blob: ebffa0e4a9c0451cfbd78bcf4835825ffd72c6de [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 Hiroyukia5d76b52007-10-16 01:26:11 -070051#include <linux/page-isolation.h>
Joonsoo Kimeefa8642014-12-12 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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212
Jan Beulich2c85f512009-09-21 17:03:07 -0700213static unsigned long __meminitdata nr_kernel_pages;
214static unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700215static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216
Tejun Heo0ee332c2011-12-08 10:22:09 -0800217#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
218static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
219static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
220static unsigned long __initdata required_kernelcore;
221static unsigned long __initdata required_movablecore;
222static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Mel Gormanc7132162006-09-27 01:49:43 -0700223
Tejun Heo0ee332c2011-12-08 10:22:09 -0800224/* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
225int movable_zone;
226EXPORT_SYMBOL(movable_zone);
227#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -0700228
Miklos Szeredi418508c2007-05-23 13:57:55 -0700229#if MAX_NUMNODES > 1
230int nr_node_ids __read_mostly = MAX_NUMNODES;
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700231int nr_online_nodes __read_mostly = 1;
Miklos Szeredi418508c2007-05-23 13:57:55 -0700232EXPORT_SYMBOL(nr_node_ids);
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700233EXPORT_SYMBOL(nr_online_nodes);
Miklos Szeredi418508c2007-05-23 13:57:55 -0700234#endif
235
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700236int page_group_by_mobility_disabled __read_mostly;
237
Minchan Kimee6f5092012-07-31 16:43:50 -0700238void set_pageblock_migratetype(struct page *page, int migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700239{
KOSAKI Motohiro5d0f3f72013-11-12 15:08:18 -0800240 if (unlikely(page_group_by_mobility_disabled &&
241 migratetype < MIGRATE_PCPTYPES))
Mel Gorman49255c62009-06-16 15:31:58 -0700242 migratetype = MIGRATE_UNMOVABLE;
243
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700244 set_pageblock_flags_group(page, (unsigned long)migratetype,
245 PB_migrate, PB_migrate_end);
246}
247
Nick Piggin13e74442006-01-06 00:10:58 -0800248#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700249static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700251 int ret = 0;
252 unsigned seq;
253 unsigned long pfn = page_to_pfn(page);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800254 unsigned long sp, start_pfn;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700255
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700256 do {
257 seq = zone_span_seqbegin(zone);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800258 start_pfn = zone->zone_start_pfn;
259 sp = zone->spanned_pages;
Cody P Schafer108bcc92013-02-22 16:35:23 -0800260 if (!zone_spans_pfn(zone, pfn))
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700261 ret = 1;
262 } while (zone_span_seqretry(zone, seq));
263
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800264 if (ret)
Dave Hansen613813e2014-06-04 16:07:27 -0700265 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
266 pfn, zone_to_nid(zone), zone->name,
267 start_pfn, start_pfn + sp);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800268
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700269 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700270}
271
272static int page_is_consistent(struct zone *zone, struct page *page)
273{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700274 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700275 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700277 return 0;
278
279 return 1;
280}
281/*
282 * Temporary debugging check for pages not lying within a given zone.
283 */
284static int bad_range(struct zone *zone, struct page *page)
285{
286 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700288 if (!page_is_consistent(zone, page))
289 return 1;
290
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 return 0;
292}
Nick Piggin13e74442006-01-06 00:10:58 -0800293#else
294static inline int bad_range(struct zone *zone, struct page *page)
295{
296 return 0;
297}
298#endif
299
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -0700300static void bad_page(struct page *page, const char *reason,
301 unsigned long bad_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800303 static unsigned long resume;
304 static unsigned long nr_shown;
305 static unsigned long nr_unshown;
306
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200307 /* Don't complain about poisoned pages */
308 if (PageHWPoison(page)) {
Mel Gorman22b751c2013-02-22 16:34:59 -0800309 page_mapcount_reset(page); /* remove PageBuddy */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200310 return;
311 }
312
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800313 /*
314 * Allow a burst of 60 reports, then keep quiet for that minute;
315 * or allow a steady drip of one report per second.
316 */
317 if (nr_shown == 60) {
318 if (time_before(jiffies, resume)) {
319 nr_unshown++;
320 goto out;
321 }
322 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800323 printk(KERN_ALERT
324 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800325 nr_unshown);
326 nr_unshown = 0;
327 }
328 nr_shown = 0;
329 }
330 if (nr_shown++ == 0)
331 resume = jiffies + 60 * HZ;
332
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800333 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800334 current->comm, page_to_pfn(page));
Dave Hansenf0b791a2014-01-23 15:52:49 -0800335 dump_page_badflags(page, reason, bad_flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700336
Dave Jones4f318882011-10-31 17:07:24 -0700337 print_modules();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800339out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800340 /* Leave bad fields for debug, except PageBuddy could make trouble */
Mel Gorman22b751c2013-02-22 16:34:59 -0800341 page_mapcount_reset(page); /* remove PageBuddy */
Rusty Russell373d4d02013-01-21 17:17:39 +1030342 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343}
344
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345/*
346 * Higher-order pages are called "compound pages". They are structured thusly:
347 *
348 * The first PAGE_SIZE page is called the "head page".
349 *
350 * The remaining PAGE_SIZE pages are called "tail pages".
351 *
Wang Sheng-Hui6416b9f2011-11-17 10:53:50 +0100352 * All pages have PG_compound set. All tail pages have their ->first_page
353 * pointing at the head page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800355 * The first tail page's ->lru.next holds the address of the compound page's
356 * put_page() function. Its ->lru.prev holds the order of allocation.
357 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800359
360static void free_compound_page(struct page *page)
361{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700362 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800363}
364
Andi Kleen01ad1c02008-07-23 21:27:46 -0700365void prep_compound_page(struct page *page, unsigned long order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366{
367 int i;
368 int nr_pages = 1 << order;
369
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800370 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700371 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700372 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800373 for (i = 1; i < nr_pages; i++) {
374 struct page *p = page + i;
Youquan Song58a84aa2011-12-08 14:34:18 -0800375 set_page_count(p, 0);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700376 p->first_page = page;
David Rientjes668f9abb2014-03-03 15:38:18 -0800377 /* Make sure p->first_page is always valid for PageTail() */
378 smp_wmb();
379 __SetPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 }
381}
382
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700383static inline void prep_zero_page(struct page *page, unsigned int order,
384 gfp_t gfp_flags)
Nick Piggin17cf4402006-03-22 00:08:41 -0800385{
386 int i;
387
Andrew Morton6626c5d2006-03-22 00:08:42 -0800388 /*
389 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
390 * and __GFP_HIGHMEM from hard or soft interrupt context.
391 */
Nick Piggin725d7042006-09-25 23:30:55 -0700392 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800393 for (i = 0; i < (1 << order); i++)
394 clear_highpage(page + i);
395}
396
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800397#ifdef CONFIG_DEBUG_PAGEALLOC
398unsigned int _debug_guardpage_minorder;
Joonsoo Kim031bc572014-12-12 16:55:52 -0800399bool _debug_pagealloc_enabled __read_mostly;
Joonsoo Kime30825f2014-12-12 16:55:49 -0800400bool _debug_guardpage_enabled __read_mostly;
401
Joonsoo Kim031bc572014-12-12 16:55:52 -0800402static int __init early_debug_pagealloc(char *buf)
403{
404 if (!buf)
405 return -EINVAL;
406
407 if (strcmp(buf, "on") == 0)
408 _debug_pagealloc_enabled = true;
409
410 return 0;
411}
412early_param("debug_pagealloc", early_debug_pagealloc);
413
Joonsoo Kime30825f2014-12-12 16:55:49 -0800414static bool need_debug_guardpage(void)
415{
Joonsoo Kim031bc572014-12-12 16:55:52 -0800416 /* If we don't use debug_pagealloc, we don't need guard page */
417 if (!debug_pagealloc_enabled())
418 return false;
419
Joonsoo Kime30825f2014-12-12 16:55:49 -0800420 return true;
421}
422
423static void init_debug_guardpage(void)
424{
Joonsoo Kim031bc572014-12-12 16:55:52 -0800425 if (!debug_pagealloc_enabled())
426 return;
427
Joonsoo Kime30825f2014-12-12 16:55:49 -0800428 _debug_guardpage_enabled = true;
429}
430
431struct page_ext_operations debug_guardpage_ops = {
432 .need = need_debug_guardpage,
433 .init = init_debug_guardpage,
434};
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800435
436static int __init debug_guardpage_minorder_setup(char *buf)
437{
438 unsigned long res;
439
440 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
441 printk(KERN_ERR "Bad debug_guardpage_minorder value\n");
442 return 0;
443 }
444 _debug_guardpage_minorder = res;
445 printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res);
446 return 0;
447}
448__setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
449
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800450static inline void set_page_guard(struct zone *zone, struct page *page,
451 unsigned int order, int migratetype)
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800452{
Joonsoo Kime30825f2014-12-12 16:55:49 -0800453 struct page_ext *page_ext;
454
455 if (!debug_guardpage_enabled())
456 return;
457
458 page_ext = lookup_page_ext(page);
459 __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
460
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800461 INIT_LIST_HEAD(&page->lru);
462 set_page_private(page, order);
463 /* Guard pages are not available for any usage */
464 __mod_zone_freepage_state(zone, -(1 << order), migratetype);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800465}
466
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800467static inline void clear_page_guard(struct zone *zone, struct page *page,
468 unsigned int order, int migratetype)
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800469{
Joonsoo Kime30825f2014-12-12 16:55:49 -0800470 struct page_ext *page_ext;
471
472 if (!debug_guardpage_enabled())
473 return;
474
475 page_ext = lookup_page_ext(page);
476 __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
477
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800478 set_page_private(page, 0);
479 if (!is_migrate_isolate(migratetype))
480 __mod_zone_freepage_state(zone, (1 << order), migratetype);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800481}
482#else
Joonsoo Kime30825f2014-12-12 16:55:49 -0800483struct page_ext_operations debug_guardpage_ops = { NULL, };
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800484static inline void set_page_guard(struct zone *zone, struct page *page,
485 unsigned int order, int migratetype) {}
486static inline void clear_page_guard(struct zone *zone, struct page *page,
487 unsigned int order, int migratetype) {}
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800488#endif
489
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700490static inline void set_page_order(struct page *page, unsigned int order)
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700491{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700492 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000493 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494}
495
496static inline void rmv_page_order(struct page *page)
497{
Nick Piggin676165a2006-04-10 11:21:48 +1000498 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700499 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500}
501
502/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 * This function checks whether a page is free && is the buddy
504 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800505 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000506 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700507 * (c) a page and its buddy have the same order &&
508 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 *
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700510 * For recording whether a page is in the buddy system, we set ->_mapcount
511 * PAGE_BUDDY_MAPCOUNT_VALUE.
512 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
513 * serialized by zone->lock.
Nick Piggin676165a2006-04-10 11:21:48 +1000514 *
515 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700517static inline int page_is_buddy(struct page *page, struct page *buddy,
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700518 unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700520 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800521 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800522
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800523 if (page_is_guard(buddy) && page_order(buddy) == order) {
Mel Gormand34c5fa2014-06-04 16:10:10 -0700524 if (page_zone_id(page) != page_zone_id(buddy))
525 return 0;
526
Weijie Yang4c5018c2015-02-10 14:11:39 -0800527 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
528
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800529 return 1;
530 }
531
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700532 if (PageBuddy(buddy) && page_order(buddy) == order) {
Mel Gormand34c5fa2014-06-04 16:10:10 -0700533 /*
534 * zone check is done late to avoid uselessly
535 * calculating zone/node ids for pages that could
536 * never merge.
537 */
538 if (page_zone_id(page) != page_zone_id(buddy))
539 return 0;
540
Weijie Yang4c5018c2015-02-10 14:11:39 -0800541 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
542
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700543 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000544 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700545 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546}
547
548/*
549 * Freeing function for a buddy system allocator.
550 *
551 * The concept of a buddy system is to maintain direct-mapped table
552 * (containing bit values) for memory blocks of various "orders".
553 * The bottom level table contains the map for the smallest allocatable
554 * units of memory (here, pages), and each level above it describes
555 * pairs of units from the levels below, hence, "buddies".
556 * At a high level, all that happens here is marking the table entry
557 * at the bottom level available, and propagating the changes upward
558 * as necessary, plus some accounting needed to play nicely with other
559 * parts of the VM system.
560 * At each level, we keep a list of pages, which are heads of continuous
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700561 * free pages of length of (1 << order) and marked with _mapcount
562 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
563 * field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 * So when we are allocating or freeing one, we can derive the state of the
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100565 * other. That is, if we allocate a small block, and both were
566 * free, the remainder of the region must be split into blocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 * If a block is freed, and its buddy is also free, then this
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100568 * triggers coalescing into a block of larger size.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +0100570 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 */
572
Nick Piggin48db57f2006-01-08 01:00:42 -0800573static inline void __free_one_page(struct page *page,
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700574 unsigned long pfn,
Mel Gormaned0ae212009-06-16 15:32:07 -0700575 struct zone *zone, unsigned int order,
576 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577{
578 unsigned long page_idx;
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700579 unsigned long combined_idx;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800580 unsigned long uninitialized_var(buddy_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700581 struct page *buddy;
Joonsoo Kim3c605092014-11-13 15:19:21 -0800582 int max_order = MAX_ORDER;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583
Cody P Schaferd29bb972013-02-22 16:35:25 -0800584 VM_BUG_ON(!zone_is_initialized(zone));
Kirill A. Shutemov6e9f0d52015-02-11 15:25:50 -0800585 VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586
Mel Gormaned0ae212009-06-16 15:32:07 -0700587 VM_BUG_ON(migratetype == -1);
Joonsoo Kim3c605092014-11-13 15:19:21 -0800588 if (is_migrate_isolate(migratetype)) {
589 /*
590 * We restrict max order of merging to prevent merge
591 * between freepages on isolate pageblock and normal
592 * pageblock. Without this, pageblock isolation
593 * could cause incorrect freepage accounting.
594 */
595 max_order = min(MAX_ORDER, pageblock_order + 1);
596 } else {
Joonsoo Kim8f82b552014-11-13 15:19:18 -0800597 __mod_zone_freepage_state(zone, 1 << order, migratetype);
Joonsoo Kim3c605092014-11-13 15:19:21 -0800598 }
Mel Gormaned0ae212009-06-16 15:32:07 -0700599
Joonsoo Kim3c605092014-11-13 15:19:21 -0800600 page_idx = pfn & ((1 << max_order) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601
Sasha Levin309381fea2014-01-23 15:52:54 -0800602 VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page);
603 VM_BUG_ON_PAGE(bad_range(zone, page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604
Joonsoo Kim3c605092014-11-13 15:19:21 -0800605 while (order < max_order - 1) {
KyongHo Cho43506fa2011-01-13 15:47:24 -0800606 buddy_idx = __find_buddy_index(page_idx, order);
607 buddy = page + (buddy_idx - page_idx);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700608 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700609 break;
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800610 /*
611 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
612 * merge with it and move up one order.
613 */
614 if (page_is_guard(buddy)) {
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800615 clear_page_guard(zone, buddy, order, migratetype);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800616 } else {
617 list_del(&buddy->lru);
618 zone->free_area[order].nr_free--;
619 rmv_page_order(buddy);
620 }
KyongHo Cho43506fa2011-01-13 15:47:24 -0800621 combined_idx = buddy_idx & page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 page = page + (combined_idx - page_idx);
623 page_idx = combined_idx;
624 order++;
625 }
626 set_page_order(page, order);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700627
628 /*
629 * If this is not the largest possible page, check if the buddy
630 * of the next-highest order is free. If it is, it's possible
631 * that pages are being freed that will coalesce soon. In case,
632 * that is happening, add the free page to the tail of the list
633 * so it's less likely to be used soon and more likely to be merged
634 * as a higher order page
635 */
Mel Gormanb7f50cf2010-10-26 14:21:11 -0700636 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700637 struct page *higher_page, *higher_buddy;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800638 combined_idx = buddy_idx & page_idx;
639 higher_page = page + (combined_idx - page_idx);
640 buddy_idx = __find_buddy_index(combined_idx, order + 1);
Li Haifeng0ba8f2d2012-09-17 14:09:21 -0700641 higher_buddy = higher_page + (buddy_idx - combined_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700642 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
643 list_add_tail(&page->lru,
644 &zone->free_area[order].free_list[migratetype]);
645 goto out;
646 }
647 }
648
649 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
650out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 zone->free_area[order].nr_free++;
652}
653
Nick Piggin224abf92006-01-06 00:11:11 -0800654static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655{
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -0700656 const char *bad_reason = NULL;
Dave Hansenf0b791a2014-01-23 15:52:49 -0800657 unsigned long bad_flags = 0;
658
659 if (unlikely(page_mapcount(page)))
660 bad_reason = "nonzero mapcount";
661 if (unlikely(page->mapping != NULL))
662 bad_reason = "non-NULL mapping";
663 if (unlikely(atomic_read(&page->_count) != 0))
664 bad_reason = "nonzero _count";
665 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
666 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
667 bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
668 }
Johannes Weiner9edad6e2014-12-10 15:44:58 -0800669#ifdef CONFIG_MEMCG
670 if (unlikely(page->mem_cgroup))
671 bad_reason = "page still charged to cgroup";
672#endif
Dave Hansenf0b791a2014-01-23 15:52:49 -0800673 if (unlikely(bad_reason)) {
674 bad_page(page, bad_reason, bad_flags);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800675 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800676 }
Peter Zijlstra90572892013-10-07 11:29:20 +0100677 page_cpupid_reset_last(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800678 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
679 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
680 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681}
682
683/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700684 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700686 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 *
688 * If the zone was previously in an "all pages pinned" state then look to
689 * see if this freeing clears that state.
690 *
691 * And clear the zone's pages_scanned counter, to hold off the "all pages are
692 * pinned" detection logic.
693 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700694static void free_pcppages_bulk(struct zone *zone, int count,
695 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696{
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700697 int migratetype = 0;
Mel Gormana6f9edd2009-09-21 17:03:20 -0700698 int batch_free = 0;
Mel Gorman72853e22010-09-09 16:38:16 -0700699 int to_free = count;
Mel Gorman0d5d8232014-08-06 16:07:16 -0700700 unsigned long nr_scanned;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700701
Nick Pigginc54ad302006-01-06 00:10:56 -0800702 spin_lock(&zone->lock);
Mel Gorman0d5d8232014-08-06 16:07:16 -0700703 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
704 if (nr_scanned)
705 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
Mel Gormanf2260e62009-06-16 15:32:13 -0700706
Mel Gorman72853e22010-09-09 16:38:16 -0700707 while (to_free) {
Nick Piggin48db57f2006-01-08 01:00:42 -0800708 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700709 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -0800710
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700711 /*
Mel Gormana6f9edd2009-09-21 17:03:20 -0700712 * Remove pages from lists in a round-robin fashion. A
713 * batch_free count is maintained that is incremented when an
714 * empty list is encountered. This is so more pages are freed
715 * off fuller lists instead of spinning excessively around empty
716 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700717 */
718 do {
Mel Gormana6f9edd2009-09-21 17:03:20 -0700719 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700720 if (++migratetype == MIGRATE_PCPTYPES)
721 migratetype = 0;
722 list = &pcp->lists[migratetype];
723 } while (list_empty(list));
724
Namhyung Kim1d168712011-03-22 16:32:45 -0700725 /* This is the only non-empty list. Free them all. */
726 if (batch_free == MIGRATE_PCPTYPES)
727 batch_free = to_free;
728
Mel Gormana6f9edd2009-09-21 17:03:20 -0700729 do {
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -0700730 int mt; /* migratetype of the to-be-freed page */
731
Mel Gormana6f9edd2009-09-21 17:03:20 -0700732 page = list_entry(list->prev, struct page, lru);
733 /* must delete as __free_one_page list manipulates */
734 list_del(&page->lru);
Minchan Kimb12c4ad2012-10-08 16:32:08 -0700735 mt = get_freepage_migratetype(page);
Joonsoo Kim8f82b552014-11-13 15:19:18 -0800736 if (unlikely(has_isolate_pageblock(zone)))
Joonsoo Kim51bb1a42014-11-13 15:19:14 -0800737 mt = get_pageblock_migratetype(page);
Joonsoo Kim51bb1a42014-11-13 15:19:14 -0800738
Hugh Dickinsa7016232010-01-29 17:46:34 +0000739 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700740 __free_one_page(page, page_to_pfn(page), zone, 0, mt);
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -0700741 trace_mm_page_pcpu_drain(page, 0, mt);
Mel Gorman72853e22010-09-09 16:38:16 -0700742 } while (--to_free && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800744 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745}
746
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700747static void free_one_page(struct zone *zone,
748 struct page *page, unsigned long pfn,
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700749 unsigned int order,
Mel Gormaned0ae212009-06-16 15:32:07 -0700750 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800751{
Mel Gorman0d5d8232014-08-06 16:07:16 -0700752 unsigned long nr_scanned;
Christoph Lameter006d22d2006-09-25 23:31:48 -0700753 spin_lock(&zone->lock);
Mel Gorman0d5d8232014-08-06 16:07:16 -0700754 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
755 if (nr_scanned)
756 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
Mel Gormanf2260e62009-06-16 15:32:13 -0700757
Joonsoo Kimad53f922014-11-13 15:19:11 -0800758 if (unlikely(has_isolate_pageblock(zone) ||
759 is_migrate_isolate(migratetype))) {
760 migratetype = get_pfnblock_migratetype(page, pfn);
Joonsoo Kimad53f922014-11-13 15:19:11 -0800761 }
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700762 __free_one_page(page, pfn, zone, order, migratetype);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700763 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800764}
765
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800766static int free_tail_pages_check(struct page *head_page, struct page *page)
767{
768 if (!IS_ENABLED(CONFIG_DEBUG_VM))
769 return 0;
770 if (unlikely(!PageTail(page))) {
771 bad_page(page, "PageTail not set", 0);
772 return 1;
773 }
774 if (unlikely(page->first_page != head_page)) {
775 bad_page(page, "first_page not consistent", 0);
776 return 1;
777 }
778 return 0;
779}
780
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700781static bool free_pages_prepare(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782{
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800783 bool compound = PageCompound(page);
784 int i, bad = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785
Yu Zhaoab1f3062014-12-10 15:43:17 -0800786 VM_BUG_ON_PAGE(PageTail(page), page);
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800787 VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
Yu Zhaoab1f3062014-12-10 15:43:17 -0800788
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -0800789 trace_mm_page_free(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100790 kmemcheck_free_shadow(page, order);
Andrey Ryabininb8c73fc2015-02-13 14:39:28 -0800791 kasan_free_pages(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100792
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -0800793 if (PageAnon(page))
794 page->mapping = NULL;
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800795 bad += free_pages_check(page);
796 for (i = 1; i < (1 << order); i++) {
797 if (compound)
798 bad += free_tail_pages_check(page, page + i);
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -0800799 bad += free_pages_check(page + i);
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800800 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800801 if (bad)
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700802 return false;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800803
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800804 reset_page_owner(page, order);
805
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700806 if (!PageHighMem(page)) {
Pintu Kumarb8af2942013-09-11 14:20:34 -0700807 debug_check_no_locks_freed(page_address(page),
808 PAGE_SIZE << order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700809 debug_check_no_obj_freed(page_address(page),
810 PAGE_SIZE << order);
811 }
Nick Piggindafb1362006-10-11 01:21:30 -0700812 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800813 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700814
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700815 return true;
816}
817
818static void __free_pages_ok(struct page *page, unsigned int order)
819{
820 unsigned long flags;
Minchan Kim95e34412012-10-08 16:32:11 -0700821 int migratetype;
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700822 unsigned long pfn = page_to_pfn(page);
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700823
824 if (!free_pages_prepare(page, order))
825 return;
826
Mel Gormancfc47a22014-06-04 16:10:19 -0700827 migratetype = get_pfnblock_migratetype(page, pfn);
Nick Pigginc54ad302006-01-06 00:10:56 -0800828 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700829 __count_vm_events(PGFREE, 1 << order);
Minchan Kim95e34412012-10-08 16:32:11 -0700830 set_freepage_migratetype(page, migratetype);
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700831 free_one_page(page_zone(page), page, pfn, order, migratetype);
Nick Pigginc54ad302006-01-06 00:10:56 -0800832 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833}
834
Jiang Liu170a5a72013-07-03 15:03:17 -0700835void __init __free_pages_bootmem(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800836{
Johannes Weinerc3993072012-01-10 15:08:10 -0800837 unsigned int nr_pages = 1 << order;
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700838 struct page *p = page;
Johannes Weinerc3993072012-01-10 15:08:10 -0800839 unsigned int loop;
David Howellsa226f6c2006-01-06 00:11:08 -0800840
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700841 prefetchw(p);
842 for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
843 prefetchw(p + 1);
Johannes Weinerc3993072012-01-10 15:08:10 -0800844 __ClearPageReserved(p);
845 set_page_count(p, 0);
David Howellsa226f6c2006-01-06 00:11:08 -0800846 }
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700847 __ClearPageReserved(p);
848 set_page_count(p, 0);
Johannes Weinerc3993072012-01-10 15:08:10 -0800849
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700850 page_zone(page)->managed_pages += nr_pages;
Johannes Weinerc3993072012-01-10 15:08:10 -0800851 set_page_refcounted(page);
852 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800853}
854
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100855#ifdef CONFIG_CMA
Li Zhong9cf510a2013-08-23 13:52:52 +0800856/* Free whole pageblock and set its migration type to MIGRATE_CMA. */
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100857void __init init_cma_reserved_pageblock(struct page *page)
858{
859 unsigned i = pageblock_nr_pages;
860 struct page *p = page;
861
862 do {
863 __ClearPageReserved(p);
864 set_page_count(p, 0);
865 } while (++p, --i);
866
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100867 set_pageblock_migratetype(page, MIGRATE_CMA);
Michal Nazarewiczdc783272014-07-02 15:22:35 -0700868
869 if (pageblock_order >= MAX_ORDER) {
870 i = pageblock_nr_pages;
871 p = page;
872 do {
873 set_page_refcounted(p);
874 __free_pages(p, MAX_ORDER - 1);
875 p += MAX_ORDER_NR_PAGES;
876 } while (i -= MAX_ORDER_NR_PAGES);
877 } else {
878 set_page_refcounted(page);
879 __free_pages(page, pageblock_order);
880 }
881
Jiang Liu3dcc0572013-07-03 15:03:21 -0700882 adjust_managed_page_count(page, pageblock_nr_pages);
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100883}
884#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885
886/*
887 * The order of subdivision here is critical for the IO subsystem.
888 * Please do not alter this order without good reasons and regression
889 * testing. Specifically, as large blocks of memory are subdivided,
890 * the order in which smaller blocks are delivered depends on the order
891 * they're subdivided in this function. This is the primary factor
892 * influencing the order in which pages are delivered to the IO
893 * subsystem according to empirical testing, and this is also justified
894 * by considering the behavior of a buddy system containing a single
895 * large block of memory acted on by a series of small allocations.
896 * This behavior is a critical factor in sglist merging's success.
897 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +0100898 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800900static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700901 int low, int high, struct free_area *area,
902 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903{
904 unsigned long size = 1 << high;
905
906 while (high > low) {
907 area--;
908 high--;
909 size >>= 1;
Sasha Levin309381fea2014-01-23 15:52:54 -0800910 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800911
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800912 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) &&
Joonsoo Kime30825f2014-12-12 16:55:49 -0800913 debug_guardpage_enabled() &&
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800914 high < debug_guardpage_minorder()) {
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800915 /*
916 * Mark as guard pages (or page), that will allow to
917 * merge back to allocator when buddy will be freed.
918 * Corresponding page table entries will not be touched,
919 * pages will stay not present in virtual address space
920 */
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800921 set_page_guard(zone, &page[size], high, migratetype);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800922 continue;
923 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700924 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 area->nr_free++;
926 set_page_order(&page[size], high);
927 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928}
929
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930/*
931 * This page is about to be returned from the page allocator
932 */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200933static inline int check_new_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934{
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -0700935 const char *bad_reason = NULL;
Dave Hansenf0b791a2014-01-23 15:52:49 -0800936 unsigned long bad_flags = 0;
937
938 if (unlikely(page_mapcount(page)))
939 bad_reason = "nonzero mapcount";
940 if (unlikely(page->mapping != NULL))
941 bad_reason = "non-NULL mapping";
942 if (unlikely(atomic_read(&page->_count) != 0))
943 bad_reason = "nonzero _count";
944 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
945 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
946 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
947 }
Johannes Weiner9edad6e2014-12-10 15:44:58 -0800948#ifdef CONFIG_MEMCG
949 if (unlikely(page->mem_cgroup))
950 bad_reason = "page still charged to cgroup";
951#endif
Dave Hansenf0b791a2014-01-23 15:52:49 -0800952 if (unlikely(bad_reason)) {
953 bad_page(page, bad_reason, bad_flags);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800954 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800955 }
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200956 return 0;
957}
958
Vlastimil Babka75379192015-02-11 15:25:38 -0800959static int prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
960 int alloc_flags)
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200961{
962 int i;
963
964 for (i = 0; i < (1 << order); i++) {
965 struct page *p = page + i;
966 if (unlikely(check_new_page(p)))
967 return 1;
968 }
Hugh Dickins689bceb2005-11-21 21:32:20 -0800969
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700970 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800971 set_page_refcounted(page);
Nick Piggincc1025092006-12-06 20:32:00 -0800972
973 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 kernel_map_pages(page, 1 << order, 1);
Andrey Ryabininb8c73fc2015-02-13 14:39:28 -0800975 kasan_alloc_pages(page, order);
Nick Piggin17cf4402006-03-22 00:08:41 -0800976
977 if (gfp_flags & __GFP_ZERO)
978 prep_zero_page(page, order, gfp_flags);
979
980 if (order && (gfp_flags & __GFP_COMP))
981 prep_compound_page(page, order);
982
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800983 set_page_owner(page, order, gfp_flags);
984
Vlastimil Babka75379192015-02-11 15:25:38 -0800985 /*
986 * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was necessary to
987 * allocate the page. The expectation is that the caller is taking
988 * steps that will free more memory. The caller should avoid the page
989 * being used for !PFMEMALLOC purposes.
