blob: b60de5603ccdceb224d2e7cce699ed04abcc6e27 [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>
Andrey Ryabininb8c73fc2015-02-13 14:39:28 -080028#include <linux/kasan.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/module.h>
30#include <linux/suspend.h>
31#include <linux/pagevec.h>
32#include <linux/blkdev.h>
33#include <linux/slab.h>
Dave Hansena238ab52011-05-24 17:12:16 -070034#include <linux/ratelimit.h>
David Rientjes5a3135c22007-10-16 23:25:53 -070035#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <linux/notifier.h>
37#include <linux/topology.h>
38#include <linux/sysctl.h>
39#include <linux/cpu.h>
40#include <linux/cpuset.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070041#include <linux/memory_hotplug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include <linux/nodemask.h>
43#include <linux/vmalloc.h>
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -070044#include <linux/vmstat.h>
Christoph Lameter4be38e32006-01-06 00:11:17 -080045#include <linux/mempolicy.h>
Yasunori Goto68113782006-06-23 02:03:11 -070046#include <linux/stop_machine.h>
Mel Gormanc7132162006-09-27 01:49:43 -070047#include <linux/sort.h>
48#include <linux/pfn.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070049#include <linux/backing-dev.h>
Akinobu Mita933e3122006-12-08 02:39:45 -080050#include <linux/fault-inject.h>
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -070051#include <linux/page-isolation.h>
Joonsoo Kimeefa864b2014-12-12 16:55:46 -080052#include <linux/page_ext.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070053#include <linux/debugobjects.h>
Catalin Marinasdbb1f812009-06-11 13:23:19 +010054#include <linux/kmemleak.h>
Mel Gorman56de7262010-05-24 14:32:30 -070055#include <linux/compaction.h>
Mel Gorman0d3d0622009-09-21 17:02:44 -070056#include <trace/events/kmem.h>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070057#include <linux/prefetch.h>
Lisa Du6e543d52013-09-11 14:22:36 -070058#include <linux/mm_inline.h>
Michal Nazarewicz041d3a82011-12-29 13:09:50 +010059#include <linux/migrate.h>
Joonsoo Kime30825f2014-12-12 16:55:49 -080060#include <linux/page_ext.h>
David Rientjes949f7ec2013-04-29 15:07:48 -070061#include <linux/hugetlb.h>
Clark Williams8bd75c72013-02-07 09:47:07 -060062#include <linux/sched/rt.h>
Joonsoo Kim48c96a32014-12-12 16:56:01 -080063#include <linux/page_owner.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_);
Joonsoo Kimad2c8142014-10-09 15:26:13 -070088int _node_numa_mem_[MAX_NUMNODES];
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070089#endif
90
Linus Torvalds1da177e2005-04-16 15:20:36 -070091/*
Christoph Lameter13808912007-10-16 01:25:27 -070092 * Array of node states.
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 */
Christoph Lameter13808912007-10-16 01:25:27 -070094nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
95 [N_POSSIBLE] = NODE_MASK_ALL,
96 [N_ONLINE] = { { [0] = 1UL } },
97#ifndef CONFIG_NUMA
98 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
99#ifdef CONFIG_HIGHMEM
100 [N_HIGH_MEMORY] = { { [0] = 1UL } },
101#endif
Lai Jiangshan20b2f522012-12-12 13:52:00 -0800102#ifdef CONFIG_MOVABLE_NODE
103 [N_MEMORY] = { { [0] = 1UL } },
104#endif
Christoph Lameter13808912007-10-16 01:25:27 -0700105 [N_CPU] = { { [0] = 1UL } },
106#endif /* NUMA */
107};
108EXPORT_SYMBOL(node_states);
109
Jiang Liuc3d5f5f2013-07-03 15:03:14 -0700110/* Protect totalram_pages and zone->managed_pages */
111static DEFINE_SPINLOCK(managed_page_count_lock);
112
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700113unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -0700114unsigned long totalreserve_pages __read_mostly;
Pintu Kumare48322a2014-12-18 16:17:15 -0800115unsigned long totalcma_pages __read_mostly;
Johannes Weinerab8fabd2012-01-10 15:07:42 -0800116/*
117 * When calculating the number of globally allowed dirty pages, there
118 * is a certain number of per-zone reserves that should not be
119 * considered dirtyable memory. This is the sum of those reserves
120 * over all existing zones that contribute dirtyable memory.
121 */
122unsigned long dirty_balance_reserve __read_mostly;
123
Hugh Dickins1b76b022012-05-11 01:00:07 -0700124int percpu_pagelist_fraction;
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000125gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800127#ifdef CONFIG_PM_SLEEP
128/*
129 * The following functions are used by the suspend/hibernate code to temporarily
130 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
131 * while devices are suspended. To avoid races with the suspend/hibernate code,
132 * they should always be called with pm_mutex held (gfp_allowed_mask also should
133 * only be modified with pm_mutex held, unless the suspend/hibernate code is
134 * guaranteed not to run in parallel with that modification).
135 */
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100136
137static gfp_t saved_gfp_mask;
138
139void pm_restore_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800140{
141 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100142 if (saved_gfp_mask) {
143 gfp_allowed_mask = saved_gfp_mask;
144 saved_gfp_mask = 0;
145 }
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800146}
147
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100148void pm_restrict_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800149{
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800150 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100151 WARN_ON(saved_gfp_mask);
152 saved_gfp_mask = gfp_allowed_mask;
153 gfp_allowed_mask &= ~GFP_IOFS;
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800154}
Mel Gormanf90ac392012-01-10 15:07:15 -0800155
156bool pm_suspended_storage(void)
157{
158 if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS)
159 return false;
160 return true;
161}
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800162#endif /* CONFIG_PM_SLEEP */
163
Mel Gormand9c23402007-10-16 01:26:01 -0700164#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
165int pageblock_order __read_mostly;
166#endif
167
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800168static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -0800169
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170/*
171 * results with 256, 32 in the lowmem_reserve sysctl:
172 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
173 * 1G machine -> (16M dma, 784M normal, 224M high)
174 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
175 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
Yaowei Bai84109e12015-02-12 15:00:22 -0800176 * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +0100177 *
178 * TBD: should special case ZONE_DMA32 machines here - in those we normally
179 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700181int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800182#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700183 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -0800184#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700185#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700186 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700187#endif
Christoph Lametere53ef382006-09-25 23:31:14 -0700188#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700189 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700190#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700191 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700192};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
194EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195
Helge Deller15ad7cd2006-12-06 20:40:36 -0800196static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800197#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700198 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800199#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700200#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700201 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700202#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700203 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700204#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700205 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700206#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700207 "Movable",
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700208};
209
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210int min_free_kbytes = 1024;
Han Pingtian42aa83c2014-01-23 15:53:28 -0800211int user_min_free_kbytes = -1;
Arve Hjønnevågc3864002009-02-17 14:51:02 -0800212int min_free_order_shift = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213
Jan Beulich2c85f512009-09-21 17:03:07 -0700214static unsigned long __meminitdata nr_kernel_pages;
215static unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700216static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217
Tejun Heo0ee332c2011-12-08 10:22:09 -0800218#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
219static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
220static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
221static unsigned long __initdata required_kernelcore;
222static unsigned long __initdata required_movablecore;
223static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Mel Gormanc7132162006-09-27 01:49:43 -0700224
Tejun Heo0ee332c2011-12-08 10:22:09 -0800225/* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
226int movable_zone;
227EXPORT_SYMBOL(movable_zone);
228#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -0700229
Miklos Szeredi418508c2007-05-23 13:57:55 -0700230#if MAX_NUMNODES > 1
231int nr_node_ids __read_mostly = MAX_NUMNODES;
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700232int nr_online_nodes __read_mostly = 1;
Miklos Szeredi418508c2007-05-23 13:57:55 -0700233EXPORT_SYMBOL(nr_node_ids);
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700234EXPORT_SYMBOL(nr_online_nodes);
Miklos Szeredi418508c2007-05-23 13:57:55 -0700235#endif
236
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700237int page_group_by_mobility_disabled __read_mostly;
238
Minchan Kimee6f5092012-07-31 16:43:50 -0700239void set_pageblock_migratetype(struct page *page, int migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700240{
KOSAKI Motohiro5d0f3f72013-11-12 15:08:18 -0800241 if (unlikely(page_group_by_mobility_disabled &&
242 migratetype < MIGRATE_PCPTYPES))
Mel Gorman49255c62009-06-16 15:31:58 -0700243 migratetype = MIGRATE_UNMOVABLE;
244
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700245 set_pageblock_flags_group(page, (unsigned long)migratetype,
246 PB_migrate, PB_migrate_end);
247}
248
Nick Piggin13e74442006-01-06 00:10:58 -0800249#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700250static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700252 int ret = 0;
253 unsigned seq;
254 unsigned long pfn = page_to_pfn(page);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800255 unsigned long sp, start_pfn;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700256
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700257 do {
258 seq = zone_span_seqbegin(zone);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800259 start_pfn = zone->zone_start_pfn;
260 sp = zone->spanned_pages;
Cody P Schafer108bcc92013-02-22 16:35:23 -0800261 if (!zone_spans_pfn(zone, pfn))
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700262 ret = 1;
263 } while (zone_span_seqretry(zone, seq));
264
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800265 if (ret)
Dave Hansen613813e2014-06-04 16:07:27 -0700266 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
267 pfn, zone_to_nid(zone), zone->name,
268 start_pfn, start_pfn + sp);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800269
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700270 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700271}
272
273static int page_is_consistent(struct zone *zone, struct page *page)
274{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700275 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700276 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700278 return 0;
279
280 return 1;
281}
282/*
283 * Temporary debugging check for pages not lying within a given zone.
284 */
285static int bad_range(struct zone *zone, struct page *page)
286{
287 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700289 if (!page_is_consistent(zone, page))
290 return 1;
291
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 return 0;
293}
Nick Piggin13e74442006-01-06 00:10:58 -0800294#else
295static inline int bad_range(struct zone *zone, struct page *page)
296{
297 return 0;
298}
299#endif
300
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -0700301static void bad_page(struct page *page, const char *reason,
302 unsigned long bad_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800304 static unsigned long resume;
305 static unsigned long nr_shown;
306 static unsigned long nr_unshown;
307
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200308 /* Don't complain about poisoned pages */
309 if (PageHWPoison(page)) {
Mel Gorman22b751c2013-02-22 16:34:59 -0800310 page_mapcount_reset(page); /* remove PageBuddy */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200311 return;
312 }
313
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800314 /*
315 * Allow a burst of 60 reports, then keep quiet for that minute;
316 * or allow a steady drip of one report per second.
317 */
318 if (nr_shown == 60) {
319 if (time_before(jiffies, resume)) {
320 nr_unshown++;
321 goto out;
322 }
323 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800324 printk(KERN_ALERT
325 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800326 nr_unshown);
327 nr_unshown = 0;
328 }
329 nr_shown = 0;
330 }
331 if (nr_shown++ == 0)
332 resume = jiffies + 60 * HZ;
333
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800334 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800335 current->comm, page_to_pfn(page));
Dave Hansenf0b791a2014-01-23 15:52:49 -0800336 dump_page_badflags(page, reason, bad_flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700337
Dave Jones4f318882011-10-31 17:07:24 -0700338 print_modules();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800340out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800341 /* Leave bad fields for debug, except PageBuddy could make trouble */
Mel Gorman22b751c2013-02-22 16:34:59 -0800342 page_mapcount_reset(page); /* remove PageBuddy */
Rusty Russell373d4d02013-01-21 17:17:39 +1030343 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344}
345
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346/*
347 * Higher-order pages are called "compound pages". They are structured thusly:
348 *
349 * The first PAGE_SIZE page is called the "head page".
350 *
351 * The remaining PAGE_SIZE pages are called "tail pages".
352 *
Wang Sheng-Hui6416b9f2011-11-17 10:53:50 +0100353 * All pages have PG_compound set. All tail pages have their ->first_page
354 * pointing at the head page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800356 * The first tail page's ->lru.next holds the address of the compound page's
357 * put_page() function. Its ->lru.prev holds the order of allocation.
358 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800360
361static void free_compound_page(struct page *page)
362{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700363 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800364}
365
Andi Kleen01ad1c02008-07-23 21:27:46 -0700366void prep_compound_page(struct page *page, unsigned long order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367{
368 int i;
369 int nr_pages = 1 << order;
370
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800371 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700372 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700373 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800374 for (i = 1; i < nr_pages; i++) {
375 struct page *p = page + i;
Youquan Song58a84aa2011-12-08 14:34:18 -0800376 set_page_count(p, 0);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700377 p->first_page = page;
David Rientjes668f9abb2014-03-03 15:38:18 -0800378 /* Make sure p->first_page is always valid for PageTail() */
379 smp_wmb();
380 __SetPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 }
382}
383
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700384static inline void prep_zero_page(struct page *page, unsigned int order,
385 gfp_t gfp_flags)
Nick Piggin17cf4402006-03-22 00:08:41 -0800386{
387 int i;
388
Andrew Morton6626c5d2006-03-22 00:08:42 -0800389 /*
390 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
391 * and __GFP_HIGHMEM from hard or soft interrupt context.
392 */
Nick Piggin725d7042006-09-25 23:30:55 -0700393 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800394 for (i = 0; i < (1 << order); i++)
395 clear_highpage(page + i);
396}
397
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800398#ifdef CONFIG_DEBUG_PAGEALLOC
399unsigned int _debug_guardpage_minorder;
Joonsoo Kim031bc572014-12-12 16:55:52 -0800400bool _debug_pagealloc_enabled __read_mostly;
Joonsoo Kime30825f2014-12-12 16:55:49 -0800401bool _debug_guardpage_enabled __read_mostly;
402
Joonsoo Kim031bc572014-12-12 16:55:52 -0800403static int __init early_debug_pagealloc(char *buf)
404{
405 if (!buf)
406 return -EINVAL;
407
408 if (strcmp(buf, "on") == 0)
409 _debug_pagealloc_enabled = true;
410
411 return 0;
412}
413early_param("debug_pagealloc", early_debug_pagealloc);
414
Joonsoo Kime30825f2014-12-12 16:55:49 -0800415static bool need_debug_guardpage(void)
416{
Joonsoo Kim031bc572014-12-12 16:55:52 -0800417 /* If we don't use debug_pagealloc, we don't need guard page */
418 if (!debug_pagealloc_enabled())
419 return false;
420
Joonsoo Kime30825f2014-12-12 16:55:49 -0800421 return true;
422}
423
424static void init_debug_guardpage(void)
425{
Joonsoo Kim031bc572014-12-12 16:55:52 -0800426 if (!debug_pagealloc_enabled())
427 return;
428
Joonsoo Kime30825f2014-12-12 16:55:49 -0800429 _debug_guardpage_enabled = true;
430}
431
432struct page_ext_operations debug_guardpage_ops = {
433 .need = need_debug_guardpage,
434 .init = init_debug_guardpage,
435};
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800436
437static int __init debug_guardpage_minorder_setup(char *buf)
438{
439 unsigned long res;
440
441 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
442 printk(KERN_ERR "Bad debug_guardpage_minorder value\n");
443 return 0;
444 }
445 _debug_guardpage_minorder = res;
446 printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res);
447 return 0;
448}
449__setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
450
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800451static inline void set_page_guard(struct zone *zone, struct page *page,
452 unsigned int order, int migratetype)
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800453{
Joonsoo Kime30825f2014-12-12 16:55:49 -0800454 struct page_ext *page_ext;
455
456 if (!debug_guardpage_enabled())
457 return;
458
459 page_ext = lookup_page_ext(page);
460 __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
461
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800462 INIT_LIST_HEAD(&page->lru);
463 set_page_private(page, order);
464 /* Guard pages are not available for any usage */
465 __mod_zone_freepage_state(zone, -(1 << order), migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800466}
467
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800468static inline void clear_page_guard(struct zone *zone, struct page *page,
469 unsigned int order, int migratetype)
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800470{
Joonsoo Kime30825f2014-12-12 16:55:49 -0800471 struct page_ext *page_ext;
472
473 if (!debug_guardpage_enabled())
474 return;
475
476 page_ext = lookup_page_ext(page);
477 __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
478
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800479 set_page_private(page, 0);
480 if (!is_migrate_isolate(migratetype))
481 __mod_zone_freepage_state(zone, (1 << order), migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800482}
483#else
Joonsoo Kime30825f2014-12-12 16:55:49 -0800484struct page_ext_operations debug_guardpage_ops = { NULL, };
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800485static inline void set_page_guard(struct zone *zone, struct page *page,
486 unsigned int order, int migratetype) {}
487static inline void clear_page_guard(struct zone *zone, struct page *page,
488 unsigned int order, int migratetype) {}
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800489#endif
490
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700491static inline void set_page_order(struct page *page, unsigned int order)
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700492{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700493 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000494 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495}
496
497static inline void rmv_page_order(struct page *page)
498{
Nick Piggin676165a2006-04-10 11:21:48 +1000499 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700500 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501}
502
503/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 * This function checks whether a page is free && is the buddy
505 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800506 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000507 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700508 * (c) a page and its buddy have the same order &&
509 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 *
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700511 * For recording whether a page is in the buddy system, we set ->_mapcount
512 * PAGE_BUDDY_MAPCOUNT_VALUE.
513 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
514 * serialized by zone->lock.
Nick Piggin676165a2006-04-10 11:21:48 +1000515 *
516 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700518static inline int page_is_buddy(struct page *page, struct page *buddy,
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700519 unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700521 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800522 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800523
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800524 if (page_is_guard(buddy) && page_order(buddy) == order) {
Mel Gormand34c5fa2014-06-04 16:10:10 -0700525 if (page_zone_id(page) != page_zone_id(buddy))
526 return 0;
527
Weijie Yang4c5018c2015-02-10 14:11:39 -0800528 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
529
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800530 return 1;
531 }
532
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700533 if (PageBuddy(buddy) && page_order(buddy) == order) {
Mel Gormand34c5fa2014-06-04 16:10:10 -0700534 /*
535 * zone check is done late to avoid uselessly
536 * calculating zone/node ids for pages that could
537 * never merge.
538 */
539 if (page_zone_id(page) != page_zone_id(buddy))
540 return 0;
541
Weijie Yang4c5018c2015-02-10 14:11:39 -0800542 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
543
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700544 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000545 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700546 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547}
548
549/*
550 * Freeing function for a buddy system allocator.
551 *
552 * The concept of a buddy system is to maintain direct-mapped table
553 * (containing bit values) for memory blocks of various "orders".
554 * The bottom level table contains the map for the smallest allocatable
555 * units of memory (here, pages), and each level above it describes
556 * pairs of units from the levels below, hence, "buddies".
557 * At a high level, all that happens here is marking the table entry
558 * at the bottom level available, and propagating the changes upward
559 * as necessary, plus some accounting needed to play nicely with other
560 * parts of the VM system.
561 * At each level, we keep a list of pages, which are heads of continuous
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700562 * free pages of length of (1 << order) and marked with _mapcount
563 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
564 * field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 * So when we are allocating or freeing one, we can derive the state of the
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100566 * other. That is, if we allocate a small block, and both were
567 * free, the remainder of the region must be split into blocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 * If a block is freed, and its buddy is also free, then this
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100569 * triggers coalescing into a block of larger size.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +0100571 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 */
573
Nick Piggin48db57f2006-01-08 01:00:42 -0800574static inline void __free_one_page(struct page *page,
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700575 unsigned long pfn,
Mel Gormaned0ae212009-06-16 15:32:07 -0700576 struct zone *zone, unsigned int order,
577 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578{
579 unsigned long page_idx;
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700580 unsigned long combined_idx;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800581 unsigned long uninitialized_var(buddy_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700582 struct page *buddy;
Joonsoo Kim3c605092014-11-13 15:19:21 -0800583 int max_order = MAX_ORDER;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
Cody P Schaferd29bb972013-02-22 16:35:25 -0800585 VM_BUG_ON(!zone_is_initialized(zone));
Kirill A. Shutemov6e9f0d52015-02-11 15:25:50 -0800586 VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
Mel Gormaned0ae212009-06-16 15:32:07 -0700588 VM_BUG_ON(migratetype == -1);
Joonsoo Kim3c605092014-11-13 15:19:21 -0800589 if (is_migrate_isolate(migratetype)) {
590 /*
591 * We restrict max order of merging to prevent merge
592 * between freepages on isolate pageblock and normal
593 * pageblock. Without this, pageblock isolation
594 * could cause incorrect freepage accounting.
595 */
596 max_order = min(MAX_ORDER, pageblock_order + 1);
597 } else {
Joonsoo Kim8f82b552014-11-13 15:19:18 -0800598 __mod_zone_freepage_state(zone, 1 << order, migratetype);
Joonsoo Kim3c605092014-11-13 15:19:21 -0800599 }
Mel Gormaned0ae212009-06-16 15:32:07 -0700600
Joonsoo Kim3c605092014-11-13 15:19:21 -0800601 page_idx = pfn & ((1 << max_order) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602
Sasha Levin309381fea2014-01-23 15:52:54 -0800603 VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page);
604 VM_BUG_ON_PAGE(bad_range(zone, page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605
Joonsoo Kim3c605092014-11-13 15:19:21 -0800606 while (order < max_order - 1) {
KyongHo Cho43506fa2011-01-13 15:47:24 -0800607 buddy_idx = __find_buddy_index(page_idx, order);
608 buddy = page + (buddy_idx - page_idx);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700609 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700610 break;
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800611 /*
612 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
613 * merge with it and move up one order.
614 */
615 if (page_is_guard(buddy)) {
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800616 clear_page_guard(zone, buddy, order, migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800617 } else {
618 list_del(&buddy->lru);
619 zone->free_area[order].nr_free--;
620 rmv_page_order(buddy);
621 }
KyongHo Cho43506fa2011-01-13 15:47:24 -0800622 combined_idx = buddy_idx & page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 page = page + (combined_idx - page_idx);
624 page_idx = combined_idx;
625 order++;
626 }
627 set_page_order(page, order);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700628
629 /*
630 * If this is not the largest possible page, check if the buddy
631 * of the next-highest order is free. If it is, it's possible
632 * that pages are being freed that will coalesce soon. In case,
633 * that is happening, add the free page to the tail of the list
634 * so it's less likely to be used soon and more likely to be merged
635 * as a higher order page
636 */
Mel Gormanb7f50cf2010-10-26 14:21:11 -0700637 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700638 struct page *higher_page, *higher_buddy;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800639 combined_idx = buddy_idx & page_idx;
640 higher_page = page + (combined_idx - page_idx);
641 buddy_idx = __find_buddy_index(combined_idx, order + 1);
Li Haifeng0ba8f2d2012-09-17 14:09:21 -0700642 higher_buddy = higher_page + (buddy_idx - combined_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700643 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
644 list_add_tail(&page->lru,
645 &zone->free_area[order].free_list[migratetype]);
646 goto out;
647 }
648 }
649
650 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
651out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 zone->free_area[order].nr_free++;
653}
654
Nick Piggin224abf92006-01-06 00:11:11 -0800655static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656{
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -0700657 const char *bad_reason = NULL;
Dave Hansenf0b791a2014-01-23 15:52:49 -0800658 unsigned long bad_flags = 0;
659
660 if (unlikely(page_mapcount(page)))
661 bad_reason = "nonzero mapcount";
662 if (unlikely(page->mapping != NULL))
663 bad_reason = "non-NULL mapping";
664 if (unlikely(atomic_read(&page->_count) != 0))
665 bad_reason = "nonzero _count";
666 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
667 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
668 bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
669 }
Johannes Weiner9edad6e2014-12-10 15:44:58 -0800670#ifdef CONFIG_MEMCG
671 if (unlikely(page->mem_cgroup))
672 bad_reason = "page still charged to cgroup";
673#endif
Dave Hansenf0b791a2014-01-23 15:52:49 -0800674 if (unlikely(bad_reason)) {
675 bad_page(page, bad_reason, bad_flags);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800676 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800677 }
Peter Zijlstra90572892013-10-07 11:29:20 +0100678 page_cpupid_reset_last(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800679 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
680 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
681 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682}
683
684/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700685 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700687 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 *
689 * If the zone was previously in an "all pages pinned" state then look to
690 * see if this freeing clears that state.
691 *
692 * And clear the zone's pages_scanned counter, to hold off the "all pages are
693 * pinned" detection logic.
694 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700695static void free_pcppages_bulk(struct zone *zone, int count,
696 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697{
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700698 int migratetype = 0;
Mel Gormana6f9edd62009-09-21 17:03:20 -0700699 int batch_free = 0;
Mel Gorman72853e22010-09-09 16:38:16 -0700700 int to_free = count;
Mel Gorman0d5d8232014-08-06 16:07:16 -0700701 unsigned long nr_scanned;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700702
Nick Pigginc54ad302006-01-06 00:10:56 -0800703 spin_lock(&zone->lock);
Mel Gorman0d5d8232014-08-06 16:07:16 -0700704 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
705 if (nr_scanned)
706 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
Mel Gormanf2260e62009-06-16 15:32:13 -0700707
Mel Gorman72853e22010-09-09 16:38:16 -0700708 while (to_free) {
Nick Piggin48db57f2006-01-08 01:00:42 -0800709 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700710 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -0800711
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700712 /*
Mel Gormana6f9edd62009-09-21 17:03:20 -0700713 * Remove pages from lists in a round-robin fashion. A
714 * batch_free count is maintained that is incremented when an
715 * empty list is encountered. This is so more pages are freed
716 * off fuller lists instead of spinning excessively around empty
717 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700718 */
719 do {
Mel Gormana6f9edd62009-09-21 17:03:20 -0700720 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700721 if (++migratetype == MIGRATE_PCPTYPES)
722 migratetype = 0;
723 list = &pcp->lists[migratetype];
724 } while (list_empty(list));
725
Namhyung Kim1d168712011-03-22 16:32:45 -0700726 /* This is the only non-empty list. Free them all. */
727 if (batch_free == MIGRATE_PCPTYPES)
728 batch_free = to_free;
729
Mel Gormana6f9edd62009-09-21 17:03:20 -0700730 do {
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -0700731 int mt; /* migratetype of the to-be-freed page */
732
Mel Gormana6f9edd62009-09-21 17:03:20 -0700733 page = list_entry(list->prev, struct page, lru);
734 /* must delete as __free_one_page list manipulates */
735 list_del(&page->lru);
Minchan Kimb12c4ad2012-10-08 16:32:08 -0700736 mt = get_freepage_migratetype(page);
Joonsoo Kim8f82b552014-11-13 15:19:18 -0800737 if (unlikely(has_isolate_pageblock(zone)))
Joonsoo Kim51bb1a42014-11-13 15:19:14 -0800738 mt = get_pageblock_migratetype(page);
Joonsoo Kim51bb1a42014-11-13 15:19:14 -0800739
Hugh Dickinsa7016232010-01-29 17:46:34 +0000740 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700741 __free_one_page(page, page_to_pfn(page), zone, 0, mt);
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -0700742 trace_mm_page_pcpu_drain(page, 0, mt);
Mel Gorman72853e22010-09-09 16:38:16 -0700743 } while (--to_free && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800745 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746}
747
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700748static void free_one_page(struct zone *zone,
749 struct page *page, unsigned long pfn,
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700750 unsigned int order,
Mel Gormaned0ae212009-06-16 15:32:07 -0700751 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800752{
Mel Gorman0d5d8232014-08-06 16:07:16 -0700753 unsigned long nr_scanned;
Christoph Lameter006d22d2006-09-25 23:31:48 -0700754 spin_lock(&zone->lock);
Mel Gorman0d5d8232014-08-06 16:07:16 -0700755 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
756 if (nr_scanned)
757 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
Mel Gormanf2260e62009-06-16 15:32:13 -0700758
Joonsoo Kimad53f922014-11-13 15:19:11 -0800759 if (unlikely(has_isolate_pageblock(zone) ||
760 is_migrate_isolate(migratetype))) {
761 migratetype = get_pfnblock_migratetype(page, pfn);
Joonsoo Kimad53f922014-11-13 15:19:11 -0800762 }
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700763 __free_one_page(page, pfn, zone, order, migratetype);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700764 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800765}
766
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800767static int free_tail_pages_check(struct page *head_page, struct page *page)
768{
769 if (!IS_ENABLED(CONFIG_DEBUG_VM))
770 return 0;
771 if (unlikely(!PageTail(page))) {
772 bad_page(page, "PageTail not set", 0);
773 return 1;
774 }
775 if (unlikely(page->first_page != head_page)) {
776 bad_page(page, "first_page not consistent", 0);
777 return 1;
778 }
779 return 0;
780}
781
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700782static bool free_pages_prepare(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783{
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800784 bool compound = PageCompound(page);
785 int i, bad = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786
Yu Zhaoab1f3062014-12-10 15:43:17 -0800787 VM_BUG_ON_PAGE(PageTail(page), page);
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800788 VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
Yu Zhaoab1f3062014-12-10 15:43:17 -0800789
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -0800790 trace_mm_page_free(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100791 kmemcheck_free_shadow(page, order);
Andrey Ryabininb8c73fc2015-02-13 14:39:28 -0800792 kasan_free_pages(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100793
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -0800794 if (PageAnon(page))
795 page->mapping = NULL;
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800796 bad += free_pages_check(page);
797 for (i = 1; i < (1 << order); i++) {
798 if (compound)
799 bad += free_tail_pages_check(page, page + i);
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -0800800 bad += free_pages_check(page + i);
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800801 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800802 if (bad)
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700803 return false;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800804
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800805 reset_page_owner(page, order);
806
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700807 if (!PageHighMem(page)) {
Pintu Kumarb8af2942013-09-11 14:20:34 -0700808 debug_check_no_locks_freed(page_address(page),
809 PAGE_SIZE << order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700810 debug_check_no_obj_freed(page_address(page),
811 PAGE_SIZE << order);
812 }
Nick Piggindafb1362006-10-11 01:21:30 -0700813 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800814 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700815
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700816 return true;
817}
818
819static void __free_pages_ok(struct page *page, unsigned int order)
820{
821 unsigned long flags;
Minchan Kim95e34412012-10-08 16:32:11 -0700822 int migratetype;
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700823 unsigned long pfn = page_to_pfn(page);
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700824
825 if (!free_pages_prepare(page, order))
826 return;
827
Mel Gormancfc47a22014-06-04 16:10:19 -0700828 migratetype = get_pfnblock_migratetype(page, pfn);
Nick Pigginc54ad302006-01-06 00:10:56 -0800829 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700830 __count_vm_events(PGFREE, 1 << order);
Minchan Kim95e34412012-10-08 16:32:11 -0700831 set_freepage_migratetype(page, migratetype);
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700832 free_one_page(page_zone(page), page, pfn, order, migratetype);
Nick Pigginc54ad302006-01-06 00:10:56 -0800833 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834}
835
Jiang Liu170a5a72013-07-03 15:03:17 -0700836void __init __free_pages_bootmem(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800837{
Johannes Weinerc3993072012-01-10 15:08:10 -0800838 unsigned int nr_pages = 1 << order;
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700839 struct page *p = page;
Johannes Weinerc3993072012-01-10 15:08:10 -0800840 unsigned int loop;
David Howellsa226f6c2006-01-06 00:11:08 -0800841
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700842 prefetchw(p);
843 for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
844 prefetchw(p + 1);
Johannes Weinerc3993072012-01-10 15:08:10 -0800845 __ClearPageReserved(p);
846 set_page_count(p, 0);
David Howellsa226f6c2006-01-06 00:11:08 -0800847 }
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700848 __ClearPageReserved(p);
849 set_page_count(p, 0);
Johannes Weinerc3993072012-01-10 15:08:10 -0800850
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700851 page_zone(page)->managed_pages += nr_pages;
Johannes Weinerc3993072012-01-10 15:08:10 -0800852 set_page_refcounted(page);
853 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800854}
855
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100856#ifdef CONFIG_CMA
Li Zhong9cf510a2013-08-23 13:52:52 +0800857/* Free whole pageblock and set its migration type to MIGRATE_CMA. */
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100858void __init init_cma_reserved_pageblock(struct page *page)
859{
860 unsigned i = pageblock_nr_pages;
861 struct page *p = page;
862
863 do {
864 __ClearPageReserved(p);
865 set_page_count(p, 0);
866 } while (++p, --i);
867
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100868 set_pageblock_migratetype(page, MIGRATE_CMA);
Michal Nazarewiczdc783272014-07-02 15:22:35 -0700869
870 if (pageblock_order >= MAX_ORDER) {
871 i = pageblock_nr_pages;
872 p = page;
873 do {
874 set_page_refcounted(p);
875 __free_pages(p, MAX_ORDER - 1);
876 p += MAX_ORDER_NR_PAGES;
877 } while (i -= MAX_ORDER_NR_PAGES);
878 } else {
879 set_page_refcounted(page);
880 __free_pages(page, pageblock_order);
881 }
882
Jiang Liu3dcc0572013-07-03 15:03:21 -0700883 adjust_managed_page_count(page, pageblock_nr_pages);
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100884}
885#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886
887/*
888 * The order of subdivision here is critical for the IO subsystem.
