blob: 5248fe070aa4e9f94b4be087aa8957e16cf16c1f [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 Gruszkac0a32fc52012-01-10 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;
Michal Hocko5f127332013-07-08 16:00:40 -0700208int user_min_free_kbytes;
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)
264 pr_err("page %lu outside zone [ %lu - %lu ]\n",
265 pfn, start_pfn, start_pfn + sp);
266
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700267 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700268}
269
270static int page_is_consistent(struct zone *zone, struct page *page)
271{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700272 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700273 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700275 return 0;
276
277 return 1;
278}
279/*
280 * Temporary debugging check for pages not lying within a given zone.
281 */
282static int bad_range(struct zone *zone, struct page *page)
283{
284 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700286 if (!page_is_consistent(zone, page))
287 return 1;
288
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 return 0;
290}
Nick Piggin13e74442006-01-06 00:10:58 -0800291#else
292static inline int bad_range(struct zone *zone, struct page *page)
293{
294 return 0;
295}
296#endif
297
Nick Piggin224abf92006-01-06 00:11:11 -0800298static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800300 static unsigned long resume;
301 static unsigned long nr_shown;
302 static unsigned long nr_unshown;
303
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200304 /* Don't complain about poisoned pages */
305 if (PageHWPoison(page)) {
Mel Gorman22b751c2013-02-22 16:34:59 -0800306 page_mapcount_reset(page); /* remove PageBuddy */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200307 return;
308 }
309
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800310 /*
311 * Allow a burst of 60 reports, then keep quiet for that minute;
312 * or allow a steady drip of one report per second.
313 */
314 if (nr_shown == 60) {
315 if (time_before(jiffies, resume)) {
316 nr_unshown++;
317 goto out;
318 }
319 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800320 printk(KERN_ALERT
321 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800322 nr_unshown);
323 nr_unshown = 0;
324 }
325 nr_shown = 0;
326 }
327 if (nr_shown++ == 0)
328 resume = jiffies + 60 * HZ;
329
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800330 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800331 current->comm, page_to_pfn(page));
Wu Fengguang718a3822010-03-10 15:20:43 -0800332 dump_page(page);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700333
Dave Jones4f318882011-10-31 17:07:24 -0700334 print_modules();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800336out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800337 /* Leave bad fields for debug, except PageBuddy could make trouble */
Mel Gorman22b751c2013-02-22 16:34:59 -0800338 page_mapcount_reset(page); /* remove PageBuddy */
Rusty Russell373d4d02013-01-21 17:17:39 +1030339 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340}
341
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342/*
343 * Higher-order pages are called "compound pages". They are structured thusly:
344 *
345 * The first PAGE_SIZE page is called the "head page".
346 *
347 * The remaining PAGE_SIZE pages are called "tail pages".
348 *
Wang Sheng-Hui6416b9f2011-11-17 10:53:50 +0100349 * All pages have PG_compound set. All tail pages have their ->first_page
350 * pointing at the head page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800352 * The first tail page's ->lru.next holds the address of the compound page's
353 * put_page() function. Its ->lru.prev holds the order of allocation.
354 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800356
357static void free_compound_page(struct page *page)
358{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700359 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800360}
361
Andi Kleen01ad1c02008-07-23 21:27:46 -0700362void prep_compound_page(struct page *page, unsigned long order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363{
364 int i;
365 int nr_pages = 1 << order;
366
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800367 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700368 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700369 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800370 for (i = 1; i < nr_pages; i++) {
371 struct page *p = page + i;
Christoph Lameterd85f3382007-05-06 14:49:39 -0700372 __SetPageTail(p);
Youquan Song58a84aa2011-12-08 14:34:18 -0800373 set_page_count(p, 0);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700374 p->first_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 }
376}
377
Andrea Arcangeli59ff4212011-01-13 15:46:38 -0800378/* update __split_huge_page_refcount if you change this function */
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800379static int destroy_compound_page(struct page *page, unsigned long order)
Andy Whitcroft18229df2008-11-06 12:53:27 -0800380{
381 int i;
382 int nr_pages = 1 << order;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800383 int bad = 0;
Andy Whitcroft18229df2008-11-06 12:53:27 -0800384
Gavin Shan0bb2c762012-12-18 14:21:32 -0800385 if (unlikely(compound_order(page) != order)) {
Nick Piggin224abf92006-01-06 00:11:11 -0800386 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800387 bad++;
388 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389
Christoph Lameter6d777952007-05-06 14:49:40 -0700390 __ClearPageHead(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800391
Andy Whitcroft18229df2008-11-06 12:53:27 -0800392 for (i = 1; i < nr_pages; i++) {
393 struct page *p = page + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394
Alexey Zaytseve713a212009-01-10 02:47:57 +0300395 if (unlikely(!PageTail(p) || (p->first_page != page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800396 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800397 bad++;
398 }
Christoph Lameterd85f3382007-05-06 14:49:39 -0700399 __ClearPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800401
402 return bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404
Nick Piggin17cf4402006-03-22 00:08:41 -0800405static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
406{
407 int i;
408
Andrew Morton6626c5d2006-03-22 00:08:42 -0800409 /*
410 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
411 * and __GFP_HIGHMEM from hard or soft interrupt context.
412 */
Nick Piggin725d7042006-09-25 23:30:55 -0700413 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800414 for (i = 0; i < (1 << order); i++)
415 clear_highpage(page + i);
416}
417
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800418#ifdef CONFIG_DEBUG_PAGEALLOC
419unsigned int _debug_guardpage_minorder;
420
421static int __init debug_guardpage_minorder_setup(char *buf)
422{
423 unsigned long res;
424
425 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
426 printk(KERN_ERR "Bad debug_guardpage_minorder value\n");
427 return 0;
428 }
429 _debug_guardpage_minorder = res;
430 printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res);
431 return 0;
432}
433__setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
434
435static inline void set_page_guard_flag(struct page *page)
436{
437 __set_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags);
438}
439
440static inline void clear_page_guard_flag(struct page *page)
441{
442 __clear_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags);
443}
444#else
445static inline void set_page_guard_flag(struct page *page) { }
446static inline void clear_page_guard_flag(struct page *page) { }
447#endif
448
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700449static inline void set_page_order(struct page *page, int order)
450{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700451 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000452 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453}
454
455static inline void rmv_page_order(struct page *page)
456{
Nick Piggin676165a2006-04-10 11:21:48 +1000457 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700458 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459}
460
461/*
462 * Locate the struct page for both the matching buddy in our
463 * pair (buddy1) and the combined O(n+1) page they form (page).
464 *
465 * 1) Any buddy B1 will have an order O twin B2 which satisfies
466 * the following equation:
467 * B2 = B1 ^ (1 << O)
468 * For example, if the starting buddy (buddy2) is #8 its order
469 * 1 buddy is #10:
470 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
471 *
472 * 2) Any buddy B will have an order O+1 parent P which
473 * satisfies the following equation:
474 * P = B & ~(1 << O)
475 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200476 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478static inline unsigned long
KyongHo Cho43506fa2011-01-13 15:47:24 -0800479__find_buddy_index(unsigned long page_idx, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480{
KyongHo Cho43506fa2011-01-13 15:47:24 -0800481 return page_idx ^ (1 << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482}
483
484/*
485 * This function checks whether a page is free && is the buddy
486 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800487 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000488 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700489 * (c) a page and its buddy have the same order &&
490 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 *
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700492 * For recording whether a page is in the buddy system, we set ->_mapcount
493 * PAGE_BUDDY_MAPCOUNT_VALUE.
494 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
495 * serialized by zone->lock.
Nick Piggin676165a2006-04-10 11:21:48 +1000496 *
497 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700499static inline int page_is_buddy(struct page *page, struct page *buddy,
500 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700502 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800503 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800504
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700505 if (page_zone_id(page) != page_zone_id(buddy))
506 return 0;
507
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800508 if (page_is_guard(buddy) && page_order(buddy) == order) {
509 VM_BUG_ON(page_count(buddy) != 0);
510 return 1;
511 }
512
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700513 if (PageBuddy(buddy) && page_order(buddy) == order) {
Nick Piggina3af9c32009-06-16 15:32:10 -0700514 VM_BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700515 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000516 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700517 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518}
519
520/*
521 * Freeing function for a buddy system allocator.
522 *
523 * The concept of a buddy system is to maintain direct-mapped table
524 * (containing bit values) for memory blocks of various "orders".
525 * The bottom level table contains the map for the smallest allocatable
526 * units of memory (here, pages), and each level above it describes
527 * pairs of units from the levels below, hence, "buddies".
528 * At a high level, all that happens here is marking the table entry
529 * at the bottom level available, and propagating the changes upward
530 * as necessary, plus some accounting needed to play nicely with other
531 * parts of the VM system.
532 * At each level, we keep a list of pages, which are heads of continuous
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700533 * free pages of length of (1 << order) and marked with _mapcount
534 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
535 * field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 * So when we are allocating or freeing one, we can derive the state of the
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100537 * other. That is, if we allocate a small block, and both were
538 * free, the remainder of the region must be split into blocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 * If a block is freed, and its buddy is also free, then this
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100540 * triggers coalescing into a block of larger size.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +0100542 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 */
544
Nick Piggin48db57f2006-01-08 01:00:42 -0800545static inline void __free_one_page(struct page *page,
Mel Gormaned0ae212009-06-16 15:32:07 -0700546 struct zone *zone, unsigned int order,
547 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548{
549 unsigned long page_idx;
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700550 unsigned long combined_idx;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800551 unsigned long uninitialized_var(buddy_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700552 struct page *buddy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553
Cody P Schaferd29bb972013-02-22 16:35:25 -0800554 VM_BUG_ON(!zone_is_initialized(zone));
555
Nick Piggin224abf92006-01-06 00:11:11 -0800556 if (unlikely(PageCompound(page)))
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800557 if (unlikely(destroy_compound_page(page, order)))
558 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559
Mel Gormaned0ae212009-06-16 15:32:07 -0700560 VM_BUG_ON(migratetype == -1);
561
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
563
Mel Gormanf2260e62009-06-16 15:32:13 -0700564 VM_BUG_ON(page_idx & ((1 << order) - 1));
Nick Piggin725d7042006-09-25 23:30:55 -0700565 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 while (order < MAX_ORDER-1) {
KyongHo Cho43506fa2011-01-13 15:47:24 -0800568 buddy_idx = __find_buddy_index(page_idx, order);
569 buddy = page + (buddy_idx - page_idx);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700570 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700571 break;
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800572 /*
573 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
574 * merge with it and move up one order.
575 */
576 if (page_is_guard(buddy)) {
577 clear_page_guard_flag(buddy);
578 set_page_private(page, 0);
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -0700579 __mod_zone_freepage_state(zone, 1 << order,
580 migratetype);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800581 } else {
582 list_del(&buddy->lru);
583 zone->free_area[order].nr_free--;
584 rmv_page_order(buddy);
585 }
KyongHo Cho43506fa2011-01-13 15:47:24 -0800586 combined_idx = buddy_idx & page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 page = page + (combined_idx - page_idx);
588 page_idx = combined_idx;
589 order++;
590 }
591 set_page_order(page, order);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700592
593 /*
594 * If this is not the largest possible page, check if the buddy
595 * of the next-highest order is free. If it is, it's possible
596 * that pages are being freed that will coalesce soon. In case,
597 * that is happening, add the free page to the tail of the list
598 * so it's less likely to be used soon and more likely to be merged
599 * as a higher order page
600 */
Mel Gormanb7f50cf2010-10-26 14:21:11 -0700601 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700602 struct page *higher_page, *higher_buddy;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800603 combined_idx = buddy_idx & page_idx;
604 higher_page = page + (combined_idx - page_idx);
605 buddy_idx = __find_buddy_index(combined_idx, order + 1);
Li Haifeng0ba8f2d2012-09-17 14:09:21 -0700606 higher_buddy = higher_page + (buddy_idx - combined_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700607 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
608 list_add_tail(&page->lru,
609 &zone->free_area[order].free_list[migratetype]);
610 goto out;
611 }
612 }
613
614 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
615out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 zone->free_area[order].nr_free++;
617}
618
Nick Piggin224abf92006-01-06 00:11:11 -0800619static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620{
Nick Piggin92be2e332006-01-06 00:10:57 -0800621 if (unlikely(page_mapcount(page) |
622 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700623 (atomic_read(&page->_count) != 0) |
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -0700624 (page->flags & PAGE_FLAGS_CHECK_AT_FREE) |
625 (mem_cgroup_bad_page_check(page)))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800626 bad_page(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800627 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800628 }
Peter Zijlstra90572892013-10-07 11:29:20 +0100629 page_cpupid_reset_last(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800630 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
631 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
632 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633}
634
635/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700636 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700638 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 *
640 * If the zone was previously in an "all pages pinned" state then look to
641 * see if this freeing clears that state.
642 *
643 * And clear the zone's pages_scanned counter, to hold off the "all pages are
644 * pinned" detection logic.
645 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700646static void free_pcppages_bulk(struct zone *zone, int count,
647 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648{
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700649 int migratetype = 0;
Mel Gormana6f9edd2009-09-21 17:03:20 -0700650 int batch_free = 0;
Mel Gorman72853e22010-09-09 16:38:16 -0700651 int to_free = count;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700652
Nick Pigginc54ad302006-01-06 00:10:56 -0800653 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700655
Mel Gorman72853e22010-09-09 16:38:16 -0700656 while (to_free) {
Nick Piggin48db57f2006-01-08 01:00:42 -0800657 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700658 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -0800659
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700660 /*
Mel Gormana6f9edd2009-09-21 17:03:20 -0700661 * Remove pages from lists in a round-robin fashion. A
662 * batch_free count is maintained that is incremented when an
663 * empty list is encountered. This is so more pages are freed
664 * off fuller lists instead of spinning excessively around empty
665 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700666 */
667 do {
Mel Gormana6f9edd2009-09-21 17:03:20 -0700668 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700669 if (++migratetype == MIGRATE_PCPTYPES)
670 migratetype = 0;
671 list = &pcp->lists[migratetype];
672 } while (list_empty(list));
673
Namhyung Kim1d168712011-03-22 16:32:45 -0700674 /* This is the only non-empty list. Free them all. */
675 if (batch_free == MIGRATE_PCPTYPES)
676 batch_free = to_free;
677
Mel Gormana6f9edd2009-09-21 17:03:20 -0700678 do {
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -0700679 int mt; /* migratetype of the to-be-freed page */
680
Mel Gormana6f9edd2009-09-21 17:03:20 -0700681 page = list_entry(list->prev, struct page, lru);
682 /* must delete as __free_one_page list manipulates */
683 list_del(&page->lru);
Minchan Kimb12c4ad2012-10-08 16:32:08 -0700684 mt = get_freepage_migratetype(page);
Hugh Dickinsa7016232010-01-29 17:46:34 +0000685 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -0700686 __free_one_page(page, zone, 0, mt);
687 trace_mm_page_pcpu_drain(page, 0, mt);
Minchan Kim194159f2013-02-22 16:33:58 -0800688 if (likely(!is_migrate_isolate_page(page))) {
Wen Congyang97d0da22012-12-11 16:00:52 -0800689 __mod_zone_page_state(zone, NR_FREE_PAGES, 1);
690 if (is_migrate_cma(mt))
691 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, 1);
692 }
Mel Gorman72853e22010-09-09 16:38:16 -0700693 } while (--to_free && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800695 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696}
697
Mel Gormaned0ae212009-06-16 15:32:07 -0700698static void free_one_page(struct zone *zone, struct page *page, int order,
699 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800700{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700701 spin_lock(&zone->lock);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700702 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700703
Mel Gormaned0ae212009-06-16 15:32:07 -0700704 __free_one_page(page, zone, order, migratetype);
Minchan Kim194159f2013-02-22 16:33:58 -0800705 if (unlikely(!is_migrate_isolate(migratetype)))
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -0700706 __mod_zone_freepage_state(zone, 1 << order, migratetype);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700707 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800708}
709
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700710static bool free_pages_prepare(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 int i;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800713 int bad = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -0800715 trace_mm_page_free(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100716 kmemcheck_free_shadow(page, order);
717
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -0800718 if (PageAnon(page))
719 page->mapping = NULL;
720 for (i = 0; i < (1 << order); i++)
721 bad += free_pages_check(page + i);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800722 if (bad)
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700723 return false;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800724
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700725 if (!PageHighMem(page)) {
Pintu Kumarb8af2942013-09-11 14:20:34 -0700726 debug_check_no_locks_freed(page_address(page),
727 PAGE_SIZE << order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700728 debug_check_no_obj_freed(page_address(page),
729 PAGE_SIZE << order);
730 }
Nick Piggindafb1362006-10-11 01:21:30 -0700731 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800732 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700733
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700734 return true;
735}
736
737static void __free_pages_ok(struct page *page, unsigned int order)
738{
739 unsigned long flags;
Minchan Kim95e34412012-10-08 16:32:11 -0700740 int migratetype;
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700741
742 if (!free_pages_prepare(page, order))
743 return;
744
Nick Pigginc54ad302006-01-06 00:10:56 -0800745 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700746 __count_vm_events(PGFREE, 1 << order);
Minchan Kim95e34412012-10-08 16:32:11 -0700747 migratetype = get_pageblock_migratetype(page);
748 set_freepage_migratetype(page, migratetype);
749 free_one_page(page_zone(page), page, order, migratetype);
Nick Pigginc54ad302006-01-06 00:10:56 -0800750 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751}
752
Jiang Liu170a5a72013-07-03 15:03:17 -0700753void __init __free_pages_bootmem(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800754{
Johannes Weinerc3993072012-01-10 15:08:10 -0800755 unsigned int nr_pages = 1 << order;
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700756 struct page *p = page;
Johannes Weinerc3993072012-01-10 15:08:10 -0800757 unsigned int loop;
David Howellsa226f6c2006-01-06 00:11:08 -0800758
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700759 prefetchw(p);
760 for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
761 prefetchw(p + 1);
Johannes Weinerc3993072012-01-10 15:08:10 -0800762 __ClearPageReserved(p);
763 set_page_count(p, 0);
David Howellsa226f6c2006-01-06 00:11:08 -0800764 }
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700765 __ClearPageReserved(p);
766 set_page_count(p, 0);
Johannes Weinerc3993072012-01-10 15:08:10 -0800767
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700768 page_zone(page)->managed_pages += nr_pages;
Johannes Weinerc3993072012-01-10 15:08:10 -0800769 set_page_refcounted(page);
770 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800771}
772
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100773#ifdef CONFIG_CMA
Li Zhong9cf510a2013-08-23 13:52:52 +0800774/* Free whole pageblock and set its migration type to MIGRATE_CMA. */
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100775void __init init_cma_reserved_pageblock(struct page *page)
776{
777 unsigned i = pageblock_nr_pages;
778 struct page *p = page;
779
780 do {
781 __ClearPageReserved(p);
782 set_page_count(p, 0);
783 } while (++p, --i);
784
785 set_page_refcounted(page);
786 set_pageblock_migratetype(page, MIGRATE_CMA);
787 __free_pages(page, pageblock_order);
Jiang Liu3dcc0572013-07-03 15:03:21 -0700788 adjust_managed_page_count(page, pageblock_nr_pages);
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100789}
790#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791
792/*
793 * The order of subdivision here is critical for the IO subsystem.
794 * Please do not alter this order without good reasons and regression
795 * testing. Specifically, as large blocks of memory are subdivided,
796 * the order in which smaller blocks are delivered depends on the order
797 * they're subdivided in this function. This is the primary factor
798 * influencing the order in which pages are delivered to the IO
799 * subsystem according to empirical testing, and this is also justified
800 * by considering the behavior of a buddy system containing a single
801 * large block of memory acted on by a series of small allocations.
802 * This behavior is a critical factor in sglist merging's success.
803 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +0100804 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800806static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700807 int low, int high, struct free_area *area,
808 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809{
810 unsigned long size = 1 << high;
811
812 while (high > low) {
813 area--;
814 high--;
815 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700816 VM_BUG_ON(bad_range(zone, &page[size]));
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800817
818#ifdef CONFIG_DEBUG_PAGEALLOC
819 if (high < debug_guardpage_minorder()) {
820 /*
821 * Mark as guard pages (or page), that will allow to
822 * merge back to allocator when buddy will be freed.
823 * Corresponding page table entries will not be touched,
824 * pages will stay not present in virtual address space
825 */
826 INIT_LIST_HEAD(&page[size].lru);
827 set_page_guard_flag(&page[size]);
828 set_page_private(&page[size], high);
829 /* Guard pages are not available for any usage */
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -0700830 __mod_zone_freepage_state(zone, -(1 << high),
831 migratetype);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800832 continue;
833 }
834#endif
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700835 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 area->nr_free++;
837 set_page_order(&page[size], high);
838 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839}
840
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841/*
842 * This page is about to be returned from the page allocator
843 */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200844static inline int check_new_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845{
Nick Piggin92be2e332006-01-06 00:10:57 -0800846 if (unlikely(page_mapcount(page) |
847 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700848 (atomic_read(&page->_count) != 0) |
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -0700849 (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
850 (mem_cgroup_bad_page_check(page)))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800851 bad_page(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800852 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800853 }
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200854 return 0;
855}
856
857static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
858{
859 int i;
860
861 for (i = 0; i < (1 << order); i++) {
862 struct page *p = page + i;
863 if (unlikely(check_new_page(p)))
864 return 1;
865 }
Hugh Dickins689bceb2005-11-21 21:32:20 -0800866
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700867 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800868 set_page_refcounted(page);
Nick Piggincc1025092006-12-06 20:32:00 -0800869
870 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800872
873 if (gfp_flags & __GFP_ZERO)
874 prep_zero_page(page, order, gfp_flags);
875
876 if (order && (gfp_flags & __GFP_COMP))
877 prep_compound_page(page, order);
878
Hugh Dickins689bceb2005-11-21 21:32:20 -0800879 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880}
881
Mel Gorman56fd56b2007-10-16 01:25:58 -0700882/*
883 * Go through the free lists for the given migratetype and remove
884 * the smallest available page from the freelists
885 */
Mel Gorman728ec982009-06-16 15:32:04 -0700886static inline
887struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -0700888 int migratetype)
889{
890 unsigned int current_order;
Pintu Kumarb8af2942013-09-11 14:20:34 -0700891 struct free_area *area;
Mel Gorman56fd56b2007-10-16 01:25:58 -0700892 struct page *page;
893
894 /* Find a page of the appropriate size in the preferred list */
895 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
896 area = &(zone->free_area[current_order]);
897 if (list_empty(&area->free_list[migratetype]))
898 continue;
899
900 page = list_entry(area->free_list[migratetype].next,
901 struct page, lru);
902 list_del(&page->lru);
903 rmv_page_order(page);
904 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -0700905 expand(zone, page, order, current_order, area, migratetype);
906 return page;
907 }
908
909 return NULL;
910}
911
912
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700913/*
914 * This array describes the order lists are fallen back to when
915 * the free lists for the desirable migrate type are depleted
916 */
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100917static int fallbacks[MIGRATE_TYPES][4] = {
918 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
919 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
920#ifdef CONFIG_CMA
921 [MIGRATE_MOVABLE] = { MIGRATE_CMA, MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
922 [MIGRATE_CMA] = { MIGRATE_RESERVE }, /* Never used */
923#else
924 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
925#endif
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +0100926 [MIGRATE_RESERVE] = { MIGRATE_RESERVE }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -0800927#ifdef CONFIG_MEMORY_ISOLATION
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +0100928 [MIGRATE_ISOLATE] = { MIGRATE_RESERVE }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -0800929#endif
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700930};
931
Mel Gormanc361be52007-10-16 01:25:51 -0700932/*
933 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -0700934 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -0700935 * boundary. If alignment is required, use move_freepages_block()
936 */
Minchan Kim435b4052012-10-08 16:32:16 -0700937int move_freepages(struct zone *zone,
Adrian Bunkb69a7282008-07-23 21:28:12 -0700938 struct page *start_page, struct page *end_page,
939 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700940{
941 struct page *page;
942 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -0700943 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -0700944
945#ifndef CONFIG_HOLES_IN_ZONE
946 /*
947 * page_zone is not safe to call in this context when
948 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
949 * anyway as we check zone boundaries in move_freepages_block().
950 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -0700951 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -0700952 */
953 BUG_ON(page_zone(start_page) != page_zone(end_page));
954#endif
955
956 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -0700957 /* Make sure we are not inadvertently changing nodes */
958 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
959
Mel Gormanc361be52007-10-16 01:25:51 -0700960 if (!pfn_valid_within(page_to_pfn(page))) {
961 page++;
962 continue;
963 }
964
965 if (!PageBuddy(page)) {
966 page++;
967 continue;
968 }
969
970 order = page_order(page);
Kirill A. Shutemov84be48d2011-03-22 16:33:41 -0700971 list_move(&page->lru,
972 &zone->free_area[order].free_list[migratetype]);
Minchan Kim95e34412012-10-08 16:32:11 -0700973 set_freepage_migratetype(page, migratetype);
Mel Gormanc361be52007-10-16 01:25:51 -0700974 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -0700975 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -0700976 }
977
Mel Gormand1003132007-10-16 01:26:00 -0700978 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -0700979}
980
Minchan Kimee6f5092012-07-31 16:43:50 -0700981int move_freepages_block(struct zone *zone, struct page *page,
Linus Torvalds68e3e922012-06-03 20:05:57 -0700982 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700983{
984 unsigned long start_pfn, end_pfn;
985 struct page *start_page, *end_page;
986
987 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -0700988 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -0700989 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -0700990 end_page = start_page + pageblock_nr_pages - 1;
991 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -0700992
993 /* Do not cross zone boundaries */
Cody P Schafer108bcc92013-02-22 16:35:23 -0800994 if (!zone_spans_pfn(zone, start_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -0700995 start_page = page;
Cody P Schafer108bcc92013-02-22 16:35:23 -0800996 if (!zone_spans_pfn(zone, end_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -0700997 return 0;
998
999 return move_freepages(zone, start_page, end_page, migratetype);
1000}
1001
Mel Gorman2f66a682009-09-21 17:02:31 -07001002static void change_pageblock_range(struct page *pageblock_page,
1003 int start_order, int migratetype)
1004{
1005 int nr_pageblocks = 1 << (start_order - pageblock_order);
1006
1007 while (nr_pageblocks--) {
1008 set_pageblock_migratetype(pageblock_page, migratetype);
1009 pageblock_page += pageblock_nr_pages;
1010 }
1011}
1012
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001013/*
1014 * If breaking a large block of pages, move all free pages to the preferred
1015 * allocation list. If falling back for a reclaimable kernel allocation, be
1016 * more aggressive about taking ownership of free pages.
1017 *
1018 * On the other hand, never change migration type of MIGRATE_CMA pageblocks
1019 * nor move CMA pages to different free lists. We don't want unmovable pages
1020 * to be allocated from MIGRATE_CMA areas.
1021 *
1022 * Returns the new migratetype of the pageblock (or the same old migratetype
1023 * if it was unchanged).
1024 */
1025static int try_to_steal_freepages(struct zone *zone, struct page *page,
1026 int start_type, int fallback_type)
1027{
1028 int current_order = page_order(page);
1029
KOSAKI Motohiro0cbef292013-11-12 15:08:20 -08001030 /*
1031 * When borrowing from MIGRATE_CMA, we need to release the excess
1032 * buddy pages to CMA itself.
1033 */
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001034 if (is_migrate_cma(fallback_type))
1035 return fallback_type;
1036
1037 /* Take ownership for orders >= pageblock_order */
1038 if (current_order >= pageblock_order) {
1039 change_pageblock_range(page, current_order, start_type);
1040 return start_type;
1041 }
1042
1043 if (current_order >= pageblock_order / 2 ||
1044 start_type == MIGRATE_RECLAIMABLE ||
1045 page_group_by_mobility_disabled) {
1046 int pages;
1047
1048 pages = move_freepages_block(zone, page, start_type);
1049
1050 /* Claim the whole block if over half of it is free */
1051 if (pages >= (1 << (pageblock_order-1)) ||
1052 page_group_by_mobility_disabled) {
1053
1054 set_pageblock_migratetype(page, start_type);
1055 return start_type;
1056 }
1057
1058 }
1059
1060 return fallback_type;
1061}
1062
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001063/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -07001064static inline struct page *
1065__rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001066{
Pintu Kumarb8af2942013-09-11 14:20:34 -07001067 struct free_area *area;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001068 int current_order;
1069 struct page *page;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001070 int migratetype, new_type, i;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001071
1072 /* Find the largest possible block of pages in the other list */
1073 for (current_order = MAX_ORDER-1; current_order >= order;
1074 --current_order) {
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001075 for (i = 0;; i++) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001076 migratetype = fallbacks[start_migratetype][i];
1077
Mel Gorman56fd56b2007-10-16 01:25:58 -07001078 /* MIGRATE_RESERVE handled later if necessary */
1079 if (migratetype == MIGRATE_RESERVE)
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001080 break;
Mel Gormane0104872007-10-16 01:25:53 -07001081
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001082 area = &(zone->free_area[current_order]);
1083 if (list_empty(&area->free_list[migratetype]))
1084 continue;
1085
1086 page = list_entry(area->free_list[migratetype].next,
1087 struct page, lru);
1088 area->nr_free--;
1089
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001090 new_type = try_to_steal_freepages(zone, page,
1091 start_migratetype,
1092 migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001093
1094 /* Remove the page from the freelists */
1095 list_del(&page->lru);
1096 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001097
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001098 expand(zone, page, order, current_order, area,
KOSAKI Motohiro0cbef292013-11-12 15:08:20 -08001099 new_type);
Mel Gormane0fff1b2009-09-21 17:02:42 -07001100
KOSAKI Motohiro52c8f6a2013-11-12 15:08:19 -08001101 trace_mm_page_alloc_extfrag(page, order, current_order,
1102 start_migratetype, migratetype, new_type);
Mel Gormane0fff1b2009-09-21 17:02:42 -07001103
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001104 return page;
1105 }
1106 }
1107
Mel Gorman728ec982009-06-16 15:32:04 -07001108 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001109}
1110
Mel Gorman56fd56b2007-10-16 01:25:58 -07001111/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 * Do the hard work of removing an element from the buddy allocator.
