blob: d2186ecb36f7cdfd5fdc75da4bbd10ef798bcbba [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/page_alloc.c
3 *
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
6 *
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
15 */
16
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/stddef.h>
18#include <linux/mm.h>
19#include <linux/swap.h>
20#include <linux/interrupt.h>
21#include <linux/pagemap.h>
KOSAKI Motohiro10ed2732008-03-04 14:28:32 -080022#include <linux/jiffies.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/bootmem.h>
Yinghai Luedbe7d22010-08-25 13:39:16 -070024#include <linux/memblock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/compiler.h>
Randy Dunlap9f158332005-09-13 01:25:16 -070026#include <linux/kernel.h>
Vegard Nossumb1eeab62008-11-25 16:55:53 +010027#include <linux/kmemcheck.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/module.h>
29#include <linux/suspend.h>
30#include <linux/pagevec.h>
31#include <linux/blkdev.h>
32#include <linux/slab.h>
Dave Hansena238ab52011-05-24 17:12:16 -070033#include <linux/ratelimit.h>
David Rientjes5a3135c22007-10-16 23:25:53 -070034#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <linux/notifier.h>
36#include <linux/topology.h>
37#include <linux/sysctl.h>
38#include <linux/cpu.h>
39#include <linux/cpuset.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070040#include <linux/memory_hotplug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include <linux/nodemask.h>
42#include <linux/vmalloc.h>
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -070043#include <linux/vmstat.h>
Christoph Lameter4be38e32006-01-06 00:11:17 -080044#include <linux/mempolicy.h>
Yasunori Goto68113782006-06-23 02:03:11 -070045#include <linux/stop_machine.h>
Mel Gormanc7132162006-09-27 01:49:43 -070046#include <linux/sort.h>
47#include <linux/pfn.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070048#include <linux/backing-dev.h>
Akinobu Mita933e3122006-12-08 02:39:45 -080049#include <linux/fault-inject.h>
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -070050#include <linux/page-isolation.h>
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070051#include <linux/page_cgroup.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070052#include <linux/debugobjects.h>
Catalin Marinasdbb1f812009-06-11 13:23:19 +010053#include <linux/kmemleak.h>
Robert Jennings925cc712009-12-17 14:44:38 +000054#include <linux/memory.h>
Mel Gorman56de7262010-05-24 14:32:30 -070055#include <linux/compaction.h>
Mel Gorman0d3d0622009-09-21 17:02:44 -070056#include <trace/events/kmem.h>
Wu Fengguang718a3822010-03-10 15:20:43 -080057#include <linux/ftrace_event.h>
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -070058#include <linux/memcontrol.h>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070059#include <linux/prefetch.h>
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -080060#include <linux/page-debug-flags.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
62#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070063#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070064#include "internal.h"
65
Lee Schermerhorn72812012010-05-26 14:44:56 -070066#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
67DEFINE_PER_CPU(int, numa_node);
68EXPORT_PER_CPU_SYMBOL(numa_node);
69#endif
70
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070071#ifdef CONFIG_HAVE_MEMORYLESS_NODES
72/*
73 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
74 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
75 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
76 * defined in <linux/topology.h>.
77 */
78DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
79EXPORT_PER_CPU_SYMBOL(_numa_mem_);
80#endif
81
Linus Torvalds1da177e2005-04-16 15:20:36 -070082/*
Christoph Lameter13808912007-10-16 01:25:27 -070083 * Array of node states.
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 */
Christoph Lameter13808912007-10-16 01:25:27 -070085nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
86 [N_POSSIBLE] = NODE_MASK_ALL,
87 [N_ONLINE] = { { [0] = 1UL } },
88#ifndef CONFIG_NUMA
89 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
90#ifdef CONFIG_HIGHMEM
91 [N_HIGH_MEMORY] = { { [0] = 1UL } },
92#endif
93 [N_CPU] = { { [0] = 1UL } },
94#endif /* NUMA */
95};
96EXPORT_SYMBOL(node_states);
97
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070098unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -070099unsigned long totalreserve_pages __read_mostly;
Johannes Weinerab8fabd2012-01-10 15:07:42 -0800100/*
101 * When calculating the number of globally allowed dirty pages, there
102 * is a certain number of per-zone reserves that should not be
103 * considered dirtyable memory. This is the sum of those reserves
104 * over all existing zones that contribute dirtyable memory.
105 */
106unsigned long dirty_balance_reserve __read_mostly;
107
Rohit Seth8ad4b1f2006-01-08 01:00:40 -0800108int percpu_pagelist_fraction;
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000109gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800111#ifdef CONFIG_PM_SLEEP
112/*
113 * The following functions are used by the suspend/hibernate code to temporarily
114 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
115 * while devices are suspended. To avoid races with the suspend/hibernate code,
116 * they should always be called with pm_mutex held (gfp_allowed_mask also should
117 * only be modified with pm_mutex held, unless the suspend/hibernate code is
118 * guaranteed not to run in parallel with that modification).
119 */
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100120
121static gfp_t saved_gfp_mask;
122
123void pm_restore_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800124{
125 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100126 if (saved_gfp_mask) {
127 gfp_allowed_mask = saved_gfp_mask;
128 saved_gfp_mask = 0;
129 }
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800130}
131
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100132void pm_restrict_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800133{
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800134 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100135 WARN_ON(saved_gfp_mask);
136 saved_gfp_mask = gfp_allowed_mask;
137 gfp_allowed_mask &= ~GFP_IOFS;
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800138}
Mel Gormanf90ac392012-01-10 15:07:15 -0800139
140bool pm_suspended_storage(void)
141{
142 if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS)
143 return false;
144 return true;
145}
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800146#endif /* CONFIG_PM_SLEEP */
147
Mel Gormand9c23402007-10-16 01:26:01 -0700148#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
149int pageblock_order __read_mostly;
150#endif
151
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800152static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -0800153
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154/*
155 * results with 256, 32 in the lowmem_reserve sysctl:
156 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
157 * 1G machine -> (16M dma, 784M normal, 224M high)
158 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
159 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
160 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +0100161 *
162 * TBD: should special case ZONE_DMA32 machines here - in those we normally
163 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700165int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800166#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700167 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -0800168#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700169#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700170 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700171#endif
Christoph Lametere53ef382006-09-25 23:31:14 -0700172#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700173 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700174#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700175 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700176};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177
178EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179
Helge Deller15ad7cd2006-12-06 20:40:36 -0800180static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800181#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700182 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800183#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700184#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700185 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700186#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700187 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700188#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700189 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700190#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700191 "Movable",
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700192};
193
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194int min_free_kbytes = 1024;
195
Jan Beulich2c85f512009-09-21 17:03:07 -0700196static unsigned long __meminitdata nr_kernel_pages;
197static unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700198static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199
Tejun Heo0ee332c2011-12-08 10:22:09 -0800200#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
201static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
202static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
203static unsigned long __initdata required_kernelcore;
204static unsigned long __initdata required_movablecore;
205static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Mel Gormanc7132162006-09-27 01:49:43 -0700206
Tejun Heo0ee332c2011-12-08 10:22:09 -0800207/* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
208int movable_zone;
209EXPORT_SYMBOL(movable_zone);
210#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -0700211
Miklos Szeredi418508c2007-05-23 13:57:55 -0700212#if MAX_NUMNODES > 1
213int nr_node_ids __read_mostly = MAX_NUMNODES;
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700214int nr_online_nodes __read_mostly = 1;
Miklos Szeredi418508c2007-05-23 13:57:55 -0700215EXPORT_SYMBOL(nr_node_ids);
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700216EXPORT_SYMBOL(nr_online_nodes);
Miklos Szeredi418508c2007-05-23 13:57:55 -0700217#endif
218
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700219int page_group_by_mobility_disabled __read_mostly;
220
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700221static void set_pageblock_migratetype(struct page *page, int migratetype)
222{
Mel Gorman49255c62009-06-16 15:31:58 -0700223
224 if (unlikely(page_group_by_mobility_disabled))
225 migratetype = MIGRATE_UNMOVABLE;
226
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700227 set_pageblock_flags_group(page, (unsigned long)migratetype,
228 PB_migrate, PB_migrate_end);
229}
230
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -0700231bool oom_killer_disabled __read_mostly;
232
Nick Piggin13e74442006-01-06 00:10:58 -0800233#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700234static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700236 int ret = 0;
237 unsigned seq;
238 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -0700239
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700240 do {
241 seq = zone_span_seqbegin(zone);
242 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
243 ret = 1;
244 else if (pfn < zone->zone_start_pfn)
245 ret = 1;
246 } while (zone_span_seqretry(zone, seq));
247
248 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700249}
250
251static int page_is_consistent(struct zone *zone, struct page *page)
252{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700253 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700254 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700256 return 0;
257
258 return 1;
259}
260/*
261 * Temporary debugging check for pages not lying within a given zone.
262 */
263static int bad_range(struct zone *zone, struct page *page)
264{
265 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700267 if (!page_is_consistent(zone, page))
268 return 1;
269
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 return 0;
271}
Nick Piggin13e74442006-01-06 00:10:58 -0800272#else
273static inline int bad_range(struct zone *zone, struct page *page)
274{
275 return 0;
276}
277#endif
278
Nick Piggin224abf92006-01-06 00:11:11 -0800279static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800281 static unsigned long resume;
282 static unsigned long nr_shown;
283 static unsigned long nr_unshown;
284
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200285 /* Don't complain about poisoned pages */
286 if (PageHWPoison(page)) {
Andrea Arcangelief2b4b92011-03-18 00:16:35 +0100287 reset_page_mapcount(page); /* remove PageBuddy */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200288 return;
289 }
290
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800291 /*
292 * Allow a burst of 60 reports, then keep quiet for that minute;
293 * or allow a steady drip of one report per second.
294 */
295 if (nr_shown == 60) {
296 if (time_before(jiffies, resume)) {
297 nr_unshown++;
298 goto out;
299 }
300 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800301 printk(KERN_ALERT
302 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800303 nr_unshown);
304 nr_unshown = 0;
305 }
306 nr_shown = 0;
307 }
308 if (nr_shown++ == 0)
309 resume = jiffies + 60 * HZ;
310
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800311 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800312 current->comm, page_to_pfn(page));
Wu Fengguang718a3822010-03-10 15:20:43 -0800313 dump_page(page);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700314
Dave Jones4f318882011-10-31 17:07:24 -0700315 print_modules();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800317out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800318 /* Leave bad fields for debug, except PageBuddy could make trouble */
Andrea Arcangelief2b4b92011-03-18 00:16:35 +0100319 reset_page_mapcount(page); /* remove PageBuddy */
Randy Dunlap9f158332005-09-13 01:25:16 -0700320 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321}
322
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323/*
324 * Higher-order pages are called "compound pages". They are structured thusly:
325 *
326 * The first PAGE_SIZE page is called the "head page".
327 *
328 * The remaining PAGE_SIZE pages are called "tail pages".
329 *
Wang Sheng-Hui6416b9f2011-11-17 10:53:50 +0100330 * All pages have PG_compound set. All tail pages have their ->first_page
331 * pointing at the head page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800333 * The first tail page's ->lru.next holds the address of the compound page's
334 * put_page() function. Its ->lru.prev holds the order of allocation.
335 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800337
338static void free_compound_page(struct page *page)
339{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700340 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800341}
342
Andi Kleen01ad1c02008-07-23 21:27:46 -0700343void prep_compound_page(struct page *page, unsigned long order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344{
345 int i;
346 int nr_pages = 1 << order;
347
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800348 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700349 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700350 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800351 for (i = 1; i < nr_pages; i++) {
352 struct page *p = page + i;
Christoph Lameterd85f3382007-05-06 14:49:39 -0700353 __SetPageTail(p);
Youquan Song58a84aa2011-12-08 14:34:18 -0800354 set_page_count(p, 0);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700355 p->first_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 }
357}
358
Andrea Arcangeli59ff4212011-01-13 15:46:38 -0800359/* update __split_huge_page_refcount if you change this function */
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800360static int destroy_compound_page(struct page *page, unsigned long order)
Andy Whitcroft18229df2008-11-06 12:53:27 -0800361{
362 int i;
363 int nr_pages = 1 << order;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800364 int bad = 0;
Andy Whitcroft18229df2008-11-06 12:53:27 -0800365
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800366 if (unlikely(compound_order(page) != order) ||
367 unlikely(!PageHead(page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800368 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800369 bad++;
370 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371
Christoph Lameter6d777952007-05-06 14:49:40 -0700372 __ClearPageHead(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800373
Andy Whitcroft18229df2008-11-06 12:53:27 -0800374 for (i = 1; i < nr_pages; i++) {
375 struct page *p = page + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376
Alexey Zaytseve713a212009-01-10 02:47:57 +0300377 if (unlikely(!PageTail(p) || (p->first_page != page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800378 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800379 bad++;
380 }
Christoph Lameterd85f3382007-05-06 14:49:39 -0700381 __ClearPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800383
384 return bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386
Nick Piggin17cf4402006-03-22 00:08:41 -0800387static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
388{
389 int i;
390
Andrew Morton6626c5d2006-03-22 00:08:42 -0800391 /*
392 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
393 * and __GFP_HIGHMEM from hard or soft interrupt context.
394 */
Nick Piggin725d7042006-09-25 23:30:55 -0700395 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800396 for (i = 0; i < (1 << order); i++)
397 clear_highpage(page + i);
398}
399
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800400#ifdef CONFIG_DEBUG_PAGEALLOC
401unsigned int _debug_guardpage_minorder;
402
403static int __init debug_guardpage_minorder_setup(char *buf)
404{
405 unsigned long res;
406
407 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
408 printk(KERN_ERR "Bad debug_guardpage_minorder value\n");
409 return 0;
410 }
411 _debug_guardpage_minorder = res;
412 printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res);
413 return 0;
414}
415__setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
416
417static inline void set_page_guard_flag(struct page *page)
418{
419 __set_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags);
420}
421
422static inline void clear_page_guard_flag(struct page *page)
423{
424 __clear_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags);
425}
426#else
427static inline void set_page_guard_flag(struct page *page) { }
428static inline void clear_page_guard_flag(struct page *page) { }
429#endif
430
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700431static inline void set_page_order(struct page *page, int order)
432{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700433 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000434 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435}
436
437static inline void rmv_page_order(struct page *page)
438{
Nick Piggin676165a2006-04-10 11:21:48 +1000439 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700440 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441}
442
443/*
444 * Locate the struct page for both the matching buddy in our
445 * pair (buddy1) and the combined O(n+1) page they form (page).
446 *
447 * 1) Any buddy B1 will have an order O twin B2 which satisfies
448 * the following equation:
449 * B2 = B1 ^ (1 << O)
450 * For example, if the starting buddy (buddy2) is #8 its order
451 * 1 buddy is #10:
452 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
453 *
454 * 2) Any buddy B will have an order O+1 parent P which
455 * satisfies the following equation:
456 * P = B & ~(1 << O)
457 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200458 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460static inline unsigned long
KyongHo Cho43506fa2011-01-13 15:47:24 -0800461__find_buddy_index(unsigned long page_idx, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462{
KyongHo Cho43506fa2011-01-13 15:47:24 -0800463 return page_idx ^ (1 << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464}
465
466/*
467 * This function checks whether a page is free && is the buddy
468 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800469 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000470 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700471 * (c) a page and its buddy have the same order &&
472 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 *
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800474 * For recording whether a page is in the buddy system, we set ->_mapcount -2.
475 * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock.
Nick Piggin676165a2006-04-10 11:21:48 +1000476 *
477 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700479static inline int page_is_buddy(struct page *page, struct page *buddy,
480 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700482 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800483 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800484
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700485 if (page_zone_id(page) != page_zone_id(buddy))
486 return 0;
487
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800488 if (page_is_guard(buddy) && page_order(buddy) == order) {
489 VM_BUG_ON(page_count(buddy) != 0);
490 return 1;
491 }
492
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700493 if (PageBuddy(buddy) && page_order(buddy) == order) {
Nick Piggina3af9c32009-06-16 15:32:10 -0700494 VM_BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700495 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000496 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700497 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498}
499
500/*
501 * Freeing function for a buddy system allocator.
502 *
503 * The concept of a buddy system is to maintain direct-mapped table
504 * (containing bit values) for memory blocks of various "orders".
505 * The bottom level table contains the map for the smallest allocatable
506 * units of memory (here, pages), and each level above it describes
507 * pairs of units from the levels below, hence, "buddies".
508 * At a high level, all that happens here is marking the table entry
509 * at the bottom level available, and propagating the changes upward
510 * as necessary, plus some accounting needed to play nicely with other
511 * parts of the VM system.
512 * At each level, we keep a list of pages, which are heads of continuous
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800513 * free pages of length of (1 << order) and marked with _mapcount -2. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700514 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 * So when we are allocating or freeing one, we can derive the state of the
516 * other. That is, if we allocate a small block, and both were
517 * free, the remainder of the region must be split into blocks.
518 * If a block is freed, and its buddy is also free, then this
519 * triggers coalescing into a block of larger size.
520 *
521 * -- wli
522 */
523
Nick Piggin48db57f2006-01-08 01:00:42 -0800524static inline void __free_one_page(struct page *page,
Mel Gormaned0ae212009-06-16 15:32:07 -0700525 struct zone *zone, unsigned int order,
526 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527{
528 unsigned long page_idx;
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700529 unsigned long combined_idx;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800530 unsigned long uninitialized_var(buddy_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700531 struct page *buddy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532
Nick Piggin224abf92006-01-06 00:11:11 -0800533 if (unlikely(PageCompound(page)))
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800534 if (unlikely(destroy_compound_page(page, order)))
535 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536
Mel Gormaned0ae212009-06-16 15:32:07 -0700537 VM_BUG_ON(migratetype == -1);
538
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
540
Mel Gormanf2260e62009-06-16 15:32:13 -0700541 VM_BUG_ON(page_idx & ((1 << order) - 1));
Nick Piggin725d7042006-09-25 23:30:55 -0700542 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 while (order < MAX_ORDER-1) {
KyongHo Cho43506fa2011-01-13 15:47:24 -0800545 buddy_idx = __find_buddy_index(page_idx, order);
546 buddy = page + (buddy_idx - page_idx);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700547 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700548 break;
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800549 /*
550 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
551 * merge with it and move up one order.
552 */
553 if (page_is_guard(buddy)) {
554 clear_page_guard_flag(buddy);
555 set_page_private(page, 0);
556 __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
557 } else {
558 list_del(&buddy->lru);
559 zone->free_area[order].nr_free--;
560 rmv_page_order(buddy);
561 }
KyongHo Cho43506fa2011-01-13 15:47:24 -0800562 combined_idx = buddy_idx & page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 page = page + (combined_idx - page_idx);
564 page_idx = combined_idx;
565 order++;
566 }
567 set_page_order(page, order);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700568
569 /*
570 * If this is not the largest possible page, check if the buddy
571 * of the next-highest order is free. If it is, it's possible
572 * that pages are being freed that will coalesce soon. In case,
573 * that is happening, add the free page to the tail of the list
574 * so it's less likely to be used soon and more likely to be merged
575 * as a higher order page
576 */
Mel Gormanb7f50cf2010-10-26 14:21:11 -0700577 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700578 struct page *higher_page, *higher_buddy;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800579 combined_idx = buddy_idx & page_idx;
580 higher_page = page + (combined_idx - page_idx);
581 buddy_idx = __find_buddy_index(combined_idx, order + 1);
582 higher_buddy = page + (buddy_idx - combined_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700583 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
584 list_add_tail(&page->lru,
585 &zone->free_area[order].free_list[migratetype]);
586 goto out;
587 }
588 }
589
590 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
591out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 zone->free_area[order].nr_free++;
593}
594
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700595/*
596 * free_page_mlock() -- clean up attempts to free and mlocked() page.
597 * Page should not be on lru, so no need to fix that up.
598 * free_pages_check() will verify...
599 */
600static inline void free_page_mlock(struct page *page)
601{
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700602 __dec_zone_page_state(page, NR_MLOCK);
603 __count_vm_event(UNEVICTABLE_MLOCKFREED);
604}
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700605
Nick Piggin224abf92006-01-06 00:11:11 -0800606static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607{
Nick Piggin92be2e332006-01-06 00:10:57 -0800608 if (unlikely(page_mapcount(page) |
609 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700610 (atomic_read(&page->_count) != 0) |
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -0700611 (page->flags & PAGE_FLAGS_CHECK_AT_FREE) |
612 (mem_cgroup_bad_page_check(page)))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800613 bad_page(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800614 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800615 }
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800616 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
617 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
618 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619}
620
621/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700622 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700624 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 *
626 * If the zone was previously in an "all pages pinned" state then look to
627 * see if this freeing clears that state.
628 *
629 * And clear the zone's pages_scanned counter, to hold off the "all pages are
630 * pinned" detection logic.
631 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700632static void free_pcppages_bulk(struct zone *zone, int count,
633 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634{
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700635 int migratetype = 0;
Mel Gormana6f9edd2009-09-21 17:03:20 -0700636 int batch_free = 0;
Mel Gorman72853e22010-09-09 16:38:16 -0700637 int to_free = count;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700638
Nick Pigginc54ad302006-01-06 00:10:56 -0800639 spin_lock(&zone->lock);
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -0800640 zone->all_unreclaimable = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700642
Mel Gorman72853e22010-09-09 16:38:16 -0700643 while (to_free) {
Nick Piggin48db57f2006-01-08 01:00:42 -0800644 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700645 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -0800646
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700647 /*
Mel Gormana6f9edd2009-09-21 17:03:20 -0700648 * Remove pages from lists in a round-robin fashion. A
649 * batch_free count is maintained that is incremented when an
650 * empty list is encountered. This is so more pages are freed
651 * off fuller lists instead of spinning excessively around empty
652 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700653 */
654 do {
Mel Gormana6f9edd2009-09-21 17:03:20 -0700655 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700656 if (++migratetype == MIGRATE_PCPTYPES)
657 migratetype = 0;
658 list = &pcp->lists[migratetype];
659 } while (list_empty(list));
660
Namhyung Kim1d168712011-03-22 16:32:45 -0700661 /* This is the only non-empty list. Free them all. */
662 if (batch_free == MIGRATE_PCPTYPES)
663 batch_free = to_free;
664
Mel Gormana6f9edd2009-09-21 17:03:20 -0700665 do {
666 page = list_entry(list->prev, struct page, lru);
667 /* must delete as __free_one_page list manipulates */
668 list_del(&page->lru);
Hugh Dickinsa7016232010-01-29 17:46:34 +0000669 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
670 __free_one_page(page, zone, 0, page_private(page));
671 trace_mm_page_pcpu_drain(page, 0, page_private(page));
Mel Gorman72853e22010-09-09 16:38:16 -0700672 } while (--to_free && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 }
Mel Gorman72853e22010-09-09 16:38:16 -0700674 __mod_zone_page_state(zone, NR_FREE_PAGES, count);
Nick Pigginc54ad302006-01-06 00:10:56 -0800675 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676}
677
Mel Gormaned0ae212009-06-16 15:32:07 -0700678static void free_one_page(struct zone *zone, struct page *page, int order,
679 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800680{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700681 spin_lock(&zone->lock);
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -0800682 zone->all_unreclaimable = 0;
Christoph Lameter006d22d2006-09-25 23:31:48 -0700683 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700684
Mel Gormaned0ae212009-06-16 15:32:07 -0700685 __free_one_page(page, zone, order, migratetype);
Mel Gorman72853e22010-09-09 16:38:16 -0700686 __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700687 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800688}
689
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700690static bool free_pages_prepare(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 int i;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800693 int bad = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -0800695 trace_mm_page_free(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100696 kmemcheck_free_shadow(page, order);
697
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -0800698 if (PageAnon(page))
699 page->mapping = NULL;
700 for (i = 0; i < (1 << order); i++)
701 bad += free_pages_check(page + i);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800702 if (bad)
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700703 return false;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800704
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700705 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -0700706 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700707 debug_check_no_obj_freed(page_address(page),
708 PAGE_SIZE << order);
709 }
Nick Piggindafb1362006-10-11 01:21:30 -0700710 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800711 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700712
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700713 return true;
714}
715
716static void __free_pages_ok(struct page *page, unsigned int order)
717{
718 unsigned long flags;
719 int wasMlocked = __TestClearPageMlocked(page);
720
721 if (!free_pages_prepare(page, order))
722 return;
723
Nick Pigginc54ad302006-01-06 00:10:56 -0800724 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +0200725 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -0700726 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700727 __count_vm_events(PGFREE, 1 << order);
Mel Gormaned0ae212009-06-16 15:32:07 -0700728 free_one_page(page_zone(page), page, order,
729 get_pageblock_migratetype(page));
Nick Pigginc54ad302006-01-06 00:10:56 -0800730 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731}
732
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700733void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800734{
Johannes Weinerc3993072012-01-10 15:08:10 -0800735 unsigned int nr_pages = 1 << order;
736 unsigned int loop;
David Howellsa226f6c2006-01-06 00:11:08 -0800737
Johannes Weinerc3993072012-01-10 15:08:10 -0800738 prefetchw(page);
739 for (loop = 0; loop < nr_pages; loop++) {
740 struct page *p = &page[loop];
David Howellsa226f6c2006-01-06 00:11:08 -0800741
Johannes Weinerc3993072012-01-10 15:08:10 -0800742 if (loop + 1 < nr_pages)
743 prefetchw(p + 1);
744 __ClearPageReserved(p);
745 set_page_count(p, 0);
David Howellsa226f6c2006-01-06 00:11:08 -0800746 }
Johannes Weinerc3993072012-01-10 15:08:10 -0800747
748 set_page_refcounted(page);
749 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800750}
751
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752
753/*
754 * The order of subdivision here is critical for the IO subsystem.
