blob: dd886fac451ab6ab7d6a3132fd2587c10538f300 [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{
Mel Gorman49255c62009-06-16 15:31:58 -0700237
238 if (unlikely(page_group_by_mobility_disabled))
239 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 }
Mel Gorman22b751c2013-02-22 16:34:59 -0800629 page_nid_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
1030 if (is_migrate_cma(fallback_type))
1031 return fallback_type;
1032
1033 /* Take ownership for orders >= pageblock_order */
1034 if (current_order >= pageblock_order) {
1035 change_pageblock_range(page, current_order, start_type);
1036 return start_type;
1037 }
1038
1039 if (current_order >= pageblock_order / 2 ||
1040 start_type == MIGRATE_RECLAIMABLE ||
1041 page_group_by_mobility_disabled) {
1042 int pages;
1043
1044 pages = move_freepages_block(zone, page, start_type);
1045
1046 /* Claim the whole block if over half of it is free */
1047 if (pages >= (1 << (pageblock_order-1)) ||
1048 page_group_by_mobility_disabled) {
1049
1050 set_pageblock_migratetype(page, start_type);
1051 return start_type;
1052 }
1053
1054 }
1055
1056 return fallback_type;
1057}
1058
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001059/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -07001060static inline struct page *
1061__rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001062{
Pintu Kumarb8af2942013-09-11 14:20:34 -07001063 struct free_area *area;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001064 int current_order;
1065 struct page *page;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001066 int migratetype, new_type, i;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001067
1068 /* Find the largest possible block of pages in the other list */
1069 for (current_order = MAX_ORDER-1; current_order >= order;
1070 --current_order) {
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001071 for (i = 0;; i++) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001072 migratetype = fallbacks[start_migratetype][i];
1073
Mel Gorman56fd56b2007-10-16 01:25:58 -07001074 /* MIGRATE_RESERVE handled later if necessary */
1075 if (migratetype == MIGRATE_RESERVE)
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001076 break;
Mel Gormane0104872007-10-16 01:25:53 -07001077
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001078 area = &(zone->free_area[current_order]);
1079 if (list_empty(&area->free_list[migratetype]))
1080 continue;
1081
1082 page = list_entry(area->free_list[migratetype].next,
1083 struct page, lru);
1084 area->nr_free--;
1085
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001086 new_type = try_to_steal_freepages(zone, page,
1087 start_migratetype,
1088 migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001089
1090 /* Remove the page from the freelists */
1091 list_del(&page->lru);
1092 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001093
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001094 /*
1095 * Borrow the excess buddy pages as well, irrespective
1096 * of whether we stole freepages, or took ownership of
1097 * the pageblock or not.
1098 *
1099 * Exception: When borrowing from MIGRATE_CMA, release
1100 * the excess buddy pages to CMA itself.
1101 */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001102 expand(zone, page, order, current_order, area,
1103 is_migrate_cma(migratetype)
1104 ? migratetype : start_migratetype);
Mel Gormane0fff1b2009-09-21 17:02:42 -07001105
Srivatsa S. Bhatf92310c2013-09-11 14:20:36 -07001106 trace_mm_page_alloc_extfrag(page, order,
1107 current_order, start_migratetype, migratetype,
1108 new_type == start_migratetype);
Mel Gormane0fff1b2009-09-21 17:02:42 -07001109
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001110 return page;
1111 }
1112 }
1113
Mel Gorman728ec982009-06-16 15:32:04 -07001114 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001115}
1116
Mel Gorman56fd56b2007-10-16 01:25:58 -07001117/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 * Do the hard work of removing an element from the buddy allocator.
1119 * Call me with the zone->lock already held.
1120 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001121static struct page *__rmqueue(struct zone *zone, unsigned int order,
1122 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 struct page *page;
1125
Mel Gorman728ec982009-06-16 15:32:04 -07001126retry_reserve:
Mel Gorman56fd56b2007-10-16 01:25:58 -07001127 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128
Mel Gorman728ec982009-06-16 15:32:04 -07001129 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07001130 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001131
Mel Gorman728ec982009-06-16 15:32:04 -07001132 /*
1133 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1134 * is used because __rmqueue_smallest is an inline function
1135 * and we want just one call site
1136 */
1137 if (!page) {
1138 migratetype = MIGRATE_RESERVE;
1139 goto retry_reserve;
1140 }
1141 }
1142
Mel Gorman0d3d0622009-09-21 17:02:44 -07001143 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001144 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145}
1146
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001147/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 * Obtain a specified number of elements from the buddy allocator, all under
1149 * a single hold of the lock, for efficiency. Add them to the supplied list.
1150 * Returns the number of new pages which were placed at *list.
1151 */
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001152static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001153 unsigned long count, struct list_head *list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001154 int migratetype, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155{
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001156 int mt = migratetype, i;
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001157
Nick Pigginc54ad302006-01-06 00:10:56 -08001158 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001160 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001161 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08001163
1164 /*
1165 * Split buddy pages returned by expand() are received here
1166 * in physical page order. The page is added to the callers and
1167 * list and the list head then moves forward. From the callers
1168 * perspective, the linked list is ordered by page number in
1169 * some conditions. This is useful for IO devices that can
1170 * merge IO requests if the physical pages are ordered
1171 * properly.
1172 */
Mel Gormane084b2d2009-07-29 15:02:04 -07001173 if (likely(cold == 0))
1174 list_add(&page->lru, list);
1175 else
1176 list_add_tail(&page->lru, list);
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001177 if (IS_ENABLED(CONFIG_CMA)) {
1178 mt = get_pageblock_migratetype(page);
Minchan Kim194159f2013-02-22 16:33:58 -08001179 if (!is_migrate_cma(mt) && !is_migrate_isolate(mt))
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001180 mt = migratetype;
1181 }
Minchan Kimb12c4ad2012-10-08 16:32:08 -07001182 set_freepage_migratetype(page, mt);
Mel Gorman81eabcb2007-12-17 16:20:05 -08001183 list = &page->lru;
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07001184 if (is_migrate_cma(mt))
1185 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
1186 -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 }
Mel Gormanf2260e62009-06-16 15:32:13 -07001188 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08001189 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001190 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191}
1192
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001193#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001194/*
Christoph Lameter4037d452007-05-09 02:35:14 -07001195 * Called from the vmstat counter updater to drain pagesets of this
1196 * currently executing processor on remote nodes after they have
1197 * expired.
1198 *
Christoph Lameter879336c2006-03-22 00:09:08 -08001199 * Note that this function must be called with the thread pinned to
1200 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001201 */
Christoph Lameter4037d452007-05-09 02:35:14 -07001202void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001203{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001204 unsigned long flags;
Christoph Lameter4037d452007-05-09 02:35:14 -07001205 int to_drain;
Cody P Schafer998d39c2013-07-03 15:01:32 -07001206 unsigned long batch;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001207
Christoph Lameter4037d452007-05-09 02:35:14 -07001208 local_irq_save(flags);
Cody P Schafer998d39c2013-07-03 15:01:32 -07001209 batch = ACCESS_ONCE(pcp->batch);
1210 if (pcp->count >= batch)
1211 to_drain = batch;
Christoph Lameter4037d452007-05-09 02:35:14 -07001212 else
1213 to_drain = pcp->count;
KOSAKI Motohiro2a135152012-07-31 16:42:53 -07001214 if (to_drain > 0) {
1215 free_pcppages_bulk(zone, to_drain, pcp);
1216 pcp->count -= to_drain;
1217 }
Christoph Lameter4037d452007-05-09 02:35:14 -07001218 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001219}
1220#endif
1221
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001222/*
1223 * Drain pages of the indicated processor.
1224 *
1225 * The processor must either be the current processor and the
1226 * thread pinned to the current processor or a processor that
1227 * is not online.
1228 */
1229static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230{
Nick Pigginc54ad302006-01-06 00:10:56 -08001231 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07001234 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 struct per_cpu_pageset *pset;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001236 struct per_cpu_pages *pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001238 local_irq_save(flags);
1239 pset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001241 pcp = &pset->pcp;
David Rientjes2ff754f2011-01-25 15:07:23 -08001242 if (pcp->count) {
1243 free_pcppages_bulk(zone, pcp->count, pcp);
1244 pcp->count = 0;
1245 }
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001246 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 }
1248}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001250/*
1251 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1252 */
1253void drain_local_pages(void *arg)
1254{
1255 drain_pages(smp_processor_id());
1256}
1257
1258/*
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001259 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1260 *
1261 * Note that this code is protected against sending an IPI to an offline
1262 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1263 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1264 * nothing keeps CPUs from showing up after we populated the cpumask and
1265 * before the call to on_each_cpu_mask().
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001266 */
1267void drain_all_pages(void)
1268{
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001269 int cpu;
1270 struct per_cpu_pageset *pcp;
1271 struct zone *zone;
1272
1273 /*
1274 * Allocate in the BSS so we wont require allocation in
1275 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1276 */
1277 static cpumask_t cpus_with_pcps;
1278
1279 /*
1280 * We don't care about racing with CPU hotplug event
1281 * as offline notification will cause the notified
1282 * cpu to drain that CPU pcps and on_each_cpu_mask
1283 * disables preemption as part of its processing
1284 */
1285 for_each_online_cpu(cpu) {
1286 bool has_pcps = false;
1287 for_each_populated_zone(zone) {
1288 pcp = per_cpu_ptr(zone->pageset, cpu);
1289 if (pcp->pcp.count) {
1290 has_pcps = true;
1291 break;
1292 }
1293 }
1294 if (has_pcps)
1295 cpumask_set_cpu(cpu, &cpus_with_pcps);
1296 else
1297 cpumask_clear_cpu(cpu, &cpus_with_pcps);
1298 }
1299 on_each_cpu_mask(&cpus_with_pcps, drain_local_pages, NULL, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001300}
1301
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02001302#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303
1304void mark_free_pages(struct zone *zone)
1305{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001306 unsigned long pfn, max_zone_pfn;
1307 unsigned long flags;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001308 int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 struct list_head *curr;
1310
Xishi Qiu8080fc02013-09-11 14:21:45 -07001311 if (zone_is_empty(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 return;
1313
1314 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001315
Cody P Schafer108bcc92013-02-22 16:35:23 -08001316 max_zone_pfn = zone_end_pfn(zone);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001317 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1318 if (pfn_valid(pfn)) {
1319 struct page *page = pfn_to_page(pfn);
1320
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001321 if (!swsusp_page_is_forbidden(page))
1322 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001323 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001325 for_each_migratetype_order(order, t) {
1326 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001327 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001329 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1330 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001331 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001332 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001333 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 spin_unlock_irqrestore(&zone->lock, flags);
1335}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001336#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337
1338/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 * Free a 0-order page
Li Hongfc916682010-03-05 13:41:54 -08001340 * cold == 1 ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 */
Li Hongfc916682010-03-05 13:41:54 -08001342void free_hot_cold_page(struct page *page, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343{
1344 struct zone *zone = page_zone(page);
1345 struct per_cpu_pages *pcp;
1346 unsigned long flags;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001347 int migratetype;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001349 if (!free_pages_prepare(page, 0))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001350 return;
1351
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001352 migratetype = get_pageblock_migratetype(page);
Minchan Kimb12c4ad2012-10-08 16:32:08 -07001353 set_freepage_migratetype(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001355 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001356
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001357 /*
1358 * We only track unmovable, reclaimable and movable on pcp lists.
1359 * Free ISOLATE pages back to the allocator because they are being
1360 * offlined but treat RESERVE as movable pages so we can get those
1361 * areas back if necessary. Otherwise, we may have to free
1362 * excessively into the page allocator
1363 */
1364 if (migratetype >= MIGRATE_PCPTYPES) {
Minchan Kim194159f2013-02-22 16:33:58 -08001365 if (unlikely(is_migrate_isolate(migratetype))) {
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001366 free_one_page(zone, page, 0, migratetype);
1367 goto out;
1368 }
1369 migratetype = MIGRATE_MOVABLE;
1370 }
1371
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001372 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001373 if (cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001374 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001375 else
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001376 list_add(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001378 if (pcp->count >= pcp->high) {
Cody P Schafer998d39c2013-07-03 15:01:32 -07001379 unsigned long batch = ACCESS_ONCE(pcp->batch);
1380 free_pcppages_bulk(zone, batch, pcp);
1381 pcp->count -= batch;
Nick Piggin48db57f2006-01-08 01:00:42 -08001382 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001383
1384out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386}
1387
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001388/*
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001389 * Free a list of 0-order pages
1390 */
1391void free_hot_cold_page_list(struct list_head *list, int cold)
1392{
1393 struct page *page, *next;
1394
1395 list_for_each_entry_safe(page, next, list, lru) {
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -08001396 trace_mm_page_free_batched(page, cold);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001397 free_hot_cold_page(page, cold);
1398 }
1399}
1400
1401/*
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001402 * split_page takes a non-compound higher-order page, and splits it into
1403 * n (1<<order) sub-pages: page[0..n]
1404 * Each sub-page must be freed individually.
1405 *
1406 * Note: this is probably too low level an operation for use in drivers.
1407 * Please consult with lkml before using this in your driver.
1408 */
1409void split_page(struct page *page, unsigned int order)
1410{
1411 int i;
1412
Nick Piggin725d7042006-09-25 23:30:55 -07001413 VM_BUG_ON(PageCompound(page));
1414 VM_BUG_ON(!page_count(page));
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001415
1416#ifdef CONFIG_KMEMCHECK
1417 /*
1418 * Split shadow pages too, because free(page[0]) would
1419 * otherwise free the whole shadow.
1420 */
1421 if (kmemcheck_page_is_tracked(page))
1422 split_page(virt_to_page(page[0].shadow), order);
1423#endif
1424
Nick Piggin7835e982006-03-22 00:08:40 -08001425 for (i = 1; i < (1 << order); i++)
1426 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001427}
K. Y. Srinivasan5853ff22013-03-25 15:47:38 -07001428EXPORT_SYMBOL_GPL(split_page);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001429
Mel Gorman8fb74b92013-01-11 14:32:16 -08001430static int __isolate_free_page(struct page *page, unsigned int order)
Mel Gorman748446b2010-05-24 14:32:27 -07001431{
Mel Gorman748446b2010-05-24 14:32:27 -07001432 unsigned long watermark;
1433 struct zone *zone;
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07001434 int mt;
Mel Gorman748446b2010-05-24 14:32:27 -07001435
1436 BUG_ON(!PageBuddy(page));
1437
1438 zone = page_zone(page);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001439 mt = get_pageblock_migratetype(page);
Mel Gorman748446b2010-05-24 14:32:27 -07001440
Minchan Kim194159f2013-02-22 16:33:58 -08001441 if (!is_migrate_isolate(mt)) {
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001442 /* Obey watermarks as if the page was being allocated */
1443 watermark = low_wmark_pages(zone) + (1 << order);
1444 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1445 return 0;
1446
Mel Gorman8fb74b92013-01-11 14:32:16 -08001447 __mod_zone_freepage_state(zone, -(1UL << order), mt);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001448 }
Mel Gorman748446b2010-05-24 14:32:27 -07001449
1450 /* Remove page from free list */
1451 list_del(&page->lru);
1452 zone->free_area[order].nr_free--;
1453 rmv_page_order(page);
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07001454
Mel Gorman8fb74b92013-01-11 14:32:16 -08001455 /* Set the pageblock if the isolated page is at least a pageblock */
Mel Gorman748446b2010-05-24 14:32:27 -07001456 if (order >= pageblock_order - 1) {
1457 struct page *endpage = page + (1 << order) - 1;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001458 for (; page < endpage; page += pageblock_nr_pages) {
1459 int mt = get_pageblock_migratetype(page);
Minchan Kim194159f2013-02-22 16:33:58 -08001460 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001461 set_pageblock_migratetype(page,
1462 MIGRATE_MOVABLE);
1463 }
Mel Gorman748446b2010-05-24 14:32:27 -07001464 }
1465
Mel Gorman8fb74b92013-01-11 14:32:16 -08001466 return 1UL << order;
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001467}
1468
1469/*
1470 * Similar to split_page except the page is already free. As this is only
1471 * being used for migration, the migratetype of the block also changes.
1472 * As this is called with interrupts disabled, the caller is responsible
1473 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1474 * are enabled.
1475 *
1476 * Note: this is probably too low level an operation for use in drivers.
1477 * Please consult with lkml before using this in your driver.
1478 */
1479int split_free_page(struct page *page)
1480{
1481 unsigned int order;
1482 int nr_pages;
1483
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001484 order = page_order(page);
1485
Mel Gorman8fb74b92013-01-11 14:32:16 -08001486 nr_pages = __isolate_free_page(page, order);
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001487 if (!nr_pages)
1488 return 0;
1489
1490 /* Split into individual pages */
1491 set_page_refcounted(page);
1492 split_page(page, order);
1493 return nr_pages;
Mel Gorman748446b2010-05-24 14:32:27 -07001494}
1495
1496/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1498 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1499 * or two.
1500 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07001501static inline
1502struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001503 struct zone *zone, int order, gfp_t gfp_flags,
1504 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505{
1506 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001507 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 int cold = !!(gfp_flags & __GFP_COLD);
1509
Hugh Dickins689bceb2005-11-21 21:32:20 -08001510again:
Nick Piggin48db57f2006-01-08 01:00:42 -08001511 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001513 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 local_irq_save(flags);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001516 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1517 list = &pcp->lists[migratetype];
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001518 if (list_empty(list)) {
Mel Gorman535131e62007-10-16 01:25:49 -07001519 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001520 pcp->batch, list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001521 migratetype, cold);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001522 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07001523 goto failed;
Mel Gorman535131e62007-10-16 01:25:49 -07001524 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001525
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001526 if (cold)
1527 page = list_entry(list->prev, struct page, lru);
1528 else
1529 page = list_entry(list->next, struct page, lru);
1530
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001531 list_del(&page->lru);
1532 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001533 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07001534 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1535 /*
1536 * __GFP_NOFAIL is not to be used in new code.
1537 *
1538 * All __GFP_NOFAIL callers should be fixed so that they
1539 * properly detect and handle allocation failures.
1540 *
1541 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07001542 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07001543 * __GFP_NOFAIL.
1544 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07001545 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07001546 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001548 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001549 spin_unlock(&zone->lock);
1550 if (!page)
1551 goto failed;
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07001552 __mod_zone_freepage_state(zone, -(1 << order),
1553 get_pageblock_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 }
1555
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001556 __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
Christoph Lameterf8891e52006-06-30 01:55:45 -07001557 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Andi Kleen78afd562011-03-22 16:33:12 -07001558 zone_statistics(preferred_zone, zone, gfp_flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001559 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560
Nick Piggin725d7042006-09-25 23:30:55 -07001561 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -08001562 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001563 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001565
1566failed:
1567 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001568 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569}
1570
Akinobu Mita933e3122006-12-08 02:39:45 -08001571#ifdef CONFIG_FAIL_PAGE_ALLOC
1572
Akinobu Mitab2588c42011-07-26 16:09:03 -07001573static struct {
Akinobu Mita933e3122006-12-08 02:39:45 -08001574 struct fault_attr attr;
1575
1576 u32 ignore_gfp_highmem;
1577 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001578 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001579} fail_page_alloc = {
1580 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001581 .ignore_gfp_wait = 1,
1582 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001583 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001584};
1585
1586static int __init setup_fail_page_alloc(char *str)
1587{
1588 return setup_fault_attr(&fail_page_alloc.attr, str);
1589}
1590__setup("fail_page_alloc=", setup_fail_page_alloc);
1591
Gavin Shandeaf3862012-07-31 16:41:51 -07001592static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08001593{
Akinobu Mita54114992007-07-15 23:40:23 -07001594 if (order < fail_page_alloc.min_order)
Gavin Shandeaf3862012-07-31 16:41:51 -07001595 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001596 if (gfp_mask & __GFP_NOFAIL)
Gavin Shandeaf3862012-07-31 16:41:51 -07001597 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001598 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
Gavin Shandeaf3862012-07-31 16:41:51 -07001599 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001600 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
Gavin Shandeaf3862012-07-31 16:41:51 -07001601 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001602
1603 return should_fail(&fail_page_alloc.attr, 1 << order);
1604}
1605
1606#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1607
1608static int __init fail_page_alloc_debugfs(void)
1609{
Al Virof4ae40a2011-07-24 04:33:43 -04001610 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
Akinobu Mita933e3122006-12-08 02:39:45 -08001611 struct dentry *dir;
Akinobu Mita933e3122006-12-08 02:39:45 -08001612
Akinobu Mitadd48c082011-08-03 16:21:01 -07001613 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
1614 &fail_page_alloc.attr);
1615 if (IS_ERR(dir))
1616 return PTR_ERR(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08001617
Akinobu Mitab2588c42011-07-26 16:09:03 -07001618 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
1619 &fail_page_alloc.ignore_gfp_wait))
1620 goto fail;
1621 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1622 &fail_page_alloc.ignore_gfp_highmem))
1623 goto fail;
1624 if (!debugfs_create_u32("min-order", mode, dir,
1625 &fail_page_alloc.min_order))
1626 goto fail;
Akinobu Mita933e3122006-12-08 02:39:45 -08001627
Akinobu Mitab2588c42011-07-26 16:09:03 -07001628 return 0;
1629fail:
Akinobu Mitadd48c082011-08-03 16:21:01 -07001630 debugfs_remove_recursive(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08001631
Akinobu Mitab2588c42011-07-26 16:09:03 -07001632 return -ENOMEM;
Akinobu Mita933e3122006-12-08 02:39:45 -08001633}
1634
1635late_initcall(fail_page_alloc_debugfs);
1636
1637#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1638
1639#else /* CONFIG_FAIL_PAGE_ALLOC */
1640
Gavin Shandeaf3862012-07-31 16:41:51 -07001641static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08001642{
Gavin Shandeaf3862012-07-31 16:41:51 -07001643 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001644}
1645
1646#endif /* CONFIG_FAIL_PAGE_ALLOC */
1647
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648/*
Mel Gorman88f5acf2011-01-13 15:45:41 -08001649 * Return true if free pages are above 'mark'. This takes into account the order
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 * of the allocation.
1651 */
Mel Gorman88f5acf2011-01-13 15:45:41 -08001652static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1653 int classzone_idx, int alloc_flags, long free_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654{
1655 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001656 long min = mark;
Minchan Kim2cfed072012-07-31 16:43:53 -07001657 long lowmem_reserve = z->lowmem_reserve[classzone_idx];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 int o;
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07001659 long free_cma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660
Michal Hockodf0a6da2012-01-10 15:08:02 -08001661 free_pages -= (1 << order) - 1;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001662 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001664 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 min -= min / 4;
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07001666#ifdef CONFIG_CMA
1667 /* If allocation can't use CMA areas don't use free CMA pages */
1668 if (!(alloc_flags & ALLOC_CMA))
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07001669 free_cma = zone_page_state(z, NR_FREE_CMA_PAGES);
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07001670#endif
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07001671
1672 if (free_pages - free_cma <= min + lowmem_reserve)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001673 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 for (o = 0; o < order; o++) {
1675 /* At the next order, this order's pages become unavailable */
1676 free_pages -= z->free_area[o].nr_free << o;
1677
1678 /* Require fewer higher order pages to be free */
1679 min >>= 1;
1680
1681 if (free_pages <= min)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001682 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 }
Mel Gorman88f5acf2011-01-13 15:45:41 -08001684 return true;
1685}
1686
1687bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1688 int classzone_idx, int alloc_flags)
1689{
1690 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1691 zone_page_state(z, NR_FREE_PAGES));
1692}
1693
1694bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
1695 int classzone_idx, int alloc_flags)
1696{
1697 long free_pages = zone_page_state(z, NR_FREE_PAGES);
1698
1699 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1700 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1701
1702 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1703 free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704}
1705
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001706#ifdef CONFIG_NUMA
1707/*
1708 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1709 * skip over zones that are not allowed by the cpuset, or that have
1710 * been recently (in last second) found to be nearly full. See further
1711 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001712 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001713 *
1714 * If the zonelist cache is present in the passed in zonelist, then
1715 * returns a pointer to the allowed node mask (either the current
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08001716 * tasks mems_allowed, or node_states[N_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001717 *
1718 * If the zonelist cache is not available for this zonelist, does
1719 * nothing and returns NULL.
1720 *
1721 * If the fullzones BITMAP in the zonelist cache is stale (more than
1722 * a second since last zap'd) then we zap it out (clear its bits.)
1723 *
1724 * We hold off even calling zlc_setup, until after we've checked the
1725 * first zone in the zonelist, on the theory that most allocations will
1726 * be satisfied from that first zone, so best to examine that zone as
1727 * quickly as we can.
1728 */
1729static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1730{
1731 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1732 nodemask_t *allowednodes; /* zonelist_cache approximation */
1733
1734 zlc = zonelist->zlcache_ptr;
1735 if (!zlc)
1736 return NULL;
1737
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001738 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001739 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1740 zlc->last_full_zap = jiffies;
1741 }
1742
1743 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1744 &cpuset_current_mems_allowed :
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08001745 &node_states[N_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001746 return allowednodes;
1747}
1748
1749/*
1750 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1751 * if it is worth looking at further for free memory:
1752 * 1) Check that the zone isn't thought to be full (doesn't have its
1753 * bit set in the zonelist_cache fullzones BITMAP).
1754 * 2) Check that the zones node (obtained from the zonelist_cache
1755 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1756 * Return true (non-zero) if zone is worth looking at further, or
1757 * else return false (zero) if it is not.
1758 *
1759 * This check -ignores- the distinction between various watermarks,
1760 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1761 * found to be full for any variation of these watermarks, it will
1762 * be considered full for up to one second by all requests, unless
1763 * we are so low on memory on all allowed nodes that we are forced
1764 * into the second scan of the zonelist.
1765 *
1766 * In the second scan we ignore this zonelist cache and exactly
1767 * apply the watermarks to all zones, even it is slower to do so.
1768 * We are low on memory in the second scan, and should leave no stone
1769 * unturned looking for a free page.