889 * Please do not alter this order without good reasons and regression
890 * testing. Specifically, as large blocks of memory are subdivided,
891 * the order in which smaller blocks are delivered depends on the order
892 * they're subdivided in this function. This is the primary factor
893 * influencing the order in which pages are delivered to the IO
894 * subsystem according to empirical testing, and this is also justified
895 * by considering the behavior of a buddy system containing a single
896 * large block of memory acted on by a series of small allocations.
897 * This behavior is a critical factor in sglist merging's success.
898 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +0100899 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800901static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700902 int low, int high, struct free_area *area,
903 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904{
905 unsigned long size = 1 << high;
906
907 while (high > low) {
908 area--;
909 high--;
910 size >>= 1;
Sasha Levin309381fea2014-01-23 15:52:54 -0800911 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800912
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800913 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) &&
Joonsoo Kime30825f2014-12-12 16:55:49 -0800914 debug_guardpage_enabled() &&
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800915 high < debug_guardpage_minorder()) {
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800916 /*
917 * Mark as guard pages (or page), that will allow to
918 * merge back to allocator when buddy will be freed.
919 * Corresponding page table entries will not be touched,
920 * pages will stay not present in virtual address space
921 */
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800922 set_page_guard(zone, &page[size], high, migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800923 continue;
924 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700925 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 area->nr_free++;
927 set_page_order(&page[size], high);
928 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929}
930
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931/*
932 * This page is about to be returned from the page allocator
933 */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200934static inline int check_new_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935{
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -0700936 const char *bad_reason = NULL;
Dave Hansenf0b791a2014-01-23 15:52:49 -0800937 unsigned long bad_flags = 0;
938
939 if (unlikely(page_mapcount(page)))
940 bad_reason = "nonzero mapcount";
941 if (unlikely(page->mapping != NULL))
942 bad_reason = "non-NULL mapping";
943 if (unlikely(atomic_read(&page->_count) != 0))
944 bad_reason = "nonzero _count";
945 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
946 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
947 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
948 }
Johannes Weiner9edad6e2014-12-10 15:44:58 -0800949#ifdef CONFIG_MEMCG
950 if (unlikely(page->mem_cgroup))
951 bad_reason = "page still charged to cgroup";
952#endif
Dave Hansenf0b791a2014-01-23 15:52:49 -0800953 if (unlikely(bad_reason)) {
954 bad_page(page, bad_reason, bad_flags);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800955 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800956 }
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200957 return 0;
958}
959
Vlastimil Babka75379192015-02-11 15:25:38 -0800960static int prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
961 int alloc_flags)
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200962{
963 int i;
964
965 for (i = 0; i < (1 << order); i++) {
966 struct page *p = page + i;
967 if (unlikely(check_new_page(p)))
968 return 1;
969 }
Hugh Dickins689bceb2005-11-21 21:32:20 -0800970
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700971 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800972 set_page_refcounted(page);
Nick Piggincc1025092006-12-06 20:32:00 -0800973
974 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 kernel_map_pages(page, 1 << order, 1);
Andrey Ryabininb8c73fc2015-02-13 14:39:28 -0800976 kasan_alloc_pages(page, order);
Nick Piggin17cf4402006-03-22 00:08:41 -0800977
978 if (gfp_flags & __GFP_ZERO)
979 prep_zero_page(page, order, gfp_flags);
980
981 if (order && (gfp_flags & __GFP_COMP))
982 prep_compound_page(page, order);
983
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800984 set_page_owner(page, order, gfp_flags);
985
Vlastimil Babka75379192015-02-11 15:25:38 -0800986 /*
987 * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was necessary to
988 * allocate the page. The expectation is that the caller is taking
989 * steps that will free more memory. The caller should avoid the page
990 * being used for !PFMEMALLOC purposes.
991 */
992 page->pfmemalloc = !!(alloc_flags & ALLOC_NO_WATERMARKS);
993
Hugh Dickins689bceb2005-11-21 21:32:20 -0800994 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995}
996
Mel Gorman56fd56b2007-10-16 01:25:58 -0700997/*
998 * Go through the free lists for the given migratetype and remove
999 * the smallest available page from the freelists
1000 */
Mel Gorman728ec982009-06-16 15:32:04 -07001001static inline
1002struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -07001003 int migratetype)
1004{
1005 unsigned int current_order;
Pintu Kumarb8af2942013-09-11 14:20:34 -07001006 struct free_area *area;
Mel Gorman56fd56b2007-10-16 01:25:58 -07001007 struct page *page;
1008
1009 /* Find a page of the appropriate size in the preferred list */
1010 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
1011 area = &(zone->free_area[current_order]);
1012 if (list_empty(&area->free_list[migratetype]))
1013 continue;
1014
1015 page = list_entry(area->free_list[migratetype].next,
1016 struct page, lru);
1017 list_del(&page->lru);
1018 rmv_page_order(page);
1019 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -07001020 expand(zone, page, order, current_order, area, migratetype);
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001021 set_freepage_migratetype(page, migratetype);
Mel Gorman56fd56b2007-10-16 01:25:58 -07001022 return page;
1023 }
1024
1025 return NULL;
1026}
1027
1028
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001029/*
1030 * This array describes the order lists are fallen back to when
1031 * the free lists for the desirable migrate type are depleted
1032 */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001033static int fallbacks[MIGRATE_TYPES][4] = {
1034 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
1035 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001036 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
Joonsoo Kimdc676472015-04-14 15:45:15 -07001037#ifdef CONFIG_CMA
1038 [MIGRATE_CMA] = { MIGRATE_RESERVE }, /* Never used */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001039#endif
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001040 [MIGRATE_RESERVE] = { MIGRATE_RESERVE }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -08001041#ifdef CONFIG_MEMORY_ISOLATION
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001042 [MIGRATE_ISOLATE] = { MIGRATE_RESERVE }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -08001043#endif
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001044};
1045
Joonsoo Kimdc676472015-04-14 15:45:15 -07001046#ifdef CONFIG_CMA
1047static struct page *__rmqueue_cma_fallback(struct zone *zone,
1048 unsigned int order)
1049{
1050 return __rmqueue_smallest(zone, order, MIGRATE_CMA);
1051}
1052#else
1053static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
1054 unsigned int order) { return NULL; }
1055#endif
1056
Mel Gormanc361be52007-10-16 01:25:51 -07001057/*
1058 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -07001059 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -07001060 * boundary. If alignment is required, use move_freepages_block()
1061 */
Minchan Kim435b4052012-10-08 16:32:16 -07001062int move_freepages(struct zone *zone,
Adrian Bunkb69a7282008-07-23 21:28:12 -07001063 struct page *start_page, struct page *end_page,
1064 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001065{
1066 struct page *page;
1067 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -07001068 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -07001069
1070#ifndef CONFIG_HOLES_IN_ZONE
1071 /*
1072 * page_zone is not safe to call in this context when
1073 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1074 * anyway as we check zone boundaries in move_freepages_block().
1075 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -07001076 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -07001077 */
Mel Gorman97ee4ba2014-10-09 15:28:28 -07001078 VM_BUG_ON(page_zone(start_page) != page_zone(end_page));
Mel Gormanc361be52007-10-16 01:25:51 -07001079#endif
1080
1081 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -07001082 /* Make sure we are not inadvertently changing nodes */
Sasha Levin309381fea2014-01-23 15:52:54 -08001083 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
Adam Litke344c7902008-09-02 14:35:38 -07001084
Mel Gormanc361be52007-10-16 01:25:51 -07001085 if (!pfn_valid_within(page_to_pfn(page))) {
1086 page++;
1087 continue;
1088 }
1089
1090 if (!PageBuddy(page)) {
1091 page++;
1092 continue;
1093 }
1094
1095 order = page_order(page);
Kirill A. Shutemov84be48d2011-03-22 16:33:41 -07001096 list_move(&page->lru,
1097 &zone->free_area[order].free_list[migratetype]);
Minchan Kim95e34412012-10-08 16:32:11 -07001098 set_freepage_migratetype(page, migratetype);
Mel Gormanc361be52007-10-16 01:25:51 -07001099 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -07001100 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -07001101 }
1102
Mel Gormand1003132007-10-16 01:26:00 -07001103 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -07001104}
1105
Minchan Kimee6f5092012-07-31 16:43:50 -07001106int move_freepages_block(struct zone *zone, struct page *page,
Linus Torvalds68e3e922012-06-03 20:05:57 -07001107 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001108{
1109 unsigned long start_pfn, end_pfn;
1110 struct page *start_page, *end_page;
1111
1112 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -07001113 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -07001114 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -07001115 end_page = start_page + pageblock_nr_pages - 1;
1116 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -07001117
1118 /* Do not cross zone boundaries */
Cody P Schafer108bcc92013-02-22 16:35:23 -08001119 if (!zone_spans_pfn(zone, start_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001120 start_page = page;
Cody P Schafer108bcc92013-02-22 16:35:23 -08001121 if (!zone_spans_pfn(zone, end_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001122 return 0;
1123
1124 return move_freepages(zone, start_page, end_page, migratetype);
1125}
1126
Mel Gorman2f66a682009-09-21 17:02:31 -07001127static void change_pageblock_range(struct page *pageblock_page,
1128 int start_order, int migratetype)
1129{
1130 int nr_pageblocks = 1 << (start_order - pageblock_order);
1131
1132 while (nr_pageblocks--) {
1133 set_pageblock_migratetype(pageblock_page, migratetype);
1134 pageblock_page += pageblock_nr_pages;
1135 }
1136}
1137
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001138/*
Vlastimil Babka9c0415e2015-02-11 15:28:21 -08001139 * When we are falling back to another migratetype during allocation, try to
1140 * steal extra free pages from the same pageblocks to satisfy further
1141 * allocations, instead of polluting multiple pageblocks.
1142 *
1143 * If we are stealing a relatively large buddy page, it is likely there will
1144 * be more free pages in the pageblock, so try to steal them all. For
1145 * reclaimable and unmovable allocations, we steal regardless of page size,
1146 * as fragmentation caused by those allocations polluting movable pageblocks
1147 * is worse than movable allocations stealing from unmovable and reclaimable
1148 * pageblocks.
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001149 */
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001150static bool can_steal_fallback(unsigned int order, int start_mt)
1151{
1152 /*
1153 * Leaving this order check is intended, although there is
1154 * relaxed order check in next check. The reason is that
1155 * we can actually steal whole pageblock if this condition met,
1156 * but, below check doesn't guarantee it and that is just heuristic
1157 * so could be changed anytime.
1158 */
1159 if (order >= pageblock_order)
1160 return true;
1161
1162 if (order >= pageblock_order / 2 ||
1163 start_mt == MIGRATE_RECLAIMABLE ||
1164 start_mt == MIGRATE_UNMOVABLE ||
1165 page_group_by_mobility_disabled)
1166 return true;
1167
1168 return false;
1169}
1170
1171/*
1172 * This function implements actual steal behaviour. If order is large enough,
1173 * we can steal whole pageblock. If not, we first move freepages in this
1174 * pageblock and check whether half of pages are moved or not. If half of
1175 * pages are moved, we can change migratetype of pageblock and permanently
1176 * use it's pages as requested migratetype in the future.
1177 */
1178static void steal_suitable_fallback(struct zone *zone, struct page *page,
1179 int start_type)
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001180{
1181 int current_order = page_order(page);
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001182 int pages;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001183
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001184 /* Take ownership for orders >= pageblock_order */
1185 if (current_order >= pageblock_order) {
1186 change_pageblock_range(page, current_order, start_type);
Vlastimil Babka3a1086f2015-02-11 15:28:18 -08001187 return;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001188 }
1189
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001190 pages = move_freepages_block(zone, page, start_type);
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001191
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001192 /* Claim the whole block if over half of it is free */
1193 if (pages >= (1 << (pageblock_order-1)) ||
1194 page_group_by_mobility_disabled)
1195 set_pageblock_migratetype(page, start_type);
1196}
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001197
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001198/*
1199 * Check whether there is a suitable fallback freepage with requested order.
1200 * If only_stealable is true, this function returns fallback_mt only if
1201 * we can steal other freepages all together. This would help to reduce
1202 * fragmentation due to mixed migratetype pages in one pageblock.
1203 */
1204int find_suitable_fallback(struct free_area *area, unsigned int order,
1205 int migratetype, bool only_stealable, bool *can_steal)
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001206{
1207 int i;
1208 int fallback_mt;
1209
1210 if (area->nr_free == 0)
1211 return -1;
1212
1213 *can_steal = false;
1214 for (i = 0;; i++) {
1215 fallback_mt = fallbacks[migratetype][i];
1216 if (fallback_mt == MIGRATE_RESERVE)
1217 break;
1218
1219 if (list_empty(&area->free_list[fallback_mt]))
1220 continue;
1221
1222 if (can_steal_fallback(order, migratetype))
1223 *can_steal = true;
1224
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001225 if (!only_stealable)
1226 return fallback_mt;
1227
1228 if (*can_steal)
1229 return fallback_mt;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001230 }
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001231
1232 return -1;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001233}
1234
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001235/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -07001236static inline struct page *
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001237__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001238{
Pintu Kumarb8af2942013-09-11 14:20:34 -07001239 struct free_area *area;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001240 unsigned int current_order;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001241 struct page *page;
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001242 int fallback_mt;
1243 bool can_steal;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001244
1245 /* Find the largest possible block of pages in the other list */
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001246 for (current_order = MAX_ORDER-1;
1247 current_order >= order && current_order <= MAX_ORDER-1;
1248 --current_order) {
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001249 area = &(zone->free_area[current_order]);
1250 fallback_mt = find_suitable_fallback(area, current_order,
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001251 start_migratetype, false, &can_steal);
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001252 if (fallback_mt == -1)
1253 continue;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001254
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001255 page = list_entry(area->free_list[fallback_mt].next,
1256 struct page, lru);
1257 if (can_steal)
1258 steal_suitable_fallback(zone, page, start_migratetype);
Mel Gormane0104872007-10-16 01:25:53 -07001259
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001260 /* Remove the page from the freelists */
1261 area->nr_free--;
1262 list_del(&page->lru);
1263 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001264
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001265 expand(zone, page, order, current_order, area,
1266 start_migratetype);
1267 /*
1268 * The freepage_migratetype may differ from pageblock's
1269 * migratetype depending on the decisions in
1270 * try_to_steal_freepages(). This is OK as long as it
1271 * does not differ for MIGRATE_CMA pageblocks. For CMA
1272 * we need to make sure unallocated pages flushed from
1273 * pcp lists are returned to the correct freelist.
1274 */
1275 set_freepage_migratetype(page, start_migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001276
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001277 trace_mm_page_alloc_extfrag(page, order, current_order,
1278 start_migratetype, fallback_mt);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001279
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001280 return page;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001281 }
1282
Mel Gorman728ec982009-06-16 15:32:04 -07001283 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001284}
1285
Mel Gorman56fd56b2007-10-16 01:25:58 -07001286/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 * Do the hard work of removing an element from the buddy allocator.
1288 * Call me with the zone->lock already held.
1289 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001290static struct page *__rmqueue(struct zone *zone, unsigned int order,
1291 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 struct page *page;
1294
Mel Gorman728ec982009-06-16 15:32:04 -07001295retry_reserve:
Mel Gorman56fd56b2007-10-16 01:25:58 -07001296 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297
Mel Gorman728ec982009-06-16 15:32:04 -07001298 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Joonsoo Kimdc676472015-04-14 15:45:15 -07001299 if (migratetype == MIGRATE_MOVABLE)
1300 page = __rmqueue_cma_fallback(zone, order);
1301
1302 if (!page)
1303 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001304
Mel Gorman728ec982009-06-16 15:32:04 -07001305 /*
1306 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1307 * is used because __rmqueue_smallest is an inline function
1308 * and we want just one call site
1309 */
1310 if (!page) {
1311 migratetype = MIGRATE_RESERVE;
1312 goto retry_reserve;
1313 }
1314 }
1315
Mel Gorman0d3d0622009-09-21 17:02:44 -07001316 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001317 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318}
1319
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001320/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 * Obtain a specified number of elements from the buddy allocator, all under
1322 * a single hold of the lock, for efficiency. Add them to the supplied list.
1323 * Returns the number of new pages which were placed at *list.
1324 */
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001325static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001326 unsigned long count, struct list_head *list,
Mel Gormanb745bc82014-06-04 16:10:22 -07001327 int migratetype, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328{
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001329 int i;
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001330
Nick Pigginc54ad302006-01-06 00:10:56 -08001331 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001333 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001334 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08001336
1337 /*
1338 * Split buddy pages returned by expand() are received here
1339 * in physical page order. The page is added to the callers and
1340 * list and the list head then moves forward. From the callers
1341 * perspective, the linked list is ordered by page number in
1342 * some conditions. This is useful for IO devices that can
1343 * merge IO requests if the physical pages are ordered
1344 * properly.
1345 */
Mel Gormanb745bc82014-06-04 16:10:22 -07001346 if (likely(!cold))
Mel Gormane084b2d2009-07-29 15:02:04 -07001347 list_add(&page->lru, list);
1348 else
1349 list_add_tail(&page->lru, list);
Mel Gorman81eabcb2007-12-17 16:20:05 -08001350 list = &page->lru;
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001351 if (is_migrate_cma(get_freepage_migratetype(page)))
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07001352 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
1353 -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 }
Mel Gormanf2260e62009-06-16 15:32:13 -07001355 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08001356 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001357 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358}
1359
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001360#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001361/*
Christoph Lameter4037d452007-05-09 02:35:14 -07001362 * Called from the vmstat counter updater to drain pagesets of this
1363 * currently executing processor on remote nodes after they have
1364 * expired.
1365 *
Christoph Lameter879336c2006-03-22 00:09:08 -08001366 * Note that this function must be called with the thread pinned to
1367 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001368 */
Christoph Lameter4037d452007-05-09 02:35:14 -07001369void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001370{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001371 unsigned long flags;
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07001372 int to_drain, batch;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001373
Christoph Lameter4037d452007-05-09 02:35:14 -07001374 local_irq_save(flags);
Jason Low4db0c3c2015-04-15 16:14:08 -07001375 batch = READ_ONCE(pcp->batch);
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07001376 to_drain = min(pcp->count, batch);
KOSAKI Motohiro2a135152012-07-31 16:42:53 -07001377 if (to_drain > 0) {
1378 free_pcppages_bulk(zone, to_drain, pcp);
1379 pcp->count -= to_drain;
1380 }
Christoph Lameter4037d452007-05-09 02:35:14 -07001381 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001382}
1383#endif
1384
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001385/*
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001386 * Drain pcplists of the indicated processor and zone.
1387 *
1388 * The processor must either be the current processor and the
1389 * thread pinned to the current processor or a processor that
1390 * is not online.
1391 */
1392static void drain_pages_zone(unsigned int cpu, struct zone *zone)
1393{
1394 unsigned long flags;
1395 struct per_cpu_pageset *pset;
1396 struct per_cpu_pages *pcp;
1397
1398 local_irq_save(flags);
1399 pset = per_cpu_ptr(zone->pageset, cpu);
1400
1401 pcp = &pset->pcp;
1402 if (pcp->count) {
1403 free_pcppages_bulk(zone, pcp->count, pcp);
1404 pcp->count = 0;
1405 }
1406 local_irq_restore(flags);
1407}
1408
1409/*
1410 * Drain pcplists of all zones on the indicated processor.
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001411 *
1412 * The processor must either be the current processor and the
1413 * thread pinned to the current processor or a processor that
1414 * is not online.
1415 */
1416static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417{
1418 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07001420 for_each_populated_zone(zone) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001421 drain_pages_zone(cpu, zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 }
1423}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001425/*
1426 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001427 *
1428 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
1429 * the single zone's pages.
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001430 */
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001431void drain_local_pages(struct zone *zone)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001432{
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001433 int cpu = smp_processor_id();
1434
1435 if (zone)
1436 drain_pages_zone(cpu, zone);
1437 else
1438 drain_pages(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001439}
1440
1441/*
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001442 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1443 *
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001444 * When zone parameter is non-NULL, spill just the single zone's pages.
1445 *
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001446 * Note that this code is protected against sending an IPI to an offline
1447 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1448 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1449 * nothing keeps CPUs from showing up after we populated the cpumask and
1450 * before the call to on_each_cpu_mask().
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001451 */
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001452void drain_all_pages(struct zone *zone)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001453{
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001454 int cpu;
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001455
1456 /*
1457 * Allocate in the BSS so we wont require allocation in
1458 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1459 */
1460 static cpumask_t cpus_with_pcps;
1461
1462 /*
1463 * We don't care about racing with CPU hotplug event
1464 * as offline notification will cause the notified
1465 * cpu to drain that CPU pcps and on_each_cpu_mask
1466 * disables preemption as part of its processing
1467 */
1468 for_each_online_cpu(cpu) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001469 struct per_cpu_pageset *pcp;
1470 struct zone *z;
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001471 bool has_pcps = false;
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001472
1473 if (zone) {
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001474 pcp = per_cpu_ptr(zone->pageset, cpu);
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001475 if (pcp->pcp.count)
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001476 has_pcps = true;
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001477 } else {
1478 for_each_populated_zone(z) {
1479 pcp = per_cpu_ptr(z->pageset, cpu);
1480 if (pcp->pcp.count) {
1481 has_pcps = true;
1482 break;
1483 }
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001484 }
1485 }
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001486
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001487 if (has_pcps)
1488 cpumask_set_cpu(cpu, &cpus_with_pcps);
1489 else
1490 cpumask_clear_cpu(cpu, &cpus_with_pcps);
1491 }
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001492 on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages,
1493 zone, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001494}
1495
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02001496#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497
1498void mark_free_pages(struct zone *zone)
1499{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001500 unsigned long pfn, max_zone_pfn;
1501 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001502 unsigned int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 struct list_head *curr;
1504
Xishi Qiu8080fc02013-09-11 14:21:45 -07001505 if (zone_is_empty(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 return;
1507
1508 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001509
Cody P Schafer108bcc92013-02-22 16:35:23 -08001510 max_zone_pfn = zone_end_pfn(zone);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001511 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1512 if (pfn_valid(pfn)) {
1513 struct page *page = pfn_to_page(pfn);
1514
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001515 if (!swsusp_page_is_forbidden(page))
1516 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001517 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001519 for_each_migratetype_order(order, t) {
1520 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001521 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001523 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1524 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001525 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001526 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001527 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 spin_unlock_irqrestore(&zone->lock, flags);
1529}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001530#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531
1532/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 * Free a 0-order page
Mel Gormanb745bc82014-06-04 16:10:22 -07001534 * cold == true ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 */
Mel Gormanb745bc82014-06-04 16:10:22 -07001536void free_hot_cold_page(struct page *page, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537{
1538 struct zone *zone = page_zone(page);
1539 struct per_cpu_pages *pcp;
1540 unsigned long flags;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001541 unsigned long pfn = page_to_pfn(page);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001542 int migratetype;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001544 if (!free_pages_prepare(page, 0))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001545 return;
1546
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001547 migratetype = get_pfnblock_migratetype(page, pfn);
Minchan Kimb12c4ad2012-10-08 16:32:08 -07001548 set_freepage_migratetype(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001550 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001551
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001552 /*
1553 * We only track unmovable, reclaimable and movable on pcp lists.
1554 * Free ISOLATE pages back to the allocator because they are being
1555 * offlined but treat RESERVE as movable pages so we can get those
1556 * areas back if necessary. Otherwise, we may have to free
1557 * excessively into the page allocator
1558 */
1559 if (migratetype >= MIGRATE_PCPTYPES) {
Minchan Kim194159f2013-02-22 16:33:58 -08001560 if (unlikely(is_migrate_isolate(migratetype))) {
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001561 free_one_page(zone, page, pfn, 0, migratetype);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001562 goto out;
1563 }
1564 migratetype = MIGRATE_MOVABLE;
1565 }
1566
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001567 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Mel Gormanb745bc82014-06-04 16:10:22 -07001568 if (!cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001569 list_add(&page->lru, &pcp->lists[migratetype]);
Mel Gormanb745bc82014-06-04 16:10:22 -07001570 else
1571 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001573 if (pcp->count >= pcp->high) {
Jason Low4db0c3c2015-04-15 16:14:08 -07001574 unsigned long batch = READ_ONCE(pcp->batch);
Cody P Schafer998d39c2013-07-03 15:01:32 -07001575 free_pcppages_bulk(zone, batch, pcp);
1576 pcp->count -= batch;
Nick Piggin48db57f2006-01-08 01:00:42 -08001577 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001578
1579out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581}
1582
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001583/*
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001584 * Free a list of 0-order pages
1585 */
Mel Gormanb745bc82014-06-04 16:10:22 -07001586void free_hot_cold_page_list(struct list_head *list, bool cold)
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001587{
1588 struct page *page, *next;
1589
1590 list_for_each_entry_safe(page, next, list, lru) {
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -08001591 trace_mm_page_free_batched(page, cold);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001592 free_hot_cold_page(page, cold);
1593 }
1594}
1595
1596/*
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001597 * split_page takes a non-compound higher-order page, and splits it into
1598 * n (1<<order) sub-pages: page[0..n]
1599 * Each sub-page must be freed individually.
1600 *
1601 * Note: this is probably too low level an operation for use in drivers.
1602 * Please consult with lkml before using this in your driver.
1603 */
1604void split_page(struct page *page, unsigned int order)
1605{
1606 int i;
1607
Sasha Levin309381fea2014-01-23 15:52:54 -08001608 VM_BUG_ON_PAGE(PageCompound(page), page);
1609 VM_BUG_ON_PAGE(!page_count(page), page);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001610
1611#ifdef CONFIG_KMEMCHECK
1612 /*
1613 * Split shadow pages too, because free(page[0]) would
1614 * otherwise free the whole shadow.
1615 */
1616 if (kmemcheck_page_is_tracked(page))
1617 split_page(virt_to_page(page[0].shadow), order);
1618#endif
1619
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001620 set_page_owner(page, 0, 0);
1621 for (i = 1; i < (1 << order); i++) {
Nick Piggin7835e982006-03-22 00:08:40 -08001622 set_page_refcounted(page + i);
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001623 set_page_owner(page + i, 0, 0);
1624 }
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001625}
K. Y. Srinivasan5853ff22013-03-25 15:47:38 -07001626EXPORT_SYMBOL_GPL(split_page);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001627
Joonsoo Kim3c605092014-11-13 15:19:21 -08001628int __isolate_free_page(struct page *page, unsigned int order)
Mel Gorman748446b2010-05-24 14:32:27 -07001629{
Mel Gorman748446b2010-05-24 14:32:27 -07001630 unsigned long watermark;
1631 struct zone *zone;
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07001632 int mt;
Mel Gorman748446b2010-05-24 14:32:27 -07001633
1634 BUG_ON(!PageBuddy(page));
1635
1636 zone = page_zone(page);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001637 mt = get_pageblock_migratetype(page);
Mel Gorman748446b2010-05-24 14:32:27 -07001638
Minchan Kim194159f2013-02-22 16:33:58 -08001639 if (!is_migrate_isolate(mt)) {
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001640 /* Obey watermarks as if the page was being allocated */
1641 watermark = low_wmark_pages(zone) + (1 << order);
1642 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1643 return 0;
1644
Mel Gorman8fb74b92013-01-11 14:32:16 -08001645 __mod_zone_freepage_state(zone, -(1UL << order), mt);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001646 }
Mel Gorman748446b2010-05-24 14:32:27 -07001647
1648 /* Remove page from free list */
1649 list_del(&page->lru);
1650 zone->free_area[order].nr_free--;
1651 rmv_page_order(page);
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07001652
Mel Gorman8fb74b92013-01-11 14:32:16 -08001653 /* Set the pageblock if the isolated page is at least a pageblock */
Mel Gorman748446b2010-05-24 14:32:27 -07001654 if (order >= pageblock_order - 1) {
1655 struct page *endpage = page + (1 << order) - 1;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001656 for (; page < endpage; page += pageblock_nr_pages) {
1657 int mt = get_pageblock_migratetype(page);
Minchan Kim194159f2013-02-22 16:33:58 -08001658 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001659 set_pageblock_migratetype(page,
1660 MIGRATE_MOVABLE);
1661 }
Mel Gorman748446b2010-05-24 14:32:27 -07001662 }
1663
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001664 set_page_owner(page, order, 0);
Mel Gorman8fb74b92013-01-11 14:32:16 -08001665 return 1UL << order;
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001666}
1667
1668/*
1669 * Similar to split_page except the page is already free. As this is only
1670 * being used for migration, the migratetype of the block also changes.
1671 * As this is called with interrupts disabled, the caller is responsible
1672 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1673 * are enabled.
1674 *
1675 * Note: this is probably too low level an operation for use in drivers.
1676 * Please consult with lkml before using this in your driver.
1677 */
1678int split_free_page(struct page *page)
1679{
1680 unsigned int order;
1681 int nr_pages;
1682
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001683 order = page_order(page);
1684
Mel Gorman8fb74b92013-01-11 14:32:16 -08001685 nr_pages = __isolate_free_page(page, order);
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001686 if (!nr_pages)
1687 return 0;
1688
1689 /* Split into individual pages */
1690 set_page_refcounted(page);
1691 split_page(page, order);
1692 return nr_pages;
Mel Gorman748446b2010-05-24 14:32:27 -07001693}
1694
1695/*
Vlastimil Babka75379192015-02-11 15:25:38 -08001696 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07001698static inline
1699struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001700 struct zone *zone, unsigned int order,
1701 gfp_t gfp_flags, int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702{
1703 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001704 struct page *page;
Mel Gormanb745bc82014-06-04 16:10:22 -07001705 bool cold = ((gfp_flags & __GFP_COLD) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706
Nick Piggin48db57f2006-01-08 01:00:42 -08001707 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001709 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 local_irq_save(flags);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001712 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1713 list = &pcp->lists[migratetype];
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001714 if (list_empty(list)) {
Mel Gorman535131e62007-10-16 01:25:49 -07001715 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001716 pcp->batch, list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001717 migratetype, cold);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001718 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07001719 goto failed;
Mel Gorman535131e62007-10-16 01:25:49 -07001720 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001721
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001722 if (cold)
1723 page = list_entry(list->prev, struct page, lru);
1724 else
1725 page = list_entry(list->next, struct page, lru);
1726
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001727 list_del(&page->lru);
1728 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001729 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07001730 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1731 /*
1732 * __GFP_NOFAIL is not to be used in new code.
1733 *
1734 * All __GFP_NOFAIL callers should be fixed so that they
1735 * properly detect and handle allocation failures.
1736 *
1737 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07001738 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07001739 * __GFP_NOFAIL.