755 * Please do not alter this order without good reasons and regression
756 * testing. Specifically, as large blocks of memory are subdivided,
757 * the order in which smaller blocks are delivered depends on the order
758 * they're subdivided in this function. This is the primary factor
759 * influencing the order in which pages are delivered to the IO
760 * subsystem according to empirical testing, and this is also justified
761 * by considering the behavior of a buddy system containing a single
762 * large block of memory acted on by a series of small allocations.
763 * This behavior is a critical factor in sglist merging's success.
764 *
765 * -- wli
766 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800767static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700768 int low, int high, struct free_area *area,
769 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770{
771 unsigned long size = 1 << high;
772
773 while (high > low) {
774 area--;
775 high--;
776 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700777 VM_BUG_ON(bad_range(zone, &page[size]));
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800778
779#ifdef CONFIG_DEBUG_PAGEALLOC
780 if (high < debug_guardpage_minorder()) {
781 /*
782 * Mark as guard pages (or page), that will allow to
783 * merge back to allocator when buddy will be freed.
784 * Corresponding page table entries will not be touched,
785 * pages will stay not present in virtual address space
786 */
787 INIT_LIST_HEAD(&page[size].lru);
788 set_page_guard_flag(&page[size]);
789 set_page_private(&page[size], high);
790 /* Guard pages are not available for any usage */
791 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << high));
792 continue;
793 }
794#endif
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700795 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 area->nr_free++;
797 set_page_order(&page[size], high);
798 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799}
800
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801/*
802 * This page is about to be returned from the page allocator
803 */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200804static inline int check_new_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805{
Nick Piggin92be2e332006-01-06 00:10:57 -0800806 if (unlikely(page_mapcount(page) |
807 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700808 (atomic_read(&page->_count) != 0) |
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -0700809 (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
810 (mem_cgroup_bad_page_check(page)))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800811 bad_page(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800812 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800813 }
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200814 return 0;
815}
816
817static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
818{
819 int i;
820
821 for (i = 0; i < (1 << order); i++) {
822 struct page *p = page + i;
823 if (unlikely(check_new_page(p)))
824 return 1;
825 }
Hugh Dickins689bceb2005-11-21 21:32:20 -0800826
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700827 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800828 set_page_refcounted(page);
Nick Piggincc1025092006-12-06 20:32:00 -0800829
830 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800832
833 if (gfp_flags & __GFP_ZERO)
834 prep_zero_page(page, order, gfp_flags);
835
836 if (order && (gfp_flags & __GFP_COMP))
837 prep_compound_page(page, order);
838
Hugh Dickins689bceb2005-11-21 21:32:20 -0800839 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840}
841
Mel Gorman56fd56b2007-10-16 01:25:58 -0700842/*
843 * Go through the free lists for the given migratetype and remove
844 * the smallest available page from the freelists
845 */
Mel Gorman728ec982009-06-16 15:32:04 -0700846static inline
847struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -0700848 int migratetype)
849{
850 unsigned int current_order;
851 struct free_area * area;
852 struct page *page;
853
854 /* Find a page of the appropriate size in the preferred list */
855 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
856 area = &(zone->free_area[current_order]);
857 if (list_empty(&area->free_list[migratetype]))
858 continue;
859
860 page = list_entry(area->free_list[migratetype].next,
861 struct page, lru);
862 list_del(&page->lru);
863 rmv_page_order(page);
864 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -0700865 expand(zone, page, order, current_order, area, migratetype);
866 return page;
867 }
868
869 return NULL;
870}
871
872
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700873/*
874 * This array describes the order lists are fallen back to when
875 * the free lists for the desirable migrate type are depleted
876 */
877static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
Mel Gorman64c5e132007-10-16 01:25:59 -0700878 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
879 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
880 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
881 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700882};
883
Mel Gormanc361be52007-10-16 01:25:51 -0700884/*
885 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -0700886 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -0700887 * boundary. If alignment is required, use move_freepages_block()
888 */
Adrian Bunkb69a7282008-07-23 21:28:12 -0700889static int move_freepages(struct zone *zone,
890 struct page *start_page, struct page *end_page,
891 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700892{
893 struct page *page;
894 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -0700895 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -0700896
897#ifndef CONFIG_HOLES_IN_ZONE
898 /*
899 * page_zone is not safe to call in this context when
900 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
901 * anyway as we check zone boundaries in move_freepages_block().
902 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -0700903 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -0700904 */
905 BUG_ON(page_zone(start_page) != page_zone(end_page));
906#endif
907
908 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -0700909 /* Make sure we are not inadvertently changing nodes */
910 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
911
Mel Gormanc361be52007-10-16 01:25:51 -0700912 if (!pfn_valid_within(page_to_pfn(page))) {
913 page++;
914 continue;
915 }
916
917 if (!PageBuddy(page)) {
918 page++;
919 continue;
920 }
921
922 order = page_order(page);
Kirill A. Shutemov84be48d2011-03-22 16:33:41 -0700923 list_move(&page->lru,
924 &zone->free_area[order].free_list[migratetype]);
Mel Gormanc361be52007-10-16 01:25:51 -0700925 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -0700926 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -0700927 }
928
Mel Gormand1003132007-10-16 01:26:00 -0700929 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -0700930}
931
Adrian Bunkb69a7282008-07-23 21:28:12 -0700932static int move_freepages_block(struct zone *zone, struct page *page,
933 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700934{
935 unsigned long start_pfn, end_pfn;
936 struct page *start_page, *end_page;
937
938 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -0700939 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -0700940 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -0700941 end_page = start_page + pageblock_nr_pages - 1;
942 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -0700943
944 /* Do not cross zone boundaries */
945 if (start_pfn < zone->zone_start_pfn)
946 start_page = page;
947 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
948 return 0;
949
950 return move_freepages(zone, start_page, end_page, migratetype);
951}
952
Mel Gorman2f66a682009-09-21 17:02:31 -0700953static void change_pageblock_range(struct page *pageblock_page,
954 int start_order, int migratetype)
955{
956 int nr_pageblocks = 1 << (start_order - pageblock_order);
957
958 while (nr_pageblocks--) {
959 set_pageblock_migratetype(pageblock_page, migratetype);
960 pageblock_page += pageblock_nr_pages;
961 }
962}
963
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700964/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -0700965static inline struct page *
966__rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700967{
968 struct free_area * area;
969 int current_order;
970 struct page *page;
971 int migratetype, i;
972
973 /* Find the largest possible block of pages in the other list */
974 for (current_order = MAX_ORDER-1; current_order >= order;
975 --current_order) {
976 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
977 migratetype = fallbacks[start_migratetype][i];
978
Mel Gorman56fd56b2007-10-16 01:25:58 -0700979 /* MIGRATE_RESERVE handled later if necessary */
980 if (migratetype == MIGRATE_RESERVE)
981 continue;
Mel Gormane0104872007-10-16 01:25:53 -0700982
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700983 area = &(zone->free_area[current_order]);
984 if (list_empty(&area->free_list[migratetype]))
985 continue;
986
987 page = list_entry(area->free_list[migratetype].next,
988 struct page, lru);
989 area->nr_free--;
990
991 /*
Mel Gormanc361be52007-10-16 01:25:51 -0700992 * If breaking a large block of pages, move all free
Mel Gorman46dafbc2007-10-16 01:25:55 -0700993 * pages to the preferred allocation list. If falling
994 * back for a reclaimable kernel allocation, be more
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300995 * aggressive about taking ownership of free pages
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700996 */
Mel Gormand9c23402007-10-16 01:26:01 -0700997 if (unlikely(current_order >= (pageblock_order >> 1)) ||
Mel Gormandd5d2412009-09-05 11:17:11 -0700998 start_migratetype == MIGRATE_RECLAIMABLE ||
999 page_group_by_mobility_disabled) {
Mel Gorman46dafbc2007-10-16 01:25:55 -07001000 unsigned long pages;
1001 pages = move_freepages_block(zone, page,
1002 start_migratetype);
1003
1004 /* Claim the whole block if over half of it is free */
Mel Gormandd5d2412009-09-05 11:17:11 -07001005 if (pages >= (1 << (pageblock_order-1)) ||
1006 page_group_by_mobility_disabled)
Mel Gorman46dafbc2007-10-16 01:25:55 -07001007 set_pageblock_migratetype(page,
1008 start_migratetype);
1009
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001010 migratetype = start_migratetype;
Mel Gormanc361be52007-10-16 01:25:51 -07001011 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001012
1013 /* Remove the page from the freelists */
1014 list_del(&page->lru);
1015 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001016
Mel Gorman2f66a682009-09-21 17:02:31 -07001017 /* Take ownership for orders >= pageblock_order */
1018 if (current_order >= pageblock_order)
1019 change_pageblock_range(page, current_order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001020 start_migratetype);
1021
1022 expand(zone, page, order, current_order, area, migratetype);
Mel Gormane0fff1b2009-09-21 17:02:42 -07001023
1024 trace_mm_page_alloc_extfrag(page, order, current_order,
1025 start_migratetype, migratetype);
1026
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001027 return page;
1028 }
1029 }
1030
Mel Gorman728ec982009-06-16 15:32:04 -07001031 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001032}
1033
Mel Gorman56fd56b2007-10-16 01:25:58 -07001034/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 * Do the hard work of removing an element from the buddy allocator.
1036 * Call me with the zone->lock already held.
1037 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001038static struct page *__rmqueue(struct zone *zone, unsigned int order,
1039 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 struct page *page;
1042
Mel Gorman728ec982009-06-16 15:32:04 -07001043retry_reserve:
Mel Gorman56fd56b2007-10-16 01:25:58 -07001044 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045
Mel Gorman728ec982009-06-16 15:32:04 -07001046 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07001047 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001048
Mel Gorman728ec982009-06-16 15:32:04 -07001049 /*
1050 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1051 * is used because __rmqueue_smallest is an inline function
1052 * and we want just one call site
1053 */
1054 if (!page) {
1055 migratetype = MIGRATE_RESERVE;
1056 goto retry_reserve;
1057 }
1058 }
1059
Mel Gorman0d3d0622009-09-21 17:02:44 -07001060 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001061 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062}
1063
1064/*
1065 * Obtain a specified number of elements from the buddy allocator, all under
1066 * a single hold of the lock, for efficiency. Add them to the supplied list.
1067 * Returns the number of new pages which were placed at *list.
1068 */
1069static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001070 unsigned long count, struct list_head *list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001071 int migratetype, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074
Nick Pigginc54ad302006-01-06 00:10:56 -08001075 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001077 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001078 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08001080
1081 /*
1082 * Split buddy pages returned by expand() are received here
1083 * in physical page order. The page is added to the callers and
1084 * list and the list head then moves forward. From the callers
1085 * perspective, the linked list is ordered by page number in
1086 * some conditions. This is useful for IO devices that can
1087 * merge IO requests if the physical pages are ordered
1088 * properly.
1089 */
Mel Gormane084b2d2009-07-29 15:02:04 -07001090 if (likely(cold == 0))
1091 list_add(&page->lru, list);
1092 else
1093 list_add_tail(&page->lru, list);
Mel Gorman535131e62007-10-16 01:25:49 -07001094 set_page_private(page, migratetype);
Mel Gorman81eabcb2007-12-17 16:20:05 -08001095 list = &page->lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 }
Mel Gormanf2260e62009-06-16 15:32:13 -07001097 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08001098 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001099 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100}
1101
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001102#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001103/*
Christoph Lameter4037d452007-05-09 02:35:14 -07001104 * Called from the vmstat counter updater to drain pagesets of this
1105 * currently executing processor on remote nodes after they have
1106 * expired.
1107 *
Christoph Lameter879336c2006-03-22 00:09:08 -08001108 * Note that this function must be called with the thread pinned to
1109 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001110 */
Christoph Lameter4037d452007-05-09 02:35:14 -07001111void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001112{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001113 unsigned long flags;
Christoph Lameter4037d452007-05-09 02:35:14 -07001114 int to_drain;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001115
Christoph Lameter4037d452007-05-09 02:35:14 -07001116 local_irq_save(flags);
1117 if (pcp->count >= pcp->batch)
1118 to_drain = pcp->batch;
1119 else
1120 to_drain = pcp->count;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001121 free_pcppages_bulk(zone, to_drain, pcp);
Christoph Lameter4037d452007-05-09 02:35:14 -07001122 pcp->count -= to_drain;
1123 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001124}
1125#endif
1126
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001127/*
1128 * Drain pages of the indicated processor.
1129 *
1130 * The processor must either be the current processor and the
1131 * thread pinned to the current processor or a processor that
1132 * is not online.
1133 */
1134static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135{
Nick Pigginc54ad302006-01-06 00:10:56 -08001136 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07001139 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 struct per_cpu_pageset *pset;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001141 struct per_cpu_pages *pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001143 local_irq_save(flags);
1144 pset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001146 pcp = &pset->pcp;
David Rientjes2ff754f2011-01-25 15:07:23 -08001147 if (pcp->count) {
1148 free_pcppages_bulk(zone, pcp->count, pcp);
1149 pcp->count = 0;
1150 }
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001151 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 }
1153}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001155/*
1156 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1157 */
1158void drain_local_pages(void *arg)
1159{
1160 drain_pages(smp_processor_id());
1161}
1162
1163/*
1164 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
1165 */
1166void drain_all_pages(void)
1167{
Jens Axboe15c8b6c2008-05-09 09:39:44 +02001168 on_each_cpu(drain_local_pages, NULL, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001169}
1170
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02001171#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172
1173void mark_free_pages(struct zone *zone)
1174{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001175 unsigned long pfn, max_zone_pfn;
1176 unsigned long flags;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001177 int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 struct list_head *curr;
1179
1180 if (!zone->spanned_pages)
1181 return;
1182
1183 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001184
1185 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1186 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1187 if (pfn_valid(pfn)) {
1188 struct page *page = pfn_to_page(pfn);
1189
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001190 if (!swsusp_page_is_forbidden(page))
1191 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001192 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001194 for_each_migratetype_order(order, t) {
1195 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001196 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001198 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1199 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001200 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001201 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001202 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 spin_unlock_irqrestore(&zone->lock, flags);
1204}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001205#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206
1207/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 * Free a 0-order page
Li Hongfc916682010-03-05 13:41:54 -08001209 * cold == 1 ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 */
Li Hongfc916682010-03-05 13:41:54 -08001211void free_hot_cold_page(struct page *page, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212{
1213 struct zone *zone = page_zone(page);
1214 struct per_cpu_pages *pcp;
1215 unsigned long flags;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001216 int migratetype;
Johannes Weiner451ea252009-09-21 17:01:48 -07001217 int wasMlocked = __TestClearPageMlocked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001219 if (!free_pages_prepare(page, 0))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001220 return;
1221
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001222 migratetype = get_pageblock_migratetype(page);
1223 set_page_private(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +02001225 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -07001226 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001227 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001228
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001229 /*
1230 * We only track unmovable, reclaimable and movable on pcp lists.
1231 * Free ISOLATE pages back to the allocator because they are being
1232 * offlined but treat RESERVE as movable pages so we can get those
1233 * areas back if necessary. Otherwise, we may have to free
1234 * excessively into the page allocator
1235 */
1236 if (migratetype >= MIGRATE_PCPTYPES) {
1237 if (unlikely(migratetype == MIGRATE_ISOLATE)) {
1238 free_one_page(zone, page, 0, migratetype);
1239 goto out;
1240 }
1241 migratetype = MIGRATE_MOVABLE;
1242 }
1243
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001244 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001245 if (cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001246 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001247 else
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001248 list_add(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001250 if (pcp->count >= pcp->high) {
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001251 free_pcppages_bulk(zone, pcp->batch, pcp);
Nick Piggin48db57f2006-01-08 01:00:42 -08001252 pcp->count -= pcp->batch;
1253 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001254
1255out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257}
1258
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001259/*
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001260 * Free a list of 0-order pages
1261 */
1262void free_hot_cold_page_list(struct list_head *list, int cold)
1263{
1264 struct page *page, *next;
1265
1266 list_for_each_entry_safe(page, next, list, lru) {
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -08001267 trace_mm_page_free_batched(page, cold);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001268 free_hot_cold_page(page, cold);
1269 }
1270}
1271
1272/*
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001273 * split_page takes a non-compound higher-order page, and splits it into
1274 * n (1<<order) sub-pages: page[0..n]
1275 * Each sub-page must be freed individually.
1276 *
1277 * Note: this is probably too low level an operation for use in drivers.
1278 * Please consult with lkml before using this in your driver.
1279 */
1280void split_page(struct page *page, unsigned int order)
1281{
1282 int i;
1283
Nick Piggin725d7042006-09-25 23:30:55 -07001284 VM_BUG_ON(PageCompound(page));
1285 VM_BUG_ON(!page_count(page));
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001286
1287#ifdef CONFIG_KMEMCHECK
1288 /*
1289 * Split shadow pages too, because free(page[0]) would
1290 * otherwise free the whole shadow.
1291 */
1292 if (kmemcheck_page_is_tracked(page))
1293 split_page(virt_to_page(page[0].shadow), order);
1294#endif
1295
Nick Piggin7835e982006-03-22 00:08:40 -08001296 for (i = 1; i < (1 << order); i++)
1297 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001298}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001299
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300/*
Mel Gorman748446b2010-05-24 14:32:27 -07001301 * Similar to split_page except the page is already free. As this is only
1302 * being used for migration, the migratetype of the block also changes.
1303 * As this is called with interrupts disabled, the caller is responsible
1304 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1305 * are enabled.
1306 *
1307 * Note: this is probably too low level an operation for use in drivers.
1308 * Please consult with lkml before using this in your driver.
1309 */
1310int split_free_page(struct page *page)
1311{
1312 unsigned int order;
1313 unsigned long watermark;
1314 struct zone *zone;
1315
1316 BUG_ON(!PageBuddy(page));
1317
1318 zone = page_zone(page);
1319 order = page_order(page);
1320
1321 /* Obey watermarks as if the page was being allocated */
1322 watermark = low_wmark_pages(zone) + (1 << order);
1323 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1324 return 0;
1325
1326 /* Remove page from free list */
1327 list_del(&page->lru);
1328 zone->free_area[order].nr_free--;
1329 rmv_page_order(page);
1330 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1UL << order));
1331
1332 /* Split into individual pages */
1333 set_page_refcounted(page);
1334 split_page(page, order);
1335
1336 if (order >= pageblock_order - 1) {
1337 struct page *endpage = page + (1 << order) - 1;
1338 for (; page < endpage; page += pageblock_nr_pages)
1339 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1340 }
1341
1342 return 1 << order;
1343}
1344
1345/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1347 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1348 * or two.
1349 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07001350static inline
1351struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001352 struct zone *zone, int order, gfp_t gfp_flags,
1353 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354{
1355 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001356 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 int cold = !!(gfp_flags & __GFP_COLD);
1358
Hugh Dickins689bceb2005-11-21 21:32:20 -08001359again:
Nick Piggin48db57f2006-01-08 01:00:42 -08001360 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001362 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 local_irq_save(flags);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001365 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1366 list = &pcp->lists[migratetype];
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001367 if (list_empty(list)) {
Mel Gorman535131e62007-10-16 01:25:49 -07001368 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001369 pcp->batch, list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001370 migratetype, cold);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001371 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07001372 goto failed;
Mel Gorman535131e62007-10-16 01:25:49 -07001373 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001374
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001375 if (cold)
1376 page = list_entry(list->prev, struct page, lru);
1377 else
1378 page = list_entry(list->next, struct page, lru);
1379
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001380 list_del(&page->lru);
1381 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001382 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07001383 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1384 /*
1385 * __GFP_NOFAIL is not to be used in new code.
1386 *
1387 * All __GFP_NOFAIL callers should be fixed so that they
1388 * properly detect and handle allocation failures.
1389 *
1390 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07001391 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07001392 * __GFP_NOFAIL.
1393 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07001394 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07001395 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001397 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001398 spin_unlock(&zone->lock);
1399 if (!page)
1400 goto failed;
KOSAKI Motohiro6ccf80e2010-01-15 17:01:18 -08001401 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 }
1403
Christoph Lameterf8891e52006-06-30 01:55:45 -07001404 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Andi Kleen78afd562011-03-22 16:33:12 -07001405 zone_statistics(preferred_zone, zone, gfp_flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001406 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407
Nick Piggin725d7042006-09-25 23:30:55 -07001408 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -08001409 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001410 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001412
1413failed:
1414 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001415 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416}
1417
Mel Gorman41858962009-06-16 15:32:12 -07001418/* The ALLOC_WMARK bits are used as an index to zone->watermark */
1419#define ALLOC_WMARK_MIN WMARK_MIN
1420#define ALLOC_WMARK_LOW WMARK_LOW
1421#define ALLOC_WMARK_HIGH WMARK_HIGH
1422#define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1423
1424/* Mask to get the watermark bits */
1425#define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1426
Nick Piggin31488902005-11-28 13:44:03 -08001427#define ALLOC_HARDER 0x10 /* try to alloc harder */
1428#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1429#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001430
Akinobu Mita933e3122006-12-08 02:39:45 -08001431#ifdef CONFIG_FAIL_PAGE_ALLOC
1432
Akinobu Mitab2588c42011-07-26 16:09:03 -07001433static struct {
Akinobu Mita933e3122006-12-08 02:39:45 -08001434 struct fault_attr attr;
1435
1436 u32 ignore_gfp_highmem;
1437 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001438 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001439} fail_page_alloc = {
1440 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001441 .ignore_gfp_wait = 1,
1442 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001443 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001444};
1445
1446static int __init setup_fail_page_alloc(char *str)
1447{
1448 return setup_fault_attr(&fail_page_alloc.attr, str);
1449}
1450__setup("fail_page_alloc=", setup_fail_page_alloc);
1451
1452static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1453{
Akinobu Mita54114992007-07-15 23:40:23 -07001454 if (order < fail_page_alloc.min_order)
1455 return 0;
Akinobu Mita933e3122006-12-08 02:39:45 -08001456 if (gfp_mask & __GFP_NOFAIL)
1457 return 0;
1458 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1459 return 0;
1460 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1461 return 0;
1462
1463 return should_fail(&fail_page_alloc.attr, 1 << order);
1464}
1465
1466#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1467
1468static int __init fail_page_alloc_debugfs(void)
1469{
Al Virof4ae40a2011-07-24 04:33:43 -04001470 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
Akinobu Mita933e3122006-12-08 02:39:45 -08001471 struct dentry *dir;
Akinobu Mita933e3122006-12-08 02:39:45 -08001472
Akinobu Mitadd48c082011-08-03 16:21:01 -07001473 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
1474 &fail_page_alloc.attr);
1475 if (IS_ERR(dir))
1476 return PTR_ERR(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08001477
Akinobu Mitab2588c42011-07-26 16:09:03 -07001478 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
1479 &fail_page_alloc.ignore_gfp_wait))
1480 goto fail;
1481 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1482 &fail_page_alloc.ignore_gfp_highmem))
1483 goto fail;
1484 if (!debugfs_create_u32("min-order", mode, dir,
1485 &fail_page_alloc.min_order))
1486 goto fail;
Akinobu Mita933e3122006-12-08 02:39:45 -08001487
Akinobu Mitab2588c42011-07-26 16:09:03 -07001488 return 0;
1489fail:
Akinobu Mitadd48c082011-08-03 16:21:01 -07001490 debugfs_remove_recursive(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08001491
Akinobu Mitab2588c42011-07-26 16:09:03 -07001492 return -ENOMEM;
Akinobu Mita933e3122006-12-08 02:39:45 -08001493}
1494
1495late_initcall(fail_page_alloc_debugfs);
1496
1497#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1498
1499#else /* CONFIG_FAIL_PAGE_ALLOC */
1500
1501static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1502{
1503 return 0;
1504}
1505
1506#endif /* CONFIG_FAIL_PAGE_ALLOC */
1507
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508/*
Mel Gorman88f5acf2011-01-13 15:45:41 -08001509 * Return true if free pages are above 'mark'. This takes into account the order
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 * of the allocation.
