blob: a2955e10171546be4c91cf38479298c37d47e2db [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/page_alloc.c
3 *
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
6 *
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
15 */
16
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/stddef.h>
18#include <linux/mm.h>
19#include <linux/swap.h>
20#include <linux/interrupt.h>
21#include <linux/pagemap.h>
KOSAKI Motohiro10ed2732008-03-04 14:28:32 -080022#include <linux/jiffies.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/bootmem.h>
Yinghai Luedbe7d22010-08-25 13:39:16 -070024#include <linux/memblock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/compiler.h>
Randy Dunlap9f158332005-09-13 01:25:16 -070026#include <linux/kernel.h>
Vegard Nossumb1eeab62008-11-25 16:55:53 +010027#include <linux/kmemcheck.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/module.h>
29#include <linux/suspend.h>
30#include <linux/pagevec.h>
31#include <linux/blkdev.h>
32#include <linux/slab.h>
Dave Hansena238ab52011-05-24 17:12:16 -070033#include <linux/ratelimit.h>
David Rientjes5a3135c22007-10-16 23:25:53 -070034#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <linux/notifier.h>
36#include <linux/topology.h>
37#include <linux/sysctl.h>
38#include <linux/cpu.h>
39#include <linux/cpuset.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070040#include <linux/memory_hotplug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include <linux/nodemask.h>
42#include <linux/vmalloc.h>
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -070043#include <linux/vmstat.h>
Christoph Lameter4be38e32006-01-06 00:11:17 -080044#include <linux/mempolicy.h>
Yasunori Goto68113782006-06-23 02:03:11 -070045#include <linux/stop_machine.h>
Mel Gormanc7132162006-09-27 01:49:43 -070046#include <linux/sort.h>
47#include <linux/pfn.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070048#include <linux/backing-dev.h>
Akinobu Mita933e3122006-12-08 02:39:45 -080049#include <linux/fault-inject.h>
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -070050#include <linux/page-isolation.h>
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070051#include <linux/page_cgroup.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070052#include <linux/debugobjects.h>
Catalin Marinasdbb1f812009-06-11 13:23:19 +010053#include <linux/kmemleak.h>
Mel Gorman56de7262010-05-24 14:32:30 -070054#include <linux/compaction.h>
Mel Gorman0d3d0622009-09-21 17:02:44 -070055#include <trace/events/kmem.h>
Wu Fengguang718a3822010-03-10 15:20:43 -080056#include <linux/ftrace_event.h>
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -070057#include <linux/memcontrol.h>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070058#include <linux/prefetch.h>
Lisa Du6e543d52013-09-11 14:22:36 -070059#include <linux/mm_inline.h>
Michal Nazarewicz041d3a82011-12-29 13:09:50 +010060#include <linux/migrate.h>
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -080061#include <linux/page-debug-flags.h>
David Rientjes949f7ec2013-04-29 15:07:48 -070062#include <linux/hugetlb.h>
Clark Williams8bd75c72013-02-07 09:47:07 -060063#include <linux/sched/rt.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
Jiang Liu7ee3d4e2013-07-03 15:03:41 -070065#include <asm/sections.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070066#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070067#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070068#include "internal.h"
69
Cody P Schaferc8e251f2013-07-03 15:01:29 -070070/* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
71static DEFINE_MUTEX(pcp_batch_high_lock);
72
Lee Schermerhorn72812012010-05-26 14:44:56 -070073#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
74DEFINE_PER_CPU(int, numa_node);
75EXPORT_PER_CPU_SYMBOL(numa_node);
76#endif
77
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070078#ifdef CONFIG_HAVE_MEMORYLESS_NODES
79/*
80 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
81 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
82 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
83 * defined in <linux/topology.h>.
84 */
85DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
86EXPORT_PER_CPU_SYMBOL(_numa_mem_);
87#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
Nick Piggin17cf4402006-03-22 00:08:41 -0800412static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
413{
414 int i;
415
Andrew Morton6626c5d2006-03-22 00:08:42 -0800416 /*
417 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
418 * and __GFP_HIGHMEM from hard or soft interrupt context.
419 */
Nick Piggin725d7042006-09-25 23:30:55 -0700420 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800421 for (i = 0; i < (1 << order); i++)
422 clear_highpage(page + i);
423}
424
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800425#ifdef CONFIG_DEBUG_PAGEALLOC
426unsigned int _debug_guardpage_minorder;
427
428static int __init debug_guardpage_minorder_setup(char *buf)
429{
430 unsigned long res;
431
432 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
433 printk(KERN_ERR "Bad debug_guardpage_minorder value\n");
434 return 0;
435 }
436 _debug_guardpage_minorder = res;
437 printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res);
438 return 0;
439}
440__setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
441
442static inline void set_page_guard_flag(struct page *page)
443{
444 __set_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags);
445}
446
447static inline void clear_page_guard_flag(struct page *page)
448{
449 __clear_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags);
450}
451#else
452static inline void set_page_guard_flag(struct page *page) { }
453static inline void clear_page_guard_flag(struct page *page) { }
454#endif
455
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700456static inline void set_page_order(struct page *page, int order)
457{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700458 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000459 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460}
461
462static inline void rmv_page_order(struct page *page)
463{
Nick Piggin676165a2006-04-10 11:21:48 +1000464 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700465 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466}
467
468/*
469 * Locate the struct page for both the matching buddy in our
470 * pair (buddy1) and the combined O(n+1) page they form (page).
471 *
472 * 1) Any buddy B1 will have an order O twin B2 which satisfies
473 * the following equation:
474 * B2 = B1 ^ (1 << O)
475 * For example, if the starting buddy (buddy2) is #8 its order
476 * 1 buddy is #10:
477 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
478 *
479 * 2) Any buddy B will have an order O+1 parent P which
480 * satisfies the following equation:
481 * P = B & ~(1 << O)
482 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200483 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485static inline unsigned long
KyongHo Cho43506fa2011-01-13 15:47:24 -0800486__find_buddy_index(unsigned long page_idx, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487{
KyongHo Cho43506fa2011-01-13 15:47:24 -0800488 return page_idx ^ (1 << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489}
490
491/*
492 * This function checks whether a page is free && is the buddy
493 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800494 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000495 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700496 * (c) a page and its buddy have the same order &&
497 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 *
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700499 * For recording whether a page is in the buddy system, we set ->_mapcount
500 * PAGE_BUDDY_MAPCOUNT_VALUE.
501 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
502 * serialized by zone->lock.
Nick Piggin676165a2006-04-10 11:21:48 +1000503 *
504 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700506static inline int page_is_buddy(struct page *page, struct page *buddy,
507 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700509 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800510 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800511
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700512 if (page_zone_id(page) != page_zone_id(buddy))
513 return 0;
514
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800515 if (page_is_guard(buddy) && page_order(buddy) == order) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800516 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800517 return 1;
518 }
519
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700520 if (PageBuddy(buddy) && page_order(buddy) == order) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800521 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700522 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000523 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700524 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525}
526
527/*
528 * Freeing function for a buddy system allocator.
529 *
530 * The concept of a buddy system is to maintain direct-mapped table
531 * (containing bit values) for memory blocks of various "orders".
532 * The bottom level table contains the map for the smallest allocatable
533 * units of memory (here, pages), and each level above it describes
534 * pairs of units from the levels below, hence, "buddies".
535 * At a high level, all that happens here is marking the table entry
536 * at the bottom level available, and propagating the changes upward
537 * as necessary, plus some accounting needed to play nicely with other
538 * parts of the VM system.
539 * At each level, we keep a list of pages, which are heads of continuous
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700540 * free pages of length of (1 << order) and marked with _mapcount
541 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
542 * field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 * So when we are allocating or freeing one, we can derive the state of the
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100544 * other. That is, if we allocate a small block, and both were
545 * free, the remainder of the region must be split into blocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 * If a block is freed, and its buddy is also free, then this
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100547 * triggers coalescing into a block of larger size.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +0100549 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 */
551
Nick Piggin48db57f2006-01-08 01:00:42 -0800552static inline void __free_one_page(struct page *page,
Mel Gormaned0ae212009-06-16 15:32:07 -0700553 struct zone *zone, unsigned int order,
554 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555{
556 unsigned long page_idx;
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700557 unsigned long combined_idx;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800558 unsigned long uninitialized_var(buddy_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700559 struct page *buddy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560
Cody P Schaferd29bb972013-02-22 16:35:25 -0800561 VM_BUG_ON(!zone_is_initialized(zone));
562
Nick Piggin224abf92006-01-06 00:11:11 -0800563 if (unlikely(PageCompound(page)))
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800564 if (unlikely(destroy_compound_page(page, order)))
565 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
Mel Gormaned0ae212009-06-16 15:32:07 -0700567 VM_BUG_ON(migratetype == -1);
568
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
570
Sasha Levin309381fea2014-01-23 15:52:54 -0800571 VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page);
572 VM_BUG_ON_PAGE(bad_range(zone, page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 while (order < MAX_ORDER-1) {
KyongHo Cho43506fa2011-01-13 15:47:24 -0800575 buddy_idx = __find_buddy_index(page_idx, order);
576 buddy = page + (buddy_idx - page_idx);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700577 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700578 break;
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800579 /*
580 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
581 * merge with it and move up one order.
582 */
583 if (page_is_guard(buddy)) {
584 clear_page_guard_flag(buddy);
585 set_page_private(page, 0);
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -0700586 __mod_zone_freepage_state(zone, 1 << order,
587 migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800588 } else {
589 list_del(&buddy->lru);
590 zone->free_area[order].nr_free--;
591 rmv_page_order(buddy);
592 }
KyongHo Cho43506fa2011-01-13 15:47:24 -0800593 combined_idx = buddy_idx & page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 page = page + (combined_idx - page_idx);
595 page_idx = combined_idx;
596 order++;
597 }
598 set_page_order(page, order);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700599
600 /*
601 * If this is not the largest possible page, check if the buddy
602 * of the next-highest order is free. If it is, it's possible
603 * that pages are being freed that will coalesce soon. In case,
604 * that is happening, add the free page to the tail of the list
605 * so it's less likely to be used soon and more likely to be merged
606 * as a higher order page
607 */
Mel Gormanb7f50cf2010-10-26 14:21:11 -0700608 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700609 struct page *higher_page, *higher_buddy;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800610 combined_idx = buddy_idx & page_idx;
611 higher_page = page + (combined_idx - page_idx);
612 buddy_idx = __find_buddy_index(combined_idx, order + 1);
Li Haifeng0ba8f2d2012-09-17 14:09:21 -0700613 higher_buddy = higher_page + (buddy_idx - combined_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700614 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
615 list_add_tail(&page->lru,
616 &zone->free_area[order].free_list[migratetype]);
617 goto out;
618 }
619 }
620
621 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
622out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 zone->free_area[order].nr_free++;
624}
625
Nick Piggin224abf92006-01-06 00:11:11 -0800626static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627{
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -0700628 const char *bad_reason = NULL;
Dave Hansenf0b791a2014-01-23 15:52:49 -0800629 unsigned long bad_flags = 0;
630
631 if (unlikely(page_mapcount(page)))
632 bad_reason = "nonzero mapcount";
633 if (unlikely(page->mapping != NULL))
634 bad_reason = "non-NULL mapping";
635 if (unlikely(atomic_read(&page->_count) != 0))
636 bad_reason = "nonzero _count";
637 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
638 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
639 bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
640 }
641 if (unlikely(mem_cgroup_bad_page_check(page)))
642 bad_reason = "cgroup check failed";
643 if (unlikely(bad_reason)) {
644 bad_page(page, bad_reason, bad_flags);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800645 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800646 }
Peter Zijlstra90572892013-10-07 11:29:20 +0100647 page_cpupid_reset_last(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800648 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
649 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
650 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651}
652
653/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700654 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700656 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 *
658 * If the zone was previously in an "all pages pinned" state then look to
659 * see if this freeing clears that state.
660 *
661 * And clear the zone's pages_scanned counter, to hold off the "all pages are
662 * pinned" detection logic.
663 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700664static void free_pcppages_bulk(struct zone *zone, int count,
665 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666{
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700667 int migratetype = 0;
Mel Gormana6f9edd2009-09-21 17:03:20 -0700668 int batch_free = 0;
Mel Gorman72853e22010-09-09 16:38:16 -0700669 int to_free = count;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700670
Nick Pigginc54ad302006-01-06 00:10:56 -0800671 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700673
Mel Gorman72853e22010-09-09 16:38:16 -0700674 while (to_free) {
Nick Piggin48db57f2006-01-08 01:00:42 -0800675 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700676 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -0800677
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700678 /*
Mel Gormana6f9edd2009-09-21 17:03:20 -0700679 * Remove pages from lists in a round-robin fashion. A
680 * batch_free count is maintained that is incremented when an
681 * empty list is encountered. This is so more pages are freed
682 * off fuller lists instead of spinning excessively around empty
683 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700684 */
685 do {
Mel Gormana6f9edd2009-09-21 17:03:20 -0700686 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700687 if (++migratetype == MIGRATE_PCPTYPES)
688 migratetype = 0;
689 list = &pcp->lists[migratetype];
690 } while (list_empty(list));
691
Namhyung Kim1d168712011-03-22 16:32:45 -0700692 /* This is the only non-empty list. Free them all. */
693 if (batch_free == MIGRATE_PCPTYPES)
694 batch_free = to_free;
695
Mel Gormana6f9edd2009-09-21 17:03:20 -0700696 do {
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -0700697 int mt; /* migratetype of the to-be-freed page */
698
Mel Gormana6f9edd2009-09-21 17:03:20 -0700699 page = list_entry(list->prev, struct page, lru);
700 /* must delete as __free_one_page list manipulates */
701 list_del(&page->lru);
Minchan Kimb12c4ad2012-10-08 16:32:08 -0700702 mt = get_freepage_migratetype(page);
Hugh Dickinsa7016232010-01-29 17:46:34 +0000703 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -0700704 __free_one_page(page, zone, 0, mt);
705 trace_mm_page_pcpu_drain(page, 0, mt);
Minchan Kim194159f2013-02-22 16:33:58 -0800706 if (likely(!is_migrate_isolate_page(page))) {
Wen Congyang97d0da22012-12-11 16:00:52 -0800707 __mod_zone_page_state(zone, NR_FREE_PAGES, 1);
708 if (is_migrate_cma(mt))
709 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, 1);
710 }
Mel Gorman72853e22010-09-09 16:38:16 -0700711 } while (--to_free && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800713 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714}
715
Mel Gormaned0ae212009-06-16 15:32:07 -0700716static void free_one_page(struct zone *zone, struct page *page, int order,
717 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800718{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700719 spin_lock(&zone->lock);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700720 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700721
Mel Gormaned0ae212009-06-16 15:32:07 -0700722 __free_one_page(page, zone, order, migratetype);
Minchan Kim194159f2013-02-22 16:33:58 -0800723 if (unlikely(!is_migrate_isolate(migratetype)))
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -0700724 __mod_zone_freepage_state(zone, 1 << order, migratetype);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700725 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800726}
727
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700728static bool free_pages_prepare(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 int i;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800731 int bad = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -0800733 trace_mm_page_free(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100734 kmemcheck_free_shadow(page, order);
735
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -0800736 if (PageAnon(page))
737 page->mapping = NULL;
738 for (i = 0; i < (1 << order); i++)
739 bad += free_pages_check(page + i);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800740 if (bad)
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700741 return false;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800742
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700743 if (!PageHighMem(page)) {
Pintu Kumarb8af2942013-09-11 14:20:34 -0700744 debug_check_no_locks_freed(page_address(page),
745 PAGE_SIZE << order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700746 debug_check_no_obj_freed(page_address(page),
747 PAGE_SIZE << order);
748 }
Nick Piggindafb1362006-10-11 01:21:30 -0700749 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800750 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700751
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700752 return true;
753}
754
755static void __free_pages_ok(struct page *page, unsigned int order)
756{
757 unsigned long flags;
Minchan Kim95e34412012-10-08 16:32:11 -0700758 int migratetype;
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700759
760 if (!free_pages_prepare(page, order))
761 return;
762
Nick Pigginc54ad302006-01-06 00:10:56 -0800763 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700764 __count_vm_events(PGFREE, 1 << order);
Minchan Kim95e34412012-10-08 16:32:11 -0700765 migratetype = get_pageblock_migratetype(page);
766 set_freepage_migratetype(page, migratetype);
767 free_one_page(page_zone(page), page, order, migratetype);
Nick Pigginc54ad302006-01-06 00:10:56 -0800768 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769}
770
Jiang Liu170a5a72013-07-03 15:03:17 -0700771void __init __free_pages_bootmem(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800772{
Johannes Weinerc3993072012-01-10 15:08:10 -0800773 unsigned int nr_pages = 1 << order;
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700774 struct page *p = page;
Johannes Weinerc3993072012-01-10 15:08:10 -0800775 unsigned int loop;
David Howellsa226f6c2006-01-06 00:11:08 -0800776
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700777 prefetchw(p);
778 for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
779 prefetchw(p + 1);
Johannes Weinerc3993072012-01-10 15:08:10 -0800780 __ClearPageReserved(p);
781 set_page_count(p, 0);
David Howellsa226f6c2006-01-06 00:11:08 -0800782 }
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700783 __ClearPageReserved(p);
784 set_page_count(p, 0);
Johannes Weinerc3993072012-01-10 15:08:10 -0800785
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700786 page_zone(page)->managed_pages += nr_pages;
Johannes Weinerc3993072012-01-10 15:08:10 -0800787 set_page_refcounted(page);
788 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800789}
790
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100791#ifdef CONFIG_CMA
Li Zhong9cf510a2013-08-23 13:52:52 +0800792/* Free whole pageblock and set its migration type to MIGRATE_CMA. */
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100793void __init init_cma_reserved_pageblock(struct page *page)
794{
795 unsigned i = pageblock_nr_pages;
796 struct page *p = page;
797
798 do {
799 __ClearPageReserved(p);
800 set_page_count(p, 0);
801 } while (++p, --i);
802
803 set_page_refcounted(page);
804 set_pageblock_migratetype(page, MIGRATE_CMA);
805 __free_pages(page, pageblock_order);
Jiang Liu3dcc0572013-07-03 15:03:21 -0700806 adjust_managed_page_count(page, pageblock_nr_pages);
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100807}
808#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809
810/*
811 * The order of subdivision here is critical for the IO subsystem.
812 * Please do not alter this order without good reasons and regression
813 * testing. Specifically, as large blocks of memory are subdivided,
814 * the order in which smaller blocks are delivered depends on the order
815 * they're subdivided in this function. This is the primary factor
816 * influencing the order in which pages are delivered to the IO
817 * subsystem according to empirical testing, and this is also justified
818 * by considering the behavior of a buddy system containing a single
819 * large block of memory acted on by a series of small allocations.
820 * This behavior is a critical factor in sglist merging's success.
821 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +0100822 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800824static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700825 int low, int high, struct free_area *area,
826 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827{
828 unsigned long size = 1 << high;
829
830 while (high > low) {
831 area--;
832 high--;
833 size >>= 1;
Sasha Levin309381fea2014-01-23 15:52:54 -0800834 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800835
836#ifdef CONFIG_DEBUG_PAGEALLOC
837 if (high < debug_guardpage_minorder()) {
838 /*
839 * Mark as guard pages (or page), that will allow to
840 * merge back to allocator when buddy will be freed.
841 * Corresponding page table entries will not be touched,
842 * pages will stay not present in virtual address space
843 */
844 INIT_LIST_HEAD(&page[size].lru);
845 set_page_guard_flag(&page[size]);
846 set_page_private(&page[size], high);
847 /* Guard pages are not available for any usage */
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -0700848 __mod_zone_freepage_state(zone, -(1 << high),
849 migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800850 continue;
851 }
852#endif
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700853 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 area->nr_free++;
855 set_page_order(&page[size], high);
856 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857}
858
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859/*
860 * This page is about to be returned from the page allocator
861 */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200862static inline int check_new_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863{
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -0700864 const char *bad_reason = NULL;
Dave Hansenf0b791a2014-01-23 15:52:49 -0800865 unsigned long bad_flags = 0;
866
867 if (unlikely(page_mapcount(page)))
868 bad_reason = "nonzero mapcount";
869 if (unlikely(page->mapping != NULL))
870 bad_reason = "non-NULL mapping";
871 if (unlikely(atomic_read(&page->_count) != 0))
872 bad_reason = "nonzero _count";
873 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
874 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
875 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
876 }
877 if (unlikely(mem_cgroup_bad_page_check(page)))
878 bad_reason = "cgroup check failed";
879 if (unlikely(bad_reason)) {
880 bad_page(page, bad_reason, bad_flags);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800881 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800882 }
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200883 return 0;
884}
885
886static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
887{
888 int i;
889
890 for (i = 0; i < (1 << order); i++) {
891 struct page *p = page + i;
892 if (unlikely(check_new_page(p)))
893 return 1;
894 }
Hugh Dickins689bceb2005-11-21 21:32:20 -0800895
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700896 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800897 set_page_refcounted(page);
Nick Piggincc102502006-12-06 20:32:00 -0800898
899 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800901
902 if (gfp_flags & __GFP_ZERO)
903 prep_zero_page(page, order, gfp_flags);
904
905 if (order && (gfp_flags & __GFP_COMP))
906 prep_compound_page(page, order);
907
Hugh Dickins689bceb2005-11-21 21:32:20 -0800908 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909}
910
Mel Gorman56fd56b2007-10-16 01:25:58 -0700911/*
912 * Go through the free lists for the given migratetype and remove
913 * the smallest available page from the freelists
914 */
Mel Gorman728ec982009-06-16 15:32:04 -0700915static inline
916struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -0700917 int migratetype)
918{
919 unsigned int current_order;
Pintu Kumarb8af2942013-09-11 14:20:34 -0700920 struct free_area *area;
Mel Gorman56fd56b2007-10-16 01:25:58 -0700921 struct page *page;
922
923 /* Find a page of the appropriate size in the preferred list */
924 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
925 area = &(zone->free_area[current_order]);
926 if (list_empty(&area->free_list[migratetype]))
927 continue;
928
929 page = list_entry(area->free_list[migratetype].next,
930 struct page, lru);
931 list_del(&page->lru);
932 rmv_page_order(page);
933 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -0700934 expand(zone, page, order, current_order, area, migratetype);
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -0700935 set_freepage_migratetype(page, migratetype);
Mel Gorman56fd56b2007-10-16 01:25:58 -0700936 return page;
937 }
938
939 return NULL;
940}
941
942
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700943/*
944 * This array describes the order lists are fallen back to when
945 * the free lists for the desirable migrate type are depleted
946 */
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100947static int fallbacks[MIGRATE_TYPES][4] = {
948 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
949 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
950#ifdef CONFIG_CMA
951 [MIGRATE_MOVABLE] = { MIGRATE_CMA, MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
952 [MIGRATE_CMA] = { MIGRATE_RESERVE }, /* Never used */
953#else
954 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
955#endif
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +0100956 [MIGRATE_RESERVE] = { MIGRATE_RESERVE }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -0800957#ifdef CONFIG_MEMORY_ISOLATION
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +0100958 [MIGRATE_ISOLATE] = { MIGRATE_RESERVE }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -0800959#endif
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700960};
961
Mel Gormanc361be52007-10-16 01:25:51 -0700962/*
963 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -0700964 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -0700965 * boundary. If alignment is required, use move_freepages_block()
966 */
Minchan Kim435b4052012-10-08 16:32:16 -0700967int move_freepages(struct zone *zone,
Adrian Bunkb69a7282008-07-23 21:28:12 -0700968 struct page *start_page, struct page *end_page,
969 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700970{
971 struct page *page;
972 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -0700973 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -0700974
975#ifndef CONFIG_HOLES_IN_ZONE
976 /*
977 * page_zone is not safe to call in this context when
978 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
979 * anyway as we check zone boundaries in move_freepages_block().
980 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -0700981 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -0700982 */
983 BUG_ON(page_zone(start_page) != page_zone(end_page));
984#endif
985
986 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -0700987 /* Make sure we are not inadvertently changing nodes */
Sasha Levin309381fea2014-01-23 15:52:54 -0800988 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
Adam Litke344c7902008-09-02 14:35:38 -0700989
Mel Gormanc361be52007-10-16 01:25:51 -0700990 if (!pfn_valid_within(page_to_pfn(page))) {
991 page++;
992 continue;
993 }
994
995 if (!PageBuddy(page)) {
996 page++;
997 continue;
998 }
999
1000 order = page_order(page);
Kirill A. Shutemov84be48d2011-03-22 16:33:41 -07001001 list_move(&page->lru,
1002 &zone->free_area[order].free_list[migratetype]);
Minchan Kim95e34412012-10-08 16:32:11 -07001003 set_freepage_migratetype(page, migratetype);
Mel Gormanc361be52007-10-16 01:25:51 -07001004 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -07001005 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -07001006 }
1007
Mel Gormand1003132007-10-16 01:26:00 -07001008 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -07001009}
1010
Minchan Kimee6f5092012-07-31 16:43:50 -07001011int move_freepages_block(struct zone *zone, struct page *page,
Linus Torvalds68e3e922012-06-03 20:05:57 -07001012 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001013{
1014 unsigned long start_pfn, end_pfn;
1015 struct page *start_page, *end_page;
1016
1017 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -07001018 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -07001019 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -07001020 end_page = start_page + pageblock_nr_pages - 1;
1021 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -07001022
1023 /* Do not cross zone boundaries */
Cody P Schafer108bcc92013-02-22 16:35:23 -08001024 if (!zone_spans_pfn(zone, start_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001025 start_page = page;
Cody P Schafer108bcc92013-02-22 16:35:23 -08001026 if (!zone_spans_pfn(zone, end_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001027 return 0;
1028
1029 return move_freepages(zone, start_page, end_page, migratetype);
1030}
1031
Mel Gorman2f66a682009-09-21 17:02:31 -07001032static void change_pageblock_range(struct page *pageblock_page,
1033 int start_order, int migratetype)
1034{
1035 int nr_pageblocks = 1 << (start_order - pageblock_order);
1036
1037 while (nr_pageblocks--) {
1038 set_pageblock_migratetype(pageblock_page, migratetype);
1039 pageblock_page += pageblock_nr_pages;
1040 }
1041}
1042
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001043/*
1044 * If breaking a large block of pages, move all free pages to the preferred
1045 * allocation list. If falling back for a reclaimable kernel allocation, be
1046 * more aggressive about taking ownership of free pages.
1047 *
1048 * On the other hand, never change migration type of MIGRATE_CMA pageblocks
1049 * nor move CMA pages to different free lists. We don't want unmovable pages
1050 * to be allocated from MIGRATE_CMA areas.
1051 *
1052 * Returns the new migratetype of the pageblock (or the same old migratetype
1053 * if it was unchanged).
1054 */
1055static int try_to_steal_freepages(struct zone *zone, struct page *page,
1056 int start_type, int fallback_type)
1057{
1058 int current_order = page_order(page);
1059
KOSAKI Motohiro0cbef292013-11-12 15:08:20 -08001060 /*
1061 * When borrowing from MIGRATE_CMA, we need to release the excess
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001062 * buddy pages to CMA itself. We also ensure the freepage_migratetype
1063 * is set to CMA so it is returned to the correct freelist in case
1064 * the page ends up being not actually allocated from the pcp lists.
KOSAKI Motohiro0cbef292013-11-12 15:08:20 -08001065 */
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001066 if (is_migrate_cma(fallback_type))
1067 return fallback_type;
1068
1069 /* Take ownership for orders >= pageblock_order */
1070 if (current_order >= pageblock_order) {
1071 change_pageblock_range(page, current_order, start_type);
1072 return start_type;
1073 }
1074
1075 if (current_order >= pageblock_order / 2 ||
1076 start_type == MIGRATE_RECLAIMABLE ||
1077 page_group_by_mobility_disabled) {
1078 int pages;
1079
1080 pages = move_freepages_block(zone, page, start_type);
1081
1082 /* Claim the whole block if over half of it is free */
1083 if (pages >= (1 << (pageblock_order-1)) ||
1084 page_group_by_mobility_disabled) {
1085
1086 set_pageblock_migratetype(page, start_type);
1087 return start_type;
1088 }
1089
1090 }
1091
1092 return fallback_type;
1093}
1094
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001095/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -07001096static inline struct page *
1097__rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001098{
Pintu Kumarb8af2942013-09-11 14:20:34 -07001099 struct free_area *area;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001100 int current_order;
1101 struct page *page;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001102 int migratetype, new_type, i;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001103
1104 /* Find the largest possible block of pages in the other list */
1105 for (current_order = MAX_ORDER-1; current_order >= order;
1106 --current_order) {
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001107 for (i = 0;; i++) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001108 migratetype = fallbacks[start_migratetype][i];
1109
Mel Gorman56fd56b2007-10-16 01:25:58 -07001110 /* MIGRATE_RESERVE handled later if necessary */
1111 if (migratetype == MIGRATE_RESERVE)
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001112 break;
Mel Gormane0104872007-10-16 01:25:53 -07001113
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001114 area = &(zone->free_area[current_order]);
1115 if (list_empty(&area->free_list[migratetype]))
1116 continue;
1117
1118 page = list_entry(area->free_list[migratetype].next,
1119 struct page, lru);
1120 area->nr_free--;
1121
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001122 new_type = try_to_steal_freepages(zone, page,
1123 start_migratetype,
1124 migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001125
1126 /* Remove the page from the freelists */
1127 list_del(&page->lru);
1128 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001129
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001130 expand(zone, page, order, current_order, area,
KOSAKI Motohiro0cbef292013-11-12 15:08:20 -08001131 new_type);
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001132 /* The freepage_migratetype may differ from pageblock's
1133 * migratetype depending on the decisions in
1134 * try_to_steal_freepages. This is OK as long as it does
1135 * not differ for MIGRATE_CMA type.
