blob: 378a15bcd64940cd4bab29335604f757543839be [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 Hiroyukia5d76b542007-10-16 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>
Michal Nazarewicz041d3a82011-12-29 13:09:50 +010059#include <linux/migrate.h>
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -080060#include <linux/page-debug-flags.h>
David Rientjes949f7ec2013-04-29 15:07:48 -070061#include <linux/hugetlb.h>
Clark Williams8bd75c72013-02-07 09:47:07 -060062#include <linux/sched/rt.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070063
64#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070065#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070066#include "internal.h"
67
Lee Schermerhorn72812012010-05-26 14:44:56 -070068#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
69DEFINE_PER_CPU(int, numa_node);
70EXPORT_PER_CPU_SYMBOL(numa_node);
71#endif
72
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070073#ifdef CONFIG_HAVE_MEMORYLESS_NODES
74/*
75 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
76 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
77 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
78 * defined in <linux/topology.h>.
79 */
80DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
81EXPORT_PER_CPU_SYMBOL(_numa_mem_);
82#endif
83
Linus Torvalds1da177e2005-04-16 15:20:36 -070084/*
Christoph Lameter13808912007-10-16 01:25:27 -070085 * Array of node states.
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 */
Christoph Lameter13808912007-10-16 01:25:27 -070087nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
88 [N_POSSIBLE] = NODE_MASK_ALL,
89 [N_ONLINE] = { { [0] = 1UL } },
90#ifndef CONFIG_NUMA
91 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
92#ifdef CONFIG_HIGHMEM
93 [N_HIGH_MEMORY] = { { [0] = 1UL } },
94#endif
Lai Jiangshan20b2f522012-12-12 13:52:00 -080095#ifdef CONFIG_MOVABLE_NODE
96 [N_MEMORY] = { { [0] = 1UL } },
97#endif
Christoph Lameter13808912007-10-16 01:25:27 -070098 [N_CPU] = { { [0] = 1UL } },
99#endif /* NUMA */
100};
101EXPORT_SYMBOL(node_states);
102
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700103unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -0700104unsigned long totalreserve_pages __read_mostly;
Johannes Weinerab8fabd2012-01-10 15:07:42 -0800105/*
106 * When calculating the number of globally allowed dirty pages, there
107 * is a certain number of per-zone reserves that should not be
108 * considered dirtyable memory. This is the sum of those reserves
109 * over all existing zones that contribute dirtyable memory.
110 */
111unsigned long dirty_balance_reserve __read_mostly;
112
Hugh Dickins1b76b022012-05-11 01:00:07 -0700113int percpu_pagelist_fraction;
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000114gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800116#ifdef CONFIG_PM_SLEEP
117/*
118 * The following functions are used by the suspend/hibernate code to temporarily
119 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
120 * while devices are suspended. To avoid races with the suspend/hibernate code,
121 * they should always be called with pm_mutex held (gfp_allowed_mask also should
122 * only be modified with pm_mutex held, unless the suspend/hibernate code is
123 * guaranteed not to run in parallel with that modification).
124 */
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100125
126static gfp_t saved_gfp_mask;
127
128void pm_restore_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800129{
130 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100131 if (saved_gfp_mask) {
132 gfp_allowed_mask = saved_gfp_mask;
133 saved_gfp_mask = 0;
134 }
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800135}
136
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100137void pm_restrict_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800138{
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800139 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100140 WARN_ON(saved_gfp_mask);
141 saved_gfp_mask = gfp_allowed_mask;
142 gfp_allowed_mask &= ~GFP_IOFS;
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800143}
Mel Gormanf90ac392012-01-10 15:07:15 -0800144
145bool pm_suspended_storage(void)
146{
147 if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS)
148 return false;
149 return true;
150}
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800151#endif /* CONFIG_PM_SLEEP */
152
Mel Gormand9c23402007-10-16 01:26:01 -0700153#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
154int pageblock_order __read_mostly;
155#endif
156
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800157static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -0800158
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159/*
160 * results with 256, 32 in the lowmem_reserve sysctl:
161 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
162 * 1G machine -> (16M dma, 784M normal, 224M high)
163 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
164 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
165 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +0100166 *
167 * TBD: should special case ZONE_DMA32 machines here - in those we normally
168 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700170int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800171#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700172 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -0800173#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700174#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700175 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700176#endif
Christoph Lametere53ef382006-09-25 23:31:14 -0700177#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700178 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700179#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700180 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700181};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182
183EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184
Helge Deller15ad7cd2006-12-06 20:40:36 -0800185static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800186#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700187 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800188#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700189#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700190 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700191#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700192 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700193#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700194 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700195#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700196 "Movable",
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700197};
198
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199int min_free_kbytes = 1024;
200
Jan Beulich2c85f512009-09-21 17:03:07 -0700201static unsigned long __meminitdata nr_kernel_pages;
202static unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700203static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
Tejun Heo0ee332c2011-12-08 10:22:09 -0800205#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
206static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
207static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
208static unsigned long __initdata required_kernelcore;
209static unsigned long __initdata required_movablecore;
210static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Mel Gormanc7132162006-09-27 01:49:43 -0700211
Tejun Heo0ee332c2011-12-08 10:22:09 -0800212/* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
213int movable_zone;
214EXPORT_SYMBOL(movable_zone);
215#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -0700216
Miklos Szeredi418508c2007-05-23 13:57:55 -0700217#if MAX_NUMNODES > 1
218int nr_node_ids __read_mostly = MAX_NUMNODES;
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700219int nr_online_nodes __read_mostly = 1;
Miklos Szeredi418508c2007-05-23 13:57:55 -0700220EXPORT_SYMBOL(nr_node_ids);
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700221EXPORT_SYMBOL(nr_online_nodes);
Miklos Szeredi418508c2007-05-23 13:57:55 -0700222#endif
223
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700224int page_group_by_mobility_disabled __read_mostly;
225
Minchan Kimee6f5092012-07-31 16:43:50 -0700226void set_pageblock_migratetype(struct page *page, int migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700227{
Mel Gorman49255c62009-06-16 15:31:58 -0700228
229 if (unlikely(page_group_by_mobility_disabled))
230 migratetype = MIGRATE_UNMOVABLE;
231
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700232 set_pageblock_flags_group(page, (unsigned long)migratetype,
233 PB_migrate, PB_migrate_end);
234}
235
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -0700236bool oom_killer_disabled __read_mostly;
237
Nick Piggin13e74442006-01-06 00:10:58 -0800238#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700239static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700241 int ret = 0;
242 unsigned seq;
243 unsigned long pfn = page_to_pfn(page);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800244 unsigned long sp, start_pfn;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700245
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700246 do {
247 seq = zone_span_seqbegin(zone);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800248 start_pfn = zone->zone_start_pfn;
249 sp = zone->spanned_pages;
Cody P Schafer108bcc92013-02-22 16:35:23 -0800250 if (!zone_spans_pfn(zone, pfn))
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700251 ret = 1;
252 } while (zone_span_seqretry(zone, seq));
253
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800254 if (ret)
255 pr_err("page %lu outside zone [ %lu - %lu ]\n",
256 pfn, start_pfn, start_pfn + sp);
257
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700258 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700259}
260
261static int page_is_consistent(struct zone *zone, struct page *page)
262{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700263 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700264 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700266 return 0;
267
268 return 1;
269}
270/*
271 * Temporary debugging check for pages not lying within a given zone.
272 */
273static int bad_range(struct zone *zone, struct page *page)
274{
275 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700277 if (!page_is_consistent(zone, page))
278 return 1;
279
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 return 0;
281}
Nick Piggin13e74442006-01-06 00:10:58 -0800282#else
283static inline int bad_range(struct zone *zone, struct page *page)
284{
285 return 0;
286}
287#endif
288
Nick Piggin224abf92006-01-06 00:11:11 -0800289static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800291 static unsigned long resume;
292 static unsigned long nr_shown;
293 static unsigned long nr_unshown;
294
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200295 /* Don't complain about poisoned pages */
296 if (PageHWPoison(page)) {
Mel Gorman22b751c2013-02-22 16:34:59 -0800297 page_mapcount_reset(page); /* remove PageBuddy */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200298 return;
299 }
300
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800301 /*
302 * Allow a burst of 60 reports, then keep quiet for that minute;
303 * or allow a steady drip of one report per second.
304 */
305 if (nr_shown == 60) {
306 if (time_before(jiffies, resume)) {
307 nr_unshown++;
308 goto out;
309 }
310 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800311 printk(KERN_ALERT
312 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800313 nr_unshown);
314 nr_unshown = 0;
315 }
316 nr_shown = 0;
317 }
318 if (nr_shown++ == 0)
319 resume = jiffies + 60 * HZ;
320
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800321 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800322 current->comm, page_to_pfn(page));
Wu Fengguang718a3822010-03-10 15:20:43 -0800323 dump_page(page);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700324
Dave Jones4f318882011-10-31 17:07:24 -0700325 print_modules();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800327out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800328 /* Leave bad fields for debug, except PageBuddy could make trouble */
Mel Gorman22b751c2013-02-22 16:34:59 -0800329 page_mapcount_reset(page); /* remove PageBuddy */
Rusty Russell373d4d02013-01-21 17:17:39 +1030330 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331}
332
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333/*
334 * Higher-order pages are called "compound pages". They are structured thusly:
335 *
336 * The first PAGE_SIZE page is called the "head page".
337 *
338 * The remaining PAGE_SIZE pages are called "tail pages".
339 *
Wang Sheng-Hui6416b9f2011-11-17 10:53:50 +0100340 * All pages have PG_compound set. All tail pages have their ->first_page
341 * pointing at the head page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800343 * The first tail page's ->lru.next holds the address of the compound page's
344 * put_page() function. Its ->lru.prev holds the order of allocation.
345 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800347
348static void free_compound_page(struct page *page)
349{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700350 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800351}
352
Andi Kleen01ad1c02008-07-23 21:27:46 -0700353void prep_compound_page(struct page *page, unsigned long order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354{
355 int i;
356 int nr_pages = 1 << order;
357
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800358 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700359 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700360 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800361 for (i = 1; i < nr_pages; i++) {
362 struct page *p = page + i;
Christoph Lameterd85f3382007-05-06 14:49:39 -0700363 __SetPageTail(p);
Youquan Song58a84aa2011-12-08 14:34:18 -0800364 set_page_count(p, 0);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700365 p->first_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 }
367}
368
Andrea Arcangeli59ff4212011-01-13 15:46:38 -0800369/* update __split_huge_page_refcount if you change this function */
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800370static int destroy_compound_page(struct page *page, unsigned long order)
Andy Whitcroft18229df2008-11-06 12:53:27 -0800371{
372 int i;
373 int nr_pages = 1 << order;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800374 int bad = 0;
Andy Whitcroft18229df2008-11-06 12:53:27 -0800375
Gavin Shan0bb2c762012-12-18 14:21:32 -0800376 if (unlikely(compound_order(page) != order)) {
Nick Piggin224abf92006-01-06 00:11:11 -0800377 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800378 bad++;
379 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380
Christoph Lameter6d777952007-05-06 14:49:40 -0700381 __ClearPageHead(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800382
Andy Whitcroft18229df2008-11-06 12:53:27 -0800383 for (i = 1; i < nr_pages; i++) {
384 struct page *p = page + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385
Alexey Zaytseve713a212009-01-10 02:47:57 +0300386 if (unlikely(!PageTail(p) || (p->first_page != page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800387 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800388 bad++;
389 }
Christoph Lameterd85f3382007-05-06 14:49:39 -0700390 __ClearPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800392
393 return bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395
Nick Piggin17cf4402006-03-22 00:08:41 -0800396static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
397{
398 int i;
399
Andrew Morton6626c5d2006-03-22 00:08:42 -0800400 /*
401 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
402 * and __GFP_HIGHMEM from hard or soft interrupt context.
403 */
Nick Piggin725d7042006-09-25 23:30:55 -0700404 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800405 for (i = 0; i < (1 << order); i++)
406 clear_highpage(page + i);
407}
408
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800409#ifdef CONFIG_DEBUG_PAGEALLOC
410unsigned int _debug_guardpage_minorder;
411
412static int __init debug_guardpage_minorder_setup(char *buf)
413{
414 unsigned long res;
415
416 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
417 printk(KERN_ERR "Bad debug_guardpage_minorder value\n");
418 return 0;
419 }
420 _debug_guardpage_minorder = res;
421 printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res);
422 return 0;
423}
424__setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
425
426static inline void set_page_guard_flag(struct page *page)
427{
428 __set_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags);
429}
430
431static inline void clear_page_guard_flag(struct page *page)
432{
433 __clear_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags);
434}
435#else
436static inline void set_page_guard_flag(struct page *page) { }
437static inline void clear_page_guard_flag(struct page *page) { }
438#endif
439
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700440static inline void set_page_order(struct page *page, int order)
441{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700442 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000443 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444}
445
446static inline void rmv_page_order(struct page *page)
447{
Nick Piggin676165a2006-04-10 11:21:48 +1000448 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700449 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450}
451
452/*
453 * Locate the struct page for both the matching buddy in our
454 * pair (buddy1) and the combined O(n+1) page they form (page).
455 *
456 * 1) Any buddy B1 will have an order O twin B2 which satisfies
457 * the following equation:
458 * B2 = B1 ^ (1 << O)
459 * For example, if the starting buddy (buddy2) is #8 its order
460 * 1 buddy is #10:
461 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
462 *
463 * 2) Any buddy B will have an order O+1 parent P which
464 * satisfies the following equation:
465 * P = B & ~(1 << O)
466 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200467 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469static inline unsigned long
KyongHo Cho43506fa2011-01-13 15:47:24 -0800470__find_buddy_index(unsigned long page_idx, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471{
KyongHo Cho43506fa2011-01-13 15:47:24 -0800472 return page_idx ^ (1 << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473}
474
475/*
476 * This function checks whether a page is free && is the buddy
477 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800478 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000479 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700480 * (c) a page and its buddy have the same order &&
481 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 *
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800483 * For recording whether a page is in the buddy system, we set ->_mapcount -2.
484 * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock.
Nick Piggin676165a2006-04-10 11:21:48 +1000485 *
486 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700488static inline int page_is_buddy(struct page *page, struct page *buddy,
489 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700491 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800492 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800493
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700494 if (page_zone_id(page) != page_zone_id(buddy))
495 return 0;
496
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800497 if (page_is_guard(buddy) && page_order(buddy) == order) {
498 VM_BUG_ON(page_count(buddy) != 0);
499 return 1;
500 }
501
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700502 if (PageBuddy(buddy) && page_order(buddy) == order) {
Nick Piggina3af9c32009-06-16 15:32:10 -0700503 VM_BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700504 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000505 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700506 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507}
508
509/*
510 * Freeing function for a buddy system allocator.
511 *
512 * The concept of a buddy system is to maintain direct-mapped table
513 * (containing bit values) for memory blocks of various "orders".
514 * The bottom level table contains the map for the smallest allocatable
515 * units of memory (here, pages), and each level above it describes
516 * pairs of units from the levels below, hence, "buddies".
517 * At a high level, all that happens here is marking the table entry
518 * at the bottom level available, and propagating the changes upward
519 * as necessary, plus some accounting needed to play nicely with other
520 * parts of the VM system.
521 * At each level, we keep a list of pages, which are heads of continuous
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800522 * free pages of length of (1 << order) and marked with _mapcount -2. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700523 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 * So when we are allocating or freeing one, we can derive the state of the
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100525 * other. That is, if we allocate a small block, and both were
526 * free, the remainder of the region must be split into blocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 * If a block is freed, and its buddy is also free, then this
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100528 * triggers coalescing into a block of larger size.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +0100530 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 */
532
Nick Piggin48db57f2006-01-08 01:00:42 -0800533static inline void __free_one_page(struct page *page,
Mel Gormaned0ae212009-06-16 15:32:07 -0700534 struct zone *zone, unsigned int order,
535 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536{
537 unsigned long page_idx;
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700538 unsigned long combined_idx;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800539 unsigned long uninitialized_var(buddy_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700540 struct page *buddy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541
Cody P Schaferd29bb972013-02-22 16:35:25 -0800542 VM_BUG_ON(!zone_is_initialized(zone));
543
Nick Piggin224abf92006-01-06 00:11:11 -0800544 if (unlikely(PageCompound(page)))
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800545 if (unlikely(destroy_compound_page(page, order)))
546 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547
Mel Gormaned0ae212009-06-16 15:32:07 -0700548 VM_BUG_ON(migratetype == -1);
549
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
551
Mel Gormanf2260e62009-06-16 15:32:13 -0700552 VM_BUG_ON(page_idx & ((1 << order) - 1));
Nick Piggin725d7042006-09-25 23:30:55 -0700553 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 while (order < MAX_ORDER-1) {
KyongHo Cho43506fa2011-01-13 15:47:24 -0800556 buddy_idx = __find_buddy_index(page_idx, order);
557 buddy = page + (buddy_idx - page_idx);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700558 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700559 break;
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800560 /*
561 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
562 * merge with it and move up one order.
563 */
564 if (page_is_guard(buddy)) {
565 clear_page_guard_flag(buddy);
566 set_page_private(page, 0);
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -0700567 __mod_zone_freepage_state(zone, 1 << order,
568 migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800569 } else {
570 list_del(&buddy->lru);
571 zone->free_area[order].nr_free--;
572 rmv_page_order(buddy);
573 }
KyongHo Cho43506fa2011-01-13 15:47:24 -0800574 combined_idx = buddy_idx & page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 page = page + (combined_idx - page_idx);
576 page_idx = combined_idx;
577 order++;
578 }
579 set_page_order(page, order);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700580
581 /*
582 * If this is not the largest possible page, check if the buddy
583 * of the next-highest order is free. If it is, it's possible
584 * that pages are being freed that will coalesce soon. In case,
585 * that is happening, add the free page to the tail of the list
586 * so it's less likely to be used soon and more likely to be merged
587 * as a higher order page
588 */
Mel Gormanb7f50cf2010-10-26 14:21:11 -0700589 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700590 struct page *higher_page, *higher_buddy;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800591 combined_idx = buddy_idx & page_idx;
592 higher_page = page + (combined_idx - page_idx);
593 buddy_idx = __find_buddy_index(combined_idx, order + 1);
Li Haifeng0ba8f2d2012-09-17 14:09:21 -0700594 higher_buddy = higher_page + (buddy_idx - combined_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700595 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
596 list_add_tail(&page->lru,
597 &zone->free_area[order].free_list[migratetype]);
598 goto out;
599 }
600 }
601
602 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
603out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 zone->free_area[order].nr_free++;
605}
606
Nick Piggin224abf92006-01-06 00:11:11 -0800607static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608{
Nick Piggin92be2e332006-01-06 00:10:57 -0800609 if (unlikely(page_mapcount(page) |
610 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700611 (atomic_read(&page->_count) != 0) |
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -0700612 (page->flags & PAGE_FLAGS_CHECK_AT_FREE) |
613 (mem_cgroup_bad_page_check(page)))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800614 bad_page(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800615 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800616 }
Mel Gorman22b751c2013-02-22 16:34:59 -0800617 page_nid_reset_last(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800618 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
619 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
620 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621}
622
623/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700624 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700626 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 *
628 * If the zone was previously in an "all pages pinned" state then look to
629 * see if this freeing clears that state.
630 *
631 * And clear the zone's pages_scanned counter, to hold off the "all pages are
632 * pinned" detection logic.
633 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700634static void free_pcppages_bulk(struct zone *zone, int count,
635 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636{
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700637 int migratetype = 0;
Mel Gormana6f9edd62009-09-21 17:03:20 -0700638 int batch_free = 0;
Mel Gorman72853e22010-09-09 16:38:16 -0700639 int to_free = count;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700640
Nick Pigginc54ad302006-01-06 00:10:56 -0800641 spin_lock(&zone->lock);
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -0800642 zone->all_unreclaimable = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700644
Mel Gorman72853e22010-09-09 16:38:16 -0700645 while (to_free) {
Nick Piggin48db57f2006-01-08 01:00:42 -0800646 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700647 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -0800648
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700649 /*
Mel Gormana6f9edd62009-09-21 17:03:20 -0700650 * Remove pages from lists in a round-robin fashion. A
651 * batch_free count is maintained that is incremented when an
652 * empty list is encountered. This is so more pages are freed
653 * off fuller lists instead of spinning excessively around empty
654 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700655 */
656 do {
Mel Gormana6f9edd62009-09-21 17:03:20 -0700657 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700658 if (++migratetype == MIGRATE_PCPTYPES)
659 migratetype = 0;
660 list = &pcp->lists[migratetype];
661 } while (list_empty(list));
662
Namhyung Kim1d168712011-03-22 16:32:45 -0700663 /* This is the only non-empty list. Free them all. */
664 if (batch_free == MIGRATE_PCPTYPES)
665 batch_free = to_free;
666
Mel Gormana6f9edd62009-09-21 17:03:20 -0700667 do {
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -0700668 int mt; /* migratetype of the to-be-freed page */
669
Mel Gormana6f9edd62009-09-21 17:03:20 -0700670 page = list_entry(list->prev, struct page, lru);
671 /* must delete as __free_one_page list manipulates */
672 list_del(&page->lru);
Minchan Kimb12c4ad2012-10-08 16:32:08 -0700673 mt = get_freepage_migratetype(page);
Hugh Dickinsa7016232010-01-29 17:46:34 +0000674 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -0700675 __free_one_page(page, zone, 0, mt);
676 trace_mm_page_pcpu_drain(page, 0, mt);
Minchan Kim194159f2013-02-22 16:33:58 -0800677 if (likely(!is_migrate_isolate_page(page))) {
Wen Congyang97d0da22012-12-11 16:00:52 -0800678 __mod_zone_page_state(zone, NR_FREE_PAGES, 1);
679 if (is_migrate_cma(mt))
680 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, 1);
681 }
Mel Gorman72853e22010-09-09 16:38:16 -0700682 } while (--to_free && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800684 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685}
686
Mel Gormaned0ae212009-06-16 15:32:07 -0700687static void free_one_page(struct zone *zone, struct page *page, int order,
688 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800689{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700690 spin_lock(&zone->lock);
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -0800691 zone->all_unreclaimable = 0;
Christoph Lameter006d22d2006-09-25 23:31:48 -0700692 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700693
Mel Gormaned0ae212009-06-16 15:32:07 -0700694 __free_one_page(page, zone, order, migratetype);
Minchan Kim194159f2013-02-22 16:33:58 -0800695 if (unlikely(!is_migrate_isolate(migratetype)))
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -0700696 __mod_zone_freepage_state(zone, 1 << order, migratetype);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700697 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800698}
699
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700700static bool free_pages_prepare(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 int i;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800703 int bad = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -0800705 trace_mm_page_free(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100706 kmemcheck_free_shadow(page, order);
707
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -0800708 if (PageAnon(page))
709 page->mapping = NULL;
710 for (i = 0; i < (1 << order); i++)
711 bad += free_pages_check(page + i);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800712 if (bad)
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700713 return false;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800714
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700715 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -0700716 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700717 debug_check_no_obj_freed(page_address(page),
718 PAGE_SIZE << order);
719 }
Nick Piggindafb1362006-10-11 01:21:30 -0700720 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800721 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700722
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700723 return true;
724}
725
726static void __free_pages_ok(struct page *page, unsigned int order)
727{
728 unsigned long flags;
Minchan Kim95e34412012-10-08 16:32:11 -0700729 int migratetype;
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700730
731 if (!free_pages_prepare(page, order))
732 return;
733
Nick Pigginc54ad302006-01-06 00:10:56 -0800734 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700735 __count_vm_events(PGFREE, 1 << order);
Minchan Kim95e34412012-10-08 16:32:11 -0700736 migratetype = get_pageblock_migratetype(page);
737 set_freepage_migratetype(page, migratetype);
738 free_one_page(page_zone(page), page, order, migratetype);
Nick Pigginc54ad302006-01-06 00:10:56 -0800739 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740}
741
Jiang Liu9feedc92012-12-12 13:52:12 -0800742/*
743 * Read access to zone->managed_pages is safe because it's unsigned long,
744 * but we still need to serialize writers. Currently all callers of
745 * __free_pages_bootmem() except put_page_bootmem() should only be used
746 * at boot time. So for shorter boot time, we shift the burden to
747 * put_page_bootmem() to serialize writers.
748 */
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700749void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800750{
Johannes Weinerc3993072012-01-10 15:08:10 -0800751 unsigned int nr_pages = 1 << order;
752 unsigned int loop;
David Howellsa226f6c2006-01-06 00:11:08 -0800753
Johannes Weinerc3993072012-01-10 15:08:10 -0800754 prefetchw(page);
755 for (loop = 0; loop < nr_pages; loop++) {
756 struct page *p = &page[loop];
David Howellsa226f6c2006-01-06 00:11:08 -0800757
Johannes Weinerc3993072012-01-10 15:08:10 -0800758 if (loop + 1 < nr_pages)
759 prefetchw(p + 1);
760 __ClearPageReserved(p);
761 set_page_count(p, 0);
David Howellsa226f6c2006-01-06 00:11:08 -0800762 }
Johannes Weinerc3993072012-01-10 15:08:10 -0800763
Jiang Liu9feedc92012-12-12 13:52:12 -0800764 page_zone(page)->managed_pages += 1 << order;
Johannes Weinerc3993072012-01-10 15:08:10 -0800765 set_page_refcounted(page);
766 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800767}
768
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100769#ifdef CONFIG_CMA
770/* Free whole pageblock and set it's migration type to MIGRATE_CMA. */
771void __init init_cma_reserved_pageblock(struct page *page)
772{
773 unsigned i = pageblock_nr_pages;
774 struct page *p = page;
775
776 do {
777 __ClearPageReserved(p);
778 set_page_count(p, 0);
779 } while (++p, --i);
780
781 set_page_refcounted(page);
782 set_pageblock_migratetype(page, MIGRATE_CMA);
783 __free_pages(page, pageblock_order);
784 totalram_pages += pageblock_nr_pages;
Marek Szyprowski41a79732013-02-12 13:46:24 -0800785#ifdef CONFIG_HIGHMEM
786 if (PageHighMem(page))
787 totalhigh_pages += pageblock_nr_pages;
788#endif
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 Gruszkac0a32fc2012-01-10 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 Gruszkac0a32fc2012-01-10 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;
891 struct free_area * area;
892 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
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001013/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -07001014static inline struct page *
1015__rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001016{
1017 struct free_area * area;
1018 int current_order;
1019 struct page *page;
1020 int migratetype, i;
1021
1022 /* Find the largest possible block of pages in the other list */
1023 for (current_order = MAX_ORDER-1; current_order >= order;
1024 --current_order) {
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001025 for (i = 0;; i++) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001026 migratetype = fallbacks[start_migratetype][i];
1027
Mel Gorman56fd56b2007-10-16 01:25:58 -07001028 /* MIGRATE_RESERVE handled later if necessary */
1029 if (migratetype == MIGRATE_RESERVE)
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001030 break;
Mel Gormane0104872007-10-16 01:25:53 -07001031
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001032 area = &(zone->free_area[current_order]);
1033 if (list_empty(&area->free_list[migratetype]))
1034 continue;
1035
1036 page = list_entry(area->free_list[migratetype].next,
1037 struct page, lru);
1038 area->nr_free--;
1039
1040 /*
Mel Gormanc361be52007-10-16 01:25:51 -07001041 * If breaking a large block of pages, move all free
Mel Gorman46dafbc2007-10-16 01:25:55 -07001042 * pages to the preferred allocation list. If falling
1043 * back for a reclaimable kernel allocation, be more
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001044 * aggressive about taking ownership of free pages
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001045 *
1046 * On the other hand, never change migration
1047 * type of MIGRATE_CMA pageblocks nor move CMA
1048 * pages on different free lists. We don't
1049 * want unmovable pages to be allocated from
1050 * MIGRATE_CMA areas.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001051 */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001052 if (!is_migrate_cma(migratetype) &&
1053 (unlikely(current_order >= pageblock_order / 2) ||
1054 start_migratetype == MIGRATE_RECLAIMABLE ||
1055 page_group_by_mobility_disabled)) {
1056 int pages;
Mel Gorman46dafbc2007-10-16 01:25:55 -07001057 pages = move_freepages_block(zone, page,
1058 start_migratetype);
1059
1060 /* Claim the whole block if over half of it is free */
Mel Gormandd5d2412009-09-05 11:17:11 -07001061 if (pages >= (1 << (pageblock_order-1)) ||
1062 page_group_by_mobility_disabled)
Mel Gorman46dafbc2007-10-16 01:25:55 -07001063 set_pageblock_migratetype(page,
1064 start_migratetype);
1065
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001066 migratetype = start_migratetype;
Mel Gormanc361be52007-10-16 01:25:51 -07001067 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001068
1069 /* Remove the page from the freelists */
1070 list_del(&page->lru);
1071 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001072
Mel Gorman2f66a682009-09-21 17:02:31 -07001073 /* Take ownership for orders >= pageblock_order */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001074 if (current_order >= pageblock_order &&
1075 !is_migrate_cma(migratetype))
Mel Gorman2f66a682009-09-21 17:02:31 -07001076 change_pageblock_range(page, current_order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001077 start_migratetype);
1078
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001079 expand(zone, page, order, current_order, area,
1080 is_migrate_cma(migratetype)
1081 ? migratetype : start_migratetype);
Mel Gormane0fff1b2009-09-21 17:02:42 -07001082
1083 trace_mm_page_alloc_extfrag(page, order, current_order,
1084 start_migratetype, migratetype);
1085
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001086 return page;
1087 }
1088 }
1089
Mel Gorman728ec982009-06-16 15:32:04 -07001090 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001091}
1092
Mel Gorman56fd56b2007-10-16 01:25:58 -07001093/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 * Do the hard work of removing an element from the buddy allocator.