1740 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07001741 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07001742 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001744 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001745 spin_unlock(&zone->lock);
1746 if (!page)
1747 goto failed;
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07001748 __mod_zone_freepage_state(zone, -(1 << order),
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001749 get_freepage_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 }
1751
Johannes Weiner3a025762014-04-07 15:37:48 -07001752 __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
Johannes Weinerabe5f972014-10-02 16:21:10 -07001753 if (atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]) <= 0 &&
Johannes Weiner57054652014-10-09 15:28:17 -07001754 !test_bit(ZONE_FAIR_DEPLETED, &zone->flags))
1755 set_bit(ZONE_FAIR_DEPLETED, &zone->flags);
Johannes Weiner27329362014-03-03 15:38:41 -08001756
Christoph Lameterf8891e52006-06-30 01:55:45 -07001757 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Andi Kleen78afd562011-03-22 16:33:12 -07001758 zone_statistics(preferred_zone, zone, gfp_flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001759 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760
Sasha Levin309381fea2014-01-23 15:52:54 -08001761 VM_BUG_ON_PAGE(bad_range(zone, page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001763
1764failed:
1765 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001766 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767}
1768
Akinobu Mita933e3122006-12-08 02:39:45 -08001769#ifdef CONFIG_FAIL_PAGE_ALLOC
1770
Akinobu Mitab2588c42011-07-26 16:09:03 -07001771static struct {
Akinobu Mita933e3122006-12-08 02:39:45 -08001772 struct fault_attr attr;
1773
1774 u32 ignore_gfp_highmem;
1775 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001776 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001777} fail_page_alloc = {
1778 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001779 .ignore_gfp_wait = 1,
1780 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001781 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001782};
1783
1784static int __init setup_fail_page_alloc(char *str)
1785{
1786 return setup_fault_attr(&fail_page_alloc.attr, str);
1787}
1788__setup("fail_page_alloc=", setup_fail_page_alloc);
1789
Gavin Shandeaf3862012-07-31 16:41:51 -07001790static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08001791{
Akinobu Mita54114992007-07-15 23:40:23 -07001792 if (order < fail_page_alloc.min_order)
Gavin Shandeaf3862012-07-31 16:41:51 -07001793 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001794 if (gfp_mask & __GFP_NOFAIL)
Gavin Shandeaf3862012-07-31 16:41:51 -07001795 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001796 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
Gavin Shandeaf3862012-07-31 16:41:51 -07001797 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001798 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
Gavin Shandeaf3862012-07-31 16:41:51 -07001799 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001800
1801 return should_fail(&fail_page_alloc.attr, 1 << order);
1802}
1803
1804#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1805
1806static int __init fail_page_alloc_debugfs(void)
1807{
Al Virof4ae40a2011-07-24 04:33:43 -04001808 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
Akinobu Mita933e3122006-12-08 02:39:45 -08001809 struct dentry *dir;
Akinobu Mita933e3122006-12-08 02:39:45 -08001810
Akinobu Mitadd48c082011-08-03 16:21:01 -07001811 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
1812 &fail_page_alloc.attr);
1813 if (IS_ERR(dir))
1814 return PTR_ERR(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08001815
Akinobu Mitab2588c42011-07-26 16:09:03 -07001816 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
1817 &fail_page_alloc.ignore_gfp_wait))
1818 goto fail;
1819 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1820 &fail_page_alloc.ignore_gfp_highmem))
1821 goto fail;
1822 if (!debugfs_create_u32("min-order", mode, dir,
1823 &fail_page_alloc.min_order))
1824 goto fail;
Akinobu Mita933e3122006-12-08 02:39:45 -08001825
Akinobu Mitab2588c42011-07-26 16:09:03 -07001826 return 0;
1827fail:
Akinobu Mitadd48c082011-08-03 16:21:01 -07001828 debugfs_remove_recursive(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08001829
Akinobu Mitab2588c42011-07-26 16:09:03 -07001830 return -ENOMEM;
Akinobu Mita933e3122006-12-08 02:39:45 -08001831}
1832
1833late_initcall(fail_page_alloc_debugfs);
1834
1835#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1836
1837#else /* CONFIG_FAIL_PAGE_ALLOC */
1838
Gavin Shandeaf3862012-07-31 16:41:51 -07001839static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08001840{
Gavin Shandeaf3862012-07-31 16:41:51 -07001841 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001842}
1843
1844#endif /* CONFIG_FAIL_PAGE_ALLOC */
1845
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846/*
Mel Gorman88f5acf2011-01-13 15:45:41 -08001847 * Return true if free pages are above 'mark'. This takes into account the order
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 * of the allocation.
1849 */
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001850static bool __zone_watermark_ok(struct zone *z, unsigned int order,
1851 unsigned long mark, int classzone_idx, int alloc_flags,
1852 long free_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853{
Wei Yuan26086de2014-12-10 15:44:44 -08001854 /* free_pages may go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001855 long min = mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 int o;
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07001857 long free_cma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858
Michal Hockodf0a6da2012-01-10 15:08:02 -08001859 free_pages -= (1 << order) - 1;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001860 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001862 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 min -= min / 4;
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07001864#ifdef CONFIG_CMA
1865 /* If allocation can't use CMA areas don't use free CMA pages */
1866 if (!(alloc_flags & ALLOC_CMA))
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07001867 free_cma = zone_page_state(z, NR_FREE_CMA_PAGES);
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07001868#endif
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07001869
Mel Gorman3484b2d2014-08-06 16:07:14 -07001870 if (free_pages - free_cma <= min + z->lowmem_reserve[classzone_idx])
Mel Gorman88f5acf2011-01-13 15:45:41 -08001871 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 for (o = 0; o < order; o++) {
1873 /* At the next order, this order's pages become unavailable */
1874 free_pages -= z->free_area[o].nr_free << o;
1875
1876 /* Require fewer higher order pages to be free */
Arve Hjønnevågc3864002009-02-17 14:51:02 -08001877 min >>= min_free_order_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878
1879 if (free_pages <= min)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001880 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 }
Mel Gorman88f5acf2011-01-13 15:45:41 -08001882 return true;
1883}
1884
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001885bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
Mel Gorman88f5acf2011-01-13 15:45:41 -08001886 int classzone_idx, int alloc_flags)
1887{
1888 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1889 zone_page_state(z, NR_FREE_PAGES));
1890}
1891
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001892bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
1893 unsigned long mark, int classzone_idx, int alloc_flags)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001894{
1895 long free_pages = zone_page_state(z, NR_FREE_PAGES);
1896
1897 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1898 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1899
1900 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1901 free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902}
1903
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001904#ifdef CONFIG_NUMA
1905/*
1906 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1907 * skip over zones that are not allowed by the cpuset, or that have
1908 * been recently (in last second) found to be nearly full. See further
1909 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001910 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001911 *
Zhi Yong Wua1aeb652013-11-12 15:08:29 -08001912 * If the zonelist cache is present in the passed zonelist, then
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001913 * returns a pointer to the allowed node mask (either the current
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08001914 * tasks mems_allowed, or node_states[N_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001915 *
1916 * If the zonelist cache is not available for this zonelist, does
1917 * nothing and returns NULL.
1918 *
1919 * If the fullzones BITMAP in the zonelist cache is stale (more than
1920 * a second since last zap'd) then we zap it out (clear its bits.)
1921 *
1922 * We hold off even calling zlc_setup, until after we've checked the
1923 * first zone in the zonelist, on the theory that most allocations will
1924 * be satisfied from that first zone, so best to examine that zone as
1925 * quickly as we can.
1926 */
1927static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1928{
1929 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1930 nodemask_t *allowednodes; /* zonelist_cache approximation */
1931
1932 zlc = zonelist->zlcache_ptr;
1933 if (!zlc)
1934 return NULL;
1935
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001936 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001937 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1938 zlc->last_full_zap = jiffies;
1939 }
1940
1941 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1942 &cpuset_current_mems_allowed :
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08001943 &node_states[N_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001944 return allowednodes;
1945}
1946
1947/*
1948 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1949 * if it is worth looking at further for free memory:
1950 * 1) Check that the zone isn't thought to be full (doesn't have its
1951 * bit set in the zonelist_cache fullzones BITMAP).
1952 * 2) Check that the zones node (obtained from the zonelist_cache
1953 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1954 * Return true (non-zero) if zone is worth looking at further, or
1955 * else return false (zero) if it is not.
1956 *
1957 * This check -ignores- the distinction between various watermarks,
1958 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1959 * found to be full for any variation of these watermarks, it will
1960 * be considered full for up to one second by all requests, unless
1961 * we are so low on memory on all allowed nodes that we are forced
1962 * into the second scan of the zonelist.
1963 *
1964 * In the second scan we ignore this zonelist cache and exactly
1965 * apply the watermarks to all zones, even it is slower to do so.
1966 * We are low on memory in the second scan, and should leave no stone
1967 * unturned looking for a free page.
1968 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001969static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001970 nodemask_t *allowednodes)
1971{
1972 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1973 int i; /* index of *z in zonelist zones */
1974 int n; /* node that zone *z is on */
1975
1976 zlc = zonelist->zlcache_ptr;
1977 if (!zlc)
1978 return 1;
1979
Mel Gormandd1a2392008-04-28 02:12:17 -07001980 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001981 n = zlc->z_to_n[i];
1982
1983 /* This zone is worth trying if it is allowed but not full */
1984 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1985}
1986
1987/*
1988 * Given 'z' scanning a zonelist, set the corresponding bit in
1989 * zlc->fullzones, so that subsequent attempts to allocate a page
1990 * from that zone don't waste time re-examining it.
1991 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001992static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001993{
1994 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1995 int i; /* index of *z in zonelist zones */
1996
1997 zlc = zonelist->zlcache_ptr;
1998 if (!zlc)
1999 return;
2000
Mel Gormandd1a2392008-04-28 02:12:17 -07002001 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002002
2003 set_bit(i, zlc->fullzones);
2004}
2005
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002006/*
2007 * clear all zones full, called after direct reclaim makes progress so that
2008 * a zone that was recently full is not skipped over for up to a second
2009 */
2010static void zlc_clear_zones_full(struct zonelist *zonelist)
2011{
2012 struct zonelist_cache *zlc; /* cached zonelist speedup info */
2013
2014 zlc = zonelist->zlcache_ptr;
2015 if (!zlc)
2016 return;
2017
2018 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2019}
2020
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002021static bool zone_local(struct zone *local_zone, struct zone *zone)
2022{
Johannes Weinerfff40682013-12-20 14:54:12 +00002023 return local_zone->node == zone->node;
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002024}
2025
David Rientjes957f8222012-10-08 16:33:24 -07002026static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2027{
Mel Gorman5f7a75a2014-06-04 16:07:15 -07002028 return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <
2029 RECLAIM_DISTANCE;
David Rientjes957f8222012-10-08 16:33:24 -07002030}
2031
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002032#else /* CONFIG_NUMA */
2033
2034static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
2035{
2036 return NULL;
2037}
2038
Mel Gormandd1a2392008-04-28 02:12:17 -07002039static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002040 nodemask_t *allowednodes)
2041{
2042 return 1;
2043}
2044
Mel Gormandd1a2392008-04-28 02:12:17 -07002045static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002046{
2047}
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002048
2049static void zlc_clear_zones_full(struct zonelist *zonelist)
2050{
2051}
David Rientjes957f8222012-10-08 16:33:24 -07002052
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002053static bool zone_local(struct zone *local_zone, struct zone *zone)
2054{
2055 return true;
2056}
2057
David Rientjes957f8222012-10-08 16:33:24 -07002058static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2059{
2060 return true;
2061}
2062
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002063#endif /* CONFIG_NUMA */
2064
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002065static void reset_alloc_batches(struct zone *preferred_zone)
2066{
2067 struct zone *zone = preferred_zone->zone_pgdat->node_zones;
2068
2069 do {
2070 mod_zone_page_state(zone, NR_ALLOC_BATCH,
2071 high_wmark_pages(zone) - low_wmark_pages(zone) -
2072 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
Johannes Weiner57054652014-10-09 15:28:17 -07002073 clear_bit(ZONE_FAIR_DEPLETED, &zone->flags);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002074 } while (zone++ != preferred_zone);
2075}
2076
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002077/*
Paul Jackson0798e512006-12-06 20:31:38 -08002078 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002079 * a page.
2080 */
2081static struct page *
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002082get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
2083 const struct alloc_context *ac)
Martin Hicks753ee722005-06-21 17:14:41 -07002084{
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002085 struct zonelist *zonelist = ac->zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07002086 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002087 struct page *page = NULL;
Mel Gorman5117f452009-06-16 15:31:59 -07002088 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002089 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
2090 int zlc_active = 0; /* set if using zonelist_cache */
2091 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gormana6e21b142014-06-04 16:10:12 -07002092 bool consider_zone_dirty = (alloc_flags & ALLOC_WMARK_LOW) &&
2093 (gfp_mask & __GFP_WRITE);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002094 int nr_fair_skipped = 0;
2095 bool zonelist_rescan;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002096
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002097zonelist_scan:
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002098 zonelist_rescan = false;
2099
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002100 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002101 * Scan zonelist, looking for a zone with enough free.
Vladimir Davydov344736f2014-10-20 15:50:30 +04002102 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002103 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002104 for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
2105 ac->nodemask) {
Johannes Weinere085dbc2013-09-11 14:20:46 -07002106 unsigned long mark;
2107
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002108 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002109 !zlc_zone_worth_trying(zonelist, z, allowednodes))
2110 continue;
Mel Gorman664eedd2014-06-04 16:10:08 -07002111 if (cpusets_enabled() &&
2112 (alloc_flags & ALLOC_CPUSET) &&
Vladimir Davydov344736f2014-10-20 15:50:30 +04002113 !cpuset_zone_allowed(zone, gfp_mask))
Mel Gormancd38b112011-07-25 17:12:29 -07002114 continue;
Johannes Weinera756cf52012-01-10 15:07:49 -08002115 /*
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002116 * Distribute pages in proportion to the individual
2117 * zone size to ensure fair page aging. The zone a
2118 * page was allocated in should have no effect on the
2119 * time the page has in memory before being reclaimed.
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002120 */
Johannes Weiner3a025762014-04-07 15:37:48 -07002121 if (alloc_flags & ALLOC_FAIR) {
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002122 if (!zone_local(ac->preferred_zone, zone))
Mel Gormanf7b5d642014-08-06 16:07:20 -07002123 break;
Johannes Weiner57054652014-10-09 15:28:17 -07002124 if (test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) {
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002125 nr_fair_skipped++;
Johannes Weiner3a025762014-04-07 15:37:48 -07002126 continue;
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002127 }
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002128 }
2129 /*
Johannes Weinera756cf52012-01-10 15:07:49 -08002130 * When allocating a page cache page for writing, we
2131 * want to get it from a zone that is within its dirty
2132 * limit, such that no single zone holds more than its
2133 * proportional share of globally allowed dirty pages.
2134 * The dirty limits take into account the zone's
2135 * lowmem reserves and high watermark so that kswapd
2136 * should be able to balance it without having to
2137 * write pages from its LRU list.
2138 *
2139 * This may look like it could increase pressure on
2140 * lower zones by failing allocations in higher zones
2141 * before they are full. But the pages that do spill
2142 * over are limited as the lower zones are protected
2143 * by this very same mechanism. It should not become
2144 * a practical burden to them.
2145 *
2146 * XXX: For now, allow allocations to potentially
2147 * exceed the per-zone dirty limit in the slowpath
2148 * (ALLOC_WMARK_LOW unset) before going into reclaim,
2149 * which is important when on a NUMA setup the allowed
2150 * zones are together not big enough to reach the
2151 * global limit. The proper fix for these situations
2152 * will require awareness of zones in the
2153 * dirty-throttling and the flusher threads.
2154 */
Mel Gormana6e21b142014-06-04 16:10:12 -07002155 if (consider_zone_dirty && !zone_dirty_ok(zone))
Mel Gorman800a1e72014-06-04 16:10:06 -07002156 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002157
Johannes Weinere085dbc2013-09-11 14:20:46 -07002158 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
2159 if (!zone_watermark_ok(zone, order, mark,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002160 ac->classzone_idx, alloc_flags)) {
Mel Gormanfa5e0842009-06-16 15:33:22 -07002161 int ret;
2162
Mel Gorman5dab2912014-06-04 16:10:14 -07002163 /* Checked here to keep the fast path fast */
2164 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
2165 if (alloc_flags & ALLOC_NO_WATERMARKS)
2166 goto try_this_zone;
2167
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002168 if (IS_ENABLED(CONFIG_NUMA) &&
2169 !did_zlc_setup && nr_online_nodes > 1) {
Mel Gormancd38b112011-07-25 17:12:29 -07002170 /*
2171 * we do zlc_setup if there are multiple nodes
2172 * and before considering the first zone allowed
2173 * by the cpuset.
2174 */
2175 allowednodes = zlc_setup(zonelist, alloc_flags);
2176 zlc_active = 1;
2177 did_zlc_setup = 1;
2178 }
2179
David Rientjes957f8222012-10-08 16:33:24 -07002180 if (zone_reclaim_mode == 0 ||
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002181 !zone_allows_reclaim(ac->preferred_zone, zone))
Mel Gormanfa5e0842009-06-16 15:33:22 -07002182 goto this_zone_full;
2183
Mel Gormancd38b112011-07-25 17:12:29 -07002184 /*
2185 * As we may have just activated ZLC, check if the first
2186 * eligible zone has failed zone_reclaim recently.
2187 */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002188 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
Mel Gormancd38b112011-07-25 17:12:29 -07002189 !zlc_zone_worth_trying(zonelist, z, allowednodes))
2190 continue;
2191
Mel Gormanfa5e0842009-06-16 15:33:22 -07002192 ret = zone_reclaim(zone, gfp_mask, order);
2193 switch (ret) {
2194 case ZONE_RECLAIM_NOSCAN:
2195 /* did not scan */
Mel Gormancd38b112011-07-25 17:12:29 -07002196 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002197 case ZONE_RECLAIM_FULL:
2198 /* scanned but unreclaimable */
Mel Gormancd38b112011-07-25 17:12:29 -07002199 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002200 default:
2201 /* did we reclaim enough */
Mel Gormanfed27192013-04-29 15:07:57 -07002202 if (zone_watermark_ok(zone, order, mark,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002203 ac->classzone_idx, alloc_flags))
Mel Gormanfed27192013-04-29 15:07:57 -07002204 goto try_this_zone;
2205
2206 /*
2207 * Failed to reclaim enough to meet watermark.
2208 * Only mark the zone full if checking the min
2209 * watermark or if we failed to reclaim just
2210 * 1<<order pages or else the page allocator
2211 * fastpath will prematurely mark zones full
2212 * when the watermark is between the low and
2213 * min watermarks.
2214 */
2215 if (((alloc_flags & ALLOC_WMARK_MASK) == ALLOC_WMARK_MIN) ||
2216 ret == ZONE_RECLAIM_SOME)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002217 goto this_zone_full;
Mel Gormanfed27192013-04-29 15:07:57 -07002218
2219 continue;
Paul Jackson0798e512006-12-06 20:31:38 -08002220 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002221 }
2222
Mel Gormanfa5e0842009-06-16 15:33:22 -07002223try_this_zone:
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002224 page = buffered_rmqueue(ac->preferred_zone, zone, order,
2225 gfp_mask, ac->migratetype);
Vlastimil Babka75379192015-02-11 15:25:38 -08002226 if (page) {
2227 if (prep_new_page(page, order, gfp_mask, alloc_flags))
2228 goto try_this_zone;
2229 return page;
2230 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002231this_zone_full:
Mel Gorman65bb3712014-06-04 16:10:05 -07002232 if (IS_ENABLED(CONFIG_NUMA) && zlc_active)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002233 zlc_mark_zone_full(zonelist, z);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002234 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002235
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002236 /*
2237 * The first pass makes sure allocations are spread fairly within the
2238 * local node. However, the local node might have free pages left
2239 * after the fairness batches are exhausted, and remote zones haven't
2240 * even been considered yet. Try once more without fairness, and
2241 * include remote zones now, before entering the slowpath and waking
2242 * kswapd: prefer spilling to a remote zone over swapping locally.
2243 */
2244 if (alloc_flags & ALLOC_FAIR) {
2245 alloc_flags &= ~ALLOC_FAIR;
2246 if (nr_fair_skipped) {
2247 zonelist_rescan = true;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002248 reset_alloc_batches(ac->preferred_zone);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002249 }
2250 if (nr_online_nodes > 1)
2251 zonelist_rescan = true;
2252 }
2253
2254 if (unlikely(IS_ENABLED(CONFIG_NUMA) && zlc_active)) {
2255 /* Disable zlc cache for second zonelist scan */
2256 zlc_active = 0;
2257 zonelist_rescan = true;
2258 }
2259
2260 if (zonelist_rescan)
2261 goto zonelist_scan;
2262
2263 return NULL;
Martin Hicks753ee722005-06-21 17:14:41 -07002264}
2265
David Rientjes29423e772011-03-22 16:30:47 -07002266/*
2267 * Large machines with many possible nodes should not always dump per-node
2268 * meminfo in irq context.
2269 */
2270static inline bool should_suppress_show_mem(void)
2271{
2272 bool ret = false;
2273
2274#if NODES_SHIFT > 8
2275 ret = in_interrupt();
2276#endif
2277 return ret;
2278}
2279
Dave Hansena238ab52011-05-24 17:12:16 -07002280static DEFINE_RATELIMIT_STATE(nopage_rs,
2281 DEFAULT_RATELIMIT_INTERVAL,
2282 DEFAULT_RATELIMIT_BURST);
2283
2284void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
2285{
Dave Hansena238ab52011-05-24 17:12:16 -07002286 unsigned int filter = SHOW_MEM_FILTER_NODES;
2287
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -08002288 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2289 debug_guardpage_minorder() > 0)
Dave Hansena238ab52011-05-24 17:12:16 -07002290 return;
2291
2292 /*
2293 * This documents exceptions given to allocations in certain
2294 * contexts that are allowed to allocate outside current's set
2295 * of allowed nodes.
2296 */
2297 if (!(gfp_mask & __GFP_NOMEMALLOC))
2298 if (test_thread_flag(TIF_MEMDIE) ||
2299 (current->flags & (PF_MEMALLOC | PF_EXITING)))
2300 filter &= ~SHOW_MEM_FILTER_NODES;
2301 if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
2302 filter &= ~SHOW_MEM_FILTER_NODES;
2303
2304 if (fmt) {
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002305 struct va_format vaf;
2306 va_list args;
2307
Dave Hansena238ab52011-05-24 17:12:16 -07002308 va_start(args, fmt);
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002309
2310 vaf.fmt = fmt;
2311 vaf.va = &args;
2312
2313 pr_warn("%pV", &vaf);
2314
Dave Hansena238ab52011-05-24 17:12:16 -07002315 va_end(args);
2316 }
2317
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002318 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2319 current->comm, order, gfp_mask);
Dave Hansena238ab52011-05-24 17:12:16 -07002320
2321 dump_stack();
2322 if (!should_suppress_show_mem())
2323 show_mem(filter);
2324}
2325
Mel Gorman11e33f62009-06-16 15:31:57 -07002326static inline int
2327should_alloc_retry(gfp_t gfp_mask, unsigned int order,
Mel Gormanf90ac392012-01-10 15:07:15 -08002328 unsigned long did_some_progress,
Mel Gorman11e33f62009-06-16 15:31:57 -07002329 unsigned long pages_reclaimed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330{
Mel Gorman11e33f62009-06-16 15:31:57 -07002331 /* Do not loop if specifically requested */
2332 if (gfp_mask & __GFP_NORETRY)
2333 return 0;
2334
Mel Gormanf90ac392012-01-10 15:07:15 -08002335 /* Always retry if specifically requested */
2336 if (gfp_mask & __GFP_NOFAIL)
2337 return 1;
2338
2339 /*
2340 * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim
2341 * making forward progress without invoking OOM. Suspend also disables
2342 * storage devices so kswapd will not help. Bail if we are suspending.
2343 */
2344 if (!did_some_progress && pm_suspended_storage())
2345 return 0;
2346
Mel Gorman11e33f62009-06-16 15:31:57 -07002347 /*
2348 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
2349 * means __GFP_NOFAIL, but that may not be true in other
2350 * implementations.
2351 */
2352 if (order <= PAGE_ALLOC_COSTLY_ORDER)
2353 return 1;
2354
2355 /*
2356 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
2357 * specified, then we retry until we no longer reclaim any pages
2358 * (above), or we've reclaimed an order of pages at least as
2359 * large as the allocation's order. In both cases, if the
2360 * allocation still fails, we stop retrying.
2361 */
2362 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
2363 return 1;
2364
Mel Gorman11e33f62009-06-16 15:31:57 -07002365 return 0;
2366}
2367
2368static inline struct page *
2369__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002370 const struct alloc_context *ac, unsigned long *did_some_progress)
Mel Gorman11e33f62009-06-16 15:31:57 -07002371{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373
Johannes Weiner9879de72015-01-26 12:58:32 -08002374 *did_some_progress = 0;
2375
Johannes Weiner9879de72015-01-26 12:58:32 -08002376 /*
2377 * Acquire the per-zone oom lock for each zone. If that
2378 * fails, somebody else is making progress for us.
2379 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002380 if (!oom_zonelist_trylock(ac->zonelist, gfp_mask)) {
Johannes Weiner9879de72015-01-26 12:58:32 -08002381 *did_some_progress = 1;
Mel Gorman11e33f62009-06-16 15:31:57 -07002382 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383 return NULL;
2384 }
Jens Axboe6b1de912005-11-17 21:35:02 +01002385
Mel Gorman11e33f62009-06-16 15:31:57 -07002386 /*
2387 * Go through the zonelist yet one more time, keep very high watermark
2388 * here, this is only to catch a parallel oom killing, we must fail if
2389 * we're still under heavy pressure.
2390 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002391 page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order,
2392 ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002393 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07002394 goto out;
2395
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002396 if (!(gfp_mask & __GFP_NOFAIL)) {
Johannes Weiner9879de72015-01-26 12:58:32 -08002397 /* Coredumps can quickly deplete all memory reserves */
2398 if (current->flags & PF_DUMPCORE)
2399 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002400 /* The OOM killer will not help higher order allocs */
2401 if (order > PAGE_ALLOC_COSTLY_ORDER)
2402 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07002403 /* The OOM killer does not needlessly kill tasks for lowmem */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002404 if (ac->high_zoneidx < ZONE_NORMAL)
David Rientjes03668b32010-08-09 17:18:54 -07002405 goto out;
Johannes Weiner9879de72015-01-26 12:58:32 -08002406 /* The OOM killer does not compensate for light reclaim */
Johannes Weinercc873172015-02-27 15:52:09 -08002407 if (!(gfp_mask & __GFP_FS)) {
2408 /*
2409 * XXX: Page reclaim didn't yield anything,
2410 * and the OOM killer can't be invoked, but
2411 * keep looping as per should_alloc_retry().
2412 */
2413 *did_some_progress = 1;
Johannes Weiner9879de72015-01-26 12:58:32 -08002414 goto out;
Johannes Weinercc873172015-02-27 15:52:09 -08002415 }
David Rientjes4167e9b2015-04-14 15:46:55 -07002416 /* The OOM killer may not free memory on a specific node */
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002417 if (gfp_mask & __GFP_THISNODE)
2418 goto out;
2419 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002420 /* Exhausted what can be done so it's blamo time */
Michal Hockoe009d5d2015-03-12 16:25:52 -07002421 if (out_of_memory(ac->zonelist, gfp_mask, order, ac->nodemask, false)
2422 || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL))
Michal Hockoc32b3cb2015-02-11 15:26:24 -08002423 *did_some_progress = 1;
Mel Gorman11e33f62009-06-16 15:31:57 -07002424out:
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002425 oom_zonelist_unlock(ac->zonelist, gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002426 return page;
2427}
2428
Mel Gorman56de7262010-05-24 14:32:30 -07002429#ifdef CONFIG_COMPACTION
2430/* Try memory compaction for high-order allocations before reclaim */
2431static struct page *
2432__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002433 int alloc_flags, const struct alloc_context *ac,
2434 enum migrate_mode mode, int *contended_compaction,
2435 bool *deferred_compaction)
Mel Gorman56de7262010-05-24 14:32:30 -07002436{
Vlastimil Babka53853e22014-10-09 15:27:02 -07002437 unsigned long compact_result;
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002438 struct page *page;
Vlastimil Babka53853e22014-10-09 15:27:02 -07002439
Mel Gorman66199712012-01-12 17:19:41 -08002440 if (!order)
Mel Gorman56de7262010-05-24 14:32:30 -07002441 return NULL;
2442
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002443 current->flags |= PF_MEMALLOC;
Vlastimil Babka1a6d53a2015-02-11 15:25:44 -08002444 compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
2445 mode, contended_compaction);
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002446 current->flags &= ~PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07002447
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002448 switch (compact_result) {
2449 case COMPACT_DEFERRED:
Vlastimil Babka53853e22014-10-09 15:27:02 -07002450 *deferred_compaction = true;
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002451 /* fall-through */
2452 case COMPACT_SKIPPED:
2453 return NULL;
2454 default:
2455 break;
Mel Gorman56de7262010-05-24 14:32:30 -07002456 }
2457
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002458 /*
2459 * At least in one zone compaction wasn't deferred or skipped, so let's
2460 * count a compaction stall
2461 */
2462 count_vm_event(COMPACTSTALL);
2463
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002464 page = get_page_from_freelist(gfp_mask, order,
2465 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002466
2467 if (page) {
2468 struct zone *zone = page_zone(page);
2469
2470 zone->compact_blockskip_flush = false;
2471 compaction_defer_reset(zone, order, true);
2472 count_vm_event(COMPACTSUCCESS);
2473 return page;
2474 }
2475
2476 /*
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002477 * It's bad if compaction run occurs and fails. The most likely reason
2478 * is that pages exist, but not enough to satisfy watermarks.
2479 */
2480 count_vm_event(COMPACTFAIL);
2481
2482 cond_resched();
2483
Mel Gorman56de7262010-05-24 14:32:30 -07002484 return NULL;
2485}
2486#else
2487static inline struct page *
2488__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002489 int alloc_flags, const struct alloc_context *ac,
2490 enum migrate_mode mode, int *contended_compaction,
2491 bool *deferred_compaction)
Mel Gorman56de7262010-05-24 14:32:30 -07002492{
2493 return NULL;
2494}
2495#endif /* CONFIG_COMPACTION */
2496
Marek Szyprowskibba90712012-01-25 12:09:52 +01002497/* Perform direct synchronous page reclaim */
2498static int
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002499__perform_reclaim(gfp_t gfp_mask, unsigned int order,
2500 const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002501{
Mel Gorman11e33f62009-06-16 15:31:57 -07002502 struct reclaim_state reclaim_state;
Marek Szyprowskibba90712012-01-25 12:09:52 +01002503 int progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07002504
2505 cond_resched();
2506
2507 /* We now go into synchronous reclaim */
2508 cpuset_memory_pressure_bump();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002509 current->flags |= PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002510 lockdep_set_current_reclaim_state(gfp_mask);
2511 reclaim_state.reclaimed_slab = 0;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002512 current->reclaim_state = &reclaim_state;
Mel Gorman11e33f62009-06-16 15:31:57 -07002513
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002514 progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
2515 ac->nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002516
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002517 current->reclaim_state = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002518 lockdep_clear_current_reclaim_state();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002519 current->flags &= ~PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002520
2521 cond_resched();
2522
Marek Szyprowskibba90712012-01-25 12:09:52 +01002523 return progress;
2524}
2525
2526/* The really slow allocator path where we enter direct reclaim */
2527static inline struct page *
2528__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002529 int alloc_flags, const struct alloc_context *ac,
2530 unsigned long *did_some_progress)
Marek Szyprowskibba90712012-01-25 12:09:52 +01002531{
2532 struct page *page = NULL;
2533 bool drained = false;
2534
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002535 *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002536 if (unlikely(!(*did_some_progress)))
2537 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002538
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002539 /* After successful reclaim, reconsider all zones for allocation */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002540 if (IS_ENABLED(CONFIG_NUMA))
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002541 zlc_clear_zones_full(ac->zonelist);
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002542
Mel Gorman9ee493c2010-09-09 16:38:18 -07002543retry:
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002544 page = get_page_from_freelist(gfp_mask, order,
2545 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002546
2547 /*
2548 * If an allocation failed after direct reclaim, it could be because
2549 * pages are pinned on the per-cpu lists. Drain them and try again
2550 */
2551 if (!page && !drained) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002552 drain_all_pages(NULL);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002553 drained = true;
2554 goto retry;
2555 }
2556
Mel Gorman11e33f62009-06-16 15:31:57 -07002557 return page;
2558}
2559
Mel Gorman11e33f62009-06-16 15:31:57 -07002560/*
2561 * This is called in the allocator slow-path if the allocation request is of
2562 * sufficient urgency to ignore watermarks and take other desperate measures
2563 */
2564static inline struct page *
2565__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002566 const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002567{
2568 struct page *page;
2569
2570 do {
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002571 page = get_page_from_freelist(gfp_mask, order,
2572 ALLOC_NO_WATERMARKS, ac);
Mel Gorman11e33f62009-06-16 15:31:57 -07002573
2574 if (!page && gfp_mask & __GFP_NOFAIL)
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002575 wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC,
2576 HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07002577 } while (!page && (gfp_mask & __GFP_NOFAIL));
2578
2579 return page;
2580}
2581
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002582static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002583{
2584 struct zoneref *z;
2585 struct zone *zone;
2586
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002587 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
2588 ac->high_zoneidx, ac->nodemask)
2589 wakeup_kswapd(zone, order, zone_idx(ac->preferred_zone));
Mel Gorman11e33f62009-06-16 15:31:57 -07002590}
2591
Peter Zijlstra341ce062009-06-16 15:32:02 -07002592static inline int
2593gfp_to_alloc_flags(gfp_t gfp_mask)
2594{
Peter Zijlstra341ce062009-06-16 15:32:02 -07002595 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
David Rientjesb104a352014-07-30 16:08:24 -07002596 const bool atomic = !(gfp_mask & (__GFP_WAIT | __GFP_NO_KSWAPD));
Peter Zijlstra341ce062009-06-16 15:32:02 -07002597
Mel Gormana56f57f2009-06-16 15:32:02 -07002598 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07002599 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07002600
Peter Zijlstra341ce062009-06-16 15:32:02 -07002601 /*
2602 * The caller may dip into page reserves a bit more if the caller
2603 * cannot run direct reclaim, or if the caller has realtime scheduling
2604 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
David Rientjesb104a352014-07-30 16:08:24 -07002605 * set both ALLOC_HARDER (atomic == true) and ALLOC_HIGH (__GFP_HIGH).
