blob: 18cee0d4c8a20705a4b3e7dd73e7d1d8a8b8d595 [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 Hiroyukia5d76b52007-10-16 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>
Lisa Du6e543d52013-09-11 14:22:36 -070059#include <linux/mm_inline.h>
Michal Nazarewicz041d3a82011-12-29 13:09:50 +010060#include <linux/migrate.h>
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -080061#include <linux/page-debug-flags.h>
David Rientjes949f7ec2013-04-29 15:07:48 -070062#include <linux/hugetlb.h>
Clark Williams8bd75c72013-02-07 09:47:07 -060063#include <linux/sched/rt.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
Jiang Liu7ee3d4e2013-07-03 15:03:41 -070065#include <asm/sections.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070066#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070067#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070068#include "internal.h"
69
Cody P Schaferc8e251f2013-07-03 15:01:29 -070070/* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
71static DEFINE_MUTEX(pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -070072#define MIN_PERCPU_PAGELIST_FRACTION (8)
Cody P Schaferc8e251f2013-07-03 15:01:29 -070073
Lee Schermerhorn72812012010-05-26 14:44:56 -070074#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
75DEFINE_PER_CPU(int, numa_node);
76EXPORT_PER_CPU_SYMBOL(numa_node);
77#endif
78
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070079#ifdef CONFIG_HAVE_MEMORYLESS_NODES
80/*
81 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
82 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
83 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
84 * defined in <linux/topology.h>.
85 */
86DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
87EXPORT_PER_CPU_SYMBOL(_numa_mem_);
88#endif
89
Linus Torvalds1da177e2005-04-16 15:20:36 -070090/*
Christoph Lameter13808912007-10-16 01:25:27 -070091 * Array of node states.
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 */
Christoph Lameter13808912007-10-16 01:25:27 -070093nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
94 [N_POSSIBLE] = NODE_MASK_ALL,
95 [N_ONLINE] = { { [0] = 1UL } },
96#ifndef CONFIG_NUMA
97 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
98#ifdef CONFIG_HIGHMEM
99 [N_HIGH_MEMORY] = { { [0] = 1UL } },
100#endif
Lai Jiangshan20b2f522012-12-12 13:52:00 -0800101#ifdef CONFIG_MOVABLE_NODE
102 [N_MEMORY] = { { [0] = 1UL } },
103#endif
Christoph Lameter13808912007-10-16 01:25:27 -0700104 [N_CPU] = { { [0] = 1UL } },
105#endif /* NUMA */
106};
107EXPORT_SYMBOL(node_states);
108
Jiang Liuc3d5f5f2013-07-03 15:03:14 -0700109/* Protect totalram_pages and zone->managed_pages */
110static DEFINE_SPINLOCK(managed_page_count_lock);
111
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700112unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -0700113unsigned long totalreserve_pages __read_mostly;
Johannes Weinerab8fabd2012-01-10 15:07:42 -0800114/*
115 * When calculating the number of globally allowed dirty pages, there
116 * is a certain number of per-zone reserves that should not be
117 * considered dirtyable memory. This is the sum of those reserves
118 * over all existing zones that contribute dirtyable memory.
119 */
120unsigned long dirty_balance_reserve __read_mostly;
121
Hugh Dickins1b76b022012-05-11 01:00:07 -0700122int percpu_pagelist_fraction;
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000123gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800125#ifdef CONFIG_PM_SLEEP
126/*
127 * The following functions are used by the suspend/hibernate code to temporarily
128 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
129 * while devices are suspended. To avoid races with the suspend/hibernate code,
130 * they should always be called with pm_mutex held (gfp_allowed_mask also should
131 * only be modified with pm_mutex held, unless the suspend/hibernate code is
132 * guaranteed not to run in parallel with that modification).
133 */
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100134
135static gfp_t saved_gfp_mask;
136
137void pm_restore_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800138{
139 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100140 if (saved_gfp_mask) {
141 gfp_allowed_mask = saved_gfp_mask;
142 saved_gfp_mask = 0;
143 }
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800144}
145
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100146void pm_restrict_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800147{
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800148 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100149 WARN_ON(saved_gfp_mask);
150 saved_gfp_mask = gfp_allowed_mask;
151 gfp_allowed_mask &= ~GFP_IOFS;
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800152}
Mel Gormanf90ac392012-01-10 15:07:15 -0800153
154bool pm_suspended_storage(void)
155{
156 if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS)
157 return false;
158 return true;
159}
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800160#endif /* CONFIG_PM_SLEEP */
161
Mel Gormand9c23402007-10-16 01:26:01 -0700162#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
163int pageblock_order __read_mostly;
164#endif
165
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800166static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -0800167
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168/*
169 * results with 256, 32 in the lowmem_reserve sysctl:
170 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
171 * 1G machine -> (16M dma, 784M normal, 224M high)
172 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
173 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
174 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +0100175 *
176 * TBD: should special case ZONE_DMA32 machines here - in those we normally
177 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700179int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800180#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700181 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -0800182#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700183#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700184 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700185#endif
Christoph Lametere53ef382006-09-25 23:31:14 -0700186#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700187 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700188#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700189 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700190};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191
192EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
Helge Deller15ad7cd2006-12-06 20:40:36 -0800194static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800195#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700196 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800197#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700198#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700199 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700200#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700201 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700202#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700203 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700204#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700205 "Movable",
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700206};
207
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208int min_free_kbytes = 1024;
Han Pingtian42aa83c2014-01-23 15:53:28 -0800209int user_min_free_kbytes = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
Jan Beulich2c85f512009-09-21 17:03:07 -0700211static unsigned long __meminitdata nr_kernel_pages;
212static unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700213static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214
Tejun Heo0ee332c2011-12-08 10:22:09 -0800215#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
216static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
217static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
218static unsigned long __initdata required_kernelcore;
219static unsigned long __initdata required_movablecore;
220static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Mel Gormanc7132162006-09-27 01:49:43 -0700221
Tejun Heo0ee332c2011-12-08 10:22:09 -0800222/* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
223int movable_zone;
224EXPORT_SYMBOL(movable_zone);
225#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -0700226
Miklos Szeredi418508c2007-05-23 13:57:55 -0700227#if MAX_NUMNODES > 1
228int nr_node_ids __read_mostly = MAX_NUMNODES;
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700229int nr_online_nodes __read_mostly = 1;
Miklos Szeredi418508c2007-05-23 13:57:55 -0700230EXPORT_SYMBOL(nr_node_ids);
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700231EXPORT_SYMBOL(nr_online_nodes);
Miklos Szeredi418508c2007-05-23 13:57:55 -0700232#endif
233
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700234int page_group_by_mobility_disabled __read_mostly;
235
Minchan Kimee6f5092012-07-31 16:43:50 -0700236void set_pageblock_migratetype(struct page *page, int migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700237{
KOSAKI Motohiro5d0f3f72013-11-12 15:08:18 -0800238 if (unlikely(page_group_by_mobility_disabled &&
239 migratetype < MIGRATE_PCPTYPES))
Mel Gorman49255c62009-06-16 15:31:58 -0700240 migratetype = MIGRATE_UNMOVABLE;
241
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700242 set_pageblock_flags_group(page, (unsigned long)migratetype,
243 PB_migrate, PB_migrate_end);
244}
245
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -0700246bool oom_killer_disabled __read_mostly;
247
Nick Piggin13e74442006-01-06 00:10:58 -0800248#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700249static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700251 int ret = 0;
252 unsigned seq;
253 unsigned long pfn = page_to_pfn(page);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800254 unsigned long sp, start_pfn;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700255
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700256 do {
257 seq = zone_span_seqbegin(zone);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800258 start_pfn = zone->zone_start_pfn;
259 sp = zone->spanned_pages;
Cody P Schafer108bcc92013-02-22 16:35:23 -0800260 if (!zone_spans_pfn(zone, pfn))
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700261 ret = 1;
262 } while (zone_span_seqretry(zone, seq));
263
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800264 if (ret)
Dave Hansen613813e2014-06-04 16:07:27 -0700265 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
266 pfn, zone_to_nid(zone), zone->name,
267 start_pfn, start_pfn + sp);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800268
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700269 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700270}
271
272static int page_is_consistent(struct zone *zone, struct page *page)
273{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700274 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700275 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700277 return 0;
278
279 return 1;
280}
281/*
282 * Temporary debugging check for pages not lying within a given zone.
283 */
284static int bad_range(struct zone *zone, struct page *page)
285{
286 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700288 if (!page_is_consistent(zone, page))
289 return 1;
290
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 return 0;
292}
Nick Piggin13e74442006-01-06 00:10:58 -0800293#else
294static inline int bad_range(struct zone *zone, struct page *page)
295{
296 return 0;
297}
298#endif
299
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -0700300static void bad_page(struct page *page, const char *reason,
301 unsigned long bad_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800303 static unsigned long resume;
304 static unsigned long nr_shown;
305 static unsigned long nr_unshown;
306
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200307 /* Don't complain about poisoned pages */
308 if (PageHWPoison(page)) {
Mel Gorman22b751c2013-02-22 16:34:59 -0800309 page_mapcount_reset(page); /* remove PageBuddy */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200310 return;
311 }
312
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800313 /*
314 * Allow a burst of 60 reports, then keep quiet for that minute;
315 * or allow a steady drip of one report per second.
316 */
317 if (nr_shown == 60) {
318 if (time_before(jiffies, resume)) {
319 nr_unshown++;
320 goto out;
321 }
322 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800323 printk(KERN_ALERT
324 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800325 nr_unshown);
326 nr_unshown = 0;
327 }
328 nr_shown = 0;
329 }
330 if (nr_shown++ == 0)
331 resume = jiffies + 60 * HZ;
332
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800333 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800334 current->comm, page_to_pfn(page));
Dave Hansenf0b791a2014-01-23 15:52:49 -0800335 dump_page_badflags(page, reason, bad_flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700336
Dave Jones4f318882011-10-31 17:07:24 -0700337 print_modules();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800339out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800340 /* Leave bad fields for debug, except PageBuddy could make trouble */
Mel Gorman22b751c2013-02-22 16:34:59 -0800341 page_mapcount_reset(page); /* remove PageBuddy */
Rusty Russell373d4d02013-01-21 17:17:39 +1030342 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343}
344
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345/*
346 * Higher-order pages are called "compound pages". They are structured thusly:
347 *
348 * The first PAGE_SIZE page is called the "head page".
349 *
350 * The remaining PAGE_SIZE pages are called "tail pages".
351 *
Wang Sheng-Hui6416b9f2011-11-17 10:53:50 +0100352 * All pages have PG_compound set. All tail pages have their ->first_page
353 * pointing at the head page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800355 * The first tail page's ->lru.next holds the address of the compound page's
356 * put_page() function. Its ->lru.prev holds the order of allocation.
357 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800359
360static void free_compound_page(struct page *page)
361{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700362 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800363}
364
Andi Kleen01ad1c02008-07-23 21:27:46 -0700365void prep_compound_page(struct page *page, unsigned long order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366{
367 int i;
368 int nr_pages = 1 << order;
369
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800370 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700371 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700372 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800373 for (i = 1; i < nr_pages; i++) {
374 struct page *p = page + i;
Youquan Song58a84aa2011-12-08 14:34:18 -0800375 set_page_count(p, 0);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700376 p->first_page = page;
David Rientjes668f9abb2014-03-03 15:38:18 -0800377 /* Make sure p->first_page is always valid for PageTail() */
378 smp_wmb();
379 __SetPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 }
381}
382
Andrea Arcangeli59ff4212011-01-13 15:46:38 -0800383/* update __split_huge_page_refcount if you change this function */
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800384static int destroy_compound_page(struct page *page, unsigned long order)
Andy Whitcroft18229df2008-11-06 12:53:27 -0800385{
386 int i;
387 int nr_pages = 1 << order;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800388 int bad = 0;
Andy Whitcroft18229df2008-11-06 12:53:27 -0800389
Gavin Shan0bb2c762012-12-18 14:21:32 -0800390 if (unlikely(compound_order(page) != order)) {
Dave Hansenf0b791a2014-01-23 15:52:49 -0800391 bad_page(page, "wrong compound order", 0);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800392 bad++;
393 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394
Christoph Lameter6d777952007-05-06 14:49:40 -0700395 __ClearPageHead(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800396
Andy Whitcroft18229df2008-11-06 12:53:27 -0800397 for (i = 1; i < nr_pages; i++) {
398 struct page *p = page + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399
Dave Hansenf0b791a2014-01-23 15:52:49 -0800400 if (unlikely(!PageTail(p))) {
401 bad_page(page, "PageTail not set", 0);
402 bad++;
403 } else if (unlikely(p->first_page != page)) {
404 bad_page(page, "first_page not consistent", 0);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800405 bad++;
406 }
Christoph Lameterd85f3382007-05-06 14:49:39 -0700407 __ClearPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800409
410 return bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700413static inline void prep_zero_page(struct page *page, unsigned int order,
414 gfp_t gfp_flags)
Nick Piggin17cf4402006-03-22 00:08:41 -0800415{
416 int i;
417
Andrew Morton6626c5d2006-03-22 00:08:42 -0800418 /*
419 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
420 * and __GFP_HIGHMEM from hard or soft interrupt context.
421 */
Nick Piggin725d7042006-09-25 23:30:55 -0700422 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800423 for (i = 0; i < (1 << order); i++)
424 clear_highpage(page + i);
425}
426
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800427#ifdef CONFIG_DEBUG_PAGEALLOC
428unsigned int _debug_guardpage_minorder;
429
430static int __init debug_guardpage_minorder_setup(char *buf)
431{
432 unsigned long res;
433
434 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
435 printk(KERN_ERR "Bad debug_guardpage_minorder value\n");
436 return 0;
437 }
438 _debug_guardpage_minorder = res;
439 printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res);
440 return 0;
441}
442__setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
443
444static inline void set_page_guard_flag(struct page *page)
445{
446 __set_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags);
447}
448
449static inline void clear_page_guard_flag(struct page *page)
450{
451 __clear_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags);
452}
453#else
454static inline void set_page_guard_flag(struct page *page) { }
455static inline void clear_page_guard_flag(struct page *page) { }
456#endif
457
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700458static inline void set_page_order(struct page *page, unsigned int order)
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700459{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700460 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000461 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462}
463
464static inline void rmv_page_order(struct page *page)
465{
Nick Piggin676165a2006-04-10 11:21:48 +1000466 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700467 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468}
469
470/*
471 * Locate the struct page for both the matching buddy in our
472 * pair (buddy1) and the combined O(n+1) page they form (page).
473 *
474 * 1) Any buddy B1 will have an order O twin B2 which satisfies
475 * the following equation:
476 * B2 = B1 ^ (1 << O)
477 * For example, if the starting buddy (buddy2) is #8 its order
478 * 1 buddy is #10:
479 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
480 *
481 * 2) Any buddy B will have an order O+1 parent P which
482 * satisfies the following equation:
483 * P = B & ~(1 << O)
484 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200485 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487static inline unsigned long
KyongHo Cho43506fa2011-01-13 15:47:24 -0800488__find_buddy_index(unsigned long page_idx, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489{
KyongHo Cho43506fa2011-01-13 15:47:24 -0800490 return page_idx ^ (1 << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491}
492
493/*
494 * This function checks whether a page is free && is the buddy
495 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800496 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000497 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700498 * (c) a page and its buddy have the same order &&
499 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 *
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700501 * For recording whether a page is in the buddy system, we set ->_mapcount
502 * PAGE_BUDDY_MAPCOUNT_VALUE.
503 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
504 * serialized by zone->lock.
Nick Piggin676165a2006-04-10 11:21:48 +1000505 *
506 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700508static inline int page_is_buddy(struct page *page, struct page *buddy,
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700509 unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700511 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800512 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800513
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800514 if (page_is_guard(buddy) && page_order(buddy) == order) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800515 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
Mel Gormand34c5fa2014-06-04 16:10:10 -0700516
517 if (page_zone_id(page) != page_zone_id(buddy))
518 return 0;
519
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800520 return 1;
521 }
522
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700523 if (PageBuddy(buddy) && page_order(buddy) == order) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800524 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
Mel Gormand34c5fa2014-06-04 16:10:10 -0700525
526 /*
527 * zone check is done late to avoid uselessly
528 * calculating zone/node ids for pages that could
529 * never merge.
530 */
531 if (page_zone_id(page) != page_zone_id(buddy))
532 return 0;
533
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700534 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000535 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700536 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537}
538
539/*
540 * Freeing function for a buddy system allocator.
541 *
542 * The concept of a buddy system is to maintain direct-mapped table
543 * (containing bit values) for memory blocks of various "orders".
544 * The bottom level table contains the map for the smallest allocatable
545 * units of memory (here, pages), and each level above it describes
546 * pairs of units from the levels below, hence, "buddies".
547 * At a high level, all that happens here is marking the table entry
548 * at the bottom level available, and propagating the changes upward
549 * as necessary, plus some accounting needed to play nicely with other
550 * parts of the VM system.
551 * At each level, we keep a list of pages, which are heads of continuous
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700552 * free pages of length of (1 << order) and marked with _mapcount
553 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
554 * field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 * So when we are allocating or freeing one, we can derive the state of the
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100556 * other. That is, if we allocate a small block, and both were
557 * free, the remainder of the region must be split into blocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 * If a block is freed, and its buddy is also free, then this
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100559 * triggers coalescing into a block of larger size.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +0100561 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 */
563
Nick Piggin48db57f2006-01-08 01:00:42 -0800564static inline void __free_one_page(struct page *page,
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700565 unsigned long pfn,
Mel Gormaned0ae212009-06-16 15:32:07 -0700566 struct zone *zone, unsigned int order,
567 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568{
569 unsigned long page_idx;
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700570 unsigned long combined_idx;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800571 unsigned long uninitialized_var(buddy_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700572 struct page *buddy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573
Cody P Schaferd29bb972013-02-22 16:35:25 -0800574 VM_BUG_ON(!zone_is_initialized(zone));
575
Nick Piggin224abf92006-01-06 00:11:11 -0800576 if (unlikely(PageCompound(page)))
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800577 if (unlikely(destroy_compound_page(page, order)))
578 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579
Mel Gormaned0ae212009-06-16 15:32:07 -0700580 VM_BUG_ON(migratetype == -1);
581
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700582 page_idx = pfn & ((1 << MAX_ORDER) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583
Sasha Levin309381fea2014-01-23 15:52:54 -0800584 VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page);
585 VM_BUG_ON_PAGE(bad_range(zone, page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 while (order < MAX_ORDER-1) {
KyongHo Cho43506fa2011-01-13 15:47:24 -0800588 buddy_idx = __find_buddy_index(page_idx, order);
589 buddy = page + (buddy_idx - page_idx);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700590 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700591 break;
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800592 /*
593 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
594 * merge with it and move up one order.
595 */
596 if (page_is_guard(buddy)) {
597 clear_page_guard_flag(buddy);
598 set_page_private(page, 0);
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -0700599 __mod_zone_freepage_state(zone, 1 << order,
600 migratetype);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800601 } else {
602 list_del(&buddy->lru);
603 zone->free_area[order].nr_free--;
604 rmv_page_order(buddy);
605 }
KyongHo Cho43506fa2011-01-13 15:47:24 -0800606 combined_idx = buddy_idx & page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 page = page + (combined_idx - page_idx);
608 page_idx = combined_idx;
609 order++;
610 }
611 set_page_order(page, order);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700612
613 /*
614 * If this is not the largest possible page, check if the buddy
615 * of the next-highest order is free. If it is, it's possible
616 * that pages are being freed that will coalesce soon. In case,
617 * that is happening, add the free page to the tail of the list
618 * so it's less likely to be used soon and more likely to be merged
619 * as a higher order page
620 */
Mel Gormanb7f50cf2010-10-26 14:21:11 -0700621 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700622 struct page *higher_page, *higher_buddy;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800623 combined_idx = buddy_idx & page_idx;
624 higher_page = page + (combined_idx - page_idx);
625 buddy_idx = __find_buddy_index(combined_idx, order + 1);
Li Haifeng0ba8f2d2012-09-17 14:09:21 -0700626 higher_buddy = higher_page + (buddy_idx - combined_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700627 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
628 list_add_tail(&page->lru,
629 &zone->free_area[order].free_list[migratetype]);
630 goto out;
631 }
632 }
633
634 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
635out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 zone->free_area[order].nr_free++;
637}
638
Nick Piggin224abf92006-01-06 00:11:11 -0800639static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640{
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -0700641 const char *bad_reason = NULL;
Dave Hansenf0b791a2014-01-23 15:52:49 -0800642 unsigned long bad_flags = 0;
643
644 if (unlikely(page_mapcount(page)))
645 bad_reason = "nonzero mapcount";
646 if (unlikely(page->mapping != NULL))
647 bad_reason = "non-NULL mapping";
648 if (unlikely(atomic_read(&page->_count) != 0))
649 bad_reason = "nonzero _count";
650 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
651 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
652 bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
653 }
654 if (unlikely(mem_cgroup_bad_page_check(page)))
655 bad_reason = "cgroup check failed";
656 if (unlikely(bad_reason)) {
657 bad_page(page, bad_reason, bad_flags);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800658 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800659 }
Peter Zijlstra90572892013-10-07 11:29:20 +0100660 page_cpupid_reset_last(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800661 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
662 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
663 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664}
665
666/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700667 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700669 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 *
671 * If the zone was previously in an "all pages pinned" state then look to
672 * see if this freeing clears that state.
673 *
674 * And clear the zone's pages_scanned counter, to hold off the "all pages are
675 * pinned" detection logic.
676 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700677static void free_pcppages_bulk(struct zone *zone, int count,
678 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679{
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700680 int migratetype = 0;
Mel Gormana6f9edd2009-09-21 17:03:20 -0700681 int batch_free = 0;
Mel Gorman72853e22010-09-09 16:38:16 -0700682 int to_free = count;
Mel Gorman0d5d8232014-08-06 16:07:16 -0700683 unsigned long nr_scanned;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700684
Nick Pigginc54ad302006-01-06 00:10:56 -0800685 spin_lock(&zone->lock);
Mel Gorman0d5d8232014-08-06 16:07:16 -0700686 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
687 if (nr_scanned)
688 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
Mel Gormanf2260e62009-06-16 15:32:13 -0700689
Mel Gorman72853e22010-09-09 16:38:16 -0700690 while (to_free) {
Nick Piggin48db57f2006-01-08 01:00:42 -0800691 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700692 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -0800693
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700694 /*
Mel Gormana6f9edd2009-09-21 17:03:20 -0700695 * Remove pages from lists in a round-robin fashion. A
696 * batch_free count is maintained that is incremented when an
697 * empty list is encountered. This is so more pages are freed
698 * off fuller lists instead of spinning excessively around empty
699 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700700 */
701 do {
Mel Gormana6f9edd2009-09-21 17:03:20 -0700702 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700703 if (++migratetype == MIGRATE_PCPTYPES)
704 migratetype = 0;
705 list = &pcp->lists[migratetype];
706 } while (list_empty(list));
707
Namhyung Kim1d168712011-03-22 16:32:45 -0700708 /* This is the only non-empty list. Free them all. */
709 if (batch_free == MIGRATE_PCPTYPES)
710 batch_free = to_free;
711
Mel Gormana6f9edd2009-09-21 17:03:20 -0700712 do {
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -0700713 int mt; /* migratetype of the to-be-freed page */
714
Mel Gormana6f9edd2009-09-21 17:03:20 -0700715 page = list_entry(list->prev, struct page, lru);
716 /* must delete as __free_one_page list manipulates */
717 list_del(&page->lru);
Minchan Kimb12c4ad2012-10-08 16:32:08 -0700718 mt = get_freepage_migratetype(page);
Hugh Dickinsa7016232010-01-29 17:46:34 +0000719 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700720 __free_one_page(page, page_to_pfn(page), zone, 0, mt);
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -0700721 trace_mm_page_pcpu_drain(page, 0, mt);
Minchan Kim194159f2013-02-22 16:33:58 -0800722 if (likely(!is_migrate_isolate_page(page))) {
Wen Congyang97d0da22012-12-11 16:00:52 -0800723 __mod_zone_page_state(zone, NR_FREE_PAGES, 1);
724 if (is_migrate_cma(mt))
725 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, 1);
726 }
Mel Gorman72853e22010-09-09 16:38:16 -0700727 } while (--to_free && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800729 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730}
731
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700732static void free_one_page(struct zone *zone,
733 struct page *page, unsigned long pfn,
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700734 unsigned int order,
Mel Gormaned0ae212009-06-16 15:32:07 -0700735 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800736{
Mel Gorman0d5d8232014-08-06 16:07:16 -0700737 unsigned long nr_scanned;
Christoph Lameter006d22d2006-09-25 23:31:48 -0700738 spin_lock(&zone->lock);
Mel Gorman0d5d8232014-08-06 16:07:16 -0700739 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
740 if (nr_scanned)
741 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
Mel Gormanf2260e62009-06-16 15:32:13 -0700742
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700743 __free_one_page(page, pfn, zone, order, migratetype);
Minchan Kim194159f2013-02-22 16:33:58 -0800744 if (unlikely(!is_migrate_isolate(migratetype)))
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -0700745 __mod_zone_freepage_state(zone, 1 << order, migratetype);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700746 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800747}
748
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700749static bool free_pages_prepare(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 int i;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800752 int bad = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -0800754 trace_mm_page_free(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100755 kmemcheck_free_shadow(page, order);
756
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -0800757 if (PageAnon(page))
758 page->mapping = NULL;
759 for (i = 0; i < (1 << order); i++)
760 bad += free_pages_check(page + i);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800761 if (bad)
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700762 return false;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800763
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700764 if (!PageHighMem(page)) {
Pintu Kumarb8af2942013-09-11 14:20:34 -0700765 debug_check_no_locks_freed(page_address(page),
766 PAGE_SIZE << order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700767 debug_check_no_obj_freed(page_address(page),
768 PAGE_SIZE << order);
769 }
Nick Piggindafb1362006-10-11 01:21:30 -0700770 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800771 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700772
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700773 return true;
774}
775
776static void __free_pages_ok(struct page *page, unsigned int order)
777{
778 unsigned long flags;
Minchan Kim95e34412012-10-08 16:32:11 -0700779 int migratetype;
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700780 unsigned long pfn = page_to_pfn(page);
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700781
782 if (!free_pages_prepare(page, order))
783 return;
784
Mel Gormancfc47a22014-06-04 16:10:19 -0700785 migratetype = get_pfnblock_migratetype(page, pfn);
Nick Pigginc54ad302006-01-06 00:10:56 -0800786 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700787 __count_vm_events(PGFREE, 1 << order);
Minchan Kim95e34412012-10-08 16:32:11 -0700788 set_freepage_migratetype(page, migratetype);
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700789 free_one_page(page_zone(page), page, pfn, order, migratetype);
Nick Pigginc54ad302006-01-06 00:10:56 -0800790 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791}
792
Jiang Liu170a5a72013-07-03 15:03:17 -0700793void __init __free_pages_bootmem(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800794{
Johannes Weinerc3993072012-01-10 15:08:10 -0800795 unsigned int nr_pages = 1 << order;
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700796 struct page *p = page;
Johannes Weinerc3993072012-01-10 15:08:10 -0800797 unsigned int loop;
David Howellsa226f6c2006-01-06 00:11:08 -0800798
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700799 prefetchw(p);
800 for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
801 prefetchw(p + 1);
Johannes Weinerc3993072012-01-10 15:08:10 -0800802 __ClearPageReserved(p);
803 set_page_count(p, 0);
David Howellsa226f6c2006-01-06 00:11:08 -0800804 }
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700805 __ClearPageReserved(p);
806 set_page_count(p, 0);
Johannes Weinerc3993072012-01-10 15:08:10 -0800807
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700808 page_zone(page)->managed_pages += nr_pages;
Johannes Weinerc3993072012-01-10 15:08:10 -0800809 set_page_refcounted(page);
810 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800811}
812
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100813#ifdef CONFIG_CMA
Li Zhong9cf510a2013-08-23 13:52:52 +0800814/* Free whole pageblock and set its migration type to MIGRATE_CMA. */
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100815void __init init_cma_reserved_pageblock(struct page *page)
816{
817 unsigned i = pageblock_nr_pages;
818 struct page *p = page;
819
820 do {
821 __ClearPageReserved(p);
822 set_page_count(p, 0);
823 } while (++p, --i);
824
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100825 set_pageblock_migratetype(page, MIGRATE_CMA);
Michal Nazarewiczdc783272014-07-02 15:22:35 -0700826
827 if (pageblock_order >= MAX_ORDER) {
828 i = pageblock_nr_pages;
829 p = page;
830 do {
831 set_page_refcounted(p);
832 __free_pages(p, MAX_ORDER - 1);
833 p += MAX_ORDER_NR_PAGES;
834 } while (i -= MAX_ORDER_NR_PAGES);
835 } else {
836 set_page_refcounted(page);
837 __free_pages(page, pageblock_order);
838 }
839
Jiang Liu3dcc0572013-07-03 15:03:21 -0700840 adjust_managed_page_count(page, pageblock_nr_pages);
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100841}
842#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843
844/*
845 * The order of subdivision here is critical for the IO subsystem.
846 * Please do not alter this order without good reasons and regression
847 * testing. Specifically, as large blocks of memory are subdivided,
848 * the order in which smaller blocks are delivered depends on the order
849 * they're subdivided in this function. This is the primary factor
850 * influencing the order in which pages are delivered to the IO
851 * subsystem according to empirical testing, and this is also justified
852 * by considering the behavior of a buddy system containing a single
853 * large block of memory acted on by a series of small allocations.
854 * This behavior is a critical factor in sglist merging's success.
855 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +0100856 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800858static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700859 int low, int high, struct free_area *area,
860 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861{
862 unsigned long size = 1 << high;
863
864 while (high > low) {
865 area--;
866 high--;
867 size >>= 1;
Sasha Levin309381fea2014-01-23 15:52:54 -0800868 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800869
870#ifdef CONFIG_DEBUG_PAGEALLOC
871 if (high < debug_guardpage_minorder()) {
872 /*
873 * Mark as guard pages (or page), that will allow to
874 * merge back to allocator when buddy will be freed.
