blob: 60ad5e9d48921108ef4271d453febfc66f8805ee [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 Rientjes5a3135c2007-10-16 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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
61#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070062#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070063#include "internal.h"
64
Lee Schermerhorn72812012010-05-26 14:44:56 -070065#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
66DEFINE_PER_CPU(int, numa_node);
67EXPORT_PER_CPU_SYMBOL(numa_node);
68#endif
69
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070070#ifdef CONFIG_HAVE_MEMORYLESS_NODES
71/*
72 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
73 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
74 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
75 * defined in <linux/topology.h>.
76 */
77DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
78EXPORT_PER_CPU_SYMBOL(_numa_mem_);
79#endif
80
Linus Torvalds1da177e2005-04-16 15:20:36 -070081/*
Christoph Lameter13808912007-10-16 01:25:27 -070082 * Array of node states.
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 */
Christoph Lameter13808912007-10-16 01:25:27 -070084nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
85 [N_POSSIBLE] = NODE_MASK_ALL,
86 [N_ONLINE] = { { [0] = 1UL } },
87#ifndef CONFIG_NUMA
88 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
89#ifdef CONFIG_HIGHMEM
90 [N_HIGH_MEMORY] = { { [0] = 1UL } },
91#endif
92 [N_CPU] = { { [0] = 1UL } },
93#endif /* NUMA */
94};
95EXPORT_SYMBOL(node_states);
96
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070097unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -070098unsigned long totalreserve_pages __read_mostly;
Jack Cheung59f9f1c2011-11-29 16:52:49 -080099#ifdef CONFIG_FIX_MOVABLE_ZONE
100unsigned long total_unmovable_pages __read_mostly;
101#endif
Rohit Seth8ad4b1f2006-01-08 01:00:40 -0800102int percpu_pagelist_fraction;
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000103gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800105#ifdef CONFIG_PM_SLEEP
106/*
107 * The following functions are used by the suspend/hibernate code to temporarily
108 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
109 * while devices are suspended. To avoid races with the suspend/hibernate code,
110 * they should always be called with pm_mutex held (gfp_allowed_mask also should
111 * only be modified with pm_mutex held, unless the suspend/hibernate code is
112 * guaranteed not to run in parallel with that modification).
113 */
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100114
115static gfp_t saved_gfp_mask;
116
117void pm_restore_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800118{
119 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100120 if (saved_gfp_mask) {
121 gfp_allowed_mask = saved_gfp_mask;
122 saved_gfp_mask = 0;
123 }
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800124}
125
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100126void pm_restrict_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800127{
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800128 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100129 WARN_ON(saved_gfp_mask);
130 saved_gfp_mask = gfp_allowed_mask;
131 gfp_allowed_mask &= ~GFP_IOFS;
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800132}
Mel Gorman2f53cb72011-10-24 16:33:42 -0700133
134static bool pm_suspending(void)
135{
136 if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS)
137 return false;
138 return true;
139}
140
141#else
142
143static bool pm_suspending(void)
144{
145 return false;
146}
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800147#endif /* CONFIG_PM_SLEEP */
148
Mel Gormand9c23402007-10-16 01:26:01 -0700149#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
150int pageblock_order __read_mostly;
151#endif
152
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800153static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -0800154
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155/*
156 * results with 256, 32 in the lowmem_reserve sysctl:
157 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
158 * 1G machine -> (16M dma, 784M normal, 224M high)
159 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
160 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
161 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +0100162 *
163 * TBD: should special case ZONE_DMA32 machines here - in those we normally
164 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700166int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800167#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700168 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -0800169#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700170#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700171 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700172#endif
Christoph Lametere53ef382006-09-25 23:31:14 -0700173#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700174 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700175#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700176 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700177};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178
179EXPORT_SYMBOL(totalram_pages);
Jack Cheung59f9f1c2011-11-29 16:52:49 -0800180#ifdef CONFIG_FIX_MOVABLE_ZONE
181EXPORT_SYMBOL(total_unmovable_pages);
182#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183
Helge Deller15ad7cd2006-12-06 20:40:36 -0800184static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800185#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700186 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800187#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700188#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700189 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700190#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700191 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700192#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700193 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700194#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700195 "Movable",
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700196};
197
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198int min_free_kbytes = 1024;
Arve Hjønnevåg83de7312009-02-17 14:51:02 -0800199int min_free_order_shift = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200
Jan Beulich2c85f512009-09-21 17:03:07 -0700201static unsigned long __meminitdata nr_kernel_pages;
202static unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700203static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
Mel Gormanc7132162006-09-27 01:49:43 -0700205#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
206 /*
Simon Arlott183ff222007-10-20 01:27:18 +0200207 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
Mel Gormanc7132162006-09-27 01:49:43 -0700208 * ranges of memory (RAM) that may be registered with add_active_range().