1770 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001771static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001772 nodemask_t *allowednodes)
1773{
1774 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1775 int i; /* index of *z in zonelist zones */
1776 int n; /* node that zone *z is on */
1777
1778 zlc = zonelist->zlcache_ptr;
1779 if (!zlc)
1780 return 1;
1781
Mel Gormandd1a2392008-04-28 02:12:17 -07001782 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001783 n = zlc->z_to_n[i];
1784
1785 /* This zone is worth trying if it is allowed but not full */
1786 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1787}
1788
1789/*
1790 * Given 'z' scanning a zonelist, set the corresponding bit in
1791 * zlc->fullzones, so that subsequent attempts to allocate a page
1792 * from that zone don't waste time re-examining it.
1793 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001794static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001795{
1796 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1797 int i; /* index of *z in zonelist zones */
1798
1799 zlc = zonelist->zlcache_ptr;
1800 if (!zlc)
1801 return;
1802
Mel Gormandd1a2392008-04-28 02:12:17 -07001803 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001804
1805 set_bit(i, zlc->fullzones);
1806}
1807
Mel Gorman76d3fbf2011-07-25 17:12:30 -07001808/*
1809 * clear all zones full, called after direct reclaim makes progress so that
1810 * a zone that was recently full is not skipped over for up to a second
1811 */
1812static void zlc_clear_zones_full(struct zonelist *zonelist)
1813{
1814 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1815
1816 zlc = zonelist->zlcache_ptr;
1817 if (!zlc)
1818 return;
1819
1820 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1821}
1822
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001823static bool zone_local(struct zone *local_zone, struct zone *zone)
1824{
1825 return node_distance(local_zone->node, zone->node) == LOCAL_DISTANCE;
1826}
1827
David Rientjes957f8222012-10-08 16:33:24 -07001828static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
1829{
1830 return node_isset(local_zone->node, zone->zone_pgdat->reclaim_nodes);
1831}
1832
1833static void __paginginit init_zone_allows_reclaim(int nid)
1834{
1835 int i;
1836
1837 for_each_online_node(i)
David Rientjes6b187d022012-10-25 13:38:08 -07001838 if (node_distance(nid, i) <= RECLAIM_DISTANCE)
David Rientjes957f8222012-10-08 16:33:24 -07001839 node_set(i, NODE_DATA(nid)->reclaim_nodes);
David Rientjes6b187d022012-10-25 13:38:08 -07001840 else
David Rientjes957f8222012-10-08 16:33:24 -07001841 zone_reclaim_mode = 1;
David Rientjes957f8222012-10-08 16:33:24 -07001842}
1843
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001844#else /* CONFIG_NUMA */
1845
1846static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1847{
1848 return NULL;
1849}
1850
Mel Gormandd1a2392008-04-28 02:12:17 -07001851static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001852 nodemask_t *allowednodes)
1853{
1854 return 1;
1855}
1856
Mel Gormandd1a2392008-04-28 02:12:17 -07001857static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001858{
1859}
Mel Gorman76d3fbf2011-07-25 17:12:30 -07001860
1861static void zlc_clear_zones_full(struct zonelist *zonelist)
1862{
1863}
David Rientjes957f8222012-10-08 16:33:24 -07001864
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001865static bool zone_local(struct zone *local_zone, struct zone *zone)
1866{
1867 return true;
1868}
1869
David Rientjes957f8222012-10-08 16:33:24 -07001870static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
1871{
1872 return true;
1873}
1874
1875static inline void init_zone_allows_reclaim(int nid)
1876{
1877}
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001878#endif /* CONFIG_NUMA */
1879
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001880/*
Paul Jackson0798e512006-12-06 20:31:38 -08001881 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001882 * a page.
1883 */
1884static struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001885get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
Mel Gorman5117f452009-06-16 15:31:59 -07001886 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
Mel Gorman3dd28262009-06-16 15:32:00 -07001887 struct zone *preferred_zone, int migratetype)
Martin Hicks753ee722005-06-21 17:14:41 -07001888{
Mel Gormandd1a2392008-04-28 02:12:17 -07001889 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001890 struct page *page = NULL;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001891 int classzone_idx;
Mel Gorman5117f452009-06-16 15:31:59 -07001892 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001893 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1894 int zlc_active = 0; /* set if using zonelist_cache */
1895 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gorman54a6eb52008-04-28 02:12:16 -07001896
Mel Gorman19770b32008-04-28 02:12:18 -07001897 classzone_idx = zone_idx(preferred_zone);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001898zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001899 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001900 * Scan zonelist, looking for a zone with enough free.
SeungHun Lee3b11f0a2013-09-11 14:22:23 -07001901 * See also __cpuset_node_allowed_softwall() comment in kernel/cpuset.c.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001902 */
Mel Gorman19770b32008-04-28 02:12:18 -07001903 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1904 high_zoneidx, nodemask) {
Johannes Weinere085dbc2013-09-11 14:20:46 -07001905 unsigned long mark;
1906
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08001907 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001908 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1909 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001910 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001911 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Mel Gormancd38b112011-07-25 17:12:29 -07001912 continue;
Johannes Weinere085dbc2013-09-11 14:20:46 -07001913 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
Joonsoo Kime66f0972013-09-11 14:21:26 -07001914 if (unlikely(alloc_flags & ALLOC_NO_WATERMARKS))
Johannes Weinere085dbc2013-09-11 14:20:46 -07001915 goto try_this_zone;
Johannes Weinera756cf52012-01-10 15:07:49 -08001916 /*
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07001917 * Distribute pages in proportion to the individual
1918 * zone size to ensure fair page aging. The zone a
1919 * page was allocated in should have no effect on the
1920 * time the page has in memory before being reclaimed.
1921 *
1922 * When zone_reclaim_mode is enabled, try to stay in
1923 * local zones in the fastpath. If that fails, the
1924 * slowpath is entered, which will do another pass
1925 * starting with the local zones, but ultimately fall
1926 * back to remote zones that do not partake in the
1927 * fairness round-robin cycle of this zonelist.
1928 */
1929 if (alloc_flags & ALLOC_WMARK_LOW) {
1930 if (zone_page_state(zone, NR_ALLOC_BATCH) <= 0)
1931 continue;
1932 if (zone_reclaim_mode &&
1933 !zone_local(preferred_zone, zone))
1934 continue;
1935 }
1936 /*
Johannes Weinera756cf52012-01-10 15:07:49 -08001937 * When allocating a page cache page for writing, we
1938 * want to get it from a zone that is within its dirty
1939 * limit, such that no single zone holds more than its
1940 * proportional share of globally allowed dirty pages.
1941 * The dirty limits take into account the zone's
1942 * lowmem reserves and high watermark so that kswapd
1943 * should be able to balance it without having to
1944 * write pages from its LRU list.
1945 *
1946 * This may look like it could increase pressure on
1947 * lower zones by failing allocations in higher zones
1948 * before they are full. But the pages that do spill
1949 * over are limited as the lower zones are protected
1950 * by this very same mechanism. It should not become
1951 * a practical burden to them.
1952 *
1953 * XXX: For now, allow allocations to potentially
1954 * exceed the per-zone dirty limit in the slowpath
1955 * (ALLOC_WMARK_LOW unset) before going into reclaim,
1956 * which is important when on a NUMA setup the allowed
1957 * zones are together not big enough to reach the
1958 * global limit. The proper fix for these situations
1959 * will require awareness of zones in the
1960 * dirty-throttling and the flusher threads.
1961 */
1962 if ((alloc_flags & ALLOC_WMARK_LOW) &&
1963 (gfp_mask & __GFP_WRITE) && !zone_dirty_ok(zone))
1964 goto this_zone_full;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001965
Johannes Weinere085dbc2013-09-11 14:20:46 -07001966 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
1967 if (!zone_watermark_ok(zone, order, mark,
1968 classzone_idx, alloc_flags)) {
Mel Gormanfa5e0842009-06-16 15:33:22 -07001969 int ret;
1970
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08001971 if (IS_ENABLED(CONFIG_NUMA) &&
1972 !did_zlc_setup && nr_online_nodes > 1) {
Mel Gormancd38b112011-07-25 17:12:29 -07001973 /*
1974 * we do zlc_setup if there are multiple nodes
1975 * and before considering the first zone allowed
1976 * by the cpuset.
1977 */
1978 allowednodes = zlc_setup(zonelist, alloc_flags);
1979 zlc_active = 1;
1980 did_zlc_setup = 1;
1981 }
1982
David Rientjes957f8222012-10-08 16:33:24 -07001983 if (zone_reclaim_mode == 0 ||
1984 !zone_allows_reclaim(preferred_zone, zone))
Mel Gormanfa5e0842009-06-16 15:33:22 -07001985 goto this_zone_full;
1986
Mel Gormancd38b112011-07-25 17:12:29 -07001987 /*
1988 * As we may have just activated ZLC, check if the first
1989 * eligible zone has failed zone_reclaim recently.
1990 */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08001991 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
Mel Gormancd38b112011-07-25 17:12:29 -07001992 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1993 continue;
1994
Mel Gormanfa5e0842009-06-16 15:33:22 -07001995 ret = zone_reclaim(zone, gfp_mask, order);
1996 switch (ret) {
1997 case ZONE_RECLAIM_NOSCAN:
1998 /* did not scan */
Mel Gormancd38b112011-07-25 17:12:29 -07001999 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002000 case ZONE_RECLAIM_FULL:
2001 /* scanned but unreclaimable */
Mel Gormancd38b112011-07-25 17:12:29 -07002002 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002003 default:
2004 /* did we reclaim enough */
Mel Gormanfed27192013-04-29 15:07:57 -07002005 if (zone_watermark_ok(zone, order, mark,
Mel Gormanfa5e0842009-06-16 15:33:22 -07002006 classzone_idx, alloc_flags))
Mel Gormanfed27192013-04-29 15:07:57 -07002007 goto try_this_zone;
2008
2009 /*
2010 * Failed to reclaim enough to meet watermark.
2011 * Only mark the zone full if checking the min
2012 * watermark or if we failed to reclaim just
2013 * 1<<order pages or else the page allocator
2014 * fastpath will prematurely mark zones full
2015 * when the watermark is between the low and
2016 * min watermarks.
2017 */
2018 if (((alloc_flags & ALLOC_WMARK_MASK) == ALLOC_WMARK_MIN) ||
2019 ret == ZONE_RECLAIM_SOME)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002020 goto this_zone_full;
Mel Gormanfed27192013-04-29 15:07:57 -07002021
2022 continue;
Paul Jackson0798e512006-12-06 20:31:38 -08002023 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002024 }
2025
Mel Gormanfa5e0842009-06-16 15:33:22 -07002026try_this_zone:
Mel Gorman3dd28262009-06-16 15:32:00 -07002027 page = buffered_rmqueue(preferred_zone, zone, order,
2028 gfp_mask, migratetype);
Paul Jackson0798e512006-12-06 20:31:38 -08002029 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002030 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002031this_zone_full:
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002032 if (IS_ENABLED(CONFIG_NUMA))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002033 zlc_mark_zone_full(zonelist, z);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002034 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002035
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002036 if (unlikely(IS_ENABLED(CONFIG_NUMA) && page == NULL && zlc_active)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002037 /* Disable zlc cache for second zonelist scan */
2038 zlc_active = 0;
2039 goto zonelist_scan;
2040 }
Alex Shib1211862012-08-21 16:16:08 -07002041
2042 if (page)
2043 /*
2044 * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was
2045 * necessary to allocate the page. The expectation is
2046 * that the caller is taking steps that will free more
2047 * memory. The caller should avoid the page being used
2048 * for !PFMEMALLOC purposes.
2049 */
2050 page->pfmemalloc = !!(alloc_flags & ALLOC_NO_WATERMARKS);
2051
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002052 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07002053}
2054
David Rientjes29423e772011-03-22 16:30:47 -07002055/*
2056 * Large machines with many possible nodes should not always dump per-node
2057 * meminfo in irq context.
2058 */
2059static inline bool should_suppress_show_mem(void)
2060{
2061 bool ret = false;
2062
2063#if NODES_SHIFT > 8
2064 ret = in_interrupt();
2065#endif
2066 return ret;
2067}
2068
Dave Hansena238ab52011-05-24 17:12:16 -07002069static DEFINE_RATELIMIT_STATE(nopage_rs,
2070 DEFAULT_RATELIMIT_INTERVAL,
2071 DEFAULT_RATELIMIT_BURST);
2072
2073void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
2074{
Dave Hansena238ab52011-05-24 17:12:16 -07002075 unsigned int filter = SHOW_MEM_FILTER_NODES;
2076
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -08002077 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2078 debug_guardpage_minorder() > 0)
Dave Hansena238ab52011-05-24 17:12:16 -07002079 return;
2080
2081 /*
David Rientjes4b59e6c2013-04-29 15:06:11 -07002082 * Walking all memory to count page types is very expensive and should
2083 * be inhibited in non-blockable contexts.
2084 */
2085 if (!(gfp_mask & __GFP_WAIT))
2086 filter |= SHOW_MEM_FILTER_PAGE_COUNT;
2087
2088 /*
Dave Hansena238ab52011-05-24 17:12:16 -07002089 * This documents exceptions given to allocations in certain
2090 * contexts that are allowed to allocate outside current's set
2091 * of allowed nodes.
2092 */
2093 if (!(gfp_mask & __GFP_NOMEMALLOC))
2094 if (test_thread_flag(TIF_MEMDIE) ||
2095 (current->flags & (PF_MEMALLOC | PF_EXITING)))
2096 filter &= ~SHOW_MEM_FILTER_NODES;
2097 if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
2098 filter &= ~SHOW_MEM_FILTER_NODES;
2099
2100 if (fmt) {
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002101 struct va_format vaf;
2102 va_list args;
2103
Dave Hansena238ab52011-05-24 17:12:16 -07002104 va_start(args, fmt);
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002105
2106 vaf.fmt = fmt;
2107 vaf.va = &args;
2108
2109 pr_warn("%pV", &vaf);
2110
Dave Hansena238ab52011-05-24 17:12:16 -07002111 va_end(args);
2112 }
2113
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002114 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2115 current->comm, order, gfp_mask);
Dave Hansena238ab52011-05-24 17:12:16 -07002116
2117 dump_stack();
2118 if (!should_suppress_show_mem())
2119 show_mem(filter);
2120}
2121
Mel Gorman11e33f62009-06-16 15:31:57 -07002122static inline int
2123should_alloc_retry(gfp_t gfp_mask, unsigned int order,
Mel Gormanf90ac392012-01-10 15:07:15 -08002124 unsigned long did_some_progress,
Mel Gorman11e33f62009-06-16 15:31:57 -07002125 unsigned long pages_reclaimed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126{
Mel Gorman11e33f62009-06-16 15:31:57 -07002127 /* Do not loop if specifically requested */
2128 if (gfp_mask & __GFP_NORETRY)
2129 return 0;
2130
Mel Gormanf90ac392012-01-10 15:07:15 -08002131 /* Always retry if specifically requested */
2132 if (gfp_mask & __GFP_NOFAIL)
2133 return 1;
2134
2135 /*
2136 * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim
2137 * making forward progress without invoking OOM. Suspend also disables
2138 * storage devices so kswapd will not help. Bail if we are suspending.
2139 */
2140 if (!did_some_progress && pm_suspended_storage())
2141 return 0;
2142
Mel Gorman11e33f62009-06-16 15:31:57 -07002143 /*
2144 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
2145 * means __GFP_NOFAIL, but that may not be true in other
2146 * implementations.
2147 */
2148 if (order <= PAGE_ALLOC_COSTLY_ORDER)
2149 return 1;
2150
2151 /*
2152 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
2153 * specified, then we retry until we no longer reclaim any pages
2154 * (above), or we've reclaimed an order of pages at least as
2155 * large as the allocation's order. In both cases, if the
2156 * allocation still fails, we stop retrying.
2157 */
2158 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
2159 return 1;
2160
Mel Gorman11e33f62009-06-16 15:31:57 -07002161 return 0;
2162}
2163
2164static inline struct page *
2165__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
2166 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002167 nodemask_t *nodemask, struct zone *preferred_zone,
2168 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002169{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171
Mel Gorman11e33f62009-06-16 15:31:57 -07002172 /* Acquire the OOM killer lock for the zones in zonelist */
Minchan Kimff321fe2010-08-09 17:18:57 -07002173 if (!try_set_zonelist_oom(zonelist, gfp_mask)) {
Mel Gorman11e33f62009-06-16 15:31:57 -07002174 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 return NULL;
2176 }
Jens Axboe6b1de912005-11-17 21:35:02 +01002177
Mel Gorman11e33f62009-06-16 15:31:57 -07002178 /*
2179 * Go through the zonelist yet one more time, keep very high watermark
2180 * here, this is only to catch a parallel oom killing, we must fail if
2181 * we're still under heavy pressure.
2182 */
2183 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
2184 order, zonelist, high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07002185 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07002186 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002187 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07002188 goto out;
2189
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002190 if (!(gfp_mask & __GFP_NOFAIL)) {
2191 /* The OOM killer will not help higher order allocs */
2192 if (order > PAGE_ALLOC_COSTLY_ORDER)
2193 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07002194 /* The OOM killer does not needlessly kill tasks for lowmem */
2195 if (high_zoneidx < ZONE_NORMAL)
2196 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002197 /*
2198 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
2199 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
2200 * The caller should handle page allocation failure by itself if
2201 * it specifies __GFP_THISNODE.
2202 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
2203 */
2204 if (gfp_mask & __GFP_THISNODE)
2205 goto out;
2206 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002207 /* Exhausted what can be done so it's blamo time */
David Rientjes08ab9b12012-03-21 16:34:04 -07002208 out_of_memory(zonelist, gfp_mask, order, nodemask, false);
Mel Gorman11e33f62009-06-16 15:31:57 -07002209
2210out:
2211 clear_zonelist_oom(zonelist, gfp_mask);
2212 return page;
2213}
2214
Mel Gorman56de7262010-05-24 14:32:30 -07002215#ifdef CONFIG_COMPACTION
2216/* Try memory compaction for high-order allocations before reclaim */
2217static struct page *
2218__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2219 struct zonelist *zonelist, enum zone_type high_zoneidx,
2220 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman66199712012-01-12 17:19:41 -08002221 int migratetype, bool sync_migration,
Mel Gormanc67fe372012-08-21 16:16:17 -07002222 bool *contended_compaction, bool *deferred_compaction,
Mel Gorman66199712012-01-12 17:19:41 -08002223 unsigned long *did_some_progress)
Mel Gorman56de7262010-05-24 14:32:30 -07002224{
Mel Gorman66199712012-01-12 17:19:41 -08002225 if (!order)
Mel Gorman56de7262010-05-24 14:32:30 -07002226 return NULL;
2227
Rik van Rielaff62242012-03-21 16:33:52 -07002228 if (compaction_deferred(preferred_zone, order)) {
Mel Gorman66199712012-01-12 17:19:41 -08002229 *deferred_compaction = true;
2230 return NULL;
2231 }
2232
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002233 current->flags |= PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07002234 *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
Mel Gormanc67fe372012-08-21 16:16:17 -07002235 nodemask, sync_migration,
Mel Gorman8fb74b92013-01-11 14:32:16 -08002236 contended_compaction);
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002237 current->flags &= ~PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07002238
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07002239 if (*did_some_progress != COMPACT_SKIPPED) {
Mel Gorman8fb74b92013-01-11 14:32:16 -08002240 struct page *page;
2241
Mel Gorman56de7262010-05-24 14:32:30 -07002242 /* Page migration frees to the PCP lists but we want merging */
2243 drain_pages(get_cpu());
2244 put_cpu();
2245
2246 page = get_page_from_freelist(gfp_mask, nodemask,
2247 order, zonelist, high_zoneidx,
Mel Gormancfd19c52012-07-31 16:44:10 -07002248 alloc_flags & ~ALLOC_NO_WATERMARKS,
2249 preferred_zone, migratetype);
Mel Gorman56de7262010-05-24 14:32:30 -07002250 if (page) {
Mel Gorman62997022012-10-08 16:32:47 -07002251 preferred_zone->compact_blockskip_flush = false;
Mel Gorman4f92e252010-05-24 14:32:32 -07002252 preferred_zone->compact_considered = 0;
2253 preferred_zone->compact_defer_shift = 0;
Rik van Rielaff62242012-03-21 16:33:52 -07002254 if (order >= preferred_zone->compact_order_failed)
2255 preferred_zone->compact_order_failed = order + 1;
Mel Gorman56de7262010-05-24 14:32:30 -07002256 count_vm_event(COMPACTSUCCESS);
2257 return page;
2258 }
2259
2260 /*
2261 * It's bad if compaction run occurs and fails.
2262 * The most likely reason is that pages exist,
2263 * but not enough to satisfy watermarks.
2264 */
2265 count_vm_event(COMPACTFAIL);
Mel Gorman66199712012-01-12 17:19:41 -08002266
2267 /*
2268 * As async compaction considers a subset of pageblocks, only
2269 * defer if the failure was a sync compaction failure.
2270 */
2271 if (sync_migration)
Rik van Rielaff62242012-03-21 16:33:52 -07002272 defer_compaction(preferred_zone, order);
Mel Gorman56de7262010-05-24 14:32:30 -07002273
2274 cond_resched();
2275 }
2276
2277 return NULL;
2278}
2279#else
2280static inline struct page *
2281__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2282 struct zonelist *zonelist, enum zone_type high_zoneidx,
2283 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman66199712012-01-12 17:19:41 -08002284 int migratetype, bool sync_migration,
Mel Gormanc67fe372012-08-21 16:16:17 -07002285 bool *contended_compaction, bool *deferred_compaction,
Mel Gorman66199712012-01-12 17:19:41 -08002286 unsigned long *did_some_progress)
Mel Gorman56de7262010-05-24 14:32:30 -07002287{
2288 return NULL;
2289}
2290#endif /* CONFIG_COMPACTION */
2291
Marek Szyprowskibba90712012-01-25 12:09:52 +01002292/* Perform direct synchronous page reclaim */
2293static int
2294__perform_reclaim(gfp_t gfp_mask, unsigned int order, struct zonelist *zonelist,
2295 nodemask_t *nodemask)
Mel Gorman11e33f62009-06-16 15:31:57 -07002296{
Mel Gorman11e33f62009-06-16 15:31:57 -07002297 struct reclaim_state reclaim_state;
Marek Szyprowskibba90712012-01-25 12:09:52 +01002298 int progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07002299
2300 cond_resched();
2301
2302 /* We now go into synchronous reclaim */
2303 cpuset_memory_pressure_bump();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002304 current->flags |= PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002305 lockdep_set_current_reclaim_state(gfp_mask);
2306 reclaim_state.reclaimed_slab = 0;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002307 current->reclaim_state = &reclaim_state;
Mel Gorman11e33f62009-06-16 15:31:57 -07002308
Marek Szyprowskibba90712012-01-25 12:09:52 +01002309 progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002310
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002311 current->reclaim_state = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002312 lockdep_clear_current_reclaim_state();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002313 current->flags &= ~PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002314
2315 cond_resched();
2316
Marek Szyprowskibba90712012-01-25 12:09:52 +01002317 return progress;
2318}
2319
2320/* The really slow allocator path where we enter direct reclaim */
2321static inline struct page *
2322__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
2323 struct zonelist *zonelist, enum zone_type high_zoneidx,
2324 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
2325 int migratetype, unsigned long *did_some_progress)
2326{
2327 struct page *page = NULL;
2328 bool drained = false;
2329
2330 *did_some_progress = __perform_reclaim(gfp_mask, order, zonelist,
2331 nodemask);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002332 if (unlikely(!(*did_some_progress)))
2333 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002334
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002335 /* After successful reclaim, reconsider all zones for allocation */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002336 if (IS_ENABLED(CONFIG_NUMA))
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002337 zlc_clear_zones_full(zonelist);
2338
Mel Gorman9ee493c2010-09-09 16:38:18 -07002339retry:
2340 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002341 zonelist, high_zoneidx,
Mel Gormancfd19c52012-07-31 16:44:10 -07002342 alloc_flags & ~ALLOC_NO_WATERMARKS,
2343 preferred_zone, migratetype);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002344
2345 /*
2346 * If an allocation failed after direct reclaim, it could be because
2347 * pages are pinned on the per-cpu lists. Drain them and try again
2348 */
2349 if (!page && !drained) {
2350 drain_all_pages();
2351 drained = true;
2352 goto retry;
2353 }
2354
Mel Gorman11e33f62009-06-16 15:31:57 -07002355 return page;
2356}
2357
Mel Gorman11e33f62009-06-16 15:31:57 -07002358/*
2359 * This is called in the allocator slow-path if the allocation request is of
2360 * sufficient urgency to ignore watermarks and take other desperate measures
2361 */
2362static inline struct page *
2363__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
2364 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002365 nodemask_t *nodemask, struct zone *preferred_zone,
2366 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002367{
2368 struct page *page;
2369
2370 do {
2371 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002372 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
Mel Gorman3dd28262009-06-16 15:32:00 -07002373 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002374
2375 if (!page && gfp_mask & __GFP_NOFAIL)
Mel Gorman0e093d992010-10-26 14:21:45 -07002376 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07002377 } while (!page && (gfp_mask & __GFP_NOFAIL));
2378
2379 return page;
2380}
2381
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002382static void prepare_slowpath(gfp_t gfp_mask, unsigned int order,
2383 struct zonelist *zonelist,
2384 enum zone_type high_zoneidx,
2385 struct zone *preferred_zone)
Mel Gorman11e33f62009-06-16 15:31:57 -07002386{
2387 struct zoneref *z;
2388 struct zone *zone;
2389
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002390 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
2391 if (!(gfp_mask & __GFP_NO_KSWAPD))
2392 wakeup_kswapd(zone, order, zone_idx(preferred_zone));
2393 /*
2394 * Only reset the batches of zones that were actually
2395 * considered in the fast path, we don't want to
2396 * thrash fairness information for zones that are not
2397 * actually part of this zonelist's round-robin cycle.