875 * Corresponding page table entries will not be touched,
876 * pages will stay not present in virtual address space
877 */
878 INIT_LIST_HEAD(&page[size].lru);
879 set_page_guard_flag(&page[size]);
880 set_page_private(&page[size], high);
881 /* Guard pages are not available for any usage */
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -0700882 __mod_zone_freepage_state(zone, -(1 << high),
883 migratetype);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800884 continue;
885 }
886#endif
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700887 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 area->nr_free++;
889 set_page_order(&page[size], high);
890 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891}
892
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893/*
894 * This page is about to be returned from the page allocator
895 */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200896static inline int check_new_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897{
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -0700898 const char *bad_reason = NULL;
Dave Hansenf0b791a2014-01-23 15:52:49 -0800899 unsigned long bad_flags = 0;
900
901 if (unlikely(page_mapcount(page)))
902 bad_reason = "nonzero mapcount";
903 if (unlikely(page->mapping != NULL))
904 bad_reason = "non-NULL mapping";
905 if (unlikely(atomic_read(&page->_count) != 0))
906 bad_reason = "nonzero _count";
907 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
908 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
909 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
910 }
911 if (unlikely(mem_cgroup_bad_page_check(page)))
912 bad_reason = "cgroup check failed";
913 if (unlikely(bad_reason)) {
914 bad_page(page, bad_reason, bad_flags);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800915 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800916 }
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200917 return 0;
918}
919
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700920static int prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags)
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200921{
922 int i;
923
924 for (i = 0; i < (1 << order); i++) {
925 struct page *p = page + i;
926 if (unlikely(check_new_page(p)))
927 return 1;
928 }
Hugh Dickins689bceb2005-11-21 21:32:20 -0800929
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700930 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800931 set_page_refcounted(page);
Nick Piggincc1025092006-12-06 20:32:00 -0800932
933 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800935
936 if (gfp_flags & __GFP_ZERO)
937 prep_zero_page(page, order, gfp_flags);
938
939 if (order && (gfp_flags & __GFP_COMP))
940 prep_compound_page(page, order);
941
Hugh Dickins689bceb2005-11-21 21:32:20 -0800942 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943}
944
Mel Gorman56fd56b2007-10-16 01:25:58 -0700945/*
946 * Go through the free lists for the given migratetype and remove
947 * the smallest available page from the freelists
948 */
Mel Gorman728ec982009-06-16 15:32:04 -0700949static inline
950struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -0700951 int migratetype)
952{
953 unsigned int current_order;
Pintu Kumarb8af2942013-09-11 14:20:34 -0700954 struct free_area *area;
Mel Gorman56fd56b2007-10-16 01:25:58 -0700955 struct page *page;
956
957 /* Find a page of the appropriate size in the preferred list */
958 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
959 area = &(zone->free_area[current_order]);
960 if (list_empty(&area->free_list[migratetype]))
961 continue;
962
963 page = list_entry(area->free_list[migratetype].next,
964 struct page, lru);
965 list_del(&page->lru);
966 rmv_page_order(page);
967 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -0700968 expand(zone, page, order, current_order, area, migratetype);
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -0700969 set_freepage_migratetype(page, migratetype);
Mel Gorman56fd56b2007-10-16 01:25:58 -0700970 return page;
971 }
972
973 return NULL;
974}
975
976
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700977/*
978 * This array describes the order lists are fallen back to when
979 * the free lists for the desirable migrate type are depleted
980 */
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100981static int fallbacks[MIGRATE_TYPES][4] = {
982 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
983 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
984#ifdef CONFIG_CMA
985 [MIGRATE_MOVABLE] = { MIGRATE_CMA, MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
986 [MIGRATE_CMA] = { MIGRATE_RESERVE }, /* Never used */
987#else
988 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
989#endif
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +0100990 [MIGRATE_RESERVE] = { MIGRATE_RESERVE }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -0800991#ifdef CONFIG_MEMORY_ISOLATION
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +0100992 [MIGRATE_ISOLATE] = { MIGRATE_RESERVE }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -0800993#endif
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700994};
995
Mel Gormanc361be52007-10-16 01:25:51 -0700996/*
997 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -0700998 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -0700999 * boundary. If alignment is required, use move_freepages_block()
1000 */
Minchan Kim435b4052012-10-08 16:32:16 -07001001int move_freepages(struct zone *zone,
Adrian Bunkb69a7282008-07-23 21:28:12 -07001002 struct page *start_page, struct page *end_page,
1003 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001004{
1005 struct page *page;
1006 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -07001007 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -07001008
1009#ifndef CONFIG_HOLES_IN_ZONE
1010 /*
1011 * page_zone is not safe to call in this context when
1012 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1013 * anyway as we check zone boundaries in move_freepages_block().
1014 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -07001015 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -07001016 */
1017 BUG_ON(page_zone(start_page) != page_zone(end_page));
1018#endif
1019
1020 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -07001021 /* Make sure we are not inadvertently changing nodes */
Sasha Levin309381fea2014-01-23 15:52:54 -08001022 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
Adam Litke344c7902008-09-02 14:35:38 -07001023
Mel Gormanc361be52007-10-16 01:25:51 -07001024 if (!pfn_valid_within(page_to_pfn(page))) {
1025 page++;
1026 continue;
1027 }
1028
1029 if (!PageBuddy(page)) {
1030 page++;
1031 continue;
1032 }
1033
1034 order = page_order(page);
Kirill A. Shutemov84be48d2011-03-22 16:33:41 -07001035 list_move(&page->lru,
1036 &zone->free_area[order].free_list[migratetype]);
Minchan Kim95e34412012-10-08 16:32:11 -07001037 set_freepage_migratetype(page, migratetype);
Mel Gormanc361be52007-10-16 01:25:51 -07001038 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -07001039 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -07001040 }
1041
Mel Gormand1003132007-10-16 01:26:00 -07001042 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -07001043}
1044
Minchan Kimee6f5092012-07-31 16:43:50 -07001045int move_freepages_block(struct zone *zone, struct page *page,
Linus Torvalds68e3e922012-06-03 20:05:57 -07001046 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001047{
1048 unsigned long start_pfn, end_pfn;
1049 struct page *start_page, *end_page;
1050
1051 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -07001052 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -07001053 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -07001054 end_page = start_page + pageblock_nr_pages - 1;
1055 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -07001056
1057 /* Do not cross zone boundaries */
Cody P Schafer108bcc92013-02-22 16:35:23 -08001058 if (!zone_spans_pfn(zone, start_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001059 start_page = page;
Cody P Schafer108bcc92013-02-22 16:35:23 -08001060 if (!zone_spans_pfn(zone, end_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001061 return 0;
1062
1063 return move_freepages(zone, start_page, end_page, migratetype);
1064}
1065
Mel Gorman2f66a682009-09-21 17:02:31 -07001066static void change_pageblock_range(struct page *pageblock_page,
1067 int start_order, int migratetype)
1068{
1069 int nr_pageblocks = 1 << (start_order - pageblock_order);
1070
1071 while (nr_pageblocks--) {
1072 set_pageblock_migratetype(pageblock_page, migratetype);
1073 pageblock_page += pageblock_nr_pages;
1074 }
1075}
1076
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001077/*
1078 * If breaking a large block of pages, move all free pages to the preferred
1079 * allocation list. If falling back for a reclaimable kernel allocation, be
1080 * more aggressive about taking ownership of free pages.
1081 *
1082 * On the other hand, never change migration type of MIGRATE_CMA pageblocks
1083 * nor move CMA pages to different free lists. We don't want unmovable pages
1084 * to be allocated from MIGRATE_CMA areas.
1085 *
1086 * Returns the new migratetype of the pageblock (or the same old migratetype
1087 * if it was unchanged).
1088 */
1089static int try_to_steal_freepages(struct zone *zone, struct page *page,
1090 int start_type, int fallback_type)
1091{
1092 int current_order = page_order(page);
1093
KOSAKI Motohiro0cbef292013-11-12 15:08:20 -08001094 /*
1095 * When borrowing from MIGRATE_CMA, we need to release the excess
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001096 * buddy pages to CMA itself. We also ensure the freepage_migratetype
1097 * is set to CMA so it is returned to the correct freelist in case
1098 * the page ends up being not actually allocated from the pcp lists.
KOSAKI Motohiro0cbef292013-11-12 15:08:20 -08001099 */
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001100 if (is_migrate_cma(fallback_type))
1101 return fallback_type;
1102
1103 /* Take ownership for orders >= pageblock_order */
1104 if (current_order >= pageblock_order) {
1105 change_pageblock_range(page, current_order, start_type);
1106 return start_type;
1107 }
1108
1109 if (current_order >= pageblock_order / 2 ||
1110 start_type == MIGRATE_RECLAIMABLE ||
1111 page_group_by_mobility_disabled) {
1112 int pages;
1113
1114 pages = move_freepages_block(zone, page, start_type);
1115
1116 /* Claim the whole block if over half of it is free */
1117 if (pages >= (1 << (pageblock_order-1)) ||
1118 page_group_by_mobility_disabled) {
1119
1120 set_pageblock_migratetype(page, start_type);
1121 return start_type;
1122 }
1123
1124 }
1125
1126 return fallback_type;
1127}
1128
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001129/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -07001130static inline struct page *
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001131__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001132{
Pintu Kumarb8af2942013-09-11 14:20:34 -07001133 struct free_area *area;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001134 unsigned int current_order;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001135 struct page *page;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001136 int migratetype, new_type, i;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001137
1138 /* Find the largest possible block of pages in the other list */
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001139 for (current_order = MAX_ORDER-1;
1140 current_order >= order && current_order <= MAX_ORDER-1;
1141 --current_order) {
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001142 for (i = 0;; i++) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001143 migratetype = fallbacks[start_migratetype][i];
1144
Mel Gorman56fd56b2007-10-16 01:25:58 -07001145 /* MIGRATE_RESERVE handled later if necessary */
1146 if (migratetype == MIGRATE_RESERVE)
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001147 break;
Mel Gormane0104872007-10-16 01:25:53 -07001148
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001149 area = &(zone->free_area[current_order]);
1150 if (list_empty(&area->free_list[migratetype]))
1151 continue;
1152
1153 page = list_entry(area->free_list[migratetype].next,
1154 struct page, lru);
1155 area->nr_free--;
1156
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001157 new_type = try_to_steal_freepages(zone, page,
1158 start_migratetype,
1159 migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001160
1161 /* Remove the page from the freelists */
1162 list_del(&page->lru);
1163 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001164
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001165 expand(zone, page, order, current_order, area,
KOSAKI Motohiro0cbef292013-11-12 15:08:20 -08001166 new_type);
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001167 /* The freepage_migratetype may differ from pageblock's
1168 * migratetype depending on the decisions in
1169 * try_to_steal_freepages. This is OK as long as it does
1170 * not differ for MIGRATE_CMA type.
1171 */
1172 set_freepage_migratetype(page, new_type);
Mel Gormane0fff1b2009-09-21 17:02:42 -07001173
KOSAKI Motohiro52c8f6a2013-11-12 15:08:19 -08001174 trace_mm_page_alloc_extfrag(page, order, current_order,
1175 start_migratetype, migratetype, new_type);
Mel Gormane0fff1b2009-09-21 17:02:42 -07001176
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001177 return page;
1178 }
1179 }
1180
Mel Gorman728ec982009-06-16 15:32:04 -07001181 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001182}
1183
Mel Gorman56fd56b2007-10-16 01:25:58 -07001184/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 * Do the hard work of removing an element from the buddy allocator.
1186 * Call me with the zone->lock already held.
1187 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001188static struct page *__rmqueue(struct zone *zone, unsigned int order,
1189 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 struct page *page;
1192
Mel Gorman728ec982009-06-16 15:32:04 -07001193retry_reserve:
Mel Gorman56fd56b2007-10-16 01:25:58 -07001194 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195
Mel Gorman728ec982009-06-16 15:32:04 -07001196 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07001197 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001198
Mel Gorman728ec982009-06-16 15:32:04 -07001199 /*
1200 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1201 * is used because __rmqueue_smallest is an inline function
1202 * and we want just one call site
1203 */
1204 if (!page) {
1205 migratetype = MIGRATE_RESERVE;
1206 goto retry_reserve;
1207 }
1208 }
1209
Mel Gorman0d3d0622009-09-21 17:02:44 -07001210 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001211 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212}
1213
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001214/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 * Obtain a specified number of elements from the buddy allocator, all under
1216 * a single hold of the lock, for efficiency. Add them to the supplied list.
1217 * Returns the number of new pages which were placed at *list.
1218 */
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001219static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001220 unsigned long count, struct list_head *list,
Mel Gormanb745bc82014-06-04 16:10:22 -07001221 int migratetype, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222{
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001223 int i;
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001224
Nick Pigginc54ad302006-01-06 00:10:56 -08001225 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001227 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001228 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08001230
1231 /*
1232 * Split buddy pages returned by expand() are received here
1233 * in physical page order. The page is added to the callers and
1234 * list and the list head then moves forward. From the callers
1235 * perspective, the linked list is ordered by page number in
1236 * some conditions. This is useful for IO devices that can
1237 * merge IO requests if the physical pages are ordered
1238 * properly.
1239 */
Mel Gormanb745bc82014-06-04 16:10:22 -07001240 if (likely(!cold))
Mel Gormane084b2d2009-07-29 15:02:04 -07001241 list_add(&page->lru, list);
1242 else
1243 list_add_tail(&page->lru, list);
Mel Gorman81eabcb2007-12-17 16:20:05 -08001244 list = &page->lru;
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001245 if (is_migrate_cma(get_freepage_migratetype(page)))
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07001246 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
1247 -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 }
Mel Gormanf2260e62009-06-16 15:32:13 -07001249 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08001250 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001251 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252}
1253
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001254#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001255/*
Christoph Lameter4037d452007-05-09 02:35:14 -07001256 * Called from the vmstat counter updater to drain pagesets of this
1257 * currently executing processor on remote nodes after they have
1258 * expired.
1259 *
Christoph Lameter879336c2006-03-22 00:09:08 -08001260 * Note that this function must be called with the thread pinned to
1261 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001262 */
Christoph Lameter4037d452007-05-09 02:35:14 -07001263void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001264{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001265 unsigned long flags;
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07001266 int to_drain, batch;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001267
Christoph Lameter4037d452007-05-09 02:35:14 -07001268 local_irq_save(flags);
Cody P Schafer998d39c2013-07-03 15:01:32 -07001269 batch = ACCESS_ONCE(pcp->batch);
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07001270 to_drain = min(pcp->count, batch);
KOSAKI Motohiro2a135152012-07-31 16:42:53 -07001271 if (to_drain > 0) {
1272 free_pcppages_bulk(zone, to_drain, pcp);
1273 pcp->count -= to_drain;
1274 }
Christoph Lameter4037d452007-05-09 02:35:14 -07001275 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001276}
1277#endif
1278
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001279/*
1280 * Drain pages of the indicated processor.
1281 *
1282 * The processor must either be the current processor and the
1283 * thread pinned to the current processor or a processor that
1284 * is not online.
1285 */
1286static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287{
Nick Pigginc54ad302006-01-06 00:10:56 -08001288 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07001291 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 struct per_cpu_pageset *pset;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001293 struct per_cpu_pages *pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001295 local_irq_save(flags);
1296 pset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001298 pcp = &pset->pcp;
David Rientjes2ff754f2011-01-25 15:07:23 -08001299 if (pcp->count) {
1300 free_pcppages_bulk(zone, pcp->count, pcp);
1301 pcp->count = 0;
1302 }
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001303 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 }
1305}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001307/*
1308 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1309 */
1310void drain_local_pages(void *arg)
1311{
1312 drain_pages(smp_processor_id());
1313}
1314
1315/*
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001316 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1317 *
1318 * Note that this code is protected against sending an IPI to an offline
1319 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1320 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1321 * nothing keeps CPUs from showing up after we populated the cpumask and
1322 * before the call to on_each_cpu_mask().
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001323 */
1324void drain_all_pages(void)
1325{
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001326 int cpu;
1327 struct per_cpu_pageset *pcp;
1328 struct zone *zone;
1329
1330 /*
1331 * Allocate in the BSS so we wont require allocation in
1332 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1333 */
1334 static cpumask_t cpus_with_pcps;
1335
1336 /*
1337 * We don't care about racing with CPU hotplug event
1338 * as offline notification will cause the notified
1339 * cpu to drain that CPU pcps and on_each_cpu_mask
1340 * disables preemption as part of its processing
1341 */
1342 for_each_online_cpu(cpu) {
1343 bool has_pcps = false;
1344 for_each_populated_zone(zone) {
1345 pcp = per_cpu_ptr(zone->pageset, cpu);
1346 if (pcp->pcp.count) {
1347 has_pcps = true;
1348 break;
1349 }
1350 }
1351 if (has_pcps)
1352 cpumask_set_cpu(cpu, &cpus_with_pcps);
1353 else
1354 cpumask_clear_cpu(cpu, &cpus_with_pcps);
1355 }
1356 on_each_cpu_mask(&cpus_with_pcps, drain_local_pages, NULL, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001357}
1358
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02001359#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360
1361void mark_free_pages(struct zone *zone)
1362{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001363 unsigned long pfn, max_zone_pfn;
1364 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001365 unsigned int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 struct list_head *curr;
1367
Xishi Qiu8080fc02013-09-11 14:21:45 -07001368 if (zone_is_empty(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 return;
1370
1371 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001372
Cody P Schafer108bcc92013-02-22 16:35:23 -08001373 max_zone_pfn = zone_end_pfn(zone);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001374 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1375 if (pfn_valid(pfn)) {
1376 struct page *page = pfn_to_page(pfn);
1377
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001378 if (!swsusp_page_is_forbidden(page))
1379 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001380 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001382 for_each_migratetype_order(order, t) {
1383 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001384 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001386 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1387 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001388 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001389 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001390 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 spin_unlock_irqrestore(&zone->lock, flags);
1392}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001393#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394
1395/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 * Free a 0-order page
Mel Gormanb745bc82014-06-04 16:10:22 -07001397 * cold == true ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 */
Mel Gormanb745bc82014-06-04 16:10:22 -07001399void free_hot_cold_page(struct page *page, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400{
1401 struct zone *zone = page_zone(page);
1402 struct per_cpu_pages *pcp;
1403 unsigned long flags;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001404 unsigned long pfn = page_to_pfn(page);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001405 int migratetype;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001407 if (!free_pages_prepare(page, 0))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001408 return;
1409
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001410 migratetype = get_pfnblock_migratetype(page, pfn);
Minchan Kimb12c4ad2012-10-08 16:32:08 -07001411 set_freepage_migratetype(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001413 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001414
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001415 /*
1416 * We only track unmovable, reclaimable and movable on pcp lists.
1417 * Free ISOLATE pages back to the allocator because they are being
1418 * offlined but treat RESERVE as movable pages so we can get those
1419 * areas back if necessary. Otherwise, we may have to free
1420 * excessively into the page allocator
1421 */
1422 if (migratetype >= MIGRATE_PCPTYPES) {
Minchan Kim194159f2013-02-22 16:33:58 -08001423 if (unlikely(is_migrate_isolate(migratetype))) {
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001424 free_one_page(zone, page, pfn, 0, migratetype);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001425 goto out;
1426 }
1427 migratetype = MIGRATE_MOVABLE;
1428 }
1429
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001430 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Mel Gormanb745bc82014-06-04 16:10:22 -07001431 if (!cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001432 list_add(&page->lru, &pcp->lists[migratetype]);
Mel Gormanb745bc82014-06-04 16:10:22 -07001433 else
1434 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001436 if (pcp->count >= pcp->high) {
Cody P Schafer998d39c2013-07-03 15:01:32 -07001437 unsigned long batch = ACCESS_ONCE(pcp->batch);
1438 free_pcppages_bulk(zone, batch, pcp);
1439 pcp->count -= batch;
Nick Piggin48db57f2006-01-08 01:00:42 -08001440 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001441
1442out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444}
1445
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001446/*
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001447 * Free a list of 0-order pages
1448 */
Mel Gormanb745bc82014-06-04 16:10:22 -07001449void free_hot_cold_page_list(struct list_head *list, bool cold)
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001450{
1451 struct page *page, *next;
1452
1453 list_for_each_entry_safe(page, next, list, lru) {
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -08001454 trace_mm_page_free_batched(page, cold);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001455 free_hot_cold_page(page, cold);
1456 }
1457}
1458
1459/*
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001460 * split_page takes a non-compound higher-order page, and splits it into
1461 * n (1<<order) sub-pages: page[0..n]
1462 * Each sub-page must be freed individually.
1463 *
1464 * Note: this is probably too low level an operation for use in drivers.
1465 * Please consult with lkml before using this in your driver.
1466 */
1467void split_page(struct page *page, unsigned int order)
1468{
1469 int i;
1470
Sasha Levin309381fea2014-01-23 15:52:54 -08001471 VM_BUG_ON_PAGE(PageCompound(page), page);
1472 VM_BUG_ON_PAGE(!page_count(page), page);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001473
1474#ifdef CONFIG_KMEMCHECK
1475 /*
1476 * Split shadow pages too, because free(page[0]) would
1477 * otherwise free the whole shadow.
1478 */
1479 if (kmemcheck_page_is_tracked(page))
1480 split_page(virt_to_page(page[0].shadow), order);
1481#endif
1482
Nick Piggin7835e982006-03-22 00:08:40 -08001483 for (i = 1; i < (1 << order); i++)
1484 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001485}
K. Y. Srinivasan5853ff22013-03-25 15:47:38 -07001486EXPORT_SYMBOL_GPL(split_page);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001487
Mel Gorman8fb74b92013-01-11 14:32:16 -08001488static int __isolate_free_page(struct page *page, unsigned int order)
Mel Gorman748446b2010-05-24 14:32:27 -07001489{
Mel Gorman748446b2010-05-24 14:32:27 -07001490 unsigned long watermark;
1491 struct zone *zone;
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07001492 int mt;
Mel Gorman748446b2010-05-24 14:32:27 -07001493
1494 BUG_ON(!PageBuddy(page));
1495
1496 zone = page_zone(page);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001497 mt = get_pageblock_migratetype(page);
Mel Gorman748446b2010-05-24 14:32:27 -07001498
Minchan Kim194159f2013-02-22 16:33:58 -08001499 if (!is_migrate_isolate(mt)) {
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001500 /* Obey watermarks as if the page was being allocated */
1501 watermark = low_wmark_pages(zone) + (1 << order);
1502 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1503 return 0;
1504
Mel Gorman8fb74b92013-01-11 14:32:16 -08001505 __mod_zone_freepage_state(zone, -(1UL << order), mt);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001506 }
Mel Gorman748446b2010-05-24 14:32:27 -07001507
1508 /* Remove page from free list */
1509 list_del(&page->lru);
1510 zone->free_area[order].nr_free--;
1511 rmv_page_order(page);
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07001512
Mel Gorman8fb74b92013-01-11 14:32:16 -08001513 /* Set the pageblock if the isolated page is at least a pageblock */
Mel Gorman748446b2010-05-24 14:32:27 -07001514 if (order >= pageblock_order - 1) {
1515 struct page *endpage = page + (1 << order) - 1;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001516 for (; page < endpage; page += pageblock_nr_pages) {
1517 int mt = get_pageblock_migratetype(page);
Minchan Kim194159f2013-02-22 16:33:58 -08001518 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001519 set_pageblock_migratetype(page,
1520 MIGRATE_MOVABLE);
1521 }
Mel Gorman748446b2010-05-24 14:32:27 -07001522 }
1523
Mel Gorman8fb74b92013-01-11 14:32:16 -08001524 return 1UL << order;
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001525}
1526
1527/*
1528 * Similar to split_page except the page is already free. As this is only
1529 * being used for migration, the migratetype of the block also changes.
1530 * As this is called with interrupts disabled, the caller is responsible
1531 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1532 * are enabled.
1533 *
1534 * Note: this is probably too low level an operation for use in drivers.
1535 * Please consult with lkml before using this in your driver.
1536 */
1537int split_free_page(struct page *page)
1538{
1539 unsigned int order;
1540 int nr_pages;
1541
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001542 order = page_order(page);
1543
Mel Gorman8fb74b92013-01-11 14:32:16 -08001544 nr_pages = __isolate_free_page(page, order);
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001545 if (!nr_pages)
1546 return 0;
1547
1548 /* Split into individual pages */
1549 set_page_refcounted(page);
1550 split_page(page, order);
1551 return nr_pages;
Mel Gorman748446b2010-05-24 14:32:27 -07001552}
1553
1554/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1556 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1557 * or two.
1558 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07001559static inline
1560struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001561 struct zone *zone, unsigned int order,
1562 gfp_t gfp_flags, int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563{
1564 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001565 struct page *page;
Mel Gormanb745bc82014-06-04 16:10:22 -07001566 bool cold = ((gfp_flags & __GFP_COLD) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567
Hugh Dickins689bceb2005-11-21 21:32:20 -08001568again:
Nick Piggin48db57f2006-01-08 01:00:42 -08001569 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001571 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 local_irq_save(flags);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001574 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1575 list = &pcp->lists[migratetype];
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001576 if (list_empty(list)) {
Mel Gorman535131e62007-10-16 01:25:49 -07001577 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001578 pcp->batch, list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001579 migratetype, cold);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001580 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07001581 goto failed;
Mel Gorman535131e62007-10-16 01:25:49 -07001582 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001583
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001584 if (cold)
1585 page = list_entry(list->prev, struct page, lru);
1586 else
1587 page = list_entry(list->next, struct page, lru);
1588
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001589 list_del(&page->lru);
1590 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001591 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07001592 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1593 /*
1594 * __GFP_NOFAIL is not to be used in new code.
1595 *
1596 * All __GFP_NOFAIL callers should be fixed so that they
1597 * properly detect and handle allocation failures.
1598 *
1599 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07001600 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07001601 * __GFP_NOFAIL.
1602 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07001603 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07001604 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001606 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001607 spin_unlock(&zone->lock);
1608 if (!page)
1609 goto failed;
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07001610 __mod_zone_freepage_state(zone, -(1 << order),
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001611 get_freepage_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 }
1613
Johannes Weiner3a025762014-04-07 15:37:48 -07001614 __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
Mel Gorman4ffeaf32014-08-06 16:07:22 -07001615 if (zone_page_state(zone, NR_ALLOC_BATCH) == 0 &&
1616 !zone_is_fair_depleted(zone))
1617 zone_set_flag(zone, ZONE_FAIR_DEPLETED);
Johannes Weiner27329362014-03-03 15:38:41 -08001618
Christoph Lameterf8891e52006-06-30 01:55:45 -07001619 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Andi Kleen78afd562011-03-22 16:33:12 -07001620 zone_statistics(preferred_zone, zone, gfp_flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001621 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622
Sasha Levin309381fea2014-01-23 15:52:54 -08001623 VM_BUG_ON_PAGE(bad_range(zone, page), page);
Nick Piggin17cf4402006-03-22 00:08:41 -08001624 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001625 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001627
1628failed:
1629 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001630 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631}
1632
Akinobu Mita933e3122006-12-08 02:39:45 -08001633#ifdef CONFIG_FAIL_PAGE_ALLOC
1634
Akinobu Mitab2588c42011-07-26 16:09:03 -07001635static struct {
Akinobu Mita933e3122006-12-08 02:39:45 -08001636 struct fault_attr attr;
1637
1638 u32 ignore_gfp_highmem;
1639 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001640 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001641} fail_page_alloc = {
1642 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001643 .ignore_gfp_wait = 1,
1644 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001645 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001646};
1647
1648static int __init setup_fail_page_alloc(char *str)
1649{
1650 return setup_fault_attr(&fail_page_alloc.attr, str);
1651}
1652__setup("fail_page_alloc=", setup_fail_page_alloc);
1653
Gavin Shandeaf3862012-07-31 16:41:51 -07001654static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08001655{
Akinobu Mita54114992007-07-15 23:40:23 -07001656 if (order < fail_page_alloc.min_order)
Gavin Shandeaf3862012-07-31 16:41:51 -07001657 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001658 if (gfp_mask & __GFP_NOFAIL)
Gavin Shandeaf3862012-07-31 16:41:51 -07001659 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001660 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
Gavin Shandeaf3862012-07-31 16:41:51 -07001661 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001662 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
Gavin Shandeaf3862012-07-31 16:41:51 -07001663 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001664
1665 return should_fail(&fail_page_alloc.attr, 1 << order);
1666}
1667
1668#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1669
1670static int __init fail_page_alloc_debugfs(void)
1671{
Al Virof4ae40a2011-07-24 04:33:43 -04001672 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
Akinobu Mita933e3122006-12-08 02:39:45 -08001673 struct dentry *dir;
Akinobu Mita933e3122006-12-08 02:39:45 -08001674
Akinobu Mitadd48c082011-08-03 16:21:01 -07001675 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
1676 &fail_page_alloc.attr);
1677 if (IS_ERR(dir))
1678 return PTR_ERR(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08001679
Akinobu Mitab2588c42011-07-26 16:09:03 -07001680 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
1681 &fail_page_alloc.ignore_gfp_wait))
1682 goto fail;
1683 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1684 &fail_page_alloc.ignore_gfp_highmem))
1685 goto fail;
1686 if (!debugfs_create_u32("min-order", mode, dir,
1687 &fail_page_alloc.min_order))
1688 goto fail;
Akinobu Mita933e3122006-12-08 02:39:45 -08001689
Akinobu Mitab2588c42011-07-26 16:09:03 -07001690 return 0;
1691fail:
Akinobu Mitadd48c082011-08-03 16:21:01 -07001692 debugfs_remove_recursive(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08001693
Akinobu Mitab2588c42011-07-26 16:09:03 -07001694 return -ENOMEM;
Akinobu Mita933e3122006-12-08 02:39:45 -08001695}
1696
1697late_initcall(fail_page_alloc_debugfs);
1698
1699#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1700
1701#else /* CONFIG_FAIL_PAGE_ALLOC */
1702
Gavin Shandeaf3862012-07-31 16:41:51 -07001703static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08001704{
Gavin Shandeaf3862012-07-31 16:41:51 -07001705 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001706}
1707
1708#endif /* CONFIG_FAIL_PAGE_ALLOC */
1709
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710/*
Mel Gorman88f5acf2011-01-13 15:45:41 -08001711 * Return true if free pages are above 'mark'. This takes into account the order
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 * of the allocation.
1713 */
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001714static bool __zone_watermark_ok(struct zone *z, unsigned int order,
1715 unsigned long mark, int classzone_idx, int alloc_flags,
1716 long free_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717{
1718 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001719 long min = mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 int o;
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07001721 long free_cma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722
Michal Hockodf0a6da2012-01-10 15:08:02 -08001723 free_pages -= (1 << order) - 1;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001724 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001726 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 min -= min / 4;
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07001728#ifdef CONFIG_CMA
1729 /* If allocation can't use CMA areas don't use free CMA pages */
1730 if (!(alloc_flags & ALLOC_CMA))
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07001731 free_cma = zone_page_state(z, NR_FREE_CMA_PAGES);
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07001732#endif
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07001733
Mel Gorman3484b2d2014-08-06 16:07:14 -07001734 if (free_pages - free_cma <= min + z->lowmem_reserve[classzone_idx])
Mel Gorman88f5acf2011-01-13 15:45:41 -08001735 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 for (o = 0; o < order; o++) {
1737 /* At the next order, this order's pages become unavailable */
1738 free_pages -= z->free_area[o].nr_free << o;
1739
1740 /* Require fewer higher order pages to be free */
1741 min >>= 1;
1742
1743 if (free_pages <= min)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001744 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 }
Mel Gorman88f5acf2011-01-13 15:45:41 -08001746 return true;
1747}
1748
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001749bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
Mel Gorman88f5acf2011-01-13 15:45:41 -08001750 int classzone_idx, int alloc_flags)
1751{
1752 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1753 zone_page_state(z, NR_FREE_PAGES));
1754}
1755
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001756bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
1757 unsigned long mark, int classzone_idx, int alloc_flags)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001758{
1759 long free_pages = zone_page_state(z, NR_FREE_PAGES);
1760
1761 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1762 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1763
1764 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1765 free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766}
1767
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001768#ifdef CONFIG_NUMA
1769/*
1770 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1771 * skip over zones that are not allowed by the cpuset, or that have
1772 * been recently (in last second) found to be nearly full. See further
1773 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001774 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001775 *
Zhi Yong Wua1aeb652013-11-12 15:08:29 -08001776 * If the zonelist cache is present in the passed zonelist, then
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001777 * returns a pointer to the allowed node mask (either the current
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08001778 * tasks mems_allowed, or node_states[N_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001779 *
1780 * If the zonelist cache is not available for this zonelist, does
1781 * nothing and returns NULL.
1782 *
1783 * If the fullzones BITMAP in the zonelist cache is stale (more than
1784 * a second since last zap'd) then we zap it out (clear its bits.)
1785 *
1786 * We hold off even calling zlc_setup, until after we've checked the
1787 * first zone in the zonelist, on the theory that most allocations will
1788 * be satisfied from that first zone, so best to examine that zone as
1789 * quickly as we can.
1790 */
1791static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1792{
1793 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1794 nodemask_t *allowednodes; /* zonelist_cache approximation */
1795
1796 zlc = zonelist->zlcache_ptr;
1797 if (!zlc)
1798 return NULL;
1799
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001800 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001801 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1802 zlc->last_full_zap = jiffies;
1803 }
1804
1805 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1806 &cpuset_current_mems_allowed :
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08001807 &node_states[N_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001808 return allowednodes;
1809}
1810
1811/*
1812 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1813 * if it is worth looking at further for free memory:
1814 * 1) Check that the zone isn't thought to be full (doesn't have its
1815 * bit set in the zonelist_cache fullzones BITMAP).
1816 * 2) Check that the zones node (obtained from the zonelist_cache
1817 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1818 * Return true (non-zero) if zone is worth looking at further, or
1819 * else return false (zero) if it is not.
1820 *
1821 * This check -ignores- the distinction between various watermarks,
1822 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1823 * found to be full for any variation of these watermarks, it will
1824 * be considered full for up to one second by all requests, unless
1825 * we are so low on memory on all allowed nodes that we are forced
1826 * into the second scan of the zonelist.