990 */
991 page->pfmemalloc = !!(alloc_flags & ALLOC_NO_WATERMARKS);
992
Hugh Dickins689bceb2005-11-21 21:32:20 -0800993 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994}
995
Mel Gorman56fd56b2007-10-16 01:25:58 -0700996/*
997 * Go through the free lists for the given migratetype and remove
998 * the smallest available page from the freelists
999 */
Mel Gorman728ec982009-06-16 15:32:04 -07001000static inline
1001struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -07001002 int migratetype)
1003{
1004 unsigned int current_order;
Pintu Kumarb8af2942013-09-11 14:20:34 -07001005 struct free_area *area;
Mel Gorman56fd56b2007-10-16 01:25:58 -07001006 struct page *page;
1007
1008 /* Find a page of the appropriate size in the preferred list */
1009 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
1010 area = &(zone->free_area[current_order]);
1011 if (list_empty(&area->free_list[migratetype]))
1012 continue;
1013
1014 page = list_entry(area->free_list[migratetype].next,
1015 struct page, lru);
1016 list_del(&page->lru);
1017 rmv_page_order(page);
1018 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -07001019 expand(zone, page, order, current_order, area, migratetype);
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001020 set_freepage_migratetype(page, migratetype);
Mel Gorman56fd56b2007-10-16 01:25:58 -07001021 return page;
1022 }
1023
1024 return NULL;
1025}
1026
1027
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001028/*
1029 * This array describes the order lists are fallen back to when
1030 * the free lists for the desirable migrate type are depleted
1031 */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001032static int fallbacks[MIGRATE_TYPES][4] = {
1033 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
1034 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001035 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
Joonsoo Kimdc676472015-04-14 15:45:15 -07001036#ifdef CONFIG_CMA
1037 [MIGRATE_CMA] = { MIGRATE_RESERVE }, /* Never used */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001038#endif
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001039 [MIGRATE_RESERVE] = { MIGRATE_RESERVE }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -08001040#ifdef CONFIG_MEMORY_ISOLATION
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001041 [MIGRATE_ISOLATE] = { MIGRATE_RESERVE }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -08001042#endif
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001043};
1044
Joonsoo Kimdc676472015-04-14 15:45:15 -07001045#ifdef CONFIG_CMA
1046static struct page *__rmqueue_cma_fallback(struct zone *zone,
1047 unsigned int order)
1048{
1049 return __rmqueue_smallest(zone, order, MIGRATE_CMA);
1050}
1051#else
1052static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
1053 unsigned int order) { return NULL; }
1054#endif
1055
Mel Gormanc361be52007-10-16 01:25:51 -07001056/*
1057 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -07001058 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -07001059 * boundary. If alignment is required, use move_freepages_block()
1060 */
Minchan Kim435b4052012-10-08 16:32:16 -07001061int move_freepages(struct zone *zone,
Adrian Bunkb69a7282008-07-23 21:28:12 -07001062 struct page *start_page, struct page *end_page,
1063 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001064{
1065 struct page *page;
1066 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -07001067 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -07001068
1069#ifndef CONFIG_HOLES_IN_ZONE
1070 /*
1071 * page_zone is not safe to call in this context when
1072 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1073 * anyway as we check zone boundaries in move_freepages_block().
1074 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -07001075 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -07001076 */
Mel Gorman97ee4ba2014-10-09 15:28:28 -07001077 VM_BUG_ON(page_zone(start_page) != page_zone(end_page));
Mel Gormanc361be52007-10-16 01:25:51 -07001078#endif
1079
1080 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -07001081 /* Make sure we are not inadvertently changing nodes */
Sasha Levin309381fea2014-01-23 15:52:54 -08001082 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
Adam Litke344c7902008-09-02 14:35:38 -07001083
Mel Gormanc361be52007-10-16 01:25:51 -07001084 if (!pfn_valid_within(page_to_pfn(page))) {
1085 page++;
1086 continue;
1087 }
1088
1089 if (!PageBuddy(page)) {
1090 page++;
1091 continue;
1092 }
1093
1094 order = page_order(page);
Kirill A. Shutemov84be48d2011-03-22 16:33:41 -07001095 list_move(&page->lru,
1096 &zone->free_area[order].free_list[migratetype]);
Minchan Kim95e34412012-10-08 16:32:11 -07001097 set_freepage_migratetype(page, migratetype);
Mel Gormanc361be52007-10-16 01:25:51 -07001098 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -07001099 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -07001100 }
1101
Mel Gormand1003132007-10-16 01:26:00 -07001102 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -07001103}
1104
Minchan Kimee6f5092012-07-31 16:43:50 -07001105int move_freepages_block(struct zone *zone, struct page *page,
Linus Torvalds68e3e922012-06-03 20:05:57 -07001106 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001107{
1108 unsigned long start_pfn, end_pfn;
1109 struct page *start_page, *end_page;
1110
1111 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -07001112 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -07001113 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -07001114 end_page = start_page + pageblock_nr_pages - 1;
1115 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -07001116
1117 /* Do not cross zone boundaries */
Cody P Schafer108bcc92013-02-22 16:35:23 -08001118 if (!zone_spans_pfn(zone, start_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001119 start_page = page;
Cody P Schafer108bcc92013-02-22 16:35:23 -08001120 if (!zone_spans_pfn(zone, end_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001121 return 0;
1122
1123 return move_freepages(zone, start_page, end_page, migratetype);
1124}
1125
Mel Gorman2f66a682009-09-21 17:02:31 -07001126static void change_pageblock_range(struct page *pageblock_page,
1127 int start_order, int migratetype)
1128{
1129 int nr_pageblocks = 1 << (start_order - pageblock_order);
1130
1131 while (nr_pageblocks--) {
1132 set_pageblock_migratetype(pageblock_page, migratetype);
1133 pageblock_page += pageblock_nr_pages;
1134 }
1135}
1136
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001137/*
Vlastimil Babka9c0415e2015-02-11 15:28:21 -08001138 * When we are falling back to another migratetype during allocation, try to
1139 * steal extra free pages from the same pageblocks to satisfy further
1140 * allocations, instead of polluting multiple pageblocks.
1141 *
1142 * If we are stealing a relatively large buddy page, it is likely there will
1143 * be more free pages in the pageblock, so try to steal them all. For
1144 * reclaimable and unmovable allocations, we steal regardless of page size,
1145 * as fragmentation caused by those allocations polluting movable pageblocks
1146 * is worse than movable allocations stealing from unmovable and reclaimable
1147 * pageblocks.
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001148 */
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001149static bool can_steal_fallback(unsigned int order, int start_mt)
1150{
1151 /*
1152 * Leaving this order check is intended, although there is
1153 * relaxed order check in next check. The reason is that
1154 * we can actually steal whole pageblock if this condition met,
1155 * but, below check doesn't guarantee it and that is just heuristic
1156 * so could be changed anytime.
1157 */
1158 if (order >= pageblock_order)
1159 return true;
1160
1161 if (order >= pageblock_order / 2 ||
1162 start_mt == MIGRATE_RECLAIMABLE ||
1163 start_mt == MIGRATE_UNMOVABLE ||
1164 page_group_by_mobility_disabled)
1165 return true;
1166
1167 return false;
1168}
1169
1170/*
1171 * This function implements actual steal behaviour. If order is large enough,
1172 * we can steal whole pageblock. If not, we first move freepages in this
1173 * pageblock and check whether half of pages are moved or not. If half of
1174 * pages are moved, we can change migratetype of pageblock and permanently
1175 * use it's pages as requested migratetype in the future.
1176 */
1177static void steal_suitable_fallback(struct zone *zone, struct page *page,
1178 int start_type)
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001179{
1180 int current_order = page_order(page);
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001181 int pages;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001182
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001183 /* Take ownership for orders >= pageblock_order */
1184 if (current_order >= pageblock_order) {
1185 change_pageblock_range(page, current_order, start_type);
Vlastimil Babka3a1086f2015-02-11 15:28:18 -08001186 return;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001187 }
1188
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001189 pages = move_freepages_block(zone, page, start_type);
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001190
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001191 /* Claim the whole block if over half of it is free */
1192 if (pages >= (1 << (pageblock_order-1)) ||
1193 page_group_by_mobility_disabled)
1194 set_pageblock_migratetype(page, start_type);
1195}
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001196
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001197/*
1198 * Check whether there is a suitable fallback freepage with requested order.
1199 * If only_stealable is true, this function returns fallback_mt only if
1200 * we can steal other freepages all together. This would help to reduce
1201 * fragmentation due to mixed migratetype pages in one pageblock.
1202 */
1203int find_suitable_fallback(struct free_area *area, unsigned int order,
1204 int migratetype, bool only_stealable, bool *can_steal)
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001205{
1206 int i;
1207 int fallback_mt;
1208
1209 if (area->nr_free == 0)
1210 return -1;
1211
1212 *can_steal = false;
1213 for (i = 0;; i++) {
1214 fallback_mt = fallbacks[migratetype][i];
1215 if (fallback_mt == MIGRATE_RESERVE)
1216 break;
1217
1218 if (list_empty(&area->free_list[fallback_mt]))
1219 continue;
1220
1221 if (can_steal_fallback(order, migratetype))
1222 *can_steal = true;
1223
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001224 if (!only_stealable)
1225 return fallback_mt;
1226
1227 if (*can_steal)
1228 return fallback_mt;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001229 }
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001230
1231 return -1;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001232}
1233
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001234/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -07001235static inline struct page *
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001236__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001237{
Pintu Kumarb8af2942013-09-11 14:20:34 -07001238 struct free_area *area;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001239 unsigned int current_order;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001240 struct page *page;
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001241 int fallback_mt;
1242 bool can_steal;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001243
1244 /* Find the largest possible block of pages in the other list */
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001245 for (current_order = MAX_ORDER-1;
1246 current_order >= order && current_order <= MAX_ORDER-1;
1247 --current_order) {
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001248 area = &(zone->free_area[current_order]);
1249 fallback_mt = find_suitable_fallback(area, current_order,
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001250 start_migratetype, false, &can_steal);
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001251 if (fallback_mt == -1)
1252 continue;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001253
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001254 page = list_entry(area->free_list[fallback_mt].next,
1255 struct page, lru);
1256 if (can_steal)
1257 steal_suitable_fallback(zone, page, start_migratetype);
Mel Gormane0104872007-10-16 01:25:53 -07001258
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001259 /* Remove the page from the freelists */
1260 area->nr_free--;
1261 list_del(&page->lru);
1262 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001263
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001264 expand(zone, page, order, current_order, area,
1265 start_migratetype);
1266 /*
1267 * The freepage_migratetype may differ from pageblock's
1268 * migratetype depending on the decisions in
1269 * try_to_steal_freepages(). This is OK as long as it
1270 * does not differ for MIGRATE_CMA pageblocks. For CMA
1271 * we need to make sure unallocated pages flushed from
1272 * pcp lists are returned to the correct freelist.
1273 */
1274 set_freepage_migratetype(page, start_migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001275
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001276 trace_mm_page_alloc_extfrag(page, order, current_order,
1277 start_migratetype, fallback_mt);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001278
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001279 return page;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001280 }
1281
Mel Gorman728ec982009-06-16 15:32:04 -07001282 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001283}
1284
Mel Gorman56fd56b2007-10-16 01:25:58 -07001285/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 * Do the hard work of removing an element from the buddy allocator.
1287 * Call me with the zone->lock already held.
1288 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001289static struct page *__rmqueue(struct zone *zone, unsigned int order,
1290 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 struct page *page;
1293
Mel Gorman728ec982009-06-16 15:32:04 -07001294retry_reserve:
Mel Gorman56fd56b2007-10-16 01:25:58 -07001295 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296
Mel Gorman728ec982009-06-16 15:32:04 -07001297 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Joonsoo Kimdc676472015-04-14 15:45:15 -07001298 if (migratetype == MIGRATE_MOVABLE)
1299 page = __rmqueue_cma_fallback(zone, order);
1300
1301 if (!page)
1302 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001303
Mel Gorman728ec982009-06-16 15:32:04 -07001304 /*
1305 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1306 * is used because __rmqueue_smallest is an inline function
1307 * and we want just one call site
1308 */
1309 if (!page) {
1310 migratetype = MIGRATE_RESERVE;
1311 goto retry_reserve;
1312 }
1313 }
1314
Mel Gorman0d3d0622009-09-21 17:02:44 -07001315 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001316 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317}
1318
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001319/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 * Obtain a specified number of elements from the buddy allocator, all under
1321 * a single hold of the lock, for efficiency. Add them to the supplied list.
1322 * Returns the number of new pages which were placed at *list.
1323 */
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001324static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001325 unsigned long count, struct list_head *list,
Mel Gormanb745bc82014-06-04 16:10:22 -07001326 int migratetype, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327{
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001328 int i;
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001329
Nick Pigginc54ad302006-01-06 00:10:56 -08001330 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001332 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001333 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08001335
1336 /*
1337 * Split buddy pages returned by expand() are received here
1338 * in physical page order. The page is added to the callers and
1339 * list and the list head then moves forward. From the callers
1340 * perspective, the linked list is ordered by page number in
1341 * some conditions. This is useful for IO devices that can
1342 * merge IO requests if the physical pages are ordered
1343 * properly.
1344 */
Mel Gormanb745bc82014-06-04 16:10:22 -07001345 if (likely(!cold))
Mel Gormane084b2d2009-07-29 15:02:04 -07001346 list_add(&page->lru, list);
1347 else
1348 list_add_tail(&page->lru, list);
Mel Gorman81eabcb2007-12-17 16:20:05 -08001349 list = &page->lru;
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001350 if (is_migrate_cma(get_freepage_migratetype(page)))
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07001351 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
1352 -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 }
Mel Gormanf2260e62009-06-16 15:32:13 -07001354 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08001355 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001356 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357}
1358
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001359#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001360/*
Christoph Lameter4037d452007-05-09 02:35:14 -07001361 * Called from the vmstat counter updater to drain pagesets of this
1362 * currently executing processor on remote nodes after they have
1363 * expired.
1364 *
Christoph Lameter879336c2006-03-22 00:09:08 -08001365 * Note that this function must be called with the thread pinned to
1366 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001367 */
Christoph Lameter4037d452007-05-09 02:35:14 -07001368void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001369{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001370 unsigned long flags;
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07001371 int to_drain, batch;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001372
Christoph Lameter4037d452007-05-09 02:35:14 -07001373 local_irq_save(flags);
Jason Low4db0c3c2015-04-15 16:14:08 -07001374 batch = READ_ONCE(pcp->batch);
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07001375 to_drain = min(pcp->count, batch);
KOSAKI Motohiro2a135152012-07-31 16:42:53 -07001376 if (to_drain > 0) {
1377 free_pcppages_bulk(zone, to_drain, pcp);
1378 pcp->count -= to_drain;
1379 }
Christoph Lameter4037d452007-05-09 02:35:14 -07001380 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001381}
1382#endif
1383
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001384/*
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001385 * Drain pcplists of the indicated processor and zone.
1386 *
1387 * The processor must either be the current processor and the
1388 * thread pinned to the current processor or a processor that
1389 * is not online.
1390 */
1391static void drain_pages_zone(unsigned int cpu, struct zone *zone)
1392{
1393 unsigned long flags;
1394 struct per_cpu_pageset *pset;
1395 struct per_cpu_pages *pcp;
1396
1397 local_irq_save(flags);
1398 pset = per_cpu_ptr(zone->pageset, cpu);
1399
1400 pcp = &pset->pcp;
1401 if (pcp->count) {
1402 free_pcppages_bulk(zone, pcp->count, pcp);
1403 pcp->count = 0;
1404 }
1405 local_irq_restore(flags);
1406}
1407
1408/*
1409 * Drain pcplists of all zones on the indicated processor.
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001410 *
1411 * The processor must either be the current processor and the
1412 * thread pinned to the current processor or a processor that
1413 * is not online.
1414 */
1415static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416{
1417 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07001419 for_each_populated_zone(zone) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001420 drain_pages_zone(cpu, zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 }
1422}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001424/*
1425 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001426 *
1427 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
1428 * the single zone's pages.
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001429 */
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001430void drain_local_pages(struct zone *zone)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001431{
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001432 int cpu = smp_processor_id();
1433
1434 if (zone)
1435 drain_pages_zone(cpu, zone);
1436 else
1437 drain_pages(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001438}
1439
1440/*
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001441 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1442 *
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001443 * When zone parameter is non-NULL, spill just the single zone's pages.
1444 *
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001445 * Note that this code is protected against sending an IPI to an offline
1446 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1447 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1448 * nothing keeps CPUs from showing up after we populated the cpumask and
1449 * before the call to on_each_cpu_mask().
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001450 */
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001451void drain_all_pages(struct zone *zone)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001452{
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001453 int cpu;
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001454
1455 /*
1456 * Allocate in the BSS so we wont require allocation in
1457 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1458 */
1459 static cpumask_t cpus_with_pcps;
1460
1461 /*
1462 * We don't care about racing with CPU hotplug event
1463 * as offline notification will cause the notified
1464 * cpu to drain that CPU pcps and on_each_cpu_mask
1465 * disables preemption as part of its processing
1466 */
1467 for_each_online_cpu(cpu) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001468 struct per_cpu_pageset *pcp;
1469 struct zone *z;
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001470 bool has_pcps = false;
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001471
1472 if (zone) {
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001473 pcp = per_cpu_ptr(zone->pageset, cpu);
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001474 if (pcp->pcp.count)
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001475 has_pcps = true;
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001476 } else {
1477 for_each_populated_zone(z) {
1478 pcp = per_cpu_ptr(z->pageset, cpu);
1479 if (pcp->pcp.count) {
1480 has_pcps = true;
1481 break;
1482 }
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001483 }
1484 }
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001485
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001486 if (has_pcps)
1487 cpumask_set_cpu(cpu, &cpus_with_pcps);
1488 else
1489 cpumask_clear_cpu(cpu, &cpus_with_pcps);
1490 }
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001491 on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages,
1492 zone, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001493}
1494
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02001495#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496
1497void mark_free_pages(struct zone *zone)
1498{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001499 unsigned long pfn, max_zone_pfn;
1500 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001501 unsigned int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 struct list_head *curr;
1503
Xishi Qiu8080fc02013-09-11 14:21:45 -07001504 if (zone_is_empty(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 return;
1506
1507 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001508
Cody P Schafer108bcc92013-02-22 16:35:23 -08001509 max_zone_pfn = zone_end_pfn(zone);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001510 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1511 if (pfn_valid(pfn)) {
1512 struct page *page = pfn_to_page(pfn);
1513
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001514 if (!swsusp_page_is_forbidden(page))
1515 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001516 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001518 for_each_migratetype_order(order, t) {
1519 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001520 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001522 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1523 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001524 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001525 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001526 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 spin_unlock_irqrestore(&zone->lock, flags);
1528}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001529#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530
1531/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 * Free a 0-order page
Mel Gormanb745bc82014-06-04 16:10:22 -07001533 * cold == true ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 */
Mel Gormanb745bc82014-06-04 16:10:22 -07001535void free_hot_cold_page(struct page *page, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536{
1537 struct zone *zone = page_zone(page);
1538 struct per_cpu_pages *pcp;
1539 unsigned long flags;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001540 unsigned long pfn = page_to_pfn(page);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001541 int migratetype;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001543 if (!free_pages_prepare(page, 0))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001544 return;
1545
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001546 migratetype = get_pfnblock_migratetype(page, pfn);
Minchan Kimb12c4ad2012-10-08 16:32:08 -07001547 set_freepage_migratetype(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001549 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001550
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001551 /*
1552 * We only track unmovable, reclaimable and movable on pcp lists.
1553 * Free ISOLATE pages back to the allocator because they are being
1554 * offlined but treat RESERVE as movable pages so we can get those
1555 * areas back if necessary. Otherwise, we may have to free
1556 * excessively into the page allocator
1557 */
1558 if (migratetype >= MIGRATE_PCPTYPES) {
Minchan Kim194159f2013-02-22 16:33:58 -08001559 if (unlikely(is_migrate_isolate(migratetype))) {
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001560 free_one_page(zone, page, pfn, 0, migratetype);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001561 goto out;
1562 }
1563 migratetype = MIGRATE_MOVABLE;
1564 }
1565
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001566 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Mel Gormanb745bc82014-06-04 16:10:22 -07001567 if (!cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001568 list_add(&page->lru, &pcp->lists[migratetype]);
Mel Gormanb745bc82014-06-04 16:10:22 -07001569 else
1570 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001572 if (pcp->count >= pcp->high) {
Jason Low4db0c3c2015-04-15 16:14:08 -07001573 unsigned long batch = READ_ONCE(pcp->batch);
Cody P Schafer998d39c2013-07-03 15:01:32 -07001574 free_pcppages_bulk(zone, batch, pcp);
1575 pcp->count -= batch;
Nick Piggin48db57f2006-01-08 01:00:42 -08001576 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001577
1578out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580}
1581
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001582/*
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001583 * Free a list of 0-order pages
1584 */
Mel Gormanb745bc82014-06-04 16:10:22 -07001585void free_hot_cold_page_list(struct list_head *list, bool cold)
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001586{
1587 struct page *page, *next;
1588
1589 list_for_each_entry_safe(page, next, list, lru) {
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -08001590 trace_mm_page_free_batched(page, cold);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001591 free_hot_cold_page(page, cold);
1592 }
1593}
1594
1595/*
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001596 * split_page takes a non-compound higher-order page, and splits it into
1597 * n (1<<order) sub-pages: page[0..n]
1598 * Each sub-page must be freed individually.
1599 *
1600 * Note: this is probably too low level an operation for use in drivers.
1601 * Please consult with lkml before using this in your driver.
1602 */
1603void split_page(struct page *page, unsigned int order)
1604{
1605 int i;
1606
Sasha Levin309381fea2014-01-23 15:52:54 -08001607 VM_BUG_ON_PAGE(PageCompound(page), page);
1608 VM_BUG_ON_PAGE(!page_count(page), page);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001609
1610#ifdef CONFIG_KMEMCHECK
1611 /*
1612 * Split shadow pages too, because free(page[0]) would
1613 * otherwise free the whole shadow.
1614 */
1615 if (kmemcheck_page_is_tracked(page))
1616 split_page(virt_to_page(page[0].shadow), order);
1617#endif
1618
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001619 set_page_owner(page, 0, 0);
1620 for (i = 1; i < (1 << order); i++) {
Nick Piggin7835e982006-03-22 00:08:40 -08001621 set_page_refcounted(page + i);
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001622 set_page_owner(page + i, 0, 0);
1623 }
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001624}
K. Y. Srinivasan5853ff22013-03-25 15:47:38 -07001625EXPORT_SYMBOL_GPL(split_page);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001626
Joonsoo Kim3c605092014-11-13 15:19:21 -08001627int __isolate_free_page(struct page *page, unsigned int order)
Mel Gorman748446b2010-05-24 14:32:27 -07001628{
Mel Gorman748446b2010-05-24 14:32:27 -07001629 unsigned long watermark;
1630 struct zone *zone;
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07001631 int mt;
Mel Gorman748446b2010-05-24 14:32:27 -07001632
1633 BUG_ON(!PageBuddy(page));
1634
1635 zone = page_zone(page);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001636 mt = get_pageblock_migratetype(page);
Mel Gorman748446b2010-05-24 14:32:27 -07001637
Minchan Kim194159f2013-02-22 16:33:58 -08001638 if (!is_migrate_isolate(mt)) {
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001639 /* Obey watermarks as if the page was being allocated */
1640 watermark = low_wmark_pages(zone) + (1 << order);
1641 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1642 return 0;
1643
Mel Gorman8fb74b92013-01-11 14:32:16 -08001644 __mod_zone_freepage_state(zone, -(1UL << order), mt);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001645 }
Mel Gorman748446b2010-05-24 14:32:27 -07001646
1647 /* Remove page from free list */
1648 list_del(&page->lru);
1649 zone->free_area[order].nr_free--;
1650 rmv_page_order(page);
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07001651
Mel Gorman8fb74b92013-01-11 14:32:16 -08001652 /* Set the pageblock if the isolated page is at least a pageblock */
Mel Gorman748446b2010-05-24 14:32:27 -07001653 if (order >= pageblock_order - 1) {
1654 struct page *endpage = page + (1 << order) - 1;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001655 for (; page < endpage; page += pageblock_nr_pages) {
1656 int mt = get_pageblock_migratetype(page);
Minchan Kim194159f2013-02-22 16:33:58 -08001657 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001658 set_pageblock_migratetype(page,
1659 MIGRATE_MOVABLE);
1660 }
Mel Gorman748446b2010-05-24 14:32:27 -07001661 }
1662
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001663 set_page_owner(page, order, 0);
Mel Gorman8fb74b92013-01-11 14:32:16 -08001664 return 1UL << order;
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001665}
1666
1667/*
1668 * Similar to split_page except the page is already free. As this is only
1669 * being used for migration, the migratetype of the block also changes.
1670 * As this is called with interrupts disabled, the caller is responsible
1671 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1672 * are enabled.
1673 *
1674 * Note: this is probably too low level an operation for use in drivers.
1675 * Please consult with lkml before using this in your driver.
1676 */
1677int split_free_page(struct page *page)
1678{
1679 unsigned int order;
1680 int nr_pages;
1681
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001682 order = page_order(page);
1683
Mel Gorman8fb74b92013-01-11 14:32:16 -08001684 nr_pages = __isolate_free_page(page, order);
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001685 if (!nr_pages)
1686 return 0;
1687
1688 /* Split into individual pages */
1689 set_page_refcounted(page);
1690 split_page(page, order);
1691 return nr_pages;
Mel Gorman748446b2010-05-24 14:32:27 -07001692}
1693
1694/*
Vlastimil Babka75379192015-02-11 15:25:38 -08001695 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07001697static inline
1698struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001699 struct zone *zone, unsigned int order,
1700 gfp_t gfp_flags, int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701{
1702 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001703 struct page *page;
Mel Gormanb745bc82014-06-04 16:10:22 -07001704 bool cold = ((gfp_flags & __GFP_COLD) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705
Nick Piggin48db57f2006-01-08 01:00:42 -08001706 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001708 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 local_irq_save(flags);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001711 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1712 list = &pcp->lists[migratetype];
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001713 if (list_empty(list)) {
Mel Gorman535131e62007-10-16 01:25:49 -07001714 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001715 pcp->batch, list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001716 migratetype, cold);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001717 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07001718 goto failed;
Mel Gorman535131e62007-10-16 01:25:49 -07001719 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001720
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001721 if (cold)
1722 page = list_entry(list->prev, struct page, lru);
1723 else
1724 page = list_entry(list->next, struct page, lru);
1725
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001726 list_del(&page->lru);
1727 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001728 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07001729 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1730 /*
1731 * __GFP_NOFAIL is not to be used in new code.
1732 *
1733 * All __GFP_NOFAIL callers should be fixed so that they
1734 * properly detect and handle allocation failures.
1735 *
1736 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07001737 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07001738 * __GFP_NOFAIL.
1739 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07001740 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07001741 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001743 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001744 spin_unlock(&zone->lock);
1745 if (!page)
1746 goto failed;
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07001747 __mod_zone_freepage_state(zone, -(1 << order),
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001748 get_freepage_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 }
1750
Johannes Weiner3a025762014-04-07 15:37:48 -07001751 __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
Johannes Weinerabe5f972014-10-02 16:21:10 -07001752 if (atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]) <= 0 &&
Johannes Weiner57054652014-10-09 15:28:17 -07001753 !test_bit(ZONE_FAIR_DEPLETED, &zone->flags))
1754 set_bit(ZONE_FAIR_DEPLETED, &zone->flags);
Johannes Weiner27329362014-03-03 15:38:41 -08001755
Christoph Lameterf8891e52006-06-30 01:55:45 -07001756 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Andi Kleen78afd562011-03-22 16:33:12 -07001757 zone_statistics(preferred_zone, zone, gfp_flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001758 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759
Sasha Levin309381fea2014-01-23 15:52:54 -08001760 VM_BUG_ON_PAGE(bad_range(zone, page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001762
1763failed:
1764 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001765 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766}
1767
Akinobu Mita933e3122006-12-08 02:39:45 -08001768#ifdef CONFIG_FAIL_PAGE_ALLOC
1769
Akinobu Mitab2588c42011-07-26 16:09:03 -07001770static struct {
Akinobu Mita933e3122006-12-08 02:39:45 -08001771 struct fault_attr attr;
1772
1773 u32 ignore_gfp_highmem;
1774 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001775 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001776} fail_page_alloc = {
1777 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001778 .ignore_gfp_wait = 1,
1779 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001780 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001781};
1782
1783static int __init setup_fail_page_alloc(char *str)
1784{
1785 return setup_fault_attr(&fail_page_alloc.attr, str);
1786}
1787__setup("fail_page_alloc=", setup_fail_page_alloc);
1788
Gavin Shandeaf3862012-07-31 16:41:51 -07001789static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08001790{
Akinobu Mita54114992007-07-15 23:40:23 -07001791 if (order < fail_page_alloc.min_order)
Gavin Shandeaf3862012-07-31 16:41:51 -07001792 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001793 if (gfp_mask & __GFP_NOFAIL)
Gavin Shandeaf3862012-07-31 16:41:51 -07001794 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001795 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
Gavin Shandeaf3862012-07-31 16:41:51 -07001796 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001797 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
Gavin Shandeaf3862012-07-31 16:41:51 -07001798 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001799
1800 return should_fail(&fail_page_alloc.attr, 1 << order);
1801}
1802
1803#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1804
1805static int __init fail_page_alloc_debugfs(void)
1806{
Al Virof4ae40a2011-07-24 04:33:43 -04001807 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
Akinobu Mita933e3122006-12-08 02:39:45 -08001808 struct dentry *dir;
Akinobu Mita933e3122006-12-08 02:39:45 -08001809
Akinobu Mitadd48c082011-08-03 16:21:01 -07001810 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
1811 &fail_page_alloc.attr);
1812 if (IS_ERR(dir))
1813 return PTR_ERR(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08001814
Akinobu Mitab2588c42011-07-26 16:09:03 -07001815 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
1816 &fail_page_alloc.ignore_gfp_wait))
1817 goto fail;
1818 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1819 &fail_page_alloc.ignore_gfp_highmem))
1820 goto fail;
1821 if (!debugfs_create_u32("min-order", mode, dir,
1822 &fail_page_alloc.min_order))
1823 goto fail;
Akinobu Mita933e3122006-12-08 02:39:45 -08001824
Akinobu Mitab2588c42011-07-26 16:09:03 -07001825 return 0;
1826fail:
Akinobu Mitadd48c082011-08-03 16:21:01 -07001827 debugfs_remove_recursive(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08001828
Akinobu Mitab2588c42011-07-26 16:09:03 -07001829 return -ENOMEM;
Akinobu Mita933e3122006-12-08 02:39:45 -08001830}
1831
1832late_initcall(fail_page_alloc_debugfs);
1833
1834#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1835
1836#else /* CONFIG_FAIL_PAGE_ALLOC */
1837
Gavin Shandeaf3862012-07-31 16:41:51 -07001838static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08001839{
Gavin Shandeaf3862012-07-31 16:41:51 -07001840 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001841}
1842
1843#endif /* CONFIG_FAIL_PAGE_ALLOC */
1844
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845/*
Mel Gorman88f5acf2011-01-13 15:45:41 -08001846 * Return true if free pages are above 'mark'. This takes into account the order
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 * of the allocation.
