blob: 736d8e1b63817fcd8c715f2081a4f44ba43c9be9 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/page_alloc.c
3 *
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
6 *
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
15 */
16
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/stddef.h>
18#include <linux/mm.h>
19#include <linux/swap.h>
20#include <linux/interrupt.h>
21#include <linux/pagemap.h>
KOSAKI Motohiro10ed2732008-03-04 14:28:32 -080022#include <linux/jiffies.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/bootmem.h>
Yinghai Luedbe7d22010-08-25 13:39:16 -070024#include <linux/memblock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/compiler.h>
Randy Dunlap9f158332005-09-13 01:25:16 -070026#include <linux/kernel.h>
Vegard Nossumb1eeab62008-11-25 16:55:53 +010027#include <linux/kmemcheck.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/module.h>
29#include <linux/suspend.h>
30#include <linux/pagevec.h>
31#include <linux/blkdev.h>
32#include <linux/slab.h>
Dave Hansena238ab52011-05-24 17:12:16 -070033#include <linux/ratelimit.h>
David Rientjes5a3135c22007-10-16 23:25:53 -070034#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <linux/notifier.h>
36#include <linux/topology.h>
37#include <linux/sysctl.h>
38#include <linux/cpu.h>
39#include <linux/cpuset.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070040#include <linux/memory_hotplug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include <linux/nodemask.h>
42#include <linux/vmalloc.h>
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -070043#include <linux/vmstat.h>
Christoph Lameter4be38e32006-01-06 00:11:17 -080044#include <linux/mempolicy.h>
Yasunori Goto68113782006-06-23 02:03:11 -070045#include <linux/stop_machine.h>
Mel Gormanc7132162006-09-27 01:49:43 -070046#include <linux/sort.h>
47#include <linux/pfn.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070048#include <linux/backing-dev.h>
Akinobu Mita933e3122006-12-08 02:39:45 -080049#include <linux/fault-inject.h>
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -070050#include <linux/page-isolation.h>
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070051#include <linux/page_cgroup.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070052#include <linux/debugobjects.h>
Catalin Marinasdbb1f812009-06-11 13:23:19 +010053#include <linux/kmemleak.h>
Mel Gorman56de7262010-05-24 14:32:30 -070054#include <linux/compaction.h>
Mel Gorman0d3d0622009-09-21 17:02:44 -070055#include <trace/events/kmem.h>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070056#include <linux/prefetch.h>
Lisa Du6e543d52013-09-11 14:22:36 -070057#include <linux/mm_inline.h>
Michal Nazarewicz041d3a82011-12-29 13:09:50 +010058#include <linux/migrate.h>
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -080059#include <linux/page-debug-flags.h>
David Rientjes949f7ec2013-04-29 15:07:48 -070060#include <linux/hugetlb.h>
Clark Williams8bd75c72013-02-07 09:47:07 -060061#include <linux/sched/rt.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
Jiang Liu7ee3d4e2013-07-03 15:03:41 -070063#include <asm/sections.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070064#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070065#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070066#include "internal.h"
67
Cody P Schaferc8e251f2013-07-03 15:01:29 -070068/* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
69static DEFINE_MUTEX(pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -070070#define MIN_PERCPU_PAGELIST_FRACTION (8)
Cody P Schaferc8e251f2013-07-03 15:01:29 -070071
Lee Schermerhorn72812012010-05-26 14:44:56 -070072#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
73DEFINE_PER_CPU(int, numa_node);
74EXPORT_PER_CPU_SYMBOL(numa_node);
75#endif
76
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070077#ifdef CONFIG_HAVE_MEMORYLESS_NODES
78/*
79 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
80 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
81 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
82 * defined in <linux/topology.h>.
83 */
84DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
85EXPORT_PER_CPU_SYMBOL(_numa_mem_);
Joonsoo Kimad2c8142014-10-09 15:26:13 -070086int _node_numa_mem_[MAX_NUMNODES];
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070087#endif
88
Linus Torvalds1da177e2005-04-16 15:20:36 -070089/*
Christoph Lameter13808912007-10-16 01:25:27 -070090 * Array of node states.
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 */
Christoph Lameter13808912007-10-16 01:25:27 -070092nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
93 [N_POSSIBLE] = NODE_MASK_ALL,
94 [N_ONLINE] = { { [0] = 1UL } },
95#ifndef CONFIG_NUMA
96 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
97#ifdef CONFIG_HIGHMEM
98 [N_HIGH_MEMORY] = { { [0] = 1UL } },
99#endif
Lai Jiangshan20b2f522012-12-12 13:52:00 -0800100#ifdef CONFIG_MOVABLE_NODE
101 [N_MEMORY] = { { [0] = 1UL } },
102#endif
Christoph Lameter13808912007-10-16 01:25:27 -0700103 [N_CPU] = { { [0] = 1UL } },
104#endif /* NUMA */
105};
106EXPORT_SYMBOL(node_states);
107
Jiang Liuc3d5f5f2013-07-03 15:03:14 -0700108/* Protect totalram_pages and zone->managed_pages */
109static DEFINE_SPINLOCK(managed_page_count_lock);
110
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700111unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -0700112unsigned long totalreserve_pages __read_mostly;
Johannes Weinerab8fabd2012-01-10 15:07:42 -0800113/*
114 * When calculating the number of globally allowed dirty pages, there
115 * is a certain number of per-zone reserves that should not be
116 * considered dirtyable memory. This is the sum of those reserves
117 * over all existing zones that contribute dirtyable memory.
118 */
119unsigned long dirty_balance_reserve __read_mostly;
120
Hugh Dickins1b76b022012-05-11 01:00:07 -0700121int percpu_pagelist_fraction;
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000122gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800124#ifdef CONFIG_PM_SLEEP
125/*
126 * The following functions are used by the suspend/hibernate code to temporarily
127 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
128 * while devices are suspended. To avoid races with the suspend/hibernate code,
129 * they should always be called with pm_mutex held (gfp_allowed_mask also should
130 * only be modified with pm_mutex held, unless the suspend/hibernate code is
131 * guaranteed not to run in parallel with that modification).
132 */
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100133
134static gfp_t saved_gfp_mask;
135
136void pm_restore_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800137{
138 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100139 if (saved_gfp_mask) {
140 gfp_allowed_mask = saved_gfp_mask;
141 saved_gfp_mask = 0;
142 }
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800143}
144
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100145void pm_restrict_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800146{
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800147 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100148 WARN_ON(saved_gfp_mask);
149 saved_gfp_mask = gfp_allowed_mask;
150 gfp_allowed_mask &= ~GFP_IOFS;
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800151}
Mel Gormanf90ac392012-01-10 15:07:15 -0800152
153bool pm_suspended_storage(void)
154{
155 if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS)
156 return false;
157 return true;
158}
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800159#endif /* CONFIG_PM_SLEEP */
160
Mel Gormand9c23402007-10-16 01:26:01 -0700161#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
162int pageblock_order __read_mostly;
163#endif
164
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800165static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -0800166
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167/*
168 * results with 256, 32 in the lowmem_reserve sysctl:
169 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
170 * 1G machine -> (16M dma, 784M normal, 224M high)
171 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
172 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
173 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +0100174 *
175 * TBD: should special case ZONE_DMA32 machines here - in those we normally
176 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700178int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800179#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700180 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -0800181#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700182#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700183 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700184#endif
Christoph Lametere53ef382006-09-25 23:31:14 -0700185#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700186 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700187#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700188 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700189};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190
191EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192
Helge Deller15ad7cd2006-12-06 20:40:36 -0800193static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800194#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700195 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800196#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700197#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700198 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700199#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700200 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700201#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700202 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700203#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700204 "Movable",
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700205};
206
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207int min_free_kbytes = 1024;
Han Pingtian42aa83c2014-01-23 15:53:28 -0800208int user_min_free_kbytes = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209
Jan Beulich2c85f512009-09-21 17:03:07 -0700210static unsigned long __meminitdata nr_kernel_pages;
211static unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700212static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213
Tejun Heo0ee332c2011-12-08 10:22:09 -0800214#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
215static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
216static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
217static unsigned long __initdata required_kernelcore;
218static unsigned long __initdata required_movablecore;
219static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Mel Gormanc7132162006-09-27 01:49:43 -0700220
Tejun Heo0ee332c2011-12-08 10:22:09 -0800221/* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
222int movable_zone;
223EXPORT_SYMBOL(movable_zone);
224#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -0700225
Miklos Szeredi418508c2007-05-23 13:57:55 -0700226#if MAX_NUMNODES > 1
227int nr_node_ids __read_mostly = MAX_NUMNODES;
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700228int nr_online_nodes __read_mostly = 1;
Miklos Szeredi418508c2007-05-23 13:57:55 -0700229EXPORT_SYMBOL(nr_node_ids);
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700230EXPORT_SYMBOL(nr_online_nodes);
Miklos Szeredi418508c2007-05-23 13:57:55 -0700231#endif
232
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700233int page_group_by_mobility_disabled __read_mostly;
234
Minchan Kimee6f5092012-07-31 16:43:50 -0700235void set_pageblock_migratetype(struct page *page, int migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700236{
KOSAKI Motohiro5d0f3f72013-11-12 15:08:18 -0800237 if (unlikely(page_group_by_mobility_disabled &&
238 migratetype < MIGRATE_PCPTYPES))
Mel Gorman49255c62009-06-16 15:31:58 -0700239 migratetype = MIGRATE_UNMOVABLE;
240
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700241 set_pageblock_flags_group(page, (unsigned long)migratetype,
242 PB_migrate, PB_migrate_end);
243}
244
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -0700245bool oom_killer_disabled __read_mostly;
246
Nick Piggin13e74442006-01-06 00:10:58 -0800247#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700248static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700250 int ret = 0;
251 unsigned seq;
252 unsigned long pfn = page_to_pfn(page);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800253 unsigned long sp, start_pfn;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700254
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700255 do {
256 seq = zone_span_seqbegin(zone);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800257 start_pfn = zone->zone_start_pfn;
258 sp = zone->spanned_pages;
Cody P Schafer108bcc92013-02-22 16:35:23 -0800259 if (!zone_spans_pfn(zone, pfn))
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700260 ret = 1;
261 } while (zone_span_seqretry(zone, seq));
262
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800263 if (ret)
Dave Hansen613813e2014-06-04 16:07:27 -0700264 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
265 pfn, zone_to_nid(zone), zone->name,
266 start_pfn, start_pfn + sp);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800267
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700268 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700269}
270
271static int page_is_consistent(struct zone *zone, struct page *page)
272{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700273 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700274 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700276 return 0;
277
278 return 1;
279}
280/*
281 * Temporary debugging check for pages not lying within a given zone.
282 */
283static int bad_range(struct zone *zone, struct page *page)
284{
285 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700287 if (!page_is_consistent(zone, page))
288 return 1;
289
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 return 0;
291}
Nick Piggin13e74442006-01-06 00:10:58 -0800292#else
293static inline int bad_range(struct zone *zone, struct page *page)
294{
295 return 0;
296}
297#endif
298
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -0700299static void bad_page(struct page *page, const char *reason,
300 unsigned long bad_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800302 static unsigned long resume;
303 static unsigned long nr_shown;
304 static unsigned long nr_unshown;
305
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200306 /* Don't complain about poisoned pages */
307 if (PageHWPoison(page)) {
Mel Gorman22b751c2013-02-22 16:34:59 -0800308 page_mapcount_reset(page); /* remove PageBuddy */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200309 return;
310 }
311
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800312 /*
313 * Allow a burst of 60 reports, then keep quiet for that minute;
314 * or allow a steady drip of one report per second.
315 */
316 if (nr_shown == 60) {
317 if (time_before(jiffies, resume)) {
318 nr_unshown++;
319 goto out;
320 }
321 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800322 printk(KERN_ALERT
323 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800324 nr_unshown);
325 nr_unshown = 0;
326 }
327 nr_shown = 0;
328 }
329 if (nr_shown++ == 0)
330 resume = jiffies + 60 * HZ;
331
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800332 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800333 current->comm, page_to_pfn(page));
Dave Hansenf0b791a2014-01-23 15:52:49 -0800334 dump_page_badflags(page, reason, bad_flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700335
Dave Jones4f318882011-10-31 17:07:24 -0700336 print_modules();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800338out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800339 /* Leave bad fields for debug, except PageBuddy could make trouble */
Mel Gorman22b751c2013-02-22 16:34:59 -0800340 page_mapcount_reset(page); /* remove PageBuddy */
Rusty Russell373d4d02013-01-21 17:17:39 +1030341 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342}
343
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344/*
345 * Higher-order pages are called "compound pages". They are structured thusly:
346 *
347 * The first PAGE_SIZE page is called the "head page".
348 *
349 * The remaining PAGE_SIZE pages are called "tail pages".
350 *
Wang Sheng-Hui6416b9f2011-11-17 10:53:50 +0100351 * All pages have PG_compound set. All tail pages have their ->first_page
352 * pointing at the head page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800354 * The first tail page's ->lru.next holds the address of the compound page's
355 * put_page() function. Its ->lru.prev holds the order of allocation.
356 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800358
359static void free_compound_page(struct page *page)
360{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700361 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800362}
363
Andi Kleen01ad1c02008-07-23 21:27:46 -0700364void prep_compound_page(struct page *page, unsigned long order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365{
366 int i;
367 int nr_pages = 1 << order;
368
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800369 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700370 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700371 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800372 for (i = 1; i < nr_pages; i++) {
373 struct page *p = page + i;
Youquan Song58a84aa2011-12-08 14:34:18 -0800374 set_page_count(p, 0);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700375 p->first_page = page;
David Rientjes668f9abb2014-03-03 15:38:18 -0800376 /* Make sure p->first_page is always valid for PageTail() */
377 smp_wmb();
378 __SetPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 }
380}
381
Andrea Arcangeli59ff4212011-01-13 15:46:38 -0800382/* update __split_huge_page_refcount if you change this function */
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800383static int destroy_compound_page(struct page *page, unsigned long order)
Andy Whitcroft18229df2008-11-06 12:53:27 -0800384{
385 int i;
386 int nr_pages = 1 << order;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800387 int bad = 0;
Andy Whitcroft18229df2008-11-06 12:53:27 -0800388
Gavin Shan0bb2c762012-12-18 14:21:32 -0800389 if (unlikely(compound_order(page) != order)) {
Dave Hansenf0b791a2014-01-23 15:52:49 -0800390 bad_page(page, "wrong compound order", 0);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800391 bad++;
392 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393
Christoph Lameter6d777952007-05-06 14:49:40 -0700394 __ClearPageHead(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800395
Andy Whitcroft18229df2008-11-06 12:53:27 -0800396 for (i = 1; i < nr_pages; i++) {
397 struct page *p = page + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398
Dave Hansenf0b791a2014-01-23 15:52:49 -0800399 if (unlikely(!PageTail(p))) {
400 bad_page(page, "PageTail not set", 0);
401 bad++;
402 } else if (unlikely(p->first_page != page)) {
403 bad_page(page, "first_page not consistent", 0);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800404 bad++;
405 }
Christoph Lameterd85f3382007-05-06 14:49:39 -0700406 __ClearPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800408
409 return bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700412static inline void prep_zero_page(struct page *page, unsigned int order,
413 gfp_t gfp_flags)
Nick Piggin17cf4402006-03-22 00:08:41 -0800414{
415 int i;
416
Andrew Morton6626c5d2006-03-22 00:08:42 -0800417 /*
418 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
419 * and __GFP_HIGHMEM from hard or soft interrupt context.
420 */
Nick Piggin725d7042006-09-25 23:30:55 -0700421 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800422 for (i = 0; i < (1 << order); i++)
423 clear_highpage(page + i);
424}
425
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800426#ifdef CONFIG_DEBUG_PAGEALLOC
427unsigned int _debug_guardpage_minorder;
428
429static int __init debug_guardpage_minorder_setup(char *buf)
430{
431 unsigned long res;
432
433 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
434 printk(KERN_ERR "Bad debug_guardpage_minorder value\n");
435 return 0;
436 }
437 _debug_guardpage_minorder = res;
438 printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res);
439 return 0;
440}
441__setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
442
443static inline void set_page_guard_flag(struct page *page)
444{
445 __set_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags);
446}
447
448static inline void clear_page_guard_flag(struct page *page)
449{
450 __clear_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags);
451}
452#else
453static inline void set_page_guard_flag(struct page *page) { }
454static inline void clear_page_guard_flag(struct page *page) { }
455#endif
456
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700457static inline void set_page_order(struct page *page, unsigned int order)
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700458{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700459 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000460 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461}
462
463static inline void rmv_page_order(struct page *page)
464{
Nick Piggin676165a2006-04-10 11:21:48 +1000465 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700466 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467}
468
469/*
470 * Locate the struct page for both the matching buddy in our
471 * pair (buddy1) and the combined O(n+1) page they form (page).
472 *
473 * 1) Any buddy B1 will have an order O twin B2 which satisfies
474 * the following equation:
475 * B2 = B1 ^ (1 << O)
476 * For example, if the starting buddy (buddy2) is #8 its order
477 * 1 buddy is #10:
478 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
479 *
480 * 2) Any buddy B will have an order O+1 parent P which
481 * satisfies the following equation:
482 * P = B & ~(1 << O)
483 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200484 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486static inline unsigned long
KyongHo Cho43506fa2011-01-13 15:47:24 -0800487__find_buddy_index(unsigned long page_idx, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488{
KyongHo Cho43506fa2011-01-13 15:47:24 -0800489 return page_idx ^ (1 << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490}
491
492/*
493 * This function checks whether a page is free && is the buddy
494 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800495 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000496 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700497 * (c) a page and its buddy have the same order &&
498 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 *
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700500 * For recording whether a page is in the buddy system, we set ->_mapcount
501 * PAGE_BUDDY_MAPCOUNT_VALUE.
502 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
503 * serialized by zone->lock.
Nick Piggin676165a2006-04-10 11:21:48 +1000504 *
505 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700507static inline int page_is_buddy(struct page *page, struct page *buddy,
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700508 unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700510 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800511 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800512
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800513 if (page_is_guard(buddy) && page_order(buddy) == order) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800514 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
Mel Gormand34c5fa2014-06-04 16:10:10 -0700515
516 if (page_zone_id(page) != page_zone_id(buddy))
517 return 0;
518
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800519 return 1;
520 }
521
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700522 if (PageBuddy(buddy) && page_order(buddy) == order) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800523 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
Mel Gormand34c5fa2014-06-04 16:10:10 -0700524
525 /*
526 * zone check is done late to avoid uselessly
527 * calculating zone/node ids for pages that could
528 * never merge.
529 */
530 if (page_zone_id(page) != page_zone_id(buddy))
531 return 0;
532
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700533 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000534 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700535 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536}
537
538/*
539 * Freeing function for a buddy system allocator.
540 *
541 * The concept of a buddy system is to maintain direct-mapped table
542 * (containing bit values) for memory blocks of various "orders".
543 * The bottom level table contains the map for the smallest allocatable
544 * units of memory (here, pages), and each level above it describes
545 * pairs of units from the levels below, hence, "buddies".
546 * At a high level, all that happens here is marking the table entry
547 * at the bottom level available, and propagating the changes upward
548 * as necessary, plus some accounting needed to play nicely with other
549 * parts of the VM system.
550 * At each level, we keep a list of pages, which are heads of continuous
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700551 * free pages of length of (1 << order) and marked with _mapcount
552 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
553 * field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 * So when we are allocating or freeing one, we can derive the state of the
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100555 * other. That is, if we allocate a small block, and both were
556 * free, the remainder of the region must be split into blocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 * If a block is freed, and its buddy is also free, then this
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100558 * triggers coalescing into a block of larger size.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +0100560 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 */
562
Nick Piggin48db57f2006-01-08 01:00:42 -0800563static inline void __free_one_page(struct page *page,
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700564 unsigned long pfn,
Mel Gormaned0ae212009-06-16 15:32:07 -0700565 struct zone *zone, unsigned int order,
566 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567{
568 unsigned long page_idx;
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700569 unsigned long combined_idx;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800570 unsigned long uninitialized_var(buddy_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700571 struct page *buddy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572
Cody P Schaferd29bb972013-02-22 16:35:25 -0800573 VM_BUG_ON(!zone_is_initialized(zone));
574
Nick Piggin224abf92006-01-06 00:11:11 -0800575 if (unlikely(PageCompound(page)))
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800576 if (unlikely(destroy_compound_page(page, order)))
577 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578
Mel Gormaned0ae212009-06-16 15:32:07 -0700579 VM_BUG_ON(migratetype == -1);
580
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700581 page_idx = pfn & ((1 << MAX_ORDER) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582
Sasha Levin309381fea2014-01-23 15:52:54 -0800583 VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page);
584 VM_BUG_ON_PAGE(bad_range(zone, page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 while (order < MAX_ORDER-1) {
KyongHo Cho43506fa2011-01-13 15:47:24 -0800587 buddy_idx = __find_buddy_index(page_idx, order);
588 buddy = page + (buddy_idx - page_idx);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700589 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700590 break;
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800591 /*
592 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
593 * merge with it and move up one order.
594 */
595 if (page_is_guard(buddy)) {
596 clear_page_guard_flag(buddy);
597 set_page_private(page, 0);
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -0700598 __mod_zone_freepage_state(zone, 1 << order,
599 migratetype);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800600 } else {
601 list_del(&buddy->lru);
602 zone->free_area[order].nr_free--;
603 rmv_page_order(buddy);
604 }
KyongHo Cho43506fa2011-01-13 15:47:24 -0800605 combined_idx = buddy_idx & page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 page = page + (combined_idx - page_idx);
607 page_idx = combined_idx;
608 order++;
609 }
610 set_page_order(page, order);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700611
612 /*
613 * If this is not the largest possible page, check if the buddy
614 * of the next-highest order is free. If it is, it's possible
615 * that pages are being freed that will coalesce soon. In case,
616 * that is happening, add the free page to the tail of the list
617 * so it's less likely to be used soon and more likely to be merged
618 * as a higher order page
619 */
Mel Gormanb7f50cf2010-10-26 14:21:11 -0700620 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700621 struct page *higher_page, *higher_buddy;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800622 combined_idx = buddy_idx & page_idx;
623 higher_page = page + (combined_idx - page_idx);
624 buddy_idx = __find_buddy_index(combined_idx, order + 1);
Li Haifeng0ba8f2d2012-09-17 14:09:21 -0700625 higher_buddy = higher_page + (buddy_idx - combined_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700626 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
627 list_add_tail(&page->lru,
628 &zone->free_area[order].free_list[migratetype]);
629 goto out;
630 }
631 }
632
633 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
634out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 zone->free_area[order].nr_free++;
636}
637
Nick Piggin224abf92006-01-06 00:11:11 -0800638static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639{
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -0700640 const char *bad_reason = NULL;
Dave Hansenf0b791a2014-01-23 15:52:49 -0800641 unsigned long bad_flags = 0;
642
643 if (unlikely(page_mapcount(page)))
644 bad_reason = "nonzero mapcount";
645 if (unlikely(page->mapping != NULL))
646 bad_reason = "non-NULL mapping";
647 if (unlikely(atomic_read(&page->_count) != 0))
648 bad_reason = "nonzero _count";
649 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
650 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
651 bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
652 }
653 if (unlikely(mem_cgroup_bad_page_check(page)))
654 bad_reason = "cgroup check failed";
655 if (unlikely(bad_reason)) {
656 bad_page(page, bad_reason, bad_flags);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800657 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800658 }
Peter Zijlstra90572892013-10-07 11:29:20 +0100659 page_cpupid_reset_last(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800660 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
661 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
662 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663}
664
665/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700666 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700668 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 *
670 * If the zone was previously in an "all pages pinned" state then look to
671 * see if this freeing clears that state.
672 *
673 * And clear the zone's pages_scanned counter, to hold off the "all pages are
674 * pinned" detection logic.
675 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700676static void free_pcppages_bulk(struct zone *zone, int count,
677 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678{
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700679 int migratetype = 0;
Mel Gormana6f9edd2009-09-21 17:03:20 -0700680 int batch_free = 0;
Mel Gorman72853e22010-09-09 16:38:16 -0700681 int to_free = count;
Mel Gorman0d5d8232014-08-06 16:07:16 -0700682 unsigned long nr_scanned;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700683
Nick Pigginc54ad302006-01-06 00:10:56 -0800684 spin_lock(&zone->lock);
Mel Gorman0d5d8232014-08-06 16:07:16 -0700685 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
686 if (nr_scanned)
687 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
Mel Gormanf2260e62009-06-16 15:32:13 -0700688
Mel Gorman72853e22010-09-09 16:38:16 -0700689 while (to_free) {
Nick Piggin48db57f2006-01-08 01:00:42 -0800690 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700691 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -0800692
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700693 /*
Mel Gormana6f9edd2009-09-21 17:03:20 -0700694 * Remove pages from lists in a round-robin fashion. A
695 * batch_free count is maintained that is incremented when an
696 * empty list is encountered. This is so more pages are freed
697 * off fuller lists instead of spinning excessively around empty
698 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700699 */
700 do {
Mel Gormana6f9edd2009-09-21 17:03:20 -0700701 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700702 if (++migratetype == MIGRATE_PCPTYPES)
703 migratetype = 0;
704 list = &pcp->lists[migratetype];
705 } while (list_empty(list));
706
Namhyung Kim1d168712011-03-22 16:32:45 -0700707 /* This is the only non-empty list. Free them all. */
708 if (batch_free == MIGRATE_PCPTYPES)
709 batch_free = to_free;
710
Mel Gormana6f9edd2009-09-21 17:03:20 -0700711 do {
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -0700712 int mt; /* migratetype of the to-be-freed page */
713
Mel Gormana6f9edd2009-09-21 17:03:20 -0700714 page = list_entry(list->prev, struct page, lru);
715 /* must delete as __free_one_page list manipulates */
716 list_del(&page->lru);
Minchan Kimb12c4ad2012-10-08 16:32:08 -0700717 mt = get_freepage_migratetype(page);
Hugh Dickinsa7016232010-01-29 17:46:34 +0000718 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700719 __free_one_page(page, page_to_pfn(page), zone, 0, mt);
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -0700720 trace_mm_page_pcpu_drain(page, 0, mt);
Minchan Kim194159f2013-02-22 16:33:58 -0800721 if (likely(!is_migrate_isolate_page(page))) {
Wen Congyang97d0da22012-12-11 16:00:52 -0800722 __mod_zone_page_state(zone, NR_FREE_PAGES, 1);
723 if (is_migrate_cma(mt))
724 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, 1);
725 }
Mel Gorman72853e22010-09-09 16:38:16 -0700726 } while (--to_free && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800728 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729}
730
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700731static void free_one_page(struct zone *zone,
732 struct page *page, unsigned long pfn,
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700733 unsigned int order,
Mel Gormaned0ae212009-06-16 15:32:07 -0700734 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800735{
Mel Gorman0d5d8232014-08-06 16:07:16 -0700736 unsigned long nr_scanned;
Christoph Lameter006d22d2006-09-25 23:31:48 -0700737 spin_lock(&zone->lock);
Mel Gorman0d5d8232014-08-06 16:07:16 -0700738 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
739 if (nr_scanned)
740 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
Mel Gormanf2260e62009-06-16 15:32:13 -0700741
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700742 __free_one_page(page, pfn, zone, order, migratetype);
Minchan Kim194159f2013-02-22 16:33:58 -0800743 if (unlikely(!is_migrate_isolate(migratetype)))
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -0700744 __mod_zone_freepage_state(zone, 1 << order, migratetype);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700745 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800746}
747
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700748static bool free_pages_prepare(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 int i;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800751 int bad = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -0800753 trace_mm_page_free(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100754 kmemcheck_free_shadow(page, order);
755
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -0800756 if (PageAnon(page))
757 page->mapping = NULL;
758 for (i = 0; i < (1 << order); i++)
759 bad += free_pages_check(page + i);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800760 if (bad)
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700761 return false;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800762
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700763 if (!PageHighMem(page)) {
Pintu Kumarb8af2942013-09-11 14:20:34 -0700764 debug_check_no_locks_freed(page_address(page),
765 PAGE_SIZE << order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700766 debug_check_no_obj_freed(page_address(page),
767 PAGE_SIZE << order);
768 }
Nick Piggindafb1362006-10-11 01:21:30 -0700769 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800770 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700771
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700772 return true;
773}
774
775static void __free_pages_ok(struct page *page, unsigned int order)
776{
777 unsigned long flags;
Minchan Kim95e34412012-10-08 16:32:11 -0700778 int migratetype;
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700779 unsigned long pfn = page_to_pfn(page);
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700780
781 if (!free_pages_prepare(page, order))
782 return;
783
Mel Gormancfc47a22014-06-04 16:10:19 -0700784 migratetype = get_pfnblock_migratetype(page, pfn);
Nick Pigginc54ad302006-01-06 00:10:56 -0800785 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700786 __count_vm_events(PGFREE, 1 << order);
Minchan Kim95e34412012-10-08 16:32:11 -0700787 set_freepage_migratetype(page, migratetype);
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700788 free_one_page(page_zone(page), page, pfn, order, migratetype);
Nick Pigginc54ad302006-01-06 00:10:56 -0800789 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790}
791
Jiang Liu170a5a72013-07-03 15:03:17 -0700792void __init __free_pages_bootmem(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800793{
Johannes Weinerc3993072012-01-10 15:08:10 -0800794 unsigned int nr_pages = 1 << order;
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700795 struct page *p = page;
Johannes Weinerc3993072012-01-10 15:08:10 -0800796 unsigned int loop;
David Howellsa226f6c2006-01-06 00:11:08 -0800797
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700798 prefetchw(p);
799 for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
800 prefetchw(p + 1);
Johannes Weinerc3993072012-01-10 15:08:10 -0800801 __ClearPageReserved(p);
802 set_page_count(p, 0);
David Howellsa226f6c2006-01-06 00:11:08 -0800803 }
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700804 __ClearPageReserved(p);
805 set_page_count(p, 0);
Johannes Weinerc3993072012-01-10 15:08:10 -0800806
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700807 page_zone(page)->managed_pages += nr_pages;
Johannes Weinerc3993072012-01-10 15:08:10 -0800808 set_page_refcounted(page);
809 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800810}
811
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100812#ifdef CONFIG_CMA
Li Zhong9cf510a2013-08-23 13:52:52 +0800813/* Free whole pageblock and set its migration type to MIGRATE_CMA. */
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100814void __init init_cma_reserved_pageblock(struct page *page)
815{
816 unsigned i = pageblock_nr_pages;
817 struct page *p = page;
818
819 do {
820 __ClearPageReserved(p);
821 set_page_count(p, 0);
822 } while (++p, --i);
823
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100824 set_pageblock_migratetype(page, MIGRATE_CMA);
Michal Nazarewiczdc783272014-07-02 15:22:35 -0700825
826 if (pageblock_order >= MAX_ORDER) {
827 i = pageblock_nr_pages;
828 p = page;
829 do {
830 set_page_refcounted(p);
831 __free_pages(p, MAX_ORDER - 1);
832 p += MAX_ORDER_NR_PAGES;
833 } while (i -= MAX_ORDER_NR_PAGES);
834 } else {
835 set_page_refcounted(page);
836 __free_pages(page, pageblock_order);
837 }
838
Jiang Liu3dcc0572013-07-03 15:03:21 -0700839 adjust_managed_page_count(page, pageblock_nr_pages);
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100840}
841#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842
843/*
844 * The order of subdivision here is critical for the IO subsystem.
845 * Please do not alter this order without good reasons and regression
846 * testing. Specifically, as large blocks of memory are subdivided,
847 * the order in which smaller blocks are delivered depends on the order
848 * they're subdivided in this function. This is the primary factor
849 * influencing the order in which pages are delivered to the IO
850 * subsystem according to empirical testing, and this is also justified
851 * by considering the behavior of a buddy system containing a single
852 * large block of memory acted on by a series of small allocations.