1511 */
Mel Gorman88f5acf2011-01-13 15:45:41 -08001512static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1513 int classzone_idx, int alloc_flags, long free_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514{
1515 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001516 long min = mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 int o;
1518
Michal Hockodf0a6da2012-01-10 15:08:02 -08001519 free_pages -= (1 << order) - 1;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001520 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001522 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 min -= min / 4;
1524
1525 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
Mel Gorman88f5acf2011-01-13 15:45:41 -08001526 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 for (o = 0; o < order; o++) {
1528 /* At the next order, this order's pages become unavailable */
1529 free_pages -= z->free_area[o].nr_free << o;
1530
1531 /* Require fewer higher order pages to be free */
1532 min >>= 1;
1533
1534 if (free_pages <= min)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001535 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 }
Mel Gorman88f5acf2011-01-13 15:45:41 -08001537 return true;
1538}
1539
1540bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1541 int classzone_idx, int alloc_flags)
1542{
1543 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1544 zone_page_state(z, NR_FREE_PAGES));
1545}
1546
1547bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
1548 int classzone_idx, int alloc_flags)
1549{
1550 long free_pages = zone_page_state(z, NR_FREE_PAGES);
1551
1552 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1553 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1554
1555 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1556 free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557}
1558
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001559#ifdef CONFIG_NUMA
1560/*
1561 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1562 * skip over zones that are not allowed by the cpuset, or that have
1563 * been recently (in last second) found to be nearly full. See further
1564 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001565 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001566 *
1567 * If the zonelist cache is present in the passed in zonelist, then
1568 * returns a pointer to the allowed node mask (either the current
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001569 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001570 *
1571 * If the zonelist cache is not available for this zonelist, does
1572 * nothing and returns NULL.
1573 *
1574 * If the fullzones BITMAP in the zonelist cache is stale (more than
1575 * a second since last zap'd) then we zap it out (clear its bits.)
1576 *
1577 * We hold off even calling zlc_setup, until after we've checked the
1578 * first zone in the zonelist, on the theory that most allocations will
1579 * be satisfied from that first zone, so best to examine that zone as
1580 * quickly as we can.
1581 */
1582static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1583{
1584 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1585 nodemask_t *allowednodes; /* zonelist_cache approximation */
1586
1587 zlc = zonelist->zlcache_ptr;
1588 if (!zlc)
1589 return NULL;
1590
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001591 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001592 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1593 zlc->last_full_zap = jiffies;
1594 }
1595
1596 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1597 &cpuset_current_mems_allowed :
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001598 &node_states[N_HIGH_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001599 return allowednodes;
1600}
1601
1602/*
1603 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1604 * if it is worth looking at further for free memory:
1605 * 1) Check that the zone isn't thought to be full (doesn't have its
1606 * bit set in the zonelist_cache fullzones BITMAP).
1607 * 2) Check that the zones node (obtained from the zonelist_cache
1608 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1609 * Return true (non-zero) if zone is worth looking at further, or
1610 * else return false (zero) if it is not.
1611 *
1612 * This check -ignores- the distinction between various watermarks,
1613 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1614 * found to be full for any variation of these watermarks, it will
1615 * be considered full for up to one second by all requests, unless
1616 * we are so low on memory on all allowed nodes that we are forced
1617 * into the second scan of the zonelist.
1618 *
1619 * In the second scan we ignore this zonelist cache and exactly
1620 * apply the watermarks to all zones, even it is slower to do so.
1621 * We are low on memory in the second scan, and should leave no stone
1622 * unturned looking for a free page.
1623 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001624static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001625 nodemask_t *allowednodes)
1626{
1627 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1628 int i; /* index of *z in zonelist zones */
1629 int n; /* node that zone *z is on */
1630
1631 zlc = zonelist->zlcache_ptr;
1632 if (!zlc)
1633 return 1;
1634
Mel Gormandd1a2392008-04-28 02:12:17 -07001635 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001636 n = zlc->z_to_n[i];
1637
1638 /* This zone is worth trying if it is allowed but not full */
1639 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1640}
1641
1642/*
1643 * Given 'z' scanning a zonelist, set the corresponding bit in
1644 * zlc->fullzones, so that subsequent attempts to allocate a page
1645 * from that zone don't waste time re-examining it.
1646 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001647static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001648{
1649 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1650 int i; /* index of *z in zonelist zones */
1651
1652 zlc = zonelist->zlcache_ptr;
1653 if (!zlc)
1654 return;
1655
Mel Gormandd1a2392008-04-28 02:12:17 -07001656 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001657
1658 set_bit(i, zlc->fullzones);
1659}
1660
Mel Gorman76d3fbf2011-07-25 17:12:30 -07001661/*
1662 * clear all zones full, called after direct reclaim makes progress so that
1663 * a zone that was recently full is not skipped over for up to a second
1664 */
1665static void zlc_clear_zones_full(struct zonelist *zonelist)
1666{
1667 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1668
1669 zlc = zonelist->zlcache_ptr;
1670 if (!zlc)
1671 return;
1672
1673 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1674}
1675
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001676#else /* CONFIG_NUMA */
1677
1678static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1679{
1680 return NULL;
1681}
1682
Mel Gormandd1a2392008-04-28 02:12:17 -07001683static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001684 nodemask_t *allowednodes)
1685{
1686 return 1;
1687}
1688
Mel Gormandd1a2392008-04-28 02:12:17 -07001689static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001690{
1691}
Mel Gorman76d3fbf2011-07-25 17:12:30 -07001692
1693static void zlc_clear_zones_full(struct zonelist *zonelist)
1694{
1695}
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001696#endif /* CONFIG_NUMA */
1697
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001698/*
Paul Jackson0798e512006-12-06 20:31:38 -08001699 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001700 * a page.
1701 */
1702static struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001703get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
Mel Gorman5117f452009-06-16 15:31:59 -07001704 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
Mel Gorman3dd28262009-06-16 15:32:00 -07001705 struct zone *preferred_zone, int migratetype)
Martin Hicks753ee722005-06-21 17:14:41 -07001706{
Mel Gormandd1a2392008-04-28 02:12:17 -07001707 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001708 struct page *page = NULL;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001709 int classzone_idx;
Mel Gorman5117f452009-06-16 15:31:59 -07001710 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001711 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1712 int zlc_active = 0; /* set if using zonelist_cache */
1713 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gorman54a6eb52008-04-28 02:12:16 -07001714
Mel Gorman19770b32008-04-28 02:12:18 -07001715 classzone_idx = zone_idx(preferred_zone);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001716zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001717 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001718 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001719 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1720 */
Mel Gorman19770b32008-04-28 02:12:18 -07001721 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1722 high_zoneidx, nodemask) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001723 if (NUMA_BUILD && zlc_active &&
1724 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1725 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001726 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001727 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Mel Gormancd38b112011-07-25 17:12:29 -07001728 continue;
Johannes Weinera756cf52012-01-10 15:07:49 -08001729 /*
1730 * When allocating a page cache page for writing, we
1731 * want to get it from a zone that is within its dirty
1732 * limit, such that no single zone holds more than its
1733 * proportional share of globally allowed dirty pages.
1734 * The dirty limits take into account the zone's
1735 * lowmem reserves and high watermark so that kswapd
1736 * should be able to balance it without having to
1737 * write pages from its LRU list.
1738 *
1739 * This may look like it could increase pressure on
1740 * lower zones by failing allocations in higher zones
1741 * before they are full. But the pages that do spill
1742 * over are limited as the lower zones are protected
1743 * by this very same mechanism. It should not become
1744 * a practical burden to them.
1745 *
1746 * XXX: For now, allow allocations to potentially
1747 * exceed the per-zone dirty limit in the slowpath
1748 * (ALLOC_WMARK_LOW unset) before going into reclaim,
1749 * which is important when on a NUMA setup the allowed
1750 * zones are together not big enough to reach the
1751 * global limit. The proper fix for these situations
1752 * will require awareness of zones in the
1753 * dirty-throttling and the flusher threads.
1754 */
1755 if ((alloc_flags & ALLOC_WMARK_LOW) &&
1756 (gfp_mask & __GFP_WRITE) && !zone_dirty_ok(zone))
1757 goto this_zone_full;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001758
Mel Gorman41858962009-06-16 15:32:12 -07001759 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001760 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001761 unsigned long mark;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001762 int ret;
1763
Mel Gorman41858962009-06-16 15:32:12 -07001764 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
Mel Gormanfa5e0842009-06-16 15:33:22 -07001765 if (zone_watermark_ok(zone, order, mark,
1766 classzone_idx, alloc_flags))
1767 goto try_this_zone;
1768
Mel Gormancd38b112011-07-25 17:12:29 -07001769 if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) {
1770 /*
1771 * we do zlc_setup if there are multiple nodes
1772 * and before considering the first zone allowed
1773 * by the cpuset.
1774 */
1775 allowednodes = zlc_setup(zonelist, alloc_flags);
1776 zlc_active = 1;
1777 did_zlc_setup = 1;
1778 }
1779
Mel Gormanfa5e0842009-06-16 15:33:22 -07001780 if (zone_reclaim_mode == 0)
1781 goto this_zone_full;
1782
Mel Gormancd38b112011-07-25 17:12:29 -07001783 /*
1784 * As we may have just activated ZLC, check if the first
1785 * eligible zone has failed zone_reclaim recently.
1786 */
1787 if (NUMA_BUILD && zlc_active &&
1788 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1789 continue;
1790
Mel Gormanfa5e0842009-06-16 15:33:22 -07001791 ret = zone_reclaim(zone, gfp_mask, order);
1792 switch (ret) {
1793 case ZONE_RECLAIM_NOSCAN:
1794 /* did not scan */
Mel Gormancd38b112011-07-25 17:12:29 -07001795 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001796 case ZONE_RECLAIM_FULL:
1797 /* scanned but unreclaimable */
Mel Gormancd38b112011-07-25 17:12:29 -07001798 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001799 default:
1800 /* did we reclaim enough */
1801 if (!zone_watermark_ok(zone, order, mark,
1802 classzone_idx, alloc_flags))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001803 goto this_zone_full;
Paul Jackson0798e512006-12-06 20:31:38 -08001804 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001805 }
1806
Mel Gormanfa5e0842009-06-16 15:33:22 -07001807try_this_zone:
Mel Gorman3dd28262009-06-16 15:32:00 -07001808 page = buffered_rmqueue(preferred_zone, zone, order,
1809 gfp_mask, migratetype);
Paul Jackson0798e512006-12-06 20:31:38 -08001810 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001811 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001812this_zone_full:
1813 if (NUMA_BUILD)
1814 zlc_mark_zone_full(zonelist, z);
Mel Gorman54a6eb52008-04-28 02:12:16 -07001815 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001816
1817 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1818 /* Disable zlc cache for second zonelist scan */
1819 zlc_active = 0;
1820 goto zonelist_scan;
1821 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001822 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07001823}
1824
David Rientjes29423e772011-03-22 16:30:47 -07001825/*
1826 * Large machines with many possible nodes should not always dump per-node
1827 * meminfo in irq context.
1828 */
1829static inline bool should_suppress_show_mem(void)
1830{
1831 bool ret = false;
1832
1833#if NODES_SHIFT > 8
1834 ret = in_interrupt();
1835#endif
1836 return ret;
1837}
1838
Dave Hansena238ab52011-05-24 17:12:16 -07001839static DEFINE_RATELIMIT_STATE(nopage_rs,
1840 DEFAULT_RATELIMIT_INTERVAL,
1841 DEFAULT_RATELIMIT_BURST);
1842
1843void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
1844{
Dave Hansena238ab52011-05-24 17:12:16 -07001845 unsigned int filter = SHOW_MEM_FILTER_NODES;
1846
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -08001847 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
1848 debug_guardpage_minorder() > 0)
Dave Hansena238ab52011-05-24 17:12:16 -07001849 return;
1850
1851 /*
1852 * This documents exceptions given to allocations in certain
1853 * contexts that are allowed to allocate outside current's set
1854 * of allowed nodes.
1855 */
1856 if (!(gfp_mask & __GFP_NOMEMALLOC))
1857 if (test_thread_flag(TIF_MEMDIE) ||
1858 (current->flags & (PF_MEMALLOC | PF_EXITING)))
1859 filter &= ~SHOW_MEM_FILTER_NODES;
1860 if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
1861 filter &= ~SHOW_MEM_FILTER_NODES;
1862
1863 if (fmt) {
Joe Perches3ee9a4f2011-10-31 17:08:35 -07001864 struct va_format vaf;
1865 va_list args;
1866
Dave Hansena238ab52011-05-24 17:12:16 -07001867 va_start(args, fmt);
Joe Perches3ee9a4f2011-10-31 17:08:35 -07001868
1869 vaf.fmt = fmt;
1870 vaf.va = &args;
1871
1872 pr_warn("%pV", &vaf);
1873
Dave Hansena238ab52011-05-24 17:12:16 -07001874 va_end(args);
1875 }
1876
Joe Perches3ee9a4f2011-10-31 17:08:35 -07001877 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
1878 current->comm, order, gfp_mask);
Dave Hansena238ab52011-05-24 17:12:16 -07001879
1880 dump_stack();
1881 if (!should_suppress_show_mem())
1882 show_mem(filter);
1883}
1884
Mel Gorman11e33f62009-06-16 15:31:57 -07001885static inline int
1886should_alloc_retry(gfp_t gfp_mask, unsigned int order,
Mel Gormanf90ac392012-01-10 15:07:15 -08001887 unsigned long did_some_progress,
Mel Gorman11e33f62009-06-16 15:31:57 -07001888 unsigned long pages_reclaimed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889{
Mel Gorman11e33f62009-06-16 15:31:57 -07001890 /* Do not loop if specifically requested */
1891 if (gfp_mask & __GFP_NORETRY)
1892 return 0;
1893
Mel Gormanf90ac392012-01-10 15:07:15 -08001894 /* Always retry if specifically requested */
1895 if (gfp_mask & __GFP_NOFAIL)
1896 return 1;
1897
1898 /*
1899 * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim
1900 * making forward progress without invoking OOM. Suspend also disables
1901 * storage devices so kswapd will not help. Bail if we are suspending.
1902 */
1903 if (!did_some_progress && pm_suspended_storage())
1904 return 0;
1905
Mel Gorman11e33f62009-06-16 15:31:57 -07001906 /*
1907 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1908 * means __GFP_NOFAIL, but that may not be true in other
1909 * implementations.
1910 */
1911 if (order <= PAGE_ALLOC_COSTLY_ORDER)
1912 return 1;
1913
1914 /*
1915 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1916 * specified, then we retry until we no longer reclaim any pages
1917 * (above), or we've reclaimed an order of pages at least as
1918 * large as the allocation's order. In both cases, if the
1919 * allocation still fails, we stop retrying.
1920 */
1921 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
1922 return 1;
1923
Mel Gorman11e33f62009-06-16 15:31:57 -07001924 return 0;
1925}
1926
1927static inline struct page *
1928__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1929 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001930 nodemask_t *nodemask, struct zone *preferred_zone,
1931 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001932{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934
Mel Gorman11e33f62009-06-16 15:31:57 -07001935 /* Acquire the OOM killer lock for the zones in zonelist */
Minchan Kimff321fe2010-08-09 17:18:57 -07001936 if (!try_set_zonelist_oom(zonelist, gfp_mask)) {
Mel Gorman11e33f62009-06-16 15:31:57 -07001937 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 return NULL;
1939 }
Jens Axboe6b1de912005-11-17 21:35:02 +01001940
Mel Gorman11e33f62009-06-16 15:31:57 -07001941 /*
1942 * Go through the zonelist yet one more time, keep very high watermark
1943 * here, this is only to catch a parallel oom killing, we must fail if
1944 * we're still under heavy pressure.
1945 */
1946 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1947 order, zonelist, high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001948 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07001949 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001950 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07001951 goto out;
1952
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001953 if (!(gfp_mask & __GFP_NOFAIL)) {
1954 /* The OOM killer will not help higher order allocs */
1955 if (order > PAGE_ALLOC_COSTLY_ORDER)
1956 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07001957 /* The OOM killer does not needlessly kill tasks for lowmem */
1958 if (high_zoneidx < ZONE_NORMAL)
1959 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001960 /*
1961 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
1962 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
1963 * The caller should handle page allocation failure by itself if
1964 * it specifies __GFP_THISNODE.
1965 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
1966 */
1967 if (gfp_mask & __GFP_THISNODE)
1968 goto out;
1969 }
Mel Gorman11e33f62009-06-16 15:31:57 -07001970 /* Exhausted what can be done so it's blamo time */
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001971 out_of_memory(zonelist, gfp_mask, order, nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07001972
1973out:
1974 clear_zonelist_oom(zonelist, gfp_mask);
1975 return page;
1976}
1977
Mel Gorman56de7262010-05-24 14:32:30 -07001978#ifdef CONFIG_COMPACTION
1979/* Try memory compaction for high-order allocations before reclaim */
1980static struct page *
1981__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
1982 struct zonelist *zonelist, enum zone_type high_zoneidx,
1983 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman66199712012-01-12 17:19:41 -08001984 int migratetype, bool sync_migration,
1985 bool *deferred_compaction,
1986 unsigned long *did_some_progress)
Mel Gorman56de7262010-05-24 14:32:30 -07001987{
1988 struct page *page;
1989
Mel Gorman66199712012-01-12 17:19:41 -08001990 if (!order)
Mel Gorman56de7262010-05-24 14:32:30 -07001991 return NULL;
1992
Mel Gorman66199712012-01-12 17:19:41 -08001993 if (compaction_deferred(preferred_zone)) {
1994 *deferred_compaction = true;
1995 return NULL;
1996 }
1997
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001998 current->flags |= PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07001999 *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
Mel Gorman77f1fe62011-01-13 15:45:57 -08002000 nodemask, sync_migration);
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002001 current->flags &= ~PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07002002 if (*did_some_progress != COMPACT_SKIPPED) {
2003
2004 /* Page migration frees to the PCP lists but we want merging */
2005 drain_pages(get_cpu());
2006 put_cpu();
2007
2008 page = get_page_from_freelist(gfp_mask, nodemask,
2009 order, zonelist, high_zoneidx,
2010 alloc_flags, preferred_zone,
2011 migratetype);
2012 if (page) {
Mel Gorman4f92e252010-05-24 14:32:32 -07002013 preferred_zone->compact_considered = 0;
2014 preferred_zone->compact_defer_shift = 0;
Mel Gorman56de7262010-05-24 14:32:30 -07002015 count_vm_event(COMPACTSUCCESS);
2016 return page;
2017 }
2018
2019 /*
2020 * It's bad if compaction run occurs and fails.
2021 * The most likely reason is that pages exist,
2022 * but not enough to satisfy watermarks.
2023 */
2024 count_vm_event(COMPACTFAIL);
Mel Gorman66199712012-01-12 17:19:41 -08002025
2026 /*
2027 * As async compaction considers a subset of pageblocks, only
2028 * defer if the failure was a sync compaction failure.
2029 */
2030 if (sync_migration)
2031 defer_compaction(preferred_zone);
Mel Gorman56de7262010-05-24 14:32:30 -07002032
2033 cond_resched();
2034 }
2035
2036 return NULL;
2037}
2038#else
2039static inline struct page *
2040__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2041 struct zonelist *zonelist, enum zone_type high_zoneidx,
2042 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman66199712012-01-12 17:19:41 -08002043 int migratetype, bool sync_migration,
2044 bool *deferred_compaction,
2045 unsigned long *did_some_progress)
Mel Gorman56de7262010-05-24 14:32:30 -07002046{
2047 return NULL;
2048}
2049#endif /* CONFIG_COMPACTION */
2050
Mel Gorman11e33f62009-06-16 15:31:57 -07002051/* The really slow allocator path where we enter direct reclaim */
2052static inline struct page *
2053__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
2054 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07002055 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07002056 int migratetype, unsigned long *did_some_progress)
Mel Gorman11e33f62009-06-16 15:31:57 -07002057{
2058 struct page *page = NULL;
2059 struct reclaim_state reclaim_state;
Mel Gorman9ee493c2010-09-09 16:38:18 -07002060 bool drained = false;
Mel Gorman11e33f62009-06-16 15:31:57 -07002061
2062 cond_resched();
2063
2064 /* We now go into synchronous reclaim */
2065 cpuset_memory_pressure_bump();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002066 current->flags |= PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002067 lockdep_set_current_reclaim_state(gfp_mask);
2068 reclaim_state.reclaimed_slab = 0;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002069 current->reclaim_state = &reclaim_state;
Mel Gorman11e33f62009-06-16 15:31:57 -07002070
2071 *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
2072
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002073 current->reclaim_state = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002074 lockdep_clear_current_reclaim_state();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002075 current->flags &= ~PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002076
2077 cond_resched();
2078
Mel Gorman9ee493c2010-09-09 16:38:18 -07002079 if (unlikely(!(*did_some_progress)))
2080 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002081
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002082 /* After successful reclaim, reconsider all zones for allocation */
2083 if (NUMA_BUILD)
2084 zlc_clear_zones_full(zonelist);
2085
Mel Gorman9ee493c2010-09-09 16:38:18 -07002086retry:
2087 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002088 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002089 alloc_flags, preferred_zone,
2090 migratetype);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002091
2092 /*
2093 * If an allocation failed after direct reclaim, it could be because
2094 * pages are pinned on the per-cpu lists. Drain them and try again
2095 */
2096 if (!page && !drained) {
2097 drain_all_pages();
2098 drained = true;
2099 goto retry;
2100 }
2101
Mel Gorman11e33f62009-06-16 15:31:57 -07002102 return page;
2103}
2104
Mel Gorman11e33f62009-06-16 15:31:57 -07002105/*
2106 * This is called in the allocator slow-path if the allocation request is of
2107 * sufficient urgency to ignore watermarks and take other desperate measures
2108 */
2109static inline struct page *
2110__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
2111 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002112 nodemask_t *nodemask, struct zone *preferred_zone,
2113 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002114{
2115 struct page *page;
2116
2117 do {
2118 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002119 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
Mel Gorman3dd28262009-06-16 15:32:00 -07002120 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002121
2122 if (!page && gfp_mask & __GFP_NOFAIL)
Mel Gorman0e093d992010-10-26 14:21:45 -07002123 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07002124 } while (!page && (gfp_mask & __GFP_NOFAIL));
2125
2126 return page;
2127}
2128
2129static inline
2130void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
Mel Gorman99504742011-01-13 15:46:20 -08002131 enum zone_type high_zoneidx,
2132 enum zone_type classzone_idx)
Mel Gorman11e33f62009-06-16 15:31:57 -07002133{
2134 struct zoneref *z;
2135 struct zone *zone;
2136
2137 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
Mel Gorman99504742011-01-13 15:46:20 -08002138 wakeup_kswapd(zone, order, classzone_idx);
Mel Gorman11e33f62009-06-16 15:31:57 -07002139}
2140
Peter Zijlstra341ce062009-06-16 15:32:02 -07002141static inline int
2142gfp_to_alloc_flags(gfp_t gfp_mask)
2143{
Peter Zijlstra341ce062009-06-16 15:32:02 -07002144 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
2145 const gfp_t wait = gfp_mask & __GFP_WAIT;
2146
Mel Gormana56f57f2009-06-16 15:32:02 -07002147 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07002148 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07002149
Peter Zijlstra341ce062009-06-16 15:32:02 -07002150 /*
2151 * The caller may dip into page reserves a bit more if the caller
2152 * cannot run direct reclaim, or if the caller has realtime scheduling
2153 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2154 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
2155 */
Namhyung Kime6223a32010-10-26 14:21:59 -07002156 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07002157
2158 if (!wait) {
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002159 /*
2160 * Not worth trying to allocate harder for
2161 * __GFP_NOMEMALLOC even if it can't schedule.