1848 */
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001849static bool __zone_watermark_ok(struct zone *z, unsigned int order,
1850 unsigned long mark, int classzone_idx, int alloc_flags,
1851 long free_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852{
Wei Yuan26086de2014-12-10 15:44:44 -08001853 /* free_pages may go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001854 long min = mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 int o;
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07001856 long free_cma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857
Michal Hockodf0a6da2012-01-10 15:08:02 -08001858 free_pages -= (1 << order) - 1;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001859 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001861 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 min -= min / 4;
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07001863#ifdef CONFIG_CMA
1864 /* If allocation can't use CMA areas don't use free CMA pages */
1865 if (!(alloc_flags & ALLOC_CMA))
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07001866 free_cma = zone_page_state(z, NR_FREE_CMA_PAGES);
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07001867#endif
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07001868
Mel Gorman3484b2d2014-08-06 16:07:14 -07001869 if (free_pages - free_cma <= min + z->lowmem_reserve[classzone_idx])
Mel Gorman88f5acf2011-01-13 15:45:41 -08001870 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 for (o = 0; o < order; o++) {
1872 /* At the next order, this order's pages become unavailable */
1873 free_pages -= z->free_area[o].nr_free << o;
1874
1875 /* Require fewer higher order pages to be free */
1876 min >>= 1;
1877
1878 if (free_pages <= min)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001879 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 }
Mel Gorman88f5acf2011-01-13 15:45:41 -08001881 return true;
1882}
1883
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001884bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
Mel Gorman88f5acf2011-01-13 15:45:41 -08001885 int classzone_idx, int alloc_flags)
1886{
1887 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1888 zone_page_state(z, NR_FREE_PAGES));
1889}
1890
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001891bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
1892 unsigned long mark, int classzone_idx, int alloc_flags)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001893{
1894 long free_pages = zone_page_state(z, NR_FREE_PAGES);
1895
1896 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1897 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1898
1899 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1900 free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901}
1902
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001903#ifdef CONFIG_NUMA
1904/*
1905 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1906 * skip over zones that are not allowed by the cpuset, or that have
1907 * been recently (in last second) found to be nearly full. See further
1908 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001909 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001910 *
Zhi Yong Wua1aeb652013-11-12 15:08:29 -08001911 * If the zonelist cache is present in the passed zonelist, then
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001912 * returns a pointer to the allowed node mask (either the current
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08001913 * tasks mems_allowed, or node_states[N_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001914 *
1915 * If the zonelist cache is not available for this zonelist, does
1916 * nothing and returns NULL.
1917 *
1918 * If the fullzones BITMAP in the zonelist cache is stale (more than
1919 * a second since last zap'd) then we zap it out (clear its bits.)
1920 *
1921 * We hold off even calling zlc_setup, until after we've checked the
1922 * first zone in the zonelist, on the theory that most allocations will
1923 * be satisfied from that first zone, so best to examine that zone as
1924 * quickly as we can.
1925 */
1926static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1927{
1928 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1929 nodemask_t *allowednodes; /* zonelist_cache approximation */
1930
1931 zlc = zonelist->zlcache_ptr;
1932 if (!zlc)
1933 return NULL;
1934
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001935 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001936 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1937 zlc->last_full_zap = jiffies;
1938 }
1939
1940 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1941 &cpuset_current_mems_allowed :
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08001942 &node_states[N_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001943 return allowednodes;
1944}
1945
1946/*
1947 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1948 * if it is worth looking at further for free memory:
1949 * 1) Check that the zone isn't thought to be full (doesn't have its
1950 * bit set in the zonelist_cache fullzones BITMAP).
1951 * 2) Check that the zones node (obtained from the zonelist_cache
1952 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1953 * Return true (non-zero) if zone is worth looking at further, or
1954 * else return false (zero) if it is not.
1955 *
1956 * This check -ignores- the distinction between various watermarks,
1957 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1958 * found to be full for any variation of these watermarks, it will
1959 * be considered full for up to one second by all requests, unless
1960 * we are so low on memory on all allowed nodes that we are forced
1961 * into the second scan of the zonelist.
1962 *
1963 * In the second scan we ignore this zonelist cache and exactly
1964 * apply the watermarks to all zones, even it is slower to do so.
1965 * We are low on memory in the second scan, and should leave no stone
1966 * unturned looking for a free page.
1967 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001968static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001969 nodemask_t *allowednodes)
1970{
1971 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1972 int i; /* index of *z in zonelist zones */
1973 int n; /* node that zone *z is on */
1974
1975 zlc = zonelist->zlcache_ptr;
1976 if (!zlc)
1977 return 1;
1978
Mel Gormandd1a2392008-04-28 02:12:17 -07001979 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001980 n = zlc->z_to_n[i];
1981
1982 /* This zone is worth trying if it is allowed but not full */
1983 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1984}
1985
1986/*
1987 * Given 'z' scanning a zonelist, set the corresponding bit in
1988 * zlc->fullzones, so that subsequent attempts to allocate a page
1989 * from that zone don't waste time re-examining it.
1990 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001991static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001992{
1993 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1994 int i; /* index of *z in zonelist zones */
1995
1996 zlc = zonelist->zlcache_ptr;
1997 if (!zlc)
1998 return;
1999
Mel Gormandd1a2392008-04-28 02:12:17 -07002000 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002001
2002 set_bit(i, zlc->fullzones);
2003}
2004
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002005/*
2006 * clear all zones full, called after direct reclaim makes progress so that
2007 * a zone that was recently full is not skipped over for up to a second
2008 */
2009static void zlc_clear_zones_full(struct zonelist *zonelist)
2010{
2011 struct zonelist_cache *zlc; /* cached zonelist speedup info */
2012
2013 zlc = zonelist->zlcache_ptr;
2014 if (!zlc)
2015 return;
2016
2017 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2018}
2019
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002020static bool zone_local(struct zone *local_zone, struct zone *zone)
2021{
Johannes Weinerfff40682013-12-20 14:54:12 +00002022 return local_zone->node == zone->node;
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002023}
2024
David Rientjes957f8222012-10-08 16:33:24 -07002025static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2026{
Mel Gorman5f7a75a2014-06-04 16:07:15 -07002027 return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <
2028 RECLAIM_DISTANCE;
David Rientjes957f8222012-10-08 16:33:24 -07002029}
2030
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002031#else /* CONFIG_NUMA */
2032
2033static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
2034{
2035 return NULL;
2036}
2037
Mel Gormandd1a2392008-04-28 02:12:17 -07002038static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002039 nodemask_t *allowednodes)
2040{
2041 return 1;
2042}
2043
Mel Gormandd1a2392008-04-28 02:12:17 -07002044static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002045{
2046}
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002047
2048static void zlc_clear_zones_full(struct zonelist *zonelist)
2049{
2050}
David Rientjes957f8222012-10-08 16:33:24 -07002051
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002052static bool zone_local(struct zone *local_zone, struct zone *zone)
2053{
2054 return true;
2055}
2056
David Rientjes957f8222012-10-08 16:33:24 -07002057static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2058{
2059 return true;
2060}
2061
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002062#endif /* CONFIG_NUMA */
2063
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002064static void reset_alloc_batches(struct zone *preferred_zone)
2065{
2066 struct zone *zone = preferred_zone->zone_pgdat->node_zones;
2067
2068 do {
2069 mod_zone_page_state(zone, NR_ALLOC_BATCH,
2070 high_wmark_pages(zone) - low_wmark_pages(zone) -
2071 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
Johannes Weiner57054652014-10-09 15:28:17 -07002072 clear_bit(ZONE_FAIR_DEPLETED, &zone->flags);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002073 } while (zone++ != preferred_zone);
2074}
2075
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002076/*
Paul Jackson0798e512006-12-06 20:31:38 -08002077 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002078 * a page.
2079 */
2080static struct page *
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002081get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
2082 const struct alloc_context *ac)
Martin Hicks753ee722005-06-21 17:14:41 -07002083{
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002084 struct zonelist *zonelist = ac->zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07002085 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002086 struct page *page = NULL;
Mel Gorman5117f452009-06-16 15:31:59 -07002087 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002088 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
2089 int zlc_active = 0; /* set if using zonelist_cache */
2090 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gormana6e21b142014-06-04 16:10:12 -07002091 bool consider_zone_dirty = (alloc_flags & ALLOC_WMARK_LOW) &&
2092 (gfp_mask & __GFP_WRITE);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002093 int nr_fair_skipped = 0;
2094 bool zonelist_rescan;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002095
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002096zonelist_scan:
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002097 zonelist_rescan = false;
2098
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002099 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002100 * Scan zonelist, looking for a zone with enough free.
Vladimir Davydov344736f2014-10-20 15:50:30 +04002101 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002102 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002103 for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
2104 ac->nodemask) {
Johannes Weinere085dbc2013-09-11 14:20:46 -07002105 unsigned long mark;
2106
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002107 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002108 !zlc_zone_worth_trying(zonelist, z, allowednodes))
2109 continue;
Mel Gorman664eedd2014-06-04 16:10:08 -07002110 if (cpusets_enabled() &&
2111 (alloc_flags & ALLOC_CPUSET) &&
Vladimir Davydov344736f2014-10-20 15:50:30 +04002112 !cpuset_zone_allowed(zone, gfp_mask))
Mel Gormancd38b112011-07-25 17:12:29 -07002113 continue;
Johannes Weinera756cf52012-01-10 15:07:49 -08002114 /*
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002115 * Distribute pages in proportion to the individual
2116 * zone size to ensure fair page aging. The zone a
2117 * page was allocated in should have no effect on the
2118 * time the page has in memory before being reclaimed.
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002119 */
Johannes Weiner3a025762014-04-07 15:37:48 -07002120 if (alloc_flags & ALLOC_FAIR) {
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002121 if (!zone_local(ac->preferred_zone, zone))
Mel Gormanf7b5d642014-08-06 16:07:20 -07002122 break;
Johannes Weiner57054652014-10-09 15:28:17 -07002123 if (test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) {
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002124 nr_fair_skipped++;
Johannes Weiner3a025762014-04-07 15:37:48 -07002125 continue;
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002126 }
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002127 }
2128 /*
Johannes Weinera756cf52012-01-10 15:07:49 -08002129 * When allocating a page cache page for writing, we
2130 * want to get it from a zone that is within its dirty
2131 * limit, such that no single zone holds more than its
2132 * proportional share of globally allowed dirty pages.
2133 * The dirty limits take into account the zone's
2134 * lowmem reserves and high watermark so that kswapd
2135 * should be able to balance it without having to
2136 * write pages from its LRU list.
2137 *
2138 * This may look like it could increase pressure on
2139 * lower zones by failing allocations in higher zones
2140 * before they are full. But the pages that do spill
2141 * over are limited as the lower zones are protected
2142 * by this very same mechanism. It should not become
2143 * a practical burden to them.
2144 *
2145 * XXX: For now, allow allocations to potentially
2146 * exceed the per-zone dirty limit in the slowpath
2147 * (ALLOC_WMARK_LOW unset) before going into reclaim,
2148 * which is important when on a NUMA setup the allowed
2149 * zones are together not big enough to reach the
2150 * global limit. The proper fix for these situations
2151 * will require awareness of zones in the
2152 * dirty-throttling and the flusher threads.
2153 */
Mel Gormana6e21b142014-06-04 16:10:12 -07002154 if (consider_zone_dirty && !zone_dirty_ok(zone))
Mel Gorman800a1e72014-06-04 16:10:06 -07002155 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002156
Johannes Weinere085dbc2013-09-11 14:20:46 -07002157 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
2158 if (!zone_watermark_ok(zone, order, mark,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002159 ac->classzone_idx, alloc_flags)) {
Mel Gormanfa5e0842009-06-16 15:33:22 -07002160 int ret;
2161
Mel Gorman5dab2912014-06-04 16:10:14 -07002162 /* Checked here to keep the fast path fast */
2163 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
2164 if (alloc_flags & ALLOC_NO_WATERMARKS)
2165 goto try_this_zone;
2166
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002167 if (IS_ENABLED(CONFIG_NUMA) &&
2168 !did_zlc_setup && nr_online_nodes > 1) {
Mel Gormancd38b112011-07-25 17:12:29 -07002169 /*
2170 * we do zlc_setup if there are multiple nodes
2171 * and before considering the first zone allowed
2172 * by the cpuset.
2173 */
2174 allowednodes = zlc_setup(zonelist, alloc_flags);
2175 zlc_active = 1;
2176 did_zlc_setup = 1;
2177 }
2178
David Rientjes957f8222012-10-08 16:33:24 -07002179 if (zone_reclaim_mode == 0 ||
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002180 !zone_allows_reclaim(ac->preferred_zone, zone))
Mel Gormanfa5e0842009-06-16 15:33:22 -07002181 goto this_zone_full;
2182
Mel Gormancd38b112011-07-25 17:12:29 -07002183 /*
2184 * As we may have just activated ZLC, check if the first
2185 * eligible zone has failed zone_reclaim recently.
2186 */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002187 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
Mel Gormancd38b112011-07-25 17:12:29 -07002188 !zlc_zone_worth_trying(zonelist, z, allowednodes))
2189 continue;
2190
Mel Gormanfa5e0842009-06-16 15:33:22 -07002191 ret = zone_reclaim(zone, gfp_mask, order);
2192 switch (ret) {
2193 case ZONE_RECLAIM_NOSCAN:
2194 /* did not scan */
Mel Gormancd38b112011-07-25 17:12:29 -07002195 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002196 case ZONE_RECLAIM_FULL:
2197 /* scanned but unreclaimable */
Mel Gormancd38b112011-07-25 17:12:29 -07002198 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002199 default:
2200 /* did we reclaim enough */
Mel Gormanfed27192013-04-29 15:07:57 -07002201 if (zone_watermark_ok(zone, order, mark,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002202 ac->classzone_idx, alloc_flags))
Mel Gormanfed27192013-04-29 15:07:57 -07002203 goto try_this_zone;
2204
2205 /*
2206 * Failed to reclaim enough to meet watermark.
2207 * Only mark the zone full if checking the min
2208 * watermark or if we failed to reclaim just
2209 * 1<<order pages or else the page allocator
2210 * fastpath will prematurely mark zones full
2211 * when the watermark is between the low and
2212 * min watermarks.
2213 */
2214 if (((alloc_flags & ALLOC_WMARK_MASK) == ALLOC_WMARK_MIN) ||
2215 ret == ZONE_RECLAIM_SOME)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002216 goto this_zone_full;
Mel Gormanfed27192013-04-29 15:07:57 -07002217
2218 continue;
Paul Jackson0798e512006-12-06 20:31:38 -08002219 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002220 }
2221
Mel Gormanfa5e0842009-06-16 15:33:22 -07002222try_this_zone:
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002223 page = buffered_rmqueue(ac->preferred_zone, zone, order,
2224 gfp_mask, ac->migratetype);
Vlastimil Babka75379192015-02-11 15:25:38 -08002225 if (page) {
2226 if (prep_new_page(page, order, gfp_mask, alloc_flags))
2227 goto try_this_zone;
2228 return page;
2229 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002230this_zone_full:
Mel Gorman65bb3712014-06-04 16:10:05 -07002231 if (IS_ENABLED(CONFIG_NUMA) && zlc_active)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002232 zlc_mark_zone_full(zonelist, z);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002233 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002234
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002235 /*
2236 * The first pass makes sure allocations are spread fairly within the
2237 * local node. However, the local node might have free pages left
2238 * after the fairness batches are exhausted, and remote zones haven't
2239 * even been considered yet. Try once more without fairness, and
2240 * include remote zones now, before entering the slowpath and waking
2241 * kswapd: prefer spilling to a remote zone over swapping locally.
2242 */
2243 if (alloc_flags & ALLOC_FAIR) {
2244 alloc_flags &= ~ALLOC_FAIR;
2245 if (nr_fair_skipped) {
2246 zonelist_rescan = true;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002247 reset_alloc_batches(ac->preferred_zone);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002248 }
2249 if (nr_online_nodes > 1)
2250 zonelist_rescan = true;
2251 }
2252
2253 if (unlikely(IS_ENABLED(CONFIG_NUMA) && zlc_active)) {
2254 /* Disable zlc cache for second zonelist scan */
2255 zlc_active = 0;
2256 zonelist_rescan = true;
2257 }
2258
2259 if (zonelist_rescan)
2260 goto zonelist_scan;
2261
2262 return NULL;
Martin Hicks753ee722005-06-21 17:14:41 -07002263}
2264
David Rientjes29423e772011-03-22 16:30:47 -07002265/*
2266 * Large machines with many possible nodes should not always dump per-node
2267 * meminfo in irq context.
2268 */
2269static inline bool should_suppress_show_mem(void)
2270{
2271 bool ret = false;
2272
2273#if NODES_SHIFT > 8
2274 ret = in_interrupt();
2275#endif
2276 return ret;
2277}
2278
Dave Hansena238ab52011-05-24 17:12:16 -07002279static DEFINE_RATELIMIT_STATE(nopage_rs,
2280 DEFAULT_RATELIMIT_INTERVAL,
2281 DEFAULT_RATELIMIT_BURST);
2282
2283void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
2284{
Dave Hansena238ab52011-05-24 17:12:16 -07002285 unsigned int filter = SHOW_MEM_FILTER_NODES;
2286
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -08002287 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2288 debug_guardpage_minorder() > 0)
Dave Hansena238ab52011-05-24 17:12:16 -07002289 return;
2290
2291 /*
2292 * This documents exceptions given to allocations in certain
2293 * contexts that are allowed to allocate outside current's set
2294 * of allowed nodes.
2295 */
2296 if (!(gfp_mask & __GFP_NOMEMALLOC))
2297 if (test_thread_flag(TIF_MEMDIE) ||
2298 (current->flags & (PF_MEMALLOC | PF_EXITING)))
2299 filter &= ~SHOW_MEM_FILTER_NODES;
2300 if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
2301 filter &= ~SHOW_MEM_FILTER_NODES;
2302
2303 if (fmt) {
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002304 struct va_format vaf;
2305 va_list args;
2306
Dave Hansena238ab52011-05-24 17:12:16 -07002307 va_start(args, fmt);
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002308
2309 vaf.fmt = fmt;
2310 vaf.va = &args;
2311
2312 pr_warn("%pV", &vaf);
2313
Dave Hansena238ab52011-05-24 17:12:16 -07002314 va_end(args);
2315 }
2316
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002317 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2318 current->comm, order, gfp_mask);
Dave Hansena238ab52011-05-24 17:12:16 -07002319
2320 dump_stack();
2321 if (!should_suppress_show_mem())
2322 show_mem(filter);
2323}
2324
Mel Gorman11e33f62009-06-16 15:31:57 -07002325static inline int
2326should_alloc_retry(gfp_t gfp_mask, unsigned int order,
Mel Gormanf90ac392012-01-10 15:07:15 -08002327 unsigned long did_some_progress,
Mel Gorman11e33f62009-06-16 15:31:57 -07002328 unsigned long pages_reclaimed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329{
Mel Gorman11e33f62009-06-16 15:31:57 -07002330 /* Do not loop if specifically requested */
2331 if (gfp_mask & __GFP_NORETRY)
2332 return 0;
2333
Mel Gormanf90ac392012-01-10 15:07:15 -08002334 /* Always retry if specifically requested */
2335 if (gfp_mask & __GFP_NOFAIL)
2336 return 1;
2337
2338 /*
2339 * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim
2340 * making forward progress without invoking OOM. Suspend also disables
2341 * storage devices so kswapd will not help. Bail if we are suspending.
2342 */
2343 if (!did_some_progress && pm_suspended_storage())
2344 return 0;
2345
Mel Gorman11e33f62009-06-16 15:31:57 -07002346 /*
2347 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
2348 * means __GFP_NOFAIL, but that may not be true in other
2349 * implementations.
2350 */
2351 if (order <= PAGE_ALLOC_COSTLY_ORDER)
2352 return 1;
2353
2354 /*
2355 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
2356 * specified, then we retry until we no longer reclaim any pages
2357 * (above), or we've reclaimed an order of pages at least as
2358 * large as the allocation's order. In both cases, if the
2359 * allocation still fails, we stop retrying.
2360 */
2361 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
2362 return 1;
2363
Mel Gorman11e33f62009-06-16 15:31:57 -07002364 return 0;
2365}
2366
2367static inline struct page *
2368__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002369 const struct alloc_context *ac, unsigned long *did_some_progress)
Mel Gorman11e33f62009-06-16 15:31:57 -07002370{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372
Johannes Weiner9879de72015-01-26 12:58:32 -08002373 *did_some_progress = 0;
2374
Johannes Weiner9879de72015-01-26 12:58:32 -08002375 /*
2376 * Acquire the per-zone oom lock for each zone. If that
2377 * fails, somebody else is making progress for us.
2378 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002379 if (!oom_zonelist_trylock(ac->zonelist, gfp_mask)) {
Johannes Weiner9879de72015-01-26 12:58:32 -08002380 *did_some_progress = 1;
Mel Gorman11e33f62009-06-16 15:31:57 -07002381 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 return NULL;
2383 }
Jens Axboe6b1de912005-11-17 21:35:02 +01002384
Mel Gorman11e33f62009-06-16 15:31:57 -07002385 /*
2386 * Go through the zonelist yet one more time, keep very high watermark
2387 * here, this is only to catch a parallel oom killing, we must fail if
2388 * we're still under heavy pressure.
2389 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002390 page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order,
2391 ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002392 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07002393 goto out;
2394
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002395 if (!(gfp_mask & __GFP_NOFAIL)) {
Johannes Weiner9879de72015-01-26 12:58:32 -08002396 /* Coredumps can quickly deplete all memory reserves */
2397 if (current->flags & PF_DUMPCORE)
2398 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002399 /* The OOM killer will not help higher order allocs */
2400 if (order > PAGE_ALLOC_COSTLY_ORDER)
2401 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07002402 /* The OOM killer does not needlessly kill tasks for lowmem */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002403 if (ac->high_zoneidx < ZONE_NORMAL)
David Rientjes03668b32010-08-09 17:18:54 -07002404 goto out;
Johannes Weiner9879de72015-01-26 12:58:32 -08002405 /* The OOM killer does not compensate for light reclaim */
Johannes Weinercc873172015-02-27 15:52:09 -08002406 if (!(gfp_mask & __GFP_FS)) {
2407 /*
2408 * XXX: Page reclaim didn't yield anything,
2409 * and the OOM killer can't be invoked, but
2410 * keep looping as per should_alloc_retry().
2411 */
2412 *did_some_progress = 1;
Johannes Weiner9879de72015-01-26 12:58:32 -08002413 goto out;
Johannes Weinercc873172015-02-27 15:52:09 -08002414 }
David Rientjes4167e9b2015-04-14 15:46:55 -07002415 /* The OOM killer may not free memory on a specific node */
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002416 if (gfp_mask & __GFP_THISNODE)
2417 goto out;
2418 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002419 /* Exhausted what can be done so it's blamo time */
Michal Hockoe009d5d2015-03-12 16:25:52 -07002420 if (out_of_memory(ac->zonelist, gfp_mask, order, ac->nodemask, false)
2421 || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL))
Michal Hockoc32b3cb2015-02-11 15:26:24 -08002422 *did_some_progress = 1;
Mel Gorman11e33f62009-06-16 15:31:57 -07002423out:
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002424 oom_zonelist_unlock(ac->zonelist, gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002425 return page;
2426}
2427
Mel Gorman56de7262010-05-24 14:32:30 -07002428#ifdef CONFIG_COMPACTION
2429/* Try memory compaction for high-order allocations before reclaim */
2430static struct page *
2431__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002432 int alloc_flags, const struct alloc_context *ac,
2433 enum migrate_mode mode, int *contended_compaction,
2434 bool *deferred_compaction)
Mel Gorman56de7262010-05-24 14:32:30 -07002435{
Vlastimil Babka53853e22014-10-09 15:27:02 -07002436 unsigned long compact_result;
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002437 struct page *page;
Vlastimil Babka53853e22014-10-09 15:27:02 -07002438
Mel Gorman66199712012-01-12 17:19:41 -08002439 if (!order)
Mel Gorman56de7262010-05-24 14:32:30 -07002440 return NULL;
2441
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002442 current->flags |= PF_MEMALLOC;
Vlastimil Babka1a6d53a2015-02-11 15:25:44 -08002443 compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
2444 mode, contended_compaction);
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002445 current->flags &= ~PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07002446
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002447 switch (compact_result) {
2448 case COMPACT_DEFERRED:
Vlastimil Babka53853e22014-10-09 15:27:02 -07002449 *deferred_compaction = true;
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002450 /* fall-through */
2451 case COMPACT_SKIPPED:
2452 return NULL;
2453 default:
2454 break;
Mel Gorman56de7262010-05-24 14:32:30 -07002455 }
2456
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002457 /*
2458 * At least in one zone compaction wasn't deferred or skipped, so let's
2459 * count a compaction stall
2460 */
2461 count_vm_event(COMPACTSTALL);
2462
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002463 page = get_page_from_freelist(gfp_mask, order,
2464 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002465
2466 if (page) {
2467 struct zone *zone = page_zone(page);
2468
2469 zone->compact_blockskip_flush = false;
2470 compaction_defer_reset(zone, order, true);
2471 count_vm_event(COMPACTSUCCESS);
2472 return page;
2473 }
2474
2475 /*
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002476 * It's bad if compaction run occurs and fails. The most likely reason
2477 * is that pages exist, but not enough to satisfy watermarks.
2478 */
2479 count_vm_event(COMPACTFAIL);
2480
2481 cond_resched();
2482
Mel Gorman56de7262010-05-24 14:32:30 -07002483 return NULL;
2484}
2485#else
2486static inline struct page *
2487__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002488 int alloc_flags, const struct alloc_context *ac,
2489 enum migrate_mode mode, int *contended_compaction,
2490 bool *deferred_compaction)
Mel Gorman56de7262010-05-24 14:32:30 -07002491{
2492 return NULL;
2493}
2494#endif /* CONFIG_COMPACTION */
2495
Marek Szyprowskibba90712012-01-25 12:09:52 +01002496/* Perform direct synchronous page reclaim */
2497static int
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002498__perform_reclaim(gfp_t gfp_mask, unsigned int order,
2499 const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002500{
Mel Gorman11e33f62009-06-16 15:31:57 -07002501 struct reclaim_state reclaim_state;
Marek Szyprowskibba90712012-01-25 12:09:52 +01002502 int progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07002503
2504 cond_resched();
2505
2506 /* We now go into synchronous reclaim */
2507 cpuset_memory_pressure_bump();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002508 current->flags |= PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002509 lockdep_set_current_reclaim_state(gfp_mask);
2510 reclaim_state.reclaimed_slab = 0;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002511 current->reclaim_state = &reclaim_state;
Mel Gorman11e33f62009-06-16 15:31:57 -07002512
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002513 progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
2514 ac->nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002515
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002516 current->reclaim_state = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002517 lockdep_clear_current_reclaim_state();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002518 current->flags &= ~PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002519
2520 cond_resched();
2521
Marek Szyprowskibba90712012-01-25 12:09:52 +01002522 return progress;
2523}
2524
2525/* The really slow allocator path where we enter direct reclaim */
2526static inline struct page *
2527__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002528 int alloc_flags, const struct alloc_context *ac,
2529 unsigned long *did_some_progress)
Marek Szyprowskibba90712012-01-25 12:09:52 +01002530{
2531 struct page *page = NULL;
2532 bool drained = false;
2533
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002534 *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002535 if (unlikely(!(*did_some_progress)))
2536 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002537
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002538 /* After successful reclaim, reconsider all zones for allocation */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002539 if (IS_ENABLED(CONFIG_NUMA))
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002540 zlc_clear_zones_full(ac->zonelist);
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002541
Mel Gorman9ee493c2010-09-09 16:38:18 -07002542retry:
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002543 page = get_page_from_freelist(gfp_mask, order,
2544 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002545
2546 /*
2547 * If an allocation failed after direct reclaim, it could be because
2548 * pages are pinned on the per-cpu lists. Drain them and try again
2549 */
2550 if (!page && !drained) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002551 drain_all_pages(NULL);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002552 drained = true;
2553 goto retry;
2554 }
2555
Mel Gorman11e33f62009-06-16 15:31:57 -07002556 return page;
2557}
2558
Mel Gorman11e33f62009-06-16 15:31:57 -07002559/*
2560 * This is called in the allocator slow-path if the allocation request is of
2561 * sufficient urgency to ignore watermarks and take other desperate measures
2562 */
2563static inline struct page *
2564__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002565 const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002566{
2567 struct page *page;
2568
2569 do {
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002570 page = get_page_from_freelist(gfp_mask, order,
2571 ALLOC_NO_WATERMARKS, ac);
Mel Gorman11e33f62009-06-16 15:31:57 -07002572
2573 if (!page && gfp_mask & __GFP_NOFAIL)
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002574 wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC,
2575 HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07002576 } while (!page && (gfp_mask & __GFP_NOFAIL));
2577
2578 return page;
2579}
2580
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002581static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002582{
2583 struct zoneref *z;
2584 struct zone *zone;
2585
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002586 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
2587 ac->high_zoneidx, ac->nodemask)
2588 wakeup_kswapd(zone, order, zone_idx(ac->preferred_zone));
Mel Gorman11e33f62009-06-16 15:31:57 -07002589}
2590
Peter Zijlstra341ce062009-06-16 15:32:02 -07002591static inline int
2592gfp_to_alloc_flags(gfp_t gfp_mask)
2593{
Peter Zijlstra341ce062009-06-16 15:32:02 -07002594 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
David Rientjesb104a352014-07-30 16:08:24 -07002595 const bool atomic = !(gfp_mask & (__GFP_WAIT | __GFP_NO_KSWAPD));
Peter Zijlstra341ce062009-06-16 15:32:02 -07002596
Mel Gormana56f57f2009-06-16 15:32:02 -07002597 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07002598 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07002599
Peter Zijlstra341ce062009-06-16 15:32:02 -07002600 /*
2601 * The caller may dip into page reserves a bit more if the caller
2602 * cannot run direct reclaim, or if the caller has realtime scheduling
2603 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
David Rientjesb104a352014-07-30 16:08:24 -07002604 * set both ALLOC_HARDER (atomic == true) and ALLOC_HIGH (__GFP_HIGH).
Peter Zijlstra341ce062009-06-16 15:32:02 -07002605 */
Namhyung Kime6223a32010-10-26 14:21:59 -07002606 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07002607
David Rientjesb104a352014-07-30 16:08:24 -07002608 if (atomic) {
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002609 /*
David Rientjesb104a352014-07-30 16:08:24 -07002610 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
2611 * if it can't schedule.
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002612 */
David Rientjesb104a352014-07-30 16:08:24 -07002613 if (!(gfp_mask & __GFP_NOMEMALLOC))
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002614 alloc_flags |= ALLOC_HARDER;
Peter Zijlstra341ce062009-06-16 15:32:02 -07002615 /*
David Rientjesb104a352014-07-30 16:08:24 -07002616 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
Vladimir Davydov344736f2014-10-20 15:50:30 +04002617 * comment for __cpuset_node_allowed().
