blob: 7abfa70cdc1ae8767fd663d905372447a7fe6864 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/page_alloc.c
3 *
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
6 *
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
15 */
16
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/stddef.h>
18#include <linux/mm.h>
19#include <linux/swap.h>
20#include <linux/interrupt.h>
21#include <linux/pagemap.h>
KOSAKI Motohiro10ed2732008-03-04 14:28:32 -080022#include <linux/jiffies.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/bootmem.h>
Yinghai Luedbe7d22010-08-25 13:39:16 -070024#include <linux/memblock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/compiler.h>
Randy Dunlap9f158332005-09-13 01:25:16 -070026#include <linux/kernel.h>
Vegard Nossumb1eeab62008-11-25 16:55:53 +010027#include <linux/kmemcheck.h>
Andrey Ryabininb8c73fc2015-02-13 14:39:28 -080028#include <linux/kasan.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/module.h>
30#include <linux/suspend.h>
31#include <linux/pagevec.h>
32#include <linux/blkdev.h>
33#include <linux/slab.h>
Dave Hansena238ab52011-05-24 17:12:16 -070034#include <linux/ratelimit.h>
David Rientjes5a3135c22007-10-16 23:25:53 -070035#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <linux/notifier.h>
37#include <linux/topology.h>
38#include <linux/sysctl.h>
39#include <linux/cpu.h>
40#include <linux/cpuset.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070041#include <linux/memory_hotplug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include <linux/nodemask.h>
43#include <linux/vmalloc.h>
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -070044#include <linux/vmstat.h>
Christoph Lameter4be38e32006-01-06 00:11:17 -080045#include <linux/mempolicy.h>
Yasunori Goto68113782006-06-23 02:03:11 -070046#include <linux/stop_machine.h>
Mel Gormanc7132162006-09-27 01:49:43 -070047#include <linux/sort.h>
48#include <linux/pfn.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070049#include <linux/backing-dev.h>
Akinobu Mita933e3122006-12-08 02:39:45 -080050#include <linux/fault-inject.h>
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -070051#include <linux/page-isolation.h>
Joonsoo Kimeefa864b2014-12-12 16:55:46 -080052#include <linux/page_ext.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070053#include <linux/debugobjects.h>
Catalin Marinasdbb1f812009-06-11 13:23:19 +010054#include <linux/kmemleak.h>
Mel Gorman56de7262010-05-24 14:32:30 -070055#include <linux/compaction.h>
Mel Gorman0d3d0622009-09-21 17:02:44 -070056#include <trace/events/kmem.h>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070057#include <linux/prefetch.h>
Lisa Du6e543d52013-09-11 14:22:36 -070058#include <linux/mm_inline.h>
Michal Nazarewicz041d3a82011-12-29 13:09:50 +010059#include <linux/migrate.h>
Joonsoo Kime30825f2014-12-12 16:55:49 -080060#include <linux/page_ext.h>
David Rientjes949f7ec2013-04-29 15:07:48 -070061#include <linux/hugetlb.h>
Clark Williams8bd75c72013-02-07 09:47:07 -060062#include <linux/sched/rt.h>
Joonsoo Kim48c96a32014-12-12 16:56:01 -080063#include <linux/page_owner.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
Jiang Liu7ee3d4e2013-07-03 15:03:41 -070065#include <asm/sections.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070066#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070067#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070068#include "internal.h"
69
Cody P Schaferc8e251f2013-07-03 15:01:29 -070070/* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
71static DEFINE_MUTEX(pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -070072#define MIN_PERCPU_PAGELIST_FRACTION (8)
Cody P Schaferc8e251f2013-07-03 15:01:29 -070073
Lee Schermerhorn72812012010-05-26 14:44:56 -070074#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
75DEFINE_PER_CPU(int, numa_node);
76EXPORT_PER_CPU_SYMBOL(numa_node);
77#endif
78
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070079#ifdef CONFIG_HAVE_MEMORYLESS_NODES
80/*
81 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
82 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
83 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
84 * defined in <linux/topology.h>.
85 */
86DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
87EXPORT_PER_CPU_SYMBOL(_numa_mem_);
Joonsoo Kimad2c8142014-10-09 15:26:13 -070088int _node_numa_mem_[MAX_NUMNODES];
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070089#endif
90
Linus Torvalds1da177e2005-04-16 15:20:36 -070091/*
Christoph Lameter13808912007-10-16 01:25:27 -070092 * Array of node states.
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 */
Christoph Lameter13808912007-10-16 01:25:27 -070094nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
95 [N_POSSIBLE] = NODE_MASK_ALL,
96 [N_ONLINE] = { { [0] = 1UL } },
97#ifndef CONFIG_NUMA
98 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
99#ifdef CONFIG_HIGHMEM
100 [N_HIGH_MEMORY] = { { [0] = 1UL } },
101#endif
Lai Jiangshan20b2f522012-12-12 13:52:00 -0800102#ifdef CONFIG_MOVABLE_NODE
103 [N_MEMORY] = { { [0] = 1UL } },
104#endif
Christoph Lameter13808912007-10-16 01:25:27 -0700105 [N_CPU] = { { [0] = 1UL } },
106#endif /* NUMA */
107};
108EXPORT_SYMBOL(node_states);
109
Jiang Liuc3d5f5f2013-07-03 15:03:14 -0700110/* Protect totalram_pages and zone->managed_pages */
111static DEFINE_SPINLOCK(managed_page_count_lock);
112
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700113unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -0700114unsigned long totalreserve_pages __read_mostly;
Pintu Kumare48322a2014-12-18 16:17:15 -0800115unsigned long totalcma_pages __read_mostly;
Johannes Weinerab8fabd2012-01-10 15:07:42 -0800116/*
117 * When calculating the number of globally allowed dirty pages, there
118 * is a certain number of per-zone reserves that should not be
119 * considered dirtyable memory. This is the sum of those reserves
120 * over all existing zones that contribute dirtyable memory.
121 */
122unsigned long dirty_balance_reserve __read_mostly;
123
Hugh Dickins1b76b022012-05-11 01:00:07 -0700124int percpu_pagelist_fraction;
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000125gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800127#ifdef CONFIG_PM_SLEEP
128/*
129 * The following functions are used by the suspend/hibernate code to temporarily
130 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
131 * while devices are suspended. To avoid races with the suspend/hibernate code,
132 * they should always be called with pm_mutex held (gfp_allowed_mask also should
133 * only be modified with pm_mutex held, unless the suspend/hibernate code is
134 * guaranteed not to run in parallel with that modification).
135 */
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100136
137static gfp_t saved_gfp_mask;
138
139void pm_restore_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800140{
141 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100142 if (saved_gfp_mask) {
143 gfp_allowed_mask = saved_gfp_mask;
144 saved_gfp_mask = 0;
145 }
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800146}
147
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100148void pm_restrict_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800149{
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800150 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100151 WARN_ON(saved_gfp_mask);
152 saved_gfp_mask = gfp_allowed_mask;
153 gfp_allowed_mask &= ~GFP_IOFS;
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800154}
Mel Gormanf90ac392012-01-10 15:07:15 -0800155
156bool pm_suspended_storage(void)
157{
158 if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS)
159 return false;
160 return true;
161}
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800162#endif /* CONFIG_PM_SLEEP */
163
Mel Gormand9c23402007-10-16 01:26:01 -0700164#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
165int pageblock_order __read_mostly;
166#endif
167
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800168static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -0800169
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170/*
171 * results with 256, 32 in the lowmem_reserve sysctl:
172 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
173 * 1G machine -> (16M dma, 784M normal, 224M high)
174 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
175 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
Yaowei Bai84109e12015-02-12 15:00:22 -0800176 * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +0100177 *
178 * TBD: should special case ZONE_DMA32 machines here - in those we normally
179 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700181int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800182#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700183 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -0800184#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700185#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700186 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700187#endif
Christoph Lametere53ef382006-09-25 23:31:14 -0700188#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700189 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700190#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700191 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700192};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
194EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195
Helge Deller15ad7cd2006-12-06 20:40:36 -0800196static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800197#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700198 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800199#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700200#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700201 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700202#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700203 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700204#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700205 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700206#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700207 "Movable",
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700208};
209
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210int min_free_kbytes = 1024;
Han Pingtian42aa83c2014-01-23 15:53:28 -0800211int user_min_free_kbytes = -1;
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 Gruszkac0a32fc2012-01-10 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 Gruszkac0a32fc2012-01-10 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 Gruszkac0a32fc2012-01-10 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 Gruszkac0a32fc2012-01-10 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 Gruszkac0a32fc2012-01-10 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 Gruszkac0a32fc2012-01-10 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 Gruszkac0a32fc2012-01-10 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 Gruszkac0a32fc2012-01-10 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 Gruszkac0a32fc2012-01-10 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 Gruszkac0a32fc2012-01-10 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 Gruszkac0a32fc2012-01-10 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 Gormana6f9edd62009-09-21 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 Gormana6f9edd62009-09-21 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 Gormana6f9edd62009-09-21 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 Gormana6f9edd62009-09-21 17:03:20 -0700729 do {
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -0700730 int mt; /* migratetype of the to-be-freed page */
731
Mel Gormana6f9edd62009-09-21 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 Gruszkac0a32fc2012-01-10 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 Gruszkac0a32fc2012-01-10 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 Gruszkac0a32fc2012-01-10 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 },
1035#ifdef CONFIG_CMA
1036 [MIGRATE_MOVABLE] = { MIGRATE_CMA, MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
1037 [MIGRATE_CMA] = { MIGRATE_RESERVE }, /* Never used */
1038#else
1039 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
1040#endif
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001041 [MIGRATE_RESERVE] = { MIGRATE_RESERVE }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -08001042#ifdef CONFIG_MEMORY_ISOLATION
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001043 [MIGRATE_ISOLATE] = { MIGRATE_RESERVE }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -08001044#endif
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001045};
1046
Mel Gormanc361be52007-10-16 01:25:51 -07001047/*
1048 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -07001049 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -07001050 * boundary. If alignment is required, use move_freepages_block()
1051 */
Minchan Kim435b4052012-10-08 16:32:16 -07001052int move_freepages(struct zone *zone,
Adrian Bunkb69a7282008-07-23 21:28:12 -07001053 struct page *start_page, struct page *end_page,
1054 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001055{
1056 struct page *page;
1057 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -07001058 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -07001059
1060#ifndef CONFIG_HOLES_IN_ZONE
1061 /*
1062 * page_zone is not safe to call in this context when
1063 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1064 * anyway as we check zone boundaries in move_freepages_block().
1065 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -07001066 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -07001067 */
Mel Gorman97ee4ba2014-10-09 15:28:28 -07001068 VM_BUG_ON(page_zone(start_page) != page_zone(end_page));
Mel Gormanc361be52007-10-16 01:25:51 -07001069#endif
1070
1071 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -07001072 /* Make sure we are not inadvertently changing nodes */
Sasha Levin309381fea2014-01-23 15:52:54 -08001073 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
Adam Litke344c7902008-09-02 14:35:38 -07001074
Mel Gormanc361be52007-10-16 01:25:51 -07001075 if (!pfn_valid_within(page_to_pfn(page))) {
1076 page++;
1077 continue;
1078 }
1079
1080 if (!PageBuddy(page)) {
1081 page++;
1082 continue;
1083 }
1084
1085 order = page_order(page);
Kirill A. Shutemov84be48d2011-03-22 16:33:41 -07001086 list_move(&page->lru,
1087 &zone->free_area[order].free_list[migratetype]);
Minchan Kim95e34412012-10-08 16:32:11 -07001088 set_freepage_migratetype(page, migratetype);
Mel Gormanc361be52007-10-16 01:25:51 -07001089 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -07001090 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -07001091 }
1092
Mel Gormand1003132007-10-16 01:26:00 -07001093 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -07001094}
1095
Minchan Kimee6f5092012-07-31 16:43:50 -07001096int move_freepages_block(struct zone *zone, struct page *page,
Linus Torvalds68e3e922012-06-03 20:05:57 -07001097 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001098{
1099 unsigned long start_pfn, end_pfn;
1100 struct page *start_page, *end_page;
1101
1102 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -07001103 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -07001104 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -07001105 end_page = start_page + pageblock_nr_pages - 1;
1106 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -07001107
1108 /* Do not cross zone boundaries */
Cody P Schafer108bcc92013-02-22 16:35:23 -08001109 if (!zone_spans_pfn(zone, start_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001110 start_page = page;
Cody P Schafer108bcc92013-02-22 16:35:23 -08001111 if (!zone_spans_pfn(zone, end_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001112 return 0;
1113
1114 return move_freepages(zone, start_page, end_page, migratetype);
1115}
1116
Mel Gorman2f66a682009-09-21 17:02:31 -07001117static void change_pageblock_range(struct page *pageblock_page,
1118 int start_order, int migratetype)
1119{
1120 int nr_pageblocks = 1 << (start_order - pageblock_order);
1121
1122 while (nr_pageblocks--) {
1123 set_pageblock_migratetype(pageblock_page, migratetype);
1124 pageblock_page += pageblock_nr_pages;
1125 }
1126}
1127
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001128/*
Vlastimil Babka9c0415e2015-02-11 15:28:21 -08001129 * When we are falling back to another migratetype during allocation, try to
1130 * steal extra free pages from the same pageblocks to satisfy further
1131 * allocations, instead of polluting multiple pageblocks.
1132 *
1133 * If we are stealing a relatively large buddy page, it is likely there will
1134 * be more free pages in the pageblock, so try to steal them all. For
1135 * reclaimable and unmovable allocations, we steal regardless of page size,
1136 * as fragmentation caused by those allocations polluting movable pageblocks
1137 * is worse than movable allocations stealing from unmovable and reclaimable
1138 * pageblocks.
1139 *
1140 * If we claim more than half of the pageblock, change pageblock's migratetype
1141 * as well.
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001142 */
Vlastimil Babka3a1086f2015-02-11 15:28:18 -08001143static void try_to_steal_freepages(struct zone *zone, struct page *page,
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001144 int start_type, int fallback_type)
1145{
1146 int current_order = page_order(page);
1147
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001148 /* Take ownership for orders >= pageblock_order */
1149 if (current_order >= pageblock_order) {
1150 change_pageblock_range(page, current_order, start_type);
Vlastimil Babka3a1086f2015-02-11 15:28:18 -08001151 return;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001152 }
1153
1154 if (current_order >= pageblock_order / 2 ||
1155 start_type == MIGRATE_RECLAIMABLE ||
Vlastimil Babka9c0415e2015-02-11 15:28:21 -08001156 start_type == MIGRATE_UNMOVABLE ||
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001157 page_group_by_mobility_disabled) {
1158 int pages;
1159
1160 pages = move_freepages_block(zone, page, start_type);
1161
1162 /* Claim the whole block if over half of it is free */
1163 if (pages >= (1 << (pageblock_order-1)) ||
Vlastimil Babka99592d52015-02-11 15:28:15 -08001164 page_group_by_mobility_disabled)
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001165 set_pageblock_migratetype(page, start_type);
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001166 }
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001167}
1168
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001169/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -07001170static inline struct page *
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001171__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001172{
Pintu Kumarb8af2942013-09-11 14:20:34 -07001173 struct free_area *area;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001174 unsigned int current_order;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001175 struct page *page;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001176
1177 /* Find the largest possible block of pages in the other list */
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001178 for (current_order = MAX_ORDER-1;
1179 current_order >= order && current_order <= MAX_ORDER-1;
1180 --current_order) {
Vlastimil Babka3a1086f2015-02-11 15:28:18 -08001181 int i;
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001182 for (i = 0;; i++) {
Vlastimil Babka3a1086f2015-02-11 15:28:18 -08001183 int migratetype = fallbacks[start_migratetype][i];
1184 int buddy_type = start_migratetype;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001185
Mel Gorman56fd56b2007-10-16 01:25:58 -07001186 /* MIGRATE_RESERVE handled later if necessary */
1187 if (migratetype == MIGRATE_RESERVE)
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001188 break;
Mel Gormane0104872007-10-16 01:25:53 -07001189
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001190 area = &(zone->free_area[current_order]);
1191 if (list_empty(&area->free_list[migratetype]))
1192 continue;
1193
1194 page = list_entry(area->free_list[migratetype].next,
1195 struct page, lru);
1196 area->nr_free--;
1197
Vlastimil Babka3a1086f2015-02-11 15:28:18 -08001198 if (!is_migrate_cma(migratetype)) {
1199 try_to_steal_freepages(zone, page,
1200 start_migratetype,
1201 migratetype);
1202 } else {
1203 /*
1204 * When borrowing from MIGRATE_CMA, we need to
1205 * release the excess buddy pages to CMA
1206 * itself, and we do not try to steal extra
1207 * free pages.
1208 */
1209 buddy_type = migratetype;
1210 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001211
1212 /* Remove the page from the freelists */
1213 list_del(&page->lru);
1214 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001215
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001216 expand(zone, page, order, current_order, area,
Vlastimil Babka3a1086f2015-02-11 15:28:18 -08001217 buddy_type);
1218
1219 /*
1220 * The freepage_migratetype may differ from pageblock's
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001221 * migratetype depending on the decisions in
Vlastimil Babka3a1086f2015-02-11 15:28:18 -08001222 * try_to_steal_freepages(). This is OK as long as it
1223 * does not differ for MIGRATE_CMA pageblocks. For CMA
1224 * we need to make sure unallocated pages flushed from
1225 * pcp lists are returned to the correct freelist.
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001226 */
Vlastimil Babka3a1086f2015-02-11 15:28:18 -08001227 set_freepage_migratetype(page, buddy_type);
Mel Gormane0fff1b2009-09-21 17:02:42 -07001228
KOSAKI Motohiro52c8f6a2013-11-12 15:08:19 -08001229 trace_mm_page_alloc_extfrag(page, order, current_order,
Vlastimil Babka99592d52015-02-11 15:28:15 -08001230 start_migratetype, migratetype);
Mel Gormane0fff1b2009-09-21 17:02:42 -07001231
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001232 return page;
1233 }
1234 }
1235
Mel Gorman728ec982009-06-16 15:32:04 -07001236 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001237}
1238
Mel Gorman56fd56b2007-10-16 01:25:58 -07001239/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 * Do the hard work of removing an element from the buddy allocator.
1241 * Call me with the zone->lock already held.
1242 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001243static struct page *__rmqueue(struct zone *zone, unsigned int order,
1244 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 struct page *page;
1247
Mel Gorman728ec982009-06-16 15:32:04 -07001248retry_reserve:
Mel Gorman56fd56b2007-10-16 01:25:58 -07001249 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250
Mel Gorman728ec982009-06-16 15:32:04 -07001251 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07001252 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001253
Mel Gorman728ec982009-06-16 15:32:04 -07001254 /*
1255 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1256 * is used because __rmqueue_smallest is an inline function
1257 * and we want just one call site
1258 */
1259 if (!page) {
1260 migratetype = MIGRATE_RESERVE;
1261 goto retry_reserve;
1262 }
1263 }
1264
Mel Gorman0d3d0622009-09-21 17:02:44 -07001265 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001266 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267}
1268
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001269/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 * Obtain a specified number of elements from the buddy allocator, all under
1271 * a single hold of the lock, for efficiency. Add them to the supplied list.
1272 * Returns the number of new pages which were placed at *list.
1273 */
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001274static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001275 unsigned long count, struct list_head *list,
Mel Gormanb745bc82014-06-04 16:10:22 -07001276 int migratetype, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277{
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001278 int i;
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001279
Nick Pigginc54ad302006-01-06 00:10:56 -08001280 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001282 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001283 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08001285
1286 /*
1287 * Split buddy pages returned by expand() are received here
1288 * in physical page order. The page is added to the callers and
1289 * list and the list head then moves forward. From the callers
1290 * perspective, the linked list is ordered by page number in
1291 * some conditions. This is useful for IO devices that can
1292 * merge IO requests if the physical pages are ordered
1293 * properly.
1294 */
Mel Gormanb745bc82014-06-04 16:10:22 -07001295 if (likely(!cold))
Mel Gormane084b2d2009-07-29 15:02:04 -07001296 list_add(&page->lru, list);
1297 else
1298 list_add_tail(&page->lru, list);
Mel Gorman81eabcb2007-12-17 16:20:05 -08001299 list = &page->lru;
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001300 if (is_migrate_cma(get_freepage_migratetype(page)))
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07001301 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
1302 -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 }
Mel Gormanf2260e62009-06-16 15:32:13 -07001304 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08001305 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001306 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307}
1308
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001309#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001310/*
Christoph Lameter4037d452007-05-09 02:35:14 -07001311 * Called from the vmstat counter updater to drain pagesets of this
1312 * currently executing processor on remote nodes after they have
1313 * expired.
1314 *
Christoph Lameter879336c2006-03-22 00:09:08 -08001315 * Note that this function must be called with the thread pinned to
1316 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001317 */
Christoph Lameter4037d452007-05-09 02:35:14 -07001318void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001319{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001320 unsigned long flags;
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07001321 int to_drain, batch;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001322
Christoph Lameter4037d452007-05-09 02:35:14 -07001323 local_irq_save(flags);
Cody P Schafer998d39c2013-07-03 15:01:32 -07001324 batch = ACCESS_ONCE(pcp->batch);
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07001325 to_drain = min(pcp->count, batch);
KOSAKI Motohiro2a135152012-07-31 16:42:53 -07001326 if (to_drain > 0) {
1327 free_pcppages_bulk(zone, to_drain, pcp);
1328 pcp->count -= to_drain;
1329 }
Christoph Lameter4037d452007-05-09 02:35:14 -07001330 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001331}
1332#endif
1333
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001334/*
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001335 * Drain pcplists of the indicated processor and zone.
1336 *
1337 * The processor must either be the current processor and the
1338 * thread pinned to the current processor or a processor that
1339 * is not online.
1340 */
1341static void drain_pages_zone(unsigned int cpu, struct zone *zone)
1342{
1343 unsigned long flags;
1344 struct per_cpu_pageset *pset;
1345 struct per_cpu_pages *pcp;
1346
1347 local_irq_save(flags);
1348 pset = per_cpu_ptr(zone->pageset, cpu);
1349
1350 pcp = &pset->pcp;
1351 if (pcp->count) {
1352 free_pcppages_bulk(zone, pcp->count, pcp);
1353 pcp->count = 0;
1354 }
1355 local_irq_restore(flags);
1356}
1357
1358/*
1359 * Drain pcplists of all zones on the indicated processor.
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001360 *
1361 * The processor must either be the current processor and the
1362 * thread pinned to the current processor or a processor that
1363 * is not online.
1364 */
1365static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366{
1367 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07001369 for_each_populated_zone(zone) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001370 drain_pages_zone(cpu, zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 }
1372}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001374/*
1375 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001376 *
1377 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
1378 * the single zone's pages.
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001379 */
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001380void drain_local_pages(struct zone *zone)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001381{
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001382 int cpu = smp_processor_id();
1383
1384 if (zone)
1385 drain_pages_zone(cpu, zone);
1386 else
1387 drain_pages(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001388}
1389
1390/*
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001391 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1392 *
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001393 * When zone parameter is non-NULL, spill just the single zone's pages.
1394 *
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001395 * Note that this code is protected against sending an IPI to an offline
1396 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1397 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1398 * nothing keeps CPUs from showing up after we populated the cpumask and
1399 * before the call to on_each_cpu_mask().
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001400 */
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001401void drain_all_pages(struct zone *zone)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001402{
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001403 int cpu;
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001404
1405 /*
1406 * Allocate in the BSS so we wont require allocation in
1407 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1408 */
1409 static cpumask_t cpus_with_pcps;
1410
1411 /*
1412 * We don't care about racing with CPU hotplug event
1413 * as offline notification will cause the notified
1414 * cpu to drain that CPU pcps and on_each_cpu_mask
1415 * disables preemption as part of its processing
1416 */
1417 for_each_online_cpu(cpu) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001418 struct per_cpu_pageset *pcp;
1419 struct zone *z;
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001420 bool has_pcps = false;
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001421
1422 if (zone) {
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001423 pcp = per_cpu_ptr(zone->pageset, cpu);
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001424 if (pcp->pcp.count)
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001425 has_pcps = true;
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001426 } else {
1427 for_each_populated_zone(z) {
1428 pcp = per_cpu_ptr(z->pageset, cpu);
1429 if (pcp->pcp.count) {
1430 has_pcps = true;
1431 break;
1432 }
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001433 }
1434 }
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001435
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001436 if (has_pcps)
1437 cpumask_set_cpu(cpu, &cpus_with_pcps);
1438 else
1439 cpumask_clear_cpu(cpu, &cpus_with_pcps);
1440 }
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001441 on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages,
1442 zone, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001443}
1444
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02001445#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446
1447void mark_free_pages(struct zone *zone)
1448{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001449 unsigned long pfn, max_zone_pfn;
1450 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001451 unsigned int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 struct list_head *curr;
1453
Xishi Qiu8080fc02013-09-11 14:21:45 -07001454 if (zone_is_empty(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 return;
1456
1457 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001458
Cody P Schafer108bcc92013-02-22 16:35:23 -08001459 max_zone_pfn = zone_end_pfn(zone);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001460 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1461 if (pfn_valid(pfn)) {
1462 struct page *page = pfn_to_page(pfn);
1463
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001464 if (!swsusp_page_is_forbidden(page))
1465 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001466 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001468 for_each_migratetype_order(order, t) {
1469 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001470 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001472 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1473 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001474 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001475 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001476 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 spin_unlock_irqrestore(&zone->lock, flags);
1478}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001479#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480
1481/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 * Free a 0-order page
Mel Gormanb745bc82014-06-04 16:10:22 -07001483 * cold == true ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 */
Mel Gormanb745bc82014-06-04 16:10:22 -07001485void free_hot_cold_page(struct page *page, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486{
1487 struct zone *zone = page_zone(page);
1488 struct per_cpu_pages *pcp;
1489 unsigned long flags;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001490 unsigned long pfn = page_to_pfn(page);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001491 int migratetype;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001493 if (!free_pages_prepare(page, 0))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001494 return;
1495
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001496 migratetype = get_pfnblock_migratetype(page, pfn);
Minchan Kimb12c4ad2012-10-08 16:32:08 -07001497 set_freepage_migratetype(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001499 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001500
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001501 /*
1502 * We only track unmovable, reclaimable and movable on pcp lists.
1503 * Free ISOLATE pages back to the allocator because they are being
1504 * offlined but treat RESERVE as movable pages so we can get those
1505 * areas back if necessary. Otherwise, we may have to free
1506 * excessively into the page allocator
1507 */
1508 if (migratetype >= MIGRATE_PCPTYPES) {
Minchan Kim194159f2013-02-22 16:33:58 -08001509 if (unlikely(is_migrate_isolate(migratetype))) {
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001510 free_one_page(zone, page, pfn, 0, migratetype);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001511 goto out;
1512 }
1513 migratetype = MIGRATE_MOVABLE;
1514 }
1515
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001516 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Mel Gormanb745bc82014-06-04 16:10:22 -07001517 if (!cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001518 list_add(&page->lru, &pcp->lists[migratetype]);
Mel Gormanb745bc82014-06-04 16:10:22 -07001519 else
1520 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001522 if (pcp->count >= pcp->high) {
Cody P Schafer998d39c2013-07-03 15:01:32 -07001523 unsigned long batch = ACCESS_ONCE(pcp->batch);
1524 free_pcppages_bulk(zone, batch, pcp);
1525 pcp->count -= batch;
Nick Piggin48db57f2006-01-08 01:00:42 -08001526 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001527
1528out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530}
1531
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001532/*
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001533 * Free a list of 0-order pages
1534 */
Mel Gormanb745bc82014-06-04 16:10:22 -07001535void free_hot_cold_page_list(struct list_head *list, bool cold)
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001536{
1537 struct page *page, *next;
1538
1539 list_for_each_entry_safe(page, next, list, lru) {
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -08001540 trace_mm_page_free_batched(page, cold);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001541 free_hot_cold_page(page, cold);
1542 }
1543}
1544
1545/*
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001546 * split_page takes a non-compound higher-order page, and splits it into
1547 * n (1<<order) sub-pages: page[0..n]
1548 * Each sub-page must be freed individually.
1549 *
1550 * Note: this is probably too low level an operation for use in drivers.