853 * This behavior is a critical factor in sglist merging's success.
854 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +0100855 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800857static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700858 int low, int high, struct free_area *area,
859 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860{
861 unsigned long size = 1 << high;
862
863 while (high > low) {
864 area--;
865 high--;
866 size >>= 1;
Sasha Levin309381fea2014-01-23 15:52:54 -0800867 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800868
869#ifdef CONFIG_DEBUG_PAGEALLOC
870 if (high < debug_guardpage_minorder()) {
871 /*
872 * Mark as guard pages (or page), that will allow to
873 * merge back to allocator when buddy will be freed.
874 * Corresponding page table entries will not be touched,
875 * pages will stay not present in virtual address space
876 */
877 INIT_LIST_HEAD(&page[size].lru);
878 set_page_guard_flag(&page[size]);
879 set_page_private(&page[size], high);
880 /* Guard pages are not available for any usage */
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -0700881 __mod_zone_freepage_state(zone, -(1 << high),
882 migratetype);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800883 continue;
884 }
885#endif
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700886 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 area->nr_free++;
888 set_page_order(&page[size], high);
889 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890}
891
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892/*
893 * This page is about to be returned from the page allocator
894 */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200895static inline int check_new_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896{
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -0700897 const char *bad_reason = NULL;
Dave Hansenf0b791a2014-01-23 15:52:49 -0800898 unsigned long bad_flags = 0;
899
900 if (unlikely(page_mapcount(page)))
901 bad_reason = "nonzero mapcount";
902 if (unlikely(page->mapping != NULL))
903 bad_reason = "non-NULL mapping";
904 if (unlikely(atomic_read(&page->_count) != 0))
905 bad_reason = "nonzero _count";
906 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
907 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
908 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
909 }
910 if (unlikely(mem_cgroup_bad_page_check(page)))
911 bad_reason = "cgroup check failed";
912 if (unlikely(bad_reason)) {
913 bad_page(page, bad_reason, bad_flags);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800914 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800915 }
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200916 return 0;
917}
918
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700919static int prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags)
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200920{
921 int i;
922
923 for (i = 0; i < (1 << order); i++) {
924 struct page *p = page + i;
925 if (unlikely(check_new_page(p)))
926 return 1;
927 }
Hugh Dickins689bceb2005-11-21 21:32:20 -0800928
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700929 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800930 set_page_refcounted(page);
Nick Piggincc1025092006-12-06 20:32:00 -0800931
932 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800934
935 if (gfp_flags & __GFP_ZERO)
936 prep_zero_page(page, order, gfp_flags);
937
938 if (order && (gfp_flags & __GFP_COMP))
939 prep_compound_page(page, order);
940
Hugh Dickins689bceb2005-11-21 21:32:20 -0800941 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942}
943
Mel Gorman56fd56b2007-10-16 01:25:58 -0700944/*
945 * Go through the free lists for the given migratetype and remove
946 * the smallest available page from the freelists
947 */
Mel Gorman728ec982009-06-16 15:32:04 -0700948static inline
949struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -0700950 int migratetype)
951{
952 unsigned int current_order;
Pintu Kumarb8af2942013-09-11 14:20:34 -0700953 struct free_area *area;
Mel Gorman56fd56b2007-10-16 01:25:58 -0700954 struct page *page;
955
956 /* Find a page of the appropriate size in the preferred list */
957 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
958 area = &(zone->free_area[current_order]);
959 if (list_empty(&area->free_list[migratetype]))
960 continue;
961
962 page = list_entry(area->free_list[migratetype].next,
963 struct page, lru);
964 list_del(&page->lru);
965 rmv_page_order(page);
966 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -0700967 expand(zone, page, order, current_order, area, migratetype);
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -0700968 set_freepage_migratetype(page, migratetype);
Mel Gorman56fd56b2007-10-16 01:25:58 -0700969 return page;
970 }
971
972 return NULL;
973}
974
975
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700976/*
977 * This array describes the order lists are fallen back to when
978 * the free lists for the desirable migrate type are depleted
979 */
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100980static int fallbacks[MIGRATE_TYPES][4] = {
981 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
982 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
983#ifdef CONFIG_CMA
984 [MIGRATE_MOVABLE] = { MIGRATE_CMA, MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
985 [MIGRATE_CMA] = { MIGRATE_RESERVE }, /* Never used */
986#else
987 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
988#endif
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +0100989 [MIGRATE_RESERVE] = { MIGRATE_RESERVE }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -0800990#ifdef CONFIG_MEMORY_ISOLATION
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +0100991 [MIGRATE_ISOLATE] = { MIGRATE_RESERVE }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -0800992#endif
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700993};
994
Mel Gormanc361be52007-10-16 01:25:51 -0700995/*
996 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -0700997 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -0700998 * boundary. If alignment is required, use move_freepages_block()
999 */
Minchan Kim435b4052012-10-08 16:32:16 -07001000int move_freepages(struct zone *zone,
Adrian Bunkb69a7282008-07-23 21:28:12 -07001001 struct page *start_page, struct page *end_page,
1002 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001003{
1004 struct page *page;
1005 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -07001006 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -07001007
1008#ifndef CONFIG_HOLES_IN_ZONE
1009 /*
1010 * page_zone is not safe to call in this context when
1011 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1012 * anyway as we check zone boundaries in move_freepages_block().
1013 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -07001014 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -07001015 */
Mel Gorman97ee4ba2014-10-09 15:28:28 -07001016 VM_BUG_ON(page_zone(start_page) != page_zone(end_page));
Mel Gormanc361be52007-10-16 01:25:51 -07001017#endif
1018
1019 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -07001020 /* Make sure we are not inadvertently changing nodes */
Sasha Levin309381fea2014-01-23 15:52:54 -08001021 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
Adam Litke344c7902008-09-02 14:35:38 -07001022
Mel Gormanc361be52007-10-16 01:25:51 -07001023 if (!pfn_valid_within(page_to_pfn(page))) {
1024 page++;
1025 continue;
1026 }
1027
1028 if (!PageBuddy(page)) {
1029 page++;
1030 continue;
1031 }
1032
1033 order = page_order(page);
Kirill A. Shutemov84be48d2011-03-22 16:33:41 -07001034 list_move(&page->lru,
1035 &zone->free_area[order].free_list[migratetype]);
Minchan Kim95e34412012-10-08 16:32:11 -07001036 set_freepage_migratetype(page, migratetype);
Mel Gormanc361be52007-10-16 01:25:51 -07001037 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -07001038 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -07001039 }
1040
Mel Gormand1003132007-10-16 01:26:00 -07001041 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -07001042}
1043
Minchan Kimee6f5092012-07-31 16:43:50 -07001044int move_freepages_block(struct zone *zone, struct page *page,
Linus Torvalds68e3e922012-06-03 20:05:57 -07001045 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001046{
1047 unsigned long start_pfn, end_pfn;
1048 struct page *start_page, *end_page;
1049
1050 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -07001051 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -07001052 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -07001053 end_page = start_page + pageblock_nr_pages - 1;
1054 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -07001055
1056 /* Do not cross zone boundaries */
Cody P Schafer108bcc92013-02-22 16:35:23 -08001057 if (!zone_spans_pfn(zone, start_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001058 start_page = page;
Cody P Schafer108bcc92013-02-22 16:35:23 -08001059 if (!zone_spans_pfn(zone, end_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001060 return 0;
1061
1062 return move_freepages(zone, start_page, end_page, migratetype);
1063}
1064
Mel Gorman2f66a682009-09-21 17:02:31 -07001065static void change_pageblock_range(struct page *pageblock_page,
1066 int start_order, int migratetype)
1067{
1068 int nr_pageblocks = 1 << (start_order - pageblock_order);
1069
1070 while (nr_pageblocks--) {
1071 set_pageblock_migratetype(pageblock_page, migratetype);
1072 pageblock_page += pageblock_nr_pages;
1073 }
1074}
1075
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001076/*
1077 * If breaking a large block of pages, move all free pages to the preferred
1078 * allocation list. If falling back for a reclaimable kernel allocation, be
1079 * more aggressive about taking ownership of free pages.
1080 *
1081 * On the other hand, never change migration type of MIGRATE_CMA pageblocks
1082 * nor move CMA pages to different free lists. We don't want unmovable pages
1083 * to be allocated from MIGRATE_CMA areas.
1084 *
1085 * Returns the new migratetype of the pageblock (or the same old migratetype
1086 * if it was unchanged).
1087 */
1088static int try_to_steal_freepages(struct zone *zone, struct page *page,
1089 int start_type, int fallback_type)
1090{
1091 int current_order = page_order(page);
1092
KOSAKI Motohiro0cbef292013-11-12 15:08:20 -08001093 /*
1094 * When borrowing from MIGRATE_CMA, we need to release the excess
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001095 * buddy pages to CMA itself. We also ensure the freepage_migratetype
1096 * is set to CMA so it is returned to the correct freelist in case
1097 * the page ends up being not actually allocated from the pcp lists.
KOSAKI Motohiro0cbef292013-11-12 15:08:20 -08001098 */
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001099 if (is_migrate_cma(fallback_type))
1100 return fallback_type;
1101
1102 /* Take ownership for orders >= pageblock_order */
1103 if (current_order >= pageblock_order) {
1104 change_pageblock_range(page, current_order, start_type);
1105 return start_type;
1106 }
1107
1108 if (current_order >= pageblock_order / 2 ||
1109 start_type == MIGRATE_RECLAIMABLE ||
1110 page_group_by_mobility_disabled) {
1111 int pages;
1112
1113 pages = move_freepages_block(zone, page, start_type);
1114
1115 /* Claim the whole block if over half of it is free */
1116 if (pages >= (1 << (pageblock_order-1)) ||
1117 page_group_by_mobility_disabled) {
1118
1119 set_pageblock_migratetype(page, start_type);
1120 return start_type;
1121 }
1122
1123 }
1124
1125 return fallback_type;
1126}
1127
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001128/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -07001129static inline struct page *
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001130__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001131{
Pintu Kumarb8af2942013-09-11 14:20:34 -07001132 struct free_area *area;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001133 unsigned int current_order;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001134 struct page *page;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001135 int migratetype, new_type, i;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001136
1137 /* Find the largest possible block of pages in the other list */
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001138 for (current_order = MAX_ORDER-1;
1139 current_order >= order && current_order <= MAX_ORDER-1;
1140 --current_order) {
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001141 for (i = 0;; i++) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001142 migratetype = fallbacks[start_migratetype][i];
1143
Mel Gorman56fd56b2007-10-16 01:25:58 -07001144 /* MIGRATE_RESERVE handled later if necessary */
1145 if (migratetype == MIGRATE_RESERVE)
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001146 break;
Mel Gormane0104872007-10-16 01:25:53 -07001147
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001148 area = &(zone->free_area[current_order]);
1149 if (list_empty(&area->free_list[migratetype]))
1150 continue;
1151
1152 page = list_entry(area->free_list[migratetype].next,
1153 struct page, lru);
1154 area->nr_free--;
1155
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001156 new_type = try_to_steal_freepages(zone, page,
1157 start_migratetype,
1158 migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001159
1160 /* Remove the page from the freelists */
1161 list_del(&page->lru);
1162 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001163
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001164 expand(zone, page, order, current_order, area,
KOSAKI Motohiro0cbef292013-11-12 15:08:20 -08001165 new_type);
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001166 /* The freepage_migratetype may differ from pageblock's
1167 * migratetype depending on the decisions in
1168 * try_to_steal_freepages. This is OK as long as it does
1169 * not differ for MIGRATE_CMA type.
1170 */
1171 set_freepage_migratetype(page, new_type);
Mel Gormane0fff1b2009-09-21 17:02:42 -07001172
KOSAKI Motohiro52c8f6a2013-11-12 15:08:19 -08001173 trace_mm_page_alloc_extfrag(page, order, current_order,
1174 start_migratetype, migratetype, new_type);
Mel Gormane0fff1b2009-09-21 17:02:42 -07001175
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001176 return page;
1177 }
1178 }
1179
Mel Gorman728ec982009-06-16 15:32:04 -07001180 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001181}
1182
Mel Gorman56fd56b2007-10-16 01:25:58 -07001183/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 * Do the hard work of removing an element from the buddy allocator.
1185 * Call me with the zone->lock already held.
1186 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001187static struct page *__rmqueue(struct zone *zone, unsigned int order,
1188 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 struct page *page;
1191
Mel Gorman728ec982009-06-16 15:32:04 -07001192retry_reserve:
Mel Gorman56fd56b2007-10-16 01:25:58 -07001193 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194
Mel Gorman728ec982009-06-16 15:32:04 -07001195 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07001196 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001197
Mel Gorman728ec982009-06-16 15:32:04 -07001198 /*
1199 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1200 * is used because __rmqueue_smallest is an inline function
1201 * and we want just one call site
1202 */
1203 if (!page) {
1204 migratetype = MIGRATE_RESERVE;
1205 goto retry_reserve;
1206 }
1207 }
1208
Mel Gorman0d3d0622009-09-21 17:02:44 -07001209 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001210 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211}
1212
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001213/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 * Obtain a specified number of elements from the buddy allocator, all under
1215 * a single hold of the lock, for efficiency. Add them to the supplied list.
1216 * Returns the number of new pages which were placed at *list.
1217 */
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001218static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001219 unsigned long count, struct list_head *list,
Mel Gormanb745bc82014-06-04 16:10:22 -07001220 int migratetype, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221{
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001222 int i;
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001223
Nick Pigginc54ad302006-01-06 00:10:56 -08001224 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001226 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001227 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08001229
1230 /*
1231 * Split buddy pages returned by expand() are received here
1232 * in physical page order. The page is added to the callers and
1233 * list and the list head then moves forward. From the callers
1234 * perspective, the linked list is ordered by page number in
1235 * some conditions. This is useful for IO devices that can
1236 * merge IO requests if the physical pages are ordered
1237 * properly.
1238 */
Mel Gormanb745bc82014-06-04 16:10:22 -07001239 if (likely(!cold))
Mel Gormane084b2d2009-07-29 15:02:04 -07001240 list_add(&page->lru, list);
1241 else
1242 list_add_tail(&page->lru, list);
Mel Gorman81eabcb2007-12-17 16:20:05 -08001243 list = &page->lru;
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001244 if (is_migrate_cma(get_freepage_migratetype(page)))
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07001245 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
1246 -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 }
Mel Gormanf2260e62009-06-16 15:32:13 -07001248 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08001249 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001250 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251}
1252
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001253#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001254/*
Christoph Lameter4037d452007-05-09 02:35:14 -07001255 * Called from the vmstat counter updater to drain pagesets of this
1256 * currently executing processor on remote nodes after they have
1257 * expired.
1258 *
Christoph Lameter879336c2006-03-22 00:09:08 -08001259 * Note that this function must be called with the thread pinned to
1260 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001261 */
Christoph Lameter4037d452007-05-09 02:35:14 -07001262void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001263{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001264 unsigned long flags;
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07001265 int to_drain, batch;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001266
Christoph Lameter4037d452007-05-09 02:35:14 -07001267 local_irq_save(flags);
Cody P Schafer998d39c2013-07-03 15:01:32 -07001268 batch = ACCESS_ONCE(pcp->batch);
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07001269 to_drain = min(pcp->count, batch);
KOSAKI Motohiro2a135152012-07-31 16:42:53 -07001270 if (to_drain > 0) {
1271 free_pcppages_bulk(zone, to_drain, pcp);
1272 pcp->count -= to_drain;
1273 }
Christoph Lameter4037d452007-05-09 02:35:14 -07001274 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001275}
1276#endif
1277
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001278/*
1279 * Drain pages of the indicated processor.
1280 *
1281 * The processor must either be the current processor and the
1282 * thread pinned to the current processor or a processor that
1283 * is not online.
1284 */
1285static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286{
Nick Pigginc54ad302006-01-06 00:10:56 -08001287 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07001290 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 struct per_cpu_pageset *pset;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001292 struct per_cpu_pages *pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001294 local_irq_save(flags);
1295 pset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001297 pcp = &pset->pcp;
David Rientjes2ff754f2011-01-25 15:07:23 -08001298 if (pcp->count) {
1299 free_pcppages_bulk(zone, pcp->count, pcp);
1300 pcp->count = 0;
1301 }
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001302 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 }
1304}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001306/*
1307 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1308 */
1309void drain_local_pages(void *arg)
1310{
1311 drain_pages(smp_processor_id());
1312}
1313
1314/*
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001315 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1316 *
1317 * Note that this code is protected against sending an IPI to an offline
1318 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1319 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1320 * nothing keeps CPUs from showing up after we populated the cpumask and
1321 * before the call to on_each_cpu_mask().
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001322 */
1323void drain_all_pages(void)
1324{
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001325 int cpu;
1326 struct per_cpu_pageset *pcp;
1327 struct zone *zone;
1328
1329 /*
1330 * Allocate in the BSS so we wont require allocation in
1331 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1332 */
1333 static cpumask_t cpus_with_pcps;
1334
1335 /*
1336 * We don't care about racing with CPU hotplug event
1337 * as offline notification will cause the notified
1338 * cpu to drain that CPU pcps and on_each_cpu_mask
1339 * disables preemption as part of its processing
1340 */
1341 for_each_online_cpu(cpu) {
1342 bool has_pcps = false;
1343 for_each_populated_zone(zone) {
1344 pcp = per_cpu_ptr(zone->pageset, cpu);
1345 if (pcp->pcp.count) {
1346 has_pcps = true;
1347 break;
1348 }
1349 }
1350 if (has_pcps)
1351 cpumask_set_cpu(cpu, &cpus_with_pcps);
1352 else
1353 cpumask_clear_cpu(cpu, &cpus_with_pcps);
1354 }
1355 on_each_cpu_mask(&cpus_with_pcps, drain_local_pages, NULL, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001356}
1357
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02001358#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359
1360void mark_free_pages(struct zone *zone)
1361{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001362 unsigned long pfn, max_zone_pfn;
1363 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001364 unsigned int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 struct list_head *curr;
1366
Xishi Qiu8080fc02013-09-11 14:21:45 -07001367 if (zone_is_empty(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 return;
1369
1370 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001371
Cody P Schafer108bcc92013-02-22 16:35:23 -08001372 max_zone_pfn = zone_end_pfn(zone);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001373 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1374 if (pfn_valid(pfn)) {
1375 struct page *page = pfn_to_page(pfn);
1376
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001377 if (!swsusp_page_is_forbidden(page))
1378 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001379 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001381 for_each_migratetype_order(order, t) {
1382 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001383 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001385 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1386 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001387 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001388 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001389 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 spin_unlock_irqrestore(&zone->lock, flags);
1391}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001392#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393
1394/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 * Free a 0-order page
Mel Gormanb745bc82014-06-04 16:10:22 -07001396 * cold == true ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 */
Mel Gormanb745bc82014-06-04 16:10:22 -07001398void free_hot_cold_page(struct page *page, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399{
1400 struct zone *zone = page_zone(page);
1401 struct per_cpu_pages *pcp;
1402 unsigned long flags;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001403 unsigned long pfn = page_to_pfn(page);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001404 int migratetype;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001406 if (!free_pages_prepare(page, 0))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001407 return;
1408
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001409 migratetype = get_pfnblock_migratetype(page, pfn);
Minchan Kimb12c4ad2012-10-08 16:32:08 -07001410 set_freepage_migratetype(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001412 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001413
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001414 /*
1415 * We only track unmovable, reclaimable and movable on pcp lists.
1416 * Free ISOLATE pages back to the allocator because they are being
1417 * offlined but treat RESERVE as movable pages so we can get those
1418 * areas back if necessary. Otherwise, we may have to free
1419 * excessively into the page allocator
1420 */
1421 if (migratetype >= MIGRATE_PCPTYPES) {
Minchan Kim194159f2013-02-22 16:33:58 -08001422 if (unlikely(is_migrate_isolate(migratetype))) {
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001423 free_one_page(zone, page, pfn, 0, migratetype);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001424 goto out;
1425 }
1426 migratetype = MIGRATE_MOVABLE;
1427 }
1428
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001429 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Mel Gormanb745bc82014-06-04 16:10:22 -07001430 if (!cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001431 list_add(&page->lru, &pcp->lists[migratetype]);
Mel Gormanb745bc82014-06-04 16:10:22 -07001432 else
1433 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001435 if (pcp->count >= pcp->high) {
Cody P Schafer998d39c2013-07-03 15:01:32 -07001436 unsigned long batch = ACCESS_ONCE(pcp->batch);
1437 free_pcppages_bulk(zone, batch, pcp);
1438 pcp->count -= batch;
Nick Piggin48db57f2006-01-08 01:00:42 -08001439 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001440
1441out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443}
1444
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001445/*
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001446 * Free a list of 0-order pages
1447 */
Mel Gormanb745bc82014-06-04 16:10:22 -07001448void free_hot_cold_page_list(struct list_head *list, bool cold)
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001449{
1450 struct page *page, *next;
1451
1452 list_for_each_entry_safe(page, next, list, lru) {
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -08001453 trace_mm_page_free_batched(page, cold);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001454 free_hot_cold_page(page, cold);
1455 }
1456}
1457
1458/*
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001459 * split_page takes a non-compound higher-order page, and splits it into
1460 * n (1<<order) sub-pages: page[0..n]
1461 * Each sub-page must be freed individually.
1462 *
1463 * Note: this is probably too low level an operation for use in drivers.
1464 * Please consult with lkml before using this in your driver.
1465 */
1466void split_page(struct page *page, unsigned int order)
1467{
1468 int i;
1469
Sasha Levin309381fea2014-01-23 15:52:54 -08001470 VM_BUG_ON_PAGE(PageCompound(page), page);
1471 VM_BUG_ON_PAGE(!page_count(page), page);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001472
1473#ifdef CONFIG_KMEMCHECK
1474 /*
1475 * Split shadow pages too, because free(page[0]) would
1476 * otherwise free the whole shadow.
1477 */
1478 if (kmemcheck_page_is_tracked(page))
1479 split_page(virt_to_page(page[0].shadow), order);
1480#endif
1481
Nick Piggin7835e982006-03-22 00:08:40 -08001482 for (i = 1; i < (1 << order); i++)
1483 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001484}
K. Y. Srinivasan5853ff22013-03-25 15:47:38 -07001485EXPORT_SYMBOL_GPL(split_page);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001486
Mel Gorman8fb74b92013-01-11 14:32:16 -08001487static int __isolate_free_page(struct page *page, unsigned int order)
Mel Gorman748446b2010-05-24 14:32:27 -07001488{
Mel Gorman748446b2010-05-24 14:32:27 -07001489 unsigned long watermark;
1490 struct zone *zone;
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07001491 int mt;
Mel Gorman748446b2010-05-24 14:32:27 -07001492
1493 BUG_ON(!PageBuddy(page));
1494
1495 zone = page_zone(page);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001496 mt = get_pageblock_migratetype(page);
Mel Gorman748446b2010-05-24 14:32:27 -07001497
Minchan Kim194159f2013-02-22 16:33:58 -08001498 if (!is_migrate_isolate(mt)) {
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001499 /* Obey watermarks as if the page was being allocated */
1500 watermark = low_wmark_pages(zone) + (1 << order);
1501 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1502 return 0;
1503
Mel Gorman8fb74b92013-01-11 14:32:16 -08001504 __mod_zone_freepage_state(zone, -(1UL << order), mt);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001505 }
Mel Gorman748446b2010-05-24 14:32:27 -07001506
1507 /* Remove page from free list */
1508 list_del(&page->lru);
1509 zone->free_area[order].nr_free--;
1510 rmv_page_order(page);
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07001511
Mel Gorman8fb74b92013-01-11 14:32:16 -08001512 /* Set the pageblock if the isolated page is at least a pageblock */
Mel Gorman748446b2010-05-24 14:32:27 -07001513 if (order >= pageblock_order - 1) {
1514 struct page *endpage = page + (1 << order) - 1;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001515 for (; page < endpage; page += pageblock_nr_pages) {
1516 int mt = get_pageblock_migratetype(page);
Minchan Kim194159f2013-02-22 16:33:58 -08001517 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001518 set_pageblock_migratetype(page,
1519 MIGRATE_MOVABLE);
1520 }
Mel Gorman748446b2010-05-24 14:32:27 -07001521 }
1522
Mel Gorman8fb74b92013-01-11 14:32:16 -08001523 return 1UL << order;
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001524}
1525
1526/*
1527 * Similar to split_page except the page is already free. As this is only
1528 * being used for migration, the migratetype of the block also changes.
1529 * As this is called with interrupts disabled, the caller is responsible
1530 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1531 * are enabled.
1532 *
1533 * Note: this is probably too low level an operation for use in drivers.
1534 * Please consult with lkml before using this in your driver.
1535 */
1536int split_free_page(struct page *page)
1537{
1538 unsigned int order;
1539 int nr_pages;
1540
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001541 order = page_order(page);
1542
Mel Gorman8fb74b92013-01-11 14:32:16 -08001543 nr_pages = __isolate_free_page(page, order);
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001544 if (!nr_pages)
1545 return 0;
1546
1547 /* Split into individual pages */
1548 set_page_refcounted(page);
1549 split_page(page, order);
1550 return nr_pages;
Mel Gorman748446b2010-05-24 14:32:27 -07001551}
1552
1553/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1555 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1556 * or two.
1557 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07001558static inline
1559struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001560 struct zone *zone, unsigned int order,
1561 gfp_t gfp_flags, int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562{
1563 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001564 struct page *page;
Mel Gormanb745bc82014-06-04 16:10:22 -07001565 bool cold = ((gfp_flags & __GFP_COLD) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566
Hugh Dickins689bceb2005-11-21 21:32:20 -08001567again:
Nick Piggin48db57f2006-01-08 01:00:42 -08001568 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001570 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 local_irq_save(flags);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001573 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1574 list = &pcp->lists[migratetype];
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001575 if (list_empty(list)) {
Mel Gorman535131e62007-10-16 01:25:49 -07001576 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001577 pcp->batch, list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001578 migratetype, cold);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001579 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07001580 goto failed;
Mel Gorman535131e62007-10-16 01:25:49 -07001581 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001582
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001583 if (cold)
1584 page = list_entry(list->prev, struct page, lru);
1585 else
1586 page = list_entry(list->next, struct page, lru);
1587
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001588 list_del(&page->lru);
1589 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001590 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07001591 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1592 /*
1593 * __GFP_NOFAIL is not to be used in new code.
1594 *
1595 * All __GFP_NOFAIL callers should be fixed so that they
1596 * properly detect and handle allocation failures.
1597 *
1598 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07001599 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07001600 * __GFP_NOFAIL.
1601 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07001602 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07001603 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001605 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001606 spin_unlock(&zone->lock);
1607 if (!page)
1608 goto failed;
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07001609 __mod_zone_freepage_state(zone, -(1 << order),
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001610 get_freepage_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 }
1612
Johannes Weiner3a025762014-04-07 15:37:48 -07001613 __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
Johannes Weinerabe5f972014-10-02 16:21:10 -07001614 if (atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]) <= 0 &&
Johannes Weiner57054652014-10-09 15:28:17 -07001615 !test_bit(ZONE_FAIR_DEPLETED, &zone->flags))
1616 set_bit(ZONE_FAIR_DEPLETED, &zone->flags);
Johannes Weiner27329362014-03-03 15:38:41 -08001617
Christoph Lameterf8891e52006-06-30 01:55:45 -07001618 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Andi Kleen78afd562011-03-22 16:33:12 -07001619 zone_statistics(preferred_zone, zone, gfp_flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001620 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621
Sasha Levin309381fea2014-01-23 15:52:54 -08001622 VM_BUG_ON_PAGE(bad_range(zone, page), page);
Nick Piggin17cf4402006-03-22 00:08:41 -08001623 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001624 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001626
1627failed:
1628 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001629 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630}
1631
Akinobu Mita933e3122006-12-08 02:39:45 -08001632#ifdef CONFIG_FAIL_PAGE_ALLOC
1633
Akinobu Mitab2588c42011-07-26 16:09:03 -07001634static struct {
Akinobu Mita933e3122006-12-08 02:39:45 -08001635 struct fault_attr attr;
1636
1637 u32 ignore_gfp_highmem;
1638 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001639 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001640} fail_page_alloc = {
1641 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001642 .ignore_gfp_wait = 1,
1643 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001644 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001645};
1646
1647static int __init setup_fail_page_alloc(char *str)
1648{
1649 return setup_fault_attr(&fail_page_alloc.attr, str);
1650}
1651__setup("fail_page_alloc=", setup_fail_page_alloc);
1652
Gavin Shandeaf3862012-07-31 16:41:51 -07001653static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08001654{
Akinobu Mita54114992007-07-15 23:40:23 -07001655 if (order < fail_page_alloc.min_order)
Gavin Shandeaf3862012-07-31 16:41:51 -07001656 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001657 if (gfp_mask & __GFP_NOFAIL)
Gavin Shandeaf3862012-07-31 16:41:51 -07001658 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001659 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
Gavin Shandeaf3862012-07-31 16:41:51 -07001660 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001661 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
Gavin Shandeaf3862012-07-31 16:41:51 -07001662 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001663
1664 return should_fail(&fail_page_alloc.attr, 1 << order);
1665}
1666
1667#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1668
1669static int __init fail_page_alloc_debugfs(void)
1670{
Al Virof4ae40a2011-07-24 04:33:43 -04001671 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
Akinobu Mita933e3122006-12-08 02:39:45 -08001672 struct dentry *dir;
Akinobu Mita933e3122006-12-08 02:39:45 -08001673
Akinobu Mitadd48c082011-08-03 16:21:01 -07001674 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
1675 &fail_page_alloc.attr);
1676 if (IS_ERR(dir))
1677 return PTR_ERR(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08001678
Akinobu Mitab2588c42011-07-26 16:09:03 -07001679 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
1680 &fail_page_alloc.ignore_gfp_wait))
1681 goto fail;
1682 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1683 &fail_page_alloc.ignore_gfp_highmem))
1684 goto fail;
1685 if (!debugfs_create_u32("min-order", mode, dir,
1686 &fail_page_alloc.min_order))
1687 goto fail;
Akinobu Mita933e3122006-12-08 02:39:45 -08001688
Akinobu Mitab2588c42011-07-26 16:09:03 -07001689 return 0;
1690fail:
Akinobu Mitadd48c082011-08-03 16:21:01 -07001691 debugfs_remove_recursive(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08001692
Akinobu Mitab2588c42011-07-26 16:09:03 -07001693 return -ENOMEM;
Akinobu Mita933e3122006-12-08 02:39:45 -08001694}
1695
1696late_initcall(fail_page_alloc_debugfs);
1697
1698#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1699
1700#else /* CONFIG_FAIL_PAGE_ALLOC */
1701
Gavin Shandeaf3862012-07-31 16:41:51 -07001702static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08001703{
Gavin Shandeaf3862012-07-31 16:41:51 -07001704 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001705}
1706
1707#endif /* CONFIG_FAIL_PAGE_ALLOC */
1708
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709/*
Mel Gorman88f5acf2011-01-13 15:45:41 -08001710 * Return true if free pages are above 'mark'. This takes into account the order
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 * of the allocation.