Peter Zijlstra341ce062009-06-16 15:32:02 -07002618 */
2619 alloc_flags &= ~ALLOC_CPUSET;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002620 } else if (unlikely(rt_task(current)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07002621 alloc_flags |= ALLOC_HARDER;
2622
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002623 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2624 if (gfp_mask & __GFP_MEMALLOC)
2625 alloc_flags |= ALLOC_NO_WATERMARKS;
Mel Gorman907aed42012-07-31 16:44:07 -07002626 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
2627 alloc_flags |= ALLOC_NO_WATERMARKS;
2628 else if (!in_interrupt() &&
2629 ((current->flags & PF_MEMALLOC) ||
2630 unlikely(test_thread_flag(TIF_MEMDIE))))
Peter Zijlstra341ce062009-06-16 15:32:02 -07002631 alloc_flags |= ALLOC_NO_WATERMARKS;
2632 }
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002633#ifdef CONFIG_CMA
David Rientjes43e7a342014-10-09 15:27:25 -07002634 if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002635 alloc_flags |= ALLOC_CMA;
2636#endif
Peter Zijlstra341ce062009-06-16 15:32:02 -07002637 return alloc_flags;
2638}
2639
Mel Gorman072bb0a2012-07-31 16:43:58 -07002640bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
2641{
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002642 return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
Mel Gorman072bb0a2012-07-31 16:43:58 -07002643}
2644
Mel Gorman11e33f62009-06-16 15:31:57 -07002645static inline struct page *
2646__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002647 struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002648{
2649 const gfp_t wait = gfp_mask & __GFP_WAIT;
2650 struct page *page = NULL;
2651 int alloc_flags;
2652 unsigned long pages_reclaimed = 0;
2653 unsigned long did_some_progress;
David Rientjese0b9dae2014-06-04 16:08:28 -07002654 enum migrate_mode migration_mode = MIGRATE_ASYNC;
Mel Gorman66199712012-01-12 17:19:41 -08002655 bool deferred_compaction = false;
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07002656 int contended_compaction = COMPACT_CONTENDED_NONE;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002657
Christoph Lameter952f3b52006-12-06 20:33:26 -08002658 /*
Mel Gorman72807a72009-06-16 15:32:18 -07002659 * In the slowpath, we sanity check order to avoid ever trying to
2660 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2661 * be using allocators in order of preference for an area that is
2662 * too large.
2663 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07002664 if (order >= MAX_ORDER) {
2665 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07002666 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07002667 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002668
Christoph Lameter952f3b52006-12-06 20:33:26 -08002669 /*
David Rientjes4167e9b2015-04-14 15:46:55 -07002670 * If this allocation cannot block and it is for a specific node, then
2671 * fail early. There's no need to wakeup kswapd or retry for a
2672 * speculative node-specific allocation.
Christoph Lameter952f3b52006-12-06 20:33:26 -08002673 */
David Rientjes4167e9b2015-04-14 15:46:55 -07002674 if (IS_ENABLED(CONFIG_NUMA) && (gfp_mask & __GFP_THISNODE) && !wait)
Christoph Lameter952f3b52006-12-06 20:33:26 -08002675 goto nopage;
2676
Johannes Weiner9879de72015-01-26 12:58:32 -08002677retry:
Johannes Weiner3a025762014-04-07 15:37:48 -07002678 if (!(gfp_mask & __GFP_NO_KSWAPD))
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002679 wake_all_kswapds(order, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002680
Paul Jackson9bf22292005-09-06 15:18:12 -07002681 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002682 * OK, we're below the kswapd watermark and have kicked background
2683 * reclaim. Now things get more complex, so set up alloc_flags according
2684 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07002685 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002686 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687
David Rientjesf33261d2011-01-25 15:07:20 -08002688 /*
2689 * Find the true preferred zone if the allocation is unconstrained by
2690 * cpusets.
2691 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002692 if (!(alloc_flags & ALLOC_CPUSET) && !ac->nodemask) {
Mel Gormand8846372014-06-04 16:10:33 -07002693 struct zoneref *preferred_zoneref;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002694 preferred_zoneref = first_zones_zonelist(ac->zonelist,
2695 ac->high_zoneidx, NULL, &ac->preferred_zone);
2696 ac->classzone_idx = zonelist_zone_idx(preferred_zoneref);
Mel Gormand8846372014-06-04 16:10:33 -07002697 }
David Rientjesf33261d2011-01-25 15:07:20 -08002698
Peter Zijlstra341ce062009-06-16 15:32:02 -07002699 /* This is the last chance, in general, before the goto nopage. */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002700 page = get_page_from_freelist(gfp_mask, order,
2701 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002702 if (page)
2703 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704
Mel Gorman11e33f62009-06-16 15:31:57 -07002705 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002706 if (alloc_flags & ALLOC_NO_WATERMARKS) {
Mel Gorman183f6372012-07-31 16:44:12 -07002707 /*
2708 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
2709 * the allocation is high priority and these type of
2710 * allocations are system rather than user orientated
2711 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002712 ac->zonelist = node_zonelist(numa_node_id(), gfp_mask);
Mel Gorman183f6372012-07-31 16:44:12 -07002713
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002714 page = __alloc_pages_high_priority(gfp_mask, order, ac);
2715
Mel Gormancfd19c52012-07-31 16:44:10 -07002716 if (page) {
Peter Zijlstra341ce062009-06-16 15:32:02 -07002717 goto got_pg;
Mel Gormancfd19c52012-07-31 16:44:10 -07002718 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719 }
2720
2721 /* Atomic allocations - we can't balance anything */
David Rientjesaed0a0e2014-01-21 15:51:12 -08002722 if (!wait) {
2723 /*
2724 * All existing users of the deprecated __GFP_NOFAIL are
2725 * blockable, so warn of any new users that actually allow this
2726 * type of allocation to fail.
2727 */
2728 WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729 goto nopage;
David Rientjesaed0a0e2014-01-21 15:51:12 -08002730 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731
Peter Zijlstra341ce062009-06-16 15:32:02 -07002732 /* Avoid recursion of direct reclaim */
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002733 if (current->flags & PF_MEMALLOC)
Peter Zijlstra341ce062009-06-16 15:32:02 -07002734 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735
David Rientjes6583bb62009-07-29 15:02:06 -07002736 /* Avoid allocations with no watermarks from looping endlessly */
2737 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2738 goto nopage;
2739
Mel Gorman77f1fe62011-01-13 15:45:57 -08002740 /*
2741 * Try direct compaction. The first pass is asynchronous. Subsequent
2742 * attempts after direct reclaim are synchronous
2743 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002744 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
2745 migration_mode,
2746 &contended_compaction,
Vlastimil Babka53853e22014-10-09 15:27:02 -07002747 &deferred_compaction);
Mel Gorman56de7262010-05-24 14:32:30 -07002748 if (page)
2749 goto got_pg;
David Rientjes75f30862014-06-04 16:08:30 -07002750
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07002751 /* Checks for THP-specific high-order allocations */
2752 if ((gfp_mask & GFP_TRANSHUGE) == GFP_TRANSHUGE) {
2753 /*
2754 * If compaction is deferred for high-order allocations, it is
2755 * because sync compaction recently failed. If this is the case
2756 * and the caller requested a THP allocation, we do not want
2757 * to heavily disrupt the system, so we fail the allocation
2758 * instead of entering direct reclaim.
2759 */
2760 if (deferred_compaction)
2761 goto nopage;
2762
2763 /*
2764 * In all zones where compaction was attempted (and not
2765 * deferred or skipped), lock contention has been detected.
2766 * For THP allocation we do not want to disrupt the others
2767 * so we fallback to base pages instead.
2768 */
2769 if (contended_compaction == COMPACT_CONTENDED_LOCK)
2770 goto nopage;
2771
2772 /*
2773 * If compaction was aborted due to need_resched(), we do not
2774 * want to further increase allocation latency, unless it is
2775 * khugepaged trying to collapse.
2776 */
2777 if (contended_compaction == COMPACT_CONTENDED_SCHED
2778 && !(current->flags & PF_KTHREAD))
2779 goto nopage;
2780 }
Mel Gorman66199712012-01-12 17:19:41 -08002781
David Rientjes8fe78042014-08-06 16:07:54 -07002782 /*
2783 * It can become very expensive to allocate transparent hugepages at
2784 * fault, so use asynchronous memory compaction for THP unless it is
2785 * khugepaged trying to collapse.
2786 */
2787 if ((gfp_mask & GFP_TRANSHUGE) != GFP_TRANSHUGE ||
2788 (current->flags & PF_KTHREAD))
2789 migration_mode = MIGRATE_SYNC_LIGHT;
2790
Mel Gorman11e33f62009-06-16 15:31:57 -07002791 /* Try direct reclaim and then allocating */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002792 page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
2793 &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07002794 if (page)
2795 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796
Mel Gorman11e33f62009-06-16 15:31:57 -07002797 /* Check if we should retry the allocation */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07002798 pages_reclaimed += did_some_progress;
Mel Gormanf90ac392012-01-10 15:07:15 -08002799 if (should_alloc_retry(gfp_mask, order, did_some_progress,
2800 pages_reclaimed)) {
Johannes Weiner9879de72015-01-26 12:58:32 -08002801 /*
2802 * If we fail to make progress by freeing individual
2803 * pages, but the allocation wants us to keep going,
2804 * start OOM killing tasks.
2805 */
2806 if (!did_some_progress) {
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002807 page = __alloc_pages_may_oom(gfp_mask, order, ac,
2808 &did_some_progress);
Johannes Weiner9879de72015-01-26 12:58:32 -08002809 if (page)
2810 goto got_pg;
2811 if (!did_some_progress)
2812 goto nopage;
2813 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002814 /* Wait for some write requests to complete then retry */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002815 wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC, HZ/50);
Johannes Weiner9879de72015-01-26 12:58:32 -08002816 goto retry;
Mel Gorman3e7d3442011-01-13 15:45:56 -08002817 } else {
2818 /*
2819 * High-order allocations do not necessarily loop after
2820 * direct reclaim and reclaim/compaction depends on compaction
2821 * being called after reclaim so call directly if necessary
2822 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002823 page = __alloc_pages_direct_compact(gfp_mask, order,
2824 alloc_flags, ac, migration_mode,
2825 &contended_compaction,
Vlastimil Babka53853e22014-10-09 15:27:02 -07002826 &deferred_compaction);
Mel Gorman3e7d3442011-01-13 15:45:56 -08002827 if (page)
2828 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829 }
2830
2831nopage:
Dave Hansena238ab52011-05-24 17:12:16 -07002832 warn_alloc_failed(gfp_mask, order, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002833got_pg:
Mel Gorman072bb0a2012-07-31 16:43:58 -07002834 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835}
Mel Gorman11e33f62009-06-16 15:31:57 -07002836
2837/*
2838 * This is the 'heart' of the zoned buddy allocator.
2839 */
2840struct page *
2841__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2842 struct zonelist *zonelist, nodemask_t *nodemask)
2843{
Mel Gormand8846372014-06-04 16:10:33 -07002844 struct zoneref *preferred_zoneref;
Mel Gormancc9a6c82012-03-21 16:34:11 -07002845 struct page *page = NULL;
Mel Gormancc9a6c82012-03-21 16:34:11 -07002846 unsigned int cpuset_mems_cookie;
Johannes Weiner3a025762014-04-07 15:37:48 -07002847 int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET|ALLOC_FAIR;
Andrew Morton91fbdc02015-02-11 15:25:04 -08002848 gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002849 struct alloc_context ac = {
2850 .high_zoneidx = gfp_zone(gfp_mask),
2851 .nodemask = nodemask,
2852 .migratetype = gfpflags_to_migratetype(gfp_mask),
2853 };
Mel Gorman11e33f62009-06-16 15:31:57 -07002854
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10002855 gfp_mask &= gfp_allowed_mask;
2856
Mel Gorman11e33f62009-06-16 15:31:57 -07002857 lockdep_trace_alloc(gfp_mask);
2858
2859 might_sleep_if(gfp_mask & __GFP_WAIT);
2860
2861 if (should_fail_alloc_page(gfp_mask, order))
2862 return NULL;
2863
2864 /*
2865 * Check the zones suitable for the gfp_mask contain at least one
2866 * valid zone. It's possible to have an empty zonelist as a result
David Rientjes4167e9b2015-04-14 15:46:55 -07002867 * of __GFP_THISNODE and a memoryless node
Mel Gorman11e33f62009-06-16 15:31:57 -07002868 */
2869 if (unlikely(!zonelist->_zonerefs->zone))
2870 return NULL;
2871
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002872 if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE)
Vlastimil Babka21bb9bd2014-10-09 15:26:51 -07002873 alloc_flags |= ALLOC_CMA;
2874
Mel Gormancc9a6c82012-03-21 16:34:11 -07002875retry_cpuset:
Mel Gormand26914d2014-04-03 14:47:24 -07002876 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07002877
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002878 /* We set it here, as __alloc_pages_slowpath might have changed it */
2879 ac.zonelist = zonelist;
Mel Gorman5117f452009-06-16 15:31:59 -07002880 /* The preferred zone is used for statistics later */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002881 preferred_zoneref = first_zones_zonelist(ac.zonelist, ac.high_zoneidx,
2882 ac.nodemask ? : &cpuset_current_mems_allowed,
2883 &ac.preferred_zone);
2884 if (!ac.preferred_zone)
Mel Gormancc9a6c82012-03-21 16:34:11 -07002885 goto out;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002886 ac.classzone_idx = zonelist_zone_idx(preferred_zoneref);
Mel Gorman5117f452009-06-16 15:31:59 -07002887
2888 /* First allocation attempt */
Andrew Morton91fbdc02015-02-11 15:25:04 -08002889 alloc_mask = gfp_mask|__GFP_HARDWALL;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002890 page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
Ming Lei21caf2f2013-02-22 16:34:08 -08002891 if (unlikely(!page)) {
2892 /*
2893 * Runtime PM, block IO and its error handling path
2894 * can deadlock because I/O on the device might not
2895 * complete.
2896 */
Andrew Morton91fbdc02015-02-11 15:25:04 -08002897 alloc_mask = memalloc_noio_flags(gfp_mask);
2898
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002899 page = __alloc_pages_slowpath(alloc_mask, order, &ac);
Ming Lei21caf2f2013-02-22 16:34:08 -08002900 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002901
Xishi Qiu23f086f2015-02-11 15:25:07 -08002902 if (kmemcheck_enabled && page)
2903 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
2904
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002905 trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002906
2907out:
2908 /*
2909 * When updating a task's mems_allowed, it is possible to race with
2910 * parallel threads in such a way that an allocation can fail while
2911 * the mask is being updated. If a page allocation is about to fail,
2912 * check if the cpuset changed during allocation and if so, retry.
2913 */
Mel Gormand26914d2014-04-03 14:47:24 -07002914 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
Mel Gormancc9a6c82012-03-21 16:34:11 -07002915 goto retry_cpuset;
2916
Mel Gorman11e33f62009-06-16 15:31:57 -07002917 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918}
Mel Gormand2391712009-06-16 15:31:52 -07002919EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002920
2921/*
2922 * Common helper functions.
2923 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08002924unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002925{
Akinobu Mita945a1112009-09-21 17:01:47 -07002926 struct page *page;
2927
2928 /*
2929 * __get_free_pages() returns a 32-bit address, which cannot represent
2930 * a highmem page
2931 */
2932 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2933
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934 page = alloc_pages(gfp_mask, order);
2935 if (!page)
2936 return 0;
2937 return (unsigned long) page_address(page);
2938}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939EXPORT_SYMBOL(__get_free_pages);
2940
Harvey Harrison920c7a52008-02-04 22:29:26 -08002941unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002942{
Akinobu Mita945a1112009-09-21 17:01:47 -07002943 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002944}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002945EXPORT_SYMBOL(get_zeroed_page);
2946
Harvey Harrison920c7a52008-02-04 22:29:26 -08002947void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002948{
Nick Pigginb5810032005-10-29 18:16:12 -07002949 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002950 if (order == 0)
Mel Gormanb745bc82014-06-04 16:10:22 -07002951 free_hot_cold_page(page, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002952 else
2953 __free_pages_ok(page, order);
2954 }
2955}
2956
2957EXPORT_SYMBOL(__free_pages);
2958
Harvey Harrison920c7a52008-02-04 22:29:26 -08002959void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960{
2961 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07002962 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002963 __free_pages(virt_to_page((void *)addr), order);
2964 }
2965}
2966
2967EXPORT_SYMBOL(free_pages);
2968
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002969/*
Vladimir Davydov52383432014-06-04 16:06:39 -07002970 * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
2971 * of the current memory cgroup.
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002972 *
Vladimir Davydov52383432014-06-04 16:06:39 -07002973 * It should be used when the caller would like to use kmalloc, but since the
2974 * allocation is large, it has to fall back to the page allocator.
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002975 */
Vladimir Davydov52383432014-06-04 16:06:39 -07002976struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order)
2977{
2978 struct page *page;
2979 struct mem_cgroup *memcg = NULL;
2980
2981 if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
2982 return NULL;
2983 page = alloc_pages(gfp_mask, order);
2984 memcg_kmem_commit_charge(page, memcg, order);
2985 return page;
2986}
2987
2988struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
2989{
2990 struct page *page;
2991 struct mem_cgroup *memcg = NULL;
2992
2993 if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
2994 return NULL;
2995 page = alloc_pages_node(nid, gfp_mask, order);
2996 memcg_kmem_commit_charge(page, memcg, order);
2997 return page;
2998}
2999
3000/*
3001 * __free_kmem_pages and free_kmem_pages will free pages allocated with
3002 * alloc_kmem_pages.
3003 */
3004void __free_kmem_pages(struct page *page, unsigned int order)
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003005{
3006 memcg_kmem_uncharge_pages(page, order);
3007 __free_pages(page, order);
3008}
3009
Vladimir Davydov52383432014-06-04 16:06:39 -07003010void free_kmem_pages(unsigned long addr, unsigned int order)
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003011{
3012 if (addr != 0) {
3013 VM_BUG_ON(!virt_addr_valid((void *)addr));
Vladimir Davydov52383432014-06-04 16:06:39 -07003014 __free_kmem_pages(virt_to_page((void *)addr), order);
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003015 }
3016}
3017
Andi Kleenee85c2e2011-05-11 15:13:34 -07003018static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
3019{
3020 if (addr) {
3021 unsigned long alloc_end = addr + (PAGE_SIZE << order);
3022 unsigned long used = addr + PAGE_ALIGN(size);
3023
3024 split_page(virt_to_page((void *)addr), order);
3025 while (used < alloc_end) {
3026 free_page(used);
3027 used += PAGE_SIZE;
3028 }
3029 }
3030 return (void *)addr;
3031}
3032
Timur Tabi2be0ffe2008-07-23 21:28:11 -07003033/**
3034 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
3035 * @size: the number of bytes to allocate
3036 * @gfp_mask: GFP flags for the allocation
3037 *
3038 * This function is similar to alloc_pages(), except that it allocates the
3039 * minimum number of pages to satisfy the request. alloc_pages() can only
3040 * allocate memory in power-of-two pages.
3041 *
3042 * This function is also limited by MAX_ORDER.
3043 *
3044 * Memory allocated by this function must be released by free_pages_exact().
3045 */
3046void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
3047{
3048 unsigned int order = get_order(size);
3049 unsigned long addr;
3050
3051 addr = __get_free_pages(gfp_mask, order);
Andi Kleenee85c2e2011-05-11 15:13:34 -07003052 return make_alloc_exact(addr, order, size);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07003053}
3054EXPORT_SYMBOL(alloc_pages_exact);
3055
3056/**
Andi Kleenee85c2e2011-05-11 15:13:34 -07003057 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
3058 * pages on a node.
Randy Dunlapb5e6ab52011-05-16 13:16:54 -07003059 * @nid: the preferred node ID where memory should be allocated
Andi Kleenee85c2e2011-05-11 15:13:34 -07003060 * @size: the number of bytes to allocate
3061 * @gfp_mask: GFP flags for the allocation
3062 *
3063 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
3064 * back.
3065 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
3066 * but is not exact.
3067 */
Fabian Fredericke1931812014-08-06 16:04:59 -07003068void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
Andi Kleenee85c2e2011-05-11 15:13:34 -07003069{
3070 unsigned order = get_order(size);
3071 struct page *p = alloc_pages_node(nid, gfp_mask, order);
3072 if (!p)
3073 return NULL;
3074 return make_alloc_exact((unsigned long)page_address(p), order, size);
3075}
Andi Kleenee85c2e2011-05-11 15:13:34 -07003076
3077/**
Timur Tabi2be0ffe2008-07-23 21:28:11 -07003078 * free_pages_exact - release memory allocated via alloc_pages_exact()
3079 * @virt: the value returned by alloc_pages_exact.
3080 * @size: size of allocation, same value as passed to alloc_pages_exact().
3081 *
3082 * Release the memory allocated by a previous call to alloc_pages_exact.
3083 */
3084void free_pages_exact(void *virt, size_t size)
3085{
3086 unsigned long addr = (unsigned long)virt;
3087 unsigned long end = addr + PAGE_ALIGN(size);
3088
3089 while (addr < end) {
3090 free_page(addr);
3091 addr += PAGE_SIZE;
3092 }
3093}
3094EXPORT_SYMBOL(free_pages_exact);
3095
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003096/**
3097 * nr_free_zone_pages - count number of pages beyond high watermark
3098 * @offset: The zone index of the highest zone
3099 *
3100 * nr_free_zone_pages() counts the number of counts pages which are beyond the
3101 * high watermark within all zones at or below a given zone index. For each
3102 * zone, the number of pages is calculated as:
Jiang Liu834405c2013-07-03 15:03:04 -07003103 * managed_pages - high_pages
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003104 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003105static unsigned long nr_free_zone_pages(int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003106{
Mel Gormandd1a2392008-04-28 02:12:17 -07003107 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003108 struct zone *zone;
3109
Martin J. Blighe310fd42005-07-29 22:59:18 -07003110 /* Just pick one node, since fallback list is circular */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003111 unsigned long sum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003112
Mel Gorman0e884602008-04-28 02:12:14 -07003113 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003114
Mel Gorman54a6eb52008-04-28 02:12:16 -07003115 for_each_zone_zonelist(zone, z, zonelist, offset) {
Jiang Liub40da042013-02-22 16:33:52 -08003116 unsigned long size = zone->managed_pages;
Mel Gorman41858962009-06-16 15:32:12 -07003117 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07003118 if (size > high)
3119 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120 }
3121
3122 return sum;
3123}
3124
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003125/**
3126 * nr_free_buffer_pages - count number of pages beyond high watermark
3127 *
3128 * nr_free_buffer_pages() counts the number of pages which are beyond the high
3129 * watermark within ZONE_DMA and ZONE_NORMAL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003130 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003131unsigned long nr_free_buffer_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003132{
Al Viroaf4ca452005-10-21 02:55:38 -04003133 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003134}
Meelap Shahc2f1a552007-07-17 04:04:39 -07003135EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003136
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003137/**
3138 * nr_free_pagecache_pages - count number of pages beyond high watermark
3139 *
3140 * nr_free_pagecache_pages() counts the number of pages which are beyond the
3141 * high watermark within all zones.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003142 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003143unsigned long nr_free_pagecache_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003144{
Mel Gorman2a1e2742007-07-17 04:03:12 -07003145 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003146}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07003147
3148static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003149{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08003150 if (IS_ENABLED(CONFIG_NUMA))
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08003151 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003152}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003153
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154void si_meminfo(struct sysinfo *val)
3155{
3156 val->totalram = totalram_pages;
Rafael Aquinicc7452b2014-08-06 16:06:38 -07003157 val->sharedram = global_page_state(NR_SHMEM);
Christoph Lameterd23ad422007-02-10 01:43:02 -08003158 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003159 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003160 val->totalhigh = totalhigh_pages;
3161 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003162 val->mem_unit = PAGE_SIZE;
3163}
3164
3165EXPORT_SYMBOL(si_meminfo);
3166
3167#ifdef CONFIG_NUMA
3168void si_meminfo_node(struct sysinfo *val, int nid)
3169{
Jiang Liucdd91a72013-07-03 15:03:27 -07003170 int zone_type; /* needs to be signed */
3171 unsigned long managed_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003172 pg_data_t *pgdat = NODE_DATA(nid);
3173
Jiang Liucdd91a72013-07-03 15:03:27 -07003174 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
3175 managed_pages += pgdat->node_zones[zone_type].managed_pages;
3176 val->totalram = managed_pages;
Rafael Aquinicc7452b2014-08-06 16:06:38 -07003177 val->sharedram = node_page_state(nid, NR_SHMEM);
Christoph Lameterd23ad422007-02-10 01:43:02 -08003178 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003179#ifdef CONFIG_HIGHMEM
Jiang Liub40da042013-02-22 16:33:52 -08003180 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08003181 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
3182 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003183#else
3184 val->totalhigh = 0;
3185 val->freehigh = 0;
3186#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003187 val->mem_unit = PAGE_SIZE;
3188}
3189#endif
3190
David Rientjesddd588b2011-03-22 16:30:46 -07003191/*
David Rientjes7bf02ea2011-05-24 17:11:16 -07003192 * Determine whether the node should be displayed or not, depending on whether
3193 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
David Rientjesddd588b2011-03-22 16:30:46 -07003194 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07003195bool skip_free_areas_node(unsigned int flags, int nid)
David Rientjesddd588b2011-03-22 16:30:46 -07003196{
3197 bool ret = false;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003198 unsigned int cpuset_mems_cookie;
David Rientjesddd588b2011-03-22 16:30:46 -07003199
3200 if (!(flags & SHOW_MEM_FILTER_NODES))
3201 goto out;
3202
Mel Gormancc9a6c82012-03-21 16:34:11 -07003203 do {
Mel Gormand26914d2014-04-03 14:47:24 -07003204 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07003205 ret = !node_isset(nid, cpuset_current_mems_allowed);
Mel Gormand26914d2014-04-03 14:47:24 -07003206 } while (read_mems_allowed_retry(cpuset_mems_cookie));
David Rientjesddd588b2011-03-22 16:30:46 -07003207out:
3208 return ret;
3209}
3210
Linus Torvalds1da177e2005-04-16 15:20:36 -07003211#define K(x) ((x) << (PAGE_SHIFT-10))
3212
Rabin Vincent377e4f12012-12-11 16:00:24 -08003213static void show_migration_types(unsigned char type)
3214{
3215 static const char types[MIGRATE_TYPES] = {
3216 [MIGRATE_UNMOVABLE] = 'U',
3217 [MIGRATE_RECLAIMABLE] = 'E',
3218 [MIGRATE_MOVABLE] = 'M',
3219 [MIGRATE_RESERVE] = 'R',
3220#ifdef CONFIG_CMA
3221 [MIGRATE_CMA] = 'C',
3222#endif
Minchan Kim194159f2013-02-22 16:33:58 -08003223#ifdef CONFIG_MEMORY_ISOLATION
Rabin Vincent377e4f12012-12-11 16:00:24 -08003224 [MIGRATE_ISOLATE] = 'I',
Minchan Kim194159f2013-02-22 16:33:58 -08003225#endif
Rabin Vincent377e4f12012-12-11 16:00:24 -08003226 };
3227 char tmp[MIGRATE_TYPES + 1];
3228 char *p = tmp;
3229 int i;
3230
3231 for (i = 0; i < MIGRATE_TYPES; i++) {
3232 if (type & (1 << i))
3233 *p++ = types[i];
3234 }
3235
3236 *p = '\0';
3237 printk("(%s) ", tmp);
3238}
3239
Linus Torvalds1da177e2005-04-16 15:20:36 -07003240/*
3241 * Show free area list (used inside shift_scroll-lock stuff)
3242 * We also calculate the percentage fragmentation. We do this by counting the
3243 * memory on each free list with the exception of the first item on the list.
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003244 *
3245 * Bits in @filter:
3246 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
3247 * cpuset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003248 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07003249void show_free_areas(unsigned int filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003250{
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003251 unsigned long free_pcp = 0;
Jes Sorensenc7241912006-09-27 01:50:05 -07003252 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003253 struct zone *zone;
3254
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003255 for_each_populated_zone(zone) {
David Rientjes7bf02ea2011-05-24 17:11:16 -07003256 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003257 continue;
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003258
Konstantin Khlebnikov761b0672015-04-14 15:45:32 -07003259 for_each_online_cpu(cpu)
3260 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003261 }
3262
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003263 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3264 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003265 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
3266 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003267 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003268 " free:%lu free_pcp:%lu free_cma:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003269 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003270 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003271 global_page_state(NR_ISOLATED_ANON),
3272 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003273 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003274 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003275 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07003276 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07003277 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07003278 global_page_state(NR_UNSTABLE_NFS),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07003279 global_page_state(NR_SLAB_RECLAIMABLE),
3280 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07003281 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003282 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08003283 global_page_state(NR_PAGETABLE),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003284 global_page_state(NR_BOUNCE),
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003285 global_page_state(NR_FREE_PAGES),
3286 free_pcp,
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003287 global_page_state(NR_FREE_CMA_PAGES));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003288
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003289 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003290 int i;
3291
David Rientjes7bf02ea2011-05-24 17:11:16 -07003292 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003293 continue;
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003294
3295 free_pcp = 0;
3296 for_each_online_cpu(cpu)
3297 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
3298
Linus Torvalds1da177e2005-04-16 15:20:36 -07003299 show_node(zone);
3300 printk("%s"
3301 " free:%lukB"
3302 " min:%lukB"
3303 " low:%lukB"
3304 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003305 " active_anon:%lukB"
3306 " inactive_anon:%lukB"
3307 " active_file:%lukB"
3308 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003309 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003310 " isolated(anon):%lukB"
3311 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003312 " present:%lukB"
Jiang Liu9feedc92012-12-12 13:52:12 -08003313 " managed:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003314 " mlocked:%lukB"
3315 " dirty:%lukB"
3316 " writeback:%lukB"
3317 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003318 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003319 " slab_reclaimable:%lukB"
3320 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003321 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003322 " pagetables:%lukB"
3323 " unstable:%lukB"
3324 " bounce:%lukB"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003325 " free_pcp:%lukB"
3326 " local_pcp:%ukB"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003327 " free_cma:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003328 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003329 " pages_scanned:%lu"
3330 " all_unreclaimable? %s"
3331 "\n",
3332 zone->name,
Mel Gorman88f5acf2011-01-13 15:45:41 -08003333 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07003334 K(min_wmark_pages(zone)),
3335 K(low_wmark_pages(zone)),
3336 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003337 K(zone_page_state(zone, NR_ACTIVE_ANON)),
3338 K(zone_page_state(zone, NR_INACTIVE_ANON)),
3339 K(zone_page_state(zone, NR_ACTIVE_FILE)),
3340 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003341 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003342 K(zone_page_state(zone, NR_ISOLATED_ANON)),
3343 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003344 K(zone->present_pages),
Jiang Liu9feedc92012-12-12 13:52:12 -08003345 K(zone->managed_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003346 K(zone_page_state(zone, NR_MLOCK)),
3347 K(zone_page_state(zone, NR_FILE_DIRTY)),
3348 K(zone_page_state(zone, NR_WRITEBACK)),
3349 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003350 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003351 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3352 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003353 zone_page_state(zone, NR_KERNEL_STACK) *
3354 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003355 K(zone_page_state(zone, NR_PAGETABLE)),
3356 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3357 K(zone_page_state(zone, NR_BOUNCE)),
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003358 K(free_pcp),
3359 K(this_cpu_read(zone->pageset->pcp.count)),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003360 K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003361 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Mel Gorman0d5d8232014-08-06 16:07:16 -07003362 K(zone_page_state(zone, NR_PAGES_SCANNED)),
Lisa Du6e543d52013-09-11 14:22:36 -07003363 (!zone_reclaimable(zone) ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07003364 );
3365 printk("lowmem_reserve[]:");
3366 for (i = 0; i < MAX_NR_ZONES; i++)
Mel Gorman3484b2d2014-08-06 16:07:14 -07003367 printk(" %ld", zone->lowmem_reserve[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003368 printk("\n");
3369 }
3370
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003371 for_each_populated_zone(zone) {
Pintu Kumarb8af2942013-09-11 14:20:34 -07003372 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003373 unsigned char types[MAX_ORDER];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003374
David Rientjes7bf02ea2011-05-24 17:11:16 -07003375 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003376 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003377 show_node(zone);
3378 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003379
3380 spin_lock_irqsave(&zone->lock, flags);
3381 for (order = 0; order < MAX_ORDER; order++) {
Rabin Vincent377e4f12012-12-11 16:00:24 -08003382 struct free_area *area = &zone->free_area[order];
3383 int type;
3384
3385 nr[order] = area->nr_free;
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003386 total += nr[order] << order;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003387
3388 types[order] = 0;
3389 for (type = 0; type < MIGRATE_TYPES; type++) {
3390 if (!list_empty(&area->free_list[type]))
3391 types[order] |= 1 << type;
3392 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003393 }
3394 spin_unlock_irqrestore(&zone->lock, flags);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003395 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003396 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003397 if (nr[order])
3398 show_migration_types(types[order]);
3399 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003400 printk("= %lukB\n", K(total));
3401 }
3402
David Rientjes949f7ec2013-04-29 15:07:48 -07003403 hugetlb_show_meminfo();
3404
Larry Woodmane6f36022008-02-04 22:29:30 -08003405 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
3406
Linus Torvalds1da177e2005-04-16 15:20:36 -07003407 show_swap_cache_info();
3408}
3409
Mel Gorman19770b32008-04-28 02:12:18 -07003410static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
3411{
3412 zoneref->zone = zone;
3413 zoneref->zone_idx = zone_idx(zone);
3414}
3415
Linus Torvalds1da177e2005-04-16 15:20:36 -07003416/*
3417 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08003418 *
3419 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003420 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003421static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003422 int nr_zones)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003423{
Christoph Lameter1a932052006-01-06 00:11:16 -08003424 struct zone *zone;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003425 enum zone_type zone_type = MAX_NR_ZONES;
Christoph Lameter02a68a52006-01-06 00:11:18 -08003426
3427 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003428 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08003429 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08003430 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07003431 zoneref_set_zone(zone,
3432 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08003433 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003434 }
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003435 } while (zone_type);
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003436
Christoph Lameter070f8032006-01-06 00:11:19 -08003437 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003438}
3439
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003440
3441/*
3442 * zonelist_order:
3443 * 0 = automatic detection of better ordering.