1113 * Call me with the zone->lock already held.
1114 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001115static struct page *__rmqueue(struct zone *zone, unsigned int order,
1116 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 struct page *page;
1119
Mel Gorman728ec982009-06-16 15:32:04 -07001120retry_reserve:
Mel Gorman56fd56b2007-10-16 01:25:58 -07001121 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122
Mel Gorman728ec982009-06-16 15:32:04 -07001123 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07001124 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001125
Mel Gorman728ec982009-06-16 15:32:04 -07001126 /*
1127 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1128 * is used because __rmqueue_smallest is an inline function
1129 * and we want just one call site
1130 */
1131 if (!page) {
1132 migratetype = MIGRATE_RESERVE;
1133 goto retry_reserve;
1134 }
1135 }
1136
Mel Gorman0d3d0622009-09-21 17:02:44 -07001137 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001138 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139}
1140
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001141/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 * Obtain a specified number of elements from the buddy allocator, all under
1143 * a single hold of the lock, for efficiency. Add them to the supplied list.
1144 * Returns the number of new pages which were placed at *list.
1145 */
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001146static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001147 unsigned long count, struct list_head *list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001148 int migratetype, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149{
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001150 int mt = migratetype, i;
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001151
Nick Pigginc54ad302006-01-06 00:10:56 -08001152 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001154 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001155 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08001157
1158 /*
1159 * Split buddy pages returned by expand() are received here
1160 * in physical page order. The page is added to the callers and
1161 * list and the list head then moves forward. From the callers
1162 * perspective, the linked list is ordered by page number in
1163 * some conditions. This is useful for IO devices that can
1164 * merge IO requests if the physical pages are ordered
1165 * properly.
1166 */
Mel Gormane084b2d2009-07-29 15:02:04 -07001167 if (likely(cold == 0))
1168 list_add(&page->lru, list);
1169 else
1170 list_add_tail(&page->lru, list);
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001171 if (IS_ENABLED(CONFIG_CMA)) {
1172 mt = get_pageblock_migratetype(page);
Minchan Kim194159f2013-02-22 16:33:58 -08001173 if (!is_migrate_cma(mt) && !is_migrate_isolate(mt))
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001174 mt = migratetype;
1175 }
Minchan Kimb12c4ad2012-10-08 16:32:08 -07001176 set_freepage_migratetype(page, mt);
Mel Gorman81eabcb2007-12-17 16:20:05 -08001177 list = &page->lru;
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07001178 if (is_migrate_cma(mt))
1179 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
1180 -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 }
Mel Gormanf2260e62009-06-16 15:32:13 -07001182 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08001183 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001184 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185}
1186
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001187#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001188/*
Christoph Lameter4037d452007-05-09 02:35:14 -07001189 * Called from the vmstat counter updater to drain pagesets of this
1190 * currently executing processor on remote nodes after they have
1191 * expired.
1192 *
Christoph Lameter879336c2006-03-22 00:09:08 -08001193 * Note that this function must be called with the thread pinned to
1194 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001195 */
Christoph Lameter4037d452007-05-09 02:35:14 -07001196void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001197{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001198 unsigned long flags;
Christoph Lameter4037d452007-05-09 02:35:14 -07001199 int to_drain;
Cody P Schafer998d39c2013-07-03 15:01:32 -07001200 unsigned long batch;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001201
Christoph Lameter4037d452007-05-09 02:35:14 -07001202 local_irq_save(flags);
Cody P Schafer998d39c2013-07-03 15:01:32 -07001203 batch = ACCESS_ONCE(pcp->batch);
1204 if (pcp->count >= batch)
1205 to_drain = batch;
Christoph Lameter4037d452007-05-09 02:35:14 -07001206 else
1207 to_drain = pcp->count;
KOSAKI Motohiro2a135152012-07-31 16:42:53 -07001208 if (to_drain > 0) {
1209 free_pcppages_bulk(zone, to_drain, pcp);
1210 pcp->count -= to_drain;
1211 }
Christoph Lameter4037d452007-05-09 02:35:14 -07001212 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001213}
1214#endif
1215
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001216/*
1217 * Drain pages of the indicated processor.
1218 *
1219 * The processor must either be the current processor and the
1220 * thread pinned to the current processor or a processor that
1221 * is not online.
1222 */
1223static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224{
Nick Pigginc54ad302006-01-06 00:10:56 -08001225 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07001228 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 struct per_cpu_pageset *pset;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001230 struct per_cpu_pages *pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001232 local_irq_save(flags);
1233 pset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001235 pcp = &pset->pcp;
David Rientjes2ff754f2011-01-25 15:07:23 -08001236 if (pcp->count) {
1237 free_pcppages_bulk(zone, pcp->count, pcp);
1238 pcp->count = 0;
1239 }
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001240 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 }
1242}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001244/*
1245 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1246 */
1247void drain_local_pages(void *arg)
1248{
1249 drain_pages(smp_processor_id());
1250}
1251
1252/*
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001253 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1254 *
1255 * Note that this code is protected against sending an IPI to an offline
1256 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1257 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1258 * nothing keeps CPUs from showing up after we populated the cpumask and
1259 * before the call to on_each_cpu_mask().
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001260 */
1261void drain_all_pages(void)
1262{
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001263 int cpu;
1264 struct per_cpu_pageset *pcp;
1265 struct zone *zone;
1266
1267 /*
1268 * Allocate in the BSS so we wont require allocation in
1269 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1270 */
1271 static cpumask_t cpus_with_pcps;
1272
1273 /*
1274 * We don't care about racing with CPU hotplug event
1275 * as offline notification will cause the notified
1276 * cpu to drain that CPU pcps and on_each_cpu_mask
1277 * disables preemption as part of its processing
1278 */
1279 for_each_online_cpu(cpu) {
1280 bool has_pcps = false;
1281 for_each_populated_zone(zone) {
1282 pcp = per_cpu_ptr(zone->pageset, cpu);
1283 if (pcp->pcp.count) {
1284 has_pcps = true;
1285 break;
1286 }
1287 }
1288 if (has_pcps)
1289 cpumask_set_cpu(cpu, &cpus_with_pcps);
1290 else
1291 cpumask_clear_cpu(cpu, &cpus_with_pcps);
1292 }
1293 on_each_cpu_mask(&cpus_with_pcps, drain_local_pages, NULL, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001294}
1295
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02001296#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297
1298void mark_free_pages(struct zone *zone)
1299{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001300 unsigned long pfn, max_zone_pfn;
1301 unsigned long flags;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001302 int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 struct list_head *curr;
1304
Xishi Qiu8080fc02013-09-11 14:21:45 -07001305 if (zone_is_empty(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 return;
1307
1308 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001309
Cody P Schafer108bcc92013-02-22 16:35:23 -08001310 max_zone_pfn = zone_end_pfn(zone);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001311 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1312 if (pfn_valid(pfn)) {
1313 struct page *page = pfn_to_page(pfn);
1314
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001315 if (!swsusp_page_is_forbidden(page))
1316 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001317 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001319 for_each_migratetype_order(order, t) {
1320 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001321 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001323 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1324 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001325 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001326 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001327 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 spin_unlock_irqrestore(&zone->lock, flags);
1329}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001330#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331
1332/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 * Free a 0-order page
Li Hongfc916682010-03-05 13:41:54 -08001334 * cold == 1 ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 */
Li Hongfc916682010-03-05 13:41:54 -08001336void free_hot_cold_page(struct page *page, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337{
1338 struct zone *zone = page_zone(page);
1339 struct per_cpu_pages *pcp;
1340 unsigned long flags;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001341 int migratetype;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001343 if (!free_pages_prepare(page, 0))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001344 return;
1345
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001346 migratetype = get_pageblock_migratetype(page);
Minchan Kimb12c4ad2012-10-08 16:32:08 -07001347 set_freepage_migratetype(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001349 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001350
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001351 /*
1352 * We only track unmovable, reclaimable and movable on pcp lists.
1353 * Free ISOLATE pages back to the allocator because they are being
1354 * offlined but treat RESERVE as movable pages so we can get those
1355 * areas back if necessary. Otherwise, we may have to free
1356 * excessively into the page allocator
1357 */
1358 if (migratetype >= MIGRATE_PCPTYPES) {
Minchan Kim194159f2013-02-22 16:33:58 -08001359 if (unlikely(is_migrate_isolate(migratetype))) {
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001360 free_one_page(zone, page, 0, migratetype);
1361 goto out;
1362 }
1363 migratetype = MIGRATE_MOVABLE;
1364 }
1365
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001366 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001367 if (cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001368 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001369 else
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001370 list_add(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001372 if (pcp->count >= pcp->high) {
Cody P Schafer998d39c2013-07-03 15:01:32 -07001373 unsigned long batch = ACCESS_ONCE(pcp->batch);
1374 free_pcppages_bulk(zone, batch, pcp);
1375 pcp->count -= batch;
Nick Piggin48db57f2006-01-08 01:00:42 -08001376 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001377
1378out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380}
1381
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001382/*
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001383 * Free a list of 0-order pages
1384 */
1385void free_hot_cold_page_list(struct list_head *list, int cold)
1386{
1387 struct page *page, *next;
1388
1389 list_for_each_entry_safe(page, next, list, lru) {
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -08001390 trace_mm_page_free_batched(page, cold);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001391 free_hot_cold_page(page, cold);
1392 }
1393}
1394
1395/*
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001396 * split_page takes a non-compound higher-order page, and splits it into
1397 * n (1<<order) sub-pages: page[0..n]
1398 * Each sub-page must be freed individually.
1399 *
1400 * Note: this is probably too low level an operation for use in drivers.
1401 * Please consult with lkml before using this in your driver.
1402 */
1403void split_page(struct page *page, unsigned int order)
1404{
1405 int i;
1406
Nick Piggin725d7042006-09-25 23:30:55 -07001407 VM_BUG_ON(PageCompound(page));
1408 VM_BUG_ON(!page_count(page));
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001409
1410#ifdef CONFIG_KMEMCHECK
1411 /*
1412 * Split shadow pages too, because free(page[0]) would
1413 * otherwise free the whole shadow.
1414 */
1415 if (kmemcheck_page_is_tracked(page))
1416 split_page(virt_to_page(page[0].shadow), order);
1417#endif
1418
Nick Piggin7835e982006-03-22 00:08:40 -08001419 for (i = 1; i < (1 << order); i++)
1420 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001421}
K. Y. Srinivasan5853ff22013-03-25 15:47:38 -07001422EXPORT_SYMBOL_GPL(split_page);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001423
Mel Gorman8fb74b92013-01-11 14:32:16 -08001424static int __isolate_free_page(struct page *page, unsigned int order)
Mel Gorman748446b2010-05-24 14:32:27 -07001425{
Mel Gorman748446b2010-05-24 14:32:27 -07001426 unsigned long watermark;
1427 struct zone *zone;
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07001428 int mt;
Mel Gorman748446b2010-05-24 14:32:27 -07001429
1430 BUG_ON(!PageBuddy(page));
1431
1432 zone = page_zone(page);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001433 mt = get_pageblock_migratetype(page);
Mel Gorman748446b2010-05-24 14:32:27 -07001434
Minchan Kim194159f2013-02-22 16:33:58 -08001435 if (!is_migrate_isolate(mt)) {
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001436 /* Obey watermarks as if the page was being allocated */
1437 watermark = low_wmark_pages(zone) + (1 << order);
1438 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1439 return 0;
1440
Mel Gorman8fb74b92013-01-11 14:32:16 -08001441 __mod_zone_freepage_state(zone, -(1UL << order), mt);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001442 }
Mel Gorman748446b2010-05-24 14:32:27 -07001443
1444 /* Remove page from free list */
1445 list_del(&page->lru);
1446 zone->free_area[order].nr_free--;
1447 rmv_page_order(page);
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07001448
Mel Gorman8fb74b92013-01-11 14:32:16 -08001449 /* Set the pageblock if the isolated page is at least a pageblock */
Mel Gorman748446b2010-05-24 14:32:27 -07001450 if (order >= pageblock_order - 1) {
1451 struct page *endpage = page + (1 << order) - 1;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001452 for (; page < endpage; page += pageblock_nr_pages) {
1453 int mt = get_pageblock_migratetype(page);
Minchan Kim194159f2013-02-22 16:33:58 -08001454 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001455 set_pageblock_migratetype(page,
1456 MIGRATE_MOVABLE);
1457 }
Mel Gorman748446b2010-05-24 14:32:27 -07001458 }
1459
Mel Gorman8fb74b92013-01-11 14:32:16 -08001460 return 1UL << order;
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001461}
1462
1463/*
1464 * Similar to split_page except the page is already free. As this is only
1465 * being used for migration, the migratetype of the block also changes.
1466 * As this is called with interrupts disabled, the caller is responsible
1467 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1468 * are enabled.
1469 *
1470 * Note: this is probably too low level an operation for use in drivers.
1471 * Please consult with lkml before using this in your driver.
1472 */
1473int split_free_page(struct page *page)
1474{
1475 unsigned int order;
1476 int nr_pages;
1477
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001478 order = page_order(page);
1479
Mel Gorman8fb74b92013-01-11 14:32:16 -08001480 nr_pages = __isolate_free_page(page, order);
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001481 if (!nr_pages)
1482 return 0;
1483
1484 /* Split into individual pages */
1485 set_page_refcounted(page);
1486 split_page(page, order);
1487 return nr_pages;
Mel Gorman748446b2010-05-24 14:32:27 -07001488}
1489
1490/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1492 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1493 * or two.
1494 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07001495static inline
1496struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001497 struct zone *zone, int order, gfp_t gfp_flags,
1498 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499{
1500 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001501 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 int cold = !!(gfp_flags & __GFP_COLD);
1503
Hugh Dickins689bceb2005-11-21 21:32:20 -08001504again:
Nick Piggin48db57f2006-01-08 01:00:42 -08001505 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001507 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 local_irq_save(flags);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001510 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1511 list = &pcp->lists[migratetype];
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001512 if (list_empty(list)) {
Mel Gorman535131e62007-10-16 01:25:49 -07001513 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001514 pcp->batch, list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001515 migratetype, cold);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001516 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07001517 goto failed;
Mel Gorman535131e62007-10-16 01:25:49 -07001518 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001519
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001520 if (cold)
1521 page = list_entry(list->prev, struct page, lru);
1522 else
1523 page = list_entry(list->next, struct page, lru);
1524
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001525 list_del(&page->lru);
1526 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001527 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07001528 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1529 /*
1530 * __GFP_NOFAIL is not to be used in new code.
1531 *
1532 * All __GFP_NOFAIL callers should be fixed so that they
1533 * properly detect and handle allocation failures.
1534 *
1535 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07001536 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07001537 * __GFP_NOFAIL.
1538 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07001539 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07001540 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001542 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001543 spin_unlock(&zone->lock);
1544 if (!page)
1545 goto failed;
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07001546 __mod_zone_freepage_state(zone, -(1 << order),
1547 get_pageblock_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 }
1549
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001550 __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
Christoph Lameterf8891e52006-06-30 01:55:45 -07001551 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Andi Kleen78afd562011-03-22 16:33:12 -07001552 zone_statistics(preferred_zone, zone, gfp_flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001553 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554
Nick Piggin725d7042006-09-25 23:30:55 -07001555 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -08001556 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001557 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001559
1560failed:
1561 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001562 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563}
1564
Akinobu Mita933e3122006-12-08 02:39:45 -08001565#ifdef CONFIG_FAIL_PAGE_ALLOC
1566
Akinobu Mitab2588c42011-07-26 16:09:03 -07001567static struct {
Akinobu Mita933e3122006-12-08 02:39:45 -08001568 struct fault_attr attr;
1569
1570 u32 ignore_gfp_highmem;
1571 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001572 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001573} fail_page_alloc = {
1574 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001575 .ignore_gfp_wait = 1,
1576 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001577 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001578};
1579
1580static int __init setup_fail_page_alloc(char *str)
1581{
1582 return setup_fault_attr(&fail_page_alloc.attr, str);
1583}
1584__setup("fail_page_alloc=", setup_fail_page_alloc);
1585
Gavin Shandeaf3862012-07-31 16:41:51 -07001586static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08001587{
Akinobu Mita54114992007-07-15 23:40:23 -07001588 if (order < fail_page_alloc.min_order)
Gavin Shandeaf3862012-07-31 16:41:51 -07001589 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001590 if (gfp_mask & __GFP_NOFAIL)
Gavin Shandeaf3862012-07-31 16:41:51 -07001591 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001592 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
Gavin Shandeaf3862012-07-31 16:41:51 -07001593 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001594 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
Gavin Shandeaf3862012-07-31 16:41:51 -07001595 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001596
1597 return should_fail(&fail_page_alloc.attr, 1 << order);
1598}
1599
1600#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1601
1602static int __init fail_page_alloc_debugfs(void)
1603{
Al Virof4ae40a2011-07-24 04:33:43 -04001604 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
Akinobu Mita933e3122006-12-08 02:39:45 -08001605 struct dentry *dir;
Akinobu Mita933e3122006-12-08 02:39:45 -08001606
Akinobu Mitadd48c082011-08-03 16:21:01 -07001607 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
1608 &fail_page_alloc.attr);
1609 if (IS_ERR(dir))
1610 return PTR_ERR(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08001611
Akinobu Mitab2588c42011-07-26 16:09:03 -07001612 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
1613 &fail_page_alloc.ignore_gfp_wait))
1614 goto fail;
1615 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1616 &fail_page_alloc.ignore_gfp_highmem))
1617 goto fail;
1618 if (!debugfs_create_u32("min-order", mode, dir,
1619 &fail_page_alloc.min_order))
1620 goto fail;
Akinobu Mita933e3122006-12-08 02:39:45 -08001621
Akinobu Mitab2588c42011-07-26 16:09:03 -07001622 return 0;
1623fail:
Akinobu Mitadd48c082011-08-03 16:21:01 -07001624 debugfs_remove_recursive(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08001625
Akinobu Mitab2588c42011-07-26 16:09:03 -07001626 return -ENOMEM;
Akinobu Mita933e3122006-12-08 02:39:45 -08001627}
1628
1629late_initcall(fail_page_alloc_debugfs);
1630
1631#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1632
1633#else /* CONFIG_FAIL_PAGE_ALLOC */
1634
Gavin Shandeaf3862012-07-31 16:41:51 -07001635static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08001636{
Gavin Shandeaf3862012-07-31 16:41:51 -07001637 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001638}
1639
1640#endif /* CONFIG_FAIL_PAGE_ALLOC */
1641
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642/*
Mel Gorman88f5acf2011-01-13 15:45:41 -08001643 * Return true if free pages are above 'mark'. This takes into account the order
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 * of the allocation.
1645 */
Mel Gorman88f5acf2011-01-13 15:45:41 -08001646static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1647 int classzone_idx, int alloc_flags, long free_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648{
1649 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001650 long min = mark;
Minchan Kim2cfed072012-07-31 16:43:53 -07001651 long lowmem_reserve = z->lowmem_reserve[classzone_idx];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 int o;
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07001653 long free_cma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654
Michal Hockodf0a6da2012-01-10 15:08:02 -08001655 free_pages -= (1 << order) - 1;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001656 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001658 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 min -= min / 4;
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07001660#ifdef CONFIG_CMA
1661 /* If allocation can't use CMA areas don't use free CMA pages */
1662 if (!(alloc_flags & ALLOC_CMA))
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07001663 free_cma = zone_page_state(z, NR_FREE_CMA_PAGES);
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07001664#endif
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07001665
1666 if (free_pages - free_cma <= min + lowmem_reserve)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001667 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 for (o = 0; o < order; o++) {
1669 /* At the next order, this order's pages become unavailable */
1670 free_pages -= z->free_area[o].nr_free << o;
1671
1672 /* Require fewer higher order pages to be free */
1673 min >>= 1;
1674
1675 if (free_pages <= min)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001676 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 }
Mel Gorman88f5acf2011-01-13 15:45:41 -08001678 return true;
1679}
1680
1681bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1682 int classzone_idx, int alloc_flags)
1683{
1684 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1685 zone_page_state(z, NR_FREE_PAGES));
1686}
1687
1688bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
1689 int classzone_idx, int alloc_flags)
1690{
1691 long free_pages = zone_page_state(z, NR_FREE_PAGES);
1692
1693 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1694 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1695
1696 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1697 free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698}
1699
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001700#ifdef CONFIG_NUMA
1701/*
1702 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1703 * skip over zones that are not allowed by the cpuset, or that have
1704 * been recently (in last second) found to be nearly full. See further
1705 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001706 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001707 *
Zhi Yong Wua1aeb652013-11-12 15:08:29 -08001708 * If the zonelist cache is present in the passed zonelist, then
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001709 * returns a pointer to the allowed node mask (either the current
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08001710 * tasks mems_allowed, or node_states[N_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001711 *
1712 * If the zonelist cache is not available for this zonelist, does
1713 * nothing and returns NULL.
1714 *
1715 * If the fullzones BITMAP in the zonelist cache is stale (more than
1716 * a second since last zap'd) then we zap it out (clear its bits.)
1717 *
1718 * We hold off even calling zlc_setup, until after we've checked the
1719 * first zone in the zonelist, on the theory that most allocations will
1720 * be satisfied from that first zone, so best to examine that zone as
1721 * quickly as we can.
1722 */
1723static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1724{
1725 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1726 nodemask_t *allowednodes; /* zonelist_cache approximation */
1727
1728 zlc = zonelist->zlcache_ptr;
1729 if (!zlc)
1730 return NULL;
1731
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001732 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001733 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1734 zlc->last_full_zap = jiffies;
1735 }
1736
1737 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1738 &cpuset_current_mems_allowed :
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08001739 &node_states[N_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001740 return allowednodes;
1741}
1742
1743/*
1744 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1745 * if it is worth looking at further for free memory:
1746 * 1) Check that the zone isn't thought to be full (doesn't have its
1747 * bit set in the zonelist_cache fullzones BITMAP).
1748 * 2) Check that the zones node (obtained from the zonelist_cache
1749 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1750 * Return true (non-zero) if zone is worth looking at further, or
1751 * else return false (zero) if it is not.
1752 *
1753 * This check -ignores- the distinction between various watermarks,
1754 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1755 * found to be full for any variation of these watermarks, it will
1756 * be considered full for up to one second by all requests, unless
1757 * we are so low on memory on all allowed nodes that we are forced
1758 * into the second scan of the zonelist.
1759 *
1760 * In the second scan we ignore this zonelist cache and exactly
1761 * apply the watermarks to all zones, even it is slower to do so.
1762 * We are low on memory in the second scan, and should leave no stone
1763 * unturned looking for a free page.
1764 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001765static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001766 nodemask_t *allowednodes)
1767{
1768 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1769 int i; /* index of *z in zonelist zones */
1770 int n; /* node that zone *z is on */
1771
1772 zlc = zonelist->zlcache_ptr;
1773 if (!zlc)
1774 return 1;
1775
Mel Gormandd1a2392008-04-28 02:12:17 -07001776 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001777 n = zlc->z_to_n[i];
1778
1779 /* This zone is worth trying if it is allowed but not full */
1780 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1781}
1782
1783/*
1784 * Given 'z' scanning a zonelist, set the corresponding bit in
1785 * zlc->fullzones, so that subsequent attempts to allocate a page
1786 * from that zone don't waste time re-examining it.
1787 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001788static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001789{
1790 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1791 int i; /* index of *z in zonelist zones */
1792
1793 zlc = zonelist->zlcache_ptr;
1794 if (!zlc)
1795 return;
1796
Mel Gormandd1a2392008-04-28 02:12:17 -07001797 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001798
1799 set_bit(i, zlc->fullzones);
1800}
1801
Mel Gorman76d3fbf2011-07-25 17:12:30 -07001802/*
1803 * clear all zones full, called after direct reclaim makes progress so that
1804 * a zone that was recently full is not skipped over for up to a second
1805 */
1806static void zlc_clear_zones_full(struct zonelist *zonelist)
1807{
1808 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1809
1810 zlc = zonelist->zlcache_ptr;
1811 if (!zlc)
1812 return;
1813
1814 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1815}
1816
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001817static bool zone_local(struct zone *local_zone, struct zone *zone)
1818{
Johannes Weinerfff40682013-12-20 14:54:12 +00001819 return local_zone->node == zone->node;
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001820}
1821
David Rientjes957f8222012-10-08 16:33:24 -07001822static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
1823{
1824 return node_isset(local_zone->node, zone->zone_pgdat->reclaim_nodes);
1825}
1826
1827static void __paginginit init_zone_allows_reclaim(int nid)
1828{
1829 int i;
1830
1831 for_each_online_node(i)
David Rientjes6b187d022012-10-25 13:38:08 -07001832 if (node_distance(nid, i) <= RECLAIM_DISTANCE)
David Rientjes957f8222012-10-08 16:33:24 -07001833 node_set(i, NODE_DATA(nid)->reclaim_nodes);
David Rientjes6b187d022012-10-25 13:38:08 -07001834 else
David Rientjes957f8222012-10-08 16:33:24 -07001835 zone_reclaim_mode = 1;
David Rientjes957f8222012-10-08 16:33:24 -07001836}
1837
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001838#else /* CONFIG_NUMA */
1839
1840static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1841{
1842 return NULL;
1843}
1844
Mel Gormandd1a2392008-04-28 02:12:17 -07001845static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001846 nodemask_t *allowednodes)
1847{
1848 return 1;
1849}
1850
Mel Gormandd1a2392008-04-28 02:12:17 -07001851static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001852{
1853}
Mel Gorman76d3fbf2011-07-25 17:12:30 -07001854
1855static void zlc_clear_zones_full(struct zonelist *zonelist)
1856{
1857}
David Rientjes957f8222012-10-08 16:33:24 -07001858
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001859static bool zone_local(struct zone *local_zone, struct zone *zone)
1860{
1861 return true;
1862}
1863
David Rientjes957f8222012-10-08 16:33:24 -07001864static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
1865{
1866 return true;
1867}
1868
1869static inline void init_zone_allows_reclaim(int nid)
1870{
1871}
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001872#endif /* CONFIG_NUMA */
1873
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001874/*
Paul Jackson0798e512006-12-06 20:31:38 -08001875 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001876 * a page.
1877 */
1878static struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001879get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
Mel Gorman5117f452009-06-16 15:31:59 -07001880 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
Mel Gorman3dd28262009-06-16 15:32:00 -07001881 struct zone *preferred_zone, int migratetype)
Martin Hicks753ee722005-06-21 17:14:41 -07001882{
Mel Gormandd1a2392008-04-28 02:12:17 -07001883 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001884 struct page *page = NULL;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001885 int classzone_idx;
Mel Gorman5117f452009-06-16 15:31:59 -07001886 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001887 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1888 int zlc_active = 0; /* set if using zonelist_cache */
1889 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gorman54a6eb52008-04-28 02:12:16 -07001890
Mel Gorman19770b32008-04-28 02:12:18 -07001891 classzone_idx = zone_idx(preferred_zone);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001892zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001893 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001894 * Scan zonelist, looking for a zone with enough free.
SeungHun Lee3b11f0a2013-09-11 14:22:23 -07001895 * See also __cpuset_node_allowed_softwall() comment in kernel/cpuset.c.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001896 */
Mel Gorman19770b32008-04-28 02:12:18 -07001897 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1898 high_zoneidx, nodemask) {
Johannes Weinere085dbc2013-09-11 14:20:46 -07001899 unsigned long mark;
1900
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08001901 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001902 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1903 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001904 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001905 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Mel Gormancd38b112011-07-25 17:12:29 -07001906 continue;
Johannes Weinere085dbc2013-09-11 14:20:46 -07001907 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
Joonsoo Kime66f0972013-09-11 14:21:26 -07001908 if (unlikely(alloc_flags & ALLOC_NO_WATERMARKS))
Johannes Weinere085dbc2013-09-11 14:20:46 -07001909 goto try_this_zone;
Johannes Weinera756cf52012-01-10 15:07:49 -08001910 /*
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001911 * Distribute pages in proportion to the individual
1912 * zone size to ensure fair page aging. The zone a
1913 * page was allocated in should have no effect on the
1914 * time the page has in memory before being reclaimed.
1915 *
Johannes Weinerfff40682013-12-20 14:54:12 +00001916 * Try to stay in local zones in the fastpath. If
1917 * that fails, the slowpath is entered, which will do
1918 * another pass starting with the local zones, but
1919 * ultimately fall back to remote zones that do not
1920 * partake in the fairness round-robin cycle of this
1921 * zonelist.