1712 */
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001713static bool __zone_watermark_ok(struct zone *z, unsigned int order,
1714 unsigned long mark, int classzone_idx, int alloc_flags,
1715 long free_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716{
1717 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001718 long min = mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 int o;
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07001720 long free_cma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721
Michal Hockodf0a6da2012-01-10 15:08:02 -08001722 free_pages -= (1 << order) - 1;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001723 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001725 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 min -= min / 4;
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07001727#ifdef CONFIG_CMA
1728 /* If allocation can't use CMA areas don't use free CMA pages */
1729 if (!(alloc_flags & ALLOC_CMA))
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07001730 free_cma = zone_page_state(z, NR_FREE_CMA_PAGES);
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07001731#endif
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07001732
Mel Gorman3484b2d2014-08-06 16:07:14 -07001733 if (free_pages - free_cma <= min + z->lowmem_reserve[classzone_idx])
Mel Gorman88f5acf2011-01-13 15:45:41 -08001734 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 for (o = 0; o < order; o++) {
1736 /* At the next order, this order's pages become unavailable */
1737 free_pages -= z->free_area[o].nr_free << o;
1738
1739 /* Require fewer higher order pages to be free */
1740 min >>= 1;
1741
1742 if (free_pages <= min)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001743 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 }
Mel Gorman88f5acf2011-01-13 15:45:41 -08001745 return true;
1746}
1747
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001748bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
Mel Gorman88f5acf2011-01-13 15:45:41 -08001749 int classzone_idx, int alloc_flags)
1750{
1751 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1752 zone_page_state(z, NR_FREE_PAGES));
1753}
1754
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001755bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
1756 unsigned long mark, int classzone_idx, int alloc_flags)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001757{
1758 long free_pages = zone_page_state(z, NR_FREE_PAGES);
1759
1760 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1761 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1762
1763 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1764 free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765}
1766
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001767#ifdef CONFIG_NUMA
1768/*
1769 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1770 * skip over zones that are not allowed by the cpuset, or that have
1771 * been recently (in last second) found to be nearly full. See further
1772 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001773 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001774 *
Zhi Yong Wua1aeb652013-11-12 15:08:29 -08001775 * If the zonelist cache is present in the passed zonelist, then
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001776 * returns a pointer to the allowed node mask (either the current
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08001777 * tasks mems_allowed, or node_states[N_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001778 *
1779 * If the zonelist cache is not available for this zonelist, does
1780 * nothing and returns NULL.
1781 *
1782 * If the fullzones BITMAP in the zonelist cache is stale (more than
1783 * a second since last zap'd) then we zap it out (clear its bits.)
1784 *
1785 * We hold off even calling zlc_setup, until after we've checked the
1786 * first zone in the zonelist, on the theory that most allocations will
1787 * be satisfied from that first zone, so best to examine that zone as
1788 * quickly as we can.
1789 */
1790static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1791{
1792 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1793 nodemask_t *allowednodes; /* zonelist_cache approximation */
1794
1795 zlc = zonelist->zlcache_ptr;
1796 if (!zlc)
1797 return NULL;
1798
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001799 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001800 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1801 zlc->last_full_zap = jiffies;
1802 }
1803
1804 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1805 &cpuset_current_mems_allowed :
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08001806 &node_states[N_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001807 return allowednodes;
1808}
1809
1810/*
1811 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1812 * if it is worth looking at further for free memory:
1813 * 1) Check that the zone isn't thought to be full (doesn't have its
1814 * bit set in the zonelist_cache fullzones BITMAP).
1815 * 2) Check that the zones node (obtained from the zonelist_cache
1816 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1817 * Return true (non-zero) if zone is worth looking at further, or
1818 * else return false (zero) if it is not.
1819 *
1820 * This check -ignores- the distinction between various watermarks,
1821 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1822 * found to be full for any variation of these watermarks, it will
1823 * be considered full for up to one second by all requests, unless
1824 * we are so low on memory on all allowed nodes that we are forced
1825 * into the second scan of the zonelist.
1826 *
1827 * In the second scan we ignore this zonelist cache and exactly
1828 * apply the watermarks to all zones, even it is slower to do so.
1829 * We are low on memory in the second scan, and should leave no stone
1830 * unturned looking for a free page.
1831 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001832static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001833 nodemask_t *allowednodes)
1834{
1835 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1836 int i; /* index of *z in zonelist zones */
1837 int n; /* node that zone *z is on */
1838
1839 zlc = zonelist->zlcache_ptr;
1840 if (!zlc)
1841 return 1;
1842
Mel Gormandd1a2392008-04-28 02:12:17 -07001843 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001844 n = zlc->z_to_n[i];
1845
1846 /* This zone is worth trying if it is allowed but not full */
1847 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1848}
1849
1850/*
1851 * Given 'z' scanning a zonelist, set the corresponding bit in
1852 * zlc->fullzones, so that subsequent attempts to allocate a page
1853 * from that zone don't waste time re-examining it.
1854 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001855static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001856{
1857 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1858 int i; /* index of *z in zonelist zones */
1859
1860 zlc = zonelist->zlcache_ptr;
1861 if (!zlc)
1862 return;
1863
Mel Gormandd1a2392008-04-28 02:12:17 -07001864 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001865
1866 set_bit(i, zlc->fullzones);
1867}
1868
Mel Gorman76d3fbf2011-07-25 17:12:30 -07001869/*
1870 * clear all zones full, called after direct reclaim makes progress so that
1871 * a zone that was recently full is not skipped over for up to a second
1872 */
1873static void zlc_clear_zones_full(struct zonelist *zonelist)
1874{
1875 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1876
1877 zlc = zonelist->zlcache_ptr;
1878 if (!zlc)
1879 return;
1880
1881 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1882}
1883
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001884static bool zone_local(struct zone *local_zone, struct zone *zone)
1885{
Johannes Weinerfff40682013-12-20 14:54:12 +00001886 return local_zone->node == zone->node;
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001887}
1888
David Rientjes957f8222012-10-08 16:33:24 -07001889static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
1890{
Mel Gorman5f7a75a2014-06-04 16:07:15 -07001891 return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <
1892 RECLAIM_DISTANCE;
David Rientjes957f8222012-10-08 16:33:24 -07001893}
1894
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001895#else /* CONFIG_NUMA */
1896
1897static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1898{
1899 return NULL;
1900}
1901
Mel Gormandd1a2392008-04-28 02:12:17 -07001902static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001903 nodemask_t *allowednodes)
1904{
1905 return 1;
1906}
1907
Mel Gormandd1a2392008-04-28 02:12:17 -07001908static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001909{
1910}
Mel Gorman76d3fbf2011-07-25 17:12:30 -07001911
1912static void zlc_clear_zones_full(struct zonelist *zonelist)
1913{
1914}
David Rientjes957f8222012-10-08 16:33:24 -07001915
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001916static bool zone_local(struct zone *local_zone, struct zone *zone)
1917{
1918 return true;
1919}
1920
David Rientjes957f8222012-10-08 16:33:24 -07001921static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
1922{
1923 return true;
1924}
1925
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001926#endif /* CONFIG_NUMA */
1927
Mel Gorman4ffeaf32014-08-06 16:07:22 -07001928static void reset_alloc_batches(struct zone *preferred_zone)
1929{
1930 struct zone *zone = preferred_zone->zone_pgdat->node_zones;
1931
1932 do {
1933 mod_zone_page_state(zone, NR_ALLOC_BATCH,
1934 high_wmark_pages(zone) - low_wmark_pages(zone) -
1935 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
Johannes Weiner57054652014-10-09 15:28:17 -07001936 clear_bit(ZONE_FAIR_DEPLETED, &zone->flags);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07001937 } while (zone++ != preferred_zone);
1938}
1939
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001940/*
Paul Jackson0798e512006-12-06 20:31:38 -08001941 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001942 * a page.
1943 */
1944static struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001945get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
Mel Gorman5117f452009-06-16 15:31:59 -07001946 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
Mel Gormand8846372014-06-04 16:10:33 -07001947 struct zone *preferred_zone, int classzone_idx, int migratetype)
Martin Hicks753ee722005-06-21 17:14:41 -07001948{
Mel Gormandd1a2392008-04-28 02:12:17 -07001949 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001950 struct page *page = NULL;
Mel Gorman5117f452009-06-16 15:31:59 -07001951 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001952 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1953 int zlc_active = 0; /* set if using zonelist_cache */
1954 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gormana6e21b142014-06-04 16:10:12 -07001955 bool consider_zone_dirty = (alloc_flags & ALLOC_WMARK_LOW) &&
1956 (gfp_mask & __GFP_WRITE);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07001957 int nr_fair_skipped = 0;
1958 bool zonelist_rescan;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001959
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001960zonelist_scan:
Mel Gorman4ffeaf32014-08-06 16:07:22 -07001961 zonelist_rescan = false;
1962
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001963 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001964 * Scan zonelist, looking for a zone with enough free.
SeungHun Lee3b11f0a2013-09-11 14:22:23 -07001965 * See also __cpuset_node_allowed_softwall() comment in kernel/cpuset.c.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001966 */
Mel Gorman19770b32008-04-28 02:12:18 -07001967 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1968 high_zoneidx, nodemask) {
Johannes Weinere085dbc2013-09-11 14:20:46 -07001969 unsigned long mark;
1970
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08001971 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001972 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1973 continue;
Mel Gorman664eedd2014-06-04 16:10:08 -07001974 if (cpusets_enabled() &&
1975 (alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001976 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Mel Gormancd38b112011-07-25 17:12:29 -07001977 continue;
Johannes Weinera756cf52012-01-10 15:07:49 -08001978 /*
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001979 * Distribute pages in proportion to the individual
1980 * zone size to ensure fair page aging. The zone a
1981 * page was allocated in should have no effect on the
1982 * time the page has in memory before being reclaimed.
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001983 */
Johannes Weiner3a025762014-04-07 15:37:48 -07001984 if (alloc_flags & ALLOC_FAIR) {
Johannes Weinerfff40682013-12-20 14:54:12 +00001985 if (!zone_local(preferred_zone, zone))
Mel Gormanf7b5d642014-08-06 16:07:20 -07001986 break;
Johannes Weiner57054652014-10-09 15:28:17 -07001987 if (test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) {
Mel Gorman4ffeaf32014-08-06 16:07:22 -07001988 nr_fair_skipped++;
Johannes Weiner3a025762014-04-07 15:37:48 -07001989 continue;
Mel Gorman4ffeaf32014-08-06 16:07:22 -07001990 }
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001991 }
1992 /*
Johannes Weinera756cf52012-01-10 15:07:49 -08001993 * When allocating a page cache page for writing, we
1994 * want to get it from a zone that is within its dirty
1995 * limit, such that no single zone holds more than its
1996 * proportional share of globally allowed dirty pages.
1997 * The dirty limits take into account the zone's
1998 * lowmem reserves and high watermark so that kswapd
1999 * should be able to balance it without having to
2000 * write pages from its LRU list.
2001 *
2002 * This may look like it could increase pressure on
2003 * lower zones by failing allocations in higher zones
2004 * before they are full. But the pages that do spill
2005 * over are limited as the lower zones are protected
2006 * by this very same mechanism. It should not become
2007 * a practical burden to them.
2008 *
2009 * XXX: For now, allow allocations to potentially
2010 * exceed the per-zone dirty limit in the slowpath
2011 * (ALLOC_WMARK_LOW unset) before going into reclaim,
2012 * which is important when on a NUMA setup the allowed
2013 * zones are together not big enough to reach the
2014 * global limit. The proper fix for these situations
2015 * will require awareness of zones in the
2016 * dirty-throttling and the flusher threads.
2017 */
Mel Gormana6e21b142014-06-04 16:10:12 -07002018 if (consider_zone_dirty && !zone_dirty_ok(zone))
Mel Gorman800a1e72014-06-04 16:10:06 -07002019 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002020
Johannes Weinere085dbc2013-09-11 14:20:46 -07002021 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
2022 if (!zone_watermark_ok(zone, order, mark,
2023 classzone_idx, alloc_flags)) {
Mel Gormanfa5e0842009-06-16 15:33:22 -07002024 int ret;
2025
Mel Gorman5dab2912014-06-04 16:10:14 -07002026 /* Checked here to keep the fast path fast */
2027 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
2028 if (alloc_flags & ALLOC_NO_WATERMARKS)
2029 goto try_this_zone;
2030
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002031 if (IS_ENABLED(CONFIG_NUMA) &&
2032 !did_zlc_setup && nr_online_nodes > 1) {
Mel Gormancd38b112011-07-25 17:12:29 -07002033 /*
2034 * we do zlc_setup if there are multiple nodes
2035 * and before considering the first zone allowed
2036 * by the cpuset.
2037 */
2038 allowednodes = zlc_setup(zonelist, alloc_flags);
2039 zlc_active = 1;
2040 did_zlc_setup = 1;
2041 }
2042
David Rientjes957f8222012-10-08 16:33:24 -07002043 if (zone_reclaim_mode == 0 ||
2044 !zone_allows_reclaim(preferred_zone, zone))
Mel Gormanfa5e0842009-06-16 15:33:22 -07002045 goto this_zone_full;
2046
Mel Gormancd38b112011-07-25 17:12:29 -07002047 /*
2048 * As we may have just activated ZLC, check if the first
2049 * eligible zone has failed zone_reclaim recently.
2050 */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002051 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
Mel Gormancd38b112011-07-25 17:12:29 -07002052 !zlc_zone_worth_trying(zonelist, z, allowednodes))
2053 continue;
2054
Mel Gormanfa5e0842009-06-16 15:33:22 -07002055 ret = zone_reclaim(zone, gfp_mask, order);
2056 switch (ret) {
2057 case ZONE_RECLAIM_NOSCAN:
2058 /* did not scan */
Mel Gormancd38b112011-07-25 17:12:29 -07002059 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002060 case ZONE_RECLAIM_FULL:
2061 /* scanned but unreclaimable */
Mel Gormancd38b112011-07-25 17:12:29 -07002062 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002063 default:
2064 /* did we reclaim enough */
Mel Gormanfed27192013-04-29 15:07:57 -07002065 if (zone_watermark_ok(zone, order, mark,
Mel Gormanfa5e0842009-06-16 15:33:22 -07002066 classzone_idx, alloc_flags))
Mel Gormanfed27192013-04-29 15:07:57 -07002067 goto try_this_zone;
2068
2069 /*
2070 * Failed to reclaim enough to meet watermark.
2071 * Only mark the zone full if checking the min
2072 * watermark or if we failed to reclaim just
2073 * 1<<order pages or else the page allocator
2074 * fastpath will prematurely mark zones full
2075 * when the watermark is between the low and
2076 * min watermarks.
2077 */
2078 if (((alloc_flags & ALLOC_WMARK_MASK) == ALLOC_WMARK_MIN) ||
2079 ret == ZONE_RECLAIM_SOME)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002080 goto this_zone_full;
Mel Gormanfed27192013-04-29 15:07:57 -07002081
2082 continue;
Paul Jackson0798e512006-12-06 20:31:38 -08002083 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002084 }
2085
Mel Gormanfa5e0842009-06-16 15:33:22 -07002086try_this_zone:
Mel Gorman3dd28262009-06-16 15:32:00 -07002087 page = buffered_rmqueue(preferred_zone, zone, order,
2088 gfp_mask, migratetype);
Paul Jackson0798e512006-12-06 20:31:38 -08002089 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002090 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002091this_zone_full:
Mel Gorman65bb3712014-06-04 16:10:05 -07002092 if (IS_ENABLED(CONFIG_NUMA) && zlc_active)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002093 zlc_mark_zone_full(zonelist, z);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002094 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002095
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002096 if (page) {
Alex Shib1211862012-08-21 16:16:08 -07002097 /*
2098 * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was
2099 * necessary to allocate the page. The expectation is
2100 * that the caller is taking steps that will free more
2101 * memory. The caller should avoid the page being used
2102 * for !PFMEMALLOC purposes.
2103 */
2104 page->pfmemalloc = !!(alloc_flags & ALLOC_NO_WATERMARKS);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002105 return page;
2106 }
Alex Shib1211862012-08-21 16:16:08 -07002107
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002108 /*
2109 * The first pass makes sure allocations are spread fairly within the
2110 * local node. However, the local node might have free pages left
2111 * after the fairness batches are exhausted, and remote zones haven't
2112 * even been considered yet. Try once more without fairness, and
2113 * include remote zones now, before entering the slowpath and waking
2114 * kswapd: prefer spilling to a remote zone over swapping locally.
2115 */
2116 if (alloc_flags & ALLOC_FAIR) {
2117 alloc_flags &= ~ALLOC_FAIR;
2118 if (nr_fair_skipped) {
2119 zonelist_rescan = true;
2120 reset_alloc_batches(preferred_zone);
2121 }
2122 if (nr_online_nodes > 1)
2123 zonelist_rescan = true;
2124 }
2125
2126 if (unlikely(IS_ENABLED(CONFIG_NUMA) && zlc_active)) {
2127 /* Disable zlc cache for second zonelist scan */
2128 zlc_active = 0;
2129 zonelist_rescan = true;
2130 }
2131
2132 if (zonelist_rescan)
2133 goto zonelist_scan;
2134
2135 return NULL;
Martin Hicks753ee722005-06-21 17:14:41 -07002136}
2137
David Rientjes29423e772011-03-22 16:30:47 -07002138/*
2139 * Large machines with many possible nodes should not always dump per-node
2140 * meminfo in irq context.
2141 */
2142static inline bool should_suppress_show_mem(void)
2143{
2144 bool ret = false;
2145
2146#if NODES_SHIFT > 8
2147 ret = in_interrupt();
2148#endif
2149 return ret;
2150}
2151
Dave Hansena238ab52011-05-24 17:12:16 -07002152static DEFINE_RATELIMIT_STATE(nopage_rs,
2153 DEFAULT_RATELIMIT_INTERVAL,
2154 DEFAULT_RATELIMIT_BURST);
2155
2156void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
2157{
Dave Hansena238ab52011-05-24 17:12:16 -07002158 unsigned int filter = SHOW_MEM_FILTER_NODES;
2159
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -08002160 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2161 debug_guardpage_minorder() > 0)
Dave Hansena238ab52011-05-24 17:12:16 -07002162 return;
2163
2164 /*
2165 * This documents exceptions given to allocations in certain
2166 * contexts that are allowed to allocate outside current's set
2167 * of allowed nodes.
2168 */
2169 if (!(gfp_mask & __GFP_NOMEMALLOC))
2170 if (test_thread_flag(TIF_MEMDIE) ||
2171 (current->flags & (PF_MEMALLOC | PF_EXITING)))
2172 filter &= ~SHOW_MEM_FILTER_NODES;
2173 if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
2174 filter &= ~SHOW_MEM_FILTER_NODES;
2175
2176 if (fmt) {
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002177 struct va_format vaf;
2178 va_list args;
2179
Dave Hansena238ab52011-05-24 17:12:16 -07002180 va_start(args, fmt);
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002181
2182 vaf.fmt = fmt;
2183 vaf.va = &args;
2184
2185 pr_warn("%pV", &vaf);
2186
Dave Hansena238ab52011-05-24 17:12:16 -07002187 va_end(args);
2188 }
2189
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002190 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2191 current->comm, order, gfp_mask);
Dave Hansena238ab52011-05-24 17:12:16 -07002192
2193 dump_stack();
2194 if (!should_suppress_show_mem())
2195 show_mem(filter);
2196}
2197
Mel Gorman11e33f62009-06-16 15:31:57 -07002198static inline int
2199should_alloc_retry(gfp_t gfp_mask, unsigned int order,
Mel Gormanf90ac392012-01-10 15:07:15 -08002200 unsigned long did_some_progress,
Mel Gorman11e33f62009-06-16 15:31:57 -07002201 unsigned long pages_reclaimed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202{
Mel Gorman11e33f62009-06-16 15:31:57 -07002203 /* Do not loop if specifically requested */
2204 if (gfp_mask & __GFP_NORETRY)
2205 return 0;
2206
Mel Gormanf90ac392012-01-10 15:07:15 -08002207 /* Always retry if specifically requested */
2208 if (gfp_mask & __GFP_NOFAIL)
2209 return 1;
2210
2211 /*
2212 * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim
2213 * making forward progress without invoking OOM. Suspend also disables
2214 * storage devices so kswapd will not help. Bail if we are suspending.
2215 */
2216 if (!did_some_progress && pm_suspended_storage())
2217 return 0;
2218
Mel Gorman11e33f62009-06-16 15:31:57 -07002219 /*
2220 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
2221 * means __GFP_NOFAIL, but that may not be true in other
2222 * implementations.
2223 */
2224 if (order <= PAGE_ALLOC_COSTLY_ORDER)
2225 return 1;
2226
2227 /*
2228 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
2229 * specified, then we retry until we no longer reclaim any pages
2230 * (above), or we've reclaimed an order of pages at least as
2231 * large as the allocation's order. In both cases, if the
2232 * allocation still fails, we stop retrying.
2233 */
2234 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
2235 return 1;
2236
Mel Gorman11e33f62009-06-16 15:31:57 -07002237 return 0;
2238}
2239
2240static inline struct page *
2241__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
2242 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002243 nodemask_t *nodemask, struct zone *preferred_zone,
Mel Gormand8846372014-06-04 16:10:33 -07002244 int classzone_idx, int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002245{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247
David Rientjese972a072014-08-06 16:07:52 -07002248 /* Acquire the per-zone oom lock for each zone */
2249 if (!oom_zonelist_trylock(zonelist, gfp_mask)) {
Mel Gorman11e33f62009-06-16 15:31:57 -07002250 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251 return NULL;
2252 }
Jens Axboe6b1de912005-11-17 21:35:02 +01002253
Mel Gorman11e33f62009-06-16 15:31:57 -07002254 /*
2255 * Go through the zonelist yet one more time, keep very high watermark
2256 * here, this is only to catch a parallel oom killing, we must fail if
2257 * we're still under heavy pressure.
2258 */
2259 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
2260 order, zonelist, high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07002261 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
Mel Gormand8846372014-06-04 16:10:33 -07002262 preferred_zone, classzone_idx, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002263 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07002264 goto out;
2265
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002266 if (!(gfp_mask & __GFP_NOFAIL)) {
2267 /* The OOM killer will not help higher order allocs */
2268 if (order > PAGE_ALLOC_COSTLY_ORDER)
2269 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07002270 /* The OOM killer does not needlessly kill tasks for lowmem */
2271 if (high_zoneidx < ZONE_NORMAL)
2272 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002273 /*
2274 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
2275 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
2276 * The caller should handle page allocation failure by itself if
2277 * it specifies __GFP_THISNODE.
2278 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
2279 */
2280 if (gfp_mask & __GFP_THISNODE)
2281 goto out;
2282 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002283 /* Exhausted what can be done so it's blamo time */
David Rientjes08ab9b12012-03-21 16:34:04 -07002284 out_of_memory(zonelist, gfp_mask, order, nodemask, false);
Mel Gorman11e33f62009-06-16 15:31:57 -07002285
2286out:
David Rientjese972a072014-08-06 16:07:52 -07002287 oom_zonelist_unlock(zonelist, gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002288 return page;
2289}
2290
Mel Gorman56de7262010-05-24 14:32:30 -07002291#ifdef CONFIG_COMPACTION
2292/* Try memory compaction for high-order allocations before reclaim */
2293static struct page *
2294__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2295 struct zonelist *zonelist, enum zone_type high_zoneidx,
2296 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gormand8846372014-06-04 16:10:33 -07002297 int classzone_idx, int migratetype, enum migrate_mode mode,
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07002298 int *contended_compaction, bool *deferred_compaction)
Mel Gorman56de7262010-05-24 14:32:30 -07002299{
Vlastimil Babka53853e22014-10-09 15:27:02 -07002300 struct zone *last_compact_zone = NULL;
2301 unsigned long compact_result;
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002302 struct page *page;
Vlastimil Babka53853e22014-10-09 15:27:02 -07002303
Mel Gorman66199712012-01-12 17:19:41 -08002304 if (!order)
Mel Gorman56de7262010-05-24 14:32:30 -07002305 return NULL;
2306
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002307 current->flags |= PF_MEMALLOC;
Vlastimil Babka53853e22014-10-09 15:27:02 -07002308 compact_result = try_to_compact_pages(zonelist, order, gfp_mask,
David Rientjese0b9dae2014-06-04 16:08:28 -07002309 nodemask, mode,
Vlastimil Babka53853e22014-10-09 15:27:02 -07002310 contended_compaction,
2311 &last_compact_zone);
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002312 current->flags &= ~PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07002313
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002314 switch (compact_result) {
2315 case COMPACT_DEFERRED:
Vlastimil Babka53853e22014-10-09 15:27:02 -07002316 *deferred_compaction = true;
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002317 /* fall-through */
2318 case COMPACT_SKIPPED:
2319 return NULL;
2320 default:
2321 break;
Mel Gorman56de7262010-05-24 14:32:30 -07002322 }
2323
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002324 /*
2325 * At least in one zone compaction wasn't deferred or skipped, so let's
2326 * count a compaction stall
2327 */
2328 count_vm_event(COMPACTSTALL);
2329
2330 /* Page migration frees to the PCP lists but we want merging */
2331 drain_pages(get_cpu());
2332 put_cpu();
2333
2334 page = get_page_from_freelist(gfp_mask, nodemask,
2335 order, zonelist, high_zoneidx,
2336 alloc_flags & ~ALLOC_NO_WATERMARKS,
2337 preferred_zone, classzone_idx, migratetype);
2338
2339 if (page) {
2340 struct zone *zone = page_zone(page);
2341
2342 zone->compact_blockskip_flush = false;
2343 compaction_defer_reset(zone, order, true);
2344 count_vm_event(COMPACTSUCCESS);
2345 return page;
2346 }
2347
2348 /*
2349 * last_compact_zone is where try_to_compact_pages thought allocation
2350 * should succeed, so it did not defer compaction. But here we know
2351 * that it didn't succeed, so we do the defer.
2352 */
2353 if (last_compact_zone && mode != MIGRATE_ASYNC)
2354 defer_compaction(last_compact_zone, order);
2355
2356 /*
2357 * It's bad if compaction run occurs and fails. The most likely reason
2358 * is that pages exist, but not enough to satisfy watermarks.
2359 */
2360 count_vm_event(COMPACTFAIL);
2361
2362 cond_resched();
2363
Mel Gorman56de7262010-05-24 14:32:30 -07002364 return NULL;
2365}
2366#else
2367static inline struct page *
2368__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2369 struct zonelist *zonelist, enum zone_type high_zoneidx,
2370 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Vlastimil Babka53853e22014-10-09 15:27:02 -07002371 int classzone_idx, int migratetype, enum migrate_mode mode,
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07002372 int *contended_compaction, bool *deferred_compaction)
Mel Gorman56de7262010-05-24 14:32:30 -07002373{
2374 return NULL;
2375}
2376#endif /* CONFIG_COMPACTION */
2377
Marek Szyprowskibba90712012-01-25 12:09:52 +01002378/* Perform direct synchronous page reclaim */
2379static int
2380__perform_reclaim(gfp_t gfp_mask, unsigned int order, struct zonelist *zonelist,
2381 nodemask_t *nodemask)
Mel Gorman11e33f62009-06-16 15:31:57 -07002382{
Mel Gorman11e33f62009-06-16 15:31:57 -07002383 struct reclaim_state reclaim_state;
Marek Szyprowskibba90712012-01-25 12:09:52 +01002384 int progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07002385
2386 cond_resched();
2387
2388 /* We now go into synchronous reclaim */
2389 cpuset_memory_pressure_bump();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002390 current->flags |= PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002391 lockdep_set_current_reclaim_state(gfp_mask);
2392 reclaim_state.reclaimed_slab = 0;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002393 current->reclaim_state = &reclaim_state;
Mel Gorman11e33f62009-06-16 15:31:57 -07002394
Marek Szyprowskibba90712012-01-25 12:09:52 +01002395 progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002396
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002397 current->reclaim_state = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002398 lockdep_clear_current_reclaim_state();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002399 current->flags &= ~PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002400
2401 cond_resched();
2402
Marek Szyprowskibba90712012-01-25 12:09:52 +01002403 return progress;
2404}
2405
2406/* The really slow allocator path where we enter direct reclaim */
2407static inline struct page *
2408__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
2409 struct zonelist *zonelist, enum zone_type high_zoneidx,
2410 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gormand8846372014-06-04 16:10:33 -07002411 int classzone_idx, int migratetype, unsigned long *did_some_progress)
Marek Szyprowskibba90712012-01-25 12:09:52 +01002412{
2413 struct page *page = NULL;
2414 bool drained = false;
2415
2416 *did_some_progress = __perform_reclaim(gfp_mask, order, zonelist,
2417 nodemask);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002418 if (unlikely(!(*did_some_progress)))
2419 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002420
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002421 /* After successful reclaim, reconsider all zones for allocation */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002422 if (IS_ENABLED(CONFIG_NUMA))
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002423 zlc_clear_zones_full(zonelist);
2424
Mel Gorman9ee493c2010-09-09 16:38:18 -07002425retry:
2426 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002427 zonelist, high_zoneidx,
Mel Gormancfd19c52012-07-31 16:44:10 -07002428 alloc_flags & ~ALLOC_NO_WATERMARKS,
Mel Gormand8846372014-06-04 16:10:33 -07002429 preferred_zone, classzone_idx,
2430 migratetype);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002431
2432 /*
2433 * If an allocation failed after direct reclaim, it could be because
2434 * pages are pinned on the per-cpu lists. Drain them and try again
2435 */
2436 if (!page && !drained) {
2437 drain_all_pages();
2438 drained = true;
2439 goto retry;
2440 }
2441
Mel Gorman11e33f62009-06-16 15:31:57 -07002442 return page;
2443}
2444
Mel Gorman11e33f62009-06-16 15:31:57 -07002445/*
2446 * This is called in the allocator slow-path if the allocation request is of
2447 * sufficient urgency to ignore watermarks and take other desperate measures
2448 */
2449static inline struct page *
2450__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
2451 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002452 nodemask_t *nodemask, struct zone *preferred_zone,
Mel Gormand8846372014-06-04 16:10:33 -07002453 int classzone_idx, int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002454{
2455 struct page *page;
2456
2457 do {
2458 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002459 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
Mel Gormand8846372014-06-04 16:10:33 -07002460 preferred_zone, classzone_idx, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002461
2462 if (!page && gfp_mask & __GFP_NOFAIL)
Mel Gorman0e093d992010-10-26 14:21:45 -07002463 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07002464 } while (!page && (gfp_mask & __GFP_NOFAIL));
2465
2466 return page;
2467}
2468
Johannes Weiner3a025762014-04-07 15:37:48 -07002469static void wake_all_kswapds(unsigned int order,
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002470 struct zonelist *zonelist,
2471 enum zone_type high_zoneidx,
Weijie Yang7ade3c92014-10-09 15:28:12 -07002472 struct zone *preferred_zone,
2473 nodemask_t *nodemask)
Mel Gorman11e33f62009-06-16 15:31:57 -07002474{
2475 struct zoneref *z;
2476 struct zone *zone;
2477
Weijie Yang7ade3c92014-10-09 15:28:12 -07002478 for_each_zone_zonelist_nodemask(zone, z, zonelist,
2479 high_zoneidx, nodemask)
Johannes Weiner3a025762014-04-07 15:37:48 -07002480 wakeup_kswapd(zone, order, zone_idx(preferred_zone));
Mel Gorman11e33f62009-06-16 15:31:57 -07002481}
2482
Peter Zijlstra341ce062009-06-16 15:32:02 -07002483static inline int
2484gfp_to_alloc_flags(gfp_t gfp_mask)
2485{
Peter Zijlstra341ce062009-06-16 15:32:02 -07002486 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
David Rientjesb104a352014-07-30 16:08:24 -07002487 const bool atomic = !(gfp_mask & (__GFP_WAIT | __GFP_NO_KSWAPD));
Peter Zijlstra341ce062009-06-16 15:32:02 -07002488
Mel Gormana56f57f2009-06-16 15:32:02 -07002489 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07002490 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07002491
Peter Zijlstra341ce062009-06-16 15:32:02 -07002492 /*
2493 * The caller may dip into page reserves a bit more if the caller
2494 * cannot run direct reclaim, or if the caller has realtime scheduling
2495 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
David Rientjesb104a352014-07-30 16:08:24 -07002496 * set both ALLOC_HARDER (atomic == true) and ALLOC_HIGH (__GFP_HIGH).