Peter Zijlstra341ce062009-06-16 15:32:02 -07002606 */
Namhyung Kime6223a32010-10-26 14:21:59 -07002607 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07002608
David Rientjesb104a352014-07-30 16:08:24 -07002609 if (atomic) {
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002610 /*
David Rientjesb104a352014-07-30 16:08:24 -07002611 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
2612 * if it can't schedule.
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002613 */
David Rientjesb104a352014-07-30 16:08:24 -07002614 if (!(gfp_mask & __GFP_NOMEMALLOC))
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002615 alloc_flags |= ALLOC_HARDER;
Peter Zijlstra341ce062009-06-16 15:32:02 -07002616 /*
David Rientjesb104a352014-07-30 16:08:24 -07002617 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
Vladimir Davydov344736f2014-10-20 15:50:30 +04002618 * comment for __cpuset_node_allowed().
Peter Zijlstra341ce062009-06-16 15:32:02 -07002619 */
2620 alloc_flags &= ~ALLOC_CPUSET;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002621 } else if (unlikely(rt_task(current)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07002622 alloc_flags |= ALLOC_HARDER;
2623
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002624 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2625 if (gfp_mask & __GFP_MEMALLOC)
2626 alloc_flags |= ALLOC_NO_WATERMARKS;
Mel Gorman907aed42012-07-31 16:44:07 -07002627 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
2628 alloc_flags |= ALLOC_NO_WATERMARKS;
2629 else if (!in_interrupt() &&
2630 ((current->flags & PF_MEMALLOC) ||
2631 unlikely(test_thread_flag(TIF_MEMDIE))))
Peter Zijlstra341ce062009-06-16 15:32:02 -07002632 alloc_flags |= ALLOC_NO_WATERMARKS;
2633 }
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002634#ifdef CONFIG_CMA
David Rientjes43e7a342014-10-09 15:27:25 -07002635 if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002636 alloc_flags |= ALLOC_CMA;
2637#endif
Peter Zijlstra341ce062009-06-16 15:32:02 -07002638 return alloc_flags;
2639}
2640
Mel Gorman072bb0a2012-07-31 16:43:58 -07002641bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
2642{
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002643 return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
Mel Gorman072bb0a2012-07-31 16:43:58 -07002644}
2645
Mel Gorman11e33f62009-06-16 15:31:57 -07002646static inline struct page *
2647__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002648 struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002649{
2650 const gfp_t wait = gfp_mask & __GFP_WAIT;
2651 struct page *page = NULL;
2652 int alloc_flags;
2653 unsigned long pages_reclaimed = 0;
2654 unsigned long did_some_progress;
David Rientjese0b9dae2014-06-04 16:08:28 -07002655 enum migrate_mode migration_mode = MIGRATE_ASYNC;
Mel Gorman66199712012-01-12 17:19:41 -08002656 bool deferred_compaction = false;
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07002657 int contended_compaction = COMPACT_CONTENDED_NONE;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002658
Christoph Lameter952f3b52006-12-06 20:33:26 -08002659 /*
Mel Gorman72807a72009-06-16 15:32:18 -07002660 * In the slowpath, we sanity check order to avoid ever trying to
2661 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2662 * be using allocators in order of preference for an area that is
2663 * too large.
2664 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07002665 if (order >= MAX_ORDER) {
2666 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07002667 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07002668 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002669
Christoph Lameter952f3b52006-12-06 20:33:26 -08002670 /*
David Rientjes4167e9b2015-04-14 15:46:55 -07002671 * If this allocation cannot block and it is for a specific node, then
2672 * fail early. There's no need to wakeup kswapd or retry for a
2673 * speculative node-specific allocation.
Christoph Lameter952f3b52006-12-06 20:33:26 -08002674 */
David Rientjes4167e9b2015-04-14 15:46:55 -07002675 if (IS_ENABLED(CONFIG_NUMA) && (gfp_mask & __GFP_THISNODE) && !wait)
Christoph Lameter952f3b52006-12-06 20:33:26 -08002676 goto nopage;
2677
Johannes Weiner9879de72015-01-26 12:58:32 -08002678retry:
Johannes Weiner3a025762014-04-07 15:37:48 -07002679 if (!(gfp_mask & __GFP_NO_KSWAPD))
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002680 wake_all_kswapds(order, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002681
Paul Jackson9bf22292005-09-06 15:18:12 -07002682 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002683 * OK, we're below the kswapd watermark and have kicked background
2684 * reclaim. Now things get more complex, so set up alloc_flags according
2685 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07002686 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002687 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688
David Rientjesf33261d2011-01-25 15:07:20 -08002689 /*
2690 * Find the true preferred zone if the allocation is unconstrained by
2691 * cpusets.
2692 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002693 if (!(alloc_flags & ALLOC_CPUSET) && !ac->nodemask) {
Mel Gormand8846372014-06-04 16:10:33 -07002694 struct zoneref *preferred_zoneref;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002695 preferred_zoneref = first_zones_zonelist(ac->zonelist,
2696 ac->high_zoneidx, NULL, &ac->preferred_zone);
2697 ac->classzone_idx = zonelist_zone_idx(preferred_zoneref);
Mel Gormand8846372014-06-04 16:10:33 -07002698 }
David Rientjesf33261d2011-01-25 15:07:20 -08002699
Peter Zijlstra341ce062009-06-16 15:32:02 -07002700 /* This is the last chance, in general, before the goto nopage. */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002701 page = get_page_from_freelist(gfp_mask, order,
2702 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002703 if (page)
2704 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705
Mel Gorman11e33f62009-06-16 15:31:57 -07002706 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002707 if (alloc_flags & ALLOC_NO_WATERMARKS) {
Mel Gorman183f6372012-07-31 16:44:12 -07002708 /*
2709 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
2710 * the allocation is high priority and these type of
2711 * allocations are system rather than user orientated
2712 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002713 ac->zonelist = node_zonelist(numa_node_id(), gfp_mask);
Mel Gorman183f6372012-07-31 16:44:12 -07002714
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002715 page = __alloc_pages_high_priority(gfp_mask, order, ac);
2716
Mel Gormancfd19c52012-07-31 16:44:10 -07002717 if (page) {
Peter Zijlstra341ce062009-06-16 15:32:02 -07002718 goto got_pg;
Mel Gormancfd19c52012-07-31 16:44:10 -07002719 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720 }
2721
2722 /* Atomic allocations - we can't balance anything */
David Rientjesaed0a0e2014-01-21 15:51:12 -08002723 if (!wait) {
2724 /*
2725 * All existing users of the deprecated __GFP_NOFAIL are
2726 * blockable, so warn of any new users that actually allow this
2727 * type of allocation to fail.
2728 */
2729 WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730 goto nopage;
David Rientjesaed0a0e2014-01-21 15:51:12 -08002731 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732
Peter Zijlstra341ce062009-06-16 15:32:02 -07002733 /* Avoid recursion of direct reclaim */
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002734 if (current->flags & PF_MEMALLOC)
Peter Zijlstra341ce062009-06-16 15:32:02 -07002735 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002736
David Rientjes6583bb62009-07-29 15:02:06 -07002737 /* Avoid allocations with no watermarks from looping endlessly */
2738 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2739 goto nopage;
2740
Mel Gorman77f1fe62011-01-13 15:45:57 -08002741 /*
2742 * Try direct compaction. The first pass is asynchronous. Subsequent
2743 * attempts after direct reclaim are synchronous
2744 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002745 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
2746 migration_mode,
2747 &contended_compaction,
Vlastimil Babka53853e22014-10-09 15:27:02 -07002748 &deferred_compaction);
Mel Gorman56de7262010-05-24 14:32:30 -07002749 if (page)
2750 goto got_pg;
David Rientjes75f30862014-06-04 16:08:30 -07002751
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07002752 /* Checks for THP-specific high-order allocations */
2753 if ((gfp_mask & GFP_TRANSHUGE) == GFP_TRANSHUGE) {
2754 /*
2755 * If compaction is deferred for high-order allocations, it is
2756 * because sync compaction recently failed. If this is the case
2757 * and the caller requested a THP allocation, we do not want
2758 * to heavily disrupt the system, so we fail the allocation
2759 * instead of entering direct reclaim.
2760 */
2761 if (deferred_compaction)
2762 goto nopage;
2763
2764 /*
2765 * In all zones where compaction was attempted (and not
2766 * deferred or skipped), lock contention has been detected.
2767 * For THP allocation we do not want to disrupt the others
2768 * so we fallback to base pages instead.
2769 */
2770 if (contended_compaction == COMPACT_CONTENDED_LOCK)
2771 goto nopage;
2772
2773 /*
2774 * If compaction was aborted due to need_resched(), we do not
2775 * want to further increase allocation latency, unless it is
2776 * khugepaged trying to collapse.
2777 */
2778 if (contended_compaction == COMPACT_CONTENDED_SCHED
2779 && !(current->flags & PF_KTHREAD))
2780 goto nopage;
2781 }
Mel Gorman66199712012-01-12 17:19:41 -08002782
David Rientjes8fe78042014-08-06 16:07:54 -07002783 /*
2784 * It can become very expensive to allocate transparent hugepages at
2785 * fault, so use asynchronous memory compaction for THP unless it is
2786 * khugepaged trying to collapse.
2787 */
2788 if ((gfp_mask & GFP_TRANSHUGE) != GFP_TRANSHUGE ||
2789 (current->flags & PF_KTHREAD))
2790 migration_mode = MIGRATE_SYNC_LIGHT;
2791
Mel Gorman11e33f62009-06-16 15:31:57 -07002792 /* Try direct reclaim and then allocating */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002793 page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
2794 &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07002795 if (page)
2796 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797
Mel Gorman11e33f62009-06-16 15:31:57 -07002798 /* Check if we should retry the allocation */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07002799 pages_reclaimed += did_some_progress;
Mel Gormanf90ac392012-01-10 15:07:15 -08002800 if (should_alloc_retry(gfp_mask, order, did_some_progress,
2801 pages_reclaimed)) {
Johannes Weiner9879de72015-01-26 12:58:32 -08002802 /*
2803 * If we fail to make progress by freeing individual
2804 * pages, but the allocation wants us to keep going,
2805 * start OOM killing tasks.
2806 */
2807 if (!did_some_progress) {
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002808 page = __alloc_pages_may_oom(gfp_mask, order, ac,
2809 &did_some_progress);
Johannes Weiner9879de72015-01-26 12:58:32 -08002810 if (page)
2811 goto got_pg;
2812 if (!did_some_progress)
2813 goto nopage;
2814 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002815 /* Wait for some write requests to complete then retry */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002816 wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC, HZ/50);
Johannes Weiner9879de72015-01-26 12:58:32 -08002817 goto retry;
Mel Gorman3e7d3442011-01-13 15:45:56 -08002818 } else {
2819 /*
2820 * High-order allocations do not necessarily loop after
2821 * direct reclaim and reclaim/compaction depends on compaction
2822 * being called after reclaim so call directly if necessary
2823 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002824 page = __alloc_pages_direct_compact(gfp_mask, order,
2825 alloc_flags, ac, migration_mode,
2826 &contended_compaction,
Vlastimil Babka53853e22014-10-09 15:27:02 -07002827 &deferred_compaction);
Mel Gorman3e7d3442011-01-13 15:45:56 -08002828 if (page)
2829 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830 }
2831
2832nopage:
Dave Hansena238ab52011-05-24 17:12:16 -07002833 warn_alloc_failed(gfp_mask, order, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834got_pg:
Mel Gorman072bb0a2012-07-31 16:43:58 -07002835 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836}
Mel Gorman11e33f62009-06-16 15:31:57 -07002837
2838/*
2839 * This is the 'heart' of the zoned buddy allocator.
2840 */
2841struct page *
2842__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2843 struct zonelist *zonelist, nodemask_t *nodemask)
2844{
Mel Gormand8846372014-06-04 16:10:33 -07002845 struct zoneref *preferred_zoneref;
Mel Gormancc9a6c82012-03-21 16:34:11 -07002846 struct page *page = NULL;
Mel Gormancc9a6c82012-03-21 16:34:11 -07002847 unsigned int cpuset_mems_cookie;
Johannes Weiner3a025762014-04-07 15:37:48 -07002848 int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET|ALLOC_FAIR;
Andrew Morton91fbdc02015-02-11 15:25:04 -08002849 gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002850 struct alloc_context ac = {
2851 .high_zoneidx = gfp_zone(gfp_mask),
2852 .nodemask = nodemask,
2853 .migratetype = gfpflags_to_migratetype(gfp_mask),
2854 };
Mel Gorman11e33f62009-06-16 15:31:57 -07002855
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10002856 gfp_mask &= gfp_allowed_mask;
2857
Mel Gorman11e33f62009-06-16 15:31:57 -07002858 lockdep_trace_alloc(gfp_mask);
2859
2860 might_sleep_if(gfp_mask & __GFP_WAIT);
2861
2862 if (should_fail_alloc_page(gfp_mask, order))
2863 return NULL;
2864
2865 /*
2866 * Check the zones suitable for the gfp_mask contain at least one
2867 * valid zone. It's possible to have an empty zonelist as a result
David Rientjes4167e9b2015-04-14 15:46:55 -07002868 * of __GFP_THISNODE and a memoryless node
Mel Gorman11e33f62009-06-16 15:31:57 -07002869 */
2870 if (unlikely(!zonelist->_zonerefs->zone))
2871 return NULL;
2872
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002873 if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE)
Vlastimil Babka21bb9bd2014-10-09 15:26:51 -07002874 alloc_flags |= ALLOC_CMA;
2875
Mel Gormancc9a6c82012-03-21 16:34:11 -07002876retry_cpuset:
Mel Gormand26914d2014-04-03 14:47:24 -07002877 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07002878
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002879 /* We set it here, as __alloc_pages_slowpath might have changed it */
2880 ac.zonelist = zonelist;
Mel Gorman5117f452009-06-16 15:31:59 -07002881 /* The preferred zone is used for statistics later */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002882 preferred_zoneref = first_zones_zonelist(ac.zonelist, ac.high_zoneidx,
2883 ac.nodemask ? : &cpuset_current_mems_allowed,
2884 &ac.preferred_zone);
2885 if (!ac.preferred_zone)
Mel Gormancc9a6c82012-03-21 16:34:11 -07002886 goto out;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002887 ac.classzone_idx = zonelist_zone_idx(preferred_zoneref);
Mel Gorman5117f452009-06-16 15:31:59 -07002888
2889 /* First allocation attempt */
Andrew Morton91fbdc02015-02-11 15:25:04 -08002890 alloc_mask = gfp_mask|__GFP_HARDWALL;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002891 page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
Ming Lei21caf2f2013-02-22 16:34:08 -08002892 if (unlikely(!page)) {
2893 /*
2894 * Runtime PM, block IO and its error handling path
2895 * can deadlock because I/O on the device might not
2896 * complete.
2897 */
Andrew Morton91fbdc02015-02-11 15:25:04 -08002898 alloc_mask = memalloc_noio_flags(gfp_mask);
2899
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002900 page = __alloc_pages_slowpath(alloc_mask, order, &ac);
Ming Lei21caf2f2013-02-22 16:34:08 -08002901 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002902
Xishi Qiu23f086f2015-02-11 15:25:07 -08002903 if (kmemcheck_enabled && page)
2904 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
2905
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002906 trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002907
2908out:
2909 /*
2910 * When updating a task's mems_allowed, it is possible to race with
2911 * parallel threads in such a way that an allocation can fail while
2912 * the mask is being updated. If a page allocation is about to fail,
2913 * check if the cpuset changed during allocation and if so, retry.
2914 */
Mel Gormand26914d2014-04-03 14:47:24 -07002915 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
Mel Gormancc9a6c82012-03-21 16:34:11 -07002916 goto retry_cpuset;
2917
Mel Gorman11e33f62009-06-16 15:31:57 -07002918 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919}
Mel Gormand2391712009-06-16 15:31:52 -07002920EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002921
2922/*
2923 * Common helper functions.
2924 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08002925unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002926{
Akinobu Mita945a1112009-09-21 17:01:47 -07002927 struct page *page;
2928
2929 /*
2930 * __get_free_pages() returns a 32-bit address, which cannot represent
2931 * a highmem page
2932 */
2933 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2934
Linus Torvalds1da177e2005-04-16 15:20:36 -07002935 page = alloc_pages(gfp_mask, order);
2936 if (!page)
2937 return 0;
2938 return (unsigned long) page_address(page);
2939}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002940EXPORT_SYMBOL(__get_free_pages);
2941
Harvey Harrison920c7a52008-02-04 22:29:26 -08002942unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002943{
Akinobu Mita945a1112009-09-21 17:01:47 -07002944 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002945}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002946EXPORT_SYMBOL(get_zeroed_page);
2947
Harvey Harrison920c7a52008-02-04 22:29:26 -08002948void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949{
Nick Pigginb5810032005-10-29 18:16:12 -07002950 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002951 if (order == 0)
Mel Gormanb745bc82014-06-04 16:10:22 -07002952 free_hot_cold_page(page, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002953 else
2954 __free_pages_ok(page, order);
2955 }
2956}
2957
2958EXPORT_SYMBOL(__free_pages);
2959
Harvey Harrison920c7a52008-02-04 22:29:26 -08002960void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002961{
2962 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07002963 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002964 __free_pages(virt_to_page((void *)addr), order);
2965 }
2966}
2967
2968EXPORT_SYMBOL(free_pages);
2969
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002970/*
Vladimir Davydov52383432014-06-04 16:06:39 -07002971 * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
2972 * of the current memory cgroup.
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002973 *
Vladimir Davydov52383432014-06-04 16:06:39 -07002974 * It should be used when the caller would like to use kmalloc, but since the
2975 * allocation is large, it has to fall back to the page allocator.
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002976 */
Vladimir Davydov52383432014-06-04 16:06:39 -07002977struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order)
2978{
2979 struct page *page;
2980 struct mem_cgroup *memcg = NULL;
2981
2982 if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
2983 return NULL;
2984 page = alloc_pages(gfp_mask, order);
2985 memcg_kmem_commit_charge(page, memcg, order);
2986 return page;
2987}
2988
2989struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
2990{
2991 struct page *page;
2992 struct mem_cgroup *memcg = NULL;
2993
2994 if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
2995 return NULL;
2996 page = alloc_pages_node(nid, gfp_mask, order);
2997 memcg_kmem_commit_charge(page, memcg, order);
2998 return page;
2999}
3000
3001/*
3002 * __free_kmem_pages and free_kmem_pages will free pages allocated with
3003 * alloc_kmem_pages.
3004 */
3005void __free_kmem_pages(struct page *page, unsigned int order)
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003006{
3007 memcg_kmem_uncharge_pages(page, order);
3008 __free_pages(page, order);
3009}
3010
Vladimir Davydov52383432014-06-04 16:06:39 -07003011void free_kmem_pages(unsigned long addr, unsigned int order)
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003012{
3013 if (addr != 0) {
3014 VM_BUG_ON(!virt_addr_valid((void *)addr));
Vladimir Davydov52383432014-06-04 16:06:39 -07003015 __free_kmem_pages(virt_to_page((void *)addr), order);
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003016 }
3017}
3018
Andi Kleenee85c2e2011-05-11 15:13:34 -07003019static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
3020{
3021 if (addr) {
3022 unsigned long alloc_end = addr + (PAGE_SIZE << order);
3023 unsigned long used = addr + PAGE_ALIGN(size);
3024
3025 split_page(virt_to_page((void *)addr), order);
3026 while (used < alloc_end) {
3027 free_page(used);
3028 used += PAGE_SIZE;
3029 }
3030 }
3031 return (void *)addr;
3032}
3033
Timur Tabi2be0ffe2008-07-23 21:28:11 -07003034/**
3035 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
3036 * @size: the number of bytes to allocate
3037 * @gfp_mask: GFP flags for the allocation
3038 *
3039 * This function is similar to alloc_pages(), except that it allocates the
3040 * minimum number of pages to satisfy the request. alloc_pages() can only
3041 * allocate memory in power-of-two pages.
3042 *
3043 * This function is also limited by MAX_ORDER.
3044 *
3045 * Memory allocated by this function must be released by free_pages_exact().
3046 */
3047void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
3048{
3049 unsigned int order = get_order(size);
3050 unsigned long addr;
3051
3052 addr = __get_free_pages(gfp_mask, order);
Andi Kleenee85c2e2011-05-11 15:13:34 -07003053 return make_alloc_exact(addr, order, size);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07003054}
3055EXPORT_SYMBOL(alloc_pages_exact);
3056
3057/**
Andi Kleenee85c2e2011-05-11 15:13:34 -07003058 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
3059 * pages on a node.
Randy Dunlapb5e6ab52011-05-16 13:16:54 -07003060 * @nid: the preferred node ID where memory should be allocated
Andi Kleenee85c2e2011-05-11 15:13:34 -07003061 * @size: the number of bytes to allocate
3062 * @gfp_mask: GFP flags for the allocation
3063 *
3064 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
3065 * back.
3066 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
3067 * but is not exact.
3068 */
Fabian Fredericke1931812014-08-06 16:04:59 -07003069void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
Andi Kleenee85c2e2011-05-11 15:13:34 -07003070{
3071 unsigned order = get_order(size);
3072 struct page *p = alloc_pages_node(nid, gfp_mask, order);
3073 if (!p)
3074 return NULL;
3075 return make_alloc_exact((unsigned long)page_address(p), order, size);
3076}
Andi Kleenee85c2e2011-05-11 15:13:34 -07003077
3078/**
Timur Tabi2be0ffe2008-07-23 21:28:11 -07003079 * free_pages_exact - release memory allocated via alloc_pages_exact()
3080 * @virt: the value returned by alloc_pages_exact.
3081 * @size: size of allocation, same value as passed to alloc_pages_exact().
3082 *
3083 * Release the memory allocated by a previous call to alloc_pages_exact.
3084 */
3085void free_pages_exact(void *virt, size_t size)
3086{
3087 unsigned long addr = (unsigned long)virt;
3088 unsigned long end = addr + PAGE_ALIGN(size);
3089
3090 while (addr < end) {
3091 free_page(addr);
3092 addr += PAGE_SIZE;
3093 }
3094}
3095EXPORT_SYMBOL(free_pages_exact);
3096
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003097/**
3098 * nr_free_zone_pages - count number of pages beyond high watermark
3099 * @offset: The zone index of the highest zone
3100 *
3101 * nr_free_zone_pages() counts the number of counts pages which are beyond the
3102 * high watermark within all zones at or below a given zone index. For each
3103 * zone, the number of pages is calculated as:
Jiang Liu834405c2013-07-03 15:03:04 -07003104 * managed_pages - high_pages
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003105 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003106static unsigned long nr_free_zone_pages(int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003107{
Mel Gormandd1a2392008-04-28 02:12:17 -07003108 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003109 struct zone *zone;
3110
Martin J. Blighe310fd42005-07-29 22:59:18 -07003111 /* Just pick one node, since fallback list is circular */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003112 unsigned long sum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003113
Mel Gorman0e884602008-04-28 02:12:14 -07003114 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003115
Mel Gorman54a6eb52008-04-28 02:12:16 -07003116 for_each_zone_zonelist(zone, z, zonelist, offset) {
Jiang Liub40da042013-02-22 16:33:52 -08003117 unsigned long size = zone->managed_pages;
Mel Gorman41858962009-06-16 15:32:12 -07003118 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07003119 if (size > high)
3120 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003121 }
3122
3123 return sum;
3124}
3125
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003126/**
3127 * nr_free_buffer_pages - count number of pages beyond high watermark
3128 *
3129 * nr_free_buffer_pages() counts the number of pages which are beyond the high
3130 * watermark within ZONE_DMA and ZONE_NORMAL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003131 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003132unsigned long nr_free_buffer_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003133{
Al Viroaf4ca452005-10-21 02:55:38 -04003134 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003135}
Meelap Shahc2f1a552007-07-17 04:04:39 -07003136EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003137
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003138/**
3139 * nr_free_pagecache_pages - count number of pages beyond high watermark
3140 *
3141 * nr_free_pagecache_pages() counts the number of pages which are beyond the
3142 * high watermark within all zones.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003143 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003144unsigned long nr_free_pagecache_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003145{
Mel Gorman2a1e2742007-07-17 04:03:12 -07003146 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003147}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07003148
3149static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003150{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08003151 if (IS_ENABLED(CONFIG_NUMA))
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08003152 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003153}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154
Linus Torvalds1da177e2005-04-16 15:20:36 -07003155void si_meminfo(struct sysinfo *val)
3156{
3157 val->totalram = totalram_pages;
Rafael Aquinicc7452b2014-08-06 16:06:38 -07003158 val->sharedram = global_page_state(NR_SHMEM);
Christoph Lameterd23ad422007-02-10 01:43:02 -08003159 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003160 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003161 val->totalhigh = totalhigh_pages;
3162 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003163 val->mem_unit = PAGE_SIZE;
3164}
3165
3166EXPORT_SYMBOL(si_meminfo);
3167
3168#ifdef CONFIG_NUMA
3169void si_meminfo_node(struct sysinfo *val, int nid)
3170{
Jiang Liucdd91a72013-07-03 15:03:27 -07003171 int zone_type; /* needs to be signed */
3172 unsigned long managed_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003173 pg_data_t *pgdat = NODE_DATA(nid);
3174
Jiang Liucdd91a72013-07-03 15:03:27 -07003175 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
3176 managed_pages += pgdat->node_zones[zone_type].managed_pages;
3177 val->totalram = managed_pages;
Rafael Aquinicc7452b2014-08-06 16:06:38 -07003178 val->sharedram = node_page_state(nid, NR_SHMEM);
Christoph Lameterd23ad422007-02-10 01:43:02 -08003179 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003180#ifdef CONFIG_HIGHMEM
Jiang Liub40da042013-02-22 16:33:52 -08003181 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08003182 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
3183 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003184#else
3185 val->totalhigh = 0;
3186 val->freehigh = 0;
3187#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003188 val->mem_unit = PAGE_SIZE;
3189}
3190#endif
3191
David Rientjesddd588b2011-03-22 16:30:46 -07003192/*
David Rientjes7bf02ea2011-05-24 17:11:16 -07003193 * Determine whether the node should be displayed or not, depending on whether
3194 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
David Rientjesddd588b2011-03-22 16:30:46 -07003195 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07003196bool skip_free_areas_node(unsigned int flags, int nid)
David Rientjesddd588b2011-03-22 16:30:46 -07003197{
3198 bool ret = false;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003199 unsigned int cpuset_mems_cookie;
David Rientjesddd588b2011-03-22 16:30:46 -07003200
3201 if (!(flags & SHOW_MEM_FILTER_NODES))
3202 goto out;
3203
Mel Gormancc9a6c82012-03-21 16:34:11 -07003204 do {
Mel Gormand26914d2014-04-03 14:47:24 -07003205 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07003206 ret = !node_isset(nid, cpuset_current_mems_allowed);
Mel Gormand26914d2014-04-03 14:47:24 -07003207 } while (read_mems_allowed_retry(cpuset_mems_cookie));
David Rientjesddd588b2011-03-22 16:30:46 -07003208out:
3209 return ret;
3210}
3211
Linus Torvalds1da177e2005-04-16 15:20:36 -07003212#define K(x) ((x) << (PAGE_SHIFT-10))
3213
Rabin Vincent377e4f12012-12-11 16:00:24 -08003214static void show_migration_types(unsigned char type)
3215{
3216 static const char types[MIGRATE_TYPES] = {
3217 [MIGRATE_UNMOVABLE] = 'U',
3218 [MIGRATE_RECLAIMABLE] = 'E',
3219 [MIGRATE_MOVABLE] = 'M',
3220 [MIGRATE_RESERVE] = 'R',
3221#ifdef CONFIG_CMA
3222 [MIGRATE_CMA] = 'C',
3223#endif
Minchan Kim194159f2013-02-22 16:33:58 -08003224#ifdef CONFIG_MEMORY_ISOLATION
Rabin Vincent377e4f12012-12-11 16:00:24 -08003225 [MIGRATE_ISOLATE] = 'I',
Minchan Kim194159f2013-02-22 16:33:58 -08003226#endif
Rabin Vincent377e4f12012-12-11 16:00:24 -08003227 };
3228 char tmp[MIGRATE_TYPES + 1];
3229 char *p = tmp;
3230 int i;
3231
3232 for (i = 0; i < MIGRATE_TYPES; i++) {
3233 if (type & (1 << i))
3234 *p++ = types[i];
3235 }
3236
3237 *p = '\0';
3238 printk("(%s) ", tmp);
3239}
3240
Linus Torvalds1da177e2005-04-16 15:20:36 -07003241/*
3242 * Show free area list (used inside shift_scroll-lock stuff)
3243 * We also calculate the percentage fragmentation. We do this by counting the
3244 * memory on each free list with the exception of the first item on the list.