2162 */
2163 if (!(gfp_mask & __GFP_NOMEMALLOC))
2164 alloc_flags |= ALLOC_HARDER;
Peter Zijlstra341ce062009-06-16 15:32:02 -07002165 /*
2166 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
2167 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
2168 */
2169 alloc_flags &= ~ALLOC_CPUSET;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002170 } else if (unlikely(rt_task(current)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07002171 alloc_flags |= ALLOC_HARDER;
2172
2173 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2174 if (!in_interrupt() &&
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002175 ((current->flags & PF_MEMALLOC) ||
Peter Zijlstra341ce062009-06-16 15:32:02 -07002176 unlikely(test_thread_flag(TIF_MEMDIE))))
2177 alloc_flags |= ALLOC_NO_WATERMARKS;
2178 }
2179
2180 return alloc_flags;
2181}
2182
Mel Gorman11e33f62009-06-16 15:31:57 -07002183static inline struct page *
2184__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
2185 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002186 nodemask_t *nodemask, struct zone *preferred_zone,
2187 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002188{
2189 const gfp_t wait = gfp_mask & __GFP_WAIT;
2190 struct page *page = NULL;
2191 int alloc_flags;
2192 unsigned long pages_reclaimed = 0;
2193 unsigned long did_some_progress;
Mel Gorman77f1fe62011-01-13 15:45:57 -08002194 bool sync_migration = false;
Mel Gorman66199712012-01-12 17:19:41 -08002195 bool deferred_compaction = false;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002196
Christoph Lameter952f3b52006-12-06 20:33:26 -08002197 /*
Mel Gorman72807a72009-06-16 15:32:18 -07002198 * In the slowpath, we sanity check order to avoid ever trying to
2199 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2200 * be using allocators in order of preference for an area that is
2201 * too large.
2202 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07002203 if (order >= MAX_ORDER) {
2204 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07002205 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07002206 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002207
Christoph Lameter952f3b52006-12-06 20:33:26 -08002208 /*
2209 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2210 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2211 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2212 * using a larger set of nodes after it has established that the
2213 * allowed per node queues are empty and that nodes are
2214 * over allocated.
2215 */
2216 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
2217 goto nopage;
2218
Mel Gormancc4a6852009-11-11 14:26:14 -08002219restart:
Andrea Arcangeli32dba982011-01-13 15:46:49 -08002220 if (!(gfp_mask & __GFP_NO_KSWAPD))
2221 wake_all_kswapd(order, zonelist, high_zoneidx,
Mel Gorman99504742011-01-13 15:46:20 -08002222 zone_idx(preferred_zone));
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002223
Paul Jackson9bf22292005-09-06 15:18:12 -07002224 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002225 * OK, we're below the kswapd watermark and have kicked background
2226 * reclaim. Now things get more complex, so set up alloc_flags according
2227 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07002228 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002229 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230
David Rientjesf33261d2011-01-25 15:07:20 -08002231 /*
2232 * Find the true preferred zone if the allocation is unconstrained by
2233 * cpusets.
2234 */
2235 if (!(alloc_flags & ALLOC_CPUSET) && !nodemask)
2236 first_zones_zonelist(zonelist, high_zoneidx, NULL,
2237 &preferred_zone);
2238
Andrew Barrycfa54a02011-05-24 17:12:52 -07002239rebalance:
Peter Zijlstra341ce062009-06-16 15:32:02 -07002240 /* This is the last chance, in general, before the goto nopage. */
Mel Gorman19770b32008-04-28 02:12:18 -07002241 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
Peter Zijlstra341ce062009-06-16 15:32:02 -07002242 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
2243 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002244 if (page)
2245 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246
Mel Gorman11e33f62009-06-16 15:31:57 -07002247 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002248 if (alloc_flags & ALLOC_NO_WATERMARKS) {
2249 page = __alloc_pages_high_priority(gfp_mask, order,
2250 zonelist, high_zoneidx, nodemask,
2251 preferred_zone, migratetype);
2252 if (page)
2253 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254 }
2255
2256 /* Atomic allocations - we can't balance anything */
2257 if (!wait)
2258 goto nopage;
2259
Peter Zijlstra341ce062009-06-16 15:32:02 -07002260 /* Avoid recursion of direct reclaim */
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002261 if (current->flags & PF_MEMALLOC)
Peter Zijlstra341ce062009-06-16 15:32:02 -07002262 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263
David Rientjes6583bb62009-07-29 15:02:06 -07002264 /* Avoid allocations with no watermarks from looping endlessly */
2265 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2266 goto nopage;
2267
Mel Gorman77f1fe62011-01-13 15:45:57 -08002268 /*
2269 * Try direct compaction. The first pass is asynchronous. Subsequent
2270 * attempts after direct reclaim are synchronous
2271 */
Mel Gorman56de7262010-05-24 14:32:30 -07002272 page = __alloc_pages_direct_compact(gfp_mask, order,
2273 zonelist, high_zoneidx,
2274 nodemask,
2275 alloc_flags, preferred_zone,
Mel Gorman66199712012-01-12 17:19:41 -08002276 migratetype, sync_migration,
2277 &deferred_compaction,
2278 &did_some_progress);
Mel Gorman56de7262010-05-24 14:32:30 -07002279 if (page)
2280 goto got_pg;
Andrea Arcangelic6a140b2011-05-24 17:11:38 -07002281 sync_migration = true;
Mel Gorman56de7262010-05-24 14:32:30 -07002282
Mel Gorman66199712012-01-12 17:19:41 -08002283 /*
2284 * If compaction is deferred for high-order allocations, it is because
2285 * sync compaction recently failed. In this is the case and the caller
2286 * has requested the system not be heavily disrupted, fail the
2287 * allocation now instead of entering direct reclaim
2288 */
2289 if (deferred_compaction && (gfp_mask & __GFP_NO_KSWAPD))
2290 goto nopage;
2291
Mel Gorman11e33f62009-06-16 15:31:57 -07002292 /* Try direct reclaim and then allocating */
2293 page = __alloc_pages_direct_reclaim(gfp_mask, order,
2294 zonelist, high_zoneidx,
2295 nodemask,
Mel Gorman5117f452009-06-16 15:31:59 -07002296 alloc_flags, preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07002297 migratetype, &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07002298 if (page)
2299 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301 /*
Mel Gorman11e33f62009-06-16 15:31:57 -07002302 * If we failed to make any progress reclaiming, then we are
2303 * running out of options and have to consider going OOM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 */
Mel Gorman11e33f62009-06-16 15:31:57 -07002305 if (!did_some_progress) {
2306 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -07002307 if (oom_killer_disabled)
2308 goto nopage;
Mel Gorman11e33f62009-06-16 15:31:57 -07002309 page = __alloc_pages_may_oom(gfp_mask, order,
2310 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002311 nodemask, preferred_zone,
2312 migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002313 if (page)
2314 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315
David Rientjes03668b32010-08-09 17:18:54 -07002316 if (!(gfp_mask & __GFP_NOFAIL)) {
2317 /*
2318 * The oom killer is not called for high-order
2319 * allocations that may fail, so if no progress
2320 * is being made, there are no other options and
2321 * retrying is unlikely to help.
2322 */
2323 if (order > PAGE_ALLOC_COSTLY_ORDER)
2324 goto nopage;
2325 /*
2326 * The oom killer is not called for lowmem
2327 * allocations to prevent needlessly killing
2328 * innocent tasks.
2329 */
2330 if (high_zoneidx < ZONE_NORMAL)
2331 goto nopage;
2332 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334 goto restart;
2335 }
2336 }
2337
Mel Gorman11e33f62009-06-16 15:31:57 -07002338 /* Check if we should retry the allocation */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07002339 pages_reclaimed += did_some_progress;
Mel Gormanf90ac392012-01-10 15:07:15 -08002340 if (should_alloc_retry(gfp_mask, order, did_some_progress,
2341 pages_reclaimed)) {
Mel Gorman11e33f62009-06-16 15:31:57 -07002342 /* Wait for some write requests to complete then retry */
Mel Gorman0e093d992010-10-26 14:21:45 -07002343 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344 goto rebalance;
Mel Gorman3e7d3442011-01-13 15:45:56 -08002345 } else {
2346 /*
2347 * High-order allocations do not necessarily loop after
2348 * direct reclaim and reclaim/compaction depends on compaction
2349 * being called after reclaim so call directly if necessary
2350 */
2351 page = __alloc_pages_direct_compact(gfp_mask, order,
2352 zonelist, high_zoneidx,
2353 nodemask,
2354 alloc_flags, preferred_zone,
Mel Gorman66199712012-01-12 17:19:41 -08002355 migratetype, sync_migration,
2356 &deferred_compaction,
2357 &did_some_progress);
Mel Gorman3e7d3442011-01-13 15:45:56 -08002358 if (page)
2359 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360 }
2361
2362nopage:
Dave Hansena238ab52011-05-24 17:12:16 -07002363 warn_alloc_failed(gfp_mask, order, NULL);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002364 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365got_pg:
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002366 if (kmemcheck_enabled)
2367 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368 return page;
Mel Gorman11e33f62009-06-16 15:31:57 -07002369
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370}
Mel Gorman11e33f62009-06-16 15:31:57 -07002371
2372/*
2373 * This is the 'heart' of the zoned buddy allocator.
2374 */
2375struct page *
2376__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2377 struct zonelist *zonelist, nodemask_t *nodemask)
2378{
2379 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Mel Gorman5117f452009-06-16 15:31:59 -07002380 struct zone *preferred_zone;
Mel Gorman11e33f62009-06-16 15:31:57 -07002381 struct page *page;
Mel Gorman3dd28262009-06-16 15:32:00 -07002382 int migratetype = allocflags_to_migratetype(gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002383
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10002384 gfp_mask &= gfp_allowed_mask;
2385
Mel Gorman11e33f62009-06-16 15:31:57 -07002386 lockdep_trace_alloc(gfp_mask);
2387
2388 might_sleep_if(gfp_mask & __GFP_WAIT);
2389
2390 if (should_fail_alloc_page(gfp_mask, order))
2391 return NULL;
2392
2393 /*
2394 * Check the zones suitable for the gfp_mask contain at least one
2395 * valid zone. It's possible to have an empty zonelist as a result
2396 * of GFP_THISNODE and a memoryless node
2397 */
2398 if (unlikely(!zonelist->_zonerefs->zone))
2399 return NULL;
2400
Miao Xiec0ff7452010-05-24 14:32:08 -07002401 get_mems_allowed();
Mel Gorman5117f452009-06-16 15:31:59 -07002402 /* The preferred zone is used for statistics later */
David Rientjesf33261d2011-01-25 15:07:20 -08002403 first_zones_zonelist(zonelist, high_zoneidx,
2404 nodemask ? : &cpuset_current_mems_allowed,
2405 &preferred_zone);
Miao Xiec0ff7452010-05-24 14:32:08 -07002406 if (!preferred_zone) {
2407 put_mems_allowed();
Mel Gorman5117f452009-06-16 15:31:59 -07002408 return NULL;
Miao Xiec0ff7452010-05-24 14:32:08 -07002409 }
Mel Gorman5117f452009-06-16 15:31:59 -07002410
2411 /* First allocation attempt */
Mel Gorman11e33f62009-06-16 15:31:57 -07002412 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002413 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07002414 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002415 if (unlikely(!page))
2416 page = __alloc_pages_slowpath(gfp_mask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002417 zonelist, high_zoneidx, nodemask,
Mel Gorman3dd28262009-06-16 15:32:00 -07002418 preferred_zone, migratetype);
Miao Xiec0ff7452010-05-24 14:32:08 -07002419 put_mems_allowed();
Mel Gorman11e33f62009-06-16 15:31:57 -07002420
Mel Gorman4b4f2782009-09-21 17:02:41 -07002421 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002422 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423}
Mel Gormand2391712009-06-16 15:31:52 -07002424EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425
2426/*
2427 * Common helper functions.
2428 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08002429unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430{
Akinobu Mita945a1112009-09-21 17:01:47 -07002431 struct page *page;
2432
2433 /*
2434 * __get_free_pages() returns a 32-bit address, which cannot represent
2435 * a highmem page
2436 */
2437 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2438
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439 page = alloc_pages(gfp_mask, order);
2440 if (!page)
2441 return 0;
2442 return (unsigned long) page_address(page);
2443}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444EXPORT_SYMBOL(__get_free_pages);
2445
Harvey Harrison920c7a52008-02-04 22:29:26 -08002446unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447{
Akinobu Mita945a1112009-09-21 17:01:47 -07002448 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450EXPORT_SYMBOL(get_zeroed_page);
2451
Harvey Harrison920c7a52008-02-04 22:29:26 -08002452void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453{
Nick Pigginb5810032005-10-29 18:16:12 -07002454 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455 if (order == 0)
Li Hongfc916682010-03-05 13:41:54 -08002456 free_hot_cold_page(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 else
2458 __free_pages_ok(page, order);
2459 }
2460}
2461
2462EXPORT_SYMBOL(__free_pages);
2463
Harvey Harrison920c7a52008-02-04 22:29:26 -08002464void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465{
2466 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07002467 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468 __free_pages(virt_to_page((void *)addr), order);
2469 }
2470}
2471
2472EXPORT_SYMBOL(free_pages);
2473
Andi Kleenee85c2e2011-05-11 15:13:34 -07002474static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
2475{
2476 if (addr) {
2477 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2478 unsigned long used = addr + PAGE_ALIGN(size);
2479
2480 split_page(virt_to_page((void *)addr), order);
2481 while (used < alloc_end) {
2482 free_page(used);
2483 used += PAGE_SIZE;
2484 }
2485 }
2486 return (void *)addr;
2487}
2488
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002489/**
2490 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2491 * @size: the number of bytes to allocate
2492 * @gfp_mask: GFP flags for the allocation
2493 *
2494 * This function is similar to alloc_pages(), except that it allocates the
2495 * minimum number of pages to satisfy the request. alloc_pages() can only
2496 * allocate memory in power-of-two pages.
2497 *
2498 * This function is also limited by MAX_ORDER.
2499 *
2500 * Memory allocated by this function must be released by free_pages_exact().
2501 */
2502void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2503{
2504 unsigned int order = get_order(size);
2505 unsigned long addr;
2506
2507 addr = __get_free_pages(gfp_mask, order);
Andi Kleenee85c2e2011-05-11 15:13:34 -07002508 return make_alloc_exact(addr, order, size);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002509}
2510EXPORT_SYMBOL(alloc_pages_exact);
2511
2512/**
Andi Kleenee85c2e2011-05-11 15:13:34 -07002513 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2514 * pages on a node.
Randy Dunlapb5e6ab52011-05-16 13:16:54 -07002515 * @nid: the preferred node ID where memory should be allocated
Andi Kleenee85c2e2011-05-11 15:13:34 -07002516 * @size: the number of bytes to allocate
2517 * @gfp_mask: GFP flags for the allocation
2518 *
2519 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2520 * back.
2521 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2522 * but is not exact.
2523 */
2524void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
2525{
2526 unsigned order = get_order(size);
2527 struct page *p = alloc_pages_node(nid, gfp_mask, order);
2528 if (!p)
2529 return NULL;
2530 return make_alloc_exact((unsigned long)page_address(p), order, size);
2531}
2532EXPORT_SYMBOL(alloc_pages_exact_nid);
2533
2534/**
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002535 * free_pages_exact - release memory allocated via alloc_pages_exact()
2536 * @virt: the value returned by alloc_pages_exact.
2537 * @size: size of allocation, same value as passed to alloc_pages_exact().
2538 *
2539 * Release the memory allocated by a previous call to alloc_pages_exact.
2540 */
2541void free_pages_exact(void *virt, size_t size)
2542{
2543 unsigned long addr = (unsigned long)virt;
2544 unsigned long end = addr + PAGE_ALIGN(size);
2545
2546 while (addr < end) {
2547 free_page(addr);
2548 addr += PAGE_SIZE;
2549 }
2550}
2551EXPORT_SYMBOL(free_pages_exact);
2552
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553static unsigned int nr_free_zone_pages(int offset)
2554{
Mel Gormandd1a2392008-04-28 02:12:17 -07002555 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002556 struct zone *zone;
2557
Martin J. Blighe310fd42005-07-29 22:59:18 -07002558 /* Just pick one node, since fallback list is circular */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 unsigned int sum = 0;
2560
Mel Gorman0e884602008-04-28 02:12:14 -07002561 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562
Mel Gorman54a6eb52008-04-28 02:12:16 -07002563 for_each_zone_zonelist(zone, z, zonelist, offset) {
Martin J. Blighe310fd42005-07-29 22:59:18 -07002564 unsigned long size = zone->present_pages;
Mel Gorman41858962009-06-16 15:32:12 -07002565 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07002566 if (size > high)
2567 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 }
2569
2570 return sum;
2571}
2572
2573/*
2574 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2575 */
2576unsigned int nr_free_buffer_pages(void)
2577{
Al Viroaf4ca452005-10-21 02:55:38 -04002578 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579}
Meelap Shahc2f1a552007-07-17 04:04:39 -07002580EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581
2582/*
2583 * Amount of free RAM allocatable within all zones
2584 */
2585unsigned int nr_free_pagecache_pages(void)
2586{
Mel Gorman2a1e2742007-07-17 04:03:12 -07002587 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002589
2590static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591{
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002592 if (NUMA_BUILD)
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08002593 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596void si_meminfo(struct sysinfo *val)
2597{
2598 val->totalram = totalram_pages;
2599 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002600 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602 val->totalhigh = totalhigh_pages;
2603 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604 val->mem_unit = PAGE_SIZE;
2605}
2606
2607EXPORT_SYMBOL(si_meminfo);
2608
2609#ifdef CONFIG_NUMA
2610void si_meminfo_node(struct sysinfo *val, int nid)
2611{
2612 pg_data_t *pgdat = NODE_DATA(nid);
2613
2614 val->totalram = pgdat->node_present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002615 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002616#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002618 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2619 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002620#else
2621 val->totalhigh = 0;
2622 val->freehigh = 0;
2623#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624 val->mem_unit = PAGE_SIZE;
2625}
2626#endif
2627
David Rientjesddd588b2011-03-22 16:30:46 -07002628/*
David Rientjes7bf02ea2011-05-24 17:11:16 -07002629 * Determine whether the node should be displayed or not, depending on whether
2630 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
David Rientjesddd588b2011-03-22 16:30:46 -07002631 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07002632bool skip_free_areas_node(unsigned int flags, int nid)
David Rientjesddd588b2011-03-22 16:30:46 -07002633{
2634 bool ret = false;
2635
2636 if (!(flags & SHOW_MEM_FILTER_NODES))
2637 goto out;
2638
2639 get_mems_allowed();
David Rientjes7bf02ea2011-05-24 17:11:16 -07002640 ret = !node_isset(nid, cpuset_current_mems_allowed);
David Rientjesddd588b2011-03-22 16:30:46 -07002641 put_mems_allowed();
2642out:
2643 return ret;
2644}
2645
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646#define K(x) ((x) << (PAGE_SHIFT-10))
2647
2648/*
2649 * Show free area list (used inside shift_scroll-lock stuff)
2650 * We also calculate the percentage fragmentation. We do this by counting the
2651 * memory on each free list with the exception of the first item on the list.
David Rientjesddd588b2011-03-22 16:30:46 -07002652 * Suppresses nodes that are not allowed by current's cpuset if
2653 * SHOW_MEM_FILTER_NODES is passed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07002655void show_free_areas(unsigned int filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656{
Jes Sorensenc7241912006-09-27 01:50:05 -07002657 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658 struct zone *zone;
2659
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002660 for_each_populated_zone(zone) {
David Rientjes7bf02ea2011-05-24 17:11:16 -07002661 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07002662 continue;
Jes Sorensenc7241912006-09-27 01:50:05 -07002663 show_node(zone);
2664 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665
Dave Jones6b482c62005-11-10 15:45:56 -05002666 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667 struct per_cpu_pageset *pageset;
2668
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002669 pageset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002671 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2672 cpu, pageset->pcp.high,
2673 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674 }
2675 }
2676
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002677 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2678 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002679 " unevictable:%lu"
Andrew Mortonb76146e2009-10-26 16:49:52 -07002680 " dirty:%lu writeback:%lu unstable:%lu\n"
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07002681 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002682 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002683 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002684 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002685 global_page_state(NR_ISOLATED_ANON),
2686 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002687 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002688 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002689 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07002690 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07002691 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07002692 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08002693 global_page_state(NR_FREE_PAGES),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07002694 global_page_state(NR_SLAB_RECLAIMABLE),
2695 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07002696 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002697 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08002698 global_page_state(NR_PAGETABLE),
2699 global_page_state(NR_BOUNCE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002701 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702 int i;
2703
David Rientjes7bf02ea2011-05-24 17:11:16 -07002704 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07002705 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706 show_node(zone);
2707 printk("%s"
2708 " free:%lukB"
2709 " min:%lukB"
2710 " low:%lukB"
2711 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002712 " active_anon:%lukB"
2713 " inactive_anon:%lukB"
2714 " active_file:%lukB"
2715 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002716 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002717 " isolated(anon):%lukB"
2718 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719 " present:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002720 " mlocked:%lukB"
2721 " dirty:%lukB"
2722 " writeback:%lukB"
2723 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002724 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002725 " slab_reclaimable:%lukB"
2726 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07002727 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002728 " pagetables:%lukB"
2729 " unstable:%lukB"
2730 " bounce:%lukB"
2731 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732 " pages_scanned:%lu"
2733 " all_unreclaimable? %s"
2734 "\n",
2735 zone->name,
Mel Gorman88f5acf2011-01-13 15:45:41 -08002736 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07002737 K(min_wmark_pages(zone)),
2738 K(low_wmark_pages(zone)),
2739 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002740 K(zone_page_state(zone, NR_ACTIVE_ANON)),
2741 K(zone_page_state(zone, NR_INACTIVE_ANON)),
2742 K(zone_page_state(zone, NR_ACTIVE_FILE)),
2743 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002744 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002745 K(zone_page_state(zone, NR_ISOLATED_ANON)),
2746 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747 K(zone->present_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002748 K(zone_page_state(zone, NR_MLOCK)),
2749 K(zone_page_state(zone, NR_FILE_DIRTY)),
2750 K(zone_page_state(zone, NR_WRITEBACK)),
2751 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002752 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002753 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
2754 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07002755 zone_page_state(zone, NR_KERNEL_STACK) *
2756 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002757 K(zone_page_state(zone, NR_PAGETABLE)),
2758 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
2759 K(zone_page_state(zone, NR_BOUNCE)),
2760 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761 zone->pages_scanned,
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -08002762 (zone->all_unreclaimable ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763 );
2764 printk("lowmem_reserve[]:");
2765 for (i = 0; i < MAX_NR_ZONES; i++)
2766 printk(" %lu", zone->lowmem_reserve[i]);
2767 printk("\n");
2768 }
2769
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002770 for_each_populated_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002771 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002772
David Rientjes7bf02ea2011-05-24 17:11:16 -07002773 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07002774 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002775 show_node(zone);
2776 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777
2778 spin_lock_irqsave(&zone->lock, flags);
2779 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002780 nr[order] = zone->free_area[order].nr_free;
2781 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782 }
2783 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002784 for (order = 0; order < MAX_ORDER; order++)
2785 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786 printk("= %lukB\n", K(total));
2787 }
2788
Larry Woodmane6f36022008-02-04 22:29:30 -08002789 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2790
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791 show_swap_cache_info();
2792}
2793
Mel Gorman19770b32008-04-28 02:12:18 -07002794static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2795{
2796 zoneref->zone = zone;
2797 zoneref->zone_idx = zone_idx(zone);
2798}
2799
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800/*
2801 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08002802 *
2803 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002805static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2806 int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807{
Christoph Lameter1a932052006-01-06 00:11:16 -08002808 struct zone *zone;
2809
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002810 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002811 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08002812
2813 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002814 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08002815 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08002816 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002817 zoneref_set_zone(zone,
2818 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08002819 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08002821
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002822 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08002823 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824}
2825
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002826
2827/*
2828 * zonelist_order:
2829 * 0 = automatic detection of better ordering.