Peter Zijlstra341ce062009-06-16 15:32:02 -07002497 */
Namhyung Kime6223a32010-10-26 14:21:59 -07002498 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07002499
David Rientjesb104a352014-07-30 16:08:24 -07002500 if (atomic) {
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002501 /*
David Rientjesb104a352014-07-30 16:08:24 -07002502 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
2503 * if it can't schedule.
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002504 */
David Rientjesb104a352014-07-30 16:08:24 -07002505 if (!(gfp_mask & __GFP_NOMEMALLOC))
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002506 alloc_flags |= ALLOC_HARDER;
Peter Zijlstra341ce062009-06-16 15:32:02 -07002507 /*
David Rientjesb104a352014-07-30 16:08:24 -07002508 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
2509 * comment for __cpuset_node_allowed_softwall().
Peter Zijlstra341ce062009-06-16 15:32:02 -07002510 */
2511 alloc_flags &= ~ALLOC_CPUSET;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002512 } else if (unlikely(rt_task(current)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07002513 alloc_flags |= ALLOC_HARDER;
2514
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002515 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2516 if (gfp_mask & __GFP_MEMALLOC)
2517 alloc_flags |= ALLOC_NO_WATERMARKS;
Mel Gorman907aed42012-07-31 16:44:07 -07002518 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
2519 alloc_flags |= ALLOC_NO_WATERMARKS;
2520 else if (!in_interrupt() &&
2521 ((current->flags & PF_MEMALLOC) ||
2522 unlikely(test_thread_flag(TIF_MEMDIE))))
Peter Zijlstra341ce062009-06-16 15:32:02 -07002523 alloc_flags |= ALLOC_NO_WATERMARKS;
2524 }
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002525#ifdef CONFIG_CMA
David Rientjes43e7a342014-10-09 15:27:25 -07002526 if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002527 alloc_flags |= ALLOC_CMA;
2528#endif
Peter Zijlstra341ce062009-06-16 15:32:02 -07002529 return alloc_flags;
2530}
2531
Mel Gorman072bb0a2012-07-31 16:43:58 -07002532bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
2533{
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002534 return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
Mel Gorman072bb0a2012-07-31 16:43:58 -07002535}
2536
Mel Gorman11e33f62009-06-16 15:31:57 -07002537static inline struct page *
2538__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
2539 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002540 nodemask_t *nodemask, struct zone *preferred_zone,
Mel Gormand8846372014-06-04 16:10:33 -07002541 int classzone_idx, int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002542{
2543 const gfp_t wait = gfp_mask & __GFP_WAIT;
2544 struct page *page = NULL;
2545 int alloc_flags;
2546 unsigned long pages_reclaimed = 0;
2547 unsigned long did_some_progress;
David Rientjese0b9dae2014-06-04 16:08:28 -07002548 enum migrate_mode migration_mode = MIGRATE_ASYNC;
Mel Gorman66199712012-01-12 17:19:41 -08002549 bool deferred_compaction = false;
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07002550 int contended_compaction = COMPACT_CONTENDED_NONE;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002551
Christoph Lameter952f3b52006-12-06 20:33:26 -08002552 /*
Mel Gorman72807a72009-06-16 15:32:18 -07002553 * In the slowpath, we sanity check order to avoid ever trying to
2554 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2555 * be using allocators in order of preference for an area that is
2556 * too large.
2557 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07002558 if (order >= MAX_ORDER) {
2559 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07002560 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07002561 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002562
Christoph Lameter952f3b52006-12-06 20:33:26 -08002563 /*
2564 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2565 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2566 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2567 * using a larger set of nodes after it has established that the
2568 * allowed per node queues are empty and that nodes are
2569 * over allocated.
2570 */
Johannes Weiner3a025762014-04-07 15:37:48 -07002571 if (IS_ENABLED(CONFIG_NUMA) &&
2572 (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
Christoph Lameter952f3b52006-12-06 20:33:26 -08002573 goto nopage;
2574
Mel Gormancc4a6852009-11-11 14:26:14 -08002575restart:
Johannes Weiner3a025762014-04-07 15:37:48 -07002576 if (!(gfp_mask & __GFP_NO_KSWAPD))
Weijie Yang7ade3c92014-10-09 15:28:12 -07002577 wake_all_kswapds(order, zonelist, high_zoneidx,
2578 preferred_zone, nodemask);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002579
Paul Jackson9bf22292005-09-06 15:18:12 -07002580 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002581 * OK, we're below the kswapd watermark and have kicked background
2582 * reclaim. Now things get more complex, so set up alloc_flags according
2583 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07002584 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002585 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586
David Rientjesf33261d2011-01-25 15:07:20 -08002587 /*
2588 * Find the true preferred zone if the allocation is unconstrained by
2589 * cpusets.
2590 */
Mel Gormand8846372014-06-04 16:10:33 -07002591 if (!(alloc_flags & ALLOC_CPUSET) && !nodemask) {
2592 struct zoneref *preferred_zoneref;
2593 preferred_zoneref = first_zones_zonelist(zonelist, high_zoneidx,
2594 NULL, &preferred_zone);
2595 classzone_idx = zonelist_zone_idx(preferred_zoneref);
2596 }
David Rientjesf33261d2011-01-25 15:07:20 -08002597
Andrew Barrycfa54a02011-05-24 17:12:52 -07002598rebalance:
Peter Zijlstra341ce062009-06-16 15:32:02 -07002599 /* This is the last chance, in general, before the goto nopage. */
Mel Gorman19770b32008-04-28 02:12:18 -07002600 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
Peter Zijlstra341ce062009-06-16 15:32:02 -07002601 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
Mel Gormand8846372014-06-04 16:10:33 -07002602 preferred_zone, classzone_idx, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002603 if (page)
2604 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605
Mel Gorman11e33f62009-06-16 15:31:57 -07002606 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002607 if (alloc_flags & ALLOC_NO_WATERMARKS) {
Mel Gorman183f6372012-07-31 16:44:12 -07002608 /*
2609 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
2610 * the allocation is high priority and these type of
2611 * allocations are system rather than user orientated
2612 */
2613 zonelist = node_zonelist(numa_node_id(), gfp_mask);
2614
Peter Zijlstra341ce062009-06-16 15:32:02 -07002615 page = __alloc_pages_high_priority(gfp_mask, order,
2616 zonelist, high_zoneidx, nodemask,
Mel Gormand8846372014-06-04 16:10:33 -07002617 preferred_zone, classzone_idx, migratetype);
Mel Gormancfd19c52012-07-31 16:44:10 -07002618 if (page) {
Peter Zijlstra341ce062009-06-16 15:32:02 -07002619 goto got_pg;
Mel Gormancfd19c52012-07-31 16:44:10 -07002620 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621 }
2622
2623 /* Atomic allocations - we can't balance anything */
David Rientjesaed0a0e2014-01-21 15:51:12 -08002624 if (!wait) {
2625 /*
2626 * All existing users of the deprecated __GFP_NOFAIL are
2627 * blockable, so warn of any new users that actually allow this
2628 * type of allocation to fail.
2629 */
2630 WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631 goto nopage;
David Rientjesaed0a0e2014-01-21 15:51:12 -08002632 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633
Peter Zijlstra341ce062009-06-16 15:32:02 -07002634 /* Avoid recursion of direct reclaim */
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002635 if (current->flags & PF_MEMALLOC)
Peter Zijlstra341ce062009-06-16 15:32:02 -07002636 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637
David Rientjes6583bb62009-07-29 15:02:06 -07002638 /* Avoid allocations with no watermarks from looping endlessly */
2639 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2640 goto nopage;
2641
Mel Gorman77f1fe62011-01-13 15:45:57 -08002642 /*
2643 * Try direct compaction. The first pass is asynchronous. Subsequent
2644 * attempts after direct reclaim are synchronous
2645 */
David Rientjese0b9dae2014-06-04 16:08:28 -07002646 page = __alloc_pages_direct_compact(gfp_mask, order, zonelist,
2647 high_zoneidx, nodemask, alloc_flags,
Mel Gormand8846372014-06-04 16:10:33 -07002648 preferred_zone,
2649 classzone_idx, migratetype,
David Rientjese0b9dae2014-06-04 16:08:28 -07002650 migration_mode, &contended_compaction,
Vlastimil Babka53853e22014-10-09 15:27:02 -07002651 &deferred_compaction);
Mel Gorman56de7262010-05-24 14:32:30 -07002652 if (page)
2653 goto got_pg;
David Rientjes75f30862014-06-04 16:08:30 -07002654
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07002655 /* Checks for THP-specific high-order allocations */
2656 if ((gfp_mask & GFP_TRANSHUGE) == GFP_TRANSHUGE) {
2657 /*
2658 * If compaction is deferred for high-order allocations, it is
2659 * because sync compaction recently failed. If this is the case
2660 * and the caller requested a THP allocation, we do not want
2661 * to heavily disrupt the system, so we fail the allocation
2662 * instead of entering direct reclaim.
2663 */
2664 if (deferred_compaction)
2665 goto nopage;
2666
2667 /*
2668 * In all zones where compaction was attempted (and not
2669 * deferred or skipped), lock contention has been detected.
2670 * For THP allocation we do not want to disrupt the others
2671 * so we fallback to base pages instead.
2672 */
2673 if (contended_compaction == COMPACT_CONTENDED_LOCK)
2674 goto nopage;
2675
2676 /*
2677 * If compaction was aborted due to need_resched(), we do not
2678 * want to further increase allocation latency, unless it is
2679 * khugepaged trying to collapse.
2680 */
2681 if (contended_compaction == COMPACT_CONTENDED_SCHED
2682 && !(current->flags & PF_KTHREAD))
2683 goto nopage;
2684 }
Mel Gorman66199712012-01-12 17:19:41 -08002685
David Rientjes8fe78042014-08-06 16:07:54 -07002686 /*
2687 * It can become very expensive to allocate transparent hugepages at
2688 * fault, so use asynchronous memory compaction for THP unless it is
2689 * khugepaged trying to collapse.
2690 */
2691 if ((gfp_mask & GFP_TRANSHUGE) != GFP_TRANSHUGE ||
2692 (current->flags & PF_KTHREAD))
2693 migration_mode = MIGRATE_SYNC_LIGHT;
2694
Mel Gorman11e33f62009-06-16 15:31:57 -07002695 /* Try direct reclaim and then allocating */
2696 page = __alloc_pages_direct_reclaim(gfp_mask, order,
2697 zonelist, high_zoneidx,
2698 nodemask,
Mel Gorman5117f452009-06-16 15:31:59 -07002699 alloc_flags, preferred_zone,
Mel Gormand8846372014-06-04 16:10:33 -07002700 classzone_idx, migratetype,
2701 &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07002702 if (page)
2703 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705 /*
Mel Gorman11e33f62009-06-16 15:31:57 -07002706 * If we failed to make any progress reclaiming, then we are
2707 * running out of options and have to consider going OOM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708 */
Mel Gorman11e33f62009-06-16 15:31:57 -07002709 if (!did_some_progress) {
Qiang Huangb9921ec2013-11-12 15:07:22 -08002710 if (oom_gfp_allowed(gfp_mask)) {
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -07002711 if (oom_killer_disabled)
2712 goto nopage;
David Rientjes29fd66d2012-03-28 14:42:41 -07002713 /* Coredumps can quickly deplete all memory reserves */
2714 if ((current->flags & PF_DUMPCORE) &&
2715 !(gfp_mask & __GFP_NOFAIL))
2716 goto nopage;
Mel Gorman11e33f62009-06-16 15:31:57 -07002717 page = __alloc_pages_may_oom(gfp_mask, order,
2718 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002719 nodemask, preferred_zone,
Mel Gormand8846372014-06-04 16:10:33 -07002720 classzone_idx, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002721 if (page)
2722 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723
David Rientjes03668b32010-08-09 17:18:54 -07002724 if (!(gfp_mask & __GFP_NOFAIL)) {
2725 /*
2726 * The oom killer is not called for high-order
2727 * allocations that may fail, so if no progress
2728 * is being made, there are no other options and
2729 * retrying is unlikely to help.
2730 */
2731 if (order > PAGE_ALLOC_COSTLY_ORDER)
2732 goto nopage;
2733 /*
2734 * The oom killer is not called for lowmem
2735 * allocations to prevent needlessly killing
2736 * innocent tasks.
2737 */
2738 if (high_zoneidx < ZONE_NORMAL)
2739 goto nopage;
2740 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002741
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742 goto restart;
2743 }
2744 }
2745
Mel Gorman11e33f62009-06-16 15:31:57 -07002746 /* Check if we should retry the allocation */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07002747 pages_reclaimed += did_some_progress;
Mel Gormanf90ac392012-01-10 15:07:15 -08002748 if (should_alloc_retry(gfp_mask, order, did_some_progress,
2749 pages_reclaimed)) {
Mel Gorman11e33f62009-06-16 15:31:57 -07002750 /* Wait for some write requests to complete then retry */
Mel Gorman0e093d992010-10-26 14:21:45 -07002751 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752 goto rebalance;
Mel Gorman3e7d3442011-01-13 15:45:56 -08002753 } else {
2754 /*
2755 * High-order allocations do not necessarily loop after
2756 * direct reclaim and reclaim/compaction depends on compaction
2757 * being called after reclaim so call directly if necessary
2758 */
David Rientjese0b9dae2014-06-04 16:08:28 -07002759 page = __alloc_pages_direct_compact(gfp_mask, order, zonelist,
2760 high_zoneidx, nodemask, alloc_flags,
Mel Gormand8846372014-06-04 16:10:33 -07002761 preferred_zone,
2762 classzone_idx, migratetype,
David Rientjese0b9dae2014-06-04 16:08:28 -07002763 migration_mode, &contended_compaction,
Vlastimil Babka53853e22014-10-09 15:27:02 -07002764 &deferred_compaction);
Mel Gorman3e7d3442011-01-13 15:45:56 -08002765 if (page)
2766 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002767 }
2768
2769nopage:
Dave Hansena238ab52011-05-24 17:12:16 -07002770 warn_alloc_failed(gfp_mask, order, NULL);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002771 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002772got_pg:
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002773 if (kmemcheck_enabled)
2774 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002775
Mel Gorman072bb0a2012-07-31 16:43:58 -07002776 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777}
Mel Gorman11e33f62009-06-16 15:31:57 -07002778
2779/*
2780 * This is the 'heart' of the zoned buddy allocator.
2781 */
2782struct page *
2783__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2784 struct zonelist *zonelist, nodemask_t *nodemask)
2785{
2786 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Mel Gorman5117f452009-06-16 15:31:59 -07002787 struct zone *preferred_zone;
Mel Gormand8846372014-06-04 16:10:33 -07002788 struct zoneref *preferred_zoneref;
Mel Gormancc9a6c82012-03-21 16:34:11 -07002789 struct page *page = NULL;
David Rientjes43e7a342014-10-09 15:27:25 -07002790 int migratetype = gfpflags_to_migratetype(gfp_mask);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002791 unsigned int cpuset_mems_cookie;
Johannes Weiner3a025762014-04-07 15:37:48 -07002792 int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET|ALLOC_FAIR;
Mel Gormand8846372014-06-04 16:10:33 -07002793 int classzone_idx;
Mel Gorman11e33f62009-06-16 15:31:57 -07002794
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10002795 gfp_mask &= gfp_allowed_mask;
2796
Mel Gorman11e33f62009-06-16 15:31:57 -07002797 lockdep_trace_alloc(gfp_mask);
2798
2799 might_sleep_if(gfp_mask & __GFP_WAIT);
2800
2801 if (should_fail_alloc_page(gfp_mask, order))
2802 return NULL;
2803
2804 /*
2805 * Check the zones suitable for the gfp_mask contain at least one
2806 * valid zone. It's possible to have an empty zonelist as a result
2807 * of GFP_THISNODE and a memoryless node
2808 */
2809 if (unlikely(!zonelist->_zonerefs->zone))
2810 return NULL;
2811
Vlastimil Babka21bb9bd2014-10-09 15:26:51 -07002812 if (IS_ENABLED(CONFIG_CMA) && migratetype == MIGRATE_MOVABLE)
2813 alloc_flags |= ALLOC_CMA;
2814
Mel Gormancc9a6c82012-03-21 16:34:11 -07002815retry_cpuset:
Mel Gormand26914d2014-04-03 14:47:24 -07002816 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07002817
Mel Gorman5117f452009-06-16 15:31:59 -07002818 /* The preferred zone is used for statistics later */
Mel Gormand8846372014-06-04 16:10:33 -07002819 preferred_zoneref = first_zones_zonelist(zonelist, high_zoneidx,
David Rientjesf33261d2011-01-25 15:07:20 -08002820 nodemask ? : &cpuset_current_mems_allowed,
2821 &preferred_zone);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002822 if (!preferred_zone)
2823 goto out;
Mel Gormand8846372014-06-04 16:10:33 -07002824 classzone_idx = zonelist_zone_idx(preferred_zoneref);
Mel Gorman5117f452009-06-16 15:31:59 -07002825
2826 /* First allocation attempt */
Mel Gorman11e33f62009-06-16 15:31:57 -07002827 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002828 zonelist, high_zoneidx, alloc_flags,
Mel Gormand8846372014-06-04 16:10:33 -07002829 preferred_zone, classzone_idx, migratetype);
Ming Lei21caf2f2013-02-22 16:34:08 -08002830 if (unlikely(!page)) {
2831 /*
2832 * Runtime PM, block IO and its error handling path
2833 * can deadlock because I/O on the device might not
2834 * complete.
2835 */
2836 gfp_mask = memalloc_noio_flags(gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002837 page = __alloc_pages_slowpath(gfp_mask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002838 zonelist, high_zoneidx, nodemask,
Mel Gormand8846372014-06-04 16:10:33 -07002839 preferred_zone, classzone_idx, migratetype);
Ming Lei21caf2f2013-02-22 16:34:08 -08002840 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002841
Mel Gorman4b4f2782009-09-21 17:02:41 -07002842 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002843
2844out:
2845 /*
2846 * When updating a task's mems_allowed, it is possible to race with
2847 * parallel threads in such a way that an allocation can fail while
2848 * the mask is being updated. If a page allocation is about to fail,
2849 * check if the cpuset changed during allocation and if so, retry.
2850 */
Mel Gormand26914d2014-04-03 14:47:24 -07002851 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
Mel Gormancc9a6c82012-03-21 16:34:11 -07002852 goto retry_cpuset;
2853
Mel Gorman11e33f62009-06-16 15:31:57 -07002854 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855}
Mel Gormand2391712009-06-16 15:31:52 -07002856EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857
2858/*
2859 * Common helper functions.
2860 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08002861unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862{
Akinobu Mita945a1112009-09-21 17:01:47 -07002863 struct page *page;
2864
2865 /*
2866 * __get_free_pages() returns a 32-bit address, which cannot represent
2867 * a highmem page
2868 */
2869 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2870
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871 page = alloc_pages(gfp_mask, order);
2872 if (!page)
2873 return 0;
2874 return (unsigned long) page_address(page);
2875}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876EXPORT_SYMBOL(__get_free_pages);
2877
Harvey Harrison920c7a52008-02-04 22:29:26 -08002878unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879{
Akinobu Mita945a1112009-09-21 17:01:47 -07002880 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882EXPORT_SYMBOL(get_zeroed_page);
2883
Harvey Harrison920c7a52008-02-04 22:29:26 -08002884void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885{
Nick Pigginb5810032005-10-29 18:16:12 -07002886 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002887 if (order == 0)
Mel Gormanb745bc82014-06-04 16:10:22 -07002888 free_hot_cold_page(page, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889 else
2890 __free_pages_ok(page, order);
2891 }
2892}
2893
2894EXPORT_SYMBOL(__free_pages);
2895
Harvey Harrison920c7a52008-02-04 22:29:26 -08002896void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897{
2898 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07002899 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002900 __free_pages(virt_to_page((void *)addr), order);
2901 }
2902}
2903
2904EXPORT_SYMBOL(free_pages);
2905
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002906/*
Vladimir Davydov52383432014-06-04 16:06:39 -07002907 * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
2908 * of the current memory cgroup.
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002909 *
Vladimir Davydov52383432014-06-04 16:06:39 -07002910 * It should be used when the caller would like to use kmalloc, but since the
2911 * allocation is large, it has to fall back to the page allocator.
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002912 */
Vladimir Davydov52383432014-06-04 16:06:39 -07002913struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order)
2914{
2915 struct page *page;
2916 struct mem_cgroup *memcg = NULL;
2917
2918 if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
2919 return NULL;
2920 page = alloc_pages(gfp_mask, order);
2921 memcg_kmem_commit_charge(page, memcg, order);
2922 return page;
2923}
2924
2925struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
2926{
2927 struct page *page;
2928 struct mem_cgroup *memcg = NULL;
2929
2930 if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
2931 return NULL;
2932 page = alloc_pages_node(nid, gfp_mask, order);
2933 memcg_kmem_commit_charge(page, memcg, order);
2934 return page;
2935}
2936
2937/*
2938 * __free_kmem_pages and free_kmem_pages will free pages allocated with
2939 * alloc_kmem_pages.
2940 */
2941void __free_kmem_pages(struct page *page, unsigned int order)
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002942{
2943 memcg_kmem_uncharge_pages(page, order);
2944 __free_pages(page, order);
2945}
2946
Vladimir Davydov52383432014-06-04 16:06:39 -07002947void free_kmem_pages(unsigned long addr, unsigned int order)
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002948{
2949 if (addr != 0) {
2950 VM_BUG_ON(!virt_addr_valid((void *)addr));
Vladimir Davydov52383432014-06-04 16:06:39 -07002951 __free_kmem_pages(virt_to_page((void *)addr), order);
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002952 }
2953}
2954
Andi Kleenee85c2e2011-05-11 15:13:34 -07002955static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
2956{
2957 if (addr) {
2958 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2959 unsigned long used = addr + PAGE_ALIGN(size);
2960
2961 split_page(virt_to_page((void *)addr), order);
2962 while (used < alloc_end) {
2963 free_page(used);
2964 used += PAGE_SIZE;
2965 }
2966 }
2967 return (void *)addr;
2968}
2969
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002970/**
2971 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2972 * @size: the number of bytes to allocate
2973 * @gfp_mask: GFP flags for the allocation
2974 *
2975 * This function is similar to alloc_pages(), except that it allocates the
2976 * minimum number of pages to satisfy the request. alloc_pages() can only
2977 * allocate memory in power-of-two pages.
2978 *
2979 * This function is also limited by MAX_ORDER.
2980 *
2981 * Memory allocated by this function must be released by free_pages_exact().
2982 */
2983void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2984{
2985 unsigned int order = get_order(size);
2986 unsigned long addr;
2987
2988 addr = __get_free_pages(gfp_mask, order);
Andi Kleenee85c2e2011-05-11 15:13:34 -07002989 return make_alloc_exact(addr, order, size);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002990}
2991EXPORT_SYMBOL(alloc_pages_exact);
2992
2993/**
Andi Kleenee85c2e2011-05-11 15:13:34 -07002994 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2995 * pages on a node.
Randy Dunlapb5e6ab52011-05-16 13:16:54 -07002996 * @nid: the preferred node ID where memory should be allocated
Andi Kleenee85c2e2011-05-11 15:13:34 -07002997 * @size: the number of bytes to allocate
2998 * @gfp_mask: GFP flags for the allocation
2999 *
3000 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
3001 * back.
3002 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
3003 * but is not exact.
3004 */
Fabian Fredericke1931812014-08-06 16:04:59 -07003005void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
Andi Kleenee85c2e2011-05-11 15:13:34 -07003006{
3007 unsigned order = get_order(size);
3008 struct page *p = alloc_pages_node(nid, gfp_mask, order);
3009 if (!p)
3010 return NULL;
3011 return make_alloc_exact((unsigned long)page_address(p), order, size);
3012}
Andi Kleenee85c2e2011-05-11 15:13:34 -07003013
3014/**
Timur Tabi2be0ffe2008-07-23 21:28:11 -07003015 * free_pages_exact - release memory allocated via alloc_pages_exact()
3016 * @virt: the value returned by alloc_pages_exact.
3017 * @size: size of allocation, same value as passed to alloc_pages_exact().
3018 *
3019 * Release the memory allocated by a previous call to alloc_pages_exact.
3020 */
3021void free_pages_exact(void *virt, size_t size)
3022{
3023 unsigned long addr = (unsigned long)virt;
3024 unsigned long end = addr + PAGE_ALIGN(size);
3025
3026 while (addr < end) {
3027 free_page(addr);
3028 addr += PAGE_SIZE;
3029 }
3030}
3031EXPORT_SYMBOL(free_pages_exact);
3032
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003033/**
3034 * nr_free_zone_pages - count number of pages beyond high watermark
3035 * @offset: The zone index of the highest zone
3036 *
3037 * nr_free_zone_pages() counts the number of counts pages which are beyond the
3038 * high watermark within all zones at or below a given zone index. For each
3039 * zone, the number of pages is calculated as:
Jiang Liu834405c2013-07-03 15:03:04 -07003040 * managed_pages - high_pages
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003041 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003042static unsigned long nr_free_zone_pages(int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003043{
Mel Gormandd1a2392008-04-28 02:12:17 -07003044 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003045 struct zone *zone;
3046
Martin J. Blighe310fd42005-07-29 22:59:18 -07003047 /* Just pick one node, since fallback list is circular */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003048 unsigned long sum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003049
Mel Gorman0e884602008-04-28 02:12:14 -07003050 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003051
Mel Gorman54a6eb52008-04-28 02:12:16 -07003052 for_each_zone_zonelist(zone, z, zonelist, offset) {
Jiang Liub40da042013-02-22 16:33:52 -08003053 unsigned long size = zone->managed_pages;
Mel Gorman41858962009-06-16 15:32:12 -07003054 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07003055 if (size > high)
3056 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057 }
3058
3059 return sum;
3060}
3061
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003062/**
3063 * nr_free_buffer_pages - count number of pages beyond high watermark
3064 *
3065 * nr_free_buffer_pages() counts the number of pages which are beyond the high
3066 * watermark within ZONE_DMA and ZONE_NORMAL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003068unsigned long nr_free_buffer_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003069{
Al Viroaf4ca452005-10-21 02:55:38 -04003070 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071}
Meelap Shahc2f1a552007-07-17 04:04:39 -07003072EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003073
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003074/**
3075 * nr_free_pagecache_pages - count number of pages beyond high watermark
3076 *
3077 * nr_free_pagecache_pages() counts the number of pages which are beyond the
3078 * high watermark within all zones.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003080unsigned long nr_free_pagecache_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003081{
Mel Gorman2a1e2742007-07-17 04:03:12 -07003082 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003083}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07003084
3085static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003086{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08003087 if (IS_ENABLED(CONFIG_NUMA))
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08003088 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003089}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003090
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091void si_meminfo(struct sysinfo *val)
3092{
3093 val->totalram = totalram_pages;
Rafael Aquinicc7452b2014-08-06 16:06:38 -07003094 val->sharedram = global_page_state(NR_SHMEM);
Christoph Lameterd23ad422007-02-10 01:43:02 -08003095 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003096 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003097 val->totalhigh = totalhigh_pages;
3098 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003099 val->mem_unit = PAGE_SIZE;
3100}
3101
3102EXPORT_SYMBOL(si_meminfo);
3103
3104#ifdef CONFIG_NUMA
3105void si_meminfo_node(struct sysinfo *val, int nid)
3106{
Jiang Liucdd91a72013-07-03 15:03:27 -07003107 int zone_type; /* needs to be signed */
3108 unsigned long managed_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003109 pg_data_t *pgdat = NODE_DATA(nid);
3110
Jiang Liucdd91a72013-07-03 15:03:27 -07003111 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
3112 managed_pages += pgdat->node_zones[zone_type].managed_pages;
3113 val->totalram = managed_pages;
Rafael Aquinicc7452b2014-08-06 16:06:38 -07003114 val->sharedram = node_page_state(nid, NR_SHMEM);
Christoph Lameterd23ad422007-02-10 01:43:02 -08003115 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003116#ifdef CONFIG_HIGHMEM
Jiang Liub40da042013-02-22 16:33:52 -08003117 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08003118 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
3119 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003120#else
3121 val->totalhigh = 0;
3122 val->freehigh = 0;
3123#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003124 val->mem_unit = PAGE_SIZE;
3125}
3126#endif
3127
David Rientjesddd588b2011-03-22 16:30:46 -07003128/*
David Rientjes7bf02ea2011-05-24 17:11:16 -07003129 * Determine whether the node should be displayed or not, depending on whether
3130 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
David Rientjesddd588b2011-03-22 16:30:46 -07003131 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07003132bool skip_free_areas_node(unsigned int flags, int nid)
David Rientjesddd588b2011-03-22 16:30:46 -07003133{
3134 bool ret = false;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003135 unsigned int cpuset_mems_cookie;
David Rientjesddd588b2011-03-22 16:30:46 -07003136
3137 if (!(flags & SHOW_MEM_FILTER_NODES))
3138 goto out;
3139
Mel Gormancc9a6c82012-03-21 16:34:11 -07003140 do {
Mel Gormand26914d2014-04-03 14:47:24 -07003141 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07003142 ret = !node_isset(nid, cpuset_current_mems_allowed);
Mel Gormand26914d2014-04-03 14:47:24 -07003143 } while (read_mems_allowed_retry(cpuset_mems_cookie));
David Rientjesddd588b2011-03-22 16:30:46 -07003144out:
3145 return ret;
3146}
3147
Linus Torvalds1da177e2005-04-16 15:20:36 -07003148#define K(x) ((x) << (PAGE_SHIFT-10))
3149
Rabin Vincent377e4f12012-12-11 16:00:24 -08003150static void show_migration_types(unsigned char type)
3151{
3152 static const char types[MIGRATE_TYPES] = {
3153 [MIGRATE_UNMOVABLE] = 'U',
3154 [MIGRATE_RECLAIMABLE] = 'E',
3155 [MIGRATE_MOVABLE] = 'M',
3156 [MIGRATE_RESERVE] = 'R',
3157#ifdef CONFIG_CMA
3158 [MIGRATE_CMA] = 'C',
3159#endif
Minchan Kim194159f2013-02-22 16:33:58 -08003160#ifdef CONFIG_MEMORY_ISOLATION
Rabin Vincent377e4f12012-12-11 16:00:24 -08003161 [MIGRATE_ISOLATE] = 'I',
Minchan Kim194159f2013-02-22 16:33:58 -08003162#endif
Rabin Vincent377e4f12012-12-11 16:00:24 -08003163 };
3164 char tmp[MIGRATE_TYPES + 1];
3165 char *p = tmp;
3166 int i;
3167
3168 for (i = 0; i < MIGRATE_TYPES; i++) {
3169 if (type & (1 << i))
3170 *p++ = types[i];
3171 }
3172
3173 *p = '\0';
3174 printk("(%s) ", tmp);
3175}
3176
Linus Torvalds1da177e2005-04-16 15:20:36 -07003177/*
3178 * Show free area list (used inside shift_scroll-lock stuff)
3179 * We also calculate the percentage fragmentation. We do this by counting the
3180 * memory on each free list with the exception of the first item on the list.