2398 */
2399 if (zone_reclaim_mode && !zone_local(preferred_zone, zone))
2400 continue;
2401 mod_zone_page_state(zone, NR_ALLOC_BATCH,
2402 high_wmark_pages(zone) -
2403 low_wmark_pages(zone) -
2404 zone_page_state(zone, NR_ALLOC_BATCH));
2405 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002406}
2407
Peter Zijlstra341ce062009-06-16 15:32:02 -07002408static inline int
2409gfp_to_alloc_flags(gfp_t gfp_mask)
2410{
Peter Zijlstra341ce062009-06-16 15:32:02 -07002411 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
2412 const gfp_t wait = gfp_mask & __GFP_WAIT;
2413
Mel Gormana56f57f2009-06-16 15:32:02 -07002414 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07002415 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07002416
Peter Zijlstra341ce062009-06-16 15:32:02 -07002417 /*
2418 * The caller may dip into page reserves a bit more if the caller
2419 * cannot run direct reclaim, or if the caller has realtime scheduling
2420 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2421 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
2422 */
Namhyung Kime6223a32010-10-26 14:21:59 -07002423 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07002424
2425 if (!wait) {
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002426 /*
2427 * Not worth trying to allocate harder for
2428 * __GFP_NOMEMALLOC even if it can't schedule.
2429 */
2430 if (!(gfp_mask & __GFP_NOMEMALLOC))
2431 alloc_flags |= ALLOC_HARDER;
Peter Zijlstra341ce062009-06-16 15:32:02 -07002432 /*
2433 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
2434 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
2435 */
2436 alloc_flags &= ~ALLOC_CPUSET;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002437 } else if (unlikely(rt_task(current)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07002438 alloc_flags |= ALLOC_HARDER;
2439
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002440 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2441 if (gfp_mask & __GFP_MEMALLOC)
2442 alloc_flags |= ALLOC_NO_WATERMARKS;
Mel Gorman907aed42012-07-31 16:44:07 -07002443 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
2444 alloc_flags |= ALLOC_NO_WATERMARKS;
2445 else if (!in_interrupt() &&
2446 ((current->flags & PF_MEMALLOC) ||
2447 unlikely(test_thread_flag(TIF_MEMDIE))))
Peter Zijlstra341ce062009-06-16 15:32:02 -07002448 alloc_flags |= ALLOC_NO_WATERMARKS;
2449 }
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002450#ifdef CONFIG_CMA
2451 if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2452 alloc_flags |= ALLOC_CMA;
2453#endif
Peter Zijlstra341ce062009-06-16 15:32:02 -07002454 return alloc_flags;
2455}
2456
Mel Gorman072bb0a2012-07-31 16:43:58 -07002457bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
2458{
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002459 return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
Mel Gorman072bb0a2012-07-31 16:43:58 -07002460}
2461
Mel Gorman11e33f62009-06-16 15:31:57 -07002462static inline struct page *
2463__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
2464 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002465 nodemask_t *nodemask, struct zone *preferred_zone,
2466 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002467{
2468 const gfp_t wait = gfp_mask & __GFP_WAIT;
2469 struct page *page = NULL;
2470 int alloc_flags;
2471 unsigned long pages_reclaimed = 0;
2472 unsigned long did_some_progress;
Mel Gorman77f1fe62011-01-13 15:45:57 -08002473 bool sync_migration = false;
Mel Gorman66199712012-01-12 17:19:41 -08002474 bool deferred_compaction = false;
Mel Gormanc67fe372012-08-21 16:16:17 -07002475 bool contended_compaction = false;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002476
Christoph Lameter952f3b52006-12-06 20:33:26 -08002477 /*
Mel Gorman72807a72009-06-16 15:32:18 -07002478 * In the slowpath, we sanity check order to avoid ever trying to
2479 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2480 * be using allocators in order of preference for an area that is
2481 * too large.
2482 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07002483 if (order >= MAX_ORDER) {
2484 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07002485 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07002486 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002487
Christoph Lameter952f3b52006-12-06 20:33:26 -08002488 /*
2489 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2490 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2491 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2492 * using a larger set of nodes after it has established that the
2493 * allowed per node queues are empty and that nodes are
2494 * over allocated.
2495 */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002496 if (IS_ENABLED(CONFIG_NUMA) &&
2497 (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
Christoph Lameter952f3b52006-12-06 20:33:26 -08002498 goto nopage;
2499
Mel Gormancc4a6852009-11-11 14:26:14 -08002500restart:
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002501 prepare_slowpath(gfp_mask, order, zonelist,
2502 high_zoneidx, preferred_zone);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002503
Paul Jackson9bf22292005-09-06 15:18:12 -07002504 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002505 * OK, we're below the kswapd watermark and have kicked background
2506 * reclaim. Now things get more complex, so set up alloc_flags according
2507 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07002508 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002509 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510
David Rientjesf33261d2011-01-25 15:07:20 -08002511 /*
2512 * Find the true preferred zone if the allocation is unconstrained by
2513 * cpusets.
2514 */
2515 if (!(alloc_flags & ALLOC_CPUSET) && !nodemask)
2516 first_zones_zonelist(zonelist, high_zoneidx, NULL,
2517 &preferred_zone);
2518
Andrew Barrycfa54a02011-05-24 17:12:52 -07002519rebalance:
Peter Zijlstra341ce062009-06-16 15:32:02 -07002520 /* This is the last chance, in general, before the goto nopage. */
Mel Gorman19770b32008-04-28 02:12:18 -07002521 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
Peter Zijlstra341ce062009-06-16 15:32:02 -07002522 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
2523 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002524 if (page)
2525 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526
Mel Gorman11e33f62009-06-16 15:31:57 -07002527 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002528 if (alloc_flags & ALLOC_NO_WATERMARKS) {
Mel Gorman183f6372012-07-31 16:44:12 -07002529 /*
2530 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
2531 * the allocation is high priority and these type of
2532 * allocations are system rather than user orientated
2533 */
2534 zonelist = node_zonelist(numa_node_id(), gfp_mask);
2535
Peter Zijlstra341ce062009-06-16 15:32:02 -07002536 page = __alloc_pages_high_priority(gfp_mask, order,
2537 zonelist, high_zoneidx, nodemask,
2538 preferred_zone, migratetype);
Mel Gormancfd19c52012-07-31 16:44:10 -07002539 if (page) {
Peter Zijlstra341ce062009-06-16 15:32:02 -07002540 goto got_pg;
Mel Gormancfd19c52012-07-31 16:44:10 -07002541 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 }
2543
2544 /* Atomic allocations - we can't balance anything */
2545 if (!wait)
2546 goto nopage;
2547
Peter Zijlstra341ce062009-06-16 15:32:02 -07002548 /* Avoid recursion of direct reclaim */
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002549 if (current->flags & PF_MEMALLOC)
Peter Zijlstra341ce062009-06-16 15:32:02 -07002550 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551
David Rientjes6583bb62009-07-29 15:02:06 -07002552 /* Avoid allocations with no watermarks from looping endlessly */
2553 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2554 goto nopage;
2555
Mel Gorman77f1fe62011-01-13 15:45:57 -08002556 /*
2557 * Try direct compaction. The first pass is asynchronous. Subsequent
2558 * attempts after direct reclaim are synchronous
2559 */
Mel Gorman56de7262010-05-24 14:32:30 -07002560 page = __alloc_pages_direct_compact(gfp_mask, order,
2561 zonelist, high_zoneidx,
2562 nodemask,
2563 alloc_flags, preferred_zone,
Mel Gorman66199712012-01-12 17:19:41 -08002564 migratetype, sync_migration,
Mel Gormanc67fe372012-08-21 16:16:17 -07002565 &contended_compaction,
Mel Gorman66199712012-01-12 17:19:41 -08002566 &deferred_compaction,
2567 &did_some_progress);
Mel Gorman56de7262010-05-24 14:32:30 -07002568 if (page)
2569 goto got_pg;
Andrea Arcangelic6a140b2011-05-24 17:11:38 -07002570 sync_migration = true;
Mel Gorman56de7262010-05-24 14:32:30 -07002571
Linus Torvalds31f8d422012-12-10 10:47:45 -08002572 /*
2573 * If compaction is deferred for high-order allocations, it is because
2574 * sync compaction recently failed. In this is the case and the caller
2575 * requested a movable allocation that does not heavily disrupt the
2576 * system then fail the allocation instead of entering direct reclaim.
2577 */
2578 if ((deferred_compaction || contended_compaction) &&
Linus Torvaldscaf49192012-12-10 10:51:16 -08002579 (gfp_mask & __GFP_NO_KSWAPD))
Linus Torvalds31f8d422012-12-10 10:47:45 -08002580 goto nopage;
Mel Gorman66199712012-01-12 17:19:41 -08002581
Mel Gorman11e33f62009-06-16 15:31:57 -07002582 /* Try direct reclaim and then allocating */
2583 page = __alloc_pages_direct_reclaim(gfp_mask, order,
2584 zonelist, high_zoneidx,
2585 nodemask,
Mel Gorman5117f452009-06-16 15:31:59 -07002586 alloc_flags, preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07002587 migratetype, &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07002588 if (page)
2589 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591 /*
Mel Gorman11e33f62009-06-16 15:31:57 -07002592 * If we failed to make any progress reclaiming, then we are
2593 * running out of options and have to consider going OOM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594 */
Mel Gorman11e33f62009-06-16 15:31:57 -07002595 if (!did_some_progress) {
2596 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -07002597 if (oom_killer_disabled)
2598 goto nopage;
David Rientjes29fd66d2012-03-28 14:42:41 -07002599 /* Coredumps can quickly deplete all memory reserves */
2600 if ((current->flags & PF_DUMPCORE) &&
2601 !(gfp_mask & __GFP_NOFAIL))
2602 goto nopage;
Mel Gorman11e33f62009-06-16 15:31:57 -07002603 page = __alloc_pages_may_oom(gfp_mask, order,
2604 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002605 nodemask, preferred_zone,
2606 migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002607 if (page)
2608 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609
David Rientjes03668b32010-08-09 17:18:54 -07002610 if (!(gfp_mask & __GFP_NOFAIL)) {
2611 /*
2612 * The oom killer is not called for high-order
2613 * allocations that may fail, so if no progress
2614 * is being made, there are no other options and
2615 * retrying is unlikely to help.
2616 */
2617 if (order > PAGE_ALLOC_COSTLY_ORDER)
2618 goto nopage;
2619 /*
2620 * The oom killer is not called for lowmem
2621 * allocations to prevent needlessly killing
2622 * innocent tasks.
2623 */
2624 if (high_zoneidx < ZONE_NORMAL)
2625 goto nopage;
2626 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 goto restart;
2629 }
2630 }
2631
Mel Gorman11e33f62009-06-16 15:31:57 -07002632 /* Check if we should retry the allocation */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07002633 pages_reclaimed += did_some_progress;
Mel Gormanf90ac392012-01-10 15:07:15 -08002634 if (should_alloc_retry(gfp_mask, order, did_some_progress,
2635 pages_reclaimed)) {
Mel Gorman11e33f62009-06-16 15:31:57 -07002636 /* Wait for some write requests to complete then retry */
Mel Gorman0e093d992010-10-26 14:21:45 -07002637 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638 goto rebalance;
Mel Gorman3e7d3442011-01-13 15:45:56 -08002639 } else {
2640 /*
2641 * High-order allocations do not necessarily loop after
2642 * direct reclaim and reclaim/compaction depends on compaction
2643 * being called after reclaim so call directly if necessary
2644 */
2645 page = __alloc_pages_direct_compact(gfp_mask, order,
2646 zonelist, high_zoneidx,
2647 nodemask,
2648 alloc_flags, preferred_zone,
Mel Gorman66199712012-01-12 17:19:41 -08002649 migratetype, sync_migration,
Mel Gormanc67fe372012-08-21 16:16:17 -07002650 &contended_compaction,
Mel Gorman66199712012-01-12 17:19:41 -08002651 &deferred_compaction,
2652 &did_some_progress);
Mel Gorman3e7d3442011-01-13 15:45:56 -08002653 if (page)
2654 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655 }
2656
2657nopage:
Dave Hansena238ab52011-05-24 17:12:16 -07002658 warn_alloc_failed(gfp_mask, order, NULL);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002659 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660got_pg:
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002661 if (kmemcheck_enabled)
2662 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002663
Mel Gorman072bb0a2012-07-31 16:43:58 -07002664 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665}
Mel Gorman11e33f62009-06-16 15:31:57 -07002666
2667/*
2668 * This is the 'heart' of the zoned buddy allocator.
2669 */
2670struct page *
2671__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2672 struct zonelist *zonelist, nodemask_t *nodemask)
2673{
2674 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Mel Gorman5117f452009-06-16 15:31:59 -07002675 struct zone *preferred_zone;
Mel Gormancc9a6c82012-03-21 16:34:11 -07002676 struct page *page = NULL;
Mel Gorman3dd28262009-06-16 15:32:00 -07002677 int migratetype = allocflags_to_migratetype(gfp_mask);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002678 unsigned int cpuset_mems_cookie;
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002679 int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET;
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002680 struct mem_cgroup *memcg = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002681
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10002682 gfp_mask &= gfp_allowed_mask;
2683
Mel Gorman11e33f62009-06-16 15:31:57 -07002684 lockdep_trace_alloc(gfp_mask);
2685
2686 might_sleep_if(gfp_mask & __GFP_WAIT);
2687
2688 if (should_fail_alloc_page(gfp_mask, order))
2689 return NULL;
2690
2691 /*
2692 * Check the zones suitable for the gfp_mask contain at least one
2693 * valid zone. It's possible to have an empty zonelist as a result
2694 * of GFP_THISNODE and a memoryless node
2695 */
2696 if (unlikely(!zonelist->_zonerefs->zone))
2697 return NULL;
2698
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002699 /*
2700 * Will only have any effect when __GFP_KMEMCG is set. This is
2701 * verified in the (always inline) callee
2702 */
2703 if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
2704 return NULL;
2705
Mel Gormancc9a6c82012-03-21 16:34:11 -07002706retry_cpuset:
2707 cpuset_mems_cookie = get_mems_allowed();
2708
Mel Gorman5117f452009-06-16 15:31:59 -07002709 /* The preferred zone is used for statistics later */
David Rientjesf33261d2011-01-25 15:07:20 -08002710 first_zones_zonelist(zonelist, high_zoneidx,
2711 nodemask ? : &cpuset_current_mems_allowed,
2712 &preferred_zone);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002713 if (!preferred_zone)
2714 goto out;
Mel Gorman5117f452009-06-16 15:31:59 -07002715
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002716#ifdef CONFIG_CMA
2717 if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2718 alloc_flags |= ALLOC_CMA;
2719#endif
Mel Gorman5117f452009-06-16 15:31:59 -07002720 /* First allocation attempt */
Mel Gorman11e33f62009-06-16 15:31:57 -07002721 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002722 zonelist, high_zoneidx, alloc_flags,
Mel Gorman3dd28262009-06-16 15:32:00 -07002723 preferred_zone, migratetype);
Ming Lei21caf2f2013-02-22 16:34:08 -08002724 if (unlikely(!page)) {
2725 /*
2726 * Runtime PM, block IO and its error handling path
2727 * can deadlock because I/O on the device might not
2728 * complete.
2729 */
2730 gfp_mask = memalloc_noio_flags(gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002731 page = __alloc_pages_slowpath(gfp_mask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002732 zonelist, high_zoneidx, nodemask,
Mel Gorman3dd28262009-06-16 15:32:00 -07002733 preferred_zone, migratetype);
Ming Lei21caf2f2013-02-22 16:34:08 -08002734 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002735
Mel Gorman4b4f2782009-09-21 17:02:41 -07002736 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002737
2738out:
2739 /*
2740 * When updating a task's mems_allowed, it is possible to race with
2741 * parallel threads in such a way that an allocation can fail while
2742 * the mask is being updated. If a page allocation is about to fail,
2743 * check if the cpuset changed during allocation and if so, retry.
2744 */
2745 if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page))
2746 goto retry_cpuset;
2747
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002748 memcg_kmem_commit_charge(page, memcg, order);
2749
Mel Gorman11e33f62009-06-16 15:31:57 -07002750 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751}
Mel Gormand2391712009-06-16 15:31:52 -07002752EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753
2754/*
2755 * Common helper functions.
2756 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08002757unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758{
Akinobu Mita945a1112009-09-21 17:01:47 -07002759 struct page *page;
2760
2761 /*
2762 * __get_free_pages() returns a 32-bit address, which cannot represent
2763 * a highmem page
2764 */
2765 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2766
Linus Torvalds1da177e2005-04-16 15:20:36 -07002767 page = alloc_pages(gfp_mask, order);
2768 if (!page)
2769 return 0;
2770 return (unsigned long) page_address(page);
2771}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002772EXPORT_SYMBOL(__get_free_pages);
2773
Harvey Harrison920c7a52008-02-04 22:29:26 -08002774unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002775{
Akinobu Mita945a1112009-09-21 17:01:47 -07002776 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778EXPORT_SYMBOL(get_zeroed_page);
2779
Harvey Harrison920c7a52008-02-04 22:29:26 -08002780void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781{
Nick Pigginb5810032005-10-29 18:16:12 -07002782 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783 if (order == 0)
Li Hongfc916682010-03-05 13:41:54 -08002784 free_hot_cold_page(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785 else
2786 __free_pages_ok(page, order);
2787 }
2788}
2789
2790EXPORT_SYMBOL(__free_pages);
2791
Harvey Harrison920c7a52008-02-04 22:29:26 -08002792void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793{
2794 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07002795 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796 __free_pages(virt_to_page((void *)addr), order);
2797 }
2798}
2799
2800EXPORT_SYMBOL(free_pages);
2801
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002802/*
2803 * __free_memcg_kmem_pages and free_memcg_kmem_pages will free
2804 * pages allocated with __GFP_KMEMCG.
2805 *
2806 * Those pages are accounted to a particular memcg, embedded in the
2807 * corresponding page_cgroup. To avoid adding a hit in the allocator to search
2808 * for that information only to find out that it is NULL for users who have no
2809 * interest in that whatsoever, we provide these functions.
2810 *
2811 * The caller knows better which flags it relies on.
2812 */
2813void __free_memcg_kmem_pages(struct page *page, unsigned int order)
2814{
2815 memcg_kmem_uncharge_pages(page, order);
2816 __free_pages(page, order);
2817}
2818
2819void free_memcg_kmem_pages(unsigned long addr, unsigned int order)
2820{
2821 if (addr != 0) {
2822 VM_BUG_ON(!virt_addr_valid((void *)addr));
2823 __free_memcg_kmem_pages(virt_to_page((void *)addr), order);
2824 }
2825}
2826
Andi Kleenee85c2e2011-05-11 15:13:34 -07002827static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
2828{
2829 if (addr) {
2830 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2831 unsigned long used = addr + PAGE_ALIGN(size);
2832
2833 split_page(virt_to_page((void *)addr), order);
2834 while (used < alloc_end) {
2835 free_page(used);
2836 used += PAGE_SIZE;
2837 }
2838 }
2839 return (void *)addr;
2840}
2841
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002842/**
2843 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2844 * @size: the number of bytes to allocate
2845 * @gfp_mask: GFP flags for the allocation
2846 *
2847 * This function is similar to alloc_pages(), except that it allocates the
2848 * minimum number of pages to satisfy the request. alloc_pages() can only
2849 * allocate memory in power-of-two pages.
2850 *
2851 * This function is also limited by MAX_ORDER.
2852 *
2853 * Memory allocated by this function must be released by free_pages_exact().
2854 */
2855void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2856{
2857 unsigned int order = get_order(size);
2858 unsigned long addr;
2859
2860 addr = __get_free_pages(gfp_mask, order);
Andi Kleenee85c2e2011-05-11 15:13:34 -07002861 return make_alloc_exact(addr, order, size);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002862}
2863EXPORT_SYMBOL(alloc_pages_exact);
2864
2865/**
Andi Kleenee85c2e2011-05-11 15:13:34 -07002866 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2867 * pages on a node.
Randy Dunlapb5e6ab52011-05-16 13:16:54 -07002868 * @nid: the preferred node ID where memory should be allocated
Andi Kleenee85c2e2011-05-11 15:13:34 -07002869 * @size: the number of bytes to allocate
2870 * @gfp_mask: GFP flags for the allocation
2871 *
2872 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2873 * back.
2874 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2875 * but is not exact.
2876 */
2877void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
2878{
2879 unsigned order = get_order(size);
2880 struct page *p = alloc_pages_node(nid, gfp_mask, order);
2881 if (!p)
2882 return NULL;
2883 return make_alloc_exact((unsigned long)page_address(p), order, size);
2884}
2885EXPORT_SYMBOL(alloc_pages_exact_nid);
2886
2887/**
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002888 * free_pages_exact - release memory allocated via alloc_pages_exact()
2889 * @virt: the value returned by alloc_pages_exact.
2890 * @size: size of allocation, same value as passed to alloc_pages_exact().
2891 *
2892 * Release the memory allocated by a previous call to alloc_pages_exact.
2893 */
2894void free_pages_exact(void *virt, size_t size)
2895{
2896 unsigned long addr = (unsigned long)virt;
2897 unsigned long end = addr + PAGE_ALIGN(size);
2898
2899 while (addr < end) {
2900 free_page(addr);
2901 addr += PAGE_SIZE;
2902 }
2903}
2904EXPORT_SYMBOL(free_pages_exact);
2905
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08002906/**
2907 * nr_free_zone_pages - count number of pages beyond high watermark
2908 * @offset: The zone index of the highest zone
2909 *
2910 * nr_free_zone_pages() counts the number of counts pages which are beyond the
2911 * high watermark within all zones at or below a given zone index. For each
2912 * zone, the number of pages is calculated as:
Jiang Liu834405c2013-07-03 15:03:04 -07002913 * managed_pages - high_pages
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08002914 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08002915static unsigned long nr_free_zone_pages(int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002916{
Mel Gormandd1a2392008-04-28 02:12:17 -07002917 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002918 struct zone *zone;
2919
Martin J. Blighe310fd42005-07-29 22:59:18 -07002920 /* Just pick one node, since fallback list is circular */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08002921 unsigned long sum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002922
Mel Gorman0e884602008-04-28 02:12:14 -07002923 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924
Mel Gorman54a6eb52008-04-28 02:12:16 -07002925 for_each_zone_zonelist(zone, z, zonelist, offset) {
Jiang Liub40da042013-02-22 16:33:52 -08002926 unsigned long size = zone->managed_pages;
Mel Gorman41858962009-06-16 15:32:12 -07002927 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07002928 if (size > high)
2929 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002930 }
2931
2932 return sum;
2933}
2934
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08002935/**
2936 * nr_free_buffer_pages - count number of pages beyond high watermark
2937 *
2938 * nr_free_buffer_pages() counts the number of pages which are beyond the high
2939 * watermark within ZONE_DMA and ZONE_NORMAL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002940 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08002941unsigned long nr_free_buffer_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002942{
Al Viroaf4ca452005-10-21 02:55:38 -04002943 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002944}
Meelap Shahc2f1a552007-07-17 04:04:39 -07002945EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002946
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08002947/**
2948 * nr_free_pagecache_pages - count number of pages beyond high watermark
2949 *
2950 * nr_free_pagecache_pages() counts the number of pages which are beyond the
2951 * high watermark within all zones.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002952 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08002953unsigned long nr_free_pagecache_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002954{
Mel Gorman2a1e2742007-07-17 04:03:12 -07002955 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002956}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002957
2958static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002959{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002960 if (IS_ENABLED(CONFIG_NUMA))
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08002961 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002962}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002963
Linus Torvalds1da177e2005-04-16 15:20:36 -07002964void si_meminfo(struct sysinfo *val)
2965{
2966 val->totalram = totalram_pages;
2967 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002968 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002969 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002970 val->totalhigh = totalhigh_pages;
2971 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002972 val->mem_unit = PAGE_SIZE;
2973}
2974
2975EXPORT_SYMBOL(si_meminfo);
2976
2977#ifdef CONFIG_NUMA
2978void si_meminfo_node(struct sysinfo *val, int nid)
2979{
Jiang Liucdd91a72013-07-03 15:03:27 -07002980 int zone_type; /* needs to be signed */
2981 unsigned long managed_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002982 pg_data_t *pgdat = NODE_DATA(nid);
2983
Jiang Liucdd91a72013-07-03 15:03:27 -07002984 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
2985 managed_pages += pgdat->node_zones[zone_type].managed_pages;
2986 val->totalram = managed_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002987 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002988#ifdef CONFIG_HIGHMEM
Jiang Liub40da042013-02-22 16:33:52 -08002989 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002990 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2991 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002992#else
2993 val->totalhigh = 0;
2994 val->freehigh = 0;
2995#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002996 val->mem_unit = PAGE_SIZE;
2997}
2998#endif
2999
David Rientjesddd588b2011-03-22 16:30:46 -07003000/*
David Rientjes7bf02ea2011-05-24 17:11:16 -07003001 * Determine whether the node should be displayed or not, depending on whether
3002 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
David Rientjesddd588b2011-03-22 16:30:46 -07003003 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07003004bool skip_free_areas_node(unsigned int flags, int nid)
David Rientjesddd588b2011-03-22 16:30:46 -07003005{
3006 bool ret = false;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003007 unsigned int cpuset_mems_cookie;
David Rientjesddd588b2011-03-22 16:30:46 -07003008
3009 if (!(flags & SHOW_MEM_FILTER_NODES))
3010 goto out;
3011
Mel Gormancc9a6c82012-03-21 16:34:11 -07003012 do {
3013 cpuset_mems_cookie = get_mems_allowed();
3014 ret = !node_isset(nid, cpuset_current_mems_allowed);
3015 } while (!put_mems_allowed(cpuset_mems_cookie));
David Rientjesddd588b2011-03-22 16:30:46 -07003016out:
3017 return ret;
3018}
3019
Linus Torvalds1da177e2005-04-16 15:20:36 -07003020#define K(x) ((x) << (PAGE_SHIFT-10))
3021
Rabin Vincent377e4f12012-12-11 16:00:24 -08003022static void show_migration_types(unsigned char type)
3023{
3024 static const char types[MIGRATE_TYPES] = {
3025 [MIGRATE_UNMOVABLE] = 'U',
3026 [MIGRATE_RECLAIMABLE] = 'E',
3027 [MIGRATE_MOVABLE] = 'M',
3028 [MIGRATE_RESERVE] = 'R',
3029#ifdef CONFIG_CMA
3030 [MIGRATE_CMA] = 'C',
3031#endif
Minchan Kim194159f2013-02-22 16:33:58 -08003032#ifdef CONFIG_MEMORY_ISOLATION
Rabin Vincent377e4f12012-12-11 16:00:24 -08003033 [MIGRATE_ISOLATE] = 'I',
Minchan Kim194159f2013-02-22 16:33:58 -08003034#endif
Rabin Vincent377e4f12012-12-11 16:00:24 -08003035 };
3036 char tmp[MIGRATE_TYPES + 1];
3037 char *p = tmp;
3038 int i;
3039
3040 for (i = 0; i < MIGRATE_TYPES; i++) {
3041 if (type & (1 << i))
3042 *p++ = types[i];
3043 }
3044
3045 *p = '\0';
3046 printk("(%s) ", tmp);
3047}
3048
Linus Torvalds1da177e2005-04-16 15:20:36 -07003049/*
3050 * Show free area list (used inside shift_scroll-lock stuff)
3051 * We also calculate the percentage fragmentation. We do this by counting the
3052 * memory on each free list with the exception of the first item on the list.