1551 * Please consult with lkml before using this in your driver.
1552 */
1553void split_page(struct page *page, unsigned int order)
1554{
1555 int i;
1556
Sasha Levin309381fea2014-01-23 15:52:54 -08001557 VM_BUG_ON_PAGE(PageCompound(page), page);
1558 VM_BUG_ON_PAGE(!page_count(page), page);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001559
1560#ifdef CONFIG_KMEMCHECK
1561 /*
1562 * Split shadow pages too, because free(page[0]) would
1563 * otherwise free the whole shadow.
1564 */
1565 if (kmemcheck_page_is_tracked(page))
1566 split_page(virt_to_page(page[0].shadow), order);
1567#endif
1568
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001569 set_page_owner(page, 0, 0);
1570 for (i = 1; i < (1 << order); i++) {
Nick Piggin7835e982006-03-22 00:08:40 -08001571 set_page_refcounted(page + i);
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001572 set_page_owner(page + i, 0, 0);
1573 }
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001574}
K. Y. Srinivasan5853ff22013-03-25 15:47:38 -07001575EXPORT_SYMBOL_GPL(split_page);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001576
Joonsoo Kim3c605092014-11-13 15:19:21 -08001577int __isolate_free_page(struct page *page, unsigned int order)
Mel Gorman748446b2010-05-24 14:32:27 -07001578{
Mel Gorman748446b2010-05-24 14:32:27 -07001579 unsigned long watermark;
1580 struct zone *zone;
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07001581 int mt;
Mel Gorman748446b2010-05-24 14:32:27 -07001582
1583 BUG_ON(!PageBuddy(page));
1584
1585 zone = page_zone(page);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001586 mt = get_pageblock_migratetype(page);
Mel Gorman748446b2010-05-24 14:32:27 -07001587
Minchan Kim194159f2013-02-22 16:33:58 -08001588 if (!is_migrate_isolate(mt)) {
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001589 /* Obey watermarks as if the page was being allocated */
1590 watermark = low_wmark_pages(zone) + (1 << order);
1591 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1592 return 0;
1593
Mel Gorman8fb74b92013-01-11 14:32:16 -08001594 __mod_zone_freepage_state(zone, -(1UL << order), mt);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001595 }
Mel Gorman748446b2010-05-24 14:32:27 -07001596
1597 /* Remove page from free list */
1598 list_del(&page->lru);
1599 zone->free_area[order].nr_free--;
1600 rmv_page_order(page);
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07001601
Mel Gorman8fb74b92013-01-11 14:32:16 -08001602 /* Set the pageblock if the isolated page is at least a pageblock */
Mel Gorman748446b2010-05-24 14:32:27 -07001603 if (order >= pageblock_order - 1) {
1604 struct page *endpage = page + (1 << order) - 1;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001605 for (; page < endpage; page += pageblock_nr_pages) {
1606 int mt = get_pageblock_migratetype(page);
Minchan Kim194159f2013-02-22 16:33:58 -08001607 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001608 set_pageblock_migratetype(page,
1609 MIGRATE_MOVABLE);
1610 }
Mel Gorman748446b2010-05-24 14:32:27 -07001611 }
1612
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001613 set_page_owner(page, order, 0);
Mel Gorman8fb74b92013-01-11 14:32:16 -08001614 return 1UL << order;
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001615}
1616
1617/*
1618 * Similar to split_page except the page is already free. As this is only
1619 * being used for migration, the migratetype of the block also changes.
1620 * As this is called with interrupts disabled, the caller is responsible
1621 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1622 * are enabled.
1623 *
1624 * Note: this is probably too low level an operation for use in drivers.
1625 * Please consult with lkml before using this in your driver.
1626 */
1627int split_free_page(struct page *page)
1628{
1629 unsigned int order;
1630 int nr_pages;
1631
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001632 order = page_order(page);
1633
Mel Gorman8fb74b92013-01-11 14:32:16 -08001634 nr_pages = __isolate_free_page(page, order);
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001635 if (!nr_pages)
1636 return 0;
1637
1638 /* Split into individual pages */
1639 set_page_refcounted(page);
1640 split_page(page, order);
1641 return nr_pages;
Mel Gorman748446b2010-05-24 14:32:27 -07001642}
1643
1644/*
Vlastimil Babka75379192015-02-11 15:25:38 -08001645 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07001647static inline
1648struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001649 struct zone *zone, unsigned int order,
1650 gfp_t gfp_flags, int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651{
1652 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001653 struct page *page;
Mel Gormanb745bc82014-06-04 16:10:22 -07001654 bool cold = ((gfp_flags & __GFP_COLD) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655
Nick Piggin48db57f2006-01-08 01:00:42 -08001656 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001658 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 local_irq_save(flags);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001661 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1662 list = &pcp->lists[migratetype];
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001663 if (list_empty(list)) {
Mel Gorman535131e62007-10-16 01:25:49 -07001664 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001665 pcp->batch, list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001666 migratetype, cold);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001667 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07001668 goto failed;
Mel Gorman535131e62007-10-16 01:25:49 -07001669 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001670
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001671 if (cold)
1672 page = list_entry(list->prev, struct page, lru);
1673 else
1674 page = list_entry(list->next, struct page, lru);
1675
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001676 list_del(&page->lru);
1677 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001678 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07001679 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1680 /*
1681 * __GFP_NOFAIL is not to be used in new code.
1682 *
1683 * All __GFP_NOFAIL callers should be fixed so that they
1684 * properly detect and handle allocation failures.
1685 *
1686 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07001687 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07001688 * __GFP_NOFAIL.
1689 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07001690 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07001691 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001693 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001694 spin_unlock(&zone->lock);
1695 if (!page)
1696 goto failed;
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07001697 __mod_zone_freepage_state(zone, -(1 << order),
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001698 get_freepage_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 }
1700
Johannes Weiner3a025762014-04-07 15:37:48 -07001701 __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
Johannes Weinerabe5f972014-10-02 16:21:10 -07001702 if (atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]) <= 0 &&
Johannes Weiner57054652014-10-09 15:28:17 -07001703 !test_bit(ZONE_FAIR_DEPLETED, &zone->flags))
1704 set_bit(ZONE_FAIR_DEPLETED, &zone->flags);
Johannes Weiner27329362014-03-03 15:38:41 -08001705
Christoph Lameterf8891e52006-06-30 01:55:45 -07001706 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Andi Kleen78afd562011-03-22 16:33:12 -07001707 zone_statistics(preferred_zone, zone, gfp_flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001708 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709
Sasha Levin309381fea2014-01-23 15:52:54 -08001710 VM_BUG_ON_PAGE(bad_range(zone, page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001712
1713failed:
1714 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001715 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716}
1717
Akinobu Mita933e3122006-12-08 02:39:45 -08001718#ifdef CONFIG_FAIL_PAGE_ALLOC
1719
Akinobu Mitab2588c42011-07-26 16:09:03 -07001720static struct {
Akinobu Mita933e3122006-12-08 02:39:45 -08001721 struct fault_attr attr;
1722
1723 u32 ignore_gfp_highmem;
1724 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001725 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001726} fail_page_alloc = {
1727 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001728 .ignore_gfp_wait = 1,
1729 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001730 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001731};
1732
1733static int __init setup_fail_page_alloc(char *str)
1734{
1735 return setup_fault_attr(&fail_page_alloc.attr, str);
1736}
1737__setup("fail_page_alloc=", setup_fail_page_alloc);
1738
Gavin Shandeaf3862012-07-31 16:41:51 -07001739static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08001740{
Akinobu Mita54114992007-07-15 23:40:23 -07001741 if (order < fail_page_alloc.min_order)
Gavin Shandeaf3862012-07-31 16:41:51 -07001742 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001743 if (gfp_mask & __GFP_NOFAIL)
Gavin Shandeaf3862012-07-31 16:41:51 -07001744 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001745 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
Gavin Shandeaf3862012-07-31 16:41:51 -07001746 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001747 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
Gavin Shandeaf3862012-07-31 16:41:51 -07001748 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001749
1750 return should_fail(&fail_page_alloc.attr, 1 << order);
1751}
1752
1753#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1754
1755static int __init fail_page_alloc_debugfs(void)
1756{
Al Virof4ae40a2011-07-24 04:33:43 -04001757 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
Akinobu Mita933e3122006-12-08 02:39:45 -08001758 struct dentry *dir;
Akinobu Mita933e3122006-12-08 02:39:45 -08001759
Akinobu Mitadd48c082011-08-03 16:21:01 -07001760 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
1761 &fail_page_alloc.attr);
1762 if (IS_ERR(dir))
1763 return PTR_ERR(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08001764
Akinobu Mitab2588c42011-07-26 16:09:03 -07001765 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
1766 &fail_page_alloc.ignore_gfp_wait))
1767 goto fail;
1768 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1769 &fail_page_alloc.ignore_gfp_highmem))
1770 goto fail;
1771 if (!debugfs_create_u32("min-order", mode, dir,
1772 &fail_page_alloc.min_order))
1773 goto fail;
Akinobu Mita933e3122006-12-08 02:39:45 -08001774
Akinobu Mitab2588c42011-07-26 16:09:03 -07001775 return 0;
1776fail:
Akinobu Mitadd48c082011-08-03 16:21:01 -07001777 debugfs_remove_recursive(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08001778
Akinobu Mitab2588c42011-07-26 16:09:03 -07001779 return -ENOMEM;
Akinobu Mita933e3122006-12-08 02:39:45 -08001780}
1781
1782late_initcall(fail_page_alloc_debugfs);
1783
1784#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1785
1786#else /* CONFIG_FAIL_PAGE_ALLOC */
1787
Gavin Shandeaf3862012-07-31 16:41:51 -07001788static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08001789{
Gavin Shandeaf3862012-07-31 16:41:51 -07001790 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001791}
1792
1793#endif /* CONFIG_FAIL_PAGE_ALLOC */
1794
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795/*
Mel Gorman88f5acf2011-01-13 15:45:41 -08001796 * Return true if free pages are above 'mark'. This takes into account the order
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 * of the allocation.
1798 */
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001799static bool __zone_watermark_ok(struct zone *z, unsigned int order,
1800 unsigned long mark, int classzone_idx, int alloc_flags,
1801 long free_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802{
Wei Yuan26086de2014-12-10 15:44:44 -08001803 /* free_pages may go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001804 long min = mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 int o;
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07001806 long free_cma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807
Michal Hockodf0a6da2012-01-10 15:08:02 -08001808 free_pages -= (1 << order) - 1;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001809 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001811 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 min -= min / 4;
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07001813#ifdef CONFIG_CMA
1814 /* If allocation can't use CMA areas don't use free CMA pages */
1815 if (!(alloc_flags & ALLOC_CMA))
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07001816 free_cma = zone_page_state(z, NR_FREE_CMA_PAGES);
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07001817#endif
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07001818
Mel Gorman3484b2d2014-08-06 16:07:14 -07001819 if (free_pages - free_cma <= min + z->lowmem_reserve[classzone_idx])
Mel Gorman88f5acf2011-01-13 15:45:41 -08001820 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 for (o = 0; o < order; o++) {
1822 /* At the next order, this order's pages become unavailable */
1823 free_pages -= z->free_area[o].nr_free << o;
1824
1825 /* Require fewer higher order pages to be free */
1826 min >>= 1;
1827
1828 if (free_pages <= min)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001829 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830 }
Mel Gorman88f5acf2011-01-13 15:45:41 -08001831 return true;
1832}
1833
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001834bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
Mel Gorman88f5acf2011-01-13 15:45:41 -08001835 int classzone_idx, int alloc_flags)
1836{
1837 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1838 zone_page_state(z, NR_FREE_PAGES));
1839}
1840
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001841bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
1842 unsigned long mark, int classzone_idx, int alloc_flags)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001843{
1844 long free_pages = zone_page_state(z, NR_FREE_PAGES);
1845
1846 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1847 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1848
1849 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1850 free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851}
1852
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001853#ifdef CONFIG_NUMA
1854/*
1855 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1856 * skip over zones that are not allowed by the cpuset, or that have
1857 * been recently (in last second) found to be nearly full. See further
1858 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001859 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001860 *
Zhi Yong Wua1aeb652013-11-12 15:08:29 -08001861 * If the zonelist cache is present in the passed zonelist, then
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001862 * returns a pointer to the allowed node mask (either the current
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08001863 * tasks mems_allowed, or node_states[N_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001864 *
1865 * If the zonelist cache is not available for this zonelist, does
1866 * nothing and returns NULL.
1867 *
1868 * If the fullzones BITMAP in the zonelist cache is stale (more than
1869 * a second since last zap'd) then we zap it out (clear its bits.)
1870 *
1871 * We hold off even calling zlc_setup, until after we've checked the
1872 * first zone in the zonelist, on the theory that most allocations will
1873 * be satisfied from that first zone, so best to examine that zone as
1874 * quickly as we can.
1875 */
1876static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1877{
1878 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1879 nodemask_t *allowednodes; /* zonelist_cache approximation */
1880
1881 zlc = zonelist->zlcache_ptr;
1882 if (!zlc)
1883 return NULL;
1884
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001885 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001886 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1887 zlc->last_full_zap = jiffies;
1888 }
1889
1890 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1891 &cpuset_current_mems_allowed :
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08001892 &node_states[N_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001893 return allowednodes;
1894}
1895
1896/*
1897 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1898 * if it is worth looking at further for free memory:
1899 * 1) Check that the zone isn't thought to be full (doesn't have its
1900 * bit set in the zonelist_cache fullzones BITMAP).
1901 * 2) Check that the zones node (obtained from the zonelist_cache
1902 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1903 * Return true (non-zero) if zone is worth looking at further, or
1904 * else return false (zero) if it is not.
1905 *
1906 * This check -ignores- the distinction between various watermarks,
1907 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1908 * found to be full for any variation of these watermarks, it will
1909 * be considered full for up to one second by all requests, unless
1910 * we are so low on memory on all allowed nodes that we are forced
1911 * into the second scan of the zonelist.
1912 *
1913 * In the second scan we ignore this zonelist cache and exactly
1914 * apply the watermarks to all zones, even it is slower to do so.
1915 * We are low on memory in the second scan, and should leave no stone
1916 * unturned looking for a free page.
1917 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001918static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001919 nodemask_t *allowednodes)
1920{
1921 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1922 int i; /* index of *z in zonelist zones */
1923 int n; /* node that zone *z is on */
1924
1925 zlc = zonelist->zlcache_ptr;
1926 if (!zlc)
1927 return 1;
1928
Mel Gormandd1a2392008-04-28 02:12:17 -07001929 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001930 n = zlc->z_to_n[i];
1931
1932 /* This zone is worth trying if it is allowed but not full */
1933 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1934}
1935
1936/*
1937 * Given 'z' scanning a zonelist, set the corresponding bit in
1938 * zlc->fullzones, so that subsequent attempts to allocate a page
1939 * from that zone don't waste time re-examining it.
1940 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001941static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001942{
1943 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1944 int i; /* index of *z in zonelist zones */
1945
1946 zlc = zonelist->zlcache_ptr;
1947 if (!zlc)
1948 return;
1949
Mel Gormandd1a2392008-04-28 02:12:17 -07001950 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001951
1952 set_bit(i, zlc->fullzones);
1953}
1954
Mel Gorman76d3fbf2011-07-25 17:12:30 -07001955/*
1956 * clear all zones full, called after direct reclaim makes progress so that
1957 * a zone that was recently full is not skipped over for up to a second
1958 */
1959static void zlc_clear_zones_full(struct zonelist *zonelist)
1960{
1961 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1962
1963 zlc = zonelist->zlcache_ptr;
1964 if (!zlc)
1965 return;
1966
1967 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1968}
1969
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001970static bool zone_local(struct zone *local_zone, struct zone *zone)
1971{
Johannes Weinerfff40682013-12-20 14:54:12 +00001972 return local_zone->node == zone->node;
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001973}
1974
David Rientjes957f8222012-10-08 16:33:24 -07001975static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
1976{
Mel Gorman5f7a75a2014-06-04 16:07:15 -07001977 return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <
1978 RECLAIM_DISTANCE;
David Rientjes957f8222012-10-08 16:33:24 -07001979}
1980
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001981#else /* CONFIG_NUMA */
1982
1983static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1984{
1985 return NULL;
1986}
1987
Mel Gormandd1a2392008-04-28 02:12:17 -07001988static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001989 nodemask_t *allowednodes)
1990{
1991 return 1;
1992}
1993
Mel Gormandd1a2392008-04-28 02:12:17 -07001994static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001995{
1996}
Mel Gorman76d3fbf2011-07-25 17:12:30 -07001997
1998static void zlc_clear_zones_full(struct zonelist *zonelist)
1999{
2000}
David Rientjes957f8222012-10-08 16:33:24 -07002001
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002002static bool zone_local(struct zone *local_zone, struct zone *zone)
2003{
2004 return true;
2005}
2006
David Rientjes957f8222012-10-08 16:33:24 -07002007static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2008{
2009 return true;
2010}
2011
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002012#endif /* CONFIG_NUMA */
2013
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002014static void reset_alloc_batches(struct zone *preferred_zone)
2015{
2016 struct zone *zone = preferred_zone->zone_pgdat->node_zones;
2017
2018 do {
2019 mod_zone_page_state(zone, NR_ALLOC_BATCH,
2020 high_wmark_pages(zone) - low_wmark_pages(zone) -
2021 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
Johannes Weiner57054652014-10-09 15:28:17 -07002022 clear_bit(ZONE_FAIR_DEPLETED, &zone->flags);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002023 } while (zone++ != preferred_zone);
2024}
2025
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002026/*
Paul Jackson0798e512006-12-06 20:31:38 -08002027 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002028 * a page.
2029 */
2030static struct page *
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002031get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
2032 const struct alloc_context *ac)
Martin Hicks753ee722005-06-21 17:14:41 -07002033{
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002034 struct zonelist *zonelist = ac->zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07002035 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002036 struct page *page = NULL;
Mel Gorman5117f452009-06-16 15:31:59 -07002037 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002038 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
2039 int zlc_active = 0; /* set if using zonelist_cache */
2040 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gormana6e21b142014-06-04 16:10:12 -07002041 bool consider_zone_dirty = (alloc_flags & ALLOC_WMARK_LOW) &&
2042 (gfp_mask & __GFP_WRITE);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002043 int nr_fair_skipped = 0;
2044 bool zonelist_rescan;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002045
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002046zonelist_scan:
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002047 zonelist_rescan = false;
2048
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002049 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002050 * Scan zonelist, looking for a zone with enough free.
Vladimir Davydov344736f2014-10-20 15:50:30 +04002051 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002052 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002053 for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
2054 ac->nodemask) {
Johannes Weinere085dbc2013-09-11 14:20:46 -07002055 unsigned long mark;
2056
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002057 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002058 !zlc_zone_worth_trying(zonelist, z, allowednodes))
2059 continue;
Mel Gorman664eedd2014-06-04 16:10:08 -07002060 if (cpusets_enabled() &&
2061 (alloc_flags & ALLOC_CPUSET) &&
Vladimir Davydov344736f2014-10-20 15:50:30 +04002062 !cpuset_zone_allowed(zone, gfp_mask))
Mel Gormancd38b112011-07-25 17:12:29 -07002063 continue;
Johannes Weinera756cf52012-01-10 15:07:49 -08002064 /*
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002065 * Distribute pages in proportion to the individual
2066 * zone size to ensure fair page aging. The zone a
2067 * page was allocated in should have no effect on the
2068 * time the page has in memory before being reclaimed.
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002069 */
Johannes Weiner3a025762014-04-07 15:37:48 -07002070 if (alloc_flags & ALLOC_FAIR) {
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002071 if (!zone_local(ac->preferred_zone, zone))
Mel Gormanf7b5d642014-08-06 16:07:20 -07002072 break;
Johannes Weiner57054652014-10-09 15:28:17 -07002073 if (test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) {
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002074 nr_fair_skipped++;
Johannes Weiner3a025762014-04-07 15:37:48 -07002075 continue;
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002076 }
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002077 }
2078 /*
Johannes Weinera756cf52012-01-10 15:07:49 -08002079 * When allocating a page cache page for writing, we
2080 * want to get it from a zone that is within its dirty
2081 * limit, such that no single zone holds more than its
2082 * proportional share of globally allowed dirty pages.
2083 * The dirty limits take into account the zone's
2084 * lowmem reserves and high watermark so that kswapd
2085 * should be able to balance it without having to
2086 * write pages from its LRU list.
2087 *
2088 * This may look like it could increase pressure on
2089 * lower zones by failing allocations in higher zones
2090 * before they are full. But the pages that do spill
2091 * over are limited as the lower zones are protected
2092 * by this very same mechanism. It should not become
2093 * a practical burden to them.
2094 *
2095 * XXX: For now, allow allocations to potentially
2096 * exceed the per-zone dirty limit in the slowpath
2097 * (ALLOC_WMARK_LOW unset) before going into reclaim,
2098 * which is important when on a NUMA setup the allowed
2099 * zones are together not big enough to reach the
2100 * global limit. The proper fix for these situations
2101 * will require awareness of zones in the
2102 * dirty-throttling and the flusher threads.
2103 */
Mel Gormana6e21b142014-06-04 16:10:12 -07002104 if (consider_zone_dirty && !zone_dirty_ok(zone))
Mel Gorman800a1e72014-06-04 16:10:06 -07002105 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002106
Johannes Weinere085dbc2013-09-11 14:20:46 -07002107 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
2108 if (!zone_watermark_ok(zone, order, mark,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002109 ac->classzone_idx, alloc_flags)) {
Mel Gormanfa5e0842009-06-16 15:33:22 -07002110 int ret;
2111
Mel Gorman5dab2912014-06-04 16:10:14 -07002112 /* Checked here to keep the fast path fast */
2113 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
2114 if (alloc_flags & ALLOC_NO_WATERMARKS)
2115 goto try_this_zone;
2116
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002117 if (IS_ENABLED(CONFIG_NUMA) &&
2118 !did_zlc_setup && nr_online_nodes > 1) {
Mel Gormancd38b112011-07-25 17:12:29 -07002119 /*
2120 * we do zlc_setup if there are multiple nodes
2121 * and before considering the first zone allowed
2122 * by the cpuset.
2123 */
2124 allowednodes = zlc_setup(zonelist, alloc_flags);
2125 zlc_active = 1;
2126 did_zlc_setup = 1;
2127 }
2128
David Rientjes957f8222012-10-08 16:33:24 -07002129 if (zone_reclaim_mode == 0 ||
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002130 !zone_allows_reclaim(ac->preferred_zone, zone))
Mel Gormanfa5e0842009-06-16 15:33:22 -07002131 goto this_zone_full;
2132
Mel Gormancd38b112011-07-25 17:12:29 -07002133 /*
2134 * As we may have just activated ZLC, check if the first
2135 * eligible zone has failed zone_reclaim recently.
2136 */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002137 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
Mel Gormancd38b112011-07-25 17:12:29 -07002138 !zlc_zone_worth_trying(zonelist, z, allowednodes))
2139 continue;
2140
Mel Gormanfa5e0842009-06-16 15:33:22 -07002141 ret = zone_reclaim(zone, gfp_mask, order);
2142 switch (ret) {
2143 case ZONE_RECLAIM_NOSCAN:
2144 /* did not scan */
Mel Gormancd38b112011-07-25 17:12:29 -07002145 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002146 case ZONE_RECLAIM_FULL:
2147 /* scanned but unreclaimable */
Mel Gormancd38b112011-07-25 17:12:29 -07002148 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002149 default:
2150 /* did we reclaim enough */
Mel Gormanfed27192013-04-29 15:07:57 -07002151 if (zone_watermark_ok(zone, order, mark,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002152 ac->classzone_idx, alloc_flags))
Mel Gormanfed27192013-04-29 15:07:57 -07002153 goto try_this_zone;
2154
2155 /*
2156 * Failed to reclaim enough to meet watermark.
2157 * Only mark the zone full if checking the min
2158 * watermark or if we failed to reclaim just
2159 * 1<<order pages or else the page allocator
2160 * fastpath will prematurely mark zones full
2161 * when the watermark is between the low and
2162 * min watermarks.
2163 */
2164 if (((alloc_flags & ALLOC_WMARK_MASK) == ALLOC_WMARK_MIN) ||
2165 ret == ZONE_RECLAIM_SOME)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002166 goto this_zone_full;
Mel Gormanfed27192013-04-29 15:07:57 -07002167
2168 continue;
Paul Jackson0798e512006-12-06 20:31:38 -08002169 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002170 }
2171
Mel Gormanfa5e0842009-06-16 15:33:22 -07002172try_this_zone:
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002173 page = buffered_rmqueue(ac->preferred_zone, zone, order,
2174 gfp_mask, ac->migratetype);
Vlastimil Babka75379192015-02-11 15:25:38 -08002175 if (page) {
2176 if (prep_new_page(page, order, gfp_mask, alloc_flags))
2177 goto try_this_zone;
2178 return page;
2179 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002180this_zone_full:
Mel Gorman65bb3712014-06-04 16:10:05 -07002181 if (IS_ENABLED(CONFIG_NUMA) && zlc_active)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002182 zlc_mark_zone_full(zonelist, z);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002183 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002184
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002185 /*
2186 * The first pass makes sure allocations are spread fairly within the
2187 * local node. However, the local node might have free pages left
2188 * after the fairness batches are exhausted, and remote zones haven't
2189 * even been considered yet. Try once more without fairness, and
2190 * include remote zones now, before entering the slowpath and waking
2191 * kswapd: prefer spilling to a remote zone over swapping locally.
2192 */
2193 if (alloc_flags & ALLOC_FAIR) {
2194 alloc_flags &= ~ALLOC_FAIR;
2195 if (nr_fair_skipped) {
2196 zonelist_rescan = true;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002197 reset_alloc_batches(ac->preferred_zone);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002198 }
2199 if (nr_online_nodes > 1)
2200 zonelist_rescan = true;
2201 }
2202
2203 if (unlikely(IS_ENABLED(CONFIG_NUMA) && zlc_active)) {
2204 /* Disable zlc cache for second zonelist scan */
2205 zlc_active = 0;
2206 zonelist_rescan = true;
2207 }
2208
2209 if (zonelist_rescan)
2210 goto zonelist_scan;
2211
2212 return NULL;
Martin Hicks753ee722005-06-21 17:14:41 -07002213}
2214
David Rientjes29423e772011-03-22 16:30:47 -07002215/*
2216 * Large machines with many possible nodes should not always dump per-node
2217 * meminfo in irq context.
2218 */
2219static inline bool should_suppress_show_mem(void)
2220{
2221 bool ret = false;
2222
2223#if NODES_SHIFT > 8
2224 ret = in_interrupt();
2225#endif
2226 return ret;
2227}
2228
Dave Hansena238ab52011-05-24 17:12:16 -07002229static DEFINE_RATELIMIT_STATE(nopage_rs,
2230 DEFAULT_RATELIMIT_INTERVAL,
2231 DEFAULT_RATELIMIT_BURST);
2232
2233void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
2234{
Dave Hansena238ab52011-05-24 17:12:16 -07002235 unsigned int filter = SHOW_MEM_FILTER_NODES;
2236
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -08002237 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2238 debug_guardpage_minorder() > 0)
Dave Hansena238ab52011-05-24 17:12:16 -07002239 return;
2240
2241 /*
2242 * This documents exceptions given to allocations in certain
2243 * contexts that are allowed to allocate outside current's set
2244 * of allowed nodes.