3444 * 1 = order by ([node] distance, -zonetype)
3445 * 2 = order by (-zonetype, [node] distance)
3446 *
3447 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3448 * the same zonelist. So only NUMA can configure this param.
3449 */
3450#define ZONELIST_ORDER_DEFAULT 0
3451#define ZONELIST_ORDER_NODE 1
3452#define ZONELIST_ORDER_ZONE 2
3453
3454/* zonelist order in the kernel.
3455 * set_zonelist_order() will set this to NODE or ZONE.
3456 */
3457static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
3458static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
3459
3460
Linus Torvalds1da177e2005-04-16 15:20:36 -07003461#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003462/* The value user specified ....changed by config */
3463static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3464/* string for sysctl */
3465#define NUMA_ZONELIST_ORDER_LEN 16
3466char numa_zonelist_order[16] = "default";
3467
3468/*
3469 * interface for configure zonelist ordering.
3470 * command line option "numa_zonelist_order"
3471 * = "[dD]efault - default, automatic configuration.
3472 * = "[nN]ode - order by node locality, then by zone within node
3473 * = "[zZ]one - order by zone, then by locality within zone
3474 */
3475
3476static int __parse_numa_zonelist_order(char *s)
3477{
3478 if (*s == 'd' || *s == 'D') {
3479 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3480 } else if (*s == 'n' || *s == 'N') {
3481 user_zonelist_order = ZONELIST_ORDER_NODE;
3482 } else if (*s == 'z' || *s == 'Z') {
3483 user_zonelist_order = ZONELIST_ORDER_ZONE;
3484 } else {
3485 printk(KERN_WARNING
3486 "Ignoring invalid numa_zonelist_order value: "
3487 "%s\n", s);
3488 return -EINVAL;
3489 }
3490 return 0;
3491}
3492
3493static __init int setup_numa_zonelist_order(char *s)
3494{
Volodymyr G. Lukiianykecb256f2011-01-13 15:46:26 -08003495 int ret;
3496
3497 if (!s)
3498 return 0;
3499
3500 ret = __parse_numa_zonelist_order(s);
3501 if (ret == 0)
3502 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
3503
3504 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003505}
3506early_param("numa_zonelist_order", setup_numa_zonelist_order);
3507
3508/*
3509 * sysctl handler for numa_zonelist_order
3510 */
Joe Perchescccad5b2014-06-06 14:38:09 -07003511int numa_zonelist_order_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003512 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003513 loff_t *ppos)
3514{
3515 char saved_string[NUMA_ZONELIST_ORDER_LEN];
3516 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01003517 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003518
Andi Kleen443c6f12009-12-23 21:00:47 +01003519 mutex_lock(&zl_order_mutex);
Chen Gangdacbde02013-07-03 15:02:35 -07003520 if (write) {
3521 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
3522 ret = -EINVAL;
3523 goto out;
3524 }
3525 strcpy(saved_string, (char *)table->data);
3526 }
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003527 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003528 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01003529 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003530 if (write) {
3531 int oldval = user_zonelist_order;
Chen Gangdacbde02013-07-03 15:02:35 -07003532
3533 ret = __parse_numa_zonelist_order((char *)table->data);
3534 if (ret) {
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003535 /*
3536 * bogus value. restore saved string
3537 */
Chen Gangdacbde02013-07-03 15:02:35 -07003538 strncpy((char *)table->data, saved_string,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003539 NUMA_ZONELIST_ORDER_LEN);
3540 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003541 } else if (oldval != user_zonelist_order) {
3542 mutex_lock(&zonelists_mutex);
Jiang Liu9adb62a2012-07-31 16:43:28 -07003543 build_all_zonelists(NULL, NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003544 mutex_unlock(&zonelists_mutex);
3545 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003546 }
Andi Kleen443c6f12009-12-23 21:00:47 +01003547out:
3548 mutex_unlock(&zl_order_mutex);
3549 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003550}
3551
3552
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003553#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003554static int node_load[MAX_NUMNODES];
3555
Linus Torvalds1da177e2005-04-16 15:20:36 -07003556/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07003557 * 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 -07003558 * @node: node whose fallback list we're appending
3559 * @used_node_mask: nodemask_t of already used nodes
3560 *
3561 * We use a number of factors to determine which is the next node that should
3562 * appear on a given node's fallback list. The node should not have appeared
3563 * already in @node's fallback list, and it should be the next closest node
3564 * according to the distance array (which contains arbitrary distance values
3565 * from each node to each node in the system), and should also prefer nodes
3566 * with no CPUs, since presumably they'll have very little allocation pressure
3567 * on them otherwise.
3568 * It returns -1 if no node is found.
3569 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003570static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003571{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01003572 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003573 int min_val = INT_MAX;
David Rientjes00ef2d22013-02-22 16:35:36 -08003574 int best_node = NUMA_NO_NODE;
Rusty Russella70f7302009-03-13 14:49:46 +10303575 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003576
Linus Torvalds4cf808e2006-02-17 20:38:21 +01003577 /* Use the local node if we haven't already */
3578 if (!node_isset(node, *used_node_mask)) {
3579 node_set(node, *used_node_mask);
3580 return node;
3581 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003582
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08003583 for_each_node_state(n, N_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003584
3585 /* Don't want a node to appear more than once */
3586 if (node_isset(n, *used_node_mask))
3587 continue;
3588
Linus Torvalds1da177e2005-04-16 15:20:36 -07003589 /* Use the distance array to find the distance */
3590 val = node_distance(node, n);
3591
Linus Torvalds4cf808e2006-02-17 20:38:21 +01003592 /* Penalize nodes under us ("prefer the next node") */
3593 val += (n < node);
3594
Linus Torvalds1da177e2005-04-16 15:20:36 -07003595 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10303596 tmp = cpumask_of_node(n);
3597 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003598 val += PENALTY_FOR_NODE_WITH_CPUS;
3599
3600 /* Slight preference for less loaded node */
3601 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
3602 val += node_load[n];
3603
3604 if (val < min_val) {
3605 min_val = val;
3606 best_node = n;
3607 }
3608 }
3609
3610 if (best_node >= 0)
3611 node_set(best_node, *used_node_mask);
3612
3613 return best_node;
3614}
3615
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003616
3617/*
3618 * Build zonelists ordered by node and zones within node.
3619 * This results in maximum locality--normal zone overflows into local
3620 * DMA zone, if any--but risks exhausting DMA zone.
3621 */
3622static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003623{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003624 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003625 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003626
Mel Gorman54a6eb52008-04-28 02:12:16 -07003627 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07003628 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07003629 ;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003630 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gormandd1a2392008-04-28 02:12:17 -07003631 zonelist->_zonerefs[j].zone = NULL;
3632 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003633}
3634
3635/*
Christoph Lameter523b9452007-10-16 01:25:37 -07003636 * Build gfp_thisnode zonelists
3637 */
3638static void build_thisnode_zonelists(pg_data_t *pgdat)
3639{
Christoph Lameter523b9452007-10-16 01:25:37 -07003640 int j;
3641 struct zonelist *zonelist;
3642
Mel Gorman54a6eb52008-04-28 02:12:16 -07003643 zonelist = &pgdat->node_zonelists[1];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003644 j = build_zonelists_node(pgdat, zonelist, 0);
Mel Gormandd1a2392008-04-28 02:12:17 -07003645 zonelist->_zonerefs[j].zone = NULL;
3646 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07003647}
3648
3649/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003650 * Build zonelists ordered by zone and nodes within zones.
3651 * This results in conserving DMA zone[s] until all Normal memory is
3652 * exhausted, but results in overflowing to remote node while memory
3653 * may still exist in local DMA zone.
3654 */
3655static int node_order[MAX_NUMNODES];
3656
3657static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
3658{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003659 int pos, j, node;
3660 int zone_type; /* needs to be signed */
3661 struct zone *z;
3662 struct zonelist *zonelist;
3663
Mel Gorman54a6eb52008-04-28 02:12:16 -07003664 zonelist = &pgdat->node_zonelists[0];
3665 pos = 0;
3666 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
3667 for (j = 0; j < nr_nodes; j++) {
3668 node = node_order[j];
3669 z = &NODE_DATA(node)->node_zones[zone_type];
3670 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07003671 zoneref_set_zone(z,
3672 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07003673 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003674 }
3675 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003676 }
Mel Gormandd1a2392008-04-28 02:12:17 -07003677 zonelist->_zonerefs[pos].zone = NULL;
3678 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003679}
3680
Mel Gorman31939132014-10-09 15:28:30 -07003681#if defined(CONFIG_64BIT)
3682/*
3683 * Devices that require DMA32/DMA are relatively rare and do not justify a
3684 * penalty to every machine in case the specialised case applies. Default
3685 * to Node-ordering on 64-bit NUMA machines
3686 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003687static int default_zonelist_order(void)
3688{
Mel Gorman31939132014-10-09 15:28:30 -07003689 return ZONELIST_ORDER_NODE;
3690}
3691#else
3692/*
3693 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
3694 * by the kernel. If processes running on node 0 deplete the low memory zone
3695 * then reclaim will occur more frequency increasing stalls and potentially
3696 * be easier to OOM if a large percentage of the zone is under writeback or
3697 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
3698 * Hence, default to zone ordering on 32-bit.
3699 */
3700static int default_zonelist_order(void)
3701{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003702 return ZONELIST_ORDER_ZONE;
3703}
Mel Gorman31939132014-10-09 15:28:30 -07003704#endif /* CONFIG_64BIT */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003705
3706static void set_zonelist_order(void)
3707{
3708 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
3709 current_zonelist_order = default_zonelist_order();
3710 else
3711 current_zonelist_order = user_zonelist_order;
3712}
3713
3714static void build_zonelists(pg_data_t *pgdat)
3715{
3716 int j, node, load;
3717 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003718 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003719 int local_node, prev_node;
3720 struct zonelist *zonelist;
3721 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003722
3723 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07003724 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003725 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07003726 zonelist->_zonerefs[0].zone = NULL;
3727 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003728 }
3729
3730 /* NUMA-aware ordering of nodes */
3731 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003732 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003733 prev_node = local_node;
3734 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003735
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003736 memset(node_order, 0, sizeof(node_order));
3737 j = 0;
3738
Linus Torvalds1da177e2005-04-16 15:20:36 -07003739 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
3740 /*
3741 * We don't want to pressure a particular node.
3742 * So adding penalty to the first node in same
3743 * distance group to make it round-robin.
3744 */
David Rientjes957f8222012-10-08 16:33:24 -07003745 if (node_distance(local_node, node) !=
3746 node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003747 node_load[node] = load;
3748
Linus Torvalds1da177e2005-04-16 15:20:36 -07003749 prev_node = node;
3750 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003751 if (order == ZONELIST_ORDER_NODE)
3752 build_zonelists_in_node_order(pgdat, node);
3753 else
3754 node_order[j++] = node; /* remember order */
3755 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003756
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003757 if (order == ZONELIST_ORDER_ZONE) {
3758 /* calculate node order -- i.e., DMA last! */
3759 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003760 }
Christoph Lameter523b9452007-10-16 01:25:37 -07003761
3762 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003763}
3764
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003765/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003766static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003767{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003768 struct zonelist *zonelist;
3769 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07003770 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003771
Mel Gorman54a6eb52008-04-28 02:12:16 -07003772 zonelist = &pgdat->node_zonelists[0];
3773 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3774 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07003775 for (z = zonelist->_zonerefs; z->zone; z++)
3776 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003777}
3778
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003779#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3780/*
3781 * Return node id of node used for "local" allocations.
3782 * I.e., first node id of first zone in arg node's generic zonelist.
3783 * Used for initializing percpu 'numa_mem', which is used primarily
3784 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3785 */
3786int local_memory_node(int node)
3787{
3788 struct zone *zone;
3789
3790 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3791 gfp_zone(GFP_KERNEL),
3792 NULL,
3793 &zone);
3794 return zone->node;
3795}
3796#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003797
Linus Torvalds1da177e2005-04-16 15:20:36 -07003798#else /* CONFIG_NUMA */
3799
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003800static void set_zonelist_order(void)
3801{
3802 current_zonelist_order = ZONELIST_ORDER_ZONE;
3803}
3804
3805static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003806{
Christoph Lameter19655d32006-09-25 23:31:19 -07003807 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003808 enum zone_type j;
3809 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003810
3811 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003812
Mel Gorman54a6eb52008-04-28 02:12:16 -07003813 zonelist = &pgdat->node_zonelists[0];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003814 j = build_zonelists_node(pgdat, zonelist, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003815
Mel Gorman54a6eb52008-04-28 02:12:16 -07003816 /*
3817 * Now we build the zonelist so that it contains the zones
3818 * of all the other nodes.
3819 * We don't want to pressure a particular node, so when
3820 * building the zones for node N, we make sure that the
3821 * zones coming right after the local ones are those from
3822 * node N+1 (modulo N)
3823 */
3824 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3825 if (!node_online(node))
3826 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003827 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003828 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07003829 for (node = 0; node < local_node; node++) {
3830 if (!node_online(node))
3831 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003832 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gorman54a6eb52008-04-28 02:12:16 -07003833 }
3834
Mel Gormandd1a2392008-04-28 02:12:17 -07003835 zonelist->_zonerefs[j].zone = NULL;
3836 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003837}
3838
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003839/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003840static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003841{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003842 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003843}
3844
Linus Torvalds1da177e2005-04-16 15:20:36 -07003845#endif /* CONFIG_NUMA */
3846
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003847/*
3848 * Boot pageset table. One per cpu which is going to be used for all
3849 * zones and all nodes. The parameters will be set in such a way
3850 * that an item put on a list will immediately be handed over to
3851 * the buddy list. This is safe since pageset manipulation is done
3852 * with interrupts disabled.
3853 *
3854 * The boot_pagesets must be kept even after bootup is complete for
3855 * unused processors and/or zones. They do play a role for bootstrapping
3856 * hotplugged processors.
3857 *
3858 * zoneinfo_show() and maybe other functions do
3859 * not check if the processor is online before following the pageset pointer.
3860 * Other parts of the kernel may not check if the zone is available.
3861 */
3862static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3863static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07003864static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003865
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003866/*
3867 * Global mutex to protect against size modification of zonelists
3868 * as well as to serialize pageset setup for the new populated zone.
3869 */
3870DEFINE_MUTEX(zonelists_mutex);
3871
Rusty Russell9b1a4d32008-07-28 12:16:30 -05003872/* return values int ....just for stop_machine() */
Jiang Liu4ed7e022012-07-31 16:43:35 -07003873static int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003874{
Yasunori Goto68113782006-06-23 02:03:11 -07003875 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003876 int cpu;
Jiang Liu9adb62a2012-07-31 16:43:28 -07003877 pg_data_t *self = data;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003878
Bo Liu7f9cfb32009-08-18 14:11:19 -07003879#ifdef CONFIG_NUMA
3880 memset(node_load, 0, sizeof(node_load));
3881#endif
Jiang Liu9adb62a2012-07-31 16:43:28 -07003882
3883 if (self && !node_online(self->node_id)) {
3884 build_zonelists(self);
3885 build_zonelist_cache(self);
3886 }
3887
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003888 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07003889 pg_data_t *pgdat = NODE_DATA(nid);
3890
3891 build_zonelists(pgdat);
3892 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003893 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003894
3895 /*
3896 * Initialize the boot_pagesets that are going to be used
3897 * for bootstrapping processors. The real pagesets for
3898 * each zone will be allocated later when the per cpu
3899 * allocator is available.
3900 *
3901 * boot_pagesets are used also for bootstrapping offline
3902 * cpus if the system is already booted because the pagesets
3903 * are needed to initialize allocators on a specific cpu too.
3904 * F.e. the percpu allocator needs the page allocator which
3905 * needs the percpu allocator in order to allocate its pagesets
3906 * (a chicken-egg dilemma).
3907 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003908 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003909 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3910
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003911#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3912 /*
3913 * We now know the "local memory node" for each node--
3914 * i.e., the node of the first zone in the generic zonelist.
3915 * Set up numa_mem percpu variable for on-line cpus. During
3916 * boot, only the boot cpu should be on-line; we'll init the
3917 * secondary cpus' numa_mem as they come on-line. During
3918 * node/memory hotplug, we'll fixup all on-line cpus.
3919 */
3920 if (cpu_online(cpu))
3921 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3922#endif
3923 }
3924
Yasunori Goto68113782006-06-23 02:03:11 -07003925 return 0;
3926}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003927
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08003928static noinline void __init
3929build_all_zonelists_init(void)
3930{
3931 __build_all_zonelists(NULL);
3932 mminit_verify_zonelist();
3933 cpuset_init_current_mems_allowed();
3934}
3935
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003936/*
3937 * Called with zonelists_mutex held always
3938 * unless system_state == SYSTEM_BOOTING.
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08003939 *
3940 * __ref due to (1) call of __meminit annotated setup_zone_pageset
3941 * [we're only called with non-NULL zone through __meminit paths] and
3942 * (2) call of __init annotated helper build_all_zonelists_init
3943 * [protected by SYSTEM_BOOTING].
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003944 */
Jiang Liu9adb62a2012-07-31 16:43:28 -07003945void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
Yasunori Goto68113782006-06-23 02:03:11 -07003946{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003947 set_zonelist_order();
3948
Yasunori Goto68113782006-06-23 02:03:11 -07003949 if (system_state == SYSTEM_BOOTING) {
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08003950 build_all_zonelists_init();
Yasunori Goto68113782006-06-23 02:03:11 -07003951 } else {
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003952#ifdef CONFIG_MEMORY_HOTPLUG
Jiang Liu9adb62a2012-07-31 16:43:28 -07003953 if (zone)
3954 setup_zone_pageset(zone);
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003955#endif
Cody P Schaferdd1895e2013-07-03 15:01:36 -07003956 /* we have to stop all cpus to guarantee there is no user
3957 of zonelist */
Jiang Liu9adb62a2012-07-31 16:43:28 -07003958 stop_machine(__build_all_zonelists, pgdat, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07003959 /* cpuset refresh routine should be here */
3960 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07003961 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003962 /*
3963 * Disable grouping by mobility if the number of pages in the
3964 * system is too low to allow the mechanism to work. It would be
3965 * more accurate, but expensive to check per-zone. This check is
3966 * made on memory-hotadd so a system can start with mobility
3967 * disabled and enable it later
3968 */
Mel Gormand9c23402007-10-16 01:26:01 -07003969 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003970 page_group_by_mobility_disabled = 1;
3971 else
3972 page_group_by_mobility_disabled = 0;
3973
Anton Blanchardf88dfff2014-12-10 15:42:53 -08003974 pr_info("Built %i zonelists in %s order, mobility grouping %s. "
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003975 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003976 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003977 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003978 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003979 vm_total_pages);
3980#ifdef CONFIG_NUMA
Anton Blanchardf88dfff2014-12-10 15:42:53 -08003981 pr_info("Policy zone: %s\n", zone_names[policy_zone]);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003982#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003983}
3984
3985/*
3986 * Helper functions to size the waitqueue hash table.
3987 * Essentially these want to choose hash table sizes sufficiently
3988 * large so that collisions trying to wait on pages are rare.
3989 * But in fact, the number of active page waitqueues on typical
3990 * systems is ridiculously low, less than 200. So this is even
3991 * conservative, even though it seems large.
3992 *
3993 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3994 * waitqueues, i.e. the size of the waitq table given the number of pages.
3995 */
3996#define PAGES_PER_WAITQUEUE 256
3997
Yasunori Gotocca448f2006-06-23 02:03:10 -07003998#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07003999static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004000{
4001 unsigned long size = 1;
4002
4003 pages /= PAGES_PER_WAITQUEUE;
4004
4005 while (size < pages)
4006 size <<= 1;
4007
4008 /*
4009 * Once we have dozens or even hundreds of threads sleeping
4010 * on IO we've got bigger problems than wait queue collision.
4011 * Limit the size of the wait table to a reasonable size.
4012 */
4013 size = min(size, 4096UL);
4014
4015 return max(size, 4UL);
4016}
Yasunori Gotocca448f2006-06-23 02:03:10 -07004017#else
4018/*
4019 * A zone's size might be changed by hot-add, so it is not possible to determine
4020 * a suitable size for its wait_table. So we use the maximum size now.
4021 *
4022 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
4023 *
4024 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
4025 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
4026 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
4027 *
4028 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
4029 * or more by the traditional way. (See above). It equals:
4030 *
4031 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
4032 * ia64(16K page size) : = ( 8G + 4M)byte.
4033 * powerpc (64K page size) : = (32G +16M)byte.
4034 */
4035static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
4036{
4037 return 4096UL;
4038}
4039#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004040
4041/*
4042 * This is an integer logarithm so that shifts can be used later
4043 * to extract the more random high bits from the multiplicative
4044 * hash function before the remainder is taken.
4045 */
4046static inline unsigned long wait_table_bits(unsigned long size)
4047{
4048 return ffz(~size);
4049}
4050
Mel Gorman56fd56b2007-10-16 01:25:58 -07004051/*
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07004052 * Check if a pageblock contains reserved pages
4053 */
4054static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
4055{
4056 unsigned long pfn;
4057
4058 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
4059 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
4060 return 1;
4061 }
4062 return 0;
4063}
4064
4065/*
Mel Gormand9c23402007-10-16 01:26:01 -07004066 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07004067 * of blocks reserved is based on min_wmark_pages(zone). The memory within
4068 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07004069 * higher will lead to a bigger reserve which will get freed as contiguous
4070 * blocks as reclaim kicks in
4071 */
4072static void setup_zone_migrate_reserve(struct zone *zone)
4073{
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07004074 unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004075 struct page *page;
Mel Gorman78986a62009-09-21 17:03:02 -07004076 unsigned long block_migratetype;
4077 int reserve;
Yasuaki Ishimatsu943dca12014-01-21 15:49:06 -08004078 int old_reserve;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004079
Michal Hockod02156382011-12-08 14:34:27 -08004080 /*
4081 * Get the start pfn, end pfn and the number of blocks to reserve
4082 * We have to be careful to be aligned to pageblock_nr_pages to
4083 * make sure that we always check pfn_valid for the first page in
4084 * the block.
4085 */
Mel Gorman56fd56b2007-10-16 01:25:58 -07004086 start_pfn = zone->zone_start_pfn;
Cody P Schafer108bcc92013-02-22 16:35:23 -08004087 end_pfn = zone_end_pfn(zone);
Michal Hockod02156382011-12-08 14:34:27 -08004088 start_pfn = roundup(start_pfn, pageblock_nr_pages);
Mel Gorman41858962009-06-16 15:32:12 -07004089 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07004090 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004091
Mel Gorman78986a62009-09-21 17:03:02 -07004092 /*
4093 * Reserve blocks are generally in place to help high-order atomic
4094 * allocations that are short-lived. A min_free_kbytes value that
4095 * would result in more than 2 reserve blocks for atomic allocations
4096 * is assumed to be in place to help anti-fragmentation for the
4097 * future allocation of hugepages at runtime.
4098 */
4099 reserve = min(2, reserve);
Yasuaki Ishimatsu943dca12014-01-21 15:49:06 -08004100 old_reserve = zone->nr_migrate_reserve_block;
4101
4102 /* When memory hot-add, we almost always need to do nothing */
4103 if (reserve == old_reserve)
4104 return;
4105 zone->nr_migrate_reserve_block = reserve;
Mel Gorman78986a62009-09-21 17:03:02 -07004106
Mel Gormand9c23402007-10-16 01:26:01 -07004107 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07004108 if (!pfn_valid(pfn))
4109 continue;
4110 page = pfn_to_page(pfn);
4111
Adam Litke344c7902008-09-02 14:35:38 -07004112 /* Watch out for overlapping nodes */
4113 if (page_to_nid(page) != zone_to_nid(zone))
4114 continue;
4115
Mel Gorman56fd56b2007-10-16 01:25:58 -07004116 block_migratetype = get_pageblock_migratetype(page);
4117
Mel Gorman938929f2012-01-10 15:07:14 -08004118 /* Only test what is necessary when the reserves are not met */
4119 if (reserve > 0) {
4120 /*
4121 * Blocks with reserved pages will never free, skip
4122 * them.
4123 */
4124 block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
4125 if (pageblock_is_reserved(pfn, block_end_pfn))
4126 continue;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004127
Mel Gorman938929f2012-01-10 15:07:14 -08004128 /* If this block is reserved, account for it */
4129 if (block_migratetype == MIGRATE_RESERVE) {
4130 reserve--;
4131 continue;
4132 }
4133
4134 /* Suitable for reserving if this block is movable */
4135 if (block_migratetype == MIGRATE_MOVABLE) {
4136 set_pageblock_migratetype(page,
4137 MIGRATE_RESERVE);
4138 move_freepages_block(zone, page,
4139 MIGRATE_RESERVE);
4140 reserve--;
4141 continue;
4142 }
Yasuaki Ishimatsu943dca12014-01-21 15:49:06 -08004143 } else if (!old_reserve) {
4144 /*
4145 * At boot time we don't need to scan the whole zone
4146 * for turning off MIGRATE_RESERVE.
4147 */
4148 break;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004149 }
4150
4151 /*
4152 * If the reserve is met and this is a previous reserved block,
4153 * take it back
4154 */
4155 if (block_migratetype == MIGRATE_RESERVE) {
4156 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4157 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4158 }
4159 }
4160}
Mel Gormanac0e5b72007-10-16 01:25:58 -07004161
Linus Torvalds1da177e2005-04-16 15:20:36 -07004162/*
4163 * Initially all pages are reserved - free ones are freed
4164 * up by free_all_bootmem() once the early boot process is
4165 * done. Non-atomic initialization, single-pass.
4166 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01004167void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08004168 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004169{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004170 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07004171 unsigned long end_pfn = start_pfn + size;
4172 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004173 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004174
Hugh Dickins22b31ee2009-01-06 14:40:09 -08004175 if (highest_memmap_pfn < end_pfn - 1)
4176 highest_memmap_pfn = end_pfn - 1;
4177
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004178 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08004179 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08004180 /*
4181 * There can be holes in boot-time mem_map[]s
4182 * handed to this function. They do not
4183 * exist on hotplugged memory.
4184 */
4185 if (context == MEMMAP_EARLY) {
4186 if (!early_pfn_valid(pfn))
4187 continue;
4188 if (!early_pfn_in_nid(pfn, nid))
4189 continue;
4190 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004191 page = pfn_to_page(pfn);
4192 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07004193 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08004194 init_page_count(page);
Mel Gorman22b751c2013-02-22 16:34:59 -08004195 page_mapcount_reset(page);
Peter Zijlstra90572892013-10-07 11:29:20 +01004196 page_cpupid_reset_last(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004197 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004198 /*
4199 * Mark the block movable so that blocks are reserved for
4200 * movable at startup. This will force kernel allocations
4201 * to reserve their blocks rather than leaking throughout
4202 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07004203 * kernel allocations are made. Later some blocks near
4204 * the start are marked MIGRATE_RESERVE by
4205 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004206 *
4207 * bitmap is created for zone's valid pfn range. but memmap
4208 * can be created for invalid pages (for alignment)
4209 * check here not to call set_pageblock_migratetype() against
4210 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004211 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004212 if ((z->zone_start_pfn <= pfn)
Cody P Schafer108bcc92013-02-22 16:35:23 -08004213 && (pfn < zone_end_pfn(z))
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004214 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07004215 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004216
Linus Torvalds1da177e2005-04-16 15:20:36 -07004217 INIT_LIST_HEAD(&page->lru);
4218#ifdef WANT_PAGE_VIRTUAL
4219 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
4220 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07004221 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004222#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004223 }
4224}
4225
Andi Kleen1e548de2008-02-04 22:29:26 -08004226static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004227{
Mel Gorman7aeb09f2014-06-04 16:10:21 -07004228 unsigned int order, t;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004229 for_each_migratetype_order(order, t) {
4230 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004231 zone->free_area[order].nr_free = 0;
4232 }
4233}
4234
4235#ifndef __HAVE_ARCH_MEMMAP_INIT
4236#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08004237 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004238#endif
4239
David Rientjes7cd2b0a2014-06-23 13:22:04 -07004240static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004241{
David Howells3a6be872009-05-06 16:03:03 -07004242#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004243 int batch;
4244
4245 /*
4246 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07004247 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004248 *
4249 * OK, so we don't know how big the cache is. So guess.
4250 */
Jiang Liub40da042013-02-22 16:33:52 -08004251 batch = zone->managed_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07004252 if (batch * PAGE_SIZE > 512 * 1024)
4253 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004254 batch /= 4; /* We effectively *= 4 below */
4255 if (batch < 1)
4256 batch = 1;
4257
4258 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004259 * Clamp the batch to a 2^n - 1 value. Having a power
4260 * of 2 value was found to be more likely to have
4261 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004262 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004263 * For example if 2 tasks are alternately allocating
4264 * batches of pages, one task can end up with a lot
4265 * of pages of one half of the possible page colors
4266 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004267 */
David Howells91552032009-05-06 16:03:02 -07004268 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07004269
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004270 return batch;
David Howells3a6be872009-05-06 16:03:03 -07004271
4272#else
4273 /* The deferral and batching of frees should be suppressed under NOMMU
4274 * conditions.