1136 */
1137 set_freepage_migratetype(page, new_type);
Mel Gormane0fff1b2009-09-21 17:02:42 -07001138
KOSAKI Motohiro52c8f6a2013-11-12 15:08:19 -08001139 trace_mm_page_alloc_extfrag(page, order, current_order,
1140 start_migratetype, migratetype, new_type);
Mel Gormane0fff1b2009-09-21 17:02:42 -07001141
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001142 return page;
1143 }
1144 }
1145
Mel Gorman728ec982009-06-16 15:32:04 -07001146 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001147}
1148
Mel Gorman56fd56b2007-10-16 01:25:58 -07001149/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 * Do the hard work of removing an element from the buddy allocator.
1151 * Call me with the zone->lock already held.
1152 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001153static struct page *__rmqueue(struct zone *zone, unsigned int order,
1154 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 struct page *page;
1157
Mel Gorman728ec982009-06-16 15:32:04 -07001158retry_reserve:
Mel Gorman56fd56b2007-10-16 01:25:58 -07001159 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160
Mel Gorman728ec982009-06-16 15:32:04 -07001161 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07001162 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001163
Mel Gorman728ec982009-06-16 15:32:04 -07001164 /*
1165 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1166 * is used because __rmqueue_smallest is an inline function
1167 * and we want just one call site
1168 */
1169 if (!page) {
1170 migratetype = MIGRATE_RESERVE;
1171 goto retry_reserve;
1172 }
1173 }
1174
Mel Gorman0d3d0622009-09-21 17:02:44 -07001175 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001176 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177}
1178
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001179/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 * Obtain a specified number of elements from the buddy allocator, all under
1181 * a single hold of the lock, for efficiency. Add them to the supplied list.
1182 * Returns the number of new pages which were placed at *list.
1183 */
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001184static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001185 unsigned long count, struct list_head *list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001186 int migratetype, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187{
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001188 int i;
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001189
Nick Pigginc54ad302006-01-06 00:10:56 -08001190 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001192 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001193 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08001195
1196 /*
1197 * Split buddy pages returned by expand() are received here
1198 * in physical page order. The page is added to the callers and
1199 * list and the list head then moves forward. From the callers
1200 * perspective, the linked list is ordered by page number in
1201 * some conditions. This is useful for IO devices that can
1202 * merge IO requests if the physical pages are ordered
1203 * properly.
1204 */
Mel Gormane084b2d2009-07-29 15:02:04 -07001205 if (likely(cold == 0))
1206 list_add(&page->lru, list);
1207 else
1208 list_add_tail(&page->lru, list);
Mel Gorman81eabcb2007-12-17 16:20:05 -08001209 list = &page->lru;
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001210 if (is_migrate_cma(get_freepage_migratetype(page)))
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07001211 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
1212 -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 }
Mel Gormanf2260e62009-06-16 15:32:13 -07001214 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08001215 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001216 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217}
1218
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001219#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001220/*
Christoph Lameter4037d452007-05-09 02:35:14 -07001221 * Called from the vmstat counter updater to drain pagesets of this
1222 * currently executing processor on remote nodes after they have
1223 * expired.
1224 *
Christoph Lameter879336c2006-03-22 00:09:08 -08001225 * Note that this function must be called with the thread pinned to
1226 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001227 */
Christoph Lameter4037d452007-05-09 02:35:14 -07001228void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001229{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001230 unsigned long flags;
Christoph Lameter4037d452007-05-09 02:35:14 -07001231 int to_drain;
Cody P Schafer998d39c2013-07-03 15:01:32 -07001232 unsigned long batch;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001233
Christoph Lameter4037d452007-05-09 02:35:14 -07001234 local_irq_save(flags);
Cody P Schafer998d39c2013-07-03 15:01:32 -07001235 batch = ACCESS_ONCE(pcp->batch);
1236 if (pcp->count >= batch)
1237 to_drain = batch;
Christoph Lameter4037d452007-05-09 02:35:14 -07001238 else
1239 to_drain = pcp->count;
KOSAKI Motohiro2a135152012-07-31 16:42:53 -07001240 if (to_drain > 0) {
1241 free_pcppages_bulk(zone, to_drain, pcp);
1242 pcp->count -= to_drain;
1243 }
Christoph Lameter4037d452007-05-09 02:35:14 -07001244 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001245}
1246#endif
1247
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001248/*
1249 * Drain pages of the indicated processor.
1250 *
1251 * The processor must either be the current processor and the
1252 * thread pinned to the current processor or a processor that
1253 * is not online.
1254 */
1255static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256{
Nick Pigginc54ad302006-01-06 00:10:56 -08001257 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07001260 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 struct per_cpu_pageset *pset;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001262 struct per_cpu_pages *pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001264 local_irq_save(flags);
1265 pset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001267 pcp = &pset->pcp;
David Rientjes2ff754f2011-01-25 15:07:23 -08001268 if (pcp->count) {
1269 free_pcppages_bulk(zone, pcp->count, pcp);
1270 pcp->count = 0;
1271 }
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001272 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 }
1274}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001276/*
1277 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1278 */
1279void drain_local_pages(void *arg)
1280{
1281 drain_pages(smp_processor_id());
1282}
1283
1284/*
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001285 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1286 *
1287 * Note that this code is protected against sending an IPI to an offline
1288 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1289 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1290 * nothing keeps CPUs from showing up after we populated the cpumask and
1291 * before the call to on_each_cpu_mask().
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001292 */
1293void drain_all_pages(void)
1294{
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001295 int cpu;
1296 struct per_cpu_pageset *pcp;
1297 struct zone *zone;
1298
1299 /*
1300 * Allocate in the BSS so we wont require allocation in
1301 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1302 */
1303 static cpumask_t cpus_with_pcps;
1304
1305 /*
1306 * We don't care about racing with CPU hotplug event
1307 * as offline notification will cause the notified
1308 * cpu to drain that CPU pcps and on_each_cpu_mask
1309 * disables preemption as part of its processing
1310 */
1311 for_each_online_cpu(cpu) {
1312 bool has_pcps = false;
1313 for_each_populated_zone(zone) {
1314 pcp = per_cpu_ptr(zone->pageset, cpu);
1315 if (pcp->pcp.count) {
1316 has_pcps = true;
1317 break;
1318 }
1319 }
1320 if (has_pcps)
1321 cpumask_set_cpu(cpu, &cpus_with_pcps);
1322 else
1323 cpumask_clear_cpu(cpu, &cpus_with_pcps);
1324 }
1325 on_each_cpu_mask(&cpus_with_pcps, drain_local_pages, NULL, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001326}
1327
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02001328#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329
1330void mark_free_pages(struct zone *zone)
1331{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001332 unsigned long pfn, max_zone_pfn;
1333 unsigned long flags;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001334 int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 struct list_head *curr;
1336
Xishi Qiu8080fc02013-09-11 14:21:45 -07001337 if (zone_is_empty(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 return;
1339
1340 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001341
Cody P Schafer108bcc92013-02-22 16:35:23 -08001342 max_zone_pfn = zone_end_pfn(zone);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001343 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1344 if (pfn_valid(pfn)) {
1345 struct page *page = pfn_to_page(pfn);
1346
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001347 if (!swsusp_page_is_forbidden(page))
1348 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001349 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001351 for_each_migratetype_order(order, t) {
1352 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001353 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001355 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1356 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001357 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001358 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001359 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 spin_unlock_irqrestore(&zone->lock, flags);
1361}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001362#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363
1364/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 * Free a 0-order page
Li Hongfc916682010-03-05 13:41:54 -08001366 * cold == 1 ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 */
Li Hongfc916682010-03-05 13:41:54 -08001368void free_hot_cold_page(struct page *page, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369{
1370 struct zone *zone = page_zone(page);
1371 struct per_cpu_pages *pcp;
1372 unsigned long flags;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001373 int migratetype;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001375 if (!free_pages_prepare(page, 0))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001376 return;
1377
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001378 migratetype = get_pageblock_migratetype(page);
Minchan Kimb12c4ad2012-10-08 16:32:08 -07001379 set_freepage_migratetype(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001381 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001382
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001383 /*
1384 * We only track unmovable, reclaimable and movable on pcp lists.
1385 * Free ISOLATE pages back to the allocator because they are being
1386 * offlined but treat RESERVE as movable pages so we can get those
1387 * areas back if necessary. Otherwise, we may have to free
1388 * excessively into the page allocator
1389 */
1390 if (migratetype >= MIGRATE_PCPTYPES) {
Minchan Kim194159f2013-02-22 16:33:58 -08001391 if (unlikely(is_migrate_isolate(migratetype))) {
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001392 free_one_page(zone, page, 0, migratetype);
1393 goto out;
1394 }
1395 migratetype = MIGRATE_MOVABLE;
1396 }
1397
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001398 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001399 if (cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001400 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001401 else
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001402 list_add(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001404 if (pcp->count >= pcp->high) {
Cody P Schafer998d39c2013-07-03 15:01:32 -07001405 unsigned long batch = ACCESS_ONCE(pcp->batch);
1406 free_pcppages_bulk(zone, batch, pcp);
1407 pcp->count -= batch;
Nick Piggin48db57f2006-01-08 01:00:42 -08001408 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001409
1410out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412}
1413
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001414/*
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001415 * Free a list of 0-order pages
1416 */
1417void free_hot_cold_page_list(struct list_head *list, int cold)
1418{
1419 struct page *page, *next;
1420
1421 list_for_each_entry_safe(page, next, list, lru) {
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -08001422 trace_mm_page_free_batched(page, cold);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001423 free_hot_cold_page(page, cold);
1424 }
1425}
1426
1427/*
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001428 * split_page takes a non-compound higher-order page, and splits it into
1429 * n (1<<order) sub-pages: page[0..n]
1430 * Each sub-page must be freed individually.
1431 *
1432 * Note: this is probably too low level an operation for use in drivers.
1433 * Please consult with lkml before using this in your driver.
1434 */
1435void split_page(struct page *page, unsigned int order)
1436{
1437 int i;
1438
Sasha Levin309381fea2014-01-23 15:52:54 -08001439 VM_BUG_ON_PAGE(PageCompound(page), page);
1440 VM_BUG_ON_PAGE(!page_count(page), page);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001441
1442#ifdef CONFIG_KMEMCHECK
1443 /*
1444 * Split shadow pages too, because free(page[0]) would
1445 * otherwise free the whole shadow.
1446 */
1447 if (kmemcheck_page_is_tracked(page))
1448 split_page(virt_to_page(page[0].shadow), order);
1449#endif
1450
Nick Piggin7835e982006-03-22 00:08:40 -08001451 for (i = 1; i < (1 << order); i++)
1452 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001453}
K. Y. Srinivasan5853ff22013-03-25 15:47:38 -07001454EXPORT_SYMBOL_GPL(split_page);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001455
Mel Gorman8fb74b92013-01-11 14:32:16 -08001456static int __isolate_free_page(struct page *page, unsigned int order)
Mel Gorman748446b2010-05-24 14:32:27 -07001457{
Mel Gorman748446b2010-05-24 14:32:27 -07001458 unsigned long watermark;
1459 struct zone *zone;
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07001460 int mt;
Mel Gorman748446b2010-05-24 14:32:27 -07001461
1462 BUG_ON(!PageBuddy(page));
1463
1464 zone = page_zone(page);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001465 mt = get_pageblock_migratetype(page);
Mel Gorman748446b2010-05-24 14:32:27 -07001466
Minchan Kim194159f2013-02-22 16:33:58 -08001467 if (!is_migrate_isolate(mt)) {
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001468 /* Obey watermarks as if the page was being allocated */
1469 watermark = low_wmark_pages(zone) + (1 << order);
1470 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1471 return 0;
1472
Mel Gorman8fb74b92013-01-11 14:32:16 -08001473 __mod_zone_freepage_state(zone, -(1UL << order), mt);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001474 }
Mel Gorman748446b2010-05-24 14:32:27 -07001475
1476 /* Remove page from free list */
1477 list_del(&page->lru);
1478 zone->free_area[order].nr_free--;
1479 rmv_page_order(page);
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07001480
Mel Gorman8fb74b92013-01-11 14:32:16 -08001481 /* Set the pageblock if the isolated page is at least a pageblock */
Mel Gorman748446b2010-05-24 14:32:27 -07001482 if (order >= pageblock_order - 1) {
1483 struct page *endpage = page + (1 << order) - 1;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001484 for (; page < endpage; page += pageblock_nr_pages) {
1485 int mt = get_pageblock_migratetype(page);
Minchan Kim194159f2013-02-22 16:33:58 -08001486 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001487 set_pageblock_migratetype(page,
1488 MIGRATE_MOVABLE);
1489 }
Mel Gorman748446b2010-05-24 14:32:27 -07001490 }
1491
Mel Gorman8fb74b92013-01-11 14:32:16 -08001492 return 1UL << order;
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001493}
1494
1495/*
1496 * Similar to split_page except the page is already free. As this is only
1497 * being used for migration, the migratetype of the block also changes.
1498 * As this is called with interrupts disabled, the caller is responsible
1499 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1500 * are enabled.
1501 *
1502 * Note: this is probably too low level an operation for use in drivers.
1503 * Please consult with lkml before using this in your driver.
1504 */
1505int split_free_page(struct page *page)
1506{
1507 unsigned int order;
1508 int nr_pages;
1509
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001510 order = page_order(page);
1511
Mel Gorman8fb74b92013-01-11 14:32:16 -08001512 nr_pages = __isolate_free_page(page, order);
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001513 if (!nr_pages)
1514 return 0;
1515
1516 /* Split into individual pages */
1517 set_page_refcounted(page);
1518 split_page(page, order);
1519 return nr_pages;
Mel Gorman748446b2010-05-24 14:32:27 -07001520}
1521
1522/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1524 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1525 * or two.
1526 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07001527static inline
1528struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001529 struct zone *zone, int order, gfp_t gfp_flags,
1530 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531{
1532 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001533 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 int cold = !!(gfp_flags & __GFP_COLD);
1535
Hugh Dickins689bceb2005-11-21 21:32:20 -08001536again:
Nick Piggin48db57f2006-01-08 01:00:42 -08001537 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001539 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 local_irq_save(flags);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001542 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1543 list = &pcp->lists[migratetype];
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001544 if (list_empty(list)) {
Mel Gorman535131e62007-10-16 01:25:49 -07001545 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001546 pcp->batch, list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001547 migratetype, cold);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001548 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07001549 goto failed;
Mel Gorman535131e62007-10-16 01:25:49 -07001550 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001551
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001552 if (cold)
1553 page = list_entry(list->prev, struct page, lru);
1554 else
1555 page = list_entry(list->next, struct page, lru);
1556
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001557 list_del(&page->lru);
1558 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001559 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07001560 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1561 /*
1562 * __GFP_NOFAIL is not to be used in new code.
1563 *
1564 * All __GFP_NOFAIL callers should be fixed so that they
1565 * properly detect and handle allocation failures.
1566 *
1567 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07001568 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07001569 * __GFP_NOFAIL.
1570 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07001571 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07001572 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001574 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001575 spin_unlock(&zone->lock);
1576 if (!page)
1577 goto failed;
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07001578 __mod_zone_freepage_state(zone, -(1 << order),
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001579 get_freepage_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 }
1581
Johannes Weiner3a025762014-04-07 15:37:48 -07001582 __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
Johannes Weiner27329362014-03-03 15:38:41 -08001583
Christoph Lameterf8891e52006-06-30 01:55:45 -07001584 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Andi Kleen78afd562011-03-22 16:33:12 -07001585 zone_statistics(preferred_zone, zone, gfp_flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001586 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587
Sasha Levin309381fea2014-01-23 15:52:54 -08001588 VM_BUG_ON_PAGE(bad_range(zone, page), page);
Nick Piggin17cf4402006-03-22 00:08:41 -08001589 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001590 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001592
1593failed:
1594 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001595 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596}
1597
Akinobu Mita933e3122006-12-08 02:39:45 -08001598#ifdef CONFIG_FAIL_PAGE_ALLOC
1599
Akinobu Mitab2588c42011-07-26 16:09:03 -07001600static struct {
Akinobu Mita933e3122006-12-08 02:39:45 -08001601 struct fault_attr attr;
1602
1603 u32 ignore_gfp_highmem;
1604 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001605 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001606} fail_page_alloc = {
1607 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001608 .ignore_gfp_wait = 1,
1609 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001610 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001611};
1612
1613static int __init setup_fail_page_alloc(char *str)
1614{
1615 return setup_fault_attr(&fail_page_alloc.attr, str);
1616}
1617__setup("fail_page_alloc=", setup_fail_page_alloc);
1618
Gavin Shandeaf3862012-07-31 16:41:51 -07001619static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08001620{
Akinobu Mita54114992007-07-15 23:40:23 -07001621 if (order < fail_page_alloc.min_order)
Gavin Shandeaf3862012-07-31 16:41:51 -07001622 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001623 if (gfp_mask & __GFP_NOFAIL)
Gavin Shandeaf3862012-07-31 16:41:51 -07001624 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001625 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
Gavin Shandeaf3862012-07-31 16:41:51 -07001626 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001627 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
Gavin Shandeaf3862012-07-31 16:41:51 -07001628 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001629
1630 return should_fail(&fail_page_alloc.attr, 1 << order);
1631}
1632
1633#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1634
1635static int __init fail_page_alloc_debugfs(void)
1636{
Al Virof4ae40a2011-07-24 04:33:43 -04001637 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
Akinobu Mita933e3122006-12-08 02:39:45 -08001638 struct dentry *dir;
Akinobu Mita933e3122006-12-08 02:39:45 -08001639
Akinobu Mitadd48c082011-08-03 16:21:01 -07001640 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
1641 &fail_page_alloc.attr);
1642 if (IS_ERR(dir))
1643 return PTR_ERR(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08001644
Akinobu Mitab2588c42011-07-26 16:09:03 -07001645 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
1646 &fail_page_alloc.ignore_gfp_wait))
1647 goto fail;
1648 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1649 &fail_page_alloc.ignore_gfp_highmem))
1650 goto fail;
1651 if (!debugfs_create_u32("min-order", mode, dir,
1652 &fail_page_alloc.min_order))
1653 goto fail;
Akinobu Mita933e3122006-12-08 02:39:45 -08001654
Akinobu Mitab2588c42011-07-26 16:09:03 -07001655 return 0;
1656fail:
Akinobu Mitadd48c082011-08-03 16:21:01 -07001657 debugfs_remove_recursive(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08001658
Akinobu Mitab2588c42011-07-26 16:09:03 -07001659 return -ENOMEM;
Akinobu Mita933e3122006-12-08 02:39:45 -08001660}
1661
1662late_initcall(fail_page_alloc_debugfs);
1663
1664#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1665
1666#else /* CONFIG_FAIL_PAGE_ALLOC */
1667
Gavin Shandeaf3862012-07-31 16:41:51 -07001668static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08001669{
Gavin Shandeaf3862012-07-31 16:41:51 -07001670 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001671}
1672
1673#endif /* CONFIG_FAIL_PAGE_ALLOC */
1674
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675/*
Mel Gorman88f5acf2011-01-13 15:45:41 -08001676 * Return true if free pages are above 'mark'. This takes into account the order
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 * of the allocation.
1678 */
Mel Gorman88f5acf2011-01-13 15:45:41 -08001679static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1680 int classzone_idx, int alloc_flags, long free_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681{
1682 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001683 long min = mark;
Minchan Kim2cfed072012-07-31 16:43:53 -07001684 long lowmem_reserve = z->lowmem_reserve[classzone_idx];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 int o;
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07001686 long free_cma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687
Michal Hockodf0a6da2012-01-10 15:08:02 -08001688 free_pages -= (1 << order) - 1;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001689 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001691 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 min -= min / 4;
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07001693#ifdef CONFIG_CMA
1694 /* If allocation can't use CMA areas don't use free CMA pages */
1695 if (!(alloc_flags & ALLOC_CMA))
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07001696 free_cma = zone_page_state(z, NR_FREE_CMA_PAGES);
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07001697#endif
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07001698
1699 if (free_pages - free_cma <= min + lowmem_reserve)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001700 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 for (o = 0; o < order; o++) {
1702 /* At the next order, this order's pages become unavailable */
1703 free_pages -= z->free_area[o].nr_free << o;
1704
1705 /* Require fewer higher order pages to be free */
1706 min >>= 1;
1707
1708 if (free_pages <= min)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001709 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 }
Mel Gorman88f5acf2011-01-13 15:45:41 -08001711 return true;
1712}
1713
1714bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1715 int classzone_idx, int alloc_flags)
1716{
1717 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1718 zone_page_state(z, NR_FREE_PAGES));
1719}
1720
1721bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
1722 int classzone_idx, int alloc_flags)
1723{
1724 long free_pages = zone_page_state(z, NR_FREE_PAGES);
1725
1726 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1727 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1728
1729 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1730 free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731}
1732
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001733#ifdef CONFIG_NUMA
1734/*
1735 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1736 * skip over zones that are not allowed by the cpuset, or that have
1737 * been recently (in last second) found to be nearly full. See further
1738 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001739 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001740 *
Zhi Yong Wua1aeb652013-11-12 15:08:29 -08001741 * If the zonelist cache is present in the passed zonelist, then
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001742 * returns a pointer to the allowed node mask (either the current
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08001743 * tasks mems_allowed, or node_states[N_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001744 *
1745 * If the zonelist cache is not available for this zonelist, does
1746 * nothing and returns NULL.
1747 *
1748 * If the fullzones BITMAP in the zonelist cache is stale (more than
1749 * a second since last zap'd) then we zap it out (clear its bits.)
1750 *
1751 * We hold off even calling zlc_setup, until after we've checked the
1752 * first zone in the zonelist, on the theory that most allocations will
1753 * be satisfied from that first zone, so best to examine that zone as
1754 * quickly as we can.
1755 */
1756static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1757{
1758 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1759 nodemask_t *allowednodes; /* zonelist_cache approximation */
1760
1761 zlc = zonelist->zlcache_ptr;
1762 if (!zlc)
1763 return NULL;
1764
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001765 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001766 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1767 zlc->last_full_zap = jiffies;
1768 }
1769
1770 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1771 &cpuset_current_mems_allowed :
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08001772 &node_states[N_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001773 return allowednodes;
1774}
1775
1776/*
1777 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1778 * if it is worth looking at further for free memory:
1779 * 1) Check that the zone isn't thought to be full (doesn't have its
1780 * bit set in the zonelist_cache fullzones BITMAP).
1781 * 2) Check that the zones node (obtained from the zonelist_cache
1782 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1783 * Return true (non-zero) if zone is worth looking at further, or
1784 * else return false (zero) if it is not.
1785 *
1786 * This check -ignores- the distinction between various watermarks,
1787 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1788 * found to be full for any variation of these watermarks, it will
1789 * be considered full for up to one second by all requests, unless
1790 * we are so low on memory on all allowed nodes that we are forced
1791 * into the second scan of the zonelist.
1792 *
1793 * In the second scan we ignore this zonelist cache and exactly
1794 * apply the watermarks to all zones, even it is slower to do so.
1795 * We are low on memory in the second scan, and should leave no stone
1796 * unturned looking for a free page.
1797 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001798static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001799 nodemask_t *allowednodes)
1800{
1801 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1802 int i; /* index of *z in zonelist zones */
1803 int n; /* node that zone *z is on */
1804
1805 zlc = zonelist->zlcache_ptr;
1806 if (!zlc)
1807 return 1;
1808
Mel Gormandd1a2392008-04-28 02:12:17 -07001809 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001810 n = zlc->z_to_n[i];
1811
1812 /* This zone is worth trying if it is allowed but not full */
1813 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1814}
1815
1816/*
1817 * Given 'z' scanning a zonelist, set the corresponding bit in
1818 * zlc->fullzones, so that subsequent attempts to allocate a page
1819 * from that zone don't waste time re-examining it.
1820 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001821static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001822{
1823 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1824 int i; /* index of *z in zonelist zones */
1825
1826 zlc = zonelist->zlcache_ptr;
1827 if (!zlc)
1828 return;
1829
Mel Gormandd1a2392008-04-28 02:12:17 -07001830 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001831
1832 set_bit(i, zlc->fullzones);
1833}
1834
Mel Gorman76d3fbf2011-07-25 17:12:30 -07001835/*
1836 * clear all zones full, called after direct reclaim makes progress so that
1837 * a zone that was recently full is not skipped over for up to a second
1838 */
1839static void zlc_clear_zones_full(struct zonelist *zonelist)
1840{
1841 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1842
1843 zlc = zonelist->zlcache_ptr;
1844 if (!zlc)
1845 return;
1846
1847 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1848}
1849
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001850static bool zone_local(struct zone *local_zone, struct zone *zone)
1851{
Johannes Weinerfff40682013-12-20 14:54:12 +00001852 return local_zone->node == zone->node;
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001853}
1854
David Rientjes957f8222012-10-08 16:33:24 -07001855static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
1856{
Mel Gorman5f7a75a2014-06-04 16:07:15 -07001857 return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <
1858 RECLAIM_DISTANCE;
David Rientjes957f8222012-10-08 16:33:24 -07001859}
1860
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001861#else /* CONFIG_NUMA */
1862
1863static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1864{
1865 return NULL;
1866}
1867
Mel Gormandd1a2392008-04-28 02:12:17 -07001868static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001869 nodemask_t *allowednodes)
1870{
1871 return 1;
1872}
1873
Mel Gormandd1a2392008-04-28 02:12:17 -07001874static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001875{
1876}
Mel Gorman76d3fbf2011-07-25 17:12:30 -07001877
1878static void zlc_clear_zones_full(struct zonelist *zonelist)
1879{
1880}
David Rientjes957f8222012-10-08 16:33:24 -07001881
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001882static bool zone_local(struct zone *local_zone, struct zone *zone)
1883{
1884 return true;
1885}
1886
David Rientjes957f8222012-10-08 16:33:24 -07001887static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
1888{
1889 return true;
1890}
1891
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001892#endif /* CONFIG_NUMA */
1893
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001894/*
Paul Jackson0798e512006-12-06 20:31:38 -08001895 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001896 * a page.
1897 */
1898static struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001899get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
Mel Gorman5117f452009-06-16 15:31:59 -07001900 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
Mel Gorman3dd28262009-06-16 15:32:00 -07001901 struct zone *preferred_zone, int migratetype)
Martin Hicks753ee722005-06-21 17:14:41 -07001902{
Mel Gormandd1a2392008-04-28 02:12:17 -07001903 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001904 struct page *page = NULL;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001905 int classzone_idx;
Mel Gorman5117f452009-06-16 15:31:59 -07001906 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001907 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1908 int zlc_active = 0; /* set if using zonelist_cache */
1909 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gorman54a6eb52008-04-28 02:12:16 -07001910
Mel Gorman19770b32008-04-28 02:12:18 -07001911 classzone_idx = zone_idx(preferred_zone);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001912zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001913 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001914 * Scan zonelist, looking for a zone with enough free.
SeungHun Lee3b11f0a2013-09-11 14:22:23 -07001915 * See also __cpuset_node_allowed_softwall() comment in kernel/cpuset.c.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001916 */
Mel Gorman19770b32008-04-28 02:12:18 -07001917 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1918 high_zoneidx, nodemask) {
Johannes Weinere085dbc2013-09-11 14:20:46 -07001919 unsigned long mark;
1920
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08001921 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001922 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1923 continue;
Mel Gorman664eedd2014-06-04 16:10:08 -07001924 if (cpusets_enabled() &&
1925 (alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001926 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Mel Gormancd38b112011-07-25 17:12:29 -07001927 continue;
Johannes Weinere085dbc2013-09-11 14:20:46 -07001928 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
Joonsoo Kime66f0972013-09-11 14:21:26 -07001929 if (unlikely(alloc_flags & ALLOC_NO_WATERMARKS))
Johannes Weinere085dbc2013-09-11 14:20:46 -07001930 goto try_this_zone;
Johannes Weinera756cf52012-01-10 15:07:49 -08001931 /*
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001932 * Distribute pages in proportion to the individual
1933 * zone size to ensure fair page aging. The zone a
1934 * page was allocated in should have no effect on the
1935 * time the page has in memory before being reclaimed.
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001936 */
Johannes Weiner3a025762014-04-07 15:37:48 -07001937 if (alloc_flags & ALLOC_FAIR) {
Johannes Weinerfff40682013-12-20 14:54:12 +00001938 if (!zone_local(preferred_zone, zone))
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001939 continue;
Johannes Weiner3a025762014-04-07 15:37:48 -07001940 if (zone_page_state(zone, NR_ALLOC_BATCH) <= 0)
1941 continue;
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001942 }
1943 /*
Johannes Weinera756cf52012-01-10 15:07:49 -08001944 * When allocating a page cache page for writing, we
1945 * want to get it from a zone that is within its dirty
1946 * limit, such that no single zone holds more than its
1947 * proportional share of globally allowed dirty pages.
1948 * The dirty limits take into account the zone's
1949 * lowmem reserves and high watermark so that kswapd
1950 * should be able to balance it without having to
1951 * write pages from its LRU list.
1952 *
1953 * This may look like it could increase pressure on
1954 * lower zones by failing allocations in higher zones
1955 * before they are full. But the pages that do spill
1956 * over are limited as the lower zones are protected
1957 * by this very same mechanism. It should not become
1958 * a practical burden to them.
1959 *
1960 * XXX: For now, allow allocations to potentially
1961 * exceed the per-zone dirty limit in the slowpath
1962 * (ALLOC_WMARK_LOW unset) before going into reclaim,
1963 * which is important when on a NUMA setup the allowed
1964 * zones are together not big enough to reach the
1965 * global limit. The proper fix for these situations
1966 * will require awareness of zones in the
1967 * dirty-throttling and the flusher threads.