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003245 *
3246 * Bits in @filter:
3247 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
3248 * cpuset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003249 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07003250void show_free_areas(unsigned int filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003251{
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003252 unsigned long free_pcp = 0;
Jes Sorensenc7241912006-09-27 01:50:05 -07003253 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003254 struct zone *zone;
3255
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003256 for_each_populated_zone(zone) {
David Rientjes7bf02ea2011-05-24 17:11:16 -07003257 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003258 continue;
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003259
Konstantin Khlebnikov761b0672015-04-14 15:45:32 -07003260 for_each_online_cpu(cpu)
3261 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003262 }
3263
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003264 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3265 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003266 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
3267 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003268 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003269 " free:%lu free_pcp:%lu free_cma:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003270 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003271 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003272 global_page_state(NR_ISOLATED_ANON),
3273 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003274 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003275 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003276 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07003277 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07003278 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07003279 global_page_state(NR_UNSTABLE_NFS),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07003280 global_page_state(NR_SLAB_RECLAIMABLE),
3281 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07003282 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003283 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08003284 global_page_state(NR_PAGETABLE),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003285 global_page_state(NR_BOUNCE),
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003286 global_page_state(NR_FREE_PAGES),
3287 free_pcp,
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003288 global_page_state(NR_FREE_CMA_PAGES));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003289
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003290 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003291 int i;
3292
David Rientjes7bf02ea2011-05-24 17:11:16 -07003293 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003294 continue;
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003295
3296 free_pcp = 0;
3297 for_each_online_cpu(cpu)
3298 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
3299
Linus Torvalds1da177e2005-04-16 15:20:36 -07003300 show_node(zone);
3301 printk("%s"
3302 " free:%lukB"
3303 " min:%lukB"
3304 " low:%lukB"
3305 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003306 " active_anon:%lukB"
3307 " inactive_anon:%lukB"
3308 " active_file:%lukB"
3309 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003310 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003311 " isolated(anon):%lukB"
3312 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003313 " present:%lukB"
Jiang Liu9feedc92012-12-12 13:52:12 -08003314 " managed:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003315 " mlocked:%lukB"
3316 " dirty:%lukB"
3317 " writeback:%lukB"
3318 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003319 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003320 " slab_reclaimable:%lukB"
3321 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003322 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003323 " pagetables:%lukB"
3324 " unstable:%lukB"
3325 " bounce:%lukB"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003326 " free_pcp:%lukB"
3327 " local_pcp:%ukB"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003328 " free_cma:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003329 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003330 " pages_scanned:%lu"
3331 " all_unreclaimable? %s"
3332 "\n",
3333 zone->name,
Mel Gorman88f5acf2011-01-13 15:45:41 -08003334 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07003335 K(min_wmark_pages(zone)),
3336 K(low_wmark_pages(zone)),
3337 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003338 K(zone_page_state(zone, NR_ACTIVE_ANON)),
3339 K(zone_page_state(zone, NR_INACTIVE_ANON)),
3340 K(zone_page_state(zone, NR_ACTIVE_FILE)),
3341 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003342 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003343 K(zone_page_state(zone, NR_ISOLATED_ANON)),
3344 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003345 K(zone->present_pages),
Jiang Liu9feedc92012-12-12 13:52:12 -08003346 K(zone->managed_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003347 K(zone_page_state(zone, NR_MLOCK)),
3348 K(zone_page_state(zone, NR_FILE_DIRTY)),
3349 K(zone_page_state(zone, NR_WRITEBACK)),
3350 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003351 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003352 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3353 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003354 zone_page_state(zone, NR_KERNEL_STACK) *
3355 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003356 K(zone_page_state(zone, NR_PAGETABLE)),
3357 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3358 K(zone_page_state(zone, NR_BOUNCE)),
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003359 K(free_pcp),
3360 K(this_cpu_read(zone->pageset->pcp.count)),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003361 K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003362 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Mel Gorman0d5d8232014-08-06 16:07:16 -07003363 K(zone_page_state(zone, NR_PAGES_SCANNED)),
Lisa Du6e543d52013-09-11 14:22:36 -07003364 (!zone_reclaimable(zone) ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07003365 );
3366 printk("lowmem_reserve[]:");
3367 for (i = 0; i < MAX_NR_ZONES; i++)
Mel Gorman3484b2d2014-08-06 16:07:14 -07003368 printk(" %ld", zone->lowmem_reserve[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003369 printk("\n");
3370 }
3371
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003372 for_each_populated_zone(zone) {
Pintu Kumarb8af2942013-09-11 14:20:34 -07003373 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003374 unsigned char types[MAX_ORDER];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003375
David Rientjes7bf02ea2011-05-24 17:11:16 -07003376 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003377 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003378 show_node(zone);
3379 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003380
3381 spin_lock_irqsave(&zone->lock, flags);
3382 for (order = 0; order < MAX_ORDER; order++) {
Rabin Vincent377e4f12012-12-11 16:00:24 -08003383 struct free_area *area = &zone->free_area[order];
3384 int type;
3385
3386 nr[order] = area->nr_free;
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003387 total += nr[order] << order;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003388
3389 types[order] = 0;
3390 for (type = 0; type < MIGRATE_TYPES; type++) {
3391 if (!list_empty(&area->free_list[type]))
3392 types[order] |= 1 << type;
3393 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003394 }
3395 spin_unlock_irqrestore(&zone->lock, flags);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003396 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003397 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003398 if (nr[order])
3399 show_migration_types(types[order]);
3400 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003401 printk("= %lukB\n", K(total));
3402 }
3403
David Rientjes949f7ec2013-04-29 15:07:48 -07003404 hugetlb_show_meminfo();
3405
Larry Woodmane6f36022008-02-04 22:29:30 -08003406 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
3407
Linus Torvalds1da177e2005-04-16 15:20:36 -07003408 show_swap_cache_info();
3409}
3410
Mel Gorman19770b32008-04-28 02:12:18 -07003411static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
3412{
3413 zoneref->zone = zone;
3414 zoneref->zone_idx = zone_idx(zone);
3415}
3416
Linus Torvalds1da177e2005-04-16 15:20:36 -07003417/*
3418 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08003419 *
3420 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003421 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003422static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003423 int nr_zones)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003424{
Christoph Lameter1a932052006-01-06 00:11:16 -08003425 struct zone *zone;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003426 enum zone_type zone_type = MAX_NR_ZONES;
Christoph Lameter02a68a52006-01-06 00:11:18 -08003427
3428 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003429 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08003430 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08003431 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07003432 zoneref_set_zone(zone,
3433 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08003434 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003435 }
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003436 } while (zone_type);
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003437
Christoph Lameter070f8032006-01-06 00:11:19 -08003438 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003439}
3440
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003441
3442/*
3443 * zonelist_order:
3444 * 0 = automatic detection of better ordering.
3445 * 1 = order by ([node] distance, -zonetype)
3446 * 2 = order by (-zonetype, [node] distance)
3447 *
3448 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3449 * the same zonelist. So only NUMA can configure this param.
3450 */
3451#define ZONELIST_ORDER_DEFAULT 0
3452#define ZONELIST_ORDER_NODE 1
3453#define ZONELIST_ORDER_ZONE 2
3454
3455/* zonelist order in the kernel.
3456 * set_zonelist_order() will set this to NODE or ZONE.
3457 */
3458static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
3459static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
3460
3461
Linus Torvalds1da177e2005-04-16 15:20:36 -07003462#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003463/* The value user specified ....changed by config */
3464static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3465/* string for sysctl */
3466#define NUMA_ZONELIST_ORDER_LEN 16
3467char numa_zonelist_order[16] = "default";
3468
3469/*
3470 * interface for configure zonelist ordering.
3471 * command line option "numa_zonelist_order"
3472 * = "[dD]efault - default, automatic configuration.
3473 * = "[nN]ode - order by node locality, then by zone within node
3474 * = "[zZ]one - order by zone, then by locality within zone
3475 */
3476
3477static int __parse_numa_zonelist_order(char *s)
3478{
3479 if (*s == 'd' || *s == 'D') {
3480 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3481 } else if (*s == 'n' || *s == 'N') {
3482 user_zonelist_order = ZONELIST_ORDER_NODE;
3483 } else if (*s == 'z' || *s == 'Z') {
3484 user_zonelist_order = ZONELIST_ORDER_ZONE;
3485 } else {
3486 printk(KERN_WARNING
3487 "Ignoring invalid numa_zonelist_order value: "
3488 "%s\n", s);
3489 return -EINVAL;
3490 }
3491 return 0;
3492}
3493
3494static __init int setup_numa_zonelist_order(char *s)
3495{
Volodymyr G. Lukiianykecb256f2011-01-13 15:46:26 -08003496 int ret;
3497
3498 if (!s)
3499 return 0;
3500
3501 ret = __parse_numa_zonelist_order(s);
3502 if (ret == 0)
3503 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
3504
3505 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003506}
3507early_param("numa_zonelist_order", setup_numa_zonelist_order);
3508
3509/*
3510 * sysctl handler for numa_zonelist_order
3511 */
Joe Perchescccad5b2014-06-06 14:38:09 -07003512int numa_zonelist_order_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003513 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003514 loff_t *ppos)
3515{
3516 char saved_string[NUMA_ZONELIST_ORDER_LEN];
3517 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01003518 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003519
Andi Kleen443c6f12009-12-23 21:00:47 +01003520 mutex_lock(&zl_order_mutex);
Chen Gangdacbde02013-07-03 15:02:35 -07003521 if (write) {
3522 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
3523 ret = -EINVAL;
3524 goto out;
3525 }
3526 strcpy(saved_string, (char *)table->data);
3527 }
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003528 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003529 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01003530 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003531 if (write) {
3532 int oldval = user_zonelist_order;
Chen Gangdacbde02013-07-03 15:02:35 -07003533
3534 ret = __parse_numa_zonelist_order((char *)table->data);
3535 if (ret) {
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003536 /*
3537 * bogus value. restore saved string
3538 */
Chen Gangdacbde02013-07-03 15:02:35 -07003539 strncpy((char *)table->data, saved_string,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003540 NUMA_ZONELIST_ORDER_LEN);
3541 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003542 } else if (oldval != user_zonelist_order) {
3543 mutex_lock(&zonelists_mutex);
Jiang Liu9adb62a2012-07-31 16:43:28 -07003544 build_all_zonelists(NULL, NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003545 mutex_unlock(&zonelists_mutex);
3546 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003547 }
Andi Kleen443c6f12009-12-23 21:00:47 +01003548out:
3549 mutex_unlock(&zl_order_mutex);
3550 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003551}
3552
3553
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003554#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003555static int node_load[MAX_NUMNODES];
3556
Linus Torvalds1da177e2005-04-16 15:20:36 -07003557/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07003558 * 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 -07003559 * @node: node whose fallback list we're appending
3560 * @used_node_mask: nodemask_t of already used nodes
3561 *
3562 * We use a number of factors to determine which is the next node that should
3563 * appear on a given node's fallback list. The node should not have appeared
3564 * already in @node's fallback list, and it should be the next closest node
3565 * according to the distance array (which contains arbitrary distance values
3566 * from each node to each node in the system), and should also prefer nodes
3567 * with no CPUs, since presumably they'll have very little allocation pressure
3568 * on them otherwise.
3569 * It returns -1 if no node is found.
3570 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003571static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003572{
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01003573 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003574 int min_val = INT_MAX;
David Rientjes00ef2d22013-02-22 16:35:36 -08003575 int best_node = NUMA_NO_NODE;
Rusty Russella70f7302009-03-13 14:49:46 +10303576 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003577
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01003578 /* Use the local node if we haven't already */
3579 if (!node_isset(node, *used_node_mask)) {
3580 node_set(node, *used_node_mask);
3581 return node;
3582 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003583
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08003584 for_each_node_state(n, N_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003585
3586 /* Don't want a node to appear more than once */
3587 if (node_isset(n, *used_node_mask))
3588 continue;
3589
Linus Torvalds1da177e2005-04-16 15:20:36 -07003590 /* Use the distance array to find the distance */
3591 val = node_distance(node, n);
3592
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01003593 /* Penalize nodes under us ("prefer the next node") */
3594 val += (n < node);
3595
Linus Torvalds1da177e2005-04-16 15:20:36 -07003596 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10303597 tmp = cpumask_of_node(n);
3598 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003599 val += PENALTY_FOR_NODE_WITH_CPUS;
3600
3601 /* Slight preference for less loaded node */
3602 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
3603 val += node_load[n];
3604
3605 if (val < min_val) {
3606 min_val = val;
3607 best_node = n;
3608 }
3609 }
3610
3611 if (best_node >= 0)
3612 node_set(best_node, *used_node_mask);
3613
3614 return best_node;
3615}
3616
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003617
3618/*
3619 * Build zonelists ordered by node and zones within node.
3620 * This results in maximum locality--normal zone overflows into local
3621 * DMA zone, if any--but risks exhausting DMA zone.
3622 */
3623static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003624{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003625 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003626 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003627
Mel Gorman54a6eb52008-04-28 02:12:16 -07003628 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07003629 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07003630 ;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003631 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gormandd1a2392008-04-28 02:12:17 -07003632 zonelist->_zonerefs[j].zone = NULL;
3633 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003634}
3635
3636/*
Christoph Lameter523b9452007-10-16 01:25:37 -07003637 * Build gfp_thisnode zonelists
3638 */
3639static void build_thisnode_zonelists(pg_data_t *pgdat)
3640{
Christoph Lameter523b9452007-10-16 01:25:37 -07003641 int j;
3642 struct zonelist *zonelist;
3643
Mel Gorman54a6eb52008-04-28 02:12:16 -07003644 zonelist = &pgdat->node_zonelists[1];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003645 j = build_zonelists_node(pgdat, zonelist, 0);
Mel Gormandd1a2392008-04-28 02:12:17 -07003646 zonelist->_zonerefs[j].zone = NULL;
3647 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07003648}
3649
3650/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003651 * Build zonelists ordered by zone and nodes within zones.
3652 * This results in conserving DMA zone[s] until all Normal memory is
3653 * exhausted, but results in overflowing to remote node while memory
3654 * may still exist in local DMA zone.
3655 */
3656static int node_order[MAX_NUMNODES];
3657
3658static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
3659{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003660 int pos, j, node;
3661 int zone_type; /* needs to be signed */
3662 struct zone *z;
3663 struct zonelist *zonelist;
3664
Mel Gorman54a6eb52008-04-28 02:12:16 -07003665 zonelist = &pgdat->node_zonelists[0];
3666 pos = 0;
3667 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
3668 for (j = 0; j < nr_nodes; j++) {
3669 node = node_order[j];
3670 z = &NODE_DATA(node)->node_zones[zone_type];
3671 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07003672 zoneref_set_zone(z,
3673 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07003674 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003675 }
3676 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003677 }
Mel Gormandd1a2392008-04-28 02:12:17 -07003678 zonelist->_zonerefs[pos].zone = NULL;
3679 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003680}
3681
Mel Gorman31939132014-10-09 15:28:30 -07003682#if defined(CONFIG_64BIT)
3683/*
3684 * Devices that require DMA32/DMA are relatively rare and do not justify a
3685 * penalty to every machine in case the specialised case applies. Default
3686 * to Node-ordering on 64-bit NUMA machines
3687 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003688static int default_zonelist_order(void)
3689{
Mel Gorman31939132014-10-09 15:28:30 -07003690 return ZONELIST_ORDER_NODE;
3691}
3692#else
3693/*
3694 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
3695 * by the kernel. If processes running on node 0 deplete the low memory zone
3696 * then reclaim will occur more frequency increasing stalls and potentially
3697 * be easier to OOM if a large percentage of the zone is under writeback or
3698 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
3699 * Hence, default to zone ordering on 32-bit.
3700 */
3701static int default_zonelist_order(void)
3702{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003703 return ZONELIST_ORDER_ZONE;
3704}
Mel Gorman31939132014-10-09 15:28:30 -07003705#endif /* CONFIG_64BIT */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003706
3707static void set_zonelist_order(void)
3708{
3709 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
3710 current_zonelist_order = default_zonelist_order();
3711 else
3712 current_zonelist_order = user_zonelist_order;
3713}
3714
3715static void build_zonelists(pg_data_t *pgdat)
3716{
3717 int j, node, load;
3718 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003719 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003720 int local_node, prev_node;
3721 struct zonelist *zonelist;
3722 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003723
3724 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07003725 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003726 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07003727 zonelist->_zonerefs[0].zone = NULL;
3728 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003729 }
3730
3731 /* NUMA-aware ordering of nodes */
3732 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003733 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003734 prev_node = local_node;
3735 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003736
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003737 memset(node_order, 0, sizeof(node_order));
3738 j = 0;
3739
Linus Torvalds1da177e2005-04-16 15:20:36 -07003740 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
3741 /*
3742 * We don't want to pressure a particular node.
3743 * So adding penalty to the first node in same
3744 * distance group to make it round-robin.
3745 */
David Rientjes957f8222012-10-08 16:33:24 -07003746 if (node_distance(local_node, node) !=
3747 node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003748 node_load[node] = load;
3749
Linus Torvalds1da177e2005-04-16 15:20:36 -07003750 prev_node = node;
3751 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003752 if (order == ZONELIST_ORDER_NODE)
3753 build_zonelists_in_node_order(pgdat, node);
3754 else
3755 node_order[j++] = node; /* remember order */
3756 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003757
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003758 if (order == ZONELIST_ORDER_ZONE) {
3759 /* calculate node order -- i.e., DMA last! */
3760 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003761 }
Christoph Lameter523b9452007-10-16 01:25:37 -07003762
3763 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003764}
3765
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003766/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003767static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003768{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003769 struct zonelist *zonelist;
3770 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07003771 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003772
Mel Gorman54a6eb52008-04-28 02:12:16 -07003773 zonelist = &pgdat->node_zonelists[0];
3774 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3775 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07003776 for (z = zonelist->_zonerefs; z->zone; z++)
3777 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003778}
3779
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003780#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3781/*
3782 * Return node id of node used for "local" allocations.
3783 * I.e., first node id of first zone in arg node's generic zonelist.
3784 * Used for initializing percpu 'numa_mem', which is used primarily
3785 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3786 */
3787int local_memory_node(int node)
3788{
3789 struct zone *zone;
3790
3791 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3792 gfp_zone(GFP_KERNEL),
3793 NULL,
3794 &zone);
3795 return zone->node;
3796}
3797#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003798
Linus Torvalds1da177e2005-04-16 15:20:36 -07003799#else /* CONFIG_NUMA */
3800
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003801static void set_zonelist_order(void)
3802{
3803 current_zonelist_order = ZONELIST_ORDER_ZONE;
3804}
3805
3806static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003807{
Christoph Lameter19655d32006-09-25 23:31:19 -07003808 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003809 enum zone_type j;
3810 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003811
3812 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003813
Mel Gorman54a6eb52008-04-28 02:12:16 -07003814 zonelist = &pgdat->node_zonelists[0];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003815 j = build_zonelists_node(pgdat, zonelist, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003816
Mel Gorman54a6eb52008-04-28 02:12:16 -07003817 /*
3818 * Now we build the zonelist so that it contains the zones
3819 * of all the other nodes.
3820 * We don't want to pressure a particular node, so when
3821 * building the zones for node N, we make sure that the
3822 * zones coming right after the local ones are those from
3823 * node N+1 (modulo N)
3824 */
3825 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3826 if (!node_online(node))
3827 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003828 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003829 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07003830 for (node = 0; node < local_node; node++) {
3831 if (!node_online(node))
3832 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003833 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gorman54a6eb52008-04-28 02:12:16 -07003834 }
3835
Mel Gormandd1a2392008-04-28 02:12:17 -07003836 zonelist->_zonerefs[j].zone = NULL;
3837 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003838}
3839
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003840/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003841static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003842{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003843 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003844}
3845
Linus Torvalds1da177e2005-04-16 15:20:36 -07003846#endif /* CONFIG_NUMA */
3847
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003848/*
3849 * Boot pageset table. One per cpu which is going to be used for all
3850 * zones and all nodes. The parameters will be set in such a way
3851 * that an item put on a list will immediately be handed over to
3852 * the buddy list. This is safe since pageset manipulation is done
3853 * with interrupts disabled.
3854 *
3855 * The boot_pagesets must be kept even after bootup is complete for
3856 * unused processors and/or zones. They do play a role for bootstrapping
3857 * hotplugged processors.
3858 *
3859 * zoneinfo_show() and maybe other functions do
3860 * not check if the processor is online before following the pageset pointer.
3861 * Other parts of the kernel may not check if the zone is available.
3862 */
3863static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3864static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07003865static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003866
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003867/*
3868 * Global mutex to protect against size modification of zonelists
3869 * as well as to serialize pageset setup for the new populated zone.
3870 */
3871DEFINE_MUTEX(zonelists_mutex);
3872
Rusty Russell9b1a4d32008-07-28 12:16:30 -05003873/* return values int ....just for stop_machine() */
Jiang Liu4ed7e022012-07-31 16:43:35 -07003874static int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003875{
Yasunori Goto68113782006-06-23 02:03:11 -07003876 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003877 int cpu;
Jiang Liu9adb62a2012-07-31 16:43:28 -07003878 pg_data_t *self = data;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003879
Bo Liu7f9cfb32009-08-18 14:11:19 -07003880#ifdef CONFIG_NUMA
3881 memset(node_load, 0, sizeof(node_load));
3882#endif
Jiang Liu9adb62a2012-07-31 16:43:28 -07003883
3884 if (self && !node_online(self->node_id)) {
3885 build_zonelists(self);
3886 build_zonelist_cache(self);
3887 }
3888
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003889 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07003890 pg_data_t *pgdat = NODE_DATA(nid);
3891
3892 build_zonelists(pgdat);
3893 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003894 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003895
3896 /*
3897 * Initialize the boot_pagesets that are going to be used
3898 * for bootstrapping processors. The real pagesets for
3899 * each zone will be allocated later when the per cpu
3900 * allocator is available.
3901 *
3902 * boot_pagesets are used also for bootstrapping offline
3903 * cpus if the system is already booted because the pagesets
3904 * are needed to initialize allocators on a specific cpu too.
3905 * F.e. the percpu allocator needs the page allocator which
3906 * needs the percpu allocator in order to allocate its pagesets
3907 * (a chicken-egg dilemma).
3908 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003909 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003910 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3911
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003912#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3913 /*
3914 * We now know the "local memory node" for each node--
3915 * i.e., the node of the first zone in the generic zonelist.
3916 * Set up numa_mem percpu variable for on-line cpus. During
3917 * boot, only the boot cpu should be on-line; we'll init the
3918 * secondary cpus' numa_mem as they come on-line. During
3919 * node/memory hotplug, we'll fixup all on-line cpus.
3920 */
3921 if (cpu_online(cpu))
3922 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3923#endif
3924 }
3925
Yasunori Goto68113782006-06-23 02:03:11 -07003926 return 0;
3927}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003928
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08003929static noinline void __init
3930build_all_zonelists_init(void)
3931{
3932 __build_all_zonelists(NULL);
3933 mminit_verify_zonelist();
3934 cpuset_init_current_mems_allowed();
3935}
3936
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003937/*
3938 * Called with zonelists_mutex held always
3939 * unless system_state == SYSTEM_BOOTING.
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08003940 *
3941 * __ref due to (1) call of __meminit annotated setup_zone_pageset
3942 * [we're only called with non-NULL zone through __meminit paths] and
3943 * (2) call of __init annotated helper build_all_zonelists_init
3944 * [protected by SYSTEM_BOOTING].
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003945 */
Jiang Liu9adb62a2012-07-31 16:43:28 -07003946void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
Yasunori Goto68113782006-06-23 02:03:11 -07003947{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003948 set_zonelist_order();
3949
Yasunori Goto68113782006-06-23 02:03:11 -07003950 if (system_state == SYSTEM_BOOTING) {
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08003951 build_all_zonelists_init();
Yasunori Goto68113782006-06-23 02:03:11 -07003952 } else {
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003953#ifdef CONFIG_MEMORY_HOTPLUG
Jiang Liu9adb62a2012-07-31 16:43:28 -07003954 if (zone)
3955 setup_zone_pageset(zone);
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003956#endif
Cody P Schaferdd1895e2013-07-03 15:01:36 -07003957 /* we have to stop all cpus to guarantee there is no user
3958 of zonelist */
Jiang Liu9adb62a2012-07-31 16:43:28 -07003959 stop_machine(__build_all_zonelists, pgdat, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07003960 /* cpuset refresh routine should be here */
3961 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07003962 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003963 /*
3964 * Disable grouping by mobility if the number of pages in the
3965 * system is too low to allow the mechanism to work. It would be
3966 * more accurate, but expensive to check per-zone. This check is
3967 * made on memory-hotadd so a system can start with mobility
3968 * disabled and enable it later
3969 */
Mel Gormand9c23402007-10-16 01:26:01 -07003970 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003971 page_group_by_mobility_disabled = 1;
3972 else
3973 page_group_by_mobility_disabled = 0;
3974
Anton Blanchardf88dfff2014-12-10 15:42:53 -08003975 pr_info("Built %i zonelists in %s order, mobility grouping %s. "
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003976 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003977 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003978 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003979 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003980 vm_total_pages);
3981#ifdef CONFIG_NUMA
Anton Blanchardf88dfff2014-12-10 15:42:53 -08003982 pr_info("Policy zone: %s\n", zone_names[policy_zone]);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003983#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003984}
3985
3986/*
3987 * Helper functions to size the waitqueue hash table.
3988 * Essentially these want to choose hash table sizes sufficiently
3989 * large so that collisions trying to wait on pages are rare.
3990 * But in fact, the number of active page waitqueues on typical
3991 * systems is ridiculously low, less than 200. So this is even
3992 * conservative, even though it seems large.
3993 *
3994 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3995 * waitqueues, i.e. the size of the waitq table given the number of pages.
3996 */
3997#define PAGES_PER_WAITQUEUE 256
3998
Yasunori Gotocca448f2006-06-23 02:03:10 -07003999#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07004000static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004001{
4002 unsigned long size = 1;
4003
4004 pages /= PAGES_PER_WAITQUEUE;
4005
4006 while (size < pages)
4007 size <<= 1;
4008
4009 /*
4010 * Once we have dozens or even hundreds of threads sleeping
4011 * on IO we've got bigger problems than wait queue collision.
4012 * Limit the size of the wait table to a reasonable size.
4013 */
4014 size = min(size, 4096UL);
4015
4016 return max(size, 4UL);
4017}
Yasunori Gotocca448f2006-06-23 02:03:10 -07004018#else
4019/*
4020 * A zone's size might be changed by hot-add, so it is not possible to determine
4021 * a suitable size for its wait_table. So we use the maximum size now.
4022 *
4023 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
4024 *
4025 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
4026 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
4027 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
4028 *
4029 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
4030 * or more by the traditional way. (See above). It equals:
4031 *
4032 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
4033 * ia64(16K page size) : = ( 8G + 4M)byte.
4034 * powerpc (64K page size) : = (32G +16M)byte.
4035 */
4036static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
4037{
4038 return 4096UL;
4039}
4040#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004041
4042/*
4043 * This is an integer logarithm so that shifts can be used later
4044 * to extract the more random high bits from the multiplicative
4045 * hash function before the remainder is taken.
4046 */
4047static inline unsigned long wait_table_bits(unsigned long size)
4048{
4049 return ffz(~size);
4050}
4051
Mel Gorman56fd56b2007-10-16 01:25:58 -07004052/*
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07004053 * Check if a pageblock contains reserved pages
4054 */
4055static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
4056{
4057 unsigned long pfn;
4058
4059 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
4060 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
4061 return 1;
4062 }
4063 return 0;
4064}
4065
4066/*
Mel Gormand9c23402007-10-16 01:26:01 -07004067 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07004068 * of blocks reserved is based on min_wmark_pages(zone). The memory within
4069 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07004070 * higher will lead to a bigger reserve which will get freed as contiguous
4071 * blocks as reclaim kicks in
4072 */
4073static void setup_zone_migrate_reserve(struct zone *zone)
4074{
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07004075 unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004076 struct page *page;
Mel Gorman78986a62009-09-21 17:03:02 -07004077 unsigned long block_migratetype;
4078 int reserve;
Yasuaki Ishimatsu943dca12014-01-21 15:49:06 -08004079 int old_reserve;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004080
Michal Hockod02156382011-12-08 14:34:27 -08004081 /*
4082 * Get the start pfn, end pfn and the number of blocks to reserve
4083 * We have to be careful to be aligned to pageblock_nr_pages to
4084 * make sure that we always check pfn_valid for the first page in
4085 * the block.
4086 */
Mel Gorman56fd56b2007-10-16 01:25:58 -07004087 start_pfn = zone->zone_start_pfn;
Cody P Schafer108bcc92013-02-22 16:35:23 -08004088 end_pfn = zone_end_pfn(zone);
Michal Hockod02156382011-12-08 14:34:27 -08004089 start_pfn = roundup(start_pfn, pageblock_nr_pages);
Mel Gorman41858962009-06-16 15:32:12 -07004090 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07004091 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004092
Mel Gorman78986a62009-09-21 17:03:02 -07004093 /*
4094 * Reserve blocks are generally in place to help high-order atomic
4095 * allocations that are short-lived. A min_free_kbytes value that
4096 * would result in more than 2 reserve blocks for atomic allocations
4097 * is assumed to be in place to help anti-fragmentation for the
4098 * future allocation of hugepages at runtime.
4099 */
4100 reserve = min(2, reserve);
Yasuaki Ishimatsu943dca12014-01-21 15:49:06 -08004101 old_reserve = zone->nr_migrate_reserve_block;
4102
4103 /* When memory hot-add, we almost always need to do nothing */
4104 if (reserve == old_reserve)
4105 return;
4106 zone->nr_migrate_reserve_block = reserve;
Mel Gorman78986a62009-09-21 17:03:02 -07004107
Mel Gormand9c23402007-10-16 01:26:01 -07004108 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07004109 if (!pfn_valid(pfn))
4110 continue;
4111 page = pfn_to_page(pfn);
4112
Adam Litke344c7902008-09-02 14:35:38 -07004113 /* Watch out for overlapping nodes */
4114 if (page_to_nid(page) != zone_to_nid(zone))
4115 continue;
4116
Mel Gorman56fd56b2007-10-16 01:25:58 -07004117 block_migratetype = get_pageblock_migratetype(page);
4118
Mel Gorman938929f2012-01-10 15:07:14 -08004119 /* Only test what is necessary when the reserves are not met */
4120 if (reserve > 0) {
4121 /*
4122 * Blocks with reserved pages will never free, skip
4123 * them.
4124 */
4125 block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
4126 if (pageblock_is_reserved(pfn, block_end_pfn))
4127 continue;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004128
Mel Gorman938929f2012-01-10 15:07:14 -08004129 /* If this block is reserved, account for it */
4130 if (block_migratetype == MIGRATE_RESERVE) {
4131 reserve--;
4132 continue;
4133 }
4134
4135 /* Suitable for reserving if this block is movable */
4136 if (block_migratetype == MIGRATE_MOVABLE) {
4137 set_pageblock_migratetype(page,
4138 MIGRATE_RESERVE);
4139 move_freepages_block(zone, page,
4140 MIGRATE_RESERVE);
4141 reserve--;
4142 continue;
4143 }
Yasuaki Ishimatsu943dca12014-01-21 15:49:06 -08004144 } else if (!old_reserve) {
4145 /*
4146 * At boot time we don't need to scan the whole zone
4147 * for turning off MIGRATE_RESERVE.
4148 */
4149 break;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004150 }
4151
4152 /*
4153 * If the reserve is met and this is a previous reserved block,
4154 * take it back
4155 */
4156 if (block_migratetype == MIGRATE_RESERVE) {
4157 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4158 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4159 }
4160 }
4161}
Mel Gormanac0e5b72007-10-16 01:25:58 -07004162
Linus Torvalds1da177e2005-04-16 15:20:36 -07004163/*
4164 * Initially all pages are reserved - free ones are freed
4165 * up by free_all_bootmem() once the early boot process is
4166 * done. Non-atomic initialization, single-pass.
4167 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01004168void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08004169 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004170{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004171 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07004172 unsigned long end_pfn = start_pfn + size;
4173 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004174 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004175
Hugh Dickins22b31ee2009-01-06 14:40:09 -08004176 if (highest_memmap_pfn < end_pfn - 1)
4177 highest_memmap_pfn = end_pfn - 1;
4178
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004179 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08004180 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08004181 /*
4182 * There can be holes in boot-time mem_map[]s
4183 * handed to this function. They do not
4184 * exist on hotplugged memory.