2830 * 1 = order by ([node] distance, -zonetype)
2831 * 2 = order by (-zonetype, [node] distance)
2832 *
2833 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2834 * the same zonelist. So only NUMA can configure this param.
2835 */
2836#define ZONELIST_ORDER_DEFAULT 0
2837#define ZONELIST_ORDER_NODE 1
2838#define ZONELIST_ORDER_ZONE 2
2839
2840/* zonelist order in the kernel.
2841 * set_zonelist_order() will set this to NODE or ZONE.
2842 */
2843static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2844static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2845
2846
Linus Torvalds1da177e2005-04-16 15:20:36 -07002847#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002848/* The value user specified ....changed by config */
2849static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2850/* string for sysctl */
2851#define NUMA_ZONELIST_ORDER_LEN 16
2852char numa_zonelist_order[16] = "default";
2853
2854/*
2855 * interface for configure zonelist ordering.
2856 * command line option "numa_zonelist_order"
2857 * = "[dD]efault - default, automatic configuration.
2858 * = "[nN]ode - order by node locality, then by zone within node
2859 * = "[zZ]one - order by zone, then by locality within zone
2860 */
2861
2862static int __parse_numa_zonelist_order(char *s)
2863{
2864 if (*s == 'd' || *s == 'D') {
2865 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2866 } else if (*s == 'n' || *s == 'N') {
2867 user_zonelist_order = ZONELIST_ORDER_NODE;
2868 } else if (*s == 'z' || *s == 'Z') {
2869 user_zonelist_order = ZONELIST_ORDER_ZONE;
2870 } else {
2871 printk(KERN_WARNING
2872 "Ignoring invalid numa_zonelist_order value: "
2873 "%s\n", s);
2874 return -EINVAL;
2875 }
2876 return 0;
2877}
2878
2879static __init int setup_numa_zonelist_order(char *s)
2880{
Volodymyr G. Lukiianykecb256f2011-01-13 15:46:26 -08002881 int ret;
2882
2883 if (!s)
2884 return 0;
2885
2886 ret = __parse_numa_zonelist_order(s);
2887 if (ret == 0)
2888 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
2889
2890 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002891}
2892early_param("numa_zonelist_order", setup_numa_zonelist_order);
2893
2894/*
2895 * sysctl handler for numa_zonelist_order
2896 */
2897int numa_zonelist_order_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07002898 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002899 loff_t *ppos)
2900{
2901 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2902 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01002903 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002904
Andi Kleen443c6f12009-12-23 21:00:47 +01002905 mutex_lock(&zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002906 if (write)
Andi Kleen443c6f12009-12-23 21:00:47 +01002907 strcpy(saved_string, (char*)table->data);
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07002908 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002909 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01002910 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002911 if (write) {
2912 int oldval = user_zonelist_order;
2913 if (__parse_numa_zonelist_order((char*)table->data)) {
2914 /*
2915 * bogus value. restore saved string
2916 */
2917 strncpy((char*)table->data, saved_string,
2918 NUMA_ZONELIST_ORDER_LEN);
2919 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07002920 } else if (oldval != user_zonelist_order) {
2921 mutex_lock(&zonelists_mutex);
Haicheng Li1f522502010-05-24 14:32:51 -07002922 build_all_zonelists(NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07002923 mutex_unlock(&zonelists_mutex);
2924 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002925 }
Andi Kleen443c6f12009-12-23 21:00:47 +01002926out:
2927 mutex_unlock(&zl_order_mutex);
2928 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002929}
2930
2931
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002932#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002933static int node_load[MAX_NUMNODES];
2934
Linus Torvalds1da177e2005-04-16 15:20:36 -07002935/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07002936 * 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 -07002937 * @node: node whose fallback list we're appending
2938 * @used_node_mask: nodemask_t of already used nodes
2939 *
2940 * We use a number of factors to determine which is the next node that should
2941 * appear on a given node's fallback list. The node should not have appeared
2942 * already in @node's fallback list, and it should be the next closest node
2943 * according to the distance array (which contains arbitrary distance values
2944 * from each node to each node in the system), and should also prefer nodes
2945 * with no CPUs, since presumably they'll have very little allocation pressure
2946 * on them otherwise.
2947 * It returns -1 if no node is found.
2948 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002949static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002950{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002951 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002952 int min_val = INT_MAX;
2953 int best_node = -1;
Rusty Russella70f7302009-03-13 14:49:46 +10302954 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002956 /* Use the local node if we haven't already */
2957 if (!node_isset(node, *used_node_mask)) {
2958 node_set(node, *used_node_mask);
2959 return node;
2960 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002961
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002962 for_each_node_state(n, N_HIGH_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002963
2964 /* Don't want a node to appear more than once */
2965 if (node_isset(n, *used_node_mask))
2966 continue;
2967
Linus Torvalds1da177e2005-04-16 15:20:36 -07002968 /* Use the distance array to find the distance */
2969 val = node_distance(node, n);
2970
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002971 /* Penalize nodes under us ("prefer the next node") */
2972 val += (n < node);
2973
Linus Torvalds1da177e2005-04-16 15:20:36 -07002974 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10302975 tmp = cpumask_of_node(n);
2976 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002977 val += PENALTY_FOR_NODE_WITH_CPUS;
2978
2979 /* Slight preference for less loaded node */
2980 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2981 val += node_load[n];
2982
2983 if (val < min_val) {
2984 min_val = val;
2985 best_node = n;
2986 }
2987 }
2988
2989 if (best_node >= 0)
2990 node_set(best_node, *used_node_mask);
2991
2992 return best_node;
2993}
2994
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002995
2996/*
2997 * Build zonelists ordered by node and zones within node.
2998 * This results in maximum locality--normal zone overflows into local
2999 * DMA zone, if any--but risks exhausting DMA zone.
3000 */
3001static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003003 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003005
Mel Gorman54a6eb52008-04-28 02:12:16 -07003006 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07003007 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07003008 ;
3009 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3010 MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07003011 zonelist->_zonerefs[j].zone = NULL;
3012 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003013}
3014
3015/*
Christoph Lameter523b9452007-10-16 01:25:37 -07003016 * Build gfp_thisnode zonelists
3017 */
3018static void build_thisnode_zonelists(pg_data_t *pgdat)
3019{
Christoph Lameter523b9452007-10-16 01:25:37 -07003020 int j;
3021 struct zonelist *zonelist;
3022
Mel Gorman54a6eb52008-04-28 02:12:16 -07003023 zonelist = &pgdat->node_zonelists[1];
3024 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07003025 zonelist->_zonerefs[j].zone = NULL;
3026 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07003027}
3028
3029/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003030 * Build zonelists ordered by zone and nodes within zones.
3031 * This results in conserving DMA zone[s] until all Normal memory is
3032 * exhausted, but results in overflowing to remote node while memory
3033 * may still exist in local DMA zone.
3034 */
3035static int node_order[MAX_NUMNODES];
3036
3037static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
3038{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003039 int pos, j, node;
3040 int zone_type; /* needs to be signed */
3041 struct zone *z;
3042 struct zonelist *zonelist;
3043
Mel Gorman54a6eb52008-04-28 02:12:16 -07003044 zonelist = &pgdat->node_zonelists[0];
3045 pos = 0;
3046 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
3047 for (j = 0; j < nr_nodes; j++) {
3048 node = node_order[j];
3049 z = &NODE_DATA(node)->node_zones[zone_type];
3050 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07003051 zoneref_set_zone(z,
3052 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07003053 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003054 }
3055 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003056 }
Mel Gormandd1a2392008-04-28 02:12:17 -07003057 zonelist->_zonerefs[pos].zone = NULL;
3058 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003059}
3060
3061static int default_zonelist_order(void)
3062{
3063 int nid, zone_type;
3064 unsigned long low_kmem_size,total_size;
3065 struct zone *z;
3066 int average_size;
3067 /*
Thomas Weber88393162010-03-16 11:47:56 +01003068 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003069 * If they are really small and used heavily, the system can fall
3070 * into OOM very easily.
David Rientjese325c902010-05-24 14:32:13 -07003071 * This function detect ZONE_DMA/DMA32 size and configures zone order.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003072 */
3073 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
3074 low_kmem_size = 0;
3075 total_size = 0;
3076 for_each_online_node(nid) {
3077 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3078 z = &NODE_DATA(nid)->node_zones[zone_type];
3079 if (populated_zone(z)) {
3080 if (zone_type < ZONE_NORMAL)
3081 low_kmem_size += z->present_pages;
3082 total_size += z->present_pages;
David Rientjese325c902010-05-24 14:32:13 -07003083 } else if (zone_type == ZONE_NORMAL) {
3084 /*
3085 * If any node has only lowmem, then node order
3086 * is preferred to allow kernel allocations
3087 * locally; otherwise, they can easily infringe
3088 * on other nodes when there is an abundance of
3089 * lowmem available to allocate from.
3090 */
3091 return ZONELIST_ORDER_NODE;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003092 }
3093 }
3094 }
3095 if (!low_kmem_size || /* there are no DMA area. */
3096 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
3097 return ZONELIST_ORDER_NODE;
3098 /*
3099 * look into each node's config.
3100 * If there is a node whose DMA/DMA32 memory is very big area on
3101 * local memory, NODE_ORDER may be suitable.
3102 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003103 average_size = total_size /
3104 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003105 for_each_online_node(nid) {
3106 low_kmem_size = 0;
3107 total_size = 0;
3108 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3109 z = &NODE_DATA(nid)->node_zones[zone_type];
3110 if (populated_zone(z)) {
3111 if (zone_type < ZONE_NORMAL)
3112 low_kmem_size += z->present_pages;
3113 total_size += z->present_pages;
3114 }
3115 }
3116 if (low_kmem_size &&
3117 total_size > average_size && /* ignore small node */
3118 low_kmem_size > total_size * 70/100)
3119 return ZONELIST_ORDER_NODE;
3120 }
3121 return ZONELIST_ORDER_ZONE;
3122}
3123
3124static void set_zonelist_order(void)
3125{
3126 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
3127 current_zonelist_order = default_zonelist_order();
3128 else
3129 current_zonelist_order = user_zonelist_order;
3130}
3131
3132static void build_zonelists(pg_data_t *pgdat)
3133{
3134 int j, node, load;
3135 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003136 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003137 int local_node, prev_node;
3138 struct zonelist *zonelist;
3139 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003140
3141 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07003142 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003143 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07003144 zonelist->_zonerefs[0].zone = NULL;
3145 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003146 }
3147
3148 /* NUMA-aware ordering of nodes */
3149 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003150 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003151 prev_node = local_node;
3152 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003153
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003154 memset(node_order, 0, sizeof(node_order));
3155 j = 0;
3156
Linus Torvalds1da177e2005-04-16 15:20:36 -07003157 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08003158 int distance = node_distance(local_node, node);
3159
3160 /*
3161 * If another node is sufficiently far away then it is better
3162 * to reclaim pages in a zone before going off node.
3163 */
3164 if (distance > RECLAIM_DISTANCE)
3165 zone_reclaim_mode = 1;
3166
Linus Torvalds1da177e2005-04-16 15:20:36 -07003167 /*
3168 * We don't want to pressure a particular node.
3169 * So adding penalty to the first node in same
3170 * distance group to make it round-robin.
3171 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08003172 if (distance != node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003173 node_load[node] = load;
3174
Linus Torvalds1da177e2005-04-16 15:20:36 -07003175 prev_node = node;
3176 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003177 if (order == ZONELIST_ORDER_NODE)
3178 build_zonelists_in_node_order(pgdat, node);
3179 else
3180 node_order[j++] = node; /* remember order */
3181 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003182
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003183 if (order == ZONELIST_ORDER_ZONE) {
3184 /* calculate node order -- i.e., DMA last! */
3185 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003186 }
Christoph Lameter523b9452007-10-16 01:25:37 -07003187
3188 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003189}
3190
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003191/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003192static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003193{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003194 struct zonelist *zonelist;
3195 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07003196 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003197
Mel Gorman54a6eb52008-04-28 02:12:16 -07003198 zonelist = &pgdat->node_zonelists[0];
3199 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3200 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07003201 for (z = zonelist->_zonerefs; z->zone; z++)
3202 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003203}
3204
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003205#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3206/*
3207 * Return node id of node used for "local" allocations.
3208 * I.e., first node id of first zone in arg node's generic zonelist.
3209 * Used for initializing percpu 'numa_mem', which is used primarily
3210 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3211 */
3212int local_memory_node(int node)
3213{
3214 struct zone *zone;
3215
3216 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3217 gfp_zone(GFP_KERNEL),
3218 NULL,
3219 &zone);
3220 return zone->node;
3221}
3222#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003223
Linus Torvalds1da177e2005-04-16 15:20:36 -07003224#else /* CONFIG_NUMA */
3225
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003226static void set_zonelist_order(void)
3227{
3228 current_zonelist_order = ZONELIST_ORDER_ZONE;
3229}
3230
3231static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003232{
Christoph Lameter19655d32006-09-25 23:31:19 -07003233 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003234 enum zone_type j;
3235 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003236
3237 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003238
Mel Gorman54a6eb52008-04-28 02:12:16 -07003239 zonelist = &pgdat->node_zonelists[0];
3240 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003241
Mel Gorman54a6eb52008-04-28 02:12:16 -07003242 /*
3243 * Now we build the zonelist so that it contains the zones
3244 * of all the other nodes.
3245 * We don't want to pressure a particular node, so when
3246 * building the zones for node N, we make sure that the
3247 * zones coming right after the local ones are those from
3248 * node N+1 (modulo N)
3249 */
3250 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3251 if (!node_online(node))
3252 continue;
3253 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3254 MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003255 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07003256 for (node = 0; node < local_node; node++) {
3257 if (!node_online(node))
3258 continue;
3259 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3260 MAX_NR_ZONES - 1);
3261 }
3262
Mel Gormandd1a2392008-04-28 02:12:17 -07003263 zonelist->_zonerefs[j].zone = NULL;
3264 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003265}
3266
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003267/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003268static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003269{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003270 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003271}
3272
Linus Torvalds1da177e2005-04-16 15:20:36 -07003273#endif /* CONFIG_NUMA */
3274
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003275/*
3276 * Boot pageset table. One per cpu which is going to be used for all
3277 * zones and all nodes. The parameters will be set in such a way
3278 * that an item put on a list will immediately be handed over to
3279 * the buddy list. This is safe since pageset manipulation is done
3280 * with interrupts disabled.
3281 *
3282 * The boot_pagesets must be kept even after bootup is complete for
3283 * unused processors and/or zones. They do play a role for bootstrapping
3284 * hotplugged processors.
3285 *
3286 * zoneinfo_show() and maybe other functions do
3287 * not check if the processor is online before following the pageset pointer.
3288 * Other parts of the kernel may not check if the zone is available.
3289 */
3290static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3291static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07003292static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003293
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003294/*
3295 * Global mutex to protect against size modification of zonelists
3296 * as well as to serialize pageset setup for the new populated zone.
3297 */
3298DEFINE_MUTEX(zonelists_mutex);
3299
Rusty Russell9b1a4d32008-07-28 12:16:30 -05003300/* return values int ....just for stop_machine() */
Haicheng Li1f522502010-05-24 14:32:51 -07003301static __init_refok int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003302{
Yasunori Goto68113782006-06-23 02:03:11 -07003303 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003304 int cpu;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003305
Bo Liu7f9cfb32009-08-18 14:11:19 -07003306#ifdef CONFIG_NUMA
3307 memset(node_load, 0, sizeof(node_load));
3308#endif
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003309 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07003310 pg_data_t *pgdat = NODE_DATA(nid);
3311
3312 build_zonelists(pgdat);
3313 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003314 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003315
3316 /*
3317 * Initialize the boot_pagesets that are going to be used
3318 * for bootstrapping processors. The real pagesets for
3319 * each zone will be allocated later when the per cpu
3320 * allocator is available.
3321 *
3322 * boot_pagesets are used also for bootstrapping offline
3323 * cpus if the system is already booted because the pagesets
3324 * are needed to initialize allocators on a specific cpu too.
3325 * F.e. the percpu allocator needs the page allocator which
3326 * needs the percpu allocator in order to allocate its pagesets
3327 * (a chicken-egg dilemma).
3328 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003329 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003330 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3331
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003332#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3333 /*
3334 * We now know the "local memory node" for each node--
3335 * i.e., the node of the first zone in the generic zonelist.
3336 * Set up numa_mem percpu variable for on-line cpus. During
3337 * boot, only the boot cpu should be on-line; we'll init the
3338 * secondary cpus' numa_mem as they come on-line. During
3339 * node/memory hotplug, we'll fixup all on-line cpus.
3340 */
3341 if (cpu_online(cpu))
3342 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3343#endif
3344 }
3345
Yasunori Goto68113782006-06-23 02:03:11 -07003346 return 0;
3347}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003348
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003349/*
3350 * Called with zonelists_mutex held always
3351 * unless system_state == SYSTEM_BOOTING.
3352 */
Paul Mundt9f6ae442011-04-14 15:21:57 -07003353void __ref build_all_zonelists(void *data)
Yasunori Goto68113782006-06-23 02:03:11 -07003354{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003355 set_zonelist_order();
3356
Yasunori Goto68113782006-06-23 02:03:11 -07003357 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07003358 __build_all_zonelists(NULL);
Mel Gorman68ad8df2008-07-23 21:26:52 -07003359 mminit_verify_zonelist();
Yasunori Goto68113782006-06-23 02:03:11 -07003360 cpuset_init_current_mems_allowed();
3361 } else {
Simon Arlott183ff222007-10-20 01:27:18 +02003362 /* we have to stop all cpus to guarantee there is no user
Yasunori Goto68113782006-06-23 02:03:11 -07003363 of zonelist */
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003364#ifdef CONFIG_MEMORY_HOTPLUG
3365 if (data)
3366 setup_zone_pageset((struct zone *)data);
3367#endif
3368 stop_machine(__build_all_zonelists, NULL, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07003369 /* cpuset refresh routine should be here */
3370 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07003371 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003372 /*
3373 * Disable grouping by mobility if the number of pages in the
3374 * system is too low to allow the mechanism to work. It would be
3375 * more accurate, but expensive to check per-zone. This check is
3376 * made on memory-hotadd so a system can start with mobility
3377 * disabled and enable it later
3378 */
Mel Gormand9c23402007-10-16 01:26:01 -07003379 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003380 page_group_by_mobility_disabled = 1;
3381 else
3382 page_group_by_mobility_disabled = 0;
3383
3384 printk("Built %i zonelists in %s order, mobility grouping %s. "
3385 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003386 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003387 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003388 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003389 vm_total_pages);
3390#ifdef CONFIG_NUMA
3391 printk("Policy zone: %s\n", zone_names[policy_zone]);
3392#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003393}
3394
3395/*
3396 * Helper functions to size the waitqueue hash table.
3397 * Essentially these want to choose hash table sizes sufficiently
3398 * large so that collisions trying to wait on pages are rare.
3399 * But in fact, the number of active page waitqueues on typical
3400 * systems is ridiculously low, less than 200. So this is even
3401 * conservative, even though it seems large.
3402 *
3403 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3404 * waitqueues, i.e. the size of the waitq table given the number of pages.
3405 */
3406#define PAGES_PER_WAITQUEUE 256
3407
Yasunori Gotocca448f2006-06-23 02:03:10 -07003408#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07003409static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003410{
3411 unsigned long size = 1;
3412
3413 pages /= PAGES_PER_WAITQUEUE;
3414
3415 while (size < pages)
3416 size <<= 1;
3417
3418 /*
3419 * Once we have dozens or even hundreds of threads sleeping
3420 * on IO we've got bigger problems than wait queue collision.
3421 * Limit the size of the wait table to a reasonable size.
3422 */
3423 size = min(size, 4096UL);
3424
3425 return max(size, 4UL);
3426}
Yasunori Gotocca448f2006-06-23 02:03:10 -07003427#else
3428/*
3429 * A zone's size might be changed by hot-add, so it is not possible to determine
3430 * a suitable size for its wait_table. So we use the maximum size now.
3431 *
3432 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3433 *
3434 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3435 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3436 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3437 *
3438 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3439 * or more by the traditional way. (See above). It equals:
3440 *
3441 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3442 * ia64(16K page size) : = ( 8G + 4M)byte.
3443 * powerpc (64K page size) : = (32G +16M)byte.
3444 */
3445static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3446{
3447 return 4096UL;
3448}
3449#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003450
3451/*
3452 * This is an integer logarithm so that shifts can be used later
3453 * to extract the more random high bits from the multiplicative
3454 * hash function before the remainder is taken.
3455 */
3456static inline unsigned long wait_table_bits(unsigned long size)
3457{
3458 return ffz(~size);
3459}
3460
3461#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3462
Mel Gorman56fd56b2007-10-16 01:25:58 -07003463/*
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07003464 * Check if a pageblock contains reserved pages
3465 */
3466static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
3467{
3468 unsigned long pfn;
3469
3470 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
3471 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
3472 return 1;
3473 }
3474 return 0;
3475}
3476
3477/*
Mel Gormand9c23402007-10-16 01:26:01 -07003478 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07003479 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3480 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07003481 * higher will lead to a bigger reserve which will get freed as contiguous
3482 * blocks as reclaim kicks in
3483 */
3484static void setup_zone_migrate_reserve(struct zone *zone)
3485{
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07003486 unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003487 struct page *page;
Mel Gorman78986a62009-09-21 17:03:02 -07003488 unsigned long block_migratetype;
3489 int reserve;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003490
Michal Hockod02156382011-12-08 14:34:27 -08003491 /*
3492 * Get the start pfn, end pfn and the number of blocks to reserve
3493 * We have to be careful to be aligned to pageblock_nr_pages to
3494 * make sure that we always check pfn_valid for the first page in
3495 * the block.
3496 */
Mel Gorman56fd56b2007-10-16 01:25:58 -07003497 start_pfn = zone->zone_start_pfn;
3498 end_pfn = start_pfn + zone->spanned_pages;
Michal Hockod02156382011-12-08 14:34:27 -08003499 start_pfn = roundup(start_pfn, pageblock_nr_pages);
Mel Gorman41858962009-06-16 15:32:12 -07003500 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07003501 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003502
Mel Gorman78986a62009-09-21 17:03:02 -07003503 /*
3504 * Reserve blocks are generally in place to help high-order atomic
3505 * allocations that are short-lived. A min_free_kbytes value that
3506 * would result in more than 2 reserve blocks for atomic allocations
3507 * is assumed to be in place to help anti-fragmentation for the
3508 * future allocation of hugepages at runtime.
3509 */
3510 reserve = min(2, reserve);
3511
Mel Gormand9c23402007-10-16 01:26:01 -07003512 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07003513 if (!pfn_valid(pfn))
3514 continue;
3515 page = pfn_to_page(pfn);
3516
Adam Litke344c7902008-09-02 14:35:38 -07003517 /* Watch out for overlapping nodes */
3518 if (page_to_nid(page) != zone_to_nid(zone))
3519 continue;
3520
Mel Gorman56fd56b2007-10-16 01:25:58 -07003521 block_migratetype = get_pageblock_migratetype(page);
3522
Mel Gorman938929f2012-01-10 15:07:14 -08003523 /* Only test what is necessary when the reserves are not met */
3524 if (reserve > 0) {
3525 /*
3526 * Blocks with reserved pages will never free, skip
3527 * them.
3528 */
3529 block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
3530 if (pageblock_is_reserved(pfn, block_end_pfn))
3531 continue;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003532
Mel Gorman938929f2012-01-10 15:07:14 -08003533 /* If this block is reserved, account for it */
3534 if (block_migratetype == MIGRATE_RESERVE) {
3535 reserve--;
3536 continue;
3537 }
3538
3539 /* Suitable for reserving if this block is movable */
3540 if (block_migratetype == MIGRATE_MOVABLE) {
3541 set_pageblock_migratetype(page,
3542 MIGRATE_RESERVE);
3543 move_freepages_block(zone, page,
3544 MIGRATE_RESERVE);
3545 reserve--;
3546 continue;
3547 }
Mel Gorman56fd56b2007-10-16 01:25:58 -07003548 }
3549
3550 /*
3551 * If the reserve is met and this is a previous reserved block,
3552 * take it back
3553 */
3554 if (block_migratetype == MIGRATE_RESERVE) {
3555 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
3556 move_freepages_block(zone, page, MIGRATE_MOVABLE);
3557 }
3558 }
3559}
Mel Gormanac0e5b72007-10-16 01:25:58 -07003560
Linus Torvalds1da177e2005-04-16 15:20:36 -07003561/*
3562 * Initially all pages are reserved - free ones are freed
3563 * up by free_all_bootmem() once the early boot process is
3564 * done. Non-atomic initialization, single-pass.