2245 */
2246 if (!(gfp_mask & __GFP_NOMEMALLOC))
2247 if (test_thread_flag(TIF_MEMDIE) ||
2248 (current->flags & (PF_MEMALLOC | PF_EXITING)))
2249 filter &= ~SHOW_MEM_FILTER_NODES;
2250 if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
2251 filter &= ~SHOW_MEM_FILTER_NODES;
2252
2253 if (fmt) {
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002254 struct va_format vaf;
2255 va_list args;
2256
Dave Hansena238ab52011-05-24 17:12:16 -07002257 va_start(args, fmt);
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002258
2259 vaf.fmt = fmt;
2260 vaf.va = &args;
2261
2262 pr_warn("%pV", &vaf);
2263
Dave Hansena238ab52011-05-24 17:12:16 -07002264 va_end(args);
2265 }
2266
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002267 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2268 current->comm, order, gfp_mask);
Dave Hansena238ab52011-05-24 17:12:16 -07002269
2270 dump_stack();
2271 if (!should_suppress_show_mem())
2272 show_mem(filter);
2273}
2274
Mel Gorman11e33f62009-06-16 15:31:57 -07002275static inline int
2276should_alloc_retry(gfp_t gfp_mask, unsigned int order,
Mel Gormanf90ac392012-01-10 15:07:15 -08002277 unsigned long did_some_progress,
Mel Gorman11e33f62009-06-16 15:31:57 -07002278 unsigned long pages_reclaimed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279{
Mel Gorman11e33f62009-06-16 15:31:57 -07002280 /* Do not loop if specifically requested */
2281 if (gfp_mask & __GFP_NORETRY)
2282 return 0;
2283
Mel Gormanf90ac392012-01-10 15:07:15 -08002284 /* Always retry if specifically requested */
2285 if (gfp_mask & __GFP_NOFAIL)
2286 return 1;
2287
2288 /*
2289 * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim
2290 * making forward progress without invoking OOM. Suspend also disables
2291 * storage devices so kswapd will not help. Bail if we are suspending.
2292 */
2293 if (!did_some_progress && pm_suspended_storage())
2294 return 0;
2295
Mel Gorman11e33f62009-06-16 15:31:57 -07002296 /*
2297 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
2298 * means __GFP_NOFAIL, but that may not be true in other
2299 * implementations.
2300 */
2301 if (order <= PAGE_ALLOC_COSTLY_ORDER)
2302 return 1;
2303
2304 /*
2305 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
2306 * specified, then we retry until we no longer reclaim any pages
2307 * (above), or we've reclaimed an order of pages at least as
2308 * large as the allocation's order. In both cases, if the
2309 * allocation still fails, we stop retrying.
2310 */
2311 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
2312 return 1;
2313
Mel Gorman11e33f62009-06-16 15:31:57 -07002314 return 0;
2315}
2316
2317static inline struct page *
2318__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002319 const struct alloc_context *ac, unsigned long *did_some_progress)
Mel Gorman11e33f62009-06-16 15:31:57 -07002320{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322
Johannes Weiner9879de72015-01-26 12:58:32 -08002323 *did_some_progress = 0;
2324
Johannes Weiner9879de72015-01-26 12:58:32 -08002325 /*
2326 * Acquire the per-zone oom lock for each zone. If that
2327 * fails, somebody else is making progress for us.
2328 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002329 if (!oom_zonelist_trylock(ac->zonelist, gfp_mask)) {
Johannes Weiner9879de72015-01-26 12:58:32 -08002330 *did_some_progress = 1;
Mel Gorman11e33f62009-06-16 15:31:57 -07002331 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332 return NULL;
2333 }
Jens Axboe6b1de912005-11-17 21:35:02 +01002334
Mel Gorman11e33f62009-06-16 15:31:57 -07002335 /*
2336 * Go through the zonelist yet one more time, keep very high watermark
2337 * here, this is only to catch a parallel oom killing, we must fail if
2338 * we're still under heavy pressure.
2339 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002340 page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order,
2341 ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002342 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07002343 goto out;
2344
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002345 if (!(gfp_mask & __GFP_NOFAIL)) {
Johannes Weiner9879de72015-01-26 12:58:32 -08002346 /* Coredumps can quickly deplete all memory reserves */
2347 if (current->flags & PF_DUMPCORE)
2348 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002349 /* The OOM killer will not help higher order allocs */
2350 if (order > PAGE_ALLOC_COSTLY_ORDER)
2351 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07002352 /* The OOM killer does not needlessly kill tasks for lowmem */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002353 if (ac->high_zoneidx < ZONE_NORMAL)
David Rientjes03668b32010-08-09 17:18:54 -07002354 goto out;
Johannes Weiner9879de72015-01-26 12:58:32 -08002355 /* The OOM killer does not compensate for light reclaim */
Johannes Weinercc873172015-02-27 15:52:09 -08002356 if (!(gfp_mask & __GFP_FS)) {
2357 /*
2358 * XXX: Page reclaim didn't yield anything,
2359 * and the OOM killer can't be invoked, but
2360 * keep looping as per should_alloc_retry().
2361 */
2362 *did_some_progress = 1;
Johannes Weiner9879de72015-01-26 12:58:32 -08002363 goto out;
Johannes Weinercc873172015-02-27 15:52:09 -08002364 }
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002365 /*
2366 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
2367 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
2368 * The caller should handle page allocation failure by itself if
2369 * it specifies __GFP_THISNODE.
2370 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
2371 */
2372 if (gfp_mask & __GFP_THISNODE)
2373 goto out;
2374 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002375 /* Exhausted what can be done so it's blamo time */
Michal Hockoc32b3cb2015-02-11 15:26:24 -08002376 if (out_of_memory(ac->zonelist, gfp_mask, order, ac->nodemask, false))
2377 *did_some_progress = 1;
Mel Gorman11e33f62009-06-16 15:31:57 -07002378out:
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002379 oom_zonelist_unlock(ac->zonelist, gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002380 return page;
2381}
2382
Mel Gorman56de7262010-05-24 14:32:30 -07002383#ifdef CONFIG_COMPACTION
2384/* Try memory compaction for high-order allocations before reclaim */
2385static struct page *
2386__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002387 int alloc_flags, const struct alloc_context *ac,
2388 enum migrate_mode mode, int *contended_compaction,
2389 bool *deferred_compaction)
Mel Gorman56de7262010-05-24 14:32:30 -07002390{
Vlastimil Babka53853e22014-10-09 15:27:02 -07002391 unsigned long compact_result;
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002392 struct page *page;
Vlastimil Babka53853e22014-10-09 15:27:02 -07002393
Mel Gorman66199712012-01-12 17:19:41 -08002394 if (!order)
Mel Gorman56de7262010-05-24 14:32:30 -07002395 return NULL;
2396
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002397 current->flags |= PF_MEMALLOC;
Vlastimil Babka1a6d53a2015-02-11 15:25:44 -08002398 compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
2399 mode, contended_compaction);
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002400 current->flags &= ~PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07002401
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002402 switch (compact_result) {
2403 case COMPACT_DEFERRED:
Vlastimil Babka53853e22014-10-09 15:27:02 -07002404 *deferred_compaction = true;
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002405 /* fall-through */
2406 case COMPACT_SKIPPED:
2407 return NULL;
2408 default:
2409 break;
Mel Gorman56de7262010-05-24 14:32:30 -07002410 }
2411
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002412 /*
2413 * At least in one zone compaction wasn't deferred or skipped, so let's
2414 * count a compaction stall
2415 */
2416 count_vm_event(COMPACTSTALL);
2417
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002418 page = get_page_from_freelist(gfp_mask, order,
2419 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002420
2421 if (page) {
2422 struct zone *zone = page_zone(page);
2423
2424 zone->compact_blockskip_flush = false;
2425 compaction_defer_reset(zone, order, true);
2426 count_vm_event(COMPACTSUCCESS);
2427 return page;
2428 }
2429
2430 /*
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002431 * It's bad if compaction run occurs and fails. The most likely reason
2432 * is that pages exist, but not enough to satisfy watermarks.
2433 */
2434 count_vm_event(COMPACTFAIL);
2435
2436 cond_resched();
2437
Mel Gorman56de7262010-05-24 14:32:30 -07002438 return NULL;
2439}
2440#else
2441static inline struct page *
2442__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002443 int alloc_flags, const struct alloc_context *ac,
2444 enum migrate_mode mode, int *contended_compaction,
2445 bool *deferred_compaction)
Mel Gorman56de7262010-05-24 14:32:30 -07002446{
2447 return NULL;
2448}
2449#endif /* CONFIG_COMPACTION */
2450
Marek Szyprowskibba90712012-01-25 12:09:52 +01002451/* Perform direct synchronous page reclaim */
2452static int
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002453__perform_reclaim(gfp_t gfp_mask, unsigned int order,
2454 const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002455{
Mel Gorman11e33f62009-06-16 15:31:57 -07002456 struct reclaim_state reclaim_state;
Marek Szyprowskibba90712012-01-25 12:09:52 +01002457 int progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07002458
2459 cond_resched();
2460
2461 /* We now go into synchronous reclaim */
2462 cpuset_memory_pressure_bump();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002463 current->flags |= PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002464 lockdep_set_current_reclaim_state(gfp_mask);
2465 reclaim_state.reclaimed_slab = 0;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002466 current->reclaim_state = &reclaim_state;
Mel Gorman11e33f62009-06-16 15:31:57 -07002467
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002468 progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
2469 ac->nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002470
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002471 current->reclaim_state = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002472 lockdep_clear_current_reclaim_state();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002473 current->flags &= ~PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002474
2475 cond_resched();
2476
Marek Szyprowskibba90712012-01-25 12:09:52 +01002477 return progress;
2478}
2479
2480/* The really slow allocator path where we enter direct reclaim */
2481static inline struct page *
2482__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002483 int alloc_flags, const struct alloc_context *ac,
2484 unsigned long *did_some_progress)
Marek Szyprowskibba90712012-01-25 12:09:52 +01002485{
2486 struct page *page = NULL;
2487 bool drained = false;
2488
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002489 *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002490 if (unlikely(!(*did_some_progress)))
2491 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002492
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002493 /* After successful reclaim, reconsider all zones for allocation */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002494 if (IS_ENABLED(CONFIG_NUMA))
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002495 zlc_clear_zones_full(ac->zonelist);
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002496
Mel Gorman9ee493c2010-09-09 16:38:18 -07002497retry:
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002498 page = get_page_from_freelist(gfp_mask, order,
2499 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002500
2501 /*
2502 * If an allocation failed after direct reclaim, it could be because
2503 * pages are pinned on the per-cpu lists. Drain them and try again
2504 */
2505 if (!page && !drained) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002506 drain_all_pages(NULL);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002507 drained = true;
2508 goto retry;
2509 }
2510
Mel Gorman11e33f62009-06-16 15:31:57 -07002511 return page;
2512}
2513
Mel Gorman11e33f62009-06-16 15:31:57 -07002514/*
2515 * This is called in the allocator slow-path if the allocation request is of
2516 * sufficient urgency to ignore watermarks and take other desperate measures
2517 */
2518static inline struct page *
2519__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002520 const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002521{
2522 struct page *page;
2523
2524 do {
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002525 page = get_page_from_freelist(gfp_mask, order,
2526 ALLOC_NO_WATERMARKS, ac);
Mel Gorman11e33f62009-06-16 15:31:57 -07002527
2528 if (!page && gfp_mask & __GFP_NOFAIL)
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002529 wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC,
2530 HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07002531 } while (!page && (gfp_mask & __GFP_NOFAIL));
2532
2533 return page;
2534}
2535
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002536static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002537{
2538 struct zoneref *z;
2539 struct zone *zone;
2540
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002541 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
2542 ac->high_zoneidx, ac->nodemask)
2543 wakeup_kswapd(zone, order, zone_idx(ac->preferred_zone));
Mel Gorman11e33f62009-06-16 15:31:57 -07002544}
2545
Peter Zijlstra341ce062009-06-16 15:32:02 -07002546static inline int
2547gfp_to_alloc_flags(gfp_t gfp_mask)
2548{
Peter Zijlstra341ce062009-06-16 15:32:02 -07002549 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
David Rientjesb104a352014-07-30 16:08:24 -07002550 const bool atomic = !(gfp_mask & (__GFP_WAIT | __GFP_NO_KSWAPD));
Peter Zijlstra341ce062009-06-16 15:32:02 -07002551
Mel Gormana56f57f2009-06-16 15:32:02 -07002552 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07002553 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07002554
Peter Zijlstra341ce062009-06-16 15:32:02 -07002555 /*
2556 * The caller may dip into page reserves a bit more if the caller
2557 * cannot run direct reclaim, or if the caller has realtime scheduling
2558 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
David Rientjesb104a352014-07-30 16:08:24 -07002559 * set both ALLOC_HARDER (atomic == true) and ALLOC_HIGH (__GFP_HIGH).
Peter Zijlstra341ce062009-06-16 15:32:02 -07002560 */
Namhyung Kime6223a32010-10-26 14:21:59 -07002561 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07002562
David Rientjesb104a352014-07-30 16:08:24 -07002563 if (atomic) {
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002564 /*
David Rientjesb104a352014-07-30 16:08:24 -07002565 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
2566 * if it can't schedule.
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002567 */
David Rientjesb104a352014-07-30 16:08:24 -07002568 if (!(gfp_mask & __GFP_NOMEMALLOC))
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002569 alloc_flags |= ALLOC_HARDER;
Peter Zijlstra341ce062009-06-16 15:32:02 -07002570 /*
David Rientjesb104a352014-07-30 16:08:24 -07002571 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
Vladimir Davydov344736f2014-10-20 15:50:30 +04002572 * comment for __cpuset_node_allowed().
Peter Zijlstra341ce062009-06-16 15:32:02 -07002573 */
2574 alloc_flags &= ~ALLOC_CPUSET;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002575 } else if (unlikely(rt_task(current)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07002576 alloc_flags |= ALLOC_HARDER;
2577
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002578 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2579 if (gfp_mask & __GFP_MEMALLOC)
2580 alloc_flags |= ALLOC_NO_WATERMARKS;
Mel Gorman907aed42012-07-31 16:44:07 -07002581 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
2582 alloc_flags |= ALLOC_NO_WATERMARKS;
2583 else if (!in_interrupt() &&
2584 ((current->flags & PF_MEMALLOC) ||
2585 unlikely(test_thread_flag(TIF_MEMDIE))))
Peter Zijlstra341ce062009-06-16 15:32:02 -07002586 alloc_flags |= ALLOC_NO_WATERMARKS;
2587 }
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002588#ifdef CONFIG_CMA
David Rientjes43e7a342014-10-09 15:27:25 -07002589 if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002590 alloc_flags |= ALLOC_CMA;
2591#endif
Peter Zijlstra341ce062009-06-16 15:32:02 -07002592 return alloc_flags;
2593}
2594
Mel Gorman072bb0a2012-07-31 16:43:58 -07002595bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
2596{
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002597 return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
Mel Gorman072bb0a2012-07-31 16:43:58 -07002598}
2599
Mel Gorman11e33f62009-06-16 15:31:57 -07002600static inline struct page *
2601__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002602 struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002603{
2604 const gfp_t wait = gfp_mask & __GFP_WAIT;
2605 struct page *page = NULL;
2606 int alloc_flags;
2607 unsigned long pages_reclaimed = 0;
2608 unsigned long did_some_progress;
David Rientjese0b9dae2014-06-04 16:08:28 -07002609 enum migrate_mode migration_mode = MIGRATE_ASYNC;
Mel Gorman66199712012-01-12 17:19:41 -08002610 bool deferred_compaction = false;
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07002611 int contended_compaction = COMPACT_CONTENDED_NONE;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002612
Christoph Lameter952f3b52006-12-06 20:33:26 -08002613 /*
Mel Gorman72807a72009-06-16 15:32:18 -07002614 * In the slowpath, we sanity check order to avoid ever trying to
2615 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2616 * be using allocators in order of preference for an area that is
2617 * too large.
2618 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07002619 if (order >= MAX_ORDER) {
2620 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07002621 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07002622 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002623
Christoph Lameter952f3b52006-12-06 20:33:26 -08002624 /*
2625 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2626 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2627 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2628 * using a larger set of nodes after it has established that the
2629 * allowed per node queues are empty and that nodes are
2630 * over allocated.
2631 */
Johannes Weiner3a025762014-04-07 15:37:48 -07002632 if (IS_ENABLED(CONFIG_NUMA) &&
2633 (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
Christoph Lameter952f3b52006-12-06 20:33:26 -08002634 goto nopage;
2635
Johannes Weiner9879de72015-01-26 12:58:32 -08002636retry:
Johannes Weiner3a025762014-04-07 15:37:48 -07002637 if (!(gfp_mask & __GFP_NO_KSWAPD))
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002638 wake_all_kswapds(order, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002639
Paul Jackson9bf22292005-09-06 15:18:12 -07002640 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002641 * OK, we're below the kswapd watermark and have kicked background
2642 * reclaim. Now things get more complex, so set up alloc_flags according
2643 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07002644 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002645 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646
David Rientjesf33261d2011-01-25 15:07:20 -08002647 /*
2648 * Find the true preferred zone if the allocation is unconstrained by
2649 * cpusets.
2650 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002651 if (!(alloc_flags & ALLOC_CPUSET) && !ac->nodemask) {
Mel Gormand8846372014-06-04 16:10:33 -07002652 struct zoneref *preferred_zoneref;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002653 preferred_zoneref = first_zones_zonelist(ac->zonelist,
2654 ac->high_zoneidx, NULL, &ac->preferred_zone);
2655 ac->classzone_idx = zonelist_zone_idx(preferred_zoneref);
Mel Gormand8846372014-06-04 16:10:33 -07002656 }
David Rientjesf33261d2011-01-25 15:07:20 -08002657
Peter Zijlstra341ce062009-06-16 15:32:02 -07002658 /* This is the last chance, in general, before the goto nopage. */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002659 page = get_page_from_freelist(gfp_mask, order,
2660 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002661 if (page)
2662 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663
Mel Gorman11e33f62009-06-16 15:31:57 -07002664 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002665 if (alloc_flags & ALLOC_NO_WATERMARKS) {
Mel Gorman183f6372012-07-31 16:44:12 -07002666 /*
2667 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
2668 * the allocation is high priority and these type of
2669 * allocations are system rather than user orientated
2670 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002671 ac->zonelist = node_zonelist(numa_node_id(), gfp_mask);
Mel Gorman183f6372012-07-31 16:44:12 -07002672
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002673 page = __alloc_pages_high_priority(gfp_mask, order, ac);
2674
Mel Gormancfd19c52012-07-31 16:44:10 -07002675 if (page) {
Peter Zijlstra341ce062009-06-16 15:32:02 -07002676 goto got_pg;
Mel Gormancfd19c52012-07-31 16:44:10 -07002677 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678 }
2679
2680 /* Atomic allocations - we can't balance anything */
David Rientjesaed0a0e2014-01-21 15:51:12 -08002681 if (!wait) {
2682 /*
2683 * All existing users of the deprecated __GFP_NOFAIL are
2684 * blockable, so warn of any new users that actually allow this
2685 * type of allocation to fail.
2686 */
2687 WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688 goto nopage;
David Rientjesaed0a0e2014-01-21 15:51:12 -08002689 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690
Peter Zijlstra341ce062009-06-16 15:32:02 -07002691 /* Avoid recursion of direct reclaim */
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002692 if (current->flags & PF_MEMALLOC)
Peter Zijlstra341ce062009-06-16 15:32:02 -07002693 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694
David Rientjes6583bb62009-07-29 15:02:06 -07002695 /* Avoid allocations with no watermarks from looping endlessly */
2696 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2697 goto nopage;
2698
Mel Gorman77f1fe62011-01-13 15:45:57 -08002699 /*
2700 * Try direct compaction. The first pass is asynchronous. Subsequent
2701 * attempts after direct reclaim are synchronous
2702 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002703 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
2704 migration_mode,
2705 &contended_compaction,
Vlastimil Babka53853e22014-10-09 15:27:02 -07002706 &deferred_compaction);
Mel Gorman56de7262010-05-24 14:32:30 -07002707 if (page)
2708 goto got_pg;
David Rientjes75f30862014-06-04 16:08:30 -07002709
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07002710 /* Checks for THP-specific high-order allocations */
2711 if ((gfp_mask & GFP_TRANSHUGE) == GFP_TRANSHUGE) {
2712 /*
2713 * If compaction is deferred for high-order allocations, it is
2714 * because sync compaction recently failed. If this is the case
2715 * and the caller requested a THP allocation, we do not want
2716 * to heavily disrupt the system, so we fail the allocation
2717 * instead of entering direct reclaim.
2718 */
2719 if (deferred_compaction)
2720 goto nopage;
2721
2722 /*
2723 * In all zones where compaction was attempted (and not
2724 * deferred or skipped), lock contention has been detected.
2725 * For THP allocation we do not want to disrupt the others
2726 * so we fallback to base pages instead.
2727 */
2728 if (contended_compaction == COMPACT_CONTENDED_LOCK)
2729 goto nopage;
2730
2731 /*
2732 * If compaction was aborted due to need_resched(), we do not
2733 * want to further increase allocation latency, unless it is
2734 * khugepaged trying to collapse.
2735 */
2736 if (contended_compaction == COMPACT_CONTENDED_SCHED
2737 && !(current->flags & PF_KTHREAD))
2738 goto nopage;
2739 }
Mel Gorman66199712012-01-12 17:19:41 -08002740
David Rientjes8fe78042014-08-06 16:07:54 -07002741 /*
2742 * It can become very expensive to allocate transparent hugepages at
2743 * fault, so use asynchronous memory compaction for THP unless it is
2744 * khugepaged trying to collapse.
2745 */
2746 if ((gfp_mask & GFP_TRANSHUGE) != GFP_TRANSHUGE ||
2747 (current->flags & PF_KTHREAD))
2748 migration_mode = MIGRATE_SYNC_LIGHT;
2749
Mel Gorman11e33f62009-06-16 15:31:57 -07002750 /* Try direct reclaim and then allocating */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002751 page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
2752 &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07002753 if (page)
2754 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755
Mel Gorman11e33f62009-06-16 15:31:57 -07002756 /* Check if we should retry the allocation */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07002757 pages_reclaimed += did_some_progress;
Mel Gormanf90ac392012-01-10 15:07:15 -08002758 if (should_alloc_retry(gfp_mask, order, did_some_progress,
2759 pages_reclaimed)) {
Johannes Weiner9879de72015-01-26 12:58:32 -08002760 /*
2761 * If we fail to make progress by freeing individual
2762 * pages, but the allocation wants us to keep going,
2763 * start OOM killing tasks.
2764 */
2765 if (!did_some_progress) {
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002766 page = __alloc_pages_may_oom(gfp_mask, order, ac,
2767 &did_some_progress);
Johannes Weiner9879de72015-01-26 12:58:32 -08002768 if (page)
2769 goto got_pg;
2770 if (!did_some_progress)
2771 goto nopage;
2772 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002773 /* Wait for some write requests to complete then retry */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002774 wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC, HZ/50);
Johannes Weiner9879de72015-01-26 12:58:32 -08002775 goto retry;
Mel Gorman3e7d3442011-01-13 15:45:56 -08002776 } else {
2777 /*
2778 * High-order allocations do not necessarily loop after
2779 * direct reclaim and reclaim/compaction depends on compaction
2780 * being called after reclaim so call directly if necessary
2781 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002782 page = __alloc_pages_direct_compact(gfp_mask, order,
2783 alloc_flags, ac, migration_mode,
2784 &contended_compaction,
Vlastimil Babka53853e22014-10-09 15:27:02 -07002785 &deferred_compaction);
Mel Gorman3e7d3442011-01-13 15:45:56 -08002786 if (page)
2787 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788 }
2789
2790nopage:
Dave Hansena238ab52011-05-24 17:12:16 -07002791 warn_alloc_failed(gfp_mask, order, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792got_pg:
Mel Gorman072bb0a2012-07-31 16:43:58 -07002793 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794}
Mel Gorman11e33f62009-06-16 15:31:57 -07002795
2796/*
2797 * This is the 'heart' of the zoned buddy allocator.
2798 */
2799struct page *
2800__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2801 struct zonelist *zonelist, nodemask_t *nodemask)
2802{
Mel Gormand8846372014-06-04 16:10:33 -07002803 struct zoneref *preferred_zoneref;
Mel Gormancc9a6c82012-03-21 16:34:11 -07002804 struct page *page = NULL;
Mel Gormancc9a6c82012-03-21 16:34:11 -07002805 unsigned int cpuset_mems_cookie;
Johannes Weiner3a025762014-04-07 15:37:48 -07002806 int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET|ALLOC_FAIR;
Andrew Morton91fbdc02015-02-11 15:25:04 -08002807 gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002808 struct alloc_context ac = {
2809 .high_zoneidx = gfp_zone(gfp_mask),
2810 .nodemask = nodemask,
2811 .migratetype = gfpflags_to_migratetype(gfp_mask),
2812 };
Mel Gorman11e33f62009-06-16 15:31:57 -07002813
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10002814 gfp_mask &= gfp_allowed_mask;
2815
Mel Gorman11e33f62009-06-16 15:31:57 -07002816 lockdep_trace_alloc(gfp_mask);
2817
2818 might_sleep_if(gfp_mask & __GFP_WAIT);
2819
2820 if (should_fail_alloc_page(gfp_mask, order))
2821 return NULL;
2822
2823 /*
2824 * Check the zones suitable for the gfp_mask contain at least one
2825 * valid zone. It's possible to have an empty zonelist as a result
2826 * of GFP_THISNODE and a memoryless node
2827 */
2828 if (unlikely(!zonelist->_zonerefs->zone))
2829 return NULL;
2830
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002831 if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE)
Vlastimil Babka21bb9bd2014-10-09 15:26:51 -07002832 alloc_flags |= ALLOC_CMA;
2833
Mel Gormancc9a6c82012-03-21 16:34:11 -07002834retry_cpuset:
Mel Gormand26914d2014-04-03 14:47:24 -07002835 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07002836
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002837 /* We set it here, as __alloc_pages_slowpath might have changed it */
2838 ac.zonelist = zonelist;
Mel Gorman5117f452009-06-16 15:31:59 -07002839 /* The preferred zone is used for statistics later */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002840 preferred_zoneref = first_zones_zonelist(ac.zonelist, ac.high_zoneidx,
2841 ac.nodemask ? : &cpuset_current_mems_allowed,
2842 &ac.preferred_zone);
2843 if (!ac.preferred_zone)
Mel Gormancc9a6c82012-03-21 16:34:11 -07002844 goto out;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002845 ac.classzone_idx = zonelist_zone_idx(preferred_zoneref);
Mel Gorman5117f452009-06-16 15:31:59 -07002846
2847 /* First allocation attempt */
Andrew Morton91fbdc02015-02-11 15:25:04 -08002848 alloc_mask = gfp_mask|__GFP_HARDWALL;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002849 page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
Ming Lei21caf2f2013-02-22 16:34:08 -08002850 if (unlikely(!page)) {
2851 /*
2852 * Runtime PM, block IO and its error handling path
2853 * can deadlock because I/O on the device might not
2854 * complete.
2855 */
Andrew Morton91fbdc02015-02-11 15:25:04 -08002856 alloc_mask = memalloc_noio_flags(gfp_mask);
2857
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002858 page = __alloc_pages_slowpath(alloc_mask, order, &ac);
Ming Lei21caf2f2013-02-22 16:34:08 -08002859 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002860
Xishi Qiu23f086f2015-02-11 15:25:07 -08002861 if (kmemcheck_enabled && page)
2862 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
2863
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002864 trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002865
2866out:
2867 /*
2868 * When updating a task's mems_allowed, it is possible to race with
2869 * parallel threads in such a way that an allocation can fail while
2870 * the mask is being updated. If a page allocation is about to fail,
2871 * check if the cpuset changed during allocation and if so, retry.
2872 */
Mel Gormand26914d2014-04-03 14:47:24 -07002873 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
Mel Gormancc9a6c82012-03-21 16:34:11 -07002874 goto retry_cpuset;
2875
Mel Gorman11e33f62009-06-16 15:31:57 -07002876 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002877}
Mel Gormand2391712009-06-16 15:31:52 -07002878EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879
2880/*
2881 * Common helper functions.