4275 *
4276 * The problem is that NOMMU needs to be able to allocate large chunks
4277 * of contiguous memory as there's no hardware page translation to
4278 * assemble apparent contiguous memory from discontiguous pages.
4279 *
4280 * Queueing large contiguous runs of pages for batching, however,
4281 * causes the pages to actually be freed in smaller chunks. As there
4282 * can be a significant delay between the individual batches being
4283 * recycled, this leads to the once large chunks of space being
4284 * fragmented and becoming unavailable for high-order allocations.
4285 */
4286 return 0;
4287#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004288}
4289
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004290/*
4291 * pcp->high and pcp->batch values are related and dependent on one another:
4292 * ->batch must never be higher then ->high.
4293 * The following function updates them in a safe manner without read side
4294 * locking.
4295 *
4296 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4297 * those fields changing asynchronously (acording the the above rule).
4298 *
4299 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4300 * outside of boot time (or some other assurance that no concurrent updaters
4301 * exist).
4302 */
4303static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
4304 unsigned long batch)
4305{
4306 /* start with a fail safe value for batch */
4307 pcp->batch = 1;
4308 smp_wmb();
4309
4310 /* Update high, then batch, in order */
4311 pcp->high = high;
4312 smp_wmb();
4313
4314 pcp->batch = batch;
4315}
4316
Cody P Schafer36640332013-07-03 15:01:40 -07004317/* a companion to pageset_set_high() */
Cody P Schafer4008bab2013-07-03 15:01:28 -07004318static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
4319{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004320 pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
Cody P Schafer4008bab2013-07-03 15:01:28 -07004321}
4322
Cody P Schafer88c90db2013-07-03 15:01:35 -07004323static void pageset_init(struct per_cpu_pageset *p)
Christoph Lameter2caaad42005-06-21 17:15:00 -07004324{
4325 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004326 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004327
Magnus Damm1c6fe942005-10-26 01:58:59 -07004328 memset(p, 0, sizeof(*p));
4329
Christoph Lameter3dfa5722008-02-04 22:29:19 -08004330 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004331 pcp->count = 0;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004332 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
4333 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07004334}
4335
Cody P Schafer88c90db2013-07-03 15:01:35 -07004336static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
4337{
4338 pageset_init(p);
4339 pageset_set_batch(p, batch);
4340}
4341
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004342/*
Cody P Schafer36640332013-07-03 15:01:40 -07004343 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004344 * to the value high for the pageset p.
4345 */
Cody P Schafer36640332013-07-03 15:01:40 -07004346static void pageset_set_high(struct per_cpu_pageset *p,
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004347 unsigned long high)
4348{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004349 unsigned long batch = max(1UL, high / 4);
4350 if ((high / 4) > (PAGE_SHIFT * 8))
4351 batch = PAGE_SHIFT * 8;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004352
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004353 pageset_update(&p->pcp, high, batch);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004354}
4355
David Rientjes7cd2b0a2014-06-23 13:22:04 -07004356static void pageset_set_high_and_batch(struct zone *zone,
4357 struct per_cpu_pageset *pcp)
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004358{
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004359 if (percpu_pagelist_fraction)
Cody P Schafer36640332013-07-03 15:01:40 -07004360 pageset_set_high(pcp,
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004361 (zone->managed_pages /
4362 percpu_pagelist_fraction));
4363 else
4364 pageset_set_batch(pcp, zone_batchsize(zone));
4365}
4366
Cody P Schafer169f6c12013-07-03 15:01:41 -07004367static void __meminit zone_pageset_init(struct zone *zone, int cpu)
4368{
4369 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4370
4371 pageset_init(pcp);
4372 pageset_set_high_and_batch(zone, pcp);
4373}
4374
Jiang Liu4ed7e022012-07-31 16:43:35 -07004375static void __meminit setup_zone_pageset(struct zone *zone)
Wu Fengguang319774e2010-05-24 14:32:49 -07004376{
4377 int cpu;
Wu Fengguang319774e2010-05-24 14:32:49 -07004378 zone->pageset = alloc_percpu(struct per_cpu_pageset);
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004379 for_each_possible_cpu(cpu)
4380 zone_pageset_init(zone, cpu);
Wu Fengguang319774e2010-05-24 14:32:49 -07004381}
4382
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004383/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004384 * Allocate per cpu pagesets and initialize them.
4385 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07004386 */
Al Viro78d99552005-12-15 09:18:25 +00004387void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004388{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004389 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004390
Wu Fengguang319774e2010-05-24 14:32:49 -07004391 for_each_populated_zone(zone)
4392 setup_zone_pageset(zone);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004393}
4394
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004395static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07004396int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07004397{
4398 int i;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004399 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07004400
4401 /*
4402 * The per-page waitqueue mechanism uses hashed waitqueues
4403 * per zone.
4404 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07004405 zone->wait_table_hash_nr_entries =
4406 wait_table_hash_nr_entries(zone_size_pages);
4407 zone->wait_table_bits =
4408 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004409 alloc_size = zone->wait_table_hash_nr_entries
4410 * sizeof(wait_queue_head_t);
4411
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07004412 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07004413 zone->wait_table = (wait_queue_head_t *)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004414 memblock_virt_alloc_node_nopanic(
4415 alloc_size, zone->zone_pgdat->node_id);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004416 } else {
4417 /*
4418 * This case means that a zone whose size was 0 gets new memory
4419 * via memory hot-add.
4420 * But it may be the case that a new node was hot-added. In
4421 * this case vmalloc() will not be able to use this new node's
4422 * memory - this wait_table must be initialized to use this new
4423 * node itself as well.
4424 * To use this new node's memory, further consideration will be
4425 * necessary.
4426 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07004427 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004428 }
4429 if (!zone->wait_table)
4430 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07004431
Pintu Kumarb8af2942013-09-11 14:20:34 -07004432 for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07004433 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004434
4435 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004436}
4437
Matt Tolentinoc09b4242006-01-17 07:03:44 +01004438static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07004439{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004440 /*
4441 * per cpu subsystem is not up at this point. The following code
4442 * relies on the ability of the linker to provide the
4443 * offset of a (static) per cpu variable into the per cpu area.
4444 */
4445 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07004446
Xishi Qiub38a8722013-11-12 15:07:20 -08004447 if (populated_zone(zone))
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004448 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
4449 zone->name, zone->present_pages,
4450 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07004451}
4452
Jiang Liu4ed7e022012-07-31 16:43:35 -07004453int __meminit init_currently_empty_zone(struct zone *zone,
Yasunori Goto718127c2006-06-23 02:03:10 -07004454 unsigned long zone_start_pfn,
Dave Hansena2f3aa022007-01-10 23:15:30 -08004455 unsigned long size,
4456 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07004457{
4458 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004459 int ret;
4460 ret = zone_wait_table_init(zone, size);
4461 if (ret)
4462 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07004463 pgdat->nr_zones = zone_idx(zone) + 1;
4464
Dave Hansened8ece22005-10-29 18:16:50 -07004465 zone->zone_start_pfn = zone_start_pfn;
4466
Mel Gorman708614e2008-07-23 21:26:51 -07004467 mminit_dprintk(MMINIT_TRACE, "memmap_init",
4468 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4469 pgdat->node_id,
4470 (unsigned long)zone_idx(zone),
4471 zone_start_pfn, (zone_start_pfn + size));
4472
Andi Kleen1e548de2008-02-04 22:29:26 -08004473 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07004474
4475 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004476}
4477
Tejun Heo0ee332c2011-12-08 10:22:09 -08004478#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07004479#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4480/*
4481 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
Mel Gormanc7132162006-09-27 01:49:43 -07004482 */
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004483int __meminit __early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004484{
Tejun Heoc13291a2011-07-12 10:46:30 +02004485 unsigned long start_pfn, end_pfn;
Yinghai Lue76b63f2013-09-11 14:22:17 -07004486 int nid;
Russ Anderson7c243c72013-04-29 15:07:59 -07004487 /*
4488 * NOTE: The following SMP-unsafe globals are only used early in boot
4489 * when the kernel is running single-threaded.
4490 */
4491 static unsigned long __meminitdata last_start_pfn, last_end_pfn;
4492 static int __meminitdata last_nid;
4493
4494 if (last_start_pfn <= pfn && pfn < last_end_pfn)
4495 return last_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004496
Yinghai Lue76b63f2013-09-11 14:22:17 -07004497 nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
4498 if (nid != -1) {
4499 last_start_pfn = start_pfn;
4500 last_end_pfn = end_pfn;
4501 last_nid = nid;
4502 }
4503
4504 return nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004505}
4506#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4507
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004508int __meminit early_pfn_to_nid(unsigned long pfn)
4509{
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08004510 int nid;
4511
4512 nid = __early_pfn_to_nid(pfn);
4513 if (nid >= 0)
4514 return nid;
4515 /* just returns 0 */
4516 return 0;
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004517}
4518
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08004519#ifdef CONFIG_NODES_SPAN_OTHER_NODES
4520bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
4521{
4522 int nid;
4523
4524 nid = __early_pfn_to_nid(pfn);
4525 if (nid >= 0 && nid != node)
4526 return false;
4527 return true;
4528}
4529#endif
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004530
Mel Gormanc7132162006-09-27 01:49:43 -07004531/**
Santosh Shilimkar67828322014-01-21 15:50:25 -08004532 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004533 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
Santosh Shilimkar67828322014-01-21 15:50:25 -08004534 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
Mel Gormanc7132162006-09-27 01:49:43 -07004535 *
Zhang Zhen7d018172014-06-04 16:10:53 -07004536 * If an architecture guarantees that all ranges registered contain no holes
4537 * and may be freed, this this function may be used instead of calling
4538 * memblock_free_early_nid() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07004539 */
Tejun Heoc13291a2011-07-12 10:46:30 +02004540void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004541{
Tejun Heoc13291a2011-07-12 10:46:30 +02004542 unsigned long start_pfn, end_pfn;
4543 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004544
Tejun Heoc13291a2011-07-12 10:46:30 +02004545 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
4546 start_pfn = min(start_pfn, max_low_pfn);
4547 end_pfn = min(end_pfn, max_low_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004548
Tejun Heoc13291a2011-07-12 10:46:30 +02004549 if (start_pfn < end_pfn)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004550 memblock_free_early_nid(PFN_PHYS(start_pfn),
4551 (end_pfn - start_pfn) << PAGE_SHIFT,
4552 this_nid);
Mel Gormanc7132162006-09-27 01:49:43 -07004553 }
4554}
4555
4556/**
4557 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004558 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07004559 *
Zhang Zhen7d018172014-06-04 16:10:53 -07004560 * If an architecture guarantees that all ranges registered contain no holes and may
4561 * be freed, this function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07004562 */
4563void __init sparse_memory_present_with_active_regions(int nid)
4564{
Tejun Heoc13291a2011-07-12 10:46:30 +02004565 unsigned long start_pfn, end_pfn;
4566 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004567
Tejun Heoc13291a2011-07-12 10:46:30 +02004568 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
4569 memory_present(this_nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004570}
4571
4572/**
4573 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004574 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4575 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4576 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07004577 *
4578 * It returns the start and end page frame of a node based on information
Zhang Zhen7d018172014-06-04 16:10:53 -07004579 * provided by memblock_set_node(). If called for a node
Mel Gormanc7132162006-09-27 01:49:43 -07004580 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004581 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07004582 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004583void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004584 unsigned long *start_pfn, unsigned long *end_pfn)
4585{
Tejun Heoc13291a2011-07-12 10:46:30 +02004586 unsigned long this_start_pfn, this_end_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004587 int i;
Tejun Heoc13291a2011-07-12 10:46:30 +02004588
Mel Gormanc7132162006-09-27 01:49:43 -07004589 *start_pfn = -1UL;
4590 *end_pfn = 0;
4591
Tejun Heoc13291a2011-07-12 10:46:30 +02004592 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
4593 *start_pfn = min(*start_pfn, this_start_pfn);
4594 *end_pfn = max(*end_pfn, this_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004595 }
4596
Christoph Lameter633c0662007-10-16 01:25:37 -07004597 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07004598 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07004599}
4600
4601/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07004602 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4603 * assumption is made that zones within a node are ordered in monotonic
4604 * increasing memory addresses so that the "highest" populated zone is used
4605 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004606static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004607{
4608 int zone_index;
4609 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
4610 if (zone_index == ZONE_MOVABLE)
4611 continue;
4612
4613 if (arch_zone_highest_possible_pfn[zone_index] >
4614 arch_zone_lowest_possible_pfn[zone_index])
4615 break;
4616 }
4617
4618 VM_BUG_ON(zone_index == -1);
4619 movable_zone = zone_index;
4620}
4621
4622/*
4623 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004624 * because it is sized independent of architecture. Unlike the other zones,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004625 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4626 * in each node depending on the size of each node and how evenly kernelcore
4627 * is distributed. This helper function adjusts the zone ranges
4628 * provided by the architecture for a given node by using the end of the
4629 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4630 * zones within a node are in order of monotonic increases memory addresses
4631 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004632static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004633 unsigned long zone_type,
4634 unsigned long node_start_pfn,
4635 unsigned long node_end_pfn,
4636 unsigned long *zone_start_pfn,
4637 unsigned long *zone_end_pfn)
4638{
4639 /* Only adjust if ZONE_MOVABLE is on this node */
4640 if (zone_movable_pfn[nid]) {
4641 /* Size ZONE_MOVABLE */
4642 if (zone_type == ZONE_MOVABLE) {
4643 *zone_start_pfn = zone_movable_pfn[nid];
4644 *zone_end_pfn = min(node_end_pfn,
4645 arch_zone_highest_possible_pfn[movable_zone]);
4646
4647 /* Adjust for ZONE_MOVABLE starting within this range */
4648 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
4649 *zone_end_pfn > zone_movable_pfn[nid]) {
4650 *zone_end_pfn = zone_movable_pfn[nid];
4651
4652 /* Check if this whole range is within ZONE_MOVABLE */
4653 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
4654 *zone_start_pfn = *zone_end_pfn;
4655 }
4656}
4657
4658/*
Mel Gormanc7132162006-09-27 01:49:43 -07004659 * Return the number of pages a zone spans in a node, including holes
4660 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4661 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004662static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004663 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004664 unsigned long node_start_pfn,
4665 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004666 unsigned long *ignored)
4667{
Mel Gormanc7132162006-09-27 01:49:43 -07004668 unsigned long zone_start_pfn, zone_end_pfn;
4669
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004670 /* Get the start and end of the zone */
Mel Gormanc7132162006-09-27 01:49:43 -07004671 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4672 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07004673 adjust_zone_range_for_zone_movable(nid, zone_type,
4674 node_start_pfn, node_end_pfn,
4675 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004676
4677 /* Check that this node has pages within the zone's required range */
4678 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4679 return 0;
4680
4681 /* Move the zone boundaries inside the node if necessary */
4682 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4683 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4684
4685 /* Return the spanned pages */
4686 return zone_end_pfn - zone_start_pfn;
4687}
4688
4689/*
4690 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004691 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07004692 */
Yinghai Lu32996252009-12-15 17:59:02 -08004693unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004694 unsigned long range_start_pfn,
4695 unsigned long range_end_pfn)
4696{
Tejun Heo96e907d2011-07-12 10:46:29 +02004697 unsigned long nr_absent = range_end_pfn - range_start_pfn;
4698 unsigned long start_pfn, end_pfn;
4699 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004700
Tejun Heo96e907d2011-07-12 10:46:29 +02004701 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
4702 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
4703 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
4704 nr_absent -= end_pfn - start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004705 }
Tejun Heo96e907d2011-07-12 10:46:29 +02004706 return nr_absent;
Mel Gormanc7132162006-09-27 01:49:43 -07004707}
4708
4709/**
4710 * absent_pages_in_range - Return number of page frames in holes within a range
4711 * @start_pfn: The start PFN to start searching for holes
4712 * @end_pfn: The end PFN to stop searching for holes
4713 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004714 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07004715 */
4716unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4717 unsigned long end_pfn)
4718{
4719 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4720}
4721
4722/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004723static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004724 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004725 unsigned long node_start_pfn,
4726 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004727 unsigned long *ignored)
4728{
Tejun Heo96e907d2011-07-12 10:46:29 +02004729 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
4730 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman9c7cd682006-09-27 01:49:58 -07004731 unsigned long zone_start_pfn, zone_end_pfn;
4732
Tejun Heo96e907d2011-07-12 10:46:29 +02004733 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
4734 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004735
Mel Gorman2a1e2742007-07-17 04:03:12 -07004736 adjust_zone_range_for_zone_movable(nid, zone_type,
4737 node_start_pfn, node_end_pfn,
4738 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004739 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004740}
Mel Gorman0e0b8642006-09-27 01:49:56 -07004741
Tejun Heo0ee332c2011-12-08 10:22:09 -08004742#else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004743static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004744 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004745 unsigned long node_start_pfn,
4746 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004747 unsigned long *zones_size)
4748{
4749 return zones_size[zone_type];
4750}
4751
Paul Mundt6ea6e682007-07-15 23:38:20 -07004752static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004753 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004754 unsigned long node_start_pfn,
4755 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004756 unsigned long *zholes_size)
4757{
4758 if (!zholes_size)
4759 return 0;
4760
4761 return zholes_size[zone_type];
4762}
Yinghai Lu20e69262013-03-01 14:51:27 -08004763
Tejun Heo0ee332c2011-12-08 10:22:09 -08004764#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07004765
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004766static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004767 unsigned long node_start_pfn,
4768 unsigned long node_end_pfn,
4769 unsigned long *zones_size,
4770 unsigned long *zholes_size)
Mel Gormanc7132162006-09-27 01:49:43 -07004771{
4772 unsigned long realtotalpages, totalpages = 0;
4773 enum zone_type i;
4774
4775 for (i = 0; i < MAX_NR_ZONES; i++)
4776 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004777 node_start_pfn,
4778 node_end_pfn,
4779 zones_size);
Mel Gormanc7132162006-09-27 01:49:43 -07004780 pgdat->node_spanned_pages = totalpages;
4781
4782 realtotalpages = totalpages;
4783 for (i = 0; i < MAX_NR_ZONES; i++)
4784 realtotalpages -=
4785 zone_absent_pages_in_node(pgdat->node_id, i,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004786 node_start_pfn, node_end_pfn,
4787 zholes_size);
Mel Gormanc7132162006-09-27 01:49:43 -07004788 pgdat->node_present_pages = realtotalpages;
4789 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4790 realtotalpages);
4791}
4792
Mel Gorman835c1342007-10-16 01:25:47 -07004793#ifndef CONFIG_SPARSEMEM
4794/*
4795 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07004796 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4797 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07004798 * round what is now in bits to nearest long in bits, then return it in
4799 * bytes.
4800 */
Linus Torvalds7c455122013-02-18 09:58:02 -08004801static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07004802{
4803 unsigned long usemapsize;
4804
Linus Torvalds7c455122013-02-18 09:58:02 -08004805 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
Mel Gormand9c23402007-10-16 01:26:01 -07004806 usemapsize = roundup(zonesize, pageblock_nr_pages);
4807 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07004808 usemapsize *= NR_PAGEBLOCK_BITS;
4809 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4810
4811 return usemapsize / 8;
4812}
4813
4814static void __init setup_usemap(struct pglist_data *pgdat,
Linus Torvalds7c455122013-02-18 09:58:02 -08004815 struct zone *zone,
4816 unsigned long zone_start_pfn,
4817 unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07004818{
Linus Torvalds7c455122013-02-18 09:58:02 -08004819 unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
Mel Gorman835c1342007-10-16 01:25:47 -07004820 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08004821 if (usemapsize)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004822 zone->pageblock_flags =
4823 memblock_virt_alloc_node_nopanic(usemapsize,
4824 pgdat->node_id);
Mel Gorman835c1342007-10-16 01:25:47 -07004825}
4826#else
Linus Torvalds7c455122013-02-18 09:58:02 -08004827static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
4828 unsigned long zone_start_pfn, unsigned long zonesize) {}
Mel Gorman835c1342007-10-16 01:25:47 -07004829#endif /* CONFIG_SPARSEMEM */
4830
Mel Gormand9c23402007-10-16 01:26:01 -07004831#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08004832
Mel Gormand9c23402007-10-16 01:26:01 -07004833/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
Chen Gang15ca2202013-09-11 14:20:27 -07004834void __paginginit set_pageblock_order(void)
Mel Gormand9c23402007-10-16 01:26:01 -07004835{
Andrew Morton955c1cd2012-05-29 15:06:31 -07004836 unsigned int order;
4837
Mel Gormand9c23402007-10-16 01:26:01 -07004838 /* Check that pageblock_nr_pages has not already been setup */
4839 if (pageblock_order)
4840 return;
4841
Andrew Morton955c1cd2012-05-29 15:06:31 -07004842 if (HPAGE_SHIFT > PAGE_SHIFT)
4843 order = HUGETLB_PAGE_ORDER;
4844 else
4845 order = MAX_ORDER - 1;
4846
Mel Gormand9c23402007-10-16 01:26:01 -07004847 /*
4848 * Assume the largest contiguous order of interest is a huge page.
Andrew Morton955c1cd2012-05-29 15:06:31 -07004849 * This value may be variable depending on boot parameters on IA64 and
4850 * powerpc.
Mel Gormand9c23402007-10-16 01:26:01 -07004851 */
4852 pageblock_order = order;
4853}
4854#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4855
Mel Gormanba72cb82007-11-28 16:21:13 -08004856/*
4857 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
Andrew Morton955c1cd2012-05-29 15:06:31 -07004858 * is unused as pageblock_order is set at compile-time. See
4859 * include/linux/pageblock-flags.h for the values of pageblock_order based on
4860 * the kernel config
Mel Gormanba72cb82007-11-28 16:21:13 -08004861 */
Chen Gang15ca2202013-09-11 14:20:27 -07004862void __paginginit set_pageblock_order(void)
Mel Gormanba72cb82007-11-28 16:21:13 -08004863{
Mel Gormanba72cb82007-11-28 16:21:13 -08004864}
Mel Gormand9c23402007-10-16 01:26:01 -07004865
4866#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4867
Jiang Liu01cefae2012-12-12 13:52:19 -08004868static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
4869 unsigned long present_pages)
4870{
4871 unsigned long pages = spanned_pages;
4872
4873 /*
4874 * Provide a more accurate estimation if there are holes within
4875 * the zone and SPARSEMEM is in use. If there are holes within the
4876 * zone, each populated memory region may cost us one or two extra
4877 * memmap pages due to alignment because memmap pages for each
4878 * populated regions may not naturally algined on page boundary.
4879 * So the (present_pages >> 4) heuristic is a tradeoff for that.
4880 */
4881 if (spanned_pages > present_pages + (present_pages >> 4) &&
4882 IS_ENABLED(CONFIG_SPARSEMEM))
4883 pages = present_pages;
4884
4885 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
4886}
4887
Linus Torvalds1da177e2005-04-16 15:20:36 -07004888/*
4889 * Set up the zone data structures:
4890 * - mark all pages reserved
4891 * - mark all memory queues empty
4892 * - clear the memory bitmaps
Minchan Kim6527af52012-07-31 16:46:16 -07004893 *
4894 * NOTE: pgdat should get zeroed by caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004895 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08004896static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004897 unsigned long node_start_pfn, unsigned long node_end_pfn,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004898 unsigned long *zones_size, unsigned long *zholes_size)
4899{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07004900 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07004901 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004902 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07004903 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004904
Dave Hansen208d54e2005-10-29 18:16:52 -07004905 pgdat_resize_init(pgdat);
Andrea Arcangeli8177a422012-03-23 20:56:34 +01004906#ifdef CONFIG_NUMA_BALANCING
4907 spin_lock_init(&pgdat->numabalancing_migrate_lock);
4908 pgdat->numabalancing_migrate_nr_pages = 0;
4909 pgdat->numabalancing_migrate_next_window = jiffies;
4910#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004911 init_waitqueue_head(&pgdat->kswapd_wait);
Mel Gorman55150612012-07-31 16:44:35 -07004912 init_waitqueue_head(&pgdat->pfmemalloc_wait);
Joonsoo Kimeefa8642014-12-12 16:55:46 -08004913 pgdat_page_ext_init(pgdat);
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01004914
Linus Torvalds1da177e2005-04-16 15:20:36 -07004915 for (j = 0; j < MAX_NR_ZONES; j++) {
4916 struct zone *zone = pgdat->node_zones + j;
Jiang Liu9feedc92012-12-12 13:52:12 -08004917 unsigned long size, realsize, freesize, memmap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004918
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004919 size = zone_spanned_pages_in_node(nid, j, node_start_pfn,
4920 node_end_pfn, zones_size);
Jiang Liu9feedc92012-12-12 13:52:12 -08004921 realsize = freesize = size - zone_absent_pages_in_node(nid, j,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004922 node_start_pfn,
4923 node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004924 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004925
Mel Gorman0e0b8642006-09-27 01:49:56 -07004926 /*
Jiang Liu9feedc92012-12-12 13:52:12 -08004927 * Adjust freesize so that it accounts for how much memory
Mel Gorman0e0b8642006-09-27 01:49:56 -07004928 * is used by this zone for memmap. This affects the watermark
4929 * and per-cpu initialisations
4930 */
Jiang Liu01cefae2012-12-12 13:52:19 -08004931 memmap_pages = calc_memmap_size(size, realsize);
Zhong Hongboba914f42014-12-12 16:56:21 -08004932 if (!is_highmem_idx(j)) {
4933 if (freesize >= memmap_pages) {
4934 freesize -= memmap_pages;
4935 if (memmap_pages)
4936 printk(KERN_DEBUG
4937 " %s zone: %lu pages used for memmap\n",
4938 zone_names[j], memmap_pages);
4939 } else
4940 printk(KERN_WARNING
4941 " %s zone: %lu pages exceeds freesize %lu\n",
4942 zone_names[j], memmap_pages, freesize);
4943 }
Mel Gorman0e0b8642006-09-27 01:49:56 -07004944
Christoph Lameter62672762007-02-10 01:43:07 -08004945 /* Account for reserved pages */
Jiang Liu9feedc92012-12-12 13:52:12 -08004946 if (j == 0 && freesize > dma_reserve) {
4947 freesize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07004948 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08004949 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004950 }
4951
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004952 if (!is_highmem_idx(j))
Jiang Liu9feedc92012-12-12 13:52:12 -08004953 nr_kernel_pages += freesize;
Jiang Liu01cefae2012-12-12 13:52:19 -08004954 /* Charge for highmem memmap if there are enough kernel pages */
4955 else if (nr_kernel_pages > memmap_pages * 2)
4956 nr_kernel_pages -= memmap_pages;
Jiang Liu9feedc92012-12-12 13:52:12 -08004957 nr_all_pages += freesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004958
4959 zone->spanned_pages = size;
Jiang Liu306f2e92013-02-22 16:33:54 -08004960 zone->present_pages = realsize;
Jiang Liu9feedc92012-12-12 13:52:12 -08004961 /*
4962 * Set an approximate value for lowmem here, it will be adjusted
4963 * when the bootmem allocator frees pages into the buddy system.
4964 * And all highmem pages will be managed by the buddy system.
4965 */
4966 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
Christoph Lameter96146342006-07-03 00:24:13 -07004967#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07004968 zone->node = nid;
Jiang Liu9feedc92012-12-12 13:52:12 -08004969 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07004970 / 100;
Jiang Liu9feedc92012-12-12 13:52:12 -08004971 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07004972#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004973 zone->name = zone_names[j];
4974 spin_lock_init(&zone->lock);
4975 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07004976 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004977 zone->zone_pgdat = pgdat;
Dave Hansened8ece22005-10-29 18:16:50 -07004978 zone_pcp_init(zone);
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07004979
4980 /* For bootup, initialized properly in watermark setup */
4981 mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages);
4982
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004983 lruvec_init(&zone->lruvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004984 if (!size)
4985 continue;
4986
Andrew Morton955c1cd2012-05-29 15:06:31 -07004987 set_pageblock_order();
Linus Torvalds7c455122013-02-18 09:58:02 -08004988 setup_usemap(pgdat, zone, zone_start_pfn, size);
Dave Hansena2f3aa022007-01-10 23:15:30 -08004989 ret = init_currently_empty_zone(zone, zone_start_pfn,
4990 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07004991 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07004992 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004993 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004994 }
4995}
4996
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004997static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004998{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004999 /* Skip empty nodes */
5000 if (!pgdat->node_spanned_pages)
5001 return;
5002
Andy Whitcroftd41dee32005-06-23 00:07:54 -07005003#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07005004 /* ia64 gets its own node_mem_map, before this, without bootmem */
5005 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07005006 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07005007 struct page *map;
5008
Bob Piccoe984bb42006-05-20 15:00:31 -07005009 /*
5010 * The zone's endpoints aren't required to be MAX_ORDER
5011 * aligned but the node_mem_map endpoints must be in order
5012 * for the buddy allocator to function correctly.
5013 */
5014 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
Cody P Schafer108bcc92013-02-22 16:35:23 -08005015 end = pgdat_end_pfn(pgdat);
Bob Piccoe984bb42006-05-20 15:00:31 -07005016 end = ALIGN(end, MAX_ORDER_NR_PAGES);
5017 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07005018 map = alloc_remap(pgdat->node_id, size);
5019 if (!map)
Santosh Shilimkar67828322014-01-21 15:50:25 -08005020 map = memblock_virt_alloc_node_nopanic(size,
5021 pgdat->node_id);
Bob Piccoe984bb42006-05-20 15:00:31 -07005022 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005023 }
Roman Zippel12d810c2007-05-31 00:40:54 -07005024#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07005025 /*
5026 * With no DISCONTIG, the global mem_map is just set as node 0's
5027 */
Mel Gormanc7132162006-09-27 01:49:43 -07005028 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005029 mem_map = NODE_DATA(0)->node_mem_map;
Tejun Heo0ee332c2011-12-08 10:22:09 -08005030#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07005031 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08005032 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Tejun Heo0ee332c2011-12-08 10:22:09 -08005033#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005034 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005035#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07005036#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005037}
5038
Johannes Weiner9109fb72008-07-23 21:27:20 -07005039void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
5040 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005041{
Johannes Weiner9109fb72008-07-23 21:27:20 -07005042 pg_data_t *pgdat = NODE_DATA(nid);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005043 unsigned long start_pfn = 0;
5044 unsigned long end_pfn = 0;
Johannes Weiner9109fb72008-07-23 21:27:20 -07005045
Minchan Kim88fdf752012-07-31 16:46:14 -07005046 /* pg_data_t should be reset to zero when it's allocated */
Linus Torvalds8783b6e2012-08-02 10:37:03 -07005047 WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
Minchan Kim88fdf752012-07-31 16:46:14 -07005048
Linus Torvalds1da177e2005-04-16 15:20:36 -07005049 pgdat->node_id = nid;
5050 pgdat->node_start_pfn = node_start_pfn;
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005051#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5052 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
Juergen Gross8d29e182015-02-11 15:26:01 -08005053 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
5054 (u64)start_pfn << PAGE_SHIFT, ((u64)end_pfn << PAGE_SHIFT) - 1);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005055#endif
5056 calculate_node_totalpages(pgdat, start_pfn, end_pfn,
5057 zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005058
5059 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07005060#ifdef CONFIG_FLAT_NODE_MEM_MAP
5061 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
5062 nid, (unsigned long)pgdat,
5063 (unsigned long)pgdat->node_mem_map);
5064#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005065
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005066 free_area_init_core(pgdat, start_pfn, end_pfn,
5067 zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005068}
5069
Tejun Heo0ee332c2011-12-08 10:22:09 -08005070#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07005071
5072#if MAX_NUMNODES > 1
5073/*
5074 * Figure out the number of possible node ids.