1968 */
1969 if ((alloc_flags & ALLOC_WMARK_LOW) &&
1970 (gfp_mask & __GFP_WRITE) && !zone_dirty_ok(zone))
Mel Gorman800a1e72014-06-04 16:10:06 -07001971 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001972
Johannes Weinere085dbc2013-09-11 14:20:46 -07001973 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
1974 if (!zone_watermark_ok(zone, order, mark,
1975 classzone_idx, alloc_flags)) {
Mel Gormanfa5e0842009-06-16 15:33:22 -07001976 int ret;
1977
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08001978 if (IS_ENABLED(CONFIG_NUMA) &&
1979 !did_zlc_setup && nr_online_nodes > 1) {
Mel Gormancd38b112011-07-25 17:12:29 -07001980 /*
1981 * we do zlc_setup if there are multiple nodes
1982 * and before considering the first zone allowed
1983 * by the cpuset.
1984 */
1985 allowednodes = zlc_setup(zonelist, alloc_flags);
1986 zlc_active = 1;
1987 did_zlc_setup = 1;
1988 }
1989
David Rientjes957f8222012-10-08 16:33:24 -07001990 if (zone_reclaim_mode == 0 ||
1991 !zone_allows_reclaim(preferred_zone, zone))
Mel Gormanfa5e0842009-06-16 15:33:22 -07001992 goto this_zone_full;
1993
Mel Gormancd38b112011-07-25 17:12:29 -07001994 /*
1995 * As we may have just activated ZLC, check if the first
1996 * eligible zone has failed zone_reclaim recently.
1997 */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08001998 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
Mel Gormancd38b112011-07-25 17:12:29 -07001999 !zlc_zone_worth_trying(zonelist, z, allowednodes))
2000 continue;
2001
Mel Gormanfa5e0842009-06-16 15:33:22 -07002002 ret = zone_reclaim(zone, gfp_mask, order);
2003 switch (ret) {
2004 case ZONE_RECLAIM_NOSCAN:
2005 /* did not scan */
Mel Gormancd38b112011-07-25 17:12:29 -07002006 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002007 case ZONE_RECLAIM_FULL:
2008 /* scanned but unreclaimable */
Mel Gormancd38b112011-07-25 17:12:29 -07002009 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002010 default:
2011 /* did we reclaim enough */
Mel Gormanfed27192013-04-29 15:07:57 -07002012 if (zone_watermark_ok(zone, order, mark,
Mel Gormanfa5e0842009-06-16 15:33:22 -07002013 classzone_idx, alloc_flags))
Mel Gormanfed27192013-04-29 15:07:57 -07002014 goto try_this_zone;
2015
2016 /*
2017 * Failed to reclaim enough to meet watermark.
2018 * Only mark the zone full if checking the min
2019 * watermark or if we failed to reclaim just
2020 * 1<<order pages or else the page allocator
2021 * fastpath will prematurely mark zones full
2022 * when the watermark is between the low and
2023 * min watermarks.
2024 */
2025 if (((alloc_flags & ALLOC_WMARK_MASK) == ALLOC_WMARK_MIN) ||
2026 ret == ZONE_RECLAIM_SOME)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002027 goto this_zone_full;
Mel Gormanfed27192013-04-29 15:07:57 -07002028
2029 continue;
Paul Jackson0798e512006-12-06 20:31:38 -08002030 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002031 }
2032
Mel Gormanfa5e0842009-06-16 15:33:22 -07002033try_this_zone:
Mel Gorman3dd28262009-06-16 15:32:00 -07002034 page = buffered_rmqueue(preferred_zone, zone, order,
2035 gfp_mask, migratetype);
Paul Jackson0798e512006-12-06 20:31:38 -08002036 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002037 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002038this_zone_full:
Mel Gorman65bb3712014-06-04 16:10:05 -07002039 if (IS_ENABLED(CONFIG_NUMA) && zlc_active)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002040 zlc_mark_zone_full(zonelist, z);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002041 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002042
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002043 if (unlikely(IS_ENABLED(CONFIG_NUMA) && page == NULL && zlc_active)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002044 /* Disable zlc cache for second zonelist scan */
2045 zlc_active = 0;
2046 goto zonelist_scan;
2047 }
Alex Shib1211862012-08-21 16:16:08 -07002048
2049 if (page)
2050 /*
2051 * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was
2052 * necessary to allocate the page. The expectation is
2053 * that the caller is taking steps that will free more
2054 * memory. The caller should avoid the page being used
2055 * for !PFMEMALLOC purposes.
2056 */
2057 page->pfmemalloc = !!(alloc_flags & ALLOC_NO_WATERMARKS);
2058
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002059 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07002060}
2061
David Rientjes29423e72011-03-22 16:30:47 -07002062/*
2063 * Large machines with many possible nodes should not always dump per-node
2064 * meminfo in irq context.
2065 */
2066static inline bool should_suppress_show_mem(void)
2067{
2068 bool ret = false;
2069
2070#if NODES_SHIFT > 8
2071 ret = in_interrupt();
2072#endif
2073 return ret;
2074}
2075
Dave Hansena238ab52011-05-24 17:12:16 -07002076static DEFINE_RATELIMIT_STATE(nopage_rs,
2077 DEFAULT_RATELIMIT_INTERVAL,
2078 DEFAULT_RATELIMIT_BURST);
2079
2080void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
2081{
Dave Hansena238ab52011-05-24 17:12:16 -07002082 unsigned int filter = SHOW_MEM_FILTER_NODES;
2083
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -08002084 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2085 debug_guardpage_minorder() > 0)
Dave Hansena238ab52011-05-24 17:12:16 -07002086 return;
2087
2088 /*
2089 * This documents exceptions given to allocations in certain
2090 * contexts that are allowed to allocate outside current's set
2091 * of allowed nodes.
2092 */
2093 if (!(gfp_mask & __GFP_NOMEMALLOC))
2094 if (test_thread_flag(TIF_MEMDIE) ||
2095 (current->flags & (PF_MEMALLOC | PF_EXITING)))
2096 filter &= ~SHOW_MEM_FILTER_NODES;
2097 if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
2098 filter &= ~SHOW_MEM_FILTER_NODES;
2099
2100 if (fmt) {
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002101 struct va_format vaf;
2102 va_list args;
2103
Dave Hansena238ab52011-05-24 17:12:16 -07002104 va_start(args, fmt);
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002105
2106 vaf.fmt = fmt;
2107 vaf.va = &args;
2108
2109 pr_warn("%pV", &vaf);
2110
Dave Hansena238ab52011-05-24 17:12:16 -07002111 va_end(args);
2112 }
2113
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002114 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2115 current->comm, order, gfp_mask);
Dave Hansena238ab52011-05-24 17:12:16 -07002116
2117 dump_stack();
2118 if (!should_suppress_show_mem())
2119 show_mem(filter);
2120}
2121
Mel Gorman11e33f62009-06-16 15:31:57 -07002122static inline int
2123should_alloc_retry(gfp_t gfp_mask, unsigned int order,
Mel Gormanf90ac392012-01-10 15:07:15 -08002124 unsigned long did_some_progress,
Mel Gorman11e33f62009-06-16 15:31:57 -07002125 unsigned long pages_reclaimed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126{
Mel Gorman11e33f62009-06-16 15:31:57 -07002127 /* Do not loop if specifically requested */
2128 if (gfp_mask & __GFP_NORETRY)
2129 return 0;
2130
Mel Gormanf90ac392012-01-10 15:07:15 -08002131 /* Always retry if specifically requested */
2132 if (gfp_mask & __GFP_NOFAIL)
2133 return 1;
2134
2135 /*
2136 * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim
2137 * making forward progress without invoking OOM. Suspend also disables
2138 * storage devices so kswapd will not help. Bail if we are suspending.
2139 */
2140 if (!did_some_progress && pm_suspended_storage())
2141 return 0;
2142
Mel Gorman11e33f62009-06-16 15:31:57 -07002143 /*
2144 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
2145 * means __GFP_NOFAIL, but that may not be true in other
2146 * implementations.
2147 */
2148 if (order <= PAGE_ALLOC_COSTLY_ORDER)
2149 return 1;
2150
2151 /*
2152 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
2153 * specified, then we retry until we no longer reclaim any pages
2154 * (above), or we've reclaimed an order of pages at least as
2155 * large as the allocation's order. In both cases, if the
2156 * allocation still fails, we stop retrying.
2157 */
2158 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
2159 return 1;
2160
Mel Gorman11e33f62009-06-16 15:31:57 -07002161 return 0;
2162}
2163
2164static inline struct page *
2165__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
2166 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002167 nodemask_t *nodemask, struct zone *preferred_zone,
2168 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002169{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171
Mel Gorman11e33f62009-06-16 15:31:57 -07002172 /* Acquire the OOM killer lock for the zones in zonelist */
Minchan Kimff321fe2010-08-09 17:18:57 -07002173 if (!try_set_zonelist_oom(zonelist, gfp_mask)) {
Mel Gorman11e33f62009-06-16 15:31:57 -07002174 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 return NULL;
2176 }
Jens Axboe6b1de912005-11-17 21:35:02 +01002177
Mel Gorman11e33f62009-06-16 15:31:57 -07002178 /*
2179 * Go through the zonelist yet one more time, keep very high watermark
2180 * here, this is only to catch a parallel oom killing, we must fail if
2181 * we're still under heavy pressure.
2182 */
2183 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
2184 order, zonelist, high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07002185 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07002186 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002187 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07002188 goto out;
2189
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002190 if (!(gfp_mask & __GFP_NOFAIL)) {
2191 /* The OOM killer will not help higher order allocs */
2192 if (order > PAGE_ALLOC_COSTLY_ORDER)
2193 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07002194 /* The OOM killer does not needlessly kill tasks for lowmem */
2195 if (high_zoneidx < ZONE_NORMAL)
2196 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002197 /*
2198 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
2199 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
2200 * The caller should handle page allocation failure by itself if
2201 * it specifies __GFP_THISNODE.
2202 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
2203 */
2204 if (gfp_mask & __GFP_THISNODE)
2205 goto out;
2206 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002207 /* Exhausted what can be done so it's blamo time */
David Rientjes08ab9b12012-03-21 16:34:04 -07002208 out_of_memory(zonelist, gfp_mask, order, nodemask, false);
Mel Gorman11e33f62009-06-16 15:31:57 -07002209
2210out:
2211 clear_zonelist_oom(zonelist, gfp_mask);
2212 return page;
2213}
2214
Mel Gorman56de7262010-05-24 14:32:30 -07002215#ifdef CONFIG_COMPACTION
2216/* Try memory compaction for high-order allocations before reclaim */
2217static struct page *
2218__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2219 struct zonelist *zonelist, enum zone_type high_zoneidx,
2220 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
David Rientjese0b9dae2014-06-04 16:08:28 -07002221 int migratetype, enum migrate_mode mode,
Mel Gormanc67fe372012-08-21 16:16:17 -07002222 bool *contended_compaction, bool *deferred_compaction,
Mel Gorman66199712012-01-12 17:19:41 -08002223 unsigned long *did_some_progress)
Mel Gorman56de7262010-05-24 14:32:30 -07002224{
Mel Gorman66199712012-01-12 17:19:41 -08002225 if (!order)
Mel Gorman56de7262010-05-24 14:32:30 -07002226 return NULL;
2227
Rik van Rielaff62242012-03-21 16:33:52 -07002228 if (compaction_deferred(preferred_zone, order)) {
Mel Gorman66199712012-01-12 17:19:41 -08002229 *deferred_compaction = true;
2230 return NULL;
2231 }
2232
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002233 current->flags |= PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07002234 *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
David Rientjese0b9dae2014-06-04 16:08:28 -07002235 nodemask, mode,
Mel Gorman8fb74b92013-01-11 14:32:16 -08002236 contended_compaction);
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002237 current->flags &= ~PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07002238
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07002239 if (*did_some_progress != COMPACT_SKIPPED) {
Mel Gorman8fb74b92013-01-11 14:32:16 -08002240 struct page *page;
2241
Mel Gorman56de7262010-05-24 14:32:30 -07002242 /* Page migration frees to the PCP lists but we want merging */
2243 drain_pages(get_cpu());
2244 put_cpu();
2245
2246 page = get_page_from_freelist(gfp_mask, nodemask,
2247 order, zonelist, high_zoneidx,
Mel Gormancfd19c52012-07-31 16:44:10 -07002248 alloc_flags & ~ALLOC_NO_WATERMARKS,
2249 preferred_zone, migratetype);
Mel Gorman56de7262010-05-24 14:32:30 -07002250 if (page) {
Mel Gorman62997022012-10-08 16:32:47 -07002251 preferred_zone->compact_blockskip_flush = false;
Vlastimil Babkade6c60a2014-01-21 15:51:07 -08002252 compaction_defer_reset(preferred_zone, order, true);
Mel Gorman56de7262010-05-24 14:32:30 -07002253 count_vm_event(COMPACTSUCCESS);
2254 return page;
2255 }
2256
2257 /*
2258 * It's bad if compaction run occurs and fails.
2259 * The most likely reason is that pages exist,
2260 * but not enough to satisfy watermarks.
2261 */
2262 count_vm_event(COMPACTFAIL);
Mel Gorman66199712012-01-12 17:19:41 -08002263
2264 /*
2265 * As async compaction considers a subset of pageblocks, only
2266 * defer if the failure was a sync compaction failure.
2267 */
David Rientjese0b9dae2014-06-04 16:08:28 -07002268 if (mode != MIGRATE_ASYNC)
Rik van Rielaff62242012-03-21 16:33:52 -07002269 defer_compaction(preferred_zone, order);
Mel Gorman56de7262010-05-24 14:32:30 -07002270
2271 cond_resched();
2272 }
2273
2274 return NULL;
2275}
2276#else
2277static inline struct page *
2278__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2279 struct zonelist *zonelist, enum zone_type high_zoneidx,
2280 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
David Rientjese0b9dae2014-06-04 16:08:28 -07002281 int migratetype, enum migrate_mode mode, bool *contended_compaction,
2282 bool *deferred_compaction, unsigned long *did_some_progress)
Mel Gorman56de7262010-05-24 14:32:30 -07002283{
2284 return NULL;
2285}
2286#endif /* CONFIG_COMPACTION */
2287
Marek Szyprowskibba90712012-01-25 12:09:52 +01002288/* Perform direct synchronous page reclaim */
2289static int
2290__perform_reclaim(gfp_t gfp_mask, unsigned int order, struct zonelist *zonelist,
2291 nodemask_t *nodemask)
Mel Gorman11e33f62009-06-16 15:31:57 -07002292{
Mel Gorman11e33f62009-06-16 15:31:57 -07002293 struct reclaim_state reclaim_state;
Marek Szyprowskibba90712012-01-25 12:09:52 +01002294 int progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07002295
2296 cond_resched();
2297
2298 /* We now go into synchronous reclaim */
2299 cpuset_memory_pressure_bump();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002300 current->flags |= PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002301 lockdep_set_current_reclaim_state(gfp_mask);
2302 reclaim_state.reclaimed_slab = 0;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002303 current->reclaim_state = &reclaim_state;
Mel Gorman11e33f62009-06-16 15:31:57 -07002304
Marek Szyprowskibba90712012-01-25 12:09:52 +01002305 progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002306
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002307 current->reclaim_state = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002308 lockdep_clear_current_reclaim_state();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002309 current->flags &= ~PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002310
2311 cond_resched();
2312
Marek Szyprowskibba90712012-01-25 12:09:52 +01002313 return progress;
2314}
2315
2316/* The really slow allocator path where we enter direct reclaim */
2317static inline struct page *
2318__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
2319 struct zonelist *zonelist, enum zone_type high_zoneidx,
2320 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
2321 int migratetype, unsigned long *did_some_progress)
2322{
2323 struct page *page = NULL;
2324 bool drained = false;
2325
2326 *did_some_progress = __perform_reclaim(gfp_mask, order, zonelist,
2327 nodemask);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002328 if (unlikely(!(*did_some_progress)))
2329 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002330
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002331 /* After successful reclaim, reconsider all zones for allocation */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002332 if (IS_ENABLED(CONFIG_NUMA))
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002333 zlc_clear_zones_full(zonelist);
2334
Mel Gorman9ee493c2010-09-09 16:38:18 -07002335retry:
2336 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002337 zonelist, high_zoneidx,
Mel Gormancfd19c52012-07-31 16:44:10 -07002338 alloc_flags & ~ALLOC_NO_WATERMARKS,
2339 preferred_zone, migratetype);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002340
2341 /*
2342 * If an allocation failed after direct reclaim, it could be because
2343 * pages are pinned on the per-cpu lists. Drain them and try again
2344 */
2345 if (!page && !drained) {
2346 drain_all_pages();
2347 drained = true;
2348 goto retry;
2349 }
2350
Mel Gorman11e33f62009-06-16 15:31:57 -07002351 return page;
2352}
2353
Mel Gorman11e33f62009-06-16 15:31:57 -07002354/*
2355 * This is called in the allocator slow-path if the allocation request is of
2356 * sufficient urgency to ignore watermarks and take other desperate measures
2357 */
2358static inline struct page *
2359__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
2360 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002361 nodemask_t *nodemask, struct zone *preferred_zone,
2362 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002363{
2364 struct page *page;
2365
2366 do {
2367 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002368 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
Mel Gorman3dd28262009-06-16 15:32:00 -07002369 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002370
2371 if (!page && gfp_mask & __GFP_NOFAIL)
Mel Gorman0e093d992010-10-26 14:21:45 -07002372 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07002373 } while (!page && (gfp_mask & __GFP_NOFAIL));
2374
2375 return page;
2376}
2377
Johannes Weiner3a025762014-04-07 15:37:48 -07002378static void reset_alloc_batches(struct zonelist *zonelist,
2379 enum zone_type high_zoneidx,
2380 struct zone *preferred_zone)
2381{
2382 struct zoneref *z;
2383 struct zone *zone;
2384
2385 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
2386 /*
2387 * Only reset the batches of zones that were actually
2388 * considered in the fairness pass, we don't want to
2389 * trash fairness information for zones that are not
2390 * actually part of this zonelist's round-robin cycle.
2391 */
2392 if (!zone_local(preferred_zone, zone))
2393 continue;
2394 mod_zone_page_state(zone, NR_ALLOC_BATCH,
2395 high_wmark_pages(zone) - low_wmark_pages(zone) -
2396 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
2397 }
2398}
2399
2400static void wake_all_kswapds(unsigned int order,
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002401 struct zonelist *zonelist,
2402 enum zone_type high_zoneidx,
2403 struct zone *preferred_zone)
Mel Gorman11e33f62009-06-16 15:31:57 -07002404{
2405 struct zoneref *z;
2406 struct zone *zone;
2407
Johannes Weiner3a025762014-04-07 15:37:48 -07002408 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
2409 wakeup_kswapd(zone, order, zone_idx(preferred_zone));
Mel Gorman11e33f62009-06-16 15:31:57 -07002410}
2411
Peter Zijlstra341ce062009-06-16 15:32:02 -07002412static inline int
2413gfp_to_alloc_flags(gfp_t gfp_mask)
2414{
Peter Zijlstra341ce062009-06-16 15:32:02 -07002415 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
2416 const gfp_t wait = gfp_mask & __GFP_WAIT;
2417
Mel Gormana56f57f2009-06-16 15:32:02 -07002418 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07002419 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07002420
Peter Zijlstra341ce062009-06-16 15:32:02 -07002421 /*
2422 * The caller may dip into page reserves a bit more if the caller
2423 * cannot run direct reclaim, or if the caller has realtime scheduling
2424 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2425 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
2426 */
Namhyung Kime6223a32010-10-26 14:21:59 -07002427 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07002428
2429 if (!wait) {
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002430 /*
2431 * Not worth trying to allocate harder for
2432 * __GFP_NOMEMALLOC even if it can't schedule.
2433 */
2434 if (!(gfp_mask & __GFP_NOMEMALLOC))
2435 alloc_flags |= ALLOC_HARDER;
Peter Zijlstra341ce062009-06-16 15:32:02 -07002436 /*
2437 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
2438 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
2439 */
2440 alloc_flags &= ~ALLOC_CPUSET;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002441 } else if (unlikely(rt_task(current)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07002442 alloc_flags |= ALLOC_HARDER;
2443
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002444 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2445 if (gfp_mask & __GFP_MEMALLOC)
2446 alloc_flags |= ALLOC_NO_WATERMARKS;
Mel Gorman907aed42012-07-31 16:44:07 -07002447 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
2448 alloc_flags |= ALLOC_NO_WATERMARKS;
2449 else if (!in_interrupt() &&
2450 ((current->flags & PF_MEMALLOC) ||
2451 unlikely(test_thread_flag(TIF_MEMDIE))))
Peter Zijlstra341ce062009-06-16 15:32:02 -07002452 alloc_flags |= ALLOC_NO_WATERMARKS;
2453 }
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002454#ifdef CONFIG_CMA
2455 if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2456 alloc_flags |= ALLOC_CMA;
2457#endif
Peter Zijlstra341ce062009-06-16 15:32:02 -07002458 return alloc_flags;
2459}
2460
Mel Gorman072bb0a2012-07-31 16:43:58 -07002461bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
2462{
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002463 return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
Mel Gorman072bb0a2012-07-31 16:43:58 -07002464}
2465
Mel Gorman11e33f62009-06-16 15:31:57 -07002466static inline struct page *
2467__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
2468 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002469 nodemask_t *nodemask, struct zone *preferred_zone,
2470 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002471{
2472 const gfp_t wait = gfp_mask & __GFP_WAIT;
2473 struct page *page = NULL;
2474 int alloc_flags;
2475 unsigned long pages_reclaimed = 0;
2476 unsigned long did_some_progress;
David Rientjese0b9dae2014-06-04 16:08:28 -07002477 enum migrate_mode migration_mode = MIGRATE_ASYNC;
Mel Gorman66199712012-01-12 17:19:41 -08002478 bool deferred_compaction = false;
Mel Gormanc67fe372012-08-21 16:16:17 -07002479 bool contended_compaction = false;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002480
Christoph Lameter952f3b52006-12-06 20:33:26 -08002481 /*
Mel Gorman72807a72009-06-16 15:32:18 -07002482 * In the slowpath, we sanity check order to avoid ever trying to
2483 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2484 * be using allocators in order of preference for an area that is
2485 * too large.
2486 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07002487 if (order >= MAX_ORDER) {
2488 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07002489 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07002490 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002491
Christoph Lameter952f3b52006-12-06 20:33:26 -08002492 /*
2493 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2494 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2495 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2496 * using a larger set of nodes after it has established that the
2497 * allowed per node queues are empty and that nodes are
2498 * over allocated.
2499 */
Johannes Weiner3a025762014-04-07 15:37:48 -07002500 if (IS_ENABLED(CONFIG_NUMA) &&
2501 (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
Christoph Lameter952f3b52006-12-06 20:33:26 -08002502 goto nopage;
2503
Mel Gormancc4a6852009-11-11 14:26:14 -08002504restart:
Johannes Weiner3a025762014-04-07 15:37:48 -07002505 if (!(gfp_mask & __GFP_NO_KSWAPD))
2506 wake_all_kswapds(order, zonelist, high_zoneidx, preferred_zone);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002507
Paul Jackson9bf22292005-09-06 15:18:12 -07002508 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002509 * OK, we're below the kswapd watermark and have kicked background
2510 * reclaim. Now things get more complex, so set up alloc_flags according
2511 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07002512 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002513 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514
David Rientjesf33261d2011-01-25 15:07:20 -08002515 /*
2516 * Find the true preferred zone if the allocation is unconstrained by
2517 * cpusets.
2518 */
2519 if (!(alloc_flags & ALLOC_CPUSET) && !nodemask)
2520 first_zones_zonelist(zonelist, high_zoneidx, NULL,
2521 &preferred_zone);
2522
Andrew Barrycfa54a02011-05-24 17:12:52 -07002523rebalance:
Peter Zijlstra341ce062009-06-16 15:32:02 -07002524 /* This is the last chance, in general, before the goto nopage. */
Mel Gorman19770b32008-04-28 02:12:18 -07002525 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
Peter Zijlstra341ce062009-06-16 15:32:02 -07002526 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
2527 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002528 if (page)
2529 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530
Mel Gorman11e33f62009-06-16 15:31:57 -07002531 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002532 if (alloc_flags & ALLOC_NO_WATERMARKS) {
Mel Gorman183f6372012-07-31 16:44:12 -07002533 /*
2534 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
2535 * the allocation is high priority and these type of
2536 * allocations are system rather than user orientated
2537 */
2538 zonelist = node_zonelist(numa_node_id(), gfp_mask);
2539
Peter Zijlstra341ce062009-06-16 15:32:02 -07002540 page = __alloc_pages_high_priority(gfp_mask, order,
2541 zonelist, high_zoneidx, nodemask,
2542 preferred_zone, migratetype);
Mel Gormancfd19c52012-07-31 16:44:10 -07002543 if (page) {
Peter Zijlstra341ce062009-06-16 15:32:02 -07002544 goto got_pg;
Mel Gormancfd19c52012-07-31 16:44:10 -07002545 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 }
2547
2548 /* Atomic allocations - we can't balance anything */
David Rientjesaed0a0e2014-01-21 15:51:12 -08002549 if (!wait) {
2550 /*
2551 * All existing users of the deprecated __GFP_NOFAIL are
2552 * blockable, so warn of any new users that actually allow this
2553 * type of allocation to fail.
2554 */
2555 WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 goto nopage;
David Rientjesaed0a0e2014-01-21 15:51:12 -08002557 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558
Peter Zijlstra341ce062009-06-16 15:32:02 -07002559 /* Avoid recursion of direct reclaim */
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002560 if (current->flags & PF_MEMALLOC)
Peter Zijlstra341ce062009-06-16 15:32:02 -07002561 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562
David Rientjes6583bb62009-07-29 15:02:06 -07002563 /* Avoid allocations with no watermarks from looping endlessly */
2564 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2565 goto nopage;
2566
Mel Gorman77f1fe62011-01-13 15:45:57 -08002567 /*
2568 * Try direct compaction. The first pass is asynchronous. Subsequent
2569 * attempts after direct reclaim are synchronous
2570 */
David Rientjese0b9dae2014-06-04 16:08:28 -07002571 page = __alloc_pages_direct_compact(gfp_mask, order, zonelist,
2572 high_zoneidx, nodemask, alloc_flags,
2573 preferred_zone, migratetype,
2574 migration_mode, &contended_compaction,
Mel Gorman66199712012-01-12 17:19:41 -08002575 &deferred_compaction,
2576 &did_some_progress);
Mel Gorman56de7262010-05-24 14:32:30 -07002577 if (page)
2578 goto got_pg;
David Rientjes75f30862014-06-04 16:08:30 -07002579
2580 /*
2581 * It can become very expensive to allocate transparent hugepages at
2582 * fault, so use asynchronous memory compaction for THP unless it is
2583 * khugepaged trying to collapse.
2584 */
2585 if (!(gfp_mask & __GFP_NO_KSWAPD) || (current->flags & PF_KTHREAD))
2586 migration_mode = MIGRATE_SYNC_LIGHT;
Mel Gorman56de7262010-05-24 14:32:30 -07002587
Linus Torvalds31f8d422012-12-10 10:47:45 -08002588 /*
2589 * If compaction is deferred for high-order allocations, it is because
2590 * sync compaction recently failed. In this is the case and the caller
2591 * requested a movable allocation that does not heavily disrupt the
2592 * system then fail the allocation instead of entering direct reclaim.
2593 */
2594 if ((deferred_compaction || contended_compaction) &&
Linus Torvaldscaf49192012-12-10 10:51:16 -08002595 (gfp_mask & __GFP_NO_KSWAPD))
Linus Torvalds31f8d422012-12-10 10:47:45 -08002596 goto nopage;
Mel Gorman66199712012-01-12 17:19:41 -08002597
Mel Gorman11e33f62009-06-16 15:31:57 -07002598 /* Try direct reclaim and then allocating */
2599 page = __alloc_pages_direct_reclaim(gfp_mask, order,
2600 zonelist, high_zoneidx,
2601 nodemask,
Mel Gorman5117f452009-06-16 15:31:59 -07002602 alloc_flags, preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07002603 migratetype, &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07002604 if (page)
2605 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607 /*
Mel Gorman11e33f62009-06-16 15:31:57 -07002608 * If we failed to make any progress reclaiming, then we are
2609 * running out of options and have to consider going OOM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610 */
Mel Gorman11e33f62009-06-16 15:31:57 -07002611 if (!did_some_progress) {
Qiang Huangb9921ec2013-11-12 15:07:22 -08002612 if (oom_gfp_allowed(gfp_mask)) {
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -07002613 if (oom_killer_disabled)
2614 goto nopage;
David Rientjes29fd66d2012-03-28 14:42:41 -07002615 /* Coredumps can quickly deplete all memory reserves */
2616 if ((current->flags & PF_DUMPCORE) &&
2617 !(gfp_mask & __GFP_NOFAIL))
2618 goto nopage;
Mel Gorman11e33f62009-06-16 15:31:57 -07002619 page = __alloc_pages_may_oom(gfp_mask, order,
2620 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002621 nodemask, preferred_zone,
2622 migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002623 if (page)
2624 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625
David Rientjes03668b32010-08-09 17:18:54 -07002626 if (!(gfp_mask & __GFP_NOFAIL)) {
2627 /*
2628 * The oom killer is not called for high-order
2629 * allocations that may fail, so if no progress
2630 * is being made, there are no other options and
2631 * retrying is unlikely to help.
2632 */
2633 if (order > PAGE_ALLOC_COSTLY_ORDER)
2634 goto nopage;
2635 /*
2636 * The oom killer is not called for lowmem
2637 * allocations to prevent needlessly killing
2638 * innocent tasks.