2882 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08002883unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884{
Akinobu Mita945a1112009-09-21 17:01:47 -07002885 struct page *page;
2886
2887 /*
2888 * __get_free_pages() returns a 32-bit address, which cannot represent
2889 * a highmem page
2890 */
2891 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2892
Linus Torvalds1da177e2005-04-16 15:20:36 -07002893 page = alloc_pages(gfp_mask, order);
2894 if (!page)
2895 return 0;
2896 return (unsigned long) page_address(page);
2897}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898EXPORT_SYMBOL(__get_free_pages);
2899
Harvey Harrison920c7a52008-02-04 22:29:26 -08002900unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002901{
Akinobu Mita945a1112009-09-21 17:01:47 -07002902 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002903}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002904EXPORT_SYMBOL(get_zeroed_page);
2905
Harvey Harrison920c7a52008-02-04 22:29:26 -08002906void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907{
Nick Pigginb5810032005-10-29 18:16:12 -07002908 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909 if (order == 0)
Mel Gormanb745bc82014-06-04 16:10:22 -07002910 free_hot_cold_page(page, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002911 else
2912 __free_pages_ok(page, order);
2913 }
2914}
2915
2916EXPORT_SYMBOL(__free_pages);
2917
Harvey Harrison920c7a52008-02-04 22:29:26 -08002918void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919{
2920 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07002921 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002922 __free_pages(virt_to_page((void *)addr), order);
2923 }
2924}
2925
2926EXPORT_SYMBOL(free_pages);
2927
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002928/*
Vladimir Davydov52383432014-06-04 16:06:39 -07002929 * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
2930 * of the current memory cgroup.
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002931 *
Vladimir Davydov52383432014-06-04 16:06:39 -07002932 * It should be used when the caller would like to use kmalloc, but since the
2933 * allocation is large, it has to fall back to the page allocator.
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002934 */
Vladimir Davydov52383432014-06-04 16:06:39 -07002935struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order)
2936{
2937 struct page *page;
2938 struct mem_cgroup *memcg = NULL;
2939
2940 if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
2941 return NULL;
2942 page = alloc_pages(gfp_mask, order);
2943 memcg_kmem_commit_charge(page, memcg, order);
2944 return page;
2945}
2946
2947struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
2948{
2949 struct page *page;
2950 struct mem_cgroup *memcg = NULL;
2951
2952 if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
2953 return NULL;
2954 page = alloc_pages_node(nid, gfp_mask, order);
2955 memcg_kmem_commit_charge(page, memcg, order);
2956 return page;
2957}
2958
2959/*
2960 * __free_kmem_pages and free_kmem_pages will free pages allocated with
2961 * alloc_kmem_pages.
2962 */
2963void __free_kmem_pages(struct page *page, unsigned int order)
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002964{
2965 memcg_kmem_uncharge_pages(page, order);
2966 __free_pages(page, order);
2967}
2968
Vladimir Davydov52383432014-06-04 16:06:39 -07002969void free_kmem_pages(unsigned long addr, unsigned int order)
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002970{
2971 if (addr != 0) {
2972 VM_BUG_ON(!virt_addr_valid((void *)addr));
Vladimir Davydov52383432014-06-04 16:06:39 -07002973 __free_kmem_pages(virt_to_page((void *)addr), order);
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002974 }
2975}
2976
Andi Kleenee85c2e2011-05-11 15:13:34 -07002977static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
2978{
2979 if (addr) {
2980 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2981 unsigned long used = addr + PAGE_ALIGN(size);
2982
2983 split_page(virt_to_page((void *)addr), order);
2984 while (used < alloc_end) {
2985 free_page(used);
2986 used += PAGE_SIZE;
2987 }
2988 }
2989 return (void *)addr;
2990}
2991
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002992/**
2993 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2994 * @size: the number of bytes to allocate
2995 * @gfp_mask: GFP flags for the allocation
2996 *
2997 * This function is similar to alloc_pages(), except that it allocates the
2998 * minimum number of pages to satisfy the request. alloc_pages() can only
2999 * allocate memory in power-of-two pages.
3000 *
3001 * This function is also limited by MAX_ORDER.
3002 *
3003 * Memory allocated by this function must be released by free_pages_exact().
3004 */
3005void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
3006{
3007 unsigned int order = get_order(size);
3008 unsigned long addr;
3009
3010 addr = __get_free_pages(gfp_mask, order);
Andi Kleenee85c2e2011-05-11 15:13:34 -07003011 return make_alloc_exact(addr, order, size);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07003012}
3013EXPORT_SYMBOL(alloc_pages_exact);
3014
3015/**
Andi Kleenee85c2e2011-05-11 15:13:34 -07003016 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
3017 * pages on a node.
Randy Dunlapb5e6ab52011-05-16 13:16:54 -07003018 * @nid: the preferred node ID where memory should be allocated
Andi Kleenee85c2e2011-05-11 15:13:34 -07003019 * @size: the number of bytes to allocate
3020 * @gfp_mask: GFP flags for the allocation
3021 *
3022 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
3023 * back.
3024 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
3025 * but is not exact.
3026 */
Fabian Fredericke1931812014-08-06 16:04:59 -07003027void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
Andi Kleenee85c2e2011-05-11 15:13:34 -07003028{
3029 unsigned order = get_order(size);
3030 struct page *p = alloc_pages_node(nid, gfp_mask, order);
3031 if (!p)
3032 return NULL;
3033 return make_alloc_exact((unsigned long)page_address(p), order, size);
3034}
Andi Kleenee85c2e2011-05-11 15:13:34 -07003035
3036/**
Timur Tabi2be0ffe2008-07-23 21:28:11 -07003037 * free_pages_exact - release memory allocated via alloc_pages_exact()
3038 * @virt: the value returned by alloc_pages_exact.
3039 * @size: size of allocation, same value as passed to alloc_pages_exact().
3040 *
3041 * Release the memory allocated by a previous call to alloc_pages_exact.
3042 */
3043void free_pages_exact(void *virt, size_t size)
3044{
3045 unsigned long addr = (unsigned long)virt;
3046 unsigned long end = addr + PAGE_ALIGN(size);
3047
3048 while (addr < end) {
3049 free_page(addr);
3050 addr += PAGE_SIZE;
3051 }
3052}
3053EXPORT_SYMBOL(free_pages_exact);
3054
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003055/**
3056 * nr_free_zone_pages - count number of pages beyond high watermark
3057 * @offset: The zone index of the highest zone
3058 *
3059 * nr_free_zone_pages() counts the number of counts pages which are beyond the
3060 * high watermark within all zones at or below a given zone index. For each
3061 * zone, the number of pages is calculated as:
Jiang Liu834405c2013-07-03 15:03:04 -07003062 * managed_pages - high_pages
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003063 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003064static unsigned long nr_free_zone_pages(int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003065{
Mel Gormandd1a2392008-04-28 02:12:17 -07003066 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003067 struct zone *zone;
3068
Martin J. Blighe310fd42005-07-29 22:59:18 -07003069 /* Just pick one node, since fallback list is circular */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003070 unsigned long sum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071
Mel Gorman0e884602008-04-28 02:12:14 -07003072 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003073
Mel Gorman54a6eb52008-04-28 02:12:16 -07003074 for_each_zone_zonelist(zone, z, zonelist, offset) {
Jiang Liub40da042013-02-22 16:33:52 -08003075 unsigned long size = zone->managed_pages;
Mel Gorman41858962009-06-16 15:32:12 -07003076 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07003077 if (size > high)
3078 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079 }
3080
3081 return sum;
3082}
3083
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003084/**
3085 * nr_free_buffer_pages - count number of pages beyond high watermark
3086 *
3087 * nr_free_buffer_pages() counts the number of pages which are beyond the high
3088 * watermark within ZONE_DMA and ZONE_NORMAL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003089 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003090unsigned long nr_free_buffer_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091{
Al Viroaf4ca452005-10-21 02:55:38 -04003092 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093}
Meelap Shahc2f1a552007-07-17 04:04:39 -07003094EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003095
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003096/**
3097 * nr_free_pagecache_pages - count number of pages beyond high watermark
3098 *
3099 * nr_free_pagecache_pages() counts the number of pages which are beyond the
3100 * high watermark within all zones.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003101 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003102unsigned long nr_free_pagecache_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003103{
Mel Gorman2a1e2742007-07-17 04:03:12 -07003104 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003105}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07003106
3107static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08003109 if (IS_ENABLED(CONFIG_NUMA))
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08003110 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003112
Linus Torvalds1da177e2005-04-16 15:20:36 -07003113void si_meminfo(struct sysinfo *val)
3114{
3115 val->totalram = totalram_pages;
Rafael Aquinicc7452b2014-08-06 16:06:38 -07003116 val->sharedram = global_page_state(NR_SHMEM);
Christoph Lameterd23ad422007-02-10 01:43:02 -08003117 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003118 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003119 val->totalhigh = totalhigh_pages;
3120 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003121 val->mem_unit = PAGE_SIZE;
3122}
3123
3124EXPORT_SYMBOL(si_meminfo);
3125
3126#ifdef CONFIG_NUMA
3127void si_meminfo_node(struct sysinfo *val, int nid)
3128{
Jiang Liucdd91a72013-07-03 15:03:27 -07003129 int zone_type; /* needs to be signed */
3130 unsigned long managed_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003131 pg_data_t *pgdat = NODE_DATA(nid);
3132
Jiang Liucdd91a72013-07-03 15:03:27 -07003133 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
3134 managed_pages += pgdat->node_zones[zone_type].managed_pages;
3135 val->totalram = managed_pages;
Rafael Aquinicc7452b2014-08-06 16:06:38 -07003136 val->sharedram = node_page_state(nid, NR_SHMEM);
Christoph Lameterd23ad422007-02-10 01:43:02 -08003137 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003138#ifdef CONFIG_HIGHMEM
Jiang Liub40da042013-02-22 16:33:52 -08003139 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08003140 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
3141 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003142#else
3143 val->totalhigh = 0;
3144 val->freehigh = 0;
3145#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003146 val->mem_unit = PAGE_SIZE;
3147}
3148#endif
3149
David Rientjesddd588b2011-03-22 16:30:46 -07003150/*
David Rientjes7bf02ea2011-05-24 17:11:16 -07003151 * Determine whether the node should be displayed or not, depending on whether
3152 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
David Rientjesddd588b2011-03-22 16:30:46 -07003153 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07003154bool skip_free_areas_node(unsigned int flags, int nid)
David Rientjesddd588b2011-03-22 16:30:46 -07003155{
3156 bool ret = false;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003157 unsigned int cpuset_mems_cookie;
David Rientjesddd588b2011-03-22 16:30:46 -07003158
3159 if (!(flags & SHOW_MEM_FILTER_NODES))
3160 goto out;
3161
Mel Gormancc9a6c82012-03-21 16:34:11 -07003162 do {
Mel Gormand26914d2014-04-03 14:47:24 -07003163 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07003164 ret = !node_isset(nid, cpuset_current_mems_allowed);
Mel Gormand26914d2014-04-03 14:47:24 -07003165 } while (read_mems_allowed_retry(cpuset_mems_cookie));
David Rientjesddd588b2011-03-22 16:30:46 -07003166out:
3167 return ret;
3168}
3169
Linus Torvalds1da177e2005-04-16 15:20:36 -07003170#define K(x) ((x) << (PAGE_SHIFT-10))
3171
Rabin Vincent377e4f12012-12-11 16:00:24 -08003172static void show_migration_types(unsigned char type)
3173{
3174 static const char types[MIGRATE_TYPES] = {
3175 [MIGRATE_UNMOVABLE] = 'U',
3176 [MIGRATE_RECLAIMABLE] = 'E',
3177 [MIGRATE_MOVABLE] = 'M',
3178 [MIGRATE_RESERVE] = 'R',
3179#ifdef CONFIG_CMA
3180 [MIGRATE_CMA] = 'C',
3181#endif
Minchan Kim194159f2013-02-22 16:33:58 -08003182#ifdef CONFIG_MEMORY_ISOLATION
Rabin Vincent377e4f12012-12-11 16:00:24 -08003183 [MIGRATE_ISOLATE] = 'I',
Minchan Kim194159f2013-02-22 16:33:58 -08003184#endif
Rabin Vincent377e4f12012-12-11 16:00:24 -08003185 };
3186 char tmp[MIGRATE_TYPES + 1];
3187 char *p = tmp;
3188 int i;
3189
3190 for (i = 0; i < MIGRATE_TYPES; i++) {
3191 if (type & (1 << i))
3192 *p++ = types[i];
3193 }
3194
3195 *p = '\0';
3196 printk("(%s) ", tmp);
3197}
3198
Linus Torvalds1da177e2005-04-16 15:20:36 -07003199/*
3200 * Show free area list (used inside shift_scroll-lock stuff)
3201 * We also calculate the percentage fragmentation. We do this by counting the
3202 * memory on each free list with the exception of the first item on the list.
David Rientjesddd588b2011-03-22 16:30:46 -07003203 * Suppresses nodes that are not allowed by current's cpuset if
3204 * SHOW_MEM_FILTER_NODES is passed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003205 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07003206void show_free_areas(unsigned int filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003207{
Jes Sorensenc7241912006-09-27 01:50:05 -07003208 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003209 struct zone *zone;
3210
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003211 for_each_populated_zone(zone) {
David Rientjes7bf02ea2011-05-24 17:11:16 -07003212 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003213 continue;
Jes Sorensenc7241912006-09-27 01:50:05 -07003214 show_node(zone);
3215 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003216
Dave Jones6b482c62005-11-10 15:45:56 -05003217 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003218 struct per_cpu_pageset *pageset;
3219
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003220 pageset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003221
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003222 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
3223 cpu, pageset->pcp.high,
3224 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003225 }
3226 }
3227
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003228 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3229 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003230 " unevictable:%lu"
Andrew Mortonb76146e2009-10-26 16:49:52 -07003231 " dirty:%lu writeback:%lu unstable:%lu\n"
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07003232 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003233 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3234 " free_cma:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003235 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003236 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003237 global_page_state(NR_ISOLATED_ANON),
3238 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003239 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003240 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003241 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07003242 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07003243 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07003244 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08003245 global_page_state(NR_FREE_PAGES),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07003246 global_page_state(NR_SLAB_RECLAIMABLE),
3247 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07003248 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003249 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08003250 global_page_state(NR_PAGETABLE),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003251 global_page_state(NR_BOUNCE),
3252 global_page_state(NR_FREE_CMA_PAGES));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003253
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003254 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003255 int i;
3256
David Rientjes7bf02ea2011-05-24 17:11:16 -07003257 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003258 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003259 show_node(zone);
3260 printk("%s"
3261 " free:%lukB"
3262 " min:%lukB"
3263 " low:%lukB"
3264 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003265 " active_anon:%lukB"
3266 " inactive_anon:%lukB"
3267 " active_file:%lukB"
3268 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003269 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003270 " isolated(anon):%lukB"
3271 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003272 " present:%lukB"
Jiang Liu9feedc92012-12-12 13:52:12 -08003273 " managed:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003274 " mlocked:%lukB"
3275 " dirty:%lukB"
3276 " writeback:%lukB"
3277 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003278 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003279 " slab_reclaimable:%lukB"
3280 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003281 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003282 " pagetables:%lukB"
3283 " unstable:%lukB"
3284 " bounce:%lukB"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003285 " free_cma:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003286 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003287 " pages_scanned:%lu"
3288 " all_unreclaimable? %s"
3289 "\n",
3290 zone->name,
Mel Gorman88f5acf2011-01-13 15:45:41 -08003291 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07003292 K(min_wmark_pages(zone)),
3293 K(low_wmark_pages(zone)),
3294 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003295 K(zone_page_state(zone, NR_ACTIVE_ANON)),
3296 K(zone_page_state(zone, NR_INACTIVE_ANON)),
3297 K(zone_page_state(zone, NR_ACTIVE_FILE)),
3298 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003299 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003300 K(zone_page_state(zone, NR_ISOLATED_ANON)),
3301 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003302 K(zone->present_pages),
Jiang Liu9feedc92012-12-12 13:52:12 -08003303 K(zone->managed_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003304 K(zone_page_state(zone, NR_MLOCK)),
3305 K(zone_page_state(zone, NR_FILE_DIRTY)),
3306 K(zone_page_state(zone, NR_WRITEBACK)),
3307 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003308 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003309 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3310 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003311 zone_page_state(zone, NR_KERNEL_STACK) *
3312 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003313 K(zone_page_state(zone, NR_PAGETABLE)),
3314 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3315 K(zone_page_state(zone, NR_BOUNCE)),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003316 K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003317 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Mel Gorman0d5d8232014-08-06 16:07:16 -07003318 K(zone_page_state(zone, NR_PAGES_SCANNED)),
Lisa Du6e543d52013-09-11 14:22:36 -07003319 (!zone_reclaimable(zone) ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07003320 );
3321 printk("lowmem_reserve[]:");
3322 for (i = 0; i < MAX_NR_ZONES; i++)
Mel Gorman3484b2d2014-08-06 16:07:14 -07003323 printk(" %ld", zone->lowmem_reserve[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003324 printk("\n");
3325 }
3326
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003327 for_each_populated_zone(zone) {
Pintu Kumarb8af2942013-09-11 14:20:34 -07003328 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003329 unsigned char types[MAX_ORDER];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003330
David Rientjes7bf02ea2011-05-24 17:11:16 -07003331 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003332 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003333 show_node(zone);
3334 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003335
3336 spin_lock_irqsave(&zone->lock, flags);
3337 for (order = 0; order < MAX_ORDER; order++) {
Rabin Vincent377e4f12012-12-11 16:00:24 -08003338 struct free_area *area = &zone->free_area[order];
3339 int type;
3340
3341 nr[order] = area->nr_free;
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003342 total += nr[order] << order;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003343
3344 types[order] = 0;
3345 for (type = 0; type < MIGRATE_TYPES; type++) {
3346 if (!list_empty(&area->free_list[type]))
3347 types[order] |= 1 << type;
3348 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003349 }
3350 spin_unlock_irqrestore(&zone->lock, flags);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003351 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003352 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003353 if (nr[order])
3354 show_migration_types(types[order]);
3355 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003356 printk("= %lukB\n", K(total));
3357 }
3358
David Rientjes949f7ec2013-04-29 15:07:48 -07003359 hugetlb_show_meminfo();
3360
Larry Woodmane6f36022008-02-04 22:29:30 -08003361 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
3362
Linus Torvalds1da177e2005-04-16 15:20:36 -07003363 show_swap_cache_info();
3364}
3365
Mel Gorman19770b32008-04-28 02:12:18 -07003366static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
3367{
3368 zoneref->zone = zone;
3369 zoneref->zone_idx = zone_idx(zone);
3370}
3371
Linus Torvalds1da177e2005-04-16 15:20:36 -07003372/*
3373 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08003374 *
3375 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003376 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003377static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003378 int nr_zones)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003379{
Christoph Lameter1a932052006-01-06 00:11:16 -08003380 struct zone *zone;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003381 enum zone_type zone_type = MAX_NR_ZONES;
Christoph Lameter02a68a52006-01-06 00:11:18 -08003382
3383 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003384 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08003385 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08003386 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07003387 zoneref_set_zone(zone,
3388 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08003389 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003390 }
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003391 } while (zone_type);
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003392
Christoph Lameter070f8032006-01-06 00:11:19 -08003393 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003394}
3395
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003396
3397/*
3398 * zonelist_order:
3399 * 0 = automatic detection of better ordering.
3400 * 1 = order by ([node] distance, -zonetype)
3401 * 2 = order by (-zonetype, [node] distance)
3402 *
3403 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3404 * the same zonelist. So only NUMA can configure this param.
3405 */
3406#define ZONELIST_ORDER_DEFAULT 0
3407#define ZONELIST_ORDER_NODE 1
3408#define ZONELIST_ORDER_ZONE 2
3409
3410/* zonelist order in the kernel.
3411 * set_zonelist_order() will set this to NODE or ZONE.
3412 */
3413static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
3414static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
3415
3416
Linus Torvalds1da177e2005-04-16 15:20:36 -07003417#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003418/* The value user specified ....changed by config */
3419static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3420/* string for sysctl */
3421#define NUMA_ZONELIST_ORDER_LEN 16
3422char numa_zonelist_order[16] = "default";
3423
3424/*
3425 * interface for configure zonelist ordering.
3426 * command line option "numa_zonelist_order"
3427 * = "[dD]efault - default, automatic configuration.
3428 * = "[nN]ode - order by node locality, then by zone within node
3429 * = "[zZ]one - order by zone, then by locality within zone
3430 */
3431
3432static int __parse_numa_zonelist_order(char *s)
3433{
3434 if (*s == 'd' || *s == 'D') {
3435 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3436 } else if (*s == 'n' || *s == 'N') {
3437 user_zonelist_order = ZONELIST_ORDER_NODE;
3438 } else if (*s == 'z' || *s == 'Z') {
3439 user_zonelist_order = ZONELIST_ORDER_ZONE;
3440 } else {
3441 printk(KERN_WARNING
3442 "Ignoring invalid numa_zonelist_order value: "
3443 "%s\n", s);
3444 return -EINVAL;
3445 }
3446 return 0;
3447}
3448
3449static __init int setup_numa_zonelist_order(char *s)
3450{
Volodymyr G. Lukiianykecb256f2011-01-13 15:46:26 -08003451 int ret;
3452
3453 if (!s)
3454 return 0;
3455
3456 ret = __parse_numa_zonelist_order(s);
3457 if (ret == 0)
3458 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
3459
3460 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003461}
3462early_param("numa_zonelist_order", setup_numa_zonelist_order);
3463
3464/*
3465 * sysctl handler for numa_zonelist_order
3466 */
Joe Perchescccad5b2014-06-06 14:38:09 -07003467int numa_zonelist_order_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003468 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003469 loff_t *ppos)
3470{
3471 char saved_string[NUMA_ZONELIST_ORDER_LEN];
3472 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01003473 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003474
Andi Kleen443c6f12009-12-23 21:00:47 +01003475 mutex_lock(&zl_order_mutex);
Chen Gangdacbde02013-07-03 15:02:35 -07003476 if (write) {
3477 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
3478 ret = -EINVAL;
3479 goto out;
3480 }
3481 strcpy(saved_string, (char *)table->data);
3482 }
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003483 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003484 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01003485 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003486 if (write) {
3487 int oldval = user_zonelist_order;
Chen Gangdacbde02013-07-03 15:02:35 -07003488
3489 ret = __parse_numa_zonelist_order((char *)table->data);
3490 if (ret) {
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003491 /*
3492 * bogus value. restore saved string
3493 */
Chen Gangdacbde02013-07-03 15:02:35 -07003494 strncpy((char *)table->data, saved_string,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003495 NUMA_ZONELIST_ORDER_LEN);
3496 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003497 } else if (oldval != user_zonelist_order) {
3498 mutex_lock(&zonelists_mutex);
Jiang Liu9adb62a2012-07-31 16:43:28 -07003499 build_all_zonelists(NULL, NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003500 mutex_unlock(&zonelists_mutex);
3501 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003502 }
Andi Kleen443c6f12009-12-23 21:00:47 +01003503out:
3504 mutex_unlock(&zl_order_mutex);
3505 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003506}
3507
3508
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003509#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003510static int node_load[MAX_NUMNODES];
3511
Linus Torvalds1da177e2005-04-16 15:20:36 -07003512/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07003513 * 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 -07003514 * @node: node whose fallback list we're appending
3515 * @used_node_mask: nodemask_t of already used nodes
3516 *
3517 * We use a number of factors to determine which is the next node that should
3518 * appear on a given node's fallback list. The node should not have appeared
3519 * already in @node's fallback list, and it should be the next closest node
3520 * according to the distance array (which contains arbitrary distance values
3521 * from each node to each node in the system), and should also prefer nodes
3522 * with no CPUs, since presumably they'll have very little allocation pressure
3523 * on them otherwise.
3524 * It returns -1 if no node is found.
3525 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003526static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003527{
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01003528 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003529 int min_val = INT_MAX;
David Rientjes00ef2d22013-02-22 16:35:36 -08003530 int best_node = NUMA_NO_NODE;
Rusty Russella70f7302009-03-13 14:49:46 +10303531 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003532
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01003533 /* Use the local node if we haven't already */
3534 if (!node_isset(node, *used_node_mask)) {
3535 node_set(node, *used_node_mask);
3536 return node;
3537 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003538
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08003539 for_each_node_state(n, N_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003540
3541 /* Don't want a node to appear more than once */
3542 if (node_isset(n, *used_node_mask))
3543 continue;
3544
Linus Torvalds1da177e2005-04-16 15:20:36 -07003545 /* Use the distance array to find the distance */
3546 val = node_distance(node, n);
3547
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01003548 /* Penalize nodes under us ("prefer the next node") */
3549 val += (n < node);
3550
Linus Torvalds1da177e2005-04-16 15:20:36 -07003551 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10303552 tmp = cpumask_of_node(n);
3553 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003554 val += PENALTY_FOR_NODE_WITH_CPUS;
3555
3556 /* Slight preference for less loaded node */
3557 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
3558 val += node_load[n];
3559
3560 if (val < min_val) {
3561 min_val = val;
3562 best_node = n;
3563 }
3564 }
3565
3566 if (best_node >= 0)
3567 node_set(best_node, *used_node_mask);
3568
3569 return best_node;
3570}
3571
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003572
3573/*
3574 * Build zonelists ordered by node and zones within node.
3575 * This results in maximum locality--normal zone overflows into local
3576 * DMA zone, if any--but risks exhausting DMA zone.
3577 */
3578static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003579{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003580 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003581 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003582
Mel Gorman54a6eb52008-04-28 02:12:16 -07003583 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07003584 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07003585 ;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003586 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gormandd1a2392008-04-28 02:12:17 -07003587 zonelist->_zonerefs[j].zone = NULL;
3588 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003589}
3590
3591/*
Christoph Lameter523b9452007-10-16 01:25:37 -07003592 * Build gfp_thisnode zonelists
3593 */
3594static void build_thisnode_zonelists(pg_data_t *pgdat)
3595{
Christoph Lameter523b9452007-10-16 01:25:37 -07003596 int j;
3597 struct zonelist *zonelist;
3598
Mel Gorman54a6eb52008-04-28 02:12:16 -07003599 zonelist = &pgdat->node_zonelists[1];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003600 j = build_zonelists_node(pgdat, zonelist, 0);
Mel Gormandd1a2392008-04-28 02:12:17 -07003601 zonelist->_zonerefs[j].zone = NULL;
3602 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07003603}
3604
3605/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003606 * Build zonelists ordered by zone and nodes within zones.
3607 * This results in conserving DMA zone[s] until all Normal memory is
3608 * exhausted, but results in overflowing to remote node while memory
3609 * may still exist in local DMA zone.
3610 */
3611static int node_order[MAX_NUMNODES];
3612
3613static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
3614{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003615 int pos, j, node;
3616 int zone_type; /* needs to be signed */
3617 struct zone *z;
3618 struct zonelist *zonelist;
3619
Mel Gorman54a6eb52008-04-28 02:12:16 -07003620 zonelist = &pgdat->node_zonelists[0];
3621 pos = 0;
3622 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
3623 for (j = 0; j < nr_nodes; j++) {
3624 node = node_order[j];
3625 z = &NODE_DATA(node)->node_zones[zone_type];
3626 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07003627 zoneref_set_zone(z,
3628 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07003629 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003630 }
3631 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003632 }
Mel Gormandd1a2392008-04-28 02:12:17 -07003633 zonelist->_zonerefs[pos].zone = NULL;
3634 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003635}
3636
Mel Gorman31939132014-10-09 15:28:30 -07003637#if defined(CONFIG_64BIT)
3638/*
3639 * Devices that require DMA32/DMA are relatively rare and do not justify a
3640 * penalty to every machine in case the specialised case applies. Default
3641 * to Node-ordering on 64-bit NUMA machines
3642 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003643static int default_zonelist_order(void)
3644{
Mel Gorman31939132014-10-09 15:28:30 -07003645 return ZONELIST_ORDER_NODE;
3646}
3647#else
3648/*
3649 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
3650 * by the kernel. If processes running on node 0 deplete the low memory zone
3651 * then reclaim will occur more frequency increasing stalls and potentially
3652 * be easier to OOM if a large percentage of the zone is under writeback or
3653 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
3654 * Hence, default to zone ordering on 32-bit.
3655 */
3656static int default_zonelist_order(void)
3657{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003658 return ZONELIST_ORDER_ZONE;
3659}
Mel Gorman31939132014-10-09 15:28:30 -07003660#endif /* CONFIG_64BIT */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003661
3662static void set_zonelist_order(void)
3663{
3664 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
3665 current_zonelist_order = default_zonelist_order();
3666 else
3667 current_zonelist_order = user_zonelist_order;
3668}
3669
3670static void build_zonelists(pg_data_t *pgdat)
3671{
3672 int j, node, load;
3673 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003674 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003675 int local_node, prev_node;
3676 struct zonelist *zonelist;
3677 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003678
3679 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07003680 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003681 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07003682 zonelist->_zonerefs[0].zone = NULL;
3683 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003684 }
3685
3686 /* NUMA-aware ordering of nodes */
3687 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003688 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003689 prev_node = local_node;
3690 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003691
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003692 memset(node_order, 0, sizeof(node_order));
3693 j = 0;
3694
Linus Torvalds1da177e2005-04-16 15:20:36 -07003695 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
3696 /*
3697 * We don't want to pressure a particular node.