4185 */
4186 if (context == MEMMAP_EARLY) {
4187 if (!early_pfn_valid(pfn))
4188 continue;
4189 if (!early_pfn_in_nid(pfn, nid))
4190 continue;
4191 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004192 page = pfn_to_page(pfn);
4193 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07004194 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08004195 init_page_count(page);
Mel Gorman22b751c2013-02-22 16:34:59 -08004196 page_mapcount_reset(page);
Peter Zijlstra90572892013-10-07 11:29:20 +01004197 page_cpupid_reset_last(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004198 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004199 /*
4200 * Mark the block movable so that blocks are reserved for
4201 * movable at startup. This will force kernel allocations
4202 * to reserve their blocks rather than leaking throughout
4203 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07004204 * kernel allocations are made. Later some blocks near
4205 * the start are marked MIGRATE_RESERVE by
4206 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004207 *
4208 * bitmap is created for zone's valid pfn range. but memmap
4209 * can be created for invalid pages (for alignment)
4210 * check here not to call set_pageblock_migratetype() against
4211 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004212 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004213 if ((z->zone_start_pfn <= pfn)
Cody P Schafer108bcc92013-02-22 16:35:23 -08004214 && (pfn < zone_end_pfn(z))
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004215 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07004216 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004217
Linus Torvalds1da177e2005-04-16 15:20:36 -07004218 INIT_LIST_HEAD(&page->lru);
4219#ifdef WANT_PAGE_VIRTUAL
4220 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
4221 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07004222 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004223#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004224 }
4225}
4226
Andi Kleen1e548de2008-02-04 22:29:26 -08004227static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004228{
Mel Gorman7aeb09f2014-06-04 16:10:21 -07004229 unsigned int order, t;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004230 for_each_migratetype_order(order, t) {
4231 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004232 zone->free_area[order].nr_free = 0;
4233 }
4234}
4235
4236#ifndef __HAVE_ARCH_MEMMAP_INIT
4237#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08004238 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004239#endif
4240
David Rientjes7cd2b0a2014-06-23 13:22:04 -07004241static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004242{
David Howells3a6be872009-05-06 16:03:03 -07004243#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004244 int batch;
4245
4246 /*
4247 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07004248 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004249 *
4250 * OK, so we don't know how big the cache is. So guess.
4251 */
Jiang Liub40da042013-02-22 16:33:52 -08004252 batch = zone->managed_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07004253 if (batch * PAGE_SIZE > 512 * 1024)
4254 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004255 batch /= 4; /* We effectively *= 4 below */
4256 if (batch < 1)
4257 batch = 1;
4258
4259 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004260 * Clamp the batch to a 2^n - 1 value. Having a power
4261 * of 2 value was found to be more likely to have
4262 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004263 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004264 * For example if 2 tasks are alternately allocating
4265 * batches of pages, one task can end up with a lot
4266 * of pages of one half of the possible page colors
4267 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004268 */
David Howells91552032009-05-06 16:03:02 -07004269 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07004270
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004271 return batch;
David Howells3a6be872009-05-06 16:03:03 -07004272
4273#else
4274 /* The deferral and batching of frees should be suppressed under NOMMU
4275 * conditions.
4276 *
4277 * The problem is that NOMMU needs to be able to allocate large chunks
4278 * of contiguous memory as there's no hardware page translation to
4279 * assemble apparent contiguous memory from discontiguous pages.
4280 *
4281 * Queueing large contiguous runs of pages for batching, however,
4282 * causes the pages to actually be freed in smaller chunks. As there
4283 * can be a significant delay between the individual batches being
4284 * recycled, this leads to the once large chunks of space being
4285 * fragmented and becoming unavailable for high-order allocations.
4286 */
4287 return 0;
4288#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004289}
4290
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004291/*
4292 * pcp->high and pcp->batch values are related and dependent on one another:
4293 * ->batch must never be higher then ->high.
4294 * The following function updates them in a safe manner without read side
4295 * locking.
4296 *
4297 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4298 * those fields changing asynchronously (acording the the above rule).
4299 *
4300 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4301 * outside of boot time (or some other assurance that no concurrent updaters
4302 * exist).
4303 */
4304static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
4305 unsigned long batch)
4306{
4307 /* start with a fail safe value for batch */
4308 pcp->batch = 1;
4309 smp_wmb();
4310
4311 /* Update high, then batch, in order */
4312 pcp->high = high;
4313 smp_wmb();
4314
4315 pcp->batch = batch;
4316}
4317
Cody P Schafer36640332013-07-03 15:01:40 -07004318/* a companion to pageset_set_high() */
Cody P Schafer4008bab2013-07-03 15:01:28 -07004319static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
4320{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004321 pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
Cody P Schafer4008bab2013-07-03 15:01:28 -07004322}
4323
Cody P Schafer88c90db2013-07-03 15:01:35 -07004324static void pageset_init(struct per_cpu_pageset *p)
Christoph Lameter2caaad42005-06-21 17:15:00 -07004325{
4326 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004327 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004328
Magnus Damm1c6fe942005-10-26 01:58:59 -07004329 memset(p, 0, sizeof(*p));
4330
Christoph Lameter3dfa5722008-02-04 22:29:19 -08004331 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004332 pcp->count = 0;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004333 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
4334 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07004335}
4336
Cody P Schafer88c90db2013-07-03 15:01:35 -07004337static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
4338{
4339 pageset_init(p);
4340 pageset_set_batch(p, batch);
4341}
4342
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004343/*
Cody P Schafer36640332013-07-03 15:01:40 -07004344 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004345 * to the value high for the pageset p.
4346 */
Cody P Schafer36640332013-07-03 15:01:40 -07004347static void pageset_set_high(struct per_cpu_pageset *p,
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004348 unsigned long high)
4349{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004350 unsigned long batch = max(1UL, high / 4);
4351 if ((high / 4) > (PAGE_SHIFT * 8))
4352 batch = PAGE_SHIFT * 8;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004353
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004354 pageset_update(&p->pcp, high, batch);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004355}
4356
David Rientjes7cd2b0a2014-06-23 13:22:04 -07004357static void pageset_set_high_and_batch(struct zone *zone,
4358 struct per_cpu_pageset *pcp)
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004359{
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004360 if (percpu_pagelist_fraction)
Cody P Schafer36640332013-07-03 15:01:40 -07004361 pageset_set_high(pcp,
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004362 (zone->managed_pages /
4363 percpu_pagelist_fraction));
4364 else
4365 pageset_set_batch(pcp, zone_batchsize(zone));
4366}
4367
Cody P Schafer169f6c12013-07-03 15:01:41 -07004368static void __meminit zone_pageset_init(struct zone *zone, int cpu)
4369{
4370 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4371
4372 pageset_init(pcp);
4373 pageset_set_high_and_batch(zone, pcp);
4374}
4375
Jiang Liu4ed7e022012-07-31 16:43:35 -07004376static void __meminit setup_zone_pageset(struct zone *zone)
Wu Fengguang319774e2010-05-24 14:32:49 -07004377{
4378 int cpu;
Wu Fengguang319774e2010-05-24 14:32:49 -07004379 zone->pageset = alloc_percpu(struct per_cpu_pageset);
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004380 for_each_possible_cpu(cpu)
4381 zone_pageset_init(zone, cpu);
Wu Fengguang319774e2010-05-24 14:32:49 -07004382}
4383
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004384/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004385 * Allocate per cpu pagesets and initialize them.
4386 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07004387 */
Al Viro78d99552005-12-15 09:18:25 +00004388void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004389{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004390 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004391
Wu Fengguang319774e2010-05-24 14:32:49 -07004392 for_each_populated_zone(zone)
4393 setup_zone_pageset(zone);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004394}
4395
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004396static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07004397int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07004398{
4399 int i;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004400 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07004401
4402 /*
4403 * The per-page waitqueue mechanism uses hashed waitqueues
4404 * per zone.
4405 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07004406 zone->wait_table_hash_nr_entries =
4407 wait_table_hash_nr_entries(zone_size_pages);
4408 zone->wait_table_bits =
4409 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004410 alloc_size = zone->wait_table_hash_nr_entries
4411 * sizeof(wait_queue_head_t);
4412
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07004413 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07004414 zone->wait_table = (wait_queue_head_t *)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004415 memblock_virt_alloc_node_nopanic(
4416 alloc_size, zone->zone_pgdat->node_id);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004417 } else {
4418 /*
4419 * This case means that a zone whose size was 0 gets new memory
4420 * via memory hot-add.
4421 * But it may be the case that a new node was hot-added. In
4422 * this case vmalloc() will not be able to use this new node's
4423 * memory - this wait_table must be initialized to use this new
4424 * node itself as well.
4425 * To use this new node's memory, further consideration will be
4426 * necessary.
4427 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07004428 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004429 }
4430 if (!zone->wait_table)
4431 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07004432
Pintu Kumarb8af2942013-09-11 14:20:34 -07004433 for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07004434 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004435
4436 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004437}
4438
Matt Tolentinoc09b4242006-01-17 07:03:44 +01004439static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07004440{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004441 /*
4442 * per cpu subsystem is not up at this point. The following code
4443 * relies on the ability of the linker to provide the
4444 * offset of a (static) per cpu variable into the per cpu area.
4445 */
4446 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07004447
Xishi Qiub38a8722013-11-12 15:07:20 -08004448 if (populated_zone(zone))
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004449 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
4450 zone->name, zone->present_pages,
4451 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07004452}
4453
Jiang Liu4ed7e022012-07-31 16:43:35 -07004454int __meminit init_currently_empty_zone(struct zone *zone,
Yasunori Goto718127c2006-06-23 02:03:10 -07004455 unsigned long zone_start_pfn,
Dave Hansena2f3aa022007-01-10 23:15:30 -08004456 unsigned long size,
4457 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07004458{
4459 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004460 int ret;
4461 ret = zone_wait_table_init(zone, size);
4462 if (ret)
4463 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07004464 pgdat->nr_zones = zone_idx(zone) + 1;
4465
Dave Hansened8ece22005-10-29 18:16:50 -07004466 zone->zone_start_pfn = zone_start_pfn;
4467
Mel Gorman708614e2008-07-23 21:26:51 -07004468 mminit_dprintk(MMINIT_TRACE, "memmap_init",
4469 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4470 pgdat->node_id,
4471 (unsigned long)zone_idx(zone),
4472 zone_start_pfn, (zone_start_pfn + size));
4473
Andi Kleen1e548de2008-02-04 22:29:26 -08004474 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07004475
4476 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004477}
4478
Tejun Heo0ee332c2011-12-08 10:22:09 -08004479#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07004480#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4481/*
4482 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
Mel Gormanc7132162006-09-27 01:49:43 -07004483 */
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004484int __meminit __early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004485{
Tejun Heoc13291a2011-07-12 10:46:30 +02004486 unsigned long start_pfn, end_pfn;
Yinghai Lue76b63f2013-09-11 14:22:17 -07004487 int nid;
Russ Anderson7c243c72013-04-29 15:07:59 -07004488 /*
4489 * NOTE: The following SMP-unsafe globals are only used early in boot
4490 * when the kernel is running single-threaded.
4491 */
4492 static unsigned long __meminitdata last_start_pfn, last_end_pfn;
4493 static int __meminitdata last_nid;
4494
4495 if (last_start_pfn <= pfn && pfn < last_end_pfn)
4496 return last_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004497
Yinghai Lue76b63f2013-09-11 14:22:17 -07004498 nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
4499 if (nid != -1) {
4500 last_start_pfn = start_pfn;
4501 last_end_pfn = end_pfn;
4502 last_nid = nid;
4503 }
4504
4505 return nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004506}
4507#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4508
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004509int __meminit early_pfn_to_nid(unsigned long pfn)
4510{
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08004511 int nid;
4512
4513 nid = __early_pfn_to_nid(pfn);
4514 if (nid >= 0)
4515 return nid;
4516 /* just returns 0 */
4517 return 0;
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004518}
4519
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08004520#ifdef CONFIG_NODES_SPAN_OTHER_NODES
4521bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
4522{
4523 int nid;
4524
4525 nid = __early_pfn_to_nid(pfn);
4526 if (nid >= 0 && nid != node)
4527 return false;
4528 return true;
4529}
4530#endif
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004531
Mel Gormanc7132162006-09-27 01:49:43 -07004532/**
Santosh Shilimkar67828322014-01-21 15:50:25 -08004533 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004534 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
Santosh Shilimkar67828322014-01-21 15:50:25 -08004535 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
Mel Gormanc7132162006-09-27 01:49:43 -07004536 *
Zhang Zhen7d018172014-06-04 16:10:53 -07004537 * If an architecture guarantees that all ranges registered contain no holes
4538 * and may be freed, this this function may be used instead of calling
4539 * memblock_free_early_nid() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07004540 */
Tejun Heoc13291a2011-07-12 10:46:30 +02004541void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004542{
Tejun Heoc13291a2011-07-12 10:46:30 +02004543 unsigned long start_pfn, end_pfn;
4544 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004545
Tejun Heoc13291a2011-07-12 10:46:30 +02004546 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
4547 start_pfn = min(start_pfn, max_low_pfn);
4548 end_pfn = min(end_pfn, max_low_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004549
Tejun Heoc13291a2011-07-12 10:46:30 +02004550 if (start_pfn < end_pfn)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004551 memblock_free_early_nid(PFN_PHYS(start_pfn),
4552 (end_pfn - start_pfn) << PAGE_SHIFT,
4553 this_nid);
Mel Gormanc7132162006-09-27 01:49:43 -07004554 }
4555}
4556
4557/**
4558 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004559 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07004560 *
Zhang Zhen7d018172014-06-04 16:10:53 -07004561 * If an architecture guarantees that all ranges registered contain no holes and may
4562 * be freed, this function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07004563 */
4564void __init sparse_memory_present_with_active_regions(int nid)
4565{
Tejun Heoc13291a2011-07-12 10:46:30 +02004566 unsigned long start_pfn, end_pfn;
4567 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004568
Tejun Heoc13291a2011-07-12 10:46:30 +02004569 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
4570 memory_present(this_nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004571}
4572
4573/**
4574 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004575 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4576 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4577 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07004578 *
4579 * It returns the start and end page frame of a node based on information
Zhang Zhen7d018172014-06-04 16:10:53 -07004580 * provided by memblock_set_node(). If called for a node
Mel Gormanc7132162006-09-27 01:49:43 -07004581 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004582 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07004583 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004584void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004585 unsigned long *start_pfn, unsigned long *end_pfn)
4586{
Tejun Heoc13291a2011-07-12 10:46:30 +02004587 unsigned long this_start_pfn, this_end_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004588 int i;
Tejun Heoc13291a2011-07-12 10:46:30 +02004589
Mel Gormanc7132162006-09-27 01:49:43 -07004590 *start_pfn = -1UL;
4591 *end_pfn = 0;
4592
Tejun Heoc13291a2011-07-12 10:46:30 +02004593 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
4594 *start_pfn = min(*start_pfn, this_start_pfn);
4595 *end_pfn = max(*end_pfn, this_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004596 }
4597
Christoph Lameter633c0662007-10-16 01:25:37 -07004598 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07004599 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07004600}
4601
4602/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07004603 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4604 * assumption is made that zones within a node are ordered in monotonic
4605 * increasing memory addresses so that the "highest" populated zone is used
4606 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004607static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004608{
4609 int zone_index;
4610 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
4611 if (zone_index == ZONE_MOVABLE)
4612 continue;
4613
4614 if (arch_zone_highest_possible_pfn[zone_index] >
4615 arch_zone_lowest_possible_pfn[zone_index])
4616 break;
4617 }
4618
4619 VM_BUG_ON(zone_index == -1);
4620 movable_zone = zone_index;
4621}
4622
4623/*
4624 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004625 * because it is sized independent of architecture. Unlike the other zones,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004626 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4627 * in each node depending on the size of each node and how evenly kernelcore
4628 * is distributed. This helper function adjusts the zone ranges
4629 * provided by the architecture for a given node by using the end of the
4630 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4631 * zones within a node are in order of monotonic increases memory addresses
4632 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004633static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004634 unsigned long zone_type,
4635 unsigned long node_start_pfn,
4636 unsigned long node_end_pfn,
4637 unsigned long *zone_start_pfn,
4638 unsigned long *zone_end_pfn)
4639{
4640 /* Only adjust if ZONE_MOVABLE is on this node */
4641 if (zone_movable_pfn[nid]) {
4642 /* Size ZONE_MOVABLE */
4643 if (zone_type == ZONE_MOVABLE) {
4644 *zone_start_pfn = zone_movable_pfn[nid];
4645 *zone_end_pfn = min(node_end_pfn,
4646 arch_zone_highest_possible_pfn[movable_zone]);
4647
4648 /* Adjust for ZONE_MOVABLE starting within this range */
4649 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
4650 *zone_end_pfn > zone_movable_pfn[nid]) {
4651 *zone_end_pfn = zone_movable_pfn[nid];
4652
4653 /* Check if this whole range is within ZONE_MOVABLE */
4654 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
4655 *zone_start_pfn = *zone_end_pfn;
4656 }
4657}
4658
4659/*
Mel Gormanc7132162006-09-27 01:49:43 -07004660 * Return the number of pages a zone spans in a node, including holes
4661 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4662 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004663static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004664 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004665 unsigned long node_start_pfn,
4666 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004667 unsigned long *ignored)
4668{
Mel Gormanc7132162006-09-27 01:49:43 -07004669 unsigned long zone_start_pfn, zone_end_pfn;
4670
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004671 /* Get the start and end of the zone */
Mel Gormanc7132162006-09-27 01:49:43 -07004672 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4673 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07004674 adjust_zone_range_for_zone_movable(nid, zone_type,
4675 node_start_pfn, node_end_pfn,
4676 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004677
4678 /* Check that this node has pages within the zone's required range */
4679 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4680 return 0;
4681
4682 /* Move the zone boundaries inside the node if necessary */
4683 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4684 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4685
4686 /* Return the spanned pages */
4687 return zone_end_pfn - zone_start_pfn;
4688}
4689
4690/*
4691 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004692 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07004693 */
Yinghai Lu32996252009-12-15 17:59:02 -08004694unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004695 unsigned long range_start_pfn,
4696 unsigned long range_end_pfn)
4697{
Tejun Heo96e907d2011-07-12 10:46:29 +02004698 unsigned long nr_absent = range_end_pfn - range_start_pfn;
4699 unsigned long start_pfn, end_pfn;
4700 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004701
Tejun Heo96e907d2011-07-12 10:46:29 +02004702 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
4703 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
4704 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
4705 nr_absent -= end_pfn - start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004706 }
Tejun Heo96e907d2011-07-12 10:46:29 +02004707 return nr_absent;
Mel Gormanc7132162006-09-27 01:49:43 -07004708}
4709
4710/**
4711 * absent_pages_in_range - Return number of page frames in holes within a range
4712 * @start_pfn: The start PFN to start searching for holes
4713 * @end_pfn: The end PFN to stop searching for holes
4714 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004715 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07004716 */
4717unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4718 unsigned long end_pfn)
4719{
4720 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4721}
4722
4723/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004724static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004725 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004726 unsigned long node_start_pfn,
4727 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004728 unsigned long *ignored)
4729{
Tejun Heo96e907d2011-07-12 10:46:29 +02004730 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
4731 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman9c7cd682006-09-27 01:49:58 -07004732 unsigned long zone_start_pfn, zone_end_pfn;
4733
Tejun Heo96e907d2011-07-12 10:46:29 +02004734 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
4735 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004736
Mel Gorman2a1e2742007-07-17 04:03:12 -07004737 adjust_zone_range_for_zone_movable(nid, zone_type,
4738 node_start_pfn, node_end_pfn,
4739 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004740 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004741}
Mel Gorman0e0b8642006-09-27 01:49:56 -07004742
Tejun Heo0ee332c2011-12-08 10:22:09 -08004743#else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004744static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004745 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004746 unsigned long node_start_pfn,
4747 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004748 unsigned long *zones_size)
4749{
4750 return zones_size[zone_type];
4751}
4752
Paul Mundt6ea6e682007-07-15 23:38:20 -07004753static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004754 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004755 unsigned long node_start_pfn,
4756 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004757 unsigned long *zholes_size)
4758{
4759 if (!zholes_size)
4760 return 0;
4761
4762 return zholes_size[zone_type];
4763}
Yinghai Lu20e69262013-03-01 14:51:27 -08004764
Tejun Heo0ee332c2011-12-08 10:22:09 -08004765#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07004766
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004767static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004768 unsigned long node_start_pfn,
4769 unsigned long node_end_pfn,
4770 unsigned long *zones_size,
4771 unsigned long *zholes_size)
Mel Gormanc7132162006-09-27 01:49:43 -07004772{
4773 unsigned long realtotalpages, totalpages = 0;
4774 enum zone_type i;
4775
4776 for (i = 0; i < MAX_NR_ZONES; i++)
4777 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004778 node_start_pfn,
4779 node_end_pfn,
4780 zones_size);
Mel Gormanc7132162006-09-27 01:49:43 -07004781 pgdat->node_spanned_pages = totalpages;
4782
4783 realtotalpages = totalpages;
4784 for (i = 0; i < MAX_NR_ZONES; i++)
4785 realtotalpages -=
4786 zone_absent_pages_in_node(pgdat->node_id, i,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004787 node_start_pfn, node_end_pfn,
4788 zholes_size);
Mel Gormanc7132162006-09-27 01:49:43 -07004789 pgdat->node_present_pages = realtotalpages;
4790 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4791 realtotalpages);
4792}
4793
Mel Gorman835c1342007-10-16 01:25:47 -07004794#ifndef CONFIG_SPARSEMEM
4795/*
4796 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07004797 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4798 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07004799 * round what is now in bits to nearest long in bits, then return it in
4800 * bytes.
4801 */
Linus Torvalds7c455122013-02-18 09:58:02 -08004802static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07004803{
4804 unsigned long usemapsize;
4805
Linus Torvalds7c455122013-02-18 09:58:02 -08004806 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
Mel Gormand9c23402007-10-16 01:26:01 -07004807 usemapsize = roundup(zonesize, pageblock_nr_pages);
4808 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07004809 usemapsize *= NR_PAGEBLOCK_BITS;
4810 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4811
4812 return usemapsize / 8;
4813}
4814
4815static void __init setup_usemap(struct pglist_data *pgdat,
Linus Torvalds7c455122013-02-18 09:58:02 -08004816 struct zone *zone,
4817 unsigned long zone_start_pfn,
4818 unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07004819{
Linus Torvalds7c455122013-02-18 09:58:02 -08004820 unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
Mel Gorman835c1342007-10-16 01:25:47 -07004821 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08004822 if (usemapsize)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004823 zone->pageblock_flags =
4824 memblock_virt_alloc_node_nopanic(usemapsize,
4825 pgdat->node_id);
Mel Gorman835c1342007-10-16 01:25:47 -07004826}
4827#else
Linus Torvalds7c455122013-02-18 09:58:02 -08004828static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
4829 unsigned long zone_start_pfn, unsigned long zonesize) {}
Mel Gorman835c1342007-10-16 01:25:47 -07004830#endif /* CONFIG_SPARSEMEM */
4831
Mel Gormand9c23402007-10-16 01:26:01 -07004832#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08004833
Mel Gormand9c23402007-10-16 01:26:01 -07004834/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
Chen Gang15ca2202013-09-11 14:20:27 -07004835void __paginginit set_pageblock_order(void)
Mel Gormand9c23402007-10-16 01:26:01 -07004836{
Andrew Morton955c1cd2012-05-29 15:06:31 -07004837 unsigned int order;
4838
Mel Gormand9c23402007-10-16 01:26:01 -07004839 /* Check that pageblock_nr_pages has not already been setup */
4840 if (pageblock_order)
4841 return;
4842
Andrew Morton955c1cd2012-05-29 15:06:31 -07004843 if (HPAGE_SHIFT > PAGE_SHIFT)
4844 order = HUGETLB_PAGE_ORDER;
4845 else
4846 order = MAX_ORDER - 1;
4847
Mel Gormand9c23402007-10-16 01:26:01 -07004848 /*
4849 * Assume the largest contiguous order of interest is a huge page.
Andrew Morton955c1cd2012-05-29 15:06:31 -07004850 * This value may be variable depending on boot parameters on IA64 and
4851 * powerpc.
Mel Gormand9c23402007-10-16 01:26:01 -07004852 */
4853 pageblock_order = order;
4854}
4855#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4856
Mel Gormanba72cb82007-11-28 16:21:13 -08004857/*
4858 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
Andrew Morton955c1cd2012-05-29 15:06:31 -07004859 * is unused as pageblock_order is set at compile-time. See
4860 * include/linux/pageblock-flags.h for the values of pageblock_order based on
4861 * the kernel config
Mel Gormanba72cb82007-11-28 16:21:13 -08004862 */
Chen Gang15ca2202013-09-11 14:20:27 -07004863void __paginginit set_pageblock_order(void)
Mel Gormanba72cb82007-11-28 16:21:13 -08004864{
Mel Gormanba72cb82007-11-28 16:21:13 -08004865}
Mel Gormand9c23402007-10-16 01:26:01 -07004866
4867#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4868
Jiang Liu01cefae2012-12-12 13:52:19 -08004869static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
4870 unsigned long present_pages)
4871{
4872 unsigned long pages = spanned_pages;
4873
4874 /*
4875 * Provide a more accurate estimation if there are holes within
4876 * the zone and SPARSEMEM is in use. If there are holes within the
4877 * zone, each populated memory region may cost us one or two extra
4878 * memmap pages due to alignment because memmap pages for each
4879 * populated regions may not naturally algined on page boundary.
4880 * So the (present_pages >> 4) heuristic is a tradeoff for that.
4881 */
4882 if (spanned_pages > present_pages + (present_pages >> 4) &&
4883 IS_ENABLED(CONFIG_SPARSEMEM))
4884 pages = present_pages;
4885
4886 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
4887}
4888
Linus Torvalds1da177e2005-04-16 15:20:36 -07004889/*
4890 * Set up the zone data structures:
4891 * - mark all pages reserved
4892 * - mark all memory queues empty
4893 * - clear the memory bitmaps
Minchan Kim6527af52012-07-31 16:46:16 -07004894 *
4895 * NOTE: pgdat should get zeroed by caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004896 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08004897static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004898 unsigned long node_start_pfn, unsigned long node_end_pfn,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004899 unsigned long *zones_size, unsigned long *zholes_size)
4900{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07004901 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07004902 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004903 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07004904 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004905
Dave Hansen208d54e2005-10-29 18:16:52 -07004906 pgdat_resize_init(pgdat);
Andrea Arcangeli8177a422012-03-23 20:56:34 +01004907#ifdef CONFIG_NUMA_BALANCING
4908 spin_lock_init(&pgdat->numabalancing_migrate_lock);
4909 pgdat->numabalancing_migrate_nr_pages = 0;
4910 pgdat->numabalancing_migrate_next_window = jiffies;
4911#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004912 init_waitqueue_head(&pgdat->kswapd_wait);
Mel Gorman55150612012-07-31 16:44:35 -07004913 init_waitqueue_head(&pgdat->pfmemalloc_wait);
Joonsoo Kimeefa864b2014-12-12 16:55:46 -08004914 pgdat_page_ext_init(pgdat);
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01004915
Linus Torvalds1da177e2005-04-16 15:20:36 -07004916 for (j = 0; j < MAX_NR_ZONES; j++) {
4917 struct zone *zone = pgdat->node_zones + j;
Jiang Liu9feedc92012-12-12 13:52:12 -08004918 unsigned long size, realsize, freesize, memmap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004919
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004920 size = zone_spanned_pages_in_node(nid, j, node_start_pfn,
4921 node_end_pfn, zones_size);
Jiang Liu9feedc92012-12-12 13:52:12 -08004922 realsize = freesize = size - zone_absent_pages_in_node(nid, j,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004923 node_start_pfn,
4924 node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004925 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004926
Mel Gorman0e0b8642006-09-27 01:49:56 -07004927 /*
Jiang Liu9feedc92012-12-12 13:52:12 -08004928 * Adjust freesize so that it accounts for how much memory
Mel Gorman0e0b8642006-09-27 01:49:56 -07004929 * is used by this zone for memmap. This affects the watermark
4930 * and per-cpu initialisations
4931 */
Jiang Liu01cefae2012-12-12 13:52:19 -08004932 memmap_pages = calc_memmap_size(size, realsize);
Zhong Hongboba914f42014-12-12 16:56:21 -08004933 if (!is_highmem_idx(j)) {
4934 if (freesize >= memmap_pages) {
4935 freesize -= memmap_pages;
4936 if (memmap_pages)
4937 printk(KERN_DEBUG
4938 " %s zone: %lu pages used for memmap\n",
4939 zone_names[j], memmap_pages);
4940 } else
4941 printk(KERN_WARNING
4942 " %s zone: %lu pages exceeds freesize %lu\n",
4943 zone_names[j], memmap_pages, freesize);
4944 }
Mel Gorman0e0b8642006-09-27 01:49:56 -07004945
Christoph Lameter62672762007-02-10 01:43:07 -08004946 /* Account for reserved pages */
Jiang Liu9feedc92012-12-12 13:52:12 -08004947 if (j == 0 && freesize > dma_reserve) {
4948 freesize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07004949 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08004950 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004951 }
4952
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004953 if (!is_highmem_idx(j))
Jiang Liu9feedc92012-12-12 13:52:12 -08004954 nr_kernel_pages += freesize;
Jiang Liu01cefae2012-12-12 13:52:19 -08004955 /* Charge for highmem memmap if there are enough kernel pages */
4956 else if (nr_kernel_pages > memmap_pages * 2)
4957 nr_kernel_pages -= memmap_pages;
Jiang Liu9feedc92012-12-12 13:52:12 -08004958 nr_all_pages += freesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004959
4960 zone->spanned_pages = size;
Jiang Liu306f2e92013-02-22 16:33:54 -08004961 zone->present_pages = realsize;
Jiang Liu9feedc92012-12-12 13:52:12 -08004962 /*
4963 * Set an approximate value for lowmem here, it will be adjusted
4964 * when the bootmem allocator frees pages into the buddy system.
4965 * And all highmem pages will be managed by the buddy system.
4966 */
4967 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
Christoph Lameter96146342006-07-03 00:24:13 -07004968#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07004969 zone->node = nid;
Jiang Liu9feedc92012-12-12 13:52:12 -08004970 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07004971 / 100;
Jiang Liu9feedc92012-12-12 13:52:12 -08004972 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07004973#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004974 zone->name = zone_names[j];
4975 spin_lock_init(&zone->lock);
4976 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07004977 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004978 zone->zone_pgdat = pgdat;
Dave Hansened8ece22005-10-29 18:16:50 -07004979 zone_pcp_init(zone);
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07004980
4981 /* For bootup, initialized properly in watermark setup */
4982 mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages);
4983
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004984 lruvec_init(&zone->lruvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004985 if (!size)
4986 continue;
4987
Andrew Morton955c1cd2012-05-29 15:06:31 -07004988 set_pageblock_order();
Linus Torvalds7c455122013-02-18 09:58:02 -08004989 setup_usemap(pgdat, zone, zone_start_pfn, size);
Dave Hansena2f3aa022007-01-10 23:15:30 -08004990 ret = init_currently_empty_zone(zone, zone_start_pfn,
4991 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07004992 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07004993 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004994 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004995 }
4996}
4997
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004998static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004999{
Linus Torvalds1da177e2005-04-16 15:20:36 -07005000 /* Skip empty nodes */
5001 if (!pgdat->node_spanned_pages)
5002 return;
5003
Andy Whitcroftd41dee32005-06-23 00:07:54 -07005004#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07005005 /* ia64 gets its own node_mem_map, before this, without bootmem */
5006 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07005007 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07005008 struct page *map;
5009
Bob Piccoe984bb42006-05-20 15:00:31 -07005010 /*
5011 * The zone's endpoints aren't required to be MAX_ORDER
5012 * aligned but the node_mem_map endpoints must be in order
5013 * for the buddy allocator to function correctly.