3565 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003566void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08003567 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003568{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003569 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07003570 unsigned long end_pfn = start_pfn + size;
3571 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003572 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003573
Hugh Dickins22b31ee2009-01-06 14:40:09 -08003574 if (highest_memmap_pfn < end_pfn - 1)
3575 highest_memmap_pfn = end_pfn - 1;
3576
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003577 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08003578 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08003579 /*
3580 * There can be holes in boot-time mem_map[]s
3581 * handed to this function. They do not
3582 * exist on hotplugged memory.
3583 */
3584 if (context == MEMMAP_EARLY) {
3585 if (!early_pfn_valid(pfn))
3586 continue;
3587 if (!early_pfn_in_nid(pfn, nid))
3588 continue;
3589 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003590 page = pfn_to_page(pfn);
3591 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07003592 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08003593 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003594 reset_page_mapcount(page);
3595 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003596 /*
3597 * Mark the block movable so that blocks are reserved for
3598 * movable at startup. This will force kernel allocations
3599 * to reserve their blocks rather than leaking throughout
3600 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07003601 * kernel allocations are made. Later some blocks near
3602 * the start are marked MIGRATE_RESERVE by
3603 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003604 *
3605 * bitmap is created for zone's valid pfn range. but memmap
3606 * can be created for invalid pages (for alignment)
3607 * check here not to call set_pageblock_migratetype() against
3608 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003609 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003610 if ((z->zone_start_pfn <= pfn)
3611 && (pfn < z->zone_start_pfn + z->spanned_pages)
3612 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07003613 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003614
Linus Torvalds1da177e2005-04-16 15:20:36 -07003615 INIT_LIST_HEAD(&page->lru);
3616#ifdef WANT_PAGE_VIRTUAL
3617 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3618 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07003619 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003620#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003621 }
3622}
3623
Andi Kleen1e548de2008-02-04 22:29:26 -08003624static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003625{
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003626 int order, t;
3627 for_each_migratetype_order(order, t) {
3628 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003629 zone->free_area[order].nr_free = 0;
3630 }
3631}
3632
3633#ifndef __HAVE_ARCH_MEMMAP_INIT
3634#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08003635 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003636#endif
3637
Sam Ravnborg1d6f4e62008-01-17 15:21:12 -08003638static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003639{
David Howells3a6be872009-05-06 16:03:03 -07003640#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003641 int batch;
3642
3643 /*
3644 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07003645 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003646 *
3647 * OK, so we don't know how big the cache is. So guess.
3648 */
3649 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07003650 if (batch * PAGE_SIZE > 512 * 1024)
3651 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003652 batch /= 4; /* We effectively *= 4 below */
3653 if (batch < 1)
3654 batch = 1;
3655
3656 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11003657 * Clamp the batch to a 2^n - 1 value. Having a power
3658 * of 2 value was found to be more likely to have
3659 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003660 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11003661 * For example if 2 tasks are alternately allocating
3662 * batches of pages, one task can end up with a lot
3663 * of pages of one half of the possible page colors
3664 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003665 */
David Howells91552032009-05-06 16:03:02 -07003666 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07003667
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003668 return batch;
David Howells3a6be872009-05-06 16:03:03 -07003669
3670#else
3671 /* The deferral and batching of frees should be suppressed under NOMMU
3672 * conditions.
3673 *
3674 * The problem is that NOMMU needs to be able to allocate large chunks
3675 * of contiguous memory as there's no hardware page translation to
3676 * assemble apparent contiguous memory from discontiguous pages.
3677 *
3678 * Queueing large contiguous runs of pages for batching, however,
3679 * causes the pages to actually be freed in smaller chunks. As there
3680 * can be a significant delay between the individual batches being
3681 * recycled, this leads to the once large chunks of space being
3682 * fragmented and becoming unavailable for high-order allocations.
3683 */
3684 return 0;
3685#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003686}
3687
Adrian Bunkb69a7282008-07-23 21:28:12 -07003688static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
Christoph Lameter2caaad42005-06-21 17:15:00 -07003689{
3690 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003691 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003692
Magnus Damm1c6fe942005-10-26 01:58:59 -07003693 memset(p, 0, sizeof(*p));
3694
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003695 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003696 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003697 pcp->high = 6 * batch;
3698 pcp->batch = max(1UL, 1 * batch);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003699 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
3700 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07003701}
3702
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003703/*
3704 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3705 * to the value high for the pageset p.
3706 */
3707
3708static void setup_pagelist_highmark(struct per_cpu_pageset *p,
3709 unsigned long high)
3710{
3711 struct per_cpu_pages *pcp;
3712
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003713 pcp = &p->pcp;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003714 pcp->high = high;
3715 pcp->batch = max(1UL, high/4);
3716 if ((high/4) > (PAGE_SHIFT * 8))
3717 pcp->batch = PAGE_SHIFT * 8;
3718}
3719
Nikanth Karthikesan58c2ee42011-03-15 10:59:02 +05303720static void setup_zone_pageset(struct zone *zone)
Wu Fengguang319774e2010-05-24 14:32:49 -07003721{
3722 int cpu;
3723
3724 zone->pageset = alloc_percpu(struct per_cpu_pageset);
3725
3726 for_each_possible_cpu(cpu) {
3727 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
3728
3729 setup_pageset(pcp, zone_batchsize(zone));
3730
3731 if (percpu_pagelist_fraction)
3732 setup_pagelist_highmark(pcp,
3733 (zone->present_pages /
3734 percpu_pagelist_fraction));
3735 }
3736}
3737
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003738/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003739 * Allocate per cpu pagesets and initialize them.
3740 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07003741 */
Al Viro78d99552005-12-15 09:18:25 +00003742void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003743{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003744 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003745
Wu Fengguang319774e2010-05-24 14:32:49 -07003746 for_each_populated_zone(zone)
3747 setup_zone_pageset(zone);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003748}
3749
Sam Ravnborg577a32f2007-05-17 23:29:25 +02003750static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07003751int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07003752{
3753 int i;
3754 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003755 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07003756
3757 /*
3758 * The per-page waitqueue mechanism uses hashed waitqueues
3759 * per zone.
3760 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07003761 zone->wait_table_hash_nr_entries =
3762 wait_table_hash_nr_entries(zone_size_pages);
3763 zone->wait_table_bits =
3764 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003765 alloc_size = zone->wait_table_hash_nr_entries
3766 * sizeof(wait_queue_head_t);
3767
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07003768 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07003769 zone->wait_table = (wait_queue_head_t *)
Yinghai Lu8f389a992011-05-11 15:13:32 -07003770 alloc_bootmem_node_nopanic(pgdat, alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003771 } else {
3772 /*
3773 * This case means that a zone whose size was 0 gets new memory
3774 * via memory hot-add.
3775 * But it may be the case that a new node was hot-added. In
3776 * this case vmalloc() will not be able to use this new node's
3777 * memory - this wait_table must be initialized to use this new
3778 * node itself as well.
3779 * To use this new node's memory, further consideration will be
3780 * necessary.
3781 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07003782 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003783 }
3784 if (!zone->wait_table)
3785 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07003786
Yasunori Goto02b694d2006-06-23 02:03:08 -07003787 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07003788 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003789
3790 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003791}
3792
Shaohua Li112067f2009-09-21 17:01:16 -07003793static int __zone_pcp_update(void *data)
3794{
3795 struct zone *zone = data;
3796 int cpu;
3797 unsigned long batch = zone_batchsize(zone), flags;
3798
Thomas Gleixner2d30a1f2010-03-10 15:20:40 -08003799 for_each_possible_cpu(cpu) {
Shaohua Li112067f2009-09-21 17:01:16 -07003800 struct per_cpu_pageset *pset;
3801 struct per_cpu_pages *pcp;
3802
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003803 pset = per_cpu_ptr(zone->pageset, cpu);
Shaohua Li112067f2009-09-21 17:01:16 -07003804 pcp = &pset->pcp;
3805
3806 local_irq_save(flags);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003807 free_pcppages_bulk(zone, pcp->count, pcp);
Shaohua Li112067f2009-09-21 17:01:16 -07003808 setup_pageset(pset, batch);
3809 local_irq_restore(flags);
3810 }
3811 return 0;
3812}
3813
3814void zone_pcp_update(struct zone *zone)
3815{
3816 stop_machine(__zone_pcp_update, zone, NULL);
3817}
3818
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003819static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07003820{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003821 /*
3822 * per cpu subsystem is not up at this point. The following code
3823 * relies on the ability of the linker to provide the
3824 * offset of a (static) per cpu variable into the per cpu area.
3825 */
3826 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07003827
Anton Blanchardf5335c02006-03-25 03:06:49 -08003828 if (zone->present_pages)
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003829 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
3830 zone->name, zone->present_pages,
3831 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07003832}
3833
Yasunori Goto718127c2006-06-23 02:03:10 -07003834__meminit int init_currently_empty_zone(struct zone *zone,
3835 unsigned long zone_start_pfn,
Dave Hansena2f3aa022007-01-10 23:15:30 -08003836 unsigned long size,
3837 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07003838{
3839 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003840 int ret;
3841 ret = zone_wait_table_init(zone, size);
3842 if (ret)
3843 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07003844 pgdat->nr_zones = zone_idx(zone) + 1;
3845
Dave Hansened8ece22005-10-29 18:16:50 -07003846 zone->zone_start_pfn = zone_start_pfn;
3847
Mel Gorman708614e2008-07-23 21:26:51 -07003848 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3849 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3850 pgdat->node_id,
3851 (unsigned long)zone_idx(zone),
3852 zone_start_pfn, (zone_start_pfn + size));
3853
Andi Kleen1e548de2008-02-04 22:29:26 -08003854 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07003855
3856 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003857}
3858
Tejun Heo0ee332c2011-12-08 10:22:09 -08003859#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07003860#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3861/*
3862 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3863 * Architectures may implement their own version but if add_active_range()
3864 * was used and there are no special requirements, this is a convenient
3865 * alternative
3866 */
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003867int __meminit __early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07003868{
Tejun Heoc13291a2011-07-12 10:46:30 +02003869 unsigned long start_pfn, end_pfn;
3870 int i, nid;
Mel Gormanc7132162006-09-27 01:49:43 -07003871
Tejun Heoc13291a2011-07-12 10:46:30 +02003872 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003873 if (start_pfn <= pfn && pfn < end_pfn)
Tejun Heoc13291a2011-07-12 10:46:30 +02003874 return nid;
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003875 /* This is a memory hole */
3876 return -1;
Mel Gormanc7132162006-09-27 01:49:43 -07003877}
3878#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3879
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003880int __meminit early_pfn_to_nid(unsigned long pfn)
3881{
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003882 int nid;
3883
3884 nid = __early_pfn_to_nid(pfn);
3885 if (nid >= 0)
3886 return nid;
3887 /* just returns 0 */
3888 return 0;
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003889}
3890
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003891#ifdef CONFIG_NODES_SPAN_OTHER_NODES
3892bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3893{
3894 int nid;
3895
3896 nid = __early_pfn_to_nid(pfn);
3897 if (nid >= 0 && nid != node)
3898 return false;
3899 return true;
3900}
3901#endif
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003902
Mel Gormanc7132162006-09-27 01:49:43 -07003903/**
3904 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003905 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3906 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07003907 *
3908 * If an architecture guarantees that all ranges registered with
3909 * add_active_ranges() contain no holes and may be freed, this
3910 * this function may be used instead of calling free_bootmem() manually.
3911 */
Tejun Heoc13291a2011-07-12 10:46:30 +02003912void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07003913{
Tejun Heoc13291a2011-07-12 10:46:30 +02003914 unsigned long start_pfn, end_pfn;
3915 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07003916
Tejun Heoc13291a2011-07-12 10:46:30 +02003917 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
3918 start_pfn = min(start_pfn, max_low_pfn);
3919 end_pfn = min(end_pfn, max_low_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003920
Tejun Heoc13291a2011-07-12 10:46:30 +02003921 if (start_pfn < end_pfn)
3922 free_bootmem_node(NODE_DATA(this_nid),
3923 PFN_PHYS(start_pfn),
3924 (end_pfn - start_pfn) << PAGE_SHIFT);
Mel Gormanc7132162006-09-27 01:49:43 -07003925 }
3926}
3927
Yinghai Lu08677212010-02-10 01:20:20 -08003928int __init add_from_early_node_map(struct range *range, int az,
3929 int nr_range, int nid)
3930{
Tejun Heoc13291a2011-07-12 10:46:30 +02003931 unsigned long start_pfn, end_pfn;
Yinghai Lu08677212010-02-10 01:20:20 -08003932 int i;
Yinghai Lu08677212010-02-10 01:20:20 -08003933
3934 /* need to go over early_node_map to find out good range for node */
Tejun Heoc13291a2011-07-12 10:46:30 +02003935 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL)
3936 nr_range = add_range(range, az, nr_range, start_pfn, end_pfn);
Yinghai Lu08677212010-02-10 01:20:20 -08003937 return nr_range;
3938}
3939
Mel Gormanc7132162006-09-27 01:49:43 -07003940/**
3941 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003942 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07003943 *
3944 * If an architecture guarantees that all ranges registered with
3945 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003946 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07003947 */
3948void __init sparse_memory_present_with_active_regions(int nid)
3949{
Tejun Heoc13291a2011-07-12 10:46:30 +02003950 unsigned long start_pfn, end_pfn;
3951 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07003952
Tejun Heoc13291a2011-07-12 10:46:30 +02003953 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
3954 memory_present(this_nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003955}
3956
3957/**
3958 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003959 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3960 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3961 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07003962 *
3963 * It returns the start and end page frame of a node based on information
3964 * provided by an arch calling add_active_range(). If called for a node
3965 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003966 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07003967 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003968void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003969 unsigned long *start_pfn, unsigned long *end_pfn)
3970{
Tejun Heoc13291a2011-07-12 10:46:30 +02003971 unsigned long this_start_pfn, this_end_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003972 int i;
Tejun Heoc13291a2011-07-12 10:46:30 +02003973
Mel Gormanc7132162006-09-27 01:49:43 -07003974 *start_pfn = -1UL;
3975 *end_pfn = 0;
3976
Tejun Heoc13291a2011-07-12 10:46:30 +02003977 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
3978 *start_pfn = min(*start_pfn, this_start_pfn);
3979 *end_pfn = max(*end_pfn, this_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003980 }
3981
Christoph Lameter633c0662007-10-16 01:25:37 -07003982 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07003983 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07003984}
3985
3986/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07003987 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3988 * assumption is made that zones within a node are ordered in monotonic
3989 * increasing memory addresses so that the "highest" populated zone is used
3990 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003991static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07003992{
3993 int zone_index;
3994 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3995 if (zone_index == ZONE_MOVABLE)
3996 continue;
3997
3998 if (arch_zone_highest_possible_pfn[zone_index] >
3999 arch_zone_lowest_possible_pfn[zone_index])
4000 break;
4001 }
4002
4003 VM_BUG_ON(zone_index == -1);
4004 movable_zone = zone_index;
4005}
4006
4007/*
4008 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004009 * because it is sized independent of architecture. Unlike the other zones,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004010 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4011 * in each node depending on the size of each node and how evenly kernelcore
4012 * is distributed. This helper function adjusts the zone ranges
4013 * provided by the architecture for a given node by using the end of the
4014 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4015 * zones within a node are in order of monotonic increases memory addresses
4016 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004017static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004018 unsigned long zone_type,
4019 unsigned long node_start_pfn,
4020 unsigned long node_end_pfn,
4021 unsigned long *zone_start_pfn,
4022 unsigned long *zone_end_pfn)
4023{
4024 /* Only adjust if ZONE_MOVABLE is on this node */
4025 if (zone_movable_pfn[nid]) {
4026 /* Size ZONE_MOVABLE */
4027 if (zone_type == ZONE_MOVABLE) {
4028 *zone_start_pfn = zone_movable_pfn[nid];
4029 *zone_end_pfn = min(node_end_pfn,
4030 arch_zone_highest_possible_pfn[movable_zone]);
4031
4032 /* Adjust for ZONE_MOVABLE starting within this range */
4033 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
4034 *zone_end_pfn > zone_movable_pfn[nid]) {
4035 *zone_end_pfn = zone_movable_pfn[nid];
4036
4037 /* Check if this whole range is within ZONE_MOVABLE */
4038 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
4039 *zone_start_pfn = *zone_end_pfn;
4040 }
4041}
4042
4043/*
Mel Gormanc7132162006-09-27 01:49:43 -07004044 * Return the number of pages a zone spans in a node, including holes
4045 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4046 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004047static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004048 unsigned long zone_type,
4049 unsigned long *ignored)
4050{
4051 unsigned long node_start_pfn, node_end_pfn;
4052 unsigned long zone_start_pfn, zone_end_pfn;
4053
4054 /* Get the start and end of the node and zone */
4055 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
4056 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4057 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07004058 adjust_zone_range_for_zone_movable(nid, zone_type,
4059 node_start_pfn, node_end_pfn,
4060 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004061
4062 /* Check that this node has pages within the zone's required range */
4063 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4064 return 0;
4065
4066 /* Move the zone boundaries inside the node if necessary */
4067 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4068 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4069
4070 /* Return the spanned pages */
4071 return zone_end_pfn - zone_start_pfn;
4072}
4073
4074/*
4075 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004076 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07004077 */
Yinghai Lu32996252009-12-15 17:59:02 -08004078unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004079 unsigned long range_start_pfn,
4080 unsigned long range_end_pfn)
4081{
Tejun Heo96e907d2011-07-12 10:46:29 +02004082 unsigned long nr_absent = range_end_pfn - range_start_pfn;
4083 unsigned long start_pfn, end_pfn;
4084 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004085
Tejun Heo96e907d2011-07-12 10:46:29 +02004086 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
4087 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
4088 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
4089 nr_absent -= end_pfn - start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004090 }
Tejun Heo96e907d2011-07-12 10:46:29 +02004091 return nr_absent;
Mel Gormanc7132162006-09-27 01:49:43 -07004092}
4093
4094/**
4095 * absent_pages_in_range - Return number of page frames in holes within a range
4096 * @start_pfn: The start PFN to start searching for holes
4097 * @end_pfn: The end PFN to stop searching for holes
4098 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004099 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07004100 */
4101unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4102 unsigned long end_pfn)
4103{
4104 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4105}
4106
4107/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004108static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004109 unsigned long zone_type,
4110 unsigned long *ignored)
4111{
Tejun Heo96e907d2011-07-12 10:46:29 +02004112 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
4113 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman9c7cd682006-09-27 01:49:58 -07004114 unsigned long node_start_pfn, node_end_pfn;
4115 unsigned long zone_start_pfn, zone_end_pfn;
4116
4117 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
Tejun Heo96e907d2011-07-12 10:46:29 +02004118 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
4119 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004120
Mel Gorman2a1e2742007-07-17 04:03:12 -07004121 adjust_zone_range_for_zone_movable(nid, zone_type,
4122 node_start_pfn, node_end_pfn,
4123 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004124 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004125}
Mel Gorman0e0b8642006-09-27 01:49:56 -07004126
Tejun Heo0ee332c2011-12-08 10:22:09 -08004127#else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004128static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004129 unsigned long zone_type,
4130 unsigned long *zones_size)
4131{
4132 return zones_size[zone_type];
4133}
4134
Paul Mundt6ea6e682007-07-15 23:38:20 -07004135static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004136 unsigned long zone_type,
4137 unsigned long *zholes_size)
4138{
4139 if (!zholes_size)
4140 return 0;
4141
4142 return zholes_size[zone_type];
4143}
Mel Gorman0e0b8642006-09-27 01:49:56 -07004144
Tejun Heo0ee332c2011-12-08 10:22:09 -08004145#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07004146
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004147static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Mel Gormanc7132162006-09-27 01:49:43 -07004148 unsigned long *zones_size, unsigned long *zholes_size)
4149{
4150 unsigned long realtotalpages, totalpages = 0;
4151 enum zone_type i;
4152
4153 for (i = 0; i < MAX_NR_ZONES; i++)
4154 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
4155 zones_size);
4156 pgdat->node_spanned_pages = totalpages;
4157
4158 realtotalpages = totalpages;
4159 for (i = 0; i < MAX_NR_ZONES; i++)
4160 realtotalpages -=
4161 zone_absent_pages_in_node(pgdat->node_id, i,
4162 zholes_size);
4163 pgdat->node_present_pages = realtotalpages;
4164 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4165 realtotalpages);
4166}
4167
Mel Gorman835c1342007-10-16 01:25:47 -07004168#ifndef CONFIG_SPARSEMEM
4169/*
4170 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07004171 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4172 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07004173 * round what is now in bits to nearest long in bits, then return it in
4174 * bytes.
4175 */
4176static unsigned long __init usemap_size(unsigned long zonesize)
4177{
4178 unsigned long usemapsize;
4179
Mel Gormand9c23402007-10-16 01:26:01 -07004180 usemapsize = roundup(zonesize, pageblock_nr_pages);
4181 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07004182 usemapsize *= NR_PAGEBLOCK_BITS;
4183 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4184
4185 return usemapsize / 8;
4186}
4187
4188static void __init setup_usemap(struct pglist_data *pgdat,
4189 struct zone *zone, unsigned long zonesize)
4190{
4191 unsigned long usemapsize = usemap_size(zonesize);
4192 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08004193 if (usemapsize)
Yinghai Lu8f389a992011-05-11 15:13:32 -07004194 zone->pageblock_flags = alloc_bootmem_node_nopanic(pgdat,
4195 usemapsize);
Mel Gorman835c1342007-10-16 01:25:47 -07004196}
4197#else
Jesper Juhlfa9f90b2010-11-28 21:39:34 +01004198static inline void setup_usemap(struct pglist_data *pgdat,
Mel Gorman835c1342007-10-16 01:25:47 -07004199 struct zone *zone, unsigned long zonesize) {}
4200#endif /* CONFIG_SPARSEMEM */
4201
Mel Gormand9c23402007-10-16 01:26:01 -07004202#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08004203
4204/* Return a sensible default order for the pageblock size. */
4205static inline int pageblock_default_order(void)
4206{
4207 if (HPAGE_SHIFT > PAGE_SHIFT)
4208 return HUGETLB_PAGE_ORDER;
4209
4210 return MAX_ORDER-1;
4211}
4212
Mel Gormand9c23402007-10-16 01:26:01 -07004213/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4214static inline void __init set_pageblock_order(unsigned int order)
4215{
4216 /* Check that pageblock_nr_pages has not already been setup */
4217 if (pageblock_order)
4218 return;
4219
4220 /*
4221 * Assume the largest contiguous order of interest is a huge page.