2639 */
2640 if (high_zoneidx < ZONE_NORMAL)
2641 goto nopage;
2642 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644 goto restart;
2645 }
2646 }
2647
Mel Gorman11e33f62009-06-16 15:31:57 -07002648 /* Check if we should retry the allocation */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07002649 pages_reclaimed += did_some_progress;
Mel Gormanf90ac392012-01-10 15:07:15 -08002650 if (should_alloc_retry(gfp_mask, order, did_some_progress,
2651 pages_reclaimed)) {
Mel Gorman11e33f62009-06-16 15:31:57 -07002652 /* Wait for some write requests to complete then retry */
Mel Gorman0e093d992010-10-26 14:21:45 -07002653 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 goto rebalance;
Mel Gorman3e7d3442011-01-13 15:45:56 -08002655 } else {
2656 /*
2657 * High-order allocations do not necessarily loop after
2658 * direct reclaim and reclaim/compaction depends on compaction
2659 * being called after reclaim so call directly if necessary
2660 */
David Rientjese0b9dae2014-06-04 16:08:28 -07002661 page = __alloc_pages_direct_compact(gfp_mask, order, zonelist,
2662 high_zoneidx, nodemask, alloc_flags,
2663 preferred_zone, migratetype,
2664 migration_mode, &contended_compaction,
Mel Gorman66199712012-01-12 17:19:41 -08002665 &deferred_compaction,
2666 &did_some_progress);
Mel Gorman3e7d3442011-01-13 15:45:56 -08002667 if (page)
2668 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669 }
2670
2671nopage:
Dave Hansena238ab52011-05-24 17:12:16 -07002672 warn_alloc_failed(gfp_mask, order, NULL);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002673 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674got_pg:
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002675 if (kmemcheck_enabled)
2676 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002677
Mel Gorman072bb0a2012-07-31 16:43:58 -07002678 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679}
Mel Gorman11e33f62009-06-16 15:31:57 -07002680
2681/*
2682 * This is the 'heart' of the zoned buddy allocator.
2683 */
2684struct page *
2685__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2686 struct zonelist *zonelist, nodemask_t *nodemask)
2687{
2688 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Mel Gorman5117f452009-06-16 15:31:59 -07002689 struct zone *preferred_zone;
Mel Gormancc9a6c82012-03-21 16:34:11 -07002690 struct page *page = NULL;
Mel Gorman3dd28262009-06-16 15:32:00 -07002691 int migratetype = allocflags_to_migratetype(gfp_mask);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002692 unsigned int cpuset_mems_cookie;
Johannes Weiner3a025762014-04-07 15:37:48 -07002693 int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET|ALLOC_FAIR;
Mel Gorman11e33f62009-06-16 15:31:57 -07002694
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10002695 gfp_mask &= gfp_allowed_mask;
2696
Mel Gorman11e33f62009-06-16 15:31:57 -07002697 lockdep_trace_alloc(gfp_mask);
2698
2699 might_sleep_if(gfp_mask & __GFP_WAIT);
2700
2701 if (should_fail_alloc_page(gfp_mask, order))
2702 return NULL;
2703
2704 /*
2705 * Check the zones suitable for the gfp_mask contain at least one
2706 * valid zone. It's possible to have an empty zonelist as a result
2707 * of GFP_THISNODE and a memoryless node
2708 */
2709 if (unlikely(!zonelist->_zonerefs->zone))
2710 return NULL;
2711
Mel Gormancc9a6c82012-03-21 16:34:11 -07002712retry_cpuset:
Mel Gormand26914d2014-04-03 14:47:24 -07002713 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07002714
Mel Gorman5117f452009-06-16 15:31:59 -07002715 /* The preferred zone is used for statistics later */
David Rientjesf33261d2011-01-25 15:07:20 -08002716 first_zones_zonelist(zonelist, high_zoneidx,
2717 nodemask ? : &cpuset_current_mems_allowed,
2718 &preferred_zone);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002719 if (!preferred_zone)
2720 goto out;
Mel Gorman5117f452009-06-16 15:31:59 -07002721
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002722#ifdef CONFIG_CMA
2723 if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2724 alloc_flags |= ALLOC_CMA;
2725#endif
Johannes Weiner3a025762014-04-07 15:37:48 -07002726retry:
Mel Gorman5117f452009-06-16 15:31:59 -07002727 /* First allocation attempt */
Mel Gorman11e33f62009-06-16 15:31:57 -07002728 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002729 zonelist, high_zoneidx, alloc_flags,
Mel Gorman3dd28262009-06-16 15:32:00 -07002730 preferred_zone, migratetype);
Ming Lei21caf2f2013-02-22 16:34:08 -08002731 if (unlikely(!page)) {
2732 /*
Johannes Weiner3a025762014-04-07 15:37:48 -07002733 * The first pass makes sure allocations are spread
2734 * fairly within the local node. However, the local
2735 * node might have free pages left after the fairness
2736 * batches are exhausted, and remote zones haven't
2737 * even been considered yet. Try once more without
2738 * fairness, and include remote zones now, before
2739 * entering the slowpath and waking kswapd: prefer
2740 * spilling to a remote zone over swapping locally.
2741 */
2742 if (alloc_flags & ALLOC_FAIR) {
2743 reset_alloc_batches(zonelist, high_zoneidx,
2744 preferred_zone);
2745 alloc_flags &= ~ALLOC_FAIR;
2746 goto retry;
2747 }
2748 /*
Ming Lei21caf2f2013-02-22 16:34:08 -08002749 * Runtime PM, block IO and its error handling path
2750 * can deadlock because I/O on the device might not
2751 * complete.
2752 */
2753 gfp_mask = memalloc_noio_flags(gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002754 page = __alloc_pages_slowpath(gfp_mask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002755 zonelist, high_zoneidx, nodemask,
Mel Gorman3dd28262009-06-16 15:32:00 -07002756 preferred_zone, migratetype);
Ming Lei21caf2f2013-02-22 16:34:08 -08002757 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002758
Mel Gorman4b4f2782009-09-21 17:02:41 -07002759 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002760
2761out:
2762 /*
2763 * When updating a task's mems_allowed, it is possible to race with
2764 * parallel threads in such a way that an allocation can fail while
2765 * the mask is being updated. If a page allocation is about to fail,
2766 * check if the cpuset changed during allocation and if so, retry.
2767 */
Mel Gormand26914d2014-04-03 14:47:24 -07002768 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
Mel Gormancc9a6c82012-03-21 16:34:11 -07002769 goto retry_cpuset;
2770
Mel Gorman11e33f62009-06-16 15:31:57 -07002771 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002772}
Mel Gormand2391712009-06-16 15:31:52 -07002773EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774
2775/*
2776 * Common helper functions.
2777 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08002778unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779{
Akinobu Mita945a1112009-09-21 17:01:47 -07002780 struct page *page;
2781
2782 /*
2783 * __get_free_pages() returns a 32-bit address, which cannot represent
2784 * a highmem page
2785 */
2786 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2787
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788 page = alloc_pages(gfp_mask, order);
2789 if (!page)
2790 return 0;
2791 return (unsigned long) page_address(page);
2792}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793EXPORT_SYMBOL(__get_free_pages);
2794
Harvey Harrison920c7a52008-02-04 22:29:26 -08002795unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796{
Akinobu Mita945a1112009-09-21 17:01:47 -07002797 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002799EXPORT_SYMBOL(get_zeroed_page);
2800
Harvey Harrison920c7a52008-02-04 22:29:26 -08002801void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802{
Nick Pigginb5810032005-10-29 18:16:12 -07002803 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804 if (order == 0)
Li Hongfc916682010-03-05 13:41:54 -08002805 free_hot_cold_page(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806 else
2807 __free_pages_ok(page, order);
2808 }
2809}
2810
2811EXPORT_SYMBOL(__free_pages);
2812
Harvey Harrison920c7a52008-02-04 22:29:26 -08002813void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814{
2815 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07002816 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817 __free_pages(virt_to_page((void *)addr), order);
2818 }
2819}
2820
2821EXPORT_SYMBOL(free_pages);
2822
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002823/*
Vladimir Davydov52383432014-06-04 16:06:39 -07002824 * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
2825 * of the current memory cgroup.
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002826 *
Vladimir Davydov52383432014-06-04 16:06:39 -07002827 * It should be used when the caller would like to use kmalloc, but since the
2828 * allocation is large, it has to fall back to the page allocator.
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002829 */
Vladimir Davydov52383432014-06-04 16:06:39 -07002830struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order)
2831{
2832 struct page *page;
2833 struct mem_cgroup *memcg = NULL;
2834
2835 if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
2836 return NULL;
2837 page = alloc_pages(gfp_mask, order);
2838 memcg_kmem_commit_charge(page, memcg, order);
2839 return page;
2840}
2841
2842struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
2843{
2844 struct page *page;
2845 struct mem_cgroup *memcg = NULL;
2846
2847 if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
2848 return NULL;
2849 page = alloc_pages_node(nid, gfp_mask, order);
2850 memcg_kmem_commit_charge(page, memcg, order);
2851 return page;
2852}
2853
2854/*
2855 * __free_kmem_pages and free_kmem_pages will free pages allocated with
2856 * alloc_kmem_pages.
2857 */
2858void __free_kmem_pages(struct page *page, unsigned int order)
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002859{
2860 memcg_kmem_uncharge_pages(page, order);
2861 __free_pages(page, order);
2862}
2863
Vladimir Davydov52383432014-06-04 16:06:39 -07002864void free_kmem_pages(unsigned long addr, unsigned int order)
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002865{
2866 if (addr != 0) {
2867 VM_BUG_ON(!virt_addr_valid((void *)addr));
Vladimir Davydov52383432014-06-04 16:06:39 -07002868 __free_kmem_pages(virt_to_page((void *)addr), order);
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002869 }
2870}
2871
Andi Kleenee85c2e2011-05-11 15:13:34 -07002872static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
2873{
2874 if (addr) {
2875 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2876 unsigned long used = addr + PAGE_ALIGN(size);
2877
2878 split_page(virt_to_page((void *)addr), order);
2879 while (used < alloc_end) {
2880 free_page(used);
2881 used += PAGE_SIZE;
2882 }
2883 }
2884 return (void *)addr;
2885}
2886
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002887/**
2888 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2889 * @size: the number of bytes to allocate
2890 * @gfp_mask: GFP flags for the allocation
2891 *
2892 * This function is similar to alloc_pages(), except that it allocates the
2893 * minimum number of pages to satisfy the request. alloc_pages() can only
2894 * allocate memory in power-of-two pages.
2895 *
2896 * This function is also limited by MAX_ORDER.
2897 *
2898 * Memory allocated by this function must be released by free_pages_exact().
2899 */
2900void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2901{
2902 unsigned int order = get_order(size);
2903 unsigned long addr;
2904
2905 addr = __get_free_pages(gfp_mask, order);
Andi Kleenee85c2e2011-05-11 15:13:34 -07002906 return make_alloc_exact(addr, order, size);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002907}
2908EXPORT_SYMBOL(alloc_pages_exact);
2909
2910/**
Andi Kleenee85c2e2011-05-11 15:13:34 -07002911 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2912 * pages on a node.
Randy Dunlapb5e6ab52011-05-16 13:16:54 -07002913 * @nid: the preferred node ID where memory should be allocated
Andi Kleenee85c2e2011-05-11 15:13:34 -07002914 * @size: the number of bytes to allocate
2915 * @gfp_mask: GFP flags for the allocation
2916 *
2917 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2918 * back.
2919 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2920 * but is not exact.
2921 */
2922void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
2923{
2924 unsigned order = get_order(size);
2925 struct page *p = alloc_pages_node(nid, gfp_mask, order);
2926 if (!p)
2927 return NULL;
2928 return make_alloc_exact((unsigned long)page_address(p), order, size);
2929}
2930EXPORT_SYMBOL(alloc_pages_exact_nid);
2931
2932/**
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002933 * free_pages_exact - release memory allocated via alloc_pages_exact()
2934 * @virt: the value returned by alloc_pages_exact.
2935 * @size: size of allocation, same value as passed to alloc_pages_exact().
2936 *
2937 * Release the memory allocated by a previous call to alloc_pages_exact.
2938 */
2939void free_pages_exact(void *virt, size_t size)
2940{
2941 unsigned long addr = (unsigned long)virt;
2942 unsigned long end = addr + PAGE_ALIGN(size);
2943
2944 while (addr < end) {
2945 free_page(addr);
2946 addr += PAGE_SIZE;
2947 }
2948}
2949EXPORT_SYMBOL(free_pages_exact);
2950
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08002951/**
2952 * nr_free_zone_pages - count number of pages beyond high watermark
2953 * @offset: The zone index of the highest zone
2954 *
2955 * nr_free_zone_pages() counts the number of counts pages which are beyond the
2956 * high watermark within all zones at or below a given zone index. For each
2957 * zone, the number of pages is calculated as:
Jiang Liu834405c2013-07-03 15:03:04 -07002958 * managed_pages - high_pages
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08002959 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08002960static unsigned long nr_free_zone_pages(int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002961{
Mel Gormandd1a2392008-04-28 02:12:17 -07002962 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002963 struct zone *zone;
2964
Martin J. Blighe310fd42005-07-29 22:59:18 -07002965 /* Just pick one node, since fallback list is circular */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08002966 unsigned long sum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002967
Mel Gorman0e884602008-04-28 02:12:14 -07002968 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002969
Mel Gorman54a6eb52008-04-28 02:12:16 -07002970 for_each_zone_zonelist(zone, z, zonelist, offset) {
Jiang Liub40da042013-02-22 16:33:52 -08002971 unsigned long size = zone->managed_pages;
Mel Gorman41858962009-06-16 15:32:12 -07002972 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07002973 if (size > high)
2974 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975 }
2976
2977 return sum;
2978}
2979
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08002980/**
2981 * nr_free_buffer_pages - count number of pages beyond high watermark
2982 *
2983 * nr_free_buffer_pages() counts the number of pages which are beyond the high
2984 * watermark within ZONE_DMA and ZONE_NORMAL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002985 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08002986unsigned long nr_free_buffer_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987{
Al Viroaf4ca452005-10-21 02:55:38 -04002988 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989}
Meelap Shahc2f1a552007-07-17 04:04:39 -07002990EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002991
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08002992/**
2993 * nr_free_pagecache_pages - count number of pages beyond high watermark
2994 *
2995 * nr_free_pagecache_pages() counts the number of pages which are beyond the
2996 * high watermark within all zones.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002997 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08002998unsigned long nr_free_pagecache_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002999{
Mel Gorman2a1e2742007-07-17 04:03:12 -07003000 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07003002
3003static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08003005 if (IS_ENABLED(CONFIG_NUMA))
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08003006 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003007}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003008
Linus Torvalds1da177e2005-04-16 15:20:36 -07003009void si_meminfo(struct sysinfo *val)
3010{
3011 val->totalram = totalram_pages;
3012 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08003013 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003015 val->totalhigh = totalhigh_pages;
3016 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017 val->mem_unit = PAGE_SIZE;
3018}
3019
3020EXPORT_SYMBOL(si_meminfo);
3021
3022#ifdef CONFIG_NUMA
3023void si_meminfo_node(struct sysinfo *val, int nid)
3024{
Jiang Liucdd91a72013-07-03 15:03:27 -07003025 int zone_type; /* needs to be signed */
3026 unsigned long managed_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003027 pg_data_t *pgdat = NODE_DATA(nid);
3028
Jiang Liucdd91a72013-07-03 15:03:27 -07003029 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
3030 managed_pages += pgdat->node_zones[zone_type].managed_pages;
3031 val->totalram = managed_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08003032 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003033#ifdef CONFIG_HIGHMEM
Jiang Liub40da042013-02-22 16:33:52 -08003034 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08003035 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
3036 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003037#else
3038 val->totalhigh = 0;
3039 val->freehigh = 0;
3040#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003041 val->mem_unit = PAGE_SIZE;
3042}
3043#endif
3044
David Rientjesddd588b2011-03-22 16:30:46 -07003045/*
David Rientjes7bf02ea2011-05-24 17:11:16 -07003046 * Determine whether the node should be displayed or not, depending on whether
3047 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
David Rientjesddd588b2011-03-22 16:30:46 -07003048 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07003049bool skip_free_areas_node(unsigned int flags, int nid)
David Rientjesddd588b2011-03-22 16:30:46 -07003050{
3051 bool ret = false;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003052 unsigned int cpuset_mems_cookie;
David Rientjesddd588b2011-03-22 16:30:46 -07003053
3054 if (!(flags & SHOW_MEM_FILTER_NODES))
3055 goto out;
3056
Mel Gormancc9a6c82012-03-21 16:34:11 -07003057 do {
Mel Gormand26914d2014-04-03 14:47:24 -07003058 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07003059 ret = !node_isset(nid, cpuset_current_mems_allowed);
Mel Gormand26914d2014-04-03 14:47:24 -07003060 } while (read_mems_allowed_retry(cpuset_mems_cookie));
David Rientjesddd588b2011-03-22 16:30:46 -07003061out:
3062 return ret;
3063}
3064
Linus Torvalds1da177e2005-04-16 15:20:36 -07003065#define K(x) ((x) << (PAGE_SHIFT-10))
3066
Rabin Vincent377e4f12012-12-11 16:00:24 -08003067static void show_migration_types(unsigned char type)
3068{
3069 static const char types[MIGRATE_TYPES] = {
3070 [MIGRATE_UNMOVABLE] = 'U',
3071 [MIGRATE_RECLAIMABLE] = 'E',
3072 [MIGRATE_MOVABLE] = 'M',
3073 [MIGRATE_RESERVE] = 'R',
3074#ifdef CONFIG_CMA
3075 [MIGRATE_CMA] = 'C',
3076#endif
Minchan Kim194159f2013-02-22 16:33:58 -08003077#ifdef CONFIG_MEMORY_ISOLATION
Rabin Vincent377e4f12012-12-11 16:00:24 -08003078 [MIGRATE_ISOLATE] = 'I',
Minchan Kim194159f2013-02-22 16:33:58 -08003079#endif
Rabin Vincent377e4f12012-12-11 16:00:24 -08003080 };
3081 char tmp[MIGRATE_TYPES + 1];
3082 char *p = tmp;
3083 int i;
3084
3085 for (i = 0; i < MIGRATE_TYPES; i++) {
3086 if (type & (1 << i))
3087 *p++ = types[i];
3088 }
3089
3090 *p = '\0';
3091 printk("(%s) ", tmp);
3092}
3093
Linus Torvalds1da177e2005-04-16 15:20:36 -07003094/*
3095 * Show free area list (used inside shift_scroll-lock stuff)
3096 * We also calculate the percentage fragmentation. We do this by counting the
3097 * memory on each free list with the exception of the first item on the list.
David Rientjesddd588b2011-03-22 16:30:46 -07003098 * Suppresses nodes that are not allowed by current's cpuset if
3099 * SHOW_MEM_FILTER_NODES is passed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003100 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07003101void show_free_areas(unsigned int filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003102{
Jes Sorensenc7241912006-09-27 01:50:05 -07003103 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003104 struct zone *zone;
3105
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003106 for_each_populated_zone(zone) {
David Rientjes7bf02ea2011-05-24 17:11:16 -07003107 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003108 continue;
Jes Sorensenc7241912006-09-27 01:50:05 -07003109 show_node(zone);
3110 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111
Dave Jones6b482c62005-11-10 15:45:56 -05003112 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003113 struct per_cpu_pageset *pageset;
3114
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003115 pageset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003116
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003117 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
3118 cpu, pageset->pcp.high,
3119 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120 }
3121 }
3122
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003123 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3124 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003125 " unevictable:%lu"
Andrew Mortonb76146e2009-10-26 16:49:52 -07003126 " dirty:%lu writeback:%lu unstable:%lu\n"
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07003127 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003128 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3129 " free_cma:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003130 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003131 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003132 global_page_state(NR_ISOLATED_ANON),
3133 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003134 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003135 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003136 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07003137 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07003138 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07003139 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08003140 global_page_state(NR_FREE_PAGES),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07003141 global_page_state(NR_SLAB_RECLAIMABLE),
3142 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07003143 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003144 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08003145 global_page_state(NR_PAGETABLE),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003146 global_page_state(NR_BOUNCE),
3147 global_page_state(NR_FREE_CMA_PAGES));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003148
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003149 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003150 int i;
3151
David Rientjes7bf02ea2011-05-24 17:11:16 -07003152 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003153 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154 show_node(zone);
3155 printk("%s"
3156 " free:%lukB"
3157 " min:%lukB"
3158 " low:%lukB"
3159 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003160 " active_anon:%lukB"
3161 " inactive_anon:%lukB"
3162 " active_file:%lukB"
3163 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003164 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003165 " isolated(anon):%lukB"
3166 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003167 " present:%lukB"
Jiang Liu9feedc92012-12-12 13:52:12 -08003168 " managed:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003169 " mlocked:%lukB"
3170 " dirty:%lukB"
3171 " writeback:%lukB"
3172 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003173 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003174 " slab_reclaimable:%lukB"
3175 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003176 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003177 " pagetables:%lukB"
3178 " unstable:%lukB"
3179 " bounce:%lukB"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003180 " free_cma:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003181 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003182 " pages_scanned:%lu"
3183 " all_unreclaimable? %s"
3184 "\n",
3185 zone->name,
Mel Gorman88f5acf2011-01-13 15:45:41 -08003186 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07003187 K(min_wmark_pages(zone)),
3188 K(low_wmark_pages(zone)),
3189 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003190 K(zone_page_state(zone, NR_ACTIVE_ANON)),
3191 K(zone_page_state(zone, NR_INACTIVE_ANON)),
3192 K(zone_page_state(zone, NR_ACTIVE_FILE)),
3193 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003194 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003195 K(zone_page_state(zone, NR_ISOLATED_ANON)),
3196 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003197 K(zone->present_pages),
Jiang Liu9feedc92012-12-12 13:52:12 -08003198 K(zone->managed_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003199 K(zone_page_state(zone, NR_MLOCK)),
3200 K(zone_page_state(zone, NR_FILE_DIRTY)),
3201 K(zone_page_state(zone, NR_WRITEBACK)),
3202 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003203 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003204 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3205 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003206 zone_page_state(zone, NR_KERNEL_STACK) *
3207 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003208 K(zone_page_state(zone, NR_PAGETABLE)),
3209 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3210 K(zone_page_state(zone, NR_BOUNCE)),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003211 K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003212 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003213 zone->pages_scanned,
Lisa Du6e543d52013-09-11 14:22:36 -07003214 (!zone_reclaimable(zone) ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07003215 );
3216 printk("lowmem_reserve[]:");
3217 for (i = 0; i < MAX_NR_ZONES; i++)
3218 printk(" %lu", zone->lowmem_reserve[i]);
3219 printk("\n");
3220 }
3221
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003222 for_each_populated_zone(zone) {
Pintu Kumarb8af2942013-09-11 14:20:34 -07003223 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003224 unsigned char types[MAX_ORDER];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003225
David Rientjes7bf02ea2011-05-24 17:11:16 -07003226 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003227 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003228 show_node(zone);
3229 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003230
3231 spin_lock_irqsave(&zone->lock, flags);
3232 for (order = 0; order < MAX_ORDER; order++) {
Rabin Vincent377e4f12012-12-11 16:00:24 -08003233 struct free_area *area = &zone->free_area[order];
3234 int type;
3235
3236 nr[order] = area->nr_free;
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003237 total += nr[order] << order;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003238
3239 types[order] = 0;
3240 for (type = 0; type < MIGRATE_TYPES; type++) {
3241 if (!list_empty(&area->free_list[type]))
3242 types[order] |= 1 << type;
3243 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003244 }
3245 spin_unlock_irqrestore(&zone->lock, flags);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003246 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003247 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003248 if (nr[order])
3249 show_migration_types(types[order]);
3250 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003251 printk("= %lukB\n", K(total));
3252 }
3253
David Rientjes949f7ec2013-04-29 15:07:48 -07003254 hugetlb_show_meminfo();
3255
Larry Woodmane6f36022008-02-04 22:29:30 -08003256 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
3257
Linus Torvalds1da177e2005-04-16 15:20:36 -07003258 show_swap_cache_info();
3259}
3260
Mel Gorman19770b32008-04-28 02:12:18 -07003261static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
3262{
3263 zoneref->zone = zone;
3264 zoneref->zone_idx = zone_idx(zone);
3265}
3266
Linus Torvalds1da177e2005-04-16 15:20:36 -07003267/*
3268 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08003269 *
3270 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003271 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003272static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003273 int nr_zones)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003274{
Christoph Lameter1a932052006-01-06 00:11:16 -08003275 struct zone *zone;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003276 enum zone_type zone_type = MAX_NR_ZONES;
Christoph Lameter02a68a52006-01-06 00:11:18 -08003277
3278 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003279 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08003280 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08003281 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07003282 zoneref_set_zone(zone,
3283 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08003284 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003285 }
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003286 } while (zone_type);
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003287
Christoph Lameter070f8032006-01-06 00:11:19 -08003288 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003289}
3290
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003291
3292/*
3293 * zonelist_order:
3294 * 0 = automatic detection of better ordering.
3295 * 1 = order by ([node] distance, -zonetype)
3296 * 2 = order by (-zonetype, [node] distance)
3297 *
3298 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3299 * the same zonelist. So only NUMA can configure this param.
3300 */
3301#define ZONELIST_ORDER_DEFAULT 0
3302#define ZONELIST_ORDER_NODE 1
3303#define ZONELIST_ORDER_ZONE 2
3304
3305/* zonelist order in the kernel.
3306 * set_zonelist_order() will set this to NODE or ZONE.
3307 */
3308static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
3309static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
3310
3311
Linus Torvalds1da177e2005-04-16 15:20:36 -07003312#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003313/* The value user specified ....changed by config */
3314static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3315/* string for sysctl */
3316#define NUMA_ZONELIST_ORDER_LEN 16
3317char numa_zonelist_order[16] = "default";
3318
3319/*
3320 * interface for configure zonelist ordering.
3321 * command line option "numa_zonelist_order"
3322 * = "[dD]efault - default, automatic configuration.
3323 * = "[nN]ode - order by node locality, then by zone within node
3324 * = "[zZ]one - order by zone, then by locality within zone
3325 */
3326
3327static int __parse_numa_zonelist_order(char *s)
3328{
3329 if (*s == 'd' || *s == 'D') {
3330 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3331 } else if (*s == 'n' || *s == 'N') {
3332 user_zonelist_order = ZONELIST_ORDER_NODE;
3333 } else if (*s == 'z' || *s == 'Z') {
3334 user_zonelist_order = ZONELIST_ORDER_ZONE;
3335 } else {
3336 printk(KERN_WARNING
3337 "Ignoring invalid numa_zonelist_order value: "
3338 "%s\n", s);
3339 return -EINVAL;
3340 }
3341 return 0;
3342}
3343
3344static __init int setup_numa_zonelist_order(char *s)
3345{
Volodymyr G. Lukiianykecb256f2011-01-13 15:46:26 -08003346 int ret;
3347
3348 if (!s)
3349 return 0;
3350
3351 ret = __parse_numa_zonelist_order(s);
3352 if (ret == 0)
3353 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
3354
3355 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003356}
3357early_param("numa_zonelist_order", setup_numa_zonelist_order);
3358
3359/*
3360 * sysctl handler for numa_zonelist_order
3361 */
3362int numa_zonelist_order_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003363 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003364 loff_t *ppos)
3365{
3366 char saved_string[NUMA_ZONELIST_ORDER_LEN];
3367 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01003368 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003369
Andi Kleen443c6f12009-12-23 21:00:47 +01003370 mutex_lock(&zl_order_mutex);
Chen Gangdacbde02013-07-03 15:02:35 -07003371 if (write) {
3372 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
3373 ret = -EINVAL;
3374 goto out;
3375 }
3376 strcpy(saved_string, (char *)table->data);
3377 }
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003378 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003379 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01003380 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003381 if (write) {
3382 int oldval = user_zonelist_order;
Chen Gangdacbde02013-07-03 15:02:35 -07003383
3384 ret = __parse_numa_zonelist_order((char *)table->data);
3385 if (ret) {
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003386 /*
3387 * bogus value. restore saved string
3388 */
Chen Gangdacbde02013-07-03 15:02:35 -07003389 strncpy((char *)table->data, saved_string,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003390 NUMA_ZONELIST_ORDER_LEN);
3391 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003392 } else if (oldval != user_zonelist_order) {
3393 mutex_lock(&zonelists_mutex);
Jiang Liu9adb62a2012-07-31 16:43:28 -07003394 build_all_zonelists(NULL, NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003395 mutex_unlock(&zonelists_mutex);
3396 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003397 }
Andi Kleen443c6f12009-12-23 21:00:47 +01003398out:
3399 mutex_unlock(&zl_order_mutex);
3400 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003401}
3402
3403
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003404#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003405static int node_load[MAX_NUMNODES];
3406
Linus Torvalds1da177e2005-04-16 15:20:36 -07003407/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07003408 * 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 -07003409 * @node: node whose fallback list we're appending
3410 * @used_node_mask: nodemask_t of already used nodes
3411 *
3412 * We use a number of factors to determine which is the next node that should
3413 * appear on a given node's fallback list. The node should not have appeared
3414 * already in @node's fallback list, and it should be the next closest node
3415 * according to the distance array (which contains arbitrary distance values
3416 * from each node to each node in the system), and should also prefer nodes
3417 * with no CPUs, since presumably they'll have very little allocation pressure
3418 * on them otherwise.
3419 * It returns -1 if no node is found.
3420 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003421static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003422{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01003423 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003424 int min_val = INT_MAX;
David Rientjes00ef2d22013-02-22 16:35:36 -08003425 int best_node = NUMA_NO_NODE;
Rusty Russella70f7302009-03-13 14:49:46 +10303426 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003427
Linus Torvalds4cf808e2006-02-17 20:38:21 +01003428 /* Use the local node if we haven't already */
3429 if (!node_isset(node, *used_node_mask)) {
3430 node_set(node, *used_node_mask);
3431 return node;
3432 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003433
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08003434 for_each_node_state(n, N_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003435
3436 /* Don't want a node to appear more than once */
3437 if (node_isset(n, *used_node_mask))
3438 continue;
3439
Linus Torvalds1da177e2005-04-16 15:20:36 -07003440 /* Use the distance array to find the distance */
3441 val = node_distance(node, n);
3442
Linus Torvalds4cf808e2006-02-17 20:38:21 +01003443 /* Penalize nodes under us ("prefer the next node") */
3444 val += (n < node);
3445
Linus Torvalds1da177e2005-04-16 15:20:36 -07003446 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10303447 tmp = cpumask_of_node(n);
3448 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003449 val += PENALTY_FOR_NODE_WITH_CPUS;
3450
3451 /* Slight preference for less loaded node */
3452 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
3453 val += node_load[n];
3454
3455 if (val < min_val) {
3456 min_val = val;
3457 best_node = n;
3458 }
3459 }
3460
3461 if (best_node >= 0)
3462 node_set(best_node, *used_node_mask);
3463
3464 return best_node;
3465}
3466
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003467
3468/*
3469 * Build zonelists ordered by node and zones within node.