3698 * So adding penalty to the first node in same
3699 * distance group to make it round-robin.
3700 */
David Rientjes957f8222012-10-08 16:33:24 -07003701 if (node_distance(local_node, node) !=
3702 node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003703 node_load[node] = load;
3704
Linus Torvalds1da177e2005-04-16 15:20:36 -07003705 prev_node = node;
3706 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003707 if (order == ZONELIST_ORDER_NODE)
3708 build_zonelists_in_node_order(pgdat, node);
3709 else
3710 node_order[j++] = node; /* remember order */
3711 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003712
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003713 if (order == ZONELIST_ORDER_ZONE) {
3714 /* calculate node order -- i.e., DMA last! */
3715 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003716 }
Christoph Lameter523b9452007-10-16 01:25:37 -07003717
3718 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003719}
3720
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003721/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003722static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003723{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003724 struct zonelist *zonelist;
3725 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07003726 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003727
Mel Gorman54a6eb52008-04-28 02:12:16 -07003728 zonelist = &pgdat->node_zonelists[0];
3729 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3730 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07003731 for (z = zonelist->_zonerefs; z->zone; z++)
3732 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003733}
3734
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003735#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3736/*
3737 * Return node id of node used for "local" allocations.
3738 * I.e., first node id of first zone in arg node's generic zonelist.
3739 * Used for initializing percpu 'numa_mem', which is used primarily
3740 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3741 */
3742int local_memory_node(int node)
3743{
3744 struct zone *zone;
3745
3746 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3747 gfp_zone(GFP_KERNEL),
3748 NULL,
3749 &zone);
3750 return zone->node;
3751}
3752#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003753
Linus Torvalds1da177e2005-04-16 15:20:36 -07003754#else /* CONFIG_NUMA */
3755
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003756static void set_zonelist_order(void)
3757{
3758 current_zonelist_order = ZONELIST_ORDER_ZONE;
3759}
3760
3761static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003762{
Christoph Lameter19655d32006-09-25 23:31:19 -07003763 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003764 enum zone_type j;
3765 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003766
3767 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003768
Mel Gorman54a6eb52008-04-28 02:12:16 -07003769 zonelist = &pgdat->node_zonelists[0];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003770 j = build_zonelists_node(pgdat, zonelist, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003771
Mel Gorman54a6eb52008-04-28 02:12:16 -07003772 /*
3773 * Now we build the zonelist so that it contains the zones
3774 * of all the other nodes.
3775 * We don't want to pressure a particular node, so when
3776 * building the zones for node N, we make sure that the
3777 * zones coming right after the local ones are those from
3778 * node N+1 (modulo N)
3779 */
3780 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3781 if (!node_online(node))
3782 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003783 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003784 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07003785 for (node = 0; node < local_node; node++) {
3786 if (!node_online(node))
3787 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003788 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gorman54a6eb52008-04-28 02:12:16 -07003789 }
3790
Mel Gormandd1a2392008-04-28 02:12:17 -07003791 zonelist->_zonerefs[j].zone = NULL;
3792 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003793}
3794
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003795/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003796static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003797{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003798 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003799}
3800
Linus Torvalds1da177e2005-04-16 15:20:36 -07003801#endif /* CONFIG_NUMA */
3802
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003803/*
3804 * Boot pageset table. One per cpu which is going to be used for all
3805 * zones and all nodes. The parameters will be set in such a way
3806 * that an item put on a list will immediately be handed over to
3807 * the buddy list. This is safe since pageset manipulation is done
3808 * with interrupts disabled.
3809 *
3810 * The boot_pagesets must be kept even after bootup is complete for
3811 * unused processors and/or zones. They do play a role for bootstrapping
3812 * hotplugged processors.
3813 *
3814 * zoneinfo_show() and maybe other functions do
3815 * not check if the processor is online before following the pageset pointer.
3816 * Other parts of the kernel may not check if the zone is available.
3817 */
3818static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3819static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07003820static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003821
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003822/*
3823 * Global mutex to protect against size modification of zonelists
3824 * as well as to serialize pageset setup for the new populated zone.
3825 */
3826DEFINE_MUTEX(zonelists_mutex);
3827
Rusty Russell9b1a4d32008-07-28 12:16:30 -05003828/* return values int ....just for stop_machine() */
Jiang Liu4ed7e022012-07-31 16:43:35 -07003829static int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003830{
Yasunori Goto68113782006-06-23 02:03:11 -07003831 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003832 int cpu;
Jiang Liu9adb62a2012-07-31 16:43:28 -07003833 pg_data_t *self = data;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003834
Bo Liu7f9cfb32009-08-18 14:11:19 -07003835#ifdef CONFIG_NUMA
3836 memset(node_load, 0, sizeof(node_load));
3837#endif
Jiang Liu9adb62a2012-07-31 16:43:28 -07003838
3839 if (self && !node_online(self->node_id)) {
3840 build_zonelists(self);
3841 build_zonelist_cache(self);
3842 }
3843
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003844 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07003845 pg_data_t *pgdat = NODE_DATA(nid);
3846
3847 build_zonelists(pgdat);
3848 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003849 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003850
3851 /*
3852 * Initialize the boot_pagesets that are going to be used
3853 * for bootstrapping processors. The real pagesets for
3854 * each zone will be allocated later when the per cpu
3855 * allocator is available.
3856 *
3857 * boot_pagesets are used also for bootstrapping offline
3858 * cpus if the system is already booted because the pagesets
3859 * are needed to initialize allocators on a specific cpu too.
3860 * F.e. the percpu allocator needs the page allocator which
3861 * needs the percpu allocator in order to allocate its pagesets
3862 * (a chicken-egg dilemma).
3863 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003864 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003865 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3866
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003867#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3868 /*
3869 * We now know the "local memory node" for each node--
3870 * i.e., the node of the first zone in the generic zonelist.
3871 * Set up numa_mem percpu variable for on-line cpus. During
3872 * boot, only the boot cpu should be on-line; we'll init the
3873 * secondary cpus' numa_mem as they come on-line. During
3874 * node/memory hotplug, we'll fixup all on-line cpus.
3875 */
3876 if (cpu_online(cpu))
3877 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3878#endif
3879 }
3880
Yasunori Goto68113782006-06-23 02:03:11 -07003881 return 0;
3882}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003883
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08003884static noinline void __init
3885build_all_zonelists_init(void)
3886{
3887 __build_all_zonelists(NULL);
3888 mminit_verify_zonelist();
3889 cpuset_init_current_mems_allowed();
3890}
3891
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003892/*
3893 * Called with zonelists_mutex held always
3894 * unless system_state == SYSTEM_BOOTING.
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08003895 *
3896 * __ref due to (1) call of __meminit annotated setup_zone_pageset
3897 * [we're only called with non-NULL zone through __meminit paths] and
3898 * (2) call of __init annotated helper build_all_zonelists_init
3899 * [protected by SYSTEM_BOOTING].
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003900 */
Jiang Liu9adb62a2012-07-31 16:43:28 -07003901void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
Yasunori Goto68113782006-06-23 02:03:11 -07003902{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003903 set_zonelist_order();
3904
Yasunori Goto68113782006-06-23 02:03:11 -07003905 if (system_state == SYSTEM_BOOTING) {
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08003906 build_all_zonelists_init();
Yasunori Goto68113782006-06-23 02:03:11 -07003907 } else {
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003908#ifdef CONFIG_MEMORY_HOTPLUG
Jiang Liu9adb62a2012-07-31 16:43:28 -07003909 if (zone)
3910 setup_zone_pageset(zone);
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003911#endif
Cody P Schaferdd1895e2013-07-03 15:01:36 -07003912 /* we have to stop all cpus to guarantee there is no user
3913 of zonelist */
Jiang Liu9adb62a2012-07-31 16:43:28 -07003914 stop_machine(__build_all_zonelists, pgdat, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07003915 /* cpuset refresh routine should be here */
3916 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07003917 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003918 /*
3919 * Disable grouping by mobility if the number of pages in the
3920 * system is too low to allow the mechanism to work. It would be
3921 * more accurate, but expensive to check per-zone. This check is
3922 * made on memory-hotadd so a system can start with mobility
3923 * disabled and enable it later
3924 */
Mel Gormand9c23402007-10-16 01:26:01 -07003925 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003926 page_group_by_mobility_disabled = 1;
3927 else
3928 page_group_by_mobility_disabled = 0;
3929
Anton Blanchardf88dfff2014-12-10 15:42:53 -08003930 pr_info("Built %i zonelists in %s order, mobility grouping %s. "
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003931 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003932 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003933 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003934 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003935 vm_total_pages);
3936#ifdef CONFIG_NUMA
Anton Blanchardf88dfff2014-12-10 15:42:53 -08003937 pr_info("Policy zone: %s\n", zone_names[policy_zone]);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003938#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003939}
3940
3941/*
3942 * Helper functions to size the waitqueue hash table.
3943 * Essentially these want to choose hash table sizes sufficiently
3944 * large so that collisions trying to wait on pages are rare.
3945 * But in fact, the number of active page waitqueues on typical
3946 * systems is ridiculously low, less than 200. So this is even
3947 * conservative, even though it seems large.
3948 *
3949 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3950 * waitqueues, i.e. the size of the waitq table given the number of pages.
3951 */
3952#define PAGES_PER_WAITQUEUE 256
3953
Yasunori Gotocca448f2006-06-23 02:03:10 -07003954#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07003955static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003956{
3957 unsigned long size = 1;
3958
3959 pages /= PAGES_PER_WAITQUEUE;
3960
3961 while (size < pages)
3962 size <<= 1;
3963
3964 /*
3965 * Once we have dozens or even hundreds of threads sleeping
3966 * on IO we've got bigger problems than wait queue collision.
3967 * Limit the size of the wait table to a reasonable size.
3968 */
3969 size = min(size, 4096UL);
3970
3971 return max(size, 4UL);
3972}
Yasunori Gotocca448f2006-06-23 02:03:10 -07003973#else
3974/*
3975 * A zone's size might be changed by hot-add, so it is not possible to determine
3976 * a suitable size for its wait_table. So we use the maximum size now.
3977 *
3978 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3979 *
3980 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3981 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3982 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3983 *
3984 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3985 * or more by the traditional way. (See above). It equals:
3986 *
3987 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3988 * ia64(16K page size) : = ( 8G + 4M)byte.
3989 * powerpc (64K page size) : = (32G +16M)byte.
3990 */
3991static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3992{
3993 return 4096UL;
3994}
3995#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003996
3997/*
3998 * This is an integer logarithm so that shifts can be used later
3999 * to extract the more random high bits from the multiplicative
4000 * hash function before the remainder is taken.
4001 */
4002static inline unsigned long wait_table_bits(unsigned long size)
4003{
4004 return ffz(~size);
4005}
4006
Mel Gorman56fd56b2007-10-16 01:25:58 -07004007/*
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07004008 * Check if a pageblock contains reserved pages
4009 */
4010static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
4011{
4012 unsigned long pfn;
4013
4014 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
4015 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
4016 return 1;
4017 }
4018 return 0;
4019}
4020
4021/*
Mel Gormand9c23402007-10-16 01:26:01 -07004022 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07004023 * of blocks reserved is based on min_wmark_pages(zone). The memory within
4024 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07004025 * higher will lead to a bigger reserve which will get freed as contiguous
4026 * blocks as reclaim kicks in
4027 */
4028static void setup_zone_migrate_reserve(struct zone *zone)
4029{
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07004030 unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004031 struct page *page;
Mel Gorman78986a62009-09-21 17:03:02 -07004032 unsigned long block_migratetype;
4033 int reserve;
Yasuaki Ishimatsu943dca12014-01-21 15:49:06 -08004034 int old_reserve;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004035
Michal Hockod02156382011-12-08 14:34:27 -08004036 /*
4037 * Get the start pfn, end pfn and the number of blocks to reserve
4038 * We have to be careful to be aligned to pageblock_nr_pages to
4039 * make sure that we always check pfn_valid for the first page in
4040 * the block.
4041 */
Mel Gorman56fd56b2007-10-16 01:25:58 -07004042 start_pfn = zone->zone_start_pfn;
Cody P Schafer108bcc92013-02-22 16:35:23 -08004043 end_pfn = zone_end_pfn(zone);
Michal Hockod02156382011-12-08 14:34:27 -08004044 start_pfn = roundup(start_pfn, pageblock_nr_pages);
Mel Gorman41858962009-06-16 15:32:12 -07004045 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07004046 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004047
Mel Gorman78986a62009-09-21 17:03:02 -07004048 /*
4049 * Reserve blocks are generally in place to help high-order atomic
4050 * allocations that are short-lived. A min_free_kbytes value that
4051 * would result in more than 2 reserve blocks for atomic allocations
4052 * is assumed to be in place to help anti-fragmentation for the
4053 * future allocation of hugepages at runtime.
4054 */
4055 reserve = min(2, reserve);
Yasuaki Ishimatsu943dca12014-01-21 15:49:06 -08004056 old_reserve = zone->nr_migrate_reserve_block;
4057
4058 /* When memory hot-add, we almost always need to do nothing */
4059 if (reserve == old_reserve)
4060 return;
4061 zone->nr_migrate_reserve_block = reserve;
Mel Gorman78986a62009-09-21 17:03:02 -07004062
Mel Gormand9c23402007-10-16 01:26:01 -07004063 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07004064 if (!pfn_valid(pfn))
4065 continue;
4066 page = pfn_to_page(pfn);
4067
Adam Litke344c7902008-09-02 14:35:38 -07004068 /* Watch out for overlapping nodes */
4069 if (page_to_nid(page) != zone_to_nid(zone))
4070 continue;
4071
Mel Gorman56fd56b2007-10-16 01:25:58 -07004072 block_migratetype = get_pageblock_migratetype(page);
4073
Mel Gorman938929f2012-01-10 15:07:14 -08004074 /* Only test what is necessary when the reserves are not met */
4075 if (reserve > 0) {
4076 /*
4077 * Blocks with reserved pages will never free, skip
4078 * them.
4079 */
4080 block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
4081 if (pageblock_is_reserved(pfn, block_end_pfn))
4082 continue;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004083
Mel Gorman938929f2012-01-10 15:07:14 -08004084 /* If this block is reserved, account for it */
4085 if (block_migratetype == MIGRATE_RESERVE) {
4086 reserve--;
4087 continue;
4088 }
4089
4090 /* Suitable for reserving if this block is movable */
4091 if (block_migratetype == MIGRATE_MOVABLE) {
4092 set_pageblock_migratetype(page,
4093 MIGRATE_RESERVE);
4094 move_freepages_block(zone, page,
4095 MIGRATE_RESERVE);
4096 reserve--;
4097 continue;
4098 }
Yasuaki Ishimatsu943dca12014-01-21 15:49:06 -08004099 } else if (!old_reserve) {
4100 /*
4101 * At boot time we don't need to scan the whole zone
4102 * for turning off MIGRATE_RESERVE.
4103 */
4104 break;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004105 }
4106
4107 /*
4108 * If the reserve is met and this is a previous reserved block,
4109 * take it back
4110 */
4111 if (block_migratetype == MIGRATE_RESERVE) {
4112 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4113 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4114 }
4115 }
4116}
Mel Gormanac0e5b72007-10-16 01:25:58 -07004117
Linus Torvalds1da177e2005-04-16 15:20:36 -07004118/*
4119 * Initially all pages are reserved - free ones are freed
4120 * up by free_all_bootmem() once the early boot process is
4121 * done. Non-atomic initialization, single-pass.
4122 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01004123void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08004124 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004125{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004126 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07004127 unsigned long end_pfn = start_pfn + size;
4128 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004129 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004130
Hugh Dickins22b31ee2009-01-06 14:40:09 -08004131 if (highest_memmap_pfn < end_pfn - 1)
4132 highest_memmap_pfn = end_pfn - 1;
4133
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004134 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08004135 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08004136 /*
4137 * There can be holes in boot-time mem_map[]s
4138 * handed to this function. They do not
4139 * exist on hotplugged memory.
4140 */
4141 if (context == MEMMAP_EARLY) {
4142 if (!early_pfn_valid(pfn))
4143 continue;
4144 if (!early_pfn_in_nid(pfn, nid))
4145 continue;
4146 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004147 page = pfn_to_page(pfn);
4148 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07004149 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08004150 init_page_count(page);
Mel Gorman22b751c2013-02-22 16:34:59 -08004151 page_mapcount_reset(page);
Peter Zijlstra90572892013-10-07 11:29:20 +01004152 page_cpupid_reset_last(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004153 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004154 /*
4155 * Mark the block movable so that blocks are reserved for
4156 * movable at startup. This will force kernel allocations
4157 * to reserve their blocks rather than leaking throughout
4158 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07004159 * kernel allocations are made. Later some blocks near
4160 * the start are marked MIGRATE_RESERVE by
4161 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004162 *
4163 * bitmap is created for zone's valid pfn range. but memmap
4164 * can be created for invalid pages (for alignment)
4165 * check here not to call set_pageblock_migratetype() against
4166 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004167 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004168 if ((z->zone_start_pfn <= pfn)
Cody P Schafer108bcc92013-02-22 16:35:23 -08004169 && (pfn < zone_end_pfn(z))
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004170 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07004171 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004172
Linus Torvalds1da177e2005-04-16 15:20:36 -07004173 INIT_LIST_HEAD(&page->lru);
4174#ifdef WANT_PAGE_VIRTUAL
4175 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
4176 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07004177 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004178#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004179 }
4180}
4181
Andi Kleen1e548de2008-02-04 22:29:26 -08004182static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004183{
Mel Gorman7aeb09f2014-06-04 16:10:21 -07004184 unsigned int order, t;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004185 for_each_migratetype_order(order, t) {
4186 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004187 zone->free_area[order].nr_free = 0;
4188 }
4189}
4190
4191#ifndef __HAVE_ARCH_MEMMAP_INIT
4192#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08004193 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004194#endif
4195
David Rientjes7cd2b0a2014-06-23 13:22:04 -07004196static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004197{
David Howells3a6be872009-05-06 16:03:03 -07004198#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004199 int batch;
4200
4201 /*
4202 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07004203 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004204 *
4205 * OK, so we don't know how big the cache is. So guess.
4206 */
Jiang Liub40da042013-02-22 16:33:52 -08004207 batch = zone->managed_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07004208 if (batch * PAGE_SIZE > 512 * 1024)
4209 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004210 batch /= 4; /* We effectively *= 4 below */
4211 if (batch < 1)
4212 batch = 1;
4213
4214 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004215 * Clamp the batch to a 2^n - 1 value. Having a power
4216 * of 2 value was found to be more likely to have
4217 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004218 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004219 * For example if 2 tasks are alternately allocating
4220 * batches of pages, one task can end up with a lot
4221 * of pages of one half of the possible page colors
4222 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004223 */
David Howells91552032009-05-06 16:03:02 -07004224 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07004225
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004226 return batch;
David Howells3a6be872009-05-06 16:03:03 -07004227
4228#else
4229 /* The deferral and batching of frees should be suppressed under NOMMU
4230 * conditions.
4231 *
4232 * The problem is that NOMMU needs to be able to allocate large chunks
4233 * of contiguous memory as there's no hardware page translation to
4234 * assemble apparent contiguous memory from discontiguous pages.
4235 *
4236 * Queueing large contiguous runs of pages for batching, however,
4237 * causes the pages to actually be freed in smaller chunks. As there
4238 * can be a significant delay between the individual batches being
4239 * recycled, this leads to the once large chunks of space being
4240 * fragmented and becoming unavailable for high-order allocations.
4241 */
4242 return 0;
4243#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004244}
4245
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004246/*
4247 * pcp->high and pcp->batch values are related and dependent on one another:
4248 * ->batch must never be higher then ->high.
4249 * The following function updates them in a safe manner without read side
4250 * locking.
4251 *
4252 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4253 * those fields changing asynchronously (acording the the above rule).
4254 *
4255 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4256 * outside of boot time (or some other assurance that no concurrent updaters
4257 * exist).
4258 */
4259static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
4260 unsigned long batch)
4261{
4262 /* start with a fail safe value for batch */
4263 pcp->batch = 1;
4264 smp_wmb();
4265
4266 /* Update high, then batch, in order */
4267 pcp->high = high;
4268 smp_wmb();
4269
4270 pcp->batch = batch;
4271}
4272
Cody P Schafer36640332013-07-03 15:01:40 -07004273/* a companion to pageset_set_high() */
Cody P Schafer4008bab2013-07-03 15:01:28 -07004274static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
4275{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004276 pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
Cody P Schafer4008bab2013-07-03 15:01:28 -07004277}
4278
Cody P Schafer88c90db2013-07-03 15:01:35 -07004279static void pageset_init(struct per_cpu_pageset *p)
Christoph Lameter2caaad42005-06-21 17:15:00 -07004280{
4281 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004282 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004283
Magnus Damm1c6fe942005-10-26 01:58:59 -07004284 memset(p, 0, sizeof(*p));
4285
Christoph Lameter3dfa5722008-02-04 22:29:19 -08004286 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004287 pcp->count = 0;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004288 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
4289 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07004290}
4291
Cody P Schafer88c90db2013-07-03 15:01:35 -07004292static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
4293{
4294 pageset_init(p);
4295 pageset_set_batch(p, batch);
4296}
4297
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004298/*
Cody P Schafer36640332013-07-03 15:01:40 -07004299 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004300 * to the value high for the pageset p.
4301 */
Cody P Schafer36640332013-07-03 15:01:40 -07004302static void pageset_set_high(struct per_cpu_pageset *p,
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004303 unsigned long high)
4304{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004305 unsigned long batch = max(1UL, high / 4);
4306 if ((high / 4) > (PAGE_SHIFT * 8))
4307 batch = PAGE_SHIFT * 8;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004308
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004309 pageset_update(&p->pcp, high, batch);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004310}
4311
David Rientjes7cd2b0a2014-06-23 13:22:04 -07004312static void pageset_set_high_and_batch(struct zone *zone,
4313 struct per_cpu_pageset *pcp)
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004314{
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004315 if (percpu_pagelist_fraction)
Cody P Schafer36640332013-07-03 15:01:40 -07004316 pageset_set_high(pcp,
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004317 (zone->managed_pages /
4318 percpu_pagelist_fraction));
4319 else
4320 pageset_set_batch(pcp, zone_batchsize(zone));
4321}
4322
Cody P Schafer169f6c12013-07-03 15:01:41 -07004323static void __meminit zone_pageset_init(struct zone *zone, int cpu)
4324{
4325 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4326
4327 pageset_init(pcp);
4328 pageset_set_high_and_batch(zone, pcp);
4329}
4330
Jiang Liu4ed7e022012-07-31 16:43:35 -07004331static void __meminit setup_zone_pageset(struct zone *zone)
Wu Fengguang319774e2010-05-24 14:32:49 -07004332{
4333 int cpu;
Wu Fengguang319774e2010-05-24 14:32:49 -07004334 zone->pageset = alloc_percpu(struct per_cpu_pageset);
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004335 for_each_possible_cpu(cpu)
4336 zone_pageset_init(zone, cpu);
Wu Fengguang319774e2010-05-24 14:32:49 -07004337}
4338
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004339/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004340 * Allocate per cpu pagesets and initialize them.
4341 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07004342 */
Al Viro78d99552005-12-15 09:18:25 +00004343void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004344{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004345 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004346
Wu Fengguang319774e2010-05-24 14:32:49 -07004347 for_each_populated_zone(zone)
4348 setup_zone_pageset(zone);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004349}
4350
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004351static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07004352int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07004353{
4354 int i;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004355 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07004356
4357 /*
4358 * The per-page waitqueue mechanism uses hashed waitqueues
4359 * per zone.
4360 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07004361 zone->wait_table_hash_nr_entries =
4362 wait_table_hash_nr_entries(zone_size_pages);
4363 zone->wait_table_bits =
4364 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004365 alloc_size = zone->wait_table_hash_nr_entries
4366 * sizeof(wait_queue_head_t);
4367
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07004368 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07004369 zone->wait_table = (wait_queue_head_t *)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004370 memblock_virt_alloc_node_nopanic(
4371 alloc_size, zone->zone_pgdat->node_id);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004372 } else {
4373 /*
4374 * This case means that a zone whose size was 0 gets new memory
4375 * via memory hot-add.
4376 * But it may be the case that a new node was hot-added. In
4377 * this case vmalloc() will not be able to use this new node's
4378 * memory - this wait_table must be initialized to use this new
4379 * node itself as well.
4380 * To use this new node's memory, further consideration will be
4381 * necessary.
4382 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07004383 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004384 }
4385 if (!zone->wait_table)
4386 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07004387
Pintu Kumarb8af2942013-09-11 14:20:34 -07004388 for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07004389 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004390
4391 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004392}
4393
Matt Tolentinoc09b4242006-01-17 07:03:44 +01004394static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07004395{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004396 /*
4397 * per cpu subsystem is not up at this point. The following code
4398 * relies on the ability of the linker to provide the
4399 * offset of a (static) per cpu variable into the per cpu area.
4400 */
4401 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07004402
Xishi Qiub38a8722013-11-12 15:07:20 -08004403 if (populated_zone(zone))
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004404 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
4405 zone->name, zone->present_pages,
4406 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07004407}
4408
Jiang Liu4ed7e022012-07-31 16:43:35 -07004409int __meminit init_currently_empty_zone(struct zone *zone,
Yasunori Goto718127c2006-06-23 02:03:10 -07004410 unsigned long zone_start_pfn,
Dave Hansena2f3aa022007-01-10 23:15:30 -08004411 unsigned long size,
4412 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07004413{
4414 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004415 int ret;
4416 ret = zone_wait_table_init(zone, size);
4417 if (ret)
4418 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07004419 pgdat->nr_zones = zone_idx(zone) + 1;
4420
Dave Hansened8ece22005-10-29 18:16:50 -07004421 zone->zone_start_pfn = zone_start_pfn;
4422
Mel Gorman708614e2008-07-23 21:26:51 -07004423 mminit_dprintk(MMINIT_TRACE, "memmap_init",
4424 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4425 pgdat->node_id,
4426 (unsigned long)zone_idx(zone),
4427 zone_start_pfn, (zone_start_pfn + size));
4428
Andi Kleen1e548de2008-02-04 22:29:26 -08004429 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07004430
4431 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004432}
4433
Tejun Heo0ee332c2011-12-08 10:22:09 -08004434#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07004435#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4436/*
4437 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
Mel Gormanc7132162006-09-27 01:49:43 -07004438 */
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004439int __meminit __early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004440{
Tejun Heoc13291a2011-07-12 10:46:30 +02004441 unsigned long start_pfn, end_pfn;
Yinghai Lue76b63f2013-09-11 14:22:17 -07004442 int nid;
Russ Anderson7c243c72013-04-29 15:07:59 -07004443 /*
4444 * NOTE: The following SMP-unsafe globals are only used early in boot
4445 * when the kernel is running single-threaded.