4222 * This value may be variable depending on boot parameters on IA64
4223 */
4224 pageblock_order = order;
4225}
4226#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4227
Mel Gormanba72cb82007-11-28 16:21:13 -08004228/*
4229 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4230 * and pageblock_default_order() are unused as pageblock_order is set
4231 * at compile-time. See include/linux/pageblock-flags.h for the values of
4232 * pageblock_order based on the kernel config
4233 */
4234static inline int pageblock_default_order(unsigned int order)
4235{
4236 return MAX_ORDER-1;
4237}
Mel Gormand9c23402007-10-16 01:26:01 -07004238#define set_pageblock_order(x) do {} while (0)
4239
4240#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4241
Linus Torvalds1da177e2005-04-16 15:20:36 -07004242/*
4243 * Set up the zone data structures:
4244 * - mark all pages reserved
4245 * - mark all memory queues empty
4246 * - clear the memory bitmaps
4247 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08004248static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004249 unsigned long *zones_size, unsigned long *zholes_size)
4250{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07004251 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07004252 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004253 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07004254 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004255
Dave Hansen208d54e2005-10-29 18:16:52 -07004256 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004257 pgdat->nr_zones = 0;
4258 init_waitqueue_head(&pgdat->kswapd_wait);
4259 pgdat->kswapd_max_order = 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004260 pgdat_page_cgroup_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004261
4262 for (j = 0; j < MAX_NR_ZONES; j++) {
4263 struct zone *zone = pgdat->node_zones + j;
Mel Gorman0e0b8642006-09-27 01:49:56 -07004264 unsigned long size, realsize, memmap_pages;
Hugh Dickins41113042012-01-12 17:20:01 -08004265 enum lru_list lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004266
Mel Gormanc7132162006-09-27 01:49:43 -07004267 size = zone_spanned_pages_in_node(nid, j, zones_size);
4268 realsize = size - zone_absent_pages_in_node(nid, j,
4269 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004270
Mel Gorman0e0b8642006-09-27 01:49:56 -07004271 /*
4272 * Adjust realsize so that it accounts for how much memory
4273 * is used by this zone for memmap. This affects the watermark
4274 * and per-cpu initialisations
4275 */
Johannes Weinerf7232152008-05-23 13:04:21 -07004276 memmap_pages =
4277 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
Mel Gorman0e0b8642006-09-27 01:49:56 -07004278 if (realsize >= memmap_pages) {
4279 realsize -= memmap_pages;
Yinghai Lu5594c8c2009-01-06 14:39:14 -08004280 if (memmap_pages)
4281 printk(KERN_DEBUG
4282 " %s zone: %lu pages used for memmap\n",
4283 zone_names[j], memmap_pages);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004284 } else
4285 printk(KERN_WARNING
4286 " %s zone: %lu pages exceeds realsize %lu\n",
4287 zone_names[j], memmap_pages, realsize);
4288
Christoph Lameter62672762007-02-10 01:43:07 -08004289 /* Account for reserved pages */
4290 if (j == 0 && realsize > dma_reserve) {
Mel Gorman0e0b8642006-09-27 01:49:56 -07004291 realsize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07004292 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08004293 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004294 }
4295
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004296 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004297 nr_kernel_pages += realsize;
4298 nr_all_pages += realsize;
4299
4300 zone->spanned_pages = size;
4301 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07004302#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07004303 zone->node = nid;
Christoph Lameter8417bba2006-09-25 23:31:51 -07004304 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07004305 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07004306 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07004307#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004308 zone->name = zone_names[j];
4309 spin_lock_init(&zone->lock);
4310 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07004311 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004312 zone->zone_pgdat = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004313
Dave Hansened8ece22005-10-29 18:16:50 -07004314 zone_pcp_init(zone);
Hugh Dickins41113042012-01-12 17:20:01 -08004315 for_each_lru(lru)
4316 INIT_LIST_HEAD(&zone->lruvec.lists[lru]);
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08004317 zone->reclaim_stat.recent_rotated[0] = 0;
4318 zone->reclaim_stat.recent_rotated[1] = 0;
4319 zone->reclaim_stat.recent_scanned[0] = 0;
4320 zone->reclaim_stat.recent_scanned[1] = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07004321 zap_zone_vm_stats(zone);
David Rientjese815af92007-10-16 23:25:54 -07004322 zone->flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004323 if (!size)
4324 continue;
4325
Mel Gormanba72cb82007-11-28 16:21:13 -08004326 set_pageblock_order(pageblock_default_order());
Mel Gorman835c1342007-10-16 01:25:47 -07004327 setup_usemap(pgdat, zone, size);
Dave Hansena2f3aa022007-01-10 23:15:30 -08004328 ret = init_currently_empty_zone(zone, zone_start_pfn,
4329 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07004330 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07004331 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004332 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004333 }
4334}
4335
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004336static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004337{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004338 /* Skip empty nodes */
4339 if (!pgdat->node_spanned_pages)
4340 return;
4341
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004342#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07004343 /* ia64 gets its own node_mem_map, before this, without bootmem */
4344 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07004345 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004346 struct page *map;
4347
Bob Piccoe984bb42006-05-20 15:00:31 -07004348 /*
4349 * The zone's endpoints aren't required to be MAX_ORDER
4350 * aligned but the node_mem_map endpoints must be in order
4351 * for the buddy allocator to function correctly.
4352 */
4353 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
4354 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
4355 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4356 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07004357 map = alloc_remap(pgdat->node_id, size);
4358 if (!map)
Yinghai Lu8f389a992011-05-11 15:13:32 -07004359 map = alloc_bootmem_node_nopanic(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07004360 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004361 }
Roman Zippel12d810c2007-05-31 00:40:54 -07004362#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07004363 /*
4364 * With no DISCONTIG, the global mem_map is just set as node 0's
4365 */
Mel Gormanc7132162006-09-27 01:49:43 -07004366 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004367 mem_map = NODE_DATA(0)->node_mem_map;
Tejun Heo0ee332c2011-12-08 10:22:09 -08004368#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07004369 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08004370 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Tejun Heo0ee332c2011-12-08 10:22:09 -08004371#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07004372 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004373#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004374#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004375}
4376
Johannes Weiner9109fb72008-07-23 21:27:20 -07004377void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4378 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004379{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004380 pg_data_t *pgdat = NODE_DATA(nid);
4381
Linus Torvalds1da177e2005-04-16 15:20:36 -07004382 pgdat->node_id = nid;
4383 pgdat->node_start_pfn = node_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004384 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004385
4386 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07004387#ifdef CONFIG_FLAT_NODE_MEM_MAP
4388 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4389 nid, (unsigned long)pgdat,
4390 (unsigned long)pgdat->node_mem_map);
4391#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004392
4393 free_area_init_core(pgdat, zones_size, zholes_size);
4394}
4395
Tejun Heo0ee332c2011-12-08 10:22:09 -08004396#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07004397
4398#if MAX_NUMNODES > 1
4399/*
4400 * Figure out the number of possible node ids.
4401 */
4402static void __init setup_nr_node_ids(void)
4403{
4404 unsigned int node;
4405 unsigned int highest = 0;
4406
4407 for_each_node_mask(node, node_possible_map)
4408 highest = node;
4409 nr_node_ids = highest + 1;
4410}
4411#else
4412static inline void setup_nr_node_ids(void)
4413{
4414}
4415#endif
4416
Mel Gormanc7132162006-09-27 01:49:43 -07004417/**
Tejun Heo1e019792011-07-12 09:45:34 +02004418 * node_map_pfn_alignment - determine the maximum internode alignment
4419 *
4420 * This function should be called after node map is populated and sorted.
4421 * It calculates the maximum power of two alignment which can distinguish
4422 * all the nodes.
4423 *
4424 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
4425 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
4426 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
4427 * shifted, 1GiB is enough and this function will indicate so.
4428 *
4429 * This is used to test whether pfn -> nid mapping of the chosen memory
4430 * model has fine enough granularity to avoid incorrect mapping for the
4431 * populated node map.
4432 *
4433 * Returns the determined alignment in pfn's. 0 if there is no alignment
4434 * requirement (single node).
4435 */
4436unsigned long __init node_map_pfn_alignment(void)
4437{
4438 unsigned long accl_mask = 0, last_end = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02004439 unsigned long start, end, mask;
Tejun Heo1e019792011-07-12 09:45:34 +02004440 int last_nid = -1;
Tejun Heoc13291a2011-07-12 10:46:30 +02004441 int i, nid;
Tejun Heo1e019792011-07-12 09:45:34 +02004442
Tejun Heoc13291a2011-07-12 10:46:30 +02004443 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
Tejun Heo1e019792011-07-12 09:45:34 +02004444 if (!start || last_nid < 0 || last_nid == nid) {
4445 last_nid = nid;
4446 last_end = end;
4447 continue;
4448 }
4449
4450 /*
4451 * Start with a mask granular enough to pin-point to the
4452 * start pfn and tick off bits one-by-one until it becomes
4453 * too coarse to separate the current node from the last.
4454 */
4455 mask = ~((1 << __ffs(start)) - 1);
4456 while (mask && last_end <= (start & (mask << 1)))
4457 mask <<= 1;
4458
4459 /* accumulate all internode masks */
4460 accl_mask |= mask;
4461 }
4462
4463 /* convert mask to number of pages */
4464 return ~accl_mask + 1;
4465}
4466
Mel Gormana6af2bc2007-02-10 01:42:57 -08004467/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004468static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07004469{
Mel Gormana6af2bc2007-02-10 01:42:57 -08004470 unsigned long min_pfn = ULONG_MAX;
Tejun Heoc13291a2011-07-12 10:46:30 +02004471 unsigned long start_pfn;
4472 int i;
Mel Gorman1abbfb42006-11-23 12:01:41 +00004473
Tejun Heoc13291a2011-07-12 10:46:30 +02004474 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
4475 min_pfn = min(min_pfn, start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004476
Mel Gormana6af2bc2007-02-10 01:42:57 -08004477 if (min_pfn == ULONG_MAX) {
4478 printk(KERN_WARNING
Paul Jackson2bc0d262008-06-22 07:22:17 -07004479 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08004480 return 0;
4481 }
4482
4483 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004484}
4485
4486/**
4487 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4488 *
4489 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004490 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07004491 */
4492unsigned long __init find_min_pfn_with_active_regions(void)
4493{
4494 return find_min_pfn_for_node(MAX_NUMNODES);
4495}
4496
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004497/*
4498 * early_calculate_totalpages()
4499 * Sum pages in active regions for movable zone.
4500 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4501 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07004502static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07004503{
Mel Gorman7e63efe2007-07-17 04:03:15 -07004504 unsigned long totalpages = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02004505 unsigned long start_pfn, end_pfn;
4506 int i, nid;
Mel Gorman7e63efe2007-07-17 04:03:15 -07004507
Tejun Heoc13291a2011-07-12 10:46:30 +02004508 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
4509 unsigned long pages = end_pfn - start_pfn;
4510
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004511 totalpages += pages;
4512 if (pages)
Tejun Heoc13291a2011-07-12 10:46:30 +02004513 node_set_state(nid, N_HIGH_MEMORY);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004514 }
4515 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07004516}
4517
Mel Gorman2a1e2742007-07-17 04:03:12 -07004518/*
4519 * Find the PFN the Movable zone begins in each node. Kernel memory
4520 * is spread evenly between nodes as long as the nodes have enough
4521 * memory. When they don't, some nodes will have more kernelcore than
4522 * others
4523 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004524static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004525{
4526 int i, nid;
4527 unsigned long usable_startpfn;
4528 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07004529 /* save the state before borrow the nodemask */
4530 nodemask_t saved_node_state = node_states[N_HIGH_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004531 unsigned long totalpages = early_calculate_totalpages();
4532 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004533
Mel Gorman7e63efe2007-07-17 04:03:15 -07004534 /*
4535 * If movablecore was specified, calculate what size of
4536 * kernelcore that corresponds so that memory usable for
4537 * any allocation type is evenly spread. If both kernelcore
4538 * and movablecore are specified, then the value of kernelcore
4539 * will be used for required_kernelcore if it's greater than
4540 * what movablecore would have allowed.
4541 */
4542 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07004543 unsigned long corepages;
4544
4545 /*
4546 * Round-up so that ZONE_MOVABLE is at least as large as what
4547 * was requested by the user
4548 */
4549 required_movablecore =
4550 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
4551 corepages = totalpages - required_movablecore;
4552
4553 required_kernelcore = max(required_kernelcore, corepages);
4554 }
4555
Mel Gorman2a1e2742007-07-17 04:03:12 -07004556 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4557 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07004558 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004559
4560 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4561 find_usable_zone_for_movable();
4562 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4563
4564restart:
4565 /* Spread kernelcore memory as evenly as possible throughout nodes */
4566 kernelcore_node = required_kernelcore / usable_nodes;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004567 for_each_node_state(nid, N_HIGH_MEMORY) {
Tejun Heoc13291a2011-07-12 10:46:30 +02004568 unsigned long start_pfn, end_pfn;
4569
Mel Gorman2a1e2742007-07-17 04:03:12 -07004570 /*
4571 * Recalculate kernelcore_node if the division per node
4572 * now exceeds what is necessary to satisfy the requested
4573 * amount of memory for the kernel
4574 */
4575 if (required_kernelcore < kernelcore_node)
4576 kernelcore_node = required_kernelcore / usable_nodes;
4577
4578 /*
4579 * As the map is walked, we track how much memory is usable
4580 * by the kernel using kernelcore_remaining. When it is
4581 * 0, the rest of the node is usable by ZONE_MOVABLE
4582 */
4583 kernelcore_remaining = kernelcore_node;
4584
4585 /* Go through each range of PFNs within this node */
Tejun Heoc13291a2011-07-12 10:46:30 +02004586 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004587 unsigned long size_pages;
4588
Tejun Heoc13291a2011-07-12 10:46:30 +02004589 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004590 if (start_pfn >= end_pfn)
4591 continue;
4592
4593 /* Account for what is only usable for kernelcore */
4594 if (start_pfn < usable_startpfn) {
4595 unsigned long kernel_pages;
4596 kernel_pages = min(end_pfn, usable_startpfn)
4597 - start_pfn;
4598
4599 kernelcore_remaining -= min(kernel_pages,
4600 kernelcore_remaining);
4601 required_kernelcore -= min(kernel_pages,
4602 required_kernelcore);
4603
4604 /* Continue if range is now fully accounted */
4605 if (end_pfn <= usable_startpfn) {
4606
4607 /*
4608 * Push zone_movable_pfn to the end so
4609 * that if we have to rebalance
4610 * kernelcore across nodes, we will
4611 * not double account here
4612 */
4613 zone_movable_pfn[nid] = end_pfn;
4614 continue;
4615 }
4616 start_pfn = usable_startpfn;
4617 }
4618
4619 /*
4620 * The usable PFN range for ZONE_MOVABLE is from
4621 * start_pfn->end_pfn. Calculate size_pages as the
4622 * number of pages used as kernelcore
4623 */
4624 size_pages = end_pfn - start_pfn;
4625 if (size_pages > kernelcore_remaining)
4626 size_pages = kernelcore_remaining;
4627 zone_movable_pfn[nid] = start_pfn + size_pages;
4628
4629 /*
4630 * Some kernelcore has been met, update counts and
4631 * break if the kernelcore for this node has been
4632 * satisified
4633 */
4634 required_kernelcore -= min(required_kernelcore,
4635 size_pages);
4636 kernelcore_remaining -= size_pages;
4637 if (!kernelcore_remaining)
4638 break;
4639 }
4640 }
4641
4642 /*
4643 * If there is still required_kernelcore, we do another pass with one
4644 * less node in the count. This will push zone_movable_pfn[nid] further
4645 * along on the nodes that still have memory until kernelcore is
4646 * satisified
4647 */
4648 usable_nodes--;
4649 if (usable_nodes && required_kernelcore > usable_nodes)
4650 goto restart;
4651
4652 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4653 for (nid = 0; nid < MAX_NUMNODES; nid++)
4654 zone_movable_pfn[nid] =
4655 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07004656
4657out:
4658 /* restore the node_state */
4659 node_states[N_HIGH_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004660}
4661
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004662/* Any regular memory on that node ? */
4663static void check_for_regular_memory(pg_data_t *pgdat)
4664{
4665#ifdef CONFIG_HIGHMEM
4666 enum zone_type zone_type;
4667
4668 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4669 struct zone *zone = &pgdat->node_zones[zone_type];
Bob Liud0048b02012-01-12 17:19:07 -08004670 if (zone->present_pages) {
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004671 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
Bob Liud0048b02012-01-12 17:19:07 -08004672 break;
4673 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004674 }
4675#endif
4676}
4677
Mel Gormanc7132162006-09-27 01:49:43 -07004678/**
4679 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004680 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07004681 *
4682 * This will call free_area_init_node() for each active node in the system.
4683 * Using the page ranges provided by add_active_range(), the size of each
4684 * zone in each node and their holes is calculated. If the maximum PFN
4685 * between two adjacent zones match, it is assumed that the zone is empty.
4686 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4687 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4688 * starts where the previous one ended. For example, ZONE_DMA32 starts
4689 * at arch_max_dma_pfn.
4690 */
4691void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4692{
Tejun Heoc13291a2011-07-12 10:46:30 +02004693 unsigned long start_pfn, end_pfn;
4694 int i, nid;
Mel Gormana6af2bc2007-02-10 01:42:57 -08004695
Mel Gormanc7132162006-09-27 01:49:43 -07004696 /* Record where the zone boundaries are */
4697 memset(arch_zone_lowest_possible_pfn, 0,
4698 sizeof(arch_zone_lowest_possible_pfn));
4699 memset(arch_zone_highest_possible_pfn, 0,
4700 sizeof(arch_zone_highest_possible_pfn));
4701 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4702 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4703 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004704 if (i == ZONE_MOVABLE)
4705 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004706 arch_zone_lowest_possible_pfn[i] =
4707 arch_zone_highest_possible_pfn[i-1];
4708 arch_zone_highest_possible_pfn[i] =
4709 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4710 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07004711 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4712 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4713
4714 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4715 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4716 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004717
Mel Gormanc7132162006-09-27 01:49:43 -07004718 /* Print out the zone ranges */
4719 printk("Zone PFN ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07004720 for (i = 0; i < MAX_NR_ZONES; i++) {
4721 if (i == ZONE_MOVABLE)
4722 continue;
David Rientjes72f0ba02010-03-05 13:42:14 -08004723 printk(" %-8s ", zone_names[i]);
4724 if (arch_zone_lowest_possible_pfn[i] ==
4725 arch_zone_highest_possible_pfn[i])
4726 printk("empty\n");
4727 else
4728 printk("%0#10lx -> %0#10lx\n",
Mel Gormanc7132162006-09-27 01:49:43 -07004729 arch_zone_lowest_possible_pfn[i],
4730 arch_zone_highest_possible_pfn[i]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004731 }
4732
4733 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4734 printk("Movable zone start PFN for each node\n");
4735 for (i = 0; i < MAX_NUMNODES; i++) {
4736 if (zone_movable_pfn[i])
4737 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4738 }
Mel Gormanc7132162006-09-27 01:49:43 -07004739
4740 /* Print out the early_node_map[] */
Tejun Heoc13291a2011-07-12 10:46:30 +02004741 printk("Early memory PFN ranges\n");
4742 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
4743 printk(" %3d: %0#10lx -> %0#10lx\n", nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004744
4745 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07004746 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08004747 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07004748 for_each_online_node(nid) {
4749 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07004750 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07004751 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004752
4753 /* Any memory on that node */
4754 if (pgdat->node_present_pages)
4755 node_set_state(nid, N_HIGH_MEMORY);
4756 check_for_regular_memory(pgdat);
Mel Gormanc7132162006-09-27 01:49:43 -07004757 }
4758}
Mel Gorman2a1e2742007-07-17 04:03:12 -07004759
Mel Gorman7e63efe2007-07-17 04:03:15 -07004760static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004761{
4762 unsigned long long coremem;
4763 if (!p)
4764 return -EINVAL;
4765
4766 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004767 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004768
Mel Gorman7e63efe2007-07-17 04:03:15 -07004769 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07004770 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4771
4772 return 0;
4773}
Mel Gormaned7ed362007-07-17 04:03:14 -07004774
Mel Gorman7e63efe2007-07-17 04:03:15 -07004775/*
4776 * kernelcore=size sets the amount of memory for use for allocations that
4777 * cannot be reclaimed or migrated.
4778 */
4779static int __init cmdline_parse_kernelcore(char *p)
4780{
4781 return cmdline_parse_core(p, &required_kernelcore);
4782}
4783
4784/*
4785 * movablecore=size sets the amount of memory for use for allocations that
4786 * can be reclaimed or migrated.
4787 */
4788static int __init cmdline_parse_movablecore(char *p)
4789{
4790 return cmdline_parse_core(p, &required_movablecore);
4791}
4792
Mel Gormaned7ed362007-07-17 04:03:14 -07004793early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004794early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07004795
Tejun Heo0ee332c2011-12-08 10:22:09 -08004796#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07004797
Mel Gorman0e0b8642006-09-27 01:49:56 -07004798/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004799 * set_dma_reserve - set the specified number of pages reserved in the first zone
4800 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07004801 *
4802 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4803 * In the DMA zone, a significant percentage may be consumed by kernel image
4804 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004805 * function may optionally be used to account for unfreeable pages in the
4806 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4807 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07004808 */
4809void __init set_dma_reserve(unsigned long new_dma_reserve)
4810{
4811 dma_reserve = new_dma_reserve;
4812}
4813
Linus Torvalds1da177e2005-04-16 15:20:36 -07004814void __init free_area_init(unsigned long *zones_size)
4815{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004816 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004817 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4818}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004819
Linus Torvalds1da177e2005-04-16 15:20:36 -07004820static int page_alloc_cpu_notify(struct notifier_block *self,
4821 unsigned long action, void *hcpu)
4822{
4823 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004824
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07004825 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004826 drain_pages(cpu);
4827
4828 /*
4829 * Spill the event counters of the dead processor
4830 * into the current processors event counters.
4831 * This artificially elevates the count of the current
4832 * processor.
4833 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07004834 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004835
4836 /*
4837 * Zero the differential counters of the dead processor
4838 * so that the vm statistics are consistent.
4839 *
4840 * This is only okay since the processor is dead and cannot
4841 * race with what we are doing.
4842 */
Christoph Lameter2244b952006-06-30 01:55:33 -07004843 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004844 }
4845 return NOTIFY_OK;
4846}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004847
4848void __init page_alloc_init(void)
4849{
4850 hotcpu_notifier(page_alloc_cpu_notify, 0);
4851}
4852
4853/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004854 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4855 * or min_free_kbytes changes.
4856 */
4857static void calculate_totalreserve_pages(void)
4858{
4859 struct pglist_data *pgdat;
4860 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004861 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004862
4863 for_each_online_pgdat(pgdat) {
4864 for (i = 0; i < MAX_NR_ZONES; i++) {
4865 struct zone *zone = pgdat->node_zones + i;
4866 unsigned long max = 0;
4867
4868 /* Find valid and maximum lowmem_reserve in the zone */
4869 for (j = i; j < MAX_NR_ZONES; j++) {
4870 if (zone->lowmem_reserve[j] > max)
4871 max = zone->lowmem_reserve[j];
4872 }
4873
Mel Gorman41858962009-06-16 15:32:12 -07004874 /* we treat the high watermark as reserved pages. */
4875 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004876
4877 if (max > zone->present_pages)
4878 max = zone->present_pages;
4879 reserve_pages += max;
Johannes Weinerab8fabd2012-01-10 15:07:42 -08004880 /*
4881 * Lowmem reserves are not available to
4882 * GFP_HIGHUSER page cache allocations and
4883 * kswapd tries to balance zones to their high
4884 * watermark. As a result, neither should be
4885 * regarded as dirtyable memory, to prevent a
4886 * situation where reclaim has to clean pages
4887 * in order to balance the zones.
4888 */
4889 zone->dirty_balance_reserve = max;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004890 }
4891 }
Johannes Weinerab8fabd2012-01-10 15:07:42 -08004892 dirty_balance_reserve = reserve_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004893 totalreserve_pages = reserve_pages;
4894}
4895
4896/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07004897 * setup_per_zone_lowmem_reserve - called whenever
4898 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4899 * has a correct pages reserved value, so an adequate number of
4900 * pages are left in the zone after a successful __alloc_pages().
4901 */
4902static void setup_per_zone_lowmem_reserve(void)
4903{
4904 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004905 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004906
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08004907 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004908 for (j = 0; j < MAX_NR_ZONES; j++) {
4909 struct zone *zone = pgdat->node_zones + j;
4910 unsigned long present_pages = zone->present_pages;
4911
4912 zone->lowmem_reserve[j] = 0;
4913
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004914 idx = j;
4915 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004916 struct zone *lower_zone;
4917
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004918 idx--;
4919
Linus Torvalds1da177e2005-04-16 15:20:36 -07004920 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4921 sysctl_lowmem_reserve_ratio[idx] = 1;
4922
4923 lower_zone = pgdat->node_zones + idx;
4924 lower_zone->lowmem_reserve[j] = present_pages /
4925 sysctl_lowmem_reserve_ratio[idx];
4926 present_pages += lower_zone->present_pages;
4927 }
4928 }
4929 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004930
4931 /* update totalreserve_pages */
4932 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004933}
4934
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004935/**
Minchan Kimbc75d332009-06-16 15:32:48 -07004936 * setup_per_zone_wmarks - called when min_free_kbytes changes
Minchan Kimbce73942009-06-16 15:32:50 -07004937 * or when memory is hot-{added|removed}
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004938 *
Minchan Kimbc75d332009-06-16 15:32:48 -07004939 * Ensures that the watermark[min,low,high] values for each zone are set
4940 * correctly with respect to min_free_kbytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004941 */
Minchan Kimbc75d332009-06-16 15:32:48 -07004942void setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004943{
4944 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4945 unsigned long lowmem_pages = 0;
4946 struct zone *zone;
4947 unsigned long flags;
4948
4949 /* Calculate total number of !ZONE_HIGHMEM pages */
4950 for_each_zone(zone) {
4951 if (!is_highmem(zone))
4952 lowmem_pages += zone->present_pages;
4953 }
4954
4955 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07004956 u64 tmp;
4957
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07004958 spin_lock_irqsave(&zone->lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07004959 tmp = (u64)pages_min * zone->present_pages;
4960 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004961 if (is_highmem(zone)) {
4962 /*
Nick Piggin669ed172005-11-13 16:06:45 -08004963 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4964 * need highmem pages, so cap pages_min to a small
4965 * value here.