1095 * Call me with the zone->lock already held.
1096 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001097static struct page *__rmqueue(struct zone *zone, unsigned int order,
1098 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 struct page *page;
1101
Mel Gorman728ec982009-06-16 15:32:04 -07001102retry_reserve:
Mel Gorman56fd56b2007-10-16 01:25:58 -07001103 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104
Mel Gorman728ec982009-06-16 15:32:04 -07001105 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07001106 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001107
Mel Gorman728ec982009-06-16 15:32:04 -07001108 /*
1109 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1110 * is used because __rmqueue_smallest is an inline function
1111 * and we want just one call site
1112 */
1113 if (!page) {
1114 migratetype = MIGRATE_RESERVE;
1115 goto retry_reserve;
1116 }
1117 }
1118
Mel Gorman0d3d0622009-09-21 17:02:44 -07001119 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001120 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121}
1122
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001123/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 * Obtain a specified number of elements from the buddy allocator, all under
1125 * a single hold of the lock, for efficiency. Add them to the supplied list.
1126 * Returns the number of new pages which were placed at *list.
1127 */
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001128static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001129 unsigned long count, struct list_head *list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001130 int migratetype, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131{
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001132 int mt = migratetype, i;
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001133
Nick Pigginc54ad302006-01-06 00:10:56 -08001134 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001136 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001137 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08001139
1140 /*
1141 * Split buddy pages returned by expand() are received here
1142 * in physical page order. The page is added to the callers and
1143 * list and the list head then moves forward. From the callers
1144 * perspective, the linked list is ordered by page number in
1145 * some conditions. This is useful for IO devices that can
1146 * merge IO requests if the physical pages are ordered
1147 * properly.
1148 */
Mel Gormane084b2d2009-07-29 15:02:04 -07001149 if (likely(cold == 0))
1150 list_add(&page->lru, list);
1151 else
1152 list_add_tail(&page->lru, list);
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001153 if (IS_ENABLED(CONFIG_CMA)) {
1154 mt = get_pageblock_migratetype(page);
Minchan Kim194159f2013-02-22 16:33:58 -08001155 if (!is_migrate_cma(mt) && !is_migrate_isolate(mt))
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001156 mt = migratetype;
1157 }
Minchan Kimb12c4ad2012-10-08 16:32:08 -07001158 set_freepage_migratetype(page, mt);
Mel Gorman81eabcb2007-12-17 16:20:05 -08001159 list = &page->lru;
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07001160 if (is_migrate_cma(mt))
1161 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
1162 -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 }
Mel Gormanf2260e62009-06-16 15:32:13 -07001164 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08001165 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001166 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167}
1168
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001169#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001170/*
Christoph Lameter4037d452007-05-09 02:35:14 -07001171 * Called from the vmstat counter updater to drain pagesets of this
1172 * currently executing processor on remote nodes after they have
1173 * expired.
1174 *
Christoph Lameter879336c2006-03-22 00:09:08 -08001175 * Note that this function must be called with the thread pinned to
1176 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001177 */
Christoph Lameter4037d452007-05-09 02:35:14 -07001178void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001179{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001180 unsigned long flags;
Christoph Lameter4037d452007-05-09 02:35:14 -07001181 int to_drain;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001182
Christoph Lameter4037d452007-05-09 02:35:14 -07001183 local_irq_save(flags);
1184 if (pcp->count >= pcp->batch)
1185 to_drain = pcp->batch;
1186 else
1187 to_drain = pcp->count;
KOSAKI Motohiro2a135152012-07-31 16:42:53 -07001188 if (to_drain > 0) {
1189 free_pcppages_bulk(zone, to_drain, pcp);
1190 pcp->count -= to_drain;
1191 }
Christoph Lameter4037d452007-05-09 02:35:14 -07001192 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001193}
1194#endif
1195
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001196/*
1197 * Drain pages of the indicated processor.
1198 *
1199 * The processor must either be the current processor and the
1200 * thread pinned to the current processor or a processor that
1201 * is not online.
1202 */
1203static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204{
Nick Pigginc54ad302006-01-06 00:10:56 -08001205 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07001208 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 struct per_cpu_pageset *pset;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001210 struct per_cpu_pages *pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001212 local_irq_save(flags);
1213 pset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001215 pcp = &pset->pcp;
David Rientjes2ff754f2011-01-25 15:07:23 -08001216 if (pcp->count) {
1217 free_pcppages_bulk(zone, pcp->count, pcp);
1218 pcp->count = 0;
1219 }
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001220 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 }
1222}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001224/*
1225 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1226 */
1227void drain_local_pages(void *arg)
1228{
1229 drain_pages(smp_processor_id());
1230}
1231
1232/*
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001233 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1234 *
1235 * Note that this code is protected against sending an IPI to an offline
1236 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1237 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1238 * nothing keeps CPUs from showing up after we populated the cpumask and
1239 * before the call to on_each_cpu_mask().
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001240 */
1241void drain_all_pages(void)
1242{
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001243 int cpu;
1244 struct per_cpu_pageset *pcp;
1245 struct zone *zone;
1246
1247 /*
1248 * Allocate in the BSS so we wont require allocation in
1249 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1250 */
1251 static cpumask_t cpus_with_pcps;
1252
1253 /*
1254 * We don't care about racing with CPU hotplug event
1255 * as offline notification will cause the notified
1256 * cpu to drain that CPU pcps and on_each_cpu_mask
1257 * disables preemption as part of its processing
1258 */
1259 for_each_online_cpu(cpu) {
1260 bool has_pcps = false;
1261 for_each_populated_zone(zone) {
1262 pcp = per_cpu_ptr(zone->pageset, cpu);
1263 if (pcp->pcp.count) {
1264 has_pcps = true;
1265 break;
1266 }
1267 }
1268 if (has_pcps)
1269 cpumask_set_cpu(cpu, &cpus_with_pcps);
1270 else
1271 cpumask_clear_cpu(cpu, &cpus_with_pcps);
1272 }
1273 on_each_cpu_mask(&cpus_with_pcps, drain_local_pages, NULL, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001274}
1275
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02001276#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277
1278void mark_free_pages(struct zone *zone)
1279{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001280 unsigned long pfn, max_zone_pfn;
1281 unsigned long flags;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001282 int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 struct list_head *curr;
1284
1285 if (!zone->spanned_pages)
1286 return;
1287
1288 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001289
Cody P Schafer108bcc92013-02-22 16:35:23 -08001290 max_zone_pfn = zone_end_pfn(zone);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001291 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1292 if (pfn_valid(pfn)) {
1293 struct page *page = pfn_to_page(pfn);
1294
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001295 if (!swsusp_page_is_forbidden(page))
1296 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001297 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001299 for_each_migratetype_order(order, t) {
1300 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001301 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001303 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1304 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001305 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001306 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001307 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 spin_unlock_irqrestore(&zone->lock, flags);
1309}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001310#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311
1312/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 * Free a 0-order page
Li Hongfc916682010-03-05 13:41:54 -08001314 * cold == 1 ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 */
Li Hongfc916682010-03-05 13:41:54 -08001316void free_hot_cold_page(struct page *page, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317{
1318 struct zone *zone = page_zone(page);
1319 struct per_cpu_pages *pcp;
1320 unsigned long flags;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001321 int migratetype;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001323 if (!free_pages_prepare(page, 0))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001324 return;
1325
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001326 migratetype = get_pageblock_migratetype(page);
Minchan Kimb12c4ad2012-10-08 16:32:08 -07001327 set_freepage_migratetype(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001329 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001330
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001331 /*
1332 * We only track unmovable, reclaimable and movable on pcp lists.
1333 * Free ISOLATE pages back to the allocator because they are being
1334 * offlined but treat RESERVE as movable pages so we can get those
1335 * areas back if necessary. Otherwise, we may have to free
1336 * excessively into the page allocator
1337 */
1338 if (migratetype >= MIGRATE_PCPTYPES) {
Minchan Kim194159f2013-02-22 16:33:58 -08001339 if (unlikely(is_migrate_isolate(migratetype))) {
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001340 free_one_page(zone, page, 0, migratetype);
1341 goto out;
1342 }
1343 migratetype = MIGRATE_MOVABLE;
1344 }
1345
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001346 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001347 if (cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001348 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001349 else
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001350 list_add(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001352 if (pcp->count >= pcp->high) {
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001353 free_pcppages_bulk(zone, pcp->batch, pcp);
Nick Piggin48db57f2006-01-08 01:00:42 -08001354 pcp->count -= pcp->batch;
1355 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001356
1357out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359}
1360
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001361/*
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001362 * Free a list of 0-order pages
1363 */
1364void free_hot_cold_page_list(struct list_head *list, int cold)
1365{
1366 struct page *page, *next;
1367
1368 list_for_each_entry_safe(page, next, list, lru) {
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -08001369 trace_mm_page_free_batched(page, cold);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001370 free_hot_cold_page(page, cold);
1371 }
1372}
1373
1374/*
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001375 * split_page takes a non-compound higher-order page, and splits it into
1376 * n (1<<order) sub-pages: page[0..n]
1377 * Each sub-page must be freed individually.
1378 *
1379 * Note: this is probably too low level an operation for use in drivers.
1380 * Please consult with lkml before using this in your driver.
1381 */
1382void split_page(struct page *page, unsigned int order)
1383{
1384 int i;
1385
Nick Piggin725d7042006-09-25 23:30:55 -07001386 VM_BUG_ON(PageCompound(page));
1387 VM_BUG_ON(!page_count(page));
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001388
1389#ifdef CONFIG_KMEMCHECK
1390 /*
1391 * Split shadow pages too, because free(page[0]) would
1392 * otherwise free the whole shadow.
1393 */
1394 if (kmemcheck_page_is_tracked(page))
1395 split_page(virt_to_page(page[0].shadow), order);
1396#endif
1397
Nick Piggin7835e982006-03-22 00:08:40 -08001398 for (i = 1; i < (1 << order); i++)
1399 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001400}
K. Y. Srinivasan5853ff22013-03-25 15:47:38 -07001401EXPORT_SYMBOL_GPL(split_page);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001402
Mel Gorman8fb74b92013-01-11 14:32:16 -08001403static int __isolate_free_page(struct page *page, unsigned int order)
Mel Gorman748446b2010-05-24 14:32:27 -07001404{
Mel Gorman748446b2010-05-24 14:32:27 -07001405 unsigned long watermark;
1406 struct zone *zone;
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07001407 int mt;
Mel Gorman748446b2010-05-24 14:32:27 -07001408
1409 BUG_ON(!PageBuddy(page));
1410
1411 zone = page_zone(page);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001412 mt = get_pageblock_migratetype(page);
Mel Gorman748446b2010-05-24 14:32:27 -07001413
Minchan Kim194159f2013-02-22 16:33:58 -08001414 if (!is_migrate_isolate(mt)) {
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001415 /* Obey watermarks as if the page was being allocated */
1416 watermark = low_wmark_pages(zone) + (1 << order);
1417 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1418 return 0;
1419
Mel Gorman8fb74b92013-01-11 14:32:16 -08001420 __mod_zone_freepage_state(zone, -(1UL << order), mt);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001421 }
Mel Gorman748446b2010-05-24 14:32:27 -07001422
1423 /* Remove page from free list */
1424 list_del(&page->lru);
1425 zone->free_area[order].nr_free--;
1426 rmv_page_order(page);
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07001427
Mel Gorman8fb74b92013-01-11 14:32:16 -08001428 /* Set the pageblock if the isolated page is at least a pageblock */
Mel Gorman748446b2010-05-24 14:32:27 -07001429 if (order >= pageblock_order - 1) {
1430 struct page *endpage = page + (1 << order) - 1;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001431 for (; page < endpage; page += pageblock_nr_pages) {
1432 int mt = get_pageblock_migratetype(page);
Minchan Kim194159f2013-02-22 16:33:58 -08001433 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001434 set_pageblock_migratetype(page,
1435 MIGRATE_MOVABLE);
1436 }
Mel Gorman748446b2010-05-24 14:32:27 -07001437 }
1438
Mel Gorman8fb74b92013-01-11 14:32:16 -08001439 return 1UL << order;
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001440}
1441
1442/*
1443 * Similar to split_page except the page is already free. As this is only
1444 * being used for migration, the migratetype of the block also changes.
1445 * As this is called with interrupts disabled, the caller is responsible
1446 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1447 * are enabled.
1448 *
1449 * Note: this is probably too low level an operation for use in drivers.
1450 * Please consult with lkml before using this in your driver.
1451 */
1452int split_free_page(struct page *page)
1453{
1454 unsigned int order;
1455 int nr_pages;
1456
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001457 order = page_order(page);
1458
Mel Gorman8fb74b92013-01-11 14:32:16 -08001459 nr_pages = __isolate_free_page(page, order);
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07001460 if (!nr_pages)
1461 return 0;
1462
1463 /* Split into individual pages */
1464 set_page_refcounted(page);
1465 split_page(page, order);
1466 return nr_pages;
Mel Gorman748446b2010-05-24 14:32:27 -07001467}
1468
1469/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1471 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1472 * or two.
1473 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07001474static inline
1475struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001476 struct zone *zone, int order, gfp_t gfp_flags,
1477 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478{
1479 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001480 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 int cold = !!(gfp_flags & __GFP_COLD);
1482
Hugh Dickins689bceb2005-11-21 21:32:20 -08001483again:
Nick Piggin48db57f2006-01-08 01:00:42 -08001484 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001486 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 local_irq_save(flags);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001489 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1490 list = &pcp->lists[migratetype];
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001491 if (list_empty(list)) {
Mel Gorman535131e62007-10-16 01:25:49 -07001492 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001493 pcp->batch, list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001494 migratetype, cold);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001495 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07001496 goto failed;
Mel Gorman535131e62007-10-16 01:25:49 -07001497 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001498
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001499 if (cold)
1500 page = list_entry(list->prev, struct page, lru);
1501 else
1502 page = list_entry(list->next, struct page, lru);
1503
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001504 list_del(&page->lru);
1505 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001506 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07001507 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1508 /*
1509 * __GFP_NOFAIL is not to be used in new code.
1510 *
1511 * All __GFP_NOFAIL callers should be fixed so that they
1512 * properly detect and handle allocation failures.
1513 *
1514 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07001515 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07001516 * __GFP_NOFAIL.
1517 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07001518 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07001519 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001521 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001522 spin_unlock(&zone->lock);
1523 if (!page)
1524 goto failed;
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07001525 __mod_zone_freepage_state(zone, -(1 << order),
1526 get_pageblock_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 }
1528
Christoph Lameterf8891e52006-06-30 01:55:45 -07001529 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Andi Kleen78afd562011-03-22 16:33:12 -07001530 zone_statistics(preferred_zone, zone, gfp_flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001531 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532
Nick Piggin725d7042006-09-25 23:30:55 -07001533 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -08001534 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001535 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001537
1538failed:
1539 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001540 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541}
1542
Akinobu Mita933e3122006-12-08 02:39:45 -08001543#ifdef CONFIG_FAIL_PAGE_ALLOC
1544
Akinobu Mitab2588c42011-07-26 16:09:03 -07001545static struct {
Akinobu Mita933e3122006-12-08 02:39:45 -08001546 struct fault_attr attr;
1547
1548 u32 ignore_gfp_highmem;
1549 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001550 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001551} fail_page_alloc = {
1552 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001553 .ignore_gfp_wait = 1,
1554 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001555 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001556};
1557
1558static int __init setup_fail_page_alloc(char *str)
1559{
1560 return setup_fault_attr(&fail_page_alloc.attr, str);
1561}
1562__setup("fail_page_alloc=", setup_fail_page_alloc);
1563
Gavin Shandeaf3862012-07-31 16:41:51 -07001564static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08001565{
Akinobu Mita54114992007-07-15 23:40:23 -07001566 if (order < fail_page_alloc.min_order)
Gavin Shandeaf3862012-07-31 16:41:51 -07001567 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001568 if (gfp_mask & __GFP_NOFAIL)
Gavin Shandeaf3862012-07-31 16:41:51 -07001569 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001570 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
Gavin Shandeaf3862012-07-31 16:41:51 -07001571 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001572 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
Gavin Shandeaf3862012-07-31 16:41:51 -07001573 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001574
1575 return should_fail(&fail_page_alloc.attr, 1 << order);
1576}
1577
1578#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1579
1580static int __init fail_page_alloc_debugfs(void)
1581{
Al Virof4ae40a2011-07-24 04:33:43 -04001582 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
Akinobu Mita933e3122006-12-08 02:39:45 -08001583 struct dentry *dir;
Akinobu Mita933e3122006-12-08 02:39:45 -08001584
Akinobu Mitadd48c082011-08-03 16:21:01 -07001585 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
1586 &fail_page_alloc.attr);
1587 if (IS_ERR(dir))
1588 return PTR_ERR(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08001589
Akinobu Mitab2588c42011-07-26 16:09:03 -07001590 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
1591 &fail_page_alloc.ignore_gfp_wait))
1592 goto fail;
1593 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1594 &fail_page_alloc.ignore_gfp_highmem))
1595 goto fail;
1596 if (!debugfs_create_u32("min-order", mode, dir,
1597 &fail_page_alloc.min_order))
1598 goto fail;
Akinobu Mita933e3122006-12-08 02:39:45 -08001599
Akinobu Mitab2588c42011-07-26 16:09:03 -07001600 return 0;
1601fail:
Akinobu Mitadd48c082011-08-03 16:21:01 -07001602 debugfs_remove_recursive(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08001603
Akinobu Mitab2588c42011-07-26 16:09:03 -07001604 return -ENOMEM;
Akinobu Mita933e3122006-12-08 02:39:45 -08001605}
1606
1607late_initcall(fail_page_alloc_debugfs);
1608
1609#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1610
1611#else /* CONFIG_FAIL_PAGE_ALLOC */
1612
Gavin Shandeaf3862012-07-31 16:41:51 -07001613static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08001614{
Gavin Shandeaf3862012-07-31 16:41:51 -07001615 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08001616}
1617
1618#endif /* CONFIG_FAIL_PAGE_ALLOC */
1619
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620/*
Mel Gorman88f5acf2011-01-13 15:45:41 -08001621 * Return true if free pages are above 'mark'. This takes into account the order
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 * of the allocation.
1623 */
Mel Gorman88f5acf2011-01-13 15:45:41 -08001624static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1625 int classzone_idx, int alloc_flags, long free_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626{
1627 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001628 long min = mark;
Minchan Kim2cfed072012-07-31 16:43:53 -07001629 long lowmem_reserve = z->lowmem_reserve[classzone_idx];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 int o;
1631
Michal Hockodf0a6da2012-01-10 15:08:02 -08001632 free_pages -= (1 << order) - 1;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001633 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001635 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 min -= min / 4;
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07001637#ifdef CONFIG_CMA
1638 /* If allocation can't use CMA areas don't use free CMA pages */
1639 if (!(alloc_flags & ALLOC_CMA))
1640 free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES);
1641#endif
Minchan Kim2cfed072012-07-31 16:43:53 -07001642 if (free_pages <= min + lowmem_reserve)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001643 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 for (o = 0; o < order; o++) {
1645 /* At the next order, this order's pages become unavailable */
1646 free_pages -= z->free_area[o].nr_free << o;
1647
1648 /* Require fewer higher order pages to be free */
1649 min >>= 1;
1650
1651 if (free_pages <= min)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001652 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 }
Mel Gorman88f5acf2011-01-13 15:45:41 -08001654 return true;
1655}
1656
1657bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1658 int classzone_idx, int alloc_flags)
1659{
1660 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1661 zone_page_state(z, NR_FREE_PAGES));
1662}
1663
1664bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
1665 int classzone_idx, int alloc_flags)
1666{
1667 long free_pages = zone_page_state(z, NR_FREE_PAGES);
1668
1669 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1670 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1671
1672 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1673 free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674}
1675
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001676#ifdef CONFIG_NUMA
1677/*
1678 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1679 * skip over zones that are not allowed by the cpuset, or that have
1680 * been recently (in last second) found to be nearly full. See further
1681 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001682 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001683 *
1684 * If the zonelist cache is present in the passed in zonelist, then
1685 * returns a pointer to the allowed node mask (either the current
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08001686 * tasks mems_allowed, or node_states[N_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001687 *
1688 * If the zonelist cache is not available for this zonelist, does
1689 * nothing and returns NULL.
1690 *
1691 * If the fullzones BITMAP in the zonelist cache is stale (more than
1692 * a second since last zap'd) then we zap it out (clear its bits.)
1693 *
1694 * We hold off even calling zlc_setup, until after we've checked the
1695 * first zone in the zonelist, on the theory that most allocations will
1696 * be satisfied from that first zone, so best to examine that zone as
1697 * quickly as we can.
1698 */
1699static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1700{
1701 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1702 nodemask_t *allowednodes; /* zonelist_cache approximation */
1703
1704 zlc = zonelist->zlcache_ptr;
1705 if (!zlc)
1706 return NULL;
1707
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001708 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001709 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1710 zlc->last_full_zap = jiffies;
1711 }
1712
1713 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1714 &cpuset_current_mems_allowed :
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08001715 &node_states[N_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001716 return allowednodes;
1717}
1718
1719/*
1720 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1721 * if it is worth looking at further for free memory:
1722 * 1) Check that the zone isn't thought to be full (doesn't have its
1723 * bit set in the zonelist_cache fullzones BITMAP).
1724 * 2) Check that the zones node (obtained from the zonelist_cache
1725 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1726 * Return true (non-zero) if zone is worth looking at further, or
1727 * else return false (zero) if it is not.
1728 *
1729 * This check -ignores- the distinction between various watermarks,
1730 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1731 * found to be full for any variation of these watermarks, it will
1732 * be considered full for up to one second by all requests, unless
1733 * we are so low on memory on all allowed nodes that we are forced
1734 * into the second scan of the zonelist.
1735 *
1736 * In the second scan we ignore this zonelist cache and exactly
1737 * apply the watermarks to all zones, even it is slower to do so.
1738 * We are low on memory in the second scan, and should leave no stone
1739 * unturned looking for a free page.
1740 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001741static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001742 nodemask_t *allowednodes)
1743{
1744 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1745 int i; /* index of *z in zonelist zones */
1746 int n; /* node that zone *z is on */
1747
1748 zlc = zonelist->zlcache_ptr;
1749 if (!zlc)
1750 return 1;
1751
Mel Gormandd1a2392008-04-28 02:12:17 -07001752 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001753 n = zlc->z_to_n[i];
1754
1755 /* This zone is worth trying if it is allowed but not full */
1756 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1757}
1758
1759/*
1760 * Given 'z' scanning a zonelist, set the corresponding bit in
1761 * zlc->fullzones, so that subsequent attempts to allocate a page
1762 * from that zone don't waste time re-examining it.
1763 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001764static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001765{
1766 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1767 int i; /* index of *z in zonelist zones */
1768
1769 zlc = zonelist->zlcache_ptr;
1770 if (!zlc)
1771 return;
1772
Mel Gormandd1a2392008-04-28 02:12:17 -07001773 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001774
1775 set_bit(i, zlc->fullzones);
1776}
1777
Mel Gorman76d3fbf2011-07-25 17:12:30 -07001778/*
1779 * clear all zones full, called after direct reclaim makes progress so that
1780 * a zone that was recently full is not skipped over for up to a second
1781 */
1782static void zlc_clear_zones_full(struct zonelist *zonelist)
1783{
1784 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1785
1786 zlc = zonelist->zlcache_ptr;
1787 if (!zlc)
1788 return;
1789
1790 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1791}
1792
David Rientjes957f8222012-10-08 16:33:24 -07001793static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
1794{
1795 return node_isset(local_zone->node, zone->zone_pgdat->reclaim_nodes);
1796}
1797
1798static void __paginginit init_zone_allows_reclaim(int nid)
1799{
1800 int i;
1801
1802 for_each_online_node(i)
David Rientjes6b187d022012-10-25 13:38:08 -07001803 if (node_distance(nid, i) <= RECLAIM_DISTANCE)
David Rientjes957f8222012-10-08 16:33:24 -07001804 node_set(i, NODE_DATA(nid)->reclaim_nodes);
David Rientjes6b187d022012-10-25 13:38:08 -07001805 else
David Rientjes957f8222012-10-08 16:33:24 -07001806 zone_reclaim_mode = 1;
David Rientjes957f8222012-10-08 16:33:24 -07001807}
1808
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001809#else /* CONFIG_NUMA */
1810
1811static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1812{
1813 return NULL;
1814}
1815
Mel Gormandd1a2392008-04-28 02:12:17 -07001816static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001817 nodemask_t *allowednodes)
1818{
1819 return 1;
1820}
1821
Mel Gormandd1a2392008-04-28 02:12:17 -07001822static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001823{
1824}
Mel Gorman76d3fbf2011-07-25 17:12:30 -07001825
1826static void zlc_clear_zones_full(struct zonelist *zonelist)
1827{
1828}
David Rientjes957f8222012-10-08 16:33:24 -07001829
1830static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
1831{
1832 return true;
1833}
1834
1835static inline void init_zone_allows_reclaim(int nid)
1836{
1837}
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001838#endif /* CONFIG_NUMA */
1839
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001840/*
Paul Jackson0798e512006-12-06 20:31:38 -08001841 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001842 * a page.
1843 */
1844static struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001845get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
Mel Gorman5117f452009-06-16 15:31:59 -07001846 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
Mel Gorman3dd28262009-06-16 15:32:00 -07001847 struct zone *preferred_zone, int migratetype)
Martin Hicks753ee722005-06-21 17:14:41 -07001848{
Mel Gormandd1a2392008-04-28 02:12:17 -07001849 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001850 struct page *page = NULL;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001851 int classzone_idx;
Mel Gorman5117f452009-06-16 15:31:59 -07001852 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001853 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1854 int zlc_active = 0; /* set if using zonelist_cache */
1855 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gorman54a6eb52008-04-28 02:12:16 -07001856
Mel Gorman19770b32008-04-28 02:12:18 -07001857 classzone_idx = zone_idx(preferred_zone);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001858zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001859 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001860 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001861 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1862 */
Mel Gorman19770b32008-04-28 02:12:18 -07001863 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1864 high_zoneidx, nodemask) {
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08001865 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001866 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1867 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001868 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001869 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Mel Gormancd38b112011-07-25 17:12:29 -07001870 continue;
Johannes Weinera756cf52012-01-10 15:07:49 -08001871 /*
1872 * When allocating a page cache page for writing, we
1873 * want to get it from a zone that is within its dirty
1874 * limit, such that no single zone holds more than its
1875 * proportional share of globally allowed dirty pages.