1827 *
1828 * In the second scan we ignore this zonelist cache and exactly
1829 * apply the watermarks to all zones, even it is slower to do so.
1830 * We are low on memory in the second scan, and should leave no stone
1831 * unturned looking for a free page.
1832 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001833static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001834 nodemask_t *allowednodes)
1835{
1836 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1837 int i; /* index of *z in zonelist zones */
1838 int n; /* node that zone *z is on */
1839
1840 zlc = zonelist->zlcache_ptr;
1841 if (!zlc)
1842 return 1;
1843
Mel Gormandd1a2392008-04-28 02:12:17 -07001844 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001845 n = zlc->z_to_n[i];
1846
1847 /* This zone is worth trying if it is allowed but not full */
1848 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1849}
1850
1851/*
1852 * Given 'z' scanning a zonelist, set the corresponding bit in
1853 * zlc->fullzones, so that subsequent attempts to allocate a page
1854 * from that zone don't waste time re-examining it.
1855 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001856static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001857{
1858 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1859 int i; /* index of *z in zonelist zones */
1860
1861 zlc = zonelist->zlcache_ptr;
1862 if (!zlc)
1863 return;
1864
Mel Gormandd1a2392008-04-28 02:12:17 -07001865 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001866
1867 set_bit(i, zlc->fullzones);
1868}
1869
Mel Gorman76d3fbf2011-07-25 17:12:30 -07001870/*
1871 * clear all zones full, called after direct reclaim makes progress so that
1872 * a zone that was recently full is not skipped over for up to a second
1873 */
1874static void zlc_clear_zones_full(struct zonelist *zonelist)
1875{
1876 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1877
1878 zlc = zonelist->zlcache_ptr;
1879 if (!zlc)
1880 return;
1881
1882 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1883}
1884
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001885static bool zone_local(struct zone *local_zone, struct zone *zone)
1886{
Johannes Weinerfff40682013-12-20 14:54:12 +00001887 return local_zone->node == zone->node;
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001888}
1889
David Rientjes957f8222012-10-08 16:33:24 -07001890static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
1891{
Mel Gorman5f7a75a2014-06-04 16:07:15 -07001892 return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <
1893 RECLAIM_DISTANCE;
David Rientjes957f8222012-10-08 16:33:24 -07001894}
1895
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001896#else /* CONFIG_NUMA */
1897
1898static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1899{
1900 return NULL;
1901}
1902
Mel Gormandd1a2392008-04-28 02:12:17 -07001903static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001904 nodemask_t *allowednodes)
1905{
1906 return 1;
1907}
1908
Mel Gormandd1a2392008-04-28 02:12:17 -07001909static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001910{
1911}
Mel Gorman76d3fbf2011-07-25 17:12:30 -07001912
1913static void zlc_clear_zones_full(struct zonelist *zonelist)
1914{
1915}
David Rientjes957f8222012-10-08 16:33:24 -07001916
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001917static bool zone_local(struct zone *local_zone, struct zone *zone)
1918{
1919 return true;
1920}
1921
David Rientjes957f8222012-10-08 16:33:24 -07001922static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
1923{
1924 return true;
1925}
1926
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001927#endif /* CONFIG_NUMA */
1928
Mel Gorman4ffeaf32014-08-06 16:07:22 -07001929static void reset_alloc_batches(struct zone *preferred_zone)
1930{
1931 struct zone *zone = preferred_zone->zone_pgdat->node_zones;
1932
1933 do {
1934 mod_zone_page_state(zone, NR_ALLOC_BATCH,
1935 high_wmark_pages(zone) - low_wmark_pages(zone) -
1936 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
1937 zone_clear_flag(zone, ZONE_FAIR_DEPLETED);
1938 } while (zone++ != preferred_zone);
1939}
1940
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001941/*
Paul Jackson0798e512006-12-06 20:31:38 -08001942 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001943 * a page.
1944 */
1945static struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001946get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
Mel Gorman5117f452009-06-16 15:31:59 -07001947 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
Mel Gormand8846372014-06-04 16:10:33 -07001948 struct zone *preferred_zone, int classzone_idx, int migratetype)
Martin Hicks753ee722005-06-21 17:14:41 -07001949{
Mel Gormandd1a2392008-04-28 02:12:17 -07001950 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001951 struct page *page = NULL;
Mel Gorman5117f452009-06-16 15:31:59 -07001952 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001953 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1954 int zlc_active = 0; /* set if using zonelist_cache */
1955 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gormana6e21b142014-06-04 16:10:12 -07001956 bool consider_zone_dirty = (alloc_flags & ALLOC_WMARK_LOW) &&
1957 (gfp_mask & __GFP_WRITE);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07001958 int nr_fair_skipped = 0;
1959 bool zonelist_rescan;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001960
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001961zonelist_scan:
Mel Gorman4ffeaf32014-08-06 16:07:22 -07001962 zonelist_rescan = false;
1963
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001964 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001965 * Scan zonelist, looking for a zone with enough free.
SeungHun Lee3b11f0a2013-09-11 14:22:23 -07001966 * See also __cpuset_node_allowed_softwall() comment in kernel/cpuset.c.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001967 */
Mel Gorman19770b32008-04-28 02:12:18 -07001968 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1969 high_zoneidx, nodemask) {
Johannes Weinere085dbc2013-09-11 14:20:46 -07001970 unsigned long mark;
1971
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08001972 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001973 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1974 continue;
Mel Gorman664eedd2014-06-04 16:10:08 -07001975 if (cpusets_enabled() &&
1976 (alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001977 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Mel Gormancd38b112011-07-25 17:12:29 -07001978 continue;
Johannes Weinera756cf52012-01-10 15:07:49 -08001979 /*
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001980 * Distribute pages in proportion to the individual
1981 * zone size to ensure fair page aging. The zone a
1982 * page was allocated in should have no effect on the
1983 * time the page has in memory before being reclaimed.
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001984 */
Johannes Weiner3a025762014-04-07 15:37:48 -07001985 if (alloc_flags & ALLOC_FAIR) {
Johannes Weinerfff40682013-12-20 14:54:12 +00001986 if (!zone_local(preferred_zone, zone))
Mel Gormanf7b5d642014-08-06 16:07:20 -07001987 break;
Mel Gorman4ffeaf32014-08-06 16:07:22 -07001988 if (zone_is_fair_depleted(zone)) {
1989 nr_fair_skipped++;
Johannes Weiner3a025762014-04-07 15:37:48 -07001990 continue;
Mel Gorman4ffeaf32014-08-06 16:07:22 -07001991 }
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001992 }
1993 /*
Johannes Weinera756cf52012-01-10 15:07:49 -08001994 * When allocating a page cache page for writing, we
1995 * want to get it from a zone that is within its dirty
1996 * limit, such that no single zone holds more than its
1997 * proportional share of globally allowed dirty pages.
1998 * The dirty limits take into account the zone's
1999 * lowmem reserves and high watermark so that kswapd
2000 * should be able to balance it without having to
2001 * write pages from its LRU list.
2002 *
2003 * This may look like it could increase pressure on
2004 * lower zones by failing allocations in higher zones
2005 * before they are full. But the pages that do spill
2006 * over are limited as the lower zones are protected
2007 * by this very same mechanism. It should not become
2008 * a practical burden to them.
2009 *
2010 * XXX: For now, allow allocations to potentially
2011 * exceed the per-zone dirty limit in the slowpath
2012 * (ALLOC_WMARK_LOW unset) before going into reclaim,
2013 * which is important when on a NUMA setup the allowed
2014 * zones are together not big enough to reach the
2015 * global limit. The proper fix for these situations
2016 * will require awareness of zones in the
2017 * dirty-throttling and the flusher threads.
2018 */
Mel Gormana6e21b142014-06-04 16:10:12 -07002019 if (consider_zone_dirty && !zone_dirty_ok(zone))
Mel Gorman800a1e72014-06-04 16:10:06 -07002020 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002021
Johannes Weinere085dbc2013-09-11 14:20:46 -07002022 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
2023 if (!zone_watermark_ok(zone, order, mark,
2024 classzone_idx, alloc_flags)) {
Mel Gormanfa5e0842009-06-16 15:33:22 -07002025 int ret;
2026
Mel Gorman5dab2912014-06-04 16:10:14 -07002027 /* Checked here to keep the fast path fast */
2028 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
2029 if (alloc_flags & ALLOC_NO_WATERMARKS)
2030 goto try_this_zone;
2031
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002032 if (IS_ENABLED(CONFIG_NUMA) &&
2033 !did_zlc_setup && nr_online_nodes > 1) {
Mel Gormancd38b112011-07-25 17:12:29 -07002034 /*
2035 * we do zlc_setup if there are multiple nodes
2036 * and before considering the first zone allowed
2037 * by the cpuset.
2038 */
2039 allowednodes = zlc_setup(zonelist, alloc_flags);
2040 zlc_active = 1;
2041 did_zlc_setup = 1;
2042 }
2043
David Rientjes957f8222012-10-08 16:33:24 -07002044 if (zone_reclaim_mode == 0 ||
2045 !zone_allows_reclaim(preferred_zone, zone))
Mel Gormanfa5e0842009-06-16 15:33:22 -07002046 goto this_zone_full;
2047
Mel Gormancd38b112011-07-25 17:12:29 -07002048 /*
2049 * As we may have just activated ZLC, check if the first
2050 * eligible zone has failed zone_reclaim recently.
2051 */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002052 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
Mel Gormancd38b112011-07-25 17:12:29 -07002053 !zlc_zone_worth_trying(zonelist, z, allowednodes))
2054 continue;
2055
Mel Gormanfa5e0842009-06-16 15:33:22 -07002056 ret = zone_reclaim(zone, gfp_mask, order);
2057 switch (ret) {
2058 case ZONE_RECLAIM_NOSCAN:
2059 /* did not scan */
Mel Gormancd38b112011-07-25 17:12:29 -07002060 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002061 case ZONE_RECLAIM_FULL:
2062 /* scanned but unreclaimable */
Mel Gormancd38b112011-07-25 17:12:29 -07002063 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002064 default:
2065 /* did we reclaim enough */
Mel Gormanfed27192013-04-29 15:07:57 -07002066 if (zone_watermark_ok(zone, order, mark,
Mel Gormanfa5e0842009-06-16 15:33:22 -07002067 classzone_idx, alloc_flags))
Mel Gormanfed27192013-04-29 15:07:57 -07002068 goto try_this_zone;
2069
2070 /*
2071 * Failed to reclaim enough to meet watermark.
2072 * Only mark the zone full if checking the min
2073 * watermark or if we failed to reclaim just
2074 * 1<<order pages or else the page allocator
2075 * fastpath will prematurely mark zones full
2076 * when the watermark is between the low and
2077 * min watermarks.
2078 */
2079 if (((alloc_flags & ALLOC_WMARK_MASK) == ALLOC_WMARK_MIN) ||
2080 ret == ZONE_RECLAIM_SOME)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002081 goto this_zone_full;
Mel Gormanfed27192013-04-29 15:07:57 -07002082
2083 continue;
Paul Jackson0798e512006-12-06 20:31:38 -08002084 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002085 }
2086
Mel Gormanfa5e0842009-06-16 15:33:22 -07002087try_this_zone:
Mel Gorman3dd28262009-06-16 15:32:00 -07002088 page = buffered_rmqueue(preferred_zone, zone, order,
2089 gfp_mask, migratetype);
Paul Jackson0798e512006-12-06 20:31:38 -08002090 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002091 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002092this_zone_full:
Mel Gorman65bb3712014-06-04 16:10:05 -07002093 if (IS_ENABLED(CONFIG_NUMA) && zlc_active)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002094 zlc_mark_zone_full(zonelist, z);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002095 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002096
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002097 if (page) {
Alex Shib1211862012-08-21 16:16:08 -07002098 /*
2099 * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was
2100 * necessary to allocate the page. The expectation is
2101 * that the caller is taking steps that will free more
2102 * memory. The caller should avoid the page being used
2103 * for !PFMEMALLOC purposes.
2104 */
2105 page->pfmemalloc = !!(alloc_flags & ALLOC_NO_WATERMARKS);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002106 return page;
2107 }
Alex Shib1211862012-08-21 16:16:08 -07002108
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002109 /*
2110 * The first pass makes sure allocations are spread fairly within the
2111 * local node. However, the local node might have free pages left
2112 * after the fairness batches are exhausted, and remote zones haven't
2113 * even been considered yet. Try once more without fairness, and
2114 * include remote zones now, before entering the slowpath and waking
2115 * kswapd: prefer spilling to a remote zone over swapping locally.
2116 */
2117 if (alloc_flags & ALLOC_FAIR) {
2118 alloc_flags &= ~ALLOC_FAIR;
2119 if (nr_fair_skipped) {
2120 zonelist_rescan = true;
2121 reset_alloc_batches(preferred_zone);
2122 }
2123 if (nr_online_nodes > 1)
2124 zonelist_rescan = true;
2125 }
2126
2127 if (unlikely(IS_ENABLED(CONFIG_NUMA) && zlc_active)) {
2128 /* Disable zlc cache for second zonelist scan */
2129 zlc_active = 0;
2130 zonelist_rescan = true;
2131 }
2132
2133 if (zonelist_rescan)
2134 goto zonelist_scan;
2135
2136 return NULL;
Martin Hicks753ee722005-06-21 17:14:41 -07002137}
2138
David Rientjes29423e772011-03-22 16:30:47 -07002139/*
2140 * Large machines with many possible nodes should not always dump per-node
2141 * meminfo in irq context.
2142 */
2143static inline bool should_suppress_show_mem(void)
2144{
2145 bool ret = false;
2146
2147#if NODES_SHIFT > 8
2148 ret = in_interrupt();
2149#endif
2150 return ret;
2151}
2152
Dave Hansena238ab52011-05-24 17:12:16 -07002153static DEFINE_RATELIMIT_STATE(nopage_rs,
2154 DEFAULT_RATELIMIT_INTERVAL,
2155 DEFAULT_RATELIMIT_BURST);
2156
2157void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
2158{
Dave Hansena238ab52011-05-24 17:12:16 -07002159 unsigned int filter = SHOW_MEM_FILTER_NODES;
2160
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -08002161 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2162 debug_guardpage_minorder() > 0)
Dave Hansena238ab52011-05-24 17:12:16 -07002163 return;
2164
2165 /*
2166 * This documents exceptions given to allocations in certain
2167 * contexts that are allowed to allocate outside current's set
2168 * of allowed nodes.
2169 */
2170 if (!(gfp_mask & __GFP_NOMEMALLOC))
2171 if (test_thread_flag(TIF_MEMDIE) ||
2172 (current->flags & (PF_MEMALLOC | PF_EXITING)))
2173 filter &= ~SHOW_MEM_FILTER_NODES;
2174 if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
2175 filter &= ~SHOW_MEM_FILTER_NODES;
2176
2177 if (fmt) {
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002178 struct va_format vaf;
2179 va_list args;
2180
Dave Hansena238ab52011-05-24 17:12:16 -07002181 va_start(args, fmt);
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002182
2183 vaf.fmt = fmt;
2184 vaf.va = &args;
2185
2186 pr_warn("%pV", &vaf);
2187
Dave Hansena238ab52011-05-24 17:12:16 -07002188 va_end(args);
2189 }
2190
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002191 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2192 current->comm, order, gfp_mask);
Dave Hansena238ab52011-05-24 17:12:16 -07002193
2194 dump_stack();
2195 if (!should_suppress_show_mem())
2196 show_mem(filter);
2197}
2198
Mel Gorman11e33f62009-06-16 15:31:57 -07002199static inline int
2200should_alloc_retry(gfp_t gfp_mask, unsigned int order,
Mel Gormanf90ac392012-01-10 15:07:15 -08002201 unsigned long did_some_progress,
Mel Gorman11e33f62009-06-16 15:31:57 -07002202 unsigned long pages_reclaimed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203{
Mel Gorman11e33f62009-06-16 15:31:57 -07002204 /* Do not loop if specifically requested */
2205 if (gfp_mask & __GFP_NORETRY)
2206 return 0;
2207
Mel Gormanf90ac392012-01-10 15:07:15 -08002208 /* Always retry if specifically requested */
2209 if (gfp_mask & __GFP_NOFAIL)
2210 return 1;
2211
2212 /*
2213 * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim
2214 * making forward progress without invoking OOM. Suspend also disables
2215 * storage devices so kswapd will not help. Bail if we are suspending.
2216 */
2217 if (!did_some_progress && pm_suspended_storage())
2218 return 0;
2219
Mel Gorman11e33f62009-06-16 15:31:57 -07002220 /*
2221 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
2222 * means __GFP_NOFAIL, but that may not be true in other
2223 * implementations.
2224 */
2225 if (order <= PAGE_ALLOC_COSTLY_ORDER)
2226 return 1;
2227
2228 /*
2229 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
2230 * specified, then we retry until we no longer reclaim any pages
2231 * (above), or we've reclaimed an order of pages at least as
2232 * large as the allocation's order. In both cases, if the
2233 * allocation still fails, we stop retrying.
2234 */
2235 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
2236 return 1;
2237
Mel Gorman11e33f62009-06-16 15:31:57 -07002238 return 0;
2239}
2240
2241static inline struct page *
2242__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
2243 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002244 nodemask_t *nodemask, struct zone *preferred_zone,
Mel Gormand8846372014-06-04 16:10:33 -07002245 int classzone_idx, int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002246{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248
David Rientjese972a072014-08-06 16:07:52 -07002249 /* Acquire the per-zone oom lock for each zone */
2250 if (!oom_zonelist_trylock(zonelist, gfp_mask)) {
Mel Gorman11e33f62009-06-16 15:31:57 -07002251 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 return NULL;
2253 }
Jens Axboe6b1de912005-11-17 21:35:02 +01002254
Mel Gorman11e33f62009-06-16 15:31:57 -07002255 /*
2256 * Go through the zonelist yet one more time, keep very high watermark
2257 * here, this is only to catch a parallel oom killing, we must fail if
2258 * we're still under heavy pressure.
2259 */
2260 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
2261 order, zonelist, high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07002262 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
Mel Gormand8846372014-06-04 16:10:33 -07002263 preferred_zone, classzone_idx, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002264 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07002265 goto out;
2266
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002267 if (!(gfp_mask & __GFP_NOFAIL)) {
2268 /* The OOM killer will not help higher order allocs */
2269 if (order > PAGE_ALLOC_COSTLY_ORDER)
2270 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07002271 /* The OOM killer does not needlessly kill tasks for lowmem */
2272 if (high_zoneidx < ZONE_NORMAL)
2273 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002274 /*
2275 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
2276 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
2277 * The caller should handle page allocation failure by itself if
2278 * it specifies __GFP_THISNODE.
2279 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
2280 */
2281 if (gfp_mask & __GFP_THISNODE)
2282 goto out;
2283 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002284 /* Exhausted what can be done so it's blamo time */
David Rientjes08ab9b12012-03-21 16:34:04 -07002285 out_of_memory(zonelist, gfp_mask, order, nodemask, false);
Mel Gorman11e33f62009-06-16 15:31:57 -07002286
2287out:
David Rientjese972a072014-08-06 16:07:52 -07002288 oom_zonelist_unlock(zonelist, gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002289 return page;
2290}
2291
Mel Gorman56de7262010-05-24 14:32:30 -07002292#ifdef CONFIG_COMPACTION
2293/* Try memory compaction for high-order allocations before reclaim */
2294static struct page *
2295__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2296 struct zonelist *zonelist, enum zone_type high_zoneidx,
2297 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gormand8846372014-06-04 16:10:33 -07002298 int classzone_idx, int migratetype, enum migrate_mode mode,
Mel Gormanc67fe372012-08-21 16:16:17 -07002299 bool *contended_compaction, bool *deferred_compaction,
Mel Gorman66199712012-01-12 17:19:41 -08002300 unsigned long *did_some_progress)
Mel Gorman56de7262010-05-24 14:32:30 -07002301{
Mel Gorman66199712012-01-12 17:19:41 -08002302 if (!order)
Mel Gorman56de7262010-05-24 14:32:30 -07002303 return NULL;
2304
Rik van Rielaff62242012-03-21 16:33:52 -07002305 if (compaction_deferred(preferred_zone, order)) {
Mel Gorman66199712012-01-12 17:19:41 -08002306 *deferred_compaction = true;
2307 return NULL;
2308 }
2309
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002310 current->flags |= PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07002311 *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
David Rientjese0b9dae2014-06-04 16:08:28 -07002312 nodemask, mode,
Mel Gorman8fb74b92013-01-11 14:32:16 -08002313 contended_compaction);
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002314 current->flags &= ~PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07002315
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07002316 if (*did_some_progress != COMPACT_SKIPPED) {
Mel Gorman8fb74b92013-01-11 14:32:16 -08002317 struct page *page;
2318
Mel Gorman56de7262010-05-24 14:32:30 -07002319 /* Page migration frees to the PCP lists but we want merging */
2320 drain_pages(get_cpu());
2321 put_cpu();
2322
2323 page = get_page_from_freelist(gfp_mask, nodemask,
2324 order, zonelist, high_zoneidx,
Mel Gormancfd19c52012-07-31 16:44:10 -07002325 alloc_flags & ~ALLOC_NO_WATERMARKS,
Mel Gormand8846372014-06-04 16:10:33 -07002326 preferred_zone, classzone_idx, migratetype);
Mel Gorman56de7262010-05-24 14:32:30 -07002327 if (page) {
Mel Gorman62997022012-10-08 16:32:47 -07002328 preferred_zone->compact_blockskip_flush = false;
Vlastimil Babkade6c60a2014-01-21 15:51:07 -08002329 compaction_defer_reset(preferred_zone, order, true);
Mel Gorman56de7262010-05-24 14:32:30 -07002330 count_vm_event(COMPACTSUCCESS);
2331 return page;
2332 }
2333
2334 /*
2335 * It's bad if compaction run occurs and fails.
2336 * The most likely reason is that pages exist,
2337 * but not enough to satisfy watermarks.
2338 */
2339 count_vm_event(COMPACTFAIL);
Mel Gorman66199712012-01-12 17:19:41 -08002340
2341 /*
2342 * As async compaction considers a subset of pageblocks, only
2343 * defer if the failure was a sync compaction failure.
2344 */
David Rientjese0b9dae2014-06-04 16:08:28 -07002345 if (mode != MIGRATE_ASYNC)
Rik van Rielaff62242012-03-21 16:33:52 -07002346 defer_compaction(preferred_zone, order);
Mel Gorman56de7262010-05-24 14:32:30 -07002347
2348 cond_resched();
2349 }
2350
2351 return NULL;
2352}
2353#else
2354static inline struct page *
2355__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2356 struct zonelist *zonelist, enum zone_type high_zoneidx,
2357 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gormand8846372014-06-04 16:10:33 -07002358 int classzone_idx, int migratetype,
2359 enum migrate_mode mode, bool *contended_compaction,
David Rientjese0b9dae2014-06-04 16:08:28 -07002360 bool *deferred_compaction, unsigned long *did_some_progress)
Mel Gorman56de7262010-05-24 14:32:30 -07002361{
2362 return NULL;
2363}
2364#endif /* CONFIG_COMPACTION */
2365
Marek Szyprowskibba90712012-01-25 12:09:52 +01002366/* Perform direct synchronous page reclaim */
2367static int
2368__perform_reclaim(gfp_t gfp_mask, unsigned int order, struct zonelist *zonelist,
2369 nodemask_t *nodemask)
Mel Gorman11e33f62009-06-16 15:31:57 -07002370{
Mel Gorman11e33f62009-06-16 15:31:57 -07002371 struct reclaim_state reclaim_state;
Marek Szyprowskibba90712012-01-25 12:09:52 +01002372 int progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07002373
2374 cond_resched();
2375
2376 /* We now go into synchronous reclaim */
2377 cpuset_memory_pressure_bump();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002378 current->flags |= PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002379 lockdep_set_current_reclaim_state(gfp_mask);
2380 reclaim_state.reclaimed_slab = 0;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002381 current->reclaim_state = &reclaim_state;
Mel Gorman11e33f62009-06-16 15:31:57 -07002382
Marek Szyprowskibba90712012-01-25 12:09:52 +01002383 progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002384
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002385 current->reclaim_state = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002386 lockdep_clear_current_reclaim_state();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002387 current->flags &= ~PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002388
2389 cond_resched();
2390
Marek Szyprowskibba90712012-01-25 12:09:52 +01002391 return progress;
2392}
2393
2394/* The really slow allocator path where we enter direct reclaim */
2395static inline struct page *
2396__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
2397 struct zonelist *zonelist, enum zone_type high_zoneidx,
2398 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gormand8846372014-06-04 16:10:33 -07002399 int classzone_idx, int migratetype, unsigned long *did_some_progress)
Marek Szyprowskibba90712012-01-25 12:09:52 +01002400{
2401 struct page *page = NULL;
2402 bool drained = false;
2403
2404 *did_some_progress = __perform_reclaim(gfp_mask, order, zonelist,
2405 nodemask);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002406 if (unlikely(!(*did_some_progress)))
2407 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002408
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002409 /* After successful reclaim, reconsider all zones for allocation */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002410 if (IS_ENABLED(CONFIG_NUMA))
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002411 zlc_clear_zones_full(zonelist);
2412
Mel Gorman9ee493c2010-09-09 16:38:18 -07002413retry:
2414 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002415 zonelist, high_zoneidx,
Mel Gormancfd19c52012-07-31 16:44:10 -07002416 alloc_flags & ~ALLOC_NO_WATERMARKS,
Mel Gormand8846372014-06-04 16:10:33 -07002417 preferred_zone, classzone_idx,
2418 migratetype);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002419
2420 /*
2421 * If an allocation failed after direct reclaim, it could be because
2422 * pages are pinned on the per-cpu lists. Drain them and try again
2423 */
2424 if (!page && !drained) {
2425 drain_all_pages();
2426 drained = true;
2427 goto retry;
2428 }
2429
Mel Gorman11e33f62009-06-16 15:31:57 -07002430 return page;
2431}
2432
Mel Gorman11e33f62009-06-16 15:31:57 -07002433/*
2434 * This is called in the allocator slow-path if the allocation request is of
2435 * sufficient urgency to ignore watermarks and take other desperate measures
2436 */
2437static inline struct page *
2438__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
2439 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002440 nodemask_t *nodemask, struct zone *preferred_zone,
Mel Gormand8846372014-06-04 16:10:33 -07002441 int classzone_idx, int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002442{
2443 struct page *page;
2444
2445 do {
2446 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002447 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
Mel Gormand8846372014-06-04 16:10:33 -07002448 preferred_zone, classzone_idx, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002449
2450 if (!page && gfp_mask & __GFP_NOFAIL)
Mel Gorman0e093d992010-10-26 14:21:45 -07002451 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07002452 } while (!page && (gfp_mask & __GFP_NOFAIL));
2453
2454 return page;
2455}
2456
Johannes Weiner3a025762014-04-07 15:37:48 -07002457static void wake_all_kswapds(unsigned int order,
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002458 struct zonelist *zonelist,
2459 enum zone_type high_zoneidx,
2460 struct zone *preferred_zone)
Mel Gorman11e33f62009-06-16 15:31:57 -07002461{
2462 struct zoneref *z;
2463 struct zone *zone;
2464
Johannes Weiner3a025762014-04-07 15:37:48 -07002465 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
2466 wakeup_kswapd(zone, order, zone_idx(preferred_zone));
Mel Gorman11e33f62009-06-16 15:31:57 -07002467}
2468
Peter Zijlstra341ce062009-06-16 15:32:02 -07002469static inline int
2470gfp_to_alloc_flags(gfp_t gfp_mask)
2471{
Peter Zijlstra341ce062009-06-16 15:32:02 -07002472 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
David Rientjesb104a352014-07-30 16:08:24 -07002473 const bool atomic = !(gfp_mask & (__GFP_WAIT | __GFP_NO_KSWAPD));
Peter Zijlstra341ce062009-06-16 15:32:02 -07002474
Mel Gormana56f57f2009-06-16 15:32:02 -07002475 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07002476 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07002477
Peter Zijlstra341ce062009-06-16 15:32:02 -07002478 /*
2479 * The caller may dip into page reserves a bit more if the caller
2480 * cannot run direct reclaim, or if the caller has realtime scheduling
2481 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
David Rientjesb104a352014-07-30 16:08:24 -07002482 * set both ALLOC_HARDER (atomic == true) and ALLOC_HIGH (__GFP_HIGH).
Peter Zijlstra341ce062009-06-16 15:32:02 -07002483 */
Namhyung Kime6223a32010-10-26 14:21:59 -07002484 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07002485
David Rientjesb104a352014-07-30 16:08:24 -07002486 if (atomic) {
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002487 /*
David Rientjesb104a352014-07-30 16:08:24 -07002488 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
2489 * if it can't schedule.
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002490 */
David Rientjesb104a352014-07-30 16:08:24 -07002491 if (!(gfp_mask & __GFP_NOMEMALLOC))
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002492 alloc_flags |= ALLOC_HARDER;
Peter Zijlstra341ce062009-06-16 15:32:02 -07002493 /*
David Rientjesb104a352014-07-30 16:08:24 -07002494 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
2495 * comment for __cpuset_node_allowed_softwall().
Peter Zijlstra341ce062009-06-16 15:32:02 -07002496 */
2497 alloc_flags &= ~ALLOC_CPUSET;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002498 } else if (unlikely(rt_task(current)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07002499 alloc_flags |= ALLOC_HARDER;
2500
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002501 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2502 if (gfp_mask & __GFP_MEMALLOC)
2503 alloc_flags |= ALLOC_NO_WATERMARKS;
Mel Gorman907aed42012-07-31 16:44:07 -07002504 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
2505 alloc_flags |= ALLOC_NO_WATERMARKS;
2506 else if (!in_interrupt() &&
2507 ((current->flags & PF_MEMALLOC) ||
2508 unlikely(test_thread_flag(TIF_MEMDIE))))
Peter Zijlstra341ce062009-06-16 15:32:02 -07002509 alloc_flags |= ALLOC_NO_WATERMARKS;
2510 }
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002511#ifdef CONFIG_CMA
2512 if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2513 alloc_flags |= ALLOC_CMA;
2514#endif
Peter Zijlstra341ce062009-06-16 15:32:02 -07002515 return alloc_flags;
2516}
2517
Mel Gorman072bb0a2012-07-31 16:43:58 -07002518bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
2519{
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002520 return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
Mel Gorman072bb0a2012-07-31 16:43:58 -07002521}
2522
Mel Gorman11e33f62009-06-16 15:31:57 -07002523static inline struct page *
2524__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
2525 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002526 nodemask_t *nodemask, struct zone *preferred_zone,
Mel Gormand8846372014-06-04 16:10:33 -07002527 int classzone_idx, int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002528{
2529 const gfp_t wait = gfp_mask & __GFP_WAIT;
2530 struct page *page = NULL;
2531 int alloc_flags;
2532 unsigned long pages_reclaimed = 0;
2533 unsigned long did_some_progress;
David Rientjese0b9dae2014-06-04 16:08:28 -07002534 enum migrate_mode migration_mode = MIGRATE_ASYNC;
Mel Gorman66199712012-01-12 17:19:41 -08002535 bool deferred_compaction = false;
Mel Gormanc67fe372012-08-21 16:16:17 -07002536 bool contended_compaction = false;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002537
Christoph Lameter952f3b52006-12-06 20:33:26 -08002538 /*
Mel Gorman72807a72009-06-16 15:32:18 -07002539 * In the slowpath, we sanity check order to avoid ever trying to
2540 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2541 * be using allocators in order of preference for an area that is
2542 * too large.
2543 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07002544 if (order >= MAX_ORDER) {
2545 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07002546 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07002547 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002548
Christoph Lameter952f3b52006-12-06 20:33:26 -08002549 /*
2550 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2551 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2552 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2553 * using a larger set of nodes after it has established that the
2554 * allowed per node queues are empty and that nodes are
2555 * over allocated.
