685 *
1686 * If the zonelist cache is present in the passed in zonelist, then
1687 * returns a pointer to the allowed node mask (either the current
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08001688 * tasks mems_allowed, or node_states[N_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001689 *
1690 * If the zonelist cache is not available for this zonelist, does
1691 * nothing and returns NULL.
1692 *
1693 * If the fullzones BITMAP in the zonelist cache is stale (more than
1694 * a second since last zap'd) then we zap it out (clear its bits.)
1695 *
1696 * We hold off even calling zlc_setup, until after we've checked the
1697 * first zone in the zonelist, on the theory that most allocations will
1698 * be satisfied from that first zone, so best to examine that zone as
1699 * quickly as we can.
1700 */
1701static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1702{
1703 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1704 nodemask_t *allowednodes; /* zonelist_cache approximation */
1705
1706 zlc = zonelist->zlcache_ptr;
1707 if (!zlc)
1708 return NULL;
1709
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001710 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001711 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1712 zlc->last_full_zap = jiffies;
1713 }
1714
1715 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1716 &cpuset_current_mems_allowed :
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08001717 &node_states[N_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001718 return allowednodes;
1719}
1720
1721/*
1722 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1723 * if it is worth looking at further for free memory:
1724 * 1) Check that the zone isn't thought to be full (doesn't have its
1725 * bit set in the zonelist_cache fullzones BITMAP).
1726 * 2) Check that the zones node (obtained from the zonelist_cache
1727 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1728 * Return true (non-zero) if zone is worth looking at further, or
1729 * else return false (zero) if it is not.
1730 *
1731 * This check -ignores- the distinction between various watermarks,
1732 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1733 * found to be full for any variation of these watermarks, it will
1734 * be considered full for up to one second by all requests, unless
1735 * we are so low on memory on all allowed nodes that we are forced
1736 * into the second scan of the zonelist.
1737 *
1738 * In the second scan we ignore this zonelist cache and exactly
1739 * apply the watermarks to all zones, even it is slower to do so.
1740 * We are low on memory in the second scan, and should leave no stone
1741 * unturned looking for a free page.
1742 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001743static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001744 nodemask_t *allowednodes)
1745{
1746 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1747 int i; /* index of *z in zonelist zones */
1748 int n; /* node that zone *z is on */
1749
1750 zlc = zonelist->zlcache_ptr;
1751 if (!zlc)
1752 return 1;
1753
Mel Gormandd1a2392008-04-28 02:12:17 -07001754 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001755 n = zlc->z_to_n[i];
1756
1757 /* This zone is worth trying if it is allowed but not full */
1758 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1759}
1760
1761/*
1762 * Given 'z' scanning a zonelist, set the corresponding bit in
1763 * zlc->fullzones, so that subsequent attempts to allocate a page
1764 * from that zone don't waste time re-examining it.
1765 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001766static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001767{
1768 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1769 int i; /* index of *z in zonelist zones */
1770
1771 zlc = zonelist->zlcache_ptr;
1772 if (!zlc)
1773 return;
1774
Mel Gormandd1a2392008-04-28 02:12:17 -07001775 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001776
1777 set_bit(i, zlc->fullzones);
1778}
1779
Mel Gorman76d3fbf2011-07-25 17:12:30 -07001780/*
1781 * clear all zones full, called after direct reclaim makes progress so that
1782 * a zone that was recently full is not skipped over for up to a second
1783 */
1784static void zlc_clear_zones_full(struct zonelist *zonelist)
1785{
1786 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1787
1788 zlc = zonelist->zlcache_ptr;
1789 if (!zlc)
1790 return;
1791
1792 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1793}
1794
David Rientjes957f8222012-10-08 16:33:24 -07001795static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
1796{
1797 return node_isset(local_zone->node, zone->zone_pgdat->reclaim_nodes);
1798}
1799
1800static void __paginginit init_zone_allows_reclaim(int nid)
1801{
1802 int i;
1803
1804 for_each_online_node(i)
David Rientjes6b187d022012-10-25 13:38:08 -07001805 if (node_distance(nid, i) <= RECLAIM_DISTANCE)
David Rientjes957f8222012-10-08 16:33:24 -07001806 node_set(i, NODE_DATA(nid)->reclaim_nodes);
David Rientjes6b187d022012-10-25 13:38:08 -07001807 else
David Rientjes957f8222012-10-08 16:33:24 -07001808 zone_reclaim_mode = 1;
David Rientjes957f8222012-10-08 16:33:24 -07001809}
1810
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001811#else /* CONFIG_NUMA */
1812
1813static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1814{
1815 return NULL;
1816}
1817
Mel Gormandd1a2392008-04-28 02:12:17 -07001818static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001819 nodemask_t *allowednodes)
1820{
1821 return 1;
1822}
1823
Mel Gormandd1a2392008-04-28 02:12:17 -07001824static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001825{
1826}
Mel Gorman76d3fbf2011-07-25 17:12:30 -07001827
1828static void zlc_clear_zones_full(struct zonelist *zonelist)
1829{
1830}
David Rientjes957f8222012-10-08 16:33:24 -07001831
1832static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
1833{
1834 return true;
1835}
1836
1837static inline void init_zone_allows_reclaim(int nid)
1838{
1839}
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001840#endif /* CONFIG_NUMA */
1841
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001842/*
Paul Jackson0798e512006-12-06 20:31:38 -08001843 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001844 * a page.
1845 */
1846static struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001847get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
Mel Gorman5117f452009-06-16 15:31:59 -07001848 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
Mel Gorman3dd28262009-06-16 15:32:00 -07001849 struct zone *preferred_zone, int migratetype)
Martin Hicks753ee722005-06-21 17:14:41 -07001850{
Mel Gormandd1a2392008-04-28 02:12:17 -07001851 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001852 struct page *page = NULL;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001853 int classzone_idx;
Mel Gorman5117f452009-06-16 15:31:59 -07001854 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001855 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1856 int zlc_active = 0; /* set if using zonelist_cache */
1857 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gorman54a6eb52008-04-28 02:12:16 -07001858
Mel Gorman19770b32008-04-28 02:12:18 -07001859 classzone_idx = zone_idx(preferred_zone);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001860zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001861 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001862 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001863 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1864 */
Mel Gorman19770b32008-04-28 02:12:18 -07001865 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1866 high_zoneidx, nodemask) {
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08001867 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001868 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1869 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001870 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001871 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Mel Gormancd38b112011-07-25 17:12:29 -07001872 continue;
Johannes Weinera756cf52012-01-10 15:07:49 -08001873 /*
1874 * When allocating a page cache page for writing, we
1875 * want to get it from a zone that is within its dirty
1876 * limit, such that no single zone holds more than its
1877 * proportional share of globally allowed dirty pages.
1878 * The dirty limits take into account the zone's
1879 * lowmem reserves and high watermark so that kswapd
1880 * should be able to balance it without having to
1881 * write pages from its LRU list.
1882 *
1883 * This may look like it could increase pressure on
1884 * lower zones by failing allocations in higher zones
1885 * before they are full. But the pages that do spill
1886 * over are limited as the lower zones are protected
1887 * by this very same mechanism. It should not become
1888 * a practical burden to them.
1889 *
1890 * XXX: For now, allow allocations to potentially
1891 * exceed the per-zone dirty limit in the slowpath
1892 * (ALLOC_WMARK_LOW unset) before going into reclaim,
1893 * which is important when on a NUMA setup the allowed
1894 * zones are together not big enough to reach the
1895 * global limit. The proper fix for these situations
1896 * will require awareness of zones in the
1897 * dirty-throttling and the flusher threads.
1898 */
1899 if ((alloc_flags & ALLOC_WMARK_LOW) &&
1900 (gfp_mask & __GFP_WRITE) && !zone_dirty_ok(zone))
1901 goto this_zone_full;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001902
Mel Gorman41858962009-06-16 15:32:12 -07001903 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001904 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001905 unsigned long mark;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001906 int ret;
1907
Mel Gorman41858962009-06-16 15:32:12 -07001908 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
Mel Gormanfa5e0842009-06-16 15:33:22 -07001909 if (zone_watermark_ok(zone, order, mark,
1910 classzone_idx, alloc_flags))
1911 goto try_this_zone;
1912
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08001913 if (IS_ENABLED(CONFIG_NUMA) &&
1914 !did_zlc_setup && nr_online_nodes > 1) {
Mel Gormancd38b112011-07-25 17:12:29 -07001915 /*
1916 * we do zlc_setup if there are multiple nodes
1917 * and before considering the first zone allowed
1918 * by the cpuset.
1919 */
1920 allowednodes = zlc_setup(zonelist, alloc_flags);
1921 zlc_active = 1;
1922 did_zlc_setup = 1;
1923 }
1924
David Rientjes957f8222012-10-08 16:33:24 -07001925 if (zone_reclaim_mode == 0 ||
1926 !zone_allows_reclaim(preferred_zone, zone))
Mel Gormanfa5e0842009-06-16 15:33:22 -07001927 goto this_zone_full;
1928
Mel Gormancd38b112011-07-25 17:12:29 -07001929 /*
1930 * As we may have just activated ZLC, check if the first
1931 * eligible zone has failed zone_reclaim recently.
1932 */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08001933 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
Mel Gormancd38b112011-07-25 17:12:29 -07001934 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1935 continue;
1936
Mel Gormanfa5e0842009-06-16 15:33:22 -07001937 ret = zone_reclaim(zone, gfp_mask, order);
1938 switch (ret) {
1939 case ZONE_RECLAIM_NOSCAN:
1940 /* did not scan */
Mel Gormancd38b112011-07-25 17:12:29 -07001941 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001942 case ZONE_RECLAIM_FULL:
1943 /* scanned but unreclaimable */
Mel Gormancd38b112011-07-25 17:12:29 -07001944 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001945 default:
1946 /* did we reclaim enough */
Mel Gormanfed27192013-04-29 15:07:57 -07001947 if (zone_watermark_ok(zone, order, mark,
Mel Gormanfa5e0842009-06-16 15:33:22 -07001948 classzone_idx, alloc_flags))
Mel Gormanfed27192013-04-29 15:07:57 -07001949 goto try_this_zone;
1950
1951 /*
1952 * Failed to reclaim enough to meet watermark.
1953 * Only mark the zone full if checking the min
1954 * watermark or if we failed to reclaim just
1955 * 1<<order pages or else the page allocator
1956 * fastpath will prematurely mark zones full
1957 * when the watermark is between the low and
1958 * min watermarks.
1959 */
1960 if (((alloc_flags & ALLOC_WMARK_MASK) == ALLOC_WMARK_MIN) ||
1961 ret == ZONE_RECLAIM_SOME)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001962 goto this_zone_full;
Mel Gormanfed27192013-04-29 15:07:57 -07001963
1964 continue;
Paul Jackson0798e512006-12-06 20:31:38 -08001965 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001966 }
1967
Mel Gormanfa5e0842009-06-16 15:33:22 -07001968try_this_zone:
Mel Gorman3dd28262009-06-16 15:32:00 -07001969 page = buffered_rmqueue(preferred_zone, zone, order,
1970 gfp_mask, migratetype);
Paul Jackson0798e512006-12-06 20:31:38 -08001971 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001972 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001973this_zone_full:
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08001974 if (IS_ENABLED(CONFIG_NUMA))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001975 zlc_mark_zone_full(zonelist, z);
Mel Gorman54a6eb52008-04-28 02:12:16 -07001976 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001977
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08001978 if (unlikely(IS_ENABLED(CONFIG_NUMA) && page == NULL && zlc_active)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001979 /* Disable zlc cache for second zonelist scan */
1980 zlc_active = 0;
1981 goto zonelist_scan;
1982 }
Alex Shib1211862012-08-21 16:16:08 -07001983
1984 if (page)
1985 /*
1986 * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was
1987 * necessary to allocate the page. The expectation is
1988 * that the caller is taking steps that will free more
1989 * memory. The caller should avoid the page being used
1990 * for !PFMEMALLOC purposes.
1991 */
1992 page->pfmemalloc = !!(alloc_flags & ALLOC_NO_WATERMARKS);
1993
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001994 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07001995}
1996
David Rientjes29423e772011-03-22 16:30:47 -07001997/*
1998 * Large machines with many possible nodes should not always dump per-node
1999 * meminfo in irq context.
2000 */
2001static inline bool should_suppress_show_mem(void)
2002{
2003 bool ret = false;
2004
2005#if NODES_SHIFT > 8
2006 ret = in_interrupt();
2007#endif
2008 return ret;
2009}
2010
Dave Hansena238ab52011-05-24 17:12:16 -07002011static DEFINE_RATELIMIT_STATE(nopage_rs,
2012 DEFAULT_RATELIMIT_INTERVAL,
2013 DEFAULT_RATELIMIT_BURST);
2014
2015void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
2016{
Dave Hansena238ab52011-05-24 17:12:16 -07002017 unsigned int filter = SHOW_MEM_FILTER_NODES;
2018
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -08002019 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2020 debug_guardpage_minorder() > 0)
Dave Hansena238ab52011-05-24 17:12:16 -07002021 return;
2022
2023 /*
David Rientjes4b59e6c2013-04-29 15:06:11 -07002024 * Walking all memory to count page types is very expensive and should
2025 * be inhibited in non-blockable contexts.
2026 */
2027 if (!(gfp_mask & __GFP_WAIT))
2028 filter |= SHOW_MEM_FILTER_PAGE_COUNT;
2029
2030 /*
Dave Hansena238ab52011-05-24 17:12:16 -07002031 * This documents exceptions given to allocations in certain
2032 * contexts that are allowed to allocate outside current's set
2033 * of allowed nodes.
2034 */
2035 if (!(gfp_mask & __GFP_NOMEMALLOC))
2036 if (test_thread_flag(TIF_MEMDIE) ||
2037 (current->flags & (PF_MEMALLOC | PF_EXITING)))
2038 filter &= ~SHOW_MEM_FILTER_NODES;
2039 if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
2040 filter &= ~SHOW_MEM_FILTER_NODES;
2041
2042 if (fmt) {
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002043 struct va_format vaf;
2044 va_list args;
2045
Dave Hansena238ab52011-05-24 17:12:16 -07002046 va_start(args, fmt);
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002047
2048 vaf.fmt = fmt;
2049 vaf.va = &args;
2050
2051 pr_warn("%pV", &vaf);
2052
Dave Hansena238ab52011-05-24 17:12:16 -07002053 va_end(args);
2054 }
2055
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002056 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2057 current->comm, order, gfp_mask);
Dave Hansena238ab52011-05-24 17:12:16 -07002058
2059 dump_stack();
2060 if (!should_suppress_show_mem())
2061 show_mem(filter);
2062}
2063
Mel Gorman11e33f62009-06-16 15:31:57 -07002064static inline int
2065should_alloc_retry(gfp_t gfp_mask, unsigned int order,
Mel Gormanf90ac392012-01-10 15:07:15 -08002066 unsigned long did_some_progress,
Mel Gorman11e33f62009-06-16 15:31:57 -07002067 unsigned long pages_reclaimed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068{
Mel Gorman11e33f62009-06-16 15:31:57 -07002069 /* Do not loop if specifically requested */
2070 if (gfp_mask & __GFP_NORETRY)
2071 return 0;
2072
Mel Gormanf90ac392012-01-10 15:07:15 -08002073 /* Always retry if specifically requested */
2074 if (gfp_mask & __GFP_NOFAIL)
2075 return 1;
2076
2077 /*
2078 * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim
2079 * making forward progress without invoking OOM. Suspend also disables
2080 * storage devices so kswapd will not help. Bail if we are suspending.
2081 */
2082 if (!did_some_progress && pm_suspended_storage())
2083 return 0;
2084
Mel Gorman11e33f62009-06-16 15:31:57 -07002085 /*
2086 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
2087 * means __GFP_NOFAIL, but that may not be true in other
2088 * implementations.
2089 */
2090 if (order <= PAGE_ALLOC_COSTLY_ORDER)
2091 return 1;
2092
2093 /*
2094 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
2095 * specified, then we retry until we no longer reclaim any pages
2096 * (above), or we've reclaimed an order of pages at least as
2097 * large as the allocation's order. In both cases, if the
2098 * allocation still fails, we stop retrying.
2099 */
2100 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
2101 return 1;
2102
Mel Gorman11e33f62009-06-16 15:31:57 -07002103 return 0;
2104}
2105
2106static inline struct page *
2107__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
2108 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002109 nodemask_t *nodemask, struct zone *preferred_zone,
2110 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002111{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113
Mel Gorman11e33f62009-06-16 15:31:57 -07002114 /* Acquire the OOM killer lock for the zones in zonelist */
Minchan Kimff321fe2010-08-09 17:18:57 -07002115 if (!try_set_zonelist_oom(zonelist, gfp_mask)) {
Mel Gorman11e33f62009-06-16 15:31:57 -07002116 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 return NULL;
2118 }
Jens Axboe6b1de912005-11-17 21:35:02 +01002119
Mel Gorman11e33f62009-06-16 15:31:57 -07002120 /*
2121 * Go through the zonelist yet one more time, keep very high watermark
2122 * here, this is only to catch a parallel oom killing, we must fail if
2123 * we're still under heavy pressure.
2124 */
2125 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
2126 order, zonelist, high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07002127 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07002128 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002129 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07002130 goto out;
2131
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002132 if (!(gfp_mask & __GFP_NOFAIL)) {
2133 /* The OOM killer will not help higher order allocs */
2134 if (order > PAGE_ALLOC_COSTLY_ORDER)
2135 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07002136 /* The OOM killer does not needlessly kill tasks for lowmem */
2137 if (high_zoneidx < ZONE_NORMAL)
2138 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002139 /*
2140 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
2141 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
2142 * The caller should handle page allocation failure by itself if
2143 * it specifies __GFP_THISNODE.
2144 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
2145 */
2146 if (gfp_mask & __GFP_THISNODE)
2147 goto out;
2148 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002149 /* Exhausted what can be done so it's blamo time */
David Rientjes08ab9b12012-03-21 16:34:04 -07002150 out_of_memory(zonelist, gfp_mask, order, nodemask, false);
Mel Gorman11e33f62009-06-16 15:31:57 -07002151
2152out:
2153 clear_zonelist_oom(zonelist, gfp_mask);
2154 return page;
2155}
2156
Mel Gorman56de7262010-05-24 14:32:30 -07002157#ifdef CONFIG_COMPACTION
2158/* Try memory compaction for high-order allocations before reclaim */
2159static struct page *
2160__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2161 struct zonelist *zonelist, enum zone_type high_zoneidx,
2162 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman66199712012-01-12 17:19:41 -08002163 int migratetype, bool sync_migration,
Mel Gormanc67fe372012-08-21 16:16:17 -07002164 bool *contended_compaction, bool *deferred_compaction,
Mel Gorman66199712012-01-12 17:19:41 -08002165 unsigned long *did_some_progress)
Mel Gorman56de7262010-05-24 14:32:30 -07002166{
Mel Gorman66199712012-01-12 17:19:41 -08002167 if (!order)
Mel Gorman56de7262010-05-24 14:32:30 -07002168 return NULL;
2169
Rik van Rielaff62242012-03-21 16:33:52 -07002170 if (compaction_deferred(preferred_zone, order)) {
Mel Gorman66199712012-01-12 17:19:41 -08002171 *deferred_compaction = true;
2172 return NULL;
2173 }
2174
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002175 current->flags |= PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07002176 *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
Mel Gormanc67fe372012-08-21 16:16:17 -07002177 nodemask, sync_migration,
Mel Gorman8fb74b92013-01-11 14:32:16 -08002178 contended_compaction);
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002179 current->flags &= ~PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07002180
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07002181 if (*did_some_progress != COMPACT_SKIPPED) {
Mel Gorman8fb74b92013-01-11 14:32:16 -08002182 struct page *page;
2183
Mel Gorman56de7262010-05-24 14:32:30 -07002184 /* Page migration frees to the PCP lists but we want merging */
2185 drain_pages(get_cpu());
2186 put_cpu();
2187
2188 page = get_page_from_freelist(gfp_mask, nodemask,
2189 order, zonelist, high_zoneidx,
Mel Gormancfd19c52012-07-31 16:44:10 -07002190 alloc_flags & ~ALLOC_NO_WATERMARKS,
2191 preferred_zone, migratetype);
Mel Gorman56de7262010-05-24 14:32:30 -07002192 if (page) {
Mel Gorman62997022012-10-08 16:32:47 -07002193 preferred_zone->compact_blockskip_flush = false;
Mel Gorman4f92e252010-05-24 14:32:32 -07002194 preferred_zone->compact_considered = 0;
2195 preferred_zone->compact_defer_shift = 0;
Rik van Rielaff62242012-03-21 16:33:52 -07002196 if (order >= preferred_zone->compact_order_failed)
2197 preferred_zone->compact_order_failed = order + 1;
Mel Gorman56de7262010-05-24 14:32:30 -07002198 count_vm_event(COMPACTSUCCESS);
2199 return page;
2200 }
2201
2202 /*
2203 * It's bad if compaction run occurs and fails.
2204 * The most likely reason is that pages exist,
2205 * but not enough to satisfy watermarks.
2206 */
2207 count_vm_event(COMPACTFAIL);
Mel Gorman66199712012-01-12 17:19:41 -08002208
2209 /*
2210 * As async compaction considers a subset of pageblocks, only
2211 * defer if the failure was a sync compaction failure.