1876 * The dirty limits take into account the zone's
1877 * lowmem reserves and high watermark so that kswapd
1878 * should be able to balance it without having to
1879 * write pages from its LRU list.
1880 *
1881 * This may look like it could increase pressure on
1882 * lower zones by failing allocations in higher zones
1883 * before they are full. But the pages that do spill
1884 * over are limited as the lower zones are protected
1885 * by this very same mechanism. It should not become
1886 * a practical burden to them.
1887 *
1888 * XXX: For now, allow allocations to potentially
1889 * exceed the per-zone dirty limit in the slowpath
1890 * (ALLOC_WMARK_LOW unset) before going into reclaim,
1891 * which is important when on a NUMA setup the allowed
1892 * zones are together not big enough to reach the
1893 * global limit. The proper fix for these situations
1894 * will require awareness of zones in the
1895 * dirty-throttling and the flusher threads.
1896 */
1897 if ((alloc_flags & ALLOC_WMARK_LOW) &&
1898 (gfp_mask & __GFP_WRITE) && !zone_dirty_ok(zone))
1899 goto this_zone_full;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001900
Mel Gorman41858962009-06-16 15:32:12 -07001901 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001902 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001903 unsigned long mark;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001904 int ret;
1905
Mel Gorman41858962009-06-16 15:32:12 -07001906 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
Mel Gormanfa5e0842009-06-16 15:33:22 -07001907 if (zone_watermark_ok(zone, order, mark,
1908 classzone_idx, alloc_flags))
1909 goto try_this_zone;
1910
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08001911 if (IS_ENABLED(CONFIG_NUMA) &&
1912 !did_zlc_setup && nr_online_nodes > 1) {
Mel Gormancd38b112011-07-25 17:12:29 -07001913 /*
1914 * we do zlc_setup if there are multiple nodes
1915 * and before considering the first zone allowed
1916 * by the cpuset.
1917 */
1918 allowednodes = zlc_setup(zonelist, alloc_flags);
1919 zlc_active = 1;
1920 did_zlc_setup = 1;
1921 }
1922
David Rientjes957f8222012-10-08 16:33:24 -07001923 if (zone_reclaim_mode == 0 ||
1924 !zone_allows_reclaim(preferred_zone, zone))
Mel Gormanfa5e0842009-06-16 15:33:22 -07001925 goto this_zone_full;
1926
Mel Gormancd38b112011-07-25 17:12:29 -07001927 /*
1928 * As we may have just activated ZLC, check if the first
1929 * eligible zone has failed zone_reclaim recently.
1930 */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08001931 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
Mel Gormancd38b112011-07-25 17:12:29 -07001932 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1933 continue;
1934
Mel Gormanfa5e0842009-06-16 15:33:22 -07001935 ret = zone_reclaim(zone, gfp_mask, order);
1936 switch (ret) {
1937 case ZONE_RECLAIM_NOSCAN:
1938 /* did not scan */
Mel Gormancd38b112011-07-25 17:12:29 -07001939 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001940 case ZONE_RECLAIM_FULL:
1941 /* scanned but unreclaimable */
Mel Gormancd38b112011-07-25 17:12:29 -07001942 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001943 default:
1944 /* did we reclaim enough */
Mel Gormanfed27192013-04-29 15:07:57 -07001945 if (zone_watermark_ok(zone, order, mark,
Mel Gormanfa5e0842009-06-16 15:33:22 -07001946 classzone_idx, alloc_flags))
Mel Gormanfed27192013-04-29 15:07:57 -07001947 goto try_this_zone;
1948
1949 /*
1950 * Failed to reclaim enough to meet watermark.
1951 * Only mark the zone full if checking the min
1952 * watermark or if we failed to reclaim just
1953 * 1<<order pages or else the page allocator
1954 * fastpath will prematurely mark zones full
1955 * when the watermark is between the low and
1956 * min watermarks.
1957 */
1958 if (((alloc_flags & ALLOC_WMARK_MASK) == ALLOC_WMARK_MIN) ||
1959 ret == ZONE_RECLAIM_SOME)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001960 goto this_zone_full;
Mel Gormanfed27192013-04-29 15:07:57 -07001961
1962 continue;
Paul Jackson0798e512006-12-06 20:31:38 -08001963 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001964 }
1965
Mel Gormanfa5e0842009-06-16 15:33:22 -07001966try_this_zone:
Mel Gorman3dd28262009-06-16 15:32:00 -07001967 page = buffered_rmqueue(preferred_zone, zone, order,
1968 gfp_mask, migratetype);
Paul Jackson0798e512006-12-06 20:31:38 -08001969 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001970 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001971this_zone_full:
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08001972 if (IS_ENABLED(CONFIG_NUMA))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001973 zlc_mark_zone_full(zonelist, z);
Mel Gorman54a6eb52008-04-28 02:12:16 -07001974 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001975
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08001976 if (unlikely(IS_ENABLED(CONFIG_NUMA) && page == NULL && zlc_active)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001977 /* Disable zlc cache for second zonelist scan */
1978 zlc_active = 0;
1979 goto zonelist_scan;
1980 }
Alex Shib1211862012-08-21 16:16:08 -07001981
1982 if (page)
1983 /*
1984 * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was
1985 * necessary to allocate the page. The expectation is
1986 * that the caller is taking steps that will free more
1987 * memory. The caller should avoid the page being used
1988 * for !PFMEMALLOC purposes.
1989 */
1990 page->pfmemalloc = !!(alloc_flags & ALLOC_NO_WATERMARKS);
1991
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001992 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07001993}
1994
David Rientjes29423e772011-03-22 16:30:47 -07001995/*
1996 * Large machines with many possible nodes should not always dump per-node
1997 * meminfo in irq context.
1998 */
1999static inline bool should_suppress_show_mem(void)
2000{
2001 bool ret = false;
2002
2003#if NODES_SHIFT > 8
2004 ret = in_interrupt();
2005#endif
2006 return ret;
2007}
2008
Dave Hansena238ab52011-05-24 17:12:16 -07002009static DEFINE_RATELIMIT_STATE(nopage_rs,
2010 DEFAULT_RATELIMIT_INTERVAL,
2011 DEFAULT_RATELIMIT_BURST);
2012
2013void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
2014{
Dave Hansena238ab52011-05-24 17:12:16 -07002015 unsigned int filter = SHOW_MEM_FILTER_NODES;
2016
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -08002017 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2018 debug_guardpage_minorder() > 0)
Dave Hansena238ab52011-05-24 17:12:16 -07002019 return;
2020
2021 /*
David Rientjes4b59e6c2013-04-29 15:06:11 -07002022 * Walking all memory to count page types is very expensive and should
2023 * be inhibited in non-blockable contexts.
2024 */
2025 if (!(gfp_mask & __GFP_WAIT))
2026 filter |= SHOW_MEM_FILTER_PAGE_COUNT;
2027
2028 /*
Dave Hansena238ab52011-05-24 17:12:16 -07002029 * This documents exceptions given to allocations in certain
2030 * contexts that are allowed to allocate outside current's set
2031 * of allowed nodes.
2032 */
2033 if (!(gfp_mask & __GFP_NOMEMALLOC))
2034 if (test_thread_flag(TIF_MEMDIE) ||
2035 (current->flags & (PF_MEMALLOC | PF_EXITING)))
2036 filter &= ~SHOW_MEM_FILTER_NODES;
2037 if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
2038 filter &= ~SHOW_MEM_FILTER_NODES;
2039
2040 if (fmt) {
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002041 struct va_format vaf;
2042 va_list args;
2043
Dave Hansena238ab52011-05-24 17:12:16 -07002044 va_start(args, fmt);
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002045
2046 vaf.fmt = fmt;
2047 vaf.va = &args;
2048
2049 pr_warn("%pV", &vaf);
2050
Dave Hansena238ab52011-05-24 17:12:16 -07002051 va_end(args);
2052 }
2053
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002054 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2055 current->comm, order, gfp_mask);
Dave Hansena238ab52011-05-24 17:12:16 -07002056
2057 dump_stack();
2058 if (!should_suppress_show_mem())
2059 show_mem(filter);
2060}
2061
Mel Gorman11e33f62009-06-16 15:31:57 -07002062static inline int
2063should_alloc_retry(gfp_t gfp_mask, unsigned int order,
Mel Gormanf90ac392012-01-10 15:07:15 -08002064 unsigned long did_some_progress,
Mel Gorman11e33f62009-06-16 15:31:57 -07002065 unsigned long pages_reclaimed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066{
Mel Gorman11e33f62009-06-16 15:31:57 -07002067 /* Do not loop if specifically requested */
2068 if (gfp_mask & __GFP_NORETRY)
2069 return 0;
2070
Mel Gormanf90ac392012-01-10 15:07:15 -08002071 /* Always retry if specifically requested */
2072 if (gfp_mask & __GFP_NOFAIL)
2073 return 1;
2074
2075 /*
2076 * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim
2077 * making forward progress without invoking OOM. Suspend also disables
2078 * storage devices so kswapd will not help. Bail if we are suspending.
2079 */
2080 if (!did_some_progress && pm_suspended_storage())
2081 return 0;
2082
Mel Gorman11e33f62009-06-16 15:31:57 -07002083 /*
2084 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
2085 * means __GFP_NOFAIL, but that may not be true in other
2086 * implementations.
2087 */
2088 if (order <= PAGE_ALLOC_COSTLY_ORDER)
2089 return 1;
2090
2091 /*
2092 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
2093 * specified, then we retry until we no longer reclaim any pages
2094 * (above), or we've reclaimed an order of pages at least as
2095 * large as the allocation's order. In both cases, if the
2096 * allocation still fails, we stop retrying.
2097 */
2098 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
2099 return 1;
2100
Mel Gorman11e33f62009-06-16 15:31:57 -07002101 return 0;
2102}
2103
2104static inline struct page *
2105__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
2106 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002107 nodemask_t *nodemask, struct zone *preferred_zone,
2108 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002109{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111
Mel Gorman11e33f62009-06-16 15:31:57 -07002112 /* Acquire the OOM killer lock for the zones in zonelist */
Minchan Kimff321fe2010-08-09 17:18:57 -07002113 if (!try_set_zonelist_oom(zonelist, gfp_mask)) {
Mel Gorman11e33f62009-06-16 15:31:57 -07002114 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 return NULL;
2116 }
Jens Axboe6b1de912005-11-17 21:35:02 +01002117
Mel Gorman11e33f62009-06-16 15:31:57 -07002118 /*
2119 * Go through the zonelist yet one more time, keep very high watermark
2120 * here, this is only to catch a parallel oom killing, we must fail if
2121 * we're still under heavy pressure.
2122 */
2123 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
2124 order, zonelist, high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07002125 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07002126 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002127 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07002128 goto out;
2129
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002130 if (!(gfp_mask & __GFP_NOFAIL)) {
2131 /* The OOM killer will not help higher order allocs */
2132 if (order > PAGE_ALLOC_COSTLY_ORDER)
2133 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07002134 /* The OOM killer does not needlessly kill tasks for lowmem */
2135 if (high_zoneidx < ZONE_NORMAL)
2136 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002137 /*
2138 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
2139 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
2140 * The caller should handle page allocation failure by itself if
2141 * it specifies __GFP_THISNODE.
2142 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
2143 */
2144 if (gfp_mask & __GFP_THISNODE)
2145 goto out;
2146 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002147 /* Exhausted what can be done so it's blamo time */
David Rientjes08ab9b12012-03-21 16:34:04 -07002148 out_of_memory(zonelist, gfp_mask, order, nodemask, false);
Mel Gorman11e33f62009-06-16 15:31:57 -07002149
2150out:
2151 clear_zonelist_oom(zonelist, gfp_mask);
2152 return page;
2153}
2154
Mel Gorman56de7262010-05-24 14:32:30 -07002155#ifdef CONFIG_COMPACTION
2156/* Try memory compaction for high-order allocations before reclaim */
2157static struct page *
2158__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2159 struct zonelist *zonelist, enum zone_type high_zoneidx,
2160 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman66199712012-01-12 17:19:41 -08002161 int migratetype, bool sync_migration,
Mel Gormanc67fe372012-08-21 16:16:17 -07002162 bool *contended_compaction, bool *deferred_compaction,
Mel Gorman66199712012-01-12 17:19:41 -08002163 unsigned long *did_some_progress)
Mel Gorman56de7262010-05-24 14:32:30 -07002164{
Mel Gorman66199712012-01-12 17:19:41 -08002165 if (!order)
Mel Gorman56de7262010-05-24 14:32:30 -07002166 return NULL;
2167
Rik van Rielaff62242012-03-21 16:33:52 -07002168 if (compaction_deferred(preferred_zone, order)) {
Mel Gorman66199712012-01-12 17:19:41 -08002169 *deferred_compaction = true;
2170 return NULL;
2171 }
2172
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002173 current->flags |= PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07002174 *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
Mel Gormanc67fe372012-08-21 16:16:17 -07002175 nodemask, sync_migration,
Mel Gorman8fb74b92013-01-11 14:32:16 -08002176 contended_compaction);
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002177 current->flags &= ~PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07002178
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07002179 if (*did_some_progress != COMPACT_SKIPPED) {
Mel Gorman8fb74b92013-01-11 14:32:16 -08002180 struct page *page;
2181
Mel Gorman56de7262010-05-24 14:32:30 -07002182 /* Page migration frees to the PCP lists but we want merging */
2183 drain_pages(get_cpu());
2184 put_cpu();
2185
2186 page = get_page_from_freelist(gfp_mask, nodemask,
2187 order, zonelist, high_zoneidx,
Mel Gormancfd19c52012-07-31 16:44:10 -07002188 alloc_flags & ~ALLOC_NO_WATERMARKS,
2189 preferred_zone, migratetype);
Mel Gorman56de7262010-05-24 14:32:30 -07002190 if (page) {
Mel Gorman62997022012-10-08 16:32:47 -07002191 preferred_zone->compact_blockskip_flush = false;
Mel Gorman4f92e252010-05-24 14:32:32 -07002192 preferred_zone->compact_considered = 0;
2193 preferred_zone->compact_defer_shift = 0;
Rik van Rielaff62242012-03-21 16:33:52 -07002194 if (order >= preferred_zone->compact_order_failed)
2195 preferred_zone->compact_order_failed = order + 1;
Mel Gorman56de7262010-05-24 14:32:30 -07002196 count_vm_event(COMPACTSUCCESS);
2197 return page;
2198 }
2199
2200 /*
2201 * It's bad if compaction run occurs and fails.
2202 * The most likely reason is that pages exist,
2203 * but not enough to satisfy watermarks.
2204 */
2205 count_vm_event(COMPACTFAIL);
Mel Gorman66199712012-01-12 17:19:41 -08002206
2207 /*
2208 * As async compaction considers a subset of pageblocks, only
2209 * defer if the failure was a sync compaction failure.
2210 */
2211 if (sync_migration)
Rik van Rielaff62242012-03-21 16:33:52 -07002212 defer_compaction(preferred_zone, order);
Mel Gorman56de7262010-05-24 14:32:30 -07002213
2214 cond_resched();
2215 }
2216
2217 return NULL;
2218}
2219#else
2220static inline struct page *
2221__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2222 struct zonelist *zonelist, enum zone_type high_zoneidx,
2223 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman66199712012-01-12 17:19:41 -08002224 int migratetype, bool sync_migration,
Mel Gormanc67fe372012-08-21 16:16:17 -07002225 bool *contended_compaction, bool *deferred_compaction,
Mel Gorman66199712012-01-12 17:19:41 -08002226 unsigned long *did_some_progress)
Mel Gorman56de7262010-05-24 14:32:30 -07002227{
2228 return NULL;
2229}
2230#endif /* CONFIG_COMPACTION */
2231
Marek Szyprowskibba90712012-01-25 12:09:52 +01002232/* Perform direct synchronous page reclaim */
2233static int
2234__perform_reclaim(gfp_t gfp_mask, unsigned int order, struct zonelist *zonelist,
2235 nodemask_t *nodemask)
Mel Gorman11e33f62009-06-16 15:31:57 -07002236{
Mel Gorman11e33f62009-06-16 15:31:57 -07002237 struct reclaim_state reclaim_state;
Marek Szyprowskibba90712012-01-25 12:09:52 +01002238 int progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07002239
2240 cond_resched();
2241
2242 /* We now go into synchronous reclaim */
2243 cpuset_memory_pressure_bump();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002244 current->flags |= PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002245 lockdep_set_current_reclaim_state(gfp_mask);
2246 reclaim_state.reclaimed_slab = 0;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002247 current->reclaim_state = &reclaim_state;
Mel Gorman11e33f62009-06-16 15:31:57 -07002248
Marek Szyprowskibba90712012-01-25 12:09:52 +01002249 progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002250
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002251 current->reclaim_state = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002252 lockdep_clear_current_reclaim_state();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002253 current->flags &= ~PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002254
2255 cond_resched();
2256
Marek Szyprowskibba90712012-01-25 12:09:52 +01002257 return progress;
2258}
2259
2260/* The really slow allocator path where we enter direct reclaim */
2261static inline struct page *
2262__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
2263 struct zonelist *zonelist, enum zone_type high_zoneidx,
2264 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
2265 int migratetype, unsigned long *did_some_progress)
2266{
2267 struct page *page = NULL;
2268 bool drained = false;
2269
2270 *did_some_progress = __perform_reclaim(gfp_mask, order, zonelist,
2271 nodemask);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002272 if (unlikely(!(*did_some_progress)))
2273 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002274
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002275 /* After successful reclaim, reconsider all zones for allocation */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002276 if (IS_ENABLED(CONFIG_NUMA))
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002277 zlc_clear_zones_full(zonelist);
2278
Mel Gorman9ee493c2010-09-09 16:38:18 -07002279retry:
2280 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002281 zonelist, high_zoneidx,
Mel Gormancfd19c52012-07-31 16:44:10 -07002282 alloc_flags & ~ALLOC_NO_WATERMARKS,
2283 preferred_zone, migratetype);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002284
2285 /*
2286 * If an allocation failed after direct reclaim, it could be because
2287 * pages are pinned on the per-cpu lists. Drain them and try again
2288 */
2289 if (!page && !drained) {
2290 drain_all_pages();
2291 drained = true;
2292 goto retry;
2293 }
2294
Mel Gorman11e33f62009-06-16 15:31:57 -07002295 return page;
2296}
2297
Mel Gorman11e33f62009-06-16 15:31:57 -07002298/*
2299 * This is called in the allocator slow-path if the allocation request is of
2300 * sufficient urgency to ignore watermarks and take other desperate measures
2301 */
2302static inline struct page *
2303__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
2304 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002305 nodemask_t *nodemask, struct zone *preferred_zone,
2306 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002307{
2308 struct page *page;
2309
2310 do {
2311 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002312 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
Mel Gorman3dd28262009-06-16 15:32:00 -07002313 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002314
2315 if (!page && gfp_mask & __GFP_NOFAIL)
Mel Gorman0e093d992010-10-26 14:21:45 -07002316 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07002317 } while (!page && (gfp_mask & __GFP_NOFAIL));
2318
2319 return page;
2320}
2321
2322static inline
2323void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
Mel Gorman99504742011-01-13 15:46:20 -08002324 enum zone_type high_zoneidx,
2325 enum zone_type classzone_idx)
Mel Gorman11e33f62009-06-16 15:31:57 -07002326{
2327 struct zoneref *z;
2328 struct zone *zone;
2329
2330 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
Mel Gorman99504742011-01-13 15:46:20 -08002331 wakeup_kswapd(zone, order, classzone_idx);
Mel Gorman11e33f62009-06-16 15:31:57 -07002332}
2333
Peter Zijlstra341ce062009-06-16 15:32:02 -07002334static inline int
2335gfp_to_alloc_flags(gfp_t gfp_mask)
2336{
Peter Zijlstra341ce062009-06-16 15:32:02 -07002337 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
2338 const gfp_t wait = gfp_mask & __GFP_WAIT;
2339
Mel Gormana56f57f2009-06-16 15:32:02 -07002340 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07002341 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07002342
Peter Zijlstra341ce062009-06-16 15:32:02 -07002343 /*
2344 * The caller may dip into page reserves a bit more if the caller
2345 * cannot run direct reclaim, or if the caller has realtime scheduling
2346 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2347 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
2348 */
Namhyung Kime6223a32010-10-26 14:21:59 -07002349 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07002350
2351 if (!wait) {
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002352 /*
2353 * Not worth trying to allocate harder for
2354 * __GFP_NOMEMALLOC even if it can't schedule.
2355 */
2356 if (!(gfp_mask & __GFP_NOMEMALLOC))
2357 alloc_flags |= ALLOC_HARDER;
Peter Zijlstra341ce062009-06-16 15:32:02 -07002358 /*
2359 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
2360 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
2361 */
2362 alloc_flags &= ~ALLOC_CPUSET;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002363 } else if (unlikely(rt_task(current)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07002364 alloc_flags |= ALLOC_HARDER;
2365
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002366 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2367 if (gfp_mask & __GFP_MEMALLOC)
2368 alloc_flags |= ALLOC_NO_WATERMARKS;
Mel Gorman907aed42012-07-31 16:44:07 -07002369 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
2370 alloc_flags |= ALLOC_NO_WATERMARKS;
2371 else if (!in_interrupt() &&
2372 ((current->flags & PF_MEMALLOC) ||
2373 unlikely(test_thread_flag(TIF_MEMDIE))))
Peter Zijlstra341ce062009-06-16 15:32:02 -07002374 alloc_flags |= ALLOC_NO_WATERMARKS;
2375 }
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002376#ifdef CONFIG_CMA
2377 if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2378 alloc_flags |= ALLOC_CMA;
2379#endif
Peter Zijlstra341ce062009-06-16 15:32:02 -07002380 return alloc_flags;
2381}
2382
Mel Gorman072bb0a2012-07-31 16:43:58 -07002383bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
2384{
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002385 return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
Mel Gorman072bb0a2012-07-31 16:43:58 -07002386}
2387
Mel Gorman11e33f62009-06-16 15:31:57 -07002388static inline struct page *
2389__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
2390 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002391 nodemask_t *nodemask, struct zone *preferred_zone,
2392 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002393{
2394 const gfp_t wait = gfp_mask & __GFP_WAIT;
2395 struct page *page = NULL;
2396 int alloc_flags;
2397 unsigned long pages_reclaimed = 0;
2398 unsigned long did_some_progress;
Mel Gorman77f1fe62011-01-13 15:45:57 -08002399 bool sync_migration = false;
Mel Gorman66199712012-01-12 17:19:41 -08002400 bool deferred_compaction = false;
Mel Gormanc67fe372012-08-21 16:16:17 -07002401 bool contended_compaction = false;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002402
Christoph Lameter952f3b52006-12-06 20:33:26 -08002403 /*
Mel Gorman72807a72009-06-16 15:32:18 -07002404 * In the slowpath, we sanity check order to avoid ever trying to
2405 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2406 * be using allocators in order of preference for an area that is
2407 * too large.
2408 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07002409 if (order >= MAX_ORDER) {
2410 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07002411 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07002412 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002413
Christoph Lameter952f3b52006-12-06 20:33:26 -08002414 /*
2415 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2416 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2417 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2418 * using a larger set of nodes after it has established that the
2419 * allowed per node queues are empty and that nodes are
2420 * over allocated.
2421 */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002422 if (IS_ENABLED(CONFIG_NUMA) &&
2423 (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
Christoph Lameter952f3b52006-12-06 20:33:26 -08002424 goto nopage;
2425
Mel Gormancc4a6852009-11-11 14:26:14 -08002426restart:
Linus Torvaldscaf49192012-12-10 10:51:16 -08002427 if (!(gfp_mask & __GFP_NO_KSWAPD))
2428 wake_all_kswapd(order, zonelist, high_zoneidx,
2429 zone_idx(preferred_zone));
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002430
Paul Jackson9bf22292005-09-06 15:18:12 -07002431 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002432 * OK, we're below the kswapd watermark and have kicked background
2433 * reclaim. Now things get more complex, so set up alloc_flags according
2434 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07002435 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002436 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437
David Rientjesf33261d2011-01-25 15:07:20 -08002438 /*
2439 * Find the true preferred zone if the allocation is unconstrained by
2440 * cpusets.
2441 */
2442 if (!(alloc_flags & ALLOC_CPUSET) && !nodemask)
2443 first_zones_zonelist(zonelist, high_zoneidx, NULL,
2444 &preferred_zone);
2445
Andrew Barrycfa54a02011-05-24 17:12:52 -07002446rebalance:
Peter Zijlstra341ce062009-06-16 15:32:02 -07002447 /* This is the last chance, in general, before the goto nopage. */
Mel Gorman19770b32008-04-28 02:12:18 -07002448 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
Peter Zijlstra341ce062009-06-16 15:32:02 -07002449 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
2450 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002451 if (page)
2452 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453
Mel Gorman11e33f62009-06-16 15:31:57 -07002454 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002455 if (alloc_flags & ALLOC_NO_WATERMARKS) {
Mel Gorman183f6372012-07-31 16:44:12 -07002456 /*
2457 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
2458 * the allocation is high priority and these type of
2459 * allocations are system rather than user orientated
2460 */
2461 zonelist = node_zonelist(numa_node_id(), gfp_mask);
2462
Peter Zijlstra341ce062009-06-16 15:32:02 -07002463 page = __alloc_pages_high_priority(gfp_mask, order,
2464 zonelist, high_zoneidx, nodemask,
2465 preferred_zone, migratetype);
Mel Gormancfd19c52012-07-31 16:44:10 -07002466 if (page) {
Peter Zijlstra341ce062009-06-16 15:32:02 -07002467 goto got_pg;
Mel Gormancfd19c52012-07-31 16:44:10 -07002468 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469 }
2470
2471 /* Atomic allocations - we can't balance anything */
2472 if (!wait)
2473 goto nopage;
2474
Peter Zijlstra341ce062009-06-16 15:32:02 -07002475 /* Avoid recursion of direct reclaim */
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002476 if (current->flags & PF_MEMALLOC)
Peter Zijlstra341ce062009-06-16 15:32:02 -07002477 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478
David Rientjes6583bb62009-07-29 15:02:06 -07002479 /* Avoid allocations with no watermarks from looping endlessly */
2480 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2481 goto nopage;
2482
Mel Gorman77f1fe62011-01-13 15:45:57 -08002483 /*
2484 * Try direct compaction. The first pass is asynchronous. Subsequent
2485 * attempts after direct reclaim are synchronous
2486 */
Mel Gorman56de7262010-05-24 14:32:30 -07002487 page = __alloc_pages_direct_compact(gfp_mask, order,
2488 zonelist, high_zoneidx,
2489 nodemask,
2490 alloc_flags, preferred_zone,
Mel Gorman66199712012-01-12 17:19:41 -08002491 migratetype, sync_migration,
Mel Gormanc67fe372012-08-21 16:16:17 -07002492 &contended_compaction,
Mel Gorman66199712012-01-12 17:19:41 -08002493 &deferred_compaction,
2494 &did_some_progress);
Mel Gorman56de7262010-05-24 14:32:30 -07002495 if (page)
2496 goto got_pg;
Andrea Arcangelic6a140b2011-05-24 17:11:38 -07002497 sync_migration = true;
Mel Gorman56de7262010-05-24 14:32:30 -07002498
Linus Torvalds31f8d422012-12-10 10:47:45 -08002499 /*
2500 * If compaction is deferred for high-order allocations, it is because
2501 * sync compaction recently failed. In this is the case and the caller
2502 * requested a movable allocation that does not heavily disrupt the
2503 * system then fail the allocation instead of entering direct reclaim.