David Rientjesddd588b2011-03-22 16:30:46 -07003181 * Suppresses nodes that are not allowed by current's cpuset if
3182 * SHOW_MEM_FILTER_NODES is passed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003183 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07003184void show_free_areas(unsigned int filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003185{
Jes Sorensenc7241912006-09-27 01:50:05 -07003186 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003187 struct zone *zone;
3188
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003189 for_each_populated_zone(zone) {
David Rientjes7bf02ea2011-05-24 17:11:16 -07003190 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003191 continue;
Jes Sorensenc7241912006-09-27 01:50:05 -07003192 show_node(zone);
3193 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003194
Dave Jones6b482c62005-11-10 15:45:56 -05003195 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003196 struct per_cpu_pageset *pageset;
3197
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003198 pageset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003199
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003200 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
3201 cpu, pageset->pcp.high,
3202 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003203 }
3204 }
3205
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003206 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3207 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003208 " unevictable:%lu"
Andrew Mortonb76146e2009-10-26 16:49:52 -07003209 " dirty:%lu writeback:%lu unstable:%lu\n"
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07003210 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003211 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3212 " free_cma:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003213 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003214 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003215 global_page_state(NR_ISOLATED_ANON),
3216 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003217 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003218 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003219 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07003220 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07003221 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07003222 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08003223 global_page_state(NR_FREE_PAGES),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07003224 global_page_state(NR_SLAB_RECLAIMABLE),
3225 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07003226 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003227 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08003228 global_page_state(NR_PAGETABLE),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003229 global_page_state(NR_BOUNCE),
3230 global_page_state(NR_FREE_CMA_PAGES));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003231
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003232 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003233 int i;
3234
David Rientjes7bf02ea2011-05-24 17:11:16 -07003235 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003236 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003237 show_node(zone);
3238 printk("%s"
3239 " free:%lukB"
3240 " min:%lukB"
3241 " low:%lukB"
3242 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003243 " active_anon:%lukB"
3244 " inactive_anon:%lukB"
3245 " active_file:%lukB"
3246 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003247 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003248 " isolated(anon):%lukB"
3249 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003250 " present:%lukB"
Jiang Liu9feedc92012-12-12 13:52:12 -08003251 " managed:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003252 " mlocked:%lukB"
3253 " dirty:%lukB"
3254 " writeback:%lukB"
3255 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003256 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003257 " slab_reclaimable:%lukB"
3258 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003259 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003260 " pagetables:%lukB"
3261 " unstable:%lukB"
3262 " bounce:%lukB"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003263 " free_cma:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003264 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003265 " pages_scanned:%lu"
3266 " all_unreclaimable? %s"
3267 "\n",
3268 zone->name,
Mel Gorman88f5acf2011-01-13 15:45:41 -08003269 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07003270 K(min_wmark_pages(zone)),
3271 K(low_wmark_pages(zone)),
3272 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003273 K(zone_page_state(zone, NR_ACTIVE_ANON)),
3274 K(zone_page_state(zone, NR_INACTIVE_ANON)),
3275 K(zone_page_state(zone, NR_ACTIVE_FILE)),
3276 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003277 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003278 K(zone_page_state(zone, NR_ISOLATED_ANON)),
3279 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003280 K(zone->present_pages),
Jiang Liu9feedc92012-12-12 13:52:12 -08003281 K(zone->managed_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003282 K(zone_page_state(zone, NR_MLOCK)),
3283 K(zone_page_state(zone, NR_FILE_DIRTY)),
3284 K(zone_page_state(zone, NR_WRITEBACK)),
3285 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003286 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003287 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3288 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003289 zone_page_state(zone, NR_KERNEL_STACK) *
3290 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003291 K(zone_page_state(zone, NR_PAGETABLE)),
3292 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3293 K(zone_page_state(zone, NR_BOUNCE)),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003294 K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003295 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Mel Gorman0d5d8232014-08-06 16:07:16 -07003296 K(zone_page_state(zone, NR_PAGES_SCANNED)),
Lisa Du6e543d52013-09-11 14:22:36 -07003297 (!zone_reclaimable(zone) ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07003298 );
3299 printk("lowmem_reserve[]:");
3300 for (i = 0; i < MAX_NR_ZONES; i++)
Mel Gorman3484b2d2014-08-06 16:07:14 -07003301 printk(" %ld", zone->lowmem_reserve[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003302 printk("\n");
3303 }
3304
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003305 for_each_populated_zone(zone) {
Pintu Kumarb8af2942013-09-11 14:20:34 -07003306 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003307 unsigned char types[MAX_ORDER];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003308
David Rientjes7bf02ea2011-05-24 17:11:16 -07003309 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003310 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003311 show_node(zone);
3312 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003313
3314 spin_lock_irqsave(&zone->lock, flags);
3315 for (order = 0; order < MAX_ORDER; order++) {
Rabin Vincent377e4f12012-12-11 16:00:24 -08003316 struct free_area *area = &zone->free_area[order];
3317 int type;
3318
3319 nr[order] = area->nr_free;
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003320 total += nr[order] << order;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003321
3322 types[order] = 0;
3323 for (type = 0; type < MIGRATE_TYPES; type++) {
3324 if (!list_empty(&area->free_list[type]))
3325 types[order] |= 1 << type;
3326 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003327 }
3328 spin_unlock_irqrestore(&zone->lock, flags);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003329 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003330 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003331 if (nr[order])
3332 show_migration_types(types[order]);
3333 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003334 printk("= %lukB\n", K(total));
3335 }
3336
David Rientjes949f7ec2013-04-29 15:07:48 -07003337 hugetlb_show_meminfo();
3338
Larry Woodmane6f36022008-02-04 22:29:30 -08003339 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
3340
Linus Torvalds1da177e2005-04-16 15:20:36 -07003341 show_swap_cache_info();
3342}
3343
Mel Gorman19770b32008-04-28 02:12:18 -07003344static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
3345{
3346 zoneref->zone = zone;
3347 zoneref->zone_idx = zone_idx(zone);
3348}
3349
Linus Torvalds1da177e2005-04-16 15:20:36 -07003350/*
3351 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08003352 *
3353 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003354 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003355static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003356 int nr_zones)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003357{
Christoph Lameter1a932052006-01-06 00:11:16 -08003358 struct zone *zone;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003359 enum zone_type zone_type = MAX_NR_ZONES;
Christoph Lameter02a68a52006-01-06 00:11:18 -08003360
3361 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003362 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08003363 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08003364 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07003365 zoneref_set_zone(zone,
3366 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08003367 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003368 }
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003369 } while (zone_type);
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003370
Christoph Lameter070f8032006-01-06 00:11:19 -08003371 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003372}
3373
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003374
3375/*
3376 * zonelist_order:
3377 * 0 = automatic detection of better ordering.
3378 * 1 = order by ([node] distance, -zonetype)
3379 * 2 = order by (-zonetype, [node] distance)
3380 *
3381 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3382 * the same zonelist. So only NUMA can configure this param.
3383 */
3384#define ZONELIST_ORDER_DEFAULT 0
3385#define ZONELIST_ORDER_NODE 1
3386#define ZONELIST_ORDER_ZONE 2
3387
3388/* zonelist order in the kernel.
3389 * set_zonelist_order() will set this to NODE or ZONE.
3390 */
3391static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
3392static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
3393
3394
Linus Torvalds1da177e2005-04-16 15:20:36 -07003395#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003396/* The value user specified ....changed by config */
3397static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3398/* string for sysctl */
3399#define NUMA_ZONELIST_ORDER_LEN 16
3400char numa_zonelist_order[16] = "default";
3401
3402/*
3403 * interface for configure zonelist ordering.
3404 * command line option "numa_zonelist_order"
3405 * = "[dD]efault - default, automatic configuration.
3406 * = "[nN]ode - order by node locality, then by zone within node
3407 * = "[zZ]one - order by zone, then by locality within zone
3408 */
3409
3410static int __parse_numa_zonelist_order(char *s)
3411{
3412 if (*s == 'd' || *s == 'D') {
3413 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3414 } else if (*s == 'n' || *s == 'N') {
3415 user_zonelist_order = ZONELIST_ORDER_NODE;
3416 } else if (*s == 'z' || *s == 'Z') {
3417 user_zonelist_order = ZONELIST_ORDER_ZONE;
3418 } else {
3419 printk(KERN_WARNING
3420 "Ignoring invalid numa_zonelist_order value: "
3421 "%s\n", s);
3422 return -EINVAL;
3423 }
3424 return 0;
3425}
3426
3427static __init int setup_numa_zonelist_order(char *s)
3428{
Volodymyr G. Lukiianykecb256f2011-01-13 15:46:26 -08003429 int ret;
3430
3431 if (!s)
3432 return 0;
3433
3434 ret = __parse_numa_zonelist_order(s);
3435 if (ret == 0)
3436 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
3437
3438 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003439}
3440early_param("numa_zonelist_order", setup_numa_zonelist_order);
3441
3442/*
3443 * sysctl handler for numa_zonelist_order
3444 */
Joe Perchescccad5b2014-06-06 14:38:09 -07003445int numa_zonelist_order_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003446 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003447 loff_t *ppos)
3448{
3449 char saved_string[NUMA_ZONELIST_ORDER_LEN];
3450 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01003451 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003452
Andi Kleen443c6f12009-12-23 21:00:47 +01003453 mutex_lock(&zl_order_mutex);
Chen Gangdacbde02013-07-03 15:02:35 -07003454 if (write) {
3455 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
3456 ret = -EINVAL;
3457 goto out;
3458 }
3459 strcpy(saved_string, (char *)table->data);
3460 }
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003461 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003462 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01003463 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003464 if (write) {
3465 int oldval = user_zonelist_order;
Chen Gangdacbde02013-07-03 15:02:35 -07003466
3467 ret = __parse_numa_zonelist_order((char *)table->data);
3468 if (ret) {
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003469 /*
3470 * bogus value. restore saved string
3471 */
Chen Gangdacbde02013-07-03 15:02:35 -07003472 strncpy((char *)table->data, saved_string,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003473 NUMA_ZONELIST_ORDER_LEN);
3474 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003475 } else if (oldval != user_zonelist_order) {
3476 mutex_lock(&zonelists_mutex);
Jiang Liu9adb62a2012-07-31 16:43:28 -07003477 build_all_zonelists(NULL, NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003478 mutex_unlock(&zonelists_mutex);
3479 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003480 }
Andi Kleen443c6f12009-12-23 21:00:47 +01003481out:
3482 mutex_unlock(&zl_order_mutex);
3483 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003484}
3485
3486
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003487#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003488static int node_load[MAX_NUMNODES];
3489
Linus Torvalds1da177e2005-04-16 15:20:36 -07003490/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07003491 * 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 -07003492 * @node: node whose fallback list we're appending
3493 * @used_node_mask: nodemask_t of already used nodes
3494 *
3495 * We use a number of factors to determine which is the next node that should
3496 * appear on a given node's fallback list. The node should not have appeared
3497 * already in @node's fallback list, and it should be the next closest node
3498 * according to the distance array (which contains arbitrary distance values
3499 * from each node to each node in the system), and should also prefer nodes
3500 * with no CPUs, since presumably they'll have very little allocation pressure
3501 * on them otherwise.
3502 * It returns -1 if no node is found.
3503 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003504static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003505{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01003506 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003507 int min_val = INT_MAX;
David Rientjes00ef2d22013-02-22 16:35:36 -08003508 int best_node = NUMA_NO_NODE;
Rusty Russella70f7302009-03-13 14:49:46 +10303509 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003510
Linus Torvalds4cf808e2006-02-17 20:38:21 +01003511 /* Use the local node if we haven't already */
3512 if (!node_isset(node, *used_node_mask)) {
3513 node_set(node, *used_node_mask);
3514 return node;
3515 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003516
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08003517 for_each_node_state(n, N_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003518
3519 /* Don't want a node to appear more than once */
3520 if (node_isset(n, *used_node_mask))
3521 continue;
3522
Linus Torvalds1da177e2005-04-16 15:20:36 -07003523 /* Use the distance array to find the distance */
3524 val = node_distance(node, n);
3525
Linus Torvalds4cf808e2006-02-17 20:38:21 +01003526 /* Penalize nodes under us ("prefer the next node") */
3527 val += (n < node);
3528
Linus Torvalds1da177e2005-04-16 15:20:36 -07003529 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10303530 tmp = cpumask_of_node(n);
3531 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003532 val += PENALTY_FOR_NODE_WITH_CPUS;
3533
3534 /* Slight preference for less loaded node */
3535 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
3536 val += node_load[n];
3537
3538 if (val < min_val) {
3539 min_val = val;
3540 best_node = n;
3541 }
3542 }
3543
3544 if (best_node >= 0)
3545 node_set(best_node, *used_node_mask);
3546
3547 return best_node;
3548}
3549
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003550
3551/*
3552 * Build zonelists ordered by node and zones within node.
3553 * This results in maximum locality--normal zone overflows into local
3554 * DMA zone, if any--but risks exhausting DMA zone.
3555 */
3556static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003557{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003558 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003559 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003560
Mel Gorman54a6eb52008-04-28 02:12:16 -07003561 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07003562 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07003563 ;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003564 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gormandd1a2392008-04-28 02:12:17 -07003565 zonelist->_zonerefs[j].zone = NULL;
3566 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003567}
3568
3569/*
Christoph Lameter523b9452007-10-16 01:25:37 -07003570 * Build gfp_thisnode zonelists
3571 */
3572static void build_thisnode_zonelists(pg_data_t *pgdat)
3573{
Christoph Lameter523b9452007-10-16 01:25:37 -07003574 int j;
3575 struct zonelist *zonelist;
3576
Mel Gorman54a6eb52008-04-28 02:12:16 -07003577 zonelist = &pgdat->node_zonelists[1];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003578 j = build_zonelists_node(pgdat, zonelist, 0);
Mel Gormandd1a2392008-04-28 02:12:17 -07003579 zonelist->_zonerefs[j].zone = NULL;
3580 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07003581}
3582
3583/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003584 * Build zonelists ordered by zone and nodes within zones.
3585 * This results in conserving DMA zone[s] until all Normal memory is
3586 * exhausted, but results in overflowing to remote node while memory
3587 * may still exist in local DMA zone.
3588 */
3589static int node_order[MAX_NUMNODES];
3590
3591static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
3592{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003593 int pos, j, node;
3594 int zone_type; /* needs to be signed */
3595 struct zone *z;
3596 struct zonelist *zonelist;
3597
Mel Gorman54a6eb52008-04-28 02:12:16 -07003598 zonelist = &pgdat->node_zonelists[0];
3599 pos = 0;
3600 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
3601 for (j = 0; j < nr_nodes; j++) {
3602 node = node_order[j];
3603 z = &NODE_DATA(node)->node_zones[zone_type];
3604 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07003605 zoneref_set_zone(z,
3606 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07003607 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003608 }
3609 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003610 }
Mel Gormandd1a2392008-04-28 02:12:17 -07003611 zonelist->_zonerefs[pos].zone = NULL;
3612 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003613}
3614
Mel Gorman31939132014-10-09 15:28:30 -07003615#if defined(CONFIG_64BIT)
3616/*
3617 * Devices that require DMA32/DMA are relatively rare and do not justify a
3618 * penalty to every machine in case the specialised case applies. Default
3619 * to Node-ordering on 64-bit NUMA machines
3620 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003621static int default_zonelist_order(void)
3622{
Mel Gorman31939132014-10-09 15:28:30 -07003623 return ZONELIST_ORDER_NODE;
3624}
3625#else
3626/*
3627 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
3628 * by the kernel. If processes running on node 0 deplete the low memory zone
3629 * then reclaim will occur more frequency increasing stalls and potentially
3630 * be easier to OOM if a large percentage of the zone is under writeback or
3631 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
3632 * Hence, default to zone ordering on 32-bit.
3633 */
3634static int default_zonelist_order(void)
3635{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003636 return ZONELIST_ORDER_ZONE;
3637}
Mel Gorman31939132014-10-09 15:28:30 -07003638#endif /* CONFIG_64BIT */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003639
3640static void set_zonelist_order(void)
3641{
3642 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
3643 current_zonelist_order = default_zonelist_order();
3644 else
3645 current_zonelist_order = user_zonelist_order;
3646}
3647
3648static void build_zonelists(pg_data_t *pgdat)
3649{
3650 int j, node, load;
3651 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003652 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003653 int local_node, prev_node;
3654 struct zonelist *zonelist;
3655 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003656
3657 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07003658 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003659 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07003660 zonelist->_zonerefs[0].zone = NULL;
3661 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003662 }
3663
3664 /* NUMA-aware ordering of nodes */
3665 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003666 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003667 prev_node = local_node;
3668 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003669
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003670 memset(node_order, 0, sizeof(node_order));
3671 j = 0;
3672
Linus Torvalds1da177e2005-04-16 15:20:36 -07003673 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
3674 /*
3675 * We don't want to pressure a particular node.
3676 * So adding penalty to the first node in same
3677 * distance group to make it round-robin.
3678 */
David Rientjes957f8222012-10-08 16:33:24 -07003679 if (node_distance(local_node, node) !=
3680 node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003681 node_load[node] = load;
3682
Linus Torvalds1da177e2005-04-16 15:20:36 -07003683 prev_node = node;
3684 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003685 if (order == ZONELIST_ORDER_NODE)
3686 build_zonelists_in_node_order(pgdat, node);
3687 else
3688 node_order[j++] = node; /* remember order */
3689 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003690
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003691 if (order == ZONELIST_ORDER_ZONE) {
3692 /* calculate node order -- i.e., DMA last! */
3693 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003694 }
Christoph Lameter523b9452007-10-16 01:25:37 -07003695
3696 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003697}
3698
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003699/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003700static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003701{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003702 struct zonelist *zonelist;
3703 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07003704 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003705
Mel Gorman54a6eb52008-04-28 02:12:16 -07003706 zonelist = &pgdat->node_zonelists[0];
3707 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3708 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07003709 for (z = zonelist->_zonerefs; z->zone; z++)
3710 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003711}
3712
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003713#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3714/*
3715 * Return node id of node used for "local" allocations.
3716 * I.e., first node id of first zone in arg node's generic zonelist.
3717 * Used for initializing percpu 'numa_mem', which is used primarily
3718 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3719 */
3720int local_memory_node(int node)
3721{
3722 struct zone *zone;
3723
3724 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3725 gfp_zone(GFP_KERNEL),
3726 NULL,
3727 &zone);
3728 return zone->node;
3729}
3730#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003731
Linus Torvalds1da177e2005-04-16 15:20:36 -07003732#else /* CONFIG_NUMA */
3733
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003734static void set_zonelist_order(void)
3735{
3736 current_zonelist_order = ZONELIST_ORDER_ZONE;
3737}
3738
3739static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003740{
Christoph Lameter19655d32006-09-25 23:31:19 -07003741 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003742 enum zone_type j;
3743 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003744
3745 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003746
Mel Gorman54a6eb52008-04-28 02:12:16 -07003747 zonelist = &pgdat->node_zonelists[0];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003748 j = build_zonelists_node(pgdat, zonelist, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003749
Mel Gorman54a6eb52008-04-28 02:12:16 -07003750 /*
3751 * Now we build the zonelist so that it contains the zones
3752 * of all the other nodes.
3753 * We don't want to pressure a particular node, so when
3754 * building the zones for node N, we make sure that the
3755 * zones coming right after the local ones are those from
3756 * node N+1 (modulo N)
3757 */
3758 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3759 if (!node_online(node))
3760 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003761 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003762 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07003763 for (node = 0; node < local_node; node++) {
3764 if (!node_online(node))
3765 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003766 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gorman54a6eb52008-04-28 02:12:16 -07003767 }
3768
Mel Gormandd1a2392008-04-28 02:12:17 -07003769 zonelist->_zonerefs[j].zone = NULL;
3770 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003771}
3772
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003773/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003774static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003775{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003776 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003777}
3778
Linus Torvalds1da177e2005-04-16 15:20:36 -07003779#endif /* CONFIG_NUMA */
3780
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003781/*
3782 * Boot pageset table. One per cpu which is going to be used for all
3783 * zones and all nodes. The parameters will be set in such a way
3784 * that an item put on a list will immediately be handed over to
3785 * the buddy list. This is safe since pageset manipulation is done
3786 * with interrupts disabled.
3787 *
3788 * The boot_pagesets must be kept even after bootup is complete for
3789 * unused processors and/or zones. They do play a role for bootstrapping
3790 * hotplugged processors.
3791 *
3792 * zoneinfo_show() and maybe other functions do
3793 * not check if the processor is online before following the pageset pointer.
3794 * Other parts of the kernel may not check if the zone is available.
3795 */
3796static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3797static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07003798static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003799
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003800/*
3801 * Global mutex to protect against size modification of zonelists
3802 * as well as to serialize pageset setup for the new populated zone.
3803 */
3804DEFINE_MUTEX(zonelists_mutex);
3805
Rusty Russell9b1a4d32008-07-28 12:16:30 -05003806/* return values int ....just for stop_machine() */
Jiang Liu4ed7e022012-07-31 16:43:35 -07003807static int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003808{
Yasunori Goto68113782006-06-23 02:03:11 -07003809 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003810 int cpu;
Jiang Liu9adb62a2012-07-31 16:43:28 -07003811 pg_data_t *self = data;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003812
Bo Liu7f9cfb32009-08-18 14:11:19 -07003813#ifdef CONFIG_NUMA
3814 memset(node_load, 0, sizeof(node_load));
3815#endif
Jiang Liu9adb62a2012-07-31 16:43:28 -07003816
3817 if (self && !node_online(self->node_id)) {
3818 build_zonelists(self);
3819 build_zonelist_cache(self);
3820 }
3821
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003822 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07003823 pg_data_t *pgdat = NODE_DATA(nid);
3824
3825 build_zonelists(pgdat);
3826 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003827 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003828
3829 /*
3830 * Initialize the boot_pagesets that are going to be used
3831 * for bootstrapping processors. The real pagesets for
3832 * each zone will be allocated later when the per cpu
3833 * allocator is available.
3834 *
3835 * boot_pagesets are used also for bootstrapping offline
3836 * cpus if the system is already booted because the pagesets
3837 * are needed to initialize allocators on a specific cpu too.
3838 * F.e. the percpu allocator needs the page allocator which
3839 * needs the percpu allocator in order to allocate its pagesets
3840 * (a chicken-egg dilemma).
3841 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003842 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003843 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3844
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003845#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3846 /*
3847 * We now know the "local memory node" for each node--
3848 * i.e., the node of the first zone in the generic zonelist.
3849 * Set up numa_mem percpu variable for on-line cpus. During
3850 * boot, only the boot cpu should be on-line; we'll init the
3851 * secondary cpus' numa_mem as they come on-line. During
3852 * node/memory hotplug, we'll fixup all on-line cpus.
3853 */
3854 if (cpu_online(cpu))
3855 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3856#endif
3857 }
3858
Yasunori Goto68113782006-06-23 02:03:11 -07003859 return 0;
3860}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003861
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003862/*
3863 * Called with zonelists_mutex held always
3864 * unless system_state == SYSTEM_BOOTING.
3865 */
Jiang Liu9adb62a2012-07-31 16:43:28 -07003866void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
Yasunori Goto68113782006-06-23 02:03:11 -07003867{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003868 set_zonelist_order();
3869
Yasunori Goto68113782006-06-23 02:03:11 -07003870 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07003871 __build_all_zonelists(NULL);
Mel Gorman68ad8df2008-07-23 21:26:52 -07003872 mminit_verify_zonelist();
Yasunori Goto68113782006-06-23 02:03:11 -07003873 cpuset_init_current_mems_allowed();
3874 } else {
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003875#ifdef CONFIG_MEMORY_HOTPLUG
Jiang Liu9adb62a2012-07-31 16:43:28 -07003876 if (zone)
3877 setup_zone_pageset(zone);
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003878#endif
Cody P Schaferdd1895e2013-07-03 15:01:36 -07003879 /* we have to stop all cpus to guarantee there is no user
3880 of zonelist */
Jiang Liu9adb62a2012-07-31 16:43:28 -07003881 stop_machine(__build_all_zonelists, pgdat, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07003882 /* cpuset refresh routine should be here */
3883 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07003884 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003885 /*
3886 * Disable grouping by mobility if the number of pages in the
3887 * system is too low to allow the mechanism to work. It would be
3888 * more accurate, but expensive to check per-zone. This check is
3889 * made on memory-hotadd so a system can start with mobility
3890 * disabled and enable it later
3891 */
Mel Gormand9c23402007-10-16 01:26:01 -07003892 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003893 page_group_by_mobility_disabled = 1;
3894 else
3895 page_group_by_mobility_disabled = 0;
3896
3897 printk("Built %i zonelists in %s order, mobility grouping %s. "
3898 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003899 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003900 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003901 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003902 vm_total_pages);
3903#ifdef CONFIG_NUMA
3904 printk("Policy zone: %s\n", zone_names[policy_zone]);
3905#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003906}
3907
3908/*
3909 * Helper functions to size the waitqueue hash table.
3910 * Essentially these want to choose hash table sizes sufficiently
3911 * large so that collisions trying to wait on pages are rare.
3912 * But in fact, the number of active page waitqueues on typical
3913 * systems is ridiculously low, less than 200. So this is even
3914 * conservative, even though it seems large.
3915 *
3916 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3917 * waitqueues, i.e. the size of the waitq table given the number of pages.
3918 */
3919#define PAGES_PER_WAITQUEUE 256
3920
Yasunori Gotocca448f2006-06-23 02:03:10 -07003921#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07003922static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003923{
3924 unsigned long size = 1;
3925
3926 pages /= PAGES_PER_WAITQUEUE;
3927
3928 while (size < pages)
3929 size <<= 1;
3930
3931 /*
3932 * Once we have dozens or even hundreds of threads sleeping
3933 * on IO we've got bigger problems than wait queue collision.
3934 * Limit the size of the wait table to a reasonable size.
3935 */
3936 size = min(size, 4096UL);
3937
3938 return max(size, 4UL);
3939}
Yasunori Gotocca448f2006-06-23 02:03:10 -07003940#else
3941/*
3942 * A zone's size might be changed by hot-add, so it is not possible to determine
3943 * a suitable size for its wait_table. So we use the maximum size now.
3944 *
3945 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3946 *
3947 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3948 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3949 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3950 *
3951 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3952 * or more by the traditional way. (See above). It equals:
3953 *
3954 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3955 * ia64(16K page size) : = ( 8G + 4M)byte.
3956 * powerpc (64K page size) : = (32G +16M)byte.
3957 */
3958static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3959{
3960 return 4096UL;
3961}
3962#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003963
3964/*
3965 * This is an integer logarithm so that shifts can be used later
3966 * to extract the more random high bits from the multiplicative
3967 * hash function before the remainder is taken.
3968 */
3969static inline unsigned long wait_table_bits(unsigned long size)
3970{
3971 return ffz(~size);
3972}
3973
Mel Gorman56fd56b2007-10-16 01:25:58 -07003974/*
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07003975 * Check if a pageblock contains reserved pages
3976 */
3977static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
3978{
3979 unsigned long pfn;
3980
3981 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
3982 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
3983 return 1;
3984 }
3985 return 0;
3986}
3987
3988/*
Mel Gormand9c23402007-10-16 01:26:01 -07003989 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07003990 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3991 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07003992 * higher will lead to a bigger reserve which will get freed as contiguous
3993 * blocks as reclaim kicks in
3994 */
3995static void setup_zone_migrate_reserve(struct zone *zone)
3996{
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07003997 unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003998 struct page *page;
Mel Gorman78986a62009-09-21 17:03:02 -07003999 unsigned long block_migratetype;
4000 int reserve;
Yasuaki Ishimatsu943dca12014-01-21 15:49:06 -08004001 int old_reserve;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004002
Michal Hockod02156382011-12-08 14:34:27 -08004003 /*
4004 * Get the start pfn, end pfn and the number of blocks to reserve
4005 * We have to be careful to be aligned to pageblock_nr_pages to
4006 * make sure that we always check pfn_valid for the first page in
4007 * the block.
4008 */
Mel Gorman56fd56b2007-10-16 01:25:58 -07004009 start_pfn = zone->zone_start_pfn;
Cody P Schafer108bcc92013-02-22 16:35:23 -08004010 end_pfn = zone_end_pfn(zone);
Michal Hockod02156382011-12-08 14:34:27 -08004011 start_pfn = roundup(start_pfn, pageblock_nr_pages);
Mel Gorman41858962009-06-16 15:32:12 -07004012 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07004013 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004014
Mel Gorman78986a62009-09-21 17:03:02 -07004015 /*
4016 * Reserve blocks are generally in place to help high-order atomic
4017 * allocations that are short-lived. A min_free_kbytes value that
4018 * would result in more than 2 reserve blocks for atomic allocations
4019 * is assumed to be in place to help anti-fragmentation for the
4020 * future allocation of hugepages at runtime.
4021 */
4022 reserve = min(2, reserve);
Yasuaki Ishimatsu943dca12014-01-21 15:49:06 -08004023 old_reserve = zone->nr_migrate_reserve_block;
4024
4025 /* When memory hot-add, we almost always need to do nothing */
4026 if (reserve == old_reserve)
4027 return;
4028 zone->nr_migrate_reserve_block = reserve;
Mel Gorman78986a62009-09-21 17:03:02 -07004029
Mel Gormand9c23402007-10-16 01:26:01 -07004030 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07004031 if (!pfn_valid(pfn))
4032 continue;
4033 page = pfn_to_page(pfn);
4034
Adam Litke344c7902008-09-02 14:35:38 -07004035 /* Watch out for overlapping nodes */
4036 if (page_to_nid(page) != zone_to_nid(zone))
4037 continue;
4038
Mel Gorman56fd56b2007-10-16 01:25:58 -07004039 block_migratetype = get_pageblock_migratetype(page);
4040
Mel Gorman938929f2012-01-10 15:07:14 -08004041 /* Only test what is necessary when the reserves are not met */
4042 if (reserve > 0) {
4043 /*
4044 * Blocks with reserved pages will never free, skip
4045 * them.
4046 */
4047 block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
4048 if (pageblock_is_reserved(pfn, block_end_pfn))
4049 continue;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004050
Mel Gorman938929f2012-01-10 15:07:14 -08004051 /* If this block is reserved, account for it */
4052 if (block_migratetype == MIGRATE_RESERVE) {
4053 reserve--;
4054 continue;
4055 }
4056
4057 /* Suitable for reserving if this block is movable */
4058 if (block_migratetype == MIGRATE_MOVABLE) {
4059 set_pageblock_migratetype(page,
4060 MIGRATE_RESERVE);
4061 move_freepages_block(zone, page,
4062 MIGRATE_RESERVE);
4063 reserve--;
4064 continue;
4065 }
Yasuaki Ishimatsu943dca12014-01-21 15:49:06 -08004066 } else if (!old_reserve) {
4067 /*
4068 * At boot time we don't need to scan the whole zone
4069 * for turning off MIGRATE_RESERVE.
4070 */
4071 break;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004072 }
4073
4074 /*
4075 * If the reserve is met and this is a previous reserved block,
4076 * take it back
4077 */
4078 if (block_migratetype == MIGRATE_RESERVE) {
4079 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4080 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4081 }
4082 }
4083}
Mel Gormanac0e5b72007-10-16 01:25:58 -07004084
Linus Torvalds1da177e2005-04-16 15:20:36 -07004085/*
4086 * Initially all pages are reserved - free ones are freed
4087 * up by free_all_bootmem() once the early boot process is
4088 * done. Non-atomic initialization, single-pass.
4089 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01004090void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08004091 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004092{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004093 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07004094 unsigned long end_pfn = start_pfn + size;
4095 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004096 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004097
Hugh Dickins22b31ee2009-01-06 14:40:09 -08004098 if (highest_memmap_pfn < end_pfn - 1)
4099 highest_memmap_pfn = end_pfn - 1;
4100
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004101 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08004102 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08004103 /*
4104 * There can be holes in boot-time mem_map[]s
4105 * handed to this function. They do not
4106 * exist on hotplugged memory.