209 * Ranges passed to add_active_range() will be merged if possible
210 * so the number of times add_active_range() can be called is
211 * related to the number of nodes and the number of holes
212 */
213 #ifdef CONFIG_MAX_ACTIVE_REGIONS
214 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
215 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
216 #else
217 #if MAX_NUMNODES >= 32
218 /* If there can be many nodes, allow up to 50 holes per node */
219 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
220 #else
221 /* By default, allow up to 256 distinct regions */
222 #define MAX_ACTIVE_REGIONS 256
223 #endif
224 #endif
225
Jan Beulich98011f52007-07-15 23:38:17 -0700226 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
227 static int __meminitdata nr_nodemap_entries;
228 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
229 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
Adrian Bunkb69a7282008-07-23 21:28:12 -0700230 static unsigned long __initdata required_kernelcore;
Adrian Bunk484f51f2007-10-16 01:26:03 -0700231 static unsigned long __initdata required_movablecore;
Adrian Bunkb69a7282008-07-23 21:28:12 -0700232 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Mel Gorman2a1e2742007-07-17 04:03:12 -0700233
234 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
235 int movable_zone;
236 EXPORT_SYMBOL(movable_zone);
Mel Gormanc7132162006-09-27 01:49:43 -0700237#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
238
Miklos Szeredi418508c2007-05-23 13:57:55 -0700239#if MAX_NUMNODES > 1
240int nr_node_ids __read_mostly = MAX_NUMNODES;
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700241int nr_online_nodes __read_mostly = 1;
Miklos Szeredi418508c2007-05-23 13:57:55 -0700242EXPORT_SYMBOL(nr_node_ids);
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700243EXPORT_SYMBOL(nr_online_nodes);
Miklos Szeredi418508c2007-05-23 13:57:55 -0700244#endif
245
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700246int page_group_by_mobility_disabled __read_mostly;
247
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700248static void set_pageblock_migratetype(struct page *page, int migratetype)
249{
Mel Gorman49255c62009-06-16 15:31:58 -0700250
251 if (unlikely(page_group_by_mobility_disabled))
252 migratetype = MIGRATE_UNMOVABLE;
253
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700254 set_pageblock_flags_group(page, (unsigned long)migratetype,
255 PB_migrate, PB_migrate_end);
256}
257
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -0700258bool oom_killer_disabled __read_mostly;
259
Nick Piggin13e74442006-01-06 00:10:58 -0800260#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700261static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700263 int ret = 0;
264 unsigned seq;
265 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -0700266
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700267 do {
268 seq = zone_span_seqbegin(zone);
269 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
270 ret = 1;
271 else if (pfn < zone->zone_start_pfn)
272 ret = 1;
273 } while (zone_span_seqretry(zone, seq));
274
275 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700276}
277
278static int page_is_consistent(struct zone *zone, struct page *page)
279{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700280 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700281 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700283 return 0;
284
285 return 1;
286}
287/*
288 * Temporary debugging check for pages not lying within a given zone.
289 */
290static int bad_range(struct zone *zone, struct page *page)
291{
292 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700294 if (!page_is_consistent(zone, page))
295 return 1;
296
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 return 0;
298}
Nick Piggin13e74442006-01-06 00:10:58 -0800299#else
300static inline int bad_range(struct zone *zone, struct page *page)
301{
302 return 0;
303}
304#endif
305
Nick Piggin224abf92006-01-06 00:11:11 -0800306static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800308 static unsigned long resume;
309 static unsigned long nr_shown;
310 static unsigned long nr_unshown;
311
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200312 /* Don't complain about poisoned pages */
313 if (PageHWPoison(page)) {
Andrea Arcangelief2b4b92011-03-18 00:16:35 +0100314 reset_page_mapcount(page); /* remove PageBuddy */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200315 return;
316 }
317
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800318 /*
319 * Allow a burst of 60 reports, then keep quiet for that minute;
320 * or allow a steady drip of one report per second.