4446 */
4447 static unsigned long __meminitdata last_start_pfn, last_end_pfn;
4448 static int __meminitdata last_nid;
4449
4450 if (last_start_pfn <= pfn && pfn < last_end_pfn)
4451 return last_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004452
Yinghai Lue76b63f2013-09-11 14:22:17 -07004453 nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
4454 if (nid != -1) {
4455 last_start_pfn = start_pfn;
4456 last_end_pfn = end_pfn;
4457 last_nid = nid;
4458 }
4459
4460 return nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004461}
4462#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4463
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004464int __meminit early_pfn_to_nid(unsigned long pfn)
4465{
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08004466 int nid;
4467
4468 nid = __early_pfn_to_nid(pfn);
4469 if (nid >= 0)
4470 return nid;
4471 /* just returns 0 */
4472 return 0;
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004473}
4474
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08004475#ifdef CONFIG_NODES_SPAN_OTHER_NODES
4476bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
4477{
4478 int nid;
4479
4480 nid = __early_pfn_to_nid(pfn);
4481 if (nid >= 0 && nid != node)
4482 return false;
4483 return true;
4484}
4485#endif
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004486
Mel Gormanc7132162006-09-27 01:49:43 -07004487/**
Santosh Shilimkar67828322014-01-21 15:50:25 -08004488 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004489 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
Santosh Shilimkar67828322014-01-21 15:50:25 -08004490 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
Mel Gormanc7132162006-09-27 01:49:43 -07004491 *
Zhang Zhen7d018172014-06-04 16:10:53 -07004492 * If an architecture guarantees that all ranges registered contain no holes
4493 * and may be freed, this this function may be used instead of calling
4494 * memblock_free_early_nid() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07004495 */
Tejun Heoc13291a2011-07-12 10:46:30 +02004496void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004497{
Tejun Heoc13291a2011-07-12 10:46:30 +02004498 unsigned long start_pfn, end_pfn;
4499 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004500
Tejun Heoc13291a2011-07-12 10:46:30 +02004501 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
4502 start_pfn = min(start_pfn, max_low_pfn);
4503 end_pfn = min(end_pfn, max_low_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004504
Tejun Heoc13291a2011-07-12 10:46:30 +02004505 if (start_pfn < end_pfn)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004506 memblock_free_early_nid(PFN_PHYS(start_pfn),
4507 (end_pfn - start_pfn) << PAGE_SHIFT,
4508 this_nid);
Mel Gormanc7132162006-09-27 01:49:43 -07004509 }
4510}
4511
4512/**
4513 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004514 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07004515 *
Zhang Zhen7d018172014-06-04 16:10:53 -07004516 * If an architecture guarantees that all ranges registered contain no holes and may
4517 * be freed, this function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07004518 */
4519void __init sparse_memory_present_with_active_regions(int nid)
4520{
Tejun Heoc13291a2011-07-12 10:46:30 +02004521 unsigned long start_pfn, end_pfn;
4522 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004523
Tejun Heoc13291a2011-07-12 10:46:30 +02004524 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
4525 memory_present(this_nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004526}
4527
4528/**
4529 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004530 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4531 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4532 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07004533 *
4534 * It returns the start and end page frame of a node based on information
Zhang Zhen7d018172014-06-04 16:10:53 -07004535 * provided by memblock_set_node(). If called for a node
Mel Gormanc7132162006-09-27 01:49:43 -07004536 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004537 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07004538 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004539void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004540 unsigned long *start_pfn, unsigned long *end_pfn)
4541{
Tejun Heoc13291a2011-07-12 10:46:30 +02004542 unsigned long this_start_pfn, this_end_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004543 int i;
Tejun Heoc13291a2011-07-12 10:46:30 +02004544
Mel Gormanc7132162006-09-27 01:49:43 -07004545 *start_pfn = -1UL;
4546 *end_pfn = 0;
4547
Tejun Heoc13291a2011-07-12 10:46:30 +02004548 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
4549 *start_pfn = min(*start_pfn, this_start_pfn);
4550 *end_pfn = max(*end_pfn, this_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004551 }
4552
Christoph Lameter633c0662007-10-16 01:25:37 -07004553 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07004554 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07004555}
4556
4557/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07004558 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4559 * assumption is made that zones within a node are ordered in monotonic
4560 * increasing memory addresses so that the "highest" populated zone is used
4561 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004562static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004563{
4564 int zone_index;
4565 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
4566 if (zone_index == ZONE_MOVABLE)
4567 continue;
4568
4569 if (arch_zone_highest_possible_pfn[zone_index] >
4570 arch_zone_lowest_possible_pfn[zone_index])
4571 break;
4572 }
4573
4574 VM_BUG_ON(zone_index == -1);
4575 movable_zone = zone_index;
4576}
4577
4578/*
4579 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004580 * because it is sized independent of architecture. Unlike the other zones,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004581 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4582 * in each node depending on the size of each node and how evenly kernelcore
4583 * is distributed. This helper function adjusts the zone ranges
4584 * provided by the architecture for a given node by using the end of the
4585 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4586 * zones within a node are in order of monotonic increases memory addresses
4587 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004588static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004589 unsigned long zone_type,
4590 unsigned long node_start_pfn,
4591 unsigned long node_end_pfn,
4592 unsigned long *zone_start_pfn,
4593 unsigned long *zone_end_pfn)
4594{
4595 /* Only adjust if ZONE_MOVABLE is on this node */
4596 if (zone_movable_pfn[nid]) {
4597 /* Size ZONE_MOVABLE */
4598 if (zone_type == ZONE_MOVABLE) {
4599 *zone_start_pfn = zone_movable_pfn[nid];
4600 *zone_end_pfn = min(node_end_pfn,
4601 arch_zone_highest_possible_pfn[movable_zone]);
4602
4603 /* Adjust for ZONE_MOVABLE starting within this range */
4604 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
4605 *zone_end_pfn > zone_movable_pfn[nid]) {
4606 *zone_end_pfn = zone_movable_pfn[nid];
4607
4608 /* Check if this whole range is within ZONE_MOVABLE */
4609 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
4610 *zone_start_pfn = *zone_end_pfn;
4611 }
4612}
4613
4614/*
Mel Gormanc7132162006-09-27 01:49:43 -07004615 * Return the number of pages a zone spans in a node, including holes
4616 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4617 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004618static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004619 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004620 unsigned long node_start_pfn,
4621 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004622 unsigned long *ignored)
4623{
Mel Gormanc7132162006-09-27 01:49:43 -07004624 unsigned long zone_start_pfn, zone_end_pfn;
4625
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004626 /* Get the start and end of the zone */
Mel Gormanc7132162006-09-27 01:49:43 -07004627 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4628 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07004629 adjust_zone_range_for_zone_movable(nid, zone_type,
4630 node_start_pfn, node_end_pfn,
4631 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004632
4633 /* Check that this node has pages within the zone's required range */
4634 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4635 return 0;
4636
4637 /* Move the zone boundaries inside the node if necessary */
4638 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4639 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4640
4641 /* Return the spanned pages */
4642 return zone_end_pfn - zone_start_pfn;
4643}
4644
4645/*
4646 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004647 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07004648 */
Yinghai Lu32996252009-12-15 17:59:02 -08004649unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004650 unsigned long range_start_pfn,
4651 unsigned long range_end_pfn)
4652{
Tejun Heo96e907d2011-07-12 10:46:29 +02004653 unsigned long nr_absent = range_end_pfn - range_start_pfn;
4654 unsigned long start_pfn, end_pfn;
4655 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004656
Tejun Heo96e907d2011-07-12 10:46:29 +02004657 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
4658 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
4659 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
4660 nr_absent -= end_pfn - start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004661 }
Tejun Heo96e907d2011-07-12 10:46:29 +02004662 return nr_absent;
Mel Gormanc7132162006-09-27 01:49:43 -07004663}
4664
4665/**
4666 * absent_pages_in_range - Return number of page frames in holes within a range
4667 * @start_pfn: The start PFN to start searching for holes
4668 * @end_pfn: The end PFN to stop searching for holes
4669 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004670 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07004671 */
4672unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4673 unsigned long end_pfn)
4674{
4675 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4676}
4677
4678/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004679static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004680 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004681 unsigned long node_start_pfn,
4682 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004683 unsigned long *ignored)
4684{
Tejun Heo96e907d2011-07-12 10:46:29 +02004685 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
4686 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman9c7cd682006-09-27 01:49:58 -07004687 unsigned long zone_start_pfn, zone_end_pfn;
4688
Tejun Heo96e907d2011-07-12 10:46:29 +02004689 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
4690 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004691
Mel Gorman2a1e2742007-07-17 04:03:12 -07004692 adjust_zone_range_for_zone_movable(nid, zone_type,
4693 node_start_pfn, node_end_pfn,
4694 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004695 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004696}
Mel Gorman0e0b8642006-09-27 01:49:56 -07004697
Tejun Heo0ee332c2011-12-08 10:22:09 -08004698#else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004699static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004700 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004701 unsigned long node_start_pfn,
4702 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004703 unsigned long *zones_size)
4704{
4705 return zones_size[zone_type];
4706}
4707
Paul Mundt6ea6e682007-07-15 23:38:20 -07004708static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004709 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004710 unsigned long node_start_pfn,
4711 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004712 unsigned long *zholes_size)
4713{
4714 if (!zholes_size)
4715 return 0;
4716
4717 return zholes_size[zone_type];
4718}
Yinghai Lu20e69262013-03-01 14:51:27 -08004719
Tejun Heo0ee332c2011-12-08 10:22:09 -08004720#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07004721
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004722static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004723 unsigned long node_start_pfn,
4724 unsigned long node_end_pfn,
4725 unsigned long *zones_size,
4726 unsigned long *zholes_size)
Mel Gormanc7132162006-09-27 01:49:43 -07004727{
4728 unsigned long realtotalpages, totalpages = 0;
4729 enum zone_type i;
4730
4731 for (i = 0; i < MAX_NR_ZONES; i++)
4732 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004733 node_start_pfn,
4734 node_end_pfn,
4735 zones_size);
Mel Gormanc7132162006-09-27 01:49:43 -07004736 pgdat->node_spanned_pages = totalpages;
4737
4738 realtotalpages = totalpages;
4739 for (i = 0; i < MAX_NR_ZONES; i++)
4740 realtotalpages -=
4741 zone_absent_pages_in_node(pgdat->node_id, i,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004742 node_start_pfn, node_end_pfn,
4743 zholes_size);
Mel Gormanc7132162006-09-27 01:49:43 -07004744 pgdat->node_present_pages = realtotalpages;
4745 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4746 realtotalpages);
4747}
4748
Mel Gorman835c1342007-10-16 01:25:47 -07004749#ifndef CONFIG_SPARSEMEM
4750/*
4751 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07004752 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4753 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07004754 * round what is now in bits to nearest long in bits, then return it in
4755 * bytes.
4756 */
Linus Torvalds7c455122013-02-18 09:58:02 -08004757static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07004758{
4759 unsigned long usemapsize;
4760
Linus Torvalds7c455122013-02-18 09:58:02 -08004761 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
Mel Gormand9c23402007-10-16 01:26:01 -07004762 usemapsize = roundup(zonesize, pageblock_nr_pages);
4763 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07004764 usemapsize *= NR_PAGEBLOCK_BITS;
4765 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4766
4767 return usemapsize / 8;
4768}
4769
4770static void __init setup_usemap(struct pglist_data *pgdat,
Linus Torvalds7c455122013-02-18 09:58:02 -08004771 struct zone *zone,
4772 unsigned long zone_start_pfn,
4773 unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07004774{
Linus Torvalds7c455122013-02-18 09:58:02 -08004775 unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
Mel Gorman835c1342007-10-16 01:25:47 -07004776 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08004777 if (usemapsize)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004778 zone->pageblock_flags =
4779 memblock_virt_alloc_node_nopanic(usemapsize,
4780 pgdat->node_id);
Mel Gorman835c1342007-10-16 01:25:47 -07004781}
4782#else
Linus Torvalds7c455122013-02-18 09:58:02 -08004783static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
4784 unsigned long zone_start_pfn, unsigned long zonesize) {}
Mel Gorman835c1342007-10-16 01:25:47 -07004785#endif /* CONFIG_SPARSEMEM */
4786
Mel Gormand9c23402007-10-16 01:26:01 -07004787#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08004788
Mel Gormand9c23402007-10-16 01:26:01 -07004789/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
Chen Gang15ca2202013-09-11 14:20:27 -07004790void __paginginit set_pageblock_order(void)
Mel Gormand9c23402007-10-16 01:26:01 -07004791{
Andrew Morton955c1cd2012-05-29 15:06:31 -07004792 unsigned int order;
4793
Mel Gormand9c23402007-10-16 01:26:01 -07004794 /* Check that pageblock_nr_pages has not already been setup */
4795 if (pageblock_order)
4796 return;
4797
Andrew Morton955c1cd2012-05-29 15:06:31 -07004798 if (HPAGE_SHIFT > PAGE_SHIFT)
4799 order = HUGETLB_PAGE_ORDER;
4800 else
4801 order = MAX_ORDER - 1;
4802
Mel Gormand9c23402007-10-16 01:26:01 -07004803 /*
4804 * Assume the largest contiguous order of interest is a huge page.
Andrew Morton955c1cd2012-05-29 15:06:31 -07004805 * This value may be variable depending on boot parameters on IA64 and
4806 * powerpc.
Mel Gormand9c23402007-10-16 01:26:01 -07004807 */
4808 pageblock_order = order;
4809}
4810#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4811
Mel Gormanba72cb82007-11-28 16:21:13 -08004812/*
4813 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
Andrew Morton955c1cd2012-05-29 15:06:31 -07004814 * is unused as pageblock_order is set at compile-time. See
4815 * include/linux/pageblock-flags.h for the values of pageblock_order based on
4816 * the kernel config
Mel Gormanba72cb82007-11-28 16:21:13 -08004817 */
Chen Gang15ca2202013-09-11 14:20:27 -07004818void __paginginit set_pageblock_order(void)
Mel Gormanba72cb82007-11-28 16:21:13 -08004819{
Mel Gormanba72cb82007-11-28 16:21:13 -08004820}
Mel Gormand9c23402007-10-16 01:26:01 -07004821
4822#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4823
Jiang Liu01cefae2012-12-12 13:52:19 -08004824static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
4825 unsigned long present_pages)
4826{
4827 unsigned long pages = spanned_pages;
4828
4829 /*
4830 * Provide a more accurate estimation if there are holes within
4831 * the zone and SPARSEMEM is in use. If there are holes within the
4832 * zone, each populated memory region may cost us one or two extra
4833 * memmap pages due to alignment because memmap pages for each
4834 * populated regions may not naturally algined on page boundary.
4835 * So the (present_pages >> 4) heuristic is a tradeoff for that.
4836 */
4837 if (spanned_pages > present_pages + (present_pages >> 4) &&
4838 IS_ENABLED(CONFIG_SPARSEMEM))
4839 pages = present_pages;
4840
4841 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
4842}
4843
Linus Torvalds1da177e2005-04-16 15:20:36 -07004844/*
4845 * Set up the zone data structures:
4846 * - mark all pages reserved
4847 * - mark all memory queues empty
4848 * - clear the memory bitmaps
Minchan Kim6527af52012-07-31 16:46:16 -07004849 *
4850 * NOTE: pgdat should get zeroed by caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004851 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08004852static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004853 unsigned long node_start_pfn, unsigned long node_end_pfn,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004854 unsigned long *zones_size, unsigned long *zholes_size)
4855{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07004856 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07004857 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004858 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07004859 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004860
Dave Hansen208d54e2005-10-29 18:16:52 -07004861 pgdat_resize_init(pgdat);
Andrea Arcangeli8177a422012-03-23 20:56:34 +01004862#ifdef CONFIG_NUMA_BALANCING
4863 spin_lock_init(&pgdat->numabalancing_migrate_lock);
4864 pgdat->numabalancing_migrate_nr_pages = 0;
4865 pgdat->numabalancing_migrate_next_window = jiffies;
4866#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004867 init_waitqueue_head(&pgdat->kswapd_wait);
Mel Gorman55150612012-07-31 16:44:35 -07004868 init_waitqueue_head(&pgdat->pfmemalloc_wait);
Joonsoo Kimeefa864b2014-12-12 16:55:46 -08004869 pgdat_page_ext_init(pgdat);
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01004870
Linus Torvalds1da177e2005-04-16 15:20:36 -07004871 for (j = 0; j < MAX_NR_ZONES; j++) {
4872 struct zone *zone = pgdat->node_zones + j;
Jiang Liu9feedc92012-12-12 13:52:12 -08004873 unsigned long size, realsize, freesize, memmap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004874
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004875 size = zone_spanned_pages_in_node(nid, j, node_start_pfn,
4876 node_end_pfn, zones_size);
Jiang Liu9feedc92012-12-12 13:52:12 -08004877 realsize = freesize = size - zone_absent_pages_in_node(nid, j,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004878 node_start_pfn,
4879 node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004880 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004881
Mel Gorman0e0b8642006-09-27 01:49:56 -07004882 /*
Jiang Liu9feedc92012-12-12 13:52:12 -08004883 * Adjust freesize so that it accounts for how much memory
Mel Gorman0e0b8642006-09-27 01:49:56 -07004884 * is used by this zone for memmap. This affects the watermark
4885 * and per-cpu initialisations
4886 */
Jiang Liu01cefae2012-12-12 13:52:19 -08004887 memmap_pages = calc_memmap_size(size, realsize);
Zhong Hongboba914f42014-12-12 16:56:21 -08004888 if (!is_highmem_idx(j)) {
4889 if (freesize >= memmap_pages) {
4890 freesize -= memmap_pages;
4891 if (memmap_pages)
4892 printk(KERN_DEBUG
4893 " %s zone: %lu pages used for memmap\n",
4894 zone_names[j], memmap_pages);
4895 } else
4896 printk(KERN_WARNING
4897 " %s zone: %lu pages exceeds freesize %lu\n",
4898 zone_names[j], memmap_pages, freesize);
4899 }
Mel Gorman0e0b8642006-09-27 01:49:56 -07004900
Christoph Lameter62672762007-02-10 01:43:07 -08004901 /* Account for reserved pages */
Jiang Liu9feedc92012-12-12 13:52:12 -08004902 if (j == 0 && freesize > dma_reserve) {
4903 freesize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07004904 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08004905 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004906 }
4907
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004908 if (!is_highmem_idx(j))
Jiang Liu9feedc92012-12-12 13:52:12 -08004909 nr_kernel_pages += freesize;
Jiang Liu01cefae2012-12-12 13:52:19 -08004910 /* Charge for highmem memmap if there are enough kernel pages */
4911 else if (nr_kernel_pages > memmap_pages * 2)
4912 nr_kernel_pages -= memmap_pages;
Jiang Liu9feedc92012-12-12 13:52:12 -08004913 nr_all_pages += freesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004914
4915 zone->spanned_pages = size;
Jiang Liu306f2e92013-02-22 16:33:54 -08004916 zone->present_pages = realsize;
Jiang Liu9feedc92012-12-12 13:52:12 -08004917 /*
4918 * Set an approximate value for lowmem here, it will be adjusted
4919 * when the bootmem allocator frees pages into the buddy system.
4920 * And all highmem pages will be managed by the buddy system.
4921 */
4922 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
Christoph Lameter96146342006-07-03 00:24:13 -07004923#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07004924 zone->node = nid;
Jiang Liu9feedc92012-12-12 13:52:12 -08004925 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07004926 / 100;
Jiang Liu9feedc92012-12-12 13:52:12 -08004927 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07004928#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004929 zone->name = zone_names[j];
4930 spin_lock_init(&zone->lock);
4931 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07004932 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004933 zone->zone_pgdat = pgdat;
Dave Hansened8ece22005-10-29 18:16:50 -07004934 zone_pcp_init(zone);
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07004935
4936 /* For bootup, initialized properly in watermark setup */
4937 mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages);
4938
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004939 lruvec_init(&zone->lruvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004940 if (!size)
4941 continue;
4942
Andrew Morton955c1cd2012-05-29 15:06:31 -07004943 set_pageblock_order();
Linus Torvalds7c455122013-02-18 09:58:02 -08004944 setup_usemap(pgdat, zone, zone_start_pfn, size);
Dave Hansena2f3aa022007-01-10 23:15:30 -08004945 ret = init_currently_empty_zone(zone, zone_start_pfn,
4946 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07004947 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07004948 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004949 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004950 }
4951}
4952
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004953static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004954{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004955 /* Skip empty nodes */
4956 if (!pgdat->node_spanned_pages)
4957 return;
4958
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004959#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07004960 /* ia64 gets its own node_mem_map, before this, without bootmem */
4961 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07004962 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004963 struct page *map;
4964
Bob Piccoe984bb42006-05-20 15:00:31 -07004965 /*
4966 * The zone's endpoints aren't required to be MAX_ORDER
4967 * aligned but the node_mem_map endpoints must be in order
4968 * for the buddy allocator to function correctly.
4969 */
4970 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
Cody P Schafer108bcc92013-02-22 16:35:23 -08004971 end = pgdat_end_pfn(pgdat);
Bob Piccoe984bb42006-05-20 15:00:31 -07004972 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4973 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07004974 map = alloc_remap(pgdat->node_id, size);
4975 if (!map)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004976 map = memblock_virt_alloc_node_nopanic(size,
4977 pgdat->node_id);
Bob Piccoe984bb42006-05-20 15:00:31 -07004978 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004979 }
Roman Zippel12d810c2007-05-31 00:40:54 -07004980#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07004981 /*
4982 * With no DISCONTIG, the global mem_map is just set as node 0's
4983 */
Mel Gormanc7132162006-09-27 01:49:43 -07004984 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004985 mem_map = NODE_DATA(0)->node_mem_map;
Tejun Heo0ee332c2011-12-08 10:22:09 -08004986#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07004987 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08004988 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Tejun Heo0ee332c2011-12-08 10:22:09 -08004989#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07004990 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004991#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004992#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004993}
4994
Johannes Weiner9109fb72008-07-23 21:27:20 -07004995void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4996 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004997{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004998 pg_data_t *pgdat = NODE_DATA(nid);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004999 unsigned long start_pfn = 0;
5000 unsigned long end_pfn = 0;
Johannes Weiner9109fb72008-07-23 21:27:20 -07005001
Minchan Kim88fdf752012-07-31 16:46:14 -07005002 /* pg_data_t should be reset to zero when it's allocated */
Linus Torvalds8783b6e2012-08-02 10:37:03 -07005003 WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
Minchan Kim88fdf752012-07-31 16:46:14 -07005004
Linus Torvalds1da177e2005-04-16 15:20:36 -07005005 pgdat->node_id = nid;
5006 pgdat->node_start_pfn = node_start_pfn;
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005007#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5008 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
Juergen Gross8d29e182015-02-11 15:26:01 -08005009 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
5010 (u64)start_pfn << PAGE_SHIFT, ((u64)end_pfn << PAGE_SHIFT) - 1);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005011#endif
5012 calculate_node_totalpages(pgdat, start_pfn, end_pfn,
5013 zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005014
5015 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07005016#ifdef CONFIG_FLAT_NODE_MEM_MAP
5017 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
5018 nid, (unsigned long)pgdat,
5019 (unsigned long)pgdat->node_mem_map);
5020#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005021
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005022 free_area_init_core(pgdat, start_pfn, end_pfn,
5023 zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005024}
5025
Tejun Heo0ee332c2011-12-08 10:22:09 -08005026#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07005027
5028#if MAX_NUMNODES > 1
5029/*
5030 * Figure out the number of possible node ids.
5031 */
Cody P Schaferf9872ca2013-04-29 15:08:01 -07005032void __init setup_nr_node_ids(void)
Miklos Szeredi418508c2007-05-23 13:57:55 -07005033{
5034 unsigned int node;
5035 unsigned int highest = 0;
5036
5037 for_each_node_mask(node, node_possible_map)
5038 highest = node;
5039 nr_node_ids = highest + 1;
5040}
Miklos Szeredi418508c2007-05-23 13:57:55 -07005041#endif
5042
Mel Gormanc7132162006-09-27 01:49:43 -07005043/**
Tejun Heo1e019792011-07-12 09:45:34 +02005044 * node_map_pfn_alignment - determine the maximum internode alignment
5045 *
5046 * This function should be called after node map is populated and sorted.
5047 * It calculates the maximum power of two alignment which can distinguish
5048 * all the nodes.
5049 *
5050 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5051 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
5052 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
5053 * shifted, 1GiB is enough and this function will indicate so.
5054 *
5055 * This is used to test whether pfn -> nid mapping of the chosen memory
5056 * model has fine enough granularity to avoid incorrect mapping for the
5057 * populated node map.
5058 *
5059 * Returns the determined alignment in pfn's. 0 if there is no alignment
5060 * requirement (single node).
5061 */
5062unsigned long __init node_map_pfn_alignment(void)
5063{
5064 unsigned long accl_mask = 0, last_end = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02005065 unsigned long start, end, mask;
Tejun Heo1e019792011-07-12 09:45:34 +02005066 int last_nid = -1;
Tejun Heoc13291a2011-07-12 10:46:30 +02005067 int i, nid;
Tejun Heo1e019792011-07-12 09:45:34 +02005068
Tejun Heoc13291a2011-07-12 10:46:30 +02005069 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
Tejun Heo1e019792011-07-12 09:45:34 +02005070 if (!start || last_nid < 0 || last_nid == nid) {
5071 last_nid = nid;
5072 last_end = end;
5073 continue;
5074 }
5075
5076 /*
5077 * Start with a mask granular enough to pin-point to the
5078 * start pfn and tick off bits one-by-one until it becomes
5079 * too coarse to separate the current node from the last.
5080 */
5081 mask = ~((1 << __ffs(start)) - 1);
5082 while (mask && last_end <= (start & (mask << 1)))
5083 mask <<= 1;
5084
5085 /* accumulate all internode masks */
5086 accl_mask |= mask;
5087 }
5088
5089 /* convert mask to number of pages */
5090 return ~accl_mask + 1;
5091}
5092
Mel Gormana6af2bc2007-02-10 01:42:57 -08005093/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07005094static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07005095{
Mel Gormana6af2bc2007-02-10 01:42:57 -08005096 unsigned long min_pfn = ULONG_MAX;
Tejun Heoc13291a2011-07-12 10:46:30 +02005097 unsigned long start_pfn;
5098 int i;
Mel Gorman1abbfb42006-11-23 12:01:41 +00005099
Tejun Heoc13291a2011-07-12 10:46:30 +02005100 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
5101 min_pfn = min(min_pfn, start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005102
Mel Gormana6af2bc2007-02-10 01:42:57 -08005103 if (min_pfn == ULONG_MAX) {
5104 printk(KERN_WARNING
Paul Jackson2bc0d2612008-06-22 07:22:17 -07005105 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08005106 return 0;
5107 }
5108
5109 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005110}
5111
5112/**
5113 * find_min_pfn_with_active_regions - Find the minimum PFN registered
5114 *
5115 * It returns the minimum PFN based on information provided via
Zhang Zhen7d018172014-06-04 16:10:53 -07005116 * memblock_set_node().
Mel Gormanc7132162006-09-27 01:49:43 -07005117 */
5118unsigned long __init find_min_pfn_with_active_regions(void)
5119{
5120 return find_min_pfn_for_node(MAX_NUMNODES);
5121}
5122
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005123/*
5124 * early_calculate_totalpages()
5125 * Sum pages in active regions for movable zone.
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005126 * Populate N_MEMORY for calculating usable_nodes.
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005127 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07005128static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07005129{
Mel Gorman7e63efe2007-07-17 04:03:15 -07005130 unsigned long totalpages = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02005131 unsigned long start_pfn, end_pfn;
5132 int i, nid;
Mel Gorman7e63efe2007-07-17 04:03:15 -07005133
Tejun Heoc13291a2011-07-12 10:46:30 +02005134 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
5135 unsigned long pages = end_pfn - start_pfn;
5136
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005137 totalpages += pages;
5138 if (pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005139 node_set_state(nid, N_MEMORY);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005140 }
Pintu Kumarb8af2942013-09-11 14:20:34 -07005141 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07005142}
5143
Mel Gorman2a1e2742007-07-17 04:03:12 -07005144/*
5145 * Find the PFN the Movable zone begins in each node. Kernel memory
5146 * is spread evenly between nodes as long as the nodes have enough
5147 * memory. When they don't, some nodes will have more kernelcore than
5148 * others
5149 */
Kautuk Consulb224ef82012-03-21 16:34:15 -07005150static void __init find_zone_movable_pfns_for_nodes(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005151{
5152 int i, nid;
5153 unsigned long usable_startpfn;
5154 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07005155 /* save the state before borrow the nodemask */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005156 nodemask_t saved_node_state = node_states[N_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005157 unsigned long totalpages = early_calculate_totalpages();
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005158 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
Emil Medve136199f2014-04-07 15:37:52 -07005159 struct memblock_region *r;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005160
5161 /* Need to find movable_zone earlier when movable_node is specified. */
5162 find_usable_zone_for_movable();
Mel Gorman2a1e2742007-07-17 04:03:12 -07005163
Mel Gorman7e63efe2007-07-17 04:03:15 -07005164 /*
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005165 * If movable_node is specified, ignore kernelcore and movablecore
5166 * options.