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001922 */
Mel Gorman8798cee2013-12-20 14:54:11 +00001923 if (alloc_flags & ALLOC_WMARK_LOW) {
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001924 if (zone_page_state(zone, NR_ALLOC_BATCH) <= 0)
1925 continue;
Johannes Weinerfff40682013-12-20 14:54:12 +00001926 if (!zone_local(preferred_zone, zone))
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001927 continue;
1928 }
1929 /*
Johannes Weinera756cf52012-01-10 15:07:49 -08001930 * When allocating a page cache page for writing, we
1931 * want to get it from a zone that is within its dirty
1932 * limit, such that no single zone holds more than its
1933 * proportional share of globally allowed dirty pages.
1934 * The dirty limits take into account the zone's
1935 * lowmem reserves and high watermark so that kswapd
1936 * should be able to balance it without having to
1937 * write pages from its LRU list.
1938 *
1939 * This may look like it could increase pressure on
1940 * lower zones by failing allocations in higher zones
1941 * before they are full. But the pages that do spill
1942 * over are limited as the lower zones are protected
1943 * by this very same mechanism. It should not become
1944 * a practical burden to them.
1945 *
1946 * XXX: For now, allow allocations to potentially
1947 * exceed the per-zone dirty limit in the slowpath
1948 * (ALLOC_WMARK_LOW unset) before going into reclaim,
1949 * which is important when on a NUMA setup the allowed
1950 * zones are together not big enough to reach the
1951 * global limit. The proper fix for these situations
1952 * will require awareness of zones in the
1953 * dirty-throttling and the flusher threads.
1954 */
1955 if ((alloc_flags & ALLOC_WMARK_LOW) &&
1956 (gfp_mask & __GFP_WRITE) && !zone_dirty_ok(zone))
1957 goto this_zone_full;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001958
Johannes Weinere085dbc2013-09-11 14:20:46 -07001959 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
1960 if (!zone_watermark_ok(zone, order, mark,
1961 classzone_idx, alloc_flags)) {
Mel Gormanfa5e0842009-06-16 15:33:22 -07001962 int ret;
1963
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08001964 if (IS_ENABLED(CONFIG_NUMA) &&
1965 !did_zlc_setup && nr_online_nodes > 1) {
Mel Gormancd38b112011-07-25 17:12:29 -07001966 /*
1967 * we do zlc_setup if there are multiple nodes
1968 * and before considering the first zone allowed
1969 * by the cpuset.
1970 */
1971 allowednodes = zlc_setup(zonelist, alloc_flags);
1972 zlc_active = 1;
1973 did_zlc_setup = 1;
1974 }
1975
David Rientjes957f8222012-10-08 16:33:24 -07001976 if (zone_reclaim_mode == 0 ||
1977 !zone_allows_reclaim(preferred_zone, zone))
Mel Gormanfa5e0842009-06-16 15:33:22 -07001978 goto this_zone_full;
1979
Mel Gormancd38b112011-07-25 17:12:29 -07001980 /*
1981 * As we may have just activated ZLC, check if the first
1982 * eligible zone has failed zone_reclaim recently.
1983 */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08001984 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
Mel Gormancd38b112011-07-25 17:12:29 -07001985 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1986 continue;
1987
Mel Gormanfa5e0842009-06-16 15:33:22 -07001988 ret = zone_reclaim(zone, gfp_mask, order);
1989 switch (ret) {
1990 case ZONE_RECLAIM_NOSCAN:
1991 /* did not scan */
Mel Gormancd38b112011-07-25 17:12:29 -07001992 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001993 case ZONE_RECLAIM_FULL:
1994 /* scanned but unreclaimable */
Mel Gormancd38b112011-07-25 17:12:29 -07001995 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001996 default:
1997 /* did we reclaim enough */
Mel Gormanfed27192013-04-29 15:07:57 -07001998 if (zone_watermark_ok(zone, order, mark,
Mel Gormanfa5e0842009-06-16 15:33:22 -07001999 classzone_idx, alloc_flags))
Mel Gormanfed27192013-04-29 15:07:57 -07002000 goto try_this_zone;
2001
2002 /*
2003 * Failed to reclaim enough to meet watermark.
2004 * Only mark the zone full if checking the min
2005 * watermark or if we failed to reclaim just
2006 * 1<<order pages or else the page allocator
2007 * fastpath will prematurely mark zones full
2008 * when the watermark is between the low and
2009 * min watermarks.
2010 */
2011 if (((alloc_flags & ALLOC_WMARK_MASK) == ALLOC_WMARK_MIN) ||
2012 ret == ZONE_RECLAIM_SOME)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002013 goto this_zone_full;
Mel Gormanfed27192013-04-29 15:07:57 -07002014
2015 continue;
Paul Jackson0798e512006-12-06 20:31:38 -08002016 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002017 }
2018
Mel Gormanfa5e0842009-06-16 15:33:22 -07002019try_this_zone:
Mel Gorman3dd28262009-06-16 15:32:00 -07002020 page = buffered_rmqueue(preferred_zone, zone, order,
2021 gfp_mask, migratetype);
Paul Jackson0798e512006-12-06 20:31:38 -08002022 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002023 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002024this_zone_full:
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002025 if (IS_ENABLED(CONFIG_NUMA))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002026 zlc_mark_zone_full(zonelist, z);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002027 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002028
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002029 if (unlikely(IS_ENABLED(CONFIG_NUMA) && page == NULL && zlc_active)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002030 /* Disable zlc cache for second zonelist scan */
2031 zlc_active = 0;
2032 goto zonelist_scan;
2033 }
Alex Shib1211862012-08-21 16:16:08 -07002034
2035 if (page)
2036 /*
2037 * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was
2038 * necessary to allocate the page. The expectation is
2039 * that the caller is taking steps that will free more
2040 * memory. The caller should avoid the page being used
2041 * for !PFMEMALLOC purposes.
2042 */
2043 page->pfmemalloc = !!(alloc_flags & ALLOC_NO_WATERMARKS);
2044
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002045 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07002046}
2047
David Rientjes29423e772011-03-22 16:30:47 -07002048/*
2049 * Large machines with many possible nodes should not always dump per-node
2050 * meminfo in irq context.
2051 */
2052static inline bool should_suppress_show_mem(void)
2053{
2054 bool ret = false;
2055
2056#if NODES_SHIFT > 8
2057 ret = in_interrupt();
2058#endif
2059 return ret;
2060}
2061
Dave Hansena238ab52011-05-24 17:12:16 -07002062static DEFINE_RATELIMIT_STATE(nopage_rs,
2063 DEFAULT_RATELIMIT_INTERVAL,
2064 DEFAULT_RATELIMIT_BURST);
2065
2066void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
2067{
Dave Hansena238ab52011-05-24 17:12:16 -07002068 unsigned int filter = SHOW_MEM_FILTER_NODES;
2069
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -08002070 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2071 debug_guardpage_minorder() > 0)
Dave Hansena238ab52011-05-24 17:12:16 -07002072 return;
2073
2074 /*
David Rientjes4b59e6c2013-04-29 15:06:11 -07002075 * Walking all memory to count page types is very expensive and should
2076 * be inhibited in non-blockable contexts.
2077 */
2078 if (!(gfp_mask & __GFP_WAIT))
2079 filter |= SHOW_MEM_FILTER_PAGE_COUNT;
2080
2081 /*
Dave Hansena238ab52011-05-24 17:12:16 -07002082 * This documents exceptions given to allocations in certain
2083 * contexts that are allowed to allocate outside current's set
2084 * of allowed nodes.
2085 */
2086 if (!(gfp_mask & __GFP_NOMEMALLOC))
2087 if (test_thread_flag(TIF_MEMDIE) ||
2088 (current->flags & (PF_MEMALLOC | PF_EXITING)))
2089 filter &= ~SHOW_MEM_FILTER_NODES;
2090 if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
2091 filter &= ~SHOW_MEM_FILTER_NODES;
2092
2093 if (fmt) {
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002094 struct va_format vaf;
2095 va_list args;
2096
Dave Hansena238ab52011-05-24 17:12:16 -07002097 va_start(args, fmt);
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002098
2099 vaf.fmt = fmt;
2100 vaf.va = &args;
2101
2102 pr_warn("%pV", &vaf);
2103
Dave Hansena238ab52011-05-24 17:12:16 -07002104 va_end(args);
2105 }
2106
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002107 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2108 current->comm, order, gfp_mask);
Dave Hansena238ab52011-05-24 17:12:16 -07002109
2110 dump_stack();
2111 if (!should_suppress_show_mem())
2112 show_mem(filter);
2113}
2114
Mel Gorman11e33f62009-06-16 15:31:57 -07002115static inline int
2116should_alloc_retry(gfp_t gfp_mask, unsigned int order,
Mel Gormanf90ac392012-01-10 15:07:15 -08002117 unsigned long did_some_progress,
Mel Gorman11e33f62009-06-16 15:31:57 -07002118 unsigned long pages_reclaimed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119{
Mel Gorman11e33f62009-06-16 15:31:57 -07002120 /* Do not loop if specifically requested */
2121 if (gfp_mask & __GFP_NORETRY)
2122 return 0;
2123
Mel Gormanf90ac392012-01-10 15:07:15 -08002124 /* Always retry if specifically requested */
2125 if (gfp_mask & __GFP_NOFAIL)
2126 return 1;
2127
2128 /*
2129 * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim
2130 * making forward progress without invoking OOM. Suspend also disables
2131 * storage devices so kswapd will not help. Bail if we are suspending.
2132 */
2133 if (!did_some_progress && pm_suspended_storage())
2134 return 0;
2135
Mel Gorman11e33f62009-06-16 15:31:57 -07002136 /*
2137 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
2138 * means __GFP_NOFAIL, but that may not be true in other
2139 * implementations.
2140 */
2141 if (order <= PAGE_ALLOC_COSTLY_ORDER)
2142 return 1;
2143
2144 /*
2145 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
2146 * specified, then we retry until we no longer reclaim any pages
2147 * (above), or we've reclaimed an order of pages at least as
2148 * large as the allocation's order. In both cases, if the
2149 * allocation still fails, we stop retrying.
2150 */
2151 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
2152 return 1;
2153
Mel Gorman11e33f62009-06-16 15:31:57 -07002154 return 0;
2155}
2156
2157static inline struct page *
2158__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
2159 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002160 nodemask_t *nodemask, struct zone *preferred_zone,
2161 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002162{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164
Mel Gorman11e33f62009-06-16 15:31:57 -07002165 /* Acquire the OOM killer lock for the zones in zonelist */
Minchan Kimff321fe2010-08-09 17:18:57 -07002166 if (!try_set_zonelist_oom(zonelist, gfp_mask)) {
Mel Gorman11e33f62009-06-16 15:31:57 -07002167 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 return NULL;
2169 }
Jens Axboe6b1de912005-11-17 21:35:02 +01002170
Mel Gorman11e33f62009-06-16 15:31:57 -07002171 /*
2172 * Go through the zonelist yet one more time, keep very high watermark
2173 * here, this is only to catch a parallel oom killing, we must fail if
2174 * we're still under heavy pressure.
2175 */
2176 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
2177 order, zonelist, high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07002178 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07002179 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002180 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07002181 goto out;
2182
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002183 if (!(gfp_mask & __GFP_NOFAIL)) {
2184 /* The OOM killer will not help higher order allocs */
2185 if (order > PAGE_ALLOC_COSTLY_ORDER)
2186 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07002187 /* The OOM killer does not needlessly kill tasks for lowmem */
2188 if (high_zoneidx < ZONE_NORMAL)
2189 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002190 /*
2191 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
2192 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
2193 * The caller should handle page allocation failure by itself if
2194 * it specifies __GFP_THISNODE.
2195 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
2196 */
2197 if (gfp_mask & __GFP_THISNODE)
2198 goto out;
2199 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002200 /* Exhausted what can be done so it's blamo time */
David Rientjes08ab9b12012-03-21 16:34:04 -07002201 out_of_memory(zonelist, gfp_mask, order, nodemask, false);
Mel Gorman11e33f62009-06-16 15:31:57 -07002202
2203out:
2204 clear_zonelist_oom(zonelist, gfp_mask);
2205 return page;
2206}
2207
Mel Gorman56de7262010-05-24 14:32:30 -07002208#ifdef CONFIG_COMPACTION
2209/* Try memory compaction for high-order allocations before reclaim */
2210static struct page *
2211__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2212 struct zonelist *zonelist, enum zone_type high_zoneidx,
2213 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman66199712012-01-12 17:19:41 -08002214 int migratetype, bool sync_migration,
Mel Gormanc67fe372012-08-21 16:16:17 -07002215 bool *contended_compaction, bool *deferred_compaction,
Mel Gorman66199712012-01-12 17:19:41 -08002216 unsigned long *did_some_progress)
Mel Gorman56de7262010-05-24 14:32:30 -07002217{
Mel Gorman66199712012-01-12 17:19:41 -08002218 if (!order)
Mel Gorman56de7262010-05-24 14:32:30 -07002219 return NULL;
2220
Rik van Rielaff62242012-03-21 16:33:52 -07002221 if (compaction_deferred(preferred_zone, order)) {
Mel Gorman66199712012-01-12 17:19:41 -08002222 *deferred_compaction = true;
2223 return NULL;
2224 }
2225
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002226 current->flags |= PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07002227 *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
Mel Gormanc67fe372012-08-21 16:16:17 -07002228 nodemask, sync_migration,
Mel Gorman8fb74b92013-01-11 14:32:16 -08002229 contended_compaction);
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002230 current->flags &= ~PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07002231
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07002232 if (*did_some_progress != COMPACT_SKIPPED) {
Mel Gorman8fb74b92013-01-11 14:32:16 -08002233 struct page *page;
2234
Mel Gorman56de7262010-05-24 14:32:30 -07002235 /* Page migration frees to the PCP lists but we want merging */
2236 drain_pages(get_cpu());
2237 put_cpu();
2238
2239 page = get_page_from_freelist(gfp_mask, nodemask,
2240 order, zonelist, high_zoneidx,
Mel Gormancfd19c52012-07-31 16:44:10 -07002241 alloc_flags & ~ALLOC_NO_WATERMARKS,
2242 preferred_zone, migratetype);
Mel Gorman56de7262010-05-24 14:32:30 -07002243 if (page) {
Mel Gorman62997022012-10-08 16:32:47 -07002244 preferred_zone->compact_blockskip_flush = false;
Mel Gorman4f92e252010-05-24 14:32:32 -07002245 preferred_zone->compact_considered = 0;
2246 preferred_zone->compact_defer_shift = 0;
Rik van Rielaff62242012-03-21 16:33:52 -07002247 if (order >= preferred_zone->compact_order_failed)
2248 preferred_zone->compact_order_failed = order + 1;
Mel Gorman56de7262010-05-24 14:32:30 -07002249 count_vm_event(COMPACTSUCCESS);
2250 return page;
2251 }
2252
2253 /*
2254 * It's bad if compaction run occurs and fails.
2255 * The most likely reason is that pages exist,
2256 * but not enough to satisfy watermarks.
2257 */
2258 count_vm_event(COMPACTFAIL);
Mel Gorman66199712012-01-12 17:19:41 -08002259
2260 /*
2261 * As async compaction considers a subset of pageblocks, only
2262 * defer if the failure was a sync compaction failure.
2263 */
2264 if (sync_migration)
Rik van Rielaff62242012-03-21 16:33:52 -07002265 defer_compaction(preferred_zone, order);
Mel Gorman56de7262010-05-24 14:32:30 -07002266
2267 cond_resched();
2268 }
2269
2270 return NULL;
2271}
2272#else
2273static inline struct page *
2274__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2275 struct zonelist *zonelist, enum zone_type high_zoneidx,
2276 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman66199712012-01-12 17:19:41 -08002277 int migratetype, bool sync_migration,
Mel Gormanc67fe372012-08-21 16:16:17 -07002278 bool *contended_compaction, bool *deferred_compaction,
Mel Gorman66199712012-01-12 17:19:41 -08002279 unsigned long *did_some_progress)
Mel Gorman56de7262010-05-24 14:32:30 -07002280{
2281 return NULL;
2282}
2283#endif /* CONFIG_COMPACTION */
2284
Marek Szyprowskibba90712012-01-25 12:09:52 +01002285/* Perform direct synchronous page reclaim */
2286static int
2287__perform_reclaim(gfp_t gfp_mask, unsigned int order, struct zonelist *zonelist,
2288 nodemask_t *nodemask)
Mel Gorman11e33f62009-06-16 15:31:57 -07002289{
Mel Gorman11e33f62009-06-16 15:31:57 -07002290 struct reclaim_state reclaim_state;
Marek Szyprowskibba90712012-01-25 12:09:52 +01002291 int progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07002292
2293 cond_resched();
2294
2295 /* We now go into synchronous reclaim */
2296 cpuset_memory_pressure_bump();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002297 current->flags |= PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002298 lockdep_set_current_reclaim_state(gfp_mask);
2299 reclaim_state.reclaimed_slab = 0;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002300 current->reclaim_state = &reclaim_state;
Mel Gorman11e33f62009-06-16 15:31:57 -07002301
Marek Szyprowskibba90712012-01-25 12:09:52 +01002302 progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002303
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002304 current->reclaim_state = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002305 lockdep_clear_current_reclaim_state();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002306 current->flags &= ~PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002307
2308 cond_resched();
2309
Marek Szyprowskibba90712012-01-25 12:09:52 +01002310 return progress;
2311}
2312
2313/* The really slow allocator path where we enter direct reclaim */
2314static inline struct page *
2315__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
2316 struct zonelist *zonelist, enum zone_type high_zoneidx,
2317 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
2318 int migratetype, unsigned long *did_some_progress)
2319{
2320 struct page *page = NULL;
2321 bool drained = false;
2322
2323 *did_some_progress = __perform_reclaim(gfp_mask, order, zonelist,
2324 nodemask);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002325 if (unlikely(!(*did_some_progress)))
2326 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002327
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002328 /* After successful reclaim, reconsider all zones for allocation */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002329 if (IS_ENABLED(CONFIG_NUMA))
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002330 zlc_clear_zones_full(zonelist);
2331
Mel Gorman9ee493c2010-09-09 16:38:18 -07002332retry:
2333 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002334 zonelist, high_zoneidx,
Mel Gormancfd19c52012-07-31 16:44:10 -07002335 alloc_flags & ~ALLOC_NO_WATERMARKS,
2336 preferred_zone, migratetype);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002337
2338 /*
2339 * If an allocation failed after direct reclaim, it could be because
2340 * pages are pinned on the per-cpu lists. Drain them and try again
2341 */
2342 if (!page && !drained) {
2343 drain_all_pages();
2344 drained = true;
2345 goto retry;
2346 }
2347
Mel Gorman11e33f62009-06-16 15:31:57 -07002348 return page;
2349}
2350
Mel Gorman11e33f62009-06-16 15:31:57 -07002351/*
2352 * This is called in the allocator slow-path if the allocation request is of
2353 * sufficient urgency to ignore watermarks and take other desperate measures
2354 */
2355static inline struct page *
2356__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
2357 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002358 nodemask_t *nodemask, struct zone *preferred_zone,
2359 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002360{
2361 struct page *page;
2362
2363 do {
2364 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002365 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
Mel Gorman3dd28262009-06-16 15:32:00 -07002366 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002367
2368 if (!page && gfp_mask & __GFP_NOFAIL)
Mel Gorman0e093d992010-10-26 14:21:45 -07002369 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07002370 } while (!page && (gfp_mask & __GFP_NOFAIL));
2371
2372 return page;
2373}
2374
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002375static void prepare_slowpath(gfp_t gfp_mask, unsigned int order,
2376 struct zonelist *zonelist,
2377 enum zone_type high_zoneidx,
2378 struct zone *preferred_zone)
Mel Gorman11e33f62009-06-16 15:31:57 -07002379{
2380 struct zoneref *z;
2381 struct zone *zone;
2382
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002383 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
2384 if (!(gfp_mask & __GFP_NO_KSWAPD))
2385 wakeup_kswapd(zone, order, zone_idx(preferred_zone));
2386 /*
2387 * Only reset the batches of zones that were actually
2388 * considered in the fast path, we don't want to
2389 * thrash fairness information for zones that are not
2390 * actually part of this zonelist's round-robin cycle.
2391 */
Johannes Weinerfff40682013-12-20 14:54:12 +00002392 if (!zone_local(preferred_zone, zone))
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002393 continue;
2394 mod_zone_page_state(zone, NR_ALLOC_BATCH,
2395 high_wmark_pages(zone) -
2396 low_wmark_pages(zone) -
2397 zone_page_state(zone, NR_ALLOC_BATCH));
2398 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002399}
2400
Peter Zijlstra341ce062009-06-16 15:32:02 -07002401static inline int
2402gfp_to_alloc_flags(gfp_t gfp_mask)
2403{
Peter Zijlstra341ce062009-06-16 15:32:02 -07002404 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
2405 const gfp_t wait = gfp_mask & __GFP_WAIT;
2406
Mel Gormana56f57f2009-06-16 15:32:02 -07002407 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07002408 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07002409
Peter Zijlstra341ce062009-06-16 15:32:02 -07002410 /*
2411 * The caller may dip into page reserves a bit more if the caller
2412 * cannot run direct reclaim, or if the caller has realtime scheduling
2413 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2414 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
2415 */
Namhyung Kime6223a32010-10-26 14:21:59 -07002416 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07002417
2418 if (!wait) {
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002419 /*
2420 * Not worth trying to allocate harder for
2421 * __GFP_NOMEMALLOC even if it can't schedule.
2422 */
2423 if (!(gfp_mask & __GFP_NOMEMALLOC))
2424 alloc_flags |= ALLOC_HARDER;
Peter Zijlstra341ce062009-06-16 15:32:02 -07002425 /*
2426 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
2427 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
2428 */
2429 alloc_flags &= ~ALLOC_CPUSET;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002430 } else if (unlikely(rt_task(current)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07002431 alloc_flags |= ALLOC_HARDER;
2432
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002433 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2434 if (gfp_mask & __GFP_MEMALLOC)
2435 alloc_flags |= ALLOC_NO_WATERMARKS;
Mel Gorman907aed42012-07-31 16:44:07 -07002436 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
2437 alloc_flags |= ALLOC_NO_WATERMARKS;
2438 else if (!in_interrupt() &&
2439 ((current->flags & PF_MEMALLOC) ||
2440 unlikely(test_thread_flag(TIF_MEMDIE))))
Peter Zijlstra341ce062009-06-16 15:32:02 -07002441 alloc_flags |= ALLOC_NO_WATERMARKS;
2442 }
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002443#ifdef CONFIG_CMA
2444 if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2445 alloc_flags |= ALLOC_CMA;
2446#endif
Peter Zijlstra341ce062009-06-16 15:32:02 -07002447 return alloc_flags;
2448}
2449
Mel Gorman072bb0a2012-07-31 16:43:58 -07002450bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
2451{
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002452 return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
Mel Gorman072bb0a2012-07-31 16:43:58 -07002453}
2454
Mel Gorman11e33f62009-06-16 15:31:57 -07002455static inline struct page *
2456__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
2457 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002458 nodemask_t *nodemask, struct zone *preferred_zone,
2459 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002460{
2461 const gfp_t wait = gfp_mask & __GFP_WAIT;
2462 struct page *page = NULL;
2463 int alloc_flags;
2464 unsigned long pages_reclaimed = 0;
2465 unsigned long did_some_progress;
Mel Gorman77f1fe62011-01-13 15:45:57 -08002466 bool sync_migration = false;
Mel Gorman66199712012-01-12 17:19:41 -08002467 bool deferred_compaction = false;
Mel Gormanc67fe372012-08-21 16:16:17 -07002468 bool contended_compaction = false;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002469
Christoph Lameter952f3b52006-12-06 20:33:26 -08002470 /*
Mel Gorman72807a72009-06-16 15:32:18 -07002471 * In the slowpath, we sanity check order to avoid ever trying to
2472 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2473 * be using allocators in order of preference for an area that is
2474 * too large.
2475 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07002476 if (order >= MAX_ORDER) {
2477 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07002478 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07002479 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002480
Christoph Lameter952f3b52006-12-06 20:33:26 -08002481 /*
2482 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2483 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2484 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2485 * using a larger set of nodes after it has established that the
2486 * allowed per node queues are empty and that nodes are
2487 * over allocated.
2488 */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002489 if (IS_ENABLED(CONFIG_NUMA) &&
2490 (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
Christoph Lameter952f3b52006-12-06 20:33:26 -08002491 goto nopage;
2492
Mel Gormancc4a6852009-11-11 14:26:14 -08002493restart:
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002494 prepare_slowpath(gfp_mask, order, zonelist,
2495 high_zoneidx, preferred_zone);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002496
Paul Jackson9bf22292005-09-06 15:18:12 -07002497 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002498 * OK, we're below the kswapd watermark and have kicked background
2499 * reclaim. Now things get more complex, so set up alloc_flags according
2500 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07002501 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002502 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503
David Rientjesf33261d2011-01-25 15:07:20 -08002504 /*
2505 * Find the true preferred zone if the allocation is unconstrained by
2506 * cpusets.
2507 */
2508 if (!(alloc_flags & ALLOC_CPUSET) && !nodemask)
2509 first_zones_zonelist(zonelist, high_zoneidx, NULL,
2510 &preferred_zone);
2511
Andrew Barrycfa54a02011-05-24 17:12:52 -07002512rebalance:
Peter Zijlstra341ce062009-06-16 15:32:02 -07002513 /* This is the last chance, in general, before the goto nopage. */
Mel Gorman19770b32008-04-28 02:12:18 -07002514 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
Peter Zijlstra341ce062009-06-16 15:32:02 -07002515 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
2516 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002517 if (page)
2518 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519
Mel Gorman11e33f62009-06-16 15:31:57 -07002520 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002521 if (alloc_flags & ALLOC_NO_WATERMARKS) {
Mel Gorman183f6372012-07-31 16:44:12 -07002522 /*
2523 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
2524 * the allocation is high priority and these type of
2525 * allocations are system rather than user orientated
2526 */
2527 zonelist = node_zonelist(numa_node_id(), gfp_mask);
2528
Peter Zijlstra341ce062009-06-16 15:32:02 -07002529 page = __alloc_pages_high_priority(gfp_mask, order,
2530 zonelist, high_zoneidx, nodemask,
2531 preferred_zone, migratetype);
Mel Gormancfd19c52012-07-31 16:44:10 -07002532 if (page) {
Peter Zijlstra341ce062009-06-16 15:32:02 -07002533 goto got_pg;
Mel Gormancfd19c52012-07-31 16:44:10 -07002534 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535 }
2536
2537 /* Atomic allocations - we can't balance anything */
2538 if (!wait)
2539 goto nopage;
2540
Peter Zijlstra341ce062009-06-16 15:32:02 -07002541 /* Avoid recursion of direct reclaim */
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002542 if (current->flags & PF_MEMALLOC)
Peter Zijlstra341ce062009-06-16 15:32:02 -07002543 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544
David Rientjes6583bb62009-07-29 15:02:06 -07002545 /* Avoid allocations with no watermarks from looping endlessly */
2546 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2547 goto nopage;
2548
Mel Gorman77f1fe62011-01-13 15:45:57 -08002549 /*
2550 * Try direct compaction. The first pass is asynchronous. Subsequent
2551 * attempts after direct reclaim are synchronous
2552 */
Mel Gorman56de7262010-05-24 14:32:30 -07002553 page = __alloc_pages_direct_compact(gfp_mask, order,
2554 zonelist, high_zoneidx,
2555 nodemask,
2556 alloc_flags, preferred_zone,
Mel Gorman66199712012-01-12 17:19:41 -08002557 migratetype, sync_migration,
Mel Gormanc67fe372012-08-21 16:16:17 -07002558 &contended_compaction,
Mel Gorman66199712012-01-12 17:19:41 -08002559 &deferred_compaction,
2560 &did_some_progress);
Mel Gorman56de7262010-05-24 14:32:30 -07002561 if (page)
2562 goto got_pg;
Andrea Arcangelic6a140b2011-05-24 17:11:38 -07002563 sync_migration = true;
Mel Gorman56de7262010-05-24 14:32:30 -07002564
Linus Torvalds31f8d422012-12-10 10:47:45 -08002565 /*
2566 * If compaction is deferred for high-order allocations, it is because
2567 * sync compaction recently failed. In this is the case and the caller
2568 * requested a movable allocation that does not heavily disrupt the
2569 * system then fail the allocation instead of entering direct reclaim.