2212 */
2213 if (sync_migration)
Rik van Rielaff62242012-03-21 16:33:52 -07002214 defer_compaction(preferred_zone, order);
Mel Gorman56de7262010-05-24 14:32:30 -07002215
2216 cond_resched();
2217 }
2218
2219 return NULL;
2220}
2221#else
2222static inline struct page *
2223__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2224 struct zonelist *zonelist, enum zone_type high_zoneidx,
2225 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman66199712012-01-12 17:19:41 -08002226 int migratetype, bool sync_migration,
Mel Gormanc67fe372012-08-21 16:16:17 -07002227 bool *contended_compaction, bool *deferred_compaction,
Mel Gorman66199712012-01-12 17:19:41 -08002228 unsigned long *did_some_progress)
Mel Gorman56de7262010-05-24 14:32:30 -07002229{
2230 return NULL;
2231}
2232#endif /* CONFIG_COMPACTION */
2233
Marek Szyprowskibba90712012-01-25 12:09:52 +01002234/* Perform direct synchronous page reclaim */
2235static int
2236__perform_reclaim(gfp_t gfp_mask, unsigned int order, struct zonelist *zonelist,
2237 nodemask_t *nodemask)
Mel Gorman11e33f62009-06-16 15:31:57 -07002238{
Mel Gorman11e33f62009-06-16 15:31:57 -07002239 struct reclaim_state reclaim_state;
Marek Szyprowskibba90712012-01-25 12:09:52 +01002240 int progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07002241
2242 cond_resched();
2243
2244 /* We now go into synchronous reclaim */
2245 cpuset_memory_pressure_bump();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002246 current->flags |= PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002247 lockdep_set_current_reclaim_state(gfp_mask);
2248 reclaim_state.reclaimed_slab = 0;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002249 current->reclaim_state = &reclaim_state;
Mel Gorman11e33f62009-06-16 15:31:57 -07002250
Marek Szyprowskibba90712012-01-25 12:09:52 +01002251 progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002252
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002253 current->reclaim_state = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002254 lockdep_clear_current_reclaim_state();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002255 current->flags &= ~PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002256
2257 cond_resched();
2258
Marek Szyprowskibba90712012-01-25 12:09:52 +01002259 return progress;
2260}
2261
2262/* The really slow allocator path where we enter direct reclaim */
2263static inline struct page *
2264__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
2265 struct zonelist *zonelist, enum zone_type high_zoneidx,
2266 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
2267 int migratetype, unsigned long *did_some_progress)
2268{
2269 struct page *page = NULL;
2270 bool drained = false;
2271
2272 *did_some_progress = __perform_reclaim(gfp_mask, order, zonelist,
2273 nodemask);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002274 if (unlikely(!(*did_some_progress)))
2275 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002276
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002277 /* After successful reclaim, reconsider all zones for allocation */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002278 if (IS_ENABLED(CONFIG_NUMA))
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002279 zlc_clear_zones_full(zonelist);
2280
Mel Gorman9ee493c2010-09-09 16:38:18 -07002281retry:
2282 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002283 zonelist, high_zoneidx,
Mel Gormancfd19c52012-07-31 16:44:10 -07002284 alloc_flags & ~ALLOC_NO_WATERMARKS,
2285 preferred_zone, migratetype);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002286
2287 /*
2288 * If an allocation failed after direct reclaim, it could be because
2289 * pages are pinned on the per-cpu lists. Drain them and try again
2290 */
2291 if (!page && !drained) {
2292 drain_all_pages();
2293 drained = true;
2294 goto retry;
2295 }
2296
Mel Gorman11e33f62009-06-16 15:31:57 -07002297 return page;
2298}
2299
Mel Gorman11e33f62009-06-16 15:31:57 -07002300/*
2301 * This is called in the allocator slow-path if the allocation request is of
2302 * sufficient urgency to ignore watermarks and take other desperate measures
2303 */
2304static inline struct page *
2305__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
2306 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002307 nodemask_t *nodemask, struct zone *preferred_zone,
2308 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002309{
2310 struct page *page;
2311
2312 do {
2313 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002314 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
Mel Gorman3dd28262009-06-16 15:32:00 -07002315 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002316
2317 if (!page && gfp_mask & __GFP_NOFAIL)
Mel Gorman0e093d992010-10-26 14:21:45 -07002318 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07002319 } while (!page && (gfp_mask & __GFP_NOFAIL));
2320
2321 return page;
2322}
2323
2324static inline
2325void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
Mel Gorman99504742011-01-13 15:46:20 -08002326 enum zone_type high_zoneidx,
2327 enum zone_type classzone_idx)
Mel Gorman11e33f62009-06-16 15:31:57 -07002328{
2329 struct zoneref *z;
2330 struct zone *zone;
2331
2332 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
Mel Gorman99504742011-01-13 15:46:20 -08002333 wakeup_kswapd(zone, order, classzone_idx);
Mel Gorman11e33f62009-06-16 15:31:57 -07002334}
2335
Peter Zijlstra341ce062009-06-16 15:32:02 -07002336static inline int
2337gfp_to_alloc_flags(gfp_t gfp_mask)
2338{
Peter Zijlstra341ce062009-06-16 15:32:02 -07002339 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
2340 const gfp_t wait = gfp_mask & __GFP_WAIT;
2341
Mel Gormana56f57f2009-06-16 15:32:02 -07002342 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07002343 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07002344
Peter Zijlstra341ce062009-06-16 15:32:02 -07002345 /*
2346 * The caller may dip into page reserves a bit more if the caller
2347 * cannot run direct reclaim, or if the caller has realtime scheduling
2348 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2349 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
2350 */
Namhyung Kime6223a32010-10-26 14:21:59 -07002351 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07002352
2353 if (!wait) {
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002354 /*
2355 * Not worth trying to allocate harder for
2356 * __GFP_NOMEMALLOC even if it can't schedule.
2357 */
2358 if (!(gfp_mask & __GFP_NOMEMALLOC))
2359 alloc_flags |= ALLOC_HARDER;
Peter Zijlstra341ce062009-06-16 15:32:02 -07002360 /*
2361 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
2362 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
2363 */
2364 alloc_flags &= ~ALLOC_CPUSET;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002365 } else if (unlikely(rt_task(current)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07002366 alloc_flags |= ALLOC_HARDER;
2367
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002368 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2369 if (gfp_mask & __GFP_MEMALLOC)
2370 alloc_flags |= ALLOC_NO_WATERMARKS;
Mel Gorman907aed42012-07-31 16:44:07 -07002371 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
2372 alloc_flags |= ALLOC_NO_WATERMARKS;
2373 else if (!in_interrupt() &&
2374 ((current->flags & PF_MEMALLOC) ||
2375 unlikely(test_thread_flag(TIF_MEMDIE))))
Peter Zijlstra341ce062009-06-16 15:32:02 -07002376 alloc_flags |= ALLOC_NO_WATERMARKS;
2377 }
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002378#ifdef CONFIG_CMA
2379 if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2380 alloc_flags |= ALLOC_CMA;
2381#endif
Peter Zijlstra341ce062009-06-16 15:32:02 -07002382 return alloc_flags;
2383}
2384
Mel Gorman072bb0a2012-07-31 16:43:58 -07002385bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
2386{
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002387 return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
Mel Gorman072bb0a2012-07-31 16:43:58 -07002388}
2389
Mel Gorman11e33f62009-06-16 15:31:57 -07002390static inline struct page *
2391__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
2392 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002393 nodemask_t *nodemask, struct zone *preferred_zone,
2394 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002395{
2396 const gfp_t wait = gfp_mask & __GFP_WAIT;
2397 struct page *page = NULL;
2398 int alloc_flags;
2399 unsigned long pages_reclaimed = 0;
2400 unsigned long did_some_progress;
Mel Gorman77f1fe62011-01-13 15:45:57 -08002401 bool sync_migration = false;
Mel Gorman66199712012-01-12 17:19:41 -08002402 bool deferred_compaction = false;
Mel Gormanc67fe372012-08-21 16:16:17 -07002403 bool contended_compaction = false;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002404
Christoph Lameter952f3b52006-12-06 20:33:26 -08002405 /*
Mel Gorman72807a72009-06-16 15:32:18 -07002406 * In the slowpath, we sanity check order to avoid ever trying to
2407 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2408 * be using allocators in order of preference for an area that is
2409 * too large.
2410 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07002411 if (order >= MAX_ORDER) {
2412 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07002413 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07002414 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002415
Christoph Lameter952f3b52006-12-06 20:33:26 -08002416 /*
2417 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2418 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2419 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2420 * using a larger set of nodes after it has established that the
2421 * allowed per node queues are empty and that nodes are
2422 * over allocated.
2423 */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002424 if (IS_ENABLED(CONFIG_NUMA) &&
2425 (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
Christoph Lameter952f3b52006-12-06 20:33:26 -08002426 goto nopage;
2427
Mel Gormancc4a6852009-11-11 14:26:14 -08002428restart:
Linus Torvaldscaf49192012-12-10 10:51:16 -08002429 if (!(gfp_mask & __GFP_NO_KSWAPD))
2430 wake_all_kswapd(order, zonelist, high_zoneidx,
2431 zone_idx(preferred_zone));
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002432
Paul Jackson9bf22292005-09-06 15:18:12 -07002433 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002434 * OK, we're below the kswapd watermark and have kicked background
2435 * reclaim. Now things get more complex, so set up alloc_flags according
2436 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07002437 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002438 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439
David Rientjesf33261d2011-01-25 15:07:20 -08002440 /*
2441 * Find the true preferred zone if the allocation is unconstrained by
2442 * cpusets.
2443 */
2444 if (!(alloc_flags & ALLOC_CPUSET) && !nodemask)
2445 first_zones_zonelist(zonelist, high_zoneidx, NULL,
2446 &preferred_zone);
2447
Andrew Barrycfa54a02011-05-24 17:12:52 -07002448rebalance:
Peter Zijlstra341ce062009-06-16 15:32:02 -07002449 /* This is the last chance, in general, before the goto nopage. */
Mel Gorman19770b32008-04-28 02:12:18 -07002450 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
Peter Zijlstra341ce062009-06-16 15:32:02 -07002451 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
2452 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002453 if (page)
2454 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455
Mel Gorman11e33f62009-06-16 15:31:57 -07002456 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002457 if (alloc_flags & ALLOC_NO_WATERMARKS) {
Mel Gorman183f6372012-07-31 16:44:12 -07002458 /*
2459 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
2460 * the allocation is high priority and these type of
2461 * allocations are system rather than user orientated
2462 */
2463 zonelist = node_zonelist(numa_node_id(), gfp_mask);
2464
Peter Zijlstra341ce062009-06-16 15:32:02 -07002465 page = __alloc_pages_high_priority(gfp_mask, order,
2466 zonelist, high_zoneidx, nodemask,
2467 preferred_zone, migratetype);
Mel Gormancfd19c52012-07-31 16:44:10 -07002468 if (page) {
Peter Zijlstra341ce062009-06-16 15:32:02 -07002469 goto got_pg;
Mel Gormancfd19c52012-07-31 16:44:10 -07002470 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471 }
2472
2473 /* Atomic allocations - we can't balance anything */
2474 if (!wait)
2475 goto nopage;
2476
Peter Zijlstra341ce062009-06-16 15:32:02 -07002477 /* Avoid recursion of direct reclaim */
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002478 if (current->flags & PF_MEMALLOC)
Peter Zijlstra341ce062009-06-16 15:32:02 -07002479 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480
David Rientjes6583bb62009-07-29 15:02:06 -07002481 /* Avoid allocations with no watermarks from looping endlessly */
2482 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2483 goto nopage;
2484
Mel Gorman77f1fe62011-01-13 15:45:57 -08002485 /*
2486 * Try direct compaction. The first pass is asynchronous. Subsequent
2487 * attempts after direct reclaim are synchronous
2488 */
Mel Gorman56de7262010-05-24 14:32:30 -07002489 page = __alloc_pages_direct_compact(gfp_mask, order,
2490 zonelist, high_zoneidx,
2491 nodemask,
2492 alloc_flags, preferred_zone,
Mel Gorman66199712012-01-12 17:19:41 -08002493 migratetype, sync_migration,
Mel Gormanc67fe372012-08-21 16:16:17 -07002494 &contended_compaction,
Mel Gorman66199712012-01-12 17:19:41 -08002495 &deferred_compaction,
2496 &did_some_progress);
Mel Gorman56de7262010-05-24 14:32:30 -07002497 if (page)
2498 goto got_pg;
Andrea Arcangelic6a140b2011-05-24 17:11:38 -07002499 sync_migration = true;
Mel Gorman56de7262010-05-24 14:32:30 -07002500
Linus Torvalds31f8d422012-12-10 10:47:45 -08002501 /*
2502 * If compaction is deferred for high-order allocations, it is because
2503 * sync compaction recently failed. In this is the case and the caller
2504 * requested a movable allocation that does not heavily disrupt the
2505 * system then fail the allocation instead of entering direct reclaim.
2506 */
2507 if ((deferred_compaction || contended_compaction) &&
Linus Torvaldscaf49192012-12-10 10:51:16 -08002508 (gfp_mask & __GFP_NO_KSWAPD))
Linus Torvalds31f8d422012-12-10 10:47:45 -08002509 goto nopage;
Mel Gorman66199712012-01-12 17:19:41 -08002510
Mel Gorman11e33f62009-06-16 15:31:57 -07002511 /* Try direct reclaim and then allocating */
2512 page = __alloc_pages_direct_reclaim(gfp_mask, order,
2513 zonelist, high_zoneidx,
2514 nodemask,
Mel Gorman5117f452009-06-16 15:31:59 -07002515 alloc_flags, preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07002516 migratetype, &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07002517 if (page)
2518 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520 /*
Mel Gorman11e33f62009-06-16 15:31:57 -07002521 * If we failed to make any progress reclaiming, then we are
2522 * running out of options and have to consider going OOM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523 */
Mel Gorman11e33f62009-06-16 15:31:57 -07002524 if (!did_some_progress) {
2525 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -07002526 if (oom_killer_disabled)
2527 goto nopage;
David Rientjes29fd66d2012-03-28 14:42:41 -07002528 /* Coredumps can quickly deplete all memory reserves */
2529 if ((current->flags & PF_DUMPCORE) &&
2530 !(gfp_mask & __GFP_NOFAIL))
2531 goto nopage;
Mel Gorman11e33f62009-06-16 15:31:57 -07002532 page = __alloc_pages_may_oom(gfp_mask, order,
2533 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002534 nodemask, preferred_zone,
2535 migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002536 if (page)
2537 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538
David Rientjes03668b32010-08-09 17:18:54 -07002539 if (!(gfp_mask & __GFP_NOFAIL)) {
2540 /*
2541 * The oom killer is not called for high-order
2542 * allocations that may fail, so if no progress
2543 * is being made, there are no other options and
2544 * retrying is unlikely to help.
2545 */
2546 if (order > PAGE_ALLOC_COSTLY_ORDER)
2547 goto nopage;
2548 /*
2549 * The oom killer is not called for lowmem
2550 * allocations to prevent needlessly killing
2551 * innocent tasks.
2552 */
2553 if (high_zoneidx < ZONE_NORMAL)
2554 goto nopage;
2555 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557 goto restart;
2558 }
2559 }
2560
Mel Gorman11e33f62009-06-16 15:31:57 -07002561 /* Check if we should retry the allocation */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07002562 pages_reclaimed += did_some_progress;
Mel Gormanf90ac392012-01-10 15:07:15 -08002563 if (should_alloc_retry(gfp_mask, order, did_some_progress,
2564 pages_reclaimed)) {
Mel Gorman11e33f62009-06-16 15:31:57 -07002565 /* Wait for some write requests to complete then retry */
Mel Gorman0e093d992010-10-26 14:21:45 -07002566 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567 goto rebalance;
Mel Gorman3e7d3442011-01-13 15:45:56 -08002568 } else {
2569 /*
2570 * High-order allocations do not necessarily loop after
2571 * direct reclaim and reclaim/compaction depends on compaction
2572 * being called after reclaim so call directly if necessary
2573 */
2574 page = __alloc_pages_direct_compact(gfp_mask, order,
2575 zonelist, high_zoneidx,
2576 nodemask,
2577 alloc_flags, preferred_zone,
Mel Gorman66199712012-01-12 17:19:41 -08002578 migratetype, sync_migration,
Mel Gormanc67fe372012-08-21 16:16:17 -07002579 &contended_compaction,
Mel Gorman66199712012-01-12 17:19:41 -08002580 &deferred_compaction,
2581 &did_some_progress);
Mel Gorman3e7d3442011-01-13 15:45:56 -08002582 if (page)
2583 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 }
2585
2586nopage:
Dave Hansena238ab52011-05-24 17:12:16 -07002587 warn_alloc_failed(gfp_mask, order, NULL);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002588 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589got_pg:
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002590 if (kmemcheck_enabled)
2591 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002592
Mel Gorman072bb0a2012-07-31 16:43:58 -07002593 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594}
Mel Gorman11e33f62009-06-16 15:31:57 -07002595
2596/*
2597 * This is the 'heart' of the zoned buddy allocator.
2598 */
2599struct page *
2600__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2601 struct zonelist *zonelist, nodemask_t *nodemask)
2602{
2603 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Mel Gorman5117f452009-06-16 15:31:59 -07002604 struct zone *preferred_zone;
Mel Gormancc9a6c82012-03-21 16:34:11 -07002605 struct page *page = NULL;
Mel Gorman3dd28262009-06-16 15:32:00 -07002606 int migratetype = allocflags_to_migratetype(gfp_mask);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002607 unsigned int cpuset_mems_cookie;
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002608 int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET;
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002609 struct mem_cgroup *memcg = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002610
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10002611 gfp_mask &= gfp_allowed_mask;
2612
Mel Gorman11e33f62009-06-16 15:31:57 -07002613 lockdep_trace_alloc(gfp_mask);
2614
2615 might_sleep_if(gfp_mask & __GFP_WAIT);
2616
2617 if (should_fail_alloc_page(gfp_mask, order))
2618 return NULL;
2619
2620 /*
2621 * Check the zones suitable for the gfp_mask contain at least one
2622 * valid zone. It's possible to have an empty zonelist as a result
2623 * of GFP_THISNODE and a memoryless node
2624 */
2625 if (unlikely(!zonelist->_zonerefs->zone))
2626 return NULL;
2627
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002628 /*
2629 * Will only have any effect when __GFP_KMEMCG is set. This is
2630 * verified in the (always inline) callee
2631 */
2632 if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
2633 return NULL;
2634
Mel Gormancc9a6c82012-03-21 16:34:11 -07002635retry_cpuset:
2636 cpuset_mems_cookie = get_mems_allowed();
2637
Mel Gorman5117f452009-06-16 15:31:59 -07002638 /* The preferred zone is used for statistics later */
David Rientjesf33261d2011-01-25 15:07:20 -08002639 first_zones_zonelist(zonelist, high_zoneidx,
2640 nodemask ? : &cpuset_current_mems_allowed,
2641 &preferred_zone);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002642 if (!preferred_zone)
2643 goto out;
Mel Gorman5117f452009-06-16 15:31:59 -07002644
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002645#ifdef CONFIG_CMA
2646 if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2647 alloc_flags |= ALLOC_CMA;
2648#endif
Mel Gorman5117f452009-06-16 15:31:59 -07002649 /* First allocation attempt */
Mel Gorman11e33f62009-06-16 15:31:57 -07002650 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002651 zonelist, high_zoneidx, alloc_flags,
Mel Gorman3dd28262009-06-16 15:32:00 -07002652 preferred_zone, migratetype);
Ming Lei21caf2f2013-02-22 16:34:08 -08002653 if (unlikely(!page)) {
2654 /*
2655 * Runtime PM, block IO and its error handling path
2656 * can deadlock because I/O on the device might not
2657 * complete.
2658 */
2659 gfp_mask = memalloc_noio_flags(gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002660 page = __alloc_pages_slowpath(gfp_mask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002661 zonelist, high_zoneidx, nodemask,
Mel Gorman3dd28262009-06-16 15:32:00 -07002662 preferred_zone, migratetype);
Ming Lei21caf2f2013-02-22 16:34:08 -08002663 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002664
Mel Gorman4b4f2782009-09-21 17:02:41 -07002665 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002666
2667out:
2668 /*
2669 * When updating a task's mems_allowed, it is possible to race with
2670 * parallel threads in such a way that an allocation can fail while
2671 * the mask is being updated. If a page allocation is about to fail,
2672 * check if the cpuset changed during allocation and if so, retry.
2673 */
2674 if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page))
2675 goto retry_cpuset;
2676
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002677 memcg_kmem_commit_charge(page, memcg, order);
2678
Mel Gorman11e33f62009-06-16 15:31:57 -07002679 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680}
Mel Gormand2391712009-06-16 15:31:52 -07002681EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682
2683/*
2684 * Common helper functions.
2685 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08002686unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687{
Akinobu Mita945a1112009-09-21 17:01:47 -07002688 struct page *page;
2689
2690 /*
2691 * __get_free_pages() returns a 32-bit address, which cannot represent
2692 * a highmem page
2693 */
2694 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2695
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696 page = alloc_pages(gfp_mask, order);
2697 if (!page)
2698 return 0;
2699 return (unsigned long) page_address(page);
2700}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002701EXPORT_SYMBOL(__get_free_pages);
2702
Harvey Harrison920c7a52008-02-04 22:29:26 -08002703unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704{
Akinobu Mita945a1112009-09-21 17:01:47 -07002705 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707EXPORT_SYMBOL(get_zeroed_page);
2708
Harvey Harrison920c7a52008-02-04 22:29:26 -08002709void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710{
Nick Pigginb5810032005-10-29 18:16:12 -07002711 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712 if (order == 0)
Li Hongfc916682010-03-05 13:41:54 -08002713 free_hot_cold_page(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714 else
2715 __free_pages_ok(page, order);
2716 }
2717}
2718
2719EXPORT_SYMBOL(__free_pages);
2720
Harvey Harrison920c7a52008-02-04 22:29:26 -08002721void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722{
2723 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07002724 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725 __free_pages(virt_to_page((void *)addr), order);
2726 }
2727}
2728
2729EXPORT_SYMBOL(free_pages);
2730
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002731/*
2732 * __free_memcg_kmem_pages and free_memcg_kmem_pages will free
2733 * pages allocated with __GFP_KMEMCG.
2734 *
2735 * Those pages are accounted to a particular memcg, embedded in the
2736 * corresponding page_cgroup. To avoid adding a hit in the allocator to search
2737 * for that information only to find out that it is NULL for users who have no
2738 * interest in that whatsoever, we provide these functions.
2739 *
2740 * The caller knows better which flags it relies on.
2741 */
2742void __free_memcg_kmem_pages(struct page *page, unsigned int order)
2743{
2744 memcg_kmem_uncharge_pages(page, order);
2745 __free_pages(page, order);
2746}
2747
2748void free_memcg_kmem_pages(unsigned long addr, unsigned int order)
2749{
2750 if (addr != 0) {
2751 VM_BUG_ON(!virt_addr_valid((void *)addr));
2752 __free_memcg_kmem_pages(virt_to_page((void *)addr), order);
2753 }
2754}
2755
Andi Kleenee85c2e2011-05-11 15:13:34 -07002756static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
2757{
2758 if (addr) {
2759 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2760 unsigned long used = addr + PAGE_ALIGN(size);
2761
2762 split_page(virt_to_page((void *)addr), order);
2763 while (used < alloc_end) {
2764 free_page(used);
2765 used += PAGE_SIZE;
2766 }
2767 }
2768 return (void *)addr;
2769}
2770
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002771/**
2772 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2773 * @size: the number of bytes to allocate
2774 * @gfp_mask: GFP flags for the allocation
2775 *
2776 * This function is similar to alloc_pages(), except that it allocates the
2777 * minimum number of pages to satisfy the request. alloc_pages() can only
2778 * allocate memory in power-of-two pages.