2556 */
Johannes Weiner3a025762014-04-07 15:37:48 -07002557 if (IS_ENABLED(CONFIG_NUMA) &&
2558 (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
Christoph Lameter952f3b52006-12-06 20:33:26 -08002559 goto nopage;
2560
Mel Gormancc4a6852009-11-11 14:26:14 -08002561restart:
Johannes Weiner3a025762014-04-07 15:37:48 -07002562 if (!(gfp_mask & __GFP_NO_KSWAPD))
2563 wake_all_kswapds(order, zonelist, high_zoneidx, preferred_zone);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002564
Paul Jackson9bf22292005-09-06 15:18:12 -07002565 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002566 * OK, we're below the kswapd watermark and have kicked background
2567 * reclaim. Now things get more complex, so set up alloc_flags according
2568 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07002569 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002570 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571
David Rientjesf33261d2011-01-25 15:07:20 -08002572 /*
2573 * Find the true preferred zone if the allocation is unconstrained by
2574 * cpusets.
2575 */
Mel Gormand8846372014-06-04 16:10:33 -07002576 if (!(alloc_flags & ALLOC_CPUSET) && !nodemask) {
2577 struct zoneref *preferred_zoneref;
2578 preferred_zoneref = first_zones_zonelist(zonelist, high_zoneidx,
2579 NULL, &preferred_zone);
2580 classzone_idx = zonelist_zone_idx(preferred_zoneref);
2581 }
David Rientjesf33261d2011-01-25 15:07:20 -08002582
Andrew Barrycfa54a02011-05-24 17:12:52 -07002583rebalance:
Peter Zijlstra341ce062009-06-16 15:32:02 -07002584 /* This is the last chance, in general, before the goto nopage. */
Mel Gorman19770b32008-04-28 02:12:18 -07002585 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
Peter Zijlstra341ce062009-06-16 15:32:02 -07002586 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
Mel Gormand8846372014-06-04 16:10:33 -07002587 preferred_zone, classzone_idx, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002588 if (page)
2589 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590
Mel Gorman11e33f62009-06-16 15:31:57 -07002591 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002592 if (alloc_flags & ALLOC_NO_WATERMARKS) {
Mel Gorman183f6372012-07-31 16:44:12 -07002593 /*
2594 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
2595 * the allocation is high priority and these type of
2596 * allocations are system rather than user orientated
2597 */
2598 zonelist = node_zonelist(numa_node_id(), gfp_mask);
2599
Peter Zijlstra341ce062009-06-16 15:32:02 -07002600 page = __alloc_pages_high_priority(gfp_mask, order,
2601 zonelist, high_zoneidx, nodemask,
Mel Gormand8846372014-06-04 16:10:33 -07002602 preferred_zone, classzone_idx, migratetype);
Mel Gormancfd19c52012-07-31 16:44:10 -07002603 if (page) {
Peter Zijlstra341ce062009-06-16 15:32:02 -07002604 goto got_pg;
Mel Gormancfd19c52012-07-31 16:44:10 -07002605 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606 }
2607
2608 /* Atomic allocations - we can't balance anything */
David Rientjesaed0a0e2014-01-21 15:51:12 -08002609 if (!wait) {
2610 /*
2611 * All existing users of the deprecated __GFP_NOFAIL are
2612 * blockable, so warn of any new users that actually allow this
2613 * type of allocation to fail.
2614 */
2615 WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616 goto nopage;
David Rientjesaed0a0e2014-01-21 15:51:12 -08002617 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618
Peter Zijlstra341ce062009-06-16 15:32:02 -07002619 /* Avoid recursion of direct reclaim */
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002620 if (current->flags & PF_MEMALLOC)
Peter Zijlstra341ce062009-06-16 15:32:02 -07002621 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622
David Rientjes6583bb62009-07-29 15:02:06 -07002623 /* Avoid allocations with no watermarks from looping endlessly */
2624 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2625 goto nopage;
2626
Mel Gorman77f1fe62011-01-13 15:45:57 -08002627 /*
2628 * Try direct compaction. The first pass is asynchronous. Subsequent
2629 * attempts after direct reclaim are synchronous
2630 */
David Rientjese0b9dae2014-06-04 16:08:28 -07002631 page = __alloc_pages_direct_compact(gfp_mask, order, zonelist,
2632 high_zoneidx, nodemask, alloc_flags,
Mel Gormand8846372014-06-04 16:10:33 -07002633 preferred_zone,
2634 classzone_idx, migratetype,
David Rientjese0b9dae2014-06-04 16:08:28 -07002635 migration_mode, &contended_compaction,
Mel Gorman66199712012-01-12 17:19:41 -08002636 &deferred_compaction,
2637 &did_some_progress);
Mel Gorman56de7262010-05-24 14:32:30 -07002638 if (page)
2639 goto got_pg;
David Rientjes75f30862014-06-04 16:08:30 -07002640
2641 /*
Linus Torvalds31f8d422012-12-10 10:47:45 -08002642 * If compaction is deferred for high-order allocations, it is because
2643 * sync compaction recently failed. In this is the case and the caller
2644 * requested a movable allocation that does not heavily disrupt the
2645 * system then fail the allocation instead of entering direct reclaim.
2646 */
2647 if ((deferred_compaction || contended_compaction) &&
Linus Torvaldscaf49192012-12-10 10:51:16 -08002648 (gfp_mask & __GFP_NO_KSWAPD))
Linus Torvalds31f8d422012-12-10 10:47:45 -08002649 goto nopage;
Mel Gorman66199712012-01-12 17:19:41 -08002650
David Rientjes8fe78042014-08-06 16:07:54 -07002651 /*
2652 * It can become very expensive to allocate transparent hugepages at
2653 * fault, so use asynchronous memory compaction for THP unless it is
2654 * khugepaged trying to collapse.
2655 */
2656 if ((gfp_mask & GFP_TRANSHUGE) != GFP_TRANSHUGE ||
2657 (current->flags & PF_KTHREAD))
2658 migration_mode = MIGRATE_SYNC_LIGHT;
2659
Mel Gorman11e33f62009-06-16 15:31:57 -07002660 /* Try direct reclaim and then allocating */
2661 page = __alloc_pages_direct_reclaim(gfp_mask, order,
2662 zonelist, high_zoneidx,
2663 nodemask,
Mel Gorman5117f452009-06-16 15:31:59 -07002664 alloc_flags, preferred_zone,
Mel Gormand8846372014-06-04 16:10:33 -07002665 classzone_idx, migratetype,
2666 &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07002667 if (page)
2668 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670 /*
Mel Gorman11e33f62009-06-16 15:31:57 -07002671 * If we failed to make any progress reclaiming, then we are
2672 * running out of options and have to consider going OOM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673 */
Mel Gorman11e33f62009-06-16 15:31:57 -07002674 if (!did_some_progress) {
Qiang Huangb9921ec2013-11-12 15:07:22 -08002675 if (oom_gfp_allowed(gfp_mask)) {
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -07002676 if (oom_killer_disabled)
2677 goto nopage;
David Rientjes29fd66d2012-03-28 14:42:41 -07002678 /* Coredumps can quickly deplete all memory reserves */
2679 if ((current->flags & PF_DUMPCORE) &&
2680 !(gfp_mask & __GFP_NOFAIL))
2681 goto nopage;
Mel Gorman11e33f62009-06-16 15:31:57 -07002682 page = __alloc_pages_may_oom(gfp_mask, order,
2683 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002684 nodemask, preferred_zone,
Mel Gormand8846372014-06-04 16:10:33 -07002685 classzone_idx, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002686 if (page)
2687 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688
David Rientjes03668b32010-08-09 17:18:54 -07002689 if (!(gfp_mask & __GFP_NOFAIL)) {
2690 /*
2691 * The oom killer is not called for high-order
2692 * allocations that may fail, so if no progress
2693 * is being made, there are no other options and
2694 * retrying is unlikely to help.
2695 */
2696 if (order > PAGE_ALLOC_COSTLY_ORDER)
2697 goto nopage;
2698 /*
2699 * The oom killer is not called for lowmem
2700 * allocations to prevent needlessly killing
2701 * innocent tasks.
2702 */
2703 if (high_zoneidx < ZONE_NORMAL)
2704 goto nopage;
2705 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707 goto restart;
2708 }
2709 }
2710
Mel Gorman11e33f62009-06-16 15:31:57 -07002711 /* Check if we should retry the allocation */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07002712 pages_reclaimed += did_some_progress;
Mel Gormanf90ac392012-01-10 15:07:15 -08002713 if (should_alloc_retry(gfp_mask, order, did_some_progress,
2714 pages_reclaimed)) {
Mel Gorman11e33f62009-06-16 15:31:57 -07002715 /* Wait for some write requests to complete then retry */
Mel Gorman0e093d992010-10-26 14:21:45 -07002716 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717 goto rebalance;
Mel Gorman3e7d3442011-01-13 15:45:56 -08002718 } else {
2719 /*
2720 * High-order allocations do not necessarily loop after
2721 * direct reclaim and reclaim/compaction depends on compaction
2722 * being called after reclaim so call directly if necessary
2723 */
David Rientjese0b9dae2014-06-04 16:08:28 -07002724 page = __alloc_pages_direct_compact(gfp_mask, order, zonelist,
2725 high_zoneidx, nodemask, alloc_flags,
Mel Gormand8846372014-06-04 16:10:33 -07002726 preferred_zone,
2727 classzone_idx, migratetype,
David Rientjese0b9dae2014-06-04 16:08:28 -07002728 migration_mode, &contended_compaction,
Mel Gorman66199712012-01-12 17:19:41 -08002729 &deferred_compaction,
2730 &did_some_progress);
Mel Gorman3e7d3442011-01-13 15:45:56 -08002731 if (page)
2732 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733 }
2734
2735nopage:
Dave Hansena238ab52011-05-24 17:12:16 -07002736 warn_alloc_failed(gfp_mask, order, NULL);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002737 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002738got_pg:
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002739 if (kmemcheck_enabled)
2740 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002741
Mel Gorman072bb0a2012-07-31 16:43:58 -07002742 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743}
Mel Gorman11e33f62009-06-16 15:31:57 -07002744
2745/*
2746 * This is the 'heart' of the zoned buddy allocator.
2747 */
2748struct page *
2749__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2750 struct zonelist *zonelist, nodemask_t *nodemask)
2751{
2752 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Mel Gorman5117f452009-06-16 15:31:59 -07002753 struct zone *preferred_zone;
Mel Gormand8846372014-06-04 16:10:33 -07002754 struct zoneref *preferred_zoneref;
Mel Gormancc9a6c82012-03-21 16:34:11 -07002755 struct page *page = NULL;
Mel Gorman3dd28262009-06-16 15:32:00 -07002756 int migratetype = allocflags_to_migratetype(gfp_mask);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002757 unsigned int cpuset_mems_cookie;
Johannes Weiner3a025762014-04-07 15:37:48 -07002758 int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET|ALLOC_FAIR;
Mel Gormand8846372014-06-04 16:10:33 -07002759 int classzone_idx;
Mel Gorman11e33f62009-06-16 15:31:57 -07002760
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10002761 gfp_mask &= gfp_allowed_mask;
2762
Mel Gorman11e33f62009-06-16 15:31:57 -07002763 lockdep_trace_alloc(gfp_mask);
2764
2765 might_sleep_if(gfp_mask & __GFP_WAIT);
2766
2767 if (should_fail_alloc_page(gfp_mask, order))
2768 return NULL;
2769
2770 /*
2771 * Check the zones suitable for the gfp_mask contain at least one
2772 * valid zone. It's possible to have an empty zonelist as a result
2773 * of GFP_THISNODE and a memoryless node
2774 */
2775 if (unlikely(!zonelist->_zonerefs->zone))
2776 return NULL;
2777
Mel Gormancc9a6c82012-03-21 16:34:11 -07002778retry_cpuset:
Mel Gormand26914d2014-04-03 14:47:24 -07002779 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07002780
Mel Gorman5117f452009-06-16 15:31:59 -07002781 /* The preferred zone is used for statistics later */
Mel Gormand8846372014-06-04 16:10:33 -07002782 preferred_zoneref = first_zones_zonelist(zonelist, high_zoneidx,
David Rientjesf33261d2011-01-25 15:07:20 -08002783 nodemask ? : &cpuset_current_mems_allowed,
2784 &preferred_zone);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002785 if (!preferred_zone)
2786 goto out;
Mel Gormand8846372014-06-04 16:10:33 -07002787 classzone_idx = zonelist_zone_idx(preferred_zoneref);
Mel Gorman5117f452009-06-16 15:31:59 -07002788
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002789#ifdef CONFIG_CMA
2790 if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2791 alloc_flags |= ALLOC_CMA;
2792#endif
Mel Gorman5117f452009-06-16 15:31:59 -07002793 /* First allocation attempt */
Mel Gorman11e33f62009-06-16 15:31:57 -07002794 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002795 zonelist, high_zoneidx, alloc_flags,
Mel Gormand8846372014-06-04 16:10:33 -07002796 preferred_zone, classzone_idx, migratetype);
Ming Lei21caf2f2013-02-22 16:34:08 -08002797 if (unlikely(!page)) {
2798 /*
2799 * Runtime PM, block IO and its error handling path
2800 * can deadlock because I/O on the device might not
2801 * complete.
2802 */
2803 gfp_mask = memalloc_noio_flags(gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002804 page = __alloc_pages_slowpath(gfp_mask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002805 zonelist, high_zoneidx, nodemask,
Mel Gormand8846372014-06-04 16:10:33 -07002806 preferred_zone, classzone_idx, migratetype);
Ming Lei21caf2f2013-02-22 16:34:08 -08002807 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002808
Mel Gorman4b4f2782009-09-21 17:02:41 -07002809 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002810
2811out:
2812 /*
2813 * When updating a task's mems_allowed, it is possible to race with
2814 * parallel threads in such a way that an allocation can fail while
2815 * the mask is being updated. If a page allocation is about to fail,
2816 * check if the cpuset changed during allocation and if so, retry.
2817 */
Mel Gormand26914d2014-04-03 14:47:24 -07002818 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
Mel Gormancc9a6c82012-03-21 16:34:11 -07002819 goto retry_cpuset;
2820
Mel Gorman11e33f62009-06-16 15:31:57 -07002821 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822}
Mel Gormand2391712009-06-16 15:31:52 -07002823EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824
2825/*
2826 * Common helper functions.
2827 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08002828unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829{
Akinobu Mita945a1112009-09-21 17:01:47 -07002830 struct page *page;
2831
2832 /*
2833 * __get_free_pages() returns a 32-bit address, which cannot represent
2834 * a highmem page
2835 */
2836 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2837
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838 page = alloc_pages(gfp_mask, order);
2839 if (!page)
2840 return 0;
2841 return (unsigned long) page_address(page);
2842}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843EXPORT_SYMBOL(__get_free_pages);
2844
Harvey Harrison920c7a52008-02-04 22:29:26 -08002845unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846{
Akinobu Mita945a1112009-09-21 17:01:47 -07002847 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849EXPORT_SYMBOL(get_zeroed_page);
2850
Harvey Harrison920c7a52008-02-04 22:29:26 -08002851void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852{
Nick Pigginb5810032005-10-29 18:16:12 -07002853 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854 if (order == 0)
Mel Gormanb745bc82014-06-04 16:10:22 -07002855 free_hot_cold_page(page, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002856 else
2857 __free_pages_ok(page, order);
2858 }
2859}
2860
2861EXPORT_SYMBOL(__free_pages);
2862
Harvey Harrison920c7a52008-02-04 22:29:26 -08002863void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864{
2865 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07002866 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 __free_pages(virt_to_page((void *)addr), order);
2868 }
2869}
2870
2871EXPORT_SYMBOL(free_pages);
2872
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002873/*
Vladimir Davydov52383432014-06-04 16:06:39 -07002874 * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
2875 * of the current memory cgroup.
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002876 *
Vladimir Davydov52383432014-06-04 16:06:39 -07002877 * It should be used when the caller would like to use kmalloc, but since the
2878 * allocation is large, it has to fall back to the page allocator.
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002879 */
Vladimir Davydov52383432014-06-04 16:06:39 -07002880struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order)
2881{
2882 struct page *page;
2883 struct mem_cgroup *memcg = NULL;
2884
2885 if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
2886 return NULL;
2887 page = alloc_pages(gfp_mask, order);
2888 memcg_kmem_commit_charge(page, memcg, order);
2889 return page;
2890}
2891
2892struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
2893{
2894 struct page *page;
2895 struct mem_cgroup *memcg = NULL;
2896
2897 if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
2898 return NULL;
2899 page = alloc_pages_node(nid, gfp_mask, order);
2900 memcg_kmem_commit_charge(page, memcg, order);
2901 return page;
2902}
2903
2904/*
2905 * __free_kmem_pages and free_kmem_pages will free pages allocated with
2906 * alloc_kmem_pages.
2907 */
2908void __free_kmem_pages(struct page *page, unsigned int order)
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002909{
2910 memcg_kmem_uncharge_pages(page, order);
2911 __free_pages(page, order);
2912}
2913
Vladimir Davydov52383432014-06-04 16:06:39 -07002914void free_kmem_pages(unsigned long addr, unsigned int order)
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002915{
2916 if (addr != 0) {
2917 VM_BUG_ON(!virt_addr_valid((void *)addr));
Vladimir Davydov52383432014-06-04 16:06:39 -07002918 __free_kmem_pages(virt_to_page((void *)addr), order);
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002919 }
2920}
2921
Andi Kleenee85c2e2011-05-11 15:13:34 -07002922static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
2923{
2924 if (addr) {
2925 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2926 unsigned long used = addr + PAGE_ALIGN(size);
2927
2928 split_page(virt_to_page((void *)addr), order);
2929 while (used < alloc_end) {
2930 free_page(used);
2931 used += PAGE_SIZE;
2932 }
2933 }
2934 return (void *)addr;
2935}
2936
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002937/**
2938 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2939 * @size: the number of bytes to allocate
2940 * @gfp_mask: GFP flags for the allocation
2941 *
2942 * This function is similar to alloc_pages(), except that it allocates the
2943 * minimum number of pages to satisfy the request. alloc_pages() can only
2944 * allocate memory in power-of-two pages.
2945 *
2946 * This function is also limited by MAX_ORDER.
2947 *
2948 * Memory allocated by this function must be released by free_pages_exact().
2949 */
2950void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2951{
2952 unsigned int order = get_order(size);
2953 unsigned long addr;
2954
2955 addr = __get_free_pages(gfp_mask, order);
Andi Kleenee85c2e2011-05-11 15:13:34 -07002956 return make_alloc_exact(addr, order, size);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002957}
2958EXPORT_SYMBOL(alloc_pages_exact);
2959
2960/**
Andi Kleenee85c2e2011-05-11 15:13:34 -07002961 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2962 * pages on a node.
Randy Dunlapb5e6ab52011-05-16 13:16:54 -07002963 * @nid: the preferred node ID where memory should be allocated
Andi Kleenee85c2e2011-05-11 15:13:34 -07002964 * @size: the number of bytes to allocate
2965 * @gfp_mask: GFP flags for the allocation
2966 *
2967 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2968 * back.
2969 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2970 * but is not exact.
2971 */
Fabian Fredericke1931812014-08-06 16:04:59 -07002972void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
Andi Kleenee85c2e2011-05-11 15:13:34 -07002973{
2974 unsigned order = get_order(size);
2975 struct page *p = alloc_pages_node(nid, gfp_mask, order);
2976 if (!p)
2977 return NULL;
2978 return make_alloc_exact((unsigned long)page_address(p), order, size);
2979}
Andi Kleenee85c2e2011-05-11 15:13:34 -07002980
2981/**
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002982 * free_pages_exact - release memory allocated via alloc_pages_exact()
2983 * @virt: the value returned by alloc_pages_exact.
2984 * @size: size of allocation, same value as passed to alloc_pages_exact().
2985 *
2986 * Release the memory allocated by a previous call to alloc_pages_exact.
2987 */
2988void free_pages_exact(void *virt, size_t size)
2989{
2990 unsigned long addr = (unsigned long)virt;
2991 unsigned long end = addr + PAGE_ALIGN(size);
2992
2993 while (addr < end) {
2994 free_page(addr);
2995 addr += PAGE_SIZE;
2996 }
2997}
2998EXPORT_SYMBOL(free_pages_exact);
2999
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003000/**
3001 * nr_free_zone_pages - count number of pages beyond high watermark
3002 * @offset: The zone index of the highest zone
3003 *
3004 * nr_free_zone_pages() counts the number of counts pages which are beyond the
3005 * high watermark within all zones at or below a given zone index. For each
3006 * zone, the number of pages is calculated as:
Jiang Liu834405c2013-07-03 15:03:04 -07003007 * managed_pages - high_pages
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003008 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003009static unsigned long nr_free_zone_pages(int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010{
Mel Gormandd1a2392008-04-28 02:12:17 -07003011 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003012 struct zone *zone;
3013
Martin J. Blighe310fd42005-07-29 22:59:18 -07003014 /* Just pick one node, since fallback list is circular */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003015 unsigned long sum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003016
Mel Gorman0e884602008-04-28 02:12:14 -07003017 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003018
Mel Gorman54a6eb52008-04-28 02:12:16 -07003019 for_each_zone_zonelist(zone, z, zonelist, offset) {
Jiang Liub40da042013-02-22 16:33:52 -08003020 unsigned long size = zone->managed_pages;
Mel Gorman41858962009-06-16 15:32:12 -07003021 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07003022 if (size > high)
3023 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024 }
3025
3026 return sum;
3027}
3028
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003029/**
3030 * nr_free_buffer_pages - count number of pages beyond high watermark
3031 *
3032 * nr_free_buffer_pages() counts the number of pages which are beyond the high
3033 * watermark within ZONE_DMA and ZONE_NORMAL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003034 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003035unsigned long nr_free_buffer_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003036{
Al Viroaf4ca452005-10-21 02:55:38 -04003037 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003038}
Meelap Shahc2f1a552007-07-17 04:04:39 -07003039EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003041/**
3042 * nr_free_pagecache_pages - count number of pages beyond high watermark
3043 *
3044 * nr_free_pagecache_pages() counts the number of pages which are beyond the
3045 * high watermark within all zones.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003047unsigned long nr_free_pagecache_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003048{
Mel Gorman2a1e2742007-07-17 04:03:12 -07003049 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003050}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07003051
3052static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003053{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08003054 if (IS_ENABLED(CONFIG_NUMA))
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08003055 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003056}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058void si_meminfo(struct sysinfo *val)
3059{
3060 val->totalram = totalram_pages;
Rafael Aquinicc7452b2014-08-06 16:06:38 -07003061 val->sharedram = global_page_state(NR_SHMEM);
Christoph Lameterd23ad422007-02-10 01:43:02 -08003062 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003063 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064 val->totalhigh = totalhigh_pages;
3065 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003066 val->mem_unit = PAGE_SIZE;
3067}
3068
3069EXPORT_SYMBOL(si_meminfo);
3070
3071#ifdef CONFIG_NUMA
3072void si_meminfo_node(struct sysinfo *val, int nid)
3073{
Jiang Liucdd91a72013-07-03 15:03:27 -07003074 int zone_type; /* needs to be signed */
3075 unsigned long managed_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003076 pg_data_t *pgdat = NODE_DATA(nid);
3077
Jiang Liucdd91a72013-07-03 15:03:27 -07003078 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
3079 managed_pages += pgdat->node_zones[zone_type].managed_pages;
3080 val->totalram = managed_pages;
Rafael Aquinicc7452b2014-08-06 16:06:38 -07003081 val->sharedram = node_page_state(nid, NR_SHMEM);
Christoph Lameterd23ad422007-02-10 01:43:02 -08003082 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003083#ifdef CONFIG_HIGHMEM
Jiang Liub40da042013-02-22 16:33:52 -08003084 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08003085 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
3086 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003087#else
3088 val->totalhigh = 0;
3089 val->freehigh = 0;
3090#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091 val->mem_unit = PAGE_SIZE;
3092}
3093#endif
3094
David Rientjesddd588b2011-03-22 16:30:46 -07003095/*
David Rientjes7bf02ea2011-05-24 17:11:16 -07003096 * Determine whether the node should be displayed or not, depending on whether
3097 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
David Rientjesddd588b2011-03-22 16:30:46 -07003098 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07003099bool skip_free_areas_node(unsigned int flags, int nid)
David Rientjesddd588b2011-03-22 16:30:46 -07003100{
3101 bool ret = false;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003102 unsigned int cpuset_mems_cookie;
David Rientjesddd588b2011-03-22 16:30:46 -07003103
3104 if (!(flags & SHOW_MEM_FILTER_NODES))
3105 goto out;
3106
Mel Gormancc9a6c82012-03-21 16:34:11 -07003107 do {
Mel Gormand26914d2014-04-03 14:47:24 -07003108 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07003109 ret = !node_isset(nid, cpuset_current_mems_allowed);
Mel Gormand26914d2014-04-03 14:47:24 -07003110 } while (read_mems_allowed_retry(cpuset_mems_cookie));
David Rientjesddd588b2011-03-22 16:30:46 -07003111out:
3112 return ret;
3113}
3114
Linus Torvalds1da177e2005-04-16 15:20:36 -07003115#define K(x) ((x) << (PAGE_SHIFT-10))
3116
Rabin Vincent377e4f12012-12-11 16:00:24 -08003117static void show_migration_types(unsigned char type)
3118{
3119 static const char types[MIGRATE_TYPES] = {
3120 [MIGRATE_UNMOVABLE] = 'U',
3121 [MIGRATE_RECLAIMABLE] = 'E',
3122 [MIGRATE_MOVABLE] = 'M',
3123 [MIGRATE_RESERVE] = 'R',
3124#ifdef CONFIG_CMA
3125 [MIGRATE_CMA] = 'C',
3126#endif
Minchan Kim194159f2013-02-22 16:33:58 -08003127#ifdef CONFIG_MEMORY_ISOLATION
Rabin Vincent377e4f12012-12-11 16:00:24 -08003128 [MIGRATE_ISOLATE] = 'I',
Minchan Kim194159f2013-02-22 16:33:58 -08003129#endif
Rabin Vincent377e4f12012-12-11 16:00:24 -08003130 };
3131 char tmp[MIGRATE_TYPES + 1];
3132 char *p = tmp;
3133 int i;
3134
3135 for (i = 0; i < MIGRATE_TYPES; i++) {
3136 if (type & (1 << i))
3137 *p++ = types[i];
3138 }
3139
3140 *p = '\0';
3141 printk("(%s) ", tmp);
3142}
3143
Linus Torvalds1da177e2005-04-16 15:20:36 -07003144/*
3145 * Show free area list (used inside shift_scroll-lock stuff)
3146 * We also calculate the percentage fragmentation. We do this by counting the
3147 * memory on each free list with the exception of the first item on the list.
David Rientjesddd588b2011-03-22 16:30:46 -07003148 * Suppresses nodes that are not allowed by current's cpuset if
3149 * SHOW_MEM_FILTER_NODES is passed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003150 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07003151void show_free_areas(unsigned int filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003152{
Jes Sorensenc7241912006-09-27 01:50:05 -07003153 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154 struct zone *zone;
3155
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003156 for_each_populated_zone(zone) {
David Rientjes7bf02ea2011-05-24 17:11:16 -07003157 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003158 continue;
Jes Sorensenc7241912006-09-27 01:50:05 -07003159 show_node(zone);
3160 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003161
Dave Jones6b482c62005-11-10 15:45:56 -05003162 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003163 struct per_cpu_pageset *pageset;
3164
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003165 pageset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003166
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003167 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
3168 cpu, pageset->pcp.high,
3169 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003170 }
3171 }
3172
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003173 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3174 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003175 " unevictable:%lu"
Andrew Mortonb76146e2009-10-26 16:49:52 -07003176 " dirty:%lu writeback:%lu unstable:%lu\n"
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07003177 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003178 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3179 " free_cma:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003180 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003181 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003182 global_page_state(NR_ISOLATED_ANON),
3183 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003184 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003185 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003186 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07003187 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07003188 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07003189 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08003190 global_page_state(NR_FREE_PAGES),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07003191 global_page_state(NR_SLAB_RECLAIMABLE),
3192 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07003193 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003194 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08003195 global_page_state(NR_PAGETABLE),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003196 global_page_state(NR_BOUNCE),
3197 global_page_state(NR_FREE_CMA_PAGES));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003198
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003199 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003200 int i;
3201
David Rientjes7bf02ea2011-05-24 17:11:16 -07003202 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003203 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003204 show_node(zone);
3205 printk("%s"
3206 " free:%lukB"
3207 " min:%lukB"
3208 " low:%lukB"
3209 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003210 " active_anon:%lukB"
3211 " inactive_anon:%lukB"
3212 " active_file:%lukB"
3213 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003214 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003215 " isolated(anon):%lukB"
3216 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003217 " present:%lukB"
Jiang Liu9feedc92012-12-12 13:52:12 -08003218 " managed:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003219 " mlocked:%lukB"
3220 " dirty:%lukB"
3221 " writeback:%lukB"
3222 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003223 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003224 " slab_reclaimable:%lukB"
3225 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003226 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003227 " pagetables:%lukB"
3228 " unstable:%lukB"
3229 " bounce:%lukB"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003230 " free_cma:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003231 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003232 " pages_scanned:%lu"
3233 " all_unreclaimable? %s"
3234 "\n",
3235 zone->name,
Mel Gorman88f5acf2011-01-13 15:45:41 -08003236 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07003237 K(min_wmark_pages(zone)),
3238 K(low_wmark_pages(zone)),
3239 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003240 K(zone_page_state(zone, NR_ACTIVE_ANON)),
3241 K(zone_page_state(zone, NR_INACTIVE_ANON)),
3242 K(zone_page_state(zone, NR_ACTIVE_FILE)),
3243 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003244 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003245 K(zone_page_state(zone, NR_ISOLATED_ANON)),
3246 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003247 K(zone->present_pages),
Jiang Liu9feedc92012-12-12 13:52:12 -08003248 K(zone->managed_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003249 K(zone_page_state(zone, NR_MLOCK)),
3250 K(zone_page_state(zone, NR_FILE_DIRTY)),
3251 K(zone_page_state(zone, NR_WRITEBACK)),
3252 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003253 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003254 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3255 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003256 zone_page_state(zone, NR_KERNEL_STACK) *
3257 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003258 K(zone_page_state(zone, NR_PAGETABLE)),
3259 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3260 K(zone_page_state(zone, NR_BOUNCE)),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003261 K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003262 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Mel Gorman0d5d8232014-08-06 16:07:16 -07003263 K(zone_page_state(zone, NR_PAGES_SCANNED)),
Lisa Du6e543d52013-09-11 14:22:36 -07003264 (!zone_reclaimable(zone) ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07003265 );
3266 printk("lowmem_reserve[]:");
3267 for (i = 0; i < MAX_NR_ZONES; i++)
Mel Gorman3484b2d2014-08-06 16:07:14 -07003268 printk(" %ld", zone->lowmem_reserve[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003269 printk("\n");
3270 }
3271
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003272 for_each_populated_zone(zone) {
Pintu Kumarb8af2942013-09-11 14:20:34 -07003273 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003274 unsigned char types[MAX_ORDER];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003275
David Rientjes7bf02ea2011-05-24 17:11:16 -07003276 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003277 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003278 show_node(zone);
3279 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003280
3281 spin_lock_irqsave(&zone->lock, flags);
3282 for (order = 0; order < MAX_ORDER; order++) {
Rabin Vincent377e4f12012-12-11 16:00:24 -08003283 struct free_area *area = &zone->free_area[order];
3284 int type;
3285
3286 nr[order] = area->nr_free;
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003287 total += nr[order] << order;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003288
3289 types[order] = 0;
3290 for (type = 0; type < MIGRATE_TYPES; type++) {
3291 if (!list_empty(&area->free_list[type]))
3292 types[order] |= 1 << type;
3293 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003294 }
3295 spin_unlock_irqrestore(&zone->lock, flags);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003296 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003297 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003298 if (nr[order])
3299 show_migration_types(types[order]);
3300 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003301 printk("= %lukB\n", K(total));
3302 }
3303
David Rientjes949f7ec2013-04-29 15:07:48 -07003304 hugetlb_show_meminfo();
3305
Larry Woodmane6f36022008-02-04 22:29:30 -08003306 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
3307
Linus Torvalds1da177e2005-04-16 15:20:36 -07003308 show_swap_cache_info();
3309}
3310
Mel Gorman19770b32008-04-28 02:12:18 -07003311static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
3312{
3313 zoneref->zone = zone;
3314 zoneref->zone_idx = zone_idx(zone);
3315}
3316
Linus Torvalds1da177e2005-04-16 15:20:36 -07003317/*
3318 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08003319 *
3320 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003321 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003322static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003323 int nr_zones)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003324{
Christoph Lameter1a932052006-01-06 00:11:16 -08003325 struct zone *zone;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003326 enum zone_type zone_type = MAX_NR_ZONES;
Christoph Lameter02a68a52006-01-06 00:11:18 -08003327
3328 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003329 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08003330 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08003331 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07003332 zoneref_set_zone(zone,
3333 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08003334 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003335 }
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003336 } while (zone_type);
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003337
Christoph Lameter070f8032006-01-06 00:11:19 -08003338 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003339}
3340
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003341
3342/*
3343 * zonelist_order:
3344 * 0 = automatic detection of better ordering.