321 */
322 if (nr_shown == 60) {
323 if (time_before(jiffies, resume)) {
324 nr_unshown++;
325 goto out;
326 }
327 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800328 printk(KERN_ALERT
329 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800330 nr_unshown);
331 nr_unshown = 0;
332 }
333 nr_shown = 0;
334 }
335 if (nr_shown++ == 0)
336 resume = jiffies + 60 * HZ;
337
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800338 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800339 current->comm, page_to_pfn(page));
Wu Fengguang718a3822010-03-10 15:20:43 -0800340 dump_page(page);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700341
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800343out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800344 /* Leave bad fields for debug, except PageBuddy could make trouble */
Andrea Arcangelief2b4b92011-03-18 00:16:35 +0100345 reset_page_mapcount(page); /* remove PageBuddy */
Randy Dunlap9f158332005-09-13 01:25:16 -0700346 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347}
348
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349/*
350 * Higher-order pages are called "compound pages". They are structured thusly:
351 *
352 * The first PAGE_SIZE page is called the "head page".
353 *
354 * The remaining PAGE_SIZE pages are called "tail pages".
355 *
356 * All pages have PG_compound set. All pages have their ->private pointing at
357 * the head page (even the head page has this).
358 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800359 * The first tail page's ->lru.next holds the address of the compound page's
360 * put_page() function. Its ->lru.prev holds the order of allocation.
361 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800363
364static void free_compound_page(struct page *page)
365{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700366 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800367}
368
Andi Kleen01ad1c02008-07-23 21:27:46 -0700369void prep_compound_page(struct page *page, unsigned long order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370{
371 int i;
372 int nr_pages = 1 << order;
373
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800374 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700375 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700376 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800377 for (i = 1; i < nr_pages; i++) {
378 struct page *p = page + i;
379
Christoph Lameterd85f3382007-05-06 14:49:39 -0700380 __SetPageTail(p);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700381 p->first_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 }
383}
384
Andrea Arcangeli59ff4212011-01-13 15:46:38 -0800385/* update __split_huge_page_refcount if you change this function */
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800386static int destroy_compound_page(struct page *page, unsigned long order)
Andy Whitcroft18229df2008-11-06 12:53:27 -0800387{
388 int i;
389 int nr_pages = 1 << order;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800390 int bad = 0;
Andy Whitcroft18229df2008-11-06 12:53:27 -0800391
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800392 if (unlikely(compound_order(page) != order) ||
393 unlikely(!PageHead(page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800394 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800395 bad++;
396 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397
Christoph Lameter6d777952007-05-06 14:49:40 -0700398 __ClearPageHead(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800399
Andy Whitcroft18229df2008-11-06 12:53:27 -0800400 for (i = 1; i < nr_pages; i++) {
401 struct page *p = page + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402
Alexey Zaytseve713a212009-01-10 02:47:57 +0300403 if (unlikely(!PageTail(p) || (p->first_page != page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800404 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800405 bad++;
406 }
Christoph Lameterd85f3382007-05-06 14:49:39 -0700407 __ClearPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800409
410 return bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412
Nick Piggin17cf4402006-03-22 00:08:41 -0800413static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
414{
415 int i;
416
Andrew Morton6626c5d2006-03-22 00:08:42 -0800417 /*
418 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
419 * and __GFP_HIGHMEM from hard or soft interrupt context.
420 */
Nick Piggin725d7042006-09-25 23:30:55 -0700421 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800422 for (i = 0; i < (1 << order); i++)
423 clear_highpage(page + i);
424}
425
Andrew Morton6aa30012006-04-18 22:20:52 -0700426static inline void set_page_order(struct page *page, int order)
427{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700428 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000429 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430}
431
432static inline void rmv_page_order(struct page *page)
433{
Nick Piggin676165a2006-04-10 11:21:48 +1000434 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700435 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436}
437
438/*
439 * Locate the struct page for both the matching buddy in our
440 * pair (buddy1) and the combined O(n+1) page they form (page).
441 *
442 * 1) Any buddy B1 will have an order O twin B2 which satisfies
443 * the following equation:
444 * B2 = B1 ^ (1 << O)
445 * For example, if the starting buddy (buddy2) is #8 its order
446 * 1 buddy is #10:
447 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
448 *
449 * 2) Any buddy B will have an order O+1 parent P which
450 * satisfies the following equation:
451 * P = B & ~(1 << O)
452 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200453 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455static inline unsigned long
KyongHo Cho43506fa2011-01-13 15:47:24 -0800456__find_buddy_index(unsigned long page_idx, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457{
KyongHo Cho43506fa2011-01-13 15:47:24 -0800458 return page_idx ^ (1 << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459}
460
461/*
462 * This function checks whether a page is free && is the buddy
463 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800464 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000465 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700466 * (c) a page and its buddy have the same order &&
467 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 *
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800469 * For recording whether a page is in the buddy system, we set ->_mapcount -2.