2570 */
2571 if ((deferred_compaction || contended_compaction) &&
Linus Torvaldscaf49192012-12-10 10:51:16 -08002572 (gfp_mask & __GFP_NO_KSWAPD))
Linus Torvalds31f8d422012-12-10 10:47:45 -08002573 goto nopage;
Mel Gorman66199712012-01-12 17:19:41 -08002574
Mel Gorman11e33f62009-06-16 15:31:57 -07002575 /* Try direct reclaim and then allocating */
2576 page = __alloc_pages_direct_reclaim(gfp_mask, order,
2577 zonelist, high_zoneidx,
2578 nodemask,
Mel Gorman5117f452009-06-16 15:31:59 -07002579 alloc_flags, preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07002580 migratetype, &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07002581 if (page)
2582 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 /*
Mel Gorman11e33f62009-06-16 15:31:57 -07002585 * If we failed to make any progress reclaiming, then we are
2586 * running out of options and have to consider going OOM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587 */
Mel Gorman11e33f62009-06-16 15:31:57 -07002588 if (!did_some_progress) {
Qiang Huangb9921ec2013-11-12 15:07:22 -08002589 if (oom_gfp_allowed(gfp_mask)) {
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -07002590 if (oom_killer_disabled)
2591 goto nopage;
David Rientjes29fd66d2012-03-28 14:42:41 -07002592 /* Coredumps can quickly deplete all memory reserves */
2593 if ((current->flags & PF_DUMPCORE) &&
2594 !(gfp_mask & __GFP_NOFAIL))
2595 goto nopage;
Mel Gorman11e33f62009-06-16 15:31:57 -07002596 page = __alloc_pages_may_oom(gfp_mask, order,
2597 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002598 nodemask, preferred_zone,
2599 migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002600 if (page)
2601 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602
David Rientjes03668b32010-08-09 17:18:54 -07002603 if (!(gfp_mask & __GFP_NOFAIL)) {
2604 /*
2605 * The oom killer is not called for high-order
2606 * allocations that may fail, so if no progress
2607 * is being made, there are no other options and
2608 * retrying is unlikely to help.
2609 */
2610 if (order > PAGE_ALLOC_COSTLY_ORDER)
2611 goto nopage;
2612 /*
2613 * The oom killer is not called for lowmem
2614 * allocations to prevent needlessly killing
2615 * innocent tasks.
2616 */
2617 if (high_zoneidx < ZONE_NORMAL)
2618 goto nopage;
2619 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621 goto restart;
2622 }
2623 }
2624
Mel Gorman11e33f62009-06-16 15:31:57 -07002625 /* Check if we should retry the allocation */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07002626 pages_reclaimed += did_some_progress;
Mel Gormanf90ac392012-01-10 15:07:15 -08002627 if (should_alloc_retry(gfp_mask, order, did_some_progress,
2628 pages_reclaimed)) {
Mel Gorman11e33f62009-06-16 15:31:57 -07002629 /* Wait for some write requests to complete then retry */
Mel Gorman0e093d992010-10-26 14:21:45 -07002630 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631 goto rebalance;
Mel Gorman3e7d3442011-01-13 15:45:56 -08002632 } else {
2633 /*
2634 * High-order allocations do not necessarily loop after
2635 * direct reclaim and reclaim/compaction depends on compaction
2636 * being called after reclaim so call directly if necessary
2637 */
2638 page = __alloc_pages_direct_compact(gfp_mask, order,
2639 zonelist, high_zoneidx,
2640 nodemask,
2641 alloc_flags, preferred_zone,
Mel Gorman66199712012-01-12 17:19:41 -08002642 migratetype, sync_migration,
Mel Gormanc67fe372012-08-21 16:16:17 -07002643 &contended_compaction,
Mel Gorman66199712012-01-12 17:19:41 -08002644 &deferred_compaction,
2645 &did_some_progress);
Mel Gorman3e7d3442011-01-13 15:45:56 -08002646 if (page)
2647 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 }
2649
2650nopage:
Dave Hansena238ab52011-05-24 17:12:16 -07002651 warn_alloc_failed(gfp_mask, order, NULL);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002652 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653got_pg:
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002654 if (kmemcheck_enabled)
2655 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002656
Mel Gorman072bb0a2012-07-31 16:43:58 -07002657 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658}
Mel Gorman11e33f62009-06-16 15:31:57 -07002659
2660/*
2661 * This is the 'heart' of the zoned buddy allocator.
2662 */
2663struct page *
2664__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2665 struct zonelist *zonelist, nodemask_t *nodemask)
2666{
2667 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Mel Gorman5117f452009-06-16 15:31:59 -07002668 struct zone *preferred_zone;
Mel Gormancc9a6c82012-03-21 16:34:11 -07002669 struct page *page = NULL;
Mel Gorman3dd28262009-06-16 15:32:00 -07002670 int migratetype = allocflags_to_migratetype(gfp_mask);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002671 unsigned int cpuset_mems_cookie;
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002672 int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET;
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002673 struct mem_cgroup *memcg = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002674
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10002675 gfp_mask &= gfp_allowed_mask;
2676
Mel Gorman11e33f62009-06-16 15:31:57 -07002677 lockdep_trace_alloc(gfp_mask);
2678
2679 might_sleep_if(gfp_mask & __GFP_WAIT);
2680
2681 if (should_fail_alloc_page(gfp_mask, order))
2682 return NULL;
2683
2684 /*
2685 * Check the zones suitable for the gfp_mask contain at least one
2686 * valid zone. It's possible to have an empty zonelist as a result
2687 * of GFP_THISNODE and a memoryless node
2688 */
2689 if (unlikely(!zonelist->_zonerefs->zone))
2690 return NULL;
2691
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002692 /*
2693 * Will only have any effect when __GFP_KMEMCG is set. This is
2694 * verified in the (always inline) callee
2695 */
2696 if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
2697 return NULL;
2698
Mel Gormancc9a6c82012-03-21 16:34:11 -07002699retry_cpuset:
2700 cpuset_mems_cookie = get_mems_allowed();
2701
Mel Gorman5117f452009-06-16 15:31:59 -07002702 /* The preferred zone is used for statistics later */
David Rientjesf33261d2011-01-25 15:07:20 -08002703 first_zones_zonelist(zonelist, high_zoneidx,
2704 nodemask ? : &cpuset_current_mems_allowed,
2705 &preferred_zone);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002706 if (!preferred_zone)
2707 goto out;
Mel Gorman5117f452009-06-16 15:31:59 -07002708
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002709#ifdef CONFIG_CMA
2710 if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2711 alloc_flags |= ALLOC_CMA;
2712#endif
Mel Gorman5117f452009-06-16 15:31:59 -07002713 /* First allocation attempt */
Mel Gorman11e33f62009-06-16 15:31:57 -07002714 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002715 zonelist, high_zoneidx, alloc_flags,
Mel Gorman3dd28262009-06-16 15:32:00 -07002716 preferred_zone, migratetype);
Ming Lei21caf2f2013-02-22 16:34:08 -08002717 if (unlikely(!page)) {
2718 /*
2719 * Runtime PM, block IO and its error handling path
2720 * can deadlock because I/O on the device might not
2721 * complete.
2722 */
2723 gfp_mask = memalloc_noio_flags(gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002724 page = __alloc_pages_slowpath(gfp_mask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002725 zonelist, high_zoneidx, nodemask,
Mel Gorman3dd28262009-06-16 15:32:00 -07002726 preferred_zone, migratetype);
Ming Lei21caf2f2013-02-22 16:34:08 -08002727 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002728
Mel Gorman4b4f2782009-09-21 17:02:41 -07002729 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002730
2731out:
2732 /*
2733 * When updating a task's mems_allowed, it is possible to race with
2734 * parallel threads in such a way that an allocation can fail while
2735 * the mask is being updated. If a page allocation is about to fail,
2736 * check if the cpuset changed during allocation and if so, retry.
2737 */
2738 if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page))
2739 goto retry_cpuset;
2740
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002741 memcg_kmem_commit_charge(page, memcg, order);
2742
Mel Gorman11e33f62009-06-16 15:31:57 -07002743 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744}
Mel Gormand2391712009-06-16 15:31:52 -07002745EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746
2747/*
2748 * Common helper functions.
2749 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08002750unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751{
Akinobu Mita945a1112009-09-21 17:01:47 -07002752 struct page *page;
2753
2754 /*
2755 * __get_free_pages() returns a 32-bit address, which cannot represent
2756 * a highmem page
2757 */
2758 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2759
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760 page = alloc_pages(gfp_mask, order);
2761 if (!page)
2762 return 0;
2763 return (unsigned long) page_address(page);
2764}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765EXPORT_SYMBOL(__get_free_pages);
2766
Harvey Harrison920c7a52008-02-04 22:29:26 -08002767unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768{
Akinobu Mita945a1112009-09-21 17:01:47 -07002769 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771EXPORT_SYMBOL(get_zeroed_page);
2772
Harvey Harrison920c7a52008-02-04 22:29:26 -08002773void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774{
Nick Pigginb5810032005-10-29 18:16:12 -07002775 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776 if (order == 0)
Li Hongfc916682010-03-05 13:41:54 -08002777 free_hot_cold_page(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778 else
2779 __free_pages_ok(page, order);
2780 }
2781}
2782
2783EXPORT_SYMBOL(__free_pages);
2784
Harvey Harrison920c7a52008-02-04 22:29:26 -08002785void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786{
2787 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07002788 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789 __free_pages(virt_to_page((void *)addr), order);
2790 }
2791}
2792
2793EXPORT_SYMBOL(free_pages);
2794
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002795/*
2796 * __free_memcg_kmem_pages and free_memcg_kmem_pages will free
2797 * pages allocated with __GFP_KMEMCG.
2798 *
2799 * Those pages are accounted to a particular memcg, embedded in the
2800 * corresponding page_cgroup. To avoid adding a hit in the allocator to search
2801 * for that information only to find out that it is NULL for users who have no
2802 * interest in that whatsoever, we provide these functions.
2803 *
2804 * The caller knows better which flags it relies on.
2805 */
2806void __free_memcg_kmem_pages(struct page *page, unsigned int order)
2807{
2808 memcg_kmem_uncharge_pages(page, order);
2809 __free_pages(page, order);
2810}
2811
2812void free_memcg_kmem_pages(unsigned long addr, unsigned int order)
2813{
2814 if (addr != 0) {
2815 VM_BUG_ON(!virt_addr_valid((void *)addr));
2816 __free_memcg_kmem_pages(virt_to_page((void *)addr), order);
2817 }
2818}
2819
Andi Kleenee85c2e2011-05-11 15:13:34 -07002820static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
2821{
2822 if (addr) {
2823 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2824 unsigned long used = addr + PAGE_ALIGN(size);
2825
2826 split_page(virt_to_page((void *)addr), order);
2827 while (used < alloc_end) {
2828 free_page(used);
2829 used += PAGE_SIZE;
2830 }
2831 }
2832 return (void *)addr;
2833}
2834
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002835/**
2836 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2837 * @size: the number of bytes to allocate
2838 * @gfp_mask: GFP flags for the allocation
2839 *
2840 * This function is similar to alloc_pages(), except that it allocates the
2841 * minimum number of pages to satisfy the request. alloc_pages() can only
2842 * allocate memory in power-of-two pages.
2843 *
2844 * This function is also limited by MAX_ORDER.
2845 *
2846 * Memory allocated by this function must be released by free_pages_exact().
2847 */
2848void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2849{
2850 unsigned int order = get_order(size);
2851 unsigned long addr;
2852
2853 addr = __get_free_pages(gfp_mask, order);
Andi Kleenee85c2e2011-05-11 15:13:34 -07002854 return make_alloc_exact(addr, order, size);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002855}
2856EXPORT_SYMBOL(alloc_pages_exact);
2857
2858/**
Andi Kleenee85c2e2011-05-11 15:13:34 -07002859 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2860 * pages on a node.
Randy Dunlapb5e6ab52011-05-16 13:16:54 -07002861 * @nid: the preferred node ID where memory should be allocated
Andi Kleenee85c2e2011-05-11 15:13:34 -07002862 * @size: the number of bytes to allocate
2863 * @gfp_mask: GFP flags for the allocation
2864 *
2865 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2866 * back.
2867 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2868 * but is not exact.
2869 */
2870void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
2871{
2872 unsigned order = get_order(size);
2873 struct page *p = alloc_pages_node(nid, gfp_mask, order);
2874 if (!p)
2875 return NULL;
2876 return make_alloc_exact((unsigned long)page_address(p), order, size);
2877}
2878EXPORT_SYMBOL(alloc_pages_exact_nid);
2879
2880/**
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002881 * free_pages_exact - release memory allocated via alloc_pages_exact()
2882 * @virt: the value returned by alloc_pages_exact.
2883 * @size: size of allocation, same value as passed to alloc_pages_exact().
2884 *
2885 * Release the memory allocated by a previous call to alloc_pages_exact.
2886 */
2887void free_pages_exact(void *virt, size_t size)
2888{
2889 unsigned long addr = (unsigned long)virt;
2890 unsigned long end = addr + PAGE_ALIGN(size);
2891
2892 while (addr < end) {
2893 free_page(addr);
2894 addr += PAGE_SIZE;
2895 }
2896}
2897EXPORT_SYMBOL(free_pages_exact);
2898
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08002899/**
2900 * nr_free_zone_pages - count number of pages beyond high watermark
2901 * @offset: The zone index of the highest zone
2902 *
2903 * nr_free_zone_pages() counts the number of counts pages which are beyond the
2904 * high watermark within all zones at or below a given zone index. For each
2905 * zone, the number of pages is calculated as:
Jiang Liu834405c2013-07-03 15:03:04 -07002906 * managed_pages - high_pages
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08002907 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08002908static unsigned long nr_free_zone_pages(int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909{
Mel Gormandd1a2392008-04-28 02:12:17 -07002910 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002911 struct zone *zone;
2912
Martin J. Blighe310fd42005-07-29 22:59:18 -07002913 /* Just pick one node, since fallback list is circular */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08002914 unsigned long sum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002915
Mel Gorman0e884602008-04-28 02:12:14 -07002916 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917
Mel Gorman54a6eb52008-04-28 02:12:16 -07002918 for_each_zone_zonelist(zone, z, zonelist, offset) {
Jiang Liub40da042013-02-22 16:33:52 -08002919 unsigned long size = zone->managed_pages;
Mel Gorman41858962009-06-16 15:32:12 -07002920 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07002921 if (size > high)
2922 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923 }
2924
2925 return sum;
2926}
2927
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08002928/**
2929 * nr_free_buffer_pages - count number of pages beyond high watermark
2930 *
2931 * nr_free_buffer_pages() counts the number of pages which are beyond the high
2932 * watermark within ZONE_DMA and ZONE_NORMAL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002933 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08002934unsigned long nr_free_buffer_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002935{
Al Viroaf4ca452005-10-21 02:55:38 -04002936 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002937}
Meelap Shahc2f1a552007-07-17 04:04:39 -07002938EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08002940/**
2941 * nr_free_pagecache_pages - count number of pages beyond high watermark
2942 *
2943 * nr_free_pagecache_pages() counts the number of pages which are beyond the
2944 * high watermark within all zones.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002945 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08002946unsigned long nr_free_pagecache_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002947{
Mel Gorman2a1e2742007-07-17 04:03:12 -07002948 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002950
2951static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002952{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002953 if (IS_ENABLED(CONFIG_NUMA))
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08002954 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002956
Linus Torvalds1da177e2005-04-16 15:20:36 -07002957void si_meminfo(struct sysinfo *val)
2958{
2959 val->totalram = totalram_pages;
2960 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002961 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002962 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002963 val->totalhigh = totalhigh_pages;
2964 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002965 val->mem_unit = PAGE_SIZE;
2966}
2967
2968EXPORT_SYMBOL(si_meminfo);
2969
2970#ifdef CONFIG_NUMA
2971void si_meminfo_node(struct sysinfo *val, int nid)
2972{
Jiang Liucdd91a72013-07-03 15:03:27 -07002973 int zone_type; /* needs to be signed */
2974 unsigned long managed_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975 pg_data_t *pgdat = NODE_DATA(nid);
2976
Jiang Liucdd91a72013-07-03 15:03:27 -07002977 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
2978 managed_pages += pgdat->node_zones[zone_type].managed_pages;
2979 val->totalram = managed_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002980 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002981#ifdef CONFIG_HIGHMEM
Jiang Liub40da042013-02-22 16:33:52 -08002982 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002983 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2984 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002985#else
2986 val->totalhigh = 0;
2987 val->freehigh = 0;
2988#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989 val->mem_unit = PAGE_SIZE;
2990}
2991#endif
2992
David Rientjesddd588b2011-03-22 16:30:46 -07002993/*
David Rientjes7bf02ea2011-05-24 17:11:16 -07002994 * Determine whether the node should be displayed or not, depending on whether
2995 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
David Rientjesddd588b2011-03-22 16:30:46 -07002996 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07002997bool skip_free_areas_node(unsigned int flags, int nid)
David Rientjesddd588b2011-03-22 16:30:46 -07002998{
2999 bool ret = false;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003000 unsigned int cpuset_mems_cookie;
David Rientjesddd588b2011-03-22 16:30:46 -07003001
3002 if (!(flags & SHOW_MEM_FILTER_NODES))
3003 goto out;
3004
Mel Gormancc9a6c82012-03-21 16:34:11 -07003005 do {
3006 cpuset_mems_cookie = get_mems_allowed();
3007 ret = !node_isset(nid, cpuset_current_mems_allowed);
3008 } while (!put_mems_allowed(cpuset_mems_cookie));
David Rientjesddd588b2011-03-22 16:30:46 -07003009out:
3010 return ret;
3011}
3012
Linus Torvalds1da177e2005-04-16 15:20:36 -07003013#define K(x) ((x) << (PAGE_SHIFT-10))
3014
Rabin Vincent377e4f12012-12-11 16:00:24 -08003015static void show_migration_types(unsigned char type)
3016{
3017 static const char types[MIGRATE_TYPES] = {
3018 [MIGRATE_UNMOVABLE] = 'U',
3019 [MIGRATE_RECLAIMABLE] = 'E',
3020 [MIGRATE_MOVABLE] = 'M',
3021 [MIGRATE_RESERVE] = 'R',
3022#ifdef CONFIG_CMA
3023 [MIGRATE_CMA] = 'C',
3024#endif
Minchan Kim194159f2013-02-22 16:33:58 -08003025#ifdef CONFIG_MEMORY_ISOLATION
Rabin Vincent377e4f12012-12-11 16:00:24 -08003026 [MIGRATE_ISOLATE] = 'I',
Minchan Kim194159f2013-02-22 16:33:58 -08003027#endif
Rabin Vincent377e4f12012-12-11 16:00:24 -08003028 };
3029 char tmp[MIGRATE_TYPES + 1];
3030 char *p = tmp;
3031 int i;
3032
3033 for (i = 0; i < MIGRATE_TYPES; i++) {
3034 if (type & (1 << i))
3035 *p++ = types[i];
3036 }
3037
3038 *p = '\0';
3039 printk("(%s) ", tmp);
3040}
3041
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042/*
3043 * Show free area list (used inside shift_scroll-lock stuff)
3044 * We also calculate the percentage fragmentation. We do this by counting the
3045 * memory on each free list with the exception of the first item on the list.
David Rientjesddd588b2011-03-22 16:30:46 -07003046 * Suppresses nodes that are not allowed by current's cpuset if
3047 * SHOW_MEM_FILTER_NODES is passed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003048 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07003049void show_free_areas(unsigned int filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003050{
Jes Sorensenc7241912006-09-27 01:50:05 -07003051 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003052 struct zone *zone;
3053
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003054 for_each_populated_zone(zone) {
David Rientjes7bf02ea2011-05-24 17:11:16 -07003055 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003056 continue;
Jes Sorensenc7241912006-09-27 01:50:05 -07003057 show_node(zone);
3058 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003059
Dave Jones6b482c62005-11-10 15:45:56 -05003060 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003061 struct per_cpu_pageset *pageset;
3062
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003063 pageset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003065 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
3066 cpu, pageset->pcp.high,
3067 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068 }
3069 }
3070
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003071 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3072 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003073 " unevictable:%lu"
Andrew Mortonb76146e2009-10-26 16:49:52 -07003074 " dirty:%lu writeback:%lu unstable:%lu\n"
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07003075 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003076 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3077 " free_cma:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003078 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003079 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003080 global_page_state(NR_ISOLATED_ANON),
3081 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003082 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003083 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003084 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07003085 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07003086 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07003087 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08003088 global_page_state(NR_FREE_PAGES),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07003089 global_page_state(NR_SLAB_RECLAIMABLE),
3090 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07003091 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003092 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08003093 global_page_state(NR_PAGETABLE),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003094 global_page_state(NR_BOUNCE),
3095 global_page_state(NR_FREE_CMA_PAGES));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003096
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003097 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003098 int i;
3099
David Rientjes7bf02ea2011-05-24 17:11:16 -07003100 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003101 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003102 show_node(zone);
3103 printk("%s"
3104 " free:%lukB"
3105 " min:%lukB"
3106 " low:%lukB"
3107 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003108 " active_anon:%lukB"
3109 " inactive_anon:%lukB"
3110 " active_file:%lukB"
3111 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003112 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003113 " isolated(anon):%lukB"
3114 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003115 " present:%lukB"
Jiang Liu9feedc92012-12-12 13:52:12 -08003116 " managed:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003117 " mlocked:%lukB"
3118 " dirty:%lukB"
3119 " writeback:%lukB"
3120 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003121 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003122 " slab_reclaimable:%lukB"
3123 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003124 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003125 " pagetables:%lukB"
3126 " unstable:%lukB"
3127 " bounce:%lukB"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003128 " free_cma:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003129 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003130 " pages_scanned:%lu"
3131 " all_unreclaimable? %s"
3132 "\n",
3133 zone->name,
Mel Gorman88f5acf2011-01-13 15:45:41 -08003134 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07003135 K(min_wmark_pages(zone)),
3136 K(low_wmark_pages(zone)),
3137 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003138 K(zone_page_state(zone, NR_ACTIVE_ANON)),
3139 K(zone_page_state(zone, NR_INACTIVE_ANON)),
3140 K(zone_page_state(zone, NR_ACTIVE_FILE)),
3141 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003142 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003143 K(zone_page_state(zone, NR_ISOLATED_ANON)),
3144 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003145 K(zone->present_pages),
Jiang Liu9feedc92012-12-12 13:52:12 -08003146 K(zone->managed_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003147 K(zone_page_state(zone, NR_MLOCK)),
3148 K(zone_page_state(zone, NR_FILE_DIRTY)),
3149 K(zone_page_state(zone, NR_WRITEBACK)),
3150 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003151 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003152 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3153 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003154 zone_page_state(zone, NR_KERNEL_STACK) *
3155 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003156 K(zone_page_state(zone, NR_PAGETABLE)),
3157 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3158 K(zone_page_state(zone, NR_BOUNCE)),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003159 K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003160 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003161 zone->pages_scanned,
Lisa Du6e543d52013-09-11 14:22:36 -07003162 (!zone_reclaimable(zone) ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07003163 );
3164 printk("lowmem_reserve[]:");
3165 for (i = 0; i < MAX_NR_ZONES; i++)
3166 printk(" %lu", zone->lowmem_reserve[i]);
3167 printk("\n");
3168 }
3169
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003170 for_each_populated_zone(zone) {
Pintu Kumarb8af2942013-09-11 14:20:34 -07003171 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003172 unsigned char types[MAX_ORDER];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003173
David Rientjes7bf02ea2011-05-24 17:11:16 -07003174 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003175 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003176 show_node(zone);
3177 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003178
3179 spin_lock_irqsave(&zone->lock, flags);
3180 for (order = 0; order < MAX_ORDER; order++) {
Rabin Vincent377e4f12012-12-11 16:00:24 -08003181 struct free_area *area = &zone->free_area[order];
3182 int type;
3183
3184 nr[order] = area->nr_free;
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003185 total += nr[order] << order;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003186
3187 types[order] = 0;
3188 for (type = 0; type < MIGRATE_TYPES; type++) {
3189 if (!list_empty(&area->free_list[type]))
3190 types[order] |= 1 << type;
3191 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003192 }
3193 spin_unlock_irqrestore(&zone->lock, flags);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003194 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003195 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003196 if (nr[order])
3197 show_migration_types(types[order]);
3198 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003199 printk("= %lukB\n", K(total));
3200 }
3201
David Rientjes949f7ec2013-04-29 15:07:48 -07003202 hugetlb_show_meminfo();
3203
Larry Woodmane6f36022008-02-04 22:29:30 -08003204 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
3205
Linus Torvalds1da177e2005-04-16 15:20:36 -07003206 show_swap_cache_info();
3207}
3208
Mel Gorman19770b32008-04-28 02:12:18 -07003209static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
3210{
3211 zoneref->zone = zone;
3212 zoneref->zone_idx = zone_idx(zone);
3213}
3214
Linus Torvalds1da177e2005-04-16 15:20:36 -07003215/*
3216 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08003217 *
3218 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003219 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003220static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003221 int nr_zones)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003222{
Christoph Lameter1a932052006-01-06 00:11:16 -08003223 struct zone *zone;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003224 enum zone_type zone_type = MAX_NR_ZONES;
Christoph Lameter02a68a52006-01-06 00:11:18 -08003225
3226 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003227 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08003228 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08003229 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07003230 zoneref_set_zone(zone,
3231 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08003232 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003233 }
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003234 } while (zone_type);
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003235
Christoph Lameter070f8032006-01-06 00:11:19 -08003236 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003237}
3238
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003239
3240/*
3241 * zonelist_order:
3242 * 0 = automatic detection of better ordering.
3243 * 1 = order by ([node] distance, -zonetype)
3244 * 2 = order by (-zonetype, [node] distance)
3245 *
3246 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3247 * the same zonelist. So only NUMA can configure this param.
3248 */
3249#define ZONELIST_ORDER_DEFAULT 0
3250#define ZONELIST_ORDER_NODE 1
3251#define ZONELIST_ORDER_ZONE 2
3252
3253/* zonelist order in the kernel.
3254 * set_zonelist_order() will set this to NODE or ZONE.
3255 */
3256static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
3257static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
3258
3259
Linus Torvalds1da177e2005-04-16 15:20:36 -07003260#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003261/* The value user specified ....changed by config */
3262static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3263/* string for sysctl */
3264#define NUMA_ZONELIST_ORDER_LEN 16
3265char numa_zonelist_order[16] = "default";
3266
3267/*
3268 * interface for configure zonelist ordering.
3269 * command line option "numa_zonelist_order"
3270 * = "[dD]efault - default, automatic configuration.
3271 * = "[nN]ode - order by node locality, then by zone within node
3272 * = "[zZ]one - order by zone, then by locality within zone
3273 */
3274
3275static int __parse_numa_zonelist_order(char *s)
3276{
3277 if (*s == 'd' || *s == 'D') {
3278 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3279 } else if (*s == 'n' || *s == 'N') {
3280 user_zonelist_order = ZONELIST_ORDER_NODE;
3281 } else if (*s == 'z' || *s == 'Z') {
3282 user_zonelist_order = ZONELIST_ORDER_ZONE;
3283 } else {
3284 printk(KERN_WARNING
3285 "Ignoring invalid numa_zonelist_order value: "
3286 "%s\n", s);
3287 return -EINVAL;
3288 }
3289 return 0;
3290}
3291
3292static __init int setup_numa_zonelist_order(char *s)
3293{
Volodymyr G. Lukiianykecb256f2011-01-13 15:46:26 -08003294 int ret;
3295
3296 if (!s)
3297 return 0;
3298
3299 ret = __parse_numa_zonelist_order(s);
3300 if (ret == 0)
3301 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
3302
3303 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003304}
3305early_param("numa_zonelist_order", setup_numa_zonelist_order);
3306
3307/*
3308 * sysctl handler for numa_zonelist_order
3309 */
3310int numa_zonelist_order_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003311 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003312 loff_t *ppos)
3313{
3314 char saved_string[NUMA_ZONELIST_ORDER_LEN];
3315 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01003316 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003317
Andi Kleen443c6f12009-12-23 21:00:47 +01003318 mutex_lock(&zl_order_mutex);
Chen Gangdacbde02013-07-03 15:02:35 -07003319 if (write) {
3320 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
3321 ret = -EINVAL;
3322 goto out;
3323 }
3324 strcpy(saved_string, (char *)table->data);
3325 }
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003326 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003327 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01003328 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003329 if (write) {
3330 int oldval = user_zonelist_order;
Chen Gangdacbde02013-07-03 15:02:35 -07003331
3332 ret = __parse_numa_zonelist_order((char *)table->data);
3333 if (ret) {
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003334 /*
3335 * bogus value. restore saved string
3336 */
Chen Gangdacbde02013-07-03 15:02:35 -07003337 strncpy((char *)table->data, saved_string,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003338 NUMA_ZONELIST_ORDER_LEN);
3339 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003340 } else if (oldval != user_zonelist_order) {
3341 mutex_lock(&zonelists_mutex);
Jiang Liu9adb62a2012-07-31 16:43:28 -07003342 build_all_zonelists(NULL, NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003343 mutex_unlock(&zonelists_mutex);
3344 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003345 }
Andi Kleen443c6f12009-12-23 21:00:47 +01003346out:
3347 mutex_unlock(&zl_order_mutex);
3348 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003349}
3350
3351
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003352#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003353static int node_load[MAX_NUMNODES];
3354
Linus Torvalds1da177e2005-04-16 15:20:36 -07003355/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07003356 * 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 -07003357 * @node: node whose fallback list we're appending
3358 * @used_node_mask: nodemask_t of already used nodes
3359 *
3360 * We use a number of factors to determine which is the next node that should
3361 * appear on a given node's fallback list. The node should not have appeared
3362 * already in @node's fallback list, and it should be the next closest node
3363 * according to the distance array (which contains arbitrary distance values
3364 * from each node to each node in the system), and should also prefer nodes
3365 * with no CPUs, since presumably they'll have very little allocation pressure
3366 * on them otherwise.
3367 * It returns -1 if no node is found.