2504 */
2505 if ((deferred_compaction || contended_compaction) &&
Linus Torvaldscaf49192012-12-10 10:51:16 -08002506 (gfp_mask & __GFP_NO_KSWAPD))
Linus Torvalds31f8d422012-12-10 10:47:45 -08002507 goto nopage;
Mel Gorman66199712012-01-12 17:19:41 -08002508
Mel Gorman11e33f62009-06-16 15:31:57 -07002509 /* Try direct reclaim and then allocating */
2510 page = __alloc_pages_direct_reclaim(gfp_mask, order,
2511 zonelist, high_zoneidx,
2512 nodemask,
Mel Gorman5117f452009-06-16 15:31:59 -07002513 alloc_flags, preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07002514 migratetype, &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07002515 if (page)
2516 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518 /*
Mel Gorman11e33f62009-06-16 15:31:57 -07002519 * If we failed to make any progress reclaiming, then we are
2520 * running out of options and have to consider going OOM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521 */
Mel Gorman11e33f62009-06-16 15:31:57 -07002522 if (!did_some_progress) {
2523 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -07002524 if (oom_killer_disabled)
2525 goto nopage;
David Rientjes29fd66d2012-03-28 14:42:41 -07002526 /* Coredumps can quickly deplete all memory reserves */
2527 if ((current->flags & PF_DUMPCORE) &&
2528 !(gfp_mask & __GFP_NOFAIL))
2529 goto nopage;
Mel Gorman11e33f62009-06-16 15:31:57 -07002530 page = __alloc_pages_may_oom(gfp_mask, order,
2531 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002532 nodemask, preferred_zone,
2533 migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002534 if (page)
2535 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536
David Rientjes03668b32010-08-09 17:18:54 -07002537 if (!(gfp_mask & __GFP_NOFAIL)) {
2538 /*
2539 * The oom killer is not called for high-order
2540 * allocations that may fail, so if no progress
2541 * is being made, there are no other options and
2542 * retrying is unlikely to help.
2543 */
2544 if (order > PAGE_ALLOC_COSTLY_ORDER)
2545 goto nopage;
2546 /*
2547 * The oom killer is not called for lowmem
2548 * allocations to prevent needlessly killing
2549 * innocent tasks.
2550 */
2551 if (high_zoneidx < ZONE_NORMAL)
2552 goto nopage;
2553 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 goto restart;
2556 }
2557 }
2558
Mel Gorman11e33f62009-06-16 15:31:57 -07002559 /* Check if we should retry the allocation */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07002560 pages_reclaimed += did_some_progress;
Mel Gormanf90ac392012-01-10 15:07:15 -08002561 if (should_alloc_retry(gfp_mask, order, did_some_progress,
2562 pages_reclaimed)) {
Mel Gorman11e33f62009-06-16 15:31:57 -07002563 /* Wait for some write requests to complete then retry */
Mel Gorman0e093d992010-10-26 14:21:45 -07002564 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 goto rebalance;
Mel Gorman3e7d3442011-01-13 15:45:56 -08002566 } else {
2567 /*
2568 * High-order allocations do not necessarily loop after
2569 * direct reclaim and reclaim/compaction depends on compaction
2570 * being called after reclaim so call directly if necessary
2571 */
2572 page = __alloc_pages_direct_compact(gfp_mask, order,
2573 zonelist, high_zoneidx,
2574 nodemask,
2575 alloc_flags, preferred_zone,
Mel Gorman66199712012-01-12 17:19:41 -08002576 migratetype, sync_migration,
Mel Gormanc67fe372012-08-21 16:16:17 -07002577 &contended_compaction,
Mel Gorman66199712012-01-12 17:19:41 -08002578 &deferred_compaction,
2579 &did_some_progress);
Mel Gorman3e7d3442011-01-13 15:45:56 -08002580 if (page)
2581 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582 }
2583
2584nopage:
Dave Hansena238ab52011-05-24 17:12:16 -07002585 warn_alloc_failed(gfp_mask, order, NULL);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002586 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587got_pg:
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002588 if (kmemcheck_enabled)
2589 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002590
Mel Gorman072bb0a2012-07-31 16:43:58 -07002591 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592}
Mel Gorman11e33f62009-06-16 15:31:57 -07002593
2594/*
2595 * This is the 'heart' of the zoned buddy allocator.
2596 */
2597struct page *
2598__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2599 struct zonelist *zonelist, nodemask_t *nodemask)
2600{
2601 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Mel Gorman5117f452009-06-16 15:31:59 -07002602 struct zone *preferred_zone;
Mel Gormancc9a6c82012-03-21 16:34:11 -07002603 struct page *page = NULL;
Mel Gorman3dd28262009-06-16 15:32:00 -07002604 int migratetype = allocflags_to_migratetype(gfp_mask);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002605 unsigned int cpuset_mems_cookie;
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002606 int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET;
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002607 struct mem_cgroup *memcg = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002608
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10002609 gfp_mask &= gfp_allowed_mask;
2610
Mel Gorman11e33f62009-06-16 15:31:57 -07002611 lockdep_trace_alloc(gfp_mask);
2612
2613 might_sleep_if(gfp_mask & __GFP_WAIT);
2614
2615 if (should_fail_alloc_page(gfp_mask, order))
2616 return NULL;
2617
2618 /*
2619 * Check the zones suitable for the gfp_mask contain at least one
2620 * valid zone. It's possible to have an empty zonelist as a result
2621 * of GFP_THISNODE and a memoryless node
2622 */
2623 if (unlikely(!zonelist->_zonerefs->zone))
2624 return NULL;
2625
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002626 /*
2627 * Will only have any effect when __GFP_KMEMCG is set. This is
2628 * verified in the (always inline) callee
2629 */
2630 if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
2631 return NULL;
2632
Mel Gormancc9a6c82012-03-21 16:34:11 -07002633retry_cpuset:
2634 cpuset_mems_cookie = get_mems_allowed();
2635
Mel Gorman5117f452009-06-16 15:31:59 -07002636 /* The preferred zone is used for statistics later */
David Rientjesf33261d2011-01-25 15:07:20 -08002637 first_zones_zonelist(zonelist, high_zoneidx,
2638 nodemask ? : &cpuset_current_mems_allowed,
2639 &preferred_zone);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002640 if (!preferred_zone)
2641 goto out;
Mel Gorman5117f452009-06-16 15:31:59 -07002642
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002643#ifdef CONFIG_CMA
2644 if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2645 alloc_flags |= ALLOC_CMA;
2646#endif
Mel Gorman5117f452009-06-16 15:31:59 -07002647 /* First allocation attempt */
Mel Gorman11e33f62009-06-16 15:31:57 -07002648 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002649 zonelist, high_zoneidx, alloc_flags,
Mel Gorman3dd28262009-06-16 15:32:00 -07002650 preferred_zone, migratetype);
Ming Lei21caf2f2013-02-22 16:34:08 -08002651 if (unlikely(!page)) {
2652 /*
2653 * Runtime PM, block IO and its error handling path
2654 * can deadlock because I/O on the device might not
2655 * complete.
2656 */
2657 gfp_mask = memalloc_noio_flags(gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002658 page = __alloc_pages_slowpath(gfp_mask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002659 zonelist, high_zoneidx, nodemask,
Mel Gorman3dd28262009-06-16 15:32:00 -07002660 preferred_zone, migratetype);
Ming Lei21caf2f2013-02-22 16:34:08 -08002661 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002662
Mel Gorman4b4f2782009-09-21 17:02:41 -07002663 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002664
2665out:
2666 /*
2667 * When updating a task's mems_allowed, it is possible to race with
2668 * parallel threads in such a way that an allocation can fail while
2669 * the mask is being updated. If a page allocation is about to fail,
2670 * check if the cpuset changed during allocation and if so, retry.
2671 */
2672 if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page))
2673 goto retry_cpuset;
2674
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002675 memcg_kmem_commit_charge(page, memcg, order);
2676
Mel Gorman11e33f62009-06-16 15:31:57 -07002677 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678}
Mel Gormand2391712009-06-16 15:31:52 -07002679EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680
2681/*
2682 * Common helper functions.
2683 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08002684unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685{
Akinobu Mita945a1112009-09-21 17:01:47 -07002686 struct page *page;
2687
2688 /*
2689 * __get_free_pages() returns a 32-bit address, which cannot represent
2690 * a highmem page
2691 */
2692 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2693
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694 page = alloc_pages(gfp_mask, order);
2695 if (!page)
2696 return 0;
2697 return (unsigned long) page_address(page);
2698}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699EXPORT_SYMBOL(__get_free_pages);
2700
Harvey Harrison920c7a52008-02-04 22:29:26 -08002701unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702{
Akinobu Mita945a1112009-09-21 17:01:47 -07002703 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705EXPORT_SYMBOL(get_zeroed_page);
2706
Harvey Harrison920c7a52008-02-04 22:29:26 -08002707void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708{
Nick Pigginb5810032005-10-29 18:16:12 -07002709 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710 if (order == 0)
Li Hongfc916682010-03-05 13:41:54 -08002711 free_hot_cold_page(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712 else
2713 __free_pages_ok(page, order);
2714 }
2715}
2716
2717EXPORT_SYMBOL(__free_pages);
2718
Harvey Harrison920c7a52008-02-04 22:29:26 -08002719void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720{
2721 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07002722 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723 __free_pages(virt_to_page((void *)addr), order);
2724 }
2725}
2726
2727EXPORT_SYMBOL(free_pages);
2728
Glauber Costa6a1a0d32012-12-18 14:22:00 -08002729/*
2730 * __free_memcg_kmem_pages and free_memcg_kmem_pages will free
2731 * pages allocated with __GFP_KMEMCG.
2732 *
2733 * Those pages are accounted to a particular memcg, embedded in the
2734 * corresponding page_cgroup. To avoid adding a hit in the allocator to search
2735 * for that information only to find out that it is NULL for users who have no
2736 * interest in that whatsoever, we provide these functions.
2737 *
2738 * The caller knows better which flags it relies on.
2739 */
2740void __free_memcg_kmem_pages(struct page *page, unsigned int order)
2741{
2742 memcg_kmem_uncharge_pages(page, order);
2743 __free_pages(page, order);
2744}
2745
2746void free_memcg_kmem_pages(unsigned long addr, unsigned int order)
2747{
2748 if (addr != 0) {
2749 VM_BUG_ON(!virt_addr_valid((void *)addr));
2750 __free_memcg_kmem_pages(virt_to_page((void *)addr), order);
2751 }
2752}
2753
Andi Kleenee85c2e2011-05-11 15:13:34 -07002754static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
2755{
2756 if (addr) {
2757 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2758 unsigned long used = addr + PAGE_ALIGN(size);
2759
2760 split_page(virt_to_page((void *)addr), order);
2761 while (used < alloc_end) {
2762 free_page(used);
2763 used += PAGE_SIZE;
2764 }
2765 }
2766 return (void *)addr;
2767}
2768
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002769/**
2770 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2771 * @size: the number of bytes to allocate
2772 * @gfp_mask: GFP flags for the allocation
2773 *
2774 * This function is similar to alloc_pages(), except that it allocates the
2775 * minimum number of pages to satisfy the request. alloc_pages() can only
2776 * allocate memory in power-of-two pages.
2777 *
2778 * This function is also limited by MAX_ORDER.
2779 *
2780 * Memory allocated by this function must be released by free_pages_exact().
2781 */
2782void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2783{
2784 unsigned int order = get_order(size);
2785 unsigned long addr;
2786
2787 addr = __get_free_pages(gfp_mask, order);
Andi Kleenee85c2e2011-05-11 15:13:34 -07002788 return make_alloc_exact(addr, order, size);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002789}
2790EXPORT_SYMBOL(alloc_pages_exact);
2791
2792/**
Andi Kleenee85c2e2011-05-11 15:13:34 -07002793 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2794 * pages on a node.
Randy Dunlapb5e6ab52011-05-16 13:16:54 -07002795 * @nid: the preferred node ID where memory should be allocated
Andi Kleenee85c2e2011-05-11 15:13:34 -07002796 * @size: the number of bytes to allocate
2797 * @gfp_mask: GFP flags for the allocation
2798 *
2799 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2800 * back.
2801 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2802 * but is not exact.
2803 */
2804void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
2805{
2806 unsigned order = get_order(size);
2807 struct page *p = alloc_pages_node(nid, gfp_mask, order);
2808 if (!p)
2809 return NULL;
2810 return make_alloc_exact((unsigned long)page_address(p), order, size);
2811}
2812EXPORT_SYMBOL(alloc_pages_exact_nid);
2813
2814/**
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002815 * free_pages_exact - release memory allocated via alloc_pages_exact()
2816 * @virt: the value returned by alloc_pages_exact.
2817 * @size: size of allocation, same value as passed to alloc_pages_exact().
2818 *
2819 * Release the memory allocated by a previous call to alloc_pages_exact.
2820 */
2821void free_pages_exact(void *virt, size_t size)
2822{
2823 unsigned long addr = (unsigned long)virt;
2824 unsigned long end = addr + PAGE_ALIGN(size);
2825
2826 while (addr < end) {
2827 free_page(addr);
2828 addr += PAGE_SIZE;
2829 }
2830}
2831EXPORT_SYMBOL(free_pages_exact);
2832
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08002833/**
2834 * nr_free_zone_pages - count number of pages beyond high watermark
2835 * @offset: The zone index of the highest zone
2836 *
2837 * nr_free_zone_pages() counts the number of counts pages which are beyond the
2838 * high watermark within all zones at or below a given zone index. For each
2839 * zone, the number of pages is calculated as:
2840 * present_pages - high_pages
2841 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08002842static unsigned long nr_free_zone_pages(int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843{
Mel Gormandd1a2392008-04-28 02:12:17 -07002844 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002845 struct zone *zone;
2846
Martin J. Blighe310fd42005-07-29 22:59:18 -07002847 /* Just pick one node, since fallback list is circular */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08002848 unsigned long sum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849
Mel Gorman0e884602008-04-28 02:12:14 -07002850 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851
Mel Gorman54a6eb52008-04-28 02:12:16 -07002852 for_each_zone_zonelist(zone, z, zonelist, offset) {
Jiang Liub40da042013-02-22 16:33:52 -08002853 unsigned long size = zone->managed_pages;
Mel Gorman41858962009-06-16 15:32:12 -07002854 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07002855 if (size > high)
2856 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857 }
2858
2859 return sum;
2860}
2861
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08002862/**
2863 * nr_free_buffer_pages - count number of pages beyond high watermark
2864 *
2865 * nr_free_buffer_pages() counts the number of pages which are beyond the high
2866 * watermark within ZONE_DMA and ZONE_NORMAL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08002868unsigned long nr_free_buffer_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002869{
Al Viroaf4ca452005-10-21 02:55:38 -04002870 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871}
Meelap Shahc2f1a552007-07-17 04:04:39 -07002872EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002873
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08002874/**
2875 * nr_free_pagecache_pages - count number of pages beyond high watermark
2876 *
2877 * nr_free_pagecache_pages() counts the number of pages which are beyond the
2878 * high watermark within all zones.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08002880unsigned long nr_free_pagecache_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881{
Mel Gorman2a1e2742007-07-17 04:03:12 -07002882 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002884
2885static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002887 if (IS_ENABLED(CONFIG_NUMA))
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08002888 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891void si_meminfo(struct sysinfo *val)
2892{
2893 val->totalram = totalram_pages;
2894 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002895 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897 val->totalhigh = totalhigh_pages;
2898 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899 val->mem_unit = PAGE_SIZE;
2900}
2901
2902EXPORT_SYMBOL(si_meminfo);
2903
2904#ifdef CONFIG_NUMA
2905void si_meminfo_node(struct sysinfo *val, int nid)
2906{
2907 pg_data_t *pgdat = NODE_DATA(nid);
2908
2909 val->totalram = pgdat->node_present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002910 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002911#ifdef CONFIG_HIGHMEM
Jiang Liub40da042013-02-22 16:33:52 -08002912 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002913 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2914 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002915#else
2916 val->totalhigh = 0;
2917 val->freehigh = 0;
2918#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919 val->mem_unit = PAGE_SIZE;
2920}
2921#endif
2922
David Rientjesddd588b2011-03-22 16:30:46 -07002923/*
David Rientjes7bf02ea2011-05-24 17:11:16 -07002924 * Determine whether the node should be displayed or not, depending on whether
2925 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
David Rientjesddd588b2011-03-22 16:30:46 -07002926 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07002927bool skip_free_areas_node(unsigned int flags, int nid)
David Rientjesddd588b2011-03-22 16:30:46 -07002928{
2929 bool ret = false;
Mel Gormancc9a6c82012-03-21 16:34:11 -07002930 unsigned int cpuset_mems_cookie;
David Rientjesddd588b2011-03-22 16:30:46 -07002931
2932 if (!(flags & SHOW_MEM_FILTER_NODES))
2933 goto out;
2934
Mel Gormancc9a6c82012-03-21 16:34:11 -07002935 do {
2936 cpuset_mems_cookie = get_mems_allowed();
2937 ret = !node_isset(nid, cpuset_current_mems_allowed);
2938 } while (!put_mems_allowed(cpuset_mems_cookie));
David Rientjesddd588b2011-03-22 16:30:46 -07002939out:
2940 return ret;
2941}
2942
Linus Torvalds1da177e2005-04-16 15:20:36 -07002943#define K(x) ((x) << (PAGE_SHIFT-10))
2944
Rabin Vincent377e4f12012-12-11 16:00:24 -08002945static void show_migration_types(unsigned char type)
2946{
2947 static const char types[MIGRATE_TYPES] = {
2948 [MIGRATE_UNMOVABLE] = 'U',
2949 [MIGRATE_RECLAIMABLE] = 'E',
2950 [MIGRATE_MOVABLE] = 'M',
2951 [MIGRATE_RESERVE] = 'R',
2952#ifdef CONFIG_CMA
2953 [MIGRATE_CMA] = 'C',
2954#endif
Minchan Kim194159f2013-02-22 16:33:58 -08002955#ifdef CONFIG_MEMORY_ISOLATION
Rabin Vincent377e4f12012-12-11 16:00:24 -08002956 [MIGRATE_ISOLATE] = 'I',
Minchan Kim194159f2013-02-22 16:33:58 -08002957#endif
Rabin Vincent377e4f12012-12-11 16:00:24 -08002958 };
2959 char tmp[MIGRATE_TYPES + 1];
2960 char *p = tmp;
2961 int i;
2962
2963 for (i = 0; i < MIGRATE_TYPES; i++) {
2964 if (type & (1 << i))
2965 *p++ = types[i];
2966 }
2967
2968 *p = '\0';
2969 printk("(%s) ", tmp);
2970}
2971
Linus Torvalds1da177e2005-04-16 15:20:36 -07002972/*
2973 * Show free area list (used inside shift_scroll-lock stuff)
2974 * We also calculate the percentage fragmentation. We do this by counting the
2975 * memory on each free list with the exception of the first item on the list.
David Rientjesddd588b2011-03-22 16:30:46 -07002976 * Suppresses nodes that are not allowed by current's cpuset if
2977 * SHOW_MEM_FILTER_NODES is passed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002978 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07002979void show_free_areas(unsigned int filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980{
Jes Sorensenc7241912006-09-27 01:50:05 -07002981 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002982 struct zone *zone;
2983
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002984 for_each_populated_zone(zone) {
David Rientjes7bf02ea2011-05-24 17:11:16 -07002985 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07002986 continue;
Jes Sorensenc7241912006-09-27 01:50:05 -07002987 show_node(zone);
2988 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989
Dave Jones6b482c62005-11-10 15:45:56 -05002990 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002991 struct per_cpu_pageset *pageset;
2992
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002993 pageset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002994
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002995 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2996 cpu, pageset->pcp.high,
2997 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998 }
2999 }
3000
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003001 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3002 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003003 " unevictable:%lu"
Andrew Mortonb76146e2009-10-26 16:49:52 -07003004 " dirty:%lu writeback:%lu unstable:%lu\n"
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07003005 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003006 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3007 " free_cma:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003008 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003009 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003010 global_page_state(NR_ISOLATED_ANON),
3011 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003012 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003013 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003014 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07003015 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07003016 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07003017 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08003018 global_page_state(NR_FREE_PAGES),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07003019 global_page_state(NR_SLAB_RECLAIMABLE),
3020 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07003021 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003022 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08003023 global_page_state(NR_PAGETABLE),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003024 global_page_state(NR_BOUNCE),
3025 global_page_state(NR_FREE_CMA_PAGES));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003026
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003027 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003028 int i;
3029
David Rientjes7bf02ea2011-05-24 17:11:16 -07003030 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003031 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003032 show_node(zone);
3033 printk("%s"
3034 " free:%lukB"
3035 " min:%lukB"
3036 " low:%lukB"
3037 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003038 " active_anon:%lukB"
3039 " inactive_anon:%lukB"
3040 " active_file:%lukB"
3041 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003042 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003043 " isolated(anon):%lukB"
3044 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045 " present:%lukB"
Jiang Liu9feedc92012-12-12 13:52:12 -08003046 " managed:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003047 " mlocked:%lukB"
3048 " dirty:%lukB"
3049 " writeback:%lukB"
3050 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003051 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003052 " slab_reclaimable:%lukB"
3053 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003054 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003055 " pagetables:%lukB"
3056 " unstable:%lukB"
3057 " bounce:%lukB"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003058 " free_cma:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003059 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060 " pages_scanned:%lu"
3061 " all_unreclaimable? %s"
3062 "\n",
3063 zone->name,
Mel Gorman88f5acf2011-01-13 15:45:41 -08003064 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07003065 K(min_wmark_pages(zone)),
3066 K(low_wmark_pages(zone)),
3067 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003068 K(zone_page_state(zone, NR_ACTIVE_ANON)),
3069 K(zone_page_state(zone, NR_INACTIVE_ANON)),
3070 K(zone_page_state(zone, NR_ACTIVE_FILE)),
3071 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003072 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003073 K(zone_page_state(zone, NR_ISOLATED_ANON)),
3074 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075 K(zone->present_pages),
Jiang Liu9feedc92012-12-12 13:52:12 -08003076 K(zone->managed_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003077 K(zone_page_state(zone, NR_MLOCK)),
3078 K(zone_page_state(zone, NR_FILE_DIRTY)),
3079 K(zone_page_state(zone, NR_WRITEBACK)),
3080 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003081 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003082 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3083 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003084 zone_page_state(zone, NR_KERNEL_STACK) *
3085 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003086 K(zone_page_state(zone, NR_PAGETABLE)),
3087 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3088 K(zone_page_state(zone, NR_BOUNCE)),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003089 K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003090 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091 zone->pages_scanned,
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -08003092 (zone->all_unreclaimable ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093 );
3094 printk("lowmem_reserve[]:");
3095 for (i = 0; i < MAX_NR_ZONES; i++)
3096 printk(" %lu", zone->lowmem_reserve[i]);
3097 printk("\n");
3098 }
3099
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003100 for_each_populated_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003101 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003102 unsigned char types[MAX_ORDER];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003103
David Rientjes7bf02ea2011-05-24 17:11:16 -07003104 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003105 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003106 show_node(zone);
3107 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108
3109 spin_lock_irqsave(&zone->lock, flags);
3110 for (order = 0; order < MAX_ORDER; order++) {
Rabin Vincent377e4f12012-12-11 16:00:24 -08003111 struct free_area *area = &zone->free_area[order];
3112 int type;
3113
3114 nr[order] = area->nr_free;
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003115 total += nr[order] << order;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003116
3117 types[order] = 0;
3118 for (type = 0; type < MIGRATE_TYPES; type++) {
3119 if (!list_empty(&area->free_list[type]))
3120 types[order] |= 1 << type;
3121 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003122 }
3123 spin_unlock_irqrestore(&zone->lock, flags);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003124 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003125 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003126 if (nr[order])
3127 show_migration_types(types[order]);
3128 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003129 printk("= %lukB\n", K(total));
3130 }
3131
David Rientjes949f7ec2013-04-29 15:07:48 -07003132 hugetlb_show_meminfo();
3133
Larry Woodmane6f36022008-02-04 22:29:30 -08003134 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
3135
Linus Torvalds1da177e2005-04-16 15:20:36 -07003136 show_swap_cache_info();
3137}
3138
Mel Gorman19770b32008-04-28 02:12:18 -07003139static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
3140{
3141 zoneref->zone = zone;
3142 zoneref->zone_idx = zone_idx(zone);
3143}
3144
Linus Torvalds1da177e2005-04-16 15:20:36 -07003145/*
3146 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08003147 *
3148 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003149 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003150static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
3151 int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003152{
Christoph Lameter1a932052006-01-06 00:11:16 -08003153 struct zone *zone;
3154
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003155 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003156 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08003157
3158 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003159 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08003160 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08003161 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07003162 zoneref_set_zone(zone,
3163 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08003164 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003165 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08003166
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003167 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08003168 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003169}
3170
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003171
3172/*
3173 * zonelist_order:
3174 * 0 = automatic detection of better ordering.
3175 * 1 = order by ([node] distance, -zonetype)
3176 * 2 = order by (-zonetype, [node] distance)
3177 *
3178 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3179 * the same zonelist. So only NUMA can configure this param.
3180 */
3181#define ZONELIST_ORDER_DEFAULT 0
3182#define ZONELIST_ORDER_NODE 1
3183#define ZONELIST_ORDER_ZONE 2
3184
3185/* zonelist order in the kernel.
3186 * set_zonelist_order() will set this to NODE or ZONE.
3187 */
3188static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
3189static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
3190
3191
Linus Torvalds1da177e2005-04-16 15:20:36 -07003192#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003193/* The value user specified ....changed by config */
3194static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3195/* string for sysctl */
3196#define NUMA_ZONELIST_ORDER_LEN 16
3197char numa_zonelist_order[16] = "default";
3198
3199/*
3200 * interface for configure zonelist ordering.
3201 * command line option "numa_zonelist_order"
3202 * = "[dD]efault - default, automatic configuration.
3203 * = "[nN]ode - order by node locality, then by zone within node
3204 * = "[zZ]one - order by zone, then by locality within zone
3205 */
3206
3207static int __parse_numa_zonelist_order(char *s)
3208{
3209 if (*s == 'd' || *s == 'D') {
3210 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3211 } else if (*s == 'n' || *s == 'N') {
3212 user_zonelist_order = ZONELIST_ORDER_NODE;
3213 } else if (*s == 'z' || *s == 'Z') {
3214 user_zonelist_order = ZONELIST_ORDER_ZONE;
3215 } else {
3216 printk(KERN_WARNING
3217 "Ignoring invalid numa_zonelist_order value: "
3218 "%s\n", s);
3219 return -EINVAL;
3220 }
3221 return 0;
3222}
3223
3224static __init int setup_numa_zonelist_order(char *s)
3225{
Volodymyr G. Lukiianykecb256f2011-01-13 15:46:26 -08003226 int ret;
3227
3228 if (!s)
3229 return 0;
3230
3231 ret = __parse_numa_zonelist_order(s);
3232 if (ret == 0)
3233 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
3234
3235 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003236}
3237early_param("numa_zonelist_order", setup_numa_zonelist_order);
3238
3239/*
3240 * sysctl handler for numa_zonelist_order
3241 */
3242int numa_zonelist_order_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003243 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003244 loff_t *ppos)
3245{
3246 char saved_string[NUMA_ZONELIST_ORDER_LEN];
3247 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01003248 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003249
Andi Kleen443c6f12009-12-23 21:00:47 +01003250 mutex_lock(&zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003251 if (write)
Andi Kleen443c6f12009-12-23 21:00:47 +01003252 strcpy(saved_string, (char*)table->data);
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003253 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003254 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01003255 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003256 if (write) {
3257 int oldval = user_zonelist_order;
3258 if (__parse_numa_zonelist_order((char*)table->data)) {
3259 /*
3260 * bogus value. restore saved string
3261 */
3262 strncpy((char*)table->data, saved_string,
3263 NUMA_ZONELIST_ORDER_LEN);
3264 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003265 } else if (oldval != user_zonelist_order) {
3266 mutex_lock(&zonelists_mutex);
Jiang Liu9adb62a2012-07-31 16:43:28 -07003267 build_all_zonelists(NULL, NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003268 mutex_unlock(&zonelists_mutex);
3269 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003270 }
Andi Kleen443c6f12009-12-23 21:00:47 +01003271out:
3272 mutex_unlock(&zl_order_mutex);
3273 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003274}
3275
3276
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003277#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003278static int node_load[MAX_NUMNODES];
3279
Linus Torvalds1da177e2005-04-16 15:20:36 -07003280/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07003281 * 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 -07003282 * @node: node whose fallback list we're appending
3283 * @used_node_mask: nodemask_t of already used nodes
3284 *
3285 * We use a number of factors to determine which is the next node that should
3286 * appear on a given node's fallback list. The node should not have appeared
3287 * already in @node's fallback list, and it should be the next closest node
3288 * according to the distance array (which contains arbitrary distance values
3289 * from each node to each node in the system), and should also prefer nodes
3290 * with no CPUs, since presumably they'll have very little allocation pressure
3291 * on them otherwise.