3345 * 1 = order by ([node] distance, -zonetype)
3346 * 2 = order by (-zonetype, [node] distance)
3347 *
3348 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3349 * the same zonelist. So only NUMA can configure this param.
3350 */
3351#define ZONELIST_ORDER_DEFAULT 0
3352#define ZONELIST_ORDER_NODE 1
3353#define ZONELIST_ORDER_ZONE 2
3354
3355/* zonelist order in the kernel.
3356 * set_zonelist_order() will set this to NODE or ZONE.
3357 */
3358static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
3359static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
3360
3361
Linus Torvalds1da177e2005-04-16 15:20:36 -07003362#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003363/* The value user specified ....changed by config */
3364static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3365/* string for sysctl */
3366#define NUMA_ZONELIST_ORDER_LEN 16
3367char numa_zonelist_order[16] = "default";
3368
3369/*
3370 * interface for configure zonelist ordering.
3371 * command line option "numa_zonelist_order"
3372 * = "[dD]efault - default, automatic configuration.
3373 * = "[nN]ode - order by node locality, then by zone within node
3374 * = "[zZ]one - order by zone, then by locality within zone
3375 */
3376
3377static int __parse_numa_zonelist_order(char *s)
3378{
3379 if (*s == 'd' || *s == 'D') {
3380 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3381 } else if (*s == 'n' || *s == 'N') {
3382 user_zonelist_order = ZONELIST_ORDER_NODE;
3383 } else if (*s == 'z' || *s == 'Z') {
3384 user_zonelist_order = ZONELIST_ORDER_ZONE;
3385 } else {
3386 printk(KERN_WARNING
3387 "Ignoring invalid numa_zonelist_order value: "
3388 "%s\n", s);
3389 return -EINVAL;
3390 }
3391 return 0;
3392}
3393
3394static __init int setup_numa_zonelist_order(char *s)
3395{
Volodymyr G. Lukiianykecb256f2011-01-13 15:46:26 -08003396 int ret;
3397
3398 if (!s)
3399 return 0;
3400
3401 ret = __parse_numa_zonelist_order(s);
3402 if (ret == 0)
3403 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
3404
3405 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003406}
3407early_param("numa_zonelist_order", setup_numa_zonelist_order);
3408
3409/*
3410 * sysctl handler for numa_zonelist_order
3411 */
Joe Perchescccad5b2014-06-06 14:38:09 -07003412int numa_zonelist_order_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003413 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003414 loff_t *ppos)
3415{
3416 char saved_string[NUMA_ZONELIST_ORDER_LEN];
3417 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01003418 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003419
Andi Kleen443c6f12009-12-23 21:00:47 +01003420 mutex_lock(&zl_order_mutex);
Chen Gangdacbde02013-07-03 15:02:35 -07003421 if (write) {
3422 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
3423 ret = -EINVAL;
3424 goto out;
3425 }
3426 strcpy(saved_string, (char *)table->data);
3427 }
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003428 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003429 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01003430 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003431 if (write) {
3432 int oldval = user_zonelist_order;
Chen Gangdacbde02013-07-03 15:02:35 -07003433
3434 ret = __parse_numa_zonelist_order((char *)table->data);
3435 if (ret) {
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003436 /*
3437 * bogus value. restore saved string
3438 */
Chen Gangdacbde02013-07-03 15:02:35 -07003439 strncpy((char *)table->data, saved_string,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003440 NUMA_ZONELIST_ORDER_LEN);
3441 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003442 } else if (oldval != user_zonelist_order) {
3443 mutex_lock(&zonelists_mutex);
Jiang Liu9adb62a2012-07-31 16:43:28 -07003444 build_all_zonelists(NULL, NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003445 mutex_unlock(&zonelists_mutex);
3446 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003447 }
Andi Kleen443c6f12009-12-23 21:00:47 +01003448out:
3449 mutex_unlock(&zl_order_mutex);
3450 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003451}
3452
3453
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003454#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003455static int node_load[MAX_NUMNODES];
3456
Linus Torvalds1da177e2005-04-16 15:20:36 -07003457/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07003458 * 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 -07003459 * @node: node whose fallback list we're appending
3460 * @used_node_mask: nodemask_t of already used nodes
3461 *
3462 * We use a number of factors to determine which is the next node that should
3463 * appear on a given node's fallback list. The node should not have appeared
3464 * already in @node's fallback list, and it should be the next closest node
3465 * according to the distance array (which contains arbitrary distance values
3466 * from each node to each node in the system), and should also prefer nodes
3467 * with no CPUs, since presumably they'll have very little allocation pressure
3468 * on them otherwise.
3469 * It returns -1 if no node is found.
3470 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003471static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003472{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01003473 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003474 int min_val = INT_MAX;
David Rientjes00ef2d22013-02-22 16:35:36 -08003475 int best_node = NUMA_NO_NODE;
Rusty Russella70f7302009-03-13 14:49:46 +10303476 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003477
Linus Torvalds4cf808e2006-02-17 20:38:21 +01003478 /* Use the local node if we haven't already */
3479 if (!node_isset(node, *used_node_mask)) {
3480 node_set(node, *used_node_mask);
3481 return node;
3482 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003483
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08003484 for_each_node_state(n, N_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003485
3486 /* Don't want a node to appear more than once */
3487 if (node_isset(n, *used_node_mask))
3488 continue;
3489
Linus Torvalds1da177e2005-04-16 15:20:36 -07003490 /* Use the distance array to find the distance */
3491 val = node_distance(node, n);
3492
Linus Torvalds4cf808e2006-02-17 20:38:21 +01003493 /* Penalize nodes under us ("prefer the next node") */
3494 val += (n < node);
3495
Linus Torvalds1da177e2005-04-16 15:20:36 -07003496 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10303497 tmp = cpumask_of_node(n);
3498 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003499 val += PENALTY_FOR_NODE_WITH_CPUS;
3500
3501 /* Slight preference for less loaded node */
3502 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
3503 val += node_load[n];
3504
3505 if (val < min_val) {
3506 min_val = val;
3507 best_node = n;
3508 }
3509 }
3510
3511 if (best_node >= 0)
3512 node_set(best_node, *used_node_mask);
3513
3514 return best_node;
3515}
3516
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003517
3518/*
3519 * Build zonelists ordered by node and zones within node.
3520 * This results in maximum locality--normal zone overflows into local
3521 * DMA zone, if any--but risks exhausting DMA zone.
3522 */
3523static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003524{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003525 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003526 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003527
Mel Gorman54a6eb52008-04-28 02:12:16 -07003528 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07003529 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07003530 ;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003531 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gormandd1a2392008-04-28 02:12:17 -07003532 zonelist->_zonerefs[j].zone = NULL;
3533 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003534}
3535
3536/*
Christoph Lameter523b9452007-10-16 01:25:37 -07003537 * Build gfp_thisnode zonelists
3538 */
3539static void build_thisnode_zonelists(pg_data_t *pgdat)
3540{
Christoph Lameter523b9452007-10-16 01:25:37 -07003541 int j;
3542 struct zonelist *zonelist;
3543
Mel Gorman54a6eb52008-04-28 02:12:16 -07003544 zonelist = &pgdat->node_zonelists[1];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003545 j = build_zonelists_node(pgdat, zonelist, 0);
Mel Gormandd1a2392008-04-28 02:12:17 -07003546 zonelist->_zonerefs[j].zone = NULL;
3547 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07003548}
3549
3550/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003551 * Build zonelists ordered by zone and nodes within zones.
3552 * This results in conserving DMA zone[s] until all Normal memory is
3553 * exhausted, but results in overflowing to remote node while memory
3554 * may still exist in local DMA zone.
3555 */
3556static int node_order[MAX_NUMNODES];
3557
3558static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
3559{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003560 int pos, j, node;
3561 int zone_type; /* needs to be signed */
3562 struct zone *z;
3563 struct zonelist *zonelist;
3564
Mel Gorman54a6eb52008-04-28 02:12:16 -07003565 zonelist = &pgdat->node_zonelists[0];
3566 pos = 0;
3567 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
3568 for (j = 0; j < nr_nodes; j++) {
3569 node = node_order[j];
3570 z = &NODE_DATA(node)->node_zones[zone_type];
3571 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07003572 zoneref_set_zone(z,
3573 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07003574 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003575 }
3576 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003577 }
Mel Gormandd1a2392008-04-28 02:12:17 -07003578 zonelist->_zonerefs[pos].zone = NULL;
3579 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003580}
3581
3582static int default_zonelist_order(void)
3583{
3584 int nid, zone_type;
Pintu Kumarb8af2942013-09-11 14:20:34 -07003585 unsigned long low_kmem_size, total_size;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003586 struct zone *z;
3587 int average_size;
3588 /*
Pintu Kumarb8af2942013-09-11 14:20:34 -07003589 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003590 * If they are really small and used heavily, the system can fall
3591 * into OOM very easily.
David Rientjese325c902010-05-24 14:32:13 -07003592 * This function detect ZONE_DMA/DMA32 size and configures zone order.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003593 */
3594 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
3595 low_kmem_size = 0;
3596 total_size = 0;
3597 for_each_online_node(nid) {
3598 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3599 z = &NODE_DATA(nid)->node_zones[zone_type];
3600 if (populated_zone(z)) {
3601 if (zone_type < ZONE_NORMAL)
Jiang Liu4f9f4772013-07-03 15:03:07 -07003602 low_kmem_size += z->managed_pages;
3603 total_size += z->managed_pages;
David Rientjese325c902010-05-24 14:32:13 -07003604 } else if (zone_type == ZONE_NORMAL) {
3605 /*
3606 * If any node has only lowmem, then node order
3607 * is preferred to allow kernel allocations
3608 * locally; otherwise, they can easily infringe
3609 * on other nodes when there is an abundance of
3610 * lowmem available to allocate from.
3611 */
3612 return ZONELIST_ORDER_NODE;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003613 }
3614 }
3615 }
3616 if (!low_kmem_size || /* there are no DMA area. */
3617 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
3618 return ZONELIST_ORDER_NODE;
3619 /*
3620 * look into each node's config.
Pintu Kumarb8af2942013-09-11 14:20:34 -07003621 * If there is a node whose DMA/DMA32 memory is very big area on
3622 * local memory, NODE_ORDER may be suitable.
3623 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003624 average_size = total_size /
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08003625 (nodes_weight(node_states[N_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003626 for_each_online_node(nid) {
3627 low_kmem_size = 0;
3628 total_size = 0;
3629 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3630 z = &NODE_DATA(nid)->node_zones[zone_type];
3631 if (populated_zone(z)) {
3632 if (zone_type < ZONE_NORMAL)
3633 low_kmem_size += z->present_pages;
3634 total_size += z->present_pages;
3635 }
3636 }
3637 if (low_kmem_size &&
3638 total_size > average_size && /* ignore small node */
3639 low_kmem_size > total_size * 70/100)
3640 return ZONELIST_ORDER_NODE;
3641 }
3642 return ZONELIST_ORDER_ZONE;
3643}
3644
3645static void set_zonelist_order(void)
3646{
3647 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
3648 current_zonelist_order = default_zonelist_order();
3649 else
3650 current_zonelist_order = user_zonelist_order;
3651}
3652
3653static void build_zonelists(pg_data_t *pgdat)
3654{
3655 int j, node, load;
3656 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003657 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003658 int local_node, prev_node;
3659 struct zonelist *zonelist;
3660 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003661
3662 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07003663 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003664 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07003665 zonelist->_zonerefs[0].zone = NULL;
3666 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003667 }
3668
3669 /* NUMA-aware ordering of nodes */
3670 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003671 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003672 prev_node = local_node;
3673 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003674
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003675 memset(node_order, 0, sizeof(node_order));
3676 j = 0;
3677
Linus Torvalds1da177e2005-04-16 15:20:36 -07003678 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
3679 /*
3680 * We don't want to pressure a particular node.
3681 * So adding penalty to the first node in same
3682 * distance group to make it round-robin.
3683 */
David Rientjes957f8222012-10-08 16:33:24 -07003684 if (node_distance(local_node, node) !=
3685 node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003686 node_load[node] = load;
3687
Linus Torvalds1da177e2005-04-16 15:20:36 -07003688 prev_node = node;
3689 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003690 if (order == ZONELIST_ORDER_NODE)
3691 build_zonelists_in_node_order(pgdat, node);
3692 else
3693 node_order[j++] = node; /* remember order */
3694 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003695
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003696 if (order == ZONELIST_ORDER_ZONE) {
3697 /* calculate node order -- i.e., DMA last! */
3698 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003699 }
Christoph Lameter523b9452007-10-16 01:25:37 -07003700
3701 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003702}
3703
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003704/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003705static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003706{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003707 struct zonelist *zonelist;
3708 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07003709 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003710
Mel Gorman54a6eb52008-04-28 02:12:16 -07003711 zonelist = &pgdat->node_zonelists[0];
3712 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3713 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07003714 for (z = zonelist->_zonerefs; z->zone; z++)
3715 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003716}
3717
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003718#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3719/*
3720 * Return node id of node used for "local" allocations.
3721 * I.e., first node id of first zone in arg node's generic zonelist.
3722 * Used for initializing percpu 'numa_mem', which is used primarily
3723 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3724 */
3725int local_memory_node(int node)
3726{
3727 struct zone *zone;
3728
3729 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3730 gfp_zone(GFP_KERNEL),
3731 NULL,
3732 &zone);
3733 return zone->node;
3734}
3735#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003736
Linus Torvalds1da177e2005-04-16 15:20:36 -07003737#else /* CONFIG_NUMA */
3738
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003739static void set_zonelist_order(void)
3740{
3741 current_zonelist_order = ZONELIST_ORDER_ZONE;
3742}
3743
3744static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003745{
Christoph Lameter19655d32006-09-25 23:31:19 -07003746 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003747 enum zone_type j;
3748 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003749
3750 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003751
Mel Gorman54a6eb52008-04-28 02:12:16 -07003752 zonelist = &pgdat->node_zonelists[0];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003753 j = build_zonelists_node(pgdat, zonelist, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003754
Mel Gorman54a6eb52008-04-28 02:12:16 -07003755 /*
3756 * Now we build the zonelist so that it contains the zones
3757 * of all the other nodes.
3758 * We don't want to pressure a particular node, so when
3759 * building the zones for node N, we make sure that the
3760 * zones coming right after the local ones are those from
3761 * node N+1 (modulo N)
3762 */
3763 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3764 if (!node_online(node))
3765 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003766 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003767 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07003768 for (node = 0; node < local_node; node++) {
3769 if (!node_online(node))
3770 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003771 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gorman54a6eb52008-04-28 02:12:16 -07003772 }
3773
Mel Gormandd1a2392008-04-28 02:12:17 -07003774 zonelist->_zonerefs[j].zone = NULL;
3775 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003776}
3777
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003778/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003779static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003780{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003781 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003782}
3783
Linus Torvalds1da177e2005-04-16 15:20:36 -07003784#endif /* CONFIG_NUMA */
3785
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003786/*
3787 * Boot pageset table. One per cpu which is going to be used for all
3788 * zones and all nodes. The parameters will be set in such a way
3789 * that an item put on a list will immediately be handed over to
3790 * the buddy list. This is safe since pageset manipulation is done
3791 * with interrupts disabled.
3792 *
3793 * The boot_pagesets must be kept even after bootup is complete for
3794 * unused processors and/or zones. They do play a role for bootstrapping
3795 * hotplugged processors.
3796 *
3797 * zoneinfo_show() and maybe other functions do
3798 * not check if the processor is online before following the pageset pointer.
3799 * Other parts of the kernel may not check if the zone is available.
3800 */
3801static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3802static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07003803static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003804
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003805/*
3806 * Global mutex to protect against size modification of zonelists
3807 * as well as to serialize pageset setup for the new populated zone.
3808 */
3809DEFINE_MUTEX(zonelists_mutex);
3810
Rusty Russell9b1a4d32008-07-28 12:16:30 -05003811/* return values int ....just for stop_machine() */
Jiang Liu4ed7e022012-07-31 16:43:35 -07003812static int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003813{
Yasunori Goto68113782006-06-23 02:03:11 -07003814 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003815 int cpu;
Jiang Liu9adb62a2012-07-31 16:43:28 -07003816 pg_data_t *self = data;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003817
Bo Liu7f9cfb32009-08-18 14:11:19 -07003818#ifdef CONFIG_NUMA
3819 memset(node_load, 0, sizeof(node_load));
3820#endif
Jiang Liu9adb62a2012-07-31 16:43:28 -07003821
3822 if (self && !node_online(self->node_id)) {
3823 build_zonelists(self);
3824 build_zonelist_cache(self);
3825 }
3826
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003827 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07003828 pg_data_t *pgdat = NODE_DATA(nid);
3829
3830 build_zonelists(pgdat);
3831 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003832 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003833
3834 /*
3835 * Initialize the boot_pagesets that are going to be used
3836 * for bootstrapping processors. The real pagesets for
3837 * each zone will be allocated later when the per cpu
3838 * allocator is available.
3839 *
3840 * boot_pagesets are used also for bootstrapping offline
3841 * cpus if the system is already booted because the pagesets
3842 * are needed to initialize allocators on a specific cpu too.
3843 * F.e. the percpu allocator needs the page allocator which
3844 * needs the percpu allocator in order to allocate its pagesets
3845 * (a chicken-egg dilemma).
3846 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003847 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003848 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3849
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003850#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3851 /*
3852 * We now know the "local memory node" for each node--
3853 * i.e., the node of the first zone in the generic zonelist.
3854 * Set up numa_mem percpu variable for on-line cpus. During
3855 * boot, only the boot cpu should be on-line; we'll init the
3856 * secondary cpus' numa_mem as they come on-line. During
3857 * node/memory hotplug, we'll fixup all on-line cpus.
3858 */
3859 if (cpu_online(cpu))
3860 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3861#endif
3862 }
3863
Yasunori Goto68113782006-06-23 02:03:11 -07003864 return 0;
3865}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003866
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003867/*
3868 * Called with zonelists_mutex held always
3869 * unless system_state == SYSTEM_BOOTING.
3870 */
Jiang Liu9adb62a2012-07-31 16:43:28 -07003871void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
Yasunori Goto68113782006-06-23 02:03:11 -07003872{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003873 set_zonelist_order();
3874
Yasunori Goto68113782006-06-23 02:03:11 -07003875 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07003876 __build_all_zonelists(NULL);
Mel Gorman68ad8df2008-07-23 21:26:52 -07003877 mminit_verify_zonelist();
Yasunori Goto68113782006-06-23 02:03:11 -07003878 cpuset_init_current_mems_allowed();
3879 } else {
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003880#ifdef CONFIG_MEMORY_HOTPLUG
Jiang Liu9adb62a2012-07-31 16:43:28 -07003881 if (zone)
3882 setup_zone_pageset(zone);
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003883#endif
Cody P Schaferdd1895e2013-07-03 15:01:36 -07003884 /* we have to stop all cpus to guarantee there is no user
3885 of zonelist */
Jiang Liu9adb62a2012-07-31 16:43:28 -07003886 stop_machine(__build_all_zonelists, pgdat, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07003887 /* cpuset refresh routine should be here */
3888 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07003889 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003890 /*
3891 * Disable grouping by mobility if the number of pages in the
3892 * system is too low to allow the mechanism to work. It would be
3893 * more accurate, but expensive to check per-zone. This check is
3894 * made on memory-hotadd so a system can start with mobility
3895 * disabled and enable it later
3896 */
Mel Gormand9c23402007-10-16 01:26:01 -07003897 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003898 page_group_by_mobility_disabled = 1;
3899 else
3900 page_group_by_mobility_disabled = 0;
3901
3902 printk("Built %i zonelists in %s order, mobility grouping %s. "
3903 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003904 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003905 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003906 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003907 vm_total_pages);
3908#ifdef CONFIG_NUMA
3909 printk("Policy zone: %s\n", zone_names[policy_zone]);
3910#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003911}
3912
3913/*
3914 * Helper functions to size the waitqueue hash table.
3915 * Essentially these want to choose hash table sizes sufficiently
3916 * large so that collisions trying to wait on pages are rare.
3917 * But in fact, the number of active page waitqueues on typical
3918 * systems is ridiculously low, less than 200. So this is even
3919 * conservative, even though it seems large.
3920 *
3921 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3922 * waitqueues, i.e. the size of the waitq table given the number of pages.
3923 */
3924#define PAGES_PER_WAITQUEUE 256
3925
Yasunori Gotocca448f2006-06-23 02:03:10 -07003926#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07003927static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003928{
3929 unsigned long size = 1;
3930
3931 pages /= PAGES_PER_WAITQUEUE;
3932
3933 while (size < pages)
3934 size <<= 1;
3935
3936 /*
3937 * Once we have dozens or even hundreds of threads sleeping
3938 * on IO we've got bigger problems than wait queue collision.
3939 * Limit the size of the wait table to a reasonable size.
3940 */
3941 size = min(size, 4096UL);
3942
3943 return max(size, 4UL);
3944}
Yasunori Gotocca448f2006-06-23 02:03:10 -07003945#else
3946/*
3947 * A zone's size might be changed by hot-add, so it is not possible to determine
3948 * a suitable size for its wait_table. So we use the maximum size now.
3949 *
3950 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3951 *
3952 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3953 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3954 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3955 *
3956 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3957 * or more by the traditional way. (See above). It equals:
3958 *
3959 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3960 * ia64(16K page size) : = ( 8G + 4M)byte.
3961 * powerpc (64K page size) : = (32G +16M)byte.
3962 */
3963static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3964{
3965 return 4096UL;
3966}
3967#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003968
3969/*
3970 * This is an integer logarithm so that shifts can be used later
3971 * to extract the more random high bits from the multiplicative
3972 * hash function before the remainder is taken.
3973 */
3974static inline unsigned long wait_table_bits(unsigned long size)
3975{
3976 return ffz(~size);
3977}
3978
Mel Gorman56fd56b2007-10-16 01:25:58 -07003979/*
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07003980 * Check if a pageblock contains reserved pages
3981 */
3982static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
3983{
3984 unsigned long pfn;
3985
3986 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
3987 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
3988 return 1;
3989 }
3990 return 0;
3991}
3992
3993/*
Mel Gormand9c23402007-10-16 01:26:01 -07003994 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07003995 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3996 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07003997 * higher will lead to a bigger reserve which will get freed as contiguous
3998 * blocks as reclaim kicks in
3999 */
4000static void setup_zone_migrate_reserve(struct zone *zone)
4001{
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07004002 unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004003 struct page *page;
Mel Gorman78986a62009-09-21 17:03:02 -07004004 unsigned long block_migratetype;
4005 int reserve;
Yasuaki Ishimatsu943dca12014-01-21 15:49:06 -08004006 int old_reserve;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004007
Michal Hockod02156382011-12-08 14:34:27 -08004008 /*
4009 * Get the start pfn, end pfn and the number of blocks to reserve
4010 * We have to be careful to be aligned to pageblock_nr_pages to
4011 * make sure that we always check pfn_valid for the first page in
4012 * the block.
4013 */
Mel Gorman56fd56b2007-10-16 01:25:58 -07004014 start_pfn = zone->zone_start_pfn;
Cody P Schafer108bcc92013-02-22 16:35:23 -08004015 end_pfn = zone_end_pfn(zone);
Michal Hockod02156382011-12-08 14:34:27 -08004016 start_pfn = roundup(start_pfn, pageblock_nr_pages);
Mel Gorman41858962009-06-16 15:32:12 -07004017 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07004018 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004019
Mel Gorman78986a62009-09-21 17:03:02 -07004020 /*
4021 * Reserve blocks are generally in place to help high-order atomic
4022 * allocations that are short-lived. A min_free_kbytes value that
4023 * would result in more than 2 reserve blocks for atomic allocations
4024 * is assumed to be in place to help anti-fragmentation for the
4025 * future allocation of hugepages at runtime.
4026 */
4027 reserve = min(2, reserve);
Yasuaki Ishimatsu943dca12014-01-21 15:49:06 -08004028 old_reserve = zone->nr_migrate_reserve_block;
4029
4030 /* When memory hot-add, we almost always need to do nothing */
4031 if (reserve == old_reserve)
4032 return;
4033 zone->nr_migrate_reserve_block = reserve;
Mel Gorman78986a62009-09-21 17:03:02 -07004034
Mel Gormand9c23402007-10-16 01:26:01 -07004035 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07004036 if (!pfn_valid(pfn))
4037 continue;
4038 page = pfn_to_page(pfn);
4039
Adam Litke344c7902008-09-02 14:35:38 -07004040 /* Watch out for overlapping nodes */
4041 if (page_to_nid(page) != zone_to_nid(zone))
4042 continue;
4043
Mel Gorman56fd56b2007-10-16 01:25:58 -07004044 block_migratetype = get_pageblock_migratetype(page);
4045
Mel Gorman938929f2012-01-10 15:07:14 -08004046 /* Only test what is necessary when the reserves are not met */
4047 if (reserve > 0) {
4048 /*
4049 * Blocks with reserved pages will never free, skip
4050 * them.
4051 */
4052 block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
4053 if (pageblock_is_reserved(pfn, block_end_pfn))
4054 continue;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004055
Mel Gorman938929f2012-01-10 15:07:14 -08004056 /* If this block is reserved, account for it */
4057 if (block_migratetype == MIGRATE_RESERVE) {
4058 reserve--;
4059 continue;
4060 }
4061
4062 /* Suitable for reserving if this block is movable */
4063 if (block_migratetype == MIGRATE_MOVABLE) {
4064 set_pageblock_migratetype(page,
4065 MIGRATE_RESERVE);
4066 move_freepages_block(zone, page,
4067 MIGRATE_RESERVE);
4068 reserve--;
4069 continue;
4070 }
Yasuaki Ishimatsu943dca12014-01-21 15:49:06 -08004071 } else if (!old_reserve) {
4072 /*
4073 * At boot time we don't need to scan the whole zone
4074 * for turning off MIGRATE_RESERVE.
4075 */
4076 break;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004077 }
4078
4079 /*
4080 * If the reserve is met and this is a previous reserved block,
4081 * take it back
4082 */
4083 if (block_migratetype == MIGRATE_RESERVE) {
4084 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4085 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4086 }
4087 }
4088}
Mel Gormanac0e5b72007-10-16 01:25:58 -07004089
Linus Torvalds1da177e2005-04-16 15:20:36 -07004090/*
4091 * Initially all pages are reserved - free ones are freed
4092 * up by free_all_bootmem() once the early boot process is
4093 * done. Non-atomic initialization, single-pass.
4094 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01004095void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08004096 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004097{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004098 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07004099 unsigned long end_pfn = start_pfn + size;
4100 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004101 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004102
Hugh Dickins22b31ee2009-01-06 14:40:09 -08004103 if (highest_memmap_pfn < end_pfn - 1)
4104 highest_memmap_pfn = end_pfn - 1;
4105
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004106 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08004107 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08004108 /*
4109 * There can be holes in boot-time mem_map[]s
4110 * handed to this function. They do not
4111 * exist on hotplugged memory.
4112 */
4113 if (context == MEMMAP_EARLY) {
4114 if (!early_pfn_valid(pfn))
4115 continue;
4116 if (!early_pfn_in_nid(pfn, nid))
4117 continue;
4118 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004119 page = pfn_to_page(pfn);
4120 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07004121 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08004122 init_page_count(page);
Mel Gorman22b751c2013-02-22 16:34:59 -08004123 page_mapcount_reset(page);
Peter Zijlstra90572892013-10-07 11:29:20 +01004124 page_cpupid_reset_last(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004125 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004126 /*
4127 * Mark the block movable so that blocks are reserved for
4128 * movable at startup. This will force kernel allocations
4129 * to reserve their blocks rather than leaking throughout
4130 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07004131 * kernel allocations are made. Later some blocks near
4132 * the start are marked MIGRATE_RESERVE by
4133 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004134 *
4135 * bitmap is created for zone's valid pfn range. but memmap
4136 * can be created for invalid pages (for alignment)
4137 * check here not to call set_pageblock_migratetype() against
4138 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004139 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004140 if ((z->zone_start_pfn <= pfn)
Cody P Schafer108bcc92013-02-22 16:35:23 -08004141 && (pfn < zone_end_pfn(z))
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004142 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07004143 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004144
Linus Torvalds1da177e2005-04-16 15:20:36 -07004145 INIT_LIST_HEAD(&page->lru);
4146#ifdef WANT_PAGE_VIRTUAL
4147 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
4148 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07004149 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004150#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004151 }
4152}
4153
Andi Kleen1e548de2008-02-04 22:29:26 -08004154static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004155{
Mel Gorman7aeb09f2014-06-04 16:10:21 -07004156 unsigned int order, t;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004157 for_each_migratetype_order(order, t) {
4158 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004159 zone->free_area[order].nr_free = 0;
4160 }
4161}
4162
4163#ifndef __HAVE_ARCH_MEMMAP_INIT
4164#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08004165 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004166#endif
4167
David Rientjes7cd2b0a2014-06-23 13:22:04 -07004168static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004169{
David Howells3a6be872009-05-06 16:03:03 -07004170#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004171 int batch;
4172
4173 /*
4174 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07004175 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004176 *
4177 * OK, so we don't know how big the cache is. So guess.
4178 */
Jiang Liub40da042013-02-22 16:33:52 -08004179 batch = zone->managed_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07004180 if (batch * PAGE_SIZE > 512 * 1024)
4181 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004182 batch /= 4; /* We effectively *= 4 below */
4183 if (batch < 1)
4184 batch = 1;
4185
4186 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004187 * Clamp the batch to a 2^n - 1 value. Having a power
4188 * of 2 value was found to be more likely to have
4189 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004190 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004191 * For example if 2 tasks are alternately allocating
4192 * batches of pages, one task can end up with a lot
4193 * of pages of one half of the possible page colors
4194 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004195 */
David Howells91552032009-05-06 16:03:02 -07004196 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07004197
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004198 return batch;
David Howells3a6be872009-05-06 16:03:03 -07004199
4200#else
4201 /* The deferral and batching of frees should be suppressed under NOMMU
4202 * conditions.
4203 *
4204 * The problem is that NOMMU needs to be able to allocate large chunks
4205 * of contiguous memory as there's no hardware page translation to
4206 * assemble apparent contiguous memory from discontiguous pages.
4207 *
4208 * Queueing large contiguous runs of pages for batching, however,
4209 * causes the pages to actually be freed in smaller chunks. As there
4210 * can be a significant delay between the individual batches being
4211 * recycled, this leads to the once large chunks of space being
4212 * fragmented and becoming unavailable for high-order allocations.
4213 */
4214 return 0;
4215#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004216}
4217
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004218/*
4219 * pcp->high and pcp->batch values are related and dependent on one another:
4220 * ->batch must never be higher then ->high.
4221 * The following function updates them in a safe manner without read side
4222 * locking.
4223 *
4224 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4225 * those fields changing asynchronously (acording the the above rule).
4226 *
4227 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4228 * outside of boot time (or some other assurance that no concurrent updaters
4229 * exist).