3470 * This results in maximum locality--normal zone overflows into local
3471 * DMA zone, if any--but risks exhausting DMA zone.
3472 */
3473static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003474{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003475 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003476 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003477
Mel Gorman54a6eb52008-04-28 02:12:16 -07003478 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07003479 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07003480 ;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003481 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gormandd1a2392008-04-28 02:12:17 -07003482 zonelist->_zonerefs[j].zone = NULL;
3483 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003484}
3485
3486/*
Christoph Lameter523b9452007-10-16 01:25:37 -07003487 * Build gfp_thisnode zonelists
3488 */
3489static void build_thisnode_zonelists(pg_data_t *pgdat)
3490{
Christoph Lameter523b9452007-10-16 01:25:37 -07003491 int j;
3492 struct zonelist *zonelist;
3493
Mel Gorman54a6eb52008-04-28 02:12:16 -07003494 zonelist = &pgdat->node_zonelists[1];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003495 j = build_zonelists_node(pgdat, zonelist, 0);
Mel Gormandd1a2392008-04-28 02:12:17 -07003496 zonelist->_zonerefs[j].zone = NULL;
3497 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07003498}
3499
3500/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003501 * Build zonelists ordered by zone and nodes within zones.
3502 * This results in conserving DMA zone[s] until all Normal memory is
3503 * exhausted, but results in overflowing to remote node while memory
3504 * may still exist in local DMA zone.
3505 */
3506static int node_order[MAX_NUMNODES];
3507
3508static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
3509{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003510 int pos, j, node;
3511 int zone_type; /* needs to be signed */
3512 struct zone *z;
3513 struct zonelist *zonelist;
3514
Mel Gorman54a6eb52008-04-28 02:12:16 -07003515 zonelist = &pgdat->node_zonelists[0];
3516 pos = 0;
3517 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
3518 for (j = 0; j < nr_nodes; j++) {
3519 node = node_order[j];
3520 z = &NODE_DATA(node)->node_zones[zone_type];
3521 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07003522 zoneref_set_zone(z,
3523 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07003524 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003525 }
3526 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003527 }
Mel Gormandd1a2392008-04-28 02:12:17 -07003528 zonelist->_zonerefs[pos].zone = NULL;
3529 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003530}
3531
3532static int default_zonelist_order(void)
3533{
3534 int nid, zone_type;
Pintu Kumarb8af2942013-09-11 14:20:34 -07003535 unsigned long low_kmem_size, total_size;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003536 struct zone *z;
3537 int average_size;
3538 /*
Pintu Kumarb8af2942013-09-11 14:20:34 -07003539 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003540 * If they are really small and used heavily, the system can fall
3541 * into OOM very easily.
David Rientjese325c902010-05-24 14:32:13 -07003542 * This function detect ZONE_DMA/DMA32 size and configures zone order.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003543 */
3544 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
3545 low_kmem_size = 0;
3546 total_size = 0;
3547 for_each_online_node(nid) {
3548 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3549 z = &NODE_DATA(nid)->node_zones[zone_type];
3550 if (populated_zone(z)) {
3551 if (zone_type < ZONE_NORMAL)
Jiang Liu4f9f4772013-07-03 15:03:07 -07003552 low_kmem_size += z->managed_pages;
3553 total_size += z->managed_pages;
David Rientjese325c902010-05-24 14:32:13 -07003554 } else if (zone_type == ZONE_NORMAL) {
3555 /*
3556 * If any node has only lowmem, then node order
3557 * is preferred to allow kernel allocations
3558 * locally; otherwise, they can easily infringe
3559 * on other nodes when there is an abundance of
3560 * lowmem available to allocate from.
3561 */
3562 return ZONELIST_ORDER_NODE;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003563 }
3564 }
3565 }
3566 if (!low_kmem_size || /* there are no DMA area. */
3567 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
3568 return ZONELIST_ORDER_NODE;
3569 /*
3570 * look into each node's config.
Pintu Kumarb8af2942013-09-11 14:20:34 -07003571 * If there is a node whose DMA/DMA32 memory is very big area on
3572 * local memory, NODE_ORDER may be suitable.
3573 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003574 average_size = total_size /
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08003575 (nodes_weight(node_states[N_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003576 for_each_online_node(nid) {
3577 low_kmem_size = 0;
3578 total_size = 0;
3579 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3580 z = &NODE_DATA(nid)->node_zones[zone_type];
3581 if (populated_zone(z)) {
3582 if (zone_type < ZONE_NORMAL)
3583 low_kmem_size += z->present_pages;
3584 total_size += z->present_pages;
3585 }
3586 }
3587 if (low_kmem_size &&
3588 total_size > average_size && /* ignore small node */
3589 low_kmem_size > total_size * 70/100)
3590 return ZONELIST_ORDER_NODE;
3591 }
3592 return ZONELIST_ORDER_ZONE;
3593}
3594
3595static void set_zonelist_order(void)
3596{
3597 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
3598 current_zonelist_order = default_zonelist_order();
3599 else
3600 current_zonelist_order = user_zonelist_order;
3601}
3602
3603static void build_zonelists(pg_data_t *pgdat)
3604{
3605 int j, node, load;
3606 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003607 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003608 int local_node, prev_node;
3609 struct zonelist *zonelist;
3610 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003611
3612 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07003613 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003614 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07003615 zonelist->_zonerefs[0].zone = NULL;
3616 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003617 }
3618
3619 /* NUMA-aware ordering of nodes */
3620 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003621 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003622 prev_node = local_node;
3623 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003624
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003625 memset(node_order, 0, sizeof(node_order));
3626 j = 0;
3627
Linus Torvalds1da177e2005-04-16 15:20:36 -07003628 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
3629 /*
3630 * We don't want to pressure a particular node.
3631 * So adding penalty to the first node in same
3632 * distance group to make it round-robin.
3633 */
David Rientjes957f8222012-10-08 16:33:24 -07003634 if (node_distance(local_node, node) !=
3635 node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003636 node_load[node] = load;
3637
Linus Torvalds1da177e2005-04-16 15:20:36 -07003638 prev_node = node;
3639 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003640 if (order == ZONELIST_ORDER_NODE)
3641 build_zonelists_in_node_order(pgdat, node);
3642 else
3643 node_order[j++] = node; /* remember order */
3644 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003645
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003646 if (order == ZONELIST_ORDER_ZONE) {
3647 /* calculate node order -- i.e., DMA last! */
3648 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003649 }
Christoph Lameter523b9452007-10-16 01:25:37 -07003650
3651 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003652}
3653
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003654/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003655static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003656{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003657 struct zonelist *zonelist;
3658 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07003659 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003660
Mel Gorman54a6eb52008-04-28 02:12:16 -07003661 zonelist = &pgdat->node_zonelists[0];
3662 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3663 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07003664 for (z = zonelist->_zonerefs; z->zone; z++)
3665 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003666}
3667
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003668#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3669/*
3670 * Return node id of node used for "local" allocations.
3671 * I.e., first node id of first zone in arg node's generic zonelist.
3672 * Used for initializing percpu 'numa_mem', which is used primarily
3673 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3674 */
3675int local_memory_node(int node)
3676{
3677 struct zone *zone;
3678
3679 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3680 gfp_zone(GFP_KERNEL),
3681 NULL,
3682 &zone);
3683 return zone->node;
3684}
3685#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003686
Linus Torvalds1da177e2005-04-16 15:20:36 -07003687#else /* CONFIG_NUMA */
3688
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003689static void set_zonelist_order(void)
3690{
3691 current_zonelist_order = ZONELIST_ORDER_ZONE;
3692}
3693
3694static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003695{
Christoph Lameter19655d32006-09-25 23:31:19 -07003696 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003697 enum zone_type j;
3698 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003699
3700 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003701
Mel Gorman54a6eb52008-04-28 02:12:16 -07003702 zonelist = &pgdat->node_zonelists[0];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003703 j = build_zonelists_node(pgdat, zonelist, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003704
Mel Gorman54a6eb52008-04-28 02:12:16 -07003705 /*
3706 * Now we build the zonelist so that it contains the zones
3707 * of all the other nodes.
3708 * We don't want to pressure a particular node, so when
3709 * building the zones for node N, we make sure that the
3710 * zones coming right after the local ones are those from
3711 * node N+1 (modulo N)
3712 */
3713 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3714 if (!node_online(node))
3715 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003716 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003717 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07003718 for (node = 0; node < local_node; node++) {
3719 if (!node_online(node))
3720 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003721 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gorman54a6eb52008-04-28 02:12:16 -07003722 }
3723
Mel Gormandd1a2392008-04-28 02:12:17 -07003724 zonelist->_zonerefs[j].zone = NULL;
3725 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003726}
3727
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003728/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003729static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003730{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003731 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003732}
3733
Linus Torvalds1da177e2005-04-16 15:20:36 -07003734#endif /* CONFIG_NUMA */
3735
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003736/*
3737 * Boot pageset table. One per cpu which is going to be used for all
3738 * zones and all nodes. The parameters will be set in such a way
3739 * that an item put on a list will immediately be handed over to
3740 * the buddy list. This is safe since pageset manipulation is done
3741 * with interrupts disabled.
3742 *
3743 * The boot_pagesets must be kept even after bootup is complete for
3744 * unused processors and/or zones. They do play a role for bootstrapping
3745 * hotplugged processors.
3746 *
3747 * zoneinfo_show() and maybe other functions do
3748 * not check if the processor is online before following the pageset pointer.
3749 * Other parts of the kernel may not check if the zone is available.
3750 */
3751static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3752static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07003753static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003754
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003755/*
3756 * Global mutex to protect against size modification of zonelists
3757 * as well as to serialize pageset setup for the new populated zone.
3758 */
3759DEFINE_MUTEX(zonelists_mutex);
3760
Rusty Russell9b1a4d32008-07-28 12:16:30 -05003761/* return values int ....just for stop_machine() */
Jiang Liu4ed7e022012-07-31 16:43:35 -07003762static int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003763{
Yasunori Goto68113782006-06-23 02:03:11 -07003764 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003765 int cpu;
Jiang Liu9adb62a2012-07-31 16:43:28 -07003766 pg_data_t *self = data;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003767
Bo Liu7f9cfb32009-08-18 14:11:19 -07003768#ifdef CONFIG_NUMA
3769 memset(node_load, 0, sizeof(node_load));
3770#endif
Jiang Liu9adb62a2012-07-31 16:43:28 -07003771
3772 if (self && !node_online(self->node_id)) {
3773 build_zonelists(self);
3774 build_zonelist_cache(self);
3775 }
3776
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003777 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07003778 pg_data_t *pgdat = NODE_DATA(nid);
3779
3780 build_zonelists(pgdat);
3781 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003782 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003783
3784 /*
3785 * Initialize the boot_pagesets that are going to be used
3786 * for bootstrapping processors. The real pagesets for
3787 * each zone will be allocated later when the per cpu
3788 * allocator is available.
3789 *
3790 * boot_pagesets are used also for bootstrapping offline
3791 * cpus if the system is already booted because the pagesets
3792 * are needed to initialize allocators on a specific cpu too.
3793 * F.e. the percpu allocator needs the page allocator which
3794 * needs the percpu allocator in order to allocate its pagesets
3795 * (a chicken-egg dilemma).
3796 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003797 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003798 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3799
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003800#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3801 /*
3802 * We now know the "local memory node" for each node--
3803 * i.e., the node of the first zone in the generic zonelist.
3804 * Set up numa_mem percpu variable for on-line cpus. During
3805 * boot, only the boot cpu should be on-line; we'll init the
3806 * secondary cpus' numa_mem as they come on-line. During
3807 * node/memory hotplug, we'll fixup all on-line cpus.
3808 */
3809 if (cpu_online(cpu))
3810 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3811#endif
3812 }
3813
Yasunori Goto68113782006-06-23 02:03:11 -07003814 return 0;
3815}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003816
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003817/*
3818 * Called with zonelists_mutex held always
3819 * unless system_state == SYSTEM_BOOTING.
3820 */
Jiang Liu9adb62a2012-07-31 16:43:28 -07003821void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
Yasunori Goto68113782006-06-23 02:03:11 -07003822{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003823 set_zonelist_order();
3824
Yasunori Goto68113782006-06-23 02:03:11 -07003825 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07003826 __build_all_zonelists(NULL);
Mel Gorman68ad8df2008-07-23 21:26:52 -07003827 mminit_verify_zonelist();
Yasunori Goto68113782006-06-23 02:03:11 -07003828 cpuset_init_current_mems_allowed();
3829 } else {
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003830#ifdef CONFIG_MEMORY_HOTPLUG
Jiang Liu9adb62a2012-07-31 16:43:28 -07003831 if (zone)
3832 setup_zone_pageset(zone);
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003833#endif
Cody P Schaferdd1895e2013-07-03 15:01:36 -07003834 /* we have to stop all cpus to guarantee there is no user
3835 of zonelist */
Jiang Liu9adb62a2012-07-31 16:43:28 -07003836 stop_machine(__build_all_zonelists, pgdat, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07003837 /* cpuset refresh routine should be here */
3838 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07003839 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003840 /*
3841 * Disable grouping by mobility if the number of pages in the
3842 * system is too low to allow the mechanism to work. It would be
3843 * more accurate, but expensive to check per-zone. This check is
3844 * made on memory-hotadd so a system can start with mobility
3845 * disabled and enable it later
3846 */
Mel Gormand9c23402007-10-16 01:26:01 -07003847 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003848 page_group_by_mobility_disabled = 1;
3849 else
3850 page_group_by_mobility_disabled = 0;
3851
3852 printk("Built %i zonelists in %s order, mobility grouping %s. "
3853 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003854 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003855 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003856 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003857 vm_total_pages);
3858#ifdef CONFIG_NUMA
3859 printk("Policy zone: %s\n", zone_names[policy_zone]);
3860#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003861}
3862
3863/*
3864 * Helper functions to size the waitqueue hash table.
3865 * Essentially these want to choose hash table sizes sufficiently
3866 * large so that collisions trying to wait on pages are rare.
3867 * But in fact, the number of active page waitqueues on typical
3868 * systems is ridiculously low, less than 200. So this is even
3869 * conservative, even though it seems large.
3870 *
3871 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3872 * waitqueues, i.e. the size of the waitq table given the number of pages.
3873 */
3874#define PAGES_PER_WAITQUEUE 256
3875
Yasunori Gotocca448f2006-06-23 02:03:10 -07003876#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07003877static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003878{
3879 unsigned long size = 1;
3880
3881 pages /= PAGES_PER_WAITQUEUE;
3882
3883 while (size < pages)
3884 size <<= 1;
3885
3886 /*
3887 * Once we have dozens or even hundreds of threads sleeping
3888 * on IO we've got bigger problems than wait queue collision.
3889 * Limit the size of the wait table to a reasonable size.
3890 */
3891 size = min(size, 4096UL);
3892
3893 return max(size, 4UL);
3894}
Yasunori Gotocca448f2006-06-23 02:03:10 -07003895#else
3896/*
3897 * A zone's size might be changed by hot-add, so it is not possible to determine
3898 * a suitable size for its wait_table. So we use the maximum size now.
3899 *
3900 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3901 *
3902 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3903 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3904 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3905 *
3906 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3907 * or more by the traditional way. (See above). It equals:
3908 *
3909 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3910 * ia64(16K page size) : = ( 8G + 4M)byte.
3911 * powerpc (64K page size) : = (32G +16M)byte.
3912 */
3913static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3914{
3915 return 4096UL;
3916}
3917#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003918
3919/*
3920 * This is an integer logarithm so that shifts can be used later
3921 * to extract the more random high bits from the multiplicative
3922 * hash function before the remainder is taken.
3923 */
3924static inline unsigned long wait_table_bits(unsigned long size)
3925{
3926 return ffz(~size);
3927}
3928
Mel Gorman56fd56b2007-10-16 01:25:58 -07003929/*
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07003930 * Check if a pageblock contains reserved pages
3931 */
3932static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
3933{
3934 unsigned long pfn;
3935
3936 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
3937 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
3938 return 1;
3939 }
3940 return 0;
3941}
3942
3943/*
Mel Gormand9c23402007-10-16 01:26:01 -07003944 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07003945 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3946 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07003947 * higher will lead to a bigger reserve which will get freed as contiguous
3948 * blocks as reclaim kicks in
3949 */
3950static void setup_zone_migrate_reserve(struct zone *zone)
3951{
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07003952 unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003953 struct page *page;
Mel Gorman78986a62009-09-21 17:03:02 -07003954 unsigned long block_migratetype;
3955 int reserve;
Yasuaki Ishimatsu943dca12014-01-21 15:49:06 -08003956 int old_reserve;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003957
Michal Hockod02156382011-12-08 14:34:27 -08003958 /*
3959 * Get the start pfn, end pfn and the number of blocks to reserve
3960 * We have to be careful to be aligned to pageblock_nr_pages to
3961 * make sure that we always check pfn_valid for the first page in
3962 * the block.
3963 */
Mel Gorman56fd56b2007-10-16 01:25:58 -07003964 start_pfn = zone->zone_start_pfn;
Cody P Schafer108bcc92013-02-22 16:35:23 -08003965 end_pfn = zone_end_pfn(zone);
Michal Hockod02156382011-12-08 14:34:27 -08003966 start_pfn = roundup(start_pfn, pageblock_nr_pages);
Mel Gorman41858962009-06-16 15:32:12 -07003967 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07003968 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003969
Mel Gorman78986a62009-09-21 17:03:02 -07003970 /*
3971 * Reserve blocks are generally in place to help high-order atomic
3972 * allocations that are short-lived. A min_free_kbytes value that
3973 * would result in more than 2 reserve blocks for atomic allocations
3974 * is assumed to be in place to help anti-fragmentation for the
3975 * future allocation of hugepages at runtime.
3976 */
3977 reserve = min(2, reserve);
Yasuaki Ishimatsu943dca12014-01-21 15:49:06 -08003978 old_reserve = zone->nr_migrate_reserve_block;
3979
3980 /* When memory hot-add, we almost always need to do nothing */
3981 if (reserve == old_reserve)
3982 return;
3983 zone->nr_migrate_reserve_block = reserve;
Mel Gorman78986a62009-09-21 17:03:02 -07003984
Mel Gormand9c23402007-10-16 01:26:01 -07003985 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07003986 if (!pfn_valid(pfn))
3987 continue;
3988 page = pfn_to_page(pfn);
3989
Adam Litke344c7902008-09-02 14:35:38 -07003990 /* Watch out for overlapping nodes */
3991 if (page_to_nid(page) != zone_to_nid(zone))
3992 continue;
3993
Mel Gorman56fd56b2007-10-16 01:25:58 -07003994 block_migratetype = get_pageblock_migratetype(page);
3995
Mel Gorman938929f2012-01-10 15:07:14 -08003996 /* Only test what is necessary when the reserves are not met */
3997 if (reserve > 0) {
3998 /*
3999 * Blocks with reserved pages will never free, skip
4000 * them.
4001 */
4002 block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
4003 if (pageblock_is_reserved(pfn, block_end_pfn))
4004 continue;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004005
Mel Gorman938929f2012-01-10 15:07:14 -08004006 /* If this block is reserved, account for it */
4007 if (block_migratetype == MIGRATE_RESERVE) {
4008 reserve--;
4009 continue;
4010 }
4011
4012 /* Suitable for reserving if this block is movable */
4013 if (block_migratetype == MIGRATE_MOVABLE) {
4014 set_pageblock_migratetype(page,
4015 MIGRATE_RESERVE);
4016 move_freepages_block(zone, page,
4017 MIGRATE_RESERVE);
4018 reserve--;
4019 continue;
4020 }
Yasuaki Ishimatsu943dca12014-01-21 15:49:06 -08004021 } else if (!old_reserve) {
4022 /*
4023 * At boot time we don't need to scan the whole zone
4024 * for turning off MIGRATE_RESERVE.
4025 */
4026 break;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004027 }
4028
4029 /*
4030 * If the reserve is met and this is a previous reserved block,
4031 * take it back
4032 */
4033 if (block_migratetype == MIGRATE_RESERVE) {
4034 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4035 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4036 }
4037 }
4038}
Mel Gormanac0e5b72007-10-16 01:25:58 -07004039
Linus Torvalds1da177e2005-04-16 15:20:36 -07004040/*
4041 * Initially all pages are reserved - free ones are freed
4042 * up by free_all_bootmem() once the early boot process is
4043 * done. Non-atomic initialization, single-pass.
4044 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01004045void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa02007-01-10 23:15:30 -08004046 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004047{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004048 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07004049 unsigned long end_pfn = start_pfn + size;
4050 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004051 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004052
Hugh Dickins22b31ee2009-01-06 14:40:09 -08004053 if (highest_memmap_pfn < end_pfn - 1)
4054 highest_memmap_pfn = end_pfn - 1;
4055
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004056 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08004057 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa02007-01-10 23:15:30 -08004058 /*
4059 * There can be holes in boot-time mem_map[]s
4060 * handed to this function. They do not
4061 * exist on hotplugged memory.
4062 */
4063 if (context == MEMMAP_EARLY) {
4064 if (!early_pfn_valid(pfn))
4065 continue;
4066 if (!early_pfn_in_nid(pfn, nid))
4067 continue;
4068 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004069 page = pfn_to_page(pfn);
4070 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07004071 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08004072 init_page_count(page);
Mel Gorman22b751c2013-02-22 16:34:59 -08004073 page_mapcount_reset(page);
Peter Zijlstra90572892013-10-07 11:29:20 +01004074 page_cpupid_reset_last(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004075 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004076 /*
4077 * Mark the block movable so that blocks are reserved for
4078 * movable at startup. This will force kernel allocations
4079 * to reserve their blocks rather than leaking throughout
4080 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07004081 * kernel allocations are made. Later some blocks near
4082 * the start are marked MIGRATE_RESERVE by
4083 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004084 *
4085 * bitmap is created for zone's valid pfn range. but memmap
4086 * can be created for invalid pages (for alignment)
4087 * check here not to call set_pageblock_migratetype() against
4088 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004089 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004090 if ((z->zone_start_pfn <= pfn)
Cody P Schafer108bcc92013-02-22 16:35:23 -08004091 && (pfn < zone_end_pfn(z))
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004092 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07004093 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004094
Linus Torvalds1da177e2005-04-16 15:20:36 -07004095 INIT_LIST_HEAD(&page->lru);
4096#ifdef WANT_PAGE_VIRTUAL
4097 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
4098 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07004099 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004100#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004101 }
4102}
4103
Andi Kleen1e548de2008-02-04 22:29:26 -08004104static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004105{
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004106 int order, t;
4107 for_each_migratetype_order(order, t) {
4108 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004109 zone->free_area[order].nr_free = 0;
4110 }
4111}
4112
4113#ifndef __HAVE_ARCH_MEMMAP_INIT
4114#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa02007-01-10 23:15:30 -08004115 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004116#endif
4117
Jiang Liu4ed7e022012-07-31 16:43:35 -07004118static int __meminit zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004119{
David Howells3a6be872009-05-06 16:03:03 -07004120#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004121 int batch;
4122
4123 /*
4124 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07004125 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004126 *
4127 * OK, so we don't know how big the cache is. So guess.
4128 */
Jiang Liub40da042013-02-22 16:33:52 -08004129 batch = zone->managed_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07004130 if (batch * PAGE_SIZE > 512 * 1024)
4131 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004132 batch /= 4; /* We effectively *= 4 below */
4133 if (batch < 1)
4134 batch = 1;
4135
4136 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004137 * Clamp the batch to a 2^n - 1 value. Having a power
4138 * of 2 value was found to be more likely to have
4139 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004140 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004141 * For example if 2 tasks are alternately allocating
4142 * batches of pages, one task can end up with a lot
4143 * of pages of one half of the possible page colors
4144 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004145 */
David Howells91552032009-05-06 16:03:02 -07004146 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07004147
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004148 return batch;
David Howells3a6be872009-05-06 16:03:03 -07004149
4150#else
4151 /* The deferral and batching of frees should be suppressed under NOMMU
4152 * conditions.
4153 *
4154 * The problem is that NOMMU needs to be able to allocate large chunks
4155 * of contiguous memory as there's no hardware page translation to
4156 * assemble apparent contiguous memory from discontiguous pages.
4157 *
4158 * Queueing large contiguous runs of pages for batching, however,
4159 * causes the pages to actually be freed in smaller chunks. As there
4160 * can be a significant delay between the individual batches being
4161 * recycled, this leads to the once large chunks of space being
4162 * fragmented and becoming unavailable for high-order allocations.
4163 */
4164 return 0;
4165#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004166}
4167
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004168/*
4169 * pcp->high and pcp->batch values are related and dependent on one another:
4170 * ->batch must never be higher then ->high.
4171 * The following function updates them in a safe manner without read side
4172 * locking.
4173 *
4174 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4175 * those fields changing asynchronously (acording the the above rule).
4176 *
4177 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4178 * outside of boot time (or some other assurance that no concurrent updaters
4179 * exist).
4180 */
4181static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
4182 unsigned long batch)
4183{
4184 /* start with a fail safe value for batch */
4185 pcp->batch = 1;
4186 smp_wmb();
4187
4188 /* Update high, then batch, in order */
4189 pcp->high = high;
4190 smp_wmb();
4191
4192 pcp->batch = batch;
4193}
4194
Cody P Schafer36640332013-07-03 15:01:40 -07004195/* a companion to pageset_set_high() */
Cody P Schafer4008bab2013-07-03 15:01:28 -07004196static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
4197{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004198 pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
Cody P Schafer4008bab2013-07-03 15:01:28 -07004199}
4200
Cody P Schafer88c90db2013-07-03 15:01:35 -07004201static void pageset_init(struct per_cpu_pageset *p)
Christoph Lameter2caaad42005-06-21 17:15:00 -07004202{
4203 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004204 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004205
Magnus Damm1c6fe942005-10-26 01:58:59 -07004206 memset(p, 0, sizeof(*p));
4207
Christoph Lameter3dfa5722008-02-04 22:29:19 -08004208 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004209 pcp->count = 0;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004210 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
4211 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07004212}
4213
Cody P Schafer88c90db2013-07-03 15:01:35 -07004214static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
4215{
4216 pageset_init(p);
4217 pageset_set_batch(p, batch);
4218}
4219
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004220/*
Cody P Schafer36640332013-07-03 15:01:40 -07004221 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004222 * to the value high for the pageset p.
4223 */
Cody P Schafer36640332013-07-03 15:01:40 -07004224static void pageset_set_high(struct per_cpu_pageset *p,
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004225 unsigned long high)
4226{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004227 unsigned long batch = max(1UL, high / 4);
4228 if ((high / 4) > (PAGE_SHIFT * 8))
4229 batch = PAGE_SHIFT * 8;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004230
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004231 pageset_update(&p->pcp, high, batch);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004232}
4233
Cody P Schafer169f6c12013-07-03 15:01:41 -07004234static void __meminit pageset_set_high_and_batch(struct zone *zone,
4235 struct per_cpu_pageset *pcp)
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004236{
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004237 if (percpu_pagelist_fraction)
Cody P Schafer36640332013-07-03 15:01:40 -07004238 pageset_set_high(pcp,
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004239 (zone->managed_pages /
4240 percpu_pagelist_fraction));
4241 else
4242 pageset_set_batch(pcp, zone_batchsize(zone));
4243}
4244
Cody P Schafer169f6c12013-07-03 15:01:41 -07004245static void __meminit zone_pageset_init(struct zone *zone, int cpu)
4246{
4247 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4248
4249 pageset_init(pcp);
4250 pageset_set_high_and_batch(zone, pcp);
4251}
4252
Jiang Liu4ed7e022012-07-31 16:43:35 -07004253static void __meminit setup_zone_pageset(struct zone *zone)
Wu Fengguang319774e2010-05-24 14:32:49 -07004254{
4255 int cpu;
Wu Fengguang319774e2010-05-24 14:32:49 -07004256 zone->pageset = alloc_percpu(struct per_cpu_pageset);
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004257 for_each_possible_cpu(cpu)
4258 zone_pageset_init(zone, cpu);
Wu Fengguang319774e2010-05-24 14:32:49 -07004259}
4260
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004261/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004262 * Allocate per cpu pagesets and initialize them.
4263 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07004264 */
Al Viro78d99552005-12-15 09:18:25 +00004265void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004266{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004267 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004268
Wu Fengguang319774e2010-05-24 14:32:49 -07004269 for_each_populated_zone(zone)
4270 setup_zone_pageset(zone);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004271}
4272
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004273static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07004274int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07004275{
4276 int i;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004277 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07004278
4279 /*
4280 * The per-page waitqueue mechanism uses hashed waitqueues
4281 * per zone.
4282 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07004283 zone->wait_table_hash_nr_entries =
4284 wait_table_hash_nr_entries(zone_size_pages);
4285 zone->wait_table_bits =
4286 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004287 alloc_size = zone->wait_table_hash_nr_entries
4288 * sizeof(wait_queue_head_t);
4289
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07004290 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07004291 zone->wait_table = (wait_queue_head_t *)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004292 memblock_virt_alloc_node_nopanic(
4293 alloc_size, zone->zone_pgdat->node_id);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004294 } else {
4295 /*
4296 * This case means that a zone whose size was 0 gets new memory
4297 * via memory hot-add.
4298 * But it may be the case that a new node was hot-added. In
4299 * this case vmalloc() will not be able to use this new node's
4300 * memory - this wait_table must be initialized to use this new
4301 * node itself as well.