5014 */
5015 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
Cody P Schafer108bcc92013-02-22 16:35:23 -08005016 end = pgdat_end_pfn(pgdat);
Bob Piccoe984bb42006-05-20 15:00:31 -07005017 end = ALIGN(end, MAX_ORDER_NR_PAGES);
5018 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07005019 map = alloc_remap(pgdat->node_id, size);
5020 if (!map)
Santosh Shilimkar67828322014-01-21 15:50:25 -08005021 map = memblock_virt_alloc_node_nopanic(size,
5022 pgdat->node_id);
Bob Piccoe984bb42006-05-20 15:00:31 -07005023 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005024 }
Roman Zippel12d810c2007-05-31 00:40:54 -07005025#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07005026 /*
5027 * With no DISCONTIG, the global mem_map is just set as node 0's
5028 */
Mel Gormanc7132162006-09-27 01:49:43 -07005029 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005030 mem_map = NODE_DATA(0)->node_mem_map;
Tejun Heo0ee332c2011-12-08 10:22:09 -08005031#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07005032 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08005033 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Tejun Heo0ee332c2011-12-08 10:22:09 -08005034#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005035 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005036#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07005037#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005038}
5039
Johannes Weiner9109fb72008-07-23 21:27:20 -07005040void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
5041 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005042{
Johannes Weiner9109fb72008-07-23 21:27:20 -07005043 pg_data_t *pgdat = NODE_DATA(nid);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005044 unsigned long start_pfn = 0;
5045 unsigned long end_pfn = 0;
Johannes Weiner9109fb72008-07-23 21:27:20 -07005046
Minchan Kim88fdf752012-07-31 16:46:14 -07005047 /* pg_data_t should be reset to zero when it's allocated */
Linus Torvalds8783b6e2012-08-02 10:37:03 -07005048 WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
Minchan Kim88fdf752012-07-31 16:46:14 -07005049
Linus Torvalds1da177e2005-04-16 15:20:36 -07005050 pgdat->node_id = nid;
5051 pgdat->node_start_pfn = node_start_pfn;
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005052#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5053 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
Juergen Gross8d29e182015-02-11 15:26:01 -08005054 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
5055 (u64)start_pfn << PAGE_SHIFT, ((u64)end_pfn << PAGE_SHIFT) - 1);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005056#endif
5057 calculate_node_totalpages(pgdat, start_pfn, end_pfn,
5058 zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005059
5060 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07005061#ifdef CONFIG_FLAT_NODE_MEM_MAP
5062 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
5063 nid, (unsigned long)pgdat,
5064 (unsigned long)pgdat->node_mem_map);
5065#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005066
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005067 free_area_init_core(pgdat, start_pfn, end_pfn,
5068 zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005069}
5070
Tejun Heo0ee332c2011-12-08 10:22:09 -08005071#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07005072
5073#if MAX_NUMNODES > 1
5074/*
5075 * Figure out the number of possible node ids.
5076 */
Cody P Schaferf9872ca2013-04-29 15:08:01 -07005077void __init setup_nr_node_ids(void)
Miklos Szeredi418508c2007-05-23 13:57:55 -07005078{
5079 unsigned int node;
5080 unsigned int highest = 0;
5081
5082 for_each_node_mask(node, node_possible_map)
5083 highest = node;
5084 nr_node_ids = highest + 1;
5085}
Miklos Szeredi418508c2007-05-23 13:57:55 -07005086#endif
5087
Mel Gormanc7132162006-09-27 01:49:43 -07005088/**
Tejun Heo1e019792011-07-12 09:45:34 +02005089 * node_map_pfn_alignment - determine the maximum internode alignment
5090 *
5091 * This function should be called after node map is populated and sorted.
5092 * It calculates the maximum power of two alignment which can distinguish
5093 * all the nodes.
5094 *
5095 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5096 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
5097 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
5098 * shifted, 1GiB is enough and this function will indicate so.
5099 *
5100 * This is used to test whether pfn -> nid mapping of the chosen memory
5101 * model has fine enough granularity to avoid incorrect mapping for the
5102 * populated node map.
5103 *
5104 * Returns the determined alignment in pfn's. 0 if there is no alignment
5105 * requirement (single node).
5106 */
5107unsigned long __init node_map_pfn_alignment(void)
5108{
5109 unsigned long accl_mask = 0, last_end = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02005110 unsigned long start, end, mask;
Tejun Heo1e019792011-07-12 09:45:34 +02005111 int last_nid = -1;
Tejun Heoc13291a2011-07-12 10:46:30 +02005112 int i, nid;
Tejun Heo1e019792011-07-12 09:45:34 +02005113
Tejun Heoc13291a2011-07-12 10:46:30 +02005114 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
Tejun Heo1e019792011-07-12 09:45:34 +02005115 if (!start || last_nid < 0 || last_nid == nid) {
5116 last_nid = nid;
5117 last_end = end;
5118 continue;
5119 }
5120
5121 /*
5122 * Start with a mask granular enough to pin-point to the
5123 * start pfn and tick off bits one-by-one until it becomes
5124 * too coarse to separate the current node from the last.
5125 */
5126 mask = ~((1 << __ffs(start)) - 1);
5127 while (mask && last_end <= (start & (mask << 1)))
5128 mask <<= 1;
5129
5130 /* accumulate all internode masks */
5131 accl_mask |= mask;
5132 }
5133
5134 /* convert mask to number of pages */
5135 return ~accl_mask + 1;
5136}
5137
Mel Gormana6af2bc2007-02-10 01:42:57 -08005138/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07005139static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07005140{
Mel Gormana6af2bc2007-02-10 01:42:57 -08005141 unsigned long min_pfn = ULONG_MAX;
Tejun Heoc13291a2011-07-12 10:46:30 +02005142 unsigned long start_pfn;
5143 int i;
Mel Gorman1abbfb42006-11-23 12:01:41 +00005144
Tejun Heoc13291a2011-07-12 10:46:30 +02005145 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
5146 min_pfn = min(min_pfn, start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005147
Mel Gormana6af2bc2007-02-10 01:42:57 -08005148 if (min_pfn == ULONG_MAX) {
5149 printk(KERN_WARNING
Paul Jackson2bc0d2612008-06-22 07:22:17 -07005150 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08005151 return 0;
5152 }
5153
5154 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005155}
5156
5157/**
5158 * find_min_pfn_with_active_regions - Find the minimum PFN registered
5159 *
5160 * It returns the minimum PFN based on information provided via
Zhang Zhen7d018172014-06-04 16:10:53 -07005161 * memblock_set_node().
Mel Gormanc7132162006-09-27 01:49:43 -07005162 */
5163unsigned long __init find_min_pfn_with_active_regions(void)
5164{
5165 return find_min_pfn_for_node(MAX_NUMNODES);
5166}
5167
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005168/*
5169 * early_calculate_totalpages()
5170 * Sum pages in active regions for movable zone.
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005171 * Populate N_MEMORY for calculating usable_nodes.
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005172 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07005173static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07005174{
Mel Gorman7e63efe2007-07-17 04:03:15 -07005175 unsigned long totalpages = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02005176 unsigned long start_pfn, end_pfn;
5177 int i, nid;
Mel Gorman7e63efe2007-07-17 04:03:15 -07005178
Tejun Heoc13291a2011-07-12 10:46:30 +02005179 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
5180 unsigned long pages = end_pfn - start_pfn;
5181
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005182 totalpages += pages;
5183 if (pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005184 node_set_state(nid, N_MEMORY);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005185 }
Pintu Kumarb8af2942013-09-11 14:20:34 -07005186 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07005187}
5188
Mel Gorman2a1e2742007-07-17 04:03:12 -07005189/*
5190 * Find the PFN the Movable zone begins in each node. Kernel memory
5191 * is spread evenly between nodes as long as the nodes have enough
5192 * memory. When they don't, some nodes will have more kernelcore than
5193 * others
5194 */
Kautuk Consulb224ef82012-03-21 16:34:15 -07005195static void __init find_zone_movable_pfns_for_nodes(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005196{
5197 int i, nid;
5198 unsigned long usable_startpfn;
5199 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07005200 /* save the state before borrow the nodemask */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005201 nodemask_t saved_node_state = node_states[N_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005202 unsigned long totalpages = early_calculate_totalpages();
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005203 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
Emil Medve136199f2014-04-07 15:37:52 -07005204 struct memblock_region *r;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005205
5206 /* Need to find movable_zone earlier when movable_node is specified. */
5207 find_usable_zone_for_movable();
Mel Gorman2a1e2742007-07-17 04:03:12 -07005208
Mel Gorman7e63efe2007-07-17 04:03:15 -07005209 /*
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005210 * If movable_node is specified, ignore kernelcore and movablecore
5211 * options.
5212 */
5213 if (movable_node_is_enabled()) {
Emil Medve136199f2014-04-07 15:37:52 -07005214 for_each_memblock(memory, r) {
5215 if (!memblock_is_hotpluggable(r))
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005216 continue;
5217
Emil Medve136199f2014-04-07 15:37:52 -07005218 nid = r->nid;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005219
Emil Medve136199f2014-04-07 15:37:52 -07005220 usable_startpfn = PFN_DOWN(r->base);
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005221 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
5222 min(usable_startpfn, zone_movable_pfn[nid]) :
5223 usable_startpfn;
5224 }
5225
5226 goto out2;
5227 }
5228
5229 /*
5230 * If movablecore=nn[KMG] was specified, calculate what size of
Mel Gorman7e63efe2007-07-17 04:03:15 -07005231 * kernelcore that corresponds so that memory usable for
5232 * any allocation type is evenly spread. If both kernelcore
5233 * and movablecore are specified, then the value of kernelcore
5234 * will be used for required_kernelcore if it's greater than
5235 * what movablecore would have allowed.
5236 */
5237 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07005238 unsigned long corepages;
5239
5240 /*
5241 * Round-up so that ZONE_MOVABLE is at least as large as what
5242 * was requested by the user
5243 */
5244 required_movablecore =
5245 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
5246 corepages = totalpages - required_movablecore;
5247
5248 required_kernelcore = max(required_kernelcore, corepages);
5249 }
5250
Yinghai Lu20e69262013-03-01 14:51:27 -08005251 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
5252 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07005253 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005254
5255 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
Mel Gorman2a1e2742007-07-17 04:03:12 -07005256 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
5257
5258restart:
5259 /* Spread kernelcore memory as evenly as possible throughout nodes */
5260 kernelcore_node = required_kernelcore / usable_nodes;
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005261 for_each_node_state(nid, N_MEMORY) {
Tejun Heoc13291a2011-07-12 10:46:30 +02005262 unsigned long start_pfn, end_pfn;
5263
Mel Gorman2a1e2742007-07-17 04:03:12 -07005264 /*
5265 * Recalculate kernelcore_node if the division per node
5266 * now exceeds what is necessary to satisfy the requested
5267 * amount of memory for the kernel
5268 */
5269 if (required_kernelcore < kernelcore_node)
5270 kernelcore_node = required_kernelcore / usable_nodes;
5271
5272 /*
5273 * As the map is walked, we track how much memory is usable
5274 * by the kernel using kernelcore_remaining. When it is
5275 * 0, the rest of the node is usable by ZONE_MOVABLE
5276 */
5277 kernelcore_remaining = kernelcore_node;
5278
5279 /* Go through each range of PFNs within this node */
Tejun Heoc13291a2011-07-12 10:46:30 +02005280 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005281 unsigned long size_pages;
5282
Tejun Heoc13291a2011-07-12 10:46:30 +02005283 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005284 if (start_pfn >= end_pfn)
5285 continue;
5286
5287 /* Account for what is only usable for kernelcore */
5288 if (start_pfn < usable_startpfn) {
5289 unsigned long kernel_pages;
5290 kernel_pages = min(end_pfn, usable_startpfn)
5291 - start_pfn;
5292
5293 kernelcore_remaining -= min(kernel_pages,
5294 kernelcore_remaining);
5295 required_kernelcore -= min(kernel_pages,
5296 required_kernelcore);
5297
5298 /* Continue if range is now fully accounted */
5299 if (end_pfn <= usable_startpfn) {
5300
5301 /*
5302 * Push zone_movable_pfn to the end so
5303 * that if we have to rebalance
5304 * kernelcore across nodes, we will
5305 * not double account here
5306 */
5307 zone_movable_pfn[nid] = end_pfn;
5308 continue;
5309 }
5310 start_pfn = usable_startpfn;
5311 }
5312
5313 /*
5314 * The usable PFN range for ZONE_MOVABLE is from
5315 * start_pfn->end_pfn. Calculate size_pages as the
5316 * number of pages used as kernelcore
5317 */
5318 size_pages = end_pfn - start_pfn;
5319 if (size_pages > kernelcore_remaining)
5320 size_pages = kernelcore_remaining;
5321 zone_movable_pfn[nid] = start_pfn + size_pages;
5322
5323 /*
5324 * Some kernelcore has been met, update counts and
5325 * break if the kernelcore for this node has been
Pintu Kumarb8af2942013-09-11 14:20:34 -07005326 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07005327 */
5328 required_kernelcore -= min(required_kernelcore,
5329 size_pages);
5330 kernelcore_remaining -= size_pages;
5331 if (!kernelcore_remaining)
5332 break;
5333 }
5334 }
5335
5336 /*
5337 * If there is still required_kernelcore, we do another pass with one
5338 * less node in the count. This will push zone_movable_pfn[nid] further
5339 * along on the nodes that still have memory until kernelcore is
Pintu Kumarb8af2942013-09-11 14:20:34 -07005340 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07005341 */
5342 usable_nodes--;
5343 if (usable_nodes && required_kernelcore > usable_nodes)
5344 goto restart;
5345
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005346out2:
Mel Gorman2a1e2742007-07-17 04:03:12 -07005347 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5348 for (nid = 0; nid < MAX_NUMNODES; nid++)
5349 zone_movable_pfn[nid] =
5350 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07005351
Yinghai Lu20e69262013-03-01 14:51:27 -08005352out:
Yinghai Lu66918dc2009-06-30 11:41:37 -07005353 /* restore the node_state */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005354 node_states[N_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005355}
5356
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005357/* Any regular or high memory on that node ? */
5358static void check_for_memory(pg_data_t *pgdat, int nid)
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005359{
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005360 enum zone_type zone_type;
5361
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005362 if (N_MEMORY == N_NORMAL_MEMORY)
5363 return;
5364
5365 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005366 struct zone *zone = &pgdat->node_zones[zone_type];
Xishi Qiub38a8722013-11-12 15:07:20 -08005367 if (populated_zone(zone)) {
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005368 node_set_state(nid, N_HIGH_MEMORY);
5369 if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
5370 zone_type <= ZONE_NORMAL)
5371 node_set_state(nid, N_NORMAL_MEMORY);
Bob Liud0048b02012-01-12 17:19:07 -08005372 break;
5373 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005374 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005375}
5376
Mel Gormanc7132162006-09-27 01:49:43 -07005377/**
5378 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005379 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07005380 *
5381 * This will call free_area_init_node() for each active node in the system.
Zhang Zhen7d018172014-06-04 16:10:53 -07005382 * Using the page ranges provided by memblock_set_node(), the size of each
Mel Gormanc7132162006-09-27 01:49:43 -07005383 * zone in each node and their holes is calculated. If the maximum PFN
5384 * between two adjacent zones match, it is assumed that the zone is empty.
5385 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5386 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5387 * starts where the previous one ended. For example, ZONE_DMA32 starts
5388 * at arch_max_dma_pfn.
5389 */
5390void __init free_area_init_nodes(unsigned long *max_zone_pfn)
5391{
Tejun Heoc13291a2011-07-12 10:46:30 +02005392 unsigned long start_pfn, end_pfn;
5393 int i, nid;
Mel Gormana6af2bc2007-02-10 01:42:57 -08005394
Mel Gormanc7132162006-09-27 01:49:43 -07005395 /* Record where the zone boundaries are */
5396 memset(arch_zone_lowest_possible_pfn, 0,
5397 sizeof(arch_zone_lowest_possible_pfn));
5398 memset(arch_zone_highest_possible_pfn, 0,
5399 sizeof(arch_zone_highest_possible_pfn));
5400 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
5401 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
5402 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005403 if (i == ZONE_MOVABLE)
5404 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07005405 arch_zone_lowest_possible_pfn[i] =
5406 arch_zone_highest_possible_pfn[i-1];
5407 arch_zone_highest_possible_pfn[i] =
5408 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
5409 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07005410 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
5411 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
5412
5413 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5414 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
Kautuk Consulb224ef82012-03-21 16:34:15 -07005415 find_zone_movable_pfns_for_nodes();
Mel Gormanc7132162006-09-27 01:49:43 -07005416
Mel Gormanc7132162006-09-27 01:49:43 -07005417 /* Print out the zone ranges */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005418 pr_info("Zone ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005419 for (i = 0; i < MAX_NR_ZONES; i++) {
5420 if (i == ZONE_MOVABLE)
5421 continue;
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005422 pr_info(" %-8s ", zone_names[i]);
David Rientjes72f0ba02010-03-05 13:42:14 -08005423 if (arch_zone_lowest_possible_pfn[i] ==
5424 arch_zone_highest_possible_pfn[i])
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005425 pr_cont("empty\n");
David Rientjes72f0ba02010-03-05 13:42:14 -08005426 else
Juergen Gross8d29e182015-02-11 15:26:01 -08005427 pr_cont("[mem %#018Lx-%#018Lx]\n",
5428 (u64)arch_zone_lowest_possible_pfn[i]
5429 << PAGE_SHIFT,
5430 ((u64)arch_zone_highest_possible_pfn[i]
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005431 << PAGE_SHIFT) - 1);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005432 }
5433
5434 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005435 pr_info("Movable zone start for each node\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005436 for (i = 0; i < MAX_NUMNODES; i++) {
5437 if (zone_movable_pfn[i])
Juergen Gross8d29e182015-02-11 15:26:01 -08005438 pr_info(" Node %d: %#018Lx\n", i,
5439 (u64)zone_movable_pfn[i] << PAGE_SHIFT);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005440 }
Mel Gormanc7132162006-09-27 01:49:43 -07005441
Wanpeng Lif2d52fe2012-10-08 16:32:24 -07005442 /* Print out the early node map */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005443 pr_info("Early memory node ranges\n");
Tejun Heoc13291a2011-07-12 10:46:30 +02005444 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
Juergen Gross8d29e182015-02-11 15:26:01 -08005445 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid,
5446 (u64)start_pfn << PAGE_SHIFT,
5447 ((u64)end_pfn << PAGE_SHIFT) - 1);
Mel Gormanc7132162006-09-27 01:49:43 -07005448
5449 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07005450 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08005451 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07005452 for_each_online_node(nid) {
5453 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07005454 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07005455 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005456
5457 /* Any memory on that node */
5458 if (pgdat->node_present_pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005459 node_set_state(nid, N_MEMORY);
5460 check_for_memory(pgdat, nid);
Mel Gormanc7132162006-09-27 01:49:43 -07005461 }
5462}
Mel Gorman2a1e2742007-07-17 04:03:12 -07005463
Mel Gorman7e63efe2007-07-17 04:03:15 -07005464static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005465{
5466 unsigned long long coremem;
5467 if (!p)
5468 return -EINVAL;
5469
5470 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005471 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005472
Mel Gorman7e63efe2007-07-17 04:03:15 -07005473 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07005474 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
5475
5476 return 0;
5477}
Mel Gormaned7ed362007-07-17 04:03:14 -07005478
Mel Gorman7e63efe2007-07-17 04:03:15 -07005479/*
5480 * kernelcore=size sets the amount of memory for use for allocations that
5481 * cannot be reclaimed or migrated.
5482 */
5483static int __init cmdline_parse_kernelcore(char *p)
5484{
5485 return cmdline_parse_core(p, &required_kernelcore);
5486}
5487
5488/*
5489 * movablecore=size sets the amount of memory for use for allocations that
5490 * can be reclaimed or migrated.
5491 */
5492static int __init cmdline_parse_movablecore(char *p)
5493{
5494 return cmdline_parse_core(p, &required_movablecore);
5495}
5496
Mel Gormaned7ed362007-07-17 04:03:14 -07005497early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005498early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07005499
Tejun Heo0ee332c2011-12-08 10:22:09 -08005500#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005501
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005502void adjust_managed_page_count(struct page *page, long count)
5503{
5504 spin_lock(&managed_page_count_lock);
5505 page_zone(page)->managed_pages += count;
5506 totalram_pages += count;
Jiang Liu3dcc0572013-07-03 15:03:21 -07005507#ifdef CONFIG_HIGHMEM
5508 if (PageHighMem(page))
5509 totalhigh_pages += count;
5510#endif
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005511 spin_unlock(&managed_page_count_lock);
5512}
Jiang Liu3dcc0572013-07-03 15:03:21 -07005513EXPORT_SYMBOL(adjust_managed_page_count);
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005514
Jiang Liu11199692013-07-03 15:02:48 -07005515unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
Jiang Liu69afade2013-04-29 15:06:21 -07005516{
Jiang Liu11199692013-07-03 15:02:48 -07005517 void *pos;
5518 unsigned long pages = 0;
Jiang Liu69afade2013-04-29 15:06:21 -07005519
Jiang Liu11199692013-07-03 15:02:48 -07005520 start = (void *)PAGE_ALIGN((unsigned long)start);
5521 end = (void *)((unsigned long)end & PAGE_MASK);
5522 for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
Jiang Liudbe67df2013-07-03 15:02:51 -07005523 if ((unsigned int)poison <= 0xFF)
Jiang Liu11199692013-07-03 15:02:48 -07005524 memset(pos, poison, PAGE_SIZE);
5525 free_reserved_page(virt_to_page(pos));
Jiang Liu69afade2013-04-29 15:06:21 -07005526 }
5527
5528 if (pages && s)
Jiang Liu11199692013-07-03 15:02:48 -07005529 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
Jiang Liu69afade2013-04-29 15:06:21 -07005530 s, pages << (PAGE_SHIFT - 10), start, end);
5531
5532 return pages;
5533}
Jiang Liu11199692013-07-03 15:02:48 -07005534EXPORT_SYMBOL(free_reserved_area);
Jiang Liu69afade2013-04-29 15:06:21 -07005535
Jiang Liucfa11e02013-04-29 15:07:00 -07005536#ifdef CONFIG_HIGHMEM
5537void free_highmem_page(struct page *page)
5538{
5539 __free_reserved_page(page);
5540 totalram_pages++;
Jiang Liu7b4b2a02013-07-03 15:03:11 -07005541 page_zone(page)->managed_pages++;
Jiang Liucfa11e02013-04-29 15:07:00 -07005542 totalhigh_pages++;
5543}
5544#endif
5545
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005546
5547void __init mem_init_print_info(const char *str)
5548{
5549 unsigned long physpages, codesize, datasize, rosize, bss_size;
5550 unsigned long init_code_size, init_data_size;
5551
5552 physpages = get_num_physpages();
5553 codesize = _etext - _stext;
5554 datasize = _edata - _sdata;
5555 rosize = __end_rodata - __start_rodata;
5556 bss_size = __bss_stop - __bss_start;
5557 init_data_size = __init_end - __init_begin;
5558 init_code_size = _einittext - _sinittext;
5559
5560 /*
5561 * Detect special cases and adjust section sizes accordingly:
5562 * 1) .init.* may be embedded into .data sections
5563 * 2) .init.text.* may be out of [__init_begin, __init_end],
5564 * please refer to arch/tile/kernel/vmlinux.lds.S.
5565 * 3) .rodata.* may be embedded into .text or .data sections.
5566 */
5567#define adj_init_size(start, end, size, pos, adj) \
Pintu Kumarb8af2942013-09-11 14:20:34 -07005568 do { \
5569 if (start <= pos && pos < end && size > adj) \
5570 size -= adj; \
5571 } while (0)
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005572
5573 adj_init_size(__init_begin, __init_end, init_data_size,
5574 _sinittext, init_code_size);
5575 adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
5576 adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
5577 adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
5578 adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
5579
5580#undef adj_init_size
5581
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005582 pr_info("Memory: %luK/%luK available "
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005583 "(%luK kernel code, %luK rwdata, %luK rodata, "
Pintu Kumare48322a2014-12-18 16:17:15 -08005584 "%luK init, %luK bss, %luK reserved, %luK cma-reserved"
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005585#ifdef CONFIG_HIGHMEM
5586 ", %luK highmem"
5587#endif
5588 "%s%s)\n",
5589 nr_free_pages() << (PAGE_SHIFT-10), physpages << (PAGE_SHIFT-10),
5590 codesize >> 10, datasize >> 10, rosize >> 10,
5591 (init_data_size + init_code_size) >> 10, bss_size >> 10,
Pintu Kumare48322a2014-12-18 16:17:15 -08005592 (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT-10),
5593 totalcma_pages << (PAGE_SHIFT-10),
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005594#ifdef CONFIG_HIGHMEM
5595 totalhigh_pages << (PAGE_SHIFT-10),
5596#endif
5597 str ? ", " : "", str ? str : "");
5598}
5599
Mel Gorman0e0b8642006-09-27 01:49:56 -07005600/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005601 * set_dma_reserve - set the specified number of pages reserved in the first zone
5602 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07005603 *
5604 * The per-cpu batchsize and zone watermarks are determined by present_pages.
5605 * In the DMA zone, a significant percentage may be consumed by kernel image
5606 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005607 * function may optionally be used to account for unfreeable pages in the
5608 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5609 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07005610 */
5611void __init set_dma_reserve(unsigned long new_dma_reserve)
5612{
5613 dma_reserve = new_dma_reserve;
5614}
5615
Linus Torvalds1da177e2005-04-16 15:20:36 -07005616void __init free_area_init(unsigned long *zones_size)
5617{
Johannes Weiner9109fb72008-07-23 21:27:20 -07005618 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005619 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
5620}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005621
Linus Torvalds1da177e2005-04-16 15:20:36 -07005622static int page_alloc_cpu_notify(struct notifier_block *self,
5623 unsigned long action, void *hcpu)
5624{
5625 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005626
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07005627 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -07005628 lru_add_drain_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005629 drain_pages(cpu);
5630
5631 /*
5632 * Spill the event counters of the dead processor
5633 * into the current processors event counters.
5634 * This artificially elevates the count of the current
5635 * processor.
5636 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07005637 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005638
5639 /*
5640 * Zero the differential counters of the dead processor
5641 * so that the vm statistics are consistent.
5642 *
5643 * This is only okay since the processor is dead and cannot
5644 * race with what we are doing.
5645 */
Christoph Lameter2bb921e2013-09-11 14:21:30 -07005646 cpu_vm_stats_fold(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005647 }
5648 return NOTIFY_OK;
5649}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005650
5651void __init page_alloc_init(void)
5652{
5653 hotcpu_notifier(page_alloc_cpu_notify, 0);
5654}
5655
5656/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005657 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5658 * or min_free_kbytes changes.
5659 */
5660static void calculate_totalreserve_pages(void)
5661{
5662 struct pglist_data *pgdat;
5663 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005664 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005665
5666 for_each_online_pgdat(pgdat) {
5667 for (i = 0; i < MAX_NR_ZONES; i++) {
5668 struct zone *zone = pgdat->node_zones + i;
Mel Gorman3484b2d2014-08-06 16:07:14 -07005669 long max = 0;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005670
5671 /* Find valid and maximum lowmem_reserve in the zone */
5672 for (j = i; j < MAX_NR_ZONES; j++) {
5673 if (zone->lowmem_reserve[j] > max)
5674 max = zone->lowmem_reserve[j];
5675 }
5676
Mel Gorman41858962009-06-16 15:32:12 -07005677 /* we treat the high watermark as reserved pages. */
5678 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005679
Jiang Liub40da042013-02-22 16:33:52 -08005680 if (max > zone->managed_pages)
5681 max = zone->managed_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005682 reserve_pages += max;
Johannes Weinerab8fabd2012-01-10 15:07:42 -08005683 /*
5684 * Lowmem reserves are not available to
5685 * GFP_HIGHUSER page cache allocations and
5686 * kswapd tries to balance zones to their high
5687 * watermark. As a result, neither should be
5688 * regarded as dirtyable memory, to prevent a
5689 * situation where reclaim has to clean pages
5690 * in order to balance the zones.
5691 */
5692 zone->dirty_balance_reserve = max;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005693 }
5694 }
Johannes Weinerab8fabd2012-01-10 15:07:42 -08005695 dirty_balance_reserve = reserve_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005696 totalreserve_pages = reserve_pages;
5697}
5698
5699/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07005700 * setup_per_zone_lowmem_reserve - called whenever
5701 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
5702 * has a correct pages reserved value, so an adequate number of
5703 * pages are left in the zone after a successful __alloc_pages().
5704 */
5705static void setup_per_zone_lowmem_reserve(void)
5706{
5707 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005708 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005709
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08005710 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005711 for (j = 0; j < MAX_NR_ZONES; j++) {
5712 struct zone *zone = pgdat->node_zones + j;
Jiang Liub40da042013-02-22 16:33:52 -08005713 unsigned long managed_pages = zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005714
5715 zone->lowmem_reserve[j] = 0;
5716
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005717 idx = j;
5718 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005719 struct zone *lower_zone;
5720
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005721 idx--;
5722
Linus Torvalds1da177e2005-04-16 15:20:36 -07005723 if (sysctl_lowmem_reserve_ratio[idx] < 1)
5724 sysctl_lowmem_reserve_ratio[idx] = 1;
5725
5726 lower_zone = pgdat->node_zones + idx;
Jiang Liub40da042013-02-22 16:33:52 -08005727 lower_zone->lowmem_reserve[j] = managed_pages /
Linus Torvalds1da177e2005-04-16 15:20:36 -07005728 sysctl_lowmem_reserve_ratio[idx];
Jiang Liub40da042013-02-22 16:33:52 -08005729 managed_pages += lower_zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005730 }
5731 }
5732 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005733
5734 /* update totalreserve_pages */
5735 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005736}
5737
Mel Gormancfd3da12011-04-25 21:36:42 +00005738static void __setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005739{
5740 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
5741 unsigned long lowmem_pages = 0;
5742 struct zone *zone;
5743 unsigned long flags;
5744
5745 /* Calculate total number of !ZONE_HIGHMEM pages */
5746 for_each_zone(zone) {
5747 if (!is_highmem(zone))
Jiang Liub40da042013-02-22 16:33:52 -08005748 lowmem_pages += zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005749 }
5750
5751 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07005752 u64 tmp;
5753
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005754 spin_lock_irqsave(&zone->lock, flags);
Jiang Liub40da042013-02-22 16:33:52 -08005755 tmp = (u64)pages_min * zone->managed_pages;
Andrew Mortonac924c62006-05-15 09:43:59 -07005756 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005757 if (is_highmem(zone)) {
5758 /*
Nick Piggin669ed172005-11-13 16:06:45 -08005759 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5760 * need highmem pages, so cap pages_min to a small
5761 * value here.
5762 *
Mel Gorman41858962009-06-16 15:32:12 -07005763 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Yaowei Bai42ff2702015-04-14 15:47:14 -07005764 * deltas control asynch page reclaim, and so should
Nick Piggin669ed172005-11-13 16:06:45 -08005765 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005766 */
Andrew Morton90ae8d62013-02-22 16:32:22 -08005767 unsigned long min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005768
Jiang Liub40da042013-02-22 16:33:52 -08005769 min_pages = zone->managed_pages / 1024;
Andrew Morton90ae8d62013-02-22 16:32:22 -08005770 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
Mel Gorman41858962009-06-16 15:32:12 -07005771 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005772 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08005773 /*
5774 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07005775 * proportionate to the zone's size.
5776 */
Mel Gorman41858962009-06-16 15:32:12 -07005777 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005778 }
5779
Mel Gorman41858962009-06-16 15:32:12 -07005780 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
5781 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Marek Szyprowski49f223a2012-01-25 12:49:24 +01005782
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07005783 __mod_zone_page_state(zone, NR_ALLOC_BATCH,
Johannes Weinerabe5f972014-10-02 16:21:10 -07005784 high_wmark_pages(zone) - low_wmark_pages(zone) -
5785 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07005786
Mel Gorman56fd56b2007-10-16 01:25:58 -07005787 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005788 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005789 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005790
5791 /* update totalreserve_pages */
5792 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005793}
5794
Mel Gormancfd3da12011-04-25 21:36:42 +00005795/**
5796 * setup_per_zone_wmarks - called when min_free_kbytes changes
5797 * or when memory is hot-{added|removed}
5798 *
5799 * Ensures that the watermark[min,low,high] values for each zone are set
5800 * correctly with respect to min_free_kbytes.