3292 * It returns -1 if no node is found.
3293 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003294static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003295{
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01003296 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003297 int min_val = INT_MAX;
David Rientjes00ef2d22013-02-22 16:35:36 -08003298 int best_node = NUMA_NO_NODE;
Rusty Russella70f7302009-03-13 14:49:46 +10303299 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003300
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01003301 /* Use the local node if we haven't already */
3302 if (!node_isset(node, *used_node_mask)) {
3303 node_set(node, *used_node_mask);
3304 return node;
3305 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003306
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08003307 for_each_node_state(n, N_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003308
3309 /* Don't want a node to appear more than once */
3310 if (node_isset(n, *used_node_mask))
3311 continue;
3312
Linus Torvalds1da177e2005-04-16 15:20:36 -07003313 /* Use the distance array to find the distance */
3314 val = node_distance(node, n);
3315
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01003316 /* Penalize nodes under us ("prefer the next node") */
3317 val += (n < node);
3318
Linus Torvalds1da177e2005-04-16 15:20:36 -07003319 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10303320 tmp = cpumask_of_node(n);
3321 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003322 val += PENALTY_FOR_NODE_WITH_CPUS;
3323
3324 /* Slight preference for less loaded node */
3325 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
3326 val += node_load[n];
3327
3328 if (val < min_val) {
3329 min_val = val;
3330 best_node = n;
3331 }
3332 }
3333
3334 if (best_node >= 0)
3335 node_set(best_node, *used_node_mask);
3336
3337 return best_node;
3338}
3339
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003340
3341/*
3342 * Build zonelists ordered by node and zones within node.
3343 * This results in maximum locality--normal zone overflows into local
3344 * DMA zone, if any--but risks exhausting DMA zone.
3345 */
3346static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003347{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003348 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003349 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003350
Mel Gorman54a6eb52008-04-28 02:12:16 -07003351 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07003352 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07003353 ;
3354 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3355 MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07003356 zonelist->_zonerefs[j].zone = NULL;
3357 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003358}
3359
3360/*
Christoph Lameter523b9452007-10-16 01:25:37 -07003361 * Build gfp_thisnode zonelists
3362 */
3363static void build_thisnode_zonelists(pg_data_t *pgdat)
3364{
Christoph Lameter523b9452007-10-16 01:25:37 -07003365 int j;
3366 struct zonelist *zonelist;
3367
Mel Gorman54a6eb52008-04-28 02:12:16 -07003368 zonelist = &pgdat->node_zonelists[1];
3369 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07003370 zonelist->_zonerefs[j].zone = NULL;
3371 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07003372}
3373
3374/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003375 * Build zonelists ordered by zone and nodes within zones.
3376 * This results in conserving DMA zone[s] until all Normal memory is
3377 * exhausted, but results in overflowing to remote node while memory
3378 * may still exist in local DMA zone.
3379 */
3380static int node_order[MAX_NUMNODES];
3381
3382static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
3383{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003384 int pos, j, node;
3385 int zone_type; /* needs to be signed */
3386 struct zone *z;
3387 struct zonelist *zonelist;
3388
Mel Gorman54a6eb52008-04-28 02:12:16 -07003389 zonelist = &pgdat->node_zonelists[0];
3390 pos = 0;
3391 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
3392 for (j = 0; j < nr_nodes; j++) {
3393 node = node_order[j];
3394 z = &NODE_DATA(node)->node_zones[zone_type];
3395 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07003396 zoneref_set_zone(z,
3397 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07003398 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003399 }
3400 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003401 }
Mel Gormandd1a2392008-04-28 02:12:17 -07003402 zonelist->_zonerefs[pos].zone = NULL;
3403 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003404}
3405
3406static int default_zonelist_order(void)
3407{
3408 int nid, zone_type;
3409 unsigned long low_kmem_size,total_size;
3410 struct zone *z;
3411 int average_size;
3412 /*
Thomas Weber88393162010-03-16 11:47:56 +01003413 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003414 * If they are really small and used heavily, the system can fall
3415 * into OOM very easily.
David Rientjese325c902010-05-24 14:32:13 -07003416 * This function detect ZONE_DMA/DMA32 size and configures zone order.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003417 */
3418 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
3419 low_kmem_size = 0;
3420 total_size = 0;
3421 for_each_online_node(nid) {
3422 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3423 z = &NODE_DATA(nid)->node_zones[zone_type];
3424 if (populated_zone(z)) {
3425 if (zone_type < ZONE_NORMAL)
3426 low_kmem_size += z->present_pages;
3427 total_size += z->present_pages;
David Rientjese325c902010-05-24 14:32:13 -07003428 } else if (zone_type == ZONE_NORMAL) {
3429 /*
3430 * If any node has only lowmem, then node order
3431 * is preferred to allow kernel allocations
3432 * locally; otherwise, they can easily infringe
3433 * on other nodes when there is an abundance of
3434 * lowmem available to allocate from.
3435 */
3436 return ZONELIST_ORDER_NODE;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003437 }
3438 }
3439 }
3440 if (!low_kmem_size || /* there are no DMA area. */
3441 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
3442 return ZONELIST_ORDER_NODE;
3443 /*
3444 * look into each node's config.
3445 * If there is a node whose DMA/DMA32 memory is very big area on
3446 * local memory, NODE_ORDER may be suitable.
3447 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003448 average_size = total_size /
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08003449 (nodes_weight(node_states[N_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003450 for_each_online_node(nid) {
3451 low_kmem_size = 0;
3452 total_size = 0;
3453 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3454 z = &NODE_DATA(nid)->node_zones[zone_type];
3455 if (populated_zone(z)) {
3456 if (zone_type < ZONE_NORMAL)
3457 low_kmem_size += z->present_pages;
3458 total_size += z->present_pages;
3459 }
3460 }
3461 if (low_kmem_size &&
3462 total_size > average_size && /* ignore small node */
3463 low_kmem_size > total_size * 70/100)
3464 return ZONELIST_ORDER_NODE;
3465 }
3466 return ZONELIST_ORDER_ZONE;
3467}
3468
3469static void set_zonelist_order(void)
3470{
3471 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
3472 current_zonelist_order = default_zonelist_order();
3473 else
3474 current_zonelist_order = user_zonelist_order;
3475}
3476
3477static void build_zonelists(pg_data_t *pgdat)
3478{
3479 int j, node, load;
3480 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003481 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003482 int local_node, prev_node;
3483 struct zonelist *zonelist;
3484 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003485
3486 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07003487 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003488 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07003489 zonelist->_zonerefs[0].zone = NULL;
3490 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003491 }
3492
3493 /* NUMA-aware ordering of nodes */
3494 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003495 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003496 prev_node = local_node;
3497 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003498
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003499 memset(node_order, 0, sizeof(node_order));
3500 j = 0;
3501
Linus Torvalds1da177e2005-04-16 15:20:36 -07003502 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
3503 /*
3504 * We don't want to pressure a particular node.
3505 * So adding penalty to the first node in same
3506 * distance group to make it round-robin.
3507 */
David Rientjes957f8222012-10-08 16:33:24 -07003508 if (node_distance(local_node, node) !=
3509 node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003510 node_load[node] = load;
3511
Linus Torvalds1da177e2005-04-16 15:20:36 -07003512 prev_node = node;
3513 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003514 if (order == ZONELIST_ORDER_NODE)
3515 build_zonelists_in_node_order(pgdat, node);
3516 else
3517 node_order[j++] = node; /* remember order */
3518 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003519
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003520 if (order == ZONELIST_ORDER_ZONE) {
3521 /* calculate node order -- i.e., DMA last! */
3522 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003523 }
Christoph Lameter523b9452007-10-16 01:25:37 -07003524
3525 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003526}
3527
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003528/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003529static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003530{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003531 struct zonelist *zonelist;
3532 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07003533 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003534
Mel Gorman54a6eb52008-04-28 02:12:16 -07003535 zonelist = &pgdat->node_zonelists[0];
3536 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3537 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07003538 for (z = zonelist->_zonerefs; z->zone; z++)
3539 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003540}
3541
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003542#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3543/*
3544 * Return node id of node used for "local" allocations.
3545 * I.e., first node id of first zone in arg node's generic zonelist.
3546 * Used for initializing percpu 'numa_mem', which is used primarily
3547 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3548 */
3549int local_memory_node(int node)
3550{
3551 struct zone *zone;
3552
3553 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3554 gfp_zone(GFP_KERNEL),
3555 NULL,
3556 &zone);
3557 return zone->node;
3558}
3559#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003560
Linus Torvalds1da177e2005-04-16 15:20:36 -07003561#else /* CONFIG_NUMA */
3562
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003563static void set_zonelist_order(void)
3564{
3565 current_zonelist_order = ZONELIST_ORDER_ZONE;
3566}
3567
3568static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003569{
Christoph Lameter19655d32006-09-25 23:31:19 -07003570 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003571 enum zone_type j;
3572 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003573
3574 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003575
Mel Gorman54a6eb52008-04-28 02:12:16 -07003576 zonelist = &pgdat->node_zonelists[0];
3577 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003578
Mel Gorman54a6eb52008-04-28 02:12:16 -07003579 /*
3580 * Now we build the zonelist so that it contains the zones
3581 * of all the other nodes.
3582 * We don't want to pressure a particular node, so when
3583 * building the zones for node N, we make sure that the
3584 * zones coming right after the local ones are those from
3585 * node N+1 (modulo N)
3586 */
3587 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3588 if (!node_online(node))
3589 continue;
3590 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3591 MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003592 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07003593 for (node = 0; node < local_node; node++) {
3594 if (!node_online(node))
3595 continue;
3596 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3597 MAX_NR_ZONES - 1);
3598 }
3599
Mel Gormandd1a2392008-04-28 02:12:17 -07003600 zonelist->_zonerefs[j].zone = NULL;
3601 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003602}
3603
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003604/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003605static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003606{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003607 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003608}
3609
Linus Torvalds1da177e2005-04-16 15:20:36 -07003610#endif /* CONFIG_NUMA */
3611
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003612/*
3613 * Boot pageset table. One per cpu which is going to be used for all
3614 * zones and all nodes. The parameters will be set in such a way
3615 * that an item put on a list will immediately be handed over to
3616 * the buddy list. This is safe since pageset manipulation is done
3617 * with interrupts disabled.
3618 *
3619 * The boot_pagesets must be kept even after bootup is complete for
3620 * unused processors and/or zones. They do play a role for bootstrapping
3621 * hotplugged processors.
3622 *
3623 * zoneinfo_show() and maybe other functions do
3624 * not check if the processor is online before following the pageset pointer.
3625 * Other parts of the kernel may not check if the zone is available.
3626 */
3627static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3628static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07003629static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003630
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003631/*
3632 * Global mutex to protect against size modification of zonelists
3633 * as well as to serialize pageset setup for the new populated zone.
3634 */
3635DEFINE_MUTEX(zonelists_mutex);
3636
Rusty Russell9b1a4d32008-07-28 12:16:30 -05003637/* return values int ....just for stop_machine() */
Jiang Liu4ed7e022012-07-31 16:43:35 -07003638static int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003639{
Yasunori Goto68113782006-06-23 02:03:11 -07003640 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003641 int cpu;
Jiang Liu9adb62a2012-07-31 16:43:28 -07003642 pg_data_t *self = data;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003643
Bo Liu7f9cfb32009-08-18 14:11:19 -07003644#ifdef CONFIG_NUMA
3645 memset(node_load, 0, sizeof(node_load));
3646#endif
Jiang Liu9adb62a2012-07-31 16:43:28 -07003647
3648 if (self && !node_online(self->node_id)) {
3649 build_zonelists(self);
3650 build_zonelist_cache(self);
3651 }
3652
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003653 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07003654 pg_data_t *pgdat = NODE_DATA(nid);
3655
3656 build_zonelists(pgdat);
3657 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003658 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003659
3660 /*
3661 * Initialize the boot_pagesets that are going to be used
3662 * for bootstrapping processors. The real pagesets for
3663 * each zone will be allocated later when the per cpu
3664 * allocator is available.
3665 *
3666 * boot_pagesets are used also for bootstrapping offline
3667 * cpus if the system is already booted because the pagesets
3668 * are needed to initialize allocators on a specific cpu too.
3669 * F.e. the percpu allocator needs the page allocator which
3670 * needs the percpu allocator in order to allocate its pagesets
3671 * (a chicken-egg dilemma).
3672 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003673 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003674 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3675
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003676#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3677 /*
3678 * We now know the "local memory node" for each node--
3679 * i.e., the node of the first zone in the generic zonelist.
3680 * Set up numa_mem percpu variable for on-line cpus. During
3681 * boot, only the boot cpu should be on-line; we'll init the
3682 * secondary cpus' numa_mem as they come on-line. During
3683 * node/memory hotplug, we'll fixup all on-line cpus.
3684 */
3685 if (cpu_online(cpu))
3686 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3687#endif
3688 }
3689
Yasunori Goto68113782006-06-23 02:03:11 -07003690 return 0;
3691}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003692
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003693/*
3694 * Called with zonelists_mutex held always
3695 * unless system_state == SYSTEM_BOOTING.
3696 */
Jiang Liu9adb62a2012-07-31 16:43:28 -07003697void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
Yasunori Goto68113782006-06-23 02:03:11 -07003698{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003699 set_zonelist_order();
3700
Yasunori Goto68113782006-06-23 02:03:11 -07003701 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07003702 __build_all_zonelists(NULL);
Mel Gorman68ad8df2008-07-23 21:26:52 -07003703 mminit_verify_zonelist();
Yasunori Goto68113782006-06-23 02:03:11 -07003704 cpuset_init_current_mems_allowed();
3705 } else {
Simon Arlott183ff222007-10-20 01:27:18 +02003706 /* we have to stop all cpus to guarantee there is no user
Yasunori Goto68113782006-06-23 02:03:11 -07003707 of zonelist */
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003708#ifdef CONFIG_MEMORY_HOTPLUG
Jiang Liu9adb62a2012-07-31 16:43:28 -07003709 if (zone)
3710 setup_zone_pageset(zone);
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003711#endif
Jiang Liu9adb62a2012-07-31 16:43:28 -07003712 stop_machine(__build_all_zonelists, pgdat, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07003713 /* cpuset refresh routine should be here */
3714 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07003715 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003716 /*
3717 * Disable grouping by mobility if the number of pages in the
3718 * system is too low to allow the mechanism to work. It would be
3719 * more accurate, but expensive to check per-zone. This check is
3720 * made on memory-hotadd so a system can start with mobility
3721 * disabled and enable it later
3722 */
Mel Gormand9c23402007-10-16 01:26:01 -07003723 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003724 page_group_by_mobility_disabled = 1;
3725 else
3726 page_group_by_mobility_disabled = 0;
3727
3728 printk("Built %i zonelists in %s order, mobility grouping %s. "
3729 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003730 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003731 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003732 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003733 vm_total_pages);
3734#ifdef CONFIG_NUMA
3735 printk("Policy zone: %s\n", zone_names[policy_zone]);
3736#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003737}
3738
3739/*
3740 * Helper functions to size the waitqueue hash table.
3741 * Essentially these want to choose hash table sizes sufficiently
3742 * large so that collisions trying to wait on pages are rare.
3743 * But in fact, the number of active page waitqueues on typical
3744 * systems is ridiculously low, less than 200. So this is even
3745 * conservative, even though it seems large.
3746 *
3747 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3748 * waitqueues, i.e. the size of the waitq table given the number of pages.
3749 */
3750#define PAGES_PER_WAITQUEUE 256
3751
Yasunori Gotocca448f2006-06-23 02:03:10 -07003752#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07003753static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003754{
3755 unsigned long size = 1;
3756
3757 pages /= PAGES_PER_WAITQUEUE;
3758
3759 while (size < pages)
3760 size <<= 1;
3761
3762 /*
3763 * Once we have dozens or even hundreds of threads sleeping
3764 * on IO we've got bigger problems than wait queue collision.
3765 * Limit the size of the wait table to a reasonable size.
3766 */
3767 size = min(size, 4096UL);
3768
3769 return max(size, 4UL);
3770}
Yasunori Gotocca448f2006-06-23 02:03:10 -07003771#else
3772/*
3773 * A zone's size might be changed by hot-add, so it is not possible to determine
3774 * a suitable size for its wait_table. So we use the maximum size now.
3775 *
3776 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3777 *
3778 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3779 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3780 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3781 *
3782 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3783 * or more by the traditional way. (See above). It equals:
3784 *
3785 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3786 * ia64(16K page size) : = ( 8G + 4M)byte.
3787 * powerpc (64K page size) : = (32G +16M)byte.
3788 */
3789static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3790{
3791 return 4096UL;
3792}
3793#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003794
3795/*
3796 * This is an integer logarithm so that shifts can be used later
3797 * to extract the more random high bits from the multiplicative
3798 * hash function before the remainder is taken.
3799 */
3800static inline unsigned long wait_table_bits(unsigned long size)
3801{
3802 return ffz(~size);
3803}
3804
3805#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3806
Mel Gorman56fd56b2007-10-16 01:25:58 -07003807/*
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07003808 * Check if a pageblock contains reserved pages
3809 */
3810static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
3811{
3812 unsigned long pfn;
3813
3814 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
3815 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
3816 return 1;
3817 }
3818 return 0;
3819}
3820
3821/*
Mel Gormand9c23402007-10-16 01:26:01 -07003822 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07003823 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3824 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07003825 * higher will lead to a bigger reserve which will get freed as contiguous
3826 * blocks as reclaim kicks in
3827 */
3828static void setup_zone_migrate_reserve(struct zone *zone)
3829{
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07003830 unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003831 struct page *page;
Mel Gorman78986a62009-09-21 17:03:02 -07003832 unsigned long block_migratetype;
3833 int reserve;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003834
Michal Hockod02156382011-12-08 14:34:27 -08003835 /*
3836 * Get the start pfn, end pfn and the number of blocks to reserve
3837 * We have to be careful to be aligned to pageblock_nr_pages to
3838 * make sure that we always check pfn_valid for the first page in
3839 * the block.
3840 */
Mel Gorman56fd56b2007-10-16 01:25:58 -07003841 start_pfn = zone->zone_start_pfn;
Cody P Schafer108bcc92013-02-22 16:35:23 -08003842 end_pfn = zone_end_pfn(zone);
Michal Hockod02156382011-12-08 14:34:27 -08003843 start_pfn = roundup(start_pfn, pageblock_nr_pages);
Mel Gorman41858962009-06-16 15:32:12 -07003844 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07003845 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003846
Mel Gorman78986a62009-09-21 17:03:02 -07003847 /*
3848 * Reserve blocks are generally in place to help high-order atomic
3849 * allocations that are short-lived. A min_free_kbytes value that
3850 * would result in more than 2 reserve blocks for atomic allocations
3851 * is assumed to be in place to help anti-fragmentation for the
3852 * future allocation of hugepages at runtime.
3853 */
3854 reserve = min(2, reserve);
3855
Mel Gormand9c23402007-10-16 01:26:01 -07003856 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07003857 if (!pfn_valid(pfn))
3858 continue;
3859 page = pfn_to_page(pfn);
3860
Adam Litke344c7902008-09-02 14:35:38 -07003861 /* Watch out for overlapping nodes */
3862 if (page_to_nid(page) != zone_to_nid(zone))
3863 continue;
3864
Mel Gorman56fd56b2007-10-16 01:25:58 -07003865 block_migratetype = get_pageblock_migratetype(page);
3866
Mel Gorman938929f2012-01-10 15:07:14 -08003867 /* Only test what is necessary when the reserves are not met */
3868 if (reserve > 0) {
3869 /*
3870 * Blocks with reserved pages will never free, skip
3871 * them.
3872 */
3873 block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
3874 if (pageblock_is_reserved(pfn, block_end_pfn))
3875 continue;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003876
Mel Gorman938929f2012-01-10 15:07:14 -08003877 /* If this block is reserved, account for it */
3878 if (block_migratetype == MIGRATE_RESERVE) {
3879 reserve--;
3880 continue;
3881 }
3882
3883 /* Suitable for reserving if this block is movable */
3884 if (block_migratetype == MIGRATE_MOVABLE) {
3885 set_pageblock_migratetype(page,
3886 MIGRATE_RESERVE);
3887 move_freepages_block(zone, page,
3888 MIGRATE_RESERVE);
3889 reserve--;
3890 continue;
3891 }
Mel Gorman56fd56b2007-10-16 01:25:58 -07003892 }
3893
3894 /*
3895 * If the reserve is met and this is a previous reserved block,
3896 * take it back
3897 */
3898 if (block_migratetype == MIGRATE_RESERVE) {
3899 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
3900 move_freepages_block(zone, page, MIGRATE_MOVABLE);
3901 }
3902 }
3903}
Mel Gormanac0e5b72007-10-16 01:25:58 -07003904
Linus Torvalds1da177e2005-04-16 15:20:36 -07003905/*
3906 * Initially all pages are reserved - free ones are freed
3907 * up by free_all_bootmem() once the early boot process is
3908 * done. Non-atomic initialization, single-pass.
3909 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003910void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08003911 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003912{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003913 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07003914 unsigned long end_pfn = start_pfn + size;
3915 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003916 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003917
Hugh Dickins22b31ee2009-01-06 14:40:09 -08003918 if (highest_memmap_pfn < end_pfn - 1)
3919 highest_memmap_pfn = end_pfn - 1;
3920
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003921 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08003922 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08003923 /*
3924 * There can be holes in boot-time mem_map[]s
3925 * handed to this function. They do not
3926 * exist on hotplugged memory.
3927 */
3928 if (context == MEMMAP_EARLY) {
3929 if (!early_pfn_valid(pfn))
3930 continue;
3931 if (!early_pfn_in_nid(pfn, nid))
3932 continue;
3933 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003934 page = pfn_to_page(pfn);
3935 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07003936 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08003937 init_page_count(page);
Mel Gorman22b751c2013-02-22 16:34:59 -08003938 page_mapcount_reset(page);
3939 page_nid_reset_last(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003940 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003941 /*
3942 * Mark the block movable so that blocks are reserved for
3943 * movable at startup. This will force kernel allocations
3944 * to reserve their blocks rather than leaking throughout
3945 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07003946 * kernel allocations are made. Later some blocks near
3947 * the start are marked MIGRATE_RESERVE by
3948 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003949 *
3950 * bitmap is created for zone's valid pfn range. but memmap
3951 * can be created for invalid pages (for alignment)
3952 * check here not to call set_pageblock_migratetype() against
3953 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003954 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003955 if ((z->zone_start_pfn <= pfn)
Cody P Schafer108bcc92013-02-22 16:35:23 -08003956 && (pfn < zone_end_pfn(z))
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003957 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07003958 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003959
Linus Torvalds1da177e2005-04-16 15:20:36 -07003960 INIT_LIST_HEAD(&page->lru);
3961#ifdef WANT_PAGE_VIRTUAL
3962 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3963 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07003964 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003965#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003966 }
3967}
3968
Andi Kleen1e548de2008-02-04 22:29:26 -08003969static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003970{
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003971 int order, t;
3972 for_each_migratetype_order(order, t) {
3973 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003974 zone->free_area[order].nr_free = 0;
3975 }
3976}
3977
3978#ifndef __HAVE_ARCH_MEMMAP_INIT
3979#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08003980 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003981#endif
3982
Jiang Liu4ed7e022012-07-31 16:43:35 -07003983static int __meminit zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003984{
David Howells3a6be872009-05-06 16:03:03 -07003985#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003986 int batch;
3987
3988 /*
3989 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07003990 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003991 *
3992 * OK, so we don't know how big the cache is. So guess.
3993 */
Jiang Liub40da042013-02-22 16:33:52 -08003994 batch = zone->managed_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07003995 if (batch * PAGE_SIZE > 512 * 1024)
3996 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003997 batch /= 4; /* We effectively *= 4 below */
3998 if (batch < 1)
3999 batch = 1;
4000
4001 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004002 * Clamp the batch to a 2^n - 1 value. Having a power
4003 * of 2 value was found to be more likely to have
4004 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004005 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004006 * For example if 2 tasks are alternately allocating
4007 * batches of pages, one task can end up with a lot
4008 * of pages of one half of the possible page colors
4009 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004010 */
David Howells91552032009-05-06 16:03:02 -07004011 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07004012
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004013 return batch;
David Howells3a6be872009-05-06 16:03:03 -07004014
4015#else
4016 /* The deferral and batching of frees should be suppressed under NOMMU
4017 * conditions.
4018 *
4019 * The problem is that NOMMU needs to be able to allocate large chunks
4020 * of contiguous memory as there's no hardware page translation to
4021 * assemble apparent contiguous memory from discontiguous pages.
4022 *
4023 * Queueing large contiguous runs of pages for batching, however,
4024 * causes the pages to actually be freed in smaller chunks. As there
4025 * can be a significant delay between the individual batches being
4026 * recycled, this leads to the once large chunks of space being
4027 * fragmented and becoming unavailable for high-order allocations.
4028 */
4029 return 0;
4030#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004031}
4032
Adrian Bunkb69a7282008-07-23 21:28:12 -07004033static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
Christoph Lameter2caaad42005-06-21 17:15:00 -07004034{
4035 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004036 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004037
Magnus Damm1c6fe942005-10-26 01:58:59 -07004038 memset(p, 0, sizeof(*p));
4039
Christoph Lameter3dfa5722008-02-04 22:29:19 -08004040 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004041 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004042 pcp->high = 6 * batch;
4043 pcp->batch = max(1UL, 1 * batch);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004044 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
4045 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07004046}
4047
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004048/*
4049 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
4050 * to the value high for the pageset p.
4051 */
4052
4053static void setup_pagelist_highmark(struct per_cpu_pageset *p,
4054 unsigned long high)
4055{
4056 struct per_cpu_pages *pcp;
4057
Christoph Lameter3dfa5722008-02-04 22:29:19 -08004058 pcp = &p->pcp;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004059 pcp->high = high;
4060 pcp->batch = max(1UL, high/4);
4061 if ((high/4) > (PAGE_SHIFT * 8))
4062 pcp->batch = PAGE_SHIFT * 8;
4063}
4064
Jiang Liu4ed7e022012-07-31 16:43:35 -07004065static void __meminit setup_zone_pageset(struct zone *zone)
Wu Fengguang319774e2010-05-24 14:32:49 -07004066{
4067 int cpu;
4068
4069 zone->pageset = alloc_percpu(struct per_cpu_pageset);
4070
4071 for_each_possible_cpu(cpu) {
4072 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4073
4074 setup_pageset(pcp, zone_batchsize(zone));
4075
4076 if (percpu_pagelist_fraction)
4077 setup_pagelist_highmark(pcp,
Jiang Liub40da042013-02-22 16:33:52 -08004078 (zone->managed_pages /
Wu Fengguang319774e2010-05-24 14:32:49 -07004079 percpu_pagelist_fraction));
4080 }
4081}
4082
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004083/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004084 * Allocate per cpu pagesets and initialize them.