2779 *
2780 * This function is also limited by MAX_ORDER.
2781 *
2782 * Memory allocated by this function must be released by free_pages_exact().
2783 */
2784void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2785{
2786 unsigned int order = get_order(size);
2787 unsigned long addr;
2788
2789 addr = __get_free_pages(gfp_mask, order);
Andi Kleenee85c2e2011-05-11 15:13:34 -07002790 return make_alloc_exact(addr, order, size);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002791}
2792EXPORT_SYMBOL(alloc_pages_exact);
2793
2794/**
Andi Kleenee85c2e2011-05-11 15:13:34 -07002795 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2796 * pages on a node.
Randy Dunlapb5e6ab52011-05-16 13:16:54 -07002797 * @nid: the preferred node ID where memory should be allocated
Andi Kleenee85c2e2011-05-11 15:13:34 -07002798 * @size: the number of bytes to allocate
2799 * @gfp_mask: GFP flags for the allocation
2800 *
2801 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2802 * back.
2803 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2804 * but is not exact.
2805 */
2806void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
2807{
2808 unsigned order = get_order(size);
2809 struct page *p = alloc_pages_node(nid, gfp_mask, order);
2810 if (!p)
2811 return NULL;
2812 return make_alloc_exact((unsigned long)page_address(p), order, size);
2813}
2814EXPORT_SYMBOL(alloc_pages_exact_nid);
2815
2816/**
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002817 * free_pages_exact - release memory allocated via alloc_pages_exact()
2818 * @virt: the value returned by alloc_pages_exact.
2819 * @size: size of allocation, same value as passed to alloc_pages_exact().
2820 *
2821 * Release the memory allocated by a previous call to alloc_pages_exact.
2822 */
2823void free_pages_exact(void *virt, size_t size)
2824{
2825 unsigned long addr = (unsigned long)virt;
2826 unsigned long end = addr + PAGE_ALIGN(size);
2827
2828 while (addr < end) {
2829 free_page(addr);
2830 addr += PAGE_SIZE;
2831 }
2832}
2833EXPORT_SYMBOL(free_pages_exact);
2834
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08002835/**
2836 * nr_free_zone_pages - count number of pages beyond high watermark
2837 * @offset: The zone index of the highest zone
2838 *
2839 * nr_free_zone_pages() counts the number of counts pages which are beyond the
2840 * high watermark within all zones at or below a given zone index. For each
2841 * zone, the number of pages is calculated as:
Jiang Liu834405c2013-07-03 15:03:04 -07002842 * managed_pages - high_pages
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08002843 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08002844static unsigned long nr_free_zone_pages(int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845{
Mel Gormandd1a2392008-04-28 02:12:17 -07002846 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002847 struct zone *zone;
2848
Martin J. Blighe310fd42005-07-29 22:59:18 -07002849 /* Just pick one node, since fallback list is circular */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08002850 unsigned long sum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851
Mel Gorman0e884602008-04-28 02:12:14 -07002852 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002853
Mel Gorman54a6eb52008-04-28 02:12:16 -07002854 for_each_zone_zonelist(zone, z, zonelist, offset) {
Jiang Liub40da042013-02-22 16:33:52 -08002855 unsigned long size = zone->managed_pages;
Mel Gorman41858962009-06-16 15:32:12 -07002856 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07002857 if (size > high)
2858 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002859 }
2860
2861 return sum;
2862}
2863
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08002864/**
2865 * nr_free_buffer_pages - count number of pages beyond high watermark
2866 *
2867 * nr_free_buffer_pages() counts the number of pages which are beyond the high
2868 * watermark within ZONE_DMA and ZONE_NORMAL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002869 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08002870unsigned long nr_free_buffer_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871{
Al Viroaf4ca452005-10-21 02:55:38 -04002872 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002873}
Meelap Shahc2f1a552007-07-17 04:04:39 -07002874EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08002876/**
2877 * nr_free_pagecache_pages - count number of pages beyond high watermark
2878 *
2879 * nr_free_pagecache_pages() counts the number of pages which are beyond the
2880 * high watermark within all zones.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08002882unsigned long nr_free_pagecache_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883{
Mel Gorman2a1e2742007-07-17 04:03:12 -07002884 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002886
2887static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002888{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002889 if (IS_ENABLED(CONFIG_NUMA))
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08002890 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002892
Linus Torvalds1da177e2005-04-16 15:20:36 -07002893void si_meminfo(struct sysinfo *val)
2894{
2895 val->totalram = totalram_pages;
2896 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002897 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899 val->totalhigh = totalhigh_pages;
2900 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002901 val->mem_unit = PAGE_SIZE;
2902}
2903
2904EXPORT_SYMBOL(si_meminfo);
2905
2906#ifdef CONFIG_NUMA
2907void si_meminfo_node(struct sysinfo *val, int nid)
2908{
Jiang Liucdd91a72013-07-03 15:03:27 -07002909 int zone_type; /* needs to be signed */
2910 unsigned long managed_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002911 pg_data_t *pgdat = NODE_DATA(nid);
2912
Jiang Liucdd91a72013-07-03 15:03:27 -07002913 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
2914 managed_pages += pgdat->node_zones[zone_type].managed_pages;
2915 val->totalram = managed_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002916 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002917#ifdef CONFIG_HIGHMEM
Jiang Liub40da042013-02-22 16:33:52 -08002918 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002919 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2920 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002921#else
2922 val->totalhigh = 0;
2923 val->freehigh = 0;
2924#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002925 val->mem_unit = PAGE_SIZE;
2926}
2927#endif
2928
David Rientjesddd588b2011-03-22 16:30:46 -07002929/*
David Rientjes7bf02ea2011-05-24 17:11:16 -07002930 * Determine whether the node should be displayed or not, depending on whether
2931 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
David Rientjesddd588b2011-03-22 16:30:46 -07002932 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07002933bool skip_free_areas_node(unsigned int flags, int nid)
David Rientjesddd588b2011-03-22 16:30:46 -07002934{
2935 bool ret = false;
Mel Gormancc9a6c82012-03-21 16:34:11 -07002936 unsigned int cpuset_mems_cookie;
David Rientjesddd588b2011-03-22 16:30:46 -07002937
2938 if (!(flags & SHOW_MEM_FILTER_NODES))
2939 goto out;
2940
Mel Gormancc9a6c82012-03-21 16:34:11 -07002941 do {
2942 cpuset_mems_cookie = get_mems_allowed();
2943 ret = !node_isset(nid, cpuset_current_mems_allowed);
2944 } while (!put_mems_allowed(cpuset_mems_cookie));
David Rientjesddd588b2011-03-22 16:30:46 -07002945out:
2946 return ret;
2947}
2948
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949#define K(x) ((x) << (PAGE_SHIFT-10))
2950
Rabin Vincent377e4f12012-12-11 16:00:24 -08002951static void show_migration_types(unsigned char type)
2952{
2953 static const char types[MIGRATE_TYPES] = {
2954 [MIGRATE_UNMOVABLE] = 'U',
2955 [MIGRATE_RECLAIMABLE] = 'E',
2956 [MIGRATE_MOVABLE] = 'M',
2957 [MIGRATE_RESERVE] = 'R',
2958#ifdef CONFIG_CMA
2959 [MIGRATE_CMA] = 'C',
2960#endif
Minchan Kim194159f2013-02-22 16:33:58 -08002961#ifdef CONFIG_MEMORY_ISOLATION
Rabin Vincent377e4f12012-12-11 16:00:24 -08002962 [MIGRATE_ISOLATE] = 'I',
Minchan Kim194159f2013-02-22 16:33:58 -08002963#endif
Rabin Vincent377e4f12012-12-11 16:00:24 -08002964 };
2965 char tmp[MIGRATE_TYPES + 1];
2966 char *p = tmp;
2967 int i;
2968
2969 for (i = 0; i < MIGRATE_TYPES; i++) {
2970 if (type & (1 << i))
2971 *p++ = types[i];
2972 }
2973
2974 *p = '\0';
2975 printk("(%s) ", tmp);
2976}
2977
Linus Torvalds1da177e2005-04-16 15:20:36 -07002978/*
2979 * Show free area list (used inside shift_scroll-lock stuff)
2980 * We also calculate the percentage fragmentation. We do this by counting the
2981 * memory on each free list with the exception of the first item on the list.
David Rientjesddd588b2011-03-22 16:30:46 -07002982 * Suppresses nodes that are not allowed by current's cpuset if
2983 * SHOW_MEM_FILTER_NODES is passed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07002985void show_free_areas(unsigned int filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002986{
Jes Sorensenc7241912006-09-27 01:50:05 -07002987 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002988 struct zone *zone;
2989
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002990 for_each_populated_zone(zone) {
David Rientjes7bf02ea2011-05-24 17:11:16 -07002991 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07002992 continue;
Jes Sorensenc7241912006-09-27 01:50:05 -07002993 show_node(zone);
2994 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995
Dave Jones6b482c62005-11-10 15:45:56 -05002996 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002997 struct per_cpu_pageset *pageset;
2998
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002999 pageset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003000
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003001 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
3002 cpu, pageset->pcp.high,
3003 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004 }
3005 }
3006
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003007 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3008 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003009 " unevictable:%lu"
Andrew Mortonb76146e2009-10-26 16:49:52 -07003010 " dirty:%lu writeback:%lu unstable:%lu\n"
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07003011 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003012 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3013 " free_cma:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003014 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003015 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003016 global_page_state(NR_ISOLATED_ANON),
3017 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003018 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003019 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003020 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07003021 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07003022 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07003023 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08003024 global_page_state(NR_FREE_PAGES),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07003025 global_page_state(NR_SLAB_RECLAIMABLE),
3026 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07003027 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003028 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08003029 global_page_state(NR_PAGETABLE),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003030 global_page_state(NR_BOUNCE),
3031 global_page_state(NR_FREE_CMA_PAGES));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003032
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003033 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003034 int i;
3035
David Rientjes7bf02ea2011-05-24 17:11:16 -07003036 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003037 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003038 show_node(zone);
3039 printk("%s"
3040 " free:%lukB"
3041 " min:%lukB"
3042 " low:%lukB"
3043 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003044 " active_anon:%lukB"
3045 " inactive_anon:%lukB"
3046 " active_file:%lukB"
3047 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003048 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003049 " isolated(anon):%lukB"
3050 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003051 " present:%lukB"
Jiang Liu9feedc92012-12-12 13:52:12 -08003052 " managed:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003053 " mlocked:%lukB"
3054 " dirty:%lukB"
3055 " writeback:%lukB"
3056 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003057 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003058 " slab_reclaimable:%lukB"
3059 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003060 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003061 " pagetables:%lukB"
3062 " unstable:%lukB"
3063 " bounce:%lukB"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003064 " free_cma:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003065 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003066 " pages_scanned:%lu"
3067 " all_unreclaimable? %s"
3068 "\n",
3069 zone->name,
Mel Gorman88f5acf2011-01-13 15:45:41 -08003070 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07003071 K(min_wmark_pages(zone)),
3072 K(low_wmark_pages(zone)),
3073 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003074 K(zone_page_state(zone, NR_ACTIVE_ANON)),
3075 K(zone_page_state(zone, NR_INACTIVE_ANON)),
3076 K(zone_page_state(zone, NR_ACTIVE_FILE)),
3077 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003078 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003079 K(zone_page_state(zone, NR_ISOLATED_ANON)),
3080 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003081 K(zone->present_pages),
Jiang Liu9feedc92012-12-12 13:52:12 -08003082 K(zone->managed_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003083 K(zone_page_state(zone, NR_MLOCK)),
3084 K(zone_page_state(zone, NR_FILE_DIRTY)),
3085 K(zone_page_state(zone, NR_WRITEBACK)),
3086 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003087 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003088 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3089 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003090 zone_page_state(zone, NR_KERNEL_STACK) *
3091 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003092 K(zone_page_state(zone, NR_PAGETABLE)),
3093 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3094 K(zone_page_state(zone, NR_BOUNCE)),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003095 K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003096 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003097 zone->pages_scanned,
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -08003098 (zone->all_unreclaimable ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07003099 );
3100 printk("lowmem_reserve[]:");
3101 for (i = 0; i < MAX_NR_ZONES; i++)
3102 printk(" %lu", zone->lowmem_reserve[i]);
3103 printk("\n");
3104 }
3105
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003106 for_each_populated_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003107 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003108 unsigned char types[MAX_ORDER];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003109
David Rientjes7bf02ea2011-05-24 17:11:16 -07003110 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003111 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003112 show_node(zone);
3113 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003114
3115 spin_lock_irqsave(&zone->lock, flags);
3116 for (order = 0; order < MAX_ORDER; order++) {
Rabin Vincent377e4f12012-12-11 16:00:24 -08003117 struct free_area *area = &zone->free_area[order];
3118 int type;
3119
3120 nr[order] = area->nr_free;
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003121 total += nr[order] << order;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003122
3123 types[order] = 0;
3124 for (type = 0; type < MIGRATE_TYPES; type++) {
3125 if (!list_empty(&area->free_list[type]))
3126 types[order] |= 1 << type;
3127 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003128 }
3129 spin_unlock_irqrestore(&zone->lock, flags);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003130 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003131 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003132 if (nr[order])
3133 show_migration_types(types[order]);
3134 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003135 printk("= %lukB\n", K(total));
3136 }
3137
David Rientjes949f7ec2013-04-29 15:07:48 -07003138 hugetlb_show_meminfo();
3139
Larry Woodmane6f36022008-02-04 22:29:30 -08003140 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
3141
Linus Torvalds1da177e2005-04-16 15:20:36 -07003142 show_swap_cache_info();
3143}
3144
Mel Gorman19770b32008-04-28 02:12:18 -07003145static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
3146{
3147 zoneref->zone = zone;
3148 zoneref->zone_idx = zone_idx(zone);
3149}
3150
Linus Torvalds1da177e2005-04-16 15:20:36 -07003151/*
3152 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08003153 *
3154 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003155 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003156static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003157 int nr_zones)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003158{
Christoph Lameter1a932052006-01-06 00:11:16 -08003159 struct zone *zone;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003160 enum zone_type zone_type = MAX_NR_ZONES;
Christoph Lameter02a68a52006-01-06 00:11:18 -08003161
3162 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003163 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08003164 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08003165 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07003166 zoneref_set_zone(zone,
3167 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08003168 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003169 }
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003170 } while (zone_type);
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003171
Christoph Lameter070f8032006-01-06 00:11:19 -08003172 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003173}
3174
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003175
3176/*
3177 * zonelist_order:
3178 * 0 = automatic detection of better ordering.
3179 * 1 = order by ([node] distance, -zonetype)
3180 * 2 = order by (-zonetype, [node] distance)
3181 *
3182 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3183 * the same zonelist. So only NUMA can configure this param.
3184 */
3185#define ZONELIST_ORDER_DEFAULT 0
3186#define ZONELIST_ORDER_NODE 1
3187#define ZONELIST_ORDER_ZONE 2
3188
3189/* zonelist order in the kernel.
3190 * set_zonelist_order() will set this to NODE or ZONE.
3191 */
3192static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
3193static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
3194
3195
Linus Torvalds1da177e2005-04-16 15:20:36 -07003196#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003197/* The value user specified ....changed by config */
3198static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3199/* string for sysctl */
3200#define NUMA_ZONELIST_ORDER_LEN 16
3201char numa_zonelist_order[16] = "default";
3202
3203/*
3204 * interface for configure zonelist ordering.
3205 * command line option "numa_zonelist_order"
3206 * = "[dD]efault - default, automatic configuration.
3207 * = "[nN]ode - order by node locality, then by zone within node
3208 * = "[zZ]one - order by zone, then by locality within zone
3209 */
3210
3211static int __parse_numa_zonelist_order(char *s)
3212{
3213 if (*s == 'd' || *s == 'D') {
3214 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3215 } else if (*s == 'n' || *s == 'N') {
3216 user_zonelist_order = ZONELIST_ORDER_NODE;
3217 } else if (*s == 'z' || *s == 'Z') {
3218 user_zonelist_order = ZONELIST_ORDER_ZONE;
3219 } else {
3220 printk(KERN_WARNING
3221 "Ignoring invalid numa_zonelist_order value: "
3222 "%s\n", s);
3223 return -EINVAL;
3224 }
3225 return 0;
3226}
3227
3228static __init int setup_numa_zonelist_order(char *s)
3229{
Volodymyr G. Lukiianykecb256f2011-01-13 15:46:26 -08003230 int ret;
3231
3232 if (!s)
3233 return 0;
3234
3235 ret = __parse_numa_zonelist_order(s);
3236 if (ret == 0)
3237 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
3238
3239 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003240}
3241early_param("numa_zonelist_order", setup_numa_zonelist_order);
3242
3243/*
3244 * sysctl handler for numa_zonelist_order
3245 */
3246int numa_zonelist_order_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003247 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003248 loff_t *ppos)
3249{
3250 char saved_string[NUMA_ZONELIST_ORDER_LEN];
3251 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01003252 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003253
Andi Kleen443c6f12009-12-23 21:00:47 +01003254 mutex_lock(&zl_order_mutex);
Chen Gangdacbde02013-07-03 15:02:35 -07003255 if (write) {
3256 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
3257 ret = -EINVAL;
3258 goto out;
3259 }
3260 strcpy(saved_string, (char *)table->data);
3261 }
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003262 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003263 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01003264 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003265 if (write) {
3266 int oldval = user_zonelist_order;
Chen Gangdacbde02013-07-03 15:02:35 -07003267
3268 ret = __parse_numa_zonelist_order((char *)table->data);
3269 if (ret) {
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003270 /*
3271 * bogus value. restore saved string
3272 */
Chen Gangdacbde02013-07-03 15:02:35 -07003273 strncpy((char *)table->data, saved_string,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003274 NUMA_ZONELIST_ORDER_LEN);
3275 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003276 } else if (oldval != user_zonelist_order) {
3277 mutex_lock(&zonelists_mutex);
Jiang Liu9adb62a2012-07-31 16:43:28 -07003278 build_all_zonelists(NULL, NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003279 mutex_unlock(&zonelists_mutex);
3280 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003281 }
Andi Kleen443c6f12009-12-23 21:00:47 +01003282out:
3283 mutex_unlock(&zl_order_mutex);
3284 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003285}
3286
3287
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003288#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003289static int node_load[MAX_NUMNODES];
3290
Linus Torvalds1da177e2005-04-16 15:20:36 -07003291/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07003292 * 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 -07003293 * @node: node whose fallback list we're appending
3294 * @used_node_mask: nodemask_t of already used nodes
3295 *
3296 * We use a number of factors to determine which is the next node that should
3297 * appear on a given node's fallback list. The node should not have appeared
3298 * already in @node's fallback list, and it should be the next closest node
3299 * according to the distance array (which contains arbitrary distance values
3300 * from each node to each node in the system), and should also prefer nodes
3301 * with no CPUs, since presumably they'll have very little allocation pressure
3302 * on them otherwise.
3303 * It returns -1 if no node is found.
3304 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003305static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003306{
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01003307 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003308 int min_val = INT_MAX;
David Rientjes00ef2d22013-02-22 16:35:36 -08003309 int best_node = NUMA_NO_NODE;
Rusty Russella70f7302009-03-13 14:49:46 +10303310 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003311
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01003312 /* Use the local node if we haven't already */
3313 if (!node_isset(node, *used_node_mask)) {
3314 node_set(node, *used_node_mask);
3315 return node;
3316 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003317
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08003318 for_each_node_state(n, N_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003319
3320 /* Don't want a node to appear more than once */
3321 if (node_isset(n, *used_node_mask))
3322 continue;
3323
Linus Torvalds1da177e2005-04-16 15:20:36 -07003324 /* Use the distance array to find the distance */
3325 val = node_distance(node, n);
3326
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01003327 /* Penalize nodes under us ("prefer the next node") */
3328 val += (n < node);
3329
Linus Torvalds1da177e2005-04-16 15:20:36 -07003330 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10303331 tmp = cpumask_of_node(n);
3332 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003333 val += PENALTY_FOR_NODE_WITH_CPUS;
3334
3335 /* Slight preference for less loaded node */
3336 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
3337 val += node_load[n];
3338
3339 if (val < min_val) {
3340 min_val = val;
3341 best_node = n;
3342 }
3343 }
3344
3345 if (best_node >= 0)
3346 node_set(best_node, *used_node_mask);
3347
3348 return best_node;
3349}
3350
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003351
3352/*
3353 * Build zonelists ordered by node and zones within node.
3354 * This results in maximum locality--normal zone overflows into local
3355 * DMA zone, if any--but risks exhausting DMA zone.
3356 */
3357static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003358{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003359 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003360 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003361
Mel Gorman54a6eb52008-04-28 02:12:16 -07003362 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07003363 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07003364 ;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003365 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gormandd1a2392008-04-28 02:12:17 -07003366 zonelist->_zonerefs[j].zone = NULL;
3367 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003368}
3369
3370/*
Christoph Lameter523b9452007-10-16 01:25:37 -07003371 * Build gfp_thisnode zonelists
3372 */
3373static void build_thisnode_zonelists(pg_data_t *pgdat)
3374{
Christoph Lameter523b9452007-10-16 01:25:37 -07003375 int j;
3376 struct zonelist *zonelist;
3377
Mel Gorman54a6eb52008-04-28 02:12:16 -07003378 zonelist = &pgdat->node_zonelists[1];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003379 j = build_zonelists_node(pgdat, zonelist, 0);
Mel Gormandd1a2392008-04-28 02:12:17 -07003380 zonelist->_zonerefs[j].zone = NULL;
3381 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07003382}
3383
3384/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003385 * Build zonelists ordered by zone and nodes within zones.
3386 * This results in conserving DMA zone[s] until all Normal memory is
3387 * exhausted, but results in overflowing to remote node while memory
3388 * may still exist in local DMA zone.
3389 */
3390static int node_order[MAX_NUMNODES];
3391
3392static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
3393{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003394 int pos, j, node;
3395 int zone_type; /* needs to be signed */
3396 struct zone *z;
3397 struct zonelist *zonelist;
3398
Mel Gorman54a6eb52008-04-28 02:12:16 -07003399 zonelist = &pgdat->node_zonelists[0];
3400 pos = 0;
3401 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
3402 for (j = 0; j < nr_nodes; j++) {
3403 node = node_order[j];
3404 z = &NODE_DATA(node)->node_zones[zone_type];
3405 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07003406 zoneref_set_zone(z,
3407 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07003408 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003409 }
3410 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003411 }
Mel Gormandd1a2392008-04-28 02:12:17 -07003412 zonelist->_zonerefs[pos].zone = NULL;
3413 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003414}
3415
3416static int default_zonelist_order(void)
3417{
3418 int nid, zone_type;
3419 unsigned long low_kmem_size,total_size;
3420 struct zone *z;
3421 int average_size;
3422 /*
Thomas Weber88393162010-03-16 11:47:56 +01003423 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003424 * If they are really small and used heavily, the system can fall
3425 * into OOM very easily.