5801 */
5802void setup_per_zone_wmarks(void)
5803{
5804 mutex_lock(&zonelists_mutex);
5805 __setup_per_zone_wmarks();
5806 mutex_unlock(&zonelists_mutex);
5807}
5808
Randy Dunlap55a44622009-09-21 17:01:20 -07005809/*
Rik van Riel556adec2008-10-18 20:26:34 -07005810 * The inactive anon list should be small enough that the VM never has to
5811 * do too much work, but large enough that each inactive page has a chance
5812 * to be referenced again before it is swapped out.
5813 *
5814 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5815 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5816 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5817 * the anonymous pages are kept on the inactive list.
5818 *
5819 * total target max
5820 * memory ratio inactive anon
5821 * -------------------------------------
5822 * 10MB 1 5MB
5823 * 100MB 1 50MB
5824 * 1GB 3 250MB
5825 * 10GB 10 0.9GB
5826 * 100GB 31 3GB
5827 * 1TB 101 10GB
5828 * 10TB 320 32GB
5829 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005830static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
Minchan Kim96cb4df2009-06-16 15:32:49 -07005831{
5832 unsigned int gb, ratio;
5833
5834 /* Zone size in gigabytes */
Jiang Liub40da042013-02-22 16:33:52 -08005835 gb = zone->managed_pages >> (30 - PAGE_SHIFT);
Minchan Kim96cb4df2009-06-16 15:32:49 -07005836 if (gb)
5837 ratio = int_sqrt(10 * gb);
5838 else
5839 ratio = 1;
5840
5841 zone->inactive_ratio = ratio;
5842}
5843
KOSAKI Motohiro839a4fc2011-05-24 17:11:31 -07005844static void __meminit setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07005845{
5846 struct zone *zone;
5847
Minchan Kim96cb4df2009-06-16 15:32:49 -07005848 for_each_zone(zone)
5849 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07005850}
5851
Linus Torvalds1da177e2005-04-16 15:20:36 -07005852/*
5853 * Initialise min_free_kbytes.
5854 *
5855 * For small machines we want it small (128k min). For large machines
5856 * we want it large (64MB max). But it is not linear, because network
5857 * bandwidth does not increase linearly with machine size. We use
5858 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07005859 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
Linus Torvalds1da177e2005-04-16 15:20:36 -07005860 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5861 *
5862 * which yields
5863 *
5864 * 16MB: 512k
5865 * 32MB: 724k
5866 * 64MB: 1024k
5867 * 128MB: 1448k
5868 * 256MB: 2048k
5869 * 512MB: 2896k
5870 * 1024MB: 4096k
5871 * 2048MB: 5792k
5872 * 4096MB: 8192k
5873 * 8192MB: 11584k
5874 * 16384MB: 16384k
5875 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005876int __meminit init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005877{
5878 unsigned long lowmem_kbytes;
Michal Hocko5f127332013-07-08 16:00:40 -07005879 int new_min_free_kbytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005880
5881 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
Michal Hocko5f127332013-07-08 16:00:40 -07005882 new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005883
Michal Hocko5f127332013-07-08 16:00:40 -07005884 if (new_min_free_kbytes > user_min_free_kbytes) {
5885 min_free_kbytes = new_min_free_kbytes;
5886 if (min_free_kbytes < 128)
5887 min_free_kbytes = 128;
5888 if (min_free_kbytes > 65536)
5889 min_free_kbytes = 65536;
5890 } else {
5891 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
5892 new_min_free_kbytes, user_min_free_kbytes);
5893 }
Minchan Kimbc75d332009-06-16 15:32:48 -07005894 setup_per_zone_wmarks();
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -07005895 refresh_zone_stat_thresholds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005896 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07005897 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005898 return 0;
5899}
Minchan Kimbc75d332009-06-16 15:32:48 -07005900module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005901
5902/*
Pintu Kumarb8af2942013-09-11 14:20:34 -07005903 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
Linus Torvalds1da177e2005-04-16 15:20:36 -07005904 * that we can call two helper functions whenever min_free_kbytes
5905 * changes.
5906 */
Joe Perchescccad5b2014-06-06 14:38:09 -07005907int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005908 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005909{
Han Pingtianda8c7572014-01-23 15:53:17 -08005910 int rc;
5911
5912 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5913 if (rc)
5914 return rc;
5915
Michal Hocko5f127332013-07-08 16:00:40 -07005916 if (write) {
5917 user_min_free_kbytes = min_free_kbytes;
Minchan Kimbc75d332009-06-16 15:32:48 -07005918 setup_per_zone_wmarks();
Michal Hocko5f127332013-07-08 16:00:40 -07005919 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005920 return 0;
5921}
5922
Christoph Lameter96146342006-07-03 00:24:13 -07005923#ifdef CONFIG_NUMA
Joe Perchescccad5b2014-06-06 14:38:09 -07005924int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005925 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07005926{
5927 struct zone *zone;
5928 int rc;
5929
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005930 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07005931 if (rc)
5932 return rc;
5933
5934 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08005935 zone->min_unmapped_pages = (zone->managed_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07005936 sysctl_min_unmapped_ratio) / 100;
5937 return 0;
5938}
Christoph Lameter0ff38492006-09-25 23:31:52 -07005939
Joe Perchescccad5b2014-06-06 14:38:09 -07005940int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005941 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07005942{
5943 struct zone *zone;
5944 int rc;
5945
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005946 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07005947 if (rc)
5948 return rc;
5949
5950 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08005951 zone->min_slab_pages = (zone->managed_pages *
Christoph Lameter0ff38492006-09-25 23:31:52 -07005952 sysctl_min_slab_ratio) / 100;
5953 return 0;
5954}
Christoph Lameter96146342006-07-03 00:24:13 -07005955#endif
5956
Linus Torvalds1da177e2005-04-16 15:20:36 -07005957/*
5958 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5959 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5960 * whenever sysctl_lowmem_reserve_ratio changes.
5961 *
5962 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07005963 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07005964 * if in function of the boot time zone sizes.
5965 */
Joe Perchescccad5b2014-06-06 14:38:09 -07005966int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005967 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005968{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005969 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005970 setup_per_zone_lowmem_reserve();
5971 return 0;
5972}
5973
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005974/*
5975 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
Pintu Kumarb8af2942013-09-11 14:20:34 -07005976 * cpu. It is the fraction of total pages in each zone that a hot per cpu
5977 * pagelist can have before it gets flushed back to buddy allocator.
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005978 */
Joe Perchescccad5b2014-06-06 14:38:09 -07005979int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005980 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005981{
5982 struct zone *zone;
David Rientjes7cd2b0a2014-06-23 13:22:04 -07005983 int old_percpu_pagelist_fraction;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005984 int ret;
5985
Cody P Schaferc8e251f2013-07-03 15:01:29 -07005986 mutex_lock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07005987 old_percpu_pagelist_fraction = percpu_pagelist_fraction;
5988
5989 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
5990 if (!write || ret < 0)
5991 goto out;
5992
5993 /* Sanity checking to avoid pcp imbalance */
5994 if (percpu_pagelist_fraction &&
5995 percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
5996 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
5997 ret = -EINVAL;
5998 goto out;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005999 }
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006000
6001 /* No change? */
6002 if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
6003 goto out;
6004
6005 for_each_populated_zone(zone) {
6006 unsigned int cpu;
6007
6008 for_each_possible_cpu(cpu)
6009 pageset_set_high_and_batch(zone,
6010 per_cpu_ptr(zone->pageset, cpu));
6011 }
6012out:
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006013 mutex_unlock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006014 return ret;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006015}
6016
David S. Millerf034b5d2006-08-24 03:08:07 -07006017int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006018
6019#ifdef CONFIG_NUMA
6020static int __init set_hashdist(char *str)
6021{
6022 if (!str)
6023 return 0;
6024 hashdist = simple_strtoul(str, &str, 0);
6025 return 1;
6026}
6027__setup("hashdist=", set_hashdist);
6028#endif
6029
6030/*
6031 * allocate a large system hash table from bootmem
6032 * - it is assumed that the hash table must contain an exact power-of-2
6033 * quantity of entries
6034 * - limit is the number of hash buckets, not the total allocation size
6035 */
6036void *__init alloc_large_system_hash(const char *tablename,
6037 unsigned long bucketsize,
6038 unsigned long numentries,
6039 int scale,
6040 int flags,
6041 unsigned int *_hash_shift,
6042 unsigned int *_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00006043 unsigned long low_limit,
6044 unsigned long high_limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006045{
Tim Bird31fe62b2012-05-23 13:33:35 +00006046 unsigned long long max = high_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006047 unsigned long log2qty, size;
6048 void *table = NULL;
6049
6050 /* allow the kernel cmdline to have a say */
6051 if (!numentries) {
6052 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08006053 numentries = nr_kernel_pages;
Jerry Zhoua7e83312013-09-11 14:20:26 -07006054
6055 /* It isn't necessary when PAGE_SIZE >= 1MB */
6056 if (PAGE_SHIFT < 20)
6057 numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006058
6059 /* limit to 1 bucket per 2^scale bytes of low memory */
6060 if (scale > PAGE_SHIFT)
6061 numentries >>= (scale - PAGE_SHIFT);
6062 else
6063 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08006064
6065 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07006066 if (unlikely(flags & HASH_SMALL)) {
6067 /* Makes no sense without HASH_EARLY */
6068 WARN_ON(!(flags & HASH_EARLY));
6069 if (!(numentries >> *_hash_shift)) {
6070 numentries = 1UL << *_hash_shift;
6071 BUG_ON(!numentries);
6072 }
6073 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08006074 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006075 }
John Hawkes6e692ed2006-03-25 03:08:02 -08006076 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006077
6078 /* limit allocation size to 1/16 total memory by default */
6079 if (max == 0) {
6080 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
6081 do_div(max, bucketsize);
6082 }
Dimitri Sivanich074b8512012-02-08 12:39:07 -08006083 max = min(max, 0x80000000ULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006084
Tim Bird31fe62b2012-05-23 13:33:35 +00006085 if (numentries < low_limit)
6086 numentries = low_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006087 if (numentries > max)
6088 numentries = max;
6089
David Howellsf0d1b0b2006-12-08 02:37:49 -08006090 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006091
6092 do {
6093 size = bucketsize << log2qty;
6094 if (flags & HASH_EARLY)
Santosh Shilimkar67828322014-01-21 15:50:25 -08006095 table = memblock_virt_alloc_nopanic(size, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006096 else if (hashdist)
6097 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
6098 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07006099 /*
6100 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07006101 * some pages at the end of hash table which
6102 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07006103 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01006104 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07006105 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01006106 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
6107 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006108 }
6109 } while (!table && size > PAGE_SIZE && --log2qty);
6110
6111 if (!table)
6112 panic("Failed to allocate %s hash table\n", tablename);
6113
Robin Holtf241e6602010-10-07 12:59:26 -07006114 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07006115 tablename,
Robin Holtf241e6602010-10-07 12:59:26 -07006116 (1UL << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08006117 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07006118 size);
6119
6120 if (_hash_shift)
6121 *_hash_shift = log2qty;
6122 if (_hash_mask)
6123 *_hash_mask = (1 << log2qty) - 1;
6124
6125 return table;
6126}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08006127
Mel Gorman835c1342007-10-16 01:25:47 -07006128/* Return a pointer to the bitmap storing bits affecting a block of pages */
6129static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
6130 unsigned long pfn)
6131{
6132#ifdef CONFIG_SPARSEMEM
6133 return __pfn_to_section(pfn)->pageblock_flags;
6134#else
6135 return zone->pageblock_flags;
6136#endif /* CONFIG_SPARSEMEM */
6137}
Andrew Morton6220ec72006-10-19 23:29:05 -07006138
Mel Gorman835c1342007-10-16 01:25:47 -07006139static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
6140{
6141#ifdef CONFIG_SPARSEMEM
6142 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07006143 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07006144#else
Laura Abbottc060f942013-01-11 14:31:51 -08006145 pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
Mel Gormand9c23402007-10-16 01:26:01 -07006146 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07006147#endif /* CONFIG_SPARSEMEM */
6148}
6149
6150/**
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006151 * 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 -07006152 * @page: The page within the block of interest
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006153 * @pfn: The target page frame number
6154 * @end_bitidx: The last bit of interest to retrieve
6155 * @mask: mask of bits that the caller is interested in
6156 *
6157 * Return: pageblock_bits flags
Mel Gorman835c1342007-10-16 01:25:47 -07006158 */
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006159unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
Mel Gormane58469b2014-06-04 16:10:16 -07006160 unsigned long end_bitidx,
6161 unsigned long mask)
Mel Gorman835c1342007-10-16 01:25:47 -07006162{
6163 struct zone *zone;
6164 unsigned long *bitmap;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006165 unsigned long bitidx, word_bitidx;
Mel Gormane58469b2014-06-04 16:10:16 -07006166 unsigned long word;
Mel Gorman835c1342007-10-16 01:25:47 -07006167
6168 zone = page_zone(page);
Mel Gorman835c1342007-10-16 01:25:47 -07006169 bitmap = get_pageblock_bitmap(zone, pfn);
6170 bitidx = pfn_to_bitidx(zone, pfn);
Mel Gormane58469b2014-06-04 16:10:16 -07006171 word_bitidx = bitidx / BITS_PER_LONG;
6172 bitidx &= (BITS_PER_LONG-1);
Mel Gorman835c1342007-10-16 01:25:47 -07006173
Mel Gormane58469b2014-06-04 16:10:16 -07006174 word = bitmap[word_bitidx];
6175 bitidx += end_bitidx;
6176 return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
Mel Gorman835c1342007-10-16 01:25:47 -07006177}
6178
6179/**
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006180 * 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 -07006181 * @page: The page within the block of interest
Mel Gorman835c1342007-10-16 01:25:47 -07006182 * @flags: The flags to set
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006183 * @pfn: The target page frame number
6184 * @end_bitidx: The last bit of interest
6185 * @mask: mask of bits that the caller is interested in
Mel Gorman835c1342007-10-16 01:25:47 -07006186 */
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006187void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
6188 unsigned long pfn,
Mel Gormane58469b2014-06-04 16:10:16 -07006189 unsigned long end_bitidx,
6190 unsigned long mask)
Mel Gorman835c1342007-10-16 01:25:47 -07006191{
6192 struct zone *zone;
6193 unsigned long *bitmap;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006194 unsigned long bitidx, word_bitidx;
Mel Gormane58469b2014-06-04 16:10:16 -07006195 unsigned long old_word, word;
6196
6197 BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
Mel Gorman835c1342007-10-16 01:25:47 -07006198
6199 zone = page_zone(page);
Mel Gorman835c1342007-10-16 01:25:47 -07006200 bitmap = get_pageblock_bitmap(zone, pfn);
6201 bitidx = pfn_to_bitidx(zone, pfn);
Mel Gormane58469b2014-06-04 16:10:16 -07006202 word_bitidx = bitidx / BITS_PER_LONG;
6203 bitidx &= (BITS_PER_LONG-1);
6204
Sasha Levin309381fea2014-01-23 15:52:54 -08006205 VM_BUG_ON_PAGE(!zone_spans_pfn(zone, pfn), page);
Mel Gorman835c1342007-10-16 01:25:47 -07006206
Mel Gormane58469b2014-06-04 16:10:16 -07006207 bitidx += end_bitidx;
6208 mask <<= (BITS_PER_LONG - bitidx - 1);
6209 flags <<= (BITS_PER_LONG - bitidx - 1);
6210
Jason Low4db0c3c2015-04-15 16:14:08 -07006211 word = READ_ONCE(bitmap[word_bitidx]);
Mel Gormane58469b2014-06-04 16:10:16 -07006212 for (;;) {
6213 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
6214 if (word == old_word)
6215 break;
6216 word = old_word;
6217 }
Mel Gorman835c1342007-10-16 01:25:47 -07006218}
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07006219
6220/*
Minchan Kim80934512012-07-31 16:43:01 -07006221 * This function checks whether pageblock includes unmovable pages or not.
6222 * If @count is not zero, it is okay to include less @count unmovable pages
6223 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07006224 * PageLRU check without isolation or lru_lock could race so that
Minchan Kim80934512012-07-31 16:43:01 -07006225 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6226 * expect this function should be exact.
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07006227 */
Wen Congyangb023f462012-12-11 16:00:45 -08006228bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
6229 bool skip_hwpoisoned_pages)
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006230{
6231 unsigned long pfn, iter, found;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01006232 int mt;
6233
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006234 /*
6235 * For avoiding noise data, lru_add_drain_all() should be called
Minchan Kim80934512012-07-31 16:43:01 -07006236 * If ZONE_MOVABLE, the zone never contains unmovable pages
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006237 */
6238 if (zone_idx(zone) == ZONE_MOVABLE)
Minchan Kim80934512012-07-31 16:43:01 -07006239 return false;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01006240 mt = get_pageblock_migratetype(page);
6241 if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
Minchan Kim80934512012-07-31 16:43:01 -07006242 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006243
6244 pfn = page_to_pfn(page);
6245 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
6246 unsigned long check = pfn + iter;
6247
Namhyung Kim29723fc2011-02-25 14:44:25 -08006248 if (!pfn_valid_within(check))
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006249 continue;
Namhyung Kim29723fc2011-02-25 14:44:25 -08006250
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006251 page = pfn_to_page(check);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07006252
6253 /*
6254 * Hugepages are not in LRU lists, but they're movable.
6255 * We need not scan over tail pages bacause we don't
6256 * handle each tail page individually in migration.
6257 */
6258 if (PageHuge(page)) {
6259 iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
6260 continue;
6261 }
6262
Minchan Kim97d255c2012-07-31 16:42:59 -07006263 /*
6264 * We can't use page_count without pin a page
6265 * because another CPU can free compound page.
6266 * This check already skips compound tails of THP
6267 * because their page->_count is zero at all time.
6268 */
6269 if (!atomic_read(&page->_count)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006270 if (PageBuddy(page))
6271 iter += (1 << page_order(page)) - 1;
6272 continue;
6273 }
Minchan Kim97d255c2012-07-31 16:42:59 -07006274
Wen Congyangb023f462012-12-11 16:00:45 -08006275 /*
6276 * The HWPoisoned page may be not in buddy system, and
6277 * page_count() is not 0.
6278 */
6279 if (skip_hwpoisoned_pages && PageHWPoison(page))
6280 continue;
6281
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006282 if (!PageLRU(page))
6283 found++;
6284 /*
Johannes Weiner6b4f7792014-12-12 16:56:13 -08006285 * If there are RECLAIMABLE pages, we need to check
6286 * it. But now, memory offline itself doesn't call
6287 * shrink_node_slabs() and it still to be fixed.
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006288 */
6289 /*
6290 * If the page is not RAM, page_count()should be 0.
6291 * we don't need more check. This is an _used_ not-movable page.
6292 *
6293 * The problematic thing here is PG_reserved pages. PG_reserved
6294 * is set to both of a memory hole page and a _used_ kernel
6295 * page at boot.
6296 */
6297 if (found > count)
Minchan Kim80934512012-07-31 16:43:01 -07006298 return true;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006299 }
Minchan Kim80934512012-07-31 16:43:01 -07006300 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006301}
6302
6303bool is_pageblock_removable_nolock(struct page *page)
6304{
Michal Hocko656a0702012-01-20 14:33:58 -08006305 struct zone *zone;
6306 unsigned long pfn;
Michal Hocko687875f2012-01-20 14:33:55 -08006307
6308 /*
6309 * We have to be careful here because we are iterating over memory
6310 * sections which are not zone aware so we might end up outside of
6311 * the zone but still within the section.
Michal Hocko656a0702012-01-20 14:33:58 -08006312 * We have to take care about the node as well. If the node is offline
6313 * its NODE_DATA will be NULL - see page_zone.
Michal Hocko687875f2012-01-20 14:33:55 -08006314 */
Michal Hocko656a0702012-01-20 14:33:58 -08006315 if (!node_online(page_to_nid(page)))
6316 return false;
6317
6318 zone = page_zone(page);
6319 pfn = page_to_pfn(page);
Cody P Schafer108bcc92013-02-22 16:35:23 -08006320 if (!zone_spans_pfn(zone, pfn))
Michal Hocko687875f2012-01-20 14:33:55 -08006321 return false;
6322
Wen Congyangb023f462012-12-11 16:00:45 -08006323 return !has_unmovable_pages(zone, page, 0, true);
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07006324}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006325
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006326#ifdef CONFIG_CMA
6327
6328static unsigned long pfn_max_align_down(unsigned long pfn)
6329{
6330 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
6331 pageblock_nr_pages) - 1);
6332}
6333
6334static unsigned long pfn_max_align_up(unsigned long pfn)
6335{
6336 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
6337 pageblock_nr_pages));
6338}
6339
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006340/* [start, end) must belong to a single zone. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006341static int __alloc_contig_migrate_range(struct compact_control *cc,
6342 unsigned long start, unsigned long end)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006343{
6344 /* This function is based on compact_zone() from compaction.c. */
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006345 unsigned long nr_reclaimed;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006346 unsigned long pfn = start;
6347 unsigned int tries = 0;
6348 int ret = 0;
6349
Marek Szyprowskibe49a6e2012-12-12 13:51:19 -08006350 migrate_prep();
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006351
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006352 while (pfn < end || !list_empty(&cc->migratepages)) {
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006353 if (fatal_signal_pending(current)) {
6354 ret = -EINTR;
6355 break;
6356 }
6357
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006358 if (list_empty(&cc->migratepages)) {
6359 cc->nr_migratepages = 0;
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07006360 pfn = isolate_migratepages_range(cc, pfn, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006361 if (!pfn) {
6362 ret = -EINTR;
6363 break;
6364 }
6365 tries = 0;
6366 } else if (++tries == 5) {
6367 ret = ret < 0 ? ret : -EBUSY;
6368 break;
6369 }
6370
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006371 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
6372 &cc->migratepages);
6373 cc->nr_migratepages -= nr_reclaimed;
Minchan Kim02c6de82012-10-08 16:31:55 -07006374
Hugh Dickins9c620e22013-02-22 16:35:14 -08006375 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
David Rientjese0b9dae2014-06-04 16:08:28 -07006376 NULL, 0, cc->mode, MR_CMA);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006377 }
Srinivas Pandruvada2a6f5122013-02-22 16:32:09 -08006378 if (ret < 0) {
6379 putback_movable_pages(&cc->migratepages);
6380 return ret;
6381 }
6382 return 0;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006383}
6384
6385/**
6386 * alloc_contig_range() -- tries to allocate given range of pages
6387 * @start: start PFN to allocate
6388 * @end: one-past-the-last PFN to allocate
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006389 * @migratetype: migratetype of the underlaying pageblocks (either
6390 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6391 * in range must have the same migratetype and it must
6392 * be either of the two.
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006393 *
6394 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6395 * aligned, however it's the caller's responsibility to guarantee that
6396 * we are the only thread that changes migrate type of pageblocks the
6397 * pages fall in.
6398 *
6399 * The PFN range must belong to a single zone.
6400 *
6401 * Returns zero on success or negative error code. On success all
6402 * pages which PFN is in [start, end) are allocated for the caller and
6403 * need to be freed with free_contig_range().
6404 */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006405int alloc_contig_range(unsigned long start, unsigned long end,
6406 unsigned migratetype)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006407{
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006408 unsigned long outer_start, outer_end;
6409 int ret = 0, order;
6410
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006411 struct compact_control cc = {
6412 .nr_migratepages = 0,
6413 .order = -1,
6414 .zone = page_zone(pfn_to_page(start)),
David Rientjese0b9dae2014-06-04 16:08:28 -07006415 .mode = MIGRATE_SYNC,
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006416 .ignore_skip_hint = true,
6417 };
6418 INIT_LIST_HEAD(&cc.migratepages);
6419
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006420 /*
6421 * What we do here is we mark all pageblocks in range as
6422 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6423 * have different sizes, and due to the way page allocator
6424 * work, we align the range to biggest of the two pages so
6425 * that page allocator won't try to merge buddies from
6426 * different pageblocks and change MIGRATE_ISOLATE to some
6427 * other migration type.
6428 *
6429 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6430 * migrate the pages from an unaligned range (ie. pages that
6431 * we are interested in). This will put all the pages in
6432 * range back to page allocator as MIGRATE_ISOLATE.
6433 *
6434 * When this is done, we take the pages in range from page
6435 * allocator removing them from the buddy system. This way
6436 * page allocator will never consider using them.
6437 *
6438 * This lets us mark the pageblocks back as
6439 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6440 * aligned range but not in the unaligned, original range are
6441 * put back to page allocator so that buddy can use them.
6442 */
6443
6444 ret = start_isolate_page_range(pfn_max_align_down(start),
Wen Congyangb023f462012-12-11 16:00:45 -08006445 pfn_max_align_up(end), migratetype,
6446 false);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006447 if (ret)
Bob Liu86a595f2012-10-25 13:37:56 -07006448 return ret;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006449
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006450 ret = __alloc_contig_migrate_range(&cc, start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006451 if (ret)
6452 goto done;
6453
6454 /*
6455 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6456 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
6457 * more, all pages in [start, end) are free in page allocator.
6458 * What we are going to do is to allocate all pages from
6459 * [start, end) (that is remove them from page allocator).
6460 *
6461 * The only problem is that pages at the beginning and at the
6462 * end of interesting range may be not aligned with pages that
6463 * page allocator holds, ie. they can be part of higher order
6464 * pages. Because of this, we reserve the bigger range and
6465 * once this is done free the pages we are not interested in.
6466 *
6467 * We don't have to hold zone->lock here because the pages are
6468 * isolated thus they won't get removed from buddy.
6469 */
6470
6471 lru_add_drain_all();
Vlastimil Babka510f5502014-12-10 15:43:07 -08006472 drain_all_pages(cc.zone);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006473
6474 order = 0;
6475 outer_start = start;
6476 while (!PageBuddy(pfn_to_page(outer_start))) {
6477 if (++order >= MAX_ORDER) {
6478 ret = -EBUSY;
6479 goto done;
6480 }
6481 outer_start &= ~0UL << order;
6482 }
6483
6484 /* Make sure the range is really isolated. */
Wen Congyangb023f462012-12-11 16:00:45 -08006485 if (test_pages_isolated(outer_start, end, false)) {
Michal Nazarewiczdae803e2014-11-13 15:19:27 -08006486 pr_info("%s: [%lx, %lx) PFNs busy\n",
6487 __func__, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006488 ret = -EBUSY;
6489 goto done;
6490 }
6491
Marek Szyprowski49f223a2012-01-25 12:49:24 +01006492 /* Grab isolated pages from freelists. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006493 outer_end = isolate_freepages_range(&cc, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006494 if (!outer_end) {
6495 ret = -EBUSY;
6496 goto done;
6497 }
6498
6499 /* Free head and tail (if any) */
6500 if (start != outer_start)
6501 free_contig_range(outer_start, start - outer_start);
6502 if (end != outer_end)
6503 free_contig_range(end, outer_end - end);
6504
6505done:
6506 undo_isolate_page_range(pfn_max_align_down(start),
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006507 pfn_max_align_up(end), migratetype);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006508 return ret;
6509}
6510
6511void free_contig_range(unsigned long pfn, unsigned nr_pages)
6512{
Marek Szyprowskibcc2b022012-12-20 15:05:18 -08006513 unsigned int count = 0;
6514
6515 for (; nr_pages--; pfn++) {
6516 struct page *page = pfn_to_page(pfn);
6517
6518 count += page_count(page) != 1;
6519 __free_page(page);
6520 }
6521 WARN(count != 0, "%d pages are still in use!\n", count);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006522}
6523#endif
6524
Jiang Liu4ed7e022012-07-31 16:43:35 -07006525#ifdef CONFIG_MEMORY_HOTPLUG
Cody P Schafer0a647f32013-07-03 15:01:33 -07006526/*
6527 * The zone indicated has a new number of managed_pages; batch sizes and percpu
6528 * page high values need to be recalulated.
6529 */
Jiang Liu4ed7e022012-07-31 16:43:35 -07006530void __meminit zone_pcp_update(struct zone *zone)
6531{
Cody P Schafer0a647f32013-07-03 15:01:33 -07006532 unsigned cpu;
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006533 mutex_lock(&pcp_batch_high_lock);
Cody P Schafer0a647f32013-07-03 15:01:33 -07006534 for_each_possible_cpu(cpu)
Cody P Schafer169f6c12013-07-03 15:01:41 -07006535 pageset_set_high_and_batch(zone,
6536 per_cpu_ptr(zone->pageset, cpu));
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006537 mutex_unlock(&pcp_batch_high_lock);
Jiang Liu4ed7e022012-07-31 16:43:35 -07006538}
6539#endif
6540
Jiang Liu340175b2012-07-31 16:43:32 -07006541void zone_pcp_reset(struct zone *zone)
6542{
6543 unsigned long flags;
Minchan Kim5a883812012-10-08 16:33:39 -07006544 int cpu;
6545 struct per_cpu_pageset *pset;
Jiang Liu340175b2012-07-31 16:43:32 -07006546
6547 /* avoid races with drain_pages() */
6548 local_irq_save(flags);
6549 if (zone->pageset != &boot_pageset) {
Minchan Kim5a883812012-10-08 16:33:39 -07006550 for_each_online_cpu(cpu) {
6551 pset = per_cpu_ptr(zone->pageset, cpu);
6552 drain_zonestat(zone, pset);
6553 }
Jiang Liu340175b2012-07-31 16:43:32 -07006554 free_percpu(zone->pageset);
6555 zone->pageset = &boot_pageset;
6556 }
6557 local_irq_restore(flags);
6558}
6559
Wen Congyang6dcd73d2012-12-11 16:01:01 -08006560#ifdef CONFIG_MEMORY_HOTREMOVE
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006561/*
6562 * All pages in the range must be isolated before calling this.
6563 */
6564void
6565__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
6566{
6567 struct page *page;
6568 struct zone *zone;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07006569 unsigned int order, i;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006570 unsigned long pfn;
6571 unsigned long flags;
6572 /* find the first valid pfn */
6573 for (pfn = start_pfn; pfn < end_pfn; pfn++)
6574 if (pfn_valid(pfn))
6575 break;
6576 if (pfn == end_pfn)
6577 return;
6578 zone = page_zone(pfn_to_page(pfn));
6579 spin_lock_irqsave(&zone->lock, flags);
6580 pfn = start_pfn;
6581 while (pfn < end_pfn) {
6582 if (!pfn_valid(pfn)) {
6583 pfn++;
6584 continue;
6585 }
6586 page = pfn_to_page(pfn);
Wen Congyangb023f462012-12-11 16:00:45 -08006587 /*
6588 * The HWPoisoned page may be not in buddy system, and
6589 * page_count() is not 0.
6590 */
6591 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
6592 pfn++;
6593 SetPageReserved(page);
6594 continue;
6595 }
6596
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006597 BUG_ON(page_count(page));
6598 BUG_ON(!PageBuddy(page));
6599 order = page_order(page);
6600#ifdef CONFIG_DEBUG_VM
6601 printk(KERN_INFO "remove from free list %lx %d %lx\n",
6602 pfn, 1 << order, end_pfn);
6603#endif
6604 list_del(&page->lru);
6605 rmv_page_order(page);
6606 zone->free_area[order].nr_free--;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006607 for (i = 0; i < (1 << order); i++)
6608 SetPageReserved((page+i));
6609 pfn += (1 << order);
6610 }
6611 spin_unlock_irqrestore(&zone->lock, flags);
6612}
6613#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01006614
6615#ifdef CONFIG_MEMORY_FAILURE
6616bool is_free_buddy_page(struct page *page)
6617{
6618 struct zone *zone = page_zone(page);
6619 unsigned long pfn = page_to_pfn(page);
6620 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07006621 unsigned int order;
Wu Fengguang8d22ba12009-12-16 12:19:58 +01006622
6623 spin_lock_irqsave(&zone->lock, flags);
6624 for (order = 0; order < MAX_ORDER; order++) {
6625 struct page *page_head = page - (pfn & ((1 << order) - 1));
6626
6627 if (PageBuddy(page_head) && page_order(page_head) >= order)
6628 break;
6629 }
6630 spin_unlock_irqrestore(&zone->lock, flags);
6631
6632 return order < MAX_ORDER;
6633}
6634#endif