5167 */
5168 if (movable_node_is_enabled()) {
Emil Medve136199f2014-04-07 15:37:52 -07005169 for_each_memblock(memory, r) {
5170 if (!memblock_is_hotpluggable(r))
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005171 continue;
5172
Emil Medve136199f2014-04-07 15:37:52 -07005173 nid = r->nid;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005174
Emil Medve136199f2014-04-07 15:37:52 -07005175 usable_startpfn = PFN_DOWN(r->base);
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005176 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
5177 min(usable_startpfn, zone_movable_pfn[nid]) :
5178 usable_startpfn;
5179 }
5180
5181 goto out2;
5182 }
5183
5184 /*
5185 * If movablecore=nn[KMG] was specified, calculate what size of
Mel Gorman7e63efe2007-07-17 04:03:15 -07005186 * kernelcore that corresponds so that memory usable for
5187 * any allocation type is evenly spread. If both kernelcore
5188 * and movablecore are specified, then the value of kernelcore
5189 * will be used for required_kernelcore if it's greater than
5190 * what movablecore would have allowed.
5191 */
5192 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07005193 unsigned long corepages;
5194
5195 /*
5196 * Round-up so that ZONE_MOVABLE is at least as large as what
5197 * was requested by the user
5198 */
5199 required_movablecore =
5200 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
5201 corepages = totalpages - required_movablecore;
5202
5203 required_kernelcore = max(required_kernelcore, corepages);
5204 }
5205
Yinghai Lu20e69262013-03-01 14:51:27 -08005206 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
5207 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07005208 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005209
5210 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
Mel Gorman2a1e2742007-07-17 04:03:12 -07005211 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
5212
5213restart:
5214 /* Spread kernelcore memory as evenly as possible throughout nodes */
5215 kernelcore_node = required_kernelcore / usable_nodes;
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005216 for_each_node_state(nid, N_MEMORY) {
Tejun Heoc13291a2011-07-12 10:46:30 +02005217 unsigned long start_pfn, end_pfn;
5218
Mel Gorman2a1e2742007-07-17 04:03:12 -07005219 /*
5220 * Recalculate kernelcore_node if the division per node
5221 * now exceeds what is necessary to satisfy the requested
5222 * amount of memory for the kernel
5223 */
5224 if (required_kernelcore < kernelcore_node)
5225 kernelcore_node = required_kernelcore / usable_nodes;
5226
5227 /*
5228 * As the map is walked, we track how much memory is usable
5229 * by the kernel using kernelcore_remaining. When it is
5230 * 0, the rest of the node is usable by ZONE_MOVABLE
5231 */
5232 kernelcore_remaining = kernelcore_node;
5233
5234 /* Go through each range of PFNs within this node */
Tejun Heoc13291a2011-07-12 10:46:30 +02005235 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005236 unsigned long size_pages;
5237
Tejun Heoc13291a2011-07-12 10:46:30 +02005238 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005239 if (start_pfn >= end_pfn)
5240 continue;
5241
5242 /* Account for what is only usable for kernelcore */
5243 if (start_pfn < usable_startpfn) {
5244 unsigned long kernel_pages;
5245 kernel_pages = min(end_pfn, usable_startpfn)
5246 - start_pfn;
5247
5248 kernelcore_remaining -= min(kernel_pages,
5249 kernelcore_remaining);
5250 required_kernelcore -= min(kernel_pages,
5251 required_kernelcore);
5252
5253 /* Continue if range is now fully accounted */
5254 if (end_pfn <= usable_startpfn) {
5255
5256 /*
5257 * Push zone_movable_pfn to the end so
5258 * that if we have to rebalance
5259 * kernelcore across nodes, we will
5260 * not double account here
5261 */
5262 zone_movable_pfn[nid] = end_pfn;
5263 continue;
5264 }
5265 start_pfn = usable_startpfn;
5266 }
5267
5268 /*
5269 * The usable PFN range for ZONE_MOVABLE is from
5270 * start_pfn->end_pfn. Calculate size_pages as the
5271 * number of pages used as kernelcore
5272 */
5273 size_pages = end_pfn - start_pfn;
5274 if (size_pages > kernelcore_remaining)
5275 size_pages = kernelcore_remaining;
5276 zone_movable_pfn[nid] = start_pfn + size_pages;
5277
5278 /*
5279 * Some kernelcore has been met, update counts and
5280 * break if the kernelcore for this node has been
Pintu Kumarb8af2942013-09-11 14:20:34 -07005281 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07005282 */
5283 required_kernelcore -= min(required_kernelcore,
5284 size_pages);
5285 kernelcore_remaining -= size_pages;
5286 if (!kernelcore_remaining)
5287 break;
5288 }
5289 }
5290
5291 /*
5292 * If there is still required_kernelcore, we do another pass with one
5293 * less node in the count. This will push zone_movable_pfn[nid] further
5294 * along on the nodes that still have memory until kernelcore is
Pintu Kumarb8af2942013-09-11 14:20:34 -07005295 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07005296 */
5297 usable_nodes--;
5298 if (usable_nodes && required_kernelcore > usable_nodes)
5299 goto restart;
5300
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005301out2:
Mel Gorman2a1e2742007-07-17 04:03:12 -07005302 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5303 for (nid = 0; nid < MAX_NUMNODES; nid++)
5304 zone_movable_pfn[nid] =
5305 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07005306
Yinghai Lu20e69262013-03-01 14:51:27 -08005307out:
Yinghai Lu66918dc2009-06-30 11:41:37 -07005308 /* restore the node_state */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005309 node_states[N_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005310}
5311
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005312/* Any regular or high memory on that node ? */
5313static void check_for_memory(pg_data_t *pgdat, int nid)
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005314{
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005315 enum zone_type zone_type;
5316
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005317 if (N_MEMORY == N_NORMAL_MEMORY)
5318 return;
5319
5320 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005321 struct zone *zone = &pgdat->node_zones[zone_type];
Xishi Qiub38a8722013-11-12 15:07:20 -08005322 if (populated_zone(zone)) {
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005323 node_set_state(nid, N_HIGH_MEMORY);
5324 if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
5325 zone_type <= ZONE_NORMAL)
5326 node_set_state(nid, N_NORMAL_MEMORY);
Bob Liud0048b02012-01-12 17:19:07 -08005327 break;
5328 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005329 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005330}
5331
Mel Gormanc7132162006-09-27 01:49:43 -07005332/**
5333 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005334 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07005335 *
5336 * This will call free_area_init_node() for each active node in the system.
Zhang Zhen7d018172014-06-04 16:10:53 -07005337 * Using the page ranges provided by memblock_set_node(), the size of each
Mel Gormanc7132162006-09-27 01:49:43 -07005338 * zone in each node and their holes is calculated. If the maximum PFN
5339 * between two adjacent zones match, it is assumed that the zone is empty.
5340 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5341 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5342 * starts where the previous one ended. For example, ZONE_DMA32 starts
5343 * at arch_max_dma_pfn.
5344 */
5345void __init free_area_init_nodes(unsigned long *max_zone_pfn)
5346{
Tejun Heoc13291a2011-07-12 10:46:30 +02005347 unsigned long start_pfn, end_pfn;
5348 int i, nid;
Mel Gormana6af2bc2007-02-10 01:42:57 -08005349
Mel Gormanc7132162006-09-27 01:49:43 -07005350 /* Record where the zone boundaries are */
5351 memset(arch_zone_lowest_possible_pfn, 0,
5352 sizeof(arch_zone_lowest_possible_pfn));
5353 memset(arch_zone_highest_possible_pfn, 0,
5354 sizeof(arch_zone_highest_possible_pfn));
5355 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
5356 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
5357 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005358 if (i == ZONE_MOVABLE)
5359 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07005360 arch_zone_lowest_possible_pfn[i] =
5361 arch_zone_highest_possible_pfn[i-1];
5362 arch_zone_highest_possible_pfn[i] =
5363 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
5364 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07005365 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
5366 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
5367
5368 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5369 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
Kautuk Consulb224ef82012-03-21 16:34:15 -07005370 find_zone_movable_pfns_for_nodes();
Mel Gormanc7132162006-09-27 01:49:43 -07005371
Mel Gormanc7132162006-09-27 01:49:43 -07005372 /* Print out the zone ranges */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005373 pr_info("Zone ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005374 for (i = 0; i < MAX_NR_ZONES; i++) {
5375 if (i == ZONE_MOVABLE)
5376 continue;
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005377 pr_info(" %-8s ", zone_names[i]);
David Rientjes72f0ba02010-03-05 13:42:14 -08005378 if (arch_zone_lowest_possible_pfn[i] ==
5379 arch_zone_highest_possible_pfn[i])
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005380 pr_cont("empty\n");
David Rientjes72f0ba02010-03-05 13:42:14 -08005381 else
Juergen Gross8d29e182015-02-11 15:26:01 -08005382 pr_cont("[mem %#018Lx-%#018Lx]\n",
5383 (u64)arch_zone_lowest_possible_pfn[i]
5384 << PAGE_SHIFT,
5385 ((u64)arch_zone_highest_possible_pfn[i]
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005386 << PAGE_SHIFT) - 1);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005387 }
5388
5389 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005390 pr_info("Movable zone start for each node\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005391 for (i = 0; i < MAX_NUMNODES; i++) {
5392 if (zone_movable_pfn[i])
Juergen Gross8d29e182015-02-11 15:26:01 -08005393 pr_info(" Node %d: %#018Lx\n", i,
5394 (u64)zone_movable_pfn[i] << PAGE_SHIFT);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005395 }
Mel Gormanc7132162006-09-27 01:49:43 -07005396
Wanpeng Lif2d52fe2012-10-08 16:32:24 -07005397 /* Print out the early node map */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005398 pr_info("Early memory node ranges\n");
Tejun Heoc13291a2011-07-12 10:46:30 +02005399 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
Juergen Gross8d29e182015-02-11 15:26:01 -08005400 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid,
5401 (u64)start_pfn << PAGE_SHIFT,
5402 ((u64)end_pfn << PAGE_SHIFT) - 1);
Mel Gormanc7132162006-09-27 01:49:43 -07005403
5404 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07005405 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08005406 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07005407 for_each_online_node(nid) {
5408 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07005409 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07005410 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005411
5412 /* Any memory on that node */
5413 if (pgdat->node_present_pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005414 node_set_state(nid, N_MEMORY);
5415 check_for_memory(pgdat, nid);
Mel Gormanc7132162006-09-27 01:49:43 -07005416 }
5417}
Mel Gorman2a1e2742007-07-17 04:03:12 -07005418
Mel Gorman7e63efe2007-07-17 04:03:15 -07005419static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005420{
5421 unsigned long long coremem;
5422 if (!p)
5423 return -EINVAL;
5424
5425 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005426 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005427
Mel Gorman7e63efe2007-07-17 04:03:15 -07005428 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07005429 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
5430
5431 return 0;
5432}
Mel Gormaned7ed362007-07-17 04:03:14 -07005433
Mel Gorman7e63efe2007-07-17 04:03:15 -07005434/*
5435 * kernelcore=size sets the amount of memory for use for allocations that
5436 * cannot be reclaimed or migrated.
5437 */
5438static int __init cmdline_parse_kernelcore(char *p)
5439{
5440 return cmdline_parse_core(p, &required_kernelcore);
5441}
5442
5443/*
5444 * movablecore=size sets the amount of memory for use for allocations that
5445 * can be reclaimed or migrated.
5446 */
5447static int __init cmdline_parse_movablecore(char *p)
5448{
5449 return cmdline_parse_core(p, &required_movablecore);
5450}
5451
Mel Gormaned7ed362007-07-17 04:03:14 -07005452early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005453early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07005454
Tejun Heo0ee332c2011-12-08 10:22:09 -08005455#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005456
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005457void adjust_managed_page_count(struct page *page, long count)
5458{
5459 spin_lock(&managed_page_count_lock);
5460 page_zone(page)->managed_pages += count;
5461 totalram_pages += count;
Jiang Liu3dcc0572013-07-03 15:03:21 -07005462#ifdef CONFIG_HIGHMEM
5463 if (PageHighMem(page))
5464 totalhigh_pages += count;
5465#endif
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005466 spin_unlock(&managed_page_count_lock);
5467}
Jiang Liu3dcc0572013-07-03 15:03:21 -07005468EXPORT_SYMBOL(adjust_managed_page_count);
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005469
Jiang Liu11199692013-07-03 15:02:48 -07005470unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
Jiang Liu69afade2013-04-29 15:06:21 -07005471{
Jiang Liu11199692013-07-03 15:02:48 -07005472 void *pos;
5473 unsigned long pages = 0;
Jiang Liu69afade2013-04-29 15:06:21 -07005474
Jiang Liu11199692013-07-03 15:02:48 -07005475 start = (void *)PAGE_ALIGN((unsigned long)start);
5476 end = (void *)((unsigned long)end & PAGE_MASK);
5477 for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
Jiang Liudbe67df2013-07-03 15:02:51 -07005478 if ((unsigned int)poison <= 0xFF)
Jiang Liu11199692013-07-03 15:02:48 -07005479 memset(pos, poison, PAGE_SIZE);
5480 free_reserved_page(virt_to_page(pos));
Jiang Liu69afade2013-04-29 15:06:21 -07005481 }
5482
5483 if (pages && s)
Jiang Liu11199692013-07-03 15:02:48 -07005484 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
Jiang Liu69afade2013-04-29 15:06:21 -07005485 s, pages << (PAGE_SHIFT - 10), start, end);
5486
5487 return pages;
5488}
Jiang Liu11199692013-07-03 15:02:48 -07005489EXPORT_SYMBOL(free_reserved_area);
Jiang Liu69afade2013-04-29 15:06:21 -07005490
Jiang Liucfa11e02013-04-29 15:07:00 -07005491#ifdef CONFIG_HIGHMEM
5492void free_highmem_page(struct page *page)
5493{
5494 __free_reserved_page(page);
5495 totalram_pages++;
Jiang Liu7b4b2a02013-07-03 15:03:11 -07005496 page_zone(page)->managed_pages++;
Jiang Liucfa11e02013-04-29 15:07:00 -07005497 totalhigh_pages++;
5498}
5499#endif
5500
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005501
5502void __init mem_init_print_info(const char *str)
5503{
5504 unsigned long physpages, codesize, datasize, rosize, bss_size;
5505 unsigned long init_code_size, init_data_size;
5506
5507 physpages = get_num_physpages();
5508 codesize = _etext - _stext;
5509 datasize = _edata - _sdata;
5510 rosize = __end_rodata - __start_rodata;
5511 bss_size = __bss_stop - __bss_start;
5512 init_data_size = __init_end - __init_begin;
5513 init_code_size = _einittext - _sinittext;
5514
5515 /*
5516 * Detect special cases and adjust section sizes accordingly:
5517 * 1) .init.* may be embedded into .data sections
5518 * 2) .init.text.* may be out of [__init_begin, __init_end],
5519 * please refer to arch/tile/kernel/vmlinux.lds.S.
5520 * 3) .rodata.* may be embedded into .text or .data sections.
5521 */
5522#define adj_init_size(start, end, size, pos, adj) \
Pintu Kumarb8af2942013-09-11 14:20:34 -07005523 do { \
5524 if (start <= pos && pos < end && size > adj) \
5525 size -= adj; \
5526 } while (0)
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005527
5528 adj_init_size(__init_begin, __init_end, init_data_size,
5529 _sinittext, init_code_size);
5530 adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
5531 adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
5532 adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
5533 adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
5534
5535#undef adj_init_size
5536
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005537 pr_info("Memory: %luK/%luK available "
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005538 "(%luK kernel code, %luK rwdata, %luK rodata, "
Pintu Kumare48322a2014-12-18 16:17:15 -08005539 "%luK init, %luK bss, %luK reserved, %luK cma-reserved"
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005540#ifdef CONFIG_HIGHMEM
5541 ", %luK highmem"
5542#endif
5543 "%s%s)\n",
5544 nr_free_pages() << (PAGE_SHIFT-10), physpages << (PAGE_SHIFT-10),
5545 codesize >> 10, datasize >> 10, rosize >> 10,
5546 (init_data_size + init_code_size) >> 10, bss_size >> 10,
Pintu Kumare48322a2014-12-18 16:17:15 -08005547 (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT-10),
5548 totalcma_pages << (PAGE_SHIFT-10),
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005549#ifdef CONFIG_HIGHMEM
5550 totalhigh_pages << (PAGE_SHIFT-10),
5551#endif
5552 str ? ", " : "", str ? str : "");
5553}
5554
Mel Gorman0e0b8642006-09-27 01:49:56 -07005555/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005556 * set_dma_reserve - set the specified number of pages reserved in the first zone
5557 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07005558 *
5559 * The per-cpu batchsize and zone watermarks are determined by present_pages.
5560 * In the DMA zone, a significant percentage may be consumed by kernel image
5561 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005562 * function may optionally be used to account for unfreeable pages in the
5563 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5564 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07005565 */
5566void __init set_dma_reserve(unsigned long new_dma_reserve)
5567{
5568 dma_reserve = new_dma_reserve;
5569}
5570
Linus Torvalds1da177e2005-04-16 15:20:36 -07005571void __init free_area_init(unsigned long *zones_size)
5572{
Johannes Weiner9109fb72008-07-23 21:27:20 -07005573 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005574 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
5575}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005576
Linus Torvalds1da177e2005-04-16 15:20:36 -07005577static int page_alloc_cpu_notify(struct notifier_block *self,
5578 unsigned long action, void *hcpu)
5579{
5580 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005581
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07005582 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -07005583 lru_add_drain_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005584 drain_pages(cpu);
5585
5586 /*
5587 * Spill the event counters of the dead processor
5588 * into the current processors event counters.
5589 * This artificially elevates the count of the current
5590 * processor.
5591 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07005592 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005593
5594 /*
5595 * Zero the differential counters of the dead processor
5596 * so that the vm statistics are consistent.
5597 *
5598 * This is only okay since the processor is dead and cannot
5599 * race with what we are doing.
5600 */
Christoph Lameter2bb921e2013-09-11 14:21:30 -07005601 cpu_vm_stats_fold(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005602 }
5603 return NOTIFY_OK;
5604}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005605
5606void __init page_alloc_init(void)
5607{
5608 hotcpu_notifier(page_alloc_cpu_notify, 0);
5609}
5610
5611/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005612 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5613 * or min_free_kbytes changes.
5614 */
5615static void calculate_totalreserve_pages(void)
5616{
5617 struct pglist_data *pgdat;
5618 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005619 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005620
5621 for_each_online_pgdat(pgdat) {
5622 for (i = 0; i < MAX_NR_ZONES; i++) {
5623 struct zone *zone = pgdat->node_zones + i;
Mel Gorman3484b2d2014-08-06 16:07:14 -07005624 long max = 0;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005625
5626 /* Find valid and maximum lowmem_reserve in the zone */
5627 for (j = i; j < MAX_NR_ZONES; j++) {
5628 if (zone->lowmem_reserve[j] > max)
5629 max = zone->lowmem_reserve[j];
5630 }
5631
Mel Gorman41858962009-06-16 15:32:12 -07005632 /* we treat the high watermark as reserved pages. */
5633 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005634
Jiang Liub40da042013-02-22 16:33:52 -08005635 if (max > zone->managed_pages)
5636 max = zone->managed_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005637 reserve_pages += max;
Johannes Weinerab8fabd2012-01-10 15:07:42 -08005638 /*
5639 * Lowmem reserves are not available to
5640 * GFP_HIGHUSER page cache allocations and
5641 * kswapd tries to balance zones to their high
5642 * watermark. As a result, neither should be
5643 * regarded as dirtyable memory, to prevent a
5644 * situation where reclaim has to clean pages
5645 * in order to balance the zones.
5646 */
5647 zone->dirty_balance_reserve = max;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005648 }
5649 }
Johannes Weinerab8fabd2012-01-10 15:07:42 -08005650 dirty_balance_reserve = reserve_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005651 totalreserve_pages = reserve_pages;
5652}
5653
5654/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07005655 * setup_per_zone_lowmem_reserve - called whenever
5656 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
5657 * has a correct pages reserved value, so an adequate number of
5658 * pages are left in the zone after a successful __alloc_pages().
5659 */
5660static void setup_per_zone_lowmem_reserve(void)
5661{
5662 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005663 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005664
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08005665 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005666 for (j = 0; j < MAX_NR_ZONES; j++) {
5667 struct zone *zone = pgdat->node_zones + j;
Jiang Liub40da042013-02-22 16:33:52 -08005668 unsigned long managed_pages = zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005669
5670 zone->lowmem_reserve[j] = 0;
5671
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005672 idx = j;
5673 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005674 struct zone *lower_zone;
5675
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005676 idx--;
5677
Linus Torvalds1da177e2005-04-16 15:20:36 -07005678 if (sysctl_lowmem_reserve_ratio[idx] < 1)
5679 sysctl_lowmem_reserve_ratio[idx] = 1;
5680
5681 lower_zone = pgdat->node_zones + idx;
Jiang Liub40da042013-02-22 16:33:52 -08005682 lower_zone->lowmem_reserve[j] = managed_pages /
Linus Torvalds1da177e2005-04-16 15:20:36 -07005683 sysctl_lowmem_reserve_ratio[idx];
Jiang Liub40da042013-02-22 16:33:52 -08005684 managed_pages += lower_zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005685 }
5686 }
5687 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005688
5689 /* update totalreserve_pages */
5690 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005691}
5692
Mel Gormancfd3da12011-04-25 21:36:42 +00005693static void __setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005694{
5695 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
5696 unsigned long lowmem_pages = 0;
5697 struct zone *zone;
5698 unsigned long flags;
5699
5700 /* Calculate total number of !ZONE_HIGHMEM pages */
5701 for_each_zone(zone) {
5702 if (!is_highmem(zone))
Jiang Liub40da042013-02-22 16:33:52 -08005703 lowmem_pages += zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005704 }
5705
5706 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07005707 u64 tmp;
5708
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005709 spin_lock_irqsave(&zone->lock, flags);
Jiang Liub40da042013-02-22 16:33:52 -08005710 tmp = (u64)pages_min * zone->managed_pages;
Andrew Mortonac924c62006-05-15 09:43:59 -07005711 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005712 if (is_highmem(zone)) {
5713 /*
Nick Piggin669ed172005-11-13 16:06:45 -08005714 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5715 * need highmem pages, so cap pages_min to a small
5716 * value here.
5717 *
Mel Gorman41858962009-06-16 15:32:12 -07005718 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Nick Piggin669ed172005-11-13 16:06:45 -08005719 * deltas controls asynch page reclaim, and so should
5720 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005721 */
Andrew Morton90ae8d62013-02-22 16:32:22 -08005722 unsigned long min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005723
Jiang Liub40da042013-02-22 16:33:52 -08005724 min_pages = zone->managed_pages / 1024;
Andrew Morton90ae8d62013-02-22 16:32:22 -08005725 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
Mel Gorman41858962009-06-16 15:32:12 -07005726 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005727 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08005728 /*
5729 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07005730 * proportionate to the zone's size.
5731 */
Mel Gorman41858962009-06-16 15:32:12 -07005732 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005733 }
5734
Mel Gorman41858962009-06-16 15:32:12 -07005735 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
5736 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Marek Szyprowski49f223a2012-01-25 12:49:24 +01005737
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07005738 __mod_zone_page_state(zone, NR_ALLOC_BATCH,
Johannes Weinerabe5f972014-10-02 16:21:10 -07005739 high_wmark_pages(zone) - low_wmark_pages(zone) -
5740 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07005741
Mel Gorman56fd56b2007-10-16 01:25:58 -07005742 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005743 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005744 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005745
5746 /* update totalreserve_pages */
5747 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005748}
5749
Mel Gormancfd3da12011-04-25 21:36:42 +00005750/**
5751 * setup_per_zone_wmarks - called when min_free_kbytes changes
5752 * or when memory is hot-{added|removed}
5753 *
5754 * Ensures that the watermark[min,low,high] values for each zone are set
5755 * correctly with respect to min_free_kbytes.
5756 */
5757void setup_per_zone_wmarks(void)
5758{
5759 mutex_lock(&zonelists_mutex);
5760 __setup_per_zone_wmarks();
5761 mutex_unlock(&zonelists_mutex);
5762}
5763
Randy Dunlap55a44622009-09-21 17:01:20 -07005764/*
Rik van Riel556adec2008-10-18 20:26:34 -07005765 * The inactive anon list should be small enough that the VM never has to
5766 * do too much work, but large enough that each inactive page has a chance
5767 * to be referenced again before it is swapped out.
5768 *
5769 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5770 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5771 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5772 * the anonymous pages are kept on the inactive list.
5773 *
5774 * total target max
5775 * memory ratio inactive anon
5776 * -------------------------------------
5777 * 10MB 1 5MB
5778 * 100MB 1 50MB
5779 * 1GB 3 250MB
5780 * 10GB 10 0.9GB
5781 * 100GB 31 3GB
5782 * 1TB 101 10GB
5783 * 10TB 320 32GB
5784 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005785static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
Minchan Kim96cb4df2009-06-16 15:32:49 -07005786{
5787 unsigned int gb, ratio;
5788
5789 /* Zone size in gigabytes */
Jiang Liub40da042013-02-22 16:33:52 -08005790 gb = zone->managed_pages >> (30 - PAGE_SHIFT);
Minchan Kim96cb4df2009-06-16 15:32:49 -07005791 if (gb)
5792 ratio = int_sqrt(10 * gb);
5793 else
5794 ratio = 1;
5795
5796 zone->inactive_ratio = ratio;
5797}
5798
KOSAKI Motohiro839a4fc2011-05-24 17:11:31 -07005799static void __meminit setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07005800{
5801 struct zone *zone;
5802
Minchan Kim96cb4df2009-06-16 15:32:49 -07005803 for_each_zone(zone)
5804 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07005805}
5806
Linus Torvalds1da177e2005-04-16 15:20:36 -07005807/*
5808 * Initialise min_free_kbytes.
5809 *
5810 * For small machines we want it small (128k min). For large machines
5811 * we want it large (64MB max). But it is not linear, because network
5812 * bandwidth does not increase linearly with machine size. We use
5813 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07005814 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
Linus Torvalds1da177e2005-04-16 15:20:36 -07005815 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5816 *
5817 * which yields
5818 *
5819 * 16MB: 512k
5820 * 32MB: 724k
5821 * 64MB: 1024k
5822 * 128MB: 1448k
5823 * 256MB: 2048k
5824 * 512MB: 2896k
5825 * 1024MB: 4096k
5826 * 2048MB: 5792k
5827 * 4096MB: 8192k
5828 * 8192MB: 11584k
5829 * 16384MB: 16384k
5830 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005831int __meminit init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005832{
5833 unsigned long lowmem_kbytes;
Michal Hocko5f127332013-07-08 16:00:40 -07005834 int new_min_free_kbytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005835
5836 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
Michal Hocko5f127332013-07-08 16:00:40 -07005837 new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005838
Michal Hocko5f127332013-07-08 16:00:40 -07005839 if (new_min_free_kbytes > user_min_free_kbytes) {
5840 min_free_kbytes = new_min_free_kbytes;
5841 if (min_free_kbytes < 128)
5842 min_free_kbytes = 128;
5843 if (min_free_kbytes > 65536)
5844 min_free_kbytes = 65536;
5845 } else {
5846 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
5847 new_min_free_kbytes, user_min_free_kbytes);
5848 }
Minchan Kimbc75d332009-06-16 15:32:48 -07005849 setup_per_zone_wmarks();
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -07005850 refresh_zone_stat_thresholds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005851 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07005852 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005853 return 0;
5854}
Minchan Kimbc75d332009-06-16 15:32:48 -07005855module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005856
5857/*
Pintu Kumarb8af2942013-09-11 14:20:34 -07005858 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
Linus Torvalds1da177e2005-04-16 15:20:36 -07005859 * that we can call two helper functions whenever min_free_kbytes
5860 * changes.