4966 *
Mel Gorman41858962009-06-16 15:32:12 -07004967 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Nick Piggin669ed172005-11-13 16:06:45 -08004968 * deltas controls asynch page reclaim, and so should
4969 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004970 */
4971 int min_pages;
4972
4973 min_pages = zone->present_pages / 1024;
4974 if (min_pages < SWAP_CLUSTER_MAX)
4975 min_pages = SWAP_CLUSTER_MAX;
4976 if (min_pages > 128)
4977 min_pages = 128;
Mel Gorman41858962009-06-16 15:32:12 -07004978 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004979 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08004980 /*
4981 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07004982 * proportionate to the zone's size.
4983 */
Mel Gorman41858962009-06-16 15:32:12 -07004984 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004985 }
4986
Mel Gorman41858962009-06-16 15:32:12 -07004987 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
4988 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Mel Gorman56fd56b2007-10-16 01:25:58 -07004989 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07004990 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004991 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004992
4993 /* update totalreserve_pages */
4994 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004995}
4996
Randy Dunlap55a44622009-09-21 17:01:20 -07004997/*
Rik van Riel556adec2008-10-18 20:26:34 -07004998 * The inactive anon list should be small enough that the VM never has to
4999 * do too much work, but large enough that each inactive page has a chance
5000 * to be referenced again before it is swapped out.
5001 *
5002 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5003 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5004 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5005 * the anonymous pages are kept on the inactive list.
5006 *
5007 * total target max
5008 * memory ratio inactive anon
5009 * -------------------------------------
5010 * 10MB 1 5MB
5011 * 100MB 1 50MB
5012 * 1GB 3 250MB
5013 * 10GB 10 0.9GB
5014 * 100GB 31 3GB
5015 * 1TB 101 10GB
5016 * 10TB 320 32GB
5017 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005018static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
Minchan Kim96cb4df2009-06-16 15:32:49 -07005019{
5020 unsigned int gb, ratio;
5021
5022 /* Zone size in gigabytes */
5023 gb = zone->present_pages >> (30 - PAGE_SHIFT);
5024 if (gb)
5025 ratio = int_sqrt(10 * gb);
5026 else
5027 ratio = 1;
5028
5029 zone->inactive_ratio = ratio;
5030}
5031
KOSAKI Motohiro839a4fc2011-05-24 17:11:31 -07005032static void __meminit setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07005033{
5034 struct zone *zone;
5035
Minchan Kim96cb4df2009-06-16 15:32:49 -07005036 for_each_zone(zone)
5037 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07005038}
5039
Linus Torvalds1da177e2005-04-16 15:20:36 -07005040/*
5041 * Initialise min_free_kbytes.
5042 *
5043 * For small machines we want it small (128k min). For large machines
5044 * we want it large (64MB max). But it is not linear, because network
5045 * bandwidth does not increase linearly with machine size. We use
5046 *
5047 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5048 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5049 *
5050 * which yields
5051 *
5052 * 16MB: 512k
5053 * 32MB: 724k
5054 * 64MB: 1024k
5055 * 128MB: 1448k
5056 * 256MB: 2048k
5057 * 512MB: 2896k
5058 * 1024MB: 4096k
5059 * 2048MB: 5792k
5060 * 4096MB: 8192k
5061 * 8192MB: 11584k
5062 * 16384MB: 16384k
5063 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005064int __meminit init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005065{
5066 unsigned long lowmem_kbytes;
5067
5068 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
5069
5070 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
5071 if (min_free_kbytes < 128)
5072 min_free_kbytes = 128;
5073 if (min_free_kbytes > 65536)
5074 min_free_kbytes = 65536;
Minchan Kimbc75d332009-06-16 15:32:48 -07005075 setup_per_zone_wmarks();
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -07005076 refresh_zone_stat_thresholds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005077 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07005078 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005079 return 0;
5080}
Minchan Kimbc75d332009-06-16 15:32:48 -07005081module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005082
5083/*
5084 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5085 * that we can call two helper functions whenever min_free_kbytes
5086 * changes.
5087 */
5088int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005089 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005090{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005091 proc_dointvec(table, write, buffer, length, ppos);
Mel Gorman3b1d92c2007-05-06 14:49:30 -07005092 if (write)
Minchan Kimbc75d332009-06-16 15:32:48 -07005093 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005094 return 0;
5095}
5096
Christoph Lameter96146342006-07-03 00:24:13 -07005097#ifdef CONFIG_NUMA
5098int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005099 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07005100{
5101 struct zone *zone;
5102 int rc;
5103
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005104 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07005105 if (rc)
5106 return rc;
5107
5108 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07005109 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07005110 sysctl_min_unmapped_ratio) / 100;
5111 return 0;
5112}
Christoph Lameter0ff38492006-09-25 23:31:52 -07005113
5114int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005115 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07005116{
5117 struct zone *zone;
5118 int rc;
5119
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005120 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07005121 if (rc)
5122 return rc;
5123
5124 for_each_zone(zone)
5125 zone->min_slab_pages = (zone->present_pages *
5126 sysctl_min_slab_ratio) / 100;
5127 return 0;
5128}
Christoph Lameter96146342006-07-03 00:24:13 -07005129#endif
5130
Linus Torvalds1da177e2005-04-16 15:20:36 -07005131/*
5132 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5133 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5134 * whenever sysctl_lowmem_reserve_ratio changes.
5135 *
5136 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07005137 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07005138 * if in function of the boot time zone sizes.
5139 */
5140int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005141 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005142{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005143 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005144 setup_per_zone_lowmem_reserve();
5145 return 0;
5146}
5147
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005148/*
5149 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5150 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5151 * can have before it gets flushed back to buddy allocator.
5152 */
5153
5154int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005155 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005156{
5157 struct zone *zone;
5158 unsigned int cpu;
5159 int ret;
5160
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005161 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005162 if (!write || (ret == -EINVAL))
5163 return ret;
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07005164 for_each_populated_zone(zone) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005165 for_each_possible_cpu(cpu) {
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005166 unsigned long high;
5167 high = zone->present_pages / percpu_pagelist_fraction;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005168 setup_pagelist_highmark(
5169 per_cpu_ptr(zone->pageset, cpu), high);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005170 }
5171 }
5172 return 0;
5173}
5174
David S. Millerf034b5d2006-08-24 03:08:07 -07005175int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005176
5177#ifdef CONFIG_NUMA
5178static int __init set_hashdist(char *str)
5179{
5180 if (!str)
5181 return 0;
5182 hashdist = simple_strtoul(str, &str, 0);
5183 return 1;
5184}
5185__setup("hashdist=", set_hashdist);
5186#endif
5187
5188/*
5189 * allocate a large system hash table from bootmem
5190 * - it is assumed that the hash table must contain an exact power-of-2
5191 * quantity of entries
5192 * - limit is the number of hash buckets, not the total allocation size
5193 */
5194void *__init alloc_large_system_hash(const char *tablename,
5195 unsigned long bucketsize,
5196 unsigned long numentries,
5197 int scale,
5198 int flags,
5199 unsigned int *_hash_shift,
5200 unsigned int *_hash_mask,
5201 unsigned long limit)
5202{
5203 unsigned long long max = limit;
5204 unsigned long log2qty, size;
5205 void *table = NULL;
5206
5207 /* allow the kernel cmdline to have a say */
5208 if (!numentries) {
5209 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08005210 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005211 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
5212 numentries >>= 20 - PAGE_SHIFT;
5213 numentries <<= 20 - PAGE_SHIFT;
5214
5215 /* limit to 1 bucket per 2^scale bytes of low memory */
5216 if (scale > PAGE_SHIFT)
5217 numentries >>= (scale - PAGE_SHIFT);
5218 else
5219 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08005220
5221 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07005222 if (unlikely(flags & HASH_SMALL)) {
5223 /* Makes no sense without HASH_EARLY */
5224 WARN_ON(!(flags & HASH_EARLY));
5225 if (!(numentries >> *_hash_shift)) {
5226 numentries = 1UL << *_hash_shift;
5227 BUG_ON(!numentries);
5228 }
5229 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08005230 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005231 }
John Hawkes6e692ed2006-03-25 03:08:02 -08005232 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005233
5234 /* limit allocation size to 1/16 total memory by default */
5235 if (max == 0) {
5236 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
5237 do_div(max, bucketsize);
5238 }
5239
5240 if (numentries > max)
5241 numentries = max;
5242
David Howellsf0d1b0b2006-12-08 02:37:49 -08005243 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005244
5245 do {
5246 size = bucketsize << log2qty;
5247 if (flags & HASH_EARLY)
Jan Beulich74768ed2008-08-12 15:08:39 -07005248 table = alloc_bootmem_nopanic(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005249 else if (hashdist)
5250 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
5251 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07005252 /*
5253 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07005254 * some pages at the end of hash table which
5255 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07005256 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005257 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07005258 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005259 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
5260 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005261 }
5262 } while (!table && size > PAGE_SIZE && --log2qty);
5263
5264 if (!table)
5265 panic("Failed to allocate %s hash table\n", tablename);
5266
Robin Holtf241e6602010-10-07 12:59:26 -07005267 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07005268 tablename,
Robin Holtf241e6602010-10-07 12:59:26 -07005269 (1UL << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08005270 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005271 size);
5272
5273 if (_hash_shift)
5274 *_hash_shift = log2qty;
5275 if (_hash_mask)
5276 *_hash_mask = (1 << log2qty) - 1;
5277
5278 return table;
5279}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08005280
Mel Gorman835c1342007-10-16 01:25:47 -07005281/* Return a pointer to the bitmap storing bits affecting a block of pages */
5282static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
5283 unsigned long pfn)
5284{
5285#ifdef CONFIG_SPARSEMEM
5286 return __pfn_to_section(pfn)->pageblock_flags;
5287#else
5288 return zone->pageblock_flags;
5289#endif /* CONFIG_SPARSEMEM */
5290}
Andrew Morton6220ec72006-10-19 23:29:05 -07005291
Mel Gorman835c1342007-10-16 01:25:47 -07005292static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
5293{
5294#ifdef CONFIG_SPARSEMEM
5295 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07005296 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005297#else
5298 pfn = pfn - zone->zone_start_pfn;
Mel Gormand9c23402007-10-16 01:26:01 -07005299 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005300#endif /* CONFIG_SPARSEMEM */
5301}
5302
5303/**
Mel Gormand9c23402007-10-16 01:26:01 -07005304 * 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 -07005305 * @page: The page within the block of interest
5306 * @start_bitidx: The first bit of interest to retrieve
5307 * @end_bitidx: The last bit of interest
5308 * returns pageblock_bits flags
5309 */
5310unsigned long get_pageblock_flags_group(struct page *page,
5311 int start_bitidx, int end_bitidx)
5312{
5313 struct zone *zone;
5314 unsigned long *bitmap;
5315 unsigned long pfn, bitidx;
5316 unsigned long flags = 0;
5317 unsigned long value = 1;
5318
5319 zone = page_zone(page);
5320 pfn = page_to_pfn(page);
5321 bitmap = get_pageblock_bitmap(zone, pfn);
5322 bitidx = pfn_to_bitidx(zone, pfn);
5323
5324 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5325 if (test_bit(bitidx + start_bitidx, bitmap))
5326 flags |= value;
5327
5328 return flags;
5329}
5330
5331/**
Mel Gormand9c23402007-10-16 01:26:01 -07005332 * 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 -07005333 * @page: The page within the block of interest
5334 * @start_bitidx: The first bit of interest
5335 * @end_bitidx: The last bit of interest
5336 * @flags: The flags to set
5337 */
5338void set_pageblock_flags_group(struct page *page, unsigned long flags,
5339 int start_bitidx, int end_bitidx)
5340{
5341 struct zone *zone;
5342 unsigned long *bitmap;
5343 unsigned long pfn, bitidx;
5344 unsigned long value = 1;
5345
5346 zone = page_zone(page);
5347 pfn = page_to_pfn(page);
5348 bitmap = get_pageblock_bitmap(zone, pfn);
5349 bitidx = pfn_to_bitidx(zone, pfn);
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07005350 VM_BUG_ON(pfn < zone->zone_start_pfn);
5351 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
Mel Gorman835c1342007-10-16 01:25:47 -07005352
5353 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5354 if (flags & value)
5355 __set_bit(bitidx + start_bitidx, bitmap);
5356 else
5357 __clear_bit(bitidx + start_bitidx, bitmap);
5358}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005359
5360/*
5361 * This is designed as sub function...plz see page_isolation.c also.
5362 * set/clear page block's type to be ISOLATE.
5363 * page allocater never alloc memory from ISOLATE block.
5364 */
5365
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005366static int
5367__count_immobile_pages(struct zone *zone, struct page *page, int count)
5368{
5369 unsigned long pfn, iter, found;
5370 /*
5371 * For avoiding noise data, lru_add_drain_all() should be called
5372 * If ZONE_MOVABLE, the zone never contains immobile pages
5373 */
5374 if (zone_idx(zone) == ZONE_MOVABLE)
5375 return true;
5376
5377 if (get_pageblock_migratetype(page) == MIGRATE_MOVABLE)
5378 return true;
5379
5380 pfn = page_to_pfn(page);
5381 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
5382 unsigned long check = pfn + iter;
5383
Namhyung Kim29723fc2011-02-25 14:44:25 -08005384 if (!pfn_valid_within(check))
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005385 continue;
Namhyung Kim29723fc2011-02-25 14:44:25 -08005386
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005387 page = pfn_to_page(check);
5388 if (!page_count(page)) {
5389 if (PageBuddy(page))
5390 iter += (1 << page_order(page)) - 1;
5391 continue;
5392 }
5393 if (!PageLRU(page))
5394 found++;
5395 /*
5396 * If there are RECLAIMABLE pages, we need to check it.
5397 * But now, memory offline itself doesn't call shrink_slab()
5398 * and it still to be fixed.
5399 */
5400 /*
5401 * If the page is not RAM, page_count()should be 0.
5402 * we don't need more check. This is an _used_ not-movable page.
5403 *
5404 * The problematic thing here is PG_reserved pages. PG_reserved
5405 * is set to both of a memory hole page and a _used_ kernel
5406 * page at boot.
5407 */
5408 if (found > count)
5409 return false;
5410 }
5411 return true;
5412}
5413
5414bool is_pageblock_removable_nolock(struct page *page)
5415{
Michal Hocko656a0702012-01-20 14:33:58 -08005416 struct zone *zone;
5417 unsigned long pfn;
Michal Hocko687875f2012-01-20 14:33:55 -08005418
5419 /*
5420 * We have to be careful here because we are iterating over memory
5421 * sections which are not zone aware so we might end up outside of
5422 * the zone but still within the section.
Michal Hocko656a0702012-01-20 14:33:58 -08005423 * We have to take care about the node as well. If the node is offline
5424 * its NODE_DATA will be NULL - see page_zone.
Michal Hocko687875f2012-01-20 14:33:55 -08005425 */
Michal Hocko656a0702012-01-20 14:33:58 -08005426 if (!node_online(page_to_nid(page)))
5427 return false;
5428
5429 zone = page_zone(page);
5430 pfn = page_to_pfn(page);
5431 if (zone->zone_start_pfn > pfn ||
Michal Hocko687875f2012-01-20 14:33:55 -08005432 zone->zone_start_pfn + zone->spanned_pages <= pfn)
5433 return false;
5434
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005435 return __count_immobile_pages(zone, page, 0);
5436}
5437
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005438int set_migratetype_isolate(struct page *page)
5439{
5440 struct zone *zone;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005441 unsigned long flags, pfn;
Robert Jennings925cc712009-12-17 14:44:38 +00005442 struct memory_isolate_notify arg;
5443 int notifier_ret;
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005444 int ret = -EBUSY;
5445
5446 zone = page_zone(page);
Robert Jennings925cc712009-12-17 14:44:38 +00005447
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005448 spin_lock_irqsave(&zone->lock, flags);
Robert Jennings925cc712009-12-17 14:44:38 +00005449
5450 pfn = page_to_pfn(page);
5451 arg.start_pfn = pfn;
5452 arg.nr_pages = pageblock_nr_pages;
5453 arg.pages_found = 0;
5454
5455 /*
5456 * It may be possible to isolate a pageblock even if the
5457 * migratetype is not MIGRATE_MOVABLE. The memory isolation
5458 * notifier chain is used by balloon drivers to return the
5459 * number of pages in a range that are held by the balloon
5460 * driver to shrink memory. If all the pages are accounted for
5461 * by balloons, are free, or on the LRU, isolation can continue.
5462 * Later, for example, when memory hotplug notifier runs, these
5463 * pages reported as "can be isolated" should be isolated(freed)
5464 * by the balloon driver through the memory notifier chain.
5465 */
5466 notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg);
5467 notifier_ret = notifier_to_errno(notifier_ret);
KAMEZAWA Hiroyuki4b204772010-10-26 14:21:29 -07005468 if (notifier_ret)
Robert Jennings925cc712009-12-17 14:44:38 +00005469 goto out;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005470 /*
5471 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
5472 * We just check MOVABLE pages.
5473 */
5474 if (__count_immobile_pages(zone, page, arg.pages_found))
Robert Jennings925cc712009-12-17 14:44:38 +00005475 ret = 0;
5476
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005477 /*
5478 * immobile means "not-on-lru" paes. If immobile is larger than
5479 * removable-by-driver pages reported by notifier, we'll fail.
5480 */
5481
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005482out:
Robert Jennings925cc712009-12-17 14:44:38 +00005483 if (!ret) {
5484 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
5485 move_freepages_block(zone, page, MIGRATE_ISOLATE);
5486 }
5487
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005488 spin_unlock_irqrestore(&zone->lock, flags);
5489 if (!ret)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005490 drain_all_pages();
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005491 return ret;
5492}
5493
5494void unset_migratetype_isolate(struct page *page)
5495{
5496 struct zone *zone;
5497 unsigned long flags;
5498 zone = page_zone(page);
5499 spin_lock_irqsave(&zone->lock, flags);
5500 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
5501 goto out;
5502 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5503 move_freepages_block(zone, page, MIGRATE_MOVABLE);
5504out:
5505 spin_unlock_irqrestore(&zone->lock, flags);
5506}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07005507
5508#ifdef CONFIG_MEMORY_HOTREMOVE
5509/*
5510 * All pages in the range must be isolated before calling this.
5511 */
5512void
5513__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
5514{
5515 struct page *page;
5516 struct zone *zone;
5517 int order, i;
5518 unsigned long pfn;
5519 unsigned long flags;
5520 /* find the first valid pfn */
5521 for (pfn = start_pfn; pfn < end_pfn; pfn++)
5522 if (pfn_valid(pfn))
5523 break;
5524 if (pfn == end_pfn)
5525 return;
5526 zone = page_zone(pfn_to_page(pfn));
5527 spin_lock_irqsave(&zone->lock, flags);
5528 pfn = start_pfn;
5529 while (pfn < end_pfn) {
5530 if (!pfn_valid(pfn)) {
5531 pfn++;
5532 continue;
5533 }
5534 page = pfn_to_page(pfn);
5535 BUG_ON(page_count(page));
5536 BUG_ON(!PageBuddy(page));
5537 order = page_order(page);
5538#ifdef CONFIG_DEBUG_VM
5539 printk(KERN_INFO "remove from free list %lx %d %lx\n",
5540 pfn, 1 << order, end_pfn);
5541#endif
5542 list_del(&page->lru);
5543 rmv_page_order(page);
5544 zone->free_area[order].nr_free--;
5545 __mod_zone_page_state(zone, NR_FREE_PAGES,
5546 - (1UL << order));
5547 for (i = 0; i < (1 << order); i++)
5548 SetPageReserved((page+i));
5549 pfn += (1 << order);
5550 }
5551 spin_unlock_irqrestore(&zone->lock, flags);
5552}
5553#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01005554
5555#ifdef CONFIG_MEMORY_FAILURE
5556bool is_free_buddy_page(struct page *page)
5557{
5558 struct zone *zone = page_zone(page);
5559 unsigned long pfn = page_to_pfn(page);
5560 unsigned long flags;
5561 int order;
5562
5563 spin_lock_irqsave(&zone->lock, flags);
5564 for (order = 0; order < MAX_ORDER; order++) {
5565 struct page *page_head = page - (pfn & ((1 << order) - 1));
5566
5567 if (PageBuddy(page_head) && page_order(page_head) >= order)
5568 break;
5569 }
5570 spin_unlock_irqrestore(&zone->lock, flags);
5571
5572 return order < MAX_ORDER;
5573}
5574#endif
Wu Fengguang718a3822010-03-10 15:20:43 -08005575
5576static struct trace_print_flags pageflag_names[] = {
5577 {1UL << PG_locked, "locked" },
5578 {1UL << PG_error, "error" },
5579 {1UL << PG_referenced, "referenced" },
5580 {1UL << PG_uptodate, "uptodate" },
5581 {1UL << PG_dirty, "dirty" },
5582 {1UL << PG_lru, "lru" },
5583 {1UL << PG_active, "active" },
5584 {1UL << PG_slab, "slab" },
5585 {1UL << PG_owner_priv_1, "owner_priv_1" },
5586 {1UL << PG_arch_1, "arch_1" },
5587 {1UL << PG_reserved, "reserved" },
5588 {1UL << PG_private, "private" },
5589 {1UL << PG_private_2, "private_2" },
5590 {1UL << PG_writeback, "writeback" },
5591#ifdef CONFIG_PAGEFLAGS_EXTENDED
5592 {1UL << PG_head, "head" },
5593 {1UL << PG_tail, "tail" },
5594#else
5595 {1UL << PG_compound, "compound" },
5596#endif
5597 {1UL << PG_swapcache, "swapcache" },
5598 {1UL << PG_mappedtodisk, "mappedtodisk" },
5599 {1UL << PG_reclaim, "reclaim" },
Wu Fengguang718a3822010-03-10 15:20:43 -08005600 {1UL << PG_swapbacked, "swapbacked" },
5601 {1UL << PG_unevictable, "unevictable" },
5602#ifdef CONFIG_MMU
5603 {1UL << PG_mlocked, "mlocked" },
5604#endif
5605#ifdef CONFIG_ARCH_USES_PG_UNCACHED
5606 {1UL << PG_uncached, "uncached" },
5607#endif
5608#ifdef CONFIG_MEMORY_FAILURE
5609 {1UL << PG_hwpoison, "hwpoison" },
5610#endif
5611 {-1UL, NULL },
5612};
5613
5614static void dump_page_flags(unsigned long flags)
5615{
5616 const char *delim = "";
5617 unsigned long mask;
5618 int i;
5619
5620 printk(KERN_ALERT "page flags: %#lx(", flags);
5621
5622 /* remove zone id */
5623 flags &= (1UL << NR_PAGEFLAGS) - 1;
5624
5625 for (i = 0; pageflag_names[i].name && flags; i++) {
5626
5627 mask = pageflag_names[i].mask;
5628 if ((flags & mask) != mask)
5629 continue;
5630
5631 flags &= ~mask;
5632 printk("%s%s", delim, pageflag_names[i].name);
5633 delim = "|";
5634 }
5635
5636 /* check for left over flags */
5637 if (flags)
5638 printk("%s%#lx", delim, flags);
5639
5640 printk(")\n");
5641}
5642
5643void dump_page(struct page *page)
5644{
5645 printk(KERN_ALERT
5646 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
Andrea Arcangeli4e9f64c2011-01-13 15:46:29 -08005647 page, atomic_read(&page->_count), page_mapcount(page),
Wu Fengguang718a3822010-03-10 15:20:43 -08005648 page->mapping, page->index);
5649 dump_page_flags(page->flags);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07005650 mem_cgroup_print_bad_page(page);
Wu Fengguang718a3822010-03-10 15:20:43 -08005651}