470 * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock.
Nick Piggin676165a2006-04-10 11:21:48 +1000471 *
472 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700474static inline int page_is_buddy(struct page *page, struct page *buddy,
475 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700477 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800478 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800479
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700480 if (page_zone_id(page) != page_zone_id(buddy))
481 return 0;
482
483 if (PageBuddy(buddy) && page_order(buddy) == order) {
Nick Piggina3af9c32009-06-16 15:32:10 -0700484 VM_BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa30012006-04-18 22:20:52 -0700485 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000486 }
Andrew Morton6aa30012006-04-18 22:20:52 -0700487 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488}
489
490/*
491 * Freeing function for a buddy system allocator.
492 *
493 * The concept of a buddy system is to maintain direct-mapped table
494 * (containing bit values) for memory blocks of various "orders".
495 * The bottom level table contains the map for the smallest allocatable
496 * units of memory (here, pages), and each level above it describes
497 * pairs of units from the levels below, hence, "buddies".
498 * At a high level, all that happens here is marking the table entry
499 * at the bottom level available, and propagating the changes upward
500 * as necessary, plus some accounting needed to play nicely with other
501 * parts of the VM system.
502 * At each level, we keep a list of pages, which are heads of continuous
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800503 * free pages of length of (1 << order) and marked with _mapcount -2. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700504 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 * So when we are allocating or freeing one, we can derive the state of the
506 * other. That is, if we allocate a small block, and both were
507 * free, the remainder of the region must be split into blocks.
508 * If a block is freed, and its buddy is also free, then this
509 * triggers coalescing into a block of larger size.
510 *
511 * -- wli
512 */
513
Nick Piggin48db57f2006-01-08 01:00:42 -0800514static inline void __free_one_page(struct page *page,
Mel Gormaned0ae212009-06-16 15:32:07 -0700515 struct zone *zone, unsigned int order,
516 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517{
518 unsigned long page_idx;
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700519 unsigned long combined_idx;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800520 unsigned long uninitialized_var(buddy_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700521 struct page *buddy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522
Nick Piggin224abf92006-01-06 00:11:11 -0800523 if (unlikely(PageCompound(page)))
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800524 if (unlikely(destroy_compound_page(page, order)))
525 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526
Mel Gormaned0ae212009-06-16 15:32:07 -0700527 VM_BUG_ON(migratetype == -1);
528
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
530
Mel Gormanf2260e62009-06-16 15:32:13 -0700531 VM_BUG_ON(page_idx & ((1 << order) - 1));
Nick Piggin725d7042006-09-25 23:30:55 -0700532 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 while (order < MAX_ORDER-1) {
KyongHo Cho43506fa2011-01-13 15:47:24 -0800535 buddy_idx = __find_buddy_index(page_idx, order);
536 buddy = page + (buddy_idx - page_idx);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700537 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700538 break;
Nick Piggin13e74442006-01-06 00:10:58 -0800539
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700540 /* Our buddy is free, merge with it and move up one order. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 list_del(&buddy->lru);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700542 zone->free_area[order].nr_free--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 rmv_page_order(buddy);
KyongHo Cho43506fa2011-01-13 15:47:24 -0800544 combined_idx = buddy_idx & page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 page = page + (combined_idx - page_idx);
546 page_idx = combined_idx;
547 order++;
548 }
549 set_page_order(page, order);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700550
551 /*
552 * If this is not the largest possible page, check if the buddy
553 * of the next-highest order is free. If it is, it's possible
554 * that pages are being freed that will coalesce soon. In case,
555 * that is happening, add the free page to the tail of the list
556 * so it's less likely to be used soon and more likely to be merged
557 * as a higher order page
558 */
Mel Gormanb7f50cf2010-10-26 14:21:11 -0700559 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700560 struct page *higher_page, *higher_buddy;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800561 combined_idx = buddy_idx & page_idx;
562 higher_page = page + (combined_idx - page_idx);
563 buddy_idx = __find_buddy_index(combined_idx, order + 1);
564 higher_buddy = page + (buddy_idx - combined_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700565 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
566 list_add_tail(&page->lru,
567 &zone->free_area[order].free_list[migratetype]);
568 goto out;
569 }
570 }
571
572 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
573out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 zone->free_area[order].nr_free++;
575}
576
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700577/*
578 * free_page_mlock() -- clean up attempts to free and mlocked() page.
579 * Page should not be on lru, so no need to fix that up.
580 * free_pages_check() will verify...