David Rientjese325c902010-05-24 14:32:13 -07003426 * This function detect ZONE_DMA/DMA32 size and configures zone order.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003427 */
3428 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
3429 low_kmem_size = 0;
3430 total_size = 0;
3431 for_each_online_node(nid) {
3432 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3433 z = &NODE_DATA(nid)->node_zones[zone_type];
3434 if (populated_zone(z)) {
3435 if (zone_type < ZONE_NORMAL)
Jiang Liu4f9f4772013-07-03 15:03:07 -07003436 low_kmem_size += z->managed_pages;
3437 total_size += z->managed_pages;
David Rientjese325c902010-05-24 14:32:13 -07003438 } else if (zone_type == ZONE_NORMAL) {
3439 /*
3440 * If any node has only lowmem, then node order
3441 * is preferred to allow kernel allocations
3442 * locally; otherwise, they can easily infringe
3443 * on other nodes when there is an abundance of
3444 * lowmem available to allocate from.
3445 */
3446 return ZONELIST_ORDER_NODE;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003447 }
3448 }
3449 }
3450 if (!low_kmem_size || /* there are no DMA area. */
3451 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
3452 return ZONELIST_ORDER_NODE;
3453 /*
3454 * look into each node's config.
3455 * If there is a node whose DMA/DMA32 memory is very big area on
3456 * local memory, NODE_ORDER may be suitable.
3457 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003458 average_size = total_size /
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08003459 (nodes_weight(node_states[N_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003460 for_each_online_node(nid) {
3461 low_kmem_size = 0;
3462 total_size = 0;
3463 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3464 z = &NODE_DATA(nid)->node_zones[zone_type];
3465 if (populated_zone(z)) {
3466 if (zone_type < ZONE_NORMAL)
3467 low_kmem_size += z->present_pages;
3468 total_size += z->present_pages;
3469 }
3470 }
3471 if (low_kmem_size &&
3472 total_size > average_size && /* ignore small node */
3473 low_kmem_size > total_size * 70/100)
3474 return ZONELIST_ORDER_NODE;
3475 }
3476 return ZONELIST_ORDER_ZONE;
3477}
3478
3479static void set_zonelist_order(void)
3480{
3481 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
3482 current_zonelist_order = default_zonelist_order();
3483 else
3484 current_zonelist_order = user_zonelist_order;
3485}
3486
3487static void build_zonelists(pg_data_t *pgdat)
3488{
3489 int j, node, load;
3490 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003491 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003492 int local_node, prev_node;
3493 struct zonelist *zonelist;
3494 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003495
3496 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07003497 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003498 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07003499 zonelist->_zonerefs[0].zone = NULL;
3500 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003501 }
3502
3503 /* NUMA-aware ordering of nodes */
3504 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003505 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003506 prev_node = local_node;
3507 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003508
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003509 memset(node_order, 0, sizeof(node_order));
3510 j = 0;
3511
Linus Torvalds1da177e2005-04-16 15:20:36 -07003512 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
3513 /*
3514 * We don't want to pressure a particular node.
3515 * So adding penalty to the first node in same
3516 * distance group to make it round-robin.
3517 */
David Rientjes957f8222012-10-08 16:33:24 -07003518 if (node_distance(local_node, node) !=
3519 node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003520 node_load[node] = load;
3521
Linus Torvalds1da177e2005-04-16 15:20:36 -07003522 prev_node = node;
3523 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003524 if (order == ZONELIST_ORDER_NODE)
3525 build_zonelists_in_node_order(pgdat, node);
3526 else
3527 node_order[j++] = node; /* remember order */
3528 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003529
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003530 if (order == ZONELIST_ORDER_ZONE) {
3531 /* calculate node order -- i.e., DMA last! */
3532 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003533 }
Christoph Lameter523b9452007-10-16 01:25:37 -07003534
3535 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003536}
3537
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003538/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003539static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003540{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003541 struct zonelist *zonelist;
3542 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07003543 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003544
Mel Gorman54a6eb52008-04-28 02:12:16 -07003545 zonelist = &pgdat->node_zonelists[0];
3546 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3547 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07003548 for (z = zonelist->_zonerefs; z->zone; z++)
3549 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003550}
3551
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003552#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3553/*
3554 * Return node id of node used for "local" allocations.
3555 * I.e., first node id of first zone in arg node's generic zonelist.
3556 * Used for initializing percpu 'numa_mem', which is used primarily
3557 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3558 */
3559int local_memory_node(int node)
3560{
3561 struct zone *zone;
3562
3563 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3564 gfp_zone(GFP_KERNEL),
3565 NULL,
3566 &zone);
3567 return zone->node;
3568}
3569#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003570
Linus Torvalds1da177e2005-04-16 15:20:36 -07003571#else /* CONFIG_NUMA */
3572
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003573static void set_zonelist_order(void)
3574{
3575 current_zonelist_order = ZONELIST_ORDER_ZONE;
3576}
3577
3578static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003579{
Christoph Lameter19655d32006-09-25 23:31:19 -07003580 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003581 enum zone_type j;
3582 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003583
3584 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003585
Mel Gorman54a6eb52008-04-28 02:12:16 -07003586 zonelist = &pgdat->node_zonelists[0];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003587 j = build_zonelists_node(pgdat, zonelist, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003588
Mel Gorman54a6eb52008-04-28 02:12:16 -07003589 /*
3590 * Now we build the zonelist so that it contains the zones
3591 * of all the other nodes.
3592 * We don't want to pressure a particular node, so when
3593 * building the zones for node N, we make sure that the
3594 * zones coming right after the local ones are those from
3595 * node N+1 (modulo N)
3596 */
3597 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3598 if (!node_online(node))
3599 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003600 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003601 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07003602 for (node = 0; node < local_node; node++) {
3603 if (!node_online(node))
3604 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003605 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gorman54a6eb52008-04-28 02:12:16 -07003606 }
3607
Mel Gormandd1a2392008-04-28 02:12:17 -07003608 zonelist->_zonerefs[j].zone = NULL;
3609 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003610}
3611
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003612/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003613static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003614{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003615 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003616}
3617
Linus Torvalds1da177e2005-04-16 15:20:36 -07003618#endif /* CONFIG_NUMA */
3619
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003620/*
3621 * Boot pageset table. One per cpu which is going to be used for all
3622 * zones and all nodes. The parameters will be set in such a way
3623 * that an item put on a list will immediately be handed over to
3624 * the buddy list. This is safe since pageset manipulation is done
3625 * with interrupts disabled.
3626 *
3627 * The boot_pagesets must be kept even after bootup is complete for
3628 * unused processors and/or zones. They do play a role for bootstrapping
3629 * hotplugged processors.
3630 *
3631 * zoneinfo_show() and maybe other functions do
3632 * not check if the processor is online before following the pageset pointer.
3633 * Other parts of the kernel may not check if the zone is available.
3634 */
3635static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3636static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07003637static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003638
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003639/*
3640 * Global mutex to protect against size modification of zonelists
3641 * as well as to serialize pageset setup for the new populated zone.
3642 */
3643DEFINE_MUTEX(zonelists_mutex);
3644
Rusty Russell9b1a4d32008-07-28 12:16:30 -05003645/* return values int ....just for stop_machine() */
Jiang Liu4ed7e022012-07-31 16:43:35 -07003646static int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003647{
Yasunori Goto68113782006-06-23 02:03:11 -07003648 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003649 int cpu;
Jiang Liu9adb62a2012-07-31 16:43:28 -07003650 pg_data_t *self = data;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003651
Bo Liu7f9cfb32009-08-18 14:11:19 -07003652#ifdef CONFIG_NUMA
3653 memset(node_load, 0, sizeof(node_load));
3654#endif
Jiang Liu9adb62a2012-07-31 16:43:28 -07003655
3656 if (self && !node_online(self->node_id)) {
3657 build_zonelists(self);
3658 build_zonelist_cache(self);
3659 }
3660
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003661 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07003662 pg_data_t *pgdat = NODE_DATA(nid);
3663
3664 build_zonelists(pgdat);
3665 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003666 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003667
3668 /*
3669 * Initialize the boot_pagesets that are going to be used
3670 * for bootstrapping processors. The real pagesets for
3671 * each zone will be allocated later when the per cpu
3672 * allocator is available.
3673 *
3674 * boot_pagesets are used also for bootstrapping offline
3675 * cpus if the system is already booted because the pagesets
3676 * are needed to initialize allocators on a specific cpu too.
3677 * F.e. the percpu allocator needs the page allocator which
3678 * needs the percpu allocator in order to allocate its pagesets
3679 * (a chicken-egg dilemma).
3680 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003681 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003682 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3683
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003684#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3685 /*
3686 * We now know the "local memory node" for each node--
3687 * i.e., the node of the first zone in the generic zonelist.
3688 * Set up numa_mem percpu variable for on-line cpus. During
3689 * boot, only the boot cpu should be on-line; we'll init the
3690 * secondary cpus' numa_mem as they come on-line. During
3691 * node/memory hotplug, we'll fixup all on-line cpus.
3692 */
3693 if (cpu_online(cpu))
3694 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3695#endif
3696 }
3697
Yasunori Goto68113782006-06-23 02:03:11 -07003698 return 0;
3699}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003700
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003701/*
3702 * Called with zonelists_mutex held always
3703 * unless system_state == SYSTEM_BOOTING.
3704 */
Jiang Liu9adb62a2012-07-31 16:43:28 -07003705void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
Yasunori Goto68113782006-06-23 02:03:11 -07003706{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003707 set_zonelist_order();
3708
Yasunori Goto68113782006-06-23 02:03:11 -07003709 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07003710 __build_all_zonelists(NULL);
Mel Gorman68ad8df2008-07-23 21:26:52 -07003711 mminit_verify_zonelist();
Yasunori Goto68113782006-06-23 02:03:11 -07003712 cpuset_init_current_mems_allowed();
3713 } else {
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003714#ifdef CONFIG_MEMORY_HOTPLUG
Jiang Liu9adb62a2012-07-31 16:43:28 -07003715 if (zone)
3716 setup_zone_pageset(zone);
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003717#endif
Cody P Schaferdd1895e2013-07-03 15:01:36 -07003718 /* we have to stop all cpus to guarantee there is no user
3719 of zonelist */
Jiang Liu9adb62a2012-07-31 16:43:28 -07003720 stop_machine(__build_all_zonelists, pgdat, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07003721 /* cpuset refresh routine should be here */
3722 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07003723 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003724 /*
3725 * Disable grouping by mobility if the number of pages in the
3726 * system is too low to allow the mechanism to work. It would be
3727 * more accurate, but expensive to check per-zone. This check is
3728 * made on memory-hotadd so a system can start with mobility
3729 * disabled and enable it later
3730 */
Mel Gormand9c23402007-10-16 01:26:01 -07003731 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003732 page_group_by_mobility_disabled = 1;
3733 else
3734 page_group_by_mobility_disabled = 0;
3735
3736 printk("Built %i zonelists in %s order, mobility grouping %s. "
3737 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003738 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003739 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003740 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003741 vm_total_pages);
3742#ifdef CONFIG_NUMA
3743 printk("Policy zone: %s\n", zone_names[policy_zone]);
3744#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003745}
3746
3747/*
3748 * Helper functions to size the waitqueue hash table.
3749 * Essentially these want to choose hash table sizes sufficiently
3750 * large so that collisions trying to wait on pages are rare.
3751 * But in fact, the number of active page waitqueues on typical
3752 * systems is ridiculously low, less than 200. So this is even
3753 * conservative, even though it seems large.
3754 *
3755 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3756 * waitqueues, i.e. the size of the waitq table given the number of pages.
3757 */
3758#define PAGES_PER_WAITQUEUE 256
3759
Yasunori Gotocca448f2006-06-23 02:03:10 -07003760#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07003761static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003762{
3763 unsigned long size = 1;
3764
3765 pages /= PAGES_PER_WAITQUEUE;
3766
3767 while (size < pages)
3768 size <<= 1;
3769
3770 /*
3771 * Once we have dozens or even hundreds of threads sleeping
3772 * on IO we've got bigger problems than wait queue collision.
3773 * Limit the size of the wait table to a reasonable size.
3774 */
3775 size = min(size, 4096UL);
3776
3777 return max(size, 4UL);
3778}
Yasunori Gotocca448f2006-06-23 02:03:10 -07003779#else
3780/*
3781 * A zone's size might be changed by hot-add, so it is not possible to determine
3782 * a suitable size for its wait_table. So we use the maximum size now.
3783 *
3784 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3785 *
3786 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3787 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3788 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3789 *
3790 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3791 * or more by the traditional way. (See above). It equals:
3792 *
3793 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3794 * ia64(16K page size) : = ( 8G + 4M)byte.
3795 * powerpc (64K page size) : = (32G +16M)byte.
3796 */
3797static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3798{
3799 return 4096UL;
3800}
3801#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003802
3803/*
3804 * This is an integer logarithm so that shifts can be used later
3805 * to extract the more random high bits from the multiplicative
3806 * hash function before the remainder is taken.
3807 */
3808static inline unsigned long wait_table_bits(unsigned long size)
3809{
3810 return ffz(~size);
3811}
3812
3813#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3814
Mel Gorman56fd56b2007-10-16 01:25:58 -07003815/*
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07003816 * Check if a pageblock contains reserved pages
3817 */
3818static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
3819{
3820 unsigned long pfn;
3821
3822 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
3823 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
3824 return 1;
3825 }
3826 return 0;
3827}
3828
3829/*
Mel Gormand9c23402007-10-16 01:26:01 -07003830 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07003831 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3832 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07003833 * higher will lead to a bigger reserve which will get freed as contiguous
3834 * blocks as reclaim kicks in
3835 */
3836static void setup_zone_migrate_reserve(struct zone *zone)
3837{
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07003838 unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003839 struct page *page;
Mel Gorman78986a62009-09-21 17:03:02 -07003840 unsigned long block_migratetype;
3841 int reserve;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003842
Michal Hockod02156382011-12-08 14:34:27 -08003843 /*
3844 * Get the start pfn, end pfn and the number of blocks to reserve
3845 * We have to be careful to be aligned to pageblock_nr_pages to
3846 * make sure that we always check pfn_valid for the first page in
3847 * the block.
3848 */
Mel Gorman56fd56b2007-10-16 01:25:58 -07003849 start_pfn = zone->zone_start_pfn;
Cody P Schafer108bcc92013-02-22 16:35:23 -08003850 end_pfn = zone_end_pfn(zone);
Michal Hockod02156382011-12-08 14:34:27 -08003851 start_pfn = roundup(start_pfn, pageblock_nr_pages);
Mel Gorman41858962009-06-16 15:32:12 -07003852 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07003853 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003854
Mel Gorman78986a62009-09-21 17:03:02 -07003855 /*
3856 * Reserve blocks are generally in place to help high-order atomic
3857 * allocations that are short-lived. A min_free_kbytes value that
3858 * would result in more than 2 reserve blocks for atomic allocations
3859 * is assumed to be in place to help anti-fragmentation for the
3860 * future allocation of hugepages at runtime.
3861 */
3862 reserve = min(2, reserve);
3863
Mel Gormand9c23402007-10-16 01:26:01 -07003864 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07003865 if (!pfn_valid(pfn))
3866 continue;
3867 page = pfn_to_page(pfn);
3868
Adam Litke344c7902008-09-02 14:35:38 -07003869 /* Watch out for overlapping nodes */
3870 if (page_to_nid(page) != zone_to_nid(zone))
3871 continue;
3872
Mel Gorman56fd56b2007-10-16 01:25:58 -07003873 block_migratetype = get_pageblock_migratetype(page);
3874
Mel Gorman938929f2012-01-10 15:07:14 -08003875 /* Only test what is necessary when the reserves are not met */
3876 if (reserve > 0) {
3877 /*
3878 * Blocks with reserved pages will never free, skip
3879 * them.
3880 */
3881 block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
3882 if (pageblock_is_reserved(pfn, block_end_pfn))
3883 continue;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003884
Mel Gorman938929f2012-01-10 15:07:14 -08003885 /* If this block is reserved, account for it */
3886 if (block_migratetype == MIGRATE_RESERVE) {
3887 reserve--;
3888 continue;
3889 }
3890
3891 /* Suitable for reserving if this block is movable */
3892 if (block_migratetype == MIGRATE_MOVABLE) {
3893 set_pageblock_migratetype(page,
3894 MIGRATE_RESERVE);
3895 move_freepages_block(zone, page,
3896 MIGRATE_RESERVE);
3897 reserve--;
3898 continue;
3899 }
Mel Gorman56fd56b2007-10-16 01:25:58 -07003900 }
3901
3902 /*
3903 * If the reserve is met and this is a previous reserved block,
3904 * take it back
3905 */
3906 if (block_migratetype == MIGRATE_RESERVE) {
3907 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
3908 move_freepages_block(zone, page, MIGRATE_MOVABLE);
3909 }
3910 }
3911}
Mel Gormanac0e5b72007-10-16 01:25:58 -07003912
Linus Torvalds1da177e2005-04-16 15:20:36 -07003913/*
3914 * Initially all pages are reserved - free ones are freed
3915 * up by free_all_bootmem() once the early boot process is
3916 * done. Non-atomic initialization, single-pass.
3917 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003918void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08003919 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003920{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003921 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07003922 unsigned long end_pfn = start_pfn + size;
3923 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003924 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003925
Hugh Dickins22b31ee2009-01-06 14:40:09 -08003926 if (highest_memmap_pfn < end_pfn - 1)
3927 highest_memmap_pfn = end_pfn - 1;
3928
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003929 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08003930 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08003931 /*
3932 * There can be holes in boot-time mem_map[]s
3933 * handed to this function. They do not
3934 * exist on hotplugged memory.
3935 */
3936 if (context == MEMMAP_EARLY) {
3937 if (!early_pfn_valid(pfn))
3938 continue;
3939 if (!early_pfn_in_nid(pfn, nid))
3940 continue;
3941 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003942 page = pfn_to_page(pfn);
3943 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07003944 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08003945 init_page_count(page);
Mel Gorman22b751c2013-02-22 16:34:59 -08003946 page_mapcount_reset(page);
3947 page_nid_reset_last(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003948 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003949 /*
3950 * Mark the block movable so that blocks are reserved for
3951 * movable at startup. This will force kernel allocations
3952 * to reserve their blocks rather than leaking throughout
3953 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07003954 * kernel allocations are made. Later some blocks near
3955 * the start are marked MIGRATE_RESERVE by
3956 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003957 *
3958 * bitmap is created for zone's valid pfn range. but memmap
3959 * can be created for invalid pages (for alignment)
3960 * check here not to call set_pageblock_migratetype() against
3961 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003962 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003963 if ((z->zone_start_pfn <= pfn)
Cody P Schafer108bcc92013-02-22 16:35:23 -08003964 && (pfn < zone_end_pfn(z))
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003965 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07003966 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003967
Linus Torvalds1da177e2005-04-16 15:20:36 -07003968 INIT_LIST_HEAD(&page->lru);
3969#ifdef WANT_PAGE_VIRTUAL
3970 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3971 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07003972 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003973#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003974 }
3975}
3976
Andi Kleen1e548de2008-02-04 22:29:26 -08003977static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003978{
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003979 int order, t;
3980 for_each_migratetype_order(order, t) {
3981 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003982 zone->free_area[order].nr_free = 0;
3983 }
3984}
3985
3986#ifndef __HAVE_ARCH_MEMMAP_INIT
3987#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08003988 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003989#endif
3990
Jiang Liu4ed7e022012-07-31 16:43:35 -07003991static int __meminit zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003992{
David Howells3a6be872009-05-06 16:03:03 -07003993#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003994 int batch;
3995
3996 /*
3997 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07003998 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003999 *
4000 * OK, so we don't know how big the cache is. So guess.
4001 */
Jiang Liub40da042013-02-22 16:33:52 -08004002 batch = zone->managed_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07004003 if (batch * PAGE_SIZE > 512 * 1024)
4004 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004005 batch /= 4; /* We effectively *= 4 below */
4006 if (batch < 1)
4007 batch = 1;
4008
4009 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004010 * Clamp the batch to a 2^n - 1 value. Having a power
4011 * of 2 value was found to be more likely to have
4012 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004013 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004014 * For example if 2 tasks are alternately allocating
4015 * batches of pages, one task can end up with a lot
4016 * of pages of one half of the possible page colors
4017 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004018 */
David Howells91552032009-05-06 16:03:02 -07004019 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07004020
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004021 return batch;
David Howells3a6be872009-05-06 16:03:03 -07004022
4023#else
4024 /* The deferral and batching of frees should be suppressed under NOMMU
4025 * conditions.
4026 *
4027 * The problem is that NOMMU needs to be able to allocate large chunks
4028 * of contiguous memory as there's no hardware page translation to
4029 * assemble apparent contiguous memory from discontiguous pages.
4030 *
4031 * Queueing large contiguous runs of pages for batching, however,
4032 * causes the pages to actually be freed in smaller chunks. As there
4033 * can be a significant delay between the individual batches being
4034 * recycled, this leads to the once large chunks of space being
4035 * fragmented and becoming unavailable for high-order allocations.
4036 */
4037 return 0;
4038#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004039}
4040
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004041/*
4042 * pcp->high and pcp->batch values are related and dependent on one another:
4043 * ->batch must never be higher then ->high.
4044 * The following function updates them in a safe manner without read side
4045 * locking.
4046 *
4047 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4048 * those fields changing asynchronously (acording the the above rule).
4049 *
4050 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4051 * outside of boot time (or some other assurance that no concurrent updaters
4052 * exist).