4230 */
4231static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
4232 unsigned long batch)
4233{
4234 /* start with a fail safe value for batch */
4235 pcp->batch = 1;
4236 smp_wmb();
4237
4238 /* Update high, then batch, in order */
4239 pcp->high = high;
4240 smp_wmb();
4241
4242 pcp->batch = batch;
4243}
4244
Cody P Schafer36640332013-07-03 15:01:40 -07004245/* a companion to pageset_set_high() */
Cody P Schafer4008bab2013-07-03 15:01:28 -07004246static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
4247{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004248 pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
Cody P Schafer4008bab2013-07-03 15:01:28 -07004249}
4250
Cody P Schafer88c90db2013-07-03 15:01:35 -07004251static void pageset_init(struct per_cpu_pageset *p)
Christoph Lameter2caaad42005-06-21 17:15:00 -07004252{
4253 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004254 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004255
Magnus Damm1c6fe942005-10-26 01:58:59 -07004256 memset(p, 0, sizeof(*p));
4257
Christoph Lameter3dfa5722008-02-04 22:29:19 -08004258 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004259 pcp->count = 0;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004260 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
4261 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07004262}
4263
Cody P Schafer88c90db2013-07-03 15:01:35 -07004264static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
4265{
4266 pageset_init(p);
4267 pageset_set_batch(p, batch);
4268}
4269
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004270/*
Cody P Schafer36640332013-07-03 15:01:40 -07004271 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004272 * to the value high for the pageset p.
4273 */
Cody P Schafer36640332013-07-03 15:01:40 -07004274static void pageset_set_high(struct per_cpu_pageset *p,
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004275 unsigned long high)
4276{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004277 unsigned long batch = max(1UL, high / 4);
4278 if ((high / 4) > (PAGE_SHIFT * 8))
4279 batch = PAGE_SHIFT * 8;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004280
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004281 pageset_update(&p->pcp, high, batch);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004282}
4283
David Rientjes7cd2b0a2014-06-23 13:22:04 -07004284static void pageset_set_high_and_batch(struct zone *zone,
4285 struct per_cpu_pageset *pcp)
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004286{
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004287 if (percpu_pagelist_fraction)
Cody P Schafer36640332013-07-03 15:01:40 -07004288 pageset_set_high(pcp,
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004289 (zone->managed_pages /
4290 percpu_pagelist_fraction));
4291 else
4292 pageset_set_batch(pcp, zone_batchsize(zone));
4293}
4294
Cody P Schafer169f6c12013-07-03 15:01:41 -07004295static void __meminit zone_pageset_init(struct zone *zone, int cpu)
4296{
4297 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4298
4299 pageset_init(pcp);
4300 pageset_set_high_and_batch(zone, pcp);
4301}
4302
Jiang Liu4ed7e022012-07-31 16:43:35 -07004303static void __meminit setup_zone_pageset(struct zone *zone)
Wu Fengguang319774e2010-05-24 14:32:49 -07004304{
4305 int cpu;
Wu Fengguang319774e2010-05-24 14:32:49 -07004306 zone->pageset = alloc_percpu(struct per_cpu_pageset);
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004307 for_each_possible_cpu(cpu)
4308 zone_pageset_init(zone, cpu);
Wu Fengguang319774e2010-05-24 14:32:49 -07004309}
4310
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004311/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004312 * Allocate per cpu pagesets and initialize them.
4313 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07004314 */
Al Viro78d99552005-12-15 09:18:25 +00004315void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004316{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004317 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004318
Wu Fengguang319774e2010-05-24 14:32:49 -07004319 for_each_populated_zone(zone)
4320 setup_zone_pageset(zone);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004321}
4322
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004323static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07004324int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07004325{
4326 int i;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004327 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07004328
4329 /*
4330 * The per-page waitqueue mechanism uses hashed waitqueues
4331 * per zone.
4332 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07004333 zone->wait_table_hash_nr_entries =
4334 wait_table_hash_nr_entries(zone_size_pages);
4335 zone->wait_table_bits =
4336 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004337 alloc_size = zone->wait_table_hash_nr_entries
4338 * sizeof(wait_queue_head_t);
4339
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07004340 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07004341 zone->wait_table = (wait_queue_head_t *)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004342 memblock_virt_alloc_node_nopanic(
4343 alloc_size, zone->zone_pgdat->node_id);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004344 } else {
4345 /*
4346 * This case means that a zone whose size was 0 gets new memory
4347 * via memory hot-add.
4348 * But it may be the case that a new node was hot-added. In
4349 * this case vmalloc() will not be able to use this new node's
4350 * memory - this wait_table must be initialized to use this new
4351 * node itself as well.
4352 * To use this new node's memory, further consideration will be
4353 * necessary.
4354 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07004355 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004356 }
4357 if (!zone->wait_table)
4358 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07004359
Pintu Kumarb8af2942013-09-11 14:20:34 -07004360 for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07004361 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004362
4363 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004364}
4365
Matt Tolentinoc09b4242006-01-17 07:03:44 +01004366static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07004367{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004368 /*
4369 * per cpu subsystem is not up at this point. The following code
4370 * relies on the ability of the linker to provide the
4371 * offset of a (static) per cpu variable into the per cpu area.
4372 */
4373 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07004374
Xishi Qiub38a8722013-11-12 15:07:20 -08004375 if (populated_zone(zone))
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004376 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
4377 zone->name, zone->present_pages,
4378 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07004379}
4380
Jiang Liu4ed7e022012-07-31 16:43:35 -07004381int __meminit init_currently_empty_zone(struct zone *zone,
Yasunori Goto718127c2006-06-23 02:03:10 -07004382 unsigned long zone_start_pfn,
Dave Hansena2f3aa022007-01-10 23:15:30 -08004383 unsigned long size,
4384 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07004385{
4386 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004387 int ret;
4388 ret = zone_wait_table_init(zone, size);
4389 if (ret)
4390 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07004391 pgdat->nr_zones = zone_idx(zone) + 1;
4392
Dave Hansened8ece22005-10-29 18:16:50 -07004393 zone->zone_start_pfn = zone_start_pfn;
4394
Mel Gorman708614e2008-07-23 21:26:51 -07004395 mminit_dprintk(MMINIT_TRACE, "memmap_init",
4396 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4397 pgdat->node_id,
4398 (unsigned long)zone_idx(zone),
4399 zone_start_pfn, (zone_start_pfn + size));
4400
Andi Kleen1e548de2008-02-04 22:29:26 -08004401 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07004402
4403 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004404}
4405
Tejun Heo0ee332c2011-12-08 10:22:09 -08004406#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07004407#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4408/*
4409 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
Mel Gormanc7132162006-09-27 01:49:43 -07004410 */
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004411int __meminit __early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004412{
Tejun Heoc13291a2011-07-12 10:46:30 +02004413 unsigned long start_pfn, end_pfn;
Yinghai Lue76b63f2013-09-11 14:22:17 -07004414 int nid;
Russ Anderson7c243c72013-04-29 15:07:59 -07004415 /*
4416 * NOTE: The following SMP-unsafe globals are only used early in boot
4417 * when the kernel is running single-threaded.
4418 */
4419 static unsigned long __meminitdata last_start_pfn, last_end_pfn;
4420 static int __meminitdata last_nid;
4421
4422 if (last_start_pfn <= pfn && pfn < last_end_pfn)
4423 return last_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004424
Yinghai Lue76b63f2013-09-11 14:22:17 -07004425 nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
4426 if (nid != -1) {
4427 last_start_pfn = start_pfn;
4428 last_end_pfn = end_pfn;
4429 last_nid = nid;
4430 }
4431
4432 return nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004433}
4434#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4435
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004436int __meminit early_pfn_to_nid(unsigned long pfn)
4437{
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08004438 int nid;
4439
4440 nid = __early_pfn_to_nid(pfn);
4441 if (nid >= 0)
4442 return nid;
4443 /* just returns 0 */
4444 return 0;
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004445}
4446
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08004447#ifdef CONFIG_NODES_SPAN_OTHER_NODES
4448bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
4449{
4450 int nid;
4451
4452 nid = __early_pfn_to_nid(pfn);
4453 if (nid >= 0 && nid != node)
4454 return false;
4455 return true;
4456}
4457#endif
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004458
Mel Gormanc7132162006-09-27 01:49:43 -07004459/**
Santosh Shilimkar67828322014-01-21 15:50:25 -08004460 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004461 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
Santosh Shilimkar67828322014-01-21 15:50:25 -08004462 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
Mel Gormanc7132162006-09-27 01:49:43 -07004463 *
Zhang Zhen7d018172014-06-04 16:10:53 -07004464 * If an architecture guarantees that all ranges registered contain no holes
4465 * and may be freed, this this function may be used instead of calling
4466 * memblock_free_early_nid() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07004467 */
Tejun Heoc13291a2011-07-12 10:46:30 +02004468void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004469{
Tejun Heoc13291a2011-07-12 10:46:30 +02004470 unsigned long start_pfn, end_pfn;
4471 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004472
Tejun Heoc13291a2011-07-12 10:46:30 +02004473 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
4474 start_pfn = min(start_pfn, max_low_pfn);
4475 end_pfn = min(end_pfn, max_low_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004476
Tejun Heoc13291a2011-07-12 10:46:30 +02004477 if (start_pfn < end_pfn)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004478 memblock_free_early_nid(PFN_PHYS(start_pfn),
4479 (end_pfn - start_pfn) << PAGE_SHIFT,
4480 this_nid);
Mel Gormanc7132162006-09-27 01:49:43 -07004481 }
4482}
4483
4484/**
4485 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004486 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07004487 *
Zhang Zhen7d018172014-06-04 16:10:53 -07004488 * If an architecture guarantees that all ranges registered contain no holes and may
4489 * be freed, this function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07004490 */
4491void __init sparse_memory_present_with_active_regions(int nid)
4492{
Tejun Heoc13291a2011-07-12 10:46:30 +02004493 unsigned long start_pfn, end_pfn;
4494 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004495
Tejun Heoc13291a2011-07-12 10:46:30 +02004496 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
4497 memory_present(this_nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004498}
4499
4500/**
4501 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004502 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4503 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4504 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07004505 *
4506 * It returns the start and end page frame of a node based on information
Zhang Zhen7d018172014-06-04 16:10:53 -07004507 * provided by memblock_set_node(). If called for a node
Mel Gormanc7132162006-09-27 01:49:43 -07004508 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004509 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07004510 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004511void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004512 unsigned long *start_pfn, unsigned long *end_pfn)
4513{
Tejun Heoc13291a2011-07-12 10:46:30 +02004514 unsigned long this_start_pfn, this_end_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004515 int i;
Tejun Heoc13291a2011-07-12 10:46:30 +02004516
Mel Gormanc7132162006-09-27 01:49:43 -07004517 *start_pfn = -1UL;
4518 *end_pfn = 0;
4519
Tejun Heoc13291a2011-07-12 10:46:30 +02004520 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
4521 *start_pfn = min(*start_pfn, this_start_pfn);
4522 *end_pfn = max(*end_pfn, this_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004523 }
4524
Christoph Lameter633c0662007-10-16 01:25:37 -07004525 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07004526 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07004527}
4528
4529/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07004530 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4531 * assumption is made that zones within a node are ordered in monotonic
4532 * increasing memory addresses so that the "highest" populated zone is used
4533 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004534static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004535{
4536 int zone_index;
4537 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
4538 if (zone_index == ZONE_MOVABLE)
4539 continue;
4540
4541 if (arch_zone_highest_possible_pfn[zone_index] >
4542 arch_zone_lowest_possible_pfn[zone_index])
4543 break;
4544 }
4545
4546 VM_BUG_ON(zone_index == -1);
4547 movable_zone = zone_index;
4548}
4549
4550/*
4551 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004552 * because it is sized independent of architecture. Unlike the other zones,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004553 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4554 * in each node depending on the size of each node and how evenly kernelcore
4555 * is distributed. This helper function adjusts the zone ranges
4556 * provided by the architecture for a given node by using the end of the
4557 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4558 * zones within a node are in order of monotonic increases memory addresses
4559 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004560static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004561 unsigned long zone_type,
4562 unsigned long node_start_pfn,
4563 unsigned long node_end_pfn,
4564 unsigned long *zone_start_pfn,
4565 unsigned long *zone_end_pfn)
4566{
4567 /* Only adjust if ZONE_MOVABLE is on this node */
4568 if (zone_movable_pfn[nid]) {
4569 /* Size ZONE_MOVABLE */
4570 if (zone_type == ZONE_MOVABLE) {
4571 *zone_start_pfn = zone_movable_pfn[nid];
4572 *zone_end_pfn = min(node_end_pfn,
4573 arch_zone_highest_possible_pfn[movable_zone]);
4574
4575 /* Adjust for ZONE_MOVABLE starting within this range */
4576 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
4577 *zone_end_pfn > zone_movable_pfn[nid]) {
4578 *zone_end_pfn = zone_movable_pfn[nid];
4579
4580 /* Check if this whole range is within ZONE_MOVABLE */
4581 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
4582 *zone_start_pfn = *zone_end_pfn;
4583 }
4584}
4585
4586/*
Mel Gormanc7132162006-09-27 01:49:43 -07004587 * Return the number of pages a zone spans in a node, including holes
4588 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4589 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004590static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004591 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004592 unsigned long node_start_pfn,
4593 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004594 unsigned long *ignored)
4595{
Mel Gormanc7132162006-09-27 01:49:43 -07004596 unsigned long zone_start_pfn, zone_end_pfn;
4597
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004598 /* Get the start and end of the zone */
Mel Gormanc7132162006-09-27 01:49:43 -07004599 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4600 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07004601 adjust_zone_range_for_zone_movable(nid, zone_type,
4602 node_start_pfn, node_end_pfn,
4603 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004604
4605 /* Check that this node has pages within the zone's required range */
4606 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4607 return 0;
4608
4609 /* Move the zone boundaries inside the node if necessary */
4610 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4611 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4612
4613 /* Return the spanned pages */
4614 return zone_end_pfn - zone_start_pfn;
4615}
4616
4617/*
4618 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004619 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07004620 */
Yinghai Lu32996252009-12-15 17:59:02 -08004621unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004622 unsigned long range_start_pfn,
4623 unsigned long range_end_pfn)
4624{
Tejun Heo96e907d2011-07-12 10:46:29 +02004625 unsigned long nr_absent = range_end_pfn - range_start_pfn;
4626 unsigned long start_pfn, end_pfn;
4627 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004628
Tejun Heo96e907d2011-07-12 10:46:29 +02004629 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
4630 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
4631 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
4632 nr_absent -= end_pfn - start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004633 }
Tejun Heo96e907d2011-07-12 10:46:29 +02004634 return nr_absent;
Mel Gormanc7132162006-09-27 01:49:43 -07004635}
4636
4637/**
4638 * absent_pages_in_range - Return number of page frames in holes within a range
4639 * @start_pfn: The start PFN to start searching for holes
4640 * @end_pfn: The end PFN to stop searching for holes
4641 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004642 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07004643 */
4644unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4645 unsigned long end_pfn)
4646{
4647 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4648}
4649
4650/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004651static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004652 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004653 unsigned long node_start_pfn,
4654 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004655 unsigned long *ignored)
4656{
Tejun Heo96e907d2011-07-12 10:46:29 +02004657 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
4658 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman9c7cd682006-09-27 01:49:58 -07004659 unsigned long zone_start_pfn, zone_end_pfn;
4660
Tejun Heo96e907d2011-07-12 10:46:29 +02004661 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
4662 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004663
Mel Gorman2a1e2742007-07-17 04:03:12 -07004664 adjust_zone_range_for_zone_movable(nid, zone_type,
4665 node_start_pfn, node_end_pfn,
4666 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004667 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004668}
Mel Gorman0e0b8642006-09-27 01:49:56 -07004669
Tejun Heo0ee332c2011-12-08 10:22:09 -08004670#else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004671static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004672 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004673 unsigned long node_start_pfn,
4674 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004675 unsigned long *zones_size)
4676{
4677 return zones_size[zone_type];
4678}
4679
Paul Mundt6ea6e682007-07-15 23:38:20 -07004680static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004681 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004682 unsigned long node_start_pfn,
4683 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004684 unsigned long *zholes_size)
4685{
4686 if (!zholes_size)
4687 return 0;
4688
4689 return zholes_size[zone_type];
4690}
Yinghai Lu20e69262013-03-01 14:51:27 -08004691
Tejun Heo0ee332c2011-12-08 10:22:09 -08004692#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07004693
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004694static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004695 unsigned long node_start_pfn,
4696 unsigned long node_end_pfn,
4697 unsigned long *zones_size,
4698 unsigned long *zholes_size)
Mel Gormanc7132162006-09-27 01:49:43 -07004699{
4700 unsigned long realtotalpages, totalpages = 0;
4701 enum zone_type i;
4702
4703 for (i = 0; i < MAX_NR_ZONES; i++)
4704 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004705 node_start_pfn,
4706 node_end_pfn,
4707 zones_size);
Mel Gormanc7132162006-09-27 01:49:43 -07004708 pgdat->node_spanned_pages = totalpages;
4709
4710 realtotalpages = totalpages;
4711 for (i = 0; i < MAX_NR_ZONES; i++)
4712 realtotalpages -=
4713 zone_absent_pages_in_node(pgdat->node_id, i,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004714 node_start_pfn, node_end_pfn,
4715 zholes_size);
Mel Gormanc7132162006-09-27 01:49:43 -07004716 pgdat->node_present_pages = realtotalpages;
4717 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4718 realtotalpages);
4719}
4720
Mel Gorman835c1342007-10-16 01:25:47 -07004721#ifndef CONFIG_SPARSEMEM
4722/*
4723 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07004724 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4725 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07004726 * round what is now in bits to nearest long in bits, then return it in
4727 * bytes.
4728 */
Linus Torvalds7c455122013-02-18 09:58:02 -08004729static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07004730{
4731 unsigned long usemapsize;
4732
Linus Torvalds7c455122013-02-18 09:58:02 -08004733 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
Mel Gormand9c23402007-10-16 01:26:01 -07004734 usemapsize = roundup(zonesize, pageblock_nr_pages);
4735 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07004736 usemapsize *= NR_PAGEBLOCK_BITS;
4737 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4738
4739 return usemapsize / 8;
4740}
4741
4742static void __init setup_usemap(struct pglist_data *pgdat,
Linus Torvalds7c455122013-02-18 09:58:02 -08004743 struct zone *zone,
4744 unsigned long zone_start_pfn,
4745 unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07004746{
Linus Torvalds7c455122013-02-18 09:58:02 -08004747 unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
Mel Gorman835c1342007-10-16 01:25:47 -07004748 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08004749 if (usemapsize)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004750 zone->pageblock_flags =
4751 memblock_virt_alloc_node_nopanic(usemapsize,
4752 pgdat->node_id);
Mel Gorman835c1342007-10-16 01:25:47 -07004753}
4754#else
Linus Torvalds7c455122013-02-18 09:58:02 -08004755static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
4756 unsigned long zone_start_pfn, unsigned long zonesize) {}
Mel Gorman835c1342007-10-16 01:25:47 -07004757#endif /* CONFIG_SPARSEMEM */
4758
Mel Gormand9c23402007-10-16 01:26:01 -07004759#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08004760
Mel Gormand9c23402007-10-16 01:26:01 -07004761/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
Chen Gang15ca2202013-09-11 14:20:27 -07004762void __paginginit set_pageblock_order(void)
Mel Gormand9c23402007-10-16 01:26:01 -07004763{
Andrew Morton955c1cd2012-05-29 15:06:31 -07004764 unsigned int order;
4765
Mel Gormand9c23402007-10-16 01:26:01 -07004766 /* Check that pageblock_nr_pages has not already been setup */
4767 if (pageblock_order)
4768 return;
4769
Andrew Morton955c1cd2012-05-29 15:06:31 -07004770 if (HPAGE_SHIFT > PAGE_SHIFT)
4771 order = HUGETLB_PAGE_ORDER;
4772 else
4773 order = MAX_ORDER - 1;
4774
Mel Gormand9c23402007-10-16 01:26:01 -07004775 /*
4776 * Assume the largest contiguous order of interest is a huge page.
Andrew Morton955c1cd2012-05-29 15:06:31 -07004777 * This value may be variable depending on boot parameters on IA64 and
4778 * powerpc.
Mel Gormand9c23402007-10-16 01:26:01 -07004779 */
4780 pageblock_order = order;
4781}
4782#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4783
Mel Gormanba72cb82007-11-28 16:21:13 -08004784/*
4785 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
Andrew Morton955c1cd2012-05-29 15:06:31 -07004786 * is unused as pageblock_order is set at compile-time. See
4787 * include/linux/pageblock-flags.h for the values of pageblock_order based on
4788 * the kernel config
Mel Gormanba72cb82007-11-28 16:21:13 -08004789 */
Chen Gang15ca2202013-09-11 14:20:27 -07004790void __paginginit set_pageblock_order(void)
Mel Gormanba72cb82007-11-28 16:21:13 -08004791{
Mel Gormanba72cb82007-11-28 16:21:13 -08004792}
Mel Gormand9c23402007-10-16 01:26:01 -07004793
4794#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4795
Jiang Liu01cefae2012-12-12 13:52:19 -08004796static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
4797 unsigned long present_pages)
4798{
4799 unsigned long pages = spanned_pages;
4800
4801 /*
4802 * Provide a more accurate estimation if there are holes within
4803 * the zone and SPARSEMEM is in use. If there are holes within the
4804 * zone, each populated memory region may cost us one or two extra
4805 * memmap pages due to alignment because memmap pages for each
4806 * populated regions may not naturally algined on page boundary.
4807 * So the (present_pages >> 4) heuristic is a tradeoff for that.
4808 */
4809 if (spanned_pages > present_pages + (present_pages >> 4) &&
4810 IS_ENABLED(CONFIG_SPARSEMEM))
4811 pages = present_pages;
4812
4813 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
4814}
4815
Linus Torvalds1da177e2005-04-16 15:20:36 -07004816/*
4817 * Set up the zone data structures:
4818 * - mark all pages reserved
4819 * - mark all memory queues empty
4820 * - clear the memory bitmaps
Minchan Kim6527af52012-07-31 16:46:16 -07004821 *
4822 * NOTE: pgdat should get zeroed by caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004823 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08004824static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004825 unsigned long node_start_pfn, unsigned long node_end_pfn,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004826 unsigned long *zones_size, unsigned long *zholes_size)
4827{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07004828 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07004829 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004830 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07004831 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004832
Dave Hansen208d54e2005-10-29 18:16:52 -07004833 pgdat_resize_init(pgdat);
Andrea Arcangeli8177a422012-03-23 20:56:34 +01004834#ifdef CONFIG_NUMA_BALANCING
4835 spin_lock_init(&pgdat->numabalancing_migrate_lock);
4836 pgdat->numabalancing_migrate_nr_pages = 0;
4837 pgdat->numabalancing_migrate_next_window = jiffies;
4838#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004839 init_waitqueue_head(&pgdat->kswapd_wait);
Mel Gorman55150612012-07-31 16:44:35 -07004840 init_waitqueue_head(&pgdat->pfmemalloc_wait);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004841 pgdat_page_cgroup_init(pgdat);
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01004842
Linus Torvalds1da177e2005-04-16 15:20:36 -07004843 for (j = 0; j < MAX_NR_ZONES; j++) {
4844 struct zone *zone = pgdat->node_zones + j;
Jiang Liu9feedc92012-12-12 13:52:12 -08004845 unsigned long size, realsize, freesize, memmap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004846
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004847 size = zone_spanned_pages_in_node(nid, j, node_start_pfn,
4848 node_end_pfn, zones_size);
Jiang Liu9feedc92012-12-12 13:52:12 -08004849 realsize = freesize = size - zone_absent_pages_in_node(nid, j,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004850 node_start_pfn,
4851 node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004852 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004853
Mel Gorman0e0b8642006-09-27 01:49:56 -07004854 /*
Jiang Liu9feedc92012-12-12 13:52:12 -08004855 * Adjust freesize so that it accounts for how much memory
Mel Gorman0e0b8642006-09-27 01:49:56 -07004856 * is used by this zone for memmap. This affects the watermark
4857 * and per-cpu initialisations
4858 */
Jiang Liu01cefae2012-12-12 13:52:19 -08004859 memmap_pages = calc_memmap_size(size, realsize);
Jiang Liu9feedc92012-12-12 13:52:12 -08004860 if (freesize >= memmap_pages) {
4861 freesize -= memmap_pages;
Yinghai Lu5594c8c2009-01-06 14:39:14 -08004862 if (memmap_pages)
4863 printk(KERN_DEBUG
4864 " %s zone: %lu pages used for memmap\n",
4865 zone_names[j], memmap_pages);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004866 } else
4867 printk(KERN_WARNING
Jiang Liu9feedc92012-12-12 13:52:12 -08004868 " %s zone: %lu pages exceeds freesize %lu\n",
4869 zone_names[j], memmap_pages, freesize);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004870
Christoph Lameter62672762007-02-10 01:43:07 -08004871 /* Account for reserved pages */
Jiang Liu9feedc92012-12-12 13:52:12 -08004872 if (j == 0 && freesize > dma_reserve) {
4873 freesize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07004874 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08004875 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004876 }
4877
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004878 if (!is_highmem_idx(j))
Jiang Liu9feedc92012-12-12 13:52:12 -08004879 nr_kernel_pages += freesize;
Jiang Liu01cefae2012-12-12 13:52:19 -08004880 /* Charge for highmem memmap if there are enough kernel pages */
4881 else if (nr_kernel_pages > memmap_pages * 2)
4882 nr_kernel_pages -= memmap_pages;
Jiang Liu9feedc92012-12-12 13:52:12 -08004883 nr_all_pages += freesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004884
4885 zone->spanned_pages = size;
Jiang Liu306f2e92013-02-22 16:33:54 -08004886 zone->present_pages = realsize;
Jiang Liu9feedc92012-12-12 13:52:12 -08004887 /*
4888 * Set an approximate value for lowmem here, it will be adjusted
4889 * when the bootmem allocator frees pages into the buddy system.
4890 * And all highmem pages will be managed by the buddy system.
4891 */
4892 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
Christoph Lameter96146342006-07-03 00:24:13 -07004893#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07004894 zone->node = nid;
Jiang Liu9feedc92012-12-12 13:52:12 -08004895 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07004896 / 100;
Jiang Liu9feedc92012-12-12 13:52:12 -08004897 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07004898#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004899 zone->name = zone_names[j];
4900 spin_lock_init(&zone->lock);
4901 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07004902 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004903 zone->zone_pgdat = pgdat;
Dave Hansened8ece22005-10-29 18:16:50 -07004904 zone_pcp_init(zone);
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07004905
4906 /* For bootup, initialized properly in watermark setup */
4907 mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages);
4908
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004909 lruvec_init(&zone->lruvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004910 if (!size)
4911 continue;
4912
Andrew Morton955c1cd2012-05-29 15:06:31 -07004913 set_pageblock_order();
Linus Torvalds7c455122013-02-18 09:58:02 -08004914 setup_usemap(pgdat, zone, zone_start_pfn, size);
Dave Hansena2f3aa022007-01-10 23:15:30 -08004915 ret = init_currently_empty_zone(zone, zone_start_pfn,
4916 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07004917 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07004918 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004919 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004920 }
4921}
4922
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004923static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004924{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004925 /* Skip empty nodes */
4926 if (!pgdat->node_spanned_pages)
4927 return;
4928
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004929#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07004930 /* ia64 gets its own node_mem_map, before this, without bootmem */
4931 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07004932 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004933 struct page *map;
4934
Bob Piccoe984bb42006-05-20 15:00:31 -07004935 /*
4936 * The zone's endpoints aren't required to be MAX_ORDER
4937 * aligned but the node_mem_map endpoints must be in order
4938 * for the buddy allocator to function correctly.
4939 */
4940 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
Cody P Schafer108bcc92013-02-22 16:35:23 -08004941 end = pgdat_end_pfn(pgdat);
Bob Piccoe984bb42006-05-20 15:00:31 -07004942 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4943 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07004944 map = alloc_remap(pgdat->node_id, size);
4945 if (!map)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004946 map = memblock_virt_alloc_node_nopanic(size,
4947 pgdat->node_id);
Bob Piccoe984bb42006-05-20 15:00:31 -07004948 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004949 }
Roman Zippel12d810c2007-05-31 00:40:54 -07004950#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07004951 /*
4952 * With no DISCONTIG, the global mem_map is just set as node 0's
4953 */
Mel Gormanc7132162006-09-27 01:49:43 -07004954 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004955 mem_map = NODE_DATA(0)->node_mem_map;
Tejun Heo0ee332c2011-12-08 10:22:09 -08004956#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07004957 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08004958 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Tejun Heo0ee332c2011-12-08 10:22:09 -08004959#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07004960 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004961#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004962#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004963}
4964
Johannes Weiner9109fb72008-07-23 21:27:20 -07004965void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4966 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004967{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004968 pg_data_t *pgdat = NODE_DATA(nid);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004969 unsigned long start_pfn = 0;
4970 unsigned long end_pfn = 0;
Johannes Weiner9109fb72008-07-23 21:27:20 -07004971
Minchan Kim88fdf752012-07-31 16:46:14 -07004972 /* pg_data_t should be reset to zero when it's allocated */
Linus Torvalds8783b6e2012-08-02 10:37:03 -07004973 WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
Minchan Kim88fdf752012-07-31 16:46:14 -07004974
Linus Torvalds1da177e2005-04-16 15:20:36 -07004975 pgdat->node_id = nid;
4976 pgdat->node_start_pfn = node_start_pfn;
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004977#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4978 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
4979#endif
4980 calculate_node_totalpages(pgdat, start_pfn, end_pfn,
4981 zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004982
4983 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07004984#ifdef CONFIG_FLAT_NODE_MEM_MAP
4985 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4986 nid, (unsigned long)pgdat,
4987 (unsigned long)pgdat->node_mem_map);
4988#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004989
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004990 free_area_init_core(pgdat, start_pfn, end_pfn,
4991 zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004992}
4993
Tejun Heo0ee332c2011-12-08 10:22:09 -08004994#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07004995
4996#if MAX_NUMNODES > 1
4997/*
4998 * Figure out the number of possible node ids.
4999 */
Cody P Schaferf9872ca2013-04-29 15:08:01 -07005000void __init setup_nr_node_ids(void)
Miklos Szeredi418508c2007-05-23 13:57:55 -07005001{
5002 unsigned int node;
5003 unsigned int highest = 0;
5004
5005 for_each_node_mask(node, node_possible_map)
5006 highest = node;
5007 nr_node_ids = highest + 1;
5008}
Miklos Szeredi418508c2007-05-23 13:57:55 -07005009#endif
5010
Mel Gormanc7132162006-09-27 01:49:43 -07005011/**
Tejun Heo1e019792011-07-12 09:45:34 +02005012 * node_map_pfn_alignment - determine the maximum internode alignment
5013 *
5014 * This function should be called after node map is populated and sorted.
5015 * It calculates the maximum power of two alignment which can distinguish
5016 * all the nodes.
5017 *
5018 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5019 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
5020 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
5021 * shifted, 1GiB is enough and this function will indicate so.
5022 *
5023 * This is used to test whether pfn -> nid mapping of the chosen memory
5024 * model has fine enough granularity to avoid incorrect mapping for the
5025 * populated node map.
5026 *
5027 * Returns the determined alignment in pfn's. 0 if there is no alignment
5028 * requirement (single node).
5029 */
5030unsigned long __init node_map_pfn_alignment(void)
5031{
5032 unsigned long accl_mask = 0, last_end = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02005033 unsigned long start, end, mask;
Tejun Heo1e019792011-07-12 09:45:34 +02005034 int last_nid = -1;
Tejun Heoc13291a2011-07-12 10:46:30 +02005035 int i, nid;
Tejun Heo1e019792011-07-12 09:45:34 +02005036
Tejun Heoc13291a2011-07-12 10:46:30 +02005037 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
Tejun Heo1e019792011-07-12 09:45:34 +02005038 if (!start || last_nid < 0 || last_nid == nid) {
5039 last_nid = nid;
5040 last_end = end;
5041 continue;
5042 }
5043
5044 /*
5045 * Start with a mask granular enough to pin-point to the
5046 * start pfn and tick off bits one-by-one until it becomes
5047 * too coarse to separate the current node from the last.