David Rientjesddd588b2011-03-22 16:30:46 -07003053 * Suppresses nodes that are not allowed by current's cpuset if
3054 * SHOW_MEM_FILTER_NODES is passed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003055 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07003056void show_free_areas(unsigned int filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057{
Jes Sorensenc7241912006-09-27 01:50:05 -07003058 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003059 struct zone *zone;
3060
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003061 for_each_populated_zone(zone) {
David Rientjes7bf02ea2011-05-24 17:11:16 -07003062 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003063 continue;
Jes Sorensenc7241912006-09-27 01:50:05 -07003064 show_node(zone);
3065 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003066
Dave Jones6b482c62005-11-10 15:45:56 -05003067 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068 struct per_cpu_pageset *pageset;
3069
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003070 pageset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003072 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
3073 cpu, pageset->pcp.high,
3074 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075 }
3076 }
3077
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003078 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3079 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003080 " unevictable:%lu"
Andrew Mortonb76146e2009-10-26 16:49:52 -07003081 " dirty:%lu writeback:%lu unstable:%lu\n"
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07003082 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003083 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3084 " free_cma:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003085 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003086 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003087 global_page_state(NR_ISOLATED_ANON),
3088 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003089 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003090 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003091 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07003092 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07003093 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07003094 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08003095 global_page_state(NR_FREE_PAGES),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07003096 global_page_state(NR_SLAB_RECLAIMABLE),
3097 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07003098 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003099 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08003100 global_page_state(NR_PAGETABLE),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003101 global_page_state(NR_BOUNCE),
3102 global_page_state(NR_FREE_CMA_PAGES));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003103
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003104 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003105 int i;
3106
David Rientjes7bf02ea2011-05-24 17:11:16 -07003107 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003108 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003109 show_node(zone);
3110 printk("%s"
3111 " free:%lukB"
3112 " min:%lukB"
3113 " low:%lukB"
3114 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003115 " active_anon:%lukB"
3116 " inactive_anon:%lukB"
3117 " active_file:%lukB"
3118 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003119 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003120 " isolated(anon):%lukB"
3121 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003122 " present:%lukB"
Jiang Liu9feedc92012-12-12 13:52:12 -08003123 " managed:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003124 " mlocked:%lukB"
3125 " dirty:%lukB"
3126 " writeback:%lukB"
3127 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003128 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003129 " slab_reclaimable:%lukB"
3130 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003131 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003132 " pagetables:%lukB"
3133 " unstable:%lukB"
3134 " bounce:%lukB"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003135 " free_cma:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003136 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003137 " pages_scanned:%lu"
3138 " all_unreclaimable? %s"
3139 "\n",
3140 zone->name,
Mel Gorman88f5acf2011-01-13 15:45:41 -08003141 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07003142 K(min_wmark_pages(zone)),
3143 K(low_wmark_pages(zone)),
3144 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003145 K(zone_page_state(zone, NR_ACTIVE_ANON)),
3146 K(zone_page_state(zone, NR_INACTIVE_ANON)),
3147 K(zone_page_state(zone, NR_ACTIVE_FILE)),
3148 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003149 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003150 K(zone_page_state(zone, NR_ISOLATED_ANON)),
3151 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003152 K(zone->present_pages),
Jiang Liu9feedc92012-12-12 13:52:12 -08003153 K(zone->managed_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003154 K(zone_page_state(zone, NR_MLOCK)),
3155 K(zone_page_state(zone, NR_FILE_DIRTY)),
3156 K(zone_page_state(zone, NR_WRITEBACK)),
3157 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003158 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003159 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3160 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003161 zone_page_state(zone, NR_KERNEL_STACK) *
3162 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003163 K(zone_page_state(zone, NR_PAGETABLE)),
3164 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3165 K(zone_page_state(zone, NR_BOUNCE)),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003166 K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003167 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003168 zone->pages_scanned,
Lisa Du6e543d52013-09-11 14:22:36 -07003169 (!zone_reclaimable(zone) ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07003170 );
3171 printk("lowmem_reserve[]:");
3172 for (i = 0; i < MAX_NR_ZONES; i++)
3173 printk(" %lu", zone->lowmem_reserve[i]);
3174 printk("\n");
3175 }
3176
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003177 for_each_populated_zone(zone) {
Pintu Kumarb8af2942013-09-11 14:20:34 -07003178 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003179 unsigned char types[MAX_ORDER];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003180
David Rientjes7bf02ea2011-05-24 17:11:16 -07003181 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003182 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003183 show_node(zone);
3184 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003185
3186 spin_lock_irqsave(&zone->lock, flags);
3187 for (order = 0; order < MAX_ORDER; order++) {
Rabin Vincent377e4f12012-12-11 16:00:24 -08003188 struct free_area *area = &zone->free_area[order];
3189 int type;
3190
3191 nr[order] = area->nr_free;
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003192 total += nr[order] << order;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003193
3194 types[order] = 0;
3195 for (type = 0; type < MIGRATE_TYPES; type++) {
3196 if (!list_empty(&area->free_list[type]))
3197 types[order] |= 1 << type;
3198 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003199 }
3200 spin_unlock_irqrestore(&zone->lock, flags);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003201 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003202 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003203 if (nr[order])
3204 show_migration_types(types[order]);
3205 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003206 printk("= %lukB\n", K(total));
3207 }
3208
David Rientjes949f7ec2013-04-29 15:07:48 -07003209 hugetlb_show_meminfo();
3210
Larry Woodmane6f36022008-02-04 22:29:30 -08003211 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
3212
Linus Torvalds1da177e2005-04-16 15:20:36 -07003213 show_swap_cache_info();
3214}
3215
Mel Gorman19770b32008-04-28 02:12:18 -07003216static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
3217{
3218 zoneref->zone = zone;
3219 zoneref->zone_idx = zone_idx(zone);
3220}
3221
Linus Torvalds1da177e2005-04-16 15:20:36 -07003222/*
3223 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08003224 *
3225 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003226 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003227static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003228 int nr_zones)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003229{
Christoph Lameter1a932052006-01-06 00:11:16 -08003230 struct zone *zone;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003231 enum zone_type zone_type = MAX_NR_ZONES;
Christoph Lameter02a68a52006-01-06 00:11:18 -08003232
3233 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003234 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08003235 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08003236 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07003237 zoneref_set_zone(zone,
3238 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08003239 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003240 }
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003241 } while (zone_type);
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003242
Christoph Lameter070f8032006-01-06 00:11:19 -08003243 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003244}
3245
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003246
3247/*
3248 * zonelist_order:
3249 * 0 = automatic detection of better ordering.
3250 * 1 = order by ([node] distance, -zonetype)
3251 * 2 = order by (-zonetype, [node] distance)
3252 *
3253 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3254 * the same zonelist. So only NUMA can configure this param.
3255 */
3256#define ZONELIST_ORDER_DEFAULT 0
3257#define ZONELIST_ORDER_NODE 1
3258#define ZONELIST_ORDER_ZONE 2
3259
3260/* zonelist order in the kernel.
3261 * set_zonelist_order() will set this to NODE or ZONE.
3262 */
3263static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
3264static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
3265
3266
Linus Torvalds1da177e2005-04-16 15:20:36 -07003267#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003268/* The value user specified ....changed by config */
3269static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3270/* string for sysctl */
3271#define NUMA_ZONELIST_ORDER_LEN 16
3272char numa_zonelist_order[16] = "default";
3273
3274/*
3275 * interface for configure zonelist ordering.
3276 * command line option "numa_zonelist_order"
3277 * = "[dD]efault - default, automatic configuration.
3278 * = "[nN]ode - order by node locality, then by zone within node
3279 * = "[zZ]one - order by zone, then by locality within zone
3280 */
3281
3282static int __parse_numa_zonelist_order(char *s)
3283{
3284 if (*s == 'd' || *s == 'D') {
3285 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3286 } else if (*s == 'n' || *s == 'N') {
3287 user_zonelist_order = ZONELIST_ORDER_NODE;
3288 } else if (*s == 'z' || *s == 'Z') {
3289 user_zonelist_order = ZONELIST_ORDER_ZONE;
3290 } else {
3291 printk(KERN_WARNING
3292 "Ignoring invalid numa_zonelist_order value: "
3293 "%s\n", s);
3294 return -EINVAL;
3295 }
3296 return 0;
3297}
3298
3299static __init int setup_numa_zonelist_order(char *s)
3300{
Volodymyr G. Lukiianykecb256f2011-01-13 15:46:26 -08003301 int ret;
3302
3303 if (!s)
3304 return 0;
3305
3306 ret = __parse_numa_zonelist_order(s);
3307 if (ret == 0)
3308 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
3309
3310 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003311}
3312early_param("numa_zonelist_order", setup_numa_zonelist_order);
3313
3314/*
3315 * sysctl handler for numa_zonelist_order
3316 */
3317int numa_zonelist_order_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003318 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003319 loff_t *ppos)
3320{
3321 char saved_string[NUMA_ZONELIST_ORDER_LEN];
3322 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01003323 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003324
Andi Kleen443c6f12009-12-23 21:00:47 +01003325 mutex_lock(&zl_order_mutex);
Chen Gangdacbde02013-07-03 15:02:35 -07003326 if (write) {
3327 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
3328 ret = -EINVAL;
3329 goto out;
3330 }
3331 strcpy(saved_string, (char *)table->data);
3332 }
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003333 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003334 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01003335 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003336 if (write) {
3337 int oldval = user_zonelist_order;
Chen Gangdacbde02013-07-03 15:02:35 -07003338
3339 ret = __parse_numa_zonelist_order((char *)table->data);
3340 if (ret) {
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003341 /*
3342 * bogus value. restore saved string
3343 */
Chen Gangdacbde02013-07-03 15:02:35 -07003344 strncpy((char *)table->data, saved_string,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003345 NUMA_ZONELIST_ORDER_LEN);
3346 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003347 } else if (oldval != user_zonelist_order) {
3348 mutex_lock(&zonelists_mutex);
Jiang Liu9adb62a2012-07-31 16:43:28 -07003349 build_all_zonelists(NULL, NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003350 mutex_unlock(&zonelists_mutex);
3351 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003352 }
Andi Kleen443c6f12009-12-23 21:00:47 +01003353out:
3354 mutex_unlock(&zl_order_mutex);
3355 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003356}
3357
3358
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003359#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003360static int node_load[MAX_NUMNODES];
3361
Linus Torvalds1da177e2005-04-16 15:20:36 -07003362/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07003363 * 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 -07003364 * @node: node whose fallback list we're appending
3365 * @used_node_mask: nodemask_t of already used nodes
3366 *
3367 * We use a number of factors to determine which is the next node that should
3368 * appear on a given node's fallback list. The node should not have appeared
3369 * already in @node's fallback list, and it should be the next closest node
3370 * according to the distance array (which contains arbitrary distance values
3371 * from each node to each node in the system), and should also prefer nodes
3372 * with no CPUs, since presumably they'll have very little allocation pressure
3373 * on them otherwise.
3374 * It returns -1 if no node is found.
3375 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003376static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003377{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01003378 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003379 int min_val = INT_MAX;
David Rientjes00ef2d22013-02-22 16:35:36 -08003380 int best_node = NUMA_NO_NODE;
Rusty Russella70f7302009-03-13 14:49:46 +10303381 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003382
Linus Torvalds4cf808e2006-02-17 20:38:21 +01003383 /* Use the local node if we haven't already */
3384 if (!node_isset(node, *used_node_mask)) {
3385 node_set(node, *used_node_mask);
3386 return node;
3387 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003388
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08003389 for_each_node_state(n, N_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003390
3391 /* Don't want a node to appear more than once */
3392 if (node_isset(n, *used_node_mask))
3393 continue;
3394
Linus Torvalds1da177e2005-04-16 15:20:36 -07003395 /* Use the distance array to find the distance */
3396 val = node_distance(node, n);
3397
Linus Torvalds4cf808e2006-02-17 20:38:21 +01003398 /* Penalize nodes under us ("prefer the next node") */
3399 val += (n < node);
3400
Linus Torvalds1da177e2005-04-16 15:20:36 -07003401 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10303402 tmp = cpumask_of_node(n);
3403 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003404 val += PENALTY_FOR_NODE_WITH_CPUS;
3405
3406 /* Slight preference for less loaded node */
3407 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
3408 val += node_load[n];
3409
3410 if (val < min_val) {
3411 min_val = val;
3412 best_node = n;
3413 }
3414 }
3415
3416 if (best_node >= 0)
3417 node_set(best_node, *used_node_mask);
3418
3419 return best_node;
3420}
3421
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003422
3423/*
3424 * Build zonelists ordered by node and zones within node.
3425 * This results in maximum locality--normal zone overflows into local
3426 * DMA zone, if any--but risks exhausting DMA zone.
3427 */
3428static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003429{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003430 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003431 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003432
Mel Gorman54a6eb52008-04-28 02:12:16 -07003433 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07003434 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07003435 ;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003436 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gormandd1a2392008-04-28 02:12:17 -07003437 zonelist->_zonerefs[j].zone = NULL;
3438 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003439}
3440
3441/*
Christoph Lameter523b9452007-10-16 01:25:37 -07003442 * Build gfp_thisnode zonelists
3443 */
3444static void build_thisnode_zonelists(pg_data_t *pgdat)
3445{
Christoph Lameter523b9452007-10-16 01:25:37 -07003446 int j;
3447 struct zonelist *zonelist;
3448
Mel Gorman54a6eb52008-04-28 02:12:16 -07003449 zonelist = &pgdat->node_zonelists[1];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003450 j = build_zonelists_node(pgdat, zonelist, 0);
Mel Gormandd1a2392008-04-28 02:12:17 -07003451 zonelist->_zonerefs[j].zone = NULL;
3452 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07003453}
3454
3455/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003456 * Build zonelists ordered by zone and nodes within zones.
3457 * This results in conserving DMA zone[s] until all Normal memory is
3458 * exhausted, but results in overflowing to remote node while memory
3459 * may still exist in local DMA zone.
3460 */
3461static int node_order[MAX_NUMNODES];
3462
3463static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
3464{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003465 int pos, j, node;
3466 int zone_type; /* needs to be signed */
3467 struct zone *z;
3468 struct zonelist *zonelist;
3469
Mel Gorman54a6eb52008-04-28 02:12:16 -07003470 zonelist = &pgdat->node_zonelists[0];
3471 pos = 0;
3472 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
3473 for (j = 0; j < nr_nodes; j++) {
3474 node = node_order[j];
3475 z = &NODE_DATA(node)->node_zones[zone_type];
3476 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07003477 zoneref_set_zone(z,
3478 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07003479 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003480 }
3481 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003482 }
Mel Gormandd1a2392008-04-28 02:12:17 -07003483 zonelist->_zonerefs[pos].zone = NULL;
3484 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003485}
3486
3487static int default_zonelist_order(void)
3488{
3489 int nid, zone_type;
Pintu Kumarb8af2942013-09-11 14:20:34 -07003490 unsigned long low_kmem_size, total_size;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003491 struct zone *z;
3492 int average_size;
3493 /*
Pintu Kumarb8af2942013-09-11 14:20:34 -07003494 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003495 * If they are really small and used heavily, the system can fall
3496 * into OOM very easily.
David Rientjese325c902010-05-24 14:32:13 -07003497 * This function detect ZONE_DMA/DMA32 size and configures zone order.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003498 */
3499 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
3500 low_kmem_size = 0;
3501 total_size = 0;
3502 for_each_online_node(nid) {
3503 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3504 z = &NODE_DATA(nid)->node_zones[zone_type];
3505 if (populated_zone(z)) {
3506 if (zone_type < ZONE_NORMAL)
Jiang Liu4f9f4772013-07-03 15:03:07 -07003507 low_kmem_size += z->managed_pages;
3508 total_size += z->managed_pages;
David Rientjese325c902010-05-24 14:32:13 -07003509 } else if (zone_type == ZONE_NORMAL) {
3510 /*
3511 * If any node has only lowmem, then node order
3512 * is preferred to allow kernel allocations
3513 * locally; otherwise, they can easily infringe
3514 * on other nodes when there is an abundance of
3515 * lowmem available to allocate from.
3516 */
3517 return ZONELIST_ORDER_NODE;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003518 }
3519 }
3520 }
3521 if (!low_kmem_size || /* there are no DMA area. */
3522 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
3523 return ZONELIST_ORDER_NODE;
3524 /*
3525 * look into each node's config.
Pintu Kumarb8af2942013-09-11 14:20:34 -07003526 * If there is a node whose DMA/DMA32 memory is very big area on
3527 * local memory, NODE_ORDER may be suitable.
3528 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003529 average_size = total_size /
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08003530 (nodes_weight(node_states[N_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003531 for_each_online_node(nid) {
3532 low_kmem_size = 0;
3533 total_size = 0;
3534 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3535 z = &NODE_DATA(nid)->node_zones[zone_type];
3536 if (populated_zone(z)) {
3537 if (zone_type < ZONE_NORMAL)
3538 low_kmem_size += z->present_pages;
3539 total_size += z->present_pages;
3540 }
3541 }
3542 if (low_kmem_size &&
3543 total_size > average_size && /* ignore small node */
3544 low_kmem_size > total_size * 70/100)
3545 return ZONELIST_ORDER_NODE;
3546 }
3547 return ZONELIST_ORDER_ZONE;
3548}
3549
3550static void set_zonelist_order(void)
3551{
3552 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
3553 current_zonelist_order = default_zonelist_order();
3554 else
3555 current_zonelist_order = user_zonelist_order;
3556}
3557
3558static void build_zonelists(pg_data_t *pgdat)
3559{
3560 int j, node, load;
3561 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003562 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003563 int local_node, prev_node;
3564 struct zonelist *zonelist;
3565 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003566
3567 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07003568 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003569 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07003570 zonelist->_zonerefs[0].zone = NULL;
3571 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003572 }
3573
3574 /* NUMA-aware ordering of nodes */
3575 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003576 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003577 prev_node = local_node;
3578 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003579
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003580 memset(node_order, 0, sizeof(node_order));
3581 j = 0;
3582
Linus Torvalds1da177e2005-04-16 15:20:36 -07003583 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
3584 /*
3585 * We don't want to pressure a particular node.
3586 * So adding penalty to the first node in same
3587 * distance group to make it round-robin.
3588 */
David Rientjes957f8222012-10-08 16:33:24 -07003589 if (node_distance(local_node, node) !=
3590 node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003591 node_load[node] = load;
3592
Linus Torvalds1da177e2005-04-16 15:20:36 -07003593 prev_node = node;
3594 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003595 if (order == ZONELIST_ORDER_NODE)
3596 build_zonelists_in_node_order(pgdat, node);
3597 else
3598 node_order[j++] = node; /* remember order */
3599 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003600
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003601 if (order == ZONELIST_ORDER_ZONE) {
3602 /* calculate node order -- i.e., DMA last! */
3603 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003604 }
Christoph Lameter523b9452007-10-16 01:25:37 -07003605
3606 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003607}
3608
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003609/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003610static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003611{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003612 struct zonelist *zonelist;
3613 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07003614 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003615
Mel Gorman54a6eb52008-04-28 02:12:16 -07003616 zonelist = &pgdat->node_zonelists[0];
3617 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3618 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07003619 for (z = zonelist->_zonerefs; z->zone; z++)
3620 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003621}
3622
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003623#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3624/*
3625 * Return node id of node used for "local" allocations.
3626 * I.e., first node id of first zone in arg node's generic zonelist.
3627 * Used for initializing percpu 'numa_mem', which is used primarily
3628 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3629 */
3630int local_memory_node(int node)
3631{
3632 struct zone *zone;
3633
3634 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3635 gfp_zone(GFP_KERNEL),
3636 NULL,
3637 &zone);
3638 return zone->node;
3639}
3640#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003641
Linus Torvalds1da177e2005-04-16 15:20:36 -07003642#else /* CONFIG_NUMA */
3643
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003644static void set_zonelist_order(void)
3645{
3646 current_zonelist_order = ZONELIST_ORDER_ZONE;
3647}
3648
3649static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003650{
Christoph Lameter19655d32006-09-25 23:31:19 -07003651 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003652 enum zone_type j;
3653 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003654
3655 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003656
Mel Gorman54a6eb52008-04-28 02:12:16 -07003657 zonelist = &pgdat->node_zonelists[0];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003658 j = build_zonelists_node(pgdat, zonelist, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003659
Mel Gorman54a6eb52008-04-28 02:12:16 -07003660 /*
3661 * Now we build the zonelist so that it contains the zones
3662 * of all the other nodes.
3663 * We don't want to pressure a particular node, so when
3664 * building the zones for node N, we make sure that the
3665 * zones coming right after the local ones are those from
3666 * node N+1 (modulo N)
3667 */
3668 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3669 if (!node_online(node))
3670 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003671 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003672 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07003673 for (node = 0; node < local_node; node++) {
3674 if (!node_online(node))
3675 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003676 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gorman54a6eb52008-04-28 02:12:16 -07003677 }
3678
Mel Gormandd1a2392008-04-28 02:12:17 -07003679 zonelist->_zonerefs[j].zone = NULL;
3680 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003681}
3682
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003683/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003684static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003685{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003686 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003687}
3688
Linus Torvalds1da177e2005-04-16 15:20:36 -07003689#endif /* CONFIG_NUMA */
3690
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003691/*
3692 * Boot pageset table. One per cpu which is going to be used for all
3693 * zones and all nodes. The parameters will be set in such a way
3694 * that an item put on a list will immediately be handed over to
3695 * the buddy list. This is safe since pageset manipulation is done
3696 * with interrupts disabled.
3697 *
3698 * The boot_pagesets must be kept even after bootup is complete for
3699 * unused processors and/or zones. They do play a role for bootstrapping
3700 * hotplugged processors.
3701 *
3702 * zoneinfo_show() and maybe other functions do
3703 * not check if the processor is online before following the pageset pointer.
3704 * Other parts of the kernel may not check if the zone is available.
3705 */
3706static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3707static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07003708static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003709
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003710/*
3711 * Global mutex to protect against size modification of zonelists
3712 * as well as to serialize pageset setup for the new populated zone.
3713 */
3714DEFINE_MUTEX(zonelists_mutex);
3715
Rusty Russell9b1a4d32008-07-28 12:16:30 -05003716/* return values int ....just for stop_machine() */
Jiang Liu4ed7e022012-07-31 16:43:35 -07003717static int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003718{
Yasunori Goto68113782006-06-23 02:03:11 -07003719 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003720 int cpu;
Jiang Liu9adb62a2012-07-31 16:43:28 -07003721 pg_data_t *self = data;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003722
Bo Liu7f9cfb32009-08-18 14:11:19 -07003723#ifdef CONFIG_NUMA
3724 memset(node_load, 0, sizeof(node_load));
3725#endif
Jiang Liu9adb62a2012-07-31 16:43:28 -07003726
3727 if (self && !node_online(self->node_id)) {
3728 build_zonelists(self);
3729 build_zonelist_cache(self);
3730 }
3731
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003732 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07003733 pg_data_t *pgdat = NODE_DATA(nid);
3734
3735 build_zonelists(pgdat);
3736 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003737 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003738
3739 /*
3740 * Initialize the boot_pagesets that are going to be used
3741 * for bootstrapping processors. The real pagesets for
3742 * each zone will be allocated later when the per cpu
3743 * allocator is available.
3744 *
3745 * boot_pagesets are used also for bootstrapping offline
3746 * cpus if the system is already booted because the pagesets
3747 * are needed to initialize allocators on a specific cpu too.
3748 * F.e. the percpu allocator needs the page allocator which
3749 * needs the percpu allocator in order to allocate its pagesets
3750 * (a chicken-egg dilemma).
3751 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003752 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003753 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3754
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003755#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3756 /*
3757 * We now know the "local memory node" for each node--
3758 * i.e., the node of the first zone in the generic zonelist.
3759 * Set up numa_mem percpu variable for on-line cpus. During
3760 * boot, only the boot cpu should be on-line; we'll init the
3761 * secondary cpus' numa_mem as they come on-line. During
3762 * node/memory hotplug, we'll fixup all on-line cpus.
3763 */
3764 if (cpu_online(cpu))
3765 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3766#endif
3767 }
3768
Yasunori Goto68113782006-06-23 02:03:11 -07003769 return 0;
3770}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003771
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003772/*
3773 * Called with zonelists_mutex held always
3774 * unless system_state == SYSTEM_BOOTING.