4053 */
4054static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
4055 unsigned long batch)
4056{
4057 /* start with a fail safe value for batch */
4058 pcp->batch = 1;
4059 smp_wmb();
4060
4061 /* Update high, then batch, in order */
4062 pcp->high = high;
4063 smp_wmb();
4064
4065 pcp->batch = batch;
4066}
4067
Cody P Schafer36640332013-07-03 15:01:40 -07004068/* a companion to pageset_set_high() */
Cody P Schafer4008bab2013-07-03 15:01:28 -07004069static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
4070{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004071 pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
Cody P Schafer4008bab2013-07-03 15:01:28 -07004072}
4073
Cody P Schafer88c90db2013-07-03 15:01:35 -07004074static void pageset_init(struct per_cpu_pageset *p)
Christoph Lameter2caaad42005-06-21 17:15:00 -07004075{
4076 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004077 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004078
Magnus Damm1c6fe942005-10-26 01:58:59 -07004079 memset(p, 0, sizeof(*p));
4080
Christoph Lameter3dfa5722008-02-04 22:29:19 -08004081 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004082 pcp->count = 0;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004083 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
4084 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07004085}
4086
Cody P Schafer88c90db2013-07-03 15:01:35 -07004087static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
4088{
4089 pageset_init(p);
4090 pageset_set_batch(p, batch);
4091}
4092
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004093/*
Cody P Schafer36640332013-07-03 15:01:40 -07004094 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004095 * to the value high for the pageset p.
4096 */
Cody P Schafer36640332013-07-03 15:01:40 -07004097static void pageset_set_high(struct per_cpu_pageset *p,
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004098 unsigned long high)
4099{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004100 unsigned long batch = max(1UL, high / 4);
4101 if ((high / 4) > (PAGE_SHIFT * 8))
4102 batch = PAGE_SHIFT * 8;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004103
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004104 pageset_update(&p->pcp, high, batch);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004105}
4106
Cody P Schafer169f6c12013-07-03 15:01:41 -07004107static void __meminit pageset_set_high_and_batch(struct zone *zone,
4108 struct per_cpu_pageset *pcp)
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004109{
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004110 if (percpu_pagelist_fraction)
Cody P Schafer36640332013-07-03 15:01:40 -07004111 pageset_set_high(pcp,
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004112 (zone->managed_pages /
4113 percpu_pagelist_fraction));
4114 else
4115 pageset_set_batch(pcp, zone_batchsize(zone));
4116}
4117
Cody P Schafer169f6c12013-07-03 15:01:41 -07004118static void __meminit zone_pageset_init(struct zone *zone, int cpu)
4119{
4120 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4121
4122 pageset_init(pcp);
4123 pageset_set_high_and_batch(zone, pcp);
4124}
4125
Jiang Liu4ed7e022012-07-31 16:43:35 -07004126static void __meminit setup_zone_pageset(struct zone *zone)
Wu Fengguang319774e2010-05-24 14:32:49 -07004127{
4128 int cpu;
Wu Fengguang319774e2010-05-24 14:32:49 -07004129 zone->pageset = alloc_percpu(struct per_cpu_pageset);
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004130 for_each_possible_cpu(cpu)
4131 zone_pageset_init(zone, cpu);
Wu Fengguang319774e2010-05-24 14:32:49 -07004132}
4133
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004134/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004135 * Allocate per cpu pagesets and initialize them.
4136 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07004137 */
Al Viro78d99552005-12-15 09:18:25 +00004138void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004139{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004140 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004141
Wu Fengguang319774e2010-05-24 14:32:49 -07004142 for_each_populated_zone(zone)
4143 setup_zone_pageset(zone);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004144}
4145
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004146static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07004147int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07004148{
4149 int i;
4150 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004151 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07004152
4153 /*
4154 * The per-page waitqueue mechanism uses hashed waitqueues
4155 * per zone.
4156 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07004157 zone->wait_table_hash_nr_entries =
4158 wait_table_hash_nr_entries(zone_size_pages);
4159 zone->wait_table_bits =
4160 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004161 alloc_size = zone->wait_table_hash_nr_entries
4162 * sizeof(wait_queue_head_t);
4163
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07004164 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07004165 zone->wait_table = (wait_queue_head_t *)
Yinghai Lu8f389a992011-05-11 15:13:32 -07004166 alloc_bootmem_node_nopanic(pgdat, alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004167 } else {
4168 /*
4169 * This case means that a zone whose size was 0 gets new memory
4170 * via memory hot-add.
4171 * But it may be the case that a new node was hot-added. In
4172 * this case vmalloc() will not be able to use this new node's
4173 * memory - this wait_table must be initialized to use this new
4174 * node itself as well.
4175 * To use this new node's memory, further consideration will be
4176 * necessary.
4177 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07004178 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004179 }
4180 if (!zone->wait_table)
4181 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07004182
Yasunori Goto02b694d2006-06-23 02:03:08 -07004183 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07004184 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004185
4186 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004187}
4188
Matt Tolentinoc09b4242006-01-17 07:03:44 +01004189static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07004190{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004191 /*
4192 * per cpu subsystem is not up at this point. The following code
4193 * relies on the ability of the linker to provide the
4194 * offset of a (static) per cpu variable into the per cpu area.
4195 */
4196 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07004197
Anton Blanchardf5335c02006-03-25 03:06:49 -08004198 if (zone->present_pages)
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004199 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
4200 zone->name, zone->present_pages,
4201 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07004202}
4203
Jiang Liu4ed7e022012-07-31 16:43:35 -07004204int __meminit init_currently_empty_zone(struct zone *zone,
Yasunori Goto718127c2006-06-23 02:03:10 -07004205 unsigned long zone_start_pfn,
Dave Hansena2f3aa022007-01-10 23:15:30 -08004206 unsigned long size,
4207 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07004208{
4209 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004210 int ret;
4211 ret = zone_wait_table_init(zone, size);
4212 if (ret)
4213 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07004214 pgdat->nr_zones = zone_idx(zone) + 1;
4215
Dave Hansened8ece22005-10-29 18:16:50 -07004216 zone->zone_start_pfn = zone_start_pfn;
4217
Mel Gorman708614e2008-07-23 21:26:51 -07004218 mminit_dprintk(MMINIT_TRACE, "memmap_init",
4219 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4220 pgdat->node_id,
4221 (unsigned long)zone_idx(zone),
4222 zone_start_pfn, (zone_start_pfn + size));
4223
Andi Kleen1e548de2008-02-04 22:29:26 -08004224 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07004225
4226 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004227}
4228
Tejun Heo0ee332c2011-12-08 10:22:09 -08004229#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07004230#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4231/*
4232 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4233 * Architectures may implement their own version but if add_active_range()
4234 * was used and there are no special requirements, this is a convenient
4235 * alternative
4236 */
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004237int __meminit __early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004238{
Tejun Heoc13291a2011-07-12 10:46:30 +02004239 unsigned long start_pfn, end_pfn;
4240 int i, nid;
Russ Anderson7c243c72013-04-29 15:07:59 -07004241 /*
4242 * NOTE: The following SMP-unsafe globals are only used early in boot
4243 * when the kernel is running single-threaded.
4244 */
4245 static unsigned long __meminitdata last_start_pfn, last_end_pfn;
4246 static int __meminitdata last_nid;
4247
4248 if (last_start_pfn <= pfn && pfn < last_end_pfn)
4249 return last_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004250
Tejun Heoc13291a2011-07-12 10:46:30 +02004251 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
Russ Anderson7c243c72013-04-29 15:07:59 -07004252 if (start_pfn <= pfn && pfn < end_pfn) {
4253 last_start_pfn = start_pfn;
4254 last_end_pfn = end_pfn;
4255 last_nid = nid;
Tejun Heoc13291a2011-07-12 10:46:30 +02004256 return nid;
Russ Anderson7c243c72013-04-29 15:07:59 -07004257 }
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08004258 /* This is a memory hole */
4259 return -1;
Mel Gormanc7132162006-09-27 01:49:43 -07004260}
4261#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4262
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004263int __meminit early_pfn_to_nid(unsigned long pfn)
4264{
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08004265 int nid;
4266
4267 nid = __early_pfn_to_nid(pfn);
4268 if (nid >= 0)
4269 return nid;
4270 /* just returns 0 */
4271 return 0;
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004272}
4273
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08004274#ifdef CONFIG_NODES_SPAN_OTHER_NODES
4275bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
4276{
4277 int nid;
4278
4279 nid = __early_pfn_to_nid(pfn);
4280 if (nid >= 0 && nid != node)
4281 return false;
4282 return true;
4283}
4284#endif
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004285
Mel Gormanc7132162006-09-27 01:49:43 -07004286/**
4287 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004288 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4289 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07004290 *
4291 * If an architecture guarantees that all ranges registered with
4292 * add_active_ranges() contain no holes and may be freed, this
4293 * this function may be used instead of calling free_bootmem() manually.
4294 */
Tejun Heoc13291a2011-07-12 10:46:30 +02004295void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004296{
Tejun Heoc13291a2011-07-12 10:46:30 +02004297 unsigned long start_pfn, end_pfn;
4298 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004299
Tejun Heoc13291a2011-07-12 10:46:30 +02004300 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
4301 start_pfn = min(start_pfn, max_low_pfn);
4302 end_pfn = min(end_pfn, max_low_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004303
Tejun Heoc13291a2011-07-12 10:46:30 +02004304 if (start_pfn < end_pfn)
4305 free_bootmem_node(NODE_DATA(this_nid),
4306 PFN_PHYS(start_pfn),
4307 (end_pfn - start_pfn) << PAGE_SHIFT);
Mel Gormanc7132162006-09-27 01:49:43 -07004308 }
4309}
4310
4311/**
4312 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004313 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07004314 *
4315 * If an architecture guarantees that all ranges registered with
4316 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004317 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07004318 */
4319void __init sparse_memory_present_with_active_regions(int nid)
4320{
Tejun Heoc13291a2011-07-12 10:46:30 +02004321 unsigned long start_pfn, end_pfn;
4322 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004323
Tejun Heoc13291a2011-07-12 10:46:30 +02004324 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
4325 memory_present(this_nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004326}
4327
4328/**
4329 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004330 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4331 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4332 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07004333 *
4334 * It returns the start and end page frame of a node based on information
4335 * provided by an arch calling add_active_range(). If called for a node
4336 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004337 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07004338 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004339void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004340 unsigned long *start_pfn, unsigned long *end_pfn)
4341{
Tejun Heoc13291a2011-07-12 10:46:30 +02004342 unsigned long this_start_pfn, this_end_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004343 int i;
Tejun Heoc13291a2011-07-12 10:46:30 +02004344
Mel Gormanc7132162006-09-27 01:49:43 -07004345 *start_pfn = -1UL;
4346 *end_pfn = 0;
4347
Tejun Heoc13291a2011-07-12 10:46:30 +02004348 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
4349 *start_pfn = min(*start_pfn, this_start_pfn);
4350 *end_pfn = max(*end_pfn, this_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004351 }
4352
Christoph Lameter633c0662007-10-16 01:25:37 -07004353 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07004354 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07004355}
4356
4357/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07004358 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4359 * assumption is made that zones within a node are ordered in monotonic
4360 * increasing memory addresses so that the "highest" populated zone is used
4361 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004362static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004363{
4364 int zone_index;
4365 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
4366 if (zone_index == ZONE_MOVABLE)
4367 continue;
4368
4369 if (arch_zone_highest_possible_pfn[zone_index] >
4370 arch_zone_lowest_possible_pfn[zone_index])
4371 break;
4372 }
4373
4374 VM_BUG_ON(zone_index == -1);
4375 movable_zone = zone_index;
4376}
4377
4378/*
4379 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004380 * because it is sized independent of architecture. Unlike the other zones,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004381 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4382 * in each node depending on the size of each node and how evenly kernelcore
4383 * is distributed. This helper function adjusts the zone ranges
4384 * provided by the architecture for a given node by using the end of the
4385 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4386 * zones within a node are in order of monotonic increases memory addresses
4387 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004388static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004389 unsigned long zone_type,
4390 unsigned long node_start_pfn,
4391 unsigned long node_end_pfn,
4392 unsigned long *zone_start_pfn,
4393 unsigned long *zone_end_pfn)
4394{
4395 /* Only adjust if ZONE_MOVABLE is on this node */
4396 if (zone_movable_pfn[nid]) {
4397 /* Size ZONE_MOVABLE */
4398 if (zone_type == ZONE_MOVABLE) {
4399 *zone_start_pfn = zone_movable_pfn[nid];
4400 *zone_end_pfn = min(node_end_pfn,
4401 arch_zone_highest_possible_pfn[movable_zone]);
4402
4403 /* Adjust for ZONE_MOVABLE starting within this range */
4404 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
4405 *zone_end_pfn > zone_movable_pfn[nid]) {
4406 *zone_end_pfn = zone_movable_pfn[nid];
4407
4408 /* Check if this whole range is within ZONE_MOVABLE */
4409 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
4410 *zone_start_pfn = *zone_end_pfn;
4411 }
4412}
4413
4414/*
Mel Gormanc7132162006-09-27 01:49:43 -07004415 * Return the number of pages a zone spans in a node, including holes
4416 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4417 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004418static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004419 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004420 unsigned long node_start_pfn,
4421 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004422 unsigned long *ignored)
4423{
Mel Gormanc7132162006-09-27 01:49:43 -07004424 unsigned long zone_start_pfn, zone_end_pfn;
4425
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004426 /* Get the start and end of the zone */
Mel Gormanc7132162006-09-27 01:49:43 -07004427 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4428 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07004429 adjust_zone_range_for_zone_movable(nid, zone_type,
4430 node_start_pfn, node_end_pfn,
4431 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004432
4433 /* Check that this node has pages within the zone's required range */
4434 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4435 return 0;
4436
4437 /* Move the zone boundaries inside the node if necessary */
4438 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4439 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4440
4441 /* Return the spanned pages */
4442 return zone_end_pfn - zone_start_pfn;
4443}
4444
4445/*
4446 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004447 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07004448 */
Yinghai Lu32996252009-12-15 17:59:02 -08004449unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004450 unsigned long range_start_pfn,
4451 unsigned long range_end_pfn)
4452{
Tejun Heo96e907d2011-07-12 10:46:29 +02004453 unsigned long nr_absent = range_end_pfn - range_start_pfn;
4454 unsigned long start_pfn, end_pfn;
4455 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004456
Tejun Heo96e907d2011-07-12 10:46:29 +02004457 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
4458 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
4459 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
4460 nr_absent -= end_pfn - start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004461 }
Tejun Heo96e907d2011-07-12 10:46:29 +02004462 return nr_absent;
Mel Gormanc7132162006-09-27 01:49:43 -07004463}
4464
4465/**
4466 * absent_pages_in_range - Return number of page frames in holes within a range
4467 * @start_pfn: The start PFN to start searching for holes
4468 * @end_pfn: The end PFN to stop searching for holes
4469 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004470 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07004471 */
4472unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4473 unsigned long end_pfn)
4474{
4475 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4476}
4477
4478/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004479static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004480 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004481 unsigned long node_start_pfn,
4482 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004483 unsigned long *ignored)
4484{
Tejun Heo96e907d2011-07-12 10:46:29 +02004485 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
4486 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman9c7cd682006-09-27 01:49:58 -07004487 unsigned long zone_start_pfn, zone_end_pfn;
4488
Tejun Heo96e907d2011-07-12 10:46:29 +02004489 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
4490 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004491
Mel Gorman2a1e2742007-07-17 04:03:12 -07004492 adjust_zone_range_for_zone_movable(nid, zone_type,
4493 node_start_pfn, node_end_pfn,
4494 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004495 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004496}
Mel Gorman0e0b8642006-09-27 01:49:56 -07004497
Tejun Heo0ee332c2011-12-08 10:22:09 -08004498#else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004499static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004500 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004501 unsigned long node_start_pfn,
4502 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004503 unsigned long *zones_size)
4504{
4505 return zones_size[zone_type];
4506}
4507
Paul Mundt6ea6e682007-07-15 23:38:20 -07004508static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004509 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004510 unsigned long node_start_pfn,
4511 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004512 unsigned long *zholes_size)
4513{
4514 if (!zholes_size)
4515 return 0;
4516
4517 return zholes_size[zone_type];
4518}
Yinghai Lu20e69262013-03-01 14:51:27 -08004519
Tejun Heo0ee332c2011-12-08 10:22:09 -08004520#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07004521
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004522static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004523 unsigned long node_start_pfn,
4524 unsigned long node_end_pfn,
4525 unsigned long *zones_size,
4526 unsigned long *zholes_size)
Mel Gormanc7132162006-09-27 01:49:43 -07004527{
4528 unsigned long realtotalpages, totalpages = 0;
4529 enum zone_type i;
4530
4531 for (i = 0; i < MAX_NR_ZONES; i++)
4532 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004533 node_start_pfn,
4534 node_end_pfn,
4535 zones_size);
Mel Gormanc7132162006-09-27 01:49:43 -07004536 pgdat->node_spanned_pages = totalpages;
4537
4538 realtotalpages = totalpages;
4539 for (i = 0; i < MAX_NR_ZONES; i++)
4540 realtotalpages -=
4541 zone_absent_pages_in_node(pgdat->node_id, i,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004542 node_start_pfn, node_end_pfn,
4543 zholes_size);
Mel Gormanc7132162006-09-27 01:49:43 -07004544 pgdat->node_present_pages = realtotalpages;
4545 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4546 realtotalpages);
4547}
4548
Mel Gorman835c1342007-10-16 01:25:47 -07004549#ifndef CONFIG_SPARSEMEM
4550/*
4551 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07004552 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4553 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07004554 * round what is now in bits to nearest long in bits, then return it in
4555 * bytes.
4556 */
Linus Torvalds7c455122013-02-18 09:58:02 -08004557static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07004558{
4559 unsigned long usemapsize;
4560
Linus Torvalds7c455122013-02-18 09:58:02 -08004561 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
Mel Gormand9c23402007-10-16 01:26:01 -07004562 usemapsize = roundup(zonesize, pageblock_nr_pages);
4563 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07004564 usemapsize *= NR_PAGEBLOCK_BITS;
4565 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4566
4567 return usemapsize / 8;
4568}
4569
4570static void __init setup_usemap(struct pglist_data *pgdat,
Linus Torvalds7c455122013-02-18 09:58:02 -08004571 struct zone *zone,
4572 unsigned long zone_start_pfn,
4573 unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07004574{
Linus Torvalds7c455122013-02-18 09:58:02 -08004575 unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
Mel Gorman835c1342007-10-16 01:25:47 -07004576 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08004577 if (usemapsize)
Yinghai Lu8f389a992011-05-11 15:13:32 -07004578 zone->pageblock_flags = alloc_bootmem_node_nopanic(pgdat,
4579 usemapsize);
Mel Gorman835c1342007-10-16 01:25:47 -07004580}
4581#else
Linus Torvalds7c455122013-02-18 09:58:02 -08004582static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
4583 unsigned long zone_start_pfn, unsigned long zonesize) {}
Mel Gorman835c1342007-10-16 01:25:47 -07004584#endif /* CONFIG_SPARSEMEM */
4585
Mel Gormand9c23402007-10-16 01:26:01 -07004586#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08004587
Mel Gormand9c23402007-10-16 01:26:01 -07004588/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
Xishi Qiuca57df72012-07-31 16:43:19 -07004589void __init set_pageblock_order(void)
Mel Gormand9c23402007-10-16 01:26:01 -07004590{
Andrew Morton955c1cd2012-05-29 15:06:31 -07004591 unsigned int order;
4592
Mel Gormand9c23402007-10-16 01:26:01 -07004593 /* Check that pageblock_nr_pages has not already been setup */
4594 if (pageblock_order)
4595 return;
4596
Andrew Morton955c1cd2012-05-29 15:06:31 -07004597 if (HPAGE_SHIFT > PAGE_SHIFT)
4598 order = HUGETLB_PAGE_ORDER;
4599 else
4600 order = MAX_ORDER - 1;
4601
Mel Gormand9c23402007-10-16 01:26:01 -07004602 /*
4603 * Assume the largest contiguous order of interest is a huge page.
Andrew Morton955c1cd2012-05-29 15:06:31 -07004604 * This value may be variable depending on boot parameters on IA64 and
4605 * powerpc.
Mel Gormand9c23402007-10-16 01:26:01 -07004606 */
4607 pageblock_order = order;
4608}
4609#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4610
Mel Gormanba72cb82007-11-28 16:21:13 -08004611/*
4612 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
Andrew Morton955c1cd2012-05-29 15:06:31 -07004613 * is unused as pageblock_order is set at compile-time. See
4614 * include/linux/pageblock-flags.h for the values of pageblock_order based on
4615 * the kernel config
Mel Gormanba72cb82007-11-28 16:21:13 -08004616 */
Xishi Qiuca57df72012-07-31 16:43:19 -07004617void __init set_pageblock_order(void)
Mel Gormanba72cb82007-11-28 16:21:13 -08004618{
Mel Gormanba72cb82007-11-28 16:21:13 -08004619}
Mel Gormand9c23402007-10-16 01:26:01 -07004620
4621#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4622
Jiang Liu01cefae2012-12-12 13:52:19 -08004623static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
4624 unsigned long present_pages)
4625{
4626 unsigned long pages = spanned_pages;
4627
4628 /*
4629 * Provide a more accurate estimation if there are holes within
4630 * the zone and SPARSEMEM is in use. If there are holes within the
4631 * zone, each populated memory region may cost us one or two extra
4632 * memmap pages due to alignment because memmap pages for each
4633 * populated regions may not naturally algined on page boundary.
4634 * So the (present_pages >> 4) heuristic is a tradeoff for that.