4107 */
4108 if (context == MEMMAP_EARLY) {
4109 if (!early_pfn_valid(pfn))
4110 continue;
4111 if (!early_pfn_in_nid(pfn, nid))
4112 continue;
4113 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004114 page = pfn_to_page(pfn);
4115 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07004116 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08004117 init_page_count(page);
Mel Gorman22b751c2013-02-22 16:34:59 -08004118 page_mapcount_reset(page);
Peter Zijlstra90572892013-10-07 11:29:20 +01004119 page_cpupid_reset_last(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004120 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004121 /*
4122 * Mark the block movable so that blocks are reserved for
4123 * movable at startup. This will force kernel allocations
4124 * to reserve their blocks rather than leaking throughout
4125 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07004126 * kernel allocations are made. Later some blocks near
4127 * the start are marked MIGRATE_RESERVE by
4128 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004129 *
4130 * bitmap is created for zone's valid pfn range. but memmap
4131 * can be created for invalid pages (for alignment)
4132 * check here not to call set_pageblock_migratetype() against
4133 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004134 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004135 if ((z->zone_start_pfn <= pfn)
Cody P Schafer108bcc92013-02-22 16:35:23 -08004136 && (pfn < zone_end_pfn(z))
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004137 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07004138 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004139
Linus Torvalds1da177e2005-04-16 15:20:36 -07004140 INIT_LIST_HEAD(&page->lru);
4141#ifdef WANT_PAGE_VIRTUAL
4142 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
4143 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07004144 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004145#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004146 }
4147}
4148
Andi Kleen1e548de2008-02-04 22:29:26 -08004149static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004150{
Mel Gorman7aeb09f2014-06-04 16:10:21 -07004151 unsigned int order, t;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004152 for_each_migratetype_order(order, t) {
4153 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004154 zone->free_area[order].nr_free = 0;
4155 }
4156}
4157
4158#ifndef __HAVE_ARCH_MEMMAP_INIT
4159#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08004160 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004161#endif
4162
David Rientjes7cd2b0a2014-06-23 13:22:04 -07004163static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004164{
David Howells3a6be872009-05-06 16:03:03 -07004165#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004166 int batch;
4167
4168 /*
4169 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07004170 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004171 *
4172 * OK, so we don't know how big the cache is. So guess.
4173 */
Jiang Liub40da042013-02-22 16:33:52 -08004174 batch = zone->managed_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07004175 if (batch * PAGE_SIZE > 512 * 1024)
4176 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004177 batch /= 4; /* We effectively *= 4 below */
4178 if (batch < 1)
4179 batch = 1;
4180
4181 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004182 * Clamp the batch to a 2^n - 1 value. Having a power
4183 * of 2 value was found to be more likely to have
4184 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004185 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004186 * For example if 2 tasks are alternately allocating
4187 * batches of pages, one task can end up with a lot
4188 * of pages of one half of the possible page colors
4189 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004190 */
David Howells91552032009-05-06 16:03:02 -07004191 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07004192
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004193 return batch;
David Howells3a6be872009-05-06 16:03:03 -07004194
4195#else
4196 /* The deferral and batching of frees should be suppressed under NOMMU
4197 * conditions.
4198 *
4199 * The problem is that NOMMU needs to be able to allocate large chunks
4200 * of contiguous memory as there's no hardware page translation to
4201 * assemble apparent contiguous memory from discontiguous pages.
4202 *
4203 * Queueing large contiguous runs of pages for batching, however,
4204 * causes the pages to actually be freed in smaller chunks. As there
4205 * can be a significant delay between the individual batches being
4206 * recycled, this leads to the once large chunks of space being
4207 * fragmented and becoming unavailable for high-order allocations.
4208 */
4209 return 0;
4210#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004211}
4212
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004213/*
4214 * pcp->high and pcp->batch values are related and dependent on one another:
4215 * ->batch must never be higher then ->high.
4216 * The following function updates them in a safe manner without read side
4217 * locking.
4218 *
4219 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4220 * those fields changing asynchronously (acording the the above rule).
4221 *
4222 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4223 * outside of boot time (or some other assurance that no concurrent updaters
4224 * exist).
4225 */
4226static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
4227 unsigned long batch)
4228{
4229 /* start with a fail safe value for batch */
4230 pcp->batch = 1;
4231 smp_wmb();
4232
4233 /* Update high, then batch, in order */
4234 pcp->high = high;
4235 smp_wmb();
4236
4237 pcp->batch = batch;
4238}
4239
Cody P Schafer36640332013-07-03 15:01:40 -07004240/* a companion to pageset_set_high() */
Cody P Schafer4008bab2013-07-03 15:01:28 -07004241static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
4242{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004243 pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
Cody P Schafer4008bab2013-07-03 15:01:28 -07004244}
4245
Cody P Schafer88c90db2013-07-03 15:01:35 -07004246static void pageset_init(struct per_cpu_pageset *p)
Christoph Lameter2caaad42005-06-21 17:15:00 -07004247{
4248 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004249 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004250
Magnus Damm1c6fe942005-10-26 01:58:59 -07004251 memset(p, 0, sizeof(*p));
4252
Christoph Lameter3dfa5722008-02-04 22:29:19 -08004253 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004254 pcp->count = 0;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004255 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
4256 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07004257}
4258
Cody P Schafer88c90db2013-07-03 15:01:35 -07004259static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
4260{
4261 pageset_init(p);
4262 pageset_set_batch(p, batch);
4263}
4264
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004265/*
Cody P Schafer36640332013-07-03 15:01:40 -07004266 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004267 * to the value high for the pageset p.
4268 */
Cody P Schafer36640332013-07-03 15:01:40 -07004269static void pageset_set_high(struct per_cpu_pageset *p,
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004270 unsigned long high)
4271{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004272 unsigned long batch = max(1UL, high / 4);
4273 if ((high / 4) > (PAGE_SHIFT * 8))
4274 batch = PAGE_SHIFT * 8;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004275
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004276 pageset_update(&p->pcp, high, batch);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004277}
4278
David Rientjes7cd2b0a2014-06-23 13:22:04 -07004279static void pageset_set_high_and_batch(struct zone *zone,
4280 struct per_cpu_pageset *pcp)
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004281{
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004282 if (percpu_pagelist_fraction)
Cody P Schafer36640332013-07-03 15:01:40 -07004283 pageset_set_high(pcp,
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004284 (zone->managed_pages /
4285 percpu_pagelist_fraction));
4286 else
4287 pageset_set_batch(pcp, zone_batchsize(zone));
4288}
4289
Cody P Schafer169f6c12013-07-03 15:01:41 -07004290static void __meminit zone_pageset_init(struct zone *zone, int cpu)
4291{
4292 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4293
4294 pageset_init(pcp);
4295 pageset_set_high_and_batch(zone, pcp);
4296}
4297
Jiang Liu4ed7e022012-07-31 16:43:35 -07004298static void __meminit setup_zone_pageset(struct zone *zone)
Wu Fengguang319774e2010-05-24 14:32:49 -07004299{
4300 int cpu;
Wu Fengguang319774e2010-05-24 14:32:49 -07004301 zone->pageset = alloc_percpu(struct per_cpu_pageset);
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004302 for_each_possible_cpu(cpu)
4303 zone_pageset_init(zone, cpu);
Wu Fengguang319774e2010-05-24 14:32:49 -07004304}
4305
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004306/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004307 * Allocate per cpu pagesets and initialize them.
4308 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07004309 */
Al Viro78d99552005-12-15 09:18:25 +00004310void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004311{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004312 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004313
Wu Fengguang319774e2010-05-24 14:32:49 -07004314 for_each_populated_zone(zone)
4315 setup_zone_pageset(zone);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004316}
4317
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004318static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07004319int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07004320{
4321 int i;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004322 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07004323
4324 /*
4325 * The per-page waitqueue mechanism uses hashed waitqueues
4326 * per zone.
4327 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07004328 zone->wait_table_hash_nr_entries =
4329 wait_table_hash_nr_entries(zone_size_pages);
4330 zone->wait_table_bits =
4331 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004332 alloc_size = zone->wait_table_hash_nr_entries
4333 * sizeof(wait_queue_head_t);
4334
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07004335 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07004336 zone->wait_table = (wait_queue_head_t *)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004337 memblock_virt_alloc_node_nopanic(
4338 alloc_size, zone->zone_pgdat->node_id);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004339 } else {
4340 /*
4341 * This case means that a zone whose size was 0 gets new memory
4342 * via memory hot-add.
4343 * But it may be the case that a new node was hot-added. In
4344 * this case vmalloc() will not be able to use this new node's
4345 * memory - this wait_table must be initialized to use this new
4346 * node itself as well.
4347 * To use this new node's memory, further consideration will be
4348 * necessary.
4349 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07004350 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004351 }
4352 if (!zone->wait_table)
4353 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07004354
Pintu Kumarb8af2942013-09-11 14:20:34 -07004355 for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07004356 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004357
4358 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004359}
4360
Matt Tolentinoc09b4242006-01-17 07:03:44 +01004361static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07004362{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004363 /*
4364 * per cpu subsystem is not up at this point. The following code
4365 * relies on the ability of the linker to provide the
4366 * offset of a (static) per cpu variable into the per cpu area.
4367 */
4368 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07004369
Xishi Qiub38a8722013-11-12 15:07:20 -08004370 if (populated_zone(zone))
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004371 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
4372 zone->name, zone->present_pages,
4373 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07004374}
4375
Jiang Liu4ed7e022012-07-31 16:43:35 -07004376int __meminit init_currently_empty_zone(struct zone *zone,
Yasunori Goto718127c2006-06-23 02:03:10 -07004377 unsigned long zone_start_pfn,
Dave Hansena2f3aa022007-01-10 23:15:30 -08004378 unsigned long size,
4379 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07004380{
4381 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004382 int ret;
4383 ret = zone_wait_table_init(zone, size);
4384 if (ret)
4385 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07004386 pgdat->nr_zones = zone_idx(zone) + 1;
4387
Dave Hansened8ece22005-10-29 18:16:50 -07004388 zone->zone_start_pfn = zone_start_pfn;
4389
Mel Gorman708614e2008-07-23 21:26:51 -07004390 mminit_dprintk(MMINIT_TRACE, "memmap_init",
4391 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4392 pgdat->node_id,
4393 (unsigned long)zone_idx(zone),
4394 zone_start_pfn, (zone_start_pfn + size));
4395
Andi Kleen1e548de2008-02-04 22:29:26 -08004396 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07004397
4398 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004399}
4400
Tejun Heo0ee332c2011-12-08 10:22:09 -08004401#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07004402#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4403/*
4404 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
Mel Gormanc7132162006-09-27 01:49:43 -07004405 */
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004406int __meminit __early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004407{
Tejun Heoc13291a2011-07-12 10:46:30 +02004408 unsigned long start_pfn, end_pfn;
Yinghai Lue76b63f2013-09-11 14:22:17 -07004409 int nid;
Russ Anderson7c243c72013-04-29 15:07:59 -07004410 /*
4411 * NOTE: The following SMP-unsafe globals are only used early in boot
4412 * when the kernel is running single-threaded.
4413 */
4414 static unsigned long __meminitdata last_start_pfn, last_end_pfn;
4415 static int __meminitdata last_nid;
4416
4417 if (last_start_pfn <= pfn && pfn < last_end_pfn)
4418 return last_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004419
Yinghai Lue76b63f2013-09-11 14:22:17 -07004420 nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
4421 if (nid != -1) {
4422 last_start_pfn = start_pfn;
4423 last_end_pfn = end_pfn;
4424 last_nid = nid;
4425 }
4426
4427 return nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004428}
4429#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4430
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004431int __meminit early_pfn_to_nid(unsigned long pfn)
4432{
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08004433 int nid;
4434
4435 nid = __early_pfn_to_nid(pfn);
4436 if (nid >= 0)
4437 return nid;
4438 /* just returns 0 */
4439 return 0;
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004440}
4441
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08004442#ifdef CONFIG_NODES_SPAN_OTHER_NODES
4443bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
4444{
4445 int nid;
4446
4447 nid = __early_pfn_to_nid(pfn);
4448 if (nid >= 0 && nid != node)
4449 return false;
4450 return true;
4451}
4452#endif
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004453
Mel Gormanc7132162006-09-27 01:49:43 -07004454/**
Santosh Shilimkar67828322014-01-21 15:50:25 -08004455 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004456 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
Santosh Shilimkar67828322014-01-21 15:50:25 -08004457 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
Mel Gormanc7132162006-09-27 01:49:43 -07004458 *
Zhang Zhen7d018172014-06-04 16:10:53 -07004459 * If an architecture guarantees that all ranges registered contain no holes
4460 * and may be freed, this this function may be used instead of calling
4461 * memblock_free_early_nid() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07004462 */
Tejun Heoc13291a2011-07-12 10:46:30 +02004463void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004464{
Tejun Heoc13291a2011-07-12 10:46:30 +02004465 unsigned long start_pfn, end_pfn;
4466 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004467
Tejun Heoc13291a2011-07-12 10:46:30 +02004468 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
4469 start_pfn = min(start_pfn, max_low_pfn);
4470 end_pfn = min(end_pfn, max_low_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004471
Tejun Heoc13291a2011-07-12 10:46:30 +02004472 if (start_pfn < end_pfn)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004473 memblock_free_early_nid(PFN_PHYS(start_pfn),
4474 (end_pfn - start_pfn) << PAGE_SHIFT,
4475 this_nid);
Mel Gormanc7132162006-09-27 01:49:43 -07004476 }
4477}
4478
4479/**
4480 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004481 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07004482 *
Zhang Zhen7d018172014-06-04 16:10:53 -07004483 * If an architecture guarantees that all ranges registered contain no holes and may
4484 * be freed, this function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07004485 */
4486void __init sparse_memory_present_with_active_regions(int nid)
4487{
Tejun Heoc13291a2011-07-12 10:46:30 +02004488 unsigned long start_pfn, end_pfn;
4489 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004490
Tejun Heoc13291a2011-07-12 10:46:30 +02004491 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
4492 memory_present(this_nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004493}
4494
4495/**
4496 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004497 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4498 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4499 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07004500 *
4501 * It returns the start and end page frame of a node based on information
Zhang Zhen7d018172014-06-04 16:10:53 -07004502 * provided by memblock_set_node(). If called for a node
Mel Gormanc7132162006-09-27 01:49:43 -07004503 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004504 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07004505 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004506void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004507 unsigned long *start_pfn, unsigned long *end_pfn)
4508{
Tejun Heoc13291a2011-07-12 10:46:30 +02004509 unsigned long this_start_pfn, this_end_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004510 int i;
Tejun Heoc13291a2011-07-12 10:46:30 +02004511
Mel Gormanc7132162006-09-27 01:49:43 -07004512 *start_pfn = -1UL;
4513 *end_pfn = 0;
4514
Tejun Heoc13291a2011-07-12 10:46:30 +02004515 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
4516 *start_pfn = min(*start_pfn, this_start_pfn);
4517 *end_pfn = max(*end_pfn, this_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004518 }
4519
Christoph Lameter633c0662007-10-16 01:25:37 -07004520 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07004521 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07004522}
4523
4524/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07004525 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4526 * assumption is made that zones within a node are ordered in monotonic
4527 * increasing memory addresses so that the "highest" populated zone is used
4528 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004529static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004530{
4531 int zone_index;
4532 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
4533 if (zone_index == ZONE_MOVABLE)
4534 continue;
4535
4536 if (arch_zone_highest_possible_pfn[zone_index] >
4537 arch_zone_lowest_possible_pfn[zone_index])
4538 break;
4539 }
4540
4541 VM_BUG_ON(zone_index == -1);
4542 movable_zone = zone_index;
4543}
4544
4545/*
4546 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004547 * because it is sized independent of architecture. Unlike the other zones,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004548 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4549 * in each node depending on the size of each node and how evenly kernelcore
4550 * is distributed. This helper function adjusts the zone ranges
4551 * provided by the architecture for a given node by using the end of the
4552 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4553 * zones within a node are in order of monotonic increases memory addresses
4554 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004555static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004556 unsigned long zone_type,
4557 unsigned long node_start_pfn,
4558 unsigned long node_end_pfn,
4559 unsigned long *zone_start_pfn,
4560 unsigned long *zone_end_pfn)
4561{
4562 /* Only adjust if ZONE_MOVABLE is on this node */
4563 if (zone_movable_pfn[nid]) {
4564 /* Size ZONE_MOVABLE */
4565 if (zone_type == ZONE_MOVABLE) {
4566 *zone_start_pfn = zone_movable_pfn[nid];
4567 *zone_end_pfn = min(node_end_pfn,
4568 arch_zone_highest_possible_pfn[movable_zone]);
4569
4570 /* Adjust for ZONE_MOVABLE starting within this range */
4571 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
4572 *zone_end_pfn > zone_movable_pfn[nid]) {
4573 *zone_end_pfn = zone_movable_pfn[nid];
4574
4575 /* Check if this whole range is within ZONE_MOVABLE */
4576 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
4577 *zone_start_pfn = *zone_end_pfn;
4578 }
4579}
4580
4581/*
Mel Gormanc7132162006-09-27 01:49:43 -07004582 * Return the number of pages a zone spans in a node, including holes
4583 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4584 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004585static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004586 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004587 unsigned long node_start_pfn,
4588 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004589 unsigned long *ignored)
4590{
Mel Gormanc7132162006-09-27 01:49:43 -07004591 unsigned long zone_start_pfn, zone_end_pfn;
4592
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004593 /* Get the start and end of the zone */
Mel Gormanc7132162006-09-27 01:49:43 -07004594 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4595 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07004596 adjust_zone_range_for_zone_movable(nid, zone_type,
4597 node_start_pfn, node_end_pfn,
4598 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004599
4600 /* Check that this node has pages within the zone's required range */
4601 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4602 return 0;
4603
4604 /* Move the zone boundaries inside the node if necessary */
4605 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4606 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4607
4608 /* Return the spanned pages */
4609 return zone_end_pfn - zone_start_pfn;
4610}
4611
4612/*
4613 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004614 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07004615 */
Yinghai Lu32996252009-12-15 17:59:02 -08004616unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004617 unsigned long range_start_pfn,
4618 unsigned long range_end_pfn)
4619{
Tejun Heo96e907d2011-07-12 10:46:29 +02004620 unsigned long nr_absent = range_end_pfn - range_start_pfn;
4621 unsigned long start_pfn, end_pfn;
4622 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004623
Tejun Heo96e907d2011-07-12 10:46:29 +02004624 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
4625 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
4626 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
4627 nr_absent -= end_pfn - start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004628 }
Tejun Heo96e907d2011-07-12 10:46:29 +02004629 return nr_absent;
Mel Gormanc7132162006-09-27 01:49:43 -07004630}
4631
4632/**
4633 * absent_pages_in_range - Return number of page frames in holes within a range
4634 * @start_pfn: The start PFN to start searching for holes
4635 * @end_pfn: The end PFN to stop searching for holes
4636 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004637 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07004638 */
4639unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4640 unsigned long end_pfn)
4641{
4642 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4643}
4644
4645/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004646static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004647 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004648 unsigned long node_start_pfn,
4649 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004650 unsigned long *ignored)
4651{
Tejun Heo96e907d2011-07-12 10:46:29 +02004652 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
4653 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman9c7cd682006-09-27 01:49:58 -07004654 unsigned long zone_start_pfn, zone_end_pfn;
4655
Tejun Heo96e907d2011-07-12 10:46:29 +02004656 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
4657 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004658
Mel Gorman2a1e2742007-07-17 04:03:12 -07004659 adjust_zone_range_for_zone_movable(nid, zone_type,
4660 node_start_pfn, node_end_pfn,
4661 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004662 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004663}
Mel Gorman0e0b8642006-09-27 01:49:56 -07004664
Tejun Heo0ee332c2011-12-08 10:22:09 -08004665#else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004666static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004667 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004668 unsigned long node_start_pfn,
4669 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004670 unsigned long *zones_size)
4671{
4672 return zones_size[zone_type];
4673}
4674
Paul Mundt6ea6e682007-07-15 23:38:20 -07004675static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004676 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004677 unsigned long node_start_pfn,
4678 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004679 unsigned long *zholes_size)
4680{
4681 if (!zholes_size)
4682 return 0;
4683
4684 return zholes_size[zone_type];
4685}
Yinghai Lu20e69262013-03-01 14:51:27 -08004686
Tejun Heo0ee332c2011-12-08 10:22:09 -08004687#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07004688
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004689static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004690 unsigned long node_start_pfn,
4691 unsigned long node_end_pfn,
4692 unsigned long *zones_size,
4693 unsigned long *zholes_size)
Mel Gormanc7132162006-09-27 01:49:43 -07004694{
4695 unsigned long realtotalpages, totalpages = 0;
4696 enum zone_type i;
4697
4698 for (i = 0; i < MAX_NR_ZONES; i++)
4699 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004700 node_start_pfn,
4701 node_end_pfn,
4702 zones_size);
Mel Gormanc7132162006-09-27 01:49:43 -07004703 pgdat->node_spanned_pages = totalpages;
4704
4705 realtotalpages = totalpages;
4706 for (i = 0; i < MAX_NR_ZONES; i++)
4707 realtotalpages -=
4708 zone_absent_pages_in_node(pgdat->node_id, i,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004709 node_start_pfn, node_end_pfn,
4710 zholes_size);
Mel Gormanc7132162006-09-27 01:49:43 -07004711 pgdat->node_present_pages = realtotalpages;
4712 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4713 realtotalpages);
4714}
4715
Mel Gorman835c1342007-10-16 01:25:47 -07004716#ifndef CONFIG_SPARSEMEM
4717/*
4718 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07004719 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4720 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07004721 * round what is now in bits to nearest long in bits, then return it in
4722 * bytes.
4723 */
Linus Torvalds7c455122013-02-18 09:58:02 -08004724static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07004725{
4726 unsigned long usemapsize;
4727
Linus Torvalds7c455122013-02-18 09:58:02 -08004728 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
Mel Gormand9c23402007-10-16 01:26:01 -07004729 usemapsize = roundup(zonesize, pageblock_nr_pages);
4730 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07004731 usemapsize *= NR_PAGEBLOCK_BITS;
4732 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4733
4734 return usemapsize / 8;
4735}
4736
4737static void __init setup_usemap(struct pglist_data *pgdat,
Linus Torvalds7c455122013-02-18 09:58:02 -08004738 struct zone *zone,
4739 unsigned long zone_start_pfn,
4740 unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07004741{
Linus Torvalds7c455122013-02-18 09:58:02 -08004742 unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
Mel Gorman835c1342007-10-16 01:25:47 -07004743 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08004744 if (usemapsize)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004745 zone->pageblock_flags =
4746 memblock_virt_alloc_node_nopanic(usemapsize,
4747 pgdat->node_id);
Mel Gorman835c1342007-10-16 01:25:47 -07004748}
4749#else
Linus Torvalds7c455122013-02-18 09:58:02 -08004750static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
4751 unsigned long zone_start_pfn, unsigned long zonesize) {}
Mel Gorman835c1342007-10-16 01:25:47 -07004752#endif /* CONFIG_SPARSEMEM */
4753
Mel Gormand9c23402007-10-16 01:26:01 -07004754#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08004755
Mel Gormand9c23402007-10-16 01:26:01 -07004756/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
Chen Gang15ca2202013-09-11 14:20:27 -07004757void __paginginit set_pageblock_order(void)
Mel Gormand9c23402007-10-16 01:26:01 -07004758{
Andrew Morton955c1cd2012-05-29 15:06:31 -07004759 unsigned int order;
4760
Mel Gormand9c23402007-10-16 01:26:01 -07004761 /* Check that pageblock_nr_pages has not already been setup */
4762 if (pageblock_order)
4763 return;
4764
Andrew Morton955c1cd2012-05-29 15:06:31 -07004765 if (HPAGE_SHIFT > PAGE_SHIFT)
4766 order = HUGETLB_PAGE_ORDER;
4767 else
4768 order = MAX_ORDER - 1;
4769
Mel Gormand9c23402007-10-16 01:26:01 -07004770 /*
4771 * Assume the largest contiguous order of interest is a huge page.
Andrew Morton955c1cd2012-05-29 15:06:31 -07004772 * This value may be variable depending on boot parameters on IA64 and
4773 * powerpc.
Mel Gormand9c23402007-10-16 01:26:01 -07004774 */
4775 pageblock_order = order;
4776}
4777#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4778
Mel Gormanba72cb82007-11-28 16:21:13 -08004779/*
4780 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
Andrew Morton955c1cd2012-05-29 15:06:31 -07004781 * is unused as pageblock_order is set at compile-time. See
4782 * include/linux/pageblock-flags.h for the values of pageblock_order based on
4783 * the kernel config
Mel Gormanba72cb82007-11-28 16:21:13 -08004784 */
Chen Gang15ca2202013-09-11 14:20:27 -07004785void __paginginit set_pageblock_order(void)
Mel Gormanba72cb82007-11-28 16:21:13 -08004786{
Mel Gormanba72cb82007-11-28 16:21:13 -08004787}
Mel Gormand9c23402007-10-16 01:26:01 -07004788
4789#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4790
Jiang Liu01cefae2012-12-12 13:52:19 -08004791static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
4792 unsigned long present_pages)
4793{
4794 unsigned long pages = spanned_pages;
4795
4796 /*
4797 * Provide a more accurate estimation if there are holes within
4798 * the zone and SPARSEMEM is in use. If there are holes within the
4799 * zone, each populated memory region may cost us one or two extra
4800 * memmap pages due to alignment because memmap pages for each
4801 * populated regions may not naturally algined on page boundary.
4802 * So the (present_pages >> 4) heuristic is a tradeoff for that.
4803 */
4804 if (spanned_pages > present_pages + (present_pages >> 4) &&
4805 IS_ENABLED(CONFIG_SPARSEMEM))
4806 pages = present_pages;
4807
4808 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
4809}
4810
Linus Torvalds1da177e2005-04-16 15:20:36 -07004811/*
4812 * Set up the zone data structures:
4813 * - mark all pages reserved
4814 * - mark all memory queues empty
4815 * - clear the memory bitmaps
Minchan Kim6527af52012-07-31 16:46:16 -07004816 *
4817 * NOTE: pgdat should get zeroed by caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004818 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08004819static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004820 unsigned long node_start_pfn, unsigned long node_end_pfn,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004821 unsigned long *zones_size, unsigned long *zholes_size)
4822{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07004823 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07004824 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004825 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07004826 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004827
Dave Hansen208d54e2005-10-29 18:16:52 -07004828 pgdat_resize_init(pgdat);
Andrea Arcangeli8177a422012-03-23 20:56:34 +01004829#ifdef CONFIG_NUMA_BALANCING
4830 spin_lock_init(&pgdat->numabalancing_migrate_lock);
4831 pgdat->numabalancing_migrate_nr_pages = 0;
4832 pgdat->numabalancing_migrate_next_window = jiffies;
4833#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004834 init_waitqueue_head(&pgdat->kswapd_wait);
Mel Gorman55150612012-07-31 16:44:35 -07004835 init_waitqueue_head(&pgdat->pfmemalloc_wait);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004836 pgdat_page_cgroup_init(pgdat);
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01004837
Linus Torvalds1da177e2005-04-16 15:20:36 -07004838 for (j = 0; j < MAX_NR_ZONES; j++) {
4839 struct zone *zone = pgdat->node_zones + j;
Jiang Liu9feedc92012-12-12 13:52:12 -08004840 unsigned long size, realsize, freesize, memmap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004841
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004842 size = zone_spanned_pages_in_node(nid, j, node_start_pfn,
4843 node_end_pfn, zones_size);
Jiang Liu9feedc92012-12-12 13:52:12 -08004844 realsize = freesize = size - zone_absent_pages_in_node(nid, j,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004845 node_start_pfn,
4846 node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004847 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004848
Mel Gorman0e0b8642006-09-27 01:49:56 -07004849 /*
Jiang Liu9feedc92012-12-12 13:52:12 -08004850 * Adjust freesize so that it accounts for how much memory
Mel Gorman0e0b8642006-09-27 01:49:56 -07004851 * is used by this zone for memmap. This affects the watermark
4852 * and per-cpu initialisations
4853 */
Jiang Liu01cefae2012-12-12 13:52:19 -08004854 memmap_pages = calc_memmap_size(size, realsize);
Jiang Liu9feedc92012-12-12 13:52:12 -08004855 if (freesize >= memmap_pages) {
4856 freesize -= memmap_pages;
Yinghai Lu5594c8c2009-01-06 14:39:14 -08004857 if (memmap_pages)
4858 printk(KERN_DEBUG
4859 " %s zone: %lu pages used for memmap\n",
4860 zone_names[j], memmap_pages);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004861 } else
4862 printk(KERN_WARNING
Jiang Liu9feedc92012-12-12 13:52:12 -08004863 " %s zone: %lu pages exceeds freesize %lu\n",
4864 zone_names[j], memmap_pages, freesize);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004865
Christoph Lameter62672762007-02-10 01:43:07 -08004866 /* Account for reserved pages */
Jiang Liu9feedc92012-12-12 13:52:12 -08004867 if (j == 0 && freesize > dma_reserve) {
4868 freesize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07004869 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08004870 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004871 }
4872
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004873 if (!is_highmem_idx(j))
Jiang Liu9feedc92012-12-12 13:52:12 -08004874 nr_kernel_pages += freesize;
Jiang Liu01cefae2012-12-12 13:52:19 -08004875 /* Charge for highmem memmap if there are enough kernel pages */
4876 else if (nr_kernel_pages > memmap_pages * 2)
4877 nr_kernel_pages -= memmap_pages;
Jiang Liu9feedc92012-12-12 13:52:12 -08004878 nr_all_pages += freesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004879
4880 zone->spanned_pages = size;
Jiang Liu306f2e92013-02-22 16:33:54 -08004881 zone->present_pages = realsize;
Jiang Liu9feedc92012-12-12 13:52:12 -08004882 /*
4883 * Set an approximate value for lowmem here, it will be adjusted
4884 * when the bootmem allocator frees pages into the buddy system.
4885 * And all highmem pages will be managed by the buddy system.
4886 */
4887 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
Christoph Lameter96146342006-07-03 00:24:13 -07004888#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07004889 zone->node = nid;
Jiang Liu9feedc92012-12-12 13:52:12 -08004890 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07004891 / 100;
Jiang Liu9feedc92012-12-12 13:52:12 -08004892 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07004893#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004894 zone->name = zone_names[j];
4895 spin_lock_init(&zone->lock);
4896 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07004897 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004898 zone->zone_pgdat = pgdat;
Dave Hansened8ece22005-10-29 18:16:50 -07004899 zone_pcp_init(zone);
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07004900
4901 /* For bootup, initialized properly in watermark setup */
4902 mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages);
4903
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004904 lruvec_init(&zone->lruvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004905 if (!size)
4906 continue;
4907
Andrew Morton955c1cd2012-05-29 15:06:31 -07004908 set_pageblock_order();
Linus Torvalds7c455122013-02-18 09:58:02 -08004909 setup_usemap(pgdat, zone, zone_start_pfn, size);
Dave Hansena2f3aa022007-01-10 23:15:30 -08004910 ret = init_currently_empty_zone(zone, zone_start_pfn,
4911 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07004912 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07004913 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004914 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004915 }
4916}
4917
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004918static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004919{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004920 /* Skip empty nodes */
4921 if (!pgdat->node_spanned_pages)
4922 return;
4923
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004924#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07004925 /* ia64 gets its own node_mem_map, before this, without bootmem */
4926 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07004927 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004928 struct page *map;
4929
Bob Piccoe984bb42006-05-20 15:00:31 -07004930 /*
4931 * The zone's endpoints aren't required to be MAX_ORDER
4932 * aligned but the node_mem_map endpoints must be in order
4933 * for the buddy allocator to function correctly.