3775 */
Jiang Liu9adb62a2012-07-31 16:43:28 -07003776void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
Yasunori Goto68113782006-06-23 02:03:11 -07003777{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003778 set_zonelist_order();
3779
Yasunori Goto68113782006-06-23 02:03:11 -07003780 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07003781 __build_all_zonelists(NULL);
Mel Gorman68ad8df2008-07-23 21:26:52 -07003782 mminit_verify_zonelist();
Yasunori Goto68113782006-06-23 02:03:11 -07003783 cpuset_init_current_mems_allowed();
3784 } else {
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003785#ifdef CONFIG_MEMORY_HOTPLUG
Jiang Liu9adb62a2012-07-31 16:43:28 -07003786 if (zone)
3787 setup_zone_pageset(zone);
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003788#endif
Cody P Schaferdd1895e2013-07-03 15:01:36 -07003789 /* we have to stop all cpus to guarantee there is no user
3790 of zonelist */
Jiang Liu9adb62a2012-07-31 16:43:28 -07003791 stop_machine(__build_all_zonelists, pgdat, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07003792 /* cpuset refresh routine should be here */
3793 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07003794 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003795 /*
3796 * Disable grouping by mobility if the number of pages in the
3797 * system is too low to allow the mechanism to work. It would be
3798 * more accurate, but expensive to check per-zone. This check is
3799 * made on memory-hotadd so a system can start with mobility
3800 * disabled and enable it later
3801 */
Mel Gormand9c23402007-10-16 01:26:01 -07003802 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003803 page_group_by_mobility_disabled = 1;
3804 else
3805 page_group_by_mobility_disabled = 0;
3806
3807 printk("Built %i zonelists in %s order, mobility grouping %s. "
3808 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003809 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003810 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003811 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003812 vm_total_pages);
3813#ifdef CONFIG_NUMA
3814 printk("Policy zone: %s\n", zone_names[policy_zone]);
3815#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003816}
3817
3818/*
3819 * Helper functions to size the waitqueue hash table.
3820 * Essentially these want to choose hash table sizes sufficiently
3821 * large so that collisions trying to wait on pages are rare.
3822 * But in fact, the number of active page waitqueues on typical
3823 * systems is ridiculously low, less than 200. So this is even
3824 * conservative, even though it seems large.
3825 *
3826 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3827 * waitqueues, i.e. the size of the waitq table given the number of pages.
3828 */
3829#define PAGES_PER_WAITQUEUE 256
3830
Yasunori Gotocca448f2006-06-23 02:03:10 -07003831#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07003832static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003833{
3834 unsigned long size = 1;
3835
3836 pages /= PAGES_PER_WAITQUEUE;
3837
3838 while (size < pages)
3839 size <<= 1;
3840
3841 /*
3842 * Once we have dozens or even hundreds of threads sleeping
3843 * on IO we've got bigger problems than wait queue collision.
3844 * Limit the size of the wait table to a reasonable size.
3845 */
3846 size = min(size, 4096UL);
3847
3848 return max(size, 4UL);
3849}
Yasunori Gotocca448f2006-06-23 02:03:10 -07003850#else
3851/*
3852 * A zone's size might be changed by hot-add, so it is not possible to determine
3853 * a suitable size for its wait_table. So we use the maximum size now.
3854 *
3855 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3856 *
3857 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3858 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3859 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3860 *
3861 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3862 * or more by the traditional way. (See above). It equals:
3863 *
3864 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3865 * ia64(16K page size) : = ( 8G + 4M)byte.
3866 * powerpc (64K page size) : = (32G +16M)byte.
3867 */
3868static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3869{
3870 return 4096UL;
3871}
3872#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003873
3874/*
3875 * This is an integer logarithm so that shifts can be used later
3876 * to extract the more random high bits from the multiplicative
3877 * hash function before the remainder is taken.
3878 */
3879static inline unsigned long wait_table_bits(unsigned long size)
3880{
3881 return ffz(~size);
3882}
3883
3884#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3885
Mel Gorman56fd56b2007-10-16 01:25:58 -07003886/*
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07003887 * Check if a pageblock contains reserved pages
3888 */
3889static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
3890{
3891 unsigned long pfn;
3892
3893 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
3894 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
3895 return 1;
3896 }
3897 return 0;
3898}
3899
3900/*
Mel Gormand9c23402007-10-16 01:26:01 -07003901 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07003902 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3903 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07003904 * higher will lead to a bigger reserve which will get freed as contiguous
3905 * blocks as reclaim kicks in
3906 */
3907static void setup_zone_migrate_reserve(struct zone *zone)
3908{
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07003909 unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003910 struct page *page;
Mel Gorman78986a62009-09-21 17:03:02 -07003911 unsigned long block_migratetype;
3912 int reserve;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003913
Michal Hockod02156382011-12-08 14:34:27 -08003914 /*
3915 * Get the start pfn, end pfn and the number of blocks to reserve
3916 * We have to be careful to be aligned to pageblock_nr_pages to
3917 * make sure that we always check pfn_valid for the first page in
3918 * the block.
3919 */
Mel Gorman56fd56b2007-10-16 01:25:58 -07003920 start_pfn = zone->zone_start_pfn;
Cody P Schafer108bcc92013-02-22 16:35:23 -08003921 end_pfn = zone_end_pfn(zone);
Michal Hockod02156382011-12-08 14:34:27 -08003922 start_pfn = roundup(start_pfn, pageblock_nr_pages);
Mel Gorman41858962009-06-16 15:32:12 -07003923 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07003924 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003925
Mel Gorman78986a62009-09-21 17:03:02 -07003926 /*
3927 * Reserve blocks are generally in place to help high-order atomic
3928 * allocations that are short-lived. A min_free_kbytes value that
3929 * would result in more than 2 reserve blocks for atomic allocations
3930 * is assumed to be in place to help anti-fragmentation for the
3931 * future allocation of hugepages at runtime.
3932 */
3933 reserve = min(2, reserve);
3934
Mel Gormand9c23402007-10-16 01:26:01 -07003935 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07003936 if (!pfn_valid(pfn))
3937 continue;
3938 page = pfn_to_page(pfn);
3939
Adam Litke344c7902008-09-02 14:35:38 -07003940 /* Watch out for overlapping nodes */
3941 if (page_to_nid(page) != zone_to_nid(zone))
3942 continue;
3943
Mel Gorman56fd56b2007-10-16 01:25:58 -07003944 block_migratetype = get_pageblock_migratetype(page);
3945
Mel Gorman938929f2012-01-10 15:07:14 -08003946 /* Only test what is necessary when the reserves are not met */
3947 if (reserve > 0) {
3948 /*
3949 * Blocks with reserved pages will never free, skip
3950 * them.
3951 */
3952 block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
3953 if (pageblock_is_reserved(pfn, block_end_pfn))
3954 continue;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003955
Mel Gorman938929f2012-01-10 15:07:14 -08003956 /* If this block is reserved, account for it */
3957 if (block_migratetype == MIGRATE_RESERVE) {
3958 reserve--;
3959 continue;
3960 }
3961
3962 /* Suitable for reserving if this block is movable */
3963 if (block_migratetype == MIGRATE_MOVABLE) {
3964 set_pageblock_migratetype(page,
3965 MIGRATE_RESERVE);
3966 move_freepages_block(zone, page,
3967 MIGRATE_RESERVE);
3968 reserve--;
3969 continue;
3970 }
Mel Gorman56fd56b2007-10-16 01:25:58 -07003971 }
3972
3973 /*
3974 * If the reserve is met and this is a previous reserved block,
3975 * take it back
3976 */
3977 if (block_migratetype == MIGRATE_RESERVE) {
3978 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
3979 move_freepages_block(zone, page, MIGRATE_MOVABLE);
3980 }
3981 }
3982}
Mel Gormanac0e5b72007-10-16 01:25:58 -07003983
Linus Torvalds1da177e2005-04-16 15:20:36 -07003984/*
3985 * Initially all pages are reserved - free ones are freed
3986 * up by free_all_bootmem() once the early boot process is
3987 * done. Non-atomic initialization, single-pass.
3988 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003989void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08003990 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003991{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003992 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07003993 unsigned long end_pfn = start_pfn + size;
3994 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003995 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003996
Hugh Dickins22b31ee2009-01-06 14:40:09 -08003997 if (highest_memmap_pfn < end_pfn - 1)
3998 highest_memmap_pfn = end_pfn - 1;
3999
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004000 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08004001 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08004002 /*
4003 * There can be holes in boot-time mem_map[]s
4004 * handed to this function. They do not
4005 * exist on hotplugged memory.
4006 */
4007 if (context == MEMMAP_EARLY) {
4008 if (!early_pfn_valid(pfn))
4009 continue;
4010 if (!early_pfn_in_nid(pfn, nid))
4011 continue;
4012 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004013 page = pfn_to_page(pfn);
4014 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07004015 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08004016 init_page_count(page);
Mel Gorman22b751c2013-02-22 16:34:59 -08004017 page_mapcount_reset(page);
4018 page_nid_reset_last(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004019 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004020 /*
4021 * Mark the block movable so that blocks are reserved for
4022 * movable at startup. This will force kernel allocations
4023 * to reserve their blocks rather than leaking throughout
4024 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07004025 * kernel allocations are made. Later some blocks near
4026 * the start are marked MIGRATE_RESERVE by
4027 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004028 *
4029 * bitmap is created for zone's valid pfn range. but memmap
4030 * can be created for invalid pages (for alignment)
4031 * check here not to call set_pageblock_migratetype() against
4032 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004033 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004034 if ((z->zone_start_pfn <= pfn)
Cody P Schafer108bcc92013-02-22 16:35:23 -08004035 && (pfn < zone_end_pfn(z))
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004036 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07004037 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004038
Linus Torvalds1da177e2005-04-16 15:20:36 -07004039 INIT_LIST_HEAD(&page->lru);
4040#ifdef WANT_PAGE_VIRTUAL
4041 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
4042 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07004043 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004044#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004045 }
4046}
4047
Andi Kleen1e548de2008-02-04 22:29:26 -08004048static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004049{
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004050 int order, t;
4051 for_each_migratetype_order(order, t) {
4052 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004053 zone->free_area[order].nr_free = 0;
4054 }
4055}
4056
4057#ifndef __HAVE_ARCH_MEMMAP_INIT
4058#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08004059 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004060#endif
4061
Jiang Liu4ed7e022012-07-31 16:43:35 -07004062static int __meminit zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004063{
David Howells3a6be872009-05-06 16:03:03 -07004064#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004065 int batch;
4066
4067 /*
4068 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07004069 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004070 *
4071 * OK, so we don't know how big the cache is. So guess.
4072 */
Jiang Liub40da042013-02-22 16:33:52 -08004073 batch = zone->managed_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07004074 if (batch * PAGE_SIZE > 512 * 1024)
4075 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004076 batch /= 4; /* We effectively *= 4 below */
4077 if (batch < 1)
4078 batch = 1;
4079
4080 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004081 * Clamp the batch to a 2^n - 1 value. Having a power
4082 * of 2 value was found to be more likely to have
4083 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004084 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004085 * For example if 2 tasks are alternately allocating
4086 * batches of pages, one task can end up with a lot
4087 * of pages of one half of the possible page colors
4088 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004089 */
David Howells91552032009-05-06 16:03:02 -07004090 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07004091
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004092 return batch;
David Howells3a6be872009-05-06 16:03:03 -07004093
4094#else
4095 /* The deferral and batching of frees should be suppressed under NOMMU
4096 * conditions.
4097 *
4098 * The problem is that NOMMU needs to be able to allocate large chunks
4099 * of contiguous memory as there's no hardware page translation to
4100 * assemble apparent contiguous memory from discontiguous pages.
4101 *
4102 * Queueing large contiguous runs of pages for batching, however,
4103 * causes the pages to actually be freed in smaller chunks. As there
4104 * can be a significant delay between the individual batches being
4105 * recycled, this leads to the once large chunks of space being
4106 * fragmented and becoming unavailable for high-order allocations.
4107 */
4108 return 0;
4109#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004110}
4111
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004112/*
4113 * pcp->high and pcp->batch values are related and dependent on one another:
4114 * ->batch must never be higher then ->high.
4115 * The following function updates them in a safe manner without read side
4116 * locking.
4117 *
4118 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4119 * those fields changing asynchronously (acording the the above rule).
4120 *
4121 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4122 * outside of boot time (or some other assurance that no concurrent updaters
4123 * exist).
4124 */
4125static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
4126 unsigned long batch)
4127{
4128 /* start with a fail safe value for batch */
4129 pcp->batch = 1;
4130 smp_wmb();
4131
4132 /* Update high, then batch, in order */
4133 pcp->high = high;
4134 smp_wmb();
4135
4136 pcp->batch = batch;
4137}
4138
Cody P Schafer36640332013-07-03 15:01:40 -07004139/* a companion to pageset_set_high() */
Cody P Schafer4008bab2013-07-03 15:01:28 -07004140static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
4141{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004142 pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
Cody P Schafer4008bab2013-07-03 15:01:28 -07004143}
4144
Cody P Schafer88c90db2013-07-03 15:01:35 -07004145static void pageset_init(struct per_cpu_pageset *p)
Christoph Lameter2caaad42005-06-21 17:15:00 -07004146{
4147 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004148 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004149
Magnus Damm1c6fe942005-10-26 01:58:59 -07004150 memset(p, 0, sizeof(*p));
4151
Christoph Lameter3dfa5722008-02-04 22:29:19 -08004152 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004153 pcp->count = 0;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004154 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
4155 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07004156}
4157
Cody P Schafer88c90db2013-07-03 15:01:35 -07004158static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
4159{
4160 pageset_init(p);
4161 pageset_set_batch(p, batch);
4162}
4163
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004164/*
Cody P Schafer36640332013-07-03 15:01:40 -07004165 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004166 * to the value high for the pageset p.
4167 */
Cody P Schafer36640332013-07-03 15:01:40 -07004168static void pageset_set_high(struct per_cpu_pageset *p,
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004169 unsigned long high)
4170{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004171 unsigned long batch = max(1UL, high / 4);
4172 if ((high / 4) > (PAGE_SHIFT * 8))
4173 batch = PAGE_SHIFT * 8;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004174
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004175 pageset_update(&p->pcp, high, batch);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004176}
4177
Cody P Schafer169f6c12013-07-03 15:01:41 -07004178static void __meminit pageset_set_high_and_batch(struct zone *zone,
4179 struct per_cpu_pageset *pcp)
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004180{
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004181 if (percpu_pagelist_fraction)
Cody P Schafer36640332013-07-03 15:01:40 -07004182 pageset_set_high(pcp,
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004183 (zone->managed_pages /
4184 percpu_pagelist_fraction));
4185 else
4186 pageset_set_batch(pcp, zone_batchsize(zone));
4187}
4188
Cody P Schafer169f6c12013-07-03 15:01:41 -07004189static void __meminit zone_pageset_init(struct zone *zone, int cpu)
4190{
4191 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4192
4193 pageset_init(pcp);
4194 pageset_set_high_and_batch(zone, pcp);
4195}
4196
Jiang Liu4ed7e022012-07-31 16:43:35 -07004197static void __meminit setup_zone_pageset(struct zone *zone)
Wu Fengguang319774e2010-05-24 14:32:49 -07004198{
4199 int cpu;
Wu Fengguang319774e2010-05-24 14:32:49 -07004200 zone->pageset = alloc_percpu(struct per_cpu_pageset);
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004201 for_each_possible_cpu(cpu)
4202 zone_pageset_init(zone, cpu);
Wu Fengguang319774e2010-05-24 14:32:49 -07004203}
4204
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004205/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004206 * Allocate per cpu pagesets and initialize them.
4207 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07004208 */
Al Viro78d99552005-12-15 09:18:25 +00004209void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004210{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004211 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004212
Wu Fengguang319774e2010-05-24 14:32:49 -07004213 for_each_populated_zone(zone)
4214 setup_zone_pageset(zone);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004215}
4216
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004217static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07004218int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07004219{
4220 int i;
4221 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004222 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07004223
4224 /*
4225 * The per-page waitqueue mechanism uses hashed waitqueues
4226 * per zone.
4227 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07004228 zone->wait_table_hash_nr_entries =
4229 wait_table_hash_nr_entries(zone_size_pages);
4230 zone->wait_table_bits =
4231 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004232 alloc_size = zone->wait_table_hash_nr_entries
4233 * sizeof(wait_queue_head_t);
4234
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07004235 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07004236 zone->wait_table = (wait_queue_head_t *)
Yinghai Lu8f389a992011-05-11 15:13:32 -07004237 alloc_bootmem_node_nopanic(pgdat, alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004238 } else {
4239 /*
4240 * This case means that a zone whose size was 0 gets new memory
4241 * via memory hot-add.
4242 * But it may be the case that a new node was hot-added. In
4243 * this case vmalloc() will not be able to use this new node's
4244 * memory - this wait_table must be initialized to use this new
4245 * node itself as well.
4246 * To use this new node's memory, further consideration will be
4247 * necessary.
4248 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07004249 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004250 }
4251 if (!zone->wait_table)
4252 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07004253
Pintu Kumarb8af2942013-09-11 14:20:34 -07004254 for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07004255 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004256
4257 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004258}
4259
Matt Tolentinoc09b4242006-01-17 07:03:44 +01004260static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07004261{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004262 /*
4263 * per cpu subsystem is not up at this point. The following code
4264 * relies on the ability of the linker to provide the
4265 * offset of a (static) per cpu variable into the per cpu area.
4266 */
4267 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07004268
Anton Blanchardf5335c02006-03-25 03:06:49 -08004269 if (zone->present_pages)
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004270 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
4271 zone->name, zone->present_pages,
4272 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07004273}
4274
Jiang Liu4ed7e022012-07-31 16:43:35 -07004275int __meminit init_currently_empty_zone(struct zone *zone,
Yasunori Goto718127c2006-06-23 02:03:10 -07004276 unsigned long zone_start_pfn,
Dave Hansena2f3aa022007-01-10 23:15:30 -08004277 unsigned long size,
4278 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07004279{
4280 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004281 int ret;
4282 ret = zone_wait_table_init(zone, size);
4283 if (ret)
4284 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07004285 pgdat->nr_zones = zone_idx(zone) + 1;
4286
Dave Hansened8ece22005-10-29 18:16:50 -07004287 zone->zone_start_pfn = zone_start_pfn;
4288
Mel Gorman708614e2008-07-23 21:26:51 -07004289 mminit_dprintk(MMINIT_TRACE, "memmap_init",
4290 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4291 pgdat->node_id,
4292 (unsigned long)zone_idx(zone),
4293 zone_start_pfn, (zone_start_pfn + size));
4294
Andi Kleen1e548de2008-02-04 22:29:26 -08004295 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07004296
4297 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004298}
4299
Tejun Heo0ee332c2011-12-08 10:22:09 -08004300#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07004301#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4302/*
4303 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4304 * Architectures may implement their own version but if add_active_range()
4305 * was used and there are no special requirements, this is a convenient
4306 * alternative
4307 */
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004308int __meminit __early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004309{
Tejun Heoc13291a2011-07-12 10:46:30 +02004310 unsigned long start_pfn, end_pfn;
Yinghai Lue76b63f2013-09-11 14:22:17 -07004311 int nid;
Russ Anderson7c243c72013-04-29 15:07:59 -07004312 /*
4313 * NOTE: The following SMP-unsafe globals are only used early in boot
4314 * when the kernel is running single-threaded.
4315 */
4316 static unsigned long __meminitdata last_start_pfn, last_end_pfn;
4317 static int __meminitdata last_nid;
4318
4319 if (last_start_pfn <= pfn && pfn < last_end_pfn)
4320 return last_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004321
Yinghai Lue76b63f2013-09-11 14:22:17 -07004322 nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
4323 if (nid != -1) {
4324 last_start_pfn = start_pfn;
4325 last_end_pfn = end_pfn;
4326 last_nid = nid;
4327 }
4328
4329 return nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004330}
4331#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4332
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004333int __meminit early_pfn_to_nid(unsigned long pfn)
4334{
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08004335 int nid;
4336
4337 nid = __early_pfn_to_nid(pfn);
4338 if (nid >= 0)
4339 return nid;
4340 /* just returns 0 */
4341 return 0;
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004342}
4343
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08004344#ifdef CONFIG_NODES_SPAN_OTHER_NODES
4345bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
4346{
4347 int nid;
4348
4349 nid = __early_pfn_to_nid(pfn);
4350 if (nid >= 0 && nid != node)
4351 return false;
4352 return true;
4353}
4354#endif
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004355
Mel Gormanc7132162006-09-27 01:49:43 -07004356/**
4357 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004358 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4359 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07004360 *
4361 * If an architecture guarantees that all ranges registered with
4362 * add_active_ranges() contain no holes and may be freed, this
4363 * this function may be used instead of calling free_bootmem() manually.
4364 */
Tejun Heoc13291a2011-07-12 10:46:30 +02004365void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004366{
Tejun Heoc13291a2011-07-12 10:46:30 +02004367 unsigned long start_pfn, end_pfn;
4368 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004369
Tejun Heoc13291a2011-07-12 10:46:30 +02004370 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
4371 start_pfn = min(start_pfn, max_low_pfn);
4372 end_pfn = min(end_pfn, max_low_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004373
Tejun Heoc13291a2011-07-12 10:46:30 +02004374 if (start_pfn < end_pfn)
4375 free_bootmem_node(NODE_DATA(this_nid),
4376 PFN_PHYS(start_pfn),
4377 (end_pfn - start_pfn) << PAGE_SHIFT);
Mel Gormanc7132162006-09-27 01:49:43 -07004378 }
4379}
4380
4381/**
4382 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004383 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07004384 *
4385 * If an architecture guarantees that all ranges registered with
4386 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004387 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07004388 */
4389void __init sparse_memory_present_with_active_regions(int nid)
4390{
Tejun Heoc13291a2011-07-12 10:46:30 +02004391 unsigned long start_pfn, end_pfn;
4392 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004393
Tejun Heoc13291a2011-07-12 10:46:30 +02004394 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
4395 memory_present(this_nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004396}
4397
4398/**
4399 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004400 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4401 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4402 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07004403 *
4404 * It returns the start and end page frame of a node based on information
4405 * provided by an arch calling add_active_range(). If called for a node
4406 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004407 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07004408 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004409void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004410 unsigned long *start_pfn, unsigned long *end_pfn)
4411{
Tejun Heoc13291a2011-07-12 10:46:30 +02004412 unsigned long this_start_pfn, this_end_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004413 int i;
Tejun Heoc13291a2011-07-12 10:46:30 +02004414
Mel Gormanc7132162006-09-27 01:49:43 -07004415 *start_pfn = -1UL;
4416 *end_pfn = 0;
4417
Tejun Heoc13291a2011-07-12 10:46:30 +02004418 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
4419 *start_pfn = min(*start_pfn, this_start_pfn);
4420 *end_pfn = max(*end_pfn, this_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004421 }
4422
Christoph Lameter633c0662007-10-16 01:25:37 -07004423 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07004424 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07004425}
4426
4427/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07004428 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4429 * assumption is made that zones within a node are ordered in monotonic
4430 * increasing memory addresses so that the "highest" populated zone is used
4431 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004432static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004433{
4434 int zone_index;
4435 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
4436 if (zone_index == ZONE_MOVABLE)
4437 continue;
4438
4439 if (arch_zone_highest_possible_pfn[zone_index] >
4440 arch_zone_lowest_possible_pfn[zone_index])
4441 break;
4442 }
4443
4444 VM_BUG_ON(zone_index == -1);
4445 movable_zone = zone_index;
4446}
4447
4448/*
4449 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004450 * because it is sized independent of architecture. Unlike the other zones,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004451 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4452 * in each node depending on the size of each node and how evenly kernelcore
4453 * is distributed. This helper function adjusts the zone ranges
4454 * provided by the architecture for a given node by using the end of the
4455 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4456 * zones within a node are in order of monotonic increases memory addresses
4457 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004458static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004459 unsigned long zone_type,
4460 unsigned long node_start_pfn,
4461 unsigned long node_end_pfn,
4462 unsigned long *zone_start_pfn,
4463 unsigned long *zone_end_pfn)
4464{
4465 /* Only adjust if ZONE_MOVABLE is on this node */
4466 if (zone_movable_pfn[nid]) {
4467 /* Size ZONE_MOVABLE */
4468 if (zone_type == ZONE_MOVABLE) {
4469 *zone_start_pfn = zone_movable_pfn[nid];
4470 *zone_end_pfn = min(node_end_pfn,
4471 arch_zone_highest_possible_pfn[movable_zone]);
4472
4473 /* Adjust for ZONE_MOVABLE starting within this range */
4474 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
4475 *zone_end_pfn > zone_movable_pfn[nid]) {
4476 *zone_end_pfn = zone_movable_pfn[nid];
4477
4478 /* Check if this whole range is within ZONE_MOVABLE */
4479 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
4480 *zone_start_pfn = *zone_end_pfn;
4481 }
4482}
4483
4484/*
Mel Gormanc7132162006-09-27 01:49:43 -07004485 * Return the number of pages a zone spans in a node, including holes
4486 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4487 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004488static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004489 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004490 unsigned long node_start_pfn,
4491 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004492 unsigned long *ignored)
4493{
Mel Gormanc7132162006-09-27 01:49:43 -07004494 unsigned long zone_start_pfn, zone_end_pfn;
4495
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004496 /* Get the start and end of the zone */
Mel Gormanc7132162006-09-27 01:49:43 -07004497 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4498 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07004499 adjust_zone_range_for_zone_movable(nid, zone_type,
4500 node_start_pfn, node_end_pfn,
4501 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004502
4503 /* Check that this node has pages within the zone's required range */
4504 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4505 return 0;
4506
4507 /* Move the zone boundaries inside the node if necessary */
4508 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4509 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4510
4511 /* Return the spanned pages */
4512 return zone_end_pfn - zone_start_pfn;
4513}
4514
4515/*
4516 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004517 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07004518 */
Yinghai Lu32996252009-12-15 17:59:02 -08004519unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004520 unsigned long range_start_pfn,
4521 unsigned long range_end_pfn)
4522{
Tejun Heo96e907d2011-07-12 10:46:29 +02004523 unsigned long nr_absent = range_end_pfn - range_start_pfn;
4524 unsigned long start_pfn, end_pfn;
4525 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004526
Tejun Heo96e907d2011-07-12 10:46:29 +02004527 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
4528 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
4529 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
4530 nr_absent -= end_pfn - start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004531 }
Tejun Heo96e907d2011-07-12 10:46:29 +02004532 return nr_absent;
Mel Gormanc7132162006-09-27 01:49:43 -07004533}
4534
4535/**
4536 * absent_pages_in_range - Return number of page frames in holes within a range
4537 * @start_pfn: The start PFN to start searching for holes
4538 * @end_pfn: The end PFN to stop searching for holes
4539 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004540 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07004541 */
4542unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4543 unsigned long end_pfn)
4544{
4545 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4546}
4547
4548/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004549static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004550 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004551 unsigned long node_start_pfn,
4552 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004553 unsigned long *ignored)
4554{
Tejun Heo96e907d2011-07-12 10:46:29 +02004555 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
4556 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman9c7cd682006-09-27 01:49:58 -07004557 unsigned long zone_start_pfn, zone_end_pfn;
4558
Tejun Heo96e907d2011-07-12 10:46:29 +02004559 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
4560 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004561
Mel Gorman2a1e2742007-07-17 04:03:12 -07004562 adjust_zone_range_for_zone_movable(nid, zone_type,
4563 node_start_pfn, node_end_pfn,
4564 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004565 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004566}
Mel Gorman0e0b8642006-09-27 01:49:56 -07004567
Tejun Heo0ee332c2011-12-08 10:22:09 -08004568#else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004569static inline unsigned long __meminit zone_spanned_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 *zones_size)
4574{
4575 return zones_size[zone_type];
4576}
4577
Paul Mundt6ea6e682007-07-15 23:38:20 -07004578static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004579 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004580 unsigned long node_start_pfn,
4581 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004582 unsigned long *zholes_size)
4583{
4584 if (!zholes_size)
4585 return 0;
4586
4587 return zholes_size[zone_type];
4588}
Yinghai Lu20e69262013-03-01 14:51:27 -08004589
Tejun Heo0ee332c2011-12-08 10:22:09 -08004590#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07004591
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004592static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004593 unsigned long node_start_pfn,
4594 unsigned long node_end_pfn,
4595 unsigned long *zones_size,
4596 unsigned long *zholes_size)
Mel Gormanc7132162006-09-27 01:49:43 -07004597{
4598 unsigned long realtotalpages, totalpages = 0;
4599 enum zone_type i;
4600
4601 for (i = 0; i < MAX_NR_ZONES; i++)
4602 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004603 node_start_pfn,
4604 node_end_pfn,
4605 zones_size);
Mel Gormanc7132162006-09-27 01:49:43 -07004606 pgdat->node_spanned_pages = totalpages;
4607
4608 realtotalpages = totalpages;
4609 for (i = 0; i < MAX_NR_ZONES; i++)
4610 realtotalpages -=
4611 zone_absent_pages_in_node(pgdat->node_id, i,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004612 node_start_pfn, node_end_pfn,
4613 zholes_size);
Mel Gormanc7132162006-09-27 01:49:43 -07004614 pgdat->node_present_pages = realtotalpages;
4615 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4616 realtotalpages);
4617}
4618
Mel Gorman835c1342007-10-16 01:25:47 -07004619#ifndef CONFIG_SPARSEMEM
4620/*
4621 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07004622 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4623 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07004624 * round what is now in bits to nearest long in bits, then return it in
4625 * bytes.