3368 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003369static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003370{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01003371 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003372 int min_val = INT_MAX;
David Rientjes00ef2d22013-02-22 16:35:36 -08003373 int best_node = NUMA_NO_NODE;
Rusty Russella70f7302009-03-13 14:49:46 +10303374 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003375
Linus Torvalds4cf808e2006-02-17 20:38:21 +01003376 /* Use the local node if we haven't already */
3377 if (!node_isset(node, *used_node_mask)) {
3378 node_set(node, *used_node_mask);
3379 return node;
3380 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003381
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08003382 for_each_node_state(n, N_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003383
3384 /* Don't want a node to appear more than once */
3385 if (node_isset(n, *used_node_mask))
3386 continue;
3387
Linus Torvalds1da177e2005-04-16 15:20:36 -07003388 /* Use the distance array to find the distance */
3389 val = node_distance(node, n);
3390
Linus Torvalds4cf808e2006-02-17 20:38:21 +01003391 /* Penalize nodes under us ("prefer the next node") */
3392 val += (n < node);
3393
Linus Torvalds1da177e2005-04-16 15:20:36 -07003394 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10303395 tmp = cpumask_of_node(n);
3396 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003397 val += PENALTY_FOR_NODE_WITH_CPUS;
3398
3399 /* Slight preference for less loaded node */
3400 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
3401 val += node_load[n];
3402
3403 if (val < min_val) {
3404 min_val = val;
3405 best_node = n;
3406 }
3407 }
3408
3409 if (best_node >= 0)
3410 node_set(best_node, *used_node_mask);
3411
3412 return best_node;
3413}
3414
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003415
3416/*
3417 * Build zonelists ordered by node and zones within node.
3418 * This results in maximum locality--normal zone overflows into local
3419 * DMA zone, if any--but risks exhausting DMA zone.
3420 */
3421static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003422{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003423 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003424 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003425
Mel Gorman54a6eb52008-04-28 02:12:16 -07003426 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07003427 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07003428 ;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003429 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gormandd1a2392008-04-28 02:12:17 -07003430 zonelist->_zonerefs[j].zone = NULL;
3431 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003432}
3433
3434/*
Christoph Lameter523b9452007-10-16 01:25:37 -07003435 * Build gfp_thisnode zonelists
3436 */
3437static void build_thisnode_zonelists(pg_data_t *pgdat)
3438{
Christoph Lameter523b9452007-10-16 01:25:37 -07003439 int j;
3440 struct zonelist *zonelist;
3441
Mel Gorman54a6eb52008-04-28 02:12:16 -07003442 zonelist = &pgdat->node_zonelists[1];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003443 j = build_zonelists_node(pgdat, zonelist, 0);
Mel Gormandd1a2392008-04-28 02:12:17 -07003444 zonelist->_zonerefs[j].zone = NULL;
3445 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07003446}
3447
3448/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003449 * Build zonelists ordered by zone and nodes within zones.
3450 * This results in conserving DMA zone[s] until all Normal memory is
3451 * exhausted, but results in overflowing to remote node while memory
3452 * may still exist in local DMA zone.
3453 */
3454static int node_order[MAX_NUMNODES];
3455
3456static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
3457{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003458 int pos, j, node;
3459 int zone_type; /* needs to be signed */
3460 struct zone *z;
3461 struct zonelist *zonelist;
3462
Mel Gorman54a6eb52008-04-28 02:12:16 -07003463 zonelist = &pgdat->node_zonelists[0];
3464 pos = 0;
3465 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
3466 for (j = 0; j < nr_nodes; j++) {
3467 node = node_order[j];
3468 z = &NODE_DATA(node)->node_zones[zone_type];
3469 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07003470 zoneref_set_zone(z,
3471 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07003472 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003473 }
3474 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003475 }
Mel Gormandd1a2392008-04-28 02:12:17 -07003476 zonelist->_zonerefs[pos].zone = NULL;
3477 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003478}
3479
3480static int default_zonelist_order(void)
3481{
3482 int nid, zone_type;
Pintu Kumarb8af2942013-09-11 14:20:34 -07003483 unsigned long low_kmem_size, total_size;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003484 struct zone *z;
3485 int average_size;
3486 /*
Pintu Kumarb8af2942013-09-11 14:20:34 -07003487 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003488 * If they are really small and used heavily, the system can fall
3489 * into OOM very easily.
David Rientjese325c902010-05-24 14:32:13 -07003490 * This function detect ZONE_DMA/DMA32 size and configures zone order.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003491 */
3492 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
3493 low_kmem_size = 0;
3494 total_size = 0;
3495 for_each_online_node(nid) {
3496 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3497 z = &NODE_DATA(nid)->node_zones[zone_type];
3498 if (populated_zone(z)) {
3499 if (zone_type < ZONE_NORMAL)
Jiang Liu4f9f4772013-07-03 15:03:07 -07003500 low_kmem_size += z->managed_pages;
3501 total_size += z->managed_pages;
David Rientjese325c902010-05-24 14:32:13 -07003502 } else if (zone_type == ZONE_NORMAL) {
3503 /*
3504 * If any node has only lowmem, then node order
3505 * is preferred to allow kernel allocations
3506 * locally; otherwise, they can easily infringe
3507 * on other nodes when there is an abundance of
3508 * lowmem available to allocate from.
3509 */
3510 return ZONELIST_ORDER_NODE;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003511 }
3512 }
3513 }
3514 if (!low_kmem_size || /* there are no DMA area. */
3515 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
3516 return ZONELIST_ORDER_NODE;
3517 /*
3518 * look into each node's config.
Pintu Kumarb8af2942013-09-11 14:20:34 -07003519 * If there is a node whose DMA/DMA32 memory is very big area on
3520 * local memory, NODE_ORDER may be suitable.
3521 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003522 average_size = total_size /
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08003523 (nodes_weight(node_states[N_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003524 for_each_online_node(nid) {
3525 low_kmem_size = 0;
3526 total_size = 0;
3527 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3528 z = &NODE_DATA(nid)->node_zones[zone_type];
3529 if (populated_zone(z)) {
3530 if (zone_type < ZONE_NORMAL)
3531 low_kmem_size += z->present_pages;
3532 total_size += z->present_pages;
3533 }
3534 }
3535 if (low_kmem_size &&
3536 total_size > average_size && /* ignore small node */
3537 low_kmem_size > total_size * 70/100)
3538 return ZONELIST_ORDER_NODE;
3539 }
3540 return ZONELIST_ORDER_ZONE;
3541}
3542
3543static void set_zonelist_order(void)
3544{
3545 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
3546 current_zonelist_order = default_zonelist_order();
3547 else
3548 current_zonelist_order = user_zonelist_order;
3549}
3550
3551static void build_zonelists(pg_data_t *pgdat)
3552{
3553 int j, node, load;
3554 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003555 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003556 int local_node, prev_node;
3557 struct zonelist *zonelist;
3558 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003559
3560 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07003561 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003562 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07003563 zonelist->_zonerefs[0].zone = NULL;
3564 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003565 }
3566
3567 /* NUMA-aware ordering of nodes */
3568 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003569 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003570 prev_node = local_node;
3571 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003572
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003573 memset(node_order, 0, sizeof(node_order));
3574 j = 0;
3575
Linus Torvalds1da177e2005-04-16 15:20:36 -07003576 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
3577 /*
3578 * We don't want to pressure a particular node.
3579 * So adding penalty to the first node in same
3580 * distance group to make it round-robin.
3581 */
David Rientjes957f8222012-10-08 16:33:24 -07003582 if (node_distance(local_node, node) !=
3583 node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003584 node_load[node] = load;
3585
Linus Torvalds1da177e2005-04-16 15:20:36 -07003586 prev_node = node;
3587 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003588 if (order == ZONELIST_ORDER_NODE)
3589 build_zonelists_in_node_order(pgdat, node);
3590 else
3591 node_order[j++] = node; /* remember order */
3592 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003593
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003594 if (order == ZONELIST_ORDER_ZONE) {
3595 /* calculate node order -- i.e., DMA last! */
3596 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003597 }
Christoph Lameter523b9452007-10-16 01:25:37 -07003598
3599 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003600}
3601
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003602/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003603static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003604{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003605 struct zonelist *zonelist;
3606 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07003607 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003608
Mel Gorman54a6eb52008-04-28 02:12:16 -07003609 zonelist = &pgdat->node_zonelists[0];
3610 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3611 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07003612 for (z = zonelist->_zonerefs; z->zone; z++)
3613 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003614}
3615
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003616#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3617/*
3618 * Return node id of node used for "local" allocations.
3619 * I.e., first node id of first zone in arg node's generic zonelist.
3620 * Used for initializing percpu 'numa_mem', which is used primarily
3621 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3622 */
3623int local_memory_node(int node)
3624{
3625 struct zone *zone;
3626
3627 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3628 gfp_zone(GFP_KERNEL),
3629 NULL,
3630 &zone);
3631 return zone->node;
3632}
3633#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003634
Linus Torvalds1da177e2005-04-16 15:20:36 -07003635#else /* CONFIG_NUMA */
3636
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003637static void set_zonelist_order(void)
3638{
3639 current_zonelist_order = ZONELIST_ORDER_ZONE;
3640}
3641
3642static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003643{
Christoph Lameter19655d32006-09-25 23:31:19 -07003644 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003645 enum zone_type j;
3646 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003647
3648 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003649
Mel Gorman54a6eb52008-04-28 02:12:16 -07003650 zonelist = &pgdat->node_zonelists[0];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003651 j = build_zonelists_node(pgdat, zonelist, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003652
Mel Gorman54a6eb52008-04-28 02:12:16 -07003653 /*
3654 * Now we build the zonelist so that it contains the zones
3655 * of all the other nodes.
3656 * We don't want to pressure a particular node, so when
3657 * building the zones for node N, we make sure that the
3658 * zones coming right after the local ones are those from
3659 * node N+1 (modulo N)
3660 */
3661 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3662 if (!node_online(node))
3663 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003664 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003665 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07003666 for (node = 0; node < local_node; node++) {
3667 if (!node_online(node))
3668 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003669 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gorman54a6eb52008-04-28 02:12:16 -07003670 }
3671
Mel Gormandd1a2392008-04-28 02:12:17 -07003672 zonelist->_zonerefs[j].zone = NULL;
3673 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003674}
3675
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003676/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003677static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003678{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003679 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003680}
3681
Linus Torvalds1da177e2005-04-16 15:20:36 -07003682#endif /* CONFIG_NUMA */
3683
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003684/*
3685 * Boot pageset table. One per cpu which is going to be used for all
3686 * zones and all nodes. The parameters will be set in such a way
3687 * that an item put on a list will immediately be handed over to
3688 * the buddy list. This is safe since pageset manipulation is done
3689 * with interrupts disabled.
3690 *
3691 * The boot_pagesets must be kept even after bootup is complete for
3692 * unused processors and/or zones. They do play a role for bootstrapping
3693 * hotplugged processors.
3694 *
3695 * zoneinfo_show() and maybe other functions do
3696 * not check if the processor is online before following the pageset pointer.
3697 * Other parts of the kernel may not check if the zone is available.
3698 */
3699static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3700static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07003701static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003702
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003703/*
3704 * Global mutex to protect against size modification of zonelists
3705 * as well as to serialize pageset setup for the new populated zone.
3706 */
3707DEFINE_MUTEX(zonelists_mutex);
3708
Rusty Russell9b1a4d32008-07-28 12:16:30 -05003709/* return values int ....just for stop_machine() */
Jiang Liu4ed7e022012-07-31 16:43:35 -07003710static int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003711{
Yasunori Goto68113782006-06-23 02:03:11 -07003712 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003713 int cpu;
Jiang Liu9adb62a2012-07-31 16:43:28 -07003714 pg_data_t *self = data;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003715
Bo Liu7f9cfb32009-08-18 14:11:19 -07003716#ifdef CONFIG_NUMA
3717 memset(node_load, 0, sizeof(node_load));
3718#endif
Jiang Liu9adb62a2012-07-31 16:43:28 -07003719
3720 if (self && !node_online(self->node_id)) {
3721 build_zonelists(self);
3722 build_zonelist_cache(self);
3723 }
3724
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003725 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07003726 pg_data_t *pgdat = NODE_DATA(nid);
3727
3728 build_zonelists(pgdat);
3729 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003730 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003731
3732 /*
3733 * Initialize the boot_pagesets that are going to be used
3734 * for bootstrapping processors. The real pagesets for
3735 * each zone will be allocated later when the per cpu
3736 * allocator is available.
3737 *
3738 * boot_pagesets are used also for bootstrapping offline
3739 * cpus if the system is already booted because the pagesets
3740 * are needed to initialize allocators on a specific cpu too.
3741 * F.e. the percpu allocator needs the page allocator which
3742 * needs the percpu allocator in order to allocate its pagesets
3743 * (a chicken-egg dilemma).
3744 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003745 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003746 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3747
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003748#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3749 /*
3750 * We now know the "local memory node" for each node--
3751 * i.e., the node of the first zone in the generic zonelist.
3752 * Set up numa_mem percpu variable for on-line cpus. During
3753 * boot, only the boot cpu should be on-line; we'll init the
3754 * secondary cpus' numa_mem as they come on-line. During
3755 * node/memory hotplug, we'll fixup all on-line cpus.
3756 */
3757 if (cpu_online(cpu))
3758 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3759#endif
3760 }
3761
Yasunori Goto68113782006-06-23 02:03:11 -07003762 return 0;
3763}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003764
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003765/*
3766 * Called with zonelists_mutex held always
3767 * unless system_state == SYSTEM_BOOTING.
3768 */
Jiang Liu9adb62a2012-07-31 16:43:28 -07003769void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
Yasunori Goto68113782006-06-23 02:03:11 -07003770{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003771 set_zonelist_order();
3772
Yasunori Goto68113782006-06-23 02:03:11 -07003773 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07003774 __build_all_zonelists(NULL);
Mel Gorman68ad8df2008-07-23 21:26:52 -07003775 mminit_verify_zonelist();
Yasunori Goto68113782006-06-23 02:03:11 -07003776 cpuset_init_current_mems_allowed();
3777 } else {
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003778#ifdef CONFIG_MEMORY_HOTPLUG
Jiang Liu9adb62a2012-07-31 16:43:28 -07003779 if (zone)
3780 setup_zone_pageset(zone);
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003781#endif
Cody P Schaferdd1895e2013-07-03 15:01:36 -07003782 /* we have to stop all cpus to guarantee there is no user
3783 of zonelist */
Jiang Liu9adb62a2012-07-31 16:43:28 -07003784 stop_machine(__build_all_zonelists, pgdat, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07003785 /* cpuset refresh routine should be here */
3786 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07003787 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003788 /*
3789 * Disable grouping by mobility if the number of pages in the
3790 * system is too low to allow the mechanism to work. It would be
3791 * more accurate, but expensive to check per-zone. This check is
3792 * made on memory-hotadd so a system can start with mobility
3793 * disabled and enable it later
3794 */
Mel Gormand9c23402007-10-16 01:26:01 -07003795 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003796 page_group_by_mobility_disabled = 1;
3797 else
3798 page_group_by_mobility_disabled = 0;
3799
3800 printk("Built %i zonelists in %s order, mobility grouping %s. "
3801 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003802 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003803 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003804 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003805 vm_total_pages);
3806#ifdef CONFIG_NUMA
3807 printk("Policy zone: %s\n", zone_names[policy_zone]);
3808#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003809}
3810
3811/*
3812 * Helper functions to size the waitqueue hash table.
3813 * Essentially these want to choose hash table sizes sufficiently
3814 * large so that collisions trying to wait on pages are rare.
3815 * But in fact, the number of active page waitqueues on typical
3816 * systems is ridiculously low, less than 200. So this is even
3817 * conservative, even though it seems large.
3818 *
3819 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3820 * waitqueues, i.e. the size of the waitq table given the number of pages.
3821 */
3822#define PAGES_PER_WAITQUEUE 256
3823
Yasunori Gotocca448f2006-06-23 02:03:10 -07003824#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07003825static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003826{
3827 unsigned long size = 1;
3828
3829 pages /= PAGES_PER_WAITQUEUE;
3830
3831 while (size < pages)
3832 size <<= 1;
3833
3834 /*
3835 * Once we have dozens or even hundreds of threads sleeping
3836 * on IO we've got bigger problems than wait queue collision.
3837 * Limit the size of the wait table to a reasonable size.
3838 */
3839 size = min(size, 4096UL);
3840
3841 return max(size, 4UL);
3842}
Yasunori Gotocca448f2006-06-23 02:03:10 -07003843#else
3844/*
3845 * A zone's size might be changed by hot-add, so it is not possible to determine
3846 * a suitable size for its wait_table. So we use the maximum size now.
3847 *
3848 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3849 *
3850 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3851 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3852 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3853 *
3854 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3855 * or more by the traditional way. (See above). It equals:
3856 *
3857 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3858 * ia64(16K page size) : = ( 8G + 4M)byte.
3859 * powerpc (64K page size) : = (32G +16M)byte.
3860 */
3861static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3862{
3863 return 4096UL;
3864}
3865#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003866
3867/*
3868 * This is an integer logarithm so that shifts can be used later
3869 * to extract the more random high bits from the multiplicative
3870 * hash function before the remainder is taken.
3871 */
3872static inline unsigned long wait_table_bits(unsigned long size)
3873{
3874 return ffz(~size);
3875}
3876
Mel Gorman56fd56b2007-10-16 01:25:58 -07003877/*
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07003878 * Check if a pageblock contains reserved pages
3879 */
3880static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
3881{
3882 unsigned long pfn;
3883
3884 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
3885 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
3886 return 1;
3887 }
3888 return 0;
3889}
3890
3891/*
Mel Gormand9c23402007-10-16 01:26:01 -07003892 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07003893 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3894 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07003895 * higher will lead to a bigger reserve which will get freed as contiguous
3896 * blocks as reclaim kicks in
3897 */
3898static void setup_zone_migrate_reserve(struct zone *zone)
3899{
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07003900 unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003901 struct page *page;
Mel Gorman78986a62009-09-21 17:03:02 -07003902 unsigned long block_migratetype;
3903 int reserve;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003904
Michal Hockod02156382011-12-08 14:34:27 -08003905 /*
3906 * Get the start pfn, end pfn and the number of blocks to reserve
3907 * We have to be careful to be aligned to pageblock_nr_pages to
3908 * make sure that we always check pfn_valid for the first page in
3909 * the block.
3910 */
Mel Gorman56fd56b2007-10-16 01:25:58 -07003911 start_pfn = zone->zone_start_pfn;
Cody P Schafer108bcc92013-02-22 16:35:23 -08003912 end_pfn = zone_end_pfn(zone);
Michal Hockod02156382011-12-08 14:34:27 -08003913 start_pfn = roundup(start_pfn, pageblock_nr_pages);
Mel Gorman41858962009-06-16 15:32:12 -07003914 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07003915 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003916
Mel Gorman78986a62009-09-21 17:03:02 -07003917 /*
3918 * Reserve blocks are generally in place to help high-order atomic
3919 * allocations that are short-lived. A min_free_kbytes value that
3920 * would result in more than 2 reserve blocks for atomic allocations
3921 * is assumed to be in place to help anti-fragmentation for the
3922 * future allocation of hugepages at runtime.
3923 */
3924 reserve = min(2, reserve);
3925
Mel Gormand9c23402007-10-16 01:26:01 -07003926 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07003927 if (!pfn_valid(pfn))
3928 continue;
3929 page = pfn_to_page(pfn);
3930
Adam Litke344c7902008-09-02 14:35:38 -07003931 /* Watch out for overlapping nodes */
3932 if (page_to_nid(page) != zone_to_nid(zone))
3933 continue;
3934
Mel Gorman56fd56b2007-10-16 01:25:58 -07003935 block_migratetype = get_pageblock_migratetype(page);
3936
Mel Gorman938929f2012-01-10 15:07:14 -08003937 /* Only test what is necessary when the reserves are not met */
3938 if (reserve > 0) {
3939 /*
3940 * Blocks with reserved pages will never free, skip
3941 * them.
3942 */
3943 block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
3944 if (pageblock_is_reserved(pfn, block_end_pfn))
3945 continue;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003946
Mel Gorman938929f2012-01-10 15:07:14 -08003947 /* If this block is reserved, account for it */
3948 if (block_migratetype == MIGRATE_RESERVE) {
3949 reserve--;
3950 continue;
3951 }
3952
3953 /* Suitable for reserving if this block is movable */
3954 if (block_migratetype == MIGRATE_MOVABLE) {
3955 set_pageblock_migratetype(page,
3956 MIGRATE_RESERVE);
3957 move_freepages_block(zone, page,
3958 MIGRATE_RESERVE);
3959 reserve--;
3960 continue;
3961 }
Mel Gorman56fd56b2007-10-16 01:25:58 -07003962 }
3963
3964 /*
3965 * If the reserve is met and this is a previous reserved block,
3966 * take it back
3967 */
3968 if (block_migratetype == MIGRATE_RESERVE) {
3969 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
3970 move_freepages_block(zone, page, MIGRATE_MOVABLE);
3971 }
3972 }
3973}
Mel Gormanac0e5b72007-10-16 01:25:58 -07003974
Linus Torvalds1da177e2005-04-16 15:20:36 -07003975/*
3976 * Initially all pages are reserved - free ones are freed
3977 * up by free_all_bootmem() once the early boot process is
3978 * done. Non-atomic initialization, single-pass.
3979 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003980void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08003981 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003982{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003983 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07003984 unsigned long end_pfn = start_pfn + size;
3985 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003986 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003987
Hugh Dickins22b31ee2009-01-06 14:40:09 -08003988 if (highest_memmap_pfn < end_pfn - 1)
3989 highest_memmap_pfn = end_pfn - 1;
3990
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003991 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08003992 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08003993 /*
3994 * There can be holes in boot-time mem_map[]s
3995 * handed to this function. They do not
3996 * exist on hotplugged memory.
3997 */
3998 if (context == MEMMAP_EARLY) {
3999 if (!early_pfn_valid(pfn))
4000 continue;
4001 if (!early_pfn_in_nid(pfn, nid))
4002 continue;
4003 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004004 page = pfn_to_page(pfn);
4005 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07004006 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08004007 init_page_count(page);
Mel Gorman22b751c2013-02-22 16:34:59 -08004008 page_mapcount_reset(page);
Peter Zijlstra90572892013-10-07 11:29:20 +01004009 page_cpupid_reset_last(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004010 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004011 /*
4012 * Mark the block movable so that blocks are reserved for
4013 * movable at startup. This will force kernel allocations
4014 * to reserve their blocks rather than leaking throughout
4015 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07004016 * kernel allocations are made. Later some blocks near
4017 * the start are marked MIGRATE_RESERVE by
4018 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004019 *
4020 * bitmap is created for zone's valid pfn range. but memmap
4021 * can be created for invalid pages (for alignment)
4022 * check here not to call set_pageblock_migratetype() against
4023 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004024 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004025 if ((z->zone_start_pfn <= pfn)
Cody P Schafer108bcc92013-02-22 16:35:23 -08004026 && (pfn < zone_end_pfn(z))
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004027 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07004028 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004029
Linus Torvalds1da177e2005-04-16 15:20:36 -07004030 INIT_LIST_HEAD(&page->lru);
4031#ifdef WANT_PAGE_VIRTUAL
4032 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
4033 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07004034 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004035#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004036 }
4037}
4038
Andi Kleen1e548de2008-02-04 22:29:26 -08004039static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004040{
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004041 int order, t;
4042 for_each_migratetype_order(order, t) {
4043 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004044 zone->free_area[order].nr_free = 0;
4045 }
4046}
4047
4048#ifndef __HAVE_ARCH_MEMMAP_INIT
4049#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08004050 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004051#endif
4052
Jiang Liu4ed7e022012-07-31 16:43:35 -07004053static int __meminit zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004054{
David Howells3a6be872009-05-06 16:03:03 -07004055#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004056 int batch;
4057
4058 /*
4059 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07004060 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004061 *
4062 * OK, so we don't know how big the cache is. So guess.
4063 */
Jiang Liub40da042013-02-22 16:33:52 -08004064 batch = zone->managed_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07004065 if (batch * PAGE_SIZE > 512 * 1024)
4066 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004067 batch /= 4; /* We effectively *= 4 below */
4068 if (batch < 1)
4069 batch = 1;
4070
4071 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004072 * Clamp the batch to a 2^n - 1 value. Having a power
4073 * of 2 value was found to be more likely to have
4074 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004075 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004076 * For example if 2 tasks are alternately allocating
4077 * batches of pages, one task can end up with a lot
4078 * of pages of one half of the possible page colors
4079 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004080 */
David Howells91552032009-05-06 16:03:02 -07004081 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07004082
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004083 return batch;
David Howells3a6be872009-05-06 16:03:03 -07004084
4085#else
4086 /* The deferral and batching of frees should be suppressed under NOMMU
4087 * conditions.
4088 *
4089 * The problem is that NOMMU needs to be able to allocate large chunks
4090 * of contiguous memory as there's no hardware page translation to
4091 * assemble apparent contiguous memory from discontiguous pages.
4092 *
4093 * Queueing large contiguous runs of pages for batching, however,
4094 * causes the pages to actually be freed in smaller chunks. As there
4095 * can be a significant delay between the individual batches being
4096 * recycled, this leads to the once large chunks of space being
4097 * fragmented and becoming unavailable for high-order allocations.
4098 */
4099 return 0;
4100#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004101}
4102
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004103/*
4104 * pcp->high and pcp->batch values are related and dependent on one another:
4105 * ->batch must never be higher then ->high.
4106 * The following function updates them in a safe manner without read side
4107 * locking.
4108 *
4109 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4110 * those fields changing asynchronously (acording the the above rule).
4111 *
4112 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4113 * outside of boot time (or some other assurance that no concurrent updaters
4114 * exist).
4115 */
4116static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
4117 unsigned long batch)
4118{
4119 /* start with a fail safe value for batch */
4120 pcp->batch = 1;
4121 smp_wmb();
4122
4123 /* Update high, then batch, in order */
4124 pcp->high = high;
4125 smp_wmb();
4126
4127 pcp->batch = batch;
4128}
4129
Cody P Schafer36640332013-07-03 15:01:40 -07004130/* a companion to pageset_set_high() */
Cody P Schafer4008bab2013-07-03 15:01:28 -07004131static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
4132{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004133 pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
Cody P Schafer4008bab2013-07-03 15:01:28 -07004134}
4135
Cody P Schafer88c90db2013-07-03 15:01:35 -07004136static void pageset_init(struct per_cpu_pageset *p)
Christoph Lameter2caaad42005-06-21 17:15:00 -07004137{
4138 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004139 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004140
Magnus Damm1c6fe942005-10-26 01:58:59 -07004141 memset(p, 0, sizeof(*p));
4142
Christoph Lameter3dfa5722008-02-04 22:29:19 -08004143 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004144 pcp->count = 0;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004145 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
4146 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07004147}
4148
Cody P Schafer88c90db2013-07-03 15:01:35 -07004149static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
4150{
4151 pageset_init(p);
4152 pageset_set_batch(p, batch);
4153}
4154
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004155/*
Cody P Schafer36640332013-07-03 15:01:40 -07004156 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004157 * to the value high for the pageset p.
4158 */
Cody P Schafer36640332013-07-03 15:01:40 -07004159static void pageset_set_high(struct per_cpu_pageset *p,
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004160 unsigned long high)
4161{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004162 unsigned long batch = max(1UL, high / 4);
4163 if ((high / 4) > (PAGE_SHIFT * 8))
4164 batch = PAGE_SHIFT * 8;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004165
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004166 pageset_update(&p->pcp, high, batch);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004167}
4168
Cody P Schafer169f6c12013-07-03 15:01:41 -07004169static void __meminit pageset_set_high_and_batch(struct zone *zone,
4170 struct per_cpu_pageset *pcp)
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004171{
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004172 if (percpu_pagelist_fraction)
Cody P Schafer36640332013-07-03 15:01:40 -07004173 pageset_set_high(pcp,
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004174 (zone->managed_pages /
4175 percpu_pagelist_fraction));
4176 else
4177 pageset_set_batch(pcp, zone_batchsize(zone));
4178}
4179
Cody P Schafer169f6c12013-07-03 15:01:41 -07004180static void __meminit zone_pageset_init(struct zone *zone, int cpu)
4181{
4182 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4183
4184 pageset_init(pcp);
4185 pageset_set_high_and_batch(zone, pcp);
4186}
4187
Jiang Liu4ed7e022012-07-31 16:43:35 -07004188static void __meminit setup_zone_pageset(struct zone *zone)
Wu Fengguang319774e2010-05-24 14:32:49 -07004189{
4190 int cpu;
Wu Fengguang319774e2010-05-24 14:32:49 -07004191 zone->pageset = alloc_percpu(struct per_cpu_pageset);
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004192 for_each_possible_cpu(cpu)
4193 zone_pageset_init(zone, cpu);
Wu Fengguang319774e2010-05-24 14:32:49 -07004194}
4195
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004196/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004197 * Allocate per cpu pagesets and initialize them.
4198 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07004199 */
Al Viro78d99552005-12-15 09:18:25 +00004200void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004201{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004202 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004203
Wu Fengguang319774e2010-05-24 14:32:49 -07004204 for_each_populated_zone(zone)
4205 setup_zone_pageset(zone);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004206}
4207
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004208static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07004209int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07004210{
4211 int i;
4212 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004213 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07004214
4215 /*
4216 * The per-page waitqueue mechanism uses hashed waitqueues
4217 * per zone.