4635 */
4636 if (spanned_pages > present_pages + (present_pages >> 4) &&
4637 IS_ENABLED(CONFIG_SPARSEMEM))
4638 pages = present_pages;
4639
4640 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
4641}
4642
Linus Torvalds1da177e2005-04-16 15:20:36 -07004643/*
4644 * Set up the zone data structures:
4645 * - mark all pages reserved
4646 * - mark all memory queues empty
4647 * - clear the memory bitmaps
Minchan Kim6527af52012-07-31 16:46:16 -07004648 *
4649 * NOTE: pgdat should get zeroed by caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004650 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08004651static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004652 unsigned long node_start_pfn, unsigned long node_end_pfn,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004653 unsigned long *zones_size, unsigned long *zholes_size)
4654{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07004655 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07004656 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004657 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07004658 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004659
Dave Hansen208d54e2005-10-29 18:16:52 -07004660 pgdat_resize_init(pgdat);
Andrea Arcangeli8177a422012-03-23 20:56:34 +01004661#ifdef CONFIG_NUMA_BALANCING
4662 spin_lock_init(&pgdat->numabalancing_migrate_lock);
4663 pgdat->numabalancing_migrate_nr_pages = 0;
4664 pgdat->numabalancing_migrate_next_window = jiffies;
4665#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004666 init_waitqueue_head(&pgdat->kswapd_wait);
Mel Gorman55150612012-07-31 16:44:35 -07004667 init_waitqueue_head(&pgdat->pfmemalloc_wait);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004668 pgdat_page_cgroup_init(pgdat);
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01004669
Linus Torvalds1da177e2005-04-16 15:20:36 -07004670 for (j = 0; j < MAX_NR_ZONES; j++) {
4671 struct zone *zone = pgdat->node_zones + j;
Jiang Liu9feedc92012-12-12 13:52:12 -08004672 unsigned long size, realsize, freesize, memmap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004673
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004674 size = zone_spanned_pages_in_node(nid, j, node_start_pfn,
4675 node_end_pfn, zones_size);
Jiang Liu9feedc92012-12-12 13:52:12 -08004676 realsize = freesize = size - zone_absent_pages_in_node(nid, j,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004677 node_start_pfn,
4678 node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004679 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004680
Mel Gorman0e0b8642006-09-27 01:49:56 -07004681 /*
Jiang Liu9feedc92012-12-12 13:52:12 -08004682 * Adjust freesize so that it accounts for how much memory
Mel Gorman0e0b8642006-09-27 01:49:56 -07004683 * is used by this zone for memmap. This affects the watermark
4684 * and per-cpu initialisations
4685 */
Jiang Liu01cefae2012-12-12 13:52:19 -08004686 memmap_pages = calc_memmap_size(size, realsize);
Jiang Liu9feedc92012-12-12 13:52:12 -08004687 if (freesize >= memmap_pages) {
4688 freesize -= memmap_pages;
Yinghai Lu5594c8c2009-01-06 14:39:14 -08004689 if (memmap_pages)
4690 printk(KERN_DEBUG
4691 " %s zone: %lu pages used for memmap\n",
4692 zone_names[j], memmap_pages);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004693 } else
4694 printk(KERN_WARNING
Jiang Liu9feedc92012-12-12 13:52:12 -08004695 " %s zone: %lu pages exceeds freesize %lu\n",
4696 zone_names[j], memmap_pages, freesize);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004697
Christoph Lameter62672762007-02-10 01:43:07 -08004698 /* Account for reserved pages */
Jiang Liu9feedc92012-12-12 13:52:12 -08004699 if (j == 0 && freesize > dma_reserve) {
4700 freesize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07004701 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08004702 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004703 }
4704
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004705 if (!is_highmem_idx(j))
Jiang Liu9feedc92012-12-12 13:52:12 -08004706 nr_kernel_pages += freesize;
Jiang Liu01cefae2012-12-12 13:52:19 -08004707 /* Charge for highmem memmap if there are enough kernel pages */
4708 else if (nr_kernel_pages > memmap_pages * 2)
4709 nr_kernel_pages -= memmap_pages;
Jiang Liu9feedc92012-12-12 13:52:12 -08004710 nr_all_pages += freesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004711
4712 zone->spanned_pages = size;
Jiang Liu306f2e92013-02-22 16:33:54 -08004713 zone->present_pages = realsize;
Jiang Liu9feedc92012-12-12 13:52:12 -08004714 /*
4715 * Set an approximate value for lowmem here, it will be adjusted
4716 * when the bootmem allocator frees pages into the buddy system.
4717 * And all highmem pages will be managed by the buddy system.
4718 */
4719 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
Christoph Lameter96146342006-07-03 00:24:13 -07004720#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07004721 zone->node = nid;
Jiang Liu9feedc92012-12-12 13:52:12 -08004722 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07004723 / 100;
Jiang Liu9feedc92012-12-12 13:52:12 -08004724 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07004725#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004726 zone->name = zone_names[j];
4727 spin_lock_init(&zone->lock);
4728 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07004729 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004730 zone->zone_pgdat = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004731
Dave Hansened8ece22005-10-29 18:16:50 -07004732 zone_pcp_init(zone);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004733 lruvec_init(&zone->lruvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004734 if (!size)
4735 continue;
4736
Andrew Morton955c1cd2012-05-29 15:06:31 -07004737 set_pageblock_order();
Linus Torvalds7c455122013-02-18 09:58:02 -08004738 setup_usemap(pgdat, zone, zone_start_pfn, size);
Dave Hansena2f3aa022007-01-10 23:15:30 -08004739 ret = init_currently_empty_zone(zone, zone_start_pfn,
4740 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07004741 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07004742 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004743 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004744 }
4745}
4746
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004747static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004748{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004749 /* Skip empty nodes */
4750 if (!pgdat->node_spanned_pages)
4751 return;
4752
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004753#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07004754 /* ia64 gets its own node_mem_map, before this, without bootmem */
4755 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07004756 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004757 struct page *map;
4758
Bob Piccoe984bb42006-05-20 15:00:31 -07004759 /*
4760 * The zone's endpoints aren't required to be MAX_ORDER
4761 * aligned but the node_mem_map endpoints must be in order
4762 * for the buddy allocator to function correctly.
4763 */
4764 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
Cody P Schafer108bcc92013-02-22 16:35:23 -08004765 end = pgdat_end_pfn(pgdat);
Bob Piccoe984bb42006-05-20 15:00:31 -07004766 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4767 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07004768 map = alloc_remap(pgdat->node_id, size);
4769 if (!map)
Yinghai Lu8f389a992011-05-11 15:13:32 -07004770 map = alloc_bootmem_node_nopanic(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07004771 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004772 }
Roman Zippel12d810c2007-05-31 00:40:54 -07004773#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07004774 /*
4775 * With no DISCONTIG, the global mem_map is just set as node 0's
4776 */
Mel Gormanc7132162006-09-27 01:49:43 -07004777 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004778 mem_map = NODE_DATA(0)->node_mem_map;
Tejun Heo0ee332c2011-12-08 10:22:09 -08004779#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07004780 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08004781 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Tejun Heo0ee332c2011-12-08 10:22:09 -08004782#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07004783 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004784#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004785#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004786}
4787
Johannes Weiner9109fb72008-07-23 21:27:20 -07004788void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4789 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004790{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004791 pg_data_t *pgdat = NODE_DATA(nid);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004792 unsigned long start_pfn = 0;
4793 unsigned long end_pfn = 0;
Johannes Weiner9109fb72008-07-23 21:27:20 -07004794
Minchan Kim88fdf752012-07-31 16:46:14 -07004795 /* pg_data_t should be reset to zero when it's allocated */
Linus Torvalds8783b6e2012-08-02 10:37:03 -07004796 WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
Minchan Kim88fdf752012-07-31 16:46:14 -07004797
Linus Torvalds1da177e2005-04-16 15:20:36 -07004798 pgdat->node_id = nid;
4799 pgdat->node_start_pfn = node_start_pfn;
David Rientjes957f8222012-10-08 16:33:24 -07004800 init_zone_allows_reclaim(nid);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004801#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4802 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
4803#endif
4804 calculate_node_totalpages(pgdat, start_pfn, end_pfn,
4805 zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004806
4807 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07004808#ifdef CONFIG_FLAT_NODE_MEM_MAP
4809 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4810 nid, (unsigned long)pgdat,
4811 (unsigned long)pgdat->node_mem_map);
4812#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004813
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004814 free_area_init_core(pgdat, start_pfn, end_pfn,
4815 zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004816}
4817
Tejun Heo0ee332c2011-12-08 10:22:09 -08004818#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07004819
4820#if MAX_NUMNODES > 1
4821/*
4822 * Figure out the number of possible node ids.
4823 */
Cody P Schaferf9872ca2013-04-29 15:08:01 -07004824void __init setup_nr_node_ids(void)
Miklos Szeredi418508c2007-05-23 13:57:55 -07004825{
4826 unsigned int node;
4827 unsigned int highest = 0;
4828
4829 for_each_node_mask(node, node_possible_map)
4830 highest = node;
4831 nr_node_ids = highest + 1;
4832}
Miklos Szeredi418508c2007-05-23 13:57:55 -07004833#endif
4834
Mel Gormanc7132162006-09-27 01:49:43 -07004835/**
Tejun Heo1e019792011-07-12 09:45:34 +02004836 * node_map_pfn_alignment - determine the maximum internode alignment
4837 *
4838 * This function should be called after node map is populated and sorted.
4839 * It calculates the maximum power of two alignment which can distinguish
4840 * all the nodes.
4841 *
4842 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
4843 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
4844 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
4845 * shifted, 1GiB is enough and this function will indicate so.
4846 *
4847 * This is used to test whether pfn -> nid mapping of the chosen memory
4848 * model has fine enough granularity to avoid incorrect mapping for the
4849 * populated node map.
4850 *
4851 * Returns the determined alignment in pfn's. 0 if there is no alignment
4852 * requirement (single node).
4853 */
4854unsigned long __init node_map_pfn_alignment(void)
4855{
4856 unsigned long accl_mask = 0, last_end = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02004857 unsigned long start, end, mask;
Tejun Heo1e019792011-07-12 09:45:34 +02004858 int last_nid = -1;
Tejun Heoc13291a2011-07-12 10:46:30 +02004859 int i, nid;
Tejun Heo1e019792011-07-12 09:45:34 +02004860
Tejun Heoc13291a2011-07-12 10:46:30 +02004861 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
Tejun Heo1e019792011-07-12 09:45:34 +02004862 if (!start || last_nid < 0 || last_nid == nid) {
4863 last_nid = nid;
4864 last_end = end;
4865 continue;
4866 }
4867
4868 /*
4869 * Start with a mask granular enough to pin-point to the
4870 * start pfn and tick off bits one-by-one until it becomes
4871 * too coarse to separate the current node from the last.
4872 */
4873 mask = ~((1 << __ffs(start)) - 1);
4874 while (mask && last_end <= (start & (mask << 1)))
4875 mask <<= 1;
4876
4877 /* accumulate all internode masks */
4878 accl_mask |= mask;
4879 }
4880
4881 /* convert mask to number of pages */
4882 return ~accl_mask + 1;
4883}
4884
Mel Gormana6af2bc2007-02-10 01:42:57 -08004885/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004886static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07004887{
Mel Gormana6af2bc2007-02-10 01:42:57 -08004888 unsigned long min_pfn = ULONG_MAX;
Tejun Heoc13291a2011-07-12 10:46:30 +02004889 unsigned long start_pfn;
4890 int i;
Mel Gorman1abbfb42006-11-23 12:01:41 +00004891
Tejun Heoc13291a2011-07-12 10:46:30 +02004892 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
4893 min_pfn = min(min_pfn, start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004894
Mel Gormana6af2bc2007-02-10 01:42:57 -08004895 if (min_pfn == ULONG_MAX) {
4896 printk(KERN_WARNING
Paul Jackson2bc0d2612008-06-22 07:22:17 -07004897 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08004898 return 0;
4899 }
4900
4901 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004902}
4903
4904/**
4905 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4906 *
4907 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004908 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07004909 */
4910unsigned long __init find_min_pfn_with_active_regions(void)
4911{
4912 return find_min_pfn_for_node(MAX_NUMNODES);
4913}
4914
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004915/*
4916 * early_calculate_totalpages()
4917 * Sum pages in active regions for movable zone.
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08004918 * Populate N_MEMORY for calculating usable_nodes.
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004919 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07004920static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07004921{
Mel Gorman7e63efe2007-07-17 04:03:15 -07004922 unsigned long totalpages = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02004923 unsigned long start_pfn, end_pfn;
4924 int i, nid;
Mel Gorman7e63efe2007-07-17 04:03:15 -07004925
Tejun Heoc13291a2011-07-12 10:46:30 +02004926 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
4927 unsigned long pages = end_pfn - start_pfn;
4928
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004929 totalpages += pages;
4930 if (pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08004931 node_set_state(nid, N_MEMORY);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004932 }
4933 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07004934}
4935
Mel Gorman2a1e2742007-07-17 04:03:12 -07004936/*
4937 * Find the PFN the Movable zone begins in each node. Kernel memory
4938 * is spread evenly between nodes as long as the nodes have enough
4939 * memory. When they don't, some nodes will have more kernelcore than
4940 * others
4941 */
Kautuk Consulb224ef82012-03-21 16:34:15 -07004942static void __init find_zone_movable_pfns_for_nodes(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004943{
4944 int i, nid;
4945 unsigned long usable_startpfn;
4946 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07004947 /* save the state before borrow the nodemask */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08004948 nodemask_t saved_node_state = node_states[N_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004949 unsigned long totalpages = early_calculate_totalpages();
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08004950 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004951
Mel Gorman7e63efe2007-07-17 04:03:15 -07004952 /*
4953 * If movablecore was specified, calculate what size of
4954 * kernelcore that corresponds so that memory usable for
4955 * any allocation type is evenly spread. If both kernelcore
4956 * and movablecore are specified, then the value of kernelcore
4957 * will be used for required_kernelcore if it's greater than
4958 * what movablecore would have allowed.
4959 */
4960 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07004961 unsigned long corepages;
4962
4963 /*
4964 * Round-up so that ZONE_MOVABLE is at least as large as what
4965 * was requested by the user
4966 */
4967 required_movablecore =
4968 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
4969 corepages = totalpages - required_movablecore;
4970
4971 required_kernelcore = max(required_kernelcore, corepages);
4972 }
4973
Yinghai Lu20e69262013-03-01 14:51:27 -08004974 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4975 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07004976 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004977
4978 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
Yinghai Lu20e69262013-03-01 14:51:27 -08004979 find_usable_zone_for_movable();
Mel Gorman2a1e2742007-07-17 04:03:12 -07004980 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4981
4982restart:
4983 /* Spread kernelcore memory as evenly as possible throughout nodes */
4984 kernelcore_node = required_kernelcore / usable_nodes;
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08004985 for_each_node_state(nid, N_MEMORY) {
Tejun Heoc13291a2011-07-12 10:46:30 +02004986 unsigned long start_pfn, end_pfn;
4987
Mel Gorman2a1e2742007-07-17 04:03:12 -07004988 /*
4989 * Recalculate kernelcore_node if the division per node
4990 * now exceeds what is necessary to satisfy the requested
4991 * amount of memory for the kernel
4992 */
4993 if (required_kernelcore < kernelcore_node)
4994 kernelcore_node = required_kernelcore / usable_nodes;
4995
4996 /*
4997 * As the map is walked, we track how much memory is usable
4998 * by the kernel using kernelcore_remaining. When it is
4999 * 0, the rest of the node is usable by ZONE_MOVABLE
5000 */
5001 kernelcore_remaining = kernelcore_node;
5002
5003 /* Go through each range of PFNs within this node */
Tejun Heoc13291a2011-07-12 10:46:30 +02005004 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005005 unsigned long size_pages;
5006
Tejun Heoc13291a2011-07-12 10:46:30 +02005007 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005008 if (start_pfn >= end_pfn)
5009 continue;
5010
5011 /* Account for what is only usable for kernelcore */
5012 if (start_pfn < usable_startpfn) {
5013 unsigned long kernel_pages;
5014 kernel_pages = min(end_pfn, usable_startpfn)
5015 - start_pfn;
5016
5017 kernelcore_remaining -= min(kernel_pages,
5018 kernelcore_remaining);
5019 required_kernelcore -= min(kernel_pages,
5020 required_kernelcore);
5021
5022 /* Continue if range is now fully accounted */
5023 if (end_pfn <= usable_startpfn) {
5024
5025 /*
5026 * Push zone_movable_pfn to the end so
5027 * that if we have to rebalance
5028 * kernelcore across nodes, we will
5029 * not double account here
5030 */
5031 zone_movable_pfn[nid] = end_pfn;
5032 continue;
5033 }
5034 start_pfn = usable_startpfn;
5035 }
5036
5037 /*
5038 * The usable PFN range for ZONE_MOVABLE is from
5039 * start_pfn->end_pfn. Calculate size_pages as the
5040 * number of pages used as kernelcore
5041 */
5042 size_pages = end_pfn - start_pfn;
5043 if (size_pages > kernelcore_remaining)
5044 size_pages = kernelcore_remaining;
5045 zone_movable_pfn[nid] = start_pfn + size_pages;
5046
5047 /*
5048 * Some kernelcore has been met, update counts and
5049 * break if the kernelcore for this node has been
5050 * satisified
5051 */
5052 required_kernelcore -= min(required_kernelcore,
5053 size_pages);
5054 kernelcore_remaining -= size_pages;
5055 if (!kernelcore_remaining)
5056 break;
5057 }
5058 }
5059
5060 /*
5061 * If there is still required_kernelcore, we do another pass with one
5062 * less node in the count. This will push zone_movable_pfn[nid] further
5063 * along on the nodes that still have memory until kernelcore is
5064 * satisified
5065 */
5066 usable_nodes--;
5067 if (usable_nodes && required_kernelcore > usable_nodes)
5068 goto restart;
5069
5070 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5071 for (nid = 0; nid < MAX_NUMNODES; nid++)
5072 zone_movable_pfn[nid] =
5073 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07005074
Yinghai Lu20e69262013-03-01 14:51:27 -08005075out:
Yinghai Lu66918dc2009-06-30 11:41:37 -07005076 /* restore the node_state */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005077 node_states[N_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005078}
5079
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005080/* Any regular or high memory on that node ? */
5081static void check_for_memory(pg_data_t *pgdat, int nid)
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005082{
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005083 enum zone_type zone_type;
5084
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005085 if (N_MEMORY == N_NORMAL_MEMORY)
5086 return;
5087
5088 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005089 struct zone *zone = &pgdat->node_zones[zone_type];
Bob Liud0048b02012-01-12 17:19:07 -08005090 if (zone->present_pages) {
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005091 node_set_state(nid, N_HIGH_MEMORY);
5092 if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
5093 zone_type <= ZONE_NORMAL)
5094 node_set_state(nid, N_NORMAL_MEMORY);
Bob Liud0048b02012-01-12 17:19:07 -08005095 break;
5096 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005097 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005098}
5099
Mel Gormanc7132162006-09-27 01:49:43 -07005100/**
5101 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005102 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07005103 *
5104 * This will call free_area_init_node() for each active node in the system.
5105 * Using the page ranges provided by add_active_range(), the size of each
5106 * zone in each node and their holes is calculated. If the maximum PFN
5107 * between two adjacent zones match, it is assumed that the zone is empty.
5108 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5109 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5110 * starts where the previous one ended. For example, ZONE_DMA32 starts
5111 * at arch_max_dma_pfn.
5112 */
5113void __init free_area_init_nodes(unsigned long *max_zone_pfn)
5114{
Tejun Heoc13291a2011-07-12 10:46:30 +02005115 unsigned long start_pfn, end_pfn;
5116 int i, nid;
Mel Gormana6af2bc2007-02-10 01:42:57 -08005117
Mel Gormanc7132162006-09-27 01:49:43 -07005118 /* Record where the zone boundaries are */
5119 memset(arch_zone_lowest_possible_pfn, 0,
5120 sizeof(arch_zone_lowest_possible_pfn));
5121 memset(arch_zone_highest_possible_pfn, 0,
5122 sizeof(arch_zone_highest_possible_pfn));
5123 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
5124 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
5125 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005126 if (i == ZONE_MOVABLE)
5127 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07005128 arch_zone_lowest_possible_pfn[i] =
5129 arch_zone_highest_possible_pfn[i-1];
5130 arch_zone_highest_possible_pfn[i] =
5131 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
5132 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07005133 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
5134 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
5135
5136 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5137 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
Kautuk Consulb224ef82012-03-21 16:34:15 -07005138 find_zone_movable_pfns_for_nodes();
Mel Gormanc7132162006-09-27 01:49:43 -07005139
Mel Gormanc7132162006-09-27 01:49:43 -07005140 /* Print out the zone ranges */
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005141 printk("Zone ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005142 for (i = 0; i < MAX_NR_ZONES; i++) {
5143 if (i == ZONE_MOVABLE)
5144 continue;
Kay Sievers155cbfc2012-05-08 17:24:14 +02005145 printk(KERN_CONT " %-8s ", zone_names[i]);
David Rientjes72f0ba02010-03-05 13:42:14 -08005146 if (arch_zone_lowest_possible_pfn[i] ==
5147 arch_zone_highest_possible_pfn[i])
Kay Sievers155cbfc2012-05-08 17:24:14 +02005148 printk(KERN_CONT "empty\n");
David Rientjes72f0ba02010-03-05 13:42:14 -08005149 else
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005150 printk(KERN_CONT "[mem %0#10lx-%0#10lx]\n",
5151 arch_zone_lowest_possible_pfn[i] << PAGE_SHIFT,
5152 (arch_zone_highest_possible_pfn[i]
5153 << PAGE_SHIFT) - 1);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005154 }
5155
5156 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005157 printk("Movable zone start for each node\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005158 for (i = 0; i < MAX_NUMNODES; i++) {
5159 if (zone_movable_pfn[i])
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005160 printk(" Node %d: %#010lx\n", i,
5161 zone_movable_pfn[i] << PAGE_SHIFT);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005162 }
Mel Gormanc7132162006-09-27 01:49:43 -07005163
Wanpeng Lif2d52fe2012-10-08 16:32:24 -07005164 /* Print out the early node map */
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005165 printk("Early memory node ranges\n");
Tejun Heoc13291a2011-07-12 10:46:30 +02005166 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005167 printk(" node %3d: [mem %#010lx-%#010lx]\n", nid,
5168 start_pfn << PAGE_SHIFT, (end_pfn << PAGE_SHIFT) - 1);
Mel Gormanc7132162006-09-27 01:49:43 -07005169
5170 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07005171 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08005172 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07005173 for_each_online_node(nid) {
5174 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07005175 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07005176 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005177
5178 /* Any memory on that node */
5179 if (pgdat->node_present_pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005180 node_set_state(nid, N_MEMORY);
5181 check_for_memory(pgdat, nid);
Mel Gormanc7132162006-09-27 01:49:43 -07005182 }
5183}
Mel Gorman2a1e2742007-07-17 04:03:12 -07005184
Mel Gorman7e63efe2007-07-17 04:03:15 -07005185static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005186{
5187 unsigned long long coremem;
5188 if (!p)
5189 return -EINVAL;
5190
5191 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005192 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005193
Mel Gorman7e63efe2007-07-17 04:03:15 -07005194 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07005195 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
5196
5197 return 0;
5198}
Mel Gormaned7ed362007-07-17 04:03:14 -07005199
Mel Gorman7e63efe2007-07-17 04:03:15 -07005200/*
5201 * kernelcore=size sets the amount of memory for use for allocations that
5202 * cannot be reclaimed or migrated.
5203 */
5204static int __init cmdline_parse_kernelcore(char *p)
5205{
5206 return cmdline_parse_core(p, &required_kernelcore);
5207}
5208
5209/*
5210 * movablecore=size sets the amount of memory for use for allocations that
5211 * can be reclaimed or migrated.