4626 */
Linus Torvalds7c455122013-02-18 09:58:02 -08004627static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07004628{
4629 unsigned long usemapsize;
4630
Linus Torvalds7c455122013-02-18 09:58:02 -08004631 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
Mel Gormand9c23402007-10-16 01:26:01 -07004632 usemapsize = roundup(zonesize, pageblock_nr_pages);
4633 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07004634 usemapsize *= NR_PAGEBLOCK_BITS;
4635 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4636
4637 return usemapsize / 8;
4638}
4639
4640static void __init setup_usemap(struct pglist_data *pgdat,
Linus Torvalds7c455122013-02-18 09:58:02 -08004641 struct zone *zone,
4642 unsigned long zone_start_pfn,
4643 unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07004644{
Linus Torvalds7c455122013-02-18 09:58:02 -08004645 unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
Mel Gorman835c1342007-10-16 01:25:47 -07004646 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08004647 if (usemapsize)
Yinghai Lu8f389a992011-05-11 15:13:32 -07004648 zone->pageblock_flags = alloc_bootmem_node_nopanic(pgdat,
4649 usemapsize);
Mel Gorman835c1342007-10-16 01:25:47 -07004650}
4651#else
Linus Torvalds7c455122013-02-18 09:58:02 -08004652static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
4653 unsigned long zone_start_pfn, unsigned long zonesize) {}
Mel Gorman835c1342007-10-16 01:25:47 -07004654#endif /* CONFIG_SPARSEMEM */
4655
Mel Gormand9c23402007-10-16 01:26:01 -07004656#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08004657
Mel Gormand9c23402007-10-16 01:26:01 -07004658/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
Chen Gang15ca2202013-09-11 14:20:27 -07004659void __paginginit set_pageblock_order(void)
Mel Gormand9c23402007-10-16 01:26:01 -07004660{
Andrew Morton955c1cd2012-05-29 15:06:31 -07004661 unsigned int order;
4662
Mel Gormand9c23402007-10-16 01:26:01 -07004663 /* Check that pageblock_nr_pages has not already been setup */
4664 if (pageblock_order)
4665 return;
4666
Andrew Morton955c1cd2012-05-29 15:06:31 -07004667 if (HPAGE_SHIFT > PAGE_SHIFT)
4668 order = HUGETLB_PAGE_ORDER;
4669 else
4670 order = MAX_ORDER - 1;
4671
Mel Gormand9c23402007-10-16 01:26:01 -07004672 /*
4673 * Assume the largest contiguous order of interest is a huge page.
Andrew Morton955c1cd2012-05-29 15:06:31 -07004674 * This value may be variable depending on boot parameters on IA64 and
4675 * powerpc.
Mel Gormand9c23402007-10-16 01:26:01 -07004676 */
4677 pageblock_order = order;
4678}
4679#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4680
Mel Gormanba72cb82007-11-28 16:21:13 -08004681/*
4682 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
Andrew Morton955c1cd2012-05-29 15:06:31 -07004683 * is unused as pageblock_order is set at compile-time. See
4684 * include/linux/pageblock-flags.h for the values of pageblock_order based on
4685 * the kernel config
Mel Gormanba72cb82007-11-28 16:21:13 -08004686 */
Chen Gang15ca2202013-09-11 14:20:27 -07004687void __paginginit set_pageblock_order(void)
Mel Gormanba72cb82007-11-28 16:21:13 -08004688{
Mel Gormanba72cb82007-11-28 16:21:13 -08004689}
Mel Gormand9c23402007-10-16 01:26:01 -07004690
4691#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4692
Jiang Liu01cefae2012-12-12 13:52:19 -08004693static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
4694 unsigned long present_pages)
4695{
4696 unsigned long pages = spanned_pages;
4697
4698 /*
4699 * Provide a more accurate estimation if there are holes within
4700 * the zone and SPARSEMEM is in use. If there are holes within the
4701 * zone, each populated memory region may cost us one or two extra
4702 * memmap pages due to alignment because memmap pages for each
4703 * populated regions may not naturally algined on page boundary.
4704 * So the (present_pages >> 4) heuristic is a tradeoff for that.
4705 */
4706 if (spanned_pages > present_pages + (present_pages >> 4) &&
4707 IS_ENABLED(CONFIG_SPARSEMEM))
4708 pages = present_pages;
4709
4710 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
4711}
4712
Linus Torvalds1da177e2005-04-16 15:20:36 -07004713/*
4714 * Set up the zone data structures:
4715 * - mark all pages reserved
4716 * - mark all memory queues empty
4717 * - clear the memory bitmaps
Minchan Kim6527af52012-07-31 16:46:16 -07004718 *
4719 * NOTE: pgdat should get zeroed by caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004720 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08004721static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004722 unsigned long node_start_pfn, unsigned long node_end_pfn,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004723 unsigned long *zones_size, unsigned long *zholes_size)
4724{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07004725 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07004726 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004727 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07004728 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004729
Dave Hansen208d54e2005-10-29 18:16:52 -07004730 pgdat_resize_init(pgdat);
Andrea Arcangeli8177a422012-03-23 20:56:34 +01004731#ifdef CONFIG_NUMA_BALANCING
4732 spin_lock_init(&pgdat->numabalancing_migrate_lock);
4733 pgdat->numabalancing_migrate_nr_pages = 0;
4734 pgdat->numabalancing_migrate_next_window = jiffies;
4735#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004736 init_waitqueue_head(&pgdat->kswapd_wait);
Mel Gorman55150612012-07-31 16:44:35 -07004737 init_waitqueue_head(&pgdat->pfmemalloc_wait);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004738 pgdat_page_cgroup_init(pgdat);
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01004739
Linus Torvalds1da177e2005-04-16 15:20:36 -07004740 for (j = 0; j < MAX_NR_ZONES; j++) {
4741 struct zone *zone = pgdat->node_zones + j;
Jiang Liu9feedc92012-12-12 13:52:12 -08004742 unsigned long size, realsize, freesize, memmap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004743
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004744 size = zone_spanned_pages_in_node(nid, j, node_start_pfn,
4745 node_end_pfn, zones_size);
Jiang Liu9feedc92012-12-12 13:52:12 -08004746 realsize = freesize = size - zone_absent_pages_in_node(nid, j,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004747 node_start_pfn,
4748 node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004749 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004750
Mel Gorman0e0b8642006-09-27 01:49:56 -07004751 /*
Jiang Liu9feedc92012-12-12 13:52:12 -08004752 * Adjust freesize so that it accounts for how much memory
Mel Gorman0e0b8642006-09-27 01:49:56 -07004753 * is used by this zone for memmap. This affects the watermark
4754 * and per-cpu initialisations
4755 */
Jiang Liu01cefae2012-12-12 13:52:19 -08004756 memmap_pages = calc_memmap_size(size, realsize);
Jiang Liu9feedc92012-12-12 13:52:12 -08004757 if (freesize >= memmap_pages) {
4758 freesize -= memmap_pages;
Yinghai Lu5594c8c2009-01-06 14:39:14 -08004759 if (memmap_pages)
4760 printk(KERN_DEBUG
4761 " %s zone: %lu pages used for memmap\n",
4762 zone_names[j], memmap_pages);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004763 } else
4764 printk(KERN_WARNING
Jiang Liu9feedc92012-12-12 13:52:12 -08004765 " %s zone: %lu pages exceeds freesize %lu\n",
4766 zone_names[j], memmap_pages, freesize);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004767
Christoph Lameter62672762007-02-10 01:43:07 -08004768 /* Account for reserved pages */
Jiang Liu9feedc92012-12-12 13:52:12 -08004769 if (j == 0 && freesize > dma_reserve) {
4770 freesize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07004771 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08004772 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004773 }
4774
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004775 if (!is_highmem_idx(j))
Jiang Liu9feedc92012-12-12 13:52:12 -08004776 nr_kernel_pages += freesize;
Jiang Liu01cefae2012-12-12 13:52:19 -08004777 /* Charge for highmem memmap if there are enough kernel pages */
4778 else if (nr_kernel_pages > memmap_pages * 2)
4779 nr_kernel_pages -= memmap_pages;
Jiang Liu9feedc92012-12-12 13:52:12 -08004780 nr_all_pages += freesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004781
4782 zone->spanned_pages = size;
Jiang Liu306f2e92013-02-22 16:33:54 -08004783 zone->present_pages = realsize;
Jiang Liu9feedc92012-12-12 13:52:12 -08004784 /*
4785 * Set an approximate value for lowmem here, it will be adjusted
4786 * when the bootmem allocator frees pages into the buddy system.
4787 * And all highmem pages will be managed by the buddy system.
4788 */
4789 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
Christoph Lameter96146342006-07-03 00:24:13 -07004790#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07004791 zone->node = nid;
Jiang Liu9feedc92012-12-12 13:52:12 -08004792 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07004793 / 100;
Jiang Liu9feedc92012-12-12 13:52:12 -08004794 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07004795#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004796 zone->name = zone_names[j];
4797 spin_lock_init(&zone->lock);
4798 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07004799 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004800 zone->zone_pgdat = pgdat;
Dave Hansened8ece22005-10-29 18:16:50 -07004801 zone_pcp_init(zone);
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07004802
4803 /* For bootup, initialized properly in watermark setup */
4804 mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages);
4805
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004806 lruvec_init(&zone->lruvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004807 if (!size)
4808 continue;
4809
Andrew Morton955c1cd2012-05-29 15:06:31 -07004810 set_pageblock_order();
Linus Torvalds7c455122013-02-18 09:58:02 -08004811 setup_usemap(pgdat, zone, zone_start_pfn, size);
Dave Hansena2f3aa022007-01-10 23:15:30 -08004812 ret = init_currently_empty_zone(zone, zone_start_pfn,
4813 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07004814 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07004815 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004816 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004817 }
4818}
4819
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004820static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004821{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004822 /* Skip empty nodes */
4823 if (!pgdat->node_spanned_pages)
4824 return;
4825
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004826#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07004827 /* ia64 gets its own node_mem_map, before this, without bootmem */
4828 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07004829 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004830 struct page *map;
4831
Bob Piccoe984bb42006-05-20 15:00:31 -07004832 /*
4833 * The zone's endpoints aren't required to be MAX_ORDER
4834 * aligned but the node_mem_map endpoints must be in order
4835 * for the buddy allocator to function correctly.
4836 */
4837 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
Cody P Schafer108bcc92013-02-22 16:35:23 -08004838 end = pgdat_end_pfn(pgdat);
Bob Piccoe984bb42006-05-20 15:00:31 -07004839 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4840 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07004841 map = alloc_remap(pgdat->node_id, size);
4842 if (!map)
Yinghai Lu8f389a992011-05-11 15:13:32 -07004843 map = alloc_bootmem_node_nopanic(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07004844 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004845 }
Roman Zippel12d810c2007-05-31 00:40:54 -07004846#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07004847 /*
4848 * With no DISCONTIG, the global mem_map is just set as node 0's
4849 */
Mel Gormanc7132162006-09-27 01:49:43 -07004850 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004851 mem_map = NODE_DATA(0)->node_mem_map;
Tejun Heo0ee332c2011-12-08 10:22:09 -08004852#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07004853 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08004854 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Tejun Heo0ee332c2011-12-08 10:22:09 -08004855#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07004856 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004857#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004858#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004859}
4860
Johannes Weiner9109fb72008-07-23 21:27:20 -07004861void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4862 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004863{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004864 pg_data_t *pgdat = NODE_DATA(nid);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004865 unsigned long start_pfn = 0;
4866 unsigned long end_pfn = 0;
Johannes Weiner9109fb72008-07-23 21:27:20 -07004867
Minchan Kim88fdf752012-07-31 16:46:14 -07004868 /* pg_data_t should be reset to zero when it's allocated */
Linus Torvalds8783b6e2012-08-02 10:37:03 -07004869 WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
Minchan Kim88fdf752012-07-31 16:46:14 -07004870
Linus Torvalds1da177e2005-04-16 15:20:36 -07004871 pgdat->node_id = nid;
4872 pgdat->node_start_pfn = node_start_pfn;
David Rientjes957f8222012-10-08 16:33:24 -07004873 init_zone_allows_reclaim(nid);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004874#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4875 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
4876#endif
4877 calculate_node_totalpages(pgdat, start_pfn, end_pfn,
4878 zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004879
4880 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07004881#ifdef CONFIG_FLAT_NODE_MEM_MAP
4882 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4883 nid, (unsigned long)pgdat,
4884 (unsigned long)pgdat->node_mem_map);
4885#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004886
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004887 free_area_init_core(pgdat, start_pfn, end_pfn,
4888 zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004889}
4890
Tejun Heo0ee332c2011-12-08 10:22:09 -08004891#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07004892
4893#if MAX_NUMNODES > 1
4894/*
4895 * Figure out the number of possible node ids.
4896 */
Cody P Schaferf9872ca2013-04-29 15:08:01 -07004897void __init setup_nr_node_ids(void)
Miklos Szeredi418508c2007-05-23 13:57:55 -07004898{
4899 unsigned int node;
4900 unsigned int highest = 0;
4901
4902 for_each_node_mask(node, node_possible_map)
4903 highest = node;
4904 nr_node_ids = highest + 1;
4905}
Miklos Szeredi418508c2007-05-23 13:57:55 -07004906#endif
4907
Mel Gormanc7132162006-09-27 01:49:43 -07004908/**
Tejun Heo1e019792011-07-12 09:45:34 +02004909 * node_map_pfn_alignment - determine the maximum internode alignment
4910 *
4911 * This function should be called after node map is populated and sorted.
4912 * It calculates the maximum power of two alignment which can distinguish
4913 * all the nodes.
4914 *
4915 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
4916 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
4917 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
4918 * shifted, 1GiB is enough and this function will indicate so.
4919 *
4920 * This is used to test whether pfn -> nid mapping of the chosen memory
4921 * model has fine enough granularity to avoid incorrect mapping for the
4922 * populated node map.
4923 *
4924 * Returns the determined alignment in pfn's. 0 if there is no alignment
4925 * requirement (single node).
4926 */
4927unsigned long __init node_map_pfn_alignment(void)
4928{
4929 unsigned long accl_mask = 0, last_end = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02004930 unsigned long start, end, mask;
Tejun Heo1e019792011-07-12 09:45:34 +02004931 int last_nid = -1;
Tejun Heoc13291a2011-07-12 10:46:30 +02004932 int i, nid;
Tejun Heo1e019792011-07-12 09:45:34 +02004933
Tejun Heoc13291a2011-07-12 10:46:30 +02004934 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
Tejun Heo1e019792011-07-12 09:45:34 +02004935 if (!start || last_nid < 0 || last_nid == nid) {
4936 last_nid = nid;
4937 last_end = end;
4938 continue;
4939 }
4940
4941 /*
4942 * Start with a mask granular enough to pin-point to the
4943 * start pfn and tick off bits one-by-one until it becomes
4944 * too coarse to separate the current node from the last.
4945 */
4946 mask = ~((1 << __ffs(start)) - 1);
4947 while (mask && last_end <= (start & (mask << 1)))
4948 mask <<= 1;
4949
4950 /* accumulate all internode masks */
4951 accl_mask |= mask;
4952 }
4953
4954 /* convert mask to number of pages */
4955 return ~accl_mask + 1;
4956}
4957
Mel Gormana6af2bc2007-02-10 01:42:57 -08004958/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004959static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07004960{
Mel Gormana6af2bc2007-02-10 01:42:57 -08004961 unsigned long min_pfn = ULONG_MAX;
Tejun Heoc13291a2011-07-12 10:46:30 +02004962 unsigned long start_pfn;
4963 int i;
Mel Gorman1abbfb42006-11-23 12:01:41 +00004964
Tejun Heoc13291a2011-07-12 10:46:30 +02004965 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
4966 min_pfn = min(min_pfn, start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004967
Mel Gormana6af2bc2007-02-10 01:42:57 -08004968 if (min_pfn == ULONG_MAX) {
4969 printk(KERN_WARNING
Paul Jackson2bc0d262008-06-22 07:22:17 -07004970 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08004971 return 0;
4972 }
4973
4974 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004975}
4976
4977/**
4978 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4979 *
4980 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004981 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07004982 */
4983unsigned long __init find_min_pfn_with_active_regions(void)
4984{
4985 return find_min_pfn_for_node(MAX_NUMNODES);
4986}
4987
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004988/*
4989 * early_calculate_totalpages()
4990 * Sum pages in active regions for movable zone.
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08004991 * Populate N_MEMORY for calculating usable_nodes.
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004992 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07004993static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07004994{
Mel Gorman7e63efe2007-07-17 04:03:15 -07004995 unsigned long totalpages = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02004996 unsigned long start_pfn, end_pfn;
4997 int i, nid;
Mel Gorman7e63efe2007-07-17 04:03:15 -07004998
Tejun Heoc13291a2011-07-12 10:46:30 +02004999 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
5000 unsigned long pages = end_pfn - start_pfn;
5001
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005002 totalpages += pages;
5003 if (pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005004 node_set_state(nid, N_MEMORY);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005005 }
Pintu Kumarb8af2942013-09-11 14:20:34 -07005006 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07005007}
5008
Mel Gorman2a1e2742007-07-17 04:03:12 -07005009/*
5010 * Find the PFN the Movable zone begins in each node. Kernel memory
5011 * is spread evenly between nodes as long as the nodes have enough
5012 * memory. When they don't, some nodes will have more kernelcore than
5013 * others
5014 */
Kautuk Consulb224ef82012-03-21 16:34:15 -07005015static void __init find_zone_movable_pfns_for_nodes(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005016{
5017 int i, nid;
5018 unsigned long usable_startpfn;
5019 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07005020 /* save the state before borrow the nodemask */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005021 nodemask_t saved_node_state = node_states[N_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005022 unsigned long totalpages = early_calculate_totalpages();
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005023 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005024
Mel Gorman7e63efe2007-07-17 04:03:15 -07005025 /*
5026 * If movablecore was specified, calculate what size of
5027 * kernelcore that corresponds so that memory usable for
5028 * any allocation type is evenly spread. If both kernelcore
5029 * and movablecore are specified, then the value of kernelcore
5030 * will be used for required_kernelcore if it's greater than
5031 * what movablecore would have allowed.
5032 */
5033 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07005034 unsigned long corepages;
5035
5036 /*
5037 * Round-up so that ZONE_MOVABLE is at least as large as what
5038 * was requested by the user
5039 */
5040 required_movablecore =
5041 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
5042 corepages = totalpages - required_movablecore;
5043
5044 required_kernelcore = max(required_kernelcore, corepages);
5045 }
5046
Yinghai Lu20e69262013-03-01 14:51:27 -08005047 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
5048 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07005049 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005050
5051 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
Yinghai Lu20e69262013-03-01 14:51:27 -08005052 find_usable_zone_for_movable();
Mel Gorman2a1e2742007-07-17 04:03:12 -07005053 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
5054
5055restart:
5056 /* Spread kernelcore memory as evenly as possible throughout nodes */
5057 kernelcore_node = required_kernelcore / usable_nodes;
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005058 for_each_node_state(nid, N_MEMORY) {
Tejun Heoc13291a2011-07-12 10:46:30 +02005059 unsigned long start_pfn, end_pfn;
5060
Mel Gorman2a1e2742007-07-17 04:03:12 -07005061 /*
5062 * Recalculate kernelcore_node if the division per node
5063 * now exceeds what is necessary to satisfy the requested
5064 * amount of memory for the kernel
5065 */
5066 if (required_kernelcore < kernelcore_node)
5067 kernelcore_node = required_kernelcore / usable_nodes;
5068
5069 /*
5070 * As the map is walked, we track how much memory is usable
5071 * by the kernel using kernelcore_remaining. When it is
5072 * 0, the rest of the node is usable by ZONE_MOVABLE
5073 */
5074 kernelcore_remaining = kernelcore_node;
5075
5076 /* Go through each range of PFNs within this node */
Tejun Heoc13291a2011-07-12 10:46:30 +02005077 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005078 unsigned long size_pages;
5079
Tejun Heoc13291a2011-07-12 10:46:30 +02005080 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005081 if (start_pfn >= end_pfn)
5082 continue;
5083
5084 /* Account for what is only usable for kernelcore */
5085 if (start_pfn < usable_startpfn) {
5086 unsigned long kernel_pages;
5087 kernel_pages = min(end_pfn, usable_startpfn)
5088 - start_pfn;
5089
5090 kernelcore_remaining -= min(kernel_pages,
5091 kernelcore_remaining);
5092 required_kernelcore -= min(kernel_pages,
5093 required_kernelcore);
5094
5095 /* Continue if range is now fully accounted */
5096 if (end_pfn <= usable_startpfn) {
5097
5098 /*
5099 * Push zone_movable_pfn to the end so
5100 * that if we have to rebalance
5101 * kernelcore across nodes, we will
5102 * not double account here
5103 */
5104 zone_movable_pfn[nid] = end_pfn;
5105 continue;
5106 }
5107 start_pfn = usable_startpfn;
5108 }
5109
5110 /*
5111 * The usable PFN range for ZONE_MOVABLE is from
5112 * start_pfn->end_pfn. Calculate size_pages as the
5113 * number of pages used as kernelcore
5114 */
5115 size_pages = end_pfn - start_pfn;
5116 if (size_pages > kernelcore_remaining)
5117 size_pages = kernelcore_remaining;
5118 zone_movable_pfn[nid] = start_pfn + size_pages;
5119
5120 /*
5121 * Some kernelcore has been met, update counts and
5122 * break if the kernelcore for this node has been
Pintu Kumarb8af2942013-09-11 14:20:34 -07005123 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07005124 */
5125 required_kernelcore -= min(required_kernelcore,
5126 size_pages);
5127 kernelcore_remaining -= size_pages;
5128 if (!kernelcore_remaining)
5129 break;
5130 }
5131 }
5132
5133 /*
5134 * If there is still required_kernelcore, we do another pass with one
5135 * less node in the count. This will push zone_movable_pfn[nid] further
5136 * along on the nodes that still have memory until kernelcore is
Pintu Kumarb8af2942013-09-11 14:20:34 -07005137 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07005138 */
5139 usable_nodes--;
5140 if (usable_nodes && required_kernelcore > usable_nodes)
5141 goto restart;
5142
5143 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5144 for (nid = 0; nid < MAX_NUMNODES; nid++)
5145 zone_movable_pfn[nid] =
5146 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07005147
Yinghai Lu20e69262013-03-01 14:51:27 -08005148out:
Yinghai Lu66918dc2009-06-30 11:41:37 -07005149 /* restore the node_state */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005150 node_states[N_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005151}
5152
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005153/* Any regular or high memory on that node ? */
5154static void check_for_memory(pg_data_t *pgdat, int nid)
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005155{
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005156 enum zone_type zone_type;
5157
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005158 if (N_MEMORY == N_NORMAL_MEMORY)
5159 return;
5160
5161 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005162 struct zone *zone = &pgdat->node_zones[zone_type];
Bob Liud0048b02012-01-12 17:19:07 -08005163 if (zone->present_pages) {
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005164 node_set_state(nid, N_HIGH_MEMORY);
5165 if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
5166 zone_type <= ZONE_NORMAL)
5167 node_set_state(nid, N_NORMAL_MEMORY);
Bob Liud0048b02012-01-12 17:19:07 -08005168 break;
5169 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005170 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005171}
5172
Mel Gormanc7132162006-09-27 01:49:43 -07005173/**
5174 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005175 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07005176 *
5177 * This will call free_area_init_node() for each active node in the system.