4302 * To use this new node's memory, further consideration will be
4303 * necessary.
4304 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07004305 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004306 }
4307 if (!zone->wait_table)
4308 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07004309
Pintu Kumarb8af2942013-09-11 14:20:34 -07004310 for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07004311 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004312
4313 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004314}
4315
Matt Tolentinoc09b4242006-01-17 07:03:44 +01004316static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07004317{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004318 /*
4319 * per cpu subsystem is not up at this point. The following code
4320 * relies on the ability of the linker to provide the
4321 * offset of a (static) per cpu variable into the per cpu area.
4322 */
4323 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07004324
Xishi Qiub38a8722013-11-12 15:07:20 -08004325 if (populated_zone(zone))
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004326 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
4327 zone->name, zone->present_pages,
4328 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07004329}
4330
Jiang Liu4ed7e022012-07-31 16:43:35 -07004331int __meminit init_currently_empty_zone(struct zone *zone,
Yasunori Goto718127c2006-06-23 02:03:10 -07004332 unsigned long zone_start_pfn,
Dave Hansena2f3aa02007-01-10 23:15:30 -08004333 unsigned long size,
4334 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07004335{
4336 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004337 int ret;
4338 ret = zone_wait_table_init(zone, size);
4339 if (ret)
4340 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07004341 pgdat->nr_zones = zone_idx(zone) + 1;
4342
Dave Hansened8ece22005-10-29 18:16:50 -07004343 zone->zone_start_pfn = zone_start_pfn;
4344
Mel Gorman708614e2008-07-23 21:26:51 -07004345 mminit_dprintk(MMINIT_TRACE, "memmap_init",
4346 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4347 pgdat->node_id,
4348 (unsigned long)zone_idx(zone),
4349 zone_start_pfn, (zone_start_pfn + size));
4350
Andi Kleen1e548de2008-02-04 22:29:26 -08004351 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07004352
4353 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004354}
4355
Tejun Heo0ee332c2011-12-08 10:22:09 -08004356#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07004357#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4358/*
4359 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4360 * Architectures may implement their own version but if add_active_range()
4361 * was used and there are no special requirements, this is a convenient
4362 * alternative
4363 */
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004364int __meminit __early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004365{
Tejun Heoc13291a2011-07-12 10:46:30 +02004366 unsigned long start_pfn, end_pfn;
Yinghai Lue76b63f2013-09-11 14:22:17 -07004367 int nid;
Russ Anderson7c243c72013-04-29 15:07:59 -07004368 /*
4369 * NOTE: The following SMP-unsafe globals are only used early in boot
4370 * when the kernel is running single-threaded.
4371 */
4372 static unsigned long __meminitdata last_start_pfn, last_end_pfn;
4373 static int __meminitdata last_nid;
4374
4375 if (last_start_pfn <= pfn && pfn < last_end_pfn)
4376 return last_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004377
Yinghai Lue76b63f2013-09-11 14:22:17 -07004378 nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
4379 if (nid != -1) {
4380 last_start_pfn = start_pfn;
4381 last_end_pfn = end_pfn;
4382 last_nid = nid;
4383 }
4384
4385 return nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004386}
4387#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4388
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004389int __meminit early_pfn_to_nid(unsigned long pfn)
4390{
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08004391 int nid;
4392
4393 nid = __early_pfn_to_nid(pfn);
4394 if (nid >= 0)
4395 return nid;
4396 /* just returns 0 */
4397 return 0;
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004398}
4399
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08004400#ifdef CONFIG_NODES_SPAN_OTHER_NODES
4401bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
4402{
4403 int nid;
4404
4405 nid = __early_pfn_to_nid(pfn);
4406 if (nid >= 0 && nid != node)
4407 return false;
4408 return true;
4409}
4410#endif
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004411
Mel Gormanc7132162006-09-27 01:49:43 -07004412/**
Santosh Shilimkar67828322014-01-21 15:50:25 -08004413 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004414 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
Santosh Shilimkar67828322014-01-21 15:50:25 -08004415 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
Mel Gormanc7132162006-09-27 01:49:43 -07004416 *
4417 * If an architecture guarantees that all ranges registered with
4418 * add_active_ranges() contain no holes and may be freed, this
Santosh Shilimkar67828322014-01-21 15:50:25 -08004419 * this function may be used instead of calling memblock_free_early_nid()
4420 * manually.
Mel Gormanc7132162006-09-27 01:49:43 -07004421 */
Tejun Heoc13291a2011-07-12 10:46:30 +02004422void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004423{
Tejun Heoc13291a2011-07-12 10:46:30 +02004424 unsigned long start_pfn, end_pfn;
4425 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004426
Tejun Heoc13291a2011-07-12 10:46:30 +02004427 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
4428 start_pfn = min(start_pfn, max_low_pfn);
4429 end_pfn = min(end_pfn, max_low_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004430
Tejun Heoc13291a2011-07-12 10:46:30 +02004431 if (start_pfn < end_pfn)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004432 memblock_free_early_nid(PFN_PHYS(start_pfn),
4433 (end_pfn - start_pfn) << PAGE_SHIFT,
4434 this_nid);
Mel Gormanc7132162006-09-27 01:49:43 -07004435 }
4436}
4437
4438/**
4439 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004440 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07004441 *
4442 * If an architecture guarantees that all ranges registered with
4443 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004444 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07004445 */
4446void __init sparse_memory_present_with_active_regions(int nid)
4447{
Tejun Heoc13291a2011-07-12 10:46:30 +02004448 unsigned long start_pfn, end_pfn;
4449 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004450
Tejun Heoc13291a2011-07-12 10:46:30 +02004451 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
4452 memory_present(this_nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004453}
4454
4455/**
4456 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004457 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4458 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4459 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07004460 *
4461 * It returns the start and end page frame of a node based on information
4462 * provided by an arch calling add_active_range(). If called for a node
4463 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004464 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07004465 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004466void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004467 unsigned long *start_pfn, unsigned long *end_pfn)
4468{
Tejun Heoc13291a2011-07-12 10:46:30 +02004469 unsigned long this_start_pfn, this_end_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004470 int i;
Tejun Heoc13291a2011-07-12 10:46:30 +02004471
Mel Gormanc7132162006-09-27 01:49:43 -07004472 *start_pfn = -1UL;
4473 *end_pfn = 0;
4474
Tejun Heoc13291a2011-07-12 10:46:30 +02004475 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
4476 *start_pfn = min(*start_pfn, this_start_pfn);
4477 *end_pfn = max(*end_pfn, this_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004478 }
4479
Christoph Lameter633c0662007-10-16 01:25:37 -07004480 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07004481 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07004482}
4483
4484/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07004485 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4486 * assumption is made that zones within a node are ordered in monotonic
4487 * increasing memory addresses so that the "highest" populated zone is used
4488 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004489static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004490{
4491 int zone_index;
4492 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
4493 if (zone_index == ZONE_MOVABLE)
4494 continue;
4495
4496 if (arch_zone_highest_possible_pfn[zone_index] >
4497 arch_zone_lowest_possible_pfn[zone_index])
4498 break;
4499 }
4500
4501 VM_BUG_ON(zone_index == -1);
4502 movable_zone = zone_index;
4503}
4504
4505/*
4506 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004507 * because it is sized independent of architecture. Unlike the other zones,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004508 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4509 * in each node depending on the size of each node and how evenly kernelcore
4510 * is distributed. This helper function adjusts the zone ranges
4511 * provided by the architecture for a given node by using the end of the
4512 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4513 * zones within a node are in order of monotonic increases memory addresses
4514 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004515static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004516 unsigned long zone_type,
4517 unsigned long node_start_pfn,
4518 unsigned long node_end_pfn,
4519 unsigned long *zone_start_pfn,
4520 unsigned long *zone_end_pfn)
4521{
4522 /* Only adjust if ZONE_MOVABLE is on this node */
4523 if (zone_movable_pfn[nid]) {
4524 /* Size ZONE_MOVABLE */
4525 if (zone_type == ZONE_MOVABLE) {
4526 *zone_start_pfn = zone_movable_pfn[nid];
4527 *zone_end_pfn = min(node_end_pfn,
4528 arch_zone_highest_possible_pfn[movable_zone]);
4529
4530 /* Adjust for ZONE_MOVABLE starting within this range */
4531 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
4532 *zone_end_pfn > zone_movable_pfn[nid]) {
4533 *zone_end_pfn = zone_movable_pfn[nid];
4534
4535 /* Check if this whole range is within ZONE_MOVABLE */
4536 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
4537 *zone_start_pfn = *zone_end_pfn;
4538 }
4539}
4540
4541/*
Mel Gormanc7132162006-09-27 01:49:43 -07004542 * Return the number of pages a zone spans in a node, including holes
4543 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4544 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004545static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004546 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004547 unsigned long node_start_pfn,
4548 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004549 unsigned long *ignored)
4550{
Mel Gormanc7132162006-09-27 01:49:43 -07004551 unsigned long zone_start_pfn, zone_end_pfn;
4552
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004553 /* Get the start and end of the zone */
Mel Gormanc7132162006-09-27 01:49:43 -07004554 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4555 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07004556 adjust_zone_range_for_zone_movable(nid, zone_type,
4557 node_start_pfn, node_end_pfn,
4558 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004559
4560 /* Check that this node has pages within the zone's required range */
4561 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4562 return 0;
4563
4564 /* Move the zone boundaries inside the node if necessary */
4565 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4566 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4567
4568 /* Return the spanned pages */
4569 return zone_end_pfn - zone_start_pfn;
4570}
4571
4572/*
4573 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004574 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07004575 */
Yinghai Lu32996252009-12-15 17:59:02 -08004576unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004577 unsigned long range_start_pfn,
4578 unsigned long range_end_pfn)
4579{
Tejun Heo96e907d2011-07-12 10:46:29 +02004580 unsigned long nr_absent = range_end_pfn - range_start_pfn;
4581 unsigned long start_pfn, end_pfn;
4582 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004583
Tejun Heo96e907d2011-07-12 10:46:29 +02004584 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
4585 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
4586 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
4587 nr_absent -= end_pfn - start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004588 }
Tejun Heo96e907d2011-07-12 10:46:29 +02004589 return nr_absent;
Mel Gormanc7132162006-09-27 01:49:43 -07004590}
4591
4592/**
4593 * absent_pages_in_range - Return number of page frames in holes within a range
4594 * @start_pfn: The start PFN to start searching for holes
4595 * @end_pfn: The end PFN to stop searching for holes
4596 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004597 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07004598 */
4599unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4600 unsigned long end_pfn)
4601{
4602 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4603}
4604
4605/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004606static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004607 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004608 unsigned long node_start_pfn,
4609 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004610 unsigned long *ignored)
4611{
Tejun Heo96e907d2011-07-12 10:46:29 +02004612 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
4613 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman9c7cd682006-09-27 01:49:58 -07004614 unsigned long zone_start_pfn, zone_end_pfn;
4615
Tejun Heo96e907d2011-07-12 10:46:29 +02004616 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
4617 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004618
Mel Gorman2a1e2742007-07-17 04:03:12 -07004619 adjust_zone_range_for_zone_movable(nid, zone_type,
4620 node_start_pfn, node_end_pfn,
4621 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004622 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004623}
Mel Gorman0e0b8642006-09-27 01:49:56 -07004624
Tejun Heo0ee332c2011-12-08 10:22:09 -08004625#else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004626static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004627 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004628 unsigned long node_start_pfn,
4629 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004630 unsigned long *zones_size)
4631{
4632 return zones_size[zone_type];
4633}
4634
Paul Mundt6ea6e682007-07-15 23:38:20 -07004635static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004636 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004637 unsigned long node_start_pfn,
4638 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004639 unsigned long *zholes_size)
4640{
4641 if (!zholes_size)
4642 return 0;
4643
4644 return zholes_size[zone_type];
4645}
Yinghai Lu20e69262013-03-01 14:51:27 -08004646
Tejun Heo0ee332c2011-12-08 10:22:09 -08004647#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07004648
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004649static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004650 unsigned long node_start_pfn,
4651 unsigned long node_end_pfn,
4652 unsigned long *zones_size,
4653 unsigned long *zholes_size)
Mel Gormanc7132162006-09-27 01:49:43 -07004654{
4655 unsigned long realtotalpages, totalpages = 0;
4656 enum zone_type i;
4657
4658 for (i = 0; i < MAX_NR_ZONES; i++)
4659 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004660 node_start_pfn,
4661 node_end_pfn,
4662 zones_size);
Mel Gormanc7132162006-09-27 01:49:43 -07004663 pgdat->node_spanned_pages = totalpages;
4664
4665 realtotalpages = totalpages;
4666 for (i = 0; i < MAX_NR_ZONES; i++)
4667 realtotalpages -=
4668 zone_absent_pages_in_node(pgdat->node_id, i,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004669 node_start_pfn, node_end_pfn,
4670 zholes_size);
Mel Gormanc7132162006-09-27 01:49:43 -07004671 pgdat->node_present_pages = realtotalpages;
4672 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4673 realtotalpages);
4674}
4675
Mel Gorman835c1342007-10-16 01:25:47 -07004676#ifndef CONFIG_SPARSEMEM
4677/*
4678 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07004679 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4680 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07004681 * round what is now in bits to nearest long in bits, then return it in
4682 * bytes.
4683 */
Linus Torvalds7c455122013-02-18 09:58:02 -08004684static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07004685{
4686 unsigned long usemapsize;
4687
Linus Torvalds7c455122013-02-18 09:58:02 -08004688 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
Mel Gormand9c23402007-10-16 01:26:01 -07004689 usemapsize = roundup(zonesize, pageblock_nr_pages);
4690 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07004691 usemapsize *= NR_PAGEBLOCK_BITS;
4692 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4693
4694 return usemapsize / 8;
4695}
4696
4697static void __init setup_usemap(struct pglist_data *pgdat,
Linus Torvalds7c455122013-02-18 09:58:02 -08004698 struct zone *zone,
4699 unsigned long zone_start_pfn,
4700 unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07004701{
Linus Torvalds7c455122013-02-18 09:58:02 -08004702 unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
Mel Gorman835c1342007-10-16 01:25:47 -07004703 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08004704 if (usemapsize)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004705 zone->pageblock_flags =
4706 memblock_virt_alloc_node_nopanic(usemapsize,
4707 pgdat->node_id);
Mel Gorman835c1342007-10-16 01:25:47 -07004708}
4709#else
Linus Torvalds7c455122013-02-18 09:58:02 -08004710static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
4711 unsigned long zone_start_pfn, unsigned long zonesize) {}
Mel Gorman835c1342007-10-16 01:25:47 -07004712#endif /* CONFIG_SPARSEMEM */
4713
Mel Gormand9c23402007-10-16 01:26:01 -07004714#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08004715
Mel Gormand9c23402007-10-16 01:26:01 -07004716/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
Chen Gang15ca2202013-09-11 14:20:27 -07004717void __paginginit set_pageblock_order(void)
Mel Gormand9c23402007-10-16 01:26:01 -07004718{
Andrew Morton955c1cd2012-05-29 15:06:31 -07004719 unsigned int order;
4720
Mel Gormand9c23402007-10-16 01:26:01 -07004721 /* Check that pageblock_nr_pages has not already been setup */
4722 if (pageblock_order)
4723 return;
4724
Andrew Morton955c1cd2012-05-29 15:06:31 -07004725 if (HPAGE_SHIFT > PAGE_SHIFT)
4726 order = HUGETLB_PAGE_ORDER;
4727 else
4728 order = MAX_ORDER - 1;
4729
Mel Gormand9c23402007-10-16 01:26:01 -07004730 /*
4731 * Assume the largest contiguous order of interest is a huge page.
Andrew Morton955c1cd2012-05-29 15:06:31 -07004732 * This value may be variable depending on boot parameters on IA64 and
4733 * powerpc.
Mel Gormand9c23402007-10-16 01:26:01 -07004734 */
4735 pageblock_order = order;
4736}
4737#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4738
Mel Gormanba72cb82007-11-28 16:21:13 -08004739/*
4740 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
Andrew Morton955c1cd2012-05-29 15:06:31 -07004741 * is unused as pageblock_order is set at compile-time. See
4742 * include/linux/pageblock-flags.h for the values of pageblock_order based on
4743 * the kernel config
Mel Gormanba72cb82007-11-28 16:21:13 -08004744 */
Chen Gang15ca2202013-09-11 14:20:27 -07004745void __paginginit set_pageblock_order(void)
Mel Gormanba72cb82007-11-28 16:21:13 -08004746{
Mel Gormanba72cb82007-11-28 16:21:13 -08004747}
Mel Gormand9c23402007-10-16 01:26:01 -07004748
4749#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4750
Jiang Liu01cefae2012-12-12 13:52:19 -08004751static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
4752 unsigned long present_pages)
4753{
4754 unsigned long pages = spanned_pages;
4755
4756 /*
4757 * Provide a more accurate estimation if there are holes within
4758 * the zone and SPARSEMEM is in use. If there are holes within the
4759 * zone, each populated memory region may cost us one or two extra
4760 * memmap pages due to alignment because memmap pages for each
4761 * populated regions may not naturally algined on page boundary.
4762 * So the (present_pages >> 4) heuristic is a tradeoff for that.
4763 */
4764 if (spanned_pages > present_pages + (present_pages >> 4) &&
4765 IS_ENABLED(CONFIG_SPARSEMEM))
4766 pages = present_pages;
4767
4768 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
4769}
4770
Linus Torvalds1da177e2005-04-16 15:20:36 -07004771/*
4772 * Set up the zone data structures:
4773 * - mark all pages reserved
4774 * - mark all memory queues empty
4775 * - clear the memory bitmaps
Minchan Kim6527af52012-07-31 16:46:16 -07004776 *
4777 * NOTE: pgdat should get zeroed by caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004778 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08004779static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004780 unsigned long node_start_pfn, unsigned long node_end_pfn,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004781 unsigned long *zones_size, unsigned long *zholes_size)
4782{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07004783 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07004784 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004785 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07004786 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004787
Dave Hansen208d54e2005-10-29 18:16:52 -07004788 pgdat_resize_init(pgdat);
Andrea Arcangeli8177a422012-03-23 20:56:34 +01004789#ifdef CONFIG_NUMA_BALANCING
4790 spin_lock_init(&pgdat->numabalancing_migrate_lock);
4791 pgdat->numabalancing_migrate_nr_pages = 0;
4792 pgdat->numabalancing_migrate_next_window = jiffies;
4793#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004794 init_waitqueue_head(&pgdat->kswapd_wait);
Mel Gorman55150612012-07-31 16:44:35 -07004795 init_waitqueue_head(&pgdat->pfmemalloc_wait);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004796 pgdat_page_cgroup_init(pgdat);
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01004797
Linus Torvalds1da177e2005-04-16 15:20:36 -07004798 for (j = 0; j < MAX_NR_ZONES; j++) {
4799 struct zone *zone = pgdat->node_zones + j;
Jiang Liu9feedc92012-12-12 13:52:12 -08004800 unsigned long size, realsize, freesize, memmap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004801
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004802 size = zone_spanned_pages_in_node(nid, j, node_start_pfn,
4803 node_end_pfn, zones_size);
Jiang Liu9feedc92012-12-12 13:52:12 -08004804 realsize = freesize = size - zone_absent_pages_in_node(nid, j,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004805 node_start_pfn,
4806 node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004807 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004808
Mel Gorman0e0b8642006-09-27 01:49:56 -07004809 /*
Jiang Liu9feedc92012-12-12 13:52:12 -08004810 * Adjust freesize so that it accounts for how much memory
Mel Gorman0e0b8642006-09-27 01:49:56 -07004811 * is used by this zone for memmap. This affects the watermark
4812 * and per-cpu initialisations
4813 */
Jiang Liu01cefae2012-12-12 13:52:19 -08004814 memmap_pages = calc_memmap_size(size, realsize);
Jiang Liu9feedc92012-12-12 13:52:12 -08004815 if (freesize >= memmap_pages) {
4816 freesize -= memmap_pages;
Yinghai Lu5594c8c2009-01-06 14:39:14 -08004817 if (memmap_pages)
4818 printk(KERN_DEBUG
4819 " %s zone: %lu pages used for memmap\n",
4820 zone_names[j], memmap_pages);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004821 } else
4822 printk(KERN_WARNING
Jiang Liu9feedc92012-12-12 13:52:12 -08004823 " %s zone: %lu pages exceeds freesize %lu\n",
4824 zone_names[j], memmap_pages, freesize);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004825
Christoph Lameter62672762007-02-10 01:43:07 -08004826 /* Account for reserved pages */
Jiang Liu9feedc92012-12-12 13:52:12 -08004827 if (j == 0 && freesize > dma_reserve) {
4828 freesize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07004829 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08004830 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004831 }
4832
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004833 if (!is_highmem_idx(j))
Jiang Liu9feedc92012-12-12 13:52:12 -08004834 nr_kernel_pages += freesize;
Jiang Liu01cefae2012-12-12 13:52:19 -08004835 /* Charge for highmem memmap if there are enough kernel pages */
4836 else if (nr_kernel_pages > memmap_pages * 2)
4837 nr_kernel_pages -= memmap_pages;
Jiang Liu9feedc92012-12-12 13:52:12 -08004838 nr_all_pages += freesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004839
4840 zone->spanned_pages = size;
Jiang Liu306f2e92013-02-22 16:33:54 -08004841 zone->present_pages = realsize;
Jiang Liu9feedc92012-12-12 13:52:12 -08004842 /*
4843 * Set an approximate value for lowmem here, it will be adjusted
4844 * when the bootmem allocator frees pages into the buddy system.
4845 * And all highmem pages will be managed by the buddy system.
4846 */
4847 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
Christoph Lameter96146342006-07-03 00:24:13 -07004848#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07004849 zone->node = nid;
Jiang Liu9feedc92012-12-12 13:52:12 -08004850 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07004851 / 100;
Jiang Liu9feedc92012-12-12 13:52:12 -08004852 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07004853#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004854 zone->name = zone_names[j];
4855 spin_lock_init(&zone->lock);
4856 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07004857 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004858 zone->zone_pgdat = pgdat;
Dave Hansened8ece22005-10-29 18:16:50 -07004859 zone_pcp_init(zone);
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07004860
4861 /* For bootup, initialized properly in watermark setup */
4862 mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages);
4863
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004864 lruvec_init(&zone->lruvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004865 if (!size)
4866 continue;
4867
Andrew Morton955c1cd2012-05-29 15:06:31 -07004868 set_pageblock_order();
Linus Torvalds7c455122013-02-18 09:58:02 -08004869 setup_usemap(pgdat, zone, zone_start_pfn, size);
Dave Hansena2f3aa02007-01-10 23:15:30 -08004870 ret = init_currently_empty_zone(zone, zone_start_pfn,
4871 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07004872 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07004873 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004874 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004875 }
4876}
4877
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004878static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004879{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004880 /* Skip empty nodes */
4881 if (!pgdat->node_spanned_pages)
4882 return;
4883
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004884#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07004885 /* ia64 gets its own node_mem_map, before this, without bootmem */
4886 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07004887 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004888 struct page *map;
4889
Bob Piccoe984bb42006-05-20 15:00:31 -07004890 /*
4891 * The zone's endpoints aren't required to be MAX_ORDER
4892 * aligned but the node_mem_map endpoints must be in order
4893 * for the buddy allocator to function correctly.
4894 */
4895 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
Cody P Schafer108bcc92013-02-22 16:35:23 -08004896 end = pgdat_end_pfn(pgdat);
Bob Piccoe984bb42006-05-20 15:00:31 -07004897 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4898 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07004899 map = alloc_remap(pgdat->node_id, size);
4900 if (!map)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004901 map = memblock_virt_alloc_node_nopanic(size,
4902 pgdat->node_id);
Bob Piccoe984bb42006-05-20 15:00:31 -07004903 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004904 }
Roman Zippel12d810c2007-05-31 00:40:54 -07004905#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07004906 /*
4907 * With no DISCONTIG, the global mem_map is just set as node 0's
4908 */
Mel Gormanc7132162006-09-27 01:49:43 -07004909 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004910 mem_map = NODE_DATA(0)->node_mem_map;
Tejun Heo0ee332c2011-12-08 10:22:09 -08004911#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07004912 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08004913 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Tejun Heo0ee332c2011-12-08 10:22:09 -08004914#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07004915 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004916#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004917#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004918}
4919
Johannes Weiner9109fb72008-07-23 21:27:20 -07004920void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4921 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004922{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004923 pg_data_t *pgdat = NODE_DATA(nid);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004924 unsigned long start_pfn = 0;
4925 unsigned long end_pfn = 0;
Johannes Weiner9109fb72008-07-23 21:27:20 -07004926
Minchan Kim88fdf752012-07-31 16:46:14 -07004927 /* pg_data_t should be reset to zero when it's allocated */
Linus Torvalds8783b6e2012-08-02 10:37:03 -07004928 WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
Minchan Kim88fdf752012-07-31 16:46:14 -07004929
Linus Torvalds1da177e2005-04-16 15:20:36 -07004930 pgdat->node_id = nid;
4931 pgdat->node_start_pfn = node_start_pfn;
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004932#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4933 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
4934#endif
4935 calculate_node_totalpages(pgdat, start_pfn, end_pfn,
4936 zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004937
4938 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07004939#ifdef CONFIG_FLAT_NODE_MEM_MAP
4940 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4941 nid, (unsigned long)pgdat,
4942 (unsigned long)pgdat->node_mem_map);
4943#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004944
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004945 free_area_init_core(pgdat, start_pfn, end_pfn,
4946 zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004947}
4948
Tejun Heo0ee332c2011-12-08 10:22:09 -08004949#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07004950
4951#if MAX_NUMNODES > 1
4952/*
4953 * Figure out the number of possible node ids.
4954 */
Cody P Schaferf9872ca2013-04-29 15:08:01 -07004955void __init setup_nr_node_ids(void)
Miklos Szeredi418508c2007-05-23 13:57:55 -07004956{
4957 unsigned int node;
4958 unsigned int highest = 0;
4959
4960 for_each_node_mask(node, node_possible_map)
4961 highest = node;
4962 nr_node_ids = highest + 1;
4963}
Miklos Szeredi418508c2007-05-23 13:57:55 -07004964#endif
4965
Mel Gormanc7132162006-09-27 01:49:43 -07004966/**
Tejun Heo1e019792011-07-12 09:45:34 +02004967 * node_map_pfn_alignment - determine the maximum internode alignment
4968 *
4969 * This function should be called after node map is populated and sorted.
4970 * It calculates the maximum power of two alignment which can distinguish
4971 * all the nodes.
4972 *
4973 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
4974 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
4975 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
4976 * shifted, 1GiB is enough and this function will indicate so.
4977 *
4978 * This is used to test whether pfn -> nid mapping of the chosen memory
4979 * model has fine enough granularity to avoid incorrect mapping for the
4980 * populated node map.
4981 *
4982 * Returns the determined alignment in pfn's. 0 if there is no alignment
4983 * requirement (single node).
4984 */
4985unsigned long __init node_map_pfn_alignment(void)
4986{
4987 unsigned long accl_mask = 0, last_end = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02004988 unsigned long start, end, mask;
Tejun Heo1e019792011-07-12 09:45:34 +02004989 int last_nid = -1;
Tejun Heoc13291a2011-07-12 10:46:30 +02004990 int i, nid;
Tejun Heo1e019792011-07-12 09:45:34 +02004991
Tejun Heoc13291a2011-07-12 10:46:30 +02004992 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
Tejun Heo1e019792011-07-12 09:45:34 +02004993 if (!start || last_nid < 0 || last_nid == nid) {
4994 last_nid = nid;
4995 last_end = end;
4996 continue;
4997 }
4998
4999 /*
5000 * Start with a mask granular enough to pin-point to the
5001 * start pfn and tick off bits one-by-one until it becomes
5002 * too coarse to separate the current node from the last.
5003 */
5004 mask = ~((1 << __ffs(start)) - 1);
5005 while (mask && last_end <= (start & (mask << 1)))
5006 mask <<= 1;
5007
5008 /* accumulate all internode masks */
5009 accl_mask |= mask;
5010 }
5011
5012 /* convert mask to number of pages */
5013 return ~accl_mask + 1;
5014}
5015
Mel Gormana6af2bc2007-02-10 01:42:57 -08005016/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07005017static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07005018{
Mel Gormana6af2bc2007-02-10 01:42:57 -08005019 unsigned long min_pfn = ULONG_MAX;
Tejun Heoc13291a2011-07-12 10:46:30 +02005020 unsigned long start_pfn;
5021 int i;
Mel Gorman1abbfb42006-11-23 12:01:41 +00005022
Tejun Heoc13291a2011-07-12 10:46:30 +02005023 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
5024 min_pfn = min(min_pfn, start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005025
Mel Gormana6af2bc2007-02-10 01:42:57 -08005026 if (min_pfn == ULONG_MAX) {
5027 printk(KERN_WARNING
Paul Jackson2bc0d262008-06-22 07:22:17 -07005028 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08005029 return 0;
5030 }
5031
5032 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005033}
5034
5035/**
5036 * find_min_pfn_with_active_regions - Find the minimum PFN registered
5037 *
5038 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005039 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07005040 */
5041unsigned long __init find_min_pfn_with_active_regions(void)
5042{
5043 return find_min_pfn_for_node(MAX_NUMNODES);
5044}
5045
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005046/*
5047 * early_calculate_totalpages()
5048 * Sum pages in active regions for movable zone.
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005049 * Populate N_MEMORY for calculating usable_nodes.