4085 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07004086 */
Al Viro78d99552005-12-15 09:18:25 +00004087void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004088{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004089 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004090
Wu Fengguang319774e2010-05-24 14:32:49 -07004091 for_each_populated_zone(zone)
4092 setup_zone_pageset(zone);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004093}
4094
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004095static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07004096int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07004097{
4098 int i;
4099 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004100 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07004101
4102 /*
4103 * The per-page waitqueue mechanism uses hashed waitqueues
4104 * per zone.
4105 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07004106 zone->wait_table_hash_nr_entries =
4107 wait_table_hash_nr_entries(zone_size_pages);
4108 zone->wait_table_bits =
4109 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004110 alloc_size = zone->wait_table_hash_nr_entries
4111 * sizeof(wait_queue_head_t);
4112
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07004113 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07004114 zone->wait_table = (wait_queue_head_t *)
Yinghai Lu8f389a992011-05-11 15:13:32 -07004115 alloc_bootmem_node_nopanic(pgdat, alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004116 } else {
4117 /*
4118 * This case means that a zone whose size was 0 gets new memory
4119 * via memory hot-add.
4120 * But it may be the case that a new node was hot-added. In
4121 * this case vmalloc() will not be able to use this new node's
4122 * memory - this wait_table must be initialized to use this new
4123 * node itself as well.
4124 * To use this new node's memory, further consideration will be
4125 * necessary.
4126 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07004127 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004128 }
4129 if (!zone->wait_table)
4130 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07004131
Yasunori Goto02b694d2006-06-23 02:03:08 -07004132 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07004133 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004134
4135 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004136}
4137
Matt Tolentinoc09b4242006-01-17 07:03:44 +01004138static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07004139{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004140 /*
4141 * per cpu subsystem is not up at this point. The following code
4142 * relies on the ability of the linker to provide the
4143 * offset of a (static) per cpu variable into the per cpu area.
4144 */
4145 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07004146
Anton Blanchardf5335c02006-03-25 03:06:49 -08004147 if (zone->present_pages)
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004148 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
4149 zone->name, zone->present_pages,
4150 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07004151}
4152
Jiang Liu4ed7e022012-07-31 16:43:35 -07004153int __meminit init_currently_empty_zone(struct zone *zone,
Yasunori Goto718127c2006-06-23 02:03:10 -07004154 unsigned long zone_start_pfn,
Dave Hansena2f3aa022007-01-10 23:15:30 -08004155 unsigned long size,
4156 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07004157{
4158 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004159 int ret;
4160 ret = zone_wait_table_init(zone, size);
4161 if (ret)
4162 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07004163 pgdat->nr_zones = zone_idx(zone) + 1;
4164
Dave Hansened8ece22005-10-29 18:16:50 -07004165 zone->zone_start_pfn = zone_start_pfn;
4166
Mel Gorman708614e2008-07-23 21:26:51 -07004167 mminit_dprintk(MMINIT_TRACE, "memmap_init",
4168 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4169 pgdat->node_id,
4170 (unsigned long)zone_idx(zone),
4171 zone_start_pfn, (zone_start_pfn + size));
4172
Andi Kleen1e548de2008-02-04 22:29:26 -08004173 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07004174
4175 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004176}
4177
Tejun Heo0ee332c2011-12-08 10:22:09 -08004178#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07004179#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4180/*
4181 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4182 * Architectures may implement their own version but if add_active_range()
4183 * was used and there are no special requirements, this is a convenient
4184 * alternative
4185 */
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004186int __meminit __early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004187{
Tejun Heoc13291a2011-07-12 10:46:30 +02004188 unsigned long start_pfn, end_pfn;
4189 int i, nid;
Russ Anderson7c243c72013-04-29 15:07:59 -07004190 /*
4191 * NOTE: The following SMP-unsafe globals are only used early in boot
4192 * when the kernel is running single-threaded.
4193 */
4194 static unsigned long __meminitdata last_start_pfn, last_end_pfn;
4195 static int __meminitdata last_nid;
4196
4197 if (last_start_pfn <= pfn && pfn < last_end_pfn)
4198 return last_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004199
Tejun Heoc13291a2011-07-12 10:46:30 +02004200 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
Russ Anderson7c243c72013-04-29 15:07:59 -07004201 if (start_pfn <= pfn && pfn < end_pfn) {
4202 last_start_pfn = start_pfn;
4203 last_end_pfn = end_pfn;
4204 last_nid = nid;
Tejun Heoc13291a2011-07-12 10:46:30 +02004205 return nid;
Russ Anderson7c243c72013-04-29 15:07:59 -07004206 }
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08004207 /* This is a memory hole */
4208 return -1;
Mel Gormanc7132162006-09-27 01:49:43 -07004209}
4210#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4211
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004212int __meminit early_pfn_to_nid(unsigned long pfn)
4213{
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08004214 int nid;
4215
4216 nid = __early_pfn_to_nid(pfn);
4217 if (nid >= 0)
4218 return nid;
4219 /* just returns 0 */
4220 return 0;
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004221}
4222
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08004223#ifdef CONFIG_NODES_SPAN_OTHER_NODES
4224bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
4225{
4226 int nid;
4227
4228 nid = __early_pfn_to_nid(pfn);
4229 if (nid >= 0 && nid != node)
4230 return false;
4231 return true;
4232}
4233#endif
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004234
Mel Gormanc7132162006-09-27 01:49:43 -07004235/**
4236 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004237 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4238 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07004239 *
4240 * If an architecture guarantees that all ranges registered with
4241 * add_active_ranges() contain no holes and may be freed, this
4242 * this function may be used instead of calling free_bootmem() manually.
4243 */
Tejun Heoc13291a2011-07-12 10:46:30 +02004244void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004245{
Tejun Heoc13291a2011-07-12 10:46:30 +02004246 unsigned long start_pfn, end_pfn;
4247 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004248
Tejun Heoc13291a2011-07-12 10:46:30 +02004249 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
4250 start_pfn = min(start_pfn, max_low_pfn);
4251 end_pfn = min(end_pfn, max_low_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004252
Tejun Heoc13291a2011-07-12 10:46:30 +02004253 if (start_pfn < end_pfn)
4254 free_bootmem_node(NODE_DATA(this_nid),
4255 PFN_PHYS(start_pfn),
4256 (end_pfn - start_pfn) << PAGE_SHIFT);
Mel Gormanc7132162006-09-27 01:49:43 -07004257 }
4258}
4259
4260/**
4261 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004262 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07004263 *
4264 * If an architecture guarantees that all ranges registered with
4265 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004266 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07004267 */
4268void __init sparse_memory_present_with_active_regions(int nid)
4269{
Tejun Heoc13291a2011-07-12 10:46:30 +02004270 unsigned long start_pfn, end_pfn;
4271 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004272
Tejun Heoc13291a2011-07-12 10:46:30 +02004273 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
4274 memory_present(this_nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004275}
4276
4277/**
4278 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004279 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4280 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4281 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07004282 *
4283 * It returns the start and end page frame of a node based on information
4284 * provided by an arch calling add_active_range(). If called for a node
4285 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004286 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07004287 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004288void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004289 unsigned long *start_pfn, unsigned long *end_pfn)
4290{
Tejun Heoc13291a2011-07-12 10:46:30 +02004291 unsigned long this_start_pfn, this_end_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004292 int i;
Tejun Heoc13291a2011-07-12 10:46:30 +02004293
Mel Gormanc7132162006-09-27 01:49:43 -07004294 *start_pfn = -1UL;
4295 *end_pfn = 0;
4296
Tejun Heoc13291a2011-07-12 10:46:30 +02004297 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
4298 *start_pfn = min(*start_pfn, this_start_pfn);
4299 *end_pfn = max(*end_pfn, this_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004300 }
4301
Christoph Lameter633c0662007-10-16 01:25:37 -07004302 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07004303 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07004304}
4305
4306/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07004307 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4308 * assumption is made that zones within a node are ordered in monotonic
4309 * increasing memory addresses so that the "highest" populated zone is used
4310 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004311static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004312{
4313 int zone_index;
4314 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
4315 if (zone_index == ZONE_MOVABLE)
4316 continue;
4317
4318 if (arch_zone_highest_possible_pfn[zone_index] >
4319 arch_zone_lowest_possible_pfn[zone_index])
4320 break;
4321 }
4322
4323 VM_BUG_ON(zone_index == -1);
4324 movable_zone = zone_index;
4325}
4326
4327/*
4328 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004329 * because it is sized independent of architecture. Unlike the other zones,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004330 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4331 * in each node depending on the size of each node and how evenly kernelcore
4332 * is distributed. This helper function adjusts the zone ranges
4333 * provided by the architecture for a given node by using the end of the
4334 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4335 * zones within a node are in order of monotonic increases memory addresses
4336 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004337static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004338 unsigned long zone_type,
4339 unsigned long node_start_pfn,
4340 unsigned long node_end_pfn,
4341 unsigned long *zone_start_pfn,
4342 unsigned long *zone_end_pfn)
4343{
4344 /* Only adjust if ZONE_MOVABLE is on this node */
4345 if (zone_movable_pfn[nid]) {
4346 /* Size ZONE_MOVABLE */
4347 if (zone_type == ZONE_MOVABLE) {
4348 *zone_start_pfn = zone_movable_pfn[nid];
4349 *zone_end_pfn = min(node_end_pfn,
4350 arch_zone_highest_possible_pfn[movable_zone]);
4351
4352 /* Adjust for ZONE_MOVABLE starting within this range */
4353 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
4354 *zone_end_pfn > zone_movable_pfn[nid]) {
4355 *zone_end_pfn = zone_movable_pfn[nid];
4356
4357 /* Check if this whole range is within ZONE_MOVABLE */
4358 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
4359 *zone_start_pfn = *zone_end_pfn;
4360 }
4361}
4362
4363/*
Mel Gormanc7132162006-09-27 01:49:43 -07004364 * Return the number of pages a zone spans in a node, including holes
4365 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4366 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004367static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004368 unsigned long zone_type,
4369 unsigned long *ignored)
4370{
4371 unsigned long node_start_pfn, node_end_pfn;
4372 unsigned long zone_start_pfn, zone_end_pfn;
4373
4374 /* Get the start and end of the node and zone */
4375 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
4376 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4377 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07004378 adjust_zone_range_for_zone_movable(nid, zone_type,
4379 node_start_pfn, node_end_pfn,
4380 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004381
4382 /* Check that this node has pages within the zone's required range */
4383 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4384 return 0;
4385
4386 /* Move the zone boundaries inside the node if necessary */
4387 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4388 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4389
4390 /* Return the spanned pages */
4391 return zone_end_pfn - zone_start_pfn;
4392}
4393
4394/*
4395 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004396 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07004397 */
Yinghai Lu32996252009-12-15 17:59:02 -08004398unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004399 unsigned long range_start_pfn,
4400 unsigned long range_end_pfn)
4401{
Tejun Heo96e907d2011-07-12 10:46:29 +02004402 unsigned long nr_absent = range_end_pfn - range_start_pfn;
4403 unsigned long start_pfn, end_pfn;
4404 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004405
Tejun Heo96e907d2011-07-12 10:46:29 +02004406 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
4407 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
4408 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
4409 nr_absent -= end_pfn - start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004410 }
Tejun Heo96e907d2011-07-12 10:46:29 +02004411 return nr_absent;
Mel Gormanc7132162006-09-27 01:49:43 -07004412}
4413
4414/**
4415 * absent_pages_in_range - Return number of page frames in holes within a range
4416 * @start_pfn: The start PFN to start searching for holes
4417 * @end_pfn: The end PFN to stop searching for holes
4418 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004419 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07004420 */
4421unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4422 unsigned long end_pfn)
4423{
4424 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4425}
4426
4427/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004428static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004429 unsigned long zone_type,
4430 unsigned long *ignored)
4431{
Tejun Heo96e907d2011-07-12 10:46:29 +02004432 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
4433 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman9c7cd682006-09-27 01:49:58 -07004434 unsigned long node_start_pfn, node_end_pfn;
4435 unsigned long zone_start_pfn, zone_end_pfn;
4436
4437 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
Tejun Heo96e907d2011-07-12 10:46:29 +02004438 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
4439 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004440
Mel Gorman2a1e2742007-07-17 04:03:12 -07004441 adjust_zone_range_for_zone_movable(nid, zone_type,
4442 node_start_pfn, node_end_pfn,
4443 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004444 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004445}
Mel Gorman0e0b8642006-09-27 01:49:56 -07004446
Tejun Heo0ee332c2011-12-08 10:22:09 -08004447#else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004448static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004449 unsigned long zone_type,
4450 unsigned long *zones_size)
4451{
4452 return zones_size[zone_type];
4453}
4454
Paul Mundt6ea6e682007-07-15 23:38:20 -07004455static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004456 unsigned long zone_type,
4457 unsigned long *zholes_size)
4458{
4459 if (!zholes_size)
4460 return 0;
4461
4462 return zholes_size[zone_type];
4463}
Yinghai Lu20e69262013-03-01 14:51:27 -08004464
Tejun Heo0ee332c2011-12-08 10:22:09 -08004465#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07004466
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004467static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Mel Gormanc7132162006-09-27 01:49:43 -07004468 unsigned long *zones_size, unsigned long *zholes_size)
4469{
4470 unsigned long realtotalpages, totalpages = 0;
4471 enum zone_type i;
4472
4473 for (i = 0; i < MAX_NR_ZONES; i++)
4474 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
4475 zones_size);
4476 pgdat->node_spanned_pages = totalpages;
4477
4478 realtotalpages = totalpages;
4479 for (i = 0; i < MAX_NR_ZONES; i++)
4480 realtotalpages -=
4481 zone_absent_pages_in_node(pgdat->node_id, i,
4482 zholes_size);
4483 pgdat->node_present_pages = realtotalpages;
4484 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4485 realtotalpages);
4486}
4487
Mel Gorman835c1342007-10-16 01:25:47 -07004488#ifndef CONFIG_SPARSEMEM
4489/*
4490 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07004491 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4492 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07004493 * round what is now in bits to nearest long in bits, then return it in
4494 * bytes.
4495 */
Linus Torvalds7c455122013-02-18 09:58:02 -08004496static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07004497{
4498 unsigned long usemapsize;
4499
Linus Torvalds7c455122013-02-18 09:58:02 -08004500 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
Mel Gormand9c23402007-10-16 01:26:01 -07004501 usemapsize = roundup(zonesize, pageblock_nr_pages);
4502 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07004503 usemapsize *= NR_PAGEBLOCK_BITS;
4504 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4505
4506 return usemapsize / 8;
4507}
4508
4509static void __init setup_usemap(struct pglist_data *pgdat,
Linus Torvalds7c455122013-02-18 09:58:02 -08004510 struct zone *zone,
4511 unsigned long zone_start_pfn,
4512 unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07004513{
Linus Torvalds7c455122013-02-18 09:58:02 -08004514 unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
Mel Gorman835c1342007-10-16 01:25:47 -07004515 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08004516 if (usemapsize)
Yinghai Lu8f389a992011-05-11 15:13:32 -07004517 zone->pageblock_flags = alloc_bootmem_node_nopanic(pgdat,
4518 usemapsize);
Mel Gorman835c1342007-10-16 01:25:47 -07004519}
4520#else
Linus Torvalds7c455122013-02-18 09:58:02 -08004521static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
4522 unsigned long zone_start_pfn, unsigned long zonesize) {}
Mel Gorman835c1342007-10-16 01:25:47 -07004523#endif /* CONFIG_SPARSEMEM */
4524
Mel Gormand9c23402007-10-16 01:26:01 -07004525#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08004526
Mel Gormand9c23402007-10-16 01:26:01 -07004527/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
Xishi Qiuca57df72012-07-31 16:43:19 -07004528void __init set_pageblock_order(void)
Mel Gormand9c23402007-10-16 01:26:01 -07004529{
Andrew Morton955c1cd2012-05-29 15:06:31 -07004530 unsigned int order;
4531
Mel Gormand9c23402007-10-16 01:26:01 -07004532 /* Check that pageblock_nr_pages has not already been setup */
4533 if (pageblock_order)
4534 return;
4535
Andrew Morton955c1cd2012-05-29 15:06:31 -07004536 if (HPAGE_SHIFT > PAGE_SHIFT)
4537 order = HUGETLB_PAGE_ORDER;
4538 else
4539 order = MAX_ORDER - 1;
4540
Mel Gormand9c23402007-10-16 01:26:01 -07004541 /*
4542 * Assume the largest contiguous order of interest is a huge page.
Andrew Morton955c1cd2012-05-29 15:06:31 -07004543 * This value may be variable depending on boot parameters on IA64 and
4544 * powerpc.
Mel Gormand9c23402007-10-16 01:26:01 -07004545 */
4546 pageblock_order = order;
4547}
4548#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4549
Mel Gormanba72cb82007-11-28 16:21:13 -08004550/*
4551 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
Andrew Morton955c1cd2012-05-29 15:06:31 -07004552 * is unused as pageblock_order is set at compile-time. See
4553 * include/linux/pageblock-flags.h for the values of pageblock_order based on
4554 * the kernel config
Mel Gormanba72cb82007-11-28 16:21:13 -08004555 */
Xishi Qiuca57df72012-07-31 16:43:19 -07004556void __init set_pageblock_order(void)
Mel Gormanba72cb82007-11-28 16:21:13 -08004557{
Mel Gormanba72cb82007-11-28 16:21:13 -08004558}
Mel Gormand9c23402007-10-16 01:26:01 -07004559
4560#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4561
Jiang Liu01cefae2012-12-12 13:52:19 -08004562static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
4563 unsigned long present_pages)
4564{
4565 unsigned long pages = spanned_pages;
4566
4567 /*
4568 * Provide a more accurate estimation if there are holes within
4569 * the zone and SPARSEMEM is in use. If there are holes within the
4570 * zone, each populated memory region may cost us one or two extra
4571 * memmap pages due to alignment because memmap pages for each
4572 * populated regions may not naturally algined on page boundary.
4573 * So the (present_pages >> 4) heuristic is a tradeoff for that.
4574 */
4575 if (spanned_pages > present_pages + (present_pages >> 4) &&
4576 IS_ENABLED(CONFIG_SPARSEMEM))
4577 pages = present_pages;
4578
4579 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
4580}
4581
Linus Torvalds1da177e2005-04-16 15:20:36 -07004582/*
4583 * Set up the zone data structures:
4584 * - mark all pages reserved
4585 * - mark all memory queues empty
4586 * - clear the memory bitmaps
Minchan Kim6527af52012-07-31 16:46:16 -07004587 *
4588 * NOTE: pgdat should get zeroed by caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004589 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08004590static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004591 unsigned long *zones_size, unsigned long *zholes_size)
4592{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07004593 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07004594 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004595 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07004596 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004597
Dave Hansen208d54e2005-10-29 18:16:52 -07004598 pgdat_resize_init(pgdat);
Andrea Arcangeli8177a422012-03-23 20:56:34 +01004599#ifdef CONFIG_NUMA_BALANCING
4600 spin_lock_init(&pgdat->numabalancing_migrate_lock);
4601 pgdat->numabalancing_migrate_nr_pages = 0;
4602 pgdat->numabalancing_migrate_next_window = jiffies;
4603#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004604 init_waitqueue_head(&pgdat->kswapd_wait);
Mel Gorman55150612012-07-31 16:44:35 -07004605 init_waitqueue_head(&pgdat->pfmemalloc_wait);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004606 pgdat_page_cgroup_init(pgdat);
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01004607
Linus Torvalds1da177e2005-04-16 15:20:36 -07004608 for (j = 0; j < MAX_NR_ZONES; j++) {
4609 struct zone *zone = pgdat->node_zones + j;
Jiang Liu9feedc92012-12-12 13:52:12 -08004610 unsigned long size, realsize, freesize, memmap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004611
Mel Gormanc7132162006-09-27 01:49:43 -07004612 size = zone_spanned_pages_in_node(nid, j, zones_size);
Jiang Liu9feedc92012-12-12 13:52:12 -08004613 realsize = freesize = size - zone_absent_pages_in_node(nid, j,
Mel Gormanc7132162006-09-27 01:49:43 -07004614 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004615
Mel Gorman0e0b8642006-09-27 01:49:56 -07004616 /*
Jiang Liu9feedc92012-12-12 13:52:12 -08004617 * Adjust freesize so that it accounts for how much memory
Mel Gorman0e0b8642006-09-27 01:49:56 -07004618 * is used by this zone for memmap. This affects the watermark
4619 * and per-cpu initialisations
4620 */
Jiang Liu01cefae2012-12-12 13:52:19 -08004621 memmap_pages = calc_memmap_size(size, realsize);
Jiang Liu9feedc92012-12-12 13:52:12 -08004622 if (freesize >= memmap_pages) {
4623 freesize -= memmap_pages;
Yinghai Lu5594c8c2009-01-06 14:39:14 -08004624 if (memmap_pages)
4625 printk(KERN_DEBUG
4626 " %s zone: %lu pages used for memmap\n",
4627 zone_names[j], memmap_pages);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004628 } else
4629 printk(KERN_WARNING
Jiang Liu9feedc92012-12-12 13:52:12 -08004630 " %s zone: %lu pages exceeds freesize %lu\n",
4631 zone_names[j], memmap_pages, freesize);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004632
Christoph Lameter62672762007-02-10 01:43:07 -08004633 /* Account for reserved pages */
Jiang Liu9feedc92012-12-12 13:52:12 -08004634 if (j == 0 && freesize > dma_reserve) {
4635 freesize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07004636 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08004637 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004638 }
4639
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004640 if (!is_highmem_idx(j))
Jiang Liu9feedc92012-12-12 13:52:12 -08004641 nr_kernel_pages += freesize;
Jiang Liu01cefae2012-12-12 13:52:19 -08004642 /* Charge for highmem memmap if there are enough kernel pages */
4643 else if (nr_kernel_pages > memmap_pages * 2)
4644 nr_kernel_pages -= memmap_pages;
Jiang Liu9feedc92012-12-12 13:52:12 -08004645 nr_all_pages += freesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004646
4647 zone->spanned_pages = size;
Jiang Liu306f2e92013-02-22 16:33:54 -08004648 zone->present_pages = realsize;
Jiang Liu9feedc92012-12-12 13:52:12 -08004649 /*
4650 * Set an approximate value for lowmem here, it will be adjusted
4651 * when the bootmem allocator frees pages into the buddy system.
4652 * And all highmem pages will be managed by the buddy system.
4653 */
4654 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
Christoph Lameter96146342006-07-03 00:24:13 -07004655#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07004656 zone->node = nid;
Jiang Liu9feedc92012-12-12 13:52:12 -08004657 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07004658 / 100;
Jiang Liu9feedc92012-12-12 13:52:12 -08004659 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07004660#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004661 zone->name = zone_names[j];
4662 spin_lock_init(&zone->lock);
4663 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07004664 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004665 zone->zone_pgdat = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004666
Dave Hansened8ece22005-10-29 18:16:50 -07004667 zone_pcp_init(zone);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004668 lruvec_init(&zone->lruvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004669 if (!size)
4670 continue;
4671
Andrew Morton955c1cd2012-05-29 15:06:31 -07004672 set_pageblock_order();
Linus Torvalds7c455122013-02-18 09:58:02 -08004673 setup_usemap(pgdat, zone, zone_start_pfn, size);
Dave Hansena2f3aa022007-01-10 23:15:30 -08004674 ret = init_currently_empty_zone(zone, zone_start_pfn,
4675 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07004676 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07004677 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004678 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004679 }
4680}
4681
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004682static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004683{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004684 /* Skip empty nodes */
4685 if (!pgdat->node_spanned_pages)
4686 return;
4687
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004688#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07004689 /* ia64 gets its own node_mem_map, before this, without bootmem */
4690 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07004691 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004692 struct page *map;
4693
Bob Piccoe984bb42006-05-20 15:00:31 -07004694 /*
4695 * The zone's endpoints aren't required to be MAX_ORDER
4696 * aligned but the node_mem_map endpoints must be in order
4697 * for the buddy allocator to function correctly.
4698 */
4699 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
Cody P Schafer108bcc92013-02-22 16:35:23 -08004700 end = pgdat_end_pfn(pgdat);
Bob Piccoe984bb42006-05-20 15:00:31 -07004701 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4702 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07004703 map = alloc_remap(pgdat->node_id, size);
4704 if (!map)
Yinghai Lu8f389a992011-05-11 15:13:32 -07004705 map = alloc_bootmem_node_nopanic(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07004706 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004707 }
Roman Zippel12d810c2007-05-31 00:40:54 -07004708#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07004709 /*
4710 * With no DISCONTIG, the global mem_map is just set as node 0's
4711 */
Mel Gormanc7132162006-09-27 01:49:43 -07004712 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004713 mem_map = NODE_DATA(0)->node_mem_map;
Tejun Heo0ee332c2011-12-08 10:22:09 -08004714#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07004715 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08004716 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Tejun Heo0ee332c2011-12-08 10:22:09 -08004717#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07004718 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004719#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004720#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004721}
4722
Johannes Weiner9109fb72008-07-23 21:27:20 -07004723void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4724 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004725{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004726 pg_data_t *pgdat = NODE_DATA(nid);
4727
Minchan Kim88fdf752012-07-31 16:46:14 -07004728 /* pg_data_t should be reset to zero when it's allocated */
Linus Torvalds8783b6e2012-08-02 10:37:03 -07004729 WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
Minchan Kim88fdf752012-07-31 16:46:14 -07004730
Linus Torvalds1da177e2005-04-16 15:20:36 -07004731 pgdat->node_id = nid;
4732 pgdat->node_start_pfn = node_start_pfn;
David Rientjes957f8222012-10-08 16:33:24 -07004733 init_zone_allows_reclaim(nid);
Mel Gormanc7132162006-09-27 01:49:43 -07004734 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004735
4736 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07004737#ifdef CONFIG_FLAT_NODE_MEM_MAP
4738 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4739 nid, (unsigned long)pgdat,
4740 (unsigned long)pgdat->node_mem_map);
4741#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004742
4743 free_area_init_core(pgdat, zones_size, zholes_size);
4744}
4745
Tejun Heo0ee332c2011-12-08 10:22:09 -08004746#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07004747
4748#if MAX_NUMNODES > 1
4749/*
4750 * Figure out the number of possible node ids.
4751 */
Cody P Schaferf9872ca2013-04-29 15:08:01 -07004752void __init setup_nr_node_ids(void)
Miklos Szeredi418508c2007-05-23 13:57:55 -07004753{
4754 unsigned int node;
4755 unsigned int highest = 0;
4756
4757 for_each_node_mask(node, node_possible_map)
4758 highest = node;
4759 nr_node_ids = highest + 1;
4760}
Miklos Szeredi418508c2007-05-23 13:57:55 -07004761#endif
4762
Mel Gormanc7132162006-09-27 01:49:43 -07004763/**
Tejun Heo1e019792011-07-12 09:45:34 +02004764 * node_map_pfn_alignment - determine the maximum internode alignment
4765 *
4766 * This function should be called after node map is populated and sorted.
4767 * It calculates the maximum power of two alignment which can distinguish
4768 * all the nodes.
4769 *
4770 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
4771 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
4772 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
4773 * shifted, 1GiB is enough and this function will indicate so.
4774 *
4775 * This is used to test whether pfn -> nid mapping of the chosen memory
4776 * model has fine enough granularity to avoid incorrect mapping for the
4777 * populated node map.