5212 */
5213static int __init cmdline_parse_movablecore(char *p)
5214{
5215 return cmdline_parse_core(p, &required_movablecore);
5216}
5217
Mel Gormaned7ed362007-07-17 04:03:14 -07005218early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005219early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07005220
Tejun Heo0ee332c2011-12-08 10:22:09 -08005221#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005222
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005223void adjust_managed_page_count(struct page *page, long count)
5224{
5225 spin_lock(&managed_page_count_lock);
5226 page_zone(page)->managed_pages += count;
5227 totalram_pages += count;
Jiang Liu3dcc0572013-07-03 15:03:21 -07005228#ifdef CONFIG_HIGHMEM
5229 if (PageHighMem(page))
5230 totalhigh_pages += count;
5231#endif
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005232 spin_unlock(&managed_page_count_lock);
5233}
Jiang Liu3dcc0572013-07-03 15:03:21 -07005234EXPORT_SYMBOL(adjust_managed_page_count);
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005235
Jiang Liu11199692013-07-03 15:02:48 -07005236unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
Jiang Liu69afade2013-04-29 15:06:21 -07005237{
Jiang Liu11199692013-07-03 15:02:48 -07005238 void *pos;
5239 unsigned long pages = 0;
Jiang Liu69afade2013-04-29 15:06:21 -07005240
Jiang Liu11199692013-07-03 15:02:48 -07005241 start = (void *)PAGE_ALIGN((unsigned long)start);
5242 end = (void *)((unsigned long)end & PAGE_MASK);
5243 for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
Jiang Liudbe67df2013-07-03 15:02:51 -07005244 if ((unsigned int)poison <= 0xFF)
Jiang Liu11199692013-07-03 15:02:48 -07005245 memset(pos, poison, PAGE_SIZE);
5246 free_reserved_page(virt_to_page(pos));
Jiang Liu69afade2013-04-29 15:06:21 -07005247 }
5248
5249 if (pages && s)
Jiang Liu11199692013-07-03 15:02:48 -07005250 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
Jiang Liu69afade2013-04-29 15:06:21 -07005251 s, pages << (PAGE_SHIFT - 10), start, end);
5252
5253 return pages;
5254}
Jiang Liu11199692013-07-03 15:02:48 -07005255EXPORT_SYMBOL(free_reserved_area);
Jiang Liu69afade2013-04-29 15:06:21 -07005256
Jiang Liucfa11e02013-04-29 15:07:00 -07005257#ifdef CONFIG_HIGHMEM
5258void free_highmem_page(struct page *page)
5259{
5260 __free_reserved_page(page);
5261 totalram_pages++;
Jiang Liu7b4b2a02013-07-03 15:03:11 -07005262 page_zone(page)->managed_pages++;
Jiang Liucfa11e02013-04-29 15:07:00 -07005263 totalhigh_pages++;
5264}
5265#endif
5266
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005267
5268void __init mem_init_print_info(const char *str)
5269{
5270 unsigned long physpages, codesize, datasize, rosize, bss_size;
5271 unsigned long init_code_size, init_data_size;
5272
5273 physpages = get_num_physpages();
5274 codesize = _etext - _stext;
5275 datasize = _edata - _sdata;
5276 rosize = __end_rodata - __start_rodata;
5277 bss_size = __bss_stop - __bss_start;
5278 init_data_size = __init_end - __init_begin;
5279 init_code_size = _einittext - _sinittext;
5280
5281 /*
5282 * Detect special cases and adjust section sizes accordingly:
5283 * 1) .init.* may be embedded into .data sections
5284 * 2) .init.text.* may be out of [__init_begin, __init_end],
5285 * please refer to arch/tile/kernel/vmlinux.lds.S.
5286 * 3) .rodata.* may be embedded into .text or .data sections.
5287 */
5288#define adj_init_size(start, end, size, pos, adj) \
5289 if (start <= pos && pos < end && size > adj) \
5290 size -= adj;
5291
5292 adj_init_size(__init_begin, __init_end, init_data_size,
5293 _sinittext, init_code_size);
5294 adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
5295 adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
5296 adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
5297 adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
5298
5299#undef adj_init_size
5300
5301 printk("Memory: %luK/%luK available "
5302 "(%luK kernel code, %luK rwdata, %luK rodata, "
5303 "%luK init, %luK bss, %luK reserved"
5304#ifdef CONFIG_HIGHMEM
5305 ", %luK highmem"
5306#endif
5307 "%s%s)\n",
5308 nr_free_pages() << (PAGE_SHIFT-10), physpages << (PAGE_SHIFT-10),
5309 codesize >> 10, datasize >> 10, rosize >> 10,
5310 (init_data_size + init_code_size) >> 10, bss_size >> 10,
5311 (physpages - totalram_pages) << (PAGE_SHIFT-10),
5312#ifdef CONFIG_HIGHMEM
5313 totalhigh_pages << (PAGE_SHIFT-10),
5314#endif
5315 str ? ", " : "", str ? str : "");
5316}
5317
Mel Gorman0e0b8642006-09-27 01:49:56 -07005318/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005319 * set_dma_reserve - set the specified number of pages reserved in the first zone
5320 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07005321 *
5322 * The per-cpu batchsize and zone watermarks are determined by present_pages.
5323 * In the DMA zone, a significant percentage may be consumed by kernel image
5324 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005325 * function may optionally be used to account for unfreeable pages in the
5326 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5327 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07005328 */
5329void __init set_dma_reserve(unsigned long new_dma_reserve)
5330{
5331 dma_reserve = new_dma_reserve;
5332}
5333
Linus Torvalds1da177e2005-04-16 15:20:36 -07005334void __init free_area_init(unsigned long *zones_size)
5335{
Johannes Weiner9109fb72008-07-23 21:27:20 -07005336 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005337 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
5338}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005339
Linus Torvalds1da177e2005-04-16 15:20:36 -07005340static int page_alloc_cpu_notify(struct notifier_block *self,
5341 unsigned long action, void *hcpu)
5342{
5343 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005344
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07005345 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -07005346 lru_add_drain_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005347 drain_pages(cpu);
5348
5349 /*
5350 * Spill the event counters of the dead processor
5351 * into the current processors event counters.
5352 * This artificially elevates the count of the current
5353 * processor.
5354 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07005355 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005356
5357 /*
5358 * Zero the differential counters of the dead processor
5359 * so that the vm statistics are consistent.
5360 *
5361 * This is only okay since the processor is dead and cannot
5362 * race with what we are doing.
5363 */
Christoph Lameter2244b952006-06-30 01:55:33 -07005364 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005365 }
5366 return NOTIFY_OK;
5367}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005368
5369void __init page_alloc_init(void)
5370{
5371 hotcpu_notifier(page_alloc_cpu_notify, 0);
5372}
5373
5374/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005375 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5376 * or min_free_kbytes changes.
5377 */
5378static void calculate_totalreserve_pages(void)
5379{
5380 struct pglist_data *pgdat;
5381 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005382 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005383
5384 for_each_online_pgdat(pgdat) {
5385 for (i = 0; i < MAX_NR_ZONES; i++) {
5386 struct zone *zone = pgdat->node_zones + i;
5387 unsigned long max = 0;
5388
5389 /* Find valid and maximum lowmem_reserve in the zone */
5390 for (j = i; j < MAX_NR_ZONES; j++) {
5391 if (zone->lowmem_reserve[j] > max)
5392 max = zone->lowmem_reserve[j];
5393 }
5394
Mel Gorman41858962009-06-16 15:32:12 -07005395 /* we treat the high watermark as reserved pages. */
5396 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005397
Jiang Liub40da042013-02-22 16:33:52 -08005398 if (max > zone->managed_pages)
5399 max = zone->managed_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005400 reserve_pages += max;
Johannes Weinerab8fabd2012-01-10 15:07:42 -08005401 /*
5402 * Lowmem reserves are not available to
5403 * GFP_HIGHUSER page cache allocations and
5404 * kswapd tries to balance zones to their high
5405 * watermark. As a result, neither should be
5406 * regarded as dirtyable memory, to prevent a
5407 * situation where reclaim has to clean pages
5408 * in order to balance the zones.
5409 */
5410 zone->dirty_balance_reserve = max;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005411 }
5412 }
Johannes Weinerab8fabd2012-01-10 15:07:42 -08005413 dirty_balance_reserve = reserve_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005414 totalreserve_pages = reserve_pages;
5415}
5416
5417/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07005418 * setup_per_zone_lowmem_reserve - called whenever
5419 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
5420 * has a correct pages reserved value, so an adequate number of
5421 * pages are left in the zone after a successful __alloc_pages().
5422 */
5423static void setup_per_zone_lowmem_reserve(void)
5424{
5425 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005426 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005427
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08005428 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005429 for (j = 0; j < MAX_NR_ZONES; j++) {
5430 struct zone *zone = pgdat->node_zones + j;
Jiang Liub40da042013-02-22 16:33:52 -08005431 unsigned long managed_pages = zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005432
5433 zone->lowmem_reserve[j] = 0;
5434
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005435 idx = j;
5436 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005437 struct zone *lower_zone;
5438
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005439 idx--;
5440
Linus Torvalds1da177e2005-04-16 15:20:36 -07005441 if (sysctl_lowmem_reserve_ratio[idx] < 1)
5442 sysctl_lowmem_reserve_ratio[idx] = 1;
5443
5444 lower_zone = pgdat->node_zones + idx;
Jiang Liub40da042013-02-22 16:33:52 -08005445 lower_zone->lowmem_reserve[j] = managed_pages /
Linus Torvalds1da177e2005-04-16 15:20:36 -07005446 sysctl_lowmem_reserve_ratio[idx];
Jiang Liub40da042013-02-22 16:33:52 -08005447 managed_pages += lower_zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005448 }
5449 }
5450 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005451
5452 /* update totalreserve_pages */
5453 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005454}
5455
Mel Gormancfd3da12011-04-25 21:36:42 +00005456static void __setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005457{
5458 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
5459 unsigned long lowmem_pages = 0;
5460 struct zone *zone;
5461 unsigned long flags;
5462
5463 /* Calculate total number of !ZONE_HIGHMEM pages */
5464 for_each_zone(zone) {
5465 if (!is_highmem(zone))
Jiang Liub40da042013-02-22 16:33:52 -08005466 lowmem_pages += zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005467 }
5468
5469 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07005470 u64 tmp;
5471
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005472 spin_lock_irqsave(&zone->lock, flags);
Jiang Liub40da042013-02-22 16:33:52 -08005473 tmp = (u64)pages_min * zone->managed_pages;
Andrew Mortonac924c62006-05-15 09:43:59 -07005474 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005475 if (is_highmem(zone)) {
5476 /*
Nick Piggin669ed172005-11-13 16:06:45 -08005477 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5478 * need highmem pages, so cap pages_min to a small
5479 * value here.
5480 *
Mel Gorman41858962009-06-16 15:32:12 -07005481 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Nick Piggin669ed172005-11-13 16:06:45 -08005482 * deltas controls asynch page reclaim, and so should
5483 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005484 */
Andrew Morton90ae8d62013-02-22 16:32:22 -08005485 unsigned long min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005486
Jiang Liub40da042013-02-22 16:33:52 -08005487 min_pages = zone->managed_pages / 1024;
Andrew Morton90ae8d62013-02-22 16:32:22 -08005488 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
Mel Gorman41858962009-06-16 15:32:12 -07005489 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005490 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08005491 /*
5492 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07005493 * proportionate to the zone's size.
5494 */
Mel Gorman41858962009-06-16 15:32:12 -07005495 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005496 }
5497
Mel Gorman41858962009-06-16 15:32:12 -07005498 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
5499 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Marek Szyprowski49f223a2012-01-25 12:49:24 +01005500
Mel Gorman56fd56b2007-10-16 01:25:58 -07005501 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005502 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005503 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005504
5505 /* update totalreserve_pages */
5506 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005507}
5508
Mel Gormancfd3da12011-04-25 21:36:42 +00005509/**
5510 * setup_per_zone_wmarks - called when min_free_kbytes changes
5511 * or when memory is hot-{added|removed}
5512 *
5513 * Ensures that the watermark[min,low,high] values for each zone are set
5514 * correctly with respect to min_free_kbytes.
5515 */
5516void setup_per_zone_wmarks(void)
5517{
5518 mutex_lock(&zonelists_mutex);
5519 __setup_per_zone_wmarks();
5520 mutex_unlock(&zonelists_mutex);
5521}
5522
Randy Dunlap55a44622009-09-21 17:01:20 -07005523/*
Rik van Riel556adec2008-10-18 20:26:34 -07005524 * The inactive anon list should be small enough that the VM never has to
5525 * do too much work, but large enough that each inactive page has a chance
5526 * to be referenced again before it is swapped out.
5527 *
5528 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5529 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5530 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5531 * the anonymous pages are kept on the inactive list.
5532 *
5533 * total target max
5534 * memory ratio inactive anon
5535 * -------------------------------------
5536 * 10MB 1 5MB
5537 * 100MB 1 50MB
5538 * 1GB 3 250MB
5539 * 10GB 10 0.9GB
5540 * 100GB 31 3GB
5541 * 1TB 101 10GB
5542 * 10TB 320 32GB
5543 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005544static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
Minchan Kim96cb4df2009-06-16 15:32:49 -07005545{
5546 unsigned int gb, ratio;
5547
5548 /* Zone size in gigabytes */
Jiang Liub40da042013-02-22 16:33:52 -08005549 gb = zone->managed_pages >> (30 - PAGE_SHIFT);
Minchan Kim96cb4df2009-06-16 15:32:49 -07005550 if (gb)
5551 ratio = int_sqrt(10 * gb);
5552 else
5553 ratio = 1;
5554
5555 zone->inactive_ratio = ratio;
5556}
5557
KOSAKI Motohiro839a4fc2011-05-24 17:11:31 -07005558static void __meminit setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07005559{
5560 struct zone *zone;
5561
Minchan Kim96cb4df2009-06-16 15:32:49 -07005562 for_each_zone(zone)
5563 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07005564}
5565
Linus Torvalds1da177e2005-04-16 15:20:36 -07005566/*
5567 * Initialise min_free_kbytes.
5568 *
5569 * For small machines we want it small (128k min). For large machines
5570 * we want it large (64MB max). But it is not linear, because network
5571 * bandwidth does not increase linearly with machine size. We use
5572 *
5573 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5574 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5575 *
5576 * which yields
5577 *
5578 * 16MB: 512k
5579 * 32MB: 724k
5580 * 64MB: 1024k
5581 * 128MB: 1448k
5582 * 256MB: 2048k
5583 * 512MB: 2896k
5584 * 1024MB: 4096k
5585 * 2048MB: 5792k
5586 * 4096MB: 8192k
5587 * 8192MB: 11584k
5588 * 16384MB: 16384k
5589 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005590int __meminit init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005591{
5592 unsigned long lowmem_kbytes;
Michal Hocko5f127332013-07-08 16:00:40 -07005593 int new_min_free_kbytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005594
5595 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
Michal Hocko5f127332013-07-08 16:00:40 -07005596 new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005597
Michal Hocko5f127332013-07-08 16:00:40 -07005598 if (new_min_free_kbytes > user_min_free_kbytes) {
5599 min_free_kbytes = new_min_free_kbytes;
5600 if (min_free_kbytes < 128)
5601 min_free_kbytes = 128;
5602 if (min_free_kbytes > 65536)
5603 min_free_kbytes = 65536;
5604 } else {
5605 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
5606 new_min_free_kbytes, user_min_free_kbytes);
5607 }
Minchan Kimbc75d332009-06-16 15:32:48 -07005608 setup_per_zone_wmarks();
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -07005609 refresh_zone_stat_thresholds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005610 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07005611 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005612 return 0;
5613}
Minchan Kimbc75d332009-06-16 15:32:48 -07005614module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005615
5616/*
5617 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5618 * that we can call two helper functions whenever min_free_kbytes
5619 * changes.
5620 */
5621int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005622 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005623{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005624 proc_dointvec(table, write, buffer, length, ppos);
Michal Hocko5f127332013-07-08 16:00:40 -07005625 if (write) {
5626 user_min_free_kbytes = min_free_kbytes;
Minchan Kimbc75d332009-06-16 15:32:48 -07005627 setup_per_zone_wmarks();
Michal Hocko5f127332013-07-08 16:00:40 -07005628 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005629 return 0;
5630}
5631
Christoph Lameter96146342006-07-03 00:24:13 -07005632#ifdef CONFIG_NUMA
5633int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005634 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07005635{
5636 struct zone *zone;
5637 int rc;
5638
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005639 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07005640 if (rc)
5641 return rc;
5642
5643 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08005644 zone->min_unmapped_pages = (zone->managed_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07005645 sysctl_min_unmapped_ratio) / 100;
5646 return 0;
5647}
Christoph Lameter0ff38492006-09-25 23:31:52 -07005648
5649int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005650 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07005651{
5652 struct zone *zone;
5653 int rc;
5654
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005655 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07005656 if (rc)
5657 return rc;
5658
5659 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08005660 zone->min_slab_pages = (zone->managed_pages *
Christoph Lameter0ff38492006-09-25 23:31:52 -07005661 sysctl_min_slab_ratio) / 100;
5662 return 0;
5663}
Christoph Lameter96146342006-07-03 00:24:13 -07005664#endif
5665
Linus Torvalds1da177e2005-04-16 15:20:36 -07005666/*
5667 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5668 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5669 * whenever sysctl_lowmem_reserve_ratio changes.
5670 *
5671 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07005672 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07005673 * if in function of the boot time zone sizes.
5674 */
5675int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005676 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005677{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005678 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005679 setup_per_zone_lowmem_reserve();
5680 return 0;
5681}
5682
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005683/*
5684 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5685 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5686 * can have before it gets flushed back to buddy allocator.
5687 */
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005688int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005689 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005690{
5691 struct zone *zone;
5692 unsigned int cpu;
5693 int ret;
5694
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005695 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
Sasha Levin93278812012-05-10 13:01:44 -07005696 if (!write || (ret < 0))
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005697 return ret;
Cody P Schaferc8e251f2013-07-03 15:01:29 -07005698
5699 mutex_lock(&pcp_batch_high_lock);
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07005700 for_each_populated_zone(zone) {
Cody P Schafer22a7f122013-07-03 15:01:34 -07005701 unsigned long high;
5702 high = zone->managed_pages / percpu_pagelist_fraction;
5703 for_each_possible_cpu(cpu)
Cody P Schafer36640332013-07-03 15:01:40 -07005704 pageset_set_high(per_cpu_ptr(zone->pageset, cpu),
5705 high);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005706 }
Cody P Schaferc8e251f2013-07-03 15:01:29 -07005707 mutex_unlock(&pcp_batch_high_lock);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005708 return 0;
5709}
5710
David S. Millerf034b5d2006-08-24 03:08:07 -07005711int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005712
5713#ifdef CONFIG_NUMA
5714static int __init set_hashdist(char *str)
5715{
5716 if (!str)
5717 return 0;
5718 hashdist = simple_strtoul(str, &str, 0);
5719 return 1;
5720}
5721__setup("hashdist=", set_hashdist);
5722#endif
5723
5724/*
5725 * allocate a large system hash table from bootmem
5726 * - it is assumed that the hash table must contain an exact power-of-2
5727 * quantity of entries
5728 * - limit is the number of hash buckets, not the total allocation size
5729 */
5730void *__init alloc_large_system_hash(const char *tablename,
5731 unsigned long bucketsize,
5732 unsigned long numentries,
5733 int scale,
5734 int flags,
5735 unsigned int *_hash_shift,
5736 unsigned int *_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00005737 unsigned long low_limit,
5738 unsigned long high_limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005739{
Tim Bird31fe62b2012-05-23 13:33:35 +00005740 unsigned long long max = high_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005741 unsigned long log2qty, size;
5742 void *table = NULL;
5743
5744 /* allow the kernel cmdline to have a say */
5745 if (!numentries) {
5746 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08005747 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005748 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
5749 numentries >>= 20 - PAGE_SHIFT;
5750 numentries <<= 20 - PAGE_SHIFT;
5751
5752 /* limit to 1 bucket per 2^scale bytes of low memory */
5753 if (scale > PAGE_SHIFT)
5754 numentries >>= (scale - PAGE_SHIFT);
5755 else
5756 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08005757
5758 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07005759 if (unlikely(flags & HASH_SMALL)) {
5760 /* Makes no sense without HASH_EARLY */
5761 WARN_ON(!(flags & HASH_EARLY));
5762 if (!(numentries >> *_hash_shift)) {
5763 numentries = 1UL << *_hash_shift;
5764 BUG_ON(!numentries);
5765 }
5766 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08005767 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005768 }
John Hawkes6e692ed2006-03-25 03:08:02 -08005769 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005770
5771 /* limit allocation size to 1/16 total memory by default */
5772 if (max == 0) {
5773 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
5774 do_div(max, bucketsize);
5775 }
Dimitri Sivanich074b8512012-02-08 12:39:07 -08005776 max = min(max, 0x80000000ULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005777
Tim Bird31fe62b2012-05-23 13:33:35 +00005778 if (numentries < low_limit)
5779 numentries = low_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005780 if (numentries > max)
5781 numentries = max;
5782
David Howellsf0d1b0b2006-12-08 02:37:49 -08005783 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005784
5785 do {
5786 size = bucketsize << log2qty;
5787 if (flags & HASH_EARLY)
Jan Beulich74768ed2008-08-12 15:08:39 -07005788 table = alloc_bootmem_nopanic(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005789 else if (hashdist)
5790 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
5791 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07005792 /*
5793 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07005794 * some pages at the end of hash table which
5795 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07005796 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005797 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07005798 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005799 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
5800 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005801 }
5802 } while (!table && size > PAGE_SIZE && --log2qty);
5803
5804 if (!table)
5805 panic("Failed to allocate %s hash table\n", tablename);
5806
Robin Holtf241e6602010-10-07 12:59:26 -07005807 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07005808 tablename,
Robin Holtf241e6602010-10-07 12:59:26 -07005809 (1UL << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08005810 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005811 size);
5812
5813 if (_hash_shift)
5814 *_hash_shift = log2qty;
5815 if (_hash_mask)
5816 *_hash_mask = (1 << log2qty) - 1;
5817
5818 return table;
5819}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08005820
Mel Gorman835c1342007-10-16 01:25:47 -07005821/* Return a pointer to the bitmap storing bits affecting a block of pages */
5822static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
5823 unsigned long pfn)
5824{
5825#ifdef CONFIG_SPARSEMEM
5826 return __pfn_to_section(pfn)->pageblock_flags;
5827#else
5828 return zone->pageblock_flags;
5829#endif /* CONFIG_SPARSEMEM */
5830}
Andrew Morton6220ec72006-10-19 23:29:05 -07005831
Mel Gorman835c1342007-10-16 01:25:47 -07005832static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
5833{
5834#ifdef CONFIG_SPARSEMEM
5835 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07005836 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005837#else
Laura Abbottc060f942013-01-11 14:31:51 -08005838 pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
Mel Gormand9c23402007-10-16 01:26:01 -07005839 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005840#endif /* CONFIG_SPARSEMEM */
5841}
5842
5843/**
Mel Gormand9c23402007-10-16 01:26:01 -07005844 * 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 -07005845 * @page: The page within the block of interest
5846 * @start_bitidx: The first bit of interest to retrieve
5847 * @end_bitidx: The last bit of interest
5848 * returns pageblock_bits flags
5849 */
5850unsigned long get_pageblock_flags_group(struct page *page,
5851 int start_bitidx, int end_bitidx)
5852{
5853 struct zone *zone;
5854 unsigned long *bitmap;
5855 unsigned long pfn, bitidx;
5856 unsigned long flags = 0;
5857 unsigned long value = 1;
5858
5859 zone = page_zone(page);
5860 pfn = page_to_pfn(page);
5861 bitmap = get_pageblock_bitmap(zone, pfn);
5862 bitidx = pfn_to_bitidx(zone, pfn);
5863
5864 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5865 if (test_bit(bitidx + start_bitidx, bitmap))
5866 flags |= value;
5867
5868 return flags;
5869}
5870
5871/**
Mel Gormand9c23402007-10-16 01:26:01 -07005872 * 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 -07005873 * @page: The page within the block of interest
5874 * @start_bitidx: The first bit of interest
5875 * @end_bitidx: The last bit of interest
5876 * @flags: The flags to set
5877 */
5878void set_pageblock_flags_group(struct page *page, unsigned long flags,
5879 int start_bitidx, int end_bitidx)
5880{
5881 struct zone *zone;
5882 unsigned long *bitmap;
5883 unsigned long pfn, bitidx;
5884 unsigned long value = 1;
5885
5886 zone = page_zone(page);
5887 pfn = page_to_pfn(page);
5888 bitmap = get_pageblock_bitmap(zone, pfn);
5889 bitidx = pfn_to_bitidx(zone, pfn);
Cody P Schafer108bcc92013-02-22 16:35:23 -08005890 VM_BUG_ON(!zone_spans_pfn(zone, pfn));
Mel Gorman835c1342007-10-16 01:25:47 -07005891
5892 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5893 if (flags & value)
5894 __set_bit(bitidx + start_bitidx, bitmap);
5895 else
5896 __clear_bit(bitidx + start_bitidx, bitmap);
5897}
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07005898
5899/*
Minchan Kim80934512012-07-31 16:43:01 -07005900 * This function checks whether pageblock includes unmovable pages or not.