5075 */
Cody P Schaferf9872ca2013-04-29 15:08:01 -07005076void __init setup_nr_node_ids(void)
Miklos Szeredi418508c2007-05-23 13:57:55 -07005077{
5078 unsigned int node;
5079 unsigned int highest = 0;
5080
5081 for_each_node_mask(node, node_possible_map)
5082 highest = node;
5083 nr_node_ids = highest + 1;
5084}
Miklos Szeredi418508c2007-05-23 13:57:55 -07005085#endif
5086
Mel Gormanc7132162006-09-27 01:49:43 -07005087/**
Tejun Heo1e019792011-07-12 09:45:34 +02005088 * node_map_pfn_alignment - determine the maximum internode alignment
5089 *
5090 * This function should be called after node map is populated and sorted.
5091 * It calculates the maximum power of two alignment which can distinguish
5092 * all the nodes.
5093 *
5094 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5095 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
5096 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
5097 * shifted, 1GiB is enough and this function will indicate so.
5098 *
5099 * This is used to test whether pfn -> nid mapping of the chosen memory
5100 * model has fine enough granularity to avoid incorrect mapping for the
5101 * populated node map.
5102 *
5103 * Returns the determined alignment in pfn's. 0 if there is no alignment
5104 * requirement (single node).
5105 */
5106unsigned long __init node_map_pfn_alignment(void)
5107{
5108 unsigned long accl_mask = 0, last_end = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02005109 unsigned long start, end, mask;
Tejun Heo1e019792011-07-12 09:45:34 +02005110 int last_nid = -1;
Tejun Heoc13291a2011-07-12 10:46:30 +02005111 int i, nid;
Tejun Heo1e019792011-07-12 09:45:34 +02005112
Tejun Heoc13291a2011-07-12 10:46:30 +02005113 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
Tejun Heo1e019792011-07-12 09:45:34 +02005114 if (!start || last_nid < 0 || last_nid == nid) {
5115 last_nid = nid;
5116 last_end = end;
5117 continue;
5118 }
5119
5120 /*
5121 * Start with a mask granular enough to pin-point to the
5122 * start pfn and tick off bits one-by-one until it becomes
5123 * too coarse to separate the current node from the last.
5124 */
5125 mask = ~((1 << __ffs(start)) - 1);
5126 while (mask && last_end <= (start & (mask << 1)))
5127 mask <<= 1;
5128
5129 /* accumulate all internode masks */
5130 accl_mask |= mask;
5131 }
5132
5133 /* convert mask to number of pages */
5134 return ~accl_mask + 1;
5135}
5136
Mel Gormana6af2bc2007-02-10 01:42:57 -08005137/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07005138static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07005139{
Mel Gormana6af2bc2007-02-10 01:42:57 -08005140 unsigned long min_pfn = ULONG_MAX;
Tejun Heoc13291a2011-07-12 10:46:30 +02005141 unsigned long start_pfn;
5142 int i;
Mel Gorman1abbfb42006-11-23 12:01:41 +00005143
Tejun Heoc13291a2011-07-12 10:46:30 +02005144 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
5145 min_pfn = min(min_pfn, start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005146
Mel Gormana6af2bc2007-02-10 01:42:57 -08005147 if (min_pfn == ULONG_MAX) {
5148 printk(KERN_WARNING
Paul Jackson2bc0d262008-06-22 07:22:17 -07005149 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08005150 return 0;
5151 }
5152
5153 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005154}
5155
5156/**
5157 * find_min_pfn_with_active_regions - Find the minimum PFN registered
5158 *
5159 * It returns the minimum PFN based on information provided via
Zhang Zhen7d018172014-06-04 16:10:53 -07005160 * memblock_set_node().
Mel Gormanc7132162006-09-27 01:49:43 -07005161 */
5162unsigned long __init find_min_pfn_with_active_regions(void)
5163{
5164 return find_min_pfn_for_node(MAX_NUMNODES);
5165}
5166
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005167/*
5168 * early_calculate_totalpages()
5169 * Sum pages in active regions for movable zone.
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005170 * Populate N_MEMORY for calculating usable_nodes.
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005171 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07005172static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07005173{
Mel Gorman7e63efe2007-07-17 04:03:15 -07005174 unsigned long totalpages = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02005175 unsigned long start_pfn, end_pfn;
5176 int i, nid;
Mel Gorman7e63efe2007-07-17 04:03:15 -07005177
Tejun Heoc13291a2011-07-12 10:46:30 +02005178 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
5179 unsigned long pages = end_pfn - start_pfn;
5180
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005181 totalpages += pages;
5182 if (pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005183 node_set_state(nid, N_MEMORY);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005184 }
Pintu Kumarb8af2942013-09-11 14:20:34 -07005185 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07005186}
5187
Mel Gorman2a1e2742007-07-17 04:03:12 -07005188/*
5189 * Find the PFN the Movable zone begins in each node. Kernel memory
5190 * is spread evenly between nodes as long as the nodes have enough
5191 * memory. When they don't, some nodes will have more kernelcore than
5192 * others
5193 */
Kautuk Consulb224ef82012-03-21 16:34:15 -07005194static void __init find_zone_movable_pfns_for_nodes(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005195{
5196 int i, nid;
5197 unsigned long usable_startpfn;
5198 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07005199 /* save the state before borrow the nodemask */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005200 nodemask_t saved_node_state = node_states[N_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005201 unsigned long totalpages = early_calculate_totalpages();
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005202 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
Emil Medve136199f2014-04-07 15:37:52 -07005203 struct memblock_region *r;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005204
5205 /* Need to find movable_zone earlier when movable_node is specified. */
5206 find_usable_zone_for_movable();
Mel Gorman2a1e2742007-07-17 04:03:12 -07005207
Mel Gorman7e63efe2007-07-17 04:03:15 -07005208 /*
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005209 * If movable_node is specified, ignore kernelcore and movablecore
5210 * options.
5211 */
5212 if (movable_node_is_enabled()) {
Emil Medve136199f2014-04-07 15:37:52 -07005213 for_each_memblock(memory, r) {
5214 if (!memblock_is_hotpluggable(r))
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005215 continue;
5216
Emil Medve136199f2014-04-07 15:37:52 -07005217 nid = r->nid;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005218
Emil Medve136199f2014-04-07 15:37:52 -07005219 usable_startpfn = PFN_DOWN(r->base);
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005220 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
5221 min(usable_startpfn, zone_movable_pfn[nid]) :
5222 usable_startpfn;
5223 }
5224
5225 goto out2;
5226 }
5227
5228 /*
5229 * If movablecore=nn[KMG] was specified, calculate what size of
Mel Gorman7e63efe2007-07-17 04:03:15 -07005230 * kernelcore that corresponds so that memory usable for
5231 * any allocation type is evenly spread. If both kernelcore
5232 * and movablecore are specified, then the value of kernelcore
5233 * will be used for required_kernelcore if it's greater than
5234 * what movablecore would have allowed.
5235 */
5236 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07005237 unsigned long corepages;
5238
5239 /*
5240 * Round-up so that ZONE_MOVABLE is at least as large as what
5241 * was requested by the user
5242 */
5243 required_movablecore =
5244 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
5245 corepages = totalpages - required_movablecore;
5246
5247 required_kernelcore = max(required_kernelcore, corepages);
5248 }
5249
Yinghai Lu20e69262013-03-01 14:51:27 -08005250 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
5251 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07005252 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005253
5254 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
Mel Gorman2a1e2742007-07-17 04:03:12 -07005255 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
5256
5257restart:
5258 /* Spread kernelcore memory as evenly as possible throughout nodes */
5259 kernelcore_node = required_kernelcore / usable_nodes;
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005260 for_each_node_state(nid, N_MEMORY) {
Tejun Heoc13291a2011-07-12 10:46:30 +02005261 unsigned long start_pfn, end_pfn;
5262
Mel Gorman2a1e2742007-07-17 04:03:12 -07005263 /*
5264 * Recalculate kernelcore_node if the division per node
5265 * now exceeds what is necessary to satisfy the requested
5266 * amount of memory for the kernel
5267 */
5268 if (required_kernelcore < kernelcore_node)
5269 kernelcore_node = required_kernelcore / usable_nodes;
5270
5271 /*
5272 * As the map is walked, we track how much memory is usable
5273 * by the kernel using kernelcore_remaining. When it is
5274 * 0, the rest of the node is usable by ZONE_MOVABLE
5275 */
5276 kernelcore_remaining = kernelcore_node;
5277
5278 /* Go through each range of PFNs within this node */
Tejun Heoc13291a2011-07-12 10:46:30 +02005279 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005280 unsigned long size_pages;
5281
Tejun Heoc13291a2011-07-12 10:46:30 +02005282 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005283 if (start_pfn >= end_pfn)
5284 continue;
5285
5286 /* Account for what is only usable for kernelcore */
5287 if (start_pfn < usable_startpfn) {
5288 unsigned long kernel_pages;
5289 kernel_pages = min(end_pfn, usable_startpfn)
5290 - start_pfn;
5291
5292 kernelcore_remaining -= min(kernel_pages,
5293 kernelcore_remaining);
5294 required_kernelcore -= min(kernel_pages,
5295 required_kernelcore);
5296
5297 /* Continue if range is now fully accounted */
5298 if (end_pfn <= usable_startpfn) {
5299
5300 /*
5301 * Push zone_movable_pfn to the end so
5302 * that if we have to rebalance
5303 * kernelcore across nodes, we will
5304 * not double account here
5305 */
5306 zone_movable_pfn[nid] = end_pfn;
5307 continue;
5308 }
5309 start_pfn = usable_startpfn;
5310 }
5311
5312 /*
5313 * The usable PFN range for ZONE_MOVABLE is from
5314 * start_pfn->end_pfn. Calculate size_pages as the
5315 * number of pages used as kernelcore
5316 */
5317 size_pages = end_pfn - start_pfn;
5318 if (size_pages > kernelcore_remaining)
5319 size_pages = kernelcore_remaining;
5320 zone_movable_pfn[nid] = start_pfn + size_pages;
5321
5322 /*
5323 * Some kernelcore has been met, update counts and
5324 * break if the kernelcore for this node has been
Pintu Kumarb8af2942013-09-11 14:20:34 -07005325 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07005326 */
5327 required_kernelcore -= min(required_kernelcore,
5328 size_pages);
5329 kernelcore_remaining -= size_pages;
5330 if (!kernelcore_remaining)
5331 break;
5332 }
5333 }
5334
5335 /*
5336 * If there is still required_kernelcore, we do another pass with one
5337 * less node in the count. This will push zone_movable_pfn[nid] further
5338 * along on the nodes that still have memory until kernelcore is
Pintu Kumarb8af2942013-09-11 14:20:34 -07005339 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07005340 */
5341 usable_nodes--;
5342 if (usable_nodes && required_kernelcore > usable_nodes)
5343 goto restart;
5344
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005345out2:
Mel Gorman2a1e2742007-07-17 04:03:12 -07005346 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5347 for (nid = 0; nid < MAX_NUMNODES; nid++)
5348 zone_movable_pfn[nid] =
5349 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07005350
Yinghai Lu20e69262013-03-01 14:51:27 -08005351out:
Yinghai Lu66918dc2009-06-30 11:41:37 -07005352 /* restore the node_state */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005353 node_states[N_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005354}
5355
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005356/* Any regular or high memory on that node ? */
5357static void check_for_memory(pg_data_t *pgdat, int nid)
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005358{
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005359 enum zone_type zone_type;
5360
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005361 if (N_MEMORY == N_NORMAL_MEMORY)
5362 return;
5363
5364 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005365 struct zone *zone = &pgdat->node_zones[zone_type];
Xishi Qiub38a8722013-11-12 15:07:20 -08005366 if (populated_zone(zone)) {
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005367 node_set_state(nid, N_HIGH_MEMORY);
5368 if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
5369 zone_type <= ZONE_NORMAL)
5370 node_set_state(nid, N_NORMAL_MEMORY);
Bob Liud0048b02012-01-12 17:19:07 -08005371 break;
5372 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005373 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005374}
5375
Mel Gormanc7132162006-09-27 01:49:43 -07005376/**
5377 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005378 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07005379 *
5380 * This will call free_area_init_node() for each active node in the system.
Zhang Zhen7d018172014-06-04 16:10:53 -07005381 * Using the page ranges provided by memblock_set_node(), the size of each
Mel Gormanc7132162006-09-27 01:49:43 -07005382 * zone in each node and their holes is calculated. If the maximum PFN
5383 * between two adjacent zones match, it is assumed that the zone is empty.
5384 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5385 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5386 * starts where the previous one ended. For example, ZONE_DMA32 starts
5387 * at arch_max_dma_pfn.
5388 */
5389void __init free_area_init_nodes(unsigned long *max_zone_pfn)
5390{
Tejun Heoc13291a2011-07-12 10:46:30 +02005391 unsigned long start_pfn, end_pfn;
5392 int i, nid;
Mel Gormana6af2bc2007-02-10 01:42:57 -08005393
Mel Gormanc7132162006-09-27 01:49:43 -07005394 /* Record where the zone boundaries are */
5395 memset(arch_zone_lowest_possible_pfn, 0,
5396 sizeof(arch_zone_lowest_possible_pfn));
5397 memset(arch_zone_highest_possible_pfn, 0,
5398 sizeof(arch_zone_highest_possible_pfn));
5399 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
5400 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
5401 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005402 if (i == ZONE_MOVABLE)
5403 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07005404 arch_zone_lowest_possible_pfn[i] =
5405 arch_zone_highest_possible_pfn[i-1];
5406 arch_zone_highest_possible_pfn[i] =
5407 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
5408 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07005409 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
5410 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
5411
5412 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5413 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
Kautuk Consulb224ef82012-03-21 16:34:15 -07005414 find_zone_movable_pfns_for_nodes();
Mel Gormanc7132162006-09-27 01:49:43 -07005415
Mel Gormanc7132162006-09-27 01:49:43 -07005416 /* Print out the zone ranges */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005417 pr_info("Zone ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005418 for (i = 0; i < MAX_NR_ZONES; i++) {
5419 if (i == ZONE_MOVABLE)
5420 continue;
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005421 pr_info(" %-8s ", zone_names[i]);
David Rientjes72f0ba02010-03-05 13:42:14 -08005422 if (arch_zone_lowest_possible_pfn[i] ==
5423 arch_zone_highest_possible_pfn[i])
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005424 pr_cont("empty\n");
David Rientjes72f0ba02010-03-05 13:42:14 -08005425 else
Juergen Gross8d29e182015-02-11 15:26:01 -08005426 pr_cont("[mem %#018Lx-%#018Lx]\n",
5427 (u64)arch_zone_lowest_possible_pfn[i]
5428 << PAGE_SHIFT,
5429 ((u64)arch_zone_highest_possible_pfn[i]
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005430 << PAGE_SHIFT) - 1);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005431 }
5432
5433 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005434 pr_info("Movable zone start for each node\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005435 for (i = 0; i < MAX_NUMNODES; i++) {
5436 if (zone_movable_pfn[i])
Juergen Gross8d29e182015-02-11 15:26:01 -08005437 pr_info(" Node %d: %#018Lx\n", i,
5438 (u64)zone_movable_pfn[i] << PAGE_SHIFT);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005439 }
Mel Gormanc7132162006-09-27 01:49:43 -07005440
Wanpeng Lif2d52fe2012-10-08 16:32:24 -07005441 /* Print out the early node map */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005442 pr_info("Early memory node ranges\n");
Tejun Heoc13291a2011-07-12 10:46:30 +02005443 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
Juergen Gross8d29e182015-02-11 15:26:01 -08005444 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid,
5445 (u64)start_pfn << PAGE_SHIFT,
5446 ((u64)end_pfn << PAGE_SHIFT) - 1);
Mel Gormanc7132162006-09-27 01:49:43 -07005447
5448 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07005449 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08005450 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07005451 for_each_online_node(nid) {
5452 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07005453 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07005454 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005455
5456 /* Any memory on that node */
5457 if (pgdat->node_present_pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005458 node_set_state(nid, N_MEMORY);
5459 check_for_memory(pgdat, nid);
Mel Gormanc7132162006-09-27 01:49:43 -07005460 }
5461}
Mel Gorman2a1e2742007-07-17 04:03:12 -07005462
Mel Gorman7e63efe2007-07-17 04:03:15 -07005463static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005464{
5465 unsigned long long coremem;
5466 if (!p)
5467 return -EINVAL;
5468
5469 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005470 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005471
Mel Gorman7e63efe2007-07-17 04:03:15 -07005472 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07005473 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
5474
5475 return 0;
5476}
Mel Gormaned7ed362007-07-17 04:03:14 -07005477
Mel Gorman7e63efe2007-07-17 04:03:15 -07005478/*
5479 * kernelcore=size sets the amount of memory for use for allocations that
5480 * cannot be reclaimed or migrated.
5481 */
5482static int __init cmdline_parse_kernelcore(char *p)
5483{
5484 return cmdline_parse_core(p, &required_kernelcore);
5485}
5486
5487/*
5488 * movablecore=size sets the amount of memory for use for allocations that
5489 * can be reclaimed or migrated.
5490 */
5491static int __init cmdline_parse_movablecore(char *p)
5492{
5493 return cmdline_parse_core(p, &required_movablecore);
5494}
5495
Mel Gormaned7ed362007-07-17 04:03:14 -07005496early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005497early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07005498
Tejun Heo0ee332c2011-12-08 10:22:09 -08005499#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005500
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005501void adjust_managed_page_count(struct page *page, long count)
5502{
5503 spin_lock(&managed_page_count_lock);
5504 page_zone(page)->managed_pages += count;
5505 totalram_pages += count;
Jiang Liu3dcc0572013-07-03 15:03:21 -07005506#ifdef CONFIG_HIGHMEM
5507 if (PageHighMem(page))
5508 totalhigh_pages += count;
5509#endif
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005510 spin_unlock(&managed_page_count_lock);
5511}
Jiang Liu3dcc0572013-07-03 15:03:21 -07005512EXPORT_SYMBOL(adjust_managed_page_count);
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005513
Jiang Liu11199692013-07-03 15:02:48 -07005514unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
Jiang Liu69afade2013-04-29 15:06:21 -07005515{
Jiang Liu11199692013-07-03 15:02:48 -07005516 void *pos;
5517 unsigned long pages = 0;
Jiang Liu69afade2013-04-29 15:06:21 -07005518
Jiang Liu11199692013-07-03 15:02:48 -07005519 start = (void *)PAGE_ALIGN((unsigned long)start);
5520 end = (void *)((unsigned long)end & PAGE_MASK);
5521 for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
Jiang Liudbe67df2013-07-03 15:02:51 -07005522 if ((unsigned int)poison <= 0xFF)
Jiang Liu11199692013-07-03 15:02:48 -07005523 memset(pos, poison, PAGE_SIZE);
5524 free_reserved_page(virt_to_page(pos));
Jiang Liu69afade2013-04-29 15:06:21 -07005525 }
5526
5527 if (pages && s)
Jiang Liu11199692013-07-03 15:02:48 -07005528 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
Jiang Liu69afade2013-04-29 15:06:21 -07005529 s, pages << (PAGE_SHIFT - 10), start, end);
5530
5531 return pages;
5532}
Jiang Liu11199692013-07-03 15:02:48 -07005533EXPORT_SYMBOL(free_reserved_area);
Jiang Liu69afade2013-04-29 15:06:21 -07005534
Jiang Liucfa11e02013-04-29 15:07:00 -07005535#ifdef CONFIG_HIGHMEM
5536void free_highmem_page(struct page *page)
5537{
5538 __free_reserved_page(page);
5539 totalram_pages++;
Jiang Liu7b4b2a02013-07-03 15:03:11 -07005540 page_zone(page)->managed_pages++;
Jiang Liucfa11e02013-04-29 15:07:00 -07005541 totalhigh_pages++;
5542}
5543#endif
5544
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005545
5546void __init mem_init_print_info(const char *str)
5547{
5548 unsigned long physpages, codesize, datasize, rosize, bss_size;
5549 unsigned long init_code_size, init_data_size;
5550
5551 physpages = get_num_physpages();
5552 codesize = _etext - _stext;
5553 datasize = _edata - _sdata;
5554 rosize = __end_rodata - __start_rodata;
5555 bss_size = __bss_stop - __bss_start;
5556 init_data_size = __init_end - __init_begin;
5557 init_code_size = _einittext - _sinittext;
5558
5559 /*
5560 * Detect special cases and adjust section sizes accordingly:
5561 * 1) .init.* may be embedded into .data sections
5562 * 2) .init.text.* may be out of [__init_begin, __init_end],
5563 * please refer to arch/tile/kernel/vmlinux.lds.S.
5564 * 3) .rodata.* may be embedded into .text or .data sections.
5565 */
5566#define adj_init_size(start, end, size, pos, adj) \
Pintu Kumarb8af2942013-09-11 14:20:34 -07005567 do { \
5568 if (start <= pos && pos < end && size > adj) \
5569 size -= adj; \
5570 } while (0)
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005571
5572 adj_init_size(__init_begin, __init_end, init_data_size,
5573 _sinittext, init_code_size);
5574 adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
5575 adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
5576 adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
5577 adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
5578
5579#undef adj_init_size
5580
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005581 pr_info("Memory: %luK/%luK available "
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005582 "(%luK kernel code, %luK rwdata, %luK rodata, "
Pintu Kumare48322a2014-12-18 16:17:15 -08005583 "%luK init, %luK bss, %luK reserved, %luK cma-reserved"
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005584#ifdef CONFIG_HIGHMEM
5585 ", %luK highmem"
5586#endif
5587 "%s%s)\n",
5588 nr_free_pages() << (PAGE_SHIFT-10), physpages << (PAGE_SHIFT-10),
5589 codesize >> 10, datasize >> 10, rosize >> 10,
5590 (init_data_size + init_code_size) >> 10, bss_size >> 10,
Pintu Kumare48322a2014-12-18 16:17:15 -08005591 (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT-10),
5592 totalcma_pages << (PAGE_SHIFT-10),
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005593#ifdef CONFIG_HIGHMEM
5594 totalhigh_pages << (PAGE_SHIFT-10),
5595#endif
5596 str ? ", " : "", str ? str : "");
5597}
5598
Mel Gorman0e0b8642006-09-27 01:49:56 -07005599/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005600 * set_dma_reserve - set the specified number of pages reserved in the first zone
5601 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07005602 *
5603 * The per-cpu batchsize and zone watermarks are determined by present_pages.
5604 * In the DMA zone, a significant percentage may be consumed by kernel image
5605 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005606 * function may optionally be used to account for unfreeable pages in the
5607 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5608 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07005609 */
5610void __init set_dma_reserve(unsigned long new_dma_reserve)
5611{
5612 dma_reserve = new_dma_reserve;
5613}
5614
Linus Torvalds1da177e2005-04-16 15:20:36 -07005615void __init free_area_init(unsigned long *zones_size)
5616{
Johannes Weiner9109fb72008-07-23 21:27:20 -07005617 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005618 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
5619}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005620
Linus Torvalds1da177e2005-04-16 15:20:36 -07005621static int page_alloc_cpu_notify(struct notifier_block *self,
5622 unsigned long action, void *hcpu)
5623{
5624 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005625
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07005626 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -07005627 lru_add_drain_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005628 drain_pages(cpu);
5629
5630 /*
5631 * Spill the event counters of the dead processor
5632 * into the current processors event counters.
5633 * This artificially elevates the count of the current
5634 * processor.
5635 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07005636 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005637
5638 /*
5639 * Zero the differential counters of the dead processor
5640 * so that the vm statistics are consistent.
5641 *
5642 * This is only okay since the processor is dead and cannot
5643 * race with what we are doing.
5644 */
Christoph Lameter2bb921e2013-09-11 14:21:30 -07005645 cpu_vm_stats_fold(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005646 }
5647 return NOTIFY_OK;
5648}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005649
5650void __init page_alloc_init(void)
5651{
5652 hotcpu_notifier(page_alloc_cpu_notify, 0);
5653}
5654
5655/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005656 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5657 * or min_free_kbytes changes.
5658 */
5659static void calculate_totalreserve_pages(void)
5660{
5661 struct pglist_data *pgdat;
5662 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005663 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005664
5665 for_each_online_pgdat(pgdat) {
5666 for (i = 0; i < MAX_NR_ZONES; i++) {
5667 struct zone *zone = pgdat->node_zones + i;
Mel Gorman3484b2d2014-08-06 16:07:14 -07005668 long max = 0;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005669
5670 /* Find valid and maximum lowmem_reserve in the zone */
5671 for (j = i; j < MAX_NR_ZONES; j++) {
5672 if (zone->lowmem_reserve[j] > max)
5673 max = zone->lowmem_reserve[j];
5674 }
5675
Mel Gorman41858962009-06-16 15:32:12 -07005676 /* we treat the high watermark as reserved pages. */
5677 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005678
Jiang Liub40da042013-02-22 16:33:52 -08005679 if (max > zone->managed_pages)
5680 max = zone->managed_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005681 reserve_pages += max;
Johannes Weinerab8fabd2012-01-10 15:07:42 -08005682 /*
5683 * Lowmem reserves are not available to
5684 * GFP_HIGHUSER page cache allocations and
5685 * kswapd tries to balance zones to their high
5686 * watermark. As a result, neither should be
5687 * regarded as dirtyable memory, to prevent a
5688 * situation where reclaim has to clean pages
5689 * in order to balance the zones.
5690 */
5691 zone->dirty_balance_reserve = max;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005692 }
5693 }
Johannes Weinerab8fabd2012-01-10 15:07:42 -08005694 dirty_balance_reserve = reserve_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005695 totalreserve_pages = reserve_pages;
5696}
5697
5698/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07005699 * setup_per_zone_lowmem_reserve - called whenever
5700 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
5701 * has a correct pages reserved value, so an adequate number of
5702 * pages are left in the zone after a successful __alloc_pages().
5703 */
5704static void setup_per_zone_lowmem_reserve(void)
5705{
5706 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005707 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005708
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08005709 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005710 for (j = 0; j < MAX_NR_ZONES; j++) {
5711 struct zone *zone = pgdat->node_zones + j;
Jiang Liub40da042013-02-22 16:33:52 -08005712 unsigned long managed_pages = zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005713
5714 zone->lowmem_reserve[j] = 0;
5715
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005716 idx = j;
5717 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005718 struct zone *lower_zone;
5719
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005720 idx--;
5721
Linus Torvalds1da177e2005-04-16 15:20:36 -07005722 if (sysctl_lowmem_reserve_ratio[idx] < 1)
5723 sysctl_lowmem_reserve_ratio[idx] = 1;
5724
5725 lower_zone = pgdat->node_zones + idx;
Jiang Liub40da042013-02-22 16:33:52 -08005726 lower_zone->lowmem_reserve[j] = managed_pages /
Linus Torvalds1da177e2005-04-16 15:20:36 -07005727 sysctl_lowmem_reserve_ratio[idx];
Jiang Liub40da042013-02-22 16:33:52 -08005728 managed_pages += lower_zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005729 }
5730 }
5731 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005732
5733 /* update totalreserve_pages */
5734 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005735}
5736
Mel Gormancfd3da12011-04-25 21:36:42 +00005737static void __setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005738{
5739 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
5740 unsigned long lowmem_pages = 0;
5741 struct zone *zone;
5742 unsigned long flags;
5743
5744 /* Calculate total number of !ZONE_HIGHMEM pages */
5745 for_each_zone(zone) {
5746 if (!is_highmem(zone))
Jiang Liub40da042013-02-22 16:33:52 -08005747 lowmem_pages += zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005748 }
5749
5750 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07005751 u64 tmp;
5752
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005753 spin_lock_irqsave(&zone->lock, flags);
Jiang Liub40da042013-02-22 16:33:52 -08005754 tmp = (u64)pages_min * zone->managed_pages;
Andrew Mortonac924c62006-05-15 09:43:59 -07005755 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005756 if (is_highmem(zone)) {
5757 /*
Nick Piggin669ed172005-11-13 16:06:45 -08005758 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5759 * need highmem pages, so cap pages_min to a small
5760 * value here.
5761 *
Mel Gorman41858962009-06-16 15:32:12 -07005762 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Yaowei Bai42ff2702015-04-14 15:47:14 -07005763 * deltas control asynch page reclaim, and so should
Nick Piggin669ed172005-11-13 16:06:45 -08005764 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005765 */
Andrew Morton90ae8d62013-02-22 16:32:22 -08005766 unsigned long min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005767
Jiang Liub40da042013-02-22 16:33:52 -08005768 min_pages = zone->managed_pages / 1024;
Andrew Morton90ae8d62013-02-22 16:32:22 -08005769 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
Mel Gorman41858962009-06-16 15:32:12 -07005770 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005771 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08005772 /*
5773 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07005774 * proportionate to the zone's size.
5775 */
Mel Gorman41858962009-06-16 15:32:12 -07005776 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005777 }
5778
Mel Gorman41858962009-06-16 15:32:12 -07005779 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
5780 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Marek Szyprowski49f223a2012-01-25 12:49:24 +01005781
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07005782 __mod_zone_page_state(zone, NR_ALLOC_BATCH,
Johannes Weinerabe5f972014-10-02 16:21:10 -07005783 high_wmark_pages(zone) - low_wmark_pages(zone) -
5784 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07005785
Mel Gorman56fd56b2007-10-16 01:25:58 -07005786 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005787 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005788 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005789
5790 /* update totalreserve_pages */
5791 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005792}
5793
Mel Gormancfd3da12011-04-25 21:36:42 +00005794/**
5795 * setup_per_zone_wmarks - called when min_free_kbytes changes
5796 * or when memory is hot-{added|removed}
5797 *
5798 * Ensures that the watermark[min,low,high] values for each zone are set
5799 * correctly with respect to min_free_kbytes.
5800 */
5801void setup_per_zone_wmarks(void)
5802{
5803 mutex_lock(&zonelists_mutex);
5804 __setup_per_zone_wmarks();
5805 mutex_unlock(&zonelists_mutex);
5806}
5807
Randy Dunlap55a44622009-09-21 17:01:20 -07005808/*
Rik van Riel556adec2008-10-18 20:26:34 -07005809 * The inactive anon list should be small enough that the VM never has to
5810 * do too much work, but large enough that each inactive page has a chance
5811 * to be referenced again before it is swapped out.
5812 *
5813 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5814 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5815 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5816 * the anonymous pages are kept on the inactive list.