4218 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07004219 zone->wait_table_hash_nr_entries =
4220 wait_table_hash_nr_entries(zone_size_pages);
4221 zone->wait_table_bits =
4222 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004223 alloc_size = zone->wait_table_hash_nr_entries
4224 * sizeof(wait_queue_head_t);
4225
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07004226 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07004227 zone->wait_table = (wait_queue_head_t *)
Yinghai Lu8f389a992011-05-11 15:13:32 -07004228 alloc_bootmem_node_nopanic(pgdat, alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004229 } else {
4230 /*
4231 * This case means that a zone whose size was 0 gets new memory
4232 * via memory hot-add.
4233 * But it may be the case that a new node was hot-added. In
4234 * this case vmalloc() will not be able to use this new node's
4235 * memory - this wait_table must be initialized to use this new
4236 * node itself as well.
4237 * To use this new node's memory, further consideration will be
4238 * necessary.
4239 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07004240 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004241 }
4242 if (!zone->wait_table)
4243 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07004244
Pintu Kumarb8af2942013-09-11 14:20:34 -07004245 for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07004246 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004247
4248 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004249}
4250
Matt Tolentinoc09b4242006-01-17 07:03:44 +01004251static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07004252{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004253 /*
4254 * per cpu subsystem is not up at this point. The following code
4255 * relies on the ability of the linker to provide the
4256 * offset of a (static) per cpu variable into the per cpu area.
4257 */
4258 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07004259
Xishi Qiub38a8722013-11-12 15:07:20 -08004260 if (populated_zone(zone))
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004261 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
4262 zone->name, zone->present_pages,
4263 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07004264}
4265
Jiang Liu4ed7e022012-07-31 16:43:35 -07004266int __meminit init_currently_empty_zone(struct zone *zone,
Yasunori Goto718127c2006-06-23 02:03:10 -07004267 unsigned long zone_start_pfn,
Dave Hansena2f3aa022007-01-10 23:15:30 -08004268 unsigned long size,
4269 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07004270{
4271 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004272 int ret;
4273 ret = zone_wait_table_init(zone, size);
4274 if (ret)
4275 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07004276 pgdat->nr_zones = zone_idx(zone) + 1;
4277
Dave Hansened8ece22005-10-29 18:16:50 -07004278 zone->zone_start_pfn = zone_start_pfn;
4279
Mel Gorman708614e2008-07-23 21:26:51 -07004280 mminit_dprintk(MMINIT_TRACE, "memmap_init",
4281 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4282 pgdat->node_id,
4283 (unsigned long)zone_idx(zone),
4284 zone_start_pfn, (zone_start_pfn + size));
4285
Andi Kleen1e548de2008-02-04 22:29:26 -08004286 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07004287
4288 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004289}
4290
Tejun Heo0ee332c2011-12-08 10:22:09 -08004291#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07004292#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4293/*
4294 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4295 * Architectures may implement their own version but if add_active_range()
4296 * was used and there are no special requirements, this is a convenient
4297 * alternative
4298 */
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004299int __meminit __early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004300{
Tejun Heoc13291a2011-07-12 10:46:30 +02004301 unsigned long start_pfn, end_pfn;
Yinghai Lue76b63f2013-09-11 14:22:17 -07004302 int nid;
Russ Anderson7c243c72013-04-29 15:07:59 -07004303 /*
4304 * NOTE: The following SMP-unsafe globals are only used early in boot
4305 * when the kernel is running single-threaded.
4306 */
4307 static unsigned long __meminitdata last_start_pfn, last_end_pfn;
4308 static int __meminitdata last_nid;
4309
4310 if (last_start_pfn <= pfn && pfn < last_end_pfn)
4311 return last_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004312
Yinghai Lue76b63f2013-09-11 14:22:17 -07004313 nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
4314 if (nid != -1) {
4315 last_start_pfn = start_pfn;
4316 last_end_pfn = end_pfn;
4317 last_nid = nid;
4318 }
4319
4320 return nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004321}
4322#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4323
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004324int __meminit early_pfn_to_nid(unsigned long pfn)
4325{
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08004326 int nid;
4327
4328 nid = __early_pfn_to_nid(pfn);
4329 if (nid >= 0)
4330 return nid;
4331 /* just returns 0 */
4332 return 0;
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004333}
4334
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08004335#ifdef CONFIG_NODES_SPAN_OTHER_NODES
4336bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
4337{
4338 int nid;
4339
4340 nid = __early_pfn_to_nid(pfn);
4341 if (nid >= 0 && nid != node)
4342 return false;
4343 return true;
4344}
4345#endif
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004346
Mel Gormanc7132162006-09-27 01:49:43 -07004347/**
4348 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004349 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4350 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07004351 *
4352 * If an architecture guarantees that all ranges registered with
4353 * add_active_ranges() contain no holes and may be freed, this
4354 * this function may be used instead of calling free_bootmem() manually.
4355 */
Tejun Heoc13291a2011-07-12 10:46:30 +02004356void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004357{
Tejun Heoc13291a2011-07-12 10:46:30 +02004358 unsigned long start_pfn, end_pfn;
4359 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004360
Tejun Heoc13291a2011-07-12 10:46:30 +02004361 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
4362 start_pfn = min(start_pfn, max_low_pfn);
4363 end_pfn = min(end_pfn, max_low_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004364
Tejun Heoc13291a2011-07-12 10:46:30 +02004365 if (start_pfn < end_pfn)
4366 free_bootmem_node(NODE_DATA(this_nid),
4367 PFN_PHYS(start_pfn),
4368 (end_pfn - start_pfn) << PAGE_SHIFT);
Mel Gormanc7132162006-09-27 01:49:43 -07004369 }
4370}
4371
4372/**
4373 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004374 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07004375 *
4376 * If an architecture guarantees that all ranges registered with
4377 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004378 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07004379 */
4380void __init sparse_memory_present_with_active_regions(int nid)
4381{
Tejun Heoc13291a2011-07-12 10:46:30 +02004382 unsigned long start_pfn, end_pfn;
4383 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004384
Tejun Heoc13291a2011-07-12 10:46:30 +02004385 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
4386 memory_present(this_nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004387}
4388
4389/**
4390 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004391 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4392 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4393 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07004394 *
4395 * It returns the start and end page frame of a node based on information
4396 * provided by an arch calling add_active_range(). If called for a node
4397 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004398 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07004399 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004400void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004401 unsigned long *start_pfn, unsigned long *end_pfn)
4402{
Tejun Heoc13291a2011-07-12 10:46:30 +02004403 unsigned long this_start_pfn, this_end_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004404 int i;
Tejun Heoc13291a2011-07-12 10:46:30 +02004405
Mel Gormanc7132162006-09-27 01:49:43 -07004406 *start_pfn = -1UL;
4407 *end_pfn = 0;
4408
Tejun Heoc13291a2011-07-12 10:46:30 +02004409 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
4410 *start_pfn = min(*start_pfn, this_start_pfn);
4411 *end_pfn = max(*end_pfn, this_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004412 }
4413
Christoph Lameter633c0662007-10-16 01:25:37 -07004414 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07004415 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07004416}
4417
4418/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07004419 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4420 * assumption is made that zones within a node are ordered in monotonic
4421 * increasing memory addresses so that the "highest" populated zone is used
4422 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004423static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004424{
4425 int zone_index;
4426 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
4427 if (zone_index == ZONE_MOVABLE)
4428 continue;
4429
4430 if (arch_zone_highest_possible_pfn[zone_index] >
4431 arch_zone_lowest_possible_pfn[zone_index])
4432 break;
4433 }
4434
4435 VM_BUG_ON(zone_index == -1);
4436 movable_zone = zone_index;
4437}
4438
4439/*
4440 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004441 * because it is sized independent of architecture. Unlike the other zones,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004442 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4443 * in each node depending on the size of each node and how evenly kernelcore
4444 * is distributed. This helper function adjusts the zone ranges
4445 * provided by the architecture for a given node by using the end of the
4446 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4447 * zones within a node are in order of monotonic increases memory addresses
4448 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004449static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004450 unsigned long zone_type,
4451 unsigned long node_start_pfn,
4452 unsigned long node_end_pfn,
4453 unsigned long *zone_start_pfn,
4454 unsigned long *zone_end_pfn)
4455{
4456 /* Only adjust if ZONE_MOVABLE is on this node */
4457 if (zone_movable_pfn[nid]) {
4458 /* Size ZONE_MOVABLE */
4459 if (zone_type == ZONE_MOVABLE) {
4460 *zone_start_pfn = zone_movable_pfn[nid];
4461 *zone_end_pfn = min(node_end_pfn,
4462 arch_zone_highest_possible_pfn[movable_zone]);
4463
4464 /* Adjust for ZONE_MOVABLE starting within this range */
4465 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
4466 *zone_end_pfn > zone_movable_pfn[nid]) {
4467 *zone_end_pfn = zone_movable_pfn[nid];
4468
4469 /* Check if this whole range is within ZONE_MOVABLE */
4470 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
4471 *zone_start_pfn = *zone_end_pfn;
4472 }
4473}
4474
4475/*
Mel Gormanc7132162006-09-27 01:49:43 -07004476 * Return the number of pages a zone spans in a node, including holes
4477 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4478 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004479static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004480 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004481 unsigned long node_start_pfn,
4482 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004483 unsigned long *ignored)
4484{
Mel Gormanc7132162006-09-27 01:49:43 -07004485 unsigned long zone_start_pfn, zone_end_pfn;
4486
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004487 /* Get the start and end of the zone */
Mel Gormanc7132162006-09-27 01:49:43 -07004488 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4489 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07004490 adjust_zone_range_for_zone_movable(nid, zone_type,
4491 node_start_pfn, node_end_pfn,
4492 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004493
4494 /* Check that this node has pages within the zone's required range */
4495 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4496 return 0;
4497
4498 /* Move the zone boundaries inside the node if necessary */
4499 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4500 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4501
4502 /* Return the spanned pages */
4503 return zone_end_pfn - zone_start_pfn;
4504}
4505
4506/*
4507 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004508 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07004509 */
Yinghai Lu32996252009-12-15 17:59:02 -08004510unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004511 unsigned long range_start_pfn,
4512 unsigned long range_end_pfn)
4513{
Tejun Heo96e907d2011-07-12 10:46:29 +02004514 unsigned long nr_absent = range_end_pfn - range_start_pfn;
4515 unsigned long start_pfn, end_pfn;
4516 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004517
Tejun Heo96e907d2011-07-12 10:46:29 +02004518 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
4519 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
4520 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
4521 nr_absent -= end_pfn - start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004522 }
Tejun Heo96e907d2011-07-12 10:46:29 +02004523 return nr_absent;
Mel Gormanc7132162006-09-27 01:49:43 -07004524}
4525
4526/**
4527 * absent_pages_in_range - Return number of page frames in holes within a range
4528 * @start_pfn: The start PFN to start searching for holes
4529 * @end_pfn: The end PFN to stop searching for holes
4530 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004531 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07004532 */
4533unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4534 unsigned long end_pfn)
4535{
4536 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4537}
4538
4539/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004540static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004541 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004542 unsigned long node_start_pfn,
4543 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004544 unsigned long *ignored)
4545{
Tejun Heo96e907d2011-07-12 10:46:29 +02004546 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
4547 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman9c7cd682006-09-27 01:49:58 -07004548 unsigned long zone_start_pfn, zone_end_pfn;
4549
Tejun Heo96e907d2011-07-12 10:46:29 +02004550 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
4551 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004552
Mel Gorman2a1e2742007-07-17 04:03:12 -07004553 adjust_zone_range_for_zone_movable(nid, zone_type,
4554 node_start_pfn, node_end_pfn,
4555 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004556 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004557}
Mel Gorman0e0b8642006-09-27 01:49:56 -07004558
Tejun Heo0ee332c2011-12-08 10:22:09 -08004559#else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004560static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004561 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004562 unsigned long node_start_pfn,
4563 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004564 unsigned long *zones_size)
4565{
4566 return zones_size[zone_type];
4567}
4568
Paul Mundt6ea6e682007-07-15 23:38:20 -07004569static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004570 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004571 unsigned long node_start_pfn,
4572 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004573 unsigned long *zholes_size)
4574{
4575 if (!zholes_size)
4576 return 0;
4577
4578 return zholes_size[zone_type];
4579}
Yinghai Lu20e69262013-03-01 14:51:27 -08004580
Tejun Heo0ee332c2011-12-08 10:22:09 -08004581#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07004582
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004583static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004584 unsigned long node_start_pfn,
4585 unsigned long node_end_pfn,
4586 unsigned long *zones_size,
4587 unsigned long *zholes_size)
Mel Gormanc7132162006-09-27 01:49:43 -07004588{
4589 unsigned long realtotalpages, totalpages = 0;
4590 enum zone_type i;
4591
4592 for (i = 0; i < MAX_NR_ZONES; i++)
4593 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004594 node_start_pfn,
4595 node_end_pfn,
4596 zones_size);
Mel Gormanc7132162006-09-27 01:49:43 -07004597 pgdat->node_spanned_pages = totalpages;
4598
4599 realtotalpages = totalpages;
4600 for (i = 0; i < MAX_NR_ZONES; i++)
4601 realtotalpages -=
4602 zone_absent_pages_in_node(pgdat->node_id, i,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004603 node_start_pfn, node_end_pfn,
4604 zholes_size);
Mel Gormanc7132162006-09-27 01:49:43 -07004605 pgdat->node_present_pages = realtotalpages;
4606 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4607 realtotalpages);
4608}
4609
Mel Gorman835c1342007-10-16 01:25:47 -07004610#ifndef CONFIG_SPARSEMEM
4611/*
4612 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07004613 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4614 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07004615 * round what is now in bits to nearest long in bits, then return it in
4616 * bytes.
4617 */
Linus Torvalds7c455122013-02-18 09:58:02 -08004618static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07004619{
4620 unsigned long usemapsize;
4621
Linus Torvalds7c455122013-02-18 09:58:02 -08004622 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
Mel Gormand9c23402007-10-16 01:26:01 -07004623 usemapsize = roundup(zonesize, pageblock_nr_pages);
4624 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07004625 usemapsize *= NR_PAGEBLOCK_BITS;
4626 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4627
4628 return usemapsize / 8;
4629}
4630
4631static void __init setup_usemap(struct pglist_data *pgdat,
Linus Torvalds7c455122013-02-18 09:58:02 -08004632 struct zone *zone,
4633 unsigned long zone_start_pfn,
4634 unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07004635{
Linus Torvalds7c455122013-02-18 09:58:02 -08004636 unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
Mel Gorman835c1342007-10-16 01:25:47 -07004637 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08004638 if (usemapsize)
Yinghai Lu8f389a992011-05-11 15:13:32 -07004639 zone->pageblock_flags = alloc_bootmem_node_nopanic(pgdat,
4640 usemapsize);
Mel Gorman835c1342007-10-16 01:25:47 -07004641}
4642#else
Linus Torvalds7c455122013-02-18 09:58:02 -08004643static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
4644 unsigned long zone_start_pfn, unsigned long zonesize) {}
Mel Gorman835c1342007-10-16 01:25:47 -07004645#endif /* CONFIG_SPARSEMEM */
4646
Mel Gormand9c23402007-10-16 01:26:01 -07004647#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08004648
Mel Gormand9c23402007-10-16 01:26:01 -07004649/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
Chen Gang15ca2202013-09-11 14:20:27 -07004650void __paginginit set_pageblock_order(void)
Mel Gormand9c23402007-10-16 01:26:01 -07004651{
Andrew Morton955c1cd2012-05-29 15:06:31 -07004652 unsigned int order;
4653
Mel Gormand9c23402007-10-16 01:26:01 -07004654 /* Check that pageblock_nr_pages has not already been setup */
4655 if (pageblock_order)
4656 return;
4657
Andrew Morton955c1cd2012-05-29 15:06:31 -07004658 if (HPAGE_SHIFT > PAGE_SHIFT)
4659 order = HUGETLB_PAGE_ORDER;
4660 else
4661 order = MAX_ORDER - 1;
4662
Mel Gormand9c23402007-10-16 01:26:01 -07004663 /*
4664 * Assume the largest contiguous order of interest is a huge page.
Andrew Morton955c1cd2012-05-29 15:06:31 -07004665 * This value may be variable depending on boot parameters on IA64 and
4666 * powerpc.
Mel Gormand9c23402007-10-16 01:26:01 -07004667 */
4668 pageblock_order = order;
4669}
4670#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4671
Mel Gormanba72cb82007-11-28 16:21:13 -08004672/*
4673 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
Andrew Morton955c1cd2012-05-29 15:06:31 -07004674 * is unused as pageblock_order is set at compile-time. See
4675 * include/linux/pageblock-flags.h for the values of pageblock_order based on
4676 * the kernel config
Mel Gormanba72cb82007-11-28 16:21:13 -08004677 */
Chen Gang15ca2202013-09-11 14:20:27 -07004678void __paginginit set_pageblock_order(void)
Mel Gormanba72cb82007-11-28 16:21:13 -08004679{
Mel Gormanba72cb82007-11-28 16:21:13 -08004680}
Mel Gormand9c23402007-10-16 01:26:01 -07004681
4682#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4683
Jiang Liu01cefae2012-12-12 13:52:19 -08004684static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
4685 unsigned long present_pages)
4686{
4687 unsigned long pages = spanned_pages;
4688
4689 /*
4690 * Provide a more accurate estimation if there are holes within
4691 * the zone and SPARSEMEM is in use. If there are holes within the
4692 * zone, each populated memory region may cost us one or two extra
4693 * memmap pages due to alignment because memmap pages for each
4694 * populated regions may not naturally algined on page boundary.
4695 * So the (present_pages >> 4) heuristic is a tradeoff for that.
4696 */
4697 if (spanned_pages > present_pages + (present_pages >> 4) &&
4698 IS_ENABLED(CONFIG_SPARSEMEM))
4699 pages = present_pages;
4700
4701 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
4702}
4703
Linus Torvalds1da177e2005-04-16 15:20:36 -07004704/*
4705 * Set up the zone data structures:
4706 * - mark all pages reserved
4707 * - mark all memory queues empty
4708 * - clear the memory bitmaps
Minchan Kim6527af52012-07-31 16:46:16 -07004709 *
4710 * NOTE: pgdat should get zeroed by caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004711 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08004712static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004713 unsigned long node_start_pfn, unsigned long node_end_pfn,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004714 unsigned long *zones_size, unsigned long *zholes_size)
4715{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07004716 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07004717 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004718 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07004719 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004720
Dave Hansen208d54e2005-10-29 18:16:52 -07004721 pgdat_resize_init(pgdat);
Andrea Arcangeli8177a422012-03-23 20:56:34 +01004722#ifdef CONFIG_NUMA_BALANCING
4723 spin_lock_init(&pgdat->numabalancing_migrate_lock);
4724 pgdat->numabalancing_migrate_nr_pages = 0;
4725 pgdat->numabalancing_migrate_next_window = jiffies;
4726#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004727 init_waitqueue_head(&pgdat->kswapd_wait);
Mel Gorman55150612012-07-31 16:44:35 -07004728 init_waitqueue_head(&pgdat->pfmemalloc_wait);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004729 pgdat_page_cgroup_init(pgdat);
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01004730
Linus Torvalds1da177e2005-04-16 15:20:36 -07004731 for (j = 0; j < MAX_NR_ZONES; j++) {
4732 struct zone *zone = pgdat->node_zones + j;
Jiang Liu9feedc92012-12-12 13:52:12 -08004733 unsigned long size, realsize, freesize, memmap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004734
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004735 size = zone_spanned_pages_in_node(nid, j, node_start_pfn,
4736 node_end_pfn, zones_size);
Jiang Liu9feedc92012-12-12 13:52:12 -08004737 realsize = freesize = size - zone_absent_pages_in_node(nid, j,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004738 node_start_pfn,
4739 node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004740 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004741
Mel Gorman0e0b8642006-09-27 01:49:56 -07004742 /*
Jiang Liu9feedc92012-12-12 13:52:12 -08004743 * Adjust freesize so that it accounts for how much memory
Mel Gorman0e0b8642006-09-27 01:49:56 -07004744 * is used by this zone for memmap. This affects the watermark
4745 * and per-cpu initialisations
4746 */
Jiang Liu01cefae2012-12-12 13:52:19 -08004747 memmap_pages = calc_memmap_size(size, realsize);
Jiang Liu9feedc92012-12-12 13:52:12 -08004748 if (freesize >= memmap_pages) {
4749 freesize -= memmap_pages;
Yinghai Lu5594c8c2009-01-06 14:39:14 -08004750 if (memmap_pages)
4751 printk(KERN_DEBUG
4752 " %s zone: %lu pages used for memmap\n",
4753 zone_names[j], memmap_pages);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004754 } else
4755 printk(KERN_WARNING
Jiang Liu9feedc92012-12-12 13:52:12 -08004756 " %s zone: %lu pages exceeds freesize %lu\n",
4757 zone_names[j], memmap_pages, freesize);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004758
Christoph Lameter62672762007-02-10 01:43:07 -08004759 /* Account for reserved pages */
Jiang Liu9feedc92012-12-12 13:52:12 -08004760 if (j == 0 && freesize > dma_reserve) {
4761 freesize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07004762 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08004763 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004764 }
4765
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004766 if (!is_highmem_idx(j))
Jiang Liu9feedc92012-12-12 13:52:12 -08004767 nr_kernel_pages += freesize;
Jiang Liu01cefae2012-12-12 13:52:19 -08004768 /* Charge for highmem memmap if there are enough kernel pages */
4769 else if (nr_kernel_pages > memmap_pages * 2)
4770 nr_kernel_pages -= memmap_pages;
Jiang Liu9feedc92012-12-12 13:52:12 -08004771 nr_all_pages += freesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004772
4773 zone->spanned_pages = size;
Jiang Liu306f2e92013-02-22 16:33:54 -08004774 zone->present_pages = realsize;
Jiang Liu9feedc92012-12-12 13:52:12 -08004775 /*
4776 * Set an approximate value for lowmem here, it will be adjusted
4777 * when the bootmem allocator frees pages into the buddy system.
4778 * And all highmem pages will be managed by the buddy system.
4779 */
4780 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
Christoph Lameter96146342006-07-03 00:24:13 -07004781#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07004782 zone->node = nid;
Jiang Liu9feedc92012-12-12 13:52:12 -08004783 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07004784 / 100;
Jiang Liu9feedc92012-12-12 13:52:12 -08004785 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07004786#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004787 zone->name = zone_names[j];
4788 spin_lock_init(&zone->lock);
4789 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07004790 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004791 zone->zone_pgdat = pgdat;
Dave Hansened8ece22005-10-29 18:16:50 -07004792 zone_pcp_init(zone);
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07004793
4794 /* For bootup, initialized properly in watermark setup */
4795 mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages);
4796
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004797 lruvec_init(&zone->lruvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004798 if (!size)
4799 continue;
4800
Andrew Morton955c1cd2012-05-29 15:06:31 -07004801 set_pageblock_order();
Linus Torvalds7c455122013-02-18 09:58:02 -08004802 setup_usemap(pgdat, zone, zone_start_pfn, size);
Dave Hansena2f3aa022007-01-10 23:15:30 -08004803 ret = init_currently_empty_zone(zone, zone_start_pfn,
4804 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07004805 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07004806 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004807 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004808 }
4809}
4810
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004811static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004812{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004813 /* Skip empty nodes */
4814 if (!pgdat->node_spanned_pages)
4815 return;
4816
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004817#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07004818 /* ia64 gets its own node_mem_map, before this, without bootmem */
4819 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07004820 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004821 struct page *map;
4822
Bob Piccoe984bb42006-05-20 15:00:31 -07004823 /*
4824 * The zone's endpoints aren't required to be MAX_ORDER
4825 * aligned but the node_mem_map endpoints must be in order
4826 * for the buddy allocator to function correctly.
4827 */
4828 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
Cody P Schafer108bcc92013-02-22 16:35:23 -08004829 end = pgdat_end_pfn(pgdat);
Bob Piccoe984bb42006-05-20 15:00:31 -07004830 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4831 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07004832 map = alloc_remap(pgdat->node_id, size);
4833 if (!map)
Yinghai Lu8f389a992011-05-11 15:13:32 -07004834 map = alloc_bootmem_node_nopanic(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07004835 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004836 }
Roman Zippel12d810c2007-05-31 00:40:54 -07004837#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07004838 /*
4839 * With no DISCONTIG, the global mem_map is just set as node 0's
4840 */
Mel Gormanc7132162006-09-27 01:49:43 -07004841 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004842 mem_map = NODE_DATA(0)->node_mem_map;
Tejun Heo0ee332c2011-12-08 10:22:09 -08004843#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07004844 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08004845 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Tejun Heo0ee332c2011-12-08 10:22:09 -08004846#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07004847 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004848#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004849#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004850}
4851
Johannes Weiner9109fb72008-07-23 21:27:20 -07004852void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4853 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004854{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004855 pg_data_t *pgdat = NODE_DATA(nid);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004856 unsigned long start_pfn = 0;
4857 unsigned long end_pfn = 0;
Johannes Weiner9109fb72008-07-23 21:27:20 -07004858
Minchan Kim88fdf752012-07-31 16:46:14 -07004859 /* pg_data_t should be reset to zero when it's allocated */
Linus Torvalds8783b6e2012-08-02 10:37:03 -07004860 WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
Minchan Kim88fdf752012-07-31 16:46:14 -07004861
Linus Torvalds1da177e2005-04-16 15:20:36 -07004862 pgdat->node_id = nid;
4863 pgdat->node_start_pfn = node_start_pfn;
David Rientjes957f8222012-10-08 16:33:24 -07004864 init_zone_allows_reclaim(nid);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004865#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4866 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
4867#endif
4868 calculate_node_totalpages(pgdat, start_pfn, end_pfn,
4869 zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004870
4871 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07004872#ifdef CONFIG_FLAT_NODE_MEM_MAP
4873 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4874 nid, (unsigned long)pgdat,
4875 (unsigned long)pgdat->node_mem_map);
4876#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004877
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004878 free_area_init_core(pgdat, start_pfn, end_pfn,
4879 zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004880}
4881
Tejun Heo0ee332c2011-12-08 10:22:09 -08004882#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07004883
4884#if MAX_NUMNODES > 1
4885/*
4886 * Figure out the number of possible node ids.
4887 */
Cody P Schaferf9872ca2013-04-29 15:08:01 -07004888void __init setup_nr_node_ids(void)
Miklos Szeredi418508c2007-05-23 13:57:55 -07004889{
4890 unsigned int node;
4891 unsigned int highest = 0;
4892
4893 for_each_node_mask(node, node_possible_map)
4894 highest = node;
4895 nr_node_ids = highest + 1;
4896}
Miklos Szeredi418508c2007-05-23 13:57:55 -07004897#endif
4898
Mel Gormanc7132162006-09-27 01:49:43 -07004899/**
Tejun Heo1e019792011-07-12 09:45:34 +02004900 * node_map_pfn_alignment - determine the maximum internode alignment
4901 *
4902 * This function should be called after node map is populated and sorted.
4903 * It calculates the maximum power of two alignment which can distinguish
4904 * all the nodes.
4905 *
4906 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
4907 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
4908 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
4909 * shifted, 1GiB is enough and this function will indicate so.
4910 *
4911 * This is used to test whether pfn -> nid mapping of the chosen memory
4912 * model has fine enough granularity to avoid incorrect mapping for the
4913 * populated node map.
4914 *
4915 * Returns the determined alignment in pfn's. 0 if there is no alignment
4916 * requirement (single node).
4917 */
4918unsigned long __init node_map_pfn_alignment(void)
4919{
4920 unsigned long accl_mask = 0, last_end = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02004921 unsigned long start, end, mask;
Tejun Heo1e019792011-07-12 09:45:34 +02004922 int last_nid = -1;
Tejun Heoc13291a2011-07-12 10:46:30 +02004923 int i, nid;
Tejun Heo1e019792011-07-12 09:45:34 +02004924
Tejun Heoc13291a2011-07-12 10:46:30 +02004925 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
Tejun Heo1e019792011-07-12 09:45:34 +02004926 if (!start || last_nid < 0 || last_nid == nid) {
4927 last_nid = nid;
4928 last_end = end;
4929 continue;
4930 }
4931
4932 /*
4933 * Start with a mask granular enough to pin-point to the
4934 * start pfn and tick off bits one-by-one until it becomes
4935 * too coarse to separate the current node from the last.
4936 */
4937 mask = ~((1 << __ffs(start)) - 1);
4938 while (mask && last_end <= (start & (mask << 1)))
4939 mask <<= 1;
4940
4941 /* accumulate all internode masks */
4942 accl_mask |= mask;
4943 }
4944
4945 /* convert mask to number of pages */
4946 return ~accl_mask + 1;
4947}
4948
Mel Gormana6af2bc2007-02-10 01:42:57 -08004949/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004950static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07004951{
Mel Gormana6af2bc2007-02-10 01:42:57 -08004952 unsigned long min_pfn = ULONG_MAX;
Tejun Heoc13291a2011-07-12 10:46:30 +02004953 unsigned long start_pfn;
4954 int i;
Mel Gorman1abbfb42006-11-23 12:01:41 +00004955
Tejun Heoc13291a2011-07-12 10:46:30 +02004956 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
4957 min_pfn = min(min_pfn, start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004958
Mel Gormana6af2bc2007-02-10 01:42:57 -08004959 if (min_pfn == ULONG_MAX) {
4960 printk(KERN_WARNING
Paul Jackson2bc0d262008-06-22 07:22:17 -07004961 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08004962 return 0;
4963 }
4964
4965 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004966}
4967
4968/**
4969 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4970 *
4971 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004972 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07004973 */
4974unsigned long __init find_min_pfn_with_active_regions(void)
4975{
4976 return find_min_pfn_for_node(MAX_NUMNODES);
4977}
4978
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004979/*
4980 * early_calculate_totalpages()
4981 * Sum pages in active regions for movable zone.