5178 * Using the page ranges provided by add_active_range(), the size of each
5179 * zone in each node and their holes is calculated. If the maximum PFN
5180 * between two adjacent zones match, it is assumed that the zone is empty.
5181 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5182 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5183 * starts where the previous one ended. For example, ZONE_DMA32 starts
5184 * at arch_max_dma_pfn.
5185 */
5186void __init free_area_init_nodes(unsigned long *max_zone_pfn)
5187{
Tejun Heoc13291a2011-07-12 10:46:30 +02005188 unsigned long start_pfn, end_pfn;
5189 int i, nid;
Mel Gormana6af2bc2007-02-10 01:42:57 -08005190
Mel Gormanc7132162006-09-27 01:49:43 -07005191 /* Record where the zone boundaries are */
5192 memset(arch_zone_lowest_possible_pfn, 0,
5193 sizeof(arch_zone_lowest_possible_pfn));
5194 memset(arch_zone_highest_possible_pfn, 0,
5195 sizeof(arch_zone_highest_possible_pfn));
5196 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
5197 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
5198 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005199 if (i == ZONE_MOVABLE)
5200 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07005201 arch_zone_lowest_possible_pfn[i] =
5202 arch_zone_highest_possible_pfn[i-1];
5203 arch_zone_highest_possible_pfn[i] =
5204 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
5205 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07005206 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
5207 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
5208
5209 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5210 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
Kautuk Consulb224ef82012-03-21 16:34:15 -07005211 find_zone_movable_pfns_for_nodes();
Mel Gormanc7132162006-09-27 01:49:43 -07005212
Mel Gormanc7132162006-09-27 01:49:43 -07005213 /* Print out the zone ranges */
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005214 printk("Zone ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005215 for (i = 0; i < MAX_NR_ZONES; i++) {
5216 if (i == ZONE_MOVABLE)
5217 continue;
Kay Sievers155cbfc2012-05-08 17:24:14 +02005218 printk(KERN_CONT " %-8s ", zone_names[i]);
David Rientjes72f0ba02010-03-05 13:42:14 -08005219 if (arch_zone_lowest_possible_pfn[i] ==
5220 arch_zone_highest_possible_pfn[i])
Kay Sievers155cbfc2012-05-08 17:24:14 +02005221 printk(KERN_CONT "empty\n");
David Rientjes72f0ba02010-03-05 13:42:14 -08005222 else
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005223 printk(KERN_CONT "[mem %0#10lx-%0#10lx]\n",
5224 arch_zone_lowest_possible_pfn[i] << PAGE_SHIFT,
5225 (arch_zone_highest_possible_pfn[i]
5226 << PAGE_SHIFT) - 1);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005227 }
5228
5229 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005230 printk("Movable zone start for each node\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005231 for (i = 0; i < MAX_NUMNODES; i++) {
5232 if (zone_movable_pfn[i])
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005233 printk(" Node %d: %#010lx\n", i,
5234 zone_movable_pfn[i] << PAGE_SHIFT);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005235 }
Mel Gormanc7132162006-09-27 01:49:43 -07005236
Wanpeng Lif2d52fe2012-10-08 16:32:24 -07005237 /* Print out the early node map */
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005238 printk("Early memory node ranges\n");
Tejun Heoc13291a2011-07-12 10:46:30 +02005239 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005240 printk(" node %3d: [mem %#010lx-%#010lx]\n", nid,
5241 start_pfn << PAGE_SHIFT, (end_pfn << PAGE_SHIFT) - 1);
Mel Gormanc7132162006-09-27 01:49:43 -07005242
5243 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07005244 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08005245 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07005246 for_each_online_node(nid) {
5247 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07005248 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07005249 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005250
5251 /* Any memory on that node */
5252 if (pgdat->node_present_pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005253 node_set_state(nid, N_MEMORY);
5254 check_for_memory(pgdat, nid);
Mel Gormanc7132162006-09-27 01:49:43 -07005255 }
5256}
Mel Gorman2a1e2742007-07-17 04:03:12 -07005257
Mel Gorman7e63efe2007-07-17 04:03:15 -07005258static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005259{
5260 unsigned long long coremem;
5261 if (!p)
5262 return -EINVAL;
5263
5264 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005265 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005266
Mel Gorman7e63efe2007-07-17 04:03:15 -07005267 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07005268 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
5269
5270 return 0;
5271}
Mel Gormaned7ed362007-07-17 04:03:14 -07005272
Mel Gorman7e63efe2007-07-17 04:03:15 -07005273/*
5274 * kernelcore=size sets the amount of memory for use for allocations that
5275 * cannot be reclaimed or migrated.
5276 */
5277static int __init cmdline_parse_kernelcore(char *p)
5278{
5279 return cmdline_parse_core(p, &required_kernelcore);
5280}
5281
5282/*
5283 * movablecore=size sets the amount of memory for use for allocations that
5284 * can be reclaimed or migrated.
5285 */
5286static int __init cmdline_parse_movablecore(char *p)
5287{
5288 return cmdline_parse_core(p, &required_movablecore);
5289}
5290
Mel Gormaned7ed362007-07-17 04:03:14 -07005291early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005292early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07005293
Tejun Heo0ee332c2011-12-08 10:22:09 -08005294#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005295
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005296void adjust_managed_page_count(struct page *page, long count)
5297{
5298 spin_lock(&managed_page_count_lock);
5299 page_zone(page)->managed_pages += count;
5300 totalram_pages += count;
Jiang Liu3dcc0572013-07-03 15:03:21 -07005301#ifdef CONFIG_HIGHMEM
5302 if (PageHighMem(page))
5303 totalhigh_pages += count;
5304#endif
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005305 spin_unlock(&managed_page_count_lock);
5306}
Jiang Liu3dcc0572013-07-03 15:03:21 -07005307EXPORT_SYMBOL(adjust_managed_page_count);
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005308
Jiang Liu11199692013-07-03 15:02:48 -07005309unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
Jiang Liu69afade2013-04-29 15:06:21 -07005310{
Jiang Liu11199692013-07-03 15:02:48 -07005311 void *pos;
5312 unsigned long pages = 0;
Jiang Liu69afade2013-04-29 15:06:21 -07005313
Jiang Liu11199692013-07-03 15:02:48 -07005314 start = (void *)PAGE_ALIGN((unsigned long)start);
5315 end = (void *)((unsigned long)end & PAGE_MASK);
5316 for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
Jiang Liudbe67df2013-07-03 15:02:51 -07005317 if ((unsigned int)poison <= 0xFF)
Jiang Liu11199692013-07-03 15:02:48 -07005318 memset(pos, poison, PAGE_SIZE);
5319 free_reserved_page(virt_to_page(pos));
Jiang Liu69afade2013-04-29 15:06:21 -07005320 }
5321
5322 if (pages && s)
Jiang Liu11199692013-07-03 15:02:48 -07005323 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
Jiang Liu69afade2013-04-29 15:06:21 -07005324 s, pages << (PAGE_SHIFT - 10), start, end);
5325
5326 return pages;
5327}
Jiang Liu11199692013-07-03 15:02:48 -07005328EXPORT_SYMBOL(free_reserved_area);
Jiang Liu69afade2013-04-29 15:06:21 -07005329
Jiang Liucfa11e02013-04-29 15:07:00 -07005330#ifdef CONFIG_HIGHMEM
5331void free_highmem_page(struct page *page)
5332{
5333 __free_reserved_page(page);
5334 totalram_pages++;
Jiang Liu7b4b2a02013-07-03 15:03:11 -07005335 page_zone(page)->managed_pages++;
Jiang Liucfa11e02013-04-29 15:07:00 -07005336 totalhigh_pages++;
5337}
5338#endif
5339
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005340
5341void __init mem_init_print_info(const char *str)
5342{
5343 unsigned long physpages, codesize, datasize, rosize, bss_size;
5344 unsigned long init_code_size, init_data_size;
5345
5346 physpages = get_num_physpages();
5347 codesize = _etext - _stext;
5348 datasize = _edata - _sdata;
5349 rosize = __end_rodata - __start_rodata;
5350 bss_size = __bss_stop - __bss_start;
5351 init_data_size = __init_end - __init_begin;
5352 init_code_size = _einittext - _sinittext;
5353
5354 /*
5355 * Detect special cases and adjust section sizes accordingly:
5356 * 1) .init.* may be embedded into .data sections
5357 * 2) .init.text.* may be out of [__init_begin, __init_end],
5358 * please refer to arch/tile/kernel/vmlinux.lds.S.
5359 * 3) .rodata.* may be embedded into .text or .data sections.
5360 */
5361#define adj_init_size(start, end, size, pos, adj) \
Pintu Kumarb8af2942013-09-11 14:20:34 -07005362 do { \
5363 if (start <= pos && pos < end && size > adj) \
5364 size -= adj; \
5365 } while (0)
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005366
5367 adj_init_size(__init_begin, __init_end, init_data_size,
5368 _sinittext, init_code_size);
5369 adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
5370 adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
5371 adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
5372 adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
5373
5374#undef adj_init_size
5375
5376 printk("Memory: %luK/%luK available "
5377 "(%luK kernel code, %luK rwdata, %luK rodata, "
5378 "%luK init, %luK bss, %luK reserved"
5379#ifdef CONFIG_HIGHMEM
5380 ", %luK highmem"
5381#endif
5382 "%s%s)\n",
5383 nr_free_pages() << (PAGE_SHIFT-10), physpages << (PAGE_SHIFT-10),
5384 codesize >> 10, datasize >> 10, rosize >> 10,
5385 (init_data_size + init_code_size) >> 10, bss_size >> 10,
5386 (physpages - totalram_pages) << (PAGE_SHIFT-10),
5387#ifdef CONFIG_HIGHMEM
5388 totalhigh_pages << (PAGE_SHIFT-10),
5389#endif
5390 str ? ", " : "", str ? str : "");
5391}
5392
Mel Gorman0e0b8642006-09-27 01:49:56 -07005393/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005394 * set_dma_reserve - set the specified number of pages reserved in the first zone
5395 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07005396 *
5397 * The per-cpu batchsize and zone watermarks are determined by present_pages.
5398 * In the DMA zone, a significant percentage may be consumed by kernel image
5399 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005400 * function may optionally be used to account for unfreeable pages in the
5401 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5402 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07005403 */
5404void __init set_dma_reserve(unsigned long new_dma_reserve)
5405{
5406 dma_reserve = new_dma_reserve;
5407}
5408
Linus Torvalds1da177e2005-04-16 15:20:36 -07005409void __init free_area_init(unsigned long *zones_size)
5410{
Johannes Weiner9109fb72008-07-23 21:27:20 -07005411 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005412 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
5413}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005414
Linus Torvalds1da177e2005-04-16 15:20:36 -07005415static int page_alloc_cpu_notify(struct notifier_block *self,
5416 unsigned long action, void *hcpu)
5417{
5418 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005419
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07005420 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -07005421 lru_add_drain_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005422 drain_pages(cpu);
5423
5424 /*
5425 * Spill the event counters of the dead processor
5426 * into the current processors event counters.
5427 * This artificially elevates the count of the current
5428 * processor.
5429 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07005430 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005431
5432 /*
5433 * Zero the differential counters of the dead processor
5434 * so that the vm statistics are consistent.
5435 *
5436 * This is only okay since the processor is dead and cannot
5437 * race with what we are doing.
5438 */
Christoph Lameter2bb921e2013-09-11 14:21:30 -07005439 cpu_vm_stats_fold(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005440 }
5441 return NOTIFY_OK;
5442}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005443
5444void __init page_alloc_init(void)
5445{
5446 hotcpu_notifier(page_alloc_cpu_notify, 0);
5447}
5448
5449/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005450 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5451 * or min_free_kbytes changes.
5452 */
5453static void calculate_totalreserve_pages(void)
5454{
5455 struct pglist_data *pgdat;
5456 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005457 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005458
5459 for_each_online_pgdat(pgdat) {
5460 for (i = 0; i < MAX_NR_ZONES; i++) {
5461 struct zone *zone = pgdat->node_zones + i;
5462 unsigned long max = 0;
5463
5464 /* Find valid and maximum lowmem_reserve in the zone */
5465 for (j = i; j < MAX_NR_ZONES; j++) {
5466 if (zone->lowmem_reserve[j] > max)
5467 max = zone->lowmem_reserve[j];
5468 }
5469
Mel Gorman41858962009-06-16 15:32:12 -07005470 /* we treat the high watermark as reserved pages. */
5471 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005472
Jiang Liub40da042013-02-22 16:33:52 -08005473 if (max > zone->managed_pages)
5474 max = zone->managed_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005475 reserve_pages += max;
Johannes Weinerab8fabd2012-01-10 15:07:42 -08005476 /*
5477 * Lowmem reserves are not available to
5478 * GFP_HIGHUSER page cache allocations and
5479 * kswapd tries to balance zones to their high
5480 * watermark. As a result, neither should be
5481 * regarded as dirtyable memory, to prevent a
5482 * situation where reclaim has to clean pages
5483 * in order to balance the zones.
5484 */
5485 zone->dirty_balance_reserve = max;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005486 }
5487 }
Johannes Weinerab8fabd2012-01-10 15:07:42 -08005488 dirty_balance_reserve = reserve_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005489 totalreserve_pages = reserve_pages;
5490}
5491
5492/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07005493 * setup_per_zone_lowmem_reserve - called whenever
5494 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
5495 * has a correct pages reserved value, so an adequate number of
5496 * pages are left in the zone after a successful __alloc_pages().
5497 */
5498static void setup_per_zone_lowmem_reserve(void)
5499{
5500 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005501 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005502
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08005503 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005504 for (j = 0; j < MAX_NR_ZONES; j++) {
5505 struct zone *zone = pgdat->node_zones + j;
Jiang Liub40da042013-02-22 16:33:52 -08005506 unsigned long managed_pages = zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005507
5508 zone->lowmem_reserve[j] = 0;
5509
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005510 idx = j;
5511 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005512 struct zone *lower_zone;
5513
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005514 idx--;
5515
Linus Torvalds1da177e2005-04-16 15:20:36 -07005516 if (sysctl_lowmem_reserve_ratio[idx] < 1)
5517 sysctl_lowmem_reserve_ratio[idx] = 1;
5518
5519 lower_zone = pgdat->node_zones + idx;
Jiang Liub40da042013-02-22 16:33:52 -08005520 lower_zone->lowmem_reserve[j] = managed_pages /
Linus Torvalds1da177e2005-04-16 15:20:36 -07005521 sysctl_lowmem_reserve_ratio[idx];
Jiang Liub40da042013-02-22 16:33:52 -08005522 managed_pages += lower_zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005523 }
5524 }
5525 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005526
5527 /* update totalreserve_pages */
5528 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005529}
5530
Mel Gormancfd3da12011-04-25 21:36:42 +00005531static void __setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005532{
5533 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
5534 unsigned long lowmem_pages = 0;
5535 struct zone *zone;
5536 unsigned long flags;
5537
5538 /* Calculate total number of !ZONE_HIGHMEM pages */
5539 for_each_zone(zone) {
5540 if (!is_highmem(zone))
Jiang Liub40da042013-02-22 16:33:52 -08005541 lowmem_pages += zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005542 }
5543
5544 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07005545 u64 tmp;
5546
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005547 spin_lock_irqsave(&zone->lock, flags);
Jiang Liub40da042013-02-22 16:33:52 -08005548 tmp = (u64)pages_min * zone->managed_pages;
Andrew Mortonac924c62006-05-15 09:43:59 -07005549 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005550 if (is_highmem(zone)) {
5551 /*
Nick Piggin669ed172005-11-13 16:06:45 -08005552 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5553 * need highmem pages, so cap pages_min to a small
5554 * value here.
5555 *
Mel Gorman41858962009-06-16 15:32:12 -07005556 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Nick Piggin669ed172005-11-13 16:06:45 -08005557 * deltas controls asynch page reclaim, and so should
5558 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005559 */
Andrew Morton90ae8d62013-02-22 16:32:22 -08005560 unsigned long min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005561
Jiang Liub40da042013-02-22 16:33:52 -08005562 min_pages = zone->managed_pages / 1024;
Andrew Morton90ae8d62013-02-22 16:32:22 -08005563 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
Mel Gorman41858962009-06-16 15:32:12 -07005564 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005565 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08005566 /*
5567 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07005568 * proportionate to the zone's size.
5569 */
Mel Gorman41858962009-06-16 15:32:12 -07005570 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005571 }
5572
Mel Gorman41858962009-06-16 15:32:12 -07005573 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
5574 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Marek Szyprowski49f223a2012-01-25 12:49:24 +01005575
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07005576 __mod_zone_page_state(zone, NR_ALLOC_BATCH,
5577 high_wmark_pages(zone) -
5578 low_wmark_pages(zone) -
5579 zone_page_state(zone, NR_ALLOC_BATCH));
5580
Mel Gorman56fd56b2007-10-16 01:25:58 -07005581 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005582 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005583 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005584
5585 /* update totalreserve_pages */
5586 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005587}
5588
Mel Gormancfd3da12011-04-25 21:36:42 +00005589/**
5590 * setup_per_zone_wmarks - called when min_free_kbytes changes
5591 * or when memory is hot-{added|removed}
5592 *
5593 * Ensures that the watermark[min,low,high] values for each zone are set
5594 * correctly with respect to min_free_kbytes.
5595 */
5596void setup_per_zone_wmarks(void)
5597{
5598 mutex_lock(&zonelists_mutex);
5599 __setup_per_zone_wmarks();
5600 mutex_unlock(&zonelists_mutex);
5601}
5602
Randy Dunlap55a44622009-09-21 17:01:20 -07005603/*
Rik van Riel556adec2008-10-18 20:26:34 -07005604 * The inactive anon list should be small enough that the VM never has to
5605 * do too much work, but large enough that each inactive page has a chance
5606 * to be referenced again before it is swapped out.
5607 *
5608 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5609 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5610 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5611 * the anonymous pages are kept on the inactive list.
5612 *
5613 * total target max
5614 * memory ratio inactive anon
5615 * -------------------------------------
5616 * 10MB 1 5MB
5617 * 100MB 1 50MB
5618 * 1GB 3 250MB
5619 * 10GB 10 0.9GB
5620 * 100GB 31 3GB
5621 * 1TB 101 10GB
5622 * 10TB 320 32GB
5623 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005624static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
Minchan Kim96cb4df2009-06-16 15:32:49 -07005625{
5626 unsigned int gb, ratio;
5627
5628 /* Zone size in gigabytes */
Jiang Liub40da042013-02-22 16:33:52 -08005629 gb = zone->managed_pages >> (30 - PAGE_SHIFT);
Minchan Kim96cb4df2009-06-16 15:32:49 -07005630 if (gb)
5631 ratio = int_sqrt(10 * gb);
5632 else
5633 ratio = 1;
5634
5635 zone->inactive_ratio = ratio;
5636}
5637
KOSAKI Motohiro839a4fc2011-05-24 17:11:31 -07005638static void __meminit setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07005639{
5640 struct zone *zone;
5641
Minchan Kim96cb4df2009-06-16 15:32:49 -07005642 for_each_zone(zone)
5643 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07005644}
5645
Linus Torvalds1da177e2005-04-16 15:20:36 -07005646/*
5647 * Initialise min_free_kbytes.
5648 *
5649 * For small machines we want it small (128k min). For large machines
5650 * we want it large (64MB max). But it is not linear, because network
5651 * bandwidth does not increase linearly with machine size. We use
5652 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07005653 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
Linus Torvalds1da177e2005-04-16 15:20:36 -07005654 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5655 *
5656 * which yields
5657 *
5658 * 16MB: 512k
5659 * 32MB: 724k
5660 * 64MB: 1024k
5661 * 128MB: 1448k
5662 * 256MB: 2048k
5663 * 512MB: 2896k
5664 * 1024MB: 4096k
5665 * 2048MB: 5792k
5666 * 4096MB: 8192k
5667 * 8192MB: 11584k
5668 * 16384MB: 16384k
5669 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005670int __meminit init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005671{
5672 unsigned long lowmem_kbytes;
Michal Hocko5f127332013-07-08 16:00:40 -07005673 int new_min_free_kbytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005674
5675 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
Michal Hocko5f127332013-07-08 16:00:40 -07005676 new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005677
Michal Hocko5f127332013-07-08 16:00:40 -07005678 if (new_min_free_kbytes > user_min_free_kbytes) {
5679 min_free_kbytes = new_min_free_kbytes;
5680 if (min_free_kbytes < 128)
5681 min_free_kbytes = 128;
5682 if (min_free_kbytes > 65536)
5683 min_free_kbytes = 65536;
5684 } else {
5685 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
5686 new_min_free_kbytes, user_min_free_kbytes);
5687 }
Minchan Kimbc75d332009-06-16 15:32:48 -07005688 setup_per_zone_wmarks();
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -07005689 refresh_zone_stat_thresholds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005690 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07005691 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005692 return 0;
5693}
Minchan Kimbc75d332009-06-16 15:32:48 -07005694module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005695
5696/*
Pintu Kumarb8af2942013-09-11 14:20:34 -07005697 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
Linus Torvalds1da177e2005-04-16 15:20:36 -07005698 * that we can call two helper functions whenever min_free_kbytes
5699 * changes.
5700 */
Pintu Kumarb8af2942013-09-11 14:20:34 -07005701int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005702 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005703{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005704 proc_dointvec(table, write, buffer, length, ppos);
Michal Hocko5f127332013-07-08 16:00:40 -07005705 if (write) {
5706 user_min_free_kbytes = min_free_kbytes;
Minchan Kimbc75d332009-06-16 15:32:48 -07005707 setup_per_zone_wmarks();
Michal Hocko5f127332013-07-08 16:00:40 -07005708 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005709 return 0;
5710}
5711
Christoph Lameter96146342006-07-03 00:24:13 -07005712#ifdef CONFIG_NUMA
5713int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005714 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07005715{
5716 struct zone *zone;
5717 int rc;
5718
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005719 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07005720 if (rc)
5721 return rc;
5722
5723 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08005724 zone->min_unmapped_pages = (zone->managed_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07005725 sysctl_min_unmapped_ratio) / 100;
5726 return 0;
5727}
Christoph Lameter0ff38492006-09-25 23:31:52 -07005728
5729int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005730 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07005731{
5732 struct zone *zone;
5733 int rc;
5734
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005735 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07005736 if (rc)
5737 return rc;
5738
5739 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08005740 zone->min_slab_pages = (zone->managed_pages *
Christoph Lameter0ff38492006-09-25 23:31:52 -07005741 sysctl_min_slab_ratio) / 100;
5742 return 0;
5743}
Christoph Lameter96146342006-07-03 00:24:13 -07005744#endif
5745
Linus Torvalds1da177e2005-04-16 15:20:36 -07005746/*
5747 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5748 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5749 * whenever sysctl_lowmem_reserve_ratio changes.
5750 *
5751 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07005752 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07005753 * if in function of the boot time zone sizes.
5754 */
5755int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005756 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005757{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005758 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005759 setup_per_zone_lowmem_reserve();
5760 return 0;
5761}
5762
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005763/*
5764 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
Pintu Kumarb8af2942013-09-11 14:20:34 -07005765 * cpu. It is the fraction of total pages in each zone that a hot per cpu
5766 * pagelist can have before it gets flushed back to buddy allocator.