4934 */
4935 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
Cody P Schafer108bcc92013-02-22 16:35:23 -08004936 end = pgdat_end_pfn(pgdat);
Bob Piccoe984bb42006-05-20 15:00:31 -07004937 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4938 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07004939 map = alloc_remap(pgdat->node_id, size);
4940 if (!map)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004941 map = memblock_virt_alloc_node_nopanic(size,
4942 pgdat->node_id);
Bob Piccoe984bb42006-05-20 15:00:31 -07004943 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004944 }
Roman Zippel12d810c2007-05-31 00:40:54 -07004945#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07004946 /*
4947 * With no DISCONTIG, the global mem_map is just set as node 0's
4948 */
Mel Gormanc7132162006-09-27 01:49:43 -07004949 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004950 mem_map = NODE_DATA(0)->node_mem_map;
Tejun Heo0ee332c2011-12-08 10:22:09 -08004951#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07004952 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08004953 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Tejun Heo0ee332c2011-12-08 10:22:09 -08004954#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07004955 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004956#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004957#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004958}
4959
Johannes Weiner9109fb72008-07-23 21:27:20 -07004960void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4961 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004962{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004963 pg_data_t *pgdat = NODE_DATA(nid);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004964 unsigned long start_pfn = 0;
4965 unsigned long end_pfn = 0;
Johannes Weiner9109fb72008-07-23 21:27:20 -07004966
Minchan Kim88fdf752012-07-31 16:46:14 -07004967 /* pg_data_t should be reset to zero when it's allocated */
Linus Torvalds8783b6e2012-08-02 10:37:03 -07004968 WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
Minchan Kim88fdf752012-07-31 16:46:14 -07004969
Linus Torvalds1da177e2005-04-16 15:20:36 -07004970 pgdat->node_id = nid;
4971 pgdat->node_start_pfn = node_start_pfn;
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004972#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4973 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
Luiz Capitulino8b375f62014-08-22 13:27:36 -07004974 printk(KERN_INFO "Initmem setup node %d [mem %#010Lx-%#010Lx]\n", nid,
4975 (u64) start_pfn << PAGE_SHIFT, (u64) (end_pfn << PAGE_SHIFT) - 1);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004976#endif
4977 calculate_node_totalpages(pgdat, start_pfn, end_pfn,
4978 zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004979
4980 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07004981#ifdef CONFIG_FLAT_NODE_MEM_MAP
4982 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4983 nid, (unsigned long)pgdat,
4984 (unsigned long)pgdat->node_mem_map);
4985#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004986
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004987 free_area_init_core(pgdat, start_pfn, end_pfn,
4988 zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004989}
4990
Tejun Heo0ee332c2011-12-08 10:22:09 -08004991#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07004992
4993#if MAX_NUMNODES > 1
4994/*
4995 * Figure out the number of possible node ids.
4996 */
Cody P Schaferf9872ca2013-04-29 15:08:01 -07004997void __init setup_nr_node_ids(void)
Miklos Szeredi418508c2007-05-23 13:57:55 -07004998{
4999 unsigned int node;
5000 unsigned int highest = 0;
5001
5002 for_each_node_mask(node, node_possible_map)
5003 highest = node;
5004 nr_node_ids = highest + 1;
5005}
Miklos Szeredi418508c2007-05-23 13:57:55 -07005006#endif
5007
Mel Gormanc7132162006-09-27 01:49:43 -07005008/**
Tejun Heo1e019792011-07-12 09:45:34 +02005009 * node_map_pfn_alignment - determine the maximum internode alignment
5010 *
5011 * This function should be called after node map is populated and sorted.
5012 * It calculates the maximum power of two alignment which can distinguish
5013 * all the nodes.
5014 *
5015 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5016 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
5017 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
5018 * shifted, 1GiB is enough and this function will indicate so.
5019 *
5020 * This is used to test whether pfn -> nid mapping of the chosen memory
5021 * model has fine enough granularity to avoid incorrect mapping for the
5022 * populated node map.
5023 *
5024 * Returns the determined alignment in pfn's. 0 if there is no alignment
5025 * requirement (single node).
5026 */
5027unsigned long __init node_map_pfn_alignment(void)
5028{
5029 unsigned long accl_mask = 0, last_end = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02005030 unsigned long start, end, mask;
Tejun Heo1e019792011-07-12 09:45:34 +02005031 int last_nid = -1;
Tejun Heoc13291a2011-07-12 10:46:30 +02005032 int i, nid;
Tejun Heo1e019792011-07-12 09:45:34 +02005033
Tejun Heoc13291a2011-07-12 10:46:30 +02005034 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
Tejun Heo1e019792011-07-12 09:45:34 +02005035 if (!start || last_nid < 0 || last_nid == nid) {
5036 last_nid = nid;
5037 last_end = end;
5038 continue;
5039 }
5040
5041 /*
5042 * Start with a mask granular enough to pin-point to the
5043 * start pfn and tick off bits one-by-one until it becomes
5044 * too coarse to separate the current node from the last.
5045 */
5046 mask = ~((1 << __ffs(start)) - 1);
5047 while (mask && last_end <= (start & (mask << 1)))
5048 mask <<= 1;
5049
5050 /* accumulate all internode masks */
5051 accl_mask |= mask;
5052 }
5053
5054 /* convert mask to number of pages */
5055 return ~accl_mask + 1;
5056}
5057
Mel Gormana6af2bc2007-02-10 01:42:57 -08005058/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07005059static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07005060{
Mel Gormana6af2bc2007-02-10 01:42:57 -08005061 unsigned long min_pfn = ULONG_MAX;
Tejun Heoc13291a2011-07-12 10:46:30 +02005062 unsigned long start_pfn;
5063 int i;
Mel Gorman1abbfb42006-11-23 12:01:41 +00005064
Tejun Heoc13291a2011-07-12 10:46:30 +02005065 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
5066 min_pfn = min(min_pfn, start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005067
Mel Gormana6af2bc2007-02-10 01:42:57 -08005068 if (min_pfn == ULONG_MAX) {
5069 printk(KERN_WARNING
Paul Jackson2bc0d262008-06-22 07:22:17 -07005070 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08005071 return 0;
5072 }
5073
5074 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005075}
5076
5077/**
5078 * find_min_pfn_with_active_regions - Find the minimum PFN registered
5079 *
5080 * It returns the minimum PFN based on information provided via
Zhang Zhen7d018172014-06-04 16:10:53 -07005081 * memblock_set_node().
Mel Gormanc7132162006-09-27 01:49:43 -07005082 */
5083unsigned long __init find_min_pfn_with_active_regions(void)
5084{
5085 return find_min_pfn_for_node(MAX_NUMNODES);
5086}
5087
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005088/*
5089 * early_calculate_totalpages()
5090 * Sum pages in active regions for movable zone.
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005091 * Populate N_MEMORY for calculating usable_nodes.
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005092 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07005093static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07005094{
Mel Gorman7e63efe2007-07-17 04:03:15 -07005095 unsigned long totalpages = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02005096 unsigned long start_pfn, end_pfn;
5097 int i, nid;
Mel Gorman7e63efe2007-07-17 04:03:15 -07005098
Tejun Heoc13291a2011-07-12 10:46:30 +02005099 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
5100 unsigned long pages = end_pfn - start_pfn;
5101
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005102 totalpages += pages;
5103 if (pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005104 node_set_state(nid, N_MEMORY);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005105 }
Pintu Kumarb8af2942013-09-11 14:20:34 -07005106 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07005107}
5108
Mel Gorman2a1e2742007-07-17 04:03:12 -07005109/*
5110 * Find the PFN the Movable zone begins in each node. Kernel memory
5111 * is spread evenly between nodes as long as the nodes have enough
5112 * memory. When they don't, some nodes will have more kernelcore than
5113 * others
5114 */
Kautuk Consulb224ef82012-03-21 16:34:15 -07005115static void __init find_zone_movable_pfns_for_nodes(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005116{
5117 int i, nid;
5118 unsigned long usable_startpfn;
5119 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07005120 /* save the state before borrow the nodemask */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005121 nodemask_t saved_node_state = node_states[N_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005122 unsigned long totalpages = early_calculate_totalpages();
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005123 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
Emil Medve136199f2014-04-07 15:37:52 -07005124 struct memblock_region *r;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005125
5126 /* Need to find movable_zone earlier when movable_node is specified. */
5127 find_usable_zone_for_movable();
Mel Gorman2a1e2742007-07-17 04:03:12 -07005128
Mel Gorman7e63efe2007-07-17 04:03:15 -07005129 /*
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005130 * If movable_node is specified, ignore kernelcore and movablecore
5131 * options.
5132 */
5133 if (movable_node_is_enabled()) {
Emil Medve136199f2014-04-07 15:37:52 -07005134 for_each_memblock(memory, r) {
5135 if (!memblock_is_hotpluggable(r))
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005136 continue;
5137
Emil Medve136199f2014-04-07 15:37:52 -07005138 nid = r->nid;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005139
Emil Medve136199f2014-04-07 15:37:52 -07005140 usable_startpfn = PFN_DOWN(r->base);
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005141 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
5142 min(usable_startpfn, zone_movable_pfn[nid]) :
5143 usable_startpfn;
5144 }
5145
5146 goto out2;
5147 }
5148
5149 /*
5150 * If movablecore=nn[KMG] was specified, calculate what size of
Mel Gorman7e63efe2007-07-17 04:03:15 -07005151 * kernelcore that corresponds so that memory usable for
5152 * any allocation type is evenly spread. If both kernelcore
5153 * and movablecore are specified, then the value of kernelcore
5154 * will be used for required_kernelcore if it's greater than
5155 * what movablecore would have allowed.
5156 */
5157 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07005158 unsigned long corepages;
5159
5160 /*
5161 * Round-up so that ZONE_MOVABLE is at least as large as what
5162 * was requested by the user
5163 */
5164 required_movablecore =
5165 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
5166 corepages = totalpages - required_movablecore;
5167
5168 required_kernelcore = max(required_kernelcore, corepages);
5169 }
5170
Yinghai Lu20e69262013-03-01 14:51:27 -08005171 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
5172 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07005173 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005174
5175 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
Mel Gorman2a1e2742007-07-17 04:03:12 -07005176 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
5177
5178restart:
5179 /* Spread kernelcore memory as evenly as possible throughout nodes */
5180 kernelcore_node = required_kernelcore / usable_nodes;
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005181 for_each_node_state(nid, N_MEMORY) {
Tejun Heoc13291a2011-07-12 10:46:30 +02005182 unsigned long start_pfn, end_pfn;
5183
Mel Gorman2a1e2742007-07-17 04:03:12 -07005184 /*
5185 * Recalculate kernelcore_node if the division per node
5186 * now exceeds what is necessary to satisfy the requested
5187 * amount of memory for the kernel
5188 */
5189 if (required_kernelcore < kernelcore_node)
5190 kernelcore_node = required_kernelcore / usable_nodes;
5191
5192 /*
5193 * As the map is walked, we track how much memory is usable
5194 * by the kernel using kernelcore_remaining. When it is
5195 * 0, the rest of the node is usable by ZONE_MOVABLE
5196 */
5197 kernelcore_remaining = kernelcore_node;
5198
5199 /* Go through each range of PFNs within this node */
Tejun Heoc13291a2011-07-12 10:46:30 +02005200 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005201 unsigned long size_pages;
5202
Tejun Heoc13291a2011-07-12 10:46:30 +02005203 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005204 if (start_pfn >= end_pfn)
5205 continue;
5206
5207 /* Account for what is only usable for kernelcore */
5208 if (start_pfn < usable_startpfn) {
5209 unsigned long kernel_pages;
5210 kernel_pages = min(end_pfn, usable_startpfn)
5211 - start_pfn;
5212
5213 kernelcore_remaining -= min(kernel_pages,
5214 kernelcore_remaining);
5215 required_kernelcore -= min(kernel_pages,
5216 required_kernelcore);
5217
5218 /* Continue if range is now fully accounted */
5219 if (end_pfn <= usable_startpfn) {
5220
5221 /*
5222 * Push zone_movable_pfn to the end so
5223 * that if we have to rebalance
5224 * kernelcore across nodes, we will
5225 * not double account here
5226 */
5227 zone_movable_pfn[nid] = end_pfn;
5228 continue;
5229 }
5230 start_pfn = usable_startpfn;
5231 }
5232
5233 /*
5234 * The usable PFN range for ZONE_MOVABLE is from
5235 * start_pfn->end_pfn. Calculate size_pages as the
5236 * number of pages used as kernelcore
5237 */
5238 size_pages = end_pfn - start_pfn;
5239 if (size_pages > kernelcore_remaining)
5240 size_pages = kernelcore_remaining;
5241 zone_movable_pfn[nid] = start_pfn + size_pages;
5242
5243 /*
5244 * Some kernelcore has been met, update counts and
5245 * break if the kernelcore for this node has been
Pintu Kumarb8af2942013-09-11 14:20:34 -07005246 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07005247 */
5248 required_kernelcore -= min(required_kernelcore,
5249 size_pages);
5250 kernelcore_remaining -= size_pages;
5251 if (!kernelcore_remaining)
5252 break;
5253 }
5254 }
5255
5256 /*
5257 * If there is still required_kernelcore, we do another pass with one
5258 * less node in the count. This will push zone_movable_pfn[nid] further
5259 * along on the nodes that still have memory until kernelcore is
Pintu Kumarb8af2942013-09-11 14:20:34 -07005260 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07005261 */
5262 usable_nodes--;
5263 if (usable_nodes && required_kernelcore > usable_nodes)
5264 goto restart;
5265
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005266out2:
Mel Gorman2a1e2742007-07-17 04:03:12 -07005267 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5268 for (nid = 0; nid < MAX_NUMNODES; nid++)
5269 zone_movable_pfn[nid] =
5270 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07005271
Yinghai Lu20e69262013-03-01 14:51:27 -08005272out:
Yinghai Lu66918dc2009-06-30 11:41:37 -07005273 /* restore the node_state */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005274 node_states[N_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005275}
5276
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005277/* Any regular or high memory on that node ? */
5278static void check_for_memory(pg_data_t *pgdat, int nid)
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005279{
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005280 enum zone_type zone_type;
5281
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005282 if (N_MEMORY == N_NORMAL_MEMORY)
5283 return;
5284
5285 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005286 struct zone *zone = &pgdat->node_zones[zone_type];
Xishi Qiub38a8722013-11-12 15:07:20 -08005287 if (populated_zone(zone)) {
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005288 node_set_state(nid, N_HIGH_MEMORY);
5289 if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
5290 zone_type <= ZONE_NORMAL)
5291 node_set_state(nid, N_NORMAL_MEMORY);
Bob Liud0048b02012-01-12 17:19:07 -08005292 break;
5293 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005294 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005295}
5296
Mel Gormanc7132162006-09-27 01:49:43 -07005297/**
5298 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005299 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07005300 *
5301 * This will call free_area_init_node() for each active node in the system.
Zhang Zhen7d018172014-06-04 16:10:53 -07005302 * Using the page ranges provided by memblock_set_node(), the size of each
Mel Gormanc7132162006-09-27 01:49:43 -07005303 * zone in each node and their holes is calculated. If the maximum PFN
5304 * between two adjacent zones match, it is assumed that the zone is empty.
5305 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5306 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5307 * starts where the previous one ended. For example, ZONE_DMA32 starts
5308 * at arch_max_dma_pfn.
5309 */
5310void __init free_area_init_nodes(unsigned long *max_zone_pfn)
5311{
Tejun Heoc13291a2011-07-12 10:46:30 +02005312 unsigned long start_pfn, end_pfn;
5313 int i, nid;
Mel Gormana6af2bc2007-02-10 01:42:57 -08005314
Mel Gormanc7132162006-09-27 01:49:43 -07005315 /* Record where the zone boundaries are */
5316 memset(arch_zone_lowest_possible_pfn, 0,
5317 sizeof(arch_zone_lowest_possible_pfn));
5318 memset(arch_zone_highest_possible_pfn, 0,
5319 sizeof(arch_zone_highest_possible_pfn));
5320 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
5321 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
5322 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005323 if (i == ZONE_MOVABLE)
5324 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07005325 arch_zone_lowest_possible_pfn[i] =
5326 arch_zone_highest_possible_pfn[i-1];
5327 arch_zone_highest_possible_pfn[i] =
5328 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
5329 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07005330 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
5331 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
5332
5333 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5334 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
Kautuk Consulb224ef82012-03-21 16:34:15 -07005335 find_zone_movable_pfns_for_nodes();
Mel Gormanc7132162006-09-27 01:49:43 -07005336
Mel Gormanc7132162006-09-27 01:49:43 -07005337 /* Print out the zone ranges */
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005338 printk("Zone ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005339 for (i = 0; i < MAX_NR_ZONES; i++) {
5340 if (i == ZONE_MOVABLE)
5341 continue;
Kay Sievers155cbfc2012-05-08 17:24:14 +02005342 printk(KERN_CONT " %-8s ", zone_names[i]);
David Rientjes72f0ba02010-03-05 13:42:14 -08005343 if (arch_zone_lowest_possible_pfn[i] ==
5344 arch_zone_highest_possible_pfn[i])
Kay Sievers155cbfc2012-05-08 17:24:14 +02005345 printk(KERN_CONT "empty\n");
David Rientjes72f0ba02010-03-05 13:42:14 -08005346 else
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005347 printk(KERN_CONT "[mem %0#10lx-%0#10lx]\n",
5348 arch_zone_lowest_possible_pfn[i] << PAGE_SHIFT,
5349 (arch_zone_highest_possible_pfn[i]
5350 << PAGE_SHIFT) - 1);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005351 }
5352
5353 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005354 printk("Movable zone start for each node\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005355 for (i = 0; i < MAX_NUMNODES; i++) {
5356 if (zone_movable_pfn[i])
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005357 printk(" Node %d: %#010lx\n", i,
5358 zone_movable_pfn[i] << PAGE_SHIFT);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005359 }
Mel Gormanc7132162006-09-27 01:49:43 -07005360
Wanpeng Lif2d52fe2012-10-08 16:32:24 -07005361 /* Print out the early node map */
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005362 printk("Early memory node ranges\n");
Tejun Heoc13291a2011-07-12 10:46:30 +02005363 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005364 printk(" node %3d: [mem %#010lx-%#010lx]\n", nid,
5365 start_pfn << PAGE_SHIFT, (end_pfn << PAGE_SHIFT) - 1);
Mel Gormanc7132162006-09-27 01:49:43 -07005366
5367 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07005368 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08005369 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07005370 for_each_online_node(nid) {
5371 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07005372 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07005373 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005374
5375 /* Any memory on that node */
5376 if (pgdat->node_present_pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005377 node_set_state(nid, N_MEMORY);
5378 check_for_memory(pgdat, nid);
Mel Gormanc7132162006-09-27 01:49:43 -07005379 }
5380}
Mel Gorman2a1e2742007-07-17 04:03:12 -07005381
Mel Gorman7e63efe2007-07-17 04:03:15 -07005382static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005383{
5384 unsigned long long coremem;
5385 if (!p)
5386 return -EINVAL;
5387
5388 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005389 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005390
Mel Gorman7e63efe2007-07-17 04:03:15 -07005391 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07005392 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
5393
5394 return 0;
5395}
Mel Gormaned7ed362007-07-17 04:03:14 -07005396
Mel Gorman7e63efe2007-07-17 04:03:15 -07005397/*
5398 * kernelcore=size sets the amount of memory for use for allocations that
5399 * cannot be reclaimed or migrated.
5400 */
5401static int __init cmdline_parse_kernelcore(char *p)
5402{
5403 return cmdline_parse_core(p, &required_kernelcore);
5404}
5405
5406/*
5407 * movablecore=size sets the amount of memory for use for allocations that
5408 * can be reclaimed or migrated.
5409 */
5410static int __init cmdline_parse_movablecore(char *p)
5411{
5412 return cmdline_parse_core(p, &required_movablecore);
5413}
5414
Mel Gormaned7ed362007-07-17 04:03:14 -07005415early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005416early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07005417
Tejun Heo0ee332c2011-12-08 10:22:09 -08005418#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005419
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005420void adjust_managed_page_count(struct page *page, long count)
5421{
5422 spin_lock(&managed_page_count_lock);
5423 page_zone(page)->managed_pages += count;
5424 totalram_pages += count;
Jiang Liu3dcc0572013-07-03 15:03:21 -07005425#ifdef CONFIG_HIGHMEM
5426 if (PageHighMem(page))
5427 totalhigh_pages += count;
5428#endif
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005429 spin_unlock(&managed_page_count_lock);
5430}
Jiang Liu3dcc0572013-07-03 15:03:21 -07005431EXPORT_SYMBOL(adjust_managed_page_count);
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005432
Jiang Liu11199692013-07-03 15:02:48 -07005433unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
Jiang Liu69afade2013-04-29 15:06:21 -07005434{
Jiang Liu11199692013-07-03 15:02:48 -07005435 void *pos;
5436 unsigned long pages = 0;
Jiang Liu69afade2013-04-29 15:06:21 -07005437
Jiang Liu11199692013-07-03 15:02:48 -07005438 start = (void *)PAGE_ALIGN((unsigned long)start);
5439 end = (void *)((unsigned long)end & PAGE_MASK);
5440 for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
Jiang Liudbe67df2013-07-03 15:02:51 -07005441 if ((unsigned int)poison <= 0xFF)
Jiang Liu11199692013-07-03 15:02:48 -07005442 memset(pos, poison, PAGE_SIZE);
5443 free_reserved_page(virt_to_page(pos));
Jiang Liu69afade2013-04-29 15:06:21 -07005444 }
5445
5446 if (pages && s)
Jiang Liu11199692013-07-03 15:02:48 -07005447 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
Jiang Liu69afade2013-04-29 15:06:21 -07005448 s, pages << (PAGE_SHIFT - 10), start, end);
5449
5450 return pages;
5451}
Jiang Liu11199692013-07-03 15:02:48 -07005452EXPORT_SYMBOL(free_reserved_area);
Jiang Liu69afade2013-04-29 15:06:21 -07005453
Jiang Liucfa11e02013-04-29 15:07:00 -07005454#ifdef CONFIG_HIGHMEM
5455void free_highmem_page(struct page *page)
5456{
5457 __free_reserved_page(page);
5458 totalram_pages++;
Jiang Liu7b4b2a02013-07-03 15:03:11 -07005459 page_zone(page)->managed_pages++;
Jiang Liucfa11e02013-04-29 15:07:00 -07005460 totalhigh_pages++;
5461}
5462#endif
5463
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005464
5465void __init mem_init_print_info(const char *str)
5466{
5467 unsigned long physpages, codesize, datasize, rosize, bss_size;
5468 unsigned long init_code_size, init_data_size;
5469
5470 physpages = get_num_physpages();
5471 codesize = _etext - _stext;
5472 datasize = _edata - _sdata;
5473 rosize = __end_rodata - __start_rodata;
5474 bss_size = __bss_stop - __bss_start;
5475 init_data_size = __init_end - __init_begin;
5476 init_code_size = _einittext - _sinittext;
5477
5478 /*
5479 * Detect special cases and adjust section sizes accordingly:
5480 * 1) .init.* may be embedded into .data sections
5481 * 2) .init.text.* may be out of [__init_begin, __init_end],
5482 * please refer to arch/tile/kernel/vmlinux.lds.S.
5483 * 3) .rodata.* may be embedded into .text or .data sections.
5484 */
5485#define adj_init_size(start, end, size, pos, adj) \
Pintu Kumarb8af2942013-09-11 14:20:34 -07005486 do { \
5487 if (start <= pos && pos < end && size > adj) \
5488 size -= adj; \
5489 } while (0)
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005490
5491 adj_init_size(__init_begin, __init_end, init_data_size,
5492 _sinittext, init_code_size);
5493 adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
5494 adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
5495 adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
5496 adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
5497
5498#undef adj_init_size
5499
5500 printk("Memory: %luK/%luK available "
5501 "(%luK kernel code, %luK rwdata, %luK rodata, "
5502 "%luK init, %luK bss, %luK reserved"
5503#ifdef CONFIG_HIGHMEM
5504 ", %luK highmem"
5505#endif
5506 "%s%s)\n",
5507 nr_free_pages() << (PAGE_SHIFT-10), physpages << (PAGE_SHIFT-10),
5508 codesize >> 10, datasize >> 10, rosize >> 10,
5509 (init_data_size + init_code_size) >> 10, bss_size >> 10,
5510 (physpages - totalram_pages) << (PAGE_SHIFT-10),
5511#ifdef CONFIG_HIGHMEM
5512 totalhigh_pages << (PAGE_SHIFT-10),
5513#endif
5514 str ? ", " : "", str ? str : "");
5515}
5516
Mel Gorman0e0b8642006-09-27 01:49:56 -07005517/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005518 * set_dma_reserve - set the specified number of pages reserved in the first zone
5519 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07005520 *
5521 * The per-cpu batchsize and zone watermarks are determined by present_pages.
5522 * In the DMA zone, a significant percentage may be consumed by kernel image
5523 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005524 * function may optionally be used to account for unfreeable pages in the
5525 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5526 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07005527 */
5528void __init set_dma_reserve(unsigned long new_dma_reserve)
5529{
5530 dma_reserve = new_dma_reserve;
5531}
5532
Linus Torvalds1da177e2005-04-16 15:20:36 -07005533void __init free_area_init(unsigned long *zones_size)
5534{
Johannes Weiner9109fb72008-07-23 21:27:20 -07005535 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005536 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
5537}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005538
Linus Torvalds1da177e2005-04-16 15:20:36 -07005539static int page_alloc_cpu_notify(struct notifier_block *self,
5540 unsigned long action, void *hcpu)
5541{
5542 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005543
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07005544 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -07005545 lru_add_drain_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005546 drain_pages(cpu);
5547
5548 /*
5549 * Spill the event counters of the dead processor
5550 * into the current processors event counters.
5551 * This artificially elevates the count of the current
5552 * processor.
5553 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07005554 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005555
5556 /*
5557 * Zero the differential counters of the dead processor
5558 * so that the vm statistics are consistent.
5559 *
5560 * This is only okay since the processor is dead and cannot
5561 * race with what we are doing.
5562 */
Christoph Lameter2bb921e2013-09-11 14:21:30 -07005563 cpu_vm_stats_fold(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005564 }
5565 return NOTIFY_OK;
5566}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005567
5568void __init page_alloc_init(void)
5569{
5570 hotcpu_notifier(page_alloc_cpu_notify, 0);
5571}
5572
5573/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005574 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5575 * or min_free_kbytes changes.
5576 */
5577static void calculate_totalreserve_pages(void)
5578{
5579 struct pglist_data *pgdat;
5580 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005581 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005582
5583 for_each_online_pgdat(pgdat) {
5584 for (i = 0; i < MAX_NR_ZONES; i++) {
5585 struct zone *zone = pgdat->node_zones + i;
Mel Gorman3484b2d2014-08-06 16:07:14 -07005586 long max = 0;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005587
5588 /* Find valid and maximum lowmem_reserve in the zone */
5589 for (j = i; j < MAX_NR_ZONES; j++) {
5590 if (zone->lowmem_reserve[j] > max)
5591 max = zone->lowmem_reserve[j];
5592 }
5593
Mel Gorman41858962009-06-16 15:32:12 -07005594 /* we treat the high watermark as reserved pages. */
5595 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005596
Jiang Liub40da042013-02-22 16:33:52 -08005597 if (max > zone->managed_pages)
5598 max = zone->managed_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005599 reserve_pages += max;
Johannes Weinerab8fabd2012-01-10 15:07:42 -08005600 /*
5601 * Lowmem reserves are not available to
5602 * GFP_HIGHUSER page cache allocations and
5603 * kswapd tries to balance zones to their high
5604 * watermark. As a result, neither should be
5605 * regarded as dirtyable memory, to prevent a
5606 * situation where reclaim has to clean pages
5607 * in order to balance the zones.
5608 */
5609 zone->dirty_balance_reserve = max;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005610 }
5611 }
Johannes Weinerab8fabd2012-01-10 15:07:42 -08005612 dirty_balance_reserve = reserve_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005613 totalreserve_pages = reserve_pages;
5614}
5615
5616/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07005617 * setup_per_zone_lowmem_reserve - called whenever
5618 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
5619 * has a correct pages reserved value, so an adequate number of
5620 * pages are left in the zone after a successful __alloc_pages().
5621 */
5622static void setup_per_zone_lowmem_reserve(void)
5623{
5624 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005625 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005626
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08005627 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005628 for (j = 0; j < MAX_NR_ZONES; j++) {
5629 struct zone *zone = pgdat->node_zones + j;
Jiang Liub40da042013-02-22 16:33:52 -08005630 unsigned long managed_pages = zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005631
5632 zone->lowmem_reserve[j] = 0;
5633
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005634 idx = j;
5635 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005636 struct zone *lower_zone;
5637
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005638 idx--;
5639
Linus Torvalds1da177e2005-04-16 15:20:36 -07005640 if (sysctl_lowmem_reserve_ratio[idx] < 1)
5641 sysctl_lowmem_reserve_ratio[idx] = 1;
5642
5643 lower_zone = pgdat->node_zones + idx;
Jiang Liub40da042013-02-22 16:33:52 -08005644 lower_zone->lowmem_reserve[j] = managed_pages /
Linus Torvalds1da177e2005-04-16 15:20:36 -07005645 sysctl_lowmem_reserve_ratio[idx];
Jiang Liub40da042013-02-22 16:33:52 -08005646 managed_pages += lower_zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005647 }
5648 }
5649 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005650
5651 /* update totalreserve_pages */
5652 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005653}
5654
Mel Gormancfd3da12011-04-25 21:36:42 +00005655static void __setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005656{
5657 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
5658 unsigned long lowmem_pages = 0;
5659 struct zone *zone;
5660 unsigned long flags;
5661
5662 /* Calculate total number of !ZONE_HIGHMEM pages */
5663 for_each_zone(zone) {
5664 if (!is_highmem(zone))
Jiang Liub40da042013-02-22 16:33:52 -08005665 lowmem_pages += zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005666 }
5667
5668 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07005669 u64 tmp;
5670
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005671 spin_lock_irqsave(&zone->lock, flags);
Jiang Liub40da042013-02-22 16:33:52 -08005672 tmp = (u64)pages_min * zone->managed_pages;
Andrew Mortonac924c62006-05-15 09:43:59 -07005673 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005674 if (is_highmem(zone)) {
5675 /*
Nick Piggin669ed172005-11-13 16:06:45 -08005676 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5677 * need highmem pages, so cap pages_min to a small
5678 * value here.
5679 *
Mel Gorman41858962009-06-16 15:32:12 -07005680 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Nick Piggin669ed172005-11-13 16:06:45 -08005681 * deltas controls asynch page reclaim, and so should
5682 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005683 */
Andrew Morton90ae8d62013-02-22 16:32:22 -08005684 unsigned long min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005685
Jiang Liub40da042013-02-22 16:33:52 -08005686 min_pages = zone->managed_pages / 1024;
Andrew Morton90ae8d62013-02-22 16:32:22 -08005687 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
Mel Gorman41858962009-06-16 15:32:12 -07005688 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005689 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08005690 /*
5691 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07005692 * proportionate to the zone's size.
5693 */
Mel Gorman41858962009-06-16 15:32:12 -07005694 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005695 }
5696
Mel Gorman41858962009-06-16 15:32:12 -07005697 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
5698 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Marek Szyprowski49f223a2012-01-25 12:49:24 +01005699
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07005700 __mod_zone_page_state(zone, NR_ALLOC_BATCH,
Johannes Weinerabe5f972014-10-02 16:21:10 -07005701 high_wmark_pages(zone) - low_wmark_pages(zone) -
5702 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07005703
Mel Gorman56fd56b2007-10-16 01:25:58 -07005704 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005705 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005706 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005707
5708 /* update totalreserve_pages */
5709 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005710}
5711
Mel Gormancfd3da12011-04-25 21:36:42 +00005712/**
5713 * setup_per_zone_wmarks - called when min_free_kbytes changes
5714 * or when memory is hot-{added|removed}
5715 *
5716 * Ensures that the watermark[min,low,high] values for each zone are set
5717 * correctly with respect to min_free_kbytes.