5817 *
5818 * total target max
5819 * memory ratio inactive anon
5820 * -------------------------------------
5821 * 10MB 1 5MB
5822 * 100MB 1 50MB
5823 * 1GB 3 250MB
5824 * 10GB 10 0.9GB
5825 * 100GB 31 3GB
5826 * 1TB 101 10GB
5827 * 10TB 320 32GB
5828 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005829static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
Minchan Kim96cb4df2009-06-16 15:32:49 -07005830{
5831 unsigned int gb, ratio;
5832
5833 /* Zone size in gigabytes */
Jiang Liub40da042013-02-22 16:33:52 -08005834 gb = zone->managed_pages >> (30 - PAGE_SHIFT);
Minchan Kim96cb4df2009-06-16 15:32:49 -07005835 if (gb)
5836 ratio = int_sqrt(10 * gb);
5837 else
5838 ratio = 1;
5839
5840 zone->inactive_ratio = ratio;
5841}
5842
KOSAKI Motohiro839a4fc2011-05-24 17:11:31 -07005843static void __meminit setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07005844{
5845 struct zone *zone;
5846
Minchan Kim96cb4df2009-06-16 15:32:49 -07005847 for_each_zone(zone)
5848 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07005849}
5850
Linus Torvalds1da177e2005-04-16 15:20:36 -07005851/*
5852 * Initialise min_free_kbytes.
5853 *
5854 * For small machines we want it small (128k min). For large machines
5855 * we want it large (64MB max). But it is not linear, because network
5856 * bandwidth does not increase linearly with machine size. We use
5857 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07005858 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
Linus Torvalds1da177e2005-04-16 15:20:36 -07005859 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5860 *
5861 * which yields
5862 *
5863 * 16MB: 512k
5864 * 32MB: 724k
5865 * 64MB: 1024k
5866 * 128MB: 1448k
5867 * 256MB: 2048k
5868 * 512MB: 2896k
5869 * 1024MB: 4096k
5870 * 2048MB: 5792k
5871 * 4096MB: 8192k
5872 * 8192MB: 11584k
5873 * 16384MB: 16384k
5874 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005875int __meminit init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005876{
5877 unsigned long lowmem_kbytes;
Michal Hocko5f127332013-07-08 16:00:40 -07005878 int new_min_free_kbytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005879
5880 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
Michal Hocko5f127332013-07-08 16:00:40 -07005881 new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005882
Michal Hocko5f127332013-07-08 16:00:40 -07005883 if (new_min_free_kbytes > user_min_free_kbytes) {
5884 min_free_kbytes = new_min_free_kbytes;
5885 if (min_free_kbytes < 128)
5886 min_free_kbytes = 128;
5887 if (min_free_kbytes > 65536)
5888 min_free_kbytes = 65536;
5889 } else {
5890 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
5891 new_min_free_kbytes, user_min_free_kbytes);
5892 }
Minchan Kimbc75d332009-06-16 15:32:48 -07005893 setup_per_zone_wmarks();
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -07005894 refresh_zone_stat_thresholds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005895 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07005896 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005897 return 0;
5898}
Minchan Kimbc75d332009-06-16 15:32:48 -07005899module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005900
5901/*
Pintu Kumarb8af2942013-09-11 14:20:34 -07005902 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
Linus Torvalds1da177e2005-04-16 15:20:36 -07005903 * that we can call two helper functions whenever min_free_kbytes
5904 * changes.
5905 */
Joe Perchescccad5b2014-06-06 14:38:09 -07005906int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005907 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005908{
Han Pingtianda8c7572014-01-23 15:53:17 -08005909 int rc;
5910
5911 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5912 if (rc)
5913 return rc;
5914
Michal Hocko5f127332013-07-08 16:00:40 -07005915 if (write) {
5916 user_min_free_kbytes = min_free_kbytes;
Minchan Kimbc75d332009-06-16 15:32:48 -07005917 setup_per_zone_wmarks();
Michal Hocko5f127332013-07-08 16:00:40 -07005918 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005919 return 0;
5920}
5921
Christoph Lameter96146342006-07-03 00:24:13 -07005922#ifdef CONFIG_NUMA
Joe Perchescccad5b2014-06-06 14:38:09 -07005923int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005924 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07005925{
5926 struct zone *zone;
5927 int rc;
5928
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005929 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07005930 if (rc)
5931 return rc;
5932
5933 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08005934 zone->min_unmapped_pages = (zone->managed_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07005935 sysctl_min_unmapped_ratio) / 100;
5936 return 0;
5937}
Christoph Lameter0ff38492006-09-25 23:31:52 -07005938
Joe Perchescccad5b2014-06-06 14:38:09 -07005939int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005940 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07005941{
5942 struct zone *zone;
5943 int rc;
5944
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005945 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07005946 if (rc)
5947 return rc;
5948
5949 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08005950 zone->min_slab_pages = (zone->managed_pages *
Christoph Lameter0ff38492006-09-25 23:31:52 -07005951 sysctl_min_slab_ratio) / 100;
5952 return 0;
5953}
Christoph Lameter96146342006-07-03 00:24:13 -07005954#endif
5955
Linus Torvalds1da177e2005-04-16 15:20:36 -07005956/*
5957 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5958 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5959 * whenever sysctl_lowmem_reserve_ratio changes.
5960 *
5961 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07005962 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07005963 * if in function of the boot time zone sizes.
5964 */
Joe Perchescccad5b2014-06-06 14:38:09 -07005965int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005966 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005967{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005968 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005969 setup_per_zone_lowmem_reserve();
5970 return 0;
5971}
5972
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005973/*
5974 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
Pintu Kumarb8af2942013-09-11 14:20:34 -07005975 * cpu. It is the fraction of total pages in each zone that a hot per cpu
5976 * pagelist can have before it gets flushed back to buddy allocator.
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005977 */
Joe Perchescccad5b2014-06-06 14:38:09 -07005978int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005979 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005980{
5981 struct zone *zone;
David Rientjes7cd2b0a2014-06-23 13:22:04 -07005982 int old_percpu_pagelist_fraction;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005983 int ret;
5984
Cody P Schaferc8e251f2013-07-03 15:01:29 -07005985 mutex_lock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07005986 old_percpu_pagelist_fraction = percpu_pagelist_fraction;
5987
5988 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
5989 if (!write || ret < 0)
5990 goto out;
5991
5992 /* Sanity checking to avoid pcp imbalance */
5993 if (percpu_pagelist_fraction &&
5994 percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
5995 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
5996 ret = -EINVAL;
5997 goto out;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005998 }
David Rientjes7cd2b0a2014-06-23 13:22:04 -07005999
6000 /* No change? */
6001 if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
6002 goto out;
6003
6004 for_each_populated_zone(zone) {
6005 unsigned int cpu;
6006
6007 for_each_possible_cpu(cpu)
6008 pageset_set_high_and_batch(zone,
6009 per_cpu_ptr(zone->pageset, cpu));
6010 }
6011out:
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006012 mutex_unlock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006013 return ret;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006014}
6015
David S. Millerf034b5d2006-08-24 03:08:07 -07006016int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006017
6018#ifdef CONFIG_NUMA
6019static int __init set_hashdist(char *str)
6020{
6021 if (!str)
6022 return 0;
6023 hashdist = simple_strtoul(str, &str, 0);
6024 return 1;
6025}
6026__setup("hashdist=", set_hashdist);
6027#endif
6028
6029/*
6030 * allocate a large system hash table from bootmem
6031 * - it is assumed that the hash table must contain an exact power-of-2
6032 * quantity of entries
6033 * - limit is the number of hash buckets, not the total allocation size
6034 */
6035void *__init alloc_large_system_hash(const char *tablename,
6036 unsigned long bucketsize,
6037 unsigned long numentries,
6038 int scale,
6039 int flags,
6040 unsigned int *_hash_shift,
6041 unsigned int *_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00006042 unsigned long low_limit,
6043 unsigned long high_limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006044{
Tim Bird31fe62b2012-05-23 13:33:35 +00006045 unsigned long long max = high_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006046 unsigned long log2qty, size;
6047 void *table = NULL;
6048
6049 /* allow the kernel cmdline to have a say */
6050 if (!numentries) {
6051 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08006052 numentries = nr_kernel_pages;
Jerry Zhoua7e83312013-09-11 14:20:26 -07006053
6054 /* It isn't necessary when PAGE_SIZE >= 1MB */
6055 if (PAGE_SHIFT < 20)
6056 numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006057
6058 /* limit to 1 bucket per 2^scale bytes of low memory */
6059 if (scale > PAGE_SHIFT)
6060 numentries >>= (scale - PAGE_SHIFT);
6061 else
6062 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08006063
6064 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07006065 if (unlikely(flags & HASH_SMALL)) {
6066 /* Makes no sense without HASH_EARLY */
6067 WARN_ON(!(flags & HASH_EARLY));
6068 if (!(numentries >> *_hash_shift)) {
6069 numentries = 1UL << *_hash_shift;
6070 BUG_ON(!numentries);
6071 }
6072 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08006073 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006074 }
John Hawkes6e692ed2006-03-25 03:08:02 -08006075 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006076
6077 /* limit allocation size to 1/16 total memory by default */
6078 if (max == 0) {
6079 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
6080 do_div(max, bucketsize);
6081 }
Dimitri Sivanich074b8512012-02-08 12:39:07 -08006082 max = min(max, 0x80000000ULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006083
Tim Bird31fe62b2012-05-23 13:33:35 +00006084 if (numentries < low_limit)
6085 numentries = low_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006086 if (numentries > max)
6087 numentries = max;
6088
David Howellsf0d1b0b2006-12-08 02:37:49 -08006089 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006090
6091 do {
6092 size = bucketsize << log2qty;
6093 if (flags & HASH_EARLY)
Santosh Shilimkar67828322014-01-21 15:50:25 -08006094 table = memblock_virt_alloc_nopanic(size, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006095 else if (hashdist)
6096 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
6097 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07006098 /*
6099 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07006100 * some pages at the end of hash table which
6101 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07006102 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01006103 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07006104 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01006105 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
6106 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006107 }
6108 } while (!table && size > PAGE_SIZE && --log2qty);
6109
6110 if (!table)
6111 panic("Failed to allocate %s hash table\n", tablename);
6112
Robin Holtf241e6602010-10-07 12:59:26 -07006113 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07006114 tablename,
Robin Holtf241e6602010-10-07 12:59:26 -07006115 (1UL << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08006116 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07006117 size);
6118
6119 if (_hash_shift)
6120 *_hash_shift = log2qty;
6121 if (_hash_mask)
6122 *_hash_mask = (1 << log2qty) - 1;
6123
6124 return table;
6125}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08006126
Mel Gorman835c1342007-10-16 01:25:47 -07006127/* Return a pointer to the bitmap storing bits affecting a block of pages */
6128static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
6129 unsigned long pfn)
6130{
6131#ifdef CONFIG_SPARSEMEM
6132 return __pfn_to_section(pfn)->pageblock_flags;
6133#else
6134 return zone->pageblock_flags;
6135#endif /* CONFIG_SPARSEMEM */
6136}
Andrew Morton6220ec72006-10-19 23:29:05 -07006137
Mel Gorman835c1342007-10-16 01:25:47 -07006138static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
6139{
6140#ifdef CONFIG_SPARSEMEM
6141 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07006142 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07006143#else
Laura Abbottc060f942013-01-11 14:31:51 -08006144 pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
Mel Gormand9c23402007-10-16 01:26:01 -07006145 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07006146#endif /* CONFIG_SPARSEMEM */
6147}
6148
6149/**
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006150 * 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 -07006151 * @page: The page within the block of interest
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006152 * @pfn: The target page frame number
6153 * @end_bitidx: The last bit of interest to retrieve
6154 * @mask: mask of bits that the caller is interested in
6155 *
6156 * Return: pageblock_bits flags
Mel Gorman835c1342007-10-16 01:25:47 -07006157 */
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006158unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
Mel Gormane58469b2014-06-04 16:10:16 -07006159 unsigned long end_bitidx,
6160 unsigned long mask)
Mel Gorman835c1342007-10-16 01:25:47 -07006161{
6162 struct zone *zone;
6163 unsigned long *bitmap;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006164 unsigned long bitidx, word_bitidx;
Mel Gormane58469b2014-06-04 16:10:16 -07006165 unsigned long word;
Mel Gorman835c1342007-10-16 01:25:47 -07006166
6167 zone = page_zone(page);
Mel Gorman835c1342007-10-16 01:25:47 -07006168 bitmap = get_pageblock_bitmap(zone, pfn);
6169 bitidx = pfn_to_bitidx(zone, pfn);
Mel Gormane58469b2014-06-04 16:10:16 -07006170 word_bitidx = bitidx / BITS_PER_LONG;
6171 bitidx &= (BITS_PER_LONG-1);
Mel Gorman835c1342007-10-16 01:25:47 -07006172
Mel Gormane58469b2014-06-04 16:10:16 -07006173 word = bitmap[word_bitidx];
6174 bitidx += end_bitidx;
6175 return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
Mel Gorman835c1342007-10-16 01:25:47 -07006176}
6177
6178/**
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006179 * 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 -07006180 * @page: The page within the block of interest
Mel Gorman835c1342007-10-16 01:25:47 -07006181 * @flags: The flags to set
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006182 * @pfn: The target page frame number
6183 * @end_bitidx: The last bit of interest
6184 * @mask: mask of bits that the caller is interested in
Mel Gorman835c1342007-10-16 01:25:47 -07006185 */
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006186void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
6187 unsigned long pfn,
Mel Gormane58469b2014-06-04 16:10:16 -07006188 unsigned long end_bitidx,
6189 unsigned long mask)
Mel Gorman835c1342007-10-16 01:25:47 -07006190{
6191 struct zone *zone;
6192 unsigned long *bitmap;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006193 unsigned long bitidx, word_bitidx;
Mel Gormane58469b2014-06-04 16:10:16 -07006194 unsigned long old_word, word;
6195
6196 BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
Mel Gorman835c1342007-10-16 01:25:47 -07006197
6198 zone = page_zone(page);
Mel Gorman835c1342007-10-16 01:25:47 -07006199 bitmap = get_pageblock_bitmap(zone, pfn);
6200 bitidx = pfn_to_bitidx(zone, pfn);
Mel Gormane58469b2014-06-04 16:10:16 -07006201 word_bitidx = bitidx / BITS_PER_LONG;
6202 bitidx &= (BITS_PER_LONG-1);
6203
Sasha Levin309381fea2014-01-23 15:52:54 -08006204 VM_BUG_ON_PAGE(!zone_spans_pfn(zone, pfn), page);
Mel Gorman835c1342007-10-16 01:25:47 -07006205
Mel Gormane58469b2014-06-04 16:10:16 -07006206 bitidx += end_bitidx;
6207 mask <<= (BITS_PER_LONG - bitidx - 1);
6208 flags <<= (BITS_PER_LONG - bitidx - 1);
6209
Jason Low4db0c3c2015-04-15 16:14:08 -07006210 word = READ_ONCE(bitmap[word_bitidx]);
Mel Gormane58469b2014-06-04 16:10:16 -07006211 for (;;) {
6212 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
6213 if (word == old_word)
6214 break;
6215 word = old_word;
6216 }
Mel Gorman835c1342007-10-16 01:25:47 -07006217}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07006218
6219/*
Minchan Kim80934512012-07-31 16:43:01 -07006220 * This function checks whether pageblock includes unmovable pages or not.
6221 * If @count is not zero, it is okay to include less @count unmovable pages
6222 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07006223 * PageLRU check without isolation or lru_lock could race so that
Minchan Kim80934512012-07-31 16:43:01 -07006224 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6225 * expect this function should be exact.
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07006226 */
Wen Congyangb023f462012-12-11 16:00:45 -08006227bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
6228 bool skip_hwpoisoned_pages)
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006229{
6230 unsigned long pfn, iter, found;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01006231 int mt;
6232
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006233 /*
6234 * For avoiding noise data, lru_add_drain_all() should be called
Minchan Kim80934512012-07-31 16:43:01 -07006235 * If ZONE_MOVABLE, the zone never contains unmovable pages
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006236 */
6237 if (zone_idx(zone) == ZONE_MOVABLE)
Minchan Kim80934512012-07-31 16:43:01 -07006238 return false;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01006239 mt = get_pageblock_migratetype(page);
6240 if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
Minchan Kim80934512012-07-31 16:43:01 -07006241 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006242
6243 pfn = page_to_pfn(page);
6244 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
6245 unsigned long check = pfn + iter;
6246
Namhyung Kim29723fc2011-02-25 14:44:25 -08006247 if (!pfn_valid_within(check))
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006248 continue;
Namhyung Kim29723fc2011-02-25 14:44:25 -08006249
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006250 page = pfn_to_page(check);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07006251
6252 /*
6253 * Hugepages are not in LRU lists, but they're movable.
6254 * We need not scan over tail pages bacause we don't
6255 * handle each tail page individually in migration.
6256 */
6257 if (PageHuge(page)) {
6258 iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
6259 continue;
6260 }
6261
Minchan Kim97d255c2012-07-31 16:42:59 -07006262 /*
6263 * We can't use page_count without pin a page
6264 * because another CPU can free compound page.
6265 * This check already skips compound tails of THP
6266 * because their page->_count is zero at all time.
6267 */
6268 if (!atomic_read(&page->_count)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006269 if (PageBuddy(page))
6270 iter += (1 << page_order(page)) - 1;
6271 continue;
6272 }
Minchan Kim97d255c2012-07-31 16:42:59 -07006273
Wen Congyangb023f462012-12-11 16:00:45 -08006274 /*
6275 * The HWPoisoned page may be not in buddy system, and
6276 * page_count() is not 0.
6277 */
6278 if (skip_hwpoisoned_pages && PageHWPoison(page))
6279 continue;
6280
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006281 if (!PageLRU(page))
6282 found++;
6283 /*
Johannes Weiner6b4f7792014-12-12 16:56:13 -08006284 * If there are RECLAIMABLE pages, we need to check
6285 * it. But now, memory offline itself doesn't call
6286 * shrink_node_slabs() and it still to be fixed.
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006287 */
6288 /*
6289 * If the page is not RAM, page_count()should be 0.
6290 * we don't need more check. This is an _used_ not-movable page.
6291 *
6292 * The problematic thing here is PG_reserved pages. PG_reserved
6293 * is set to both of a memory hole page and a _used_ kernel
6294 * page at boot.
6295 */
6296 if (found > count)
Minchan Kim80934512012-07-31 16:43:01 -07006297 return true;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006298 }
Minchan Kim80934512012-07-31 16:43:01 -07006299 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006300}
6301
6302bool is_pageblock_removable_nolock(struct page *page)
6303{
Michal Hocko656a0702012-01-20 14:33:58 -08006304 struct zone *zone;
6305 unsigned long pfn;
Michal Hocko687875f2012-01-20 14:33:55 -08006306
6307 /*
6308 * We have to be careful here because we are iterating over memory
6309 * sections which are not zone aware so we might end up outside of
6310 * the zone but still within the section.
Michal Hocko656a0702012-01-20 14:33:58 -08006311 * We have to take care about the node as well. If the node is offline
6312 * its NODE_DATA will be NULL - see page_zone.
Michal Hocko687875f2012-01-20 14:33:55 -08006313 */
Michal Hocko656a0702012-01-20 14:33:58 -08006314 if (!node_online(page_to_nid(page)))
6315 return false;
6316
6317 zone = page_zone(page);
6318 pfn = page_to_pfn(page);
Cody P Schafer108bcc92013-02-22 16:35:23 -08006319 if (!zone_spans_pfn(zone, pfn))
Michal Hocko687875f2012-01-20 14:33:55 -08006320 return false;
6321
Wen Congyangb023f462012-12-11 16:00:45 -08006322 return !has_unmovable_pages(zone, page, 0, true);
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07006323}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006324
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006325#ifdef CONFIG_CMA
6326
6327static unsigned long pfn_max_align_down(unsigned long pfn)
6328{
6329 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
6330 pageblock_nr_pages) - 1);
6331}
6332
6333static unsigned long pfn_max_align_up(unsigned long pfn)
6334{
6335 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
6336 pageblock_nr_pages));
6337}
6338
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006339/* [start, end) must belong to a single zone. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006340static int __alloc_contig_migrate_range(struct compact_control *cc,
6341 unsigned long start, unsigned long end)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006342{
6343 /* This function is based on compact_zone() from compaction.c. */
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006344 unsigned long nr_reclaimed;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006345 unsigned long pfn = start;
6346 unsigned int tries = 0;
6347 int ret = 0;
6348
Marek Szyprowskibe49a6e2012-12-12 13:51:19 -08006349 migrate_prep();
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006350
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006351 while (pfn < end || !list_empty(&cc->migratepages)) {
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006352 if (fatal_signal_pending(current)) {
6353 ret = -EINTR;
6354 break;
6355 }
6356
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006357 if (list_empty(&cc->migratepages)) {
6358 cc->nr_migratepages = 0;
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07006359 pfn = isolate_migratepages_range(cc, pfn, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006360 if (!pfn) {
6361 ret = -EINTR;
6362 break;
6363 }
6364 tries = 0;
6365 } else if (++tries == 5) {
6366 ret = ret < 0 ? ret : -EBUSY;
6367 break;
6368 }
6369
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006370 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
6371 &cc->migratepages);
6372 cc->nr_migratepages -= nr_reclaimed;
Minchan Kim02c6de82012-10-08 16:31:55 -07006373
Hugh Dickins9c620e22013-02-22 16:35:14 -08006374 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
David Rientjese0b9dae2014-06-04 16:08:28 -07006375 NULL, 0, cc->mode, MR_CMA);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006376 }
Srinivas Pandruvada2a6f5122013-02-22 16:32:09 -08006377 if (ret < 0) {
6378 putback_movable_pages(&cc->migratepages);
6379 return ret;
6380 }
6381 return 0;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006382}
6383
6384/**
6385 * alloc_contig_range() -- tries to allocate given range of pages
6386 * @start: start PFN to allocate
6387 * @end: one-past-the-last PFN to allocate
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006388 * @migratetype: migratetype of the underlaying pageblocks (either
6389 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6390 * in range must have the same migratetype and it must
6391 * be either of the two.
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006392 *
6393 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6394 * aligned, however it's the caller's responsibility to guarantee that
6395 * we are the only thread that changes migrate type of pageblocks the
6396 * pages fall in.
6397 *
6398 * The PFN range must belong to a single zone.
6399 *
6400 * Returns zero on success or negative error code. On success all
6401 * pages which PFN is in [start, end) are allocated for the caller and
6402 * need to be freed with free_contig_range().
6403 */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006404int alloc_contig_range(unsigned long start, unsigned long end,
6405 unsigned migratetype)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006406{
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006407 unsigned long outer_start, outer_end;
6408 int ret = 0, order;
6409
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006410 struct compact_control cc = {
6411 .nr_migratepages = 0,
6412 .order = -1,
6413 .zone = page_zone(pfn_to_page(start)),
David Rientjese0b9dae2014-06-04 16:08:28 -07006414 .mode = MIGRATE_SYNC,
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006415 .ignore_skip_hint = true,
6416 };
6417 INIT_LIST_HEAD(&cc.migratepages);
6418
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006419 /*
6420 * What we do here is we mark all pageblocks in range as
6421 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6422 * have different sizes, and due to the way page allocator
6423 * work, we align the range to biggest of the two pages so
6424 * that page allocator won't try to merge buddies from
6425 * different pageblocks and change MIGRATE_ISOLATE to some
6426 * other migration type.
6427 *
6428 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6429 * migrate the pages from an unaligned range (ie. pages that
6430 * we are interested in). This will put all the pages in
6431 * range back to page allocator as MIGRATE_ISOLATE.
6432 *
6433 * When this is done, we take the pages in range from page
6434 * allocator removing them from the buddy system. This way
6435 * page allocator will never consider using them.
6436 *
6437 * This lets us mark the pageblocks back as
6438 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6439 * aligned range but not in the unaligned, original range are
6440 * put back to page allocator so that buddy can use them.
6441 */
6442
6443 ret = start_isolate_page_range(pfn_max_align_down(start),
Wen Congyangb023f462012-12-11 16:00:45 -08006444 pfn_max_align_up(end), migratetype,
6445 false);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006446 if (ret)
Bob Liu86a595f2012-10-25 13:37:56 -07006447 return ret;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006448
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006449 ret = __alloc_contig_migrate_range(&cc, start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006450 if (ret)
6451 goto done;
6452
6453 /*
6454 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6455 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
6456 * more, all pages in [start, end) are free in page allocator.
6457 * What we are going to do is to allocate all pages from
6458 * [start, end) (that is remove them from page allocator).
6459 *
6460 * The only problem is that pages at the beginning and at the
6461 * end of interesting range may be not aligned with pages that
6462 * page allocator holds, ie. they can be part of higher order
6463 * pages. Because of this, we reserve the bigger range and
6464 * once this is done free the pages we are not interested in.
6465 *
6466 * We don't have to hold zone->lock here because the pages are
6467 * isolated thus they won't get removed from buddy.
6468 */
6469
6470 lru_add_drain_all();
Vlastimil Babka510f5502014-12-10 15:43:07 -08006471 drain_all_pages(cc.zone);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006472
6473 order = 0;
6474 outer_start = start;
6475 while (!PageBuddy(pfn_to_page(outer_start))) {
6476 if (++order >= MAX_ORDER) {
6477 ret = -EBUSY;
6478 goto done;
6479 }
6480 outer_start &= ~0UL << order;
6481 }
6482
6483 /* Make sure the range is really isolated. */
Wen Congyangb023f462012-12-11 16:00:45 -08006484 if (test_pages_isolated(outer_start, end, false)) {
Michal Nazarewiczdae803e2014-11-13 15:19:27 -08006485 pr_info("%s: [%lx, %lx) PFNs busy\n",
6486 __func__, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006487 ret = -EBUSY;
6488 goto done;
6489 }
6490
Marek Szyprowski49f223a2012-01-25 12:49:24 +01006491 /* Grab isolated pages from freelists. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006492 outer_end = isolate_freepages_range(&cc, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006493 if (!outer_end) {
6494 ret = -EBUSY;
6495 goto done;
6496 }
6497
6498 /* Free head and tail (if any) */
6499 if (start != outer_start)
6500 free_contig_range(outer_start, start - outer_start);
6501 if (end != outer_end)
6502 free_contig_range(end, outer_end - end);
6503
6504done:
6505 undo_isolate_page_range(pfn_max_align_down(start),
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006506 pfn_max_align_up(end), migratetype);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006507 return ret;
6508}
6509
6510void free_contig_range(unsigned long pfn, unsigned nr_pages)
6511{
Marek Szyprowskibcc2b022012-12-20 15:05:18 -08006512 unsigned int count = 0;
6513
6514 for (; nr_pages--; pfn++) {
6515 struct page *page = pfn_to_page(pfn);
6516
6517 count += page_count(page) != 1;
6518 __free_page(page);
6519 }
6520 WARN(count != 0, "%d pages are still in use!\n", count);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006521}
6522#endif
6523
Jiang Liu4ed7e022012-07-31 16:43:35 -07006524#ifdef CONFIG_MEMORY_HOTPLUG
Cody P Schafer0a647f32013-07-03 15:01:33 -07006525/*
6526 * The zone indicated has a new number of managed_pages; batch sizes and percpu
6527 * page high values need to be recalulated.
6528 */
Jiang Liu4ed7e022012-07-31 16:43:35 -07006529void __meminit zone_pcp_update(struct zone *zone)
6530{
Cody P Schafer0a647f32013-07-03 15:01:33 -07006531 unsigned cpu;
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006532 mutex_lock(&pcp_batch_high_lock);
Cody P Schafer0a647f32013-07-03 15:01:33 -07006533 for_each_possible_cpu(cpu)
Cody P Schafer169f6c12013-07-03 15:01:41 -07006534 pageset_set_high_and_batch(zone,
6535 per_cpu_ptr(zone->pageset, cpu));
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006536 mutex_unlock(&pcp_batch_high_lock);
Jiang Liu4ed7e022012-07-31 16:43:35 -07006537}
6538#endif
6539
Jiang Liu340175b2012-07-31 16:43:32 -07006540void zone_pcp_reset(struct zone *zone)
6541{
6542 unsigned long flags;
Minchan Kim5a883812012-10-08 16:33:39 -07006543 int cpu;
6544 struct per_cpu_pageset *pset;
Jiang Liu340175b2012-07-31 16:43:32 -07006545
6546 /* avoid races with drain_pages() */
6547 local_irq_save(flags);
6548 if (zone->pageset != &boot_pageset) {
Minchan Kim5a883812012-10-08 16:33:39 -07006549 for_each_online_cpu(cpu) {
6550 pset = per_cpu_ptr(zone->pageset, cpu);
6551 drain_zonestat(zone, pset);
6552 }
Jiang Liu340175b2012-07-31 16:43:32 -07006553 free_percpu(zone->pageset);
6554 zone->pageset = &boot_pageset;
6555 }
6556 local_irq_restore(flags);
6557}
6558
Wen Congyang6dcd73d2012-12-11 16:01:01 -08006559#ifdef CONFIG_MEMORY_HOTREMOVE
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006560/*
6561 * All pages in the range must be isolated before calling this.
6562 */
6563void
6564__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
6565{
6566 struct page *page;
6567 struct zone *zone;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07006568 unsigned int order, i;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006569 unsigned long pfn;
6570 unsigned long flags;
6571 /* find the first valid pfn */
6572 for (pfn = start_pfn; pfn < end_pfn; pfn++)
6573 if (pfn_valid(pfn))
6574 break;
6575 if (pfn == end_pfn)
6576 return;
6577 zone = page_zone(pfn_to_page(pfn));
6578 spin_lock_irqsave(&zone->lock, flags);
6579 pfn = start_pfn;
6580 while (pfn < end_pfn) {
6581 if (!pfn_valid(pfn)) {
6582 pfn++;
6583 continue;
6584 }
6585 page = pfn_to_page(pfn);
Wen Congyangb023f462012-12-11 16:00:45 -08006586 /*
6587 * The HWPoisoned page may be not in buddy system, and
6588 * page_count() is not 0.
6589 */
6590 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
6591 pfn++;
6592 SetPageReserved(page);
6593 continue;
6594 }
6595
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006596 BUG_ON(page_count(page));
6597 BUG_ON(!PageBuddy(page));
6598 order = page_order(page);
6599#ifdef CONFIG_DEBUG_VM
6600 printk(KERN_INFO "remove from free list %lx %d %lx\n",
6601 pfn, 1 << order, end_pfn);
6602#endif
6603 list_del(&page->lru);
6604 rmv_page_order(page);
6605 zone->free_area[order].nr_free--;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006606 for (i = 0; i < (1 << order); i++)
6607 SetPageReserved((page+i));
6608 pfn += (1 << order);
6609 }
6610 spin_unlock_irqrestore(&zone->lock, flags);
6611}
6612#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01006613
6614#ifdef CONFIG_MEMORY_FAILURE
6615bool is_free_buddy_page(struct page *page)
6616{
6617 struct zone *zone = page_zone(page);
6618 unsigned long pfn = page_to_pfn(page);
6619 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07006620 unsigned int order;
Wu Fengguang8d22ba12009-12-16 12:19:58 +01006621
6622 spin_lock_irqsave(&zone->lock, flags);
6623 for (order = 0; order < MAX_ORDER; order++) {
6624 struct page *page_head = page - (pfn & ((1 << order) - 1));
6625
6626 if (PageBuddy(page_head) && page_order(page_head) >= order)
6627 break;
6628 }
6629 spin_unlock_irqrestore(&zone->lock, flags);
6630
6631 return order < MAX_ORDER;
6632}
6633#endif