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005767 */
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005768int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005769 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005770{
5771 struct zone *zone;
5772 unsigned int cpu;
5773 int ret;
5774
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005775 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
Sasha Levin93278812012-05-10 13:01:44 -07005776 if (!write || (ret < 0))
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005777 return ret;
Cody P Schaferc8e251f2013-07-03 15:01:29 -07005778
5779 mutex_lock(&pcp_batch_high_lock);
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07005780 for_each_populated_zone(zone) {
Cody P Schafer22a7f122013-07-03 15:01:34 -07005781 unsigned long high;
5782 high = zone->managed_pages / percpu_pagelist_fraction;
5783 for_each_possible_cpu(cpu)
Cody P Schafer36640332013-07-03 15:01:40 -07005784 pageset_set_high(per_cpu_ptr(zone->pageset, cpu),
5785 high);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005786 }
Cody P Schaferc8e251f2013-07-03 15:01:29 -07005787 mutex_unlock(&pcp_batch_high_lock);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005788 return 0;
5789}
5790
David S. Millerf034b5d2006-08-24 03:08:07 -07005791int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005792
5793#ifdef CONFIG_NUMA
5794static int __init set_hashdist(char *str)
5795{
5796 if (!str)
5797 return 0;
5798 hashdist = simple_strtoul(str, &str, 0);
5799 return 1;
5800}
5801__setup("hashdist=", set_hashdist);
5802#endif
5803
5804/*
5805 * allocate a large system hash table from bootmem
5806 * - it is assumed that the hash table must contain an exact power-of-2
5807 * quantity of entries
5808 * - limit is the number of hash buckets, not the total allocation size
5809 */
5810void *__init alloc_large_system_hash(const char *tablename,
5811 unsigned long bucketsize,
5812 unsigned long numentries,
5813 int scale,
5814 int flags,
5815 unsigned int *_hash_shift,
5816 unsigned int *_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00005817 unsigned long low_limit,
5818 unsigned long high_limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005819{
Tim Bird31fe62b2012-05-23 13:33:35 +00005820 unsigned long long max = high_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005821 unsigned long log2qty, size;
5822 void *table = NULL;
5823
5824 /* allow the kernel cmdline to have a say */
5825 if (!numentries) {
5826 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08005827 numentries = nr_kernel_pages;
Jerry Zhoua7e83312013-09-11 14:20:26 -07005828
5829 /* It isn't necessary when PAGE_SIZE >= 1MB */
5830 if (PAGE_SHIFT < 20)
5831 numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005832
5833 /* limit to 1 bucket per 2^scale bytes of low memory */
5834 if (scale > PAGE_SHIFT)
5835 numentries >>= (scale - PAGE_SHIFT);
5836 else
5837 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08005838
5839 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07005840 if (unlikely(flags & HASH_SMALL)) {
5841 /* Makes no sense without HASH_EARLY */
5842 WARN_ON(!(flags & HASH_EARLY));
5843 if (!(numentries >> *_hash_shift)) {
5844 numentries = 1UL << *_hash_shift;
5845 BUG_ON(!numentries);
5846 }
5847 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08005848 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005849 }
John Hawkes6e692ed2006-03-25 03:08:02 -08005850 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005851
5852 /* limit allocation size to 1/16 total memory by default */
5853 if (max == 0) {
5854 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
5855 do_div(max, bucketsize);
5856 }
Dimitri Sivanich074b8512012-02-08 12:39:07 -08005857 max = min(max, 0x80000000ULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005858
Tim Bird31fe62b2012-05-23 13:33:35 +00005859 if (numentries < low_limit)
5860 numentries = low_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005861 if (numentries > max)
5862 numentries = max;
5863
David Howellsf0d1b0b2006-12-08 02:37:49 -08005864 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005865
5866 do {
5867 size = bucketsize << log2qty;
5868 if (flags & HASH_EARLY)
Jan Beulich74768ed2008-08-12 15:08:39 -07005869 table = alloc_bootmem_nopanic(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005870 else if (hashdist)
5871 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
5872 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07005873 /*
5874 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07005875 * some pages at the end of hash table which
5876 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07005877 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005878 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07005879 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005880 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
5881 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005882 }
5883 } while (!table && size > PAGE_SIZE && --log2qty);
5884
5885 if (!table)
5886 panic("Failed to allocate %s hash table\n", tablename);
5887
Robin Holtf241e6602010-10-07 12:59:26 -07005888 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07005889 tablename,
Robin Holtf241e6602010-10-07 12:59:26 -07005890 (1UL << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08005891 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005892 size);
5893
5894 if (_hash_shift)
5895 *_hash_shift = log2qty;
5896 if (_hash_mask)
5897 *_hash_mask = (1 << log2qty) - 1;
5898
5899 return table;
5900}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08005901
Mel Gorman835c1342007-10-16 01:25:47 -07005902/* Return a pointer to the bitmap storing bits affecting a block of pages */
5903static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
5904 unsigned long pfn)
5905{
5906#ifdef CONFIG_SPARSEMEM
5907 return __pfn_to_section(pfn)->pageblock_flags;
5908#else
5909 return zone->pageblock_flags;
5910#endif /* CONFIG_SPARSEMEM */
5911}
Andrew Morton6220ec72006-10-19 23:29:05 -07005912
Mel Gorman835c1342007-10-16 01:25:47 -07005913static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
5914{
5915#ifdef CONFIG_SPARSEMEM
5916 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07005917 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005918#else
Laura Abbottc060f942013-01-11 14:31:51 -08005919 pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
Mel Gormand9c23402007-10-16 01:26:01 -07005920 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005921#endif /* CONFIG_SPARSEMEM */
5922}
5923
5924/**
Mel Gormand9c23402007-10-16 01:26:01 -07005925 * 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 -07005926 * @page: The page within the block of interest
5927 * @start_bitidx: The first bit of interest to retrieve
5928 * @end_bitidx: The last bit of interest
5929 * returns pageblock_bits flags
5930 */
5931unsigned long get_pageblock_flags_group(struct page *page,
5932 int start_bitidx, int end_bitidx)
5933{
5934 struct zone *zone;
5935 unsigned long *bitmap;
5936 unsigned long pfn, bitidx;
5937 unsigned long flags = 0;
5938 unsigned long value = 1;
5939
5940 zone = page_zone(page);
5941 pfn = page_to_pfn(page);
5942 bitmap = get_pageblock_bitmap(zone, pfn);
5943 bitidx = pfn_to_bitidx(zone, pfn);
5944
5945 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5946 if (test_bit(bitidx + start_bitidx, bitmap))
5947 flags |= value;
5948
5949 return flags;
5950}
5951
5952/**
Mel Gormand9c23402007-10-16 01:26:01 -07005953 * 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 -07005954 * @page: The page within the block of interest
5955 * @start_bitidx: The first bit of interest
5956 * @end_bitidx: The last bit of interest
5957 * @flags: The flags to set
5958 */
5959void set_pageblock_flags_group(struct page *page, unsigned long flags,
5960 int start_bitidx, int end_bitidx)
5961{
5962 struct zone *zone;
5963 unsigned long *bitmap;
5964 unsigned long pfn, bitidx;
5965 unsigned long value = 1;
5966
5967 zone = page_zone(page);
5968 pfn = page_to_pfn(page);
5969 bitmap = get_pageblock_bitmap(zone, pfn);
5970 bitidx = pfn_to_bitidx(zone, pfn);
Cody P Schafer108bcc92013-02-22 16:35:23 -08005971 VM_BUG_ON(!zone_spans_pfn(zone, pfn));
Mel Gorman835c1342007-10-16 01:25:47 -07005972
5973 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5974 if (flags & value)
5975 __set_bit(bitidx + start_bitidx, bitmap);
5976 else
5977 __clear_bit(bitidx + start_bitidx, bitmap);
5978}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005979
5980/*
Minchan Kim80934512012-07-31 16:43:01 -07005981 * This function checks whether pageblock includes unmovable pages or not.
5982 * If @count is not zero, it is okay to include less @count unmovable pages
5983 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07005984 * PageLRU check without isolation or lru_lock could race so that
Minchan Kim80934512012-07-31 16:43:01 -07005985 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
5986 * expect this function should be exact.
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005987 */
Wen Congyangb023f462012-12-11 16:00:45 -08005988bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
5989 bool skip_hwpoisoned_pages)
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005990{
5991 unsigned long pfn, iter, found;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01005992 int mt;
5993
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005994 /*
5995 * For avoiding noise data, lru_add_drain_all() should be called
Minchan Kim80934512012-07-31 16:43:01 -07005996 * If ZONE_MOVABLE, the zone never contains unmovable pages
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005997 */
5998 if (zone_idx(zone) == ZONE_MOVABLE)
Minchan Kim80934512012-07-31 16:43:01 -07005999 return false;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01006000 mt = get_pageblock_migratetype(page);
6001 if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
Minchan Kim80934512012-07-31 16:43:01 -07006002 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006003
6004 pfn = page_to_pfn(page);
6005 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
6006 unsigned long check = pfn + iter;
6007
Namhyung Kim29723fc2011-02-25 14:44:25 -08006008 if (!pfn_valid_within(check))
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006009 continue;
Namhyung Kim29723fc2011-02-25 14:44:25 -08006010
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006011 page = pfn_to_page(check);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07006012
6013 /*
6014 * Hugepages are not in LRU lists, but they're movable.
6015 * We need not scan over tail pages bacause we don't
6016 * handle each tail page individually in migration.
6017 */
6018 if (PageHuge(page)) {
6019 iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
6020 continue;
6021 }
6022
Minchan Kim97d255c2012-07-31 16:42:59 -07006023 /*
6024 * We can't use page_count without pin a page
6025 * because another CPU can free compound page.
6026 * This check already skips compound tails of THP
6027 * because their page->_count is zero at all time.
6028 */
6029 if (!atomic_read(&page->_count)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006030 if (PageBuddy(page))
6031 iter += (1 << page_order(page)) - 1;
6032 continue;
6033 }
Minchan Kim97d255c2012-07-31 16:42:59 -07006034
Wen Congyangb023f462012-12-11 16:00:45 -08006035 /*
6036 * The HWPoisoned page may be not in buddy system, and
6037 * page_count() is not 0.
6038 */
6039 if (skip_hwpoisoned_pages && PageHWPoison(page))
6040 continue;
6041
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006042 if (!PageLRU(page))
6043 found++;
6044 /*
6045 * If there are RECLAIMABLE pages, we need to check it.
6046 * But now, memory offline itself doesn't call shrink_slab()
6047 * and it still to be fixed.
6048 */
6049 /*
6050 * If the page is not RAM, page_count()should be 0.
6051 * we don't need more check. This is an _used_ not-movable page.
6052 *
6053 * The problematic thing here is PG_reserved pages. PG_reserved
6054 * is set to both of a memory hole page and a _used_ kernel
6055 * page at boot.
6056 */
6057 if (found > count)
Minchan Kim80934512012-07-31 16:43:01 -07006058 return true;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006059 }
Minchan Kim80934512012-07-31 16:43:01 -07006060 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006061}
6062
6063bool is_pageblock_removable_nolock(struct page *page)
6064{
Michal Hocko656a0702012-01-20 14:33:58 -08006065 struct zone *zone;
6066 unsigned long pfn;
Michal Hocko687875f2012-01-20 14:33:55 -08006067
6068 /*
6069 * We have to be careful here because we are iterating over memory
6070 * sections which are not zone aware so we might end up outside of
6071 * the zone but still within the section.
Michal Hocko656a0702012-01-20 14:33:58 -08006072 * We have to take care about the node as well. If the node is offline
6073 * its NODE_DATA will be NULL - see page_zone.
Michal Hocko687875f2012-01-20 14:33:55 -08006074 */
Michal Hocko656a0702012-01-20 14:33:58 -08006075 if (!node_online(page_to_nid(page)))
6076 return false;
6077
6078 zone = page_zone(page);
6079 pfn = page_to_pfn(page);
Cody P Schafer108bcc92013-02-22 16:35:23 -08006080 if (!zone_spans_pfn(zone, pfn))
Michal Hocko687875f2012-01-20 14:33:55 -08006081 return false;
6082
Wen Congyangb023f462012-12-11 16:00:45 -08006083 return !has_unmovable_pages(zone, page, 0, true);
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07006084}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006085
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006086#ifdef CONFIG_CMA
6087
6088static unsigned long pfn_max_align_down(unsigned long pfn)
6089{
6090 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
6091 pageblock_nr_pages) - 1);
6092}
6093
6094static unsigned long pfn_max_align_up(unsigned long pfn)
6095{
6096 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
6097 pageblock_nr_pages));
6098}
6099
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006100/* [start, end) must belong to a single zone. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006101static int __alloc_contig_migrate_range(struct compact_control *cc,
6102 unsigned long start, unsigned long end)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006103{
6104 /* This function is based on compact_zone() from compaction.c. */
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006105 unsigned long nr_reclaimed;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006106 unsigned long pfn = start;
6107 unsigned int tries = 0;
6108 int ret = 0;
6109
Marek Szyprowskibe49a6e2012-12-12 13:51:19 -08006110 migrate_prep();
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006111
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006112 while (pfn < end || !list_empty(&cc->migratepages)) {
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006113 if (fatal_signal_pending(current)) {
6114 ret = -EINTR;
6115 break;
6116 }
6117
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006118 if (list_empty(&cc->migratepages)) {
6119 cc->nr_migratepages = 0;
6120 pfn = isolate_migratepages_range(cc->zone, cc,
Minchan Kime46a2872012-10-08 16:33:48 -07006121 pfn, end, true);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006122 if (!pfn) {
6123 ret = -EINTR;
6124 break;
6125 }
6126 tries = 0;
6127 } else if (++tries == 5) {
6128 ret = ret < 0 ? ret : -EBUSY;
6129 break;
6130 }
6131
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006132 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
6133 &cc->migratepages);
6134 cc->nr_migratepages -= nr_reclaimed;
Minchan Kim02c6de82012-10-08 16:31:55 -07006135
Hugh Dickins9c620e22013-02-22 16:35:14 -08006136 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
6137 0, MIGRATE_SYNC, MR_CMA);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006138 }
Srinivas Pandruvada2a6f5122013-02-22 16:32:09 -08006139 if (ret < 0) {
6140 putback_movable_pages(&cc->migratepages);
6141 return ret;
6142 }
6143 return 0;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006144}
6145
6146/**
6147 * alloc_contig_range() -- tries to allocate given range of pages
6148 * @start: start PFN to allocate
6149 * @end: one-past-the-last PFN to allocate
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006150 * @migratetype: migratetype of the underlaying pageblocks (either
6151 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6152 * in range must have the same migratetype and it must
6153 * be either of the two.
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006154 *
6155 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6156 * aligned, however it's the caller's responsibility to guarantee that
6157 * we are the only thread that changes migrate type of pageblocks the
6158 * pages fall in.
6159 *
6160 * The PFN range must belong to a single zone.
6161 *
6162 * Returns zero on success or negative error code. On success all
6163 * pages which PFN is in [start, end) are allocated for the caller and
6164 * need to be freed with free_contig_range().
6165 */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006166int alloc_contig_range(unsigned long start, unsigned long end,
6167 unsigned migratetype)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006168{
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006169 unsigned long outer_start, outer_end;
6170 int ret = 0, order;
6171
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006172 struct compact_control cc = {
6173 .nr_migratepages = 0,
6174 .order = -1,
6175 .zone = page_zone(pfn_to_page(start)),
6176 .sync = true,
6177 .ignore_skip_hint = true,
6178 };
6179 INIT_LIST_HEAD(&cc.migratepages);
6180
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006181 /*
6182 * What we do here is we mark all pageblocks in range as
6183 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6184 * have different sizes, and due to the way page allocator
6185 * work, we align the range to biggest of the two pages so
6186 * that page allocator won't try to merge buddies from
6187 * different pageblocks and change MIGRATE_ISOLATE to some
6188 * other migration type.
6189 *
6190 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6191 * migrate the pages from an unaligned range (ie. pages that
6192 * we are interested in). This will put all the pages in
6193 * range back to page allocator as MIGRATE_ISOLATE.
6194 *
6195 * When this is done, we take the pages in range from page
6196 * allocator removing them from the buddy system. This way
6197 * page allocator will never consider using them.
6198 *
6199 * This lets us mark the pageblocks back as
6200 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6201 * aligned range but not in the unaligned, original range are
6202 * put back to page allocator so that buddy can use them.
6203 */
6204
6205 ret = start_isolate_page_range(pfn_max_align_down(start),
Wen Congyangb023f462012-12-11 16:00:45 -08006206 pfn_max_align_up(end), migratetype,
6207 false);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006208 if (ret)
Bob Liu86a595f2012-10-25 13:37:56 -07006209 return ret;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006210
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006211 ret = __alloc_contig_migrate_range(&cc, start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006212 if (ret)
6213 goto done;
6214
6215 /*
6216 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6217 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
6218 * more, all pages in [start, end) are free in page allocator.
6219 * What we are going to do is to allocate all pages from
6220 * [start, end) (that is remove them from page allocator).
6221 *
6222 * The only problem is that pages at the beginning and at the
6223 * end of interesting range may be not aligned with pages that
6224 * page allocator holds, ie. they can be part of higher order
6225 * pages. Because of this, we reserve the bigger range and
6226 * once this is done free the pages we are not interested in.
6227 *
6228 * We don't have to hold zone->lock here because the pages are
6229 * isolated thus they won't get removed from buddy.
6230 */
6231
6232 lru_add_drain_all();
6233 drain_all_pages();
6234
6235 order = 0;
6236 outer_start = start;
6237 while (!PageBuddy(pfn_to_page(outer_start))) {
6238 if (++order >= MAX_ORDER) {
6239 ret = -EBUSY;
6240 goto done;
6241 }
6242 outer_start &= ~0UL << order;
6243 }
6244
6245 /* Make sure the range is really isolated. */
Wen Congyangb023f462012-12-11 16:00:45 -08006246 if (test_pages_isolated(outer_start, end, false)) {
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006247 pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n",
6248 outer_start, end);
6249 ret = -EBUSY;
6250 goto done;
6251 }
6252
Marek Szyprowski49f223a2012-01-25 12:49:24 +01006253
6254 /* Grab isolated pages from freelists. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006255 outer_end = isolate_freepages_range(&cc, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006256 if (!outer_end) {
6257 ret = -EBUSY;
6258 goto done;
6259 }
6260
6261 /* Free head and tail (if any) */
6262 if (start != outer_start)
6263 free_contig_range(outer_start, start - outer_start);
6264 if (end != outer_end)
6265 free_contig_range(end, outer_end - end);
6266
6267done:
6268 undo_isolate_page_range(pfn_max_align_down(start),
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006269 pfn_max_align_up(end), migratetype);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006270 return ret;
6271}
6272
6273void free_contig_range(unsigned long pfn, unsigned nr_pages)
6274{
Marek Szyprowskibcc2b022012-12-20 15:05:18 -08006275 unsigned int count = 0;
6276
6277 for (; nr_pages--; pfn++) {
6278 struct page *page = pfn_to_page(pfn);
6279
6280 count += page_count(page) != 1;
6281 __free_page(page);
6282 }
6283 WARN(count != 0, "%d pages are still in use!\n", count);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006284}
6285#endif
6286
Jiang Liu4ed7e022012-07-31 16:43:35 -07006287#ifdef CONFIG_MEMORY_HOTPLUG
Cody P Schafer0a647f32013-07-03 15:01:33 -07006288/*
6289 * The zone indicated has a new number of managed_pages; batch sizes and percpu
6290 * page high values need to be recalulated.
6291 */
Jiang Liu4ed7e022012-07-31 16:43:35 -07006292void __meminit zone_pcp_update(struct zone *zone)
6293{
Cody P Schafer0a647f32013-07-03 15:01:33 -07006294 unsigned cpu;
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006295 mutex_lock(&pcp_batch_high_lock);
Cody P Schafer0a647f32013-07-03 15:01:33 -07006296 for_each_possible_cpu(cpu)
Cody P Schafer169f6c12013-07-03 15:01:41 -07006297 pageset_set_high_and_batch(zone,
6298 per_cpu_ptr(zone->pageset, cpu));
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006299 mutex_unlock(&pcp_batch_high_lock);
Jiang Liu4ed7e022012-07-31 16:43:35 -07006300}
6301#endif
6302
Jiang Liu340175b2012-07-31 16:43:32 -07006303void zone_pcp_reset(struct zone *zone)
6304{
6305 unsigned long flags;
Minchan Kim5a883812012-10-08 16:33:39 -07006306 int cpu;
6307 struct per_cpu_pageset *pset;
Jiang Liu340175b2012-07-31 16:43:32 -07006308
6309 /* avoid races with drain_pages() */
6310 local_irq_save(flags);
6311 if (zone->pageset != &boot_pageset) {
Minchan Kim5a883812012-10-08 16:33:39 -07006312 for_each_online_cpu(cpu) {
6313 pset = per_cpu_ptr(zone->pageset, cpu);
6314 drain_zonestat(zone, pset);
6315 }
Jiang Liu340175b2012-07-31 16:43:32 -07006316 free_percpu(zone->pageset);
6317 zone->pageset = &boot_pageset;
6318 }
6319 local_irq_restore(flags);
6320}
6321
Wen Congyang6dcd73d2012-12-11 16:01:01 -08006322#ifdef CONFIG_MEMORY_HOTREMOVE
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006323/*
6324 * All pages in the range must be isolated before calling this.
6325 */
6326void
6327__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
6328{
6329 struct page *page;
6330 struct zone *zone;
6331 int order, i;
6332 unsigned long pfn;
6333 unsigned long flags;
6334 /* find the first valid pfn */
6335 for (pfn = start_pfn; pfn < end_pfn; pfn++)
6336 if (pfn_valid(pfn))
6337 break;
6338 if (pfn == end_pfn)
6339 return;
6340 zone = page_zone(pfn_to_page(pfn));
6341 spin_lock_irqsave(&zone->lock, flags);
6342 pfn = start_pfn;
6343 while (pfn < end_pfn) {
6344 if (!pfn_valid(pfn)) {
6345 pfn++;
6346 continue;
6347 }
6348 page = pfn_to_page(pfn);
Wen Congyangb023f462012-12-11 16:00:45 -08006349 /*
6350 * The HWPoisoned page may be not in buddy system, and
6351 * page_count() is not 0.
6352 */
6353 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
6354 pfn++;
6355 SetPageReserved(page);
6356 continue;
6357 }
6358
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006359 BUG_ON(page_count(page));
6360 BUG_ON(!PageBuddy(page));
6361 order = page_order(page);
6362#ifdef CONFIG_DEBUG_VM
6363 printk(KERN_INFO "remove from free list %lx %d %lx\n",
6364 pfn, 1 << order, end_pfn);
6365#endif
6366 list_del(&page->lru);
6367 rmv_page_order(page);
6368 zone->free_area[order].nr_free--;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006369 for (i = 0; i < (1 << order); i++)
6370 SetPageReserved((page+i));
6371 pfn += (1 << order);
6372 }
6373 spin_unlock_irqrestore(&zone->lock, flags);
6374}
6375#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01006376
6377#ifdef CONFIG_MEMORY_FAILURE
6378bool is_free_buddy_page(struct page *page)
6379{
6380 struct zone *zone = page_zone(page);
6381 unsigned long pfn = page_to_pfn(page);
6382 unsigned long flags;
6383 int order;
6384
6385 spin_lock_irqsave(&zone->lock, flags);
6386 for (order = 0; order < MAX_ORDER; order++) {
6387 struct page *page_head = page - (pfn & ((1 << order) - 1));
6388
6389 if (PageBuddy(page_head) && page_order(page_head) >= order)
6390 break;
6391 }
6392 spin_unlock_irqrestore(&zone->lock, flags);
6393
6394 return order < MAX_ORDER;
6395}
6396#endif
Wu Fengguang718a3822010-03-10 15:20:43 -08006397
Andrew Morton51300ce2012-05-29 15:06:44 -07006398static const struct trace_print_flags pageflag_names[] = {
Wu Fengguang718a3822010-03-10 15:20:43 -08006399 {1UL << PG_locked, "locked" },
6400 {1UL << PG_error, "error" },
6401 {1UL << PG_referenced, "referenced" },
6402 {1UL << PG_uptodate, "uptodate" },
6403 {1UL << PG_dirty, "dirty" },
6404 {1UL << PG_lru, "lru" },
6405 {1UL << PG_active, "active" },
6406 {1UL << PG_slab, "slab" },
6407 {1UL << PG_owner_priv_1, "owner_priv_1" },
6408 {1UL << PG_arch_1, "arch_1" },
6409 {1UL << PG_reserved, "reserved" },
6410 {1UL << PG_private, "private" },
6411 {1UL << PG_private_2, "private_2" },
6412 {1UL << PG_writeback, "writeback" },
6413#ifdef CONFIG_PAGEFLAGS_EXTENDED
6414 {1UL << PG_head, "head" },
6415 {1UL << PG_tail, "tail" },
6416#else
6417 {1UL << PG_compound, "compound" },
6418#endif
6419 {1UL << PG_swapcache, "swapcache" },
6420 {1UL << PG_mappedtodisk, "mappedtodisk" },
6421 {1UL << PG_reclaim, "reclaim" },
Wu Fengguang718a3822010-03-10 15:20:43 -08006422 {1UL << PG_swapbacked, "swapbacked" },
6423 {1UL << PG_unevictable, "unevictable" },
6424#ifdef CONFIG_MMU
6425 {1UL << PG_mlocked, "mlocked" },
6426#endif
6427#ifdef CONFIG_ARCH_USES_PG_UNCACHED
6428 {1UL << PG_uncached, "uncached" },
6429#endif
6430#ifdef CONFIG_MEMORY_FAILURE
6431 {1UL << PG_hwpoison, "hwpoison" },
6432#endif
Gavin Shanbe9cd872012-05-29 15:06:44 -07006433#ifdef CONFIG_TRANSPARENT_HUGEPAGE
6434 {1UL << PG_compound_lock, "compound_lock" },
6435#endif
Wu Fengguang718a3822010-03-10 15:20:43 -08006436};
6437
6438static void dump_page_flags(unsigned long flags)
6439{
6440 const char *delim = "";
6441 unsigned long mask;
6442 int i;
6443
Andrew Morton51300ce2012-05-29 15:06:44 -07006444 BUILD_BUG_ON(ARRAY_SIZE(pageflag_names) != __NR_PAGEFLAGS);
Johannes Weineracc50c12012-05-29 15:06:44 -07006445
Wu Fengguang718a3822010-03-10 15:20:43 -08006446 printk(KERN_ALERT "page flags: %#lx(", flags);
6447
6448 /* remove zone id */
6449 flags &= (1UL << NR_PAGEFLAGS) - 1;
6450
Andrew Morton51300ce2012-05-29 15:06:44 -07006451 for (i = 0; i < ARRAY_SIZE(pageflag_names) && flags; i++) {
Wu Fengguang718a3822010-03-10 15:20:43 -08006452
6453 mask = pageflag_names[i].mask;
6454 if ((flags & mask) != mask)
6455 continue;
6456
6457 flags &= ~mask;
6458 printk("%s%s", delim, pageflag_names[i].name);
6459 delim = "|";
6460 }
6461
6462 /* check for left over flags */
6463 if (flags)
6464 printk("%s%#lx", delim, flags);
6465
6466 printk(")\n");
6467}
6468
6469void dump_page(struct page *page)
6470{
6471 printk(KERN_ALERT
6472 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
Andrea Arcangeli4e9f64c2011-01-13 15:46:29 -08006473 page, atomic_read(&page->_count), page_mapcount(page),
Wu Fengguang718a3822010-03-10 15:20:43 -08006474 page->mapping, page->index);
6475 dump_page_flags(page->flags);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07006476 mem_cgroup_print_bad_page(page);
Wu Fengguang718a3822010-03-10 15:20:43 -08006477}