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005050 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07005051static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07005052{
Mel Gorman7e63efe2007-07-17 04:03:15 -07005053 unsigned long totalpages = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02005054 unsigned long start_pfn, end_pfn;
5055 int i, nid;
Mel Gorman7e63efe2007-07-17 04:03:15 -07005056
Tejun Heoc13291a2011-07-12 10:46:30 +02005057 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
5058 unsigned long pages = end_pfn - start_pfn;
5059
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005060 totalpages += pages;
5061 if (pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005062 node_set_state(nid, N_MEMORY);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005063 }
Pintu Kumarb8af2942013-09-11 14:20:34 -07005064 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07005065}
5066
Mel Gorman2a1e2742007-07-17 04:03:12 -07005067/*
5068 * Find the PFN the Movable zone begins in each node. Kernel memory
5069 * is spread evenly between nodes as long as the nodes have enough
5070 * memory. When they don't, some nodes will have more kernelcore than
5071 * others
5072 */
Kautuk Consulb224ef82012-03-21 16:34:15 -07005073static void __init find_zone_movable_pfns_for_nodes(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005074{
5075 int i, nid;
5076 unsigned long usable_startpfn;
5077 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07005078 /* save the state before borrow the nodemask */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005079 nodemask_t saved_node_state = node_states[N_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005080 unsigned long totalpages = early_calculate_totalpages();
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005081 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
Emil Medve136199f2014-04-07 15:37:52 -07005082 struct memblock_region *r;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005083
5084 /* Need to find movable_zone earlier when movable_node is specified. */
5085 find_usable_zone_for_movable();
Mel Gorman2a1e2742007-07-17 04:03:12 -07005086
Mel Gorman7e63efe2007-07-17 04:03:15 -07005087 /*
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005088 * If movable_node is specified, ignore kernelcore and movablecore
5089 * options.
5090 */
5091 if (movable_node_is_enabled()) {
Emil Medve136199f2014-04-07 15:37:52 -07005092 for_each_memblock(memory, r) {
5093 if (!memblock_is_hotpluggable(r))
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005094 continue;
5095
Emil Medve136199f2014-04-07 15:37:52 -07005096 nid = r->nid;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005097
Emil Medve136199f2014-04-07 15:37:52 -07005098 usable_startpfn = PFN_DOWN(r->base);
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005099 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
5100 min(usable_startpfn, zone_movable_pfn[nid]) :
5101 usable_startpfn;
5102 }
5103
5104 goto out2;
5105 }
5106
5107 /*
5108 * If movablecore=nn[KMG] was specified, calculate what size of
Mel Gorman7e63efe2007-07-17 04:03:15 -07005109 * kernelcore that corresponds so that memory usable for
5110 * any allocation type is evenly spread. If both kernelcore
5111 * and movablecore are specified, then the value of kernelcore
5112 * will be used for required_kernelcore if it's greater than
5113 * what movablecore would have allowed.
5114 */
5115 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07005116 unsigned long corepages;
5117
5118 /*
5119 * Round-up so that ZONE_MOVABLE is at least as large as what
5120 * was requested by the user
5121 */
5122 required_movablecore =
5123 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
5124 corepages = totalpages - required_movablecore;
5125
5126 required_kernelcore = max(required_kernelcore, corepages);
5127 }
5128
Yinghai Lu20e69262013-03-01 14:51:27 -08005129 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
5130 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07005131 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005132
5133 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
Mel Gorman2a1e2742007-07-17 04:03:12 -07005134 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
5135
5136restart:
5137 /* Spread kernelcore memory as evenly as possible throughout nodes */
5138 kernelcore_node = required_kernelcore / usable_nodes;
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005139 for_each_node_state(nid, N_MEMORY) {
Tejun Heoc13291a2011-07-12 10:46:30 +02005140 unsigned long start_pfn, end_pfn;
5141
Mel Gorman2a1e2742007-07-17 04:03:12 -07005142 /*
5143 * Recalculate kernelcore_node if the division per node
5144 * now exceeds what is necessary to satisfy the requested
5145 * amount of memory for the kernel
5146 */
5147 if (required_kernelcore < kernelcore_node)
5148 kernelcore_node = required_kernelcore / usable_nodes;
5149
5150 /*
5151 * As the map is walked, we track how much memory is usable
5152 * by the kernel using kernelcore_remaining. When it is
5153 * 0, the rest of the node is usable by ZONE_MOVABLE
5154 */
5155 kernelcore_remaining = kernelcore_node;
5156
5157 /* Go through each range of PFNs within this node */
Tejun Heoc13291a2011-07-12 10:46:30 +02005158 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005159 unsigned long size_pages;
5160
Tejun Heoc13291a2011-07-12 10:46:30 +02005161 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005162 if (start_pfn >= end_pfn)
5163 continue;
5164
5165 /* Account for what is only usable for kernelcore */
5166 if (start_pfn < usable_startpfn) {
5167 unsigned long kernel_pages;
5168 kernel_pages = min(end_pfn, usable_startpfn)
5169 - start_pfn;
5170
5171 kernelcore_remaining -= min(kernel_pages,
5172 kernelcore_remaining);
5173 required_kernelcore -= min(kernel_pages,
5174 required_kernelcore);
5175
5176 /* Continue if range is now fully accounted */
5177 if (end_pfn <= usable_startpfn) {
5178
5179 /*
5180 * Push zone_movable_pfn to the end so
5181 * that if we have to rebalance
5182 * kernelcore across nodes, we will
5183 * not double account here
5184 */
5185 zone_movable_pfn[nid] = end_pfn;
5186 continue;
5187 }
5188 start_pfn = usable_startpfn;
5189 }
5190
5191 /*
5192 * The usable PFN range for ZONE_MOVABLE is from
5193 * start_pfn->end_pfn. Calculate size_pages as the
5194 * number of pages used as kernelcore
5195 */
5196 size_pages = end_pfn - start_pfn;
5197 if (size_pages > kernelcore_remaining)
5198 size_pages = kernelcore_remaining;
5199 zone_movable_pfn[nid] = start_pfn + size_pages;
5200
5201 /*
5202 * Some kernelcore has been met, update counts and
5203 * break if the kernelcore for this node has been
Pintu Kumarb8af2942013-09-11 14:20:34 -07005204 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07005205 */
5206 required_kernelcore -= min(required_kernelcore,
5207 size_pages);
5208 kernelcore_remaining -= size_pages;
5209 if (!kernelcore_remaining)
5210 break;
5211 }
5212 }
5213
5214 /*
5215 * If there is still required_kernelcore, we do another pass with one
5216 * less node in the count. This will push zone_movable_pfn[nid] further
5217 * along on the nodes that still have memory until kernelcore is
Pintu Kumarb8af2942013-09-11 14:20:34 -07005218 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07005219 */
5220 usable_nodes--;
5221 if (usable_nodes && required_kernelcore > usable_nodes)
5222 goto restart;
5223
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005224out2:
Mel Gorman2a1e2742007-07-17 04:03:12 -07005225 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5226 for (nid = 0; nid < MAX_NUMNODES; nid++)
5227 zone_movable_pfn[nid] =
5228 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07005229
Yinghai Lu20e69262013-03-01 14:51:27 -08005230out:
Yinghai Lu66918dc2009-06-30 11:41:37 -07005231 /* restore the node_state */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005232 node_states[N_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005233}
5234
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005235/* Any regular or high memory on that node ? */
5236static void check_for_memory(pg_data_t *pgdat, int nid)
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005237{
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005238 enum zone_type zone_type;
5239
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005240 if (N_MEMORY == N_NORMAL_MEMORY)
5241 return;
5242
5243 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005244 struct zone *zone = &pgdat->node_zones[zone_type];
Xishi Qiub38a8722013-11-12 15:07:20 -08005245 if (populated_zone(zone)) {
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005246 node_set_state(nid, N_HIGH_MEMORY);
5247 if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
5248 zone_type <= ZONE_NORMAL)
5249 node_set_state(nid, N_NORMAL_MEMORY);
Bob Liud0048b02012-01-12 17:19:07 -08005250 break;
5251 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005252 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005253}
5254
Mel Gormanc7132162006-09-27 01:49:43 -07005255/**
5256 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005257 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07005258 *
5259 * This will call free_area_init_node() for each active node in the system.
5260 * Using the page ranges provided by add_active_range(), the size of each
5261 * zone in each node and their holes is calculated. If the maximum PFN
5262 * between two adjacent zones match, it is assumed that the zone is empty.
5263 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5264 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5265 * starts where the previous one ended. For example, ZONE_DMA32 starts
5266 * at arch_max_dma_pfn.
5267 */
5268void __init free_area_init_nodes(unsigned long *max_zone_pfn)
5269{
Tejun Heoc13291a2011-07-12 10:46:30 +02005270 unsigned long start_pfn, end_pfn;
5271 int i, nid;
Mel Gormana6af2bc2007-02-10 01:42:57 -08005272
Mel Gormanc7132162006-09-27 01:49:43 -07005273 /* Record where the zone boundaries are */
5274 memset(arch_zone_lowest_possible_pfn, 0,
5275 sizeof(arch_zone_lowest_possible_pfn));
5276 memset(arch_zone_highest_possible_pfn, 0,
5277 sizeof(arch_zone_highest_possible_pfn));
5278 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
5279 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
5280 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005281 if (i == ZONE_MOVABLE)
5282 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07005283 arch_zone_lowest_possible_pfn[i] =
5284 arch_zone_highest_possible_pfn[i-1];
5285 arch_zone_highest_possible_pfn[i] =
5286 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
5287 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07005288 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
5289 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
5290
5291 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5292 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
Kautuk Consulb224ef82012-03-21 16:34:15 -07005293 find_zone_movable_pfns_for_nodes();
Mel Gormanc7132162006-09-27 01:49:43 -07005294
Mel Gormanc7132162006-09-27 01:49:43 -07005295 /* Print out the zone ranges */
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005296 printk("Zone ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005297 for (i = 0; i < MAX_NR_ZONES; i++) {
5298 if (i == ZONE_MOVABLE)
5299 continue;
Kay Sievers155cbfc2012-05-08 17:24:14 +02005300 printk(KERN_CONT " %-8s ", zone_names[i]);
David Rientjes72f0ba02010-03-05 13:42:14 -08005301 if (arch_zone_lowest_possible_pfn[i] ==
5302 arch_zone_highest_possible_pfn[i])
Kay Sievers155cbfc2012-05-08 17:24:14 +02005303 printk(KERN_CONT "empty\n");
David Rientjes72f0ba02010-03-05 13:42:14 -08005304 else
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005305 printk(KERN_CONT "[mem %0#10lx-%0#10lx]\n",
5306 arch_zone_lowest_possible_pfn[i] << PAGE_SHIFT,
5307 (arch_zone_highest_possible_pfn[i]
5308 << PAGE_SHIFT) - 1);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005309 }
5310
5311 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005312 printk("Movable zone start for each node\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005313 for (i = 0; i < MAX_NUMNODES; i++) {
5314 if (zone_movable_pfn[i])
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005315 printk(" Node %d: %#010lx\n", i,
5316 zone_movable_pfn[i] << PAGE_SHIFT);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005317 }
Mel Gormanc7132162006-09-27 01:49:43 -07005318
Wanpeng Lif2d52fe2012-10-08 16:32:24 -07005319 /* Print out the early node map */
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005320 printk("Early memory node ranges\n");
Tejun Heoc13291a2011-07-12 10:46:30 +02005321 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005322 printk(" node %3d: [mem %#010lx-%#010lx]\n", nid,
5323 start_pfn << PAGE_SHIFT, (end_pfn << PAGE_SHIFT) - 1);
Mel Gormanc7132162006-09-27 01:49:43 -07005324
5325 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07005326 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08005327 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07005328 for_each_online_node(nid) {
5329 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07005330 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07005331 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005332
5333 /* Any memory on that node */
5334 if (pgdat->node_present_pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005335 node_set_state(nid, N_MEMORY);
5336 check_for_memory(pgdat, nid);
Mel Gormanc7132162006-09-27 01:49:43 -07005337 }
5338}
Mel Gorman2a1e2742007-07-17 04:03:12 -07005339
Mel Gorman7e63efe2007-07-17 04:03:15 -07005340static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005341{
5342 unsigned long long coremem;
5343 if (!p)
5344 return -EINVAL;
5345
5346 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005347 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005348
Mel Gorman7e63efe2007-07-17 04:03:15 -07005349 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07005350 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
5351
5352 return 0;
5353}
Mel Gormaned7ed362007-07-17 04:03:14 -07005354
Mel Gorman7e63efe2007-07-17 04:03:15 -07005355/*
5356 * kernelcore=size sets the amount of memory for use for allocations that
5357 * cannot be reclaimed or migrated.
5358 */
5359static int __init cmdline_parse_kernelcore(char *p)
5360{
5361 return cmdline_parse_core(p, &required_kernelcore);
5362}
5363
5364/*
5365 * movablecore=size sets the amount of memory for use for allocations that
5366 * can be reclaimed or migrated.
5367 */
5368static int __init cmdline_parse_movablecore(char *p)
5369{
5370 return cmdline_parse_core(p, &required_movablecore);
5371}
5372
Mel Gormaned7ed362007-07-17 04:03:14 -07005373early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005374early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07005375
Tejun Heo0ee332c2011-12-08 10:22:09 -08005376#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005377
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005378void adjust_managed_page_count(struct page *page, long count)
5379{
5380 spin_lock(&managed_page_count_lock);
5381 page_zone(page)->managed_pages += count;
5382 totalram_pages += count;
Jiang Liu3dcc0572013-07-03 15:03:21 -07005383#ifdef CONFIG_HIGHMEM
5384 if (PageHighMem(page))
5385 totalhigh_pages += count;
5386#endif
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005387 spin_unlock(&managed_page_count_lock);
5388}
Jiang Liu3dcc0572013-07-03 15:03:21 -07005389EXPORT_SYMBOL(adjust_managed_page_count);
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005390
Jiang Liu11199692013-07-03 15:02:48 -07005391unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
Jiang Liu69afade2013-04-29 15:06:21 -07005392{
Jiang Liu11199692013-07-03 15:02:48 -07005393 void *pos;
5394 unsigned long pages = 0;
Jiang Liu69afade2013-04-29 15:06:21 -07005395
Jiang Liu11199692013-07-03 15:02:48 -07005396 start = (void *)PAGE_ALIGN((unsigned long)start);
5397 end = (void *)((unsigned long)end & PAGE_MASK);
5398 for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
Jiang Liudbe67df2013-07-03 15:02:51 -07005399 if ((unsigned int)poison <= 0xFF)
Jiang Liu11199692013-07-03 15:02:48 -07005400 memset(pos, poison, PAGE_SIZE);
5401 free_reserved_page(virt_to_page(pos));
Jiang Liu69afade2013-04-29 15:06:21 -07005402 }
5403
5404 if (pages && s)
Jiang Liu11199692013-07-03 15:02:48 -07005405 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
Jiang Liu69afade2013-04-29 15:06:21 -07005406 s, pages << (PAGE_SHIFT - 10), start, end);
5407
5408 return pages;
5409}
Jiang Liu11199692013-07-03 15:02:48 -07005410EXPORT_SYMBOL(free_reserved_area);
Jiang Liu69afade2013-04-29 15:06:21 -07005411
Jiang Liucfa11e02013-04-29 15:07:00 -07005412#ifdef CONFIG_HIGHMEM
5413void free_highmem_page(struct page *page)
5414{
5415 __free_reserved_page(page);
5416 totalram_pages++;
Jiang Liu7b4b2a02013-07-03 15:03:11 -07005417 page_zone(page)->managed_pages++;
Jiang Liucfa11e02013-04-29 15:07:00 -07005418 totalhigh_pages++;
5419}
5420#endif
5421
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005422
5423void __init mem_init_print_info(const char *str)
5424{
5425 unsigned long physpages, codesize, datasize, rosize, bss_size;
5426 unsigned long init_code_size, init_data_size;
5427
5428 physpages = get_num_physpages();
5429 codesize = _etext - _stext;
5430 datasize = _edata - _sdata;
5431 rosize = __end_rodata - __start_rodata;
5432 bss_size = __bss_stop - __bss_start;
5433 init_data_size = __init_end - __init_begin;
5434 init_code_size = _einittext - _sinittext;
5435
5436 /*
5437 * Detect special cases and adjust section sizes accordingly:
5438 * 1) .init.* may be embedded into .data sections
5439 * 2) .init.text.* may be out of [__init_begin, __init_end],
5440 * please refer to arch/tile/kernel/vmlinux.lds.S.
5441 * 3) .rodata.* may be embedded into .text or .data sections.
5442 */
5443#define adj_init_size(start, end, size, pos, adj) \
Pintu Kumarb8af2942013-09-11 14:20:34 -07005444 do { \
5445 if (start <= pos && pos < end && size > adj) \
5446 size -= adj; \
5447 } while (0)
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005448
5449 adj_init_size(__init_begin, __init_end, init_data_size,
5450 _sinittext, init_code_size);
5451 adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
5452 adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
5453 adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
5454 adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
5455
5456#undef adj_init_size
5457
5458 printk("Memory: %luK/%luK available "
5459 "(%luK kernel code, %luK rwdata, %luK rodata, "
5460 "%luK init, %luK bss, %luK reserved"
5461#ifdef CONFIG_HIGHMEM
5462 ", %luK highmem"
5463#endif
5464 "%s%s)\n",
5465 nr_free_pages() << (PAGE_SHIFT-10), physpages << (PAGE_SHIFT-10),
5466 codesize >> 10, datasize >> 10, rosize >> 10,
5467 (init_data_size + init_code_size) >> 10, bss_size >> 10,
5468 (physpages - totalram_pages) << (PAGE_SHIFT-10),
5469#ifdef CONFIG_HIGHMEM
5470 totalhigh_pages << (PAGE_SHIFT-10),
5471#endif
5472 str ? ", " : "", str ? str : "");
5473}
5474
Mel Gorman0e0b8642006-09-27 01:49:56 -07005475/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005476 * set_dma_reserve - set the specified number of pages reserved in the first zone
5477 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07005478 *
5479 * The per-cpu batchsize and zone watermarks are determined by present_pages.
5480 * In the DMA zone, a significant percentage may be consumed by kernel image
5481 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005482 * function may optionally be used to account for unfreeable pages in the
5483 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5484 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07005485 */
5486void __init set_dma_reserve(unsigned long new_dma_reserve)
5487{
5488 dma_reserve = new_dma_reserve;
5489}
5490
Linus Torvalds1da177e2005-04-16 15:20:36 -07005491void __init free_area_init(unsigned long *zones_size)
5492{
Johannes Weiner9109fb72008-07-23 21:27:20 -07005493 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005494 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
5495}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005496
Linus Torvalds1da177e2005-04-16 15:20:36 -07005497static int page_alloc_cpu_notify(struct notifier_block *self,
5498 unsigned long action, void *hcpu)
5499{
5500 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005501
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07005502 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -07005503 lru_add_drain_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005504 drain_pages(cpu);
5505
5506 /*
5507 * Spill the event counters of the dead processor
5508 * into the current processors event counters.
5509 * This artificially elevates the count of the current
5510 * processor.
5511 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07005512 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005513
5514 /*
5515 * Zero the differential counters of the dead processor
5516 * so that the vm statistics are consistent.
5517 *
5518 * This is only okay since the processor is dead and cannot
5519 * race with what we are doing.
5520 */
Christoph Lameter2bb921e2013-09-11 14:21:30 -07005521 cpu_vm_stats_fold(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005522 }
5523 return NOTIFY_OK;
5524}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005525
5526void __init page_alloc_init(void)
5527{
5528 hotcpu_notifier(page_alloc_cpu_notify, 0);
5529}
5530
5531/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005532 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5533 * or min_free_kbytes changes.
5534 */
5535static void calculate_totalreserve_pages(void)
5536{
5537 struct pglist_data *pgdat;
5538 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005539 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005540
5541 for_each_online_pgdat(pgdat) {
5542 for (i = 0; i < MAX_NR_ZONES; i++) {
5543 struct zone *zone = pgdat->node_zones + i;
5544 unsigned long max = 0;
5545
5546 /* Find valid and maximum lowmem_reserve in the zone */
5547 for (j = i; j < MAX_NR_ZONES; j++) {
5548 if (zone->lowmem_reserve[j] > max)
5549 max = zone->lowmem_reserve[j];
5550 }
5551
Mel Gorman41858962009-06-16 15:32:12 -07005552 /* we treat the high watermark as reserved pages. */
5553 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005554
Jiang Liub40da042013-02-22 16:33:52 -08005555 if (max > zone->managed_pages)
5556 max = zone->managed_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005557 reserve_pages += max;
Johannes Weinerab8fabd2012-01-10 15:07:42 -08005558 /*
5559 * Lowmem reserves are not available to
5560 * GFP_HIGHUSER page cache allocations and
5561 * kswapd tries to balance zones to their high
5562 * watermark. As a result, neither should be
5563 * regarded as dirtyable memory, to prevent a
5564 * situation where reclaim has to clean pages
5565 * in order to balance the zones.
5566 */
5567 zone->dirty_balance_reserve = max;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005568 }
5569 }
Johannes Weinerab8fabd2012-01-10 15:07:42 -08005570 dirty_balance_reserve = reserve_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005571 totalreserve_pages = reserve_pages;
5572}
5573
5574/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07005575 * setup_per_zone_lowmem_reserve - called whenever
5576 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
5577 * has a correct pages reserved value, so an adequate number of
5578 * pages are left in the zone after a successful __alloc_pages().
5579 */
5580static void setup_per_zone_lowmem_reserve(void)
5581{
5582 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005583 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005584
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08005585 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005586 for (j = 0; j < MAX_NR_ZONES; j++) {
5587 struct zone *zone = pgdat->node_zones + j;
Jiang Liub40da042013-02-22 16:33:52 -08005588 unsigned long managed_pages = zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005589
5590 zone->lowmem_reserve[j] = 0;
5591
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005592 idx = j;
5593 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005594 struct zone *lower_zone;
5595
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005596 idx--;
5597
Linus Torvalds1da177e2005-04-16 15:20:36 -07005598 if (sysctl_lowmem_reserve_ratio[idx] < 1)
5599 sysctl_lowmem_reserve_ratio[idx] = 1;
5600
5601 lower_zone = pgdat->node_zones + idx;
Jiang Liub40da042013-02-22 16:33:52 -08005602 lower_zone->lowmem_reserve[j] = managed_pages /
Linus Torvalds1da177e2005-04-16 15:20:36 -07005603 sysctl_lowmem_reserve_ratio[idx];
Jiang Liub40da042013-02-22 16:33:52 -08005604 managed_pages += lower_zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005605 }
5606 }
5607 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005608
5609 /* update totalreserve_pages */
5610 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005611}
5612
Mel Gormancfd3da12011-04-25 21:36:42 +00005613static void __setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005614{
5615 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
5616 unsigned long lowmem_pages = 0;
5617 struct zone *zone;
5618 unsigned long flags;
5619
5620 /* Calculate total number of !ZONE_HIGHMEM pages */
5621 for_each_zone(zone) {
5622 if (!is_highmem(zone))
Jiang Liub40da042013-02-22 16:33:52 -08005623 lowmem_pages += zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005624 }
5625
5626 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07005627 u64 tmp;
5628
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005629 spin_lock_irqsave(&zone->lock, flags);
Jiang Liub40da042013-02-22 16:33:52 -08005630 tmp = (u64)pages_min * zone->managed_pages;
Andrew Mortonac924c62006-05-15 09:43:59 -07005631 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005632 if (is_highmem(zone)) {
5633 /*
Nick Piggin669ed172005-11-13 16:06:45 -08005634 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5635 * need highmem pages, so cap pages_min to a small
5636 * value here.
5637 *
Mel Gorman41858962009-06-16 15:32:12 -07005638 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Nick Piggin669ed172005-11-13 16:06:45 -08005639 * deltas controls asynch page reclaim, and so should
5640 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005641 */
Andrew Morton90ae8d62013-02-22 16:32:22 -08005642 unsigned long min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005643
Jiang Liub40da042013-02-22 16:33:52 -08005644 min_pages = zone->managed_pages / 1024;
Andrew Morton90ae8d62013-02-22 16:32:22 -08005645 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
Mel Gorman41858962009-06-16 15:32:12 -07005646 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005647 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08005648 /*
5649 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07005650 * proportionate to the zone's size.
5651 */
Mel Gorman41858962009-06-16 15:32:12 -07005652 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005653 }
5654
Mel Gorman41858962009-06-16 15:32:12 -07005655 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
5656 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Marek Szyprowski49f223a2012-01-25 12:49:24 +01005657
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07005658 __mod_zone_page_state(zone, NR_ALLOC_BATCH,
5659 high_wmark_pages(zone) -
5660 low_wmark_pages(zone) -
5661 zone_page_state(zone, NR_ALLOC_BATCH));
5662
Mel Gorman56fd56b2007-10-16 01:25:58 -07005663 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005664 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005665 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005666
5667 /* update totalreserve_pages */
5668 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005669}
5670
Mel Gormancfd3da12011-04-25 21:36:42 +00005671/**
5672 * setup_per_zone_wmarks - called when min_free_kbytes changes
5673 * or when memory is hot-{added|removed}
5674 *
5675 * Ensures that the watermark[min,low,high] values for each zone are set
5676 * correctly with respect to min_free_kbytes.
5677 */
5678void setup_per_zone_wmarks(void)
5679{
5680 mutex_lock(&zonelists_mutex);
5681 __setup_per_zone_wmarks();
5682 mutex_unlock(&zonelists_mutex);
5683}
5684
Randy Dunlap55a44622009-09-21 17:01:20 -07005685/*
Rik van Riel556adec2008-10-18 20:26:34 -07005686 * The inactive anon list should be small enough that the VM never has to
5687 * do too much work, but large enough that each inactive page has a chance
5688 * to be referenced again before it is swapped out.
5689 *
5690 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5691 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5692 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5693 * the anonymous pages are kept on the inactive list.
5694 *
5695 * total target max
5696 * memory ratio inactive anon
5697 * -------------------------------------
5698 * 10MB 1 5MB
5699 * 100MB 1 50MB
5700 * 1GB 3 250MB
5701 * 10GB 10 0.9GB
5702 * 100GB 31 3GB
5703 * 1TB 101 10GB
5704 * 10TB 320 32GB
5705 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005706static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
Minchan Kim96cb4df2009-06-16 15:32:49 -07005707{
5708 unsigned int gb, ratio;
5709
5710 /* Zone size in gigabytes */
Jiang Liub40da042013-02-22 16:33:52 -08005711 gb = zone->managed_pages >> (30 - PAGE_SHIFT);
Minchan Kim96cb4df2009-06-16 15:32:49 -07005712 if (gb)
5713 ratio = int_sqrt(10 * gb);
5714 else
5715 ratio = 1;
5716
5717 zone->inactive_ratio = ratio;
5718}
5719
KOSAKI Motohiro839a4fc2011-05-24 17:11:31 -07005720static void __meminit setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07005721{
5722 struct zone *zone;
5723
Minchan Kim96cb4df2009-06-16 15:32:49 -07005724 for_each_zone(zone)
5725 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07005726}
5727
Linus Torvalds1da177e2005-04-16 15:20:36 -07005728/*
5729 * Initialise min_free_kbytes.
5730 *
5731 * For small machines we want it small (128k min). For large machines
5732 * we want it large (64MB max). But it is not linear, because network
5733 * bandwidth does not increase linearly with machine size. We use
5734 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07005735 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
Linus Torvalds1da177e2005-04-16 15:20:36 -07005736 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5737 *
5738 * which yields
5739 *
5740 * 16MB: 512k
5741 * 32MB: 724k
5742 * 64MB: 1024k
5743 * 128MB: 1448k
5744 * 256MB: 2048k
5745 * 512MB: 2896k
5746 * 1024MB: 4096k
5747 * 2048MB: 5792k
5748 * 4096MB: 8192k
5749 * 8192MB: 11584k
5750 * 16384MB: 16384k
5751 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005752int __meminit init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005753{
5754 unsigned long lowmem_kbytes;
Michal Hocko5f127332013-07-08 16:00:40 -07005755 int new_min_free_kbytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005756
5757 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
Michal Hocko5f127332013-07-08 16:00:40 -07005758 new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005759
Michal Hocko5f127332013-07-08 16:00:40 -07005760 if (new_min_free_kbytes > user_min_free_kbytes) {
5761 min_free_kbytes = new_min_free_kbytes;
5762 if (min_free_kbytes < 128)
5763 min_free_kbytes = 128;
5764 if (min_free_kbytes > 65536)
5765 min_free_kbytes = 65536;
5766 } else {
5767 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
5768 new_min_free_kbytes, user_min_free_kbytes);
5769 }
Minchan Kimbc75d332009-06-16 15:32:48 -07005770 setup_per_zone_wmarks();
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -07005771 refresh_zone_stat_thresholds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005772 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07005773 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005774 return 0;
5775}
Minchan Kimbc75d332009-06-16 15:32:48 -07005776module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005777
5778/*
Pintu Kumarb8af2942013-09-11 14:20:34 -07005779 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
Linus Torvalds1da177e2005-04-16 15:20:36 -07005780 * that we can call two helper functions whenever min_free_kbytes
5781 * changes.
5782 */
Pintu Kumarb8af2942013-09-11 14:20:34 -07005783int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005784 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005785{
Han Pingtianda8c7572014-01-23 15:53:17 -08005786 int rc;
5787
5788 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5789 if (rc)
5790 return rc;
5791
Michal Hocko5f127332013-07-08 16:00:40 -07005792 if (write) {
5793 user_min_free_kbytes = min_free_kbytes;
Minchan Kimbc75d332009-06-16 15:32:48 -07005794 setup_per_zone_wmarks();
Michal Hocko5f127332013-07-08 16:00:40 -07005795 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005796 return 0;
5797}
5798
Christoph Lameter96146342006-07-03 00:24:13 -07005799#ifdef CONFIG_NUMA
5800int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005801 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07005802{
5803 struct zone *zone;
5804 int rc;
5805
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005806 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07005807 if (rc)
5808 return rc;
5809
5810 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08005811 zone->min_unmapped_pages = (zone->managed_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07005812 sysctl_min_unmapped_ratio) / 100;
5813 return 0;
5814}
Christoph Lameter0ff38492006-09-25 23:31:52 -07005815
5816int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005817 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07005818{
5819 struct zone *zone;
5820 int rc;
5821
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005822 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07005823 if (rc)
5824 return rc;
5825
5826 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08005827 zone->min_slab_pages = (zone->managed_pages *
Christoph Lameter0ff38492006-09-25 23:31:52 -07005828 sysctl_min_slab_ratio) / 100;
5829 return 0;
5830}
Christoph Lameter96146342006-07-03 00:24:13 -07005831#endif
5832
Linus Torvalds1da177e2005-04-16 15:20:36 -07005833/*
5834 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5835 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5836 * whenever sysctl_lowmem_reserve_ratio changes.