4778 *
4779 * Returns the determined alignment in pfn's. 0 if there is no alignment
4780 * requirement (single node).
4781 */
4782unsigned long __init node_map_pfn_alignment(void)
4783{
4784 unsigned long accl_mask = 0, last_end = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02004785 unsigned long start, end, mask;
Tejun Heo1e019792011-07-12 09:45:34 +02004786 int last_nid = -1;
Tejun Heoc13291a2011-07-12 10:46:30 +02004787 int i, nid;
Tejun Heo1e019792011-07-12 09:45:34 +02004788
Tejun Heoc13291a2011-07-12 10:46:30 +02004789 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
Tejun Heo1e019792011-07-12 09:45:34 +02004790 if (!start || last_nid < 0 || last_nid == nid) {
4791 last_nid = nid;
4792 last_end = end;
4793 continue;
4794 }
4795
4796 /*
4797 * Start with a mask granular enough to pin-point to the
4798 * start pfn and tick off bits one-by-one until it becomes
4799 * too coarse to separate the current node from the last.
4800 */
4801 mask = ~((1 << __ffs(start)) - 1);
4802 while (mask && last_end <= (start & (mask << 1)))
4803 mask <<= 1;
4804
4805 /* accumulate all internode masks */
4806 accl_mask |= mask;
4807 }
4808
4809 /* convert mask to number of pages */
4810 return ~accl_mask + 1;
4811}
4812
Mel Gormana6af2bc2007-02-10 01:42:57 -08004813/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004814static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07004815{
Mel Gormana6af2bc2007-02-10 01:42:57 -08004816 unsigned long min_pfn = ULONG_MAX;
Tejun Heoc13291a2011-07-12 10:46:30 +02004817 unsigned long start_pfn;
4818 int i;
Mel Gorman1abbfb42006-11-23 12:01:41 +00004819
Tejun Heoc13291a2011-07-12 10:46:30 +02004820 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
4821 min_pfn = min(min_pfn, start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004822
Mel Gormana6af2bc2007-02-10 01:42:57 -08004823 if (min_pfn == ULONG_MAX) {
4824 printk(KERN_WARNING
Paul Jackson2bc0d2612008-06-22 07:22:17 -07004825 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08004826 return 0;
4827 }
4828
4829 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004830}
4831
4832/**
4833 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4834 *
4835 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004836 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07004837 */
4838unsigned long __init find_min_pfn_with_active_regions(void)
4839{
4840 return find_min_pfn_for_node(MAX_NUMNODES);
4841}
4842
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004843/*
4844 * early_calculate_totalpages()
4845 * Sum pages in active regions for movable zone.
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08004846 * Populate N_MEMORY for calculating usable_nodes.
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004847 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07004848static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07004849{
Mel Gorman7e63efe2007-07-17 04:03:15 -07004850 unsigned long totalpages = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02004851 unsigned long start_pfn, end_pfn;
4852 int i, nid;
Mel Gorman7e63efe2007-07-17 04:03:15 -07004853
Tejun Heoc13291a2011-07-12 10:46:30 +02004854 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
4855 unsigned long pages = end_pfn - start_pfn;
4856
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004857 totalpages += pages;
4858 if (pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08004859 node_set_state(nid, N_MEMORY);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004860 }
4861 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07004862}
4863
Mel Gorman2a1e2742007-07-17 04:03:12 -07004864/*
4865 * Find the PFN the Movable zone begins in each node. Kernel memory
4866 * is spread evenly between nodes as long as the nodes have enough
4867 * memory. When they don't, some nodes will have more kernelcore than
4868 * others
4869 */
Kautuk Consulb224ef82012-03-21 16:34:15 -07004870static void __init find_zone_movable_pfns_for_nodes(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004871{
4872 int i, nid;
4873 unsigned long usable_startpfn;
4874 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07004875 /* save the state before borrow the nodemask */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08004876 nodemask_t saved_node_state = node_states[N_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004877 unsigned long totalpages = early_calculate_totalpages();
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08004878 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004879
Mel Gorman7e63efe2007-07-17 04:03:15 -07004880 /*
4881 * If movablecore was specified, calculate what size of
4882 * kernelcore that corresponds so that memory usable for
4883 * any allocation type is evenly spread. If both kernelcore
4884 * and movablecore are specified, then the value of kernelcore
4885 * will be used for required_kernelcore if it's greater than
4886 * what movablecore would have allowed.
4887 */
4888 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07004889 unsigned long corepages;
4890
4891 /*
4892 * Round-up so that ZONE_MOVABLE is at least as large as what
4893 * was requested by the user
4894 */
4895 required_movablecore =
4896 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
4897 corepages = totalpages - required_movablecore;
4898
4899 required_kernelcore = max(required_kernelcore, corepages);
4900 }
4901
Yinghai Lu20e69262013-03-01 14:51:27 -08004902 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4903 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07004904 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004905
4906 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
Yinghai Lu20e69262013-03-01 14:51:27 -08004907 find_usable_zone_for_movable();
Mel Gorman2a1e2742007-07-17 04:03:12 -07004908 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4909
4910restart:
4911 /* Spread kernelcore memory as evenly as possible throughout nodes */
4912 kernelcore_node = required_kernelcore / usable_nodes;
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08004913 for_each_node_state(nid, N_MEMORY) {
Tejun Heoc13291a2011-07-12 10:46:30 +02004914 unsigned long start_pfn, end_pfn;
4915
Mel Gorman2a1e2742007-07-17 04:03:12 -07004916 /*
4917 * Recalculate kernelcore_node if the division per node
4918 * now exceeds what is necessary to satisfy the requested
4919 * amount of memory for the kernel
4920 */
4921 if (required_kernelcore < kernelcore_node)
4922 kernelcore_node = required_kernelcore / usable_nodes;
4923
4924 /*
4925 * As the map is walked, we track how much memory is usable
4926 * by the kernel using kernelcore_remaining. When it is
4927 * 0, the rest of the node is usable by ZONE_MOVABLE
4928 */
4929 kernelcore_remaining = kernelcore_node;
4930
4931 /* Go through each range of PFNs within this node */
Tejun Heoc13291a2011-07-12 10:46:30 +02004932 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004933 unsigned long size_pages;
4934
Tejun Heoc13291a2011-07-12 10:46:30 +02004935 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004936 if (start_pfn >= end_pfn)
4937 continue;
4938
4939 /* Account for what is only usable for kernelcore */
4940 if (start_pfn < usable_startpfn) {
4941 unsigned long kernel_pages;
4942 kernel_pages = min(end_pfn, usable_startpfn)
4943 - start_pfn;
4944
4945 kernelcore_remaining -= min(kernel_pages,
4946 kernelcore_remaining);
4947 required_kernelcore -= min(kernel_pages,
4948 required_kernelcore);
4949
4950 /* Continue if range is now fully accounted */
4951 if (end_pfn <= usable_startpfn) {
4952
4953 /*
4954 * Push zone_movable_pfn to the end so
4955 * that if we have to rebalance
4956 * kernelcore across nodes, we will
4957 * not double account here
4958 */
4959 zone_movable_pfn[nid] = end_pfn;
4960 continue;
4961 }
4962 start_pfn = usable_startpfn;
4963 }
4964
4965 /*
4966 * The usable PFN range for ZONE_MOVABLE is from
4967 * start_pfn->end_pfn. Calculate size_pages as the
4968 * number of pages used as kernelcore
4969 */
4970 size_pages = end_pfn - start_pfn;
4971 if (size_pages > kernelcore_remaining)
4972 size_pages = kernelcore_remaining;
4973 zone_movable_pfn[nid] = start_pfn + size_pages;
4974
4975 /*
4976 * Some kernelcore has been met, update counts and
4977 * break if the kernelcore for this node has been
4978 * satisified
4979 */
4980 required_kernelcore -= min(required_kernelcore,
4981 size_pages);
4982 kernelcore_remaining -= size_pages;
4983 if (!kernelcore_remaining)
4984 break;
4985 }
4986 }
4987
4988 /*
4989 * If there is still required_kernelcore, we do another pass with one
4990 * less node in the count. This will push zone_movable_pfn[nid] further
4991 * along on the nodes that still have memory until kernelcore is
4992 * satisified
4993 */
4994 usable_nodes--;
4995 if (usable_nodes && required_kernelcore > usable_nodes)
4996 goto restart;
4997
4998 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4999 for (nid = 0; nid < MAX_NUMNODES; nid++)
5000 zone_movable_pfn[nid] =
5001 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07005002
Yinghai Lu20e69262013-03-01 14:51:27 -08005003out:
Yinghai Lu66918dc2009-06-30 11:41:37 -07005004 /* restore the node_state */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005005 node_states[N_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005006}
5007
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005008/* Any regular or high memory on that node ? */
5009static void check_for_memory(pg_data_t *pgdat, int nid)
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005010{
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005011 enum zone_type zone_type;
5012
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005013 if (N_MEMORY == N_NORMAL_MEMORY)
5014 return;
5015
5016 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005017 struct zone *zone = &pgdat->node_zones[zone_type];
Bob Liud0048b02012-01-12 17:19:07 -08005018 if (zone->present_pages) {
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005019 node_set_state(nid, N_HIGH_MEMORY);
5020 if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
5021 zone_type <= ZONE_NORMAL)
5022 node_set_state(nid, N_NORMAL_MEMORY);
Bob Liud0048b02012-01-12 17:19:07 -08005023 break;
5024 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005025 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005026}
5027
Mel Gormanc7132162006-09-27 01:49:43 -07005028/**
5029 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005030 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07005031 *
5032 * This will call free_area_init_node() for each active node in the system.
5033 * Using the page ranges provided by add_active_range(), the size of each
5034 * zone in each node and their holes is calculated. If the maximum PFN
5035 * between two adjacent zones match, it is assumed that the zone is empty.
5036 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5037 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5038 * starts where the previous one ended. For example, ZONE_DMA32 starts
5039 * at arch_max_dma_pfn.
5040 */
5041void __init free_area_init_nodes(unsigned long *max_zone_pfn)
5042{
Tejun Heoc13291a2011-07-12 10:46:30 +02005043 unsigned long start_pfn, end_pfn;
5044 int i, nid;
Mel Gormana6af2bc2007-02-10 01:42:57 -08005045
Mel Gormanc7132162006-09-27 01:49:43 -07005046 /* Record where the zone boundaries are */
5047 memset(arch_zone_lowest_possible_pfn, 0,
5048 sizeof(arch_zone_lowest_possible_pfn));
5049 memset(arch_zone_highest_possible_pfn, 0,
5050 sizeof(arch_zone_highest_possible_pfn));
5051 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
5052 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
5053 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005054 if (i == ZONE_MOVABLE)
5055 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07005056 arch_zone_lowest_possible_pfn[i] =
5057 arch_zone_highest_possible_pfn[i-1];
5058 arch_zone_highest_possible_pfn[i] =
5059 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
5060 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07005061 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
5062 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
5063
5064 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5065 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
Kautuk Consulb224ef82012-03-21 16:34:15 -07005066 find_zone_movable_pfns_for_nodes();
Mel Gormanc7132162006-09-27 01:49:43 -07005067
Mel Gormanc7132162006-09-27 01:49:43 -07005068 /* Print out the zone ranges */
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005069 printk("Zone ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005070 for (i = 0; i < MAX_NR_ZONES; i++) {
5071 if (i == ZONE_MOVABLE)
5072 continue;
Kay Sievers155cbfc2012-05-08 17:24:14 +02005073 printk(KERN_CONT " %-8s ", zone_names[i]);
David Rientjes72f0ba02010-03-05 13:42:14 -08005074 if (arch_zone_lowest_possible_pfn[i] ==
5075 arch_zone_highest_possible_pfn[i])
Kay Sievers155cbfc2012-05-08 17:24:14 +02005076 printk(KERN_CONT "empty\n");
David Rientjes72f0ba02010-03-05 13:42:14 -08005077 else
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005078 printk(KERN_CONT "[mem %0#10lx-%0#10lx]\n",
5079 arch_zone_lowest_possible_pfn[i] << PAGE_SHIFT,
5080 (arch_zone_highest_possible_pfn[i]
5081 << PAGE_SHIFT) - 1);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005082 }
5083
5084 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005085 printk("Movable zone start for each node\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005086 for (i = 0; i < MAX_NUMNODES; i++) {
5087 if (zone_movable_pfn[i])
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005088 printk(" Node %d: %#010lx\n", i,
5089 zone_movable_pfn[i] << PAGE_SHIFT);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005090 }
Mel Gormanc7132162006-09-27 01:49:43 -07005091
Wanpeng Lif2d52fe2012-10-08 16:32:24 -07005092 /* Print out the early node map */
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005093 printk("Early memory node ranges\n");
Tejun Heoc13291a2011-07-12 10:46:30 +02005094 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005095 printk(" node %3d: [mem %#010lx-%#010lx]\n", nid,
5096 start_pfn << PAGE_SHIFT, (end_pfn << PAGE_SHIFT) - 1);
Mel Gormanc7132162006-09-27 01:49:43 -07005097
5098 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07005099 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08005100 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07005101 for_each_online_node(nid) {
5102 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07005103 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07005104 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005105
5106 /* Any memory on that node */
5107 if (pgdat->node_present_pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005108 node_set_state(nid, N_MEMORY);
5109 check_for_memory(pgdat, nid);
Mel Gormanc7132162006-09-27 01:49:43 -07005110 }
5111}
Mel Gorman2a1e2742007-07-17 04:03:12 -07005112
Mel Gorman7e63efe2007-07-17 04:03:15 -07005113static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005114{
5115 unsigned long long coremem;
5116 if (!p)
5117 return -EINVAL;
5118
5119 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005120 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005121
Mel Gorman7e63efe2007-07-17 04:03:15 -07005122 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07005123 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
5124
5125 return 0;
5126}
Mel Gormaned7ed362007-07-17 04:03:14 -07005127
Mel Gorman7e63efe2007-07-17 04:03:15 -07005128/*
5129 * kernelcore=size sets the amount of memory for use for allocations that
5130 * cannot be reclaimed or migrated.
5131 */
5132static int __init cmdline_parse_kernelcore(char *p)
5133{
5134 return cmdline_parse_core(p, &required_kernelcore);
5135}
5136
5137/*
5138 * movablecore=size sets the amount of memory for use for allocations that
5139 * can be reclaimed or migrated.
5140 */
5141static int __init cmdline_parse_movablecore(char *p)
5142{
5143 return cmdline_parse_core(p, &required_movablecore);
5144}
5145
Mel Gormaned7ed362007-07-17 04:03:14 -07005146early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005147early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07005148
Tejun Heo0ee332c2011-12-08 10:22:09 -08005149#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005150
Jiang Liu69afade2013-04-29 15:06:21 -07005151unsigned long free_reserved_area(unsigned long start, unsigned long end,
5152 int poison, char *s)
5153{
5154 unsigned long pages, pos;
5155
5156 pos = start = PAGE_ALIGN(start);
5157 end &= PAGE_MASK;
5158 for (pages = 0; pos < end; pos += PAGE_SIZE, pages++) {
5159 if (poison)
5160 memset((void *)pos, poison, PAGE_SIZE);
Ralf Baechlebb3ec6b2013-05-22 12:18:47 +02005161 free_reserved_page(virt_to_page((void *)pos));
Jiang Liu69afade2013-04-29 15:06:21 -07005162 }
5163
5164 if (pages && s)
5165 pr_info("Freeing %s memory: %ldK (%lx - %lx)\n",
5166 s, pages << (PAGE_SHIFT - 10), start, end);
5167
5168 return pages;
5169}
5170
Jiang Liucfa11e02013-04-29 15:07:00 -07005171#ifdef CONFIG_HIGHMEM
5172void free_highmem_page(struct page *page)
5173{
5174 __free_reserved_page(page);
5175 totalram_pages++;
5176 totalhigh_pages++;
5177}
5178#endif
5179
Mel Gorman0e0b8642006-09-27 01:49:56 -07005180/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005181 * set_dma_reserve - set the specified number of pages reserved in the first zone
5182 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07005183 *
5184 * The per-cpu batchsize and zone watermarks are determined by present_pages.
5185 * In the DMA zone, a significant percentage may be consumed by kernel image
5186 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005187 * function may optionally be used to account for unfreeable pages in the
5188 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5189 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07005190 */
5191void __init set_dma_reserve(unsigned long new_dma_reserve)
5192{
5193 dma_reserve = new_dma_reserve;
5194}
5195
Linus Torvalds1da177e2005-04-16 15:20:36 -07005196void __init free_area_init(unsigned long *zones_size)
5197{
Johannes Weiner9109fb72008-07-23 21:27:20 -07005198 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005199 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
5200}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005201
Linus Torvalds1da177e2005-04-16 15:20:36 -07005202static int page_alloc_cpu_notify(struct notifier_block *self,
5203 unsigned long action, void *hcpu)
5204{
5205 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005206
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07005207 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -07005208 lru_add_drain_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005209 drain_pages(cpu);
5210
5211 /*
5212 * Spill the event counters of the dead processor
5213 * into the current processors event counters.
5214 * This artificially elevates the count of the current
5215 * processor.
5216 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07005217 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005218
5219 /*
5220 * Zero the differential counters of the dead processor
5221 * so that the vm statistics are consistent.
5222 *
5223 * This is only okay since the processor is dead and cannot
5224 * race with what we are doing.
5225 */
Christoph Lameter2244b952006-06-30 01:55:33 -07005226 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005227 }
5228 return NOTIFY_OK;
5229}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005230
5231void __init page_alloc_init(void)
5232{
5233 hotcpu_notifier(page_alloc_cpu_notify, 0);
5234}
5235
5236/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005237 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5238 * or min_free_kbytes changes.
5239 */
5240static void calculate_totalreserve_pages(void)
5241{
5242 struct pglist_data *pgdat;
5243 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005244 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005245
5246 for_each_online_pgdat(pgdat) {
5247 for (i = 0; i < MAX_NR_ZONES; i++) {
5248 struct zone *zone = pgdat->node_zones + i;
5249 unsigned long max = 0;
5250
5251 /* Find valid and maximum lowmem_reserve in the zone */
5252 for (j = i; j < MAX_NR_ZONES; j++) {
5253 if (zone->lowmem_reserve[j] > max)
5254 max = zone->lowmem_reserve[j];
5255 }
5256
Mel Gorman41858962009-06-16 15:32:12 -07005257 /* we treat the high watermark as reserved pages. */
5258 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005259
Jiang Liub40da042013-02-22 16:33:52 -08005260 if (max > zone->managed_pages)
5261 max = zone->managed_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005262 reserve_pages += max;
Johannes Weinerab8fabd2012-01-10 15:07:42 -08005263 /*
5264 * Lowmem reserves are not available to
5265 * GFP_HIGHUSER page cache allocations and
5266 * kswapd tries to balance zones to their high
5267 * watermark. As a result, neither should be
5268 * regarded as dirtyable memory, to prevent a
5269 * situation where reclaim has to clean pages
5270 * in order to balance the zones.
5271 */
5272 zone->dirty_balance_reserve = max;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005273 }
5274 }
Johannes Weinerab8fabd2012-01-10 15:07:42 -08005275 dirty_balance_reserve = reserve_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005276 totalreserve_pages = reserve_pages;
5277}
5278
5279/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07005280 * setup_per_zone_lowmem_reserve - called whenever
5281 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
5282 * has a correct pages reserved value, so an adequate number of
5283 * pages are left in the zone after a successful __alloc_pages().
5284 */
5285static void setup_per_zone_lowmem_reserve(void)
5286{
5287 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005288 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005289
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08005290 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005291 for (j = 0; j < MAX_NR_ZONES; j++) {
5292 struct zone *zone = pgdat->node_zones + j;
Jiang Liub40da042013-02-22 16:33:52 -08005293 unsigned long managed_pages = zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005294
5295 zone->lowmem_reserve[j] = 0;
5296
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005297 idx = j;
5298 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005299 struct zone *lower_zone;
5300
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005301 idx--;
5302
Linus Torvalds1da177e2005-04-16 15:20:36 -07005303 if (sysctl_lowmem_reserve_ratio[idx] < 1)
5304 sysctl_lowmem_reserve_ratio[idx] = 1;
5305
5306 lower_zone = pgdat->node_zones + idx;
Jiang Liub40da042013-02-22 16:33:52 -08005307 lower_zone->lowmem_reserve[j] = managed_pages /
Linus Torvalds1da177e2005-04-16 15:20:36 -07005308 sysctl_lowmem_reserve_ratio[idx];
Jiang Liub40da042013-02-22 16:33:52 -08005309 managed_pages += lower_zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005310 }
5311 }
5312 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005313
5314 /* update totalreserve_pages */
5315 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005316}
5317
Mel Gormancfd3da12011-04-25 21:36:42 +00005318static void __setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005319{
5320 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
5321 unsigned long lowmem_pages = 0;
5322 struct zone *zone;
5323 unsigned long flags;
5324
5325 /* Calculate total number of !ZONE_HIGHMEM pages */
5326 for_each_zone(zone) {
5327 if (!is_highmem(zone))
Jiang Liub40da042013-02-22 16:33:52 -08005328 lowmem_pages += zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005329 }
5330
5331 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07005332 u64 tmp;
5333
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005334 spin_lock_irqsave(&zone->lock, flags);
Jiang Liub40da042013-02-22 16:33:52 -08005335 tmp = (u64)pages_min * zone->managed_pages;
Andrew Mortonac924c62006-05-15 09:43:59 -07005336 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005337 if (is_highmem(zone)) {
5338 /*
Nick Piggin669ed172005-11-13 16:06:45 -08005339 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5340 * need highmem pages, so cap pages_min to a small
5341 * value here.
5342 *
Mel Gorman41858962009-06-16 15:32:12 -07005343 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Nick Piggin669ed172005-11-13 16:06:45 -08005344 * deltas controls asynch page reclaim, and so should
5345 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005346 */
Andrew Morton90ae8d62013-02-22 16:32:22 -08005347 unsigned long min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005348
Jiang Liub40da042013-02-22 16:33:52 -08005349 min_pages = zone->managed_pages / 1024;
Andrew Morton90ae8d62013-02-22 16:32:22 -08005350 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
Mel Gorman41858962009-06-16 15:32:12 -07005351 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005352 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08005353 /*
5354 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07005355 * proportionate to the zone's size.
5356 */
Mel Gorman41858962009-06-16 15:32:12 -07005357 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005358 }
5359
Mel Gorman41858962009-06-16 15:32:12 -07005360 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
5361 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Marek Szyprowski49f223a2012-01-25 12:49:24 +01005362
Mel Gorman56fd56b2007-10-16 01:25:58 -07005363 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005364 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005365 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005366
5367 /* update totalreserve_pages */
5368 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005369}
5370
Mel Gormancfd3da12011-04-25 21:36:42 +00005371/**
5372 * setup_per_zone_wmarks - called when min_free_kbytes changes
5373 * or when memory is hot-{added|removed}
5374 *
5375 * Ensures that the watermark[min,low,high] values for each zone are set
5376 * correctly with respect to min_free_kbytes.
5377 */
5378void setup_per_zone_wmarks(void)
5379{
5380 mutex_lock(&zonelists_mutex);
5381 __setup_per_zone_wmarks();
5382 mutex_unlock(&zonelists_mutex);
5383}
5384
Randy Dunlap55a44622009-09-21 17:01:20 -07005385/*
Rik van Riel556adec2008-10-18 20:26:34 -07005386 * The inactive anon list should be small enough that the VM never has to
5387 * do too much work, but large enough that each inactive page has a chance
5388 * to be referenced again before it is swapped out.
5389 *
5390 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5391 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5392 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5393 * the anonymous pages are kept on the inactive list.
5394 *
5395 * total target max
5396 * memory ratio inactive anon
5397 * -------------------------------------
5398 * 10MB 1 5MB
5399 * 100MB 1 50MB
5400 * 1GB 3 250MB
5401 * 10GB 10 0.9GB
5402 * 100GB 31 3GB
5403 * 1TB 101 10GB
5404 * 10TB 320 32GB
5405 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005406static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
Minchan Kim96cb4df2009-06-16 15:32:49 -07005407{
5408 unsigned int gb, ratio;
5409
5410 /* Zone size in gigabytes */
Jiang Liub40da042013-02-22 16:33:52 -08005411 gb = zone->managed_pages >> (30 - PAGE_SHIFT);
Minchan Kim96cb4df2009-06-16 15:32:49 -07005412 if (gb)
5413 ratio = int_sqrt(10 * gb);
5414 else
5415 ratio = 1;
5416
5417 zone->inactive_ratio = ratio;
5418}
5419
KOSAKI Motohiro839a4fc2011-05-24 17:11:31 -07005420static void __meminit setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07005421{
5422 struct zone *zone;
5423
Minchan Kim96cb4df2009-06-16 15:32:49 -07005424 for_each_zone(zone)
5425 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07005426}
5427
Linus Torvalds1da177e2005-04-16 15:20:36 -07005428/*
5429 * Initialise min_free_kbytes.
5430 *
5431 * For small machines we want it small (128k min). For large machines
5432 * we want it large (64MB max). But it is not linear, because network
5433 * bandwidth does not increase linearly with machine size. We use
5434 *
5435 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5436 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5437 *
5438 * which yields
5439 *
5440 * 16MB: 512k
5441 * 32MB: 724k
5442 * 64MB: 1024k
5443 * 128MB: 1448k
5444 * 256MB: 2048k
5445 * 512MB: 2896k
5446 * 1024MB: 4096k
5447 * 2048MB: 5792k
5448 * 4096MB: 8192k
5449 * 8192MB: 11584k
5450 * 16384MB: 16384k
5451 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005452int __meminit init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005453{
5454 unsigned long lowmem_kbytes;
5455
5456 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
5457
5458 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
5459 if (min_free_kbytes < 128)
5460 min_free_kbytes = 128;
5461 if (min_free_kbytes > 65536)
5462 min_free_kbytes = 65536;
Minchan Kimbc75d332009-06-16 15:32:48 -07005463 setup_per_zone_wmarks();
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -07005464 refresh_zone_stat_thresholds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005465 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07005466 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005467 return 0;
5468}
Minchan Kimbc75d332009-06-16 15:32:48 -07005469module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005470
5471/*
5472 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5473 * that we can call two helper functions whenever min_free_kbytes
5474 * changes.
5475 */
5476int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005477 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005478{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005479 proc_dointvec(table, write, buffer, length, ppos);
Mel Gorman3b1d92c2007-05-06 14:49:30 -07005480 if (write)
Minchan Kimbc75d332009-06-16 15:32:48 -07005481 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005482 return 0;
5483}
5484
Christoph Lameter96146342006-07-03 00:24:13 -07005485#ifdef CONFIG_NUMA
5486int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005487 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07005488{
5489 struct zone *zone;
5490 int rc;
5491
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005492 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07005493 if (rc)
5494 return rc;
5495
5496 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08005497 zone->min_unmapped_pages = (zone->managed_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07005498 sysctl_min_unmapped_ratio) / 100;
5499 return 0;
5500}
Christoph Lameter0ff38492006-09-25 23:31:52 -07005501
5502int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005503 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07005504{
5505 struct zone *zone;
5506 int rc;
5507
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005508 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07005509 if (rc)
5510 return rc;
5511
5512 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08005513 zone->min_slab_pages = (zone->managed_pages *
Christoph Lameter0ff38492006-09-25 23:31:52 -07005514 sysctl_min_slab_ratio) / 100;
5515 return 0;
5516}
Christoph Lameter96146342006-07-03 00:24:13 -07005517#endif
5518
Linus Torvalds1da177e2005-04-16 15:20:36 -07005519/*
5520 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5521 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5522 * whenever sysctl_lowmem_reserve_ratio changes.
5523 *
5524 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07005525 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07005526 * if in function of the boot time zone sizes.