5048 */
5049 mask = ~((1 << __ffs(start)) - 1);
5050 while (mask && last_end <= (start & (mask << 1)))
5051 mask <<= 1;
5052
5053 /* accumulate all internode masks */
5054 accl_mask |= mask;
5055 }
5056
5057 /* convert mask to number of pages */
5058 return ~accl_mask + 1;
5059}
5060
Mel Gormana6af2bc2007-02-10 01:42:57 -08005061/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07005062static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07005063{
Mel Gormana6af2bc2007-02-10 01:42:57 -08005064 unsigned long min_pfn = ULONG_MAX;
Tejun Heoc13291a2011-07-12 10:46:30 +02005065 unsigned long start_pfn;
5066 int i;
Mel Gorman1abbfb42006-11-23 12:01:41 +00005067
Tejun Heoc13291a2011-07-12 10:46:30 +02005068 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
5069 min_pfn = min(min_pfn, start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005070
Mel Gormana6af2bc2007-02-10 01:42:57 -08005071 if (min_pfn == ULONG_MAX) {
5072 printk(KERN_WARNING
Paul Jackson2bc0d262008-06-22 07:22:17 -07005073 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08005074 return 0;
5075 }
5076
5077 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005078}
5079
5080/**
5081 * find_min_pfn_with_active_regions - Find the minimum PFN registered
5082 *
5083 * It returns the minimum PFN based on information provided via
Zhang Zhen7d018172014-06-04 16:10:53 -07005084 * memblock_set_node().
Mel Gormanc7132162006-09-27 01:49:43 -07005085 */
5086unsigned long __init find_min_pfn_with_active_regions(void)
5087{
5088 return find_min_pfn_for_node(MAX_NUMNODES);
5089}
5090
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005091/*
5092 * early_calculate_totalpages()
5093 * Sum pages in active regions for movable zone.
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005094 * Populate N_MEMORY for calculating usable_nodes.
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005095 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07005096static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07005097{
Mel Gorman7e63efe2007-07-17 04:03:15 -07005098 unsigned long totalpages = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02005099 unsigned long start_pfn, end_pfn;
5100 int i, nid;
Mel Gorman7e63efe2007-07-17 04:03:15 -07005101
Tejun Heoc13291a2011-07-12 10:46:30 +02005102 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
5103 unsigned long pages = end_pfn - start_pfn;
5104
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005105 totalpages += pages;
5106 if (pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005107 node_set_state(nid, N_MEMORY);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005108 }
Pintu Kumarb8af2942013-09-11 14:20:34 -07005109 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07005110}
5111
Mel Gorman2a1e2742007-07-17 04:03:12 -07005112/*
5113 * Find the PFN the Movable zone begins in each node. Kernel memory
5114 * is spread evenly between nodes as long as the nodes have enough
5115 * memory. When they don't, some nodes will have more kernelcore than
5116 * others
5117 */
Kautuk Consulb224ef82012-03-21 16:34:15 -07005118static void __init find_zone_movable_pfns_for_nodes(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005119{
5120 int i, nid;
5121 unsigned long usable_startpfn;
5122 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07005123 /* save the state before borrow the nodemask */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005124 nodemask_t saved_node_state = node_states[N_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005125 unsigned long totalpages = early_calculate_totalpages();
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005126 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
Emil Medve136199f2014-04-07 15:37:52 -07005127 struct memblock_region *r;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005128
5129 /* Need to find movable_zone earlier when movable_node is specified. */
5130 find_usable_zone_for_movable();
Mel Gorman2a1e2742007-07-17 04:03:12 -07005131
Mel Gorman7e63efe2007-07-17 04:03:15 -07005132 /*
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005133 * If movable_node is specified, ignore kernelcore and movablecore
5134 * options.
5135 */
5136 if (movable_node_is_enabled()) {
Emil Medve136199f2014-04-07 15:37:52 -07005137 for_each_memblock(memory, r) {
5138 if (!memblock_is_hotpluggable(r))
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005139 continue;
5140
Emil Medve136199f2014-04-07 15:37:52 -07005141 nid = r->nid;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005142
Emil Medve136199f2014-04-07 15:37:52 -07005143 usable_startpfn = PFN_DOWN(r->base);
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005144 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
5145 min(usable_startpfn, zone_movable_pfn[nid]) :
5146 usable_startpfn;
5147 }
5148
5149 goto out2;
5150 }
5151
5152 /*
5153 * If movablecore=nn[KMG] was specified, calculate what size of
Mel Gorman7e63efe2007-07-17 04:03:15 -07005154 * kernelcore that corresponds so that memory usable for
5155 * any allocation type is evenly spread. If both kernelcore
5156 * and movablecore are specified, then the value of kernelcore
5157 * will be used for required_kernelcore if it's greater than
5158 * what movablecore would have allowed.
5159 */
5160 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07005161 unsigned long corepages;
5162
5163 /*
5164 * Round-up so that ZONE_MOVABLE is at least as large as what
5165 * was requested by the user
5166 */
5167 required_movablecore =
5168 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
5169 corepages = totalpages - required_movablecore;
5170
5171 required_kernelcore = max(required_kernelcore, corepages);
5172 }
5173
Yinghai Lu20e69262013-03-01 14:51:27 -08005174 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
5175 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07005176 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005177
5178 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
Mel Gorman2a1e2742007-07-17 04:03:12 -07005179 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
5180
5181restart:
5182 /* Spread kernelcore memory as evenly as possible throughout nodes */
5183 kernelcore_node = required_kernelcore / usable_nodes;
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005184 for_each_node_state(nid, N_MEMORY) {
Tejun Heoc13291a2011-07-12 10:46:30 +02005185 unsigned long start_pfn, end_pfn;
5186
Mel Gorman2a1e2742007-07-17 04:03:12 -07005187 /*
5188 * Recalculate kernelcore_node if the division per node
5189 * now exceeds what is necessary to satisfy the requested
5190 * amount of memory for the kernel
5191 */
5192 if (required_kernelcore < kernelcore_node)
5193 kernelcore_node = required_kernelcore / usable_nodes;
5194
5195 /*
5196 * As the map is walked, we track how much memory is usable
5197 * by the kernel using kernelcore_remaining. When it is
5198 * 0, the rest of the node is usable by ZONE_MOVABLE
5199 */
5200 kernelcore_remaining = kernelcore_node;
5201
5202 /* Go through each range of PFNs within this node */
Tejun Heoc13291a2011-07-12 10:46:30 +02005203 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005204 unsigned long size_pages;
5205
Tejun Heoc13291a2011-07-12 10:46:30 +02005206 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005207 if (start_pfn >= end_pfn)
5208 continue;
5209
5210 /* Account for what is only usable for kernelcore */
5211 if (start_pfn < usable_startpfn) {
5212 unsigned long kernel_pages;
5213 kernel_pages = min(end_pfn, usable_startpfn)
5214 - start_pfn;
5215
5216 kernelcore_remaining -= min(kernel_pages,
5217 kernelcore_remaining);
5218 required_kernelcore -= min(kernel_pages,
5219 required_kernelcore);
5220
5221 /* Continue if range is now fully accounted */
5222 if (end_pfn <= usable_startpfn) {
5223
5224 /*
5225 * Push zone_movable_pfn to the end so
5226 * that if we have to rebalance
5227 * kernelcore across nodes, we will
5228 * not double account here
5229 */
5230 zone_movable_pfn[nid] = end_pfn;
5231 continue;
5232 }
5233 start_pfn = usable_startpfn;
5234 }
5235
5236 /*
5237 * The usable PFN range for ZONE_MOVABLE is from
5238 * start_pfn->end_pfn. Calculate size_pages as the
5239 * number of pages used as kernelcore
5240 */
5241 size_pages = end_pfn - start_pfn;
5242 if (size_pages > kernelcore_remaining)
5243 size_pages = kernelcore_remaining;
5244 zone_movable_pfn[nid] = start_pfn + size_pages;
5245
5246 /*
5247 * Some kernelcore has been met, update counts and
5248 * break if the kernelcore for this node has been
Pintu Kumarb8af2942013-09-11 14:20:34 -07005249 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07005250 */
5251 required_kernelcore -= min(required_kernelcore,
5252 size_pages);
5253 kernelcore_remaining -= size_pages;
5254 if (!kernelcore_remaining)
5255 break;
5256 }
5257 }
5258
5259 /*
5260 * If there is still required_kernelcore, we do another pass with one
5261 * less node in the count. This will push zone_movable_pfn[nid] further
5262 * along on the nodes that still have memory until kernelcore is
Pintu Kumarb8af2942013-09-11 14:20:34 -07005263 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07005264 */
5265 usable_nodes--;
5266 if (usable_nodes && required_kernelcore > usable_nodes)
5267 goto restart;
5268
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005269out2:
Mel Gorman2a1e2742007-07-17 04:03:12 -07005270 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5271 for (nid = 0; nid < MAX_NUMNODES; nid++)
5272 zone_movable_pfn[nid] =
5273 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07005274
Yinghai Lu20e69262013-03-01 14:51:27 -08005275out:
Yinghai Lu66918dc2009-06-30 11:41:37 -07005276 /* restore the node_state */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005277 node_states[N_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005278}
5279
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005280/* Any regular or high memory on that node ? */
5281static void check_for_memory(pg_data_t *pgdat, int nid)
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005282{
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005283 enum zone_type zone_type;
5284
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005285 if (N_MEMORY == N_NORMAL_MEMORY)
5286 return;
5287
5288 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005289 struct zone *zone = &pgdat->node_zones[zone_type];
Xishi Qiub38a8722013-11-12 15:07:20 -08005290 if (populated_zone(zone)) {
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005291 node_set_state(nid, N_HIGH_MEMORY);
5292 if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
5293 zone_type <= ZONE_NORMAL)
5294 node_set_state(nid, N_NORMAL_MEMORY);
Bob Liud0048b02012-01-12 17:19:07 -08005295 break;
5296 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005297 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005298}
5299
Mel Gormanc7132162006-09-27 01:49:43 -07005300/**
5301 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005302 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07005303 *
5304 * This will call free_area_init_node() for each active node in the system.
Zhang Zhen7d018172014-06-04 16:10:53 -07005305 * Using the page ranges provided by memblock_set_node(), the size of each
Mel Gormanc7132162006-09-27 01:49:43 -07005306 * zone in each node and their holes is calculated. If the maximum PFN
5307 * between two adjacent zones match, it is assumed that the zone is empty.
5308 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5309 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5310 * starts where the previous one ended. For example, ZONE_DMA32 starts
5311 * at arch_max_dma_pfn.
5312 */
5313void __init free_area_init_nodes(unsigned long *max_zone_pfn)
5314{
Tejun Heoc13291a2011-07-12 10:46:30 +02005315 unsigned long start_pfn, end_pfn;
5316 int i, nid;
Mel Gormana6af2bc2007-02-10 01:42:57 -08005317
Mel Gormanc7132162006-09-27 01:49:43 -07005318 /* Record where the zone boundaries are */
5319 memset(arch_zone_lowest_possible_pfn, 0,
5320 sizeof(arch_zone_lowest_possible_pfn));
5321 memset(arch_zone_highest_possible_pfn, 0,
5322 sizeof(arch_zone_highest_possible_pfn));
5323 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
5324 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
5325 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005326 if (i == ZONE_MOVABLE)
5327 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07005328 arch_zone_lowest_possible_pfn[i] =
5329 arch_zone_highest_possible_pfn[i-1];
5330 arch_zone_highest_possible_pfn[i] =
5331 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
5332 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07005333 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
5334 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
5335
5336 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5337 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
Kautuk Consulb224ef82012-03-21 16:34:15 -07005338 find_zone_movable_pfns_for_nodes();
Mel Gormanc7132162006-09-27 01:49:43 -07005339
Mel Gormanc7132162006-09-27 01:49:43 -07005340 /* Print out the zone ranges */
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005341 printk("Zone ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005342 for (i = 0; i < MAX_NR_ZONES; i++) {
5343 if (i == ZONE_MOVABLE)
5344 continue;
Kay Sievers155cbfc2012-05-08 17:24:14 +02005345 printk(KERN_CONT " %-8s ", zone_names[i]);
David Rientjes72f0ba02010-03-05 13:42:14 -08005346 if (arch_zone_lowest_possible_pfn[i] ==
5347 arch_zone_highest_possible_pfn[i])
Kay Sievers155cbfc2012-05-08 17:24:14 +02005348 printk(KERN_CONT "empty\n");
David Rientjes72f0ba02010-03-05 13:42:14 -08005349 else
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005350 printk(KERN_CONT "[mem %0#10lx-%0#10lx]\n",
5351 arch_zone_lowest_possible_pfn[i] << PAGE_SHIFT,
5352 (arch_zone_highest_possible_pfn[i]
5353 << PAGE_SHIFT) - 1);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005354 }
5355
5356 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005357 printk("Movable zone start for each node\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005358 for (i = 0; i < MAX_NUMNODES; i++) {
5359 if (zone_movable_pfn[i])
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005360 printk(" Node %d: %#010lx\n", i,
5361 zone_movable_pfn[i] << PAGE_SHIFT);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005362 }
Mel Gormanc7132162006-09-27 01:49:43 -07005363
Wanpeng Lif2d52fe2012-10-08 16:32:24 -07005364 /* Print out the early node map */
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005365 printk("Early memory node ranges\n");
Tejun Heoc13291a2011-07-12 10:46:30 +02005366 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005367 printk(" node %3d: [mem %#010lx-%#010lx]\n", nid,
5368 start_pfn << PAGE_SHIFT, (end_pfn << PAGE_SHIFT) - 1);
Mel Gormanc7132162006-09-27 01:49:43 -07005369
5370 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07005371 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08005372 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07005373 for_each_online_node(nid) {
5374 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07005375 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07005376 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005377
5378 /* Any memory on that node */
5379 if (pgdat->node_present_pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005380 node_set_state(nid, N_MEMORY);
5381 check_for_memory(pgdat, nid);
Mel Gormanc7132162006-09-27 01:49:43 -07005382 }
5383}
Mel Gorman2a1e2742007-07-17 04:03:12 -07005384
Mel Gorman7e63efe2007-07-17 04:03:15 -07005385static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005386{
5387 unsigned long long coremem;
5388 if (!p)
5389 return -EINVAL;
5390
5391 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005392 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005393
Mel Gorman7e63efe2007-07-17 04:03:15 -07005394 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07005395 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
5396
5397 return 0;
5398}
Mel Gormaned7ed362007-07-17 04:03:14 -07005399
Mel Gorman7e63efe2007-07-17 04:03:15 -07005400/*
5401 * kernelcore=size sets the amount of memory for use for allocations that
5402 * cannot be reclaimed or migrated.
5403 */
5404static int __init cmdline_parse_kernelcore(char *p)
5405{
5406 return cmdline_parse_core(p, &required_kernelcore);
5407}
5408
5409/*
5410 * movablecore=size sets the amount of memory for use for allocations that
5411 * can be reclaimed or migrated.
5412 */
5413static int __init cmdline_parse_movablecore(char *p)
5414{
5415 return cmdline_parse_core(p, &required_movablecore);
5416}
5417
Mel Gormaned7ed362007-07-17 04:03:14 -07005418early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005419early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07005420
Tejun Heo0ee332c2011-12-08 10:22:09 -08005421#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005422
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005423void adjust_managed_page_count(struct page *page, long count)
5424{
5425 spin_lock(&managed_page_count_lock);
5426 page_zone(page)->managed_pages += count;
5427 totalram_pages += count;
Jiang Liu3dcc0572013-07-03 15:03:21 -07005428#ifdef CONFIG_HIGHMEM
5429 if (PageHighMem(page))
5430 totalhigh_pages += count;
5431#endif
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005432 spin_unlock(&managed_page_count_lock);
5433}
Jiang Liu3dcc0572013-07-03 15:03:21 -07005434EXPORT_SYMBOL(adjust_managed_page_count);
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005435
Jiang Liu11199692013-07-03 15:02:48 -07005436unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
Jiang Liu69afade2013-04-29 15:06:21 -07005437{
Jiang Liu11199692013-07-03 15:02:48 -07005438 void *pos;
5439 unsigned long pages = 0;
Jiang Liu69afade2013-04-29 15:06:21 -07005440
Jiang Liu11199692013-07-03 15:02:48 -07005441 start = (void *)PAGE_ALIGN((unsigned long)start);
5442 end = (void *)((unsigned long)end & PAGE_MASK);
5443 for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
Jiang Liudbe67df2013-07-03 15:02:51 -07005444 if ((unsigned int)poison <= 0xFF)
Jiang Liu11199692013-07-03 15:02:48 -07005445 memset(pos, poison, PAGE_SIZE);
5446 free_reserved_page(virt_to_page(pos));
Jiang Liu69afade2013-04-29 15:06:21 -07005447 }
5448
5449 if (pages && s)
Jiang Liu11199692013-07-03 15:02:48 -07005450 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
Jiang Liu69afade2013-04-29 15:06:21 -07005451 s, pages << (PAGE_SHIFT - 10), start, end);
5452
5453 return pages;
5454}
Jiang Liu11199692013-07-03 15:02:48 -07005455EXPORT_SYMBOL(free_reserved_area);
Jiang Liu69afade2013-04-29 15:06:21 -07005456
Jiang Liucfa11e02013-04-29 15:07:00 -07005457#ifdef CONFIG_HIGHMEM
5458void free_highmem_page(struct page *page)
5459{
5460 __free_reserved_page(page);
5461 totalram_pages++;
Jiang Liu7b4b2a02013-07-03 15:03:11 -07005462 page_zone(page)->managed_pages++;
Jiang Liucfa11e02013-04-29 15:07:00 -07005463 totalhigh_pages++;
5464}
5465#endif
5466
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005467
5468void __init mem_init_print_info(const char *str)
5469{
5470 unsigned long physpages, codesize, datasize, rosize, bss_size;
5471 unsigned long init_code_size, init_data_size;
5472
5473 physpages = get_num_physpages();
5474 codesize = _etext - _stext;
5475 datasize = _edata - _sdata;
5476 rosize = __end_rodata - __start_rodata;
5477 bss_size = __bss_stop - __bss_start;
5478 init_data_size = __init_end - __init_begin;
5479 init_code_size = _einittext - _sinittext;
5480
5481 /*
5482 * Detect special cases and adjust section sizes accordingly:
5483 * 1) .init.* may be embedded into .data sections
5484 * 2) .init.text.* may be out of [__init_begin, __init_end],
5485 * please refer to arch/tile/kernel/vmlinux.lds.S.
5486 * 3) .rodata.* may be embedded into .text or .data sections.
5487 */
5488#define adj_init_size(start, end, size, pos, adj) \
Pintu Kumarb8af2942013-09-11 14:20:34 -07005489 do { \
5490 if (start <= pos && pos < end && size > adj) \
5491 size -= adj; \
5492 } while (0)
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005493
5494 adj_init_size(__init_begin, __init_end, init_data_size,
5495 _sinittext, init_code_size);
5496 adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
5497 adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
5498 adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
5499 adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
5500
5501#undef adj_init_size
5502
5503 printk("Memory: %luK/%luK available "
5504 "(%luK kernel code, %luK rwdata, %luK rodata, "
5505 "%luK init, %luK bss, %luK reserved"
5506#ifdef CONFIG_HIGHMEM
5507 ", %luK highmem"
5508#endif
5509 "%s%s)\n",
5510 nr_free_pages() << (PAGE_SHIFT-10), physpages << (PAGE_SHIFT-10),
5511 codesize >> 10, datasize >> 10, rosize >> 10,
5512 (init_data_size + init_code_size) >> 10, bss_size >> 10,
5513 (physpages - totalram_pages) << (PAGE_SHIFT-10),
5514#ifdef CONFIG_HIGHMEM
5515 totalhigh_pages << (PAGE_SHIFT-10),
5516#endif
5517 str ? ", " : "", str ? str : "");
5518}
5519
Mel Gorman0e0b8642006-09-27 01:49:56 -07005520/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005521 * set_dma_reserve - set the specified number of pages reserved in the first zone
5522 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07005523 *
5524 * The per-cpu batchsize and zone watermarks are determined by present_pages.
5525 * In the DMA zone, a significant percentage may be consumed by kernel image
5526 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005527 * function may optionally be used to account for unfreeable pages in the
5528 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5529 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07005530 */
5531void __init set_dma_reserve(unsigned long new_dma_reserve)
5532{
5533 dma_reserve = new_dma_reserve;
5534}
5535
Linus Torvalds1da177e2005-04-16 15:20:36 -07005536void __init free_area_init(unsigned long *zones_size)
5537{
Johannes Weiner9109fb72008-07-23 21:27:20 -07005538 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005539 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
5540}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005541
Linus Torvalds1da177e2005-04-16 15:20:36 -07005542static int page_alloc_cpu_notify(struct notifier_block *self,
5543 unsigned long action, void *hcpu)
5544{
5545 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005546
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07005547 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -07005548 lru_add_drain_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005549 drain_pages(cpu);
5550
5551 /*
5552 * Spill the event counters of the dead processor
5553 * into the current processors event counters.
5554 * This artificially elevates the count of the current
5555 * processor.
5556 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07005557 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005558
5559 /*
5560 * Zero the differential counters of the dead processor
5561 * so that the vm statistics are consistent.
5562 *
5563 * This is only okay since the processor is dead and cannot
5564 * race with what we are doing.
5565 */
Christoph Lameter2bb921e2013-09-11 14:21:30 -07005566 cpu_vm_stats_fold(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005567 }
5568 return NOTIFY_OK;
5569}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005570
5571void __init page_alloc_init(void)
5572{
5573 hotcpu_notifier(page_alloc_cpu_notify, 0);
5574}
5575
5576/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005577 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5578 * or min_free_kbytes changes.
5579 */
5580static void calculate_totalreserve_pages(void)
5581{
5582 struct pglist_data *pgdat;
5583 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005584 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005585
5586 for_each_online_pgdat(pgdat) {
5587 for (i = 0; i < MAX_NR_ZONES; i++) {
5588 struct zone *zone = pgdat->node_zones + i;
Mel Gorman3484b2d2014-08-06 16:07:14 -07005589 long max = 0;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005590
5591 /* Find valid and maximum lowmem_reserve in the zone */
5592 for (j = i; j < MAX_NR_ZONES; j++) {
5593 if (zone->lowmem_reserve[j] > max)
5594 max = zone->lowmem_reserve[j];
5595 }
5596
Mel Gorman41858962009-06-16 15:32:12 -07005597 /* we treat the high watermark as reserved pages. */
5598 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005599
Jiang Liub40da042013-02-22 16:33:52 -08005600 if (max > zone->managed_pages)
5601 max = zone->managed_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005602 reserve_pages += max;
Johannes Weinerab8fabd2012-01-10 15:07:42 -08005603 /*
5604 * Lowmem reserves are not available to
5605 * GFP_HIGHUSER page cache allocations and
5606 * kswapd tries to balance zones to their high
5607 * watermark. As a result, neither should be
5608 * regarded as dirtyable memory, to prevent a
5609 * situation where reclaim has to clean pages
5610 * in order to balance the zones.
5611 */
5612 zone->dirty_balance_reserve = max;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005613 }
5614 }
Johannes Weinerab8fabd2012-01-10 15:07:42 -08005615 dirty_balance_reserve = reserve_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005616 totalreserve_pages = reserve_pages;
5617}
5618
5619/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07005620 * setup_per_zone_lowmem_reserve - called whenever
5621 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
5622 * has a correct pages reserved value, so an adequate number of
5623 * pages are left in the zone after a successful __alloc_pages().
5624 */
5625static void setup_per_zone_lowmem_reserve(void)
5626{
5627 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005628 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005629
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08005630 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005631 for (j = 0; j < MAX_NR_ZONES; j++) {
5632 struct zone *zone = pgdat->node_zones + j;
Jiang Liub40da042013-02-22 16:33:52 -08005633 unsigned long managed_pages = zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005634
5635 zone->lowmem_reserve[j] = 0;
5636
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005637 idx = j;
5638 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005639 struct zone *lower_zone;
5640
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005641 idx--;
5642
Linus Torvalds1da177e2005-04-16 15:20:36 -07005643 if (sysctl_lowmem_reserve_ratio[idx] < 1)
5644 sysctl_lowmem_reserve_ratio[idx] = 1;
5645
5646 lower_zone = pgdat->node_zones + idx;
Jiang Liub40da042013-02-22 16:33:52 -08005647 lower_zone->lowmem_reserve[j] = managed_pages /
Linus Torvalds1da177e2005-04-16 15:20:36 -07005648 sysctl_lowmem_reserve_ratio[idx];
Jiang Liub40da042013-02-22 16:33:52 -08005649 managed_pages += lower_zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005650 }
5651 }
5652 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005653
5654 /* update totalreserve_pages */
5655 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005656}
5657
Mel Gormancfd3da12011-04-25 21:36:42 +00005658static void __setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005659{
5660 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
5661 unsigned long lowmem_pages = 0;
5662 struct zone *zone;
5663 unsigned long flags;
5664
5665 /* Calculate total number of !ZONE_HIGHMEM pages */
5666 for_each_zone(zone) {
5667 if (!is_highmem(zone))
Jiang Liub40da042013-02-22 16:33:52 -08005668 lowmem_pages += zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005669 }
5670
5671 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07005672 u64 tmp;
5673
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005674 spin_lock_irqsave(&zone->lock, flags);
Jiang Liub40da042013-02-22 16:33:52 -08005675 tmp = (u64)pages_min * zone->managed_pages;
Andrew Mortonac924c62006-05-15 09:43:59 -07005676 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005677 if (is_highmem(zone)) {
5678 /*
Nick Piggin669ed172005-11-13 16:06:45 -08005679 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5680 * need highmem pages, so cap pages_min to a small
5681 * value here.
5682 *
Mel Gorman41858962009-06-16 15:32:12 -07005683 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Nick Piggin669ed172005-11-13 16:06:45 -08005684 * deltas controls asynch page reclaim, and so should
5685 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005686 */
Andrew Morton90ae8d62013-02-22 16:32:22 -08005687 unsigned long min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005688
Jiang Liub40da042013-02-22 16:33:52 -08005689 min_pages = zone->managed_pages / 1024;
Andrew Morton90ae8d62013-02-22 16:32:22 -08005690 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
Mel Gorman41858962009-06-16 15:32:12 -07005691 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005692 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08005693 /*
5694 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07005695 * proportionate to the zone's size.
5696 */
Mel Gorman41858962009-06-16 15:32:12 -07005697 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005698 }
5699
Mel Gorman41858962009-06-16 15:32:12 -07005700 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
5701 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Marek Szyprowski49f223a2012-01-25 12:49:24 +01005702
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07005703 __mod_zone_page_state(zone, NR_ALLOC_BATCH,
5704 high_wmark_pages(zone) -
5705 low_wmark_pages(zone) -
5706 zone_page_state(zone, NR_ALLOC_BATCH));
5707
Mel Gorman56fd56b2007-10-16 01:25:58 -07005708 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005709 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005710 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005711
5712 /* update totalreserve_pages */
5713 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005714}
5715
Mel Gormancfd3da12011-04-25 21:36:42 +00005716/**
5717 * setup_per_zone_wmarks - called when min_free_kbytes changes
5718 * or when memory is hot-{added|removed}
5719 *
5720 * Ensures that the watermark[min,low,high] values for each zone are set
5721 * correctly with respect to min_free_kbytes.
5722 */
5723void setup_per_zone_wmarks(void)
5724{
5725 mutex_lock(&zonelists_mutex);
5726 __setup_per_zone_wmarks();
5727 mutex_unlock(&zonelists_mutex);
5728}
5729
Randy Dunlap55a44622009-09-21 17:01:20 -07005730/*
Rik van Riel556adec2008-10-18 20:26:34 -07005731 * The inactive anon list should be small enough that the VM never has to
5732 * do too much work, but large enough that each inactive page has a chance
5733 * to be referenced again before it is swapped out.
5734 *
5735 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5736 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5737 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5738 * the anonymous pages are kept on the inactive list.
5739 *
5740 * total target max
5741 * memory ratio inactive anon
5742 * -------------------------------------
5743 * 10MB 1 5MB
5744 * 100MB 1 50MB
5745 * 1GB 3 250MB
5746 * 10GB 10 0.9GB
5747 * 100GB 31 3GB
5748 * 1TB 101 10GB
5749 * 10TB 320 32GB
5750 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005751static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
Minchan Kim96cb4df2009-06-16 15:32:49 -07005752{
5753 unsigned int gb, ratio;
5754
5755 /* Zone size in gigabytes */
Jiang Liub40da042013-02-22 16:33:52 -08005756 gb = zone->managed_pages >> (30 - PAGE_SHIFT);
Minchan Kim96cb4df2009-06-16 15:32:49 -07005757 if (gb)
5758 ratio = int_sqrt(10 * gb);
5759 else
5760 ratio = 1;
5761
5762 zone->inactive_ratio = ratio;
5763}
5764
KOSAKI Motohiro839a4fc2011-05-24 17:11:31 -07005765static void __meminit setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07005766{
5767 struct zone *zone;
5768
Minchan Kim96cb4df2009-06-16 15:32:49 -07005769 for_each_zone(zone)
5770 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07005771}
5772
Linus Torvalds1da177e2005-04-16 15:20:36 -07005773/*
5774 * Initialise min_free_kbytes.
5775 *
5776 * For small machines we want it small (128k min). For large machines
5777 * we want it large (64MB max). But it is not linear, because network
5778 * bandwidth does not increase linearly with machine size. We use
5779 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07005780 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
Linus Torvalds1da177e2005-04-16 15:20:36 -07005781 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5782 *
5783 * which yields
5784 *
5785 * 16MB: 512k
5786 * 32MB: 724k
5787 * 64MB: 1024k
5788 * 128MB: 1448k
5789 * 256MB: 2048k
5790 * 512MB: 2896k
5791 * 1024MB: 4096k
5792 * 2048MB: 5792k
5793 * 4096MB: 8192k
5794 * 8192MB: 11584k
5795 * 16384MB: 16384k
5796 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005797int __meminit init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005798{
5799 unsigned long lowmem_kbytes;
Michal Hocko5f127332013-07-08 16:00:40 -07005800 int new_min_free_kbytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005801
5802 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
Michal Hocko5f127332013-07-08 16:00:40 -07005803 new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005804
Michal Hocko5f127332013-07-08 16:00:40 -07005805 if (new_min_free_kbytes > user_min_free_kbytes) {
5806 min_free_kbytes = new_min_free_kbytes;
5807 if (min_free_kbytes < 128)
5808 min_free_kbytes = 128;
5809 if (min_free_kbytes > 65536)
5810 min_free_kbytes = 65536;
5811 } else {
5812 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
5813 new_min_free_kbytes, user_min_free_kbytes);
5814 }
Minchan Kimbc75d332009-06-16 15:32:48 -07005815 setup_per_zone_wmarks();
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -07005816 refresh_zone_stat_thresholds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005817 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07005818 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005819 return 0;
5820}
Minchan Kimbc75d332009-06-16 15:32:48 -07005821module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005822
5823/*
Pintu Kumarb8af2942013-09-11 14:20:34 -07005824 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
Linus Torvalds1da177e2005-04-16 15:20:36 -07005825 * that we can call two helper functions whenever min_free_kbytes
5826 * changes.
5827 */
Joe Perchescccad5b2014-06-06 14:38:09 -07005828int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005829 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005830{
Han Pingtianda8c7572014-01-23 15:53:17 -08005831 int rc;
5832
5833 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5834 if (rc)
5835 return rc;
5836
Michal Hocko5f127332013-07-08 16:00:40 -07005837 if (write) {
5838 user_min_free_kbytes = min_free_kbytes;
Minchan Kimbc75d332009-06-16 15:32:48 -07005839 setup_per_zone_wmarks();
Michal Hocko5f127332013-07-08 16:00:40 -07005840 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005841 return 0;
5842}
5843
Christoph Lameter96146342006-07-03 00:24:13 -07005844#ifdef CONFIG_NUMA
Joe Perchescccad5b2014-06-06 14:38:09 -07005845int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005846 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07005847{
5848 struct zone *zone;
5849 int rc;
5850
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005851 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07005852 if (rc)
5853 return rc;
5854
5855 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08005856 zone->min_unmapped_pages = (zone->managed_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07005857 sysctl_min_unmapped_ratio) / 100;
5858 return 0;
5859}
Christoph Lameter0ff38492006-09-25 23:31:52 -07005860
Joe Perchescccad5b2014-06-06 14:38:09 -07005861int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005862 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07005863{
5864 struct zone *zone;
5865 int rc;
5866
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005867 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07005868 if (rc)
5869 return rc;
5870
5871 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08005872 zone->min_slab_pages = (zone->managed_pages *
Christoph Lameter0ff38492006-09-25 23:31:52 -07005873 sysctl_min_slab_ratio) / 100;
5874 return 0;
5875}
Christoph Lameter96146342006-07-03 00:24:13 -07005876#endif
5877
Linus Torvalds1da177e2005-04-16 15:20:36 -07005878/*
5879 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5880 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5881 * whenever sysctl_lowmem_reserve_ratio changes.