5837 *
5838 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07005839 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07005840 * if in function of the boot time zone sizes.
5841 */
5842int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005843 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005844{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005845 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005846 setup_per_zone_lowmem_reserve();
5847 return 0;
5848}
5849
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005850/*
5851 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
Pintu Kumarb8af2942013-09-11 14:20:34 -07005852 * cpu. It is the fraction of total pages in each zone that a hot per cpu
5853 * pagelist can have before it gets flushed back to buddy allocator.
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005854 */
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005855int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005856 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005857{
5858 struct zone *zone;
5859 unsigned int cpu;
5860 int ret;
5861
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005862 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
Sasha Levin93278812012-05-10 13:01:44 -07005863 if (!write || (ret < 0))
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005864 return ret;
Cody P Schaferc8e251f2013-07-03 15:01:29 -07005865
5866 mutex_lock(&pcp_batch_high_lock);
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07005867 for_each_populated_zone(zone) {
Cody P Schafer22a7f122013-07-03 15:01:34 -07005868 unsigned long high;
5869 high = zone->managed_pages / percpu_pagelist_fraction;
5870 for_each_possible_cpu(cpu)
Cody P Schafer36640332013-07-03 15:01:40 -07005871 pageset_set_high(per_cpu_ptr(zone->pageset, cpu),
5872 high);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005873 }
Cody P Schaferc8e251f2013-07-03 15:01:29 -07005874 mutex_unlock(&pcp_batch_high_lock);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005875 return 0;
5876}
5877
David S. Millerf034b5d2006-08-24 03:08:07 -07005878int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005879
5880#ifdef CONFIG_NUMA
5881static int __init set_hashdist(char *str)
5882{
5883 if (!str)
5884 return 0;
5885 hashdist = simple_strtoul(str, &str, 0);
5886 return 1;
5887}
5888__setup("hashdist=", set_hashdist);
5889#endif
5890
5891/*
5892 * allocate a large system hash table from bootmem
5893 * - it is assumed that the hash table must contain an exact power-of-2
5894 * quantity of entries
5895 * - limit is the number of hash buckets, not the total allocation size
5896 */
5897void *__init alloc_large_system_hash(const char *tablename,
5898 unsigned long bucketsize,
5899 unsigned long numentries,
5900 int scale,
5901 int flags,
5902 unsigned int *_hash_shift,
5903 unsigned int *_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00005904 unsigned long low_limit,
5905 unsigned long high_limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005906{
Tim Bird31fe62b2012-05-23 13:33:35 +00005907 unsigned long long max = high_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005908 unsigned long log2qty, size;
5909 void *table = NULL;
5910
5911 /* allow the kernel cmdline to have a say */
5912 if (!numentries) {
5913 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08005914 numentries = nr_kernel_pages;
Jerry Zhoua7e83312013-09-11 14:20:26 -07005915
5916 /* It isn't necessary when PAGE_SIZE >= 1MB */
5917 if (PAGE_SHIFT < 20)
5918 numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005919
5920 /* limit to 1 bucket per 2^scale bytes of low memory */
5921 if (scale > PAGE_SHIFT)
5922 numentries >>= (scale - PAGE_SHIFT);
5923 else
5924 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08005925
5926 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07005927 if (unlikely(flags & HASH_SMALL)) {
5928 /* Makes no sense without HASH_EARLY */
5929 WARN_ON(!(flags & HASH_EARLY));
5930 if (!(numentries >> *_hash_shift)) {
5931 numentries = 1UL << *_hash_shift;
5932 BUG_ON(!numentries);
5933 }
5934 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08005935 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005936 }
John Hawkes6e692ed2006-03-25 03:08:02 -08005937 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005938
5939 /* limit allocation size to 1/16 total memory by default */
5940 if (max == 0) {
5941 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
5942 do_div(max, bucketsize);
5943 }
Dimitri Sivanich074b8512012-02-08 12:39:07 -08005944 max = min(max, 0x80000000ULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005945
Tim Bird31fe62b2012-05-23 13:33:35 +00005946 if (numentries < low_limit)
5947 numentries = low_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005948 if (numentries > max)
5949 numentries = max;
5950
David Howellsf0d1b0b2006-12-08 02:37:49 -08005951 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005952
5953 do {
5954 size = bucketsize << log2qty;
5955 if (flags & HASH_EARLY)
Santosh Shilimkar67828322014-01-21 15:50:25 -08005956 table = memblock_virt_alloc_nopanic(size, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005957 else if (hashdist)
5958 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
5959 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07005960 /*
5961 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07005962 * some pages at the end of hash table which
5963 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07005964 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005965 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07005966 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005967 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
5968 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005969 }
5970 } while (!table && size > PAGE_SIZE && --log2qty);
5971
5972 if (!table)
5973 panic("Failed to allocate %s hash table\n", tablename);
5974
Robin Holtf241e6602010-10-07 12:59:26 -07005975 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07005976 tablename,
Robin Holtf241e6602010-10-07 12:59:26 -07005977 (1UL << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08005978 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005979 size);
5980
5981 if (_hash_shift)
5982 *_hash_shift = log2qty;
5983 if (_hash_mask)
5984 *_hash_mask = (1 << log2qty) - 1;
5985
5986 return table;
5987}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08005988
Mel Gorman835c1342007-10-16 01:25:47 -07005989/* Return a pointer to the bitmap storing bits affecting a block of pages */
5990static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
5991 unsigned long pfn)
5992{
5993#ifdef CONFIG_SPARSEMEM
5994 return __pfn_to_section(pfn)->pageblock_flags;
5995#else
5996 return zone->pageblock_flags;
5997#endif /* CONFIG_SPARSEMEM */
5998}
Andrew Morton6220ec72006-10-19 23:29:05 -07005999
Mel Gorman835c1342007-10-16 01:25:47 -07006000static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
6001{
6002#ifdef CONFIG_SPARSEMEM
6003 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07006004 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07006005#else
Laura Abbottc060f942013-01-11 14:31:51 -08006006 pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
Mel Gormand9c23402007-10-16 01:26:01 -07006007 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07006008#endif /* CONFIG_SPARSEMEM */
6009}
6010
6011/**
Mel Gormand9c23402007-10-16 01:26:01 -07006012 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
Mel Gorman835c1342007-10-16 01:25:47 -07006013 * @page: The page within the block of interest
6014 * @start_bitidx: The first bit of interest to retrieve
6015 * @end_bitidx: The last bit of interest
6016 * returns pageblock_bits flags
6017 */
6018unsigned long get_pageblock_flags_group(struct page *page,
6019 int start_bitidx, int end_bitidx)
6020{
6021 struct zone *zone;
6022 unsigned long *bitmap;
6023 unsigned long pfn, bitidx;
6024 unsigned long flags = 0;
6025 unsigned long value = 1;
6026
6027 zone = page_zone(page);
6028 pfn = page_to_pfn(page);
6029 bitmap = get_pageblock_bitmap(zone, pfn);
6030 bitidx = pfn_to_bitidx(zone, pfn);
6031
6032 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
6033 if (test_bit(bitidx + start_bitidx, bitmap))
6034 flags |= value;
6035
6036 return flags;
6037}
6038
6039/**
Mel Gormand9c23402007-10-16 01:26:01 -07006040 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
Mel Gorman835c1342007-10-16 01:25:47 -07006041 * @page: The page within the block of interest
6042 * @start_bitidx: The first bit of interest
6043 * @end_bitidx: The last bit of interest
6044 * @flags: The flags to set
6045 */
6046void set_pageblock_flags_group(struct page *page, unsigned long flags,
6047 int start_bitidx, int end_bitidx)
6048{
6049 struct zone *zone;
6050 unsigned long *bitmap;
6051 unsigned long pfn, bitidx;
6052 unsigned long value = 1;
6053
6054 zone = page_zone(page);
6055 pfn = page_to_pfn(page);
6056 bitmap = get_pageblock_bitmap(zone, pfn);
6057 bitidx = pfn_to_bitidx(zone, pfn);
Sasha Levin309381fea2014-01-23 15:52:54 -08006058 VM_BUG_ON_PAGE(!zone_spans_pfn(zone, pfn), page);
Mel Gorman835c1342007-10-16 01:25:47 -07006059
6060 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
6061 if (flags & value)
6062 __set_bit(bitidx + start_bitidx, bitmap);
6063 else
6064 __clear_bit(bitidx + start_bitidx, bitmap);
6065}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07006066
6067/*
Minchan Kim80934512012-07-31 16:43:01 -07006068 * This function checks whether pageblock includes unmovable pages or not.
6069 * If @count is not zero, it is okay to include less @count unmovable pages
6070 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07006071 * PageLRU check without isolation or lru_lock could race so that
Minchan Kim80934512012-07-31 16:43:01 -07006072 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6073 * expect this function should be exact.
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07006074 */
Wen Congyangb023f462012-12-11 16:00:45 -08006075bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
6076 bool skip_hwpoisoned_pages)
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006077{
6078 unsigned long pfn, iter, found;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01006079 int mt;
6080
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006081 /*
6082 * For avoiding noise data, lru_add_drain_all() should be called
Minchan Kim80934512012-07-31 16:43:01 -07006083 * If ZONE_MOVABLE, the zone never contains unmovable pages
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006084 */
6085 if (zone_idx(zone) == ZONE_MOVABLE)
Minchan Kim80934512012-07-31 16:43:01 -07006086 return false;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01006087 mt = get_pageblock_migratetype(page);
6088 if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
Minchan Kim80934512012-07-31 16:43:01 -07006089 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006090
6091 pfn = page_to_pfn(page);
6092 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
6093 unsigned long check = pfn + iter;
6094
Namhyung Kim29723fc2011-02-25 14:44:25 -08006095 if (!pfn_valid_within(check))
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006096 continue;
Namhyung Kim29723fc2011-02-25 14:44:25 -08006097
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006098 page = pfn_to_page(check);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07006099
6100 /*
6101 * Hugepages are not in LRU lists, but they're movable.
6102 * We need not scan over tail pages bacause we don't
6103 * handle each tail page individually in migration.
6104 */
6105 if (PageHuge(page)) {
6106 iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
6107 continue;
6108 }
6109
Minchan Kim97d255c2012-07-31 16:42:59 -07006110 /*
6111 * We can't use page_count without pin a page
6112 * because another CPU can free compound page.
6113 * This check already skips compound tails of THP
6114 * because their page->_count is zero at all time.
6115 */
6116 if (!atomic_read(&page->_count)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006117 if (PageBuddy(page))
6118 iter += (1 << page_order(page)) - 1;
6119 continue;
6120 }
Minchan Kim97d255c2012-07-31 16:42:59 -07006121
Wen Congyangb023f462012-12-11 16:00:45 -08006122 /*
6123 * The HWPoisoned page may be not in buddy system, and
6124 * page_count() is not 0.
6125 */
6126 if (skip_hwpoisoned_pages && PageHWPoison(page))
6127 continue;
6128
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006129 if (!PageLRU(page))
6130 found++;
6131 /*
6132 * If there are RECLAIMABLE pages, we need to check it.
6133 * But now, memory offline itself doesn't call shrink_slab()
6134 * and it still to be fixed.
6135 */
6136 /*
6137 * If the page is not RAM, page_count()should be 0.
6138 * we don't need more check. This is an _used_ not-movable page.
6139 *
6140 * The problematic thing here is PG_reserved pages. PG_reserved
6141 * is set to both of a memory hole page and a _used_ kernel
6142 * page at boot.
6143 */
6144 if (found > count)
Minchan Kim80934512012-07-31 16:43:01 -07006145 return true;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006146 }
Minchan Kim80934512012-07-31 16:43:01 -07006147 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006148}
6149
6150bool is_pageblock_removable_nolock(struct page *page)
6151{
Michal Hocko656a0702012-01-20 14:33:58 -08006152 struct zone *zone;
6153 unsigned long pfn;
Michal Hocko687875f2012-01-20 14:33:55 -08006154
6155 /*
6156 * We have to be careful here because we are iterating over memory
6157 * sections which are not zone aware so we might end up outside of
6158 * the zone but still within the section.
Michal Hocko656a0702012-01-20 14:33:58 -08006159 * We have to take care about the node as well. If the node is offline
6160 * its NODE_DATA will be NULL - see page_zone.
Michal Hocko687875f2012-01-20 14:33:55 -08006161 */
Michal Hocko656a0702012-01-20 14:33:58 -08006162 if (!node_online(page_to_nid(page)))
6163 return false;
6164
6165 zone = page_zone(page);
6166 pfn = page_to_pfn(page);
Cody P Schafer108bcc92013-02-22 16:35:23 -08006167 if (!zone_spans_pfn(zone, pfn))
Michal Hocko687875f2012-01-20 14:33:55 -08006168 return false;
6169
Wen Congyangb023f462012-12-11 16:00:45 -08006170 return !has_unmovable_pages(zone, page, 0, true);
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07006171}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006172
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006173#ifdef CONFIG_CMA
6174
6175static unsigned long pfn_max_align_down(unsigned long pfn)
6176{
6177 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
6178 pageblock_nr_pages) - 1);
6179}
6180
6181static unsigned long pfn_max_align_up(unsigned long pfn)
6182{
6183 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
6184 pageblock_nr_pages));
6185}
6186
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006187/* [start, end) must belong to a single zone. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006188static int __alloc_contig_migrate_range(struct compact_control *cc,
6189 unsigned long start, unsigned long end)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006190{
6191 /* This function is based on compact_zone() from compaction.c. */
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006192 unsigned long nr_reclaimed;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006193 unsigned long pfn = start;
6194 unsigned int tries = 0;
6195 int ret = 0;
6196
Marek Szyprowskibe49a6e2012-12-12 13:51:19 -08006197 migrate_prep();
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006198
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006199 while (pfn < end || !list_empty(&cc->migratepages)) {
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006200 if (fatal_signal_pending(current)) {
6201 ret = -EINTR;
6202 break;
6203 }
6204
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006205 if (list_empty(&cc->migratepages)) {
6206 cc->nr_migratepages = 0;
6207 pfn = isolate_migratepages_range(cc->zone, cc,
Minchan Kime46a2872012-10-08 16:33:48 -07006208 pfn, end, true);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006209 if (!pfn) {
6210 ret = -EINTR;
6211 break;
6212 }
6213 tries = 0;
6214 } else if (++tries == 5) {
6215 ret = ret < 0 ? ret : -EBUSY;
6216 break;
6217 }
6218
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006219 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
6220 &cc->migratepages);
6221 cc->nr_migratepages -= nr_reclaimed;
Minchan Kim02c6de82012-10-08 16:31:55 -07006222
Hugh Dickins9c620e22013-02-22 16:35:14 -08006223 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
David Rientjese0b9dae2014-06-04 16:08:28 -07006224 NULL, 0, cc->mode, MR_CMA);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006225 }
Srinivas Pandruvada2a6f5122013-02-22 16:32:09 -08006226 if (ret < 0) {
6227 putback_movable_pages(&cc->migratepages);
6228 return ret;
6229 }
6230 return 0;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006231}
6232
6233/**
6234 * alloc_contig_range() -- tries to allocate given range of pages
6235 * @start: start PFN to allocate
6236 * @end: one-past-the-last PFN to allocate
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006237 * @migratetype: migratetype of the underlaying pageblocks (either
6238 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6239 * in range must have the same migratetype and it must
6240 * be either of the two.
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006241 *
6242 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6243 * aligned, however it's the caller's responsibility to guarantee that
6244 * we are the only thread that changes migrate type of pageblocks the
6245 * pages fall in.
6246 *
6247 * The PFN range must belong to a single zone.
6248 *
6249 * Returns zero on success or negative error code. On success all
6250 * pages which PFN is in [start, end) are allocated for the caller and
6251 * need to be freed with free_contig_range().
6252 */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006253int alloc_contig_range(unsigned long start, unsigned long end,
6254 unsigned migratetype)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006255{
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006256 unsigned long outer_start, outer_end;
6257 int ret = 0, order;
6258
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006259 struct compact_control cc = {
6260 .nr_migratepages = 0,
6261 .order = -1,
6262 .zone = page_zone(pfn_to_page(start)),
David Rientjese0b9dae2014-06-04 16:08:28 -07006263 .mode = MIGRATE_SYNC,
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006264 .ignore_skip_hint = true,
6265 };
6266 INIT_LIST_HEAD(&cc.migratepages);
6267
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006268 /*
6269 * What we do here is we mark all pageblocks in range as
6270 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6271 * have different sizes, and due to the way page allocator
6272 * work, we align the range to biggest of the two pages so
6273 * that page allocator won't try to merge buddies from
6274 * different pageblocks and change MIGRATE_ISOLATE to some
6275 * other migration type.
6276 *
6277 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6278 * migrate the pages from an unaligned range (ie. pages that
6279 * we are interested in). This will put all the pages in
6280 * range back to page allocator as MIGRATE_ISOLATE.
6281 *
6282 * When this is done, we take the pages in range from page
6283 * allocator removing them from the buddy system. This way
6284 * page allocator will never consider using them.
6285 *
6286 * This lets us mark the pageblocks back as
6287 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6288 * aligned range but not in the unaligned, original range are
6289 * put back to page allocator so that buddy can use them.
6290 */
6291
6292 ret = start_isolate_page_range(pfn_max_align_down(start),
Wen Congyangb023f462012-12-11 16:00:45 -08006293 pfn_max_align_up(end), migratetype,
6294 false);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006295 if (ret)
Bob Liu86a595f2012-10-25 13:37:56 -07006296 return ret;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006297
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006298 ret = __alloc_contig_migrate_range(&cc, start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006299 if (ret)
6300 goto done;
6301
6302 /*
6303 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6304 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
6305 * more, all pages in [start, end) are free in page allocator.
6306 * What we are going to do is to allocate all pages from
6307 * [start, end) (that is remove them from page allocator).
6308 *
6309 * The only problem is that pages at the beginning and at the
6310 * end of interesting range may be not aligned with pages that
6311 * page allocator holds, ie. they can be part of higher order
6312 * pages. Because of this, we reserve the bigger range and
6313 * once this is done free the pages we are not interested in.
6314 *
6315 * We don't have to hold zone->lock here because the pages are
6316 * isolated thus they won't get removed from buddy.
6317 */
6318
6319 lru_add_drain_all();
6320 drain_all_pages();
6321
6322 order = 0;
6323 outer_start = start;
6324 while (!PageBuddy(pfn_to_page(outer_start))) {
6325 if (++order >= MAX_ORDER) {
6326 ret = -EBUSY;
6327 goto done;
6328 }
6329 outer_start &= ~0UL << order;
6330 }
6331
6332 /* Make sure the range is really isolated. */
Wen Congyangb023f462012-12-11 16:00:45 -08006333 if (test_pages_isolated(outer_start, end, false)) {
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006334 pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n",
6335 outer_start, end);
6336 ret = -EBUSY;
6337 goto done;
6338 }
6339
Marek Szyprowski49f223a2012-01-25 12:49:24 +01006340
6341 /* Grab isolated pages from freelists. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006342 outer_end = isolate_freepages_range(&cc, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006343 if (!outer_end) {
6344 ret = -EBUSY;
6345 goto done;
6346 }
6347
6348 /* Free head and tail (if any) */
6349 if (start != outer_start)
6350 free_contig_range(outer_start, start - outer_start);
6351 if (end != outer_end)
6352 free_contig_range(end, outer_end - end);
6353
6354done:
6355 undo_isolate_page_range(pfn_max_align_down(start),
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006356 pfn_max_align_up(end), migratetype);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006357 return ret;
6358}
6359
6360void free_contig_range(unsigned long pfn, unsigned nr_pages)
6361{
Marek Szyprowskibcc2b022012-12-20 15:05:18 -08006362 unsigned int count = 0;
6363
6364 for (; nr_pages--; pfn++) {
6365 struct page *page = pfn_to_page(pfn);
6366
6367 count += page_count(page) != 1;
6368 __free_page(page);
6369 }
6370 WARN(count != 0, "%d pages are still in use!\n", count);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006371}
6372#endif
6373
Jiang Liu4ed7e022012-07-31 16:43:35 -07006374#ifdef CONFIG_MEMORY_HOTPLUG
Cody P Schafer0a647f32013-07-03 15:01:33 -07006375/*
6376 * The zone indicated has a new number of managed_pages; batch sizes and percpu
6377 * page high values need to be recalulated.
6378 */
Jiang Liu4ed7e022012-07-31 16:43:35 -07006379void __meminit zone_pcp_update(struct zone *zone)
6380{
Cody P Schafer0a647f32013-07-03 15:01:33 -07006381 unsigned cpu;
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006382 mutex_lock(&pcp_batch_high_lock);
Cody P Schafer0a647f32013-07-03 15:01:33 -07006383 for_each_possible_cpu(cpu)
Cody P Schafer169f6c12013-07-03 15:01:41 -07006384 pageset_set_high_and_batch(zone,
6385 per_cpu_ptr(zone->pageset, cpu));
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006386 mutex_unlock(&pcp_batch_high_lock);
Jiang Liu4ed7e022012-07-31 16:43:35 -07006387}
6388#endif
6389
Jiang Liu340175b2012-07-31 16:43:32 -07006390void zone_pcp_reset(struct zone *zone)
6391{
6392 unsigned long flags;
Minchan Kim5a883812012-10-08 16:33:39 -07006393 int cpu;
6394 struct per_cpu_pageset *pset;
Jiang Liu340175b2012-07-31 16:43:32 -07006395
6396 /* avoid races with drain_pages() */
6397 local_irq_save(flags);
6398 if (zone->pageset != &boot_pageset) {
Minchan Kim5a883812012-10-08 16:33:39 -07006399 for_each_online_cpu(cpu) {
6400 pset = per_cpu_ptr(zone->pageset, cpu);
6401 drain_zonestat(zone, pset);
6402 }
Jiang Liu340175b2012-07-31 16:43:32 -07006403 free_percpu(zone->pageset);
6404 zone->pageset = &boot_pageset;
6405 }
6406 local_irq_restore(flags);
6407}
6408
Wen Congyang6dcd73d2012-12-11 16:01:01 -08006409#ifdef CONFIG_MEMORY_HOTREMOVE
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006410/*
6411 * All pages in the range must be isolated before calling this.
6412 */
6413void
6414__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
6415{
6416 struct page *page;
6417 struct zone *zone;
6418 int order, i;
6419 unsigned long pfn;
6420 unsigned long flags;
6421 /* find the first valid pfn */
6422 for (pfn = start_pfn; pfn < end_pfn; pfn++)
6423 if (pfn_valid(pfn))
6424 break;
6425 if (pfn == end_pfn)
6426 return;
6427 zone = page_zone(pfn_to_page(pfn));
6428 spin_lock_irqsave(&zone->lock, flags);
6429 pfn = start_pfn;
6430 while (pfn < end_pfn) {
6431 if (!pfn_valid(pfn)) {
6432 pfn++;
6433 continue;
6434 }
6435 page = pfn_to_page(pfn);
Wen Congyangb023f462012-12-11 16:00:45 -08006436 /*
6437 * The HWPoisoned page may be not in buddy system, and
6438 * page_count() is not 0.
6439 */
6440 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
6441 pfn++;
6442 SetPageReserved(page);
6443 continue;
6444 }
6445
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006446 BUG_ON(page_count(page));
6447 BUG_ON(!PageBuddy(page));
6448 order = page_order(page);
6449#ifdef CONFIG_DEBUG_VM
6450 printk(KERN_INFO "remove from free list %lx %d %lx\n",
6451 pfn, 1 << order, end_pfn);
6452#endif
6453 list_del(&page->lru);
6454 rmv_page_order(page);
6455 zone->free_area[order].nr_free--;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006456 for (i = 0; i < (1 << order); i++)
6457 SetPageReserved((page+i));
6458 pfn += (1 << order);
6459 }
6460 spin_unlock_irqrestore(&zone->lock, flags);
6461}
6462#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01006463
6464#ifdef CONFIG_MEMORY_FAILURE
6465bool is_free_buddy_page(struct page *page)
6466{
6467 struct zone *zone = page_zone(page);
6468 unsigned long pfn = page_to_pfn(page);
6469 unsigned long flags;
6470 int order;
6471
6472 spin_lock_irqsave(&zone->lock, flags);
6473 for (order = 0; order < MAX_ORDER; order++) {
6474 struct page *page_head = page - (pfn & ((1 << order) - 1));
6475
6476 if (PageBuddy(page_head) && page_order(page_head) >= order)
6477 break;
6478 }
6479 spin_unlock_irqrestore(&zone->lock, flags);
6480
6481 return order < MAX_ORDER;
6482}
6483#endif
Wu Fengguang718a3822010-03-10 15:20:43 -08006484
Andrew Morton51300ce2012-05-29 15:06:44 -07006485static const struct trace_print_flags pageflag_names[] = {
Wu Fengguang718a3822010-03-10 15:20:43 -08006486 {1UL << PG_locked, "locked" },
6487 {1UL << PG_error, "error" },
6488 {1UL << PG_referenced, "referenced" },
6489 {1UL << PG_uptodate, "uptodate" },
6490 {1UL << PG_dirty, "dirty" },
6491 {1UL << PG_lru, "lru" },
6492 {1UL << PG_active, "active" },
6493 {1UL << PG_slab, "slab" },
6494 {1UL << PG_owner_priv_1, "owner_priv_1" },
6495 {1UL << PG_arch_1, "arch_1" },
6496 {1UL << PG_reserved, "reserved" },
6497 {1UL << PG_private, "private" },
6498 {1UL << PG_private_2, "private_2" },
6499 {1UL << PG_writeback, "writeback" },
6500#ifdef CONFIG_PAGEFLAGS_EXTENDED
6501 {1UL << PG_head, "head" },
6502 {1UL << PG_tail, "tail" },
6503#else
6504 {1UL << PG_compound, "compound" },
6505#endif
6506 {1UL << PG_swapcache, "swapcache" },
6507 {1UL << PG_mappedtodisk, "mappedtodisk" },
6508 {1UL << PG_reclaim, "reclaim" },
Wu Fengguang718a3822010-03-10 15:20:43 -08006509 {1UL << PG_swapbacked, "swapbacked" },
6510 {1UL << PG_unevictable, "unevictable" },
6511#ifdef CONFIG_MMU
6512 {1UL << PG_mlocked, "mlocked" },
6513#endif
6514#ifdef CONFIG_ARCH_USES_PG_UNCACHED
6515 {1UL << PG_uncached, "uncached" },
6516#endif
6517#ifdef CONFIG_MEMORY_FAILURE
6518 {1UL << PG_hwpoison, "hwpoison" },
6519#endif
Gavin Shanbe9cd872012-05-29 15:06:44 -07006520#ifdef CONFIG_TRANSPARENT_HUGEPAGE
6521 {1UL << PG_compound_lock, "compound_lock" },
6522#endif
Wu Fengguang718a3822010-03-10 15:20:43 -08006523};
6524
6525static void dump_page_flags(unsigned long flags)
6526{
6527 const char *delim = "";
6528 unsigned long mask;
6529 int i;
6530
Andrew Morton51300ce2012-05-29 15:06:44 -07006531 BUILD_BUG_ON(ARRAY_SIZE(pageflag_names) != __NR_PAGEFLAGS);
Johannes Weineracc50c12012-05-29 15:06:44 -07006532
Wu Fengguang718a3822010-03-10 15:20:43 -08006533 printk(KERN_ALERT "page flags: %#lx(", flags);
6534
6535 /* remove zone id */
6536 flags &= (1UL << NR_PAGEFLAGS) - 1;
6537
Andrew Morton51300ce2012-05-29 15:06:44 -07006538 for (i = 0; i < ARRAY_SIZE(pageflag_names) && flags; i++) {
Wu Fengguang718a3822010-03-10 15:20:43 -08006539
6540 mask = pageflag_names[i].mask;
6541 if ((flags & mask) != mask)
6542 continue;
6543
6544 flags &= ~mask;
6545 printk("%s%s", delim, pageflag_names[i].name);
6546 delim = "|";
6547 }
6548
6549 /* check for left over flags */
6550 if (flags)
6551 printk("%s%#lx", delim, flags);
6552
6553 printk(")\n");
6554}
6555
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -07006556void dump_page_badflags(struct page *page, const char *reason,
6557 unsigned long badflags)
Wu Fengguang718a3822010-03-10 15:20:43 -08006558{
6559 printk(KERN_ALERT
6560 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
Andrea Arcangeli4e9f64c2011-01-13 15:46:29 -08006561 page, atomic_read(&page->_count), page_mapcount(page),
Wu Fengguang718a3822010-03-10 15:20:43 -08006562 page->mapping, page->index);
6563 dump_page_flags(page->flags);
Dave Hansenf0b791a2014-01-23 15:52:49 -08006564 if (reason)
6565 pr_alert("page dumped because: %s\n", reason);
6566 if (page->flags & badflags) {
6567 pr_alert("bad because of flags:\n");
6568 dump_page_flags(page->flags & badflags);
6569 }
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07006570 mem_cgroup_print_bad_page(page);
Wu Fengguang718a3822010-03-10 15:20:43 -08006571}
Dave Hansenf0b791a2014-01-23 15:52:49 -08006572
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -07006573void dump_page(struct page *page, const char *reason)
Dave Hansenf0b791a2014-01-23 15:52:49 -08006574{
6575 dump_page_badflags(page, reason, 0);
6576}
John Hubbarded12d842014-04-07 15:37:59 -07006577EXPORT_SYMBOL(dump_page);