5861 */
Joe Perchescccad5b2014-06-06 14:38:09 -07005862int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005863 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005864{
Han Pingtianda8c7572014-01-23 15:53:17 -08005865 int rc;
5866
5867 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5868 if (rc)
5869 return rc;
5870
Michal Hocko5f127332013-07-08 16:00:40 -07005871 if (write) {
5872 user_min_free_kbytes = min_free_kbytes;
Minchan Kimbc75d332009-06-16 15:32:48 -07005873 setup_per_zone_wmarks();
Michal Hocko5f127332013-07-08 16:00:40 -07005874 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005875 return 0;
5876}
5877
Christoph Lameter96146342006-07-03 00:24:13 -07005878#ifdef CONFIG_NUMA
Joe Perchescccad5b2014-06-06 14:38:09 -07005879int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005880 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07005881{
5882 struct zone *zone;
5883 int rc;
5884
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005885 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07005886 if (rc)
5887 return rc;
5888
5889 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08005890 zone->min_unmapped_pages = (zone->managed_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07005891 sysctl_min_unmapped_ratio) / 100;
5892 return 0;
5893}
Christoph Lameter0ff38492006-09-25 23:31:52 -07005894
Joe Perchescccad5b2014-06-06 14:38:09 -07005895int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005896 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07005897{
5898 struct zone *zone;
5899 int rc;
5900
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005901 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07005902 if (rc)
5903 return rc;
5904
5905 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08005906 zone->min_slab_pages = (zone->managed_pages *
Christoph Lameter0ff38492006-09-25 23:31:52 -07005907 sysctl_min_slab_ratio) / 100;
5908 return 0;
5909}
Christoph Lameter96146342006-07-03 00:24:13 -07005910#endif
5911
Linus Torvalds1da177e2005-04-16 15:20:36 -07005912/*
5913 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5914 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5915 * whenever sysctl_lowmem_reserve_ratio changes.
5916 *
5917 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07005918 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07005919 * if in function of the boot time zone sizes.
5920 */
Joe Perchescccad5b2014-06-06 14:38:09 -07005921int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005922 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005923{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005924 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005925 setup_per_zone_lowmem_reserve();
5926 return 0;
5927}
5928
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005929/*
5930 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
Pintu Kumarb8af2942013-09-11 14:20:34 -07005931 * cpu. It is the fraction of total pages in each zone that a hot per cpu
5932 * pagelist can have before it gets flushed back to buddy allocator.
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005933 */
Joe Perchescccad5b2014-06-06 14:38:09 -07005934int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005935 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005936{
5937 struct zone *zone;
David Rientjes7cd2b0a2014-06-23 13:22:04 -07005938 int old_percpu_pagelist_fraction;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005939 int ret;
5940
Cody P Schaferc8e251f2013-07-03 15:01:29 -07005941 mutex_lock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07005942 old_percpu_pagelist_fraction = percpu_pagelist_fraction;
5943
5944 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
5945 if (!write || ret < 0)
5946 goto out;
5947
5948 /* Sanity checking to avoid pcp imbalance */
5949 if (percpu_pagelist_fraction &&
5950 percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
5951 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
5952 ret = -EINVAL;
5953 goto out;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005954 }
David Rientjes7cd2b0a2014-06-23 13:22:04 -07005955
5956 /* No change? */
5957 if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
5958 goto out;
5959
5960 for_each_populated_zone(zone) {
5961 unsigned int cpu;
5962
5963 for_each_possible_cpu(cpu)
5964 pageset_set_high_and_batch(zone,
5965 per_cpu_ptr(zone->pageset, cpu));
5966 }
5967out:
Cody P Schaferc8e251f2013-07-03 15:01:29 -07005968 mutex_unlock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07005969 return ret;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005970}
5971
David S. Millerf034b5d2006-08-24 03:08:07 -07005972int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005973
5974#ifdef CONFIG_NUMA
5975static int __init set_hashdist(char *str)
5976{
5977 if (!str)
5978 return 0;
5979 hashdist = simple_strtoul(str, &str, 0);
5980 return 1;
5981}
5982__setup("hashdist=", set_hashdist);
5983#endif
5984
5985/*
5986 * allocate a large system hash table from bootmem
5987 * - it is assumed that the hash table must contain an exact power-of-2
5988 * quantity of entries
5989 * - limit is the number of hash buckets, not the total allocation size
5990 */
5991void *__init alloc_large_system_hash(const char *tablename,
5992 unsigned long bucketsize,
5993 unsigned long numentries,
5994 int scale,
5995 int flags,
5996 unsigned int *_hash_shift,
5997 unsigned int *_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00005998 unsigned long low_limit,
5999 unsigned long high_limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006000{
Tim Bird31fe62b2012-05-23 13:33:35 +00006001 unsigned long long max = high_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006002 unsigned long log2qty, size;
6003 void *table = NULL;
6004
6005 /* allow the kernel cmdline to have a say */
6006 if (!numentries) {
6007 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08006008 numentries = nr_kernel_pages;
Jerry Zhoua7e83312013-09-11 14:20:26 -07006009
6010 /* It isn't necessary when PAGE_SIZE >= 1MB */
6011 if (PAGE_SHIFT < 20)
6012 numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006013
6014 /* limit to 1 bucket per 2^scale bytes of low memory */
6015 if (scale > PAGE_SHIFT)
6016 numentries >>= (scale - PAGE_SHIFT);
6017 else
6018 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08006019
6020 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07006021 if (unlikely(flags & HASH_SMALL)) {
6022 /* Makes no sense without HASH_EARLY */
6023 WARN_ON(!(flags & HASH_EARLY));
6024 if (!(numentries >> *_hash_shift)) {
6025 numentries = 1UL << *_hash_shift;
6026 BUG_ON(!numentries);
6027 }
6028 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08006029 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006030 }
John Hawkes6e692ed2006-03-25 03:08:02 -08006031 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006032
6033 /* limit allocation size to 1/16 total memory by default */
6034 if (max == 0) {
6035 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
6036 do_div(max, bucketsize);
6037 }
Dimitri Sivanich074b8512012-02-08 12:39:07 -08006038 max = min(max, 0x80000000ULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006039
Tim Bird31fe62b2012-05-23 13:33:35 +00006040 if (numentries < low_limit)
6041 numentries = low_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006042 if (numentries > max)
6043 numentries = max;
6044
David Howellsf0d1b0b2006-12-08 02:37:49 -08006045 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006046
6047 do {
6048 size = bucketsize << log2qty;
6049 if (flags & HASH_EARLY)
Santosh Shilimkar67828322014-01-21 15:50:25 -08006050 table = memblock_virt_alloc_nopanic(size, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006051 else if (hashdist)
6052 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
6053 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07006054 /*
6055 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07006056 * some pages at the end of hash table which
6057 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07006058 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01006059 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07006060 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01006061 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
6062 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006063 }
6064 } while (!table && size > PAGE_SIZE && --log2qty);
6065
6066 if (!table)
6067 panic("Failed to allocate %s hash table\n", tablename);
6068
Robin Holtf241e6602010-10-07 12:59:26 -07006069 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07006070 tablename,
Robin Holtf241e6602010-10-07 12:59:26 -07006071 (1UL << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08006072 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07006073 size);
6074
6075 if (_hash_shift)
6076 *_hash_shift = log2qty;
6077 if (_hash_mask)
6078 *_hash_mask = (1 << log2qty) - 1;
6079
6080 return table;
6081}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08006082
Mel Gorman835c1342007-10-16 01:25:47 -07006083/* Return a pointer to the bitmap storing bits affecting a block of pages */
6084static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
6085 unsigned long pfn)
6086{
6087#ifdef CONFIG_SPARSEMEM
6088 return __pfn_to_section(pfn)->pageblock_flags;
6089#else
6090 return zone->pageblock_flags;
6091#endif /* CONFIG_SPARSEMEM */
6092}
Andrew Morton6220ec72006-10-19 23:29:05 -07006093
Mel Gorman835c1342007-10-16 01:25:47 -07006094static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
6095{
6096#ifdef CONFIG_SPARSEMEM
6097 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07006098 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07006099#else
Laura Abbottc060f942013-01-11 14:31:51 -08006100 pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
Mel Gormand9c23402007-10-16 01:26:01 -07006101 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07006102#endif /* CONFIG_SPARSEMEM */
6103}
6104
6105/**
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006106 * 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 -07006107 * @page: The page within the block of interest
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006108 * @pfn: The target page frame number
6109 * @end_bitidx: The last bit of interest to retrieve
6110 * @mask: mask of bits that the caller is interested in
6111 *
6112 * Return: pageblock_bits flags
Mel Gorman835c1342007-10-16 01:25:47 -07006113 */
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006114unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
Mel Gormane58469b2014-06-04 16:10:16 -07006115 unsigned long end_bitidx,
6116 unsigned long mask)
Mel Gorman835c1342007-10-16 01:25:47 -07006117{
6118 struct zone *zone;
6119 unsigned long *bitmap;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006120 unsigned long bitidx, word_bitidx;
Mel Gormane58469b2014-06-04 16:10:16 -07006121 unsigned long word;
Mel Gorman835c1342007-10-16 01:25:47 -07006122
6123 zone = page_zone(page);
Mel Gorman835c1342007-10-16 01:25:47 -07006124 bitmap = get_pageblock_bitmap(zone, pfn);
6125 bitidx = pfn_to_bitidx(zone, pfn);
Mel Gormane58469b2014-06-04 16:10:16 -07006126 word_bitidx = bitidx / BITS_PER_LONG;
6127 bitidx &= (BITS_PER_LONG-1);
Mel Gorman835c1342007-10-16 01:25:47 -07006128
Mel Gormane58469b2014-06-04 16:10:16 -07006129 word = bitmap[word_bitidx];
6130 bitidx += end_bitidx;
6131 return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
Mel Gorman835c1342007-10-16 01:25:47 -07006132}
6133
6134/**
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006135 * 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 -07006136 * @page: The page within the block of interest
Mel Gorman835c1342007-10-16 01:25:47 -07006137 * @flags: The flags to set
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006138 * @pfn: The target page frame number
6139 * @end_bitidx: The last bit of interest
6140 * @mask: mask of bits that the caller is interested in
Mel Gorman835c1342007-10-16 01:25:47 -07006141 */
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006142void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
6143 unsigned long pfn,
Mel Gormane58469b2014-06-04 16:10:16 -07006144 unsigned long end_bitidx,
6145 unsigned long mask)
Mel Gorman835c1342007-10-16 01:25:47 -07006146{
6147 struct zone *zone;
6148 unsigned long *bitmap;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006149 unsigned long bitidx, word_bitidx;
Mel Gormane58469b2014-06-04 16:10:16 -07006150 unsigned long old_word, word;
6151
6152 BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
Mel Gorman835c1342007-10-16 01:25:47 -07006153
6154 zone = page_zone(page);
Mel Gorman835c1342007-10-16 01:25:47 -07006155 bitmap = get_pageblock_bitmap(zone, pfn);
6156 bitidx = pfn_to_bitidx(zone, pfn);
Mel Gormane58469b2014-06-04 16:10:16 -07006157 word_bitidx = bitidx / BITS_PER_LONG;
6158 bitidx &= (BITS_PER_LONG-1);
6159
Sasha Levin309381fea2014-01-23 15:52:54 -08006160 VM_BUG_ON_PAGE(!zone_spans_pfn(zone, pfn), page);
Mel Gorman835c1342007-10-16 01:25:47 -07006161
Mel Gormane58469b2014-06-04 16:10:16 -07006162 bitidx += end_bitidx;
6163 mask <<= (BITS_PER_LONG - bitidx - 1);
6164 flags <<= (BITS_PER_LONG - bitidx - 1);
6165
6166 word = ACCESS_ONCE(bitmap[word_bitidx]);
6167 for (;;) {
6168 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
6169 if (word == old_word)
6170 break;
6171 word = old_word;
6172 }
Mel Gorman835c1342007-10-16 01:25:47 -07006173}
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07006174
6175/*
Minchan Kim80934512012-07-31 16:43:01 -07006176 * This function checks whether pageblock includes unmovable pages or not.
6177 * If @count is not zero, it is okay to include less @count unmovable pages
6178 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07006179 * PageLRU check without isolation or lru_lock could race so that
Minchan Kim80934512012-07-31 16:43:01 -07006180 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6181 * expect this function should be exact.
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07006182 */
Wen Congyangb023f462012-12-11 16:00:45 -08006183bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
6184 bool skip_hwpoisoned_pages)
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006185{
6186 unsigned long pfn, iter, found;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01006187 int mt;
6188
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006189 /*
6190 * For avoiding noise data, lru_add_drain_all() should be called
Minchan Kim80934512012-07-31 16:43:01 -07006191 * If ZONE_MOVABLE, the zone never contains unmovable pages
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006192 */
6193 if (zone_idx(zone) == ZONE_MOVABLE)
Minchan Kim80934512012-07-31 16:43:01 -07006194 return false;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01006195 mt = get_pageblock_migratetype(page);
6196 if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
Minchan Kim80934512012-07-31 16:43:01 -07006197 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006198
6199 pfn = page_to_pfn(page);
6200 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
6201 unsigned long check = pfn + iter;
6202
Namhyung Kim29723fc2011-02-25 14:44:25 -08006203 if (!pfn_valid_within(check))
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006204 continue;
Namhyung Kim29723fc2011-02-25 14:44:25 -08006205
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006206 page = pfn_to_page(check);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07006207
6208 /*
6209 * Hugepages are not in LRU lists, but they're movable.
6210 * We need not scan over tail pages bacause we don't
6211 * handle each tail page individually in migration.
6212 */
6213 if (PageHuge(page)) {
6214 iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
6215 continue;
6216 }
6217
Minchan Kim97d255c2012-07-31 16:42:59 -07006218 /*
6219 * We can't use page_count without pin a page
6220 * because another CPU can free compound page.
6221 * This check already skips compound tails of THP
6222 * because their page->_count is zero at all time.
6223 */
6224 if (!atomic_read(&page->_count)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006225 if (PageBuddy(page))
6226 iter += (1 << page_order(page)) - 1;
6227 continue;
6228 }
Minchan Kim97d255c2012-07-31 16:42:59 -07006229
Wen Congyangb023f462012-12-11 16:00:45 -08006230 /*
6231 * The HWPoisoned page may be not in buddy system, and
6232 * page_count() is not 0.
6233 */
6234 if (skip_hwpoisoned_pages && PageHWPoison(page))
6235 continue;
6236
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006237 if (!PageLRU(page))
6238 found++;
6239 /*
Johannes Weiner6b4f7792014-12-12 16:56:13 -08006240 * If there are RECLAIMABLE pages, we need to check
6241 * it. But now, memory offline itself doesn't call
6242 * shrink_node_slabs() and it still to be fixed.
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006243 */
6244 /*
6245 * If the page is not RAM, page_count()should be 0.
6246 * we don't need more check. This is an _used_ not-movable page.
6247 *
6248 * The problematic thing here is PG_reserved pages. PG_reserved
6249 * is set to both of a memory hole page and a _used_ kernel
6250 * page at boot.
6251 */
6252 if (found > count)
Minchan Kim80934512012-07-31 16:43:01 -07006253 return true;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006254 }
Minchan Kim80934512012-07-31 16:43:01 -07006255 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006256}
6257
6258bool is_pageblock_removable_nolock(struct page *page)
6259{
Michal Hocko656a0702012-01-20 14:33:58 -08006260 struct zone *zone;
6261 unsigned long pfn;
Michal Hocko687875f2012-01-20 14:33:55 -08006262
6263 /*
6264 * We have to be careful here because we are iterating over memory
6265 * sections which are not zone aware so we might end up outside of
6266 * the zone but still within the section.
Michal Hocko656a0702012-01-20 14:33:58 -08006267 * We have to take care about the node as well. If the node is offline
6268 * its NODE_DATA will be NULL - see page_zone.
Michal Hocko687875f2012-01-20 14:33:55 -08006269 */
Michal Hocko656a0702012-01-20 14:33:58 -08006270 if (!node_online(page_to_nid(page)))
6271 return false;
6272
6273 zone = page_zone(page);
6274 pfn = page_to_pfn(page);
Cody P Schafer108bcc92013-02-22 16:35:23 -08006275 if (!zone_spans_pfn(zone, pfn))
Michal Hocko687875f2012-01-20 14:33:55 -08006276 return false;
6277
Wen Congyangb023f462012-12-11 16:00:45 -08006278 return !has_unmovable_pages(zone, page, 0, true);
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07006279}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006280
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006281#ifdef CONFIG_CMA
6282
6283static unsigned long pfn_max_align_down(unsigned long pfn)
6284{
6285 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
6286 pageblock_nr_pages) - 1);
6287}
6288
6289static unsigned long pfn_max_align_up(unsigned long pfn)
6290{
6291 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
6292 pageblock_nr_pages));
6293}
6294
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006295/* [start, end) must belong to a single zone. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006296static int __alloc_contig_migrate_range(struct compact_control *cc,
6297 unsigned long start, unsigned long end)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006298{
6299 /* This function is based on compact_zone() from compaction.c. */
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006300 unsigned long nr_reclaimed;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006301 unsigned long pfn = start;
6302 unsigned int tries = 0;
6303 int ret = 0;
6304
Marek Szyprowskibe49a6e2012-12-12 13:51:19 -08006305 migrate_prep();
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006306
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006307 while (pfn < end || !list_empty(&cc->migratepages)) {
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006308 if (fatal_signal_pending(current)) {
6309 ret = -EINTR;
6310 break;
6311 }
6312
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006313 if (list_empty(&cc->migratepages)) {
6314 cc->nr_migratepages = 0;
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07006315 pfn = isolate_migratepages_range(cc, pfn, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006316 if (!pfn) {
6317 ret = -EINTR;
6318 break;
6319 }
6320 tries = 0;
6321 } else if (++tries == 5) {
6322 ret = ret < 0 ? ret : -EBUSY;
6323 break;
6324 }
6325
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006326 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
6327 &cc->migratepages);
6328 cc->nr_migratepages -= nr_reclaimed;
Minchan Kim02c6de82012-10-08 16:31:55 -07006329
Hugh Dickins9c620e22013-02-22 16:35:14 -08006330 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
David Rientjese0b9dae2014-06-04 16:08:28 -07006331 NULL, 0, cc->mode, MR_CMA);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006332 }
Srinivas Pandruvada2a6f5122013-02-22 16:32:09 -08006333 if (ret < 0) {
6334 putback_movable_pages(&cc->migratepages);
6335 return ret;
6336 }
6337 return 0;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006338}
6339
6340/**
6341 * alloc_contig_range() -- tries to allocate given range of pages
6342 * @start: start PFN to allocate
6343 * @end: one-past-the-last PFN to allocate
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006344 * @migratetype: migratetype of the underlaying pageblocks (either
6345 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6346 * in range must have the same migratetype and it must
6347 * be either of the two.
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006348 *
6349 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6350 * aligned, however it's the caller's responsibility to guarantee that
6351 * we are the only thread that changes migrate type of pageblocks the
6352 * pages fall in.
6353 *
6354 * The PFN range must belong to a single zone.
6355 *
6356 * Returns zero on success or negative error code. On success all
6357 * pages which PFN is in [start, end) are allocated for the caller and
6358 * need to be freed with free_contig_range().
6359 */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006360int alloc_contig_range(unsigned long start, unsigned long end,
6361 unsigned migratetype)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006362{
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006363 unsigned long outer_start, outer_end;
6364 int ret = 0, order;
6365
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006366 struct compact_control cc = {
6367 .nr_migratepages = 0,
6368 .order = -1,
6369 .zone = page_zone(pfn_to_page(start)),
David Rientjese0b9dae2014-06-04 16:08:28 -07006370 .mode = MIGRATE_SYNC,
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006371 .ignore_skip_hint = true,
6372 };
6373 INIT_LIST_HEAD(&cc.migratepages);
6374
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006375 /*
6376 * What we do here is we mark all pageblocks in range as
6377 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6378 * have different sizes, and due to the way page allocator
6379 * work, we align the range to biggest of the two pages so
6380 * that page allocator won't try to merge buddies from
6381 * different pageblocks and change MIGRATE_ISOLATE to some
6382 * other migration type.
6383 *
6384 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6385 * migrate the pages from an unaligned range (ie. pages that
6386 * we are interested in). This will put all the pages in
6387 * range back to page allocator as MIGRATE_ISOLATE.
6388 *
6389 * When this is done, we take the pages in range from page
6390 * allocator removing them from the buddy system. This way
6391 * page allocator will never consider using them.
6392 *
6393 * This lets us mark the pageblocks back as
6394 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6395 * aligned range but not in the unaligned, original range are
6396 * put back to page allocator so that buddy can use them.
6397 */
6398
6399 ret = start_isolate_page_range(pfn_max_align_down(start),
Wen Congyangb023f462012-12-11 16:00:45 -08006400 pfn_max_align_up(end), migratetype,
6401 false);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006402 if (ret)
Bob Liu86a595f2012-10-25 13:37:56 -07006403 return ret;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006404
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006405 ret = __alloc_contig_migrate_range(&cc, start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006406 if (ret)
6407 goto done;
6408
6409 /*
6410 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6411 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
6412 * more, all pages in [start, end) are free in page allocator.
6413 * What we are going to do is to allocate all pages from
6414 * [start, end) (that is remove them from page allocator).
6415 *
6416 * The only problem is that pages at the beginning and at the
6417 * end of interesting range may be not aligned with pages that
6418 * page allocator holds, ie. they can be part of higher order
6419 * pages. Because of this, we reserve the bigger range and
6420 * once this is done free the pages we are not interested in.
6421 *
6422 * We don't have to hold zone->lock here because the pages are
6423 * isolated thus they won't get removed from buddy.
6424 */
6425
6426 lru_add_drain_all();
Vlastimil Babka510f5502014-12-10 15:43:07 -08006427 drain_all_pages(cc.zone);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006428
6429 order = 0;
6430 outer_start = start;
6431 while (!PageBuddy(pfn_to_page(outer_start))) {
6432 if (++order >= MAX_ORDER) {
6433 ret = -EBUSY;
6434 goto done;
6435 }
6436 outer_start &= ~0UL << order;
6437 }
6438
6439 /* Make sure the range is really isolated. */
Wen Congyangb023f462012-12-11 16:00:45 -08006440 if (test_pages_isolated(outer_start, end, false)) {
Michal Nazarewiczdae803e2014-11-13 15:19:27 -08006441 pr_info("%s: [%lx, %lx) PFNs busy\n",
6442 __func__, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006443 ret = -EBUSY;
6444 goto done;
6445 }
6446
Marek Szyprowski49f223a2012-01-25 12:49:24 +01006447 /* Grab isolated pages from freelists. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006448 outer_end = isolate_freepages_range(&cc, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006449 if (!outer_end) {
6450 ret = -EBUSY;
6451 goto done;
6452 }
6453
6454 /* Free head and tail (if any) */
6455 if (start != outer_start)
6456 free_contig_range(outer_start, start - outer_start);
6457 if (end != outer_end)
6458 free_contig_range(end, outer_end - end);
6459
6460done:
6461 undo_isolate_page_range(pfn_max_align_down(start),
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006462 pfn_max_align_up(end), migratetype);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006463 return ret;
6464}
6465
6466void free_contig_range(unsigned long pfn, unsigned nr_pages)
6467{
Marek Szyprowskibcc2b022012-12-20 15:05:18 -08006468 unsigned int count = 0;
6469
6470 for (; nr_pages--; pfn++) {
6471 struct page *page = pfn_to_page(pfn);
6472
6473 count += page_count(page) != 1;
6474 __free_page(page);
6475 }
6476 WARN(count != 0, "%d pages are still in use!\n", count);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006477}
6478#endif
6479
Jiang Liu4ed7e022012-07-31 16:43:35 -07006480#ifdef CONFIG_MEMORY_HOTPLUG
Cody P Schafer0a647f32013-07-03 15:01:33 -07006481/*
6482 * The zone indicated has a new number of managed_pages; batch sizes and percpu
6483 * page high values need to be recalulated.
6484 */
Jiang Liu4ed7e022012-07-31 16:43:35 -07006485void __meminit zone_pcp_update(struct zone *zone)
6486{
Cody P Schafer0a647f32013-07-03 15:01:33 -07006487 unsigned cpu;
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006488 mutex_lock(&pcp_batch_high_lock);
Cody P Schafer0a647f32013-07-03 15:01:33 -07006489 for_each_possible_cpu(cpu)
Cody P Schafer169f6c12013-07-03 15:01:41 -07006490 pageset_set_high_and_batch(zone,
6491 per_cpu_ptr(zone->pageset, cpu));
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006492 mutex_unlock(&pcp_batch_high_lock);
Jiang Liu4ed7e022012-07-31 16:43:35 -07006493}
6494#endif
6495
Jiang Liu340175b2012-07-31 16:43:32 -07006496void zone_pcp_reset(struct zone *zone)
6497{
6498 unsigned long flags;
Minchan Kim5a883812012-10-08 16:33:39 -07006499 int cpu;
6500 struct per_cpu_pageset *pset;
Jiang Liu340175b2012-07-31 16:43:32 -07006501
6502 /* avoid races with drain_pages() */
6503 local_irq_save(flags);
6504 if (zone->pageset != &boot_pageset) {
Minchan Kim5a883812012-10-08 16:33:39 -07006505 for_each_online_cpu(cpu) {
6506 pset = per_cpu_ptr(zone->pageset, cpu);
6507 drain_zonestat(zone, pset);
6508 }
Jiang Liu340175b2012-07-31 16:43:32 -07006509 free_percpu(zone->pageset);
6510 zone->pageset = &boot_pageset;
6511 }
6512 local_irq_restore(flags);
6513}
6514
Wen Congyang6dcd73d2012-12-11 16:01:01 -08006515#ifdef CONFIG_MEMORY_HOTREMOVE
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006516/*
6517 * All pages in the range must be isolated before calling this.
6518 */
6519void
6520__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
6521{
6522 struct page *page;
6523 struct zone *zone;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07006524 unsigned int order, i;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006525 unsigned long pfn;
6526 unsigned long flags;
6527 /* find the first valid pfn */
6528 for (pfn = start_pfn; pfn < end_pfn; pfn++)
6529 if (pfn_valid(pfn))
6530 break;
6531 if (pfn == end_pfn)
6532 return;
6533 zone = page_zone(pfn_to_page(pfn));
6534 spin_lock_irqsave(&zone->lock, flags);
6535 pfn = start_pfn;
6536 while (pfn < end_pfn) {
6537 if (!pfn_valid(pfn)) {
6538 pfn++;
6539 continue;
6540 }
6541 page = pfn_to_page(pfn);
Wen Congyangb023f462012-12-11 16:00:45 -08006542 /*
6543 * The HWPoisoned page may be not in buddy system, and
6544 * page_count() is not 0.
6545 */
6546 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
6547 pfn++;
6548 SetPageReserved(page);
6549 continue;
6550 }
6551
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006552 BUG_ON(page_count(page));
6553 BUG_ON(!PageBuddy(page));
6554 order = page_order(page);
6555#ifdef CONFIG_DEBUG_VM
6556 printk(KERN_INFO "remove from free list %lx %d %lx\n",
6557 pfn, 1 << order, end_pfn);
6558#endif
6559 list_del(&page->lru);
6560 rmv_page_order(page);
6561 zone->free_area[order].nr_free--;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006562 for (i = 0; i < (1 << order); i++)
6563 SetPageReserved((page+i));
6564 pfn += (1 << order);
6565 }
6566 spin_unlock_irqrestore(&zone->lock, flags);
6567}
6568#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01006569
6570#ifdef CONFIG_MEMORY_FAILURE
6571bool is_free_buddy_page(struct page *page)
6572{
6573 struct zone *zone = page_zone(page);
6574 unsigned long pfn = page_to_pfn(page);
6575 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07006576 unsigned int order;
Wu Fengguang8d22ba12009-12-16 12:19:58 +01006577
6578 spin_lock_irqsave(&zone->lock, flags);
6579 for (order = 0; order < MAX_ORDER; order++) {
6580 struct page *page_head = page - (pfn & ((1 << order) - 1));
6581
6582 if (PageBuddy(page_head) && page_order(page_head) >= order)
6583 break;
6584 }
6585 spin_unlock_irqrestore(&zone->lock, flags);
6586
6587 return order < MAX_ORDER;
6588}
6589#endif