5718 */
5719void setup_per_zone_wmarks(void)
5720{
5721 mutex_lock(&zonelists_mutex);
5722 __setup_per_zone_wmarks();
5723 mutex_unlock(&zonelists_mutex);
5724}
5725
Randy Dunlap55a44622009-09-21 17:01:20 -07005726/*
Rik van Riel556adec2008-10-18 20:26:34 -07005727 * The inactive anon list should be small enough that the VM never has to
5728 * do too much work, but large enough that each inactive page has a chance
5729 * to be referenced again before it is swapped out.
5730 *
5731 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5732 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5733 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5734 * the anonymous pages are kept on the inactive list.
5735 *
5736 * total target max
5737 * memory ratio inactive anon
5738 * -------------------------------------
5739 * 10MB 1 5MB
5740 * 100MB 1 50MB
5741 * 1GB 3 250MB
5742 * 10GB 10 0.9GB
5743 * 100GB 31 3GB
5744 * 1TB 101 10GB
5745 * 10TB 320 32GB
5746 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005747static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
Minchan Kim96cb4df2009-06-16 15:32:49 -07005748{
5749 unsigned int gb, ratio;
5750
5751 /* Zone size in gigabytes */
Jiang Liub40da042013-02-22 16:33:52 -08005752 gb = zone->managed_pages >> (30 - PAGE_SHIFT);
Minchan Kim96cb4df2009-06-16 15:32:49 -07005753 if (gb)
5754 ratio = int_sqrt(10 * gb);
5755 else
5756 ratio = 1;
5757
5758 zone->inactive_ratio = ratio;
5759}
5760
KOSAKI Motohiro839a4fc2011-05-24 17:11:31 -07005761static void __meminit setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07005762{
5763 struct zone *zone;
5764
Minchan Kim96cb4df2009-06-16 15:32:49 -07005765 for_each_zone(zone)
5766 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07005767}
5768
Linus Torvalds1da177e2005-04-16 15:20:36 -07005769/*
5770 * Initialise min_free_kbytes.
5771 *
5772 * For small machines we want it small (128k min). For large machines
5773 * we want it large (64MB max). But it is not linear, because network
5774 * bandwidth does not increase linearly with machine size. We use
5775 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07005776 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
Linus Torvalds1da177e2005-04-16 15:20:36 -07005777 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5778 *
5779 * which yields
5780 *
5781 * 16MB: 512k
5782 * 32MB: 724k
5783 * 64MB: 1024k
5784 * 128MB: 1448k
5785 * 256MB: 2048k
5786 * 512MB: 2896k
5787 * 1024MB: 4096k
5788 * 2048MB: 5792k
5789 * 4096MB: 8192k
5790 * 8192MB: 11584k
5791 * 16384MB: 16384k
5792 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005793int __meminit init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005794{
5795 unsigned long lowmem_kbytes;
Michal Hocko5f127332013-07-08 16:00:40 -07005796 int new_min_free_kbytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005797
5798 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
Michal Hocko5f127332013-07-08 16:00:40 -07005799 new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005800
Michal Hocko5f127332013-07-08 16:00:40 -07005801 if (new_min_free_kbytes > user_min_free_kbytes) {
5802 min_free_kbytes = new_min_free_kbytes;
5803 if (min_free_kbytes < 128)
5804 min_free_kbytes = 128;
5805 if (min_free_kbytes > 65536)
5806 min_free_kbytes = 65536;
5807 } else {
5808 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
5809 new_min_free_kbytes, user_min_free_kbytes);
5810 }
Minchan Kimbc75d332009-06-16 15:32:48 -07005811 setup_per_zone_wmarks();
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -07005812 refresh_zone_stat_thresholds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005813 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07005814 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005815 return 0;
5816}
Minchan Kimbc75d332009-06-16 15:32:48 -07005817module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005818
5819/*
Pintu Kumarb8af2942013-09-11 14:20:34 -07005820 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
Linus Torvalds1da177e2005-04-16 15:20:36 -07005821 * that we can call two helper functions whenever min_free_kbytes
5822 * changes.
5823 */
Joe Perchescccad5b2014-06-06 14:38:09 -07005824int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005825 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005826{
Han Pingtianda8c7572014-01-23 15:53:17 -08005827 int rc;
5828
5829 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5830 if (rc)
5831 return rc;
5832
Michal Hocko5f127332013-07-08 16:00:40 -07005833 if (write) {
5834 user_min_free_kbytes = min_free_kbytes;
Minchan Kimbc75d332009-06-16 15:32:48 -07005835 setup_per_zone_wmarks();
Michal Hocko5f127332013-07-08 16:00:40 -07005836 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005837 return 0;
5838}
5839
Christoph Lameter96146342006-07-03 00:24:13 -07005840#ifdef CONFIG_NUMA
Joe Perchescccad5b2014-06-06 14:38:09 -07005841int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005842 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07005843{
5844 struct zone *zone;
5845 int rc;
5846
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005847 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07005848 if (rc)
5849 return rc;
5850
5851 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08005852 zone->min_unmapped_pages = (zone->managed_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07005853 sysctl_min_unmapped_ratio) / 100;
5854 return 0;
5855}
Christoph Lameter0ff38492006-09-25 23:31:52 -07005856
Joe Perchescccad5b2014-06-06 14:38:09 -07005857int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005858 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07005859{
5860 struct zone *zone;
5861 int rc;
5862
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005863 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07005864 if (rc)
5865 return rc;
5866
5867 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08005868 zone->min_slab_pages = (zone->managed_pages *
Christoph Lameter0ff38492006-09-25 23:31:52 -07005869 sysctl_min_slab_ratio) / 100;
5870 return 0;
5871}
Christoph Lameter96146342006-07-03 00:24:13 -07005872#endif
5873
Linus Torvalds1da177e2005-04-16 15:20:36 -07005874/*
5875 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5876 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5877 * whenever sysctl_lowmem_reserve_ratio changes.
5878 *
5879 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07005880 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07005881 * if in function of the boot time zone sizes.
5882 */
Joe Perchescccad5b2014-06-06 14:38:09 -07005883int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005884 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005885{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005886 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005887 setup_per_zone_lowmem_reserve();
5888 return 0;
5889}
5890
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005891/*
5892 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
Pintu Kumarb8af2942013-09-11 14:20:34 -07005893 * cpu. It is the fraction of total pages in each zone that a hot per cpu
5894 * pagelist can have before it gets flushed back to buddy allocator.
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005895 */
Joe Perchescccad5b2014-06-06 14:38:09 -07005896int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005897 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005898{
5899 struct zone *zone;
David Rientjes7cd2b0a2014-06-23 13:22:04 -07005900 int old_percpu_pagelist_fraction;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005901 int ret;
5902
Cody P Schaferc8e251f2013-07-03 15:01:29 -07005903 mutex_lock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07005904 old_percpu_pagelist_fraction = percpu_pagelist_fraction;
5905
5906 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
5907 if (!write || ret < 0)
5908 goto out;
5909
5910 /* Sanity checking to avoid pcp imbalance */
5911 if (percpu_pagelist_fraction &&
5912 percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
5913 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
5914 ret = -EINVAL;
5915 goto out;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005916 }
David Rientjes7cd2b0a2014-06-23 13:22:04 -07005917
5918 /* No change? */
5919 if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
5920 goto out;
5921
5922 for_each_populated_zone(zone) {
5923 unsigned int cpu;
5924
5925 for_each_possible_cpu(cpu)
5926 pageset_set_high_and_batch(zone,
5927 per_cpu_ptr(zone->pageset, cpu));
5928 }
5929out:
Cody P Schaferc8e251f2013-07-03 15:01:29 -07005930 mutex_unlock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07005931 return ret;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005932}
5933
David S. Millerf034b5d2006-08-24 03:08:07 -07005934int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005935
5936#ifdef CONFIG_NUMA
5937static int __init set_hashdist(char *str)
5938{
5939 if (!str)
5940 return 0;
5941 hashdist = simple_strtoul(str, &str, 0);
5942 return 1;
5943}
5944__setup("hashdist=", set_hashdist);
5945#endif
5946
5947/*
5948 * allocate a large system hash table from bootmem
5949 * - it is assumed that the hash table must contain an exact power-of-2
5950 * quantity of entries
5951 * - limit is the number of hash buckets, not the total allocation size
5952 */
5953void *__init alloc_large_system_hash(const char *tablename,
5954 unsigned long bucketsize,
5955 unsigned long numentries,
5956 int scale,
5957 int flags,
5958 unsigned int *_hash_shift,
5959 unsigned int *_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00005960 unsigned long low_limit,
5961 unsigned long high_limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005962{
Tim Bird31fe62b2012-05-23 13:33:35 +00005963 unsigned long long max = high_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005964 unsigned long log2qty, size;
5965 void *table = NULL;
5966
5967 /* allow the kernel cmdline to have a say */
5968 if (!numentries) {
5969 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08005970 numentries = nr_kernel_pages;
Jerry Zhoua7e83312013-09-11 14:20:26 -07005971
5972 /* It isn't necessary when PAGE_SIZE >= 1MB */
5973 if (PAGE_SHIFT < 20)
5974 numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005975
5976 /* limit to 1 bucket per 2^scale bytes of low memory */
5977 if (scale > PAGE_SHIFT)
5978 numentries >>= (scale - PAGE_SHIFT);
5979 else
5980 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08005981
5982 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07005983 if (unlikely(flags & HASH_SMALL)) {
5984 /* Makes no sense without HASH_EARLY */
5985 WARN_ON(!(flags & HASH_EARLY));
5986 if (!(numentries >> *_hash_shift)) {
5987 numentries = 1UL << *_hash_shift;
5988 BUG_ON(!numentries);
5989 }
5990 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08005991 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005992 }
John Hawkes6e692ed2006-03-25 03:08:02 -08005993 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005994
5995 /* limit allocation size to 1/16 total memory by default */
5996 if (max == 0) {
5997 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
5998 do_div(max, bucketsize);
5999 }
Dimitri Sivanich074b8512012-02-08 12:39:07 -08006000 max = min(max, 0x80000000ULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006001
Tim Bird31fe62b2012-05-23 13:33:35 +00006002 if (numentries < low_limit)
6003 numentries = low_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006004 if (numentries > max)
6005 numentries = max;
6006
David Howellsf0d1b0b2006-12-08 02:37:49 -08006007 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006008
6009 do {
6010 size = bucketsize << log2qty;
6011 if (flags & HASH_EARLY)
Santosh Shilimkar67828322014-01-21 15:50:25 -08006012 table = memblock_virt_alloc_nopanic(size, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006013 else if (hashdist)
6014 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
6015 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07006016 /*
6017 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07006018 * some pages at the end of hash table which
6019 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07006020 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01006021 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07006022 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01006023 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
6024 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006025 }
6026 } while (!table && size > PAGE_SIZE && --log2qty);
6027
6028 if (!table)
6029 panic("Failed to allocate %s hash table\n", tablename);
6030
Robin Holtf241e6602010-10-07 12:59:26 -07006031 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07006032 tablename,
Robin Holtf241e6602010-10-07 12:59:26 -07006033 (1UL << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08006034 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07006035 size);
6036
6037 if (_hash_shift)
6038 *_hash_shift = log2qty;
6039 if (_hash_mask)
6040 *_hash_mask = (1 << log2qty) - 1;
6041
6042 return table;
6043}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08006044
Mel Gorman835c1342007-10-16 01:25:47 -07006045/* Return a pointer to the bitmap storing bits affecting a block of pages */
6046static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
6047 unsigned long pfn)
6048{
6049#ifdef CONFIG_SPARSEMEM
6050 return __pfn_to_section(pfn)->pageblock_flags;
6051#else
6052 return zone->pageblock_flags;
6053#endif /* CONFIG_SPARSEMEM */
6054}
Andrew Morton6220ec72006-10-19 23:29:05 -07006055
Mel Gorman835c1342007-10-16 01:25:47 -07006056static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
6057{
6058#ifdef CONFIG_SPARSEMEM
6059 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07006060 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07006061#else
Laura Abbottc060f942013-01-11 14:31:51 -08006062 pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
Mel Gormand9c23402007-10-16 01:26:01 -07006063 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07006064#endif /* CONFIG_SPARSEMEM */
6065}
6066
6067/**
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006068 * 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 -07006069 * @page: The page within the block of interest
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006070 * @pfn: The target page frame number
6071 * @end_bitidx: The last bit of interest to retrieve
6072 * @mask: mask of bits that the caller is interested in
6073 *
6074 * Return: pageblock_bits flags
Mel Gorman835c1342007-10-16 01:25:47 -07006075 */
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006076unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
Mel Gormane58469b2014-06-04 16:10:16 -07006077 unsigned long end_bitidx,
6078 unsigned long mask)
Mel Gorman835c1342007-10-16 01:25:47 -07006079{
6080 struct zone *zone;
6081 unsigned long *bitmap;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006082 unsigned long bitidx, word_bitidx;
Mel Gormane58469b2014-06-04 16:10:16 -07006083 unsigned long word;
Mel Gorman835c1342007-10-16 01:25:47 -07006084
6085 zone = page_zone(page);
Mel Gorman835c1342007-10-16 01:25:47 -07006086 bitmap = get_pageblock_bitmap(zone, pfn);
6087 bitidx = pfn_to_bitidx(zone, pfn);
Mel Gormane58469b2014-06-04 16:10:16 -07006088 word_bitidx = bitidx / BITS_PER_LONG;
6089 bitidx &= (BITS_PER_LONG-1);
Mel Gorman835c1342007-10-16 01:25:47 -07006090
Mel Gormane58469b2014-06-04 16:10:16 -07006091 word = bitmap[word_bitidx];
6092 bitidx += end_bitidx;
6093 return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
Mel Gorman835c1342007-10-16 01:25:47 -07006094}
6095
6096/**
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006097 * 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 -07006098 * @page: The page within the block of interest
Mel Gorman835c1342007-10-16 01:25:47 -07006099 * @flags: The flags to set
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006100 * @pfn: The target page frame number
6101 * @end_bitidx: The last bit of interest
6102 * @mask: mask of bits that the caller is interested in
Mel Gorman835c1342007-10-16 01:25:47 -07006103 */
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006104void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
6105 unsigned long pfn,
Mel Gormane58469b2014-06-04 16:10:16 -07006106 unsigned long end_bitidx,
6107 unsigned long mask)
Mel Gorman835c1342007-10-16 01:25:47 -07006108{
6109 struct zone *zone;
6110 unsigned long *bitmap;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006111 unsigned long bitidx, word_bitidx;
Mel Gormane58469b2014-06-04 16:10:16 -07006112 unsigned long old_word, word;
6113
6114 BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
Mel Gorman835c1342007-10-16 01:25:47 -07006115
6116 zone = page_zone(page);
Mel Gorman835c1342007-10-16 01:25:47 -07006117 bitmap = get_pageblock_bitmap(zone, pfn);
6118 bitidx = pfn_to_bitidx(zone, pfn);
Mel Gormane58469b2014-06-04 16:10:16 -07006119 word_bitidx = bitidx / BITS_PER_LONG;
6120 bitidx &= (BITS_PER_LONG-1);
6121
Sasha Levin309381fea2014-01-23 15:52:54 -08006122 VM_BUG_ON_PAGE(!zone_spans_pfn(zone, pfn), page);
Mel Gorman835c1342007-10-16 01:25:47 -07006123
Mel Gormane58469b2014-06-04 16:10:16 -07006124 bitidx += end_bitidx;
6125 mask <<= (BITS_PER_LONG - bitidx - 1);
6126 flags <<= (BITS_PER_LONG - bitidx - 1);
6127
6128 word = ACCESS_ONCE(bitmap[word_bitidx]);
6129 for (;;) {
6130 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
6131 if (word == old_word)
6132 break;
6133 word = old_word;
6134 }
Mel Gorman835c1342007-10-16 01:25:47 -07006135}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07006136
6137/*
Minchan Kim80934512012-07-31 16:43:01 -07006138 * This function checks whether pageblock includes unmovable pages or not.
6139 * If @count is not zero, it is okay to include less @count unmovable pages
6140 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07006141 * PageLRU check without isolation or lru_lock could race so that
Minchan Kim80934512012-07-31 16:43:01 -07006142 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6143 * expect this function should be exact.
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07006144 */
Wen Congyangb023f462012-12-11 16:00:45 -08006145bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
6146 bool skip_hwpoisoned_pages)
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006147{
6148 unsigned long pfn, iter, found;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01006149 int mt;
6150
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006151 /*
6152 * For avoiding noise data, lru_add_drain_all() should be called
Minchan Kim80934512012-07-31 16:43:01 -07006153 * If ZONE_MOVABLE, the zone never contains unmovable pages
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006154 */
6155 if (zone_idx(zone) == ZONE_MOVABLE)
Minchan Kim80934512012-07-31 16:43:01 -07006156 return false;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01006157 mt = get_pageblock_migratetype(page);
6158 if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
Minchan Kim80934512012-07-31 16:43:01 -07006159 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006160
6161 pfn = page_to_pfn(page);
6162 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
6163 unsigned long check = pfn + iter;
6164
Namhyung Kim29723fc2011-02-25 14:44:25 -08006165 if (!pfn_valid_within(check))
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006166 continue;
Namhyung Kim29723fc2011-02-25 14:44:25 -08006167
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006168 page = pfn_to_page(check);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07006169
6170 /*
6171 * Hugepages are not in LRU lists, but they're movable.
6172 * We need not scan over tail pages bacause we don't
6173 * handle each tail page individually in migration.
6174 */
6175 if (PageHuge(page)) {
6176 iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
6177 continue;
6178 }
6179
Minchan Kim97d255c2012-07-31 16:42:59 -07006180 /*
6181 * We can't use page_count without pin a page
6182 * because another CPU can free compound page.
6183 * This check already skips compound tails of THP
6184 * because their page->_count is zero at all time.
6185 */
6186 if (!atomic_read(&page->_count)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006187 if (PageBuddy(page))
6188 iter += (1 << page_order(page)) - 1;
6189 continue;
6190 }
Minchan Kim97d255c2012-07-31 16:42:59 -07006191
Wen Congyangb023f462012-12-11 16:00:45 -08006192 /*
6193 * The HWPoisoned page may be not in buddy system, and
6194 * page_count() is not 0.
6195 */
6196 if (skip_hwpoisoned_pages && PageHWPoison(page))
6197 continue;
6198
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006199 if (!PageLRU(page))
6200 found++;
6201 /*
6202 * If there are RECLAIMABLE pages, we need to check it.
6203 * But now, memory offline itself doesn't call shrink_slab()
6204 * and it still to be fixed.
6205 */
6206 /*
6207 * If the page is not RAM, page_count()should be 0.
6208 * we don't need more check. This is an _used_ not-movable page.
6209 *
6210 * The problematic thing here is PG_reserved pages. PG_reserved
6211 * is set to both of a memory hole page and a _used_ kernel
6212 * page at boot.
6213 */
6214 if (found > count)
Minchan Kim80934512012-07-31 16:43:01 -07006215 return true;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006216 }
Minchan Kim80934512012-07-31 16:43:01 -07006217 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006218}
6219
6220bool is_pageblock_removable_nolock(struct page *page)
6221{
Michal Hocko656a0702012-01-20 14:33:58 -08006222 struct zone *zone;
6223 unsigned long pfn;
Michal Hocko687875f2012-01-20 14:33:55 -08006224
6225 /*
6226 * We have to be careful here because we are iterating over memory
6227 * sections which are not zone aware so we might end up outside of
6228 * the zone but still within the section.
Michal Hocko656a0702012-01-20 14:33:58 -08006229 * We have to take care about the node as well. If the node is offline
6230 * its NODE_DATA will be NULL - see page_zone.
Michal Hocko687875f2012-01-20 14:33:55 -08006231 */
Michal Hocko656a0702012-01-20 14:33:58 -08006232 if (!node_online(page_to_nid(page)))
6233 return false;
6234
6235 zone = page_zone(page);
6236 pfn = page_to_pfn(page);
Cody P Schafer108bcc92013-02-22 16:35:23 -08006237 if (!zone_spans_pfn(zone, pfn))
Michal Hocko687875f2012-01-20 14:33:55 -08006238 return false;
6239
Wen Congyangb023f462012-12-11 16:00:45 -08006240 return !has_unmovable_pages(zone, page, 0, true);
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07006241}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006242
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006243#ifdef CONFIG_CMA
6244
6245static unsigned long pfn_max_align_down(unsigned long pfn)
6246{
6247 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
6248 pageblock_nr_pages) - 1);
6249}
6250
6251static unsigned long pfn_max_align_up(unsigned long pfn)
6252{
6253 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
6254 pageblock_nr_pages));
6255}
6256
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006257/* [start, end) must belong to a single zone. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006258static int __alloc_contig_migrate_range(struct compact_control *cc,
6259 unsigned long start, unsigned long end)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006260{
6261 /* This function is based on compact_zone() from compaction.c. */
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006262 unsigned long nr_reclaimed;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006263 unsigned long pfn = start;
6264 unsigned int tries = 0;
6265 int ret = 0;
6266
Marek Szyprowskibe49a6e2012-12-12 13:51:19 -08006267 migrate_prep();
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006268
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006269 while (pfn < end || !list_empty(&cc->migratepages)) {
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006270 if (fatal_signal_pending(current)) {
6271 ret = -EINTR;
6272 break;
6273 }
6274
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006275 if (list_empty(&cc->migratepages)) {
6276 cc->nr_migratepages = 0;
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07006277 pfn = isolate_migratepages_range(cc, pfn, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006278 if (!pfn) {
6279 ret = -EINTR;
6280 break;
6281 }
6282 tries = 0;
6283 } else if (++tries == 5) {
6284 ret = ret < 0 ? ret : -EBUSY;
6285 break;
6286 }
6287
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006288 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
6289 &cc->migratepages);
6290 cc->nr_migratepages -= nr_reclaimed;
Minchan Kim02c6de82012-10-08 16:31:55 -07006291
Hugh Dickins9c620e22013-02-22 16:35:14 -08006292 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
David Rientjese0b9dae2014-06-04 16:08:28 -07006293 NULL, 0, cc->mode, MR_CMA);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006294 }
Srinivas Pandruvada2a6f5122013-02-22 16:32:09 -08006295 if (ret < 0) {
6296 putback_movable_pages(&cc->migratepages);
6297 return ret;
6298 }
6299 return 0;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006300}
6301
6302/**
6303 * alloc_contig_range() -- tries to allocate given range of pages
6304 * @start: start PFN to allocate
6305 * @end: one-past-the-last PFN to allocate
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006306 * @migratetype: migratetype of the underlaying pageblocks (either
6307 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6308 * in range must have the same migratetype and it must
6309 * be either of the two.
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006310 *
6311 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6312 * aligned, however it's the caller's responsibility to guarantee that
6313 * we are the only thread that changes migrate type of pageblocks the
6314 * pages fall in.
6315 *
6316 * The PFN range must belong to a single zone.
6317 *
6318 * Returns zero on success or negative error code. On success all
6319 * pages which PFN is in [start, end) are allocated for the caller and
6320 * need to be freed with free_contig_range().
6321 */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006322int alloc_contig_range(unsigned long start, unsigned long end,
6323 unsigned migratetype)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006324{
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006325 unsigned long outer_start, outer_end;
6326 int ret = 0, order;
6327
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006328 struct compact_control cc = {
6329 .nr_migratepages = 0,
6330 .order = -1,
6331 .zone = page_zone(pfn_to_page(start)),
David Rientjese0b9dae2014-06-04 16:08:28 -07006332 .mode = MIGRATE_SYNC,
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006333 .ignore_skip_hint = true,
6334 };
6335 INIT_LIST_HEAD(&cc.migratepages);
6336
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006337 /*
6338 * What we do here is we mark all pageblocks in range as
6339 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6340 * have different sizes, and due to the way page allocator
6341 * work, we align the range to biggest of the two pages so
6342 * that page allocator won't try to merge buddies from
6343 * different pageblocks and change MIGRATE_ISOLATE to some
6344 * other migration type.
6345 *
6346 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6347 * migrate the pages from an unaligned range (ie. pages that
6348 * we are interested in). This will put all the pages in
6349 * range back to page allocator as MIGRATE_ISOLATE.
6350 *
6351 * When this is done, we take the pages in range from page
6352 * allocator removing them from the buddy system. This way
6353 * page allocator will never consider using them.
6354 *
6355 * This lets us mark the pageblocks back as
6356 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6357 * aligned range but not in the unaligned, original range are
6358 * put back to page allocator so that buddy can use them.
6359 */
6360
6361 ret = start_isolate_page_range(pfn_max_align_down(start),
Wen Congyangb023f462012-12-11 16:00:45 -08006362 pfn_max_align_up(end), migratetype,
6363 false);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006364 if (ret)
Bob Liu86a595f2012-10-25 13:37:56 -07006365 return ret;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006366
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006367 ret = __alloc_contig_migrate_range(&cc, start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006368 if (ret)
6369 goto done;
6370
6371 /*
6372 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6373 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
6374 * more, all pages in [start, end) are free in page allocator.
6375 * What we are going to do is to allocate all pages from
6376 * [start, end) (that is remove them from page allocator).
6377 *
6378 * The only problem is that pages at the beginning and at the
6379 * end of interesting range may be not aligned with pages that
6380 * page allocator holds, ie. they can be part of higher order
6381 * pages. Because of this, we reserve the bigger range and
6382 * once this is done free the pages we are not interested in.
6383 *
6384 * We don't have to hold zone->lock here because the pages are
6385 * isolated thus they won't get removed from buddy.
6386 */
6387
6388 lru_add_drain_all();
6389 drain_all_pages();
6390
6391 order = 0;
6392 outer_start = start;
6393 while (!PageBuddy(pfn_to_page(outer_start))) {
6394 if (++order >= MAX_ORDER) {
6395 ret = -EBUSY;
6396 goto done;
6397 }
6398 outer_start &= ~0UL << order;
6399 }
6400
6401 /* Make sure the range is really isolated. */
Wen Congyangb023f462012-12-11 16:00:45 -08006402 if (test_pages_isolated(outer_start, end, false)) {
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006403 pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n",
6404 outer_start, end);
6405 ret = -EBUSY;
6406 goto done;
6407 }
6408
Marek Szyprowski49f223a2012-01-25 12:49:24 +01006409
6410 /* Grab isolated pages from freelists. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006411 outer_end = isolate_freepages_range(&cc, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006412 if (!outer_end) {
6413 ret = -EBUSY;
6414 goto done;
6415 }
6416
6417 /* Free head and tail (if any) */
6418 if (start != outer_start)
6419 free_contig_range(outer_start, start - outer_start);
6420 if (end != outer_end)
6421 free_contig_range(end, outer_end - end);
6422
6423done:
6424 undo_isolate_page_range(pfn_max_align_down(start),
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006425 pfn_max_align_up(end), migratetype);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006426 return ret;
6427}
6428
6429void free_contig_range(unsigned long pfn, unsigned nr_pages)
6430{
Marek Szyprowskibcc2b022012-12-20 15:05:18 -08006431 unsigned int count = 0;
6432
6433 for (; nr_pages--; pfn++) {
6434 struct page *page = pfn_to_page(pfn);
6435
6436 count += page_count(page) != 1;
6437 __free_page(page);
6438 }
6439 WARN(count != 0, "%d pages are still in use!\n", count);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006440}
6441#endif
6442
Jiang Liu4ed7e022012-07-31 16:43:35 -07006443#ifdef CONFIG_MEMORY_HOTPLUG
Cody P Schafer0a647f32013-07-03 15:01:33 -07006444/*
6445 * The zone indicated has a new number of managed_pages; batch sizes and percpu
6446 * page high values need to be recalulated.
6447 */
Jiang Liu4ed7e022012-07-31 16:43:35 -07006448void __meminit zone_pcp_update(struct zone *zone)
6449{
Cody P Schafer0a647f32013-07-03 15:01:33 -07006450 unsigned cpu;
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006451 mutex_lock(&pcp_batch_high_lock);
Cody P Schafer0a647f32013-07-03 15:01:33 -07006452 for_each_possible_cpu(cpu)
Cody P Schafer169f6c12013-07-03 15:01:41 -07006453 pageset_set_high_and_batch(zone,
6454 per_cpu_ptr(zone->pageset, cpu));
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006455 mutex_unlock(&pcp_batch_high_lock);
Jiang Liu4ed7e022012-07-31 16:43:35 -07006456}
6457#endif
6458
Jiang Liu340175b2012-07-31 16:43:32 -07006459void zone_pcp_reset(struct zone *zone)
6460{
6461 unsigned long flags;
Minchan Kim5a883812012-10-08 16:33:39 -07006462 int cpu;
6463 struct per_cpu_pageset *pset;
Jiang Liu340175b2012-07-31 16:43:32 -07006464
6465 /* avoid races with drain_pages() */
6466 local_irq_save(flags);
6467 if (zone->pageset != &boot_pageset) {
Minchan Kim5a883812012-10-08 16:33:39 -07006468 for_each_online_cpu(cpu) {
6469 pset = per_cpu_ptr(zone->pageset, cpu);
6470 drain_zonestat(zone, pset);
6471 }
Jiang Liu340175b2012-07-31 16:43:32 -07006472 free_percpu(zone->pageset);
6473 zone->pageset = &boot_pageset;
6474 }
6475 local_irq_restore(flags);
6476}
6477
Wen Congyang6dcd73d2012-12-11 16:01:01 -08006478#ifdef CONFIG_MEMORY_HOTREMOVE
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006479/*
6480 * All pages in the range must be isolated before calling this.
6481 */
6482void
6483__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
6484{
6485 struct page *page;
6486 struct zone *zone;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07006487 unsigned int order, i;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006488 unsigned long pfn;
6489 unsigned long flags;
6490 /* find the first valid pfn */
6491 for (pfn = start_pfn; pfn < end_pfn; pfn++)
6492 if (pfn_valid(pfn))
6493 break;
6494 if (pfn == end_pfn)
6495 return;
6496 zone = page_zone(pfn_to_page(pfn));
6497 spin_lock_irqsave(&zone->lock, flags);
6498 pfn = start_pfn;
6499 while (pfn < end_pfn) {
6500 if (!pfn_valid(pfn)) {
6501 pfn++;
6502 continue;
6503 }
6504 page = pfn_to_page(pfn);
Wen Congyangb023f462012-12-11 16:00:45 -08006505 /*
6506 * The HWPoisoned page may be not in buddy system, and
6507 * page_count() is not 0.
6508 */
6509 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
6510 pfn++;
6511 SetPageReserved(page);
6512 continue;
6513 }
6514
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006515 BUG_ON(page_count(page));
6516 BUG_ON(!PageBuddy(page));
6517 order = page_order(page);
6518#ifdef CONFIG_DEBUG_VM
6519 printk(KERN_INFO "remove from free list %lx %d %lx\n",
6520 pfn, 1 << order, end_pfn);
6521#endif
6522 list_del(&page->lru);
6523 rmv_page_order(page);
6524 zone->free_area[order].nr_free--;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006525 for (i = 0; i < (1 << order); i++)
6526 SetPageReserved((page+i));
6527 pfn += (1 << order);
6528 }
6529 spin_unlock_irqrestore(&zone->lock, flags);
6530}
6531#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01006532
6533#ifdef CONFIG_MEMORY_FAILURE
6534bool is_free_buddy_page(struct page *page)
6535{
6536 struct zone *zone = page_zone(page);
6537 unsigned long pfn = page_to_pfn(page);
6538 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07006539 unsigned int order;
Wu Fengguang8d22ba12009-12-16 12:19:58 +01006540
6541 spin_lock_irqsave(&zone->lock, flags);
6542 for (order = 0; order < MAX_ORDER; order++) {
6543 struct page *page_head = page - (pfn & ((1 << order) - 1));
6544
6545 if (PageBuddy(page_head) && page_order(page_head) >= order)
6546 break;
6547 }
6548 spin_unlock_irqrestore(&zone->lock, flags);
6549
6550 return order < MAX_ORDER;
6551}
6552#endif