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08004982 * Populate N_MEMORY for calculating usable_nodes.
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004983 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07004984static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07004985{
Mel Gorman7e63efe2007-07-17 04:03:15 -07004986 unsigned long totalpages = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02004987 unsigned long start_pfn, end_pfn;
4988 int i, nid;
Mel Gorman7e63efe2007-07-17 04:03:15 -07004989
Tejun Heoc13291a2011-07-12 10:46:30 +02004990 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
4991 unsigned long pages = end_pfn - start_pfn;
4992
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004993 totalpages += pages;
4994 if (pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08004995 node_set_state(nid, N_MEMORY);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004996 }
Pintu Kumarb8af2942013-09-11 14:20:34 -07004997 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07004998}
4999
Mel Gorman2a1e2742007-07-17 04:03:12 -07005000/*
5001 * Find the PFN the Movable zone begins in each node. Kernel memory
5002 * is spread evenly between nodes as long as the nodes have enough
5003 * memory. When they don't, some nodes will have more kernelcore than
5004 * others
5005 */
Kautuk Consulb224ef82012-03-21 16:34:15 -07005006static void __init find_zone_movable_pfns_for_nodes(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005007{
5008 int i, nid;
5009 unsigned long usable_startpfn;
5010 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07005011 /* save the state before borrow the nodemask */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005012 nodemask_t saved_node_state = node_states[N_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005013 unsigned long totalpages = early_calculate_totalpages();
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005014 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005015
Mel Gorman7e63efe2007-07-17 04:03:15 -07005016 /*
5017 * If movablecore was specified, calculate what size of
5018 * kernelcore that corresponds so that memory usable for
5019 * any allocation type is evenly spread. If both kernelcore
5020 * and movablecore are specified, then the value of kernelcore
5021 * will be used for required_kernelcore if it's greater than
5022 * what movablecore would have allowed.
5023 */
5024 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07005025 unsigned long corepages;
5026
5027 /*
5028 * Round-up so that ZONE_MOVABLE is at least as large as what
5029 * was requested by the user
5030 */
5031 required_movablecore =
5032 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
5033 corepages = totalpages - required_movablecore;
5034
5035 required_kernelcore = max(required_kernelcore, corepages);
5036 }
5037
Yinghai Lu20e69262013-03-01 14:51:27 -08005038 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
5039 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07005040 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005041
5042 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
Yinghai Lu20e69262013-03-01 14:51:27 -08005043 find_usable_zone_for_movable();
Mel Gorman2a1e2742007-07-17 04:03:12 -07005044 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
5045
5046restart:
5047 /* Spread kernelcore memory as evenly as possible throughout nodes */
5048 kernelcore_node = required_kernelcore / usable_nodes;
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005049 for_each_node_state(nid, N_MEMORY) {
Tejun Heoc13291a2011-07-12 10:46:30 +02005050 unsigned long start_pfn, end_pfn;
5051
Mel Gorman2a1e2742007-07-17 04:03:12 -07005052 /*
5053 * Recalculate kernelcore_node if the division per node
5054 * now exceeds what is necessary to satisfy the requested
5055 * amount of memory for the kernel
5056 */
5057 if (required_kernelcore < kernelcore_node)
5058 kernelcore_node = required_kernelcore / usable_nodes;
5059
5060 /*
5061 * As the map is walked, we track how much memory is usable
5062 * by the kernel using kernelcore_remaining. When it is
5063 * 0, the rest of the node is usable by ZONE_MOVABLE
5064 */
5065 kernelcore_remaining = kernelcore_node;
5066
5067 /* Go through each range of PFNs within this node */
Tejun Heoc13291a2011-07-12 10:46:30 +02005068 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005069 unsigned long size_pages;
5070
Tejun Heoc13291a2011-07-12 10:46:30 +02005071 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005072 if (start_pfn >= end_pfn)
5073 continue;
5074
5075 /* Account for what is only usable for kernelcore */
5076 if (start_pfn < usable_startpfn) {
5077 unsigned long kernel_pages;
5078 kernel_pages = min(end_pfn, usable_startpfn)
5079 - start_pfn;
5080
5081 kernelcore_remaining -= min(kernel_pages,
5082 kernelcore_remaining);
5083 required_kernelcore -= min(kernel_pages,
5084 required_kernelcore);
5085
5086 /* Continue if range is now fully accounted */
5087 if (end_pfn <= usable_startpfn) {
5088
5089 /*
5090 * Push zone_movable_pfn to the end so
5091 * that if we have to rebalance
5092 * kernelcore across nodes, we will
5093 * not double account here
5094 */
5095 zone_movable_pfn[nid] = end_pfn;
5096 continue;
5097 }
5098 start_pfn = usable_startpfn;
5099 }
5100
5101 /*
5102 * The usable PFN range for ZONE_MOVABLE is from
5103 * start_pfn->end_pfn. Calculate size_pages as the
5104 * number of pages used as kernelcore
5105 */
5106 size_pages = end_pfn - start_pfn;
5107 if (size_pages > kernelcore_remaining)
5108 size_pages = kernelcore_remaining;
5109 zone_movable_pfn[nid] = start_pfn + size_pages;
5110
5111 /*
5112 * Some kernelcore has been met, update counts and
5113 * break if the kernelcore for this node has been
Pintu Kumarb8af2942013-09-11 14:20:34 -07005114 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07005115 */
5116 required_kernelcore -= min(required_kernelcore,
5117 size_pages);
5118 kernelcore_remaining -= size_pages;
5119 if (!kernelcore_remaining)
5120 break;
5121 }
5122 }
5123
5124 /*
5125 * If there is still required_kernelcore, we do another pass with one
5126 * less node in the count. This will push zone_movable_pfn[nid] further
5127 * along on the nodes that still have memory until kernelcore is
Pintu Kumarb8af2942013-09-11 14:20:34 -07005128 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07005129 */
5130 usable_nodes--;
5131 if (usable_nodes && required_kernelcore > usable_nodes)
5132 goto restart;
5133
5134 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5135 for (nid = 0; nid < MAX_NUMNODES; nid++)
5136 zone_movable_pfn[nid] =
5137 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07005138
Yinghai Lu20e69262013-03-01 14:51:27 -08005139out:
Yinghai Lu66918dc2009-06-30 11:41:37 -07005140 /* restore the node_state */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005141 node_states[N_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005142}
5143
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005144/* Any regular or high memory on that node ? */
5145static void check_for_memory(pg_data_t *pgdat, int nid)
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005146{
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005147 enum zone_type zone_type;
5148
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005149 if (N_MEMORY == N_NORMAL_MEMORY)
5150 return;
5151
5152 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005153 struct zone *zone = &pgdat->node_zones[zone_type];
Xishi Qiub38a8722013-11-12 15:07:20 -08005154 if (populated_zone(zone)) {
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005155 node_set_state(nid, N_HIGH_MEMORY);
5156 if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
5157 zone_type <= ZONE_NORMAL)
5158 node_set_state(nid, N_NORMAL_MEMORY);
Bob Liud0048b02012-01-12 17:19:07 -08005159 break;
5160 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005161 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005162}
5163
Mel Gormanc7132162006-09-27 01:49:43 -07005164/**
5165 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005166 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07005167 *
5168 * This will call free_area_init_node() for each active node in the system.
5169 * Using the page ranges provided by add_active_range(), the size of each
5170 * zone in each node and their holes is calculated. If the maximum PFN
5171 * between two adjacent zones match, it is assumed that the zone is empty.
5172 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5173 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5174 * starts where the previous one ended. For example, ZONE_DMA32 starts
5175 * at arch_max_dma_pfn.
5176 */
5177void __init free_area_init_nodes(unsigned long *max_zone_pfn)
5178{
Tejun Heoc13291a2011-07-12 10:46:30 +02005179 unsigned long start_pfn, end_pfn;
5180 int i, nid;
Mel Gormana6af2bc2007-02-10 01:42:57 -08005181
Mel Gormanc7132162006-09-27 01:49:43 -07005182 /* Record where the zone boundaries are */
5183 memset(arch_zone_lowest_possible_pfn, 0,
5184 sizeof(arch_zone_lowest_possible_pfn));
5185 memset(arch_zone_highest_possible_pfn, 0,
5186 sizeof(arch_zone_highest_possible_pfn));
5187 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
5188 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
5189 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005190 if (i == ZONE_MOVABLE)
5191 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07005192 arch_zone_lowest_possible_pfn[i] =
5193 arch_zone_highest_possible_pfn[i-1];
5194 arch_zone_highest_possible_pfn[i] =
5195 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
5196 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07005197 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
5198 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
5199
5200 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5201 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
Kautuk Consulb224ef82012-03-21 16:34:15 -07005202 find_zone_movable_pfns_for_nodes();
Mel Gormanc7132162006-09-27 01:49:43 -07005203
Mel Gormanc7132162006-09-27 01:49:43 -07005204 /* Print out the zone ranges */
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005205 printk("Zone ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005206 for (i = 0; i < MAX_NR_ZONES; i++) {
5207 if (i == ZONE_MOVABLE)
5208 continue;
Kay Sievers155cbfc2012-05-08 17:24:14 +02005209 printk(KERN_CONT " %-8s ", zone_names[i]);
David Rientjes72f0ba02010-03-05 13:42:14 -08005210 if (arch_zone_lowest_possible_pfn[i] ==
5211 arch_zone_highest_possible_pfn[i])
Kay Sievers155cbfc2012-05-08 17:24:14 +02005212 printk(KERN_CONT "empty\n");
David Rientjes72f0ba02010-03-05 13:42:14 -08005213 else
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005214 printk(KERN_CONT "[mem %0#10lx-%0#10lx]\n",
5215 arch_zone_lowest_possible_pfn[i] << PAGE_SHIFT,
5216 (arch_zone_highest_possible_pfn[i]
5217 << PAGE_SHIFT) - 1);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005218 }
5219
5220 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005221 printk("Movable zone start for each node\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005222 for (i = 0; i < MAX_NUMNODES; i++) {
5223 if (zone_movable_pfn[i])
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005224 printk(" Node %d: %#010lx\n", i,
5225 zone_movable_pfn[i] << PAGE_SHIFT);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005226 }
Mel Gormanc7132162006-09-27 01:49:43 -07005227
Wanpeng Lif2d52fe2012-10-08 16:32:24 -07005228 /* Print out the early node map */
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005229 printk("Early memory node ranges\n");
Tejun Heoc13291a2011-07-12 10:46:30 +02005230 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005231 printk(" node %3d: [mem %#010lx-%#010lx]\n", nid,
5232 start_pfn << PAGE_SHIFT, (end_pfn << PAGE_SHIFT) - 1);
Mel Gormanc7132162006-09-27 01:49:43 -07005233
5234 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07005235 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08005236 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07005237 for_each_online_node(nid) {
5238 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07005239 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07005240 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005241
5242 /* Any memory on that node */
5243 if (pgdat->node_present_pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005244 node_set_state(nid, N_MEMORY);
5245 check_for_memory(pgdat, nid);
Mel Gormanc7132162006-09-27 01:49:43 -07005246 }
5247}
Mel Gorman2a1e2742007-07-17 04:03:12 -07005248
Mel Gorman7e63efe2007-07-17 04:03:15 -07005249static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005250{
5251 unsigned long long coremem;
5252 if (!p)
5253 return -EINVAL;
5254
5255 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005256 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005257
Mel Gorman7e63efe2007-07-17 04:03:15 -07005258 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07005259 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
5260
5261 return 0;
5262}
Mel Gormaned7ed362007-07-17 04:03:14 -07005263
Mel Gorman7e63efe2007-07-17 04:03:15 -07005264/*
5265 * kernelcore=size sets the amount of memory for use for allocations that
5266 * cannot be reclaimed or migrated.
5267 */
5268static int __init cmdline_parse_kernelcore(char *p)
5269{
5270 return cmdline_parse_core(p, &required_kernelcore);
5271}
5272
5273/*
5274 * movablecore=size sets the amount of memory for use for allocations that
5275 * can be reclaimed or migrated.
5276 */
5277static int __init cmdline_parse_movablecore(char *p)
5278{
5279 return cmdline_parse_core(p, &required_movablecore);
5280}
5281
Mel Gormaned7ed362007-07-17 04:03:14 -07005282early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005283early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07005284
Tejun Heo0ee332c2011-12-08 10:22:09 -08005285#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005286
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005287void adjust_managed_page_count(struct page *page, long count)
5288{
5289 spin_lock(&managed_page_count_lock);
5290 page_zone(page)->managed_pages += count;
5291 totalram_pages += count;
Jiang Liu3dcc0572013-07-03 15:03:21 -07005292#ifdef CONFIG_HIGHMEM
5293 if (PageHighMem(page))
5294 totalhigh_pages += count;
5295#endif
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005296 spin_unlock(&managed_page_count_lock);
5297}
Jiang Liu3dcc0572013-07-03 15:03:21 -07005298EXPORT_SYMBOL(adjust_managed_page_count);
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005299
Jiang Liu11199692013-07-03 15:02:48 -07005300unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
Jiang Liu69afade2013-04-29 15:06:21 -07005301{
Jiang Liu11199692013-07-03 15:02:48 -07005302 void *pos;
5303 unsigned long pages = 0;
Jiang Liu69afade2013-04-29 15:06:21 -07005304
Jiang Liu11199692013-07-03 15:02:48 -07005305 start = (void *)PAGE_ALIGN((unsigned long)start);
5306 end = (void *)((unsigned long)end & PAGE_MASK);
5307 for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
Jiang Liudbe67df2013-07-03 15:02:51 -07005308 if ((unsigned int)poison <= 0xFF)
Jiang Liu11199692013-07-03 15:02:48 -07005309 memset(pos, poison, PAGE_SIZE);
5310 free_reserved_page(virt_to_page(pos));
Jiang Liu69afade2013-04-29 15:06:21 -07005311 }
5312
5313 if (pages && s)
Jiang Liu11199692013-07-03 15:02:48 -07005314 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
Jiang Liu69afade2013-04-29 15:06:21 -07005315 s, pages << (PAGE_SHIFT - 10), start, end);
5316
5317 return pages;
5318}
Jiang Liu11199692013-07-03 15:02:48 -07005319EXPORT_SYMBOL(free_reserved_area);
Jiang Liu69afade2013-04-29 15:06:21 -07005320
Jiang Liucfa11e02013-04-29 15:07:00 -07005321#ifdef CONFIG_HIGHMEM
5322void free_highmem_page(struct page *page)
5323{
5324 __free_reserved_page(page);
5325 totalram_pages++;
Jiang Liu7b4b2a02013-07-03 15:03:11 -07005326 page_zone(page)->managed_pages++;
Jiang Liucfa11e02013-04-29 15:07:00 -07005327 totalhigh_pages++;
5328}
5329#endif
5330
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005331
5332void __init mem_init_print_info(const char *str)
5333{
5334 unsigned long physpages, codesize, datasize, rosize, bss_size;
5335 unsigned long init_code_size, init_data_size;
5336
5337 physpages = get_num_physpages();
5338 codesize = _etext - _stext;
5339 datasize = _edata - _sdata;
5340 rosize = __end_rodata - __start_rodata;
5341 bss_size = __bss_stop - __bss_start;
5342 init_data_size = __init_end - __init_begin;
5343 init_code_size = _einittext - _sinittext;
5344
5345 /*
5346 * Detect special cases and adjust section sizes accordingly:
5347 * 1) .init.* may be embedded into .data sections
5348 * 2) .init.text.* may be out of [__init_begin, __init_end],
5349 * please refer to arch/tile/kernel/vmlinux.lds.S.
5350 * 3) .rodata.* may be embedded into .text or .data sections.
5351 */
5352#define adj_init_size(start, end, size, pos, adj) \
Pintu Kumarb8af2942013-09-11 14:20:34 -07005353 do { \
5354 if (start <= pos && pos < end && size > adj) \
5355 size -= adj; \
5356 } while (0)
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005357
5358 adj_init_size(__init_begin, __init_end, init_data_size,
5359 _sinittext, init_code_size);
5360 adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
5361 adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
5362 adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
5363 adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
5364
5365#undef adj_init_size
5366
5367 printk("Memory: %luK/%luK available "
5368 "(%luK kernel code, %luK rwdata, %luK rodata, "
5369 "%luK init, %luK bss, %luK reserved"
5370#ifdef CONFIG_HIGHMEM
5371 ", %luK highmem"
5372#endif
5373 "%s%s)\n",
5374 nr_free_pages() << (PAGE_SHIFT-10), physpages << (PAGE_SHIFT-10),
5375 codesize >> 10, datasize >> 10, rosize >> 10,
5376 (init_data_size + init_code_size) >> 10, bss_size >> 10,
5377 (physpages - totalram_pages) << (PAGE_SHIFT-10),
5378#ifdef CONFIG_HIGHMEM
5379 totalhigh_pages << (PAGE_SHIFT-10),
5380#endif
5381 str ? ", " : "", str ? str : "");
5382}
5383
Mel Gorman0e0b8642006-09-27 01:49:56 -07005384/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005385 * set_dma_reserve - set the specified number of pages reserved in the first zone
5386 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07005387 *
5388 * The per-cpu batchsize and zone watermarks are determined by present_pages.
5389 * In the DMA zone, a significant percentage may be consumed by kernel image
5390 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005391 * function may optionally be used to account for unfreeable pages in the
5392 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5393 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07005394 */
5395void __init set_dma_reserve(unsigned long new_dma_reserve)
5396{
5397 dma_reserve = new_dma_reserve;
5398}
5399
Linus Torvalds1da177e2005-04-16 15:20:36 -07005400void __init free_area_init(unsigned long *zones_size)
5401{
Johannes Weiner9109fb72008-07-23 21:27:20 -07005402 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005403 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
5404}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005405
Linus Torvalds1da177e2005-04-16 15:20:36 -07005406static int page_alloc_cpu_notify(struct notifier_block *self,
5407 unsigned long action, void *hcpu)
5408{
5409 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005410
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07005411 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -07005412 lru_add_drain_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005413 drain_pages(cpu);
5414
5415 /*
5416 * Spill the event counters of the dead processor
5417 * into the current processors event counters.
5418 * This artificially elevates the count of the current
5419 * processor.
5420 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07005421 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005422
5423 /*
5424 * Zero the differential counters of the dead processor
5425 * so that the vm statistics are consistent.
5426 *
5427 * This is only okay since the processor is dead and cannot
5428 * race with what we are doing.
5429 */
Christoph Lameter2bb921e2013-09-11 14:21:30 -07005430 cpu_vm_stats_fold(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005431 }
5432 return NOTIFY_OK;
5433}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005434
5435void __init page_alloc_init(void)
5436{
5437 hotcpu_notifier(page_alloc_cpu_notify, 0);
5438}
5439
5440/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005441 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5442 * or min_free_kbytes changes.
5443 */
5444static void calculate_totalreserve_pages(void)
5445{
5446 struct pglist_data *pgdat;
5447 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005448 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005449
5450 for_each_online_pgdat(pgdat) {
5451 for (i = 0; i < MAX_NR_ZONES; i++) {
5452 struct zone *zone = pgdat->node_zones + i;
5453 unsigned long max = 0;
5454
5455 /* Find valid and maximum lowmem_reserve in the zone */
5456 for (j = i; j < MAX_NR_ZONES; j++) {
5457 if (zone->lowmem_reserve[j] > max)
5458 max = zone->lowmem_reserve[j];
5459 }
5460
Mel Gorman41858962009-06-16 15:32:12 -07005461 /* we treat the high watermark as reserved pages. */
5462 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005463
Jiang Liub40da042013-02-22 16:33:52 -08005464 if (max > zone->managed_pages)
5465 max = zone->managed_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005466 reserve_pages += max;
Johannes Weinerab8fabd2012-01-10 15:07:42 -08005467 /*
5468 * Lowmem reserves are not available to
5469 * GFP_HIGHUSER page cache allocations and
5470 * kswapd tries to balance zones to their high
5471 * watermark. As a result, neither should be
5472 * regarded as dirtyable memory, to prevent a
5473 * situation where reclaim has to clean pages
5474 * in order to balance the zones.
5475 */
5476 zone->dirty_balance_reserve = max;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005477 }
5478 }
Johannes Weinerab8fabd2012-01-10 15:07:42 -08005479 dirty_balance_reserve = reserve_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005480 totalreserve_pages = reserve_pages;
5481}
5482
5483/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07005484 * setup_per_zone_lowmem_reserve - called whenever
5485 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
5486 * has a correct pages reserved value, so an adequate number of
5487 * pages are left in the zone after a successful __alloc_pages().
5488 */
5489static void setup_per_zone_lowmem_reserve(void)
5490{
5491 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005492 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005493
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08005494 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005495 for (j = 0; j < MAX_NR_ZONES; j++) {
5496 struct zone *zone = pgdat->node_zones + j;
Jiang Liub40da042013-02-22 16:33:52 -08005497 unsigned long managed_pages = zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005498
5499 zone->lowmem_reserve[j] = 0;
5500
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005501 idx = j;
5502 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005503 struct zone *lower_zone;
5504
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005505 idx--;
5506
Linus Torvalds1da177e2005-04-16 15:20:36 -07005507 if (sysctl_lowmem_reserve_ratio[idx] < 1)
5508 sysctl_lowmem_reserve_ratio[idx] = 1;
5509
5510 lower_zone = pgdat->node_zones + idx;
Jiang Liub40da042013-02-22 16:33:52 -08005511 lower_zone->lowmem_reserve[j] = managed_pages /
Linus Torvalds1da177e2005-04-16 15:20:36 -07005512 sysctl_lowmem_reserve_ratio[idx];
Jiang Liub40da042013-02-22 16:33:52 -08005513 managed_pages += lower_zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005514 }
5515 }
5516 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005517
5518 /* update totalreserve_pages */
5519 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005520}
5521
Mel Gormancfd3da12011-04-25 21:36:42 +00005522static void __setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005523{
5524 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
5525 unsigned long lowmem_pages = 0;
5526 struct zone *zone;
5527 unsigned long flags;
5528
5529 /* Calculate total number of !ZONE_HIGHMEM pages */
5530 for_each_zone(zone) {
5531 if (!is_highmem(zone))
Jiang Liub40da042013-02-22 16:33:52 -08005532 lowmem_pages += zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005533 }
5534
5535 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07005536 u64 tmp;
5537
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005538 spin_lock_irqsave(&zone->lock, flags);
Jiang Liub40da042013-02-22 16:33:52 -08005539 tmp = (u64)pages_min * zone->managed_pages;
Andrew Mortonac924c62006-05-15 09:43:59 -07005540 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005541 if (is_highmem(zone)) {
5542 /*
Nick Piggin669ed172005-11-13 16:06:45 -08005543 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5544 * need highmem pages, so cap pages_min to a small
5545 * value here.
5546 *
Mel Gorman41858962009-06-16 15:32:12 -07005547 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Nick Piggin669ed172005-11-13 16:06:45 -08005548 * deltas controls asynch page reclaim, and so should
5549 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005550 */
Andrew Morton90ae8d62013-02-22 16:32:22 -08005551 unsigned long min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005552
Jiang Liub40da042013-02-22 16:33:52 -08005553 min_pages = zone->managed_pages / 1024;
Andrew Morton90ae8d62013-02-22 16:32:22 -08005554 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
Mel Gorman41858962009-06-16 15:32:12 -07005555 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005556 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08005557 /*
5558 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07005559 * proportionate to the zone's size.
5560 */
Mel Gorman41858962009-06-16 15:32:12 -07005561 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005562 }
5563
Mel Gorman41858962009-06-16 15:32:12 -07005564 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
5565 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Marek Szyprowski49f223a2012-01-25 12:49:24 +01005566
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07005567 __mod_zone_page_state(zone, NR_ALLOC_BATCH,
5568 high_wmark_pages(zone) -
5569 low_wmark_pages(zone) -
5570 zone_page_state(zone, NR_ALLOC_BATCH));
5571
Mel Gorman56fd56b2007-10-16 01:25:58 -07005572 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005573 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005574 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005575
5576 /* update totalreserve_pages */
5577 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005578}
5579
Mel Gormancfd3da12011-04-25 21:36:42 +00005580/**
5581 * setup_per_zone_wmarks - called when min_free_kbytes changes
5582 * or when memory is hot-{added|removed}
5583 *
5584 * Ensures that the watermark[min,low,high] values for each zone are set
5585 * correctly with respect to min_free_kbytes.
5586 */
5587void setup_per_zone_wmarks(void)
5588{
5589 mutex_lock(&zonelists_mutex);
5590 __setup_per_zone_wmarks();
5591 mutex_unlock(&zonelists_mutex);
5592}
5593
Randy Dunlap55a44622009-09-21 17:01:20 -07005594/*
Rik van Riel556adec2008-10-18 20:26:34 -07005595 * The inactive anon list should be small enough that the VM never has to
5596 * do too much work, but large enough that each inactive page has a chance
5597 * to be referenced again before it is swapped out.
5598 *
5599 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5600 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5601 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5602 * the anonymous pages are kept on the inactive list.
5603 *
5604 * total target max
5605 * memory ratio inactive anon
5606 * -------------------------------------
5607 * 10MB 1 5MB
5608 * 100MB 1 50MB
5609 * 1GB 3 250MB
5610 * 10GB 10 0.9GB
5611 * 100GB 31 3GB
5612 * 1TB 101 10GB
5613 * 10TB 320 32GB
5614 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005615static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
Minchan Kim96cb4df2009-06-16 15:32:49 -07005616{
5617 unsigned int gb, ratio;
5618
5619 /* Zone size in gigabytes */
Jiang Liub40da042013-02-22 16:33:52 -08005620 gb = zone->managed_pages >> (30 - PAGE_SHIFT);
Minchan Kim96cb4df2009-06-16 15:32:49 -07005621 if (gb)
5622 ratio = int_sqrt(10 * gb);
5623 else
5624 ratio = 1;
5625
5626 zone->inactive_ratio = ratio;
5627}
5628
KOSAKI Motohiro839a4fc2011-05-24 17:11:31 -07005629static void __meminit setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07005630{
5631 struct zone *zone;
5632
Minchan Kim96cb4df2009-06-16 15:32:49 -07005633 for_each_zone(zone)
5634 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07005635}
5636
Linus Torvalds1da177e2005-04-16 15:20:36 -07005637/*
5638 * Initialise min_free_kbytes.
5639 *
5640 * For small machines we want it small (128k min). For large machines
5641 * we want it large (64MB max). But it is not linear, because network
5642 * bandwidth does not increase linearly with machine size. We use
5643 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07005644 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
Linus Torvalds1da177e2005-04-16 15:20:36 -07005645 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5646 *
5647 * which yields
5648 *
5649 * 16MB: 512k
5650 * 32MB: 724k
5651 * 64MB: 1024k
5652 * 128MB: 1448k
5653 * 256MB: 2048k
5654 * 512MB: 2896k
5655 * 1024MB: 4096k
5656 * 2048MB: 5792k
5657 * 4096MB: 8192k
5658 * 8192MB: 11584k
5659 * 16384MB: 16384k
5660 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005661int __meminit init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005662{
5663 unsigned long lowmem_kbytes;
Michal Hocko5f127332013-07-08 16:00:40 -07005664 int new_min_free_kbytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005665
5666 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
Michal Hocko5f127332013-07-08 16:00:40 -07005667 new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005668
Michal Hocko5f127332013-07-08 16:00:40 -07005669 if (new_min_free_kbytes > user_min_free_kbytes) {
5670 min_free_kbytes = new_min_free_kbytes;
5671 if (min_free_kbytes < 128)
5672 min_free_kbytes = 128;
5673 if (min_free_kbytes > 65536)
5674 min_free_kbytes = 65536;
5675 } else {
5676 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
5677 new_min_free_kbytes, user_min_free_kbytes);
5678 }
Minchan Kimbc75d332009-06-16 15:32:48 -07005679 setup_per_zone_wmarks();
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -07005680 refresh_zone_stat_thresholds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005681 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07005682 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005683 return 0;
5684}
Minchan Kimbc75d332009-06-16 15:32:48 -07005685module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005686
5687/*
Pintu Kumarb8af2942013-09-11 14:20:34 -07005688 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
Linus Torvalds1da177e2005-04-16 15:20:36 -07005689 * that we can call two helper functions whenever min_free_kbytes
5690 * changes.