5901 * If @count is not zero, it is okay to include less @count unmovable pages
5902 *
5903 * PageLRU check wihtout isolation or lru_lock could race so that
5904 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
5905 * expect this function should be exact.
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07005906 */
Wen Congyangb023f462012-12-11 16:00:45 -08005907bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
5908 bool skip_hwpoisoned_pages)
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005909{
5910 unsigned long pfn, iter, found;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01005911 int mt;
5912
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005913 /*
5914 * For avoiding noise data, lru_add_drain_all() should be called
Minchan Kim80934512012-07-31 16:43:01 -07005915 * If ZONE_MOVABLE, the zone never contains unmovable pages
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005916 */
5917 if (zone_idx(zone) == ZONE_MOVABLE)
Minchan Kim80934512012-07-31 16:43:01 -07005918 return false;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01005919 mt = get_pageblock_migratetype(page);
5920 if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
Minchan Kim80934512012-07-31 16:43:01 -07005921 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005922
5923 pfn = page_to_pfn(page);
5924 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
5925 unsigned long check = pfn + iter;
5926
Namhyung Kim29723fc2011-02-25 14:44:25 -08005927 if (!pfn_valid_within(check))
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005928 continue;
Namhyung Kim29723fc2011-02-25 14:44:25 -08005929
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005930 page = pfn_to_page(check);
Minchan Kim97d255c2012-07-31 16:42:59 -07005931 /*
5932 * We can't use page_count without pin a page
5933 * because another CPU can free compound page.
5934 * This check already skips compound tails of THP
5935 * because their page->_count is zero at all time.
5936 */
5937 if (!atomic_read(&page->_count)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005938 if (PageBuddy(page))
5939 iter += (1 << page_order(page)) - 1;
5940 continue;
5941 }
Minchan Kim97d255c2012-07-31 16:42:59 -07005942
Wen Congyangb023f462012-12-11 16:00:45 -08005943 /*
5944 * The HWPoisoned page may be not in buddy system, and
5945 * page_count() is not 0.
5946 */
5947 if (skip_hwpoisoned_pages && PageHWPoison(page))
5948 continue;
5949
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005950 if (!PageLRU(page))
5951 found++;
5952 /*
5953 * If there are RECLAIMABLE pages, we need to check it.
5954 * But now, memory offline itself doesn't call shrink_slab()
5955 * and it still to be fixed.
5956 */
5957 /*
5958 * If the page is not RAM, page_count()should be 0.
5959 * we don't need more check. This is an _used_ not-movable page.
5960 *
5961 * The problematic thing here is PG_reserved pages. PG_reserved
5962 * is set to both of a memory hole page and a _used_ kernel
5963 * page at boot.
5964 */
5965 if (found > count)
Minchan Kim80934512012-07-31 16:43:01 -07005966 return true;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005967 }
Minchan Kim80934512012-07-31 16:43:01 -07005968 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005969}
5970
5971bool is_pageblock_removable_nolock(struct page *page)
5972{
Michal Hocko656a0702012-01-20 14:33:58 -08005973 struct zone *zone;
5974 unsigned long pfn;
Michal Hocko687875f2012-01-20 14:33:55 -08005975
5976 /*
5977 * We have to be careful here because we are iterating over memory
5978 * sections which are not zone aware so we might end up outside of
5979 * the zone but still within the section.
Michal Hocko656a0702012-01-20 14:33:58 -08005980 * We have to take care about the node as well. If the node is offline
5981 * its NODE_DATA will be NULL - see page_zone.
Michal Hocko687875f2012-01-20 14:33:55 -08005982 */
Michal Hocko656a0702012-01-20 14:33:58 -08005983 if (!node_online(page_to_nid(page)))
5984 return false;
5985
5986 zone = page_zone(page);
5987 pfn = page_to_pfn(page);
Cody P Schafer108bcc92013-02-22 16:35:23 -08005988 if (!zone_spans_pfn(zone, pfn))
Michal Hocko687875f2012-01-20 14:33:55 -08005989 return false;
5990
Wen Congyangb023f462012-12-11 16:00:45 -08005991 return !has_unmovable_pages(zone, page, 0, true);
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07005992}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07005993
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01005994#ifdef CONFIG_CMA
5995
5996static unsigned long pfn_max_align_down(unsigned long pfn)
5997{
5998 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
5999 pageblock_nr_pages) - 1);
6000}
6001
6002static unsigned long pfn_max_align_up(unsigned long pfn)
6003{
6004 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
6005 pageblock_nr_pages));
6006}
6007
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006008/* [start, end) must belong to a single zone. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006009static int __alloc_contig_migrate_range(struct compact_control *cc,
6010 unsigned long start, unsigned long end)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006011{
6012 /* This function is based on compact_zone() from compaction.c. */
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006013 unsigned long nr_reclaimed;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006014 unsigned long pfn = start;
6015 unsigned int tries = 0;
6016 int ret = 0;
6017
Marek Szyprowskibe49a6e2012-12-12 13:51:19 -08006018 migrate_prep();
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006019
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006020 while (pfn < end || !list_empty(&cc->migratepages)) {
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006021 if (fatal_signal_pending(current)) {
6022 ret = -EINTR;
6023 break;
6024 }
6025
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006026 if (list_empty(&cc->migratepages)) {
6027 cc->nr_migratepages = 0;
6028 pfn = isolate_migratepages_range(cc->zone, cc,
Minchan Kime46a2872012-10-08 16:33:48 -07006029 pfn, end, true);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006030 if (!pfn) {
6031 ret = -EINTR;
6032 break;
6033 }
6034 tries = 0;
6035 } else if (++tries == 5) {
6036 ret = ret < 0 ? ret : -EBUSY;
6037 break;
6038 }
6039
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006040 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
6041 &cc->migratepages);
6042 cc->nr_migratepages -= nr_reclaimed;
Minchan Kim02c6de82012-10-08 16:31:55 -07006043
Hugh Dickins9c620e22013-02-22 16:35:14 -08006044 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
6045 0, MIGRATE_SYNC, MR_CMA);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006046 }
Srinivas Pandruvada2a6f5122013-02-22 16:32:09 -08006047 if (ret < 0) {
6048 putback_movable_pages(&cc->migratepages);
6049 return ret;
6050 }
6051 return 0;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006052}
6053
6054/**
6055 * alloc_contig_range() -- tries to allocate given range of pages
6056 * @start: start PFN to allocate
6057 * @end: one-past-the-last PFN to allocate
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006058 * @migratetype: migratetype of the underlaying pageblocks (either
6059 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6060 * in range must have the same migratetype and it must
6061 * be either of the two.
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006062 *
6063 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6064 * aligned, however it's the caller's responsibility to guarantee that
6065 * we are the only thread that changes migrate type of pageblocks the
6066 * pages fall in.
6067 *
6068 * The PFN range must belong to a single zone.
6069 *
6070 * Returns zero on success or negative error code. On success all
6071 * pages which PFN is in [start, end) are allocated for the caller and
6072 * need to be freed with free_contig_range().
6073 */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006074int alloc_contig_range(unsigned long start, unsigned long end,
6075 unsigned migratetype)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006076{
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006077 unsigned long outer_start, outer_end;
6078 int ret = 0, order;
6079
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006080 struct compact_control cc = {
6081 .nr_migratepages = 0,
6082 .order = -1,
6083 .zone = page_zone(pfn_to_page(start)),
6084 .sync = true,
6085 .ignore_skip_hint = true,
6086 };
6087 INIT_LIST_HEAD(&cc.migratepages);
6088
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006089 /*
6090 * What we do here is we mark all pageblocks in range as
6091 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6092 * have different sizes, and due to the way page allocator
6093 * work, we align the range to biggest of the two pages so
6094 * that page allocator won't try to merge buddies from
6095 * different pageblocks and change MIGRATE_ISOLATE to some
6096 * other migration type.
6097 *
6098 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6099 * migrate the pages from an unaligned range (ie. pages that
6100 * we are interested in). This will put all the pages in
6101 * range back to page allocator as MIGRATE_ISOLATE.
6102 *
6103 * When this is done, we take the pages in range from page
6104 * allocator removing them from the buddy system. This way
6105 * page allocator will never consider using them.
6106 *
6107 * This lets us mark the pageblocks back as
6108 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6109 * aligned range but not in the unaligned, original range are
6110 * put back to page allocator so that buddy can use them.
6111 */
6112
6113 ret = start_isolate_page_range(pfn_max_align_down(start),
Wen Congyangb023f462012-12-11 16:00:45 -08006114 pfn_max_align_up(end), migratetype,
6115 false);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006116 if (ret)
Bob Liu86a595f2012-10-25 13:37:56 -07006117 return ret;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006118
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006119 ret = __alloc_contig_migrate_range(&cc, start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006120 if (ret)
6121 goto done;
6122
6123 /*
6124 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6125 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
6126 * more, all pages in [start, end) are free in page allocator.
6127 * What we are going to do is to allocate all pages from
6128 * [start, end) (that is remove them from page allocator).
6129 *
6130 * The only problem is that pages at the beginning and at the
6131 * end of interesting range may be not aligned with pages that
6132 * page allocator holds, ie. they can be part of higher order
6133 * pages. Because of this, we reserve the bigger range and
6134 * once this is done free the pages we are not interested in.
6135 *
6136 * We don't have to hold zone->lock here because the pages are
6137 * isolated thus they won't get removed from buddy.
6138 */
6139
6140 lru_add_drain_all();
6141 drain_all_pages();
6142
6143 order = 0;
6144 outer_start = start;
6145 while (!PageBuddy(pfn_to_page(outer_start))) {
6146 if (++order >= MAX_ORDER) {
6147 ret = -EBUSY;
6148 goto done;
6149 }
6150 outer_start &= ~0UL << order;
6151 }
6152
6153 /* Make sure the range is really isolated. */
Wen Congyangb023f462012-12-11 16:00:45 -08006154 if (test_pages_isolated(outer_start, end, false)) {
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006155 pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n",
6156 outer_start, end);
6157 ret = -EBUSY;
6158 goto done;
6159 }
6160
Marek Szyprowski49f223a2012-01-25 12:49:24 +01006161
6162 /* Grab isolated pages from freelists. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006163 outer_end = isolate_freepages_range(&cc, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006164 if (!outer_end) {
6165 ret = -EBUSY;
6166 goto done;
6167 }
6168
6169 /* Free head and tail (if any) */
6170 if (start != outer_start)
6171 free_contig_range(outer_start, start - outer_start);
6172 if (end != outer_end)
6173 free_contig_range(end, outer_end - end);
6174
6175done:
6176 undo_isolate_page_range(pfn_max_align_down(start),
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006177 pfn_max_align_up(end), migratetype);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006178 return ret;
6179}
6180
6181void free_contig_range(unsigned long pfn, unsigned nr_pages)
6182{
Marek Szyprowskibcc2b022012-12-20 15:05:18 -08006183 unsigned int count = 0;
6184
6185 for (; nr_pages--; pfn++) {
6186 struct page *page = pfn_to_page(pfn);
6187
6188 count += page_count(page) != 1;
6189 __free_page(page);
6190 }
6191 WARN(count != 0, "%d pages are still in use!\n", count);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006192}
6193#endif
6194
Jiang Liu4ed7e022012-07-31 16:43:35 -07006195#ifdef CONFIG_MEMORY_HOTPLUG
Cody P Schafer0a647f32013-07-03 15:01:33 -07006196/*
6197 * The zone indicated has a new number of managed_pages; batch sizes and percpu
6198 * page high values need to be recalulated.
6199 */
Jiang Liu4ed7e022012-07-31 16:43:35 -07006200void __meminit zone_pcp_update(struct zone *zone)
6201{
Cody P Schafer0a647f32013-07-03 15:01:33 -07006202 unsigned cpu;
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006203 mutex_lock(&pcp_batch_high_lock);
Cody P Schafer0a647f32013-07-03 15:01:33 -07006204 for_each_possible_cpu(cpu)
Cody P Schafer169f6c12013-07-03 15:01:41 -07006205 pageset_set_high_and_batch(zone,
6206 per_cpu_ptr(zone->pageset, cpu));
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006207 mutex_unlock(&pcp_batch_high_lock);
Jiang Liu4ed7e022012-07-31 16:43:35 -07006208}
6209#endif
6210
Jiang Liu340175b2012-07-31 16:43:32 -07006211void zone_pcp_reset(struct zone *zone)
6212{
6213 unsigned long flags;
Minchan Kim5a883812012-10-08 16:33:39 -07006214 int cpu;
6215 struct per_cpu_pageset *pset;
Jiang Liu340175b2012-07-31 16:43:32 -07006216
6217 /* avoid races with drain_pages() */
6218 local_irq_save(flags);
6219 if (zone->pageset != &boot_pageset) {
Minchan Kim5a883812012-10-08 16:33:39 -07006220 for_each_online_cpu(cpu) {
6221 pset = per_cpu_ptr(zone->pageset, cpu);
6222 drain_zonestat(zone, pset);
6223 }
Jiang Liu340175b2012-07-31 16:43:32 -07006224 free_percpu(zone->pageset);
6225 zone->pageset = &boot_pageset;
6226 }
6227 local_irq_restore(flags);
6228}
6229
Wen Congyang6dcd73d2012-12-11 16:01:01 -08006230#ifdef CONFIG_MEMORY_HOTREMOVE
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006231/*
6232 * All pages in the range must be isolated before calling this.
6233 */
6234void
6235__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
6236{
6237 struct page *page;
6238 struct zone *zone;
6239 int order, i;
6240 unsigned long pfn;
6241 unsigned long flags;
6242 /* find the first valid pfn */
6243 for (pfn = start_pfn; pfn < end_pfn; pfn++)
6244 if (pfn_valid(pfn))
6245 break;
6246 if (pfn == end_pfn)
6247 return;
6248 zone = page_zone(pfn_to_page(pfn));
6249 spin_lock_irqsave(&zone->lock, flags);
6250 pfn = start_pfn;
6251 while (pfn < end_pfn) {
6252 if (!pfn_valid(pfn)) {
6253 pfn++;
6254 continue;
6255 }
6256 page = pfn_to_page(pfn);
Wen Congyangb023f462012-12-11 16:00:45 -08006257 /*
6258 * The HWPoisoned page may be not in buddy system, and
6259 * page_count() is not 0.
6260 */
6261 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
6262 pfn++;
6263 SetPageReserved(page);
6264 continue;
6265 }
6266
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006267 BUG_ON(page_count(page));
6268 BUG_ON(!PageBuddy(page));
6269 order = page_order(page);
6270#ifdef CONFIG_DEBUG_VM
6271 printk(KERN_INFO "remove from free list %lx %d %lx\n",
6272 pfn, 1 << order, end_pfn);
6273#endif
6274 list_del(&page->lru);
6275 rmv_page_order(page);
6276 zone->free_area[order].nr_free--;
Wanpeng Licea27eb2013-07-03 15:02:40 -07006277#ifdef CONFIG_HIGHMEM
6278 if (PageHighMem(page))
6279 totalhigh_pages -= 1 << order;
6280#endif
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006281 for (i = 0; i < (1 << order); i++)
6282 SetPageReserved((page+i));
6283 pfn += (1 << order);
6284 }
6285 spin_unlock_irqrestore(&zone->lock, flags);
6286}
6287#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01006288
6289#ifdef CONFIG_MEMORY_FAILURE
6290bool is_free_buddy_page(struct page *page)
6291{
6292 struct zone *zone = page_zone(page);
6293 unsigned long pfn = page_to_pfn(page);
6294 unsigned long flags;
6295 int order;
6296
6297 spin_lock_irqsave(&zone->lock, flags);
6298 for (order = 0; order < MAX_ORDER; order++) {
6299 struct page *page_head = page - (pfn & ((1 << order) - 1));
6300
6301 if (PageBuddy(page_head) && page_order(page_head) >= order)
6302 break;
6303 }
6304 spin_unlock_irqrestore(&zone->lock, flags);
6305
6306 return order < MAX_ORDER;
6307}
6308#endif
Wu Fengguang718a3822010-03-10 15:20:43 -08006309
Andrew Morton51300ce2012-05-29 15:06:44 -07006310static const struct trace_print_flags pageflag_names[] = {
Wu Fengguang718a3822010-03-10 15:20:43 -08006311 {1UL << PG_locked, "locked" },
6312 {1UL << PG_error, "error" },
6313 {1UL << PG_referenced, "referenced" },
6314 {1UL << PG_uptodate, "uptodate" },
6315 {1UL << PG_dirty, "dirty" },
6316 {1UL << PG_lru, "lru" },
6317 {1UL << PG_active, "active" },
6318 {1UL << PG_slab, "slab" },
6319 {1UL << PG_owner_priv_1, "owner_priv_1" },
6320 {1UL << PG_arch_1, "arch_1" },
6321 {1UL << PG_reserved, "reserved" },
6322 {1UL << PG_private, "private" },
6323 {1UL << PG_private_2, "private_2" },
6324 {1UL << PG_writeback, "writeback" },
6325#ifdef CONFIG_PAGEFLAGS_EXTENDED
6326 {1UL << PG_head, "head" },
6327 {1UL << PG_tail, "tail" },
6328#else
6329 {1UL << PG_compound, "compound" },
6330#endif
6331 {1UL << PG_swapcache, "swapcache" },
6332 {1UL << PG_mappedtodisk, "mappedtodisk" },
6333 {1UL << PG_reclaim, "reclaim" },
Wu Fengguang718a3822010-03-10 15:20:43 -08006334 {1UL << PG_swapbacked, "swapbacked" },
6335 {1UL << PG_unevictable, "unevictable" },
6336#ifdef CONFIG_MMU
6337 {1UL << PG_mlocked, "mlocked" },
6338#endif
6339#ifdef CONFIG_ARCH_USES_PG_UNCACHED
6340 {1UL << PG_uncached, "uncached" },
6341#endif
6342#ifdef CONFIG_MEMORY_FAILURE
6343 {1UL << PG_hwpoison, "hwpoison" },
6344#endif
Gavin Shanbe9cd872012-05-29 15:06:44 -07006345#ifdef CONFIG_TRANSPARENT_HUGEPAGE
6346 {1UL << PG_compound_lock, "compound_lock" },
6347#endif
Wu Fengguang718a3822010-03-10 15:20:43 -08006348};
6349
6350static void dump_page_flags(unsigned long flags)
6351{
6352 const char *delim = "";
6353 unsigned long mask;
6354 int i;
6355
Andrew Morton51300ce2012-05-29 15:06:44 -07006356 BUILD_BUG_ON(ARRAY_SIZE(pageflag_names) != __NR_PAGEFLAGS);
Johannes Weineracc50c12012-05-29 15:06:44 -07006357
Wu Fengguang718a3822010-03-10 15:20:43 -08006358 printk(KERN_ALERT "page flags: %#lx(", flags);
6359
6360 /* remove zone id */
6361 flags &= (1UL << NR_PAGEFLAGS) - 1;
6362
Andrew Morton51300ce2012-05-29 15:06:44 -07006363 for (i = 0; i < ARRAY_SIZE(pageflag_names) && flags; i++) {
Wu Fengguang718a3822010-03-10 15:20:43 -08006364
6365 mask = pageflag_names[i].mask;
6366 if ((flags & mask) != mask)
6367 continue;
6368
6369 flags &= ~mask;
6370 printk("%s%s", delim, pageflag_names[i].name);
6371 delim = "|";
6372 }
6373
6374 /* check for left over flags */
6375 if (flags)
6376 printk("%s%#lx", delim, flags);
6377
6378 printk(")\n");
6379}
6380
6381void dump_page(struct page *page)
6382{
6383 printk(KERN_ALERT
6384 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
Andrea Arcangeli4e9f64c2011-01-13 15:46:29 -08006385 page, atomic_read(&page->_count), page_mapcount(page),
Wu Fengguang718a3822010-03-10 15:20:43 -08006386 page->mapping, page->index);
6387 dump_page_flags(page->flags);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07006388 mem_cgroup_print_bad_page(page);
Wu Fengguang718a3822010-03-10 15:20:43 -08006389}