5882 *
5883 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07005884 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07005885 * if in function of the boot time zone sizes.
5886 */
Joe Perchescccad5b2014-06-06 14:38:09 -07005887int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005888 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005889{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005890 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005891 setup_per_zone_lowmem_reserve();
5892 return 0;
5893}
5894
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005895/*
5896 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
Pintu Kumarb8af2942013-09-11 14:20:34 -07005897 * cpu. It is the fraction of total pages in each zone that a hot per cpu
5898 * pagelist can have before it gets flushed back to buddy allocator.
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005899 */
Joe Perchescccad5b2014-06-06 14:38:09 -07005900int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005901 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005902{
5903 struct zone *zone;
David Rientjes7cd2b0a2014-06-23 13:22:04 -07005904 int old_percpu_pagelist_fraction;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005905 int ret;
5906
Cody P Schaferc8e251f2013-07-03 15:01:29 -07005907 mutex_lock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07005908 old_percpu_pagelist_fraction = percpu_pagelist_fraction;
5909
5910 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
5911 if (!write || ret < 0)
5912 goto out;
5913
5914 /* Sanity checking to avoid pcp imbalance */
5915 if (percpu_pagelist_fraction &&
5916 percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
5917 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
5918 ret = -EINVAL;
5919 goto out;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005920 }
David Rientjes7cd2b0a2014-06-23 13:22:04 -07005921
5922 /* No change? */
5923 if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
5924 goto out;
5925
5926 for_each_populated_zone(zone) {
5927 unsigned int cpu;
5928
5929 for_each_possible_cpu(cpu)
5930 pageset_set_high_and_batch(zone,
5931 per_cpu_ptr(zone->pageset, cpu));
5932 }
5933out:
Cody P Schaferc8e251f2013-07-03 15:01:29 -07005934 mutex_unlock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07005935 return ret;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005936}
5937
David S. Millerf034b5d2006-08-24 03:08:07 -07005938int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005939
5940#ifdef CONFIG_NUMA
5941static int __init set_hashdist(char *str)
5942{
5943 if (!str)
5944 return 0;
5945 hashdist = simple_strtoul(str, &str, 0);
5946 return 1;
5947}
5948__setup("hashdist=", set_hashdist);
5949#endif
5950
5951/*
5952 * allocate a large system hash table from bootmem
5953 * - it is assumed that the hash table must contain an exact power-of-2
5954 * quantity of entries
5955 * - limit is the number of hash buckets, not the total allocation size
5956 */
5957void *__init alloc_large_system_hash(const char *tablename,
5958 unsigned long bucketsize,
5959 unsigned long numentries,
5960 int scale,
5961 int flags,
5962 unsigned int *_hash_shift,
5963 unsigned int *_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00005964 unsigned long low_limit,
5965 unsigned long high_limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005966{
Tim Bird31fe62b2012-05-23 13:33:35 +00005967 unsigned long long max = high_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005968 unsigned long log2qty, size;
5969 void *table = NULL;
5970
5971 /* allow the kernel cmdline to have a say */
5972 if (!numentries) {
5973 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08005974 numentries = nr_kernel_pages;
Jerry Zhoua7e83312013-09-11 14:20:26 -07005975
5976 /* It isn't necessary when PAGE_SIZE >= 1MB */
5977 if (PAGE_SHIFT < 20)
5978 numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005979
5980 /* limit to 1 bucket per 2^scale bytes of low memory */
5981 if (scale > PAGE_SHIFT)
5982 numentries >>= (scale - PAGE_SHIFT);
5983 else
5984 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08005985
5986 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07005987 if (unlikely(flags & HASH_SMALL)) {
5988 /* Makes no sense without HASH_EARLY */
5989 WARN_ON(!(flags & HASH_EARLY));
5990 if (!(numentries >> *_hash_shift)) {
5991 numentries = 1UL << *_hash_shift;
5992 BUG_ON(!numentries);
5993 }
5994 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08005995 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005996 }
John Hawkes6e692ed2006-03-25 03:08:02 -08005997 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005998
5999 /* limit allocation size to 1/16 total memory by default */
6000 if (max == 0) {
6001 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
6002 do_div(max, bucketsize);
6003 }
Dimitri Sivanich074b8512012-02-08 12:39:07 -08006004 max = min(max, 0x80000000ULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006005
Tim Bird31fe62b2012-05-23 13:33:35 +00006006 if (numentries < low_limit)
6007 numentries = low_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006008 if (numentries > max)
6009 numentries = max;
6010
David Howellsf0d1b0b2006-12-08 02:37:49 -08006011 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006012
6013 do {
6014 size = bucketsize << log2qty;
6015 if (flags & HASH_EARLY)
Santosh Shilimkar67828322014-01-21 15:50:25 -08006016 table = memblock_virt_alloc_nopanic(size, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006017 else if (hashdist)
6018 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
6019 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07006020 /*
6021 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07006022 * some pages at the end of hash table which
6023 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07006024 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01006025 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07006026 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01006027 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
6028 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006029 }
6030 } while (!table && size > PAGE_SIZE && --log2qty);
6031
6032 if (!table)
6033 panic("Failed to allocate %s hash table\n", tablename);
6034
Robin Holtf241e6602010-10-07 12:59:26 -07006035 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07006036 tablename,
Robin Holtf241e6602010-10-07 12:59:26 -07006037 (1UL << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08006038 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07006039 size);
6040
6041 if (_hash_shift)
6042 *_hash_shift = log2qty;
6043 if (_hash_mask)
6044 *_hash_mask = (1 << log2qty) - 1;
6045
6046 return table;
6047}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08006048
Mel Gorman835c1342007-10-16 01:25:47 -07006049/* Return a pointer to the bitmap storing bits affecting a block of pages */
6050static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
6051 unsigned long pfn)
6052{
6053#ifdef CONFIG_SPARSEMEM
6054 return __pfn_to_section(pfn)->pageblock_flags;
6055#else
6056 return zone->pageblock_flags;
6057#endif /* CONFIG_SPARSEMEM */
6058}
Andrew Morton6220ec72006-10-19 23:29:05 -07006059
Mel Gorman835c1342007-10-16 01:25:47 -07006060static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
6061{
6062#ifdef CONFIG_SPARSEMEM
6063 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07006064 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07006065#else
Laura Abbottc060f942013-01-11 14:31:51 -08006066 pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
Mel Gormand9c23402007-10-16 01:26:01 -07006067 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07006068#endif /* CONFIG_SPARSEMEM */
6069}
6070
6071/**
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006072 * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
Mel Gorman835c1342007-10-16 01:25:47 -07006073 * @page: The page within the block of interest
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006074 * @pfn: The target page frame number
6075 * @end_bitidx: The last bit of interest to retrieve
6076 * @mask: mask of bits that the caller is interested in
6077 *
6078 * Return: pageblock_bits flags
Mel Gorman835c1342007-10-16 01:25:47 -07006079 */
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006080unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
Mel Gormane58469b2014-06-04 16:10:16 -07006081 unsigned long end_bitidx,
6082 unsigned long mask)
Mel Gorman835c1342007-10-16 01:25:47 -07006083{
6084 struct zone *zone;
6085 unsigned long *bitmap;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006086 unsigned long bitidx, word_bitidx;
Mel Gormane58469b2014-06-04 16:10:16 -07006087 unsigned long word;
Mel Gorman835c1342007-10-16 01:25:47 -07006088
6089 zone = page_zone(page);
Mel Gorman835c1342007-10-16 01:25:47 -07006090 bitmap = get_pageblock_bitmap(zone, pfn);
6091 bitidx = pfn_to_bitidx(zone, pfn);
Mel Gormane58469b2014-06-04 16:10:16 -07006092 word_bitidx = bitidx / BITS_PER_LONG;
6093 bitidx &= (BITS_PER_LONG-1);
Mel Gorman835c1342007-10-16 01:25:47 -07006094
Mel Gormane58469b2014-06-04 16:10:16 -07006095 word = bitmap[word_bitidx];
6096 bitidx += end_bitidx;
6097 return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
Mel Gorman835c1342007-10-16 01:25:47 -07006098}
6099
6100/**
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006101 * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
Mel Gorman835c1342007-10-16 01:25:47 -07006102 * @page: The page within the block of interest
Mel Gorman835c1342007-10-16 01:25:47 -07006103 * @flags: The flags to set
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006104 * @pfn: The target page frame number
6105 * @end_bitidx: The last bit of interest
6106 * @mask: mask of bits that the caller is interested in
Mel Gorman835c1342007-10-16 01:25:47 -07006107 */
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006108void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
6109 unsigned long pfn,
Mel Gormane58469b2014-06-04 16:10:16 -07006110 unsigned long end_bitidx,
6111 unsigned long mask)
Mel Gorman835c1342007-10-16 01:25:47 -07006112{
6113 struct zone *zone;
6114 unsigned long *bitmap;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006115 unsigned long bitidx, word_bitidx;
Mel Gormane58469b2014-06-04 16:10:16 -07006116 unsigned long old_word, word;
6117
6118 BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
Mel Gorman835c1342007-10-16 01:25:47 -07006119
6120 zone = page_zone(page);
Mel Gorman835c1342007-10-16 01:25:47 -07006121 bitmap = get_pageblock_bitmap(zone, pfn);
6122 bitidx = pfn_to_bitidx(zone, pfn);
Mel Gormane58469b2014-06-04 16:10:16 -07006123 word_bitidx = bitidx / BITS_PER_LONG;
6124 bitidx &= (BITS_PER_LONG-1);
6125
Sasha Levin309381fea2014-01-23 15:52:54 -08006126 VM_BUG_ON_PAGE(!zone_spans_pfn(zone, pfn), page);
Mel Gorman835c1342007-10-16 01:25:47 -07006127
Mel Gormane58469b2014-06-04 16:10:16 -07006128 bitidx += end_bitidx;
6129 mask <<= (BITS_PER_LONG - bitidx - 1);
6130 flags <<= (BITS_PER_LONG - bitidx - 1);
6131
6132 word = ACCESS_ONCE(bitmap[word_bitidx]);
6133 for (;;) {
6134 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
6135 if (word == old_word)
6136 break;
6137 word = old_word;
6138 }
Mel Gorman835c1342007-10-16 01:25:47 -07006139}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07006140
6141/*
Minchan Kim80934512012-07-31 16:43:01 -07006142 * This function checks whether pageblock includes unmovable pages or not.
6143 * If @count is not zero, it is okay to include less @count unmovable pages
6144 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07006145 * PageLRU check without isolation or lru_lock could race so that
Minchan Kim80934512012-07-31 16:43:01 -07006146 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6147 * expect this function should be exact.
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07006148 */
Wen Congyangb023f462012-12-11 16:00:45 -08006149bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
6150 bool skip_hwpoisoned_pages)
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006151{
6152 unsigned long pfn, iter, found;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01006153 int mt;
6154
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006155 /*
6156 * For avoiding noise data, lru_add_drain_all() should be called
Minchan Kim80934512012-07-31 16:43:01 -07006157 * If ZONE_MOVABLE, the zone never contains unmovable pages
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006158 */
6159 if (zone_idx(zone) == ZONE_MOVABLE)
Minchan Kim80934512012-07-31 16:43:01 -07006160 return false;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01006161 mt = get_pageblock_migratetype(page);
6162 if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
Minchan Kim80934512012-07-31 16:43:01 -07006163 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006164
6165 pfn = page_to_pfn(page);
6166 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
6167 unsigned long check = pfn + iter;
6168
Namhyung Kim29723fc2011-02-25 14:44:25 -08006169 if (!pfn_valid_within(check))
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006170 continue;
Namhyung Kim29723fc2011-02-25 14:44:25 -08006171
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006172 page = pfn_to_page(check);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07006173
6174 /*
6175 * Hugepages are not in LRU lists, but they're movable.
6176 * We need not scan over tail pages bacause we don't
6177 * handle each tail page individually in migration.
6178 */
6179 if (PageHuge(page)) {
6180 iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
6181 continue;
6182 }
6183
Minchan Kim97d255c2012-07-31 16:42:59 -07006184 /*
6185 * We can't use page_count without pin a page
6186 * because another CPU can free compound page.
6187 * This check already skips compound tails of THP
6188 * because their page->_count is zero at all time.
6189 */
6190 if (!atomic_read(&page->_count)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006191 if (PageBuddy(page))
6192 iter += (1 << page_order(page)) - 1;
6193 continue;
6194 }
Minchan Kim97d255c2012-07-31 16:42:59 -07006195
Wen Congyangb023f462012-12-11 16:00:45 -08006196 /*
6197 * The HWPoisoned page may be not in buddy system, and
6198 * page_count() is not 0.
6199 */
6200 if (skip_hwpoisoned_pages && PageHWPoison(page))
6201 continue;
6202
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006203 if (!PageLRU(page))
6204 found++;
6205 /*
6206 * If there are RECLAIMABLE pages, we need to check it.
6207 * But now, memory offline itself doesn't call shrink_slab()
6208 * and it still to be fixed.
6209 */
6210 /*
6211 * If the page is not RAM, page_count()should be 0.
6212 * we don't need more check. This is an _used_ not-movable page.
6213 *
6214 * The problematic thing here is PG_reserved pages. PG_reserved
6215 * is set to both of a memory hole page and a _used_ kernel
6216 * page at boot.
6217 */
6218 if (found > count)
Minchan Kim80934512012-07-31 16:43:01 -07006219 return true;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006220 }
Minchan Kim80934512012-07-31 16:43:01 -07006221 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006222}
6223
6224bool is_pageblock_removable_nolock(struct page *page)
6225{
Michal Hocko656a0702012-01-20 14:33:58 -08006226 struct zone *zone;
6227 unsigned long pfn;
Michal Hocko687875f2012-01-20 14:33:55 -08006228
6229 /*
6230 * We have to be careful here because we are iterating over memory
6231 * sections which are not zone aware so we might end up outside of
6232 * the zone but still within the section.
Michal Hocko656a0702012-01-20 14:33:58 -08006233 * We have to take care about the node as well. If the node is offline
6234 * its NODE_DATA will be NULL - see page_zone.
Michal Hocko687875f2012-01-20 14:33:55 -08006235 */
Michal Hocko656a0702012-01-20 14:33:58 -08006236 if (!node_online(page_to_nid(page)))
6237 return false;
6238
6239 zone = page_zone(page);
6240 pfn = page_to_pfn(page);
Cody P Schafer108bcc92013-02-22 16:35:23 -08006241 if (!zone_spans_pfn(zone, pfn))
Michal Hocko687875f2012-01-20 14:33:55 -08006242 return false;
6243
Wen Congyangb023f462012-12-11 16:00:45 -08006244 return !has_unmovable_pages(zone, page, 0, true);
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07006245}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006246
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006247#ifdef CONFIG_CMA
6248
6249static unsigned long pfn_max_align_down(unsigned long pfn)
6250{
6251 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
6252 pageblock_nr_pages) - 1);
6253}
6254
6255static unsigned long pfn_max_align_up(unsigned long pfn)
6256{
6257 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
6258 pageblock_nr_pages));
6259}
6260
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006261/* [start, end) must belong to a single zone. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006262static int __alloc_contig_migrate_range(struct compact_control *cc,
6263 unsigned long start, unsigned long end)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006264{
6265 /* This function is based on compact_zone() from compaction.c. */
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006266 unsigned long nr_reclaimed;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006267 unsigned long pfn = start;
6268 unsigned int tries = 0;
6269 int ret = 0;
6270
Marek Szyprowskibe49a6e2012-12-12 13:51:19 -08006271 migrate_prep();
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006272
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006273 while (pfn < end || !list_empty(&cc->migratepages)) {
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006274 if (fatal_signal_pending(current)) {
6275 ret = -EINTR;
6276 break;
6277 }
6278
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006279 if (list_empty(&cc->migratepages)) {
6280 cc->nr_migratepages = 0;
6281 pfn = isolate_migratepages_range(cc->zone, cc,
Minchan Kime46a2872012-10-08 16:33:48 -07006282 pfn, end, true);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006283 if (!pfn) {
6284 ret = -EINTR;
6285 break;
6286 }
6287 tries = 0;
6288 } else if (++tries == 5) {
6289 ret = ret < 0 ? ret : -EBUSY;
6290 break;
6291 }
6292
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006293 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
6294 &cc->migratepages);
6295 cc->nr_migratepages -= nr_reclaimed;
Minchan Kim02c6de82012-10-08 16:31:55 -07006296
Hugh Dickins9c620e22013-02-22 16:35:14 -08006297 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
David Rientjese0b9dae2014-06-04 16:08:28 -07006298 NULL, 0, cc->mode, MR_CMA);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006299 }
Srinivas Pandruvada2a6f5122013-02-22 16:32:09 -08006300 if (ret < 0) {
6301 putback_movable_pages(&cc->migratepages);
6302 return ret;
6303 }
6304 return 0;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006305}
6306
6307/**
6308 * alloc_contig_range() -- tries to allocate given range of pages
6309 * @start: start PFN to allocate
6310 * @end: one-past-the-last PFN to allocate
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006311 * @migratetype: migratetype of the underlaying pageblocks (either
6312 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6313 * in range must have the same migratetype and it must
6314 * be either of the two.
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006315 *
6316 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6317 * aligned, however it's the caller's responsibility to guarantee that
6318 * we are the only thread that changes migrate type of pageblocks the
6319 * pages fall in.
6320 *
6321 * The PFN range must belong to a single zone.
6322 *
6323 * Returns zero on success or negative error code. On success all
6324 * pages which PFN is in [start, end) are allocated for the caller and
6325 * need to be freed with free_contig_range().
6326 */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006327int alloc_contig_range(unsigned long start, unsigned long end,
6328 unsigned migratetype)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006329{
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006330 unsigned long outer_start, outer_end;
6331 int ret = 0, order;
6332
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006333 struct compact_control cc = {
6334 .nr_migratepages = 0,
6335 .order = -1,
6336 .zone = page_zone(pfn_to_page(start)),
David Rientjese0b9dae2014-06-04 16:08:28 -07006337 .mode = MIGRATE_SYNC,
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006338 .ignore_skip_hint = true,
6339 };
6340 INIT_LIST_HEAD(&cc.migratepages);
6341
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006342 /*
6343 * What we do here is we mark all pageblocks in range as
6344 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6345 * have different sizes, and due to the way page allocator
6346 * work, we align the range to biggest of the two pages so
6347 * that page allocator won't try to merge buddies from
6348 * different pageblocks and change MIGRATE_ISOLATE to some
6349 * other migration type.
6350 *
6351 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6352 * migrate the pages from an unaligned range (ie. pages that
6353 * we are interested in). This will put all the pages in
6354 * range back to page allocator as MIGRATE_ISOLATE.
6355 *
6356 * When this is done, we take the pages in range from page
6357 * allocator removing them from the buddy system. This way
6358 * page allocator will never consider using them.
6359 *
6360 * This lets us mark the pageblocks back as
6361 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6362 * aligned range but not in the unaligned, original range are
6363 * put back to page allocator so that buddy can use them.
6364 */
6365
6366 ret = start_isolate_page_range(pfn_max_align_down(start),
Wen Congyangb023f462012-12-11 16:00:45 -08006367 pfn_max_align_up(end), migratetype,
6368 false);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006369 if (ret)
Bob Liu86a595f2012-10-25 13:37:56 -07006370 return ret;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006371
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006372 ret = __alloc_contig_migrate_range(&cc, start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006373 if (ret)
6374 goto done;
6375
6376 /*
6377 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6378 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
6379 * more, all pages in [start, end) are free in page allocator.
6380 * What we are going to do is to allocate all pages from
6381 * [start, end) (that is remove them from page allocator).
6382 *
6383 * The only problem is that pages at the beginning and at the
6384 * end of interesting range may be not aligned with pages that
6385 * page allocator holds, ie. they can be part of higher order
6386 * pages. Because of this, we reserve the bigger range and
6387 * once this is done free the pages we are not interested in.
6388 *
6389 * We don't have to hold zone->lock here because the pages are
6390 * isolated thus they won't get removed from buddy.
6391 */
6392
6393 lru_add_drain_all();
6394 drain_all_pages();
6395
6396 order = 0;
6397 outer_start = start;
6398 while (!PageBuddy(pfn_to_page(outer_start))) {
6399 if (++order >= MAX_ORDER) {
6400 ret = -EBUSY;
6401 goto done;
6402 }
6403 outer_start &= ~0UL << order;
6404 }
6405
6406 /* Make sure the range is really isolated. */
Wen Congyangb023f462012-12-11 16:00:45 -08006407 if (test_pages_isolated(outer_start, end, false)) {
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006408 pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n",
6409 outer_start, end);
6410 ret = -EBUSY;
6411 goto done;
6412 }
6413
Marek Szyprowski49f223a2012-01-25 12:49:24 +01006414
6415 /* Grab isolated pages from freelists. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006416 outer_end = isolate_freepages_range(&cc, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006417 if (!outer_end) {
6418 ret = -EBUSY;
6419 goto done;
6420 }
6421
6422 /* Free head and tail (if any) */
6423 if (start != outer_start)
6424 free_contig_range(outer_start, start - outer_start);
6425 if (end != outer_end)
6426 free_contig_range(end, outer_end - end);
6427
6428done:
6429 undo_isolate_page_range(pfn_max_align_down(start),
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006430 pfn_max_align_up(end), migratetype);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006431 return ret;
6432}
6433
6434void free_contig_range(unsigned long pfn, unsigned nr_pages)
6435{
Marek Szyprowskibcc2b022012-12-20 15:05:18 -08006436 unsigned int count = 0;
6437
6438 for (; nr_pages--; pfn++) {
6439 struct page *page = pfn_to_page(pfn);
6440
6441 count += page_count(page) != 1;
6442 __free_page(page);
6443 }
6444 WARN(count != 0, "%d pages are still in use!\n", count);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006445}
6446#endif
6447
Jiang Liu4ed7e022012-07-31 16:43:35 -07006448#ifdef CONFIG_MEMORY_HOTPLUG
Cody P Schafer0a647f32013-07-03 15:01:33 -07006449/*
6450 * The zone indicated has a new number of managed_pages; batch sizes and percpu
6451 * page high values need to be recalulated.
6452 */
Jiang Liu4ed7e022012-07-31 16:43:35 -07006453void __meminit zone_pcp_update(struct zone *zone)
6454{
Cody P Schafer0a647f32013-07-03 15:01:33 -07006455 unsigned cpu;
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006456 mutex_lock(&pcp_batch_high_lock);
Cody P Schafer0a647f32013-07-03 15:01:33 -07006457 for_each_possible_cpu(cpu)
Cody P Schafer169f6c12013-07-03 15:01:41 -07006458 pageset_set_high_and_batch(zone,
6459 per_cpu_ptr(zone->pageset, cpu));
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006460 mutex_unlock(&pcp_batch_high_lock);
Jiang Liu4ed7e022012-07-31 16:43:35 -07006461}
6462#endif
6463
Jiang Liu340175b2012-07-31 16:43:32 -07006464void zone_pcp_reset(struct zone *zone)
6465{
6466 unsigned long flags;
Minchan Kim5a883812012-10-08 16:33:39 -07006467 int cpu;
6468 struct per_cpu_pageset *pset;
Jiang Liu340175b2012-07-31 16:43:32 -07006469
6470 /* avoid races with drain_pages() */
6471 local_irq_save(flags);
6472 if (zone->pageset != &boot_pageset) {
Minchan Kim5a883812012-10-08 16:33:39 -07006473 for_each_online_cpu(cpu) {
6474 pset = per_cpu_ptr(zone->pageset, cpu);
6475 drain_zonestat(zone, pset);
6476 }
Jiang Liu340175b2012-07-31 16:43:32 -07006477 free_percpu(zone->pageset);
6478 zone->pageset = &boot_pageset;
6479 }
6480 local_irq_restore(flags);
6481}
6482
Wen Congyang6dcd73d2012-12-11 16:01:01 -08006483#ifdef CONFIG_MEMORY_HOTREMOVE
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006484/*
6485 * All pages in the range must be isolated before calling this.
6486 */
6487void
6488__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
6489{
6490 struct page *page;
6491 struct zone *zone;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07006492 unsigned int order, i;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006493 unsigned long pfn;
6494 unsigned long flags;
6495 /* find the first valid pfn */
6496 for (pfn = start_pfn; pfn < end_pfn; pfn++)
6497 if (pfn_valid(pfn))
6498 break;
6499 if (pfn == end_pfn)
6500 return;
6501 zone = page_zone(pfn_to_page(pfn));
6502 spin_lock_irqsave(&zone->lock, flags);
6503 pfn = start_pfn;
6504 while (pfn < end_pfn) {
6505 if (!pfn_valid(pfn)) {
6506 pfn++;
6507 continue;
6508 }
6509 page = pfn_to_page(pfn);
Wen Congyangb023f462012-12-11 16:00:45 -08006510 /*
6511 * The HWPoisoned page may be not in buddy system, and
6512 * page_count() is not 0.
6513 */
6514 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
6515 pfn++;
6516 SetPageReserved(page);
6517 continue;
6518 }
6519
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006520 BUG_ON(page_count(page));
6521 BUG_ON(!PageBuddy(page));
6522 order = page_order(page);
6523#ifdef CONFIG_DEBUG_VM
6524 printk(KERN_INFO "remove from free list %lx %d %lx\n",
6525 pfn, 1 << order, end_pfn);
6526#endif
6527 list_del(&page->lru);
6528 rmv_page_order(page);
6529 zone->free_area[order].nr_free--;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006530 for (i = 0; i < (1 << order); i++)
6531 SetPageReserved((page+i));
6532 pfn += (1 << order);
6533 }
6534 spin_unlock_irqrestore(&zone->lock, flags);
6535}
6536#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01006537
6538#ifdef CONFIG_MEMORY_FAILURE
6539bool is_free_buddy_page(struct page *page)
6540{
6541 struct zone *zone = page_zone(page);
6542 unsigned long pfn = page_to_pfn(page);
6543 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07006544 unsigned int order;
Wu Fengguang8d22ba12009-12-16 12:19:58 +01006545
6546 spin_lock_irqsave(&zone->lock, flags);
6547 for (order = 0; order < MAX_ORDER; order++) {
6548 struct page *page_head = page - (pfn & ((1 << order) - 1));
6549
6550 if (PageBuddy(page_head) && page_order(page_head) >= order)
6551 break;
6552 }
6553 spin_unlock_irqrestore(&zone->lock, flags);
6554
6555 return order < MAX_ORDER;
6556}
6557#endif
Wu Fengguang718a3822010-03-10 15:20:43 -08006558
Andrew Morton51300ce2012-05-29 15:06:44 -07006559static const struct trace_print_flags pageflag_names[] = {
Wu Fengguang718a3822010-03-10 15:20:43 -08006560 {1UL << PG_locked, "locked" },
6561 {1UL << PG_error, "error" },
6562 {1UL << PG_referenced, "referenced" },
6563 {1UL << PG_uptodate, "uptodate" },
6564 {1UL << PG_dirty, "dirty" },
6565 {1UL << PG_lru, "lru" },
6566 {1UL << PG_active, "active" },
6567 {1UL << PG_slab, "slab" },
6568 {1UL << PG_owner_priv_1, "owner_priv_1" },
6569 {1UL << PG_arch_1, "arch_1" },
6570 {1UL << PG_reserved, "reserved" },
6571 {1UL << PG_private, "private" },
6572 {1UL << PG_private_2, "private_2" },
6573 {1UL << PG_writeback, "writeback" },
6574#ifdef CONFIG_PAGEFLAGS_EXTENDED
6575 {1UL << PG_head, "head" },
6576 {1UL << PG_tail, "tail" },
6577#else
6578 {1UL << PG_compound, "compound" },
6579#endif
6580 {1UL << PG_swapcache, "swapcache" },
6581 {1UL << PG_mappedtodisk, "mappedtodisk" },
6582 {1UL << PG_reclaim, "reclaim" },
Wu Fengguang718a3822010-03-10 15:20:43 -08006583 {1UL << PG_swapbacked, "swapbacked" },
6584 {1UL << PG_unevictable, "unevictable" },
6585#ifdef CONFIG_MMU
6586 {1UL << PG_mlocked, "mlocked" },
6587#endif
6588#ifdef CONFIG_ARCH_USES_PG_UNCACHED
6589 {1UL << PG_uncached, "uncached" },
6590#endif
6591#ifdef CONFIG_MEMORY_FAILURE
6592 {1UL << PG_hwpoison, "hwpoison" },
6593#endif
Gavin Shanbe9cd872012-05-29 15:06:44 -07006594#ifdef CONFIG_TRANSPARENT_HUGEPAGE
6595 {1UL << PG_compound_lock, "compound_lock" },
6596#endif
Wu Fengguang718a3822010-03-10 15:20:43 -08006597};
6598
6599static void dump_page_flags(unsigned long flags)
6600{
6601 const char *delim = "";
6602 unsigned long mask;
6603 int i;
6604
Andrew Morton51300ce2012-05-29 15:06:44 -07006605 BUILD_BUG_ON(ARRAY_SIZE(pageflag_names) != __NR_PAGEFLAGS);
Johannes Weineracc50c12012-05-29 15:06:44 -07006606
Wu Fengguang718a3822010-03-10 15:20:43 -08006607 printk(KERN_ALERT "page flags: %#lx(", flags);
6608
6609 /* remove zone id */
6610 flags &= (1UL << NR_PAGEFLAGS) - 1;
6611
Andrew Morton51300ce2012-05-29 15:06:44 -07006612 for (i = 0; i < ARRAY_SIZE(pageflag_names) && flags; i++) {
Wu Fengguang718a3822010-03-10 15:20:43 -08006613
6614 mask = pageflag_names[i].mask;
6615 if ((flags & mask) != mask)
6616 continue;
6617
6618 flags &= ~mask;
6619 printk("%s%s", delim, pageflag_names[i].name);
6620 delim = "|";
6621 }
6622
6623 /* check for left over flags */
6624 if (flags)
6625 printk("%s%#lx", delim, flags);
6626
6627 printk(")\n");
6628}
6629
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -07006630void dump_page_badflags(struct page *page, const char *reason,
6631 unsigned long badflags)
Wu Fengguang718a3822010-03-10 15:20:43 -08006632{
6633 printk(KERN_ALERT
6634 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
Andrea Arcangeli4e9f64c2011-01-13 15:46:29 -08006635 page, atomic_read(&page->_count), page_mapcount(page),
Wu Fengguang718a3822010-03-10 15:20:43 -08006636 page->mapping, page->index);
6637 dump_page_flags(page->flags);
Dave Hansenf0b791a2014-01-23 15:52:49 -08006638 if (reason)
6639 pr_alert("page dumped because: %s\n", reason);
6640 if (page->flags & badflags) {
6641 pr_alert("bad because of flags:\n");
6642 dump_page_flags(page->flags & badflags);
6643 }
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07006644 mem_cgroup_print_bad_page(page);
Wu Fengguang718a3822010-03-10 15:20:43 -08006645}
Dave Hansenf0b791a2014-01-23 15:52:49 -08006646
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -07006647void dump_page(struct page *page, const char *reason)
Dave Hansenf0b791a2014-01-23 15:52:49 -08006648{
6649 dump_page_badflags(page, reason, 0);
6650}
John Hubbarded12d842014-04-07 15:37:59 -07006651EXPORT_SYMBOL(dump_page);