5691 */
Pintu Kumarb8af2942013-09-11 14:20:34 -07005692int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005693 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005694{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005695 proc_dointvec(table, write, buffer, length, ppos);
Michal Hocko5f127332013-07-08 16:00:40 -07005696 if (write) {
5697 user_min_free_kbytes = min_free_kbytes;
Minchan Kimbc75d332009-06-16 15:32:48 -07005698 setup_per_zone_wmarks();
Michal Hocko5f127332013-07-08 16:00:40 -07005699 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005700 return 0;
5701}
5702
Christoph Lameter96146342006-07-03 00:24:13 -07005703#ifdef CONFIG_NUMA
5704int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005705 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07005706{
5707 struct zone *zone;
5708 int rc;
5709
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005710 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07005711 if (rc)
5712 return rc;
5713
5714 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08005715 zone->min_unmapped_pages = (zone->managed_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07005716 sysctl_min_unmapped_ratio) / 100;
5717 return 0;
5718}
Christoph Lameter0ff38492006-09-25 23:31:52 -07005719
5720int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005721 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07005722{
5723 struct zone *zone;
5724 int rc;
5725
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005726 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07005727 if (rc)
5728 return rc;
5729
5730 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08005731 zone->min_slab_pages = (zone->managed_pages *
Christoph Lameter0ff38492006-09-25 23:31:52 -07005732 sysctl_min_slab_ratio) / 100;
5733 return 0;
5734}
Christoph Lameter96146342006-07-03 00:24:13 -07005735#endif
5736
Linus Torvalds1da177e2005-04-16 15:20:36 -07005737/*
5738 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5739 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5740 * whenever sysctl_lowmem_reserve_ratio changes.
5741 *
5742 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07005743 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07005744 * if in function of the boot time zone sizes.
5745 */
5746int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005747 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005748{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005749 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005750 setup_per_zone_lowmem_reserve();
5751 return 0;
5752}
5753
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005754/*
5755 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
Pintu Kumarb8af2942013-09-11 14:20:34 -07005756 * cpu. It is the fraction of total pages in each zone that a hot per cpu
5757 * pagelist can have before it gets flushed back to buddy allocator.
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005758 */
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005759int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005760 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005761{
5762 struct zone *zone;
5763 unsigned int cpu;
5764 int ret;
5765
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005766 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
Sasha Levin93278812012-05-10 13:01:44 -07005767 if (!write || (ret < 0))
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005768 return ret;
Cody P Schaferc8e251f2013-07-03 15:01:29 -07005769
5770 mutex_lock(&pcp_batch_high_lock);
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07005771 for_each_populated_zone(zone) {
Cody P Schafer22a7f122013-07-03 15:01:34 -07005772 unsigned long high;
5773 high = zone->managed_pages / percpu_pagelist_fraction;
5774 for_each_possible_cpu(cpu)
Cody P Schafer36640332013-07-03 15:01:40 -07005775 pageset_set_high(per_cpu_ptr(zone->pageset, cpu),
5776 high);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005777 }
Cody P Schaferc8e251f2013-07-03 15:01:29 -07005778 mutex_unlock(&pcp_batch_high_lock);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005779 return 0;
5780}
5781
David S. Millerf034b5d2006-08-24 03:08:07 -07005782int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005783
5784#ifdef CONFIG_NUMA
5785static int __init set_hashdist(char *str)
5786{
5787 if (!str)
5788 return 0;
5789 hashdist = simple_strtoul(str, &str, 0);
5790 return 1;
5791}
5792__setup("hashdist=", set_hashdist);
5793#endif
5794
5795/*
5796 * allocate a large system hash table from bootmem
5797 * - it is assumed that the hash table must contain an exact power-of-2
5798 * quantity of entries
5799 * - limit is the number of hash buckets, not the total allocation size
5800 */
5801void *__init alloc_large_system_hash(const char *tablename,
5802 unsigned long bucketsize,
5803 unsigned long numentries,
5804 int scale,
5805 int flags,
5806 unsigned int *_hash_shift,
5807 unsigned int *_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00005808 unsigned long low_limit,
5809 unsigned long high_limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005810{
Tim Bird31fe62b2012-05-23 13:33:35 +00005811 unsigned long long max = high_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005812 unsigned long log2qty, size;
5813 void *table = NULL;
5814
5815 /* allow the kernel cmdline to have a say */
5816 if (!numentries) {
5817 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08005818 numentries = nr_kernel_pages;
Jerry Zhoua7e83312013-09-11 14:20:26 -07005819
5820 /* It isn't necessary when PAGE_SIZE >= 1MB */
5821 if (PAGE_SHIFT < 20)
5822 numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005823
5824 /* limit to 1 bucket per 2^scale bytes of low memory */
5825 if (scale > PAGE_SHIFT)
5826 numentries >>= (scale - PAGE_SHIFT);
5827 else
5828 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08005829
5830 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07005831 if (unlikely(flags & HASH_SMALL)) {
5832 /* Makes no sense without HASH_EARLY */
5833 WARN_ON(!(flags & HASH_EARLY));
5834 if (!(numentries >> *_hash_shift)) {
5835 numentries = 1UL << *_hash_shift;
5836 BUG_ON(!numentries);
5837 }
5838 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08005839 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005840 }
John Hawkes6e692ed2006-03-25 03:08:02 -08005841 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005842
5843 /* limit allocation size to 1/16 total memory by default */
5844 if (max == 0) {
5845 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
5846 do_div(max, bucketsize);
5847 }
Dimitri Sivanich074b8512012-02-08 12:39:07 -08005848 max = min(max, 0x80000000ULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005849
Tim Bird31fe62b2012-05-23 13:33:35 +00005850 if (numentries < low_limit)
5851 numentries = low_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005852 if (numentries > max)
5853 numentries = max;
5854
David Howellsf0d1b0b2006-12-08 02:37:49 -08005855 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005856
5857 do {
5858 size = bucketsize << log2qty;
5859 if (flags & HASH_EARLY)
Jan Beulich74768ed2008-08-12 15:08:39 -07005860 table = alloc_bootmem_nopanic(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005861 else if (hashdist)
5862 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
5863 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07005864 /*
5865 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07005866 * some pages at the end of hash table which
5867 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07005868 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005869 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07005870 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005871 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
5872 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005873 }
5874 } while (!table && size > PAGE_SIZE && --log2qty);
5875
5876 if (!table)
5877 panic("Failed to allocate %s hash table\n", tablename);
5878
Robin Holtf241e6602010-10-07 12:59:26 -07005879 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07005880 tablename,
Robin Holtf241e6602010-10-07 12:59:26 -07005881 (1UL << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08005882 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005883 size);
5884
5885 if (_hash_shift)
5886 *_hash_shift = log2qty;
5887 if (_hash_mask)
5888 *_hash_mask = (1 << log2qty) - 1;
5889
5890 return table;
5891}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08005892
Mel Gorman835c1342007-10-16 01:25:47 -07005893/* Return a pointer to the bitmap storing bits affecting a block of pages */
5894static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
5895 unsigned long pfn)
5896{
5897#ifdef CONFIG_SPARSEMEM
5898 return __pfn_to_section(pfn)->pageblock_flags;
5899#else
5900 return zone->pageblock_flags;
5901#endif /* CONFIG_SPARSEMEM */
5902}
Andrew Morton6220ec72006-10-19 23:29:05 -07005903
Mel Gorman835c1342007-10-16 01:25:47 -07005904static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
5905{
5906#ifdef CONFIG_SPARSEMEM
5907 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07005908 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005909#else
Laura Abbottc060f942013-01-11 14:31:51 -08005910 pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
Mel Gormand9c23402007-10-16 01:26:01 -07005911 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005912#endif /* CONFIG_SPARSEMEM */
5913}
5914
5915/**
Mel Gormand9c23402007-10-16 01:26:01 -07005916 * 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 -07005917 * @page: The page within the block of interest
5918 * @start_bitidx: The first bit of interest to retrieve
5919 * @end_bitidx: The last bit of interest
5920 * returns pageblock_bits flags
5921 */
5922unsigned long get_pageblock_flags_group(struct page *page,
5923 int start_bitidx, int end_bitidx)
5924{
5925 struct zone *zone;
5926 unsigned long *bitmap;
5927 unsigned long pfn, bitidx;
5928 unsigned long flags = 0;
5929 unsigned long value = 1;
5930
5931 zone = page_zone(page);
5932 pfn = page_to_pfn(page);
5933 bitmap = get_pageblock_bitmap(zone, pfn);
5934 bitidx = pfn_to_bitidx(zone, pfn);
5935
5936 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5937 if (test_bit(bitidx + start_bitidx, bitmap))
5938 flags |= value;
5939
5940 return flags;
5941}
5942
5943/**
Mel Gormand9c23402007-10-16 01:26:01 -07005944 * 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 -07005945 * @page: The page within the block of interest
5946 * @start_bitidx: The first bit of interest
5947 * @end_bitidx: The last bit of interest
5948 * @flags: The flags to set
5949 */
5950void set_pageblock_flags_group(struct page *page, unsigned long flags,
5951 int start_bitidx, int end_bitidx)
5952{
5953 struct zone *zone;
5954 unsigned long *bitmap;
5955 unsigned long pfn, bitidx;
5956 unsigned long value = 1;
5957
5958 zone = page_zone(page);
5959 pfn = page_to_pfn(page);
5960 bitmap = get_pageblock_bitmap(zone, pfn);
5961 bitidx = pfn_to_bitidx(zone, pfn);
Cody P Schafer108bcc92013-02-22 16:35:23 -08005962 VM_BUG_ON(!zone_spans_pfn(zone, pfn));
Mel Gorman835c1342007-10-16 01:25:47 -07005963
5964 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5965 if (flags & value)
5966 __set_bit(bitidx + start_bitidx, bitmap);
5967 else
5968 __clear_bit(bitidx + start_bitidx, bitmap);
5969}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005970
5971/*
Minchan Kim80934512012-07-31 16:43:01 -07005972 * This function checks whether pageblock includes unmovable pages or not.
5973 * If @count is not zero, it is okay to include less @count unmovable pages
5974 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07005975 * PageLRU check without isolation or lru_lock could race so that
Minchan Kim80934512012-07-31 16:43:01 -07005976 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
5977 * expect this function should be exact.
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005978 */
Wen Congyangb023f462012-12-11 16:00:45 -08005979bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
5980 bool skip_hwpoisoned_pages)
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005981{
5982 unsigned long pfn, iter, found;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01005983 int mt;
5984
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005985 /*
5986 * For avoiding noise data, lru_add_drain_all() should be called
Minchan Kim80934512012-07-31 16:43:01 -07005987 * If ZONE_MOVABLE, the zone never contains unmovable pages
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005988 */
5989 if (zone_idx(zone) == ZONE_MOVABLE)
Minchan Kim80934512012-07-31 16:43:01 -07005990 return false;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01005991 mt = get_pageblock_migratetype(page);
5992 if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
Minchan Kim80934512012-07-31 16:43:01 -07005993 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005994
5995 pfn = page_to_pfn(page);
5996 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
5997 unsigned long check = pfn + iter;
5998
Namhyung Kim29723fc2011-02-25 14:44:25 -08005999 if (!pfn_valid_within(check))
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006000 continue;
Namhyung Kim29723fc2011-02-25 14:44:25 -08006001
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006002 page = pfn_to_page(check);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07006003
6004 /*
6005 * Hugepages are not in LRU lists, but they're movable.
6006 * We need not scan over tail pages bacause we don't
6007 * handle each tail page individually in migration.
6008 */
6009 if (PageHuge(page)) {
6010 iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
6011 continue;
6012 }
6013
Minchan Kim97d255c2012-07-31 16:42:59 -07006014 /*
6015 * We can't use page_count without pin a page
6016 * because another CPU can free compound page.
6017 * This check already skips compound tails of THP
6018 * because their page->_count is zero at all time.
6019 */
6020 if (!atomic_read(&page->_count)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006021 if (PageBuddy(page))
6022 iter += (1 << page_order(page)) - 1;
6023 continue;
6024 }
Minchan Kim97d255c2012-07-31 16:42:59 -07006025
Wen Congyangb023f462012-12-11 16:00:45 -08006026 /*
6027 * The HWPoisoned page may be not in buddy system, and
6028 * page_count() is not 0.
6029 */
6030 if (skip_hwpoisoned_pages && PageHWPoison(page))
6031 continue;
6032
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006033 if (!PageLRU(page))
6034 found++;
6035 /*
6036 * If there are RECLAIMABLE pages, we need to check it.
6037 * But now, memory offline itself doesn't call shrink_slab()
6038 * and it still to be fixed.
6039 */
6040 /*
6041 * If the page is not RAM, page_count()should be 0.
6042 * we don't need more check. This is an _used_ not-movable page.
6043 *
6044 * The problematic thing here is PG_reserved pages. PG_reserved
6045 * is set to both of a memory hole page and a _used_ kernel
6046 * page at boot.
6047 */
6048 if (found > count)
Minchan Kim80934512012-07-31 16:43:01 -07006049 return true;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006050 }
Minchan Kim80934512012-07-31 16:43:01 -07006051 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006052}
6053
6054bool is_pageblock_removable_nolock(struct page *page)
6055{
Michal Hocko656a0702012-01-20 14:33:58 -08006056 struct zone *zone;
6057 unsigned long pfn;
Michal Hocko687875f2012-01-20 14:33:55 -08006058
6059 /*
6060 * We have to be careful here because we are iterating over memory
6061 * sections which are not zone aware so we might end up outside of
6062 * the zone but still within the section.
Michal Hocko656a0702012-01-20 14:33:58 -08006063 * We have to take care about the node as well. If the node is offline
6064 * its NODE_DATA will be NULL - see page_zone.
Michal Hocko687875f2012-01-20 14:33:55 -08006065 */
Michal Hocko656a0702012-01-20 14:33:58 -08006066 if (!node_online(page_to_nid(page)))
6067 return false;
6068
6069 zone = page_zone(page);
6070 pfn = page_to_pfn(page);
Cody P Schafer108bcc92013-02-22 16:35:23 -08006071 if (!zone_spans_pfn(zone, pfn))
Michal Hocko687875f2012-01-20 14:33:55 -08006072 return false;
6073
Wen Congyangb023f462012-12-11 16:00:45 -08006074 return !has_unmovable_pages(zone, page, 0, true);
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07006075}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006076
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006077#ifdef CONFIG_CMA
6078
6079static unsigned long pfn_max_align_down(unsigned long pfn)
6080{
6081 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
6082 pageblock_nr_pages) - 1);
6083}
6084
6085static unsigned long pfn_max_align_up(unsigned long pfn)
6086{
6087 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
6088 pageblock_nr_pages));
6089}
6090
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006091/* [start, end) must belong to a single zone. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006092static int __alloc_contig_migrate_range(struct compact_control *cc,
6093 unsigned long start, unsigned long end)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006094{
6095 /* This function is based on compact_zone() from compaction.c. */
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006096 unsigned long nr_reclaimed;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006097 unsigned long pfn = start;
6098 unsigned int tries = 0;
6099 int ret = 0;
6100
Marek Szyprowskibe49a6e2012-12-12 13:51:19 -08006101 migrate_prep();
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006102
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006103 while (pfn < end || !list_empty(&cc->migratepages)) {
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006104 if (fatal_signal_pending(current)) {
6105 ret = -EINTR;
6106 break;
6107 }
6108
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006109 if (list_empty(&cc->migratepages)) {
6110 cc->nr_migratepages = 0;
6111 pfn = isolate_migratepages_range(cc->zone, cc,
Minchan Kime46a2872012-10-08 16:33:48 -07006112 pfn, end, true);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006113 if (!pfn) {
6114 ret = -EINTR;
6115 break;
6116 }
6117 tries = 0;
6118 } else if (++tries == 5) {
6119 ret = ret < 0 ? ret : -EBUSY;
6120 break;
6121 }
6122
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006123 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
6124 &cc->migratepages);
6125 cc->nr_migratepages -= nr_reclaimed;
Minchan Kim02c6de82012-10-08 16:31:55 -07006126
Hugh Dickins9c620e22013-02-22 16:35:14 -08006127 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
6128 0, MIGRATE_SYNC, MR_CMA);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006129 }
Srinivas Pandruvada2a6f5122013-02-22 16:32:09 -08006130 if (ret < 0) {
6131 putback_movable_pages(&cc->migratepages);
6132 return ret;
6133 }
6134 return 0;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006135}
6136
6137/**
6138 * alloc_contig_range() -- tries to allocate given range of pages
6139 * @start: start PFN to allocate
6140 * @end: one-past-the-last PFN to allocate
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006141 * @migratetype: migratetype of the underlaying pageblocks (either
6142 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6143 * in range must have the same migratetype and it must
6144 * be either of the two.
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006145 *
6146 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6147 * aligned, however it's the caller's responsibility to guarantee that
6148 * we are the only thread that changes migrate type of pageblocks the
6149 * pages fall in.
6150 *
6151 * The PFN range must belong to a single zone.
6152 *
6153 * Returns zero on success or negative error code. On success all
6154 * pages which PFN is in [start, end) are allocated for the caller and
6155 * need to be freed with free_contig_range().
6156 */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006157int alloc_contig_range(unsigned long start, unsigned long end,
6158 unsigned migratetype)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006159{
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006160 unsigned long outer_start, outer_end;
6161 int ret = 0, order;
6162
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006163 struct compact_control cc = {
6164 .nr_migratepages = 0,
6165 .order = -1,
6166 .zone = page_zone(pfn_to_page(start)),
6167 .sync = true,
6168 .ignore_skip_hint = true,
6169 };
6170 INIT_LIST_HEAD(&cc.migratepages);
6171
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006172 /*
6173 * What we do here is we mark all pageblocks in range as
6174 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6175 * have different sizes, and due to the way page allocator
6176 * work, we align the range to biggest of the two pages so
6177 * that page allocator won't try to merge buddies from
6178 * different pageblocks and change MIGRATE_ISOLATE to some
6179 * other migration type.
6180 *
6181 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6182 * migrate the pages from an unaligned range (ie. pages that
6183 * we are interested in). This will put all the pages in
6184 * range back to page allocator as MIGRATE_ISOLATE.
6185 *
6186 * When this is done, we take the pages in range from page
6187 * allocator removing them from the buddy system. This way
6188 * page allocator will never consider using them.
6189 *
6190 * This lets us mark the pageblocks back as
6191 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6192 * aligned range but not in the unaligned, original range are
6193 * put back to page allocator so that buddy can use them.
6194 */
6195
6196 ret = start_isolate_page_range(pfn_max_align_down(start),
Wen Congyangb023f462012-12-11 16:00:45 -08006197 pfn_max_align_up(end), migratetype,
6198 false);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006199 if (ret)
Bob Liu86a595f2012-10-25 13:37:56 -07006200 return ret;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006201
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006202 ret = __alloc_contig_migrate_range(&cc, start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006203 if (ret)
6204 goto done;
6205
6206 /*
6207 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6208 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
6209 * more, all pages in [start, end) are free in page allocator.
6210 * What we are going to do is to allocate all pages from
6211 * [start, end) (that is remove them from page allocator).
6212 *
6213 * The only problem is that pages at the beginning and at the
6214 * end of interesting range may be not aligned with pages that
6215 * page allocator holds, ie. they can be part of higher order
6216 * pages. Because of this, we reserve the bigger range and
6217 * once this is done free the pages we are not interested in.
6218 *
6219 * We don't have to hold zone->lock here because the pages are
6220 * isolated thus they won't get removed from buddy.
6221 */
6222
6223 lru_add_drain_all();
6224 drain_all_pages();
6225
6226 order = 0;
6227 outer_start = start;
6228 while (!PageBuddy(pfn_to_page(outer_start))) {
6229 if (++order >= MAX_ORDER) {
6230 ret = -EBUSY;
6231 goto done;
6232 }
6233 outer_start &= ~0UL << order;
6234 }
6235
6236 /* Make sure the range is really isolated. */
Wen Congyangb023f462012-12-11 16:00:45 -08006237 if (test_pages_isolated(outer_start, end, false)) {
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006238 pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n",
6239 outer_start, end);
6240 ret = -EBUSY;
6241 goto done;
6242 }
6243
Marek Szyprowski49f223a2012-01-25 12:49:24 +01006244
6245 /* Grab isolated pages from freelists. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006246 outer_end = isolate_freepages_range(&cc, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006247 if (!outer_end) {
6248 ret = -EBUSY;
6249 goto done;
6250 }
6251
6252 /* Free head and tail (if any) */
6253 if (start != outer_start)
6254 free_contig_range(outer_start, start - outer_start);
6255 if (end != outer_end)
6256 free_contig_range(end, outer_end - end);
6257
6258done:
6259 undo_isolate_page_range(pfn_max_align_down(start),
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006260 pfn_max_align_up(end), migratetype);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006261 return ret;
6262}
6263
6264void free_contig_range(unsigned long pfn, unsigned nr_pages)
6265{
Marek Szyprowskibcc2b022012-12-20 15:05:18 -08006266 unsigned int count = 0;
6267
6268 for (; nr_pages--; pfn++) {
6269 struct page *page = pfn_to_page(pfn);
6270
6271 count += page_count(page) != 1;
6272 __free_page(page);
6273 }
6274 WARN(count != 0, "%d pages are still in use!\n", count);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006275}
6276#endif
6277
Jiang Liu4ed7e022012-07-31 16:43:35 -07006278#ifdef CONFIG_MEMORY_HOTPLUG
Cody P Schafer0a647f32013-07-03 15:01:33 -07006279/*
6280 * The zone indicated has a new number of managed_pages; batch sizes and percpu
6281 * page high values need to be recalulated.
6282 */
Jiang Liu4ed7e022012-07-31 16:43:35 -07006283void __meminit zone_pcp_update(struct zone *zone)
6284{
Cody P Schafer0a647f32013-07-03 15:01:33 -07006285 unsigned cpu;
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006286 mutex_lock(&pcp_batch_high_lock);
Cody P Schafer0a647f32013-07-03 15:01:33 -07006287 for_each_possible_cpu(cpu)
Cody P Schafer169f6c12013-07-03 15:01:41 -07006288 pageset_set_high_and_batch(zone,
6289 per_cpu_ptr(zone->pageset, cpu));
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006290 mutex_unlock(&pcp_batch_high_lock);
Jiang Liu4ed7e022012-07-31 16:43:35 -07006291}
6292#endif
6293
Jiang Liu340175b2012-07-31 16:43:32 -07006294void zone_pcp_reset(struct zone *zone)
6295{
6296 unsigned long flags;
Minchan Kim5a883812012-10-08 16:33:39 -07006297 int cpu;
6298 struct per_cpu_pageset *pset;
Jiang Liu340175b2012-07-31 16:43:32 -07006299
6300 /* avoid races with drain_pages() */
6301 local_irq_save(flags);
6302 if (zone->pageset != &boot_pageset) {
Minchan Kim5a883812012-10-08 16:33:39 -07006303 for_each_online_cpu(cpu) {
6304 pset = per_cpu_ptr(zone->pageset, cpu);
6305 drain_zonestat(zone, pset);
6306 }
Jiang Liu340175b2012-07-31 16:43:32 -07006307 free_percpu(zone->pageset);
6308 zone->pageset = &boot_pageset;
6309 }
6310 local_irq_restore(flags);
6311}
6312
Wen Congyang6dcd73d2012-12-11 16:01:01 -08006313#ifdef CONFIG_MEMORY_HOTREMOVE
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006314/*
6315 * All pages in the range must be isolated before calling this.
6316 */
6317void
6318__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
6319{
6320 struct page *page;
6321 struct zone *zone;
6322 int order, i;
6323 unsigned long pfn;
6324 unsigned long flags;
6325 /* find the first valid pfn */
6326 for (pfn = start_pfn; pfn < end_pfn; pfn++)
6327 if (pfn_valid(pfn))
6328 break;
6329 if (pfn == end_pfn)
6330 return;
6331 zone = page_zone(pfn_to_page(pfn));
6332 spin_lock_irqsave(&zone->lock, flags);
6333 pfn = start_pfn;
6334 while (pfn < end_pfn) {
6335 if (!pfn_valid(pfn)) {
6336 pfn++;
6337 continue;
6338 }
6339 page = pfn_to_page(pfn);
Wen Congyangb023f462012-12-11 16:00:45 -08006340 /*
6341 * The HWPoisoned page may be not in buddy system, and
6342 * page_count() is not 0.
6343 */
6344 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
6345 pfn++;
6346 SetPageReserved(page);
6347 continue;
6348 }
6349
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006350 BUG_ON(page_count(page));
6351 BUG_ON(!PageBuddy(page));
6352 order = page_order(page);
6353#ifdef CONFIG_DEBUG_VM
6354 printk(KERN_INFO "remove from free list %lx %d %lx\n",
6355 pfn, 1 << order, end_pfn);
6356#endif
6357 list_del(&page->lru);
6358 rmv_page_order(page);
6359 zone->free_area[order].nr_free--;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006360 for (i = 0; i < (1 << order); i++)
6361 SetPageReserved((page+i));
6362 pfn += (1 << order);
6363 }
6364 spin_unlock_irqrestore(&zone->lock, flags);
6365}
6366#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01006367
6368#ifdef CONFIG_MEMORY_FAILURE
6369bool is_free_buddy_page(struct page *page)
6370{
6371 struct zone *zone = page_zone(page);
6372 unsigned long pfn = page_to_pfn(page);
6373 unsigned long flags;
6374 int order;
6375
6376 spin_lock_irqsave(&zone->lock, flags);
6377 for (order = 0; order < MAX_ORDER; order++) {
6378 struct page *page_head = page - (pfn & ((1 << order) - 1));
6379
6380 if (PageBuddy(page_head) && page_order(page_head) >= order)
6381 break;
6382 }
6383 spin_unlock_irqrestore(&zone->lock, flags);
6384
6385 return order < MAX_ORDER;
6386}
6387#endif
Wu Fengguang718a3822010-03-10 15:20:43 -08006388
Andrew Morton51300ce2012-05-29 15:06:44 -07006389static const struct trace_print_flags pageflag_names[] = {
Wu Fengguang718a3822010-03-10 15:20:43 -08006390 {1UL << PG_locked, "locked" },
6391 {1UL << PG_error, "error" },
6392 {1UL << PG_referenced, "referenced" },
6393 {1UL << PG_uptodate, "uptodate" },
6394 {1UL << PG_dirty, "dirty" },
6395 {1UL << PG_lru, "lru" },
6396 {1UL << PG_active, "active" },
6397 {1UL << PG_slab, "slab" },
6398 {1UL << PG_owner_priv_1, "owner_priv_1" },
6399 {1UL << PG_arch_1, "arch_1" },
6400 {1UL << PG_reserved, "reserved" },
6401 {1UL << PG_private, "private" },
6402 {1UL << PG_private_2, "private_2" },
6403 {1UL << PG_writeback, "writeback" },
6404#ifdef CONFIG_PAGEFLAGS_EXTENDED
6405 {1UL << PG_head, "head" },
6406 {1UL << PG_tail, "tail" },
6407#else
6408 {1UL << PG_compound, "compound" },
6409#endif
6410 {1UL << PG_swapcache, "swapcache" },
6411 {1UL << PG_mappedtodisk, "mappedtodisk" },
6412 {1UL << PG_reclaim, "reclaim" },
Wu Fengguang718a3822010-03-10 15:20:43 -08006413 {1UL << PG_swapbacked, "swapbacked" },
6414 {1UL << PG_unevictable, "unevictable" },
6415#ifdef CONFIG_MMU
6416 {1UL << PG_mlocked, "mlocked" },
6417#endif
6418#ifdef CONFIG_ARCH_USES_PG_UNCACHED
6419 {1UL << PG_uncached, "uncached" },
6420#endif
6421#ifdef CONFIG_MEMORY_FAILURE
6422 {1UL << PG_hwpoison, "hwpoison" },
6423#endif
Gavin Shanbe9cd872012-05-29 15:06:44 -07006424#ifdef CONFIG_TRANSPARENT_HUGEPAGE
6425 {1UL << PG_compound_lock, "compound_lock" },
6426#endif
Wu Fengguang718a3822010-03-10 15:20:43 -08006427};
6428
6429static void dump_page_flags(unsigned long flags)
6430{
6431 const char *delim = "";
6432 unsigned long mask;
6433 int i;
6434
Andrew Morton51300ce2012-05-29 15:06:44 -07006435 BUILD_BUG_ON(ARRAY_SIZE(pageflag_names) != __NR_PAGEFLAGS);
Johannes Weineracc50c12012-05-29 15:06:44 -07006436
Wu Fengguang718a3822010-03-10 15:20:43 -08006437 printk(KERN_ALERT "page flags: %#lx(", flags);
6438
6439 /* remove zone id */
6440 flags &= (1UL << NR_PAGEFLAGS) - 1;
6441
Andrew Morton51300ce2012-05-29 15:06:44 -07006442 for (i = 0; i < ARRAY_SIZE(pageflag_names) && flags; i++) {
Wu Fengguang718a3822010-03-10 15:20:43 -08006443
6444 mask = pageflag_names[i].mask;
6445 if ((flags & mask) != mask)
6446 continue;
6447
6448 flags &= ~mask;
6449 printk("%s%s", delim, pageflag_names[i].name);
6450 delim = "|";
6451 }
6452
6453 /* check for left over flags */
6454 if (flags)
6455 printk("%s%#lx", delim, flags);
6456
6457 printk(")\n");
6458}
6459
6460void dump_page(struct page *page)
6461{
6462 printk(KERN_ALERT
6463 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
Andrea Arcangeli4e9f64c2011-01-13 15:46:29 -08006464 page, atomic_read(&page->_count), page_mapcount(page),
Wu Fengguang718a3822010-03-10 15:20:43 -08006465 page->mapping, page->index);
6466 dump_page_flags(page->flags);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07006467 mem_cgroup_print_bad_page(page);
Wu Fengguang718a3822010-03-10 15:20:43 -08006468}