5527 */
5528int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005529 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005530{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005531 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005532 setup_per_zone_lowmem_reserve();
5533 return 0;
5534}
5535
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005536/*
5537 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5538 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5539 * can have before it gets flushed back to buddy allocator.
5540 */
5541
5542int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005543 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005544{
5545 struct zone *zone;
5546 unsigned int cpu;
5547 int ret;
5548
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005549 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
Sasha Levin93278812012-05-10 13:01:44 -07005550 if (!write || (ret < 0))
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005551 return ret;
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07005552 for_each_populated_zone(zone) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005553 for_each_possible_cpu(cpu) {
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005554 unsigned long high;
Jiang Liub40da042013-02-22 16:33:52 -08005555 high = zone->managed_pages / percpu_pagelist_fraction;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005556 setup_pagelist_highmark(
5557 per_cpu_ptr(zone->pageset, cpu), high);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005558 }
5559 }
5560 return 0;
5561}
5562
David S. Millerf034b5d2006-08-24 03:08:07 -07005563int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005564
5565#ifdef CONFIG_NUMA
5566static int __init set_hashdist(char *str)
5567{
5568 if (!str)
5569 return 0;
5570 hashdist = simple_strtoul(str, &str, 0);
5571 return 1;
5572}
5573__setup("hashdist=", set_hashdist);
5574#endif
5575
5576/*
5577 * allocate a large system hash table from bootmem
5578 * - it is assumed that the hash table must contain an exact power-of-2
5579 * quantity of entries
5580 * - limit is the number of hash buckets, not the total allocation size
5581 */
5582void *__init alloc_large_system_hash(const char *tablename,
5583 unsigned long bucketsize,
5584 unsigned long numentries,
5585 int scale,
5586 int flags,
5587 unsigned int *_hash_shift,
5588 unsigned int *_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00005589 unsigned long low_limit,
5590 unsigned long high_limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005591{
Tim Bird31fe62b2012-05-23 13:33:35 +00005592 unsigned long long max = high_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005593 unsigned long log2qty, size;
5594 void *table = NULL;
5595
5596 /* allow the kernel cmdline to have a say */
5597 if (!numentries) {
5598 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08005599 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005600 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
5601 numentries >>= 20 - PAGE_SHIFT;
5602 numentries <<= 20 - PAGE_SHIFT;
5603
5604 /* limit to 1 bucket per 2^scale bytes of low memory */
5605 if (scale > PAGE_SHIFT)
5606 numentries >>= (scale - PAGE_SHIFT);
5607 else
5608 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08005609
5610 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07005611 if (unlikely(flags & HASH_SMALL)) {
5612 /* Makes no sense without HASH_EARLY */
5613 WARN_ON(!(flags & HASH_EARLY));
5614 if (!(numentries >> *_hash_shift)) {
5615 numentries = 1UL << *_hash_shift;
5616 BUG_ON(!numentries);
5617 }
5618 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08005619 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005620 }
John Hawkes6e692ed2006-03-25 03:08:02 -08005621 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005622
5623 /* limit allocation size to 1/16 total memory by default */
5624 if (max == 0) {
5625 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
5626 do_div(max, bucketsize);
5627 }
Dimitri Sivanich074b8512012-02-08 12:39:07 -08005628 max = min(max, 0x80000000ULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005629
Tim Bird31fe62b2012-05-23 13:33:35 +00005630 if (numentries < low_limit)
5631 numentries = low_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005632 if (numentries > max)
5633 numentries = max;
5634
David Howellsf0d1b0b2006-12-08 02:37:49 -08005635 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005636
5637 do {
5638 size = bucketsize << log2qty;
5639 if (flags & HASH_EARLY)
Jan Beulich74768ed2008-08-12 15:08:39 -07005640 table = alloc_bootmem_nopanic(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005641 else if (hashdist)
5642 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
5643 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07005644 /*
5645 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07005646 * some pages at the end of hash table which
5647 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07005648 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005649 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07005650 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005651 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
5652 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005653 }
5654 } while (!table && size > PAGE_SIZE && --log2qty);
5655
5656 if (!table)
5657 panic("Failed to allocate %s hash table\n", tablename);
5658
Robin Holtf241e6602010-10-07 12:59:26 -07005659 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07005660 tablename,
Robin Holtf241e6602010-10-07 12:59:26 -07005661 (1UL << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08005662 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005663 size);
5664
5665 if (_hash_shift)
5666 *_hash_shift = log2qty;
5667 if (_hash_mask)
5668 *_hash_mask = (1 << log2qty) - 1;
5669
5670 return table;
5671}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08005672
Mel Gorman835c1342007-10-16 01:25:47 -07005673/* Return a pointer to the bitmap storing bits affecting a block of pages */
5674static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
5675 unsigned long pfn)
5676{
5677#ifdef CONFIG_SPARSEMEM
5678 return __pfn_to_section(pfn)->pageblock_flags;
5679#else
5680 return zone->pageblock_flags;
5681#endif /* CONFIG_SPARSEMEM */
5682}
Andrew Morton6220ec72006-10-19 23:29:05 -07005683
Mel Gorman835c1342007-10-16 01:25:47 -07005684static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
5685{
5686#ifdef CONFIG_SPARSEMEM
5687 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07005688 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005689#else
Laura Abbottc060f942013-01-11 14:31:51 -08005690 pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
Mel Gormand9c23402007-10-16 01:26:01 -07005691 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005692#endif /* CONFIG_SPARSEMEM */
5693}
5694
5695/**
Mel Gormand9c23402007-10-16 01:26:01 -07005696 * 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 -07005697 * @page: The page within the block of interest
5698 * @start_bitidx: The first bit of interest to retrieve
5699 * @end_bitidx: The last bit of interest
5700 * returns pageblock_bits flags
5701 */
5702unsigned long get_pageblock_flags_group(struct page *page,
5703 int start_bitidx, int end_bitidx)
5704{
5705 struct zone *zone;
5706 unsigned long *bitmap;
5707 unsigned long pfn, bitidx;
5708 unsigned long flags = 0;
5709 unsigned long value = 1;
5710
5711 zone = page_zone(page);
5712 pfn = page_to_pfn(page);
5713 bitmap = get_pageblock_bitmap(zone, pfn);
5714 bitidx = pfn_to_bitidx(zone, pfn);
5715
5716 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5717 if (test_bit(bitidx + start_bitidx, bitmap))
5718 flags |= value;
5719
5720 return flags;
5721}
5722
5723/**
Mel Gormand9c23402007-10-16 01:26:01 -07005724 * 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 -07005725 * @page: The page within the block of interest
5726 * @start_bitidx: The first bit of interest
5727 * @end_bitidx: The last bit of interest
5728 * @flags: The flags to set
5729 */
5730void set_pageblock_flags_group(struct page *page, unsigned long flags,
5731 int start_bitidx, int end_bitidx)
5732{
5733 struct zone *zone;
5734 unsigned long *bitmap;
5735 unsigned long pfn, bitidx;
5736 unsigned long value = 1;
5737
5738 zone = page_zone(page);
5739 pfn = page_to_pfn(page);
5740 bitmap = get_pageblock_bitmap(zone, pfn);
5741 bitidx = pfn_to_bitidx(zone, pfn);
Cody P Schafer108bcc92013-02-22 16:35:23 -08005742 VM_BUG_ON(!zone_spans_pfn(zone, pfn));
Mel Gorman835c1342007-10-16 01:25:47 -07005743
5744 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5745 if (flags & value)
5746 __set_bit(bitidx + start_bitidx, bitmap);
5747 else
5748 __clear_bit(bitidx + start_bitidx, bitmap);
5749}
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07005750
5751/*
Minchan Kim80934512012-07-31 16:43:01 -07005752 * This function checks whether pageblock includes unmovable pages or not.
5753 * If @count is not zero, it is okay to include less @count unmovable pages
5754 *
5755 * PageLRU check wihtout isolation or lru_lock could race so that
5756 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
5757 * expect this function should be exact.
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07005758 */
Wen Congyangb023f462012-12-11 16:00:45 -08005759bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
5760 bool skip_hwpoisoned_pages)
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005761{
5762 unsigned long pfn, iter, found;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01005763 int mt;
5764
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005765 /*
5766 * For avoiding noise data, lru_add_drain_all() should be called
Minchan Kim80934512012-07-31 16:43:01 -07005767 * If ZONE_MOVABLE, the zone never contains unmovable pages
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005768 */
5769 if (zone_idx(zone) == ZONE_MOVABLE)
Minchan Kim80934512012-07-31 16:43:01 -07005770 return false;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01005771 mt = get_pageblock_migratetype(page);
5772 if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
Minchan Kim80934512012-07-31 16:43:01 -07005773 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005774
5775 pfn = page_to_pfn(page);
5776 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
5777 unsigned long check = pfn + iter;
5778
Namhyung Kim29723fc2011-02-25 14:44:25 -08005779 if (!pfn_valid_within(check))
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005780 continue;
Namhyung Kim29723fc2011-02-25 14:44:25 -08005781
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005782 page = pfn_to_page(check);
Minchan Kim97d255c2012-07-31 16:42:59 -07005783 /*
5784 * We can't use page_count without pin a page
5785 * because another CPU can free compound page.
5786 * This check already skips compound tails of THP
5787 * because their page->_count is zero at all time.
5788 */
5789 if (!atomic_read(&page->_count)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005790 if (PageBuddy(page))
5791 iter += (1 << page_order(page)) - 1;
5792 continue;
5793 }
Minchan Kim97d255c2012-07-31 16:42:59 -07005794
Wen Congyangb023f462012-12-11 16:00:45 -08005795 /*
5796 * The HWPoisoned page may be not in buddy system, and
5797 * page_count() is not 0.
5798 */
5799 if (skip_hwpoisoned_pages && PageHWPoison(page))
5800 continue;
5801
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005802 if (!PageLRU(page))
5803 found++;
5804 /*
5805 * If there are RECLAIMABLE pages, we need to check it.
5806 * But now, memory offline itself doesn't call shrink_slab()
5807 * and it still to be fixed.
5808 */
5809 /*
5810 * If the page is not RAM, page_count()should be 0.
5811 * we don't need more check. This is an _used_ not-movable page.
5812 *
5813 * The problematic thing here is PG_reserved pages. PG_reserved
5814 * is set to both of a memory hole page and a _used_ kernel
5815 * page at boot.
5816 */
5817 if (found > count)
Minchan Kim80934512012-07-31 16:43:01 -07005818 return true;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005819 }
Minchan Kim80934512012-07-31 16:43:01 -07005820 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005821}
5822
5823bool is_pageblock_removable_nolock(struct page *page)
5824{
Michal Hocko656a0702012-01-20 14:33:58 -08005825 struct zone *zone;
5826 unsigned long pfn;
Michal Hocko687875f2012-01-20 14:33:55 -08005827
5828 /*
5829 * We have to be careful here because we are iterating over memory
5830 * sections which are not zone aware so we might end up outside of
5831 * the zone but still within the section.
Michal Hocko656a0702012-01-20 14:33:58 -08005832 * We have to take care about the node as well. If the node is offline
5833 * its NODE_DATA will be NULL - see page_zone.
Michal Hocko687875f2012-01-20 14:33:55 -08005834 */
Michal Hocko656a0702012-01-20 14:33:58 -08005835 if (!node_online(page_to_nid(page)))
5836 return false;
5837
5838 zone = page_zone(page);
5839 pfn = page_to_pfn(page);
Cody P Schafer108bcc92013-02-22 16:35:23 -08005840 if (!zone_spans_pfn(zone, pfn))
Michal Hocko687875f2012-01-20 14:33:55 -08005841 return false;
5842
Wen Congyangb023f462012-12-11 16:00:45 -08005843 return !has_unmovable_pages(zone, page, 0, true);
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07005844}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07005845
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01005846#ifdef CONFIG_CMA
5847
5848static unsigned long pfn_max_align_down(unsigned long pfn)
5849{
5850 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
5851 pageblock_nr_pages) - 1);
5852}
5853
5854static unsigned long pfn_max_align_up(unsigned long pfn)
5855{
5856 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
5857 pageblock_nr_pages));
5858}
5859
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01005860/* [start, end) must belong to a single zone. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07005861static int __alloc_contig_migrate_range(struct compact_control *cc,
5862 unsigned long start, unsigned long end)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01005863{
5864 /* This function is based on compact_zone() from compaction.c. */
Minchan Kimbeb51ea2012-10-08 16:33:51 -07005865 unsigned long nr_reclaimed;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01005866 unsigned long pfn = start;
5867 unsigned int tries = 0;
5868 int ret = 0;
5869
Marek Szyprowskibe49a6e2012-12-12 13:51:19 -08005870 migrate_prep();
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01005871
Mel Gormanbb13ffe2012-10-08 16:32:41 -07005872 while (pfn < end || !list_empty(&cc->migratepages)) {
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01005873 if (fatal_signal_pending(current)) {
5874 ret = -EINTR;
5875 break;
5876 }
5877
Mel Gormanbb13ffe2012-10-08 16:32:41 -07005878 if (list_empty(&cc->migratepages)) {
5879 cc->nr_migratepages = 0;
5880 pfn = isolate_migratepages_range(cc->zone, cc,
Minchan Kime46a2872012-10-08 16:33:48 -07005881 pfn, end, true);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01005882 if (!pfn) {
5883 ret = -EINTR;
5884 break;
5885 }
5886 tries = 0;
5887 } else if (++tries == 5) {
5888 ret = ret < 0 ? ret : -EBUSY;
5889 break;
5890 }
5891
Minchan Kimbeb51ea2012-10-08 16:33:51 -07005892 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
5893 &cc->migratepages);
5894 cc->nr_migratepages -= nr_reclaimed;
Minchan Kim02c6de82012-10-08 16:31:55 -07005895
Hugh Dickins9c620e22013-02-22 16:35:14 -08005896 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
5897 0, MIGRATE_SYNC, MR_CMA);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01005898 }
Srinivas Pandruvada2a6f5122013-02-22 16:32:09 -08005899 if (ret < 0) {
5900 putback_movable_pages(&cc->migratepages);
5901 return ret;
5902 }
5903 return 0;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01005904}
5905
5906/**
5907 * alloc_contig_range() -- tries to allocate given range of pages
5908 * @start: start PFN to allocate
5909 * @end: one-past-the-last PFN to allocate
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02005910 * @migratetype: migratetype of the underlaying pageblocks (either
5911 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
5912 * in range must have the same migratetype and it must
5913 * be either of the two.
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01005914 *
5915 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
5916 * aligned, however it's the caller's responsibility to guarantee that
5917 * we are the only thread that changes migrate type of pageblocks the
5918 * pages fall in.
5919 *
5920 * The PFN range must belong to a single zone.
5921 *
5922 * Returns zero on success or negative error code. On success all
5923 * pages which PFN is in [start, end) are allocated for the caller and
5924 * need to be freed with free_contig_range().
5925 */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02005926int alloc_contig_range(unsigned long start, unsigned long end,
5927 unsigned migratetype)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01005928{
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01005929 unsigned long outer_start, outer_end;
5930 int ret = 0, order;
5931
Mel Gormanbb13ffe2012-10-08 16:32:41 -07005932 struct compact_control cc = {
5933 .nr_migratepages = 0,
5934 .order = -1,
5935 .zone = page_zone(pfn_to_page(start)),
5936 .sync = true,
5937 .ignore_skip_hint = true,
5938 };
5939 INIT_LIST_HEAD(&cc.migratepages);
5940
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01005941 /*
5942 * What we do here is we mark all pageblocks in range as
5943 * MIGRATE_ISOLATE. Because pageblock and max order pages may
5944 * have different sizes, and due to the way page allocator
5945 * work, we align the range to biggest of the two pages so
5946 * that page allocator won't try to merge buddies from
5947 * different pageblocks and change MIGRATE_ISOLATE to some
5948 * other migration type.
5949 *
5950 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
5951 * migrate the pages from an unaligned range (ie. pages that
5952 * we are interested in). This will put all the pages in
5953 * range back to page allocator as MIGRATE_ISOLATE.
5954 *
5955 * When this is done, we take the pages in range from page
5956 * allocator removing them from the buddy system. This way
5957 * page allocator will never consider using them.
5958 *
5959 * This lets us mark the pageblocks back as
5960 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
5961 * aligned range but not in the unaligned, original range are
5962 * put back to page allocator so that buddy can use them.
5963 */
5964
5965 ret = start_isolate_page_range(pfn_max_align_down(start),
Wen Congyangb023f462012-12-11 16:00:45 -08005966 pfn_max_align_up(end), migratetype,
5967 false);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01005968 if (ret)
Bob Liu86a595f2012-10-25 13:37:56 -07005969 return ret;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01005970
Mel Gormanbb13ffe2012-10-08 16:32:41 -07005971 ret = __alloc_contig_migrate_range(&cc, start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01005972 if (ret)
5973 goto done;
5974
5975 /*
5976 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
5977 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
5978 * more, all pages in [start, end) are free in page allocator.
5979 * What we are going to do is to allocate all pages from
5980 * [start, end) (that is remove them from page allocator).
5981 *
5982 * The only problem is that pages at the beginning and at the
5983 * end of interesting range may be not aligned with pages that
5984 * page allocator holds, ie. they can be part of higher order
5985 * pages. Because of this, we reserve the bigger range and
5986 * once this is done free the pages we are not interested in.
5987 *
5988 * We don't have to hold zone->lock here because the pages are
5989 * isolated thus they won't get removed from buddy.
5990 */
5991
5992 lru_add_drain_all();
5993 drain_all_pages();
5994
5995 order = 0;
5996 outer_start = start;
5997 while (!PageBuddy(pfn_to_page(outer_start))) {
5998 if (++order >= MAX_ORDER) {
5999 ret = -EBUSY;
6000 goto done;
6001 }
6002 outer_start &= ~0UL << order;
6003 }
6004
6005 /* Make sure the range is really isolated. */
Wen Congyangb023f462012-12-11 16:00:45 -08006006 if (test_pages_isolated(outer_start, end, false)) {
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006007 pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n",
6008 outer_start, end);
6009 ret = -EBUSY;
6010 goto done;
6011 }
6012
Marek Szyprowski49f223a2012-01-25 12:49:24 +01006013
6014 /* Grab isolated pages from freelists. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006015 outer_end = isolate_freepages_range(&cc, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006016 if (!outer_end) {
6017 ret = -EBUSY;
6018 goto done;
6019 }
6020
6021 /* Free head and tail (if any) */
6022 if (start != outer_start)
6023 free_contig_range(outer_start, start - outer_start);
6024 if (end != outer_end)
6025 free_contig_range(end, outer_end - end);
6026
6027done:
6028 undo_isolate_page_range(pfn_max_align_down(start),
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006029 pfn_max_align_up(end), migratetype);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006030 return ret;
6031}
6032
6033void free_contig_range(unsigned long pfn, unsigned nr_pages)
6034{
Marek Szyprowskibcc2b022012-12-20 15:05:18 -08006035 unsigned int count = 0;
6036
6037 for (; nr_pages--; pfn++) {
6038 struct page *page = pfn_to_page(pfn);
6039
6040 count += page_count(page) != 1;
6041 __free_page(page);
6042 }
6043 WARN(count != 0, "%d pages are still in use!\n", count);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006044}
6045#endif
6046
Jiang Liu4ed7e022012-07-31 16:43:35 -07006047#ifdef CONFIG_MEMORY_HOTPLUG
6048static int __meminit __zone_pcp_update(void *data)
6049{
6050 struct zone *zone = data;
6051 int cpu;
6052 unsigned long batch = zone_batchsize(zone), flags;
6053
6054 for_each_possible_cpu(cpu) {
6055 struct per_cpu_pageset *pset;
6056 struct per_cpu_pages *pcp;
6057
6058 pset = per_cpu_ptr(zone->pageset, cpu);
6059 pcp = &pset->pcp;
6060
6061 local_irq_save(flags);
6062 if (pcp->count > 0)
6063 free_pcppages_bulk(zone, pcp->count, pcp);
Minchan Kim5a883812012-10-08 16:33:39 -07006064 drain_zonestat(zone, pset);
Jiang Liu4ed7e022012-07-31 16:43:35 -07006065 setup_pageset(pset, batch);
6066 local_irq_restore(flags);
6067 }
6068 return 0;
6069}
6070
6071void __meminit zone_pcp_update(struct zone *zone)
6072{
6073 stop_machine(__zone_pcp_update, zone, NULL);
6074}
6075#endif
6076
Jiang Liu340175b2012-07-31 16:43:32 -07006077void zone_pcp_reset(struct zone *zone)
6078{
6079 unsigned long flags;
Minchan Kim5a883812012-10-08 16:33:39 -07006080 int cpu;
6081 struct per_cpu_pageset *pset;
Jiang Liu340175b2012-07-31 16:43:32 -07006082
6083 /* avoid races with drain_pages() */
6084 local_irq_save(flags);
6085 if (zone->pageset != &boot_pageset) {
Minchan Kim5a883812012-10-08 16:33:39 -07006086 for_each_online_cpu(cpu) {
6087 pset = per_cpu_ptr(zone->pageset, cpu);
6088 drain_zonestat(zone, pset);
6089 }
Jiang Liu340175b2012-07-31 16:43:32 -07006090 free_percpu(zone->pageset);
6091 zone->pageset = &boot_pageset;
6092 }
6093 local_irq_restore(flags);
6094}
6095
Wen Congyang6dcd73d2012-12-11 16:01:01 -08006096#ifdef CONFIG_MEMORY_HOTREMOVE
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006097/*
6098 * All pages in the range must be isolated before calling this.
6099 */
6100void
6101__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
6102{
6103 struct page *page;
6104 struct zone *zone;
6105 int order, i;
6106 unsigned long pfn;
6107 unsigned long flags;
6108 /* find the first valid pfn */
6109 for (pfn = start_pfn; pfn < end_pfn; pfn++)
6110 if (pfn_valid(pfn))
6111 break;
6112 if (pfn == end_pfn)
6113 return;
6114 zone = page_zone(pfn_to_page(pfn));
6115 spin_lock_irqsave(&zone->lock, flags);
6116 pfn = start_pfn;
6117 while (pfn < end_pfn) {
6118 if (!pfn_valid(pfn)) {
6119 pfn++;
6120 continue;
6121 }
6122 page = pfn_to_page(pfn);
Wen Congyangb023f462012-12-11 16:00:45 -08006123 /*
6124 * The HWPoisoned page may be not in buddy system, and
6125 * page_count() is not 0.
6126 */
6127 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
6128 pfn++;
6129 SetPageReserved(page);
6130 continue;
6131 }
6132
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006133 BUG_ON(page_count(page));
6134 BUG_ON(!PageBuddy(page));
6135 order = page_order(page);
6136#ifdef CONFIG_DEBUG_VM
6137 printk(KERN_INFO "remove from free list %lx %d %lx\n",
6138 pfn, 1 << order, end_pfn);
6139#endif
6140 list_del(&page->lru);
6141 rmv_page_order(page);
6142 zone->free_area[order].nr_free--;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006143 for (i = 0; i < (1 << order); i++)
6144 SetPageReserved((page+i));
6145 pfn += (1 << order);
6146 }
6147 spin_unlock_irqrestore(&zone->lock, flags);
6148}
6149#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01006150
6151#ifdef CONFIG_MEMORY_FAILURE
6152bool is_free_buddy_page(struct page *page)
6153{
6154 struct zone *zone = page_zone(page);
6155 unsigned long pfn = page_to_pfn(page);
6156 unsigned long flags;
6157 int order;
6158
6159 spin_lock_irqsave(&zone->lock, flags);
6160 for (order = 0; order < MAX_ORDER; order++) {
6161 struct page *page_head = page - (pfn & ((1 << order) - 1));
6162
6163 if (PageBuddy(page_head) && page_order(page_head) >= order)
6164 break;
6165 }
6166 spin_unlock_irqrestore(&zone->lock, flags);
6167
6168 return order < MAX_ORDER;
6169}
6170#endif
Wu Fengguang718a3822010-03-10 15:20:43 -08006171
Andrew Morton51300ce2012-05-29 15:06:44 -07006172static const struct trace_print_flags pageflag_names[] = {
Wu Fengguang718a3822010-03-10 15:20:43 -08006173 {1UL << PG_locked, "locked" },
6174 {1UL << PG_error, "error" },
6175 {1UL << PG_referenced, "referenced" },
6176 {1UL << PG_uptodate, "uptodate" },
6177 {1UL << PG_dirty, "dirty" },
6178 {1UL << PG_lru, "lru" },
6179 {1UL << PG_active, "active" },
6180 {1UL << PG_slab, "slab" },
6181 {1UL << PG_owner_priv_1, "owner_priv_1" },
6182 {1UL << PG_arch_1, "arch_1" },
6183 {1UL << PG_reserved, "reserved" },
6184 {1UL << PG_private, "private" },
6185 {1UL << PG_private_2, "private_2" },
6186 {1UL << PG_writeback, "writeback" },
6187#ifdef CONFIG_PAGEFLAGS_EXTENDED
6188 {1UL << PG_head, "head" },
6189 {1UL << PG_tail, "tail" },
6190#else
6191 {1UL << PG_compound, "compound" },
6192#endif
6193 {1UL << PG_swapcache, "swapcache" },
6194 {1UL << PG_mappedtodisk, "mappedtodisk" },
6195 {1UL << PG_reclaim, "reclaim" },
Wu Fengguang718a3822010-03-10 15:20:43 -08006196 {1UL << PG_swapbacked, "swapbacked" },
6197 {1UL << PG_unevictable, "unevictable" },
6198#ifdef CONFIG_MMU
6199 {1UL << PG_mlocked, "mlocked" },
6200#endif
6201#ifdef CONFIG_ARCH_USES_PG_UNCACHED
6202 {1UL << PG_uncached, "uncached" },
6203#endif
6204#ifdef CONFIG_MEMORY_FAILURE
6205 {1UL << PG_hwpoison, "hwpoison" },
6206#endif
Gavin Shanbe9cd872012-05-29 15:06:44 -07006207#ifdef CONFIG_TRANSPARENT_HUGEPAGE
6208 {1UL << PG_compound_lock, "compound_lock" },
6209#endif
Wu Fengguang718a3822010-03-10 15:20:43 -08006210};
6211
6212static void dump_page_flags(unsigned long flags)
6213{
6214 const char *delim = "";
6215 unsigned long mask;
6216 int i;
6217
Andrew Morton51300ce2012-05-29 15:06:44 -07006218 BUILD_BUG_ON(ARRAY_SIZE(pageflag_names) != __NR_PAGEFLAGS);
Johannes Weineracc50c12012-05-29 15:06:44 -07006219
Wu Fengguang718a3822010-03-10 15:20:43 -08006220 printk(KERN_ALERT "page flags: %#lx(", flags);
6221
6222 /* remove zone id */
6223 flags &= (1UL << NR_PAGEFLAGS) - 1;
6224
Andrew Morton51300ce2012-05-29 15:06:44 -07006225 for (i = 0; i < ARRAY_SIZE(pageflag_names) && flags; i++) {
Wu Fengguang718a3822010-03-10 15:20:43 -08006226
6227 mask = pageflag_names[i].mask;
6228 if ((flags & mask) != mask)
6229 continue;
6230
6231 flags &= ~mask;
6232 printk("%s%s", delim, pageflag_names[i].name);
6233 delim = "|";
6234 }
6235
6236 /* check for left over flags */
6237 if (flags)
6238 printk("%s%#lx", delim, flags);
6239
6240 printk(")\n");
6241}
6242
6243void dump_page(struct page *page)
6244{
6245 printk(KERN_ALERT
6246 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
Andrea Arcangeli4e9f64c2011-01-13 15:46:29 -08006247 page, atomic_read(&page->_count), page_mapcount(page),
Wu Fengguang718a3822010-03-10 15:20:43 -08006248 page->mapping, page->index);
6249 dump_page_flags(page->flags);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07006250 mem_cgroup_print_bad_page(page);
Wu Fengguang718a3822010-03-10 15:20:43 -08006251}