581 */
582static inline void free_page_mlock(struct page *page)
583{
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700584 __dec_zone_page_state(page, NR_MLOCK);
585 __count_vm_event(UNEVICTABLE_MLOCKFREED);
586}
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700587
Nick Piggin224abf92006-01-06 00:11:11 -0800588static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589{
Nick Piggin92be2e32006-01-06 00:10:57 -0800590 if (unlikely(page_mapcount(page) |
591 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700592 (atomic_read(&page->_count) != 0) |
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -0700593 (page->flags & PAGE_FLAGS_CHECK_AT_FREE) |
594 (mem_cgroup_bad_page_check(page)))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800595 bad_page(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800596 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800597 }
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800598 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
599 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
600 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601}
602
603/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700604 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700606 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 *
608 * If the zone was previously in an "all pages pinned" state then look to
609 * see if this freeing clears that state.
610 *
611 * And clear the zone's pages_scanned counter, to hold off the "all pages are
612 * pinned" detection logic.
613 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700614static void free_pcppages_bulk(struct zone *zone, int count,
615 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616{
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700617 int migratetype = 0;
Mel Gormana6f9edd2009-09-21 17:03:20 -0700618 int batch_free = 0;
Mel Gorman72853e22010-09-09 16:38:16 -0700619 int to_free = count;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700620
Nick Pigginc54ad302006-01-06 00:10:56 -0800621 spin_lock(&zone->lock);
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -0800622 zone->all_unreclaimable = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700624
Mel Gorman72853e22010-09-09 16:38:16 -0700625 while (to_free) {
Nick Piggin48db57f2006-01-08 01:00:42 -0800626 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700627 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -0800628
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700629 /*
Mel Gormana6f9edd2009-09-21 17:03:20 -0700630 * Remove pages from lists in a round-robin fashion. A
631 * batch_free count is maintained that is incremented when an
632 * empty list is encountered. This is so more pages are freed
633 * off fuller lists instead of spinning excessively around empty
634 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700635 */
636 do {
Mel Gormana6f9edd2009-09-21 17:03:20 -0700637 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700638 if (++migratetype == MIGRATE_PCPTYPES)
639 migratetype = 0;
640 list = &pcp->lists[migratetype];
641 } while (list_empty(list));
642
Namhyung Kim1d168712011-03-22 16:32:45 -0700643 /* This is the only non-empty list. Free them all. */
644 if (batch_free == MIGRATE_PCPTYPES)
645 batch_free = to_free;
646
Mel Gormana6f9edd2009-09-21 17:03:20 -0700647 do {
648 page = list_entry(list->prev, struct page, lru);
649 /* must delete as __free_one_page list manipulates */
650 list_del(&page->lru);
Hugh Dickinsa7016232010-01-29 17:46:34 +0000651 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
652 __free_one_page(page, zone, 0, page_private(page));
653 trace_mm_page_pcpu_drain(page, 0, page_private(page));
Mel Gorman72853e22010-09-09 16:38:16 -0700654 } while (--to_free && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 }
Mel Gorman72853e22010-09-09 16:38:16 -0700656 __mod_zone_page_state(zone, NR_FREE_PAGES, count);
Nick Pigginc54ad302006-01-06 00:10:56 -0800657 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658}
659
Mel Gormaned0ae212009-06-16 15:32:07 -0700660static void free_one_page(struct zone *zone, struct page *page, int order,
661 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800662{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700663 spin_lock(&zone->lock);
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -0800664 zone->all_unreclaimable = 0;
Christoph Lameter006d22d2006-09-25 23:31:48 -0700665 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700666
Mel Gormaned0ae212009-06-16 15:32:07 -0700667 __free_one_page(page, zone, order, migratetype);
Mel Gorman72853e22010-09-09 16:38:16 -0700668 __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700669 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800670}
671
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700672static bool free_pages_prepare(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 int i;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800675 int bad = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676
Li Hongf6503162010-03-05 13:41:52 -0800677 trace_mm_page_free_direct(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100678 kmemcheck_free_shadow(page, order);
679
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -0800680 if (PageAnon(page))
681 page->mapping = NULL;
682 for (i = 0; i < (1 << order); i++)
683 bad += free_pages_check(page + i);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800684 if (bad)
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700685 return false;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800686
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700687 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -0700688 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700689 debug_check_no_obj_freed(page_address(page),
690 PAGE_SIZE << order);
691 }
Nick Piggindafb1362006-10-11 01:21:30 -0700692 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800693 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700694
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700695 return true;
696}
697
698static void __free_pages_ok(struct page *page, unsigned int order)
699{
700 unsigned long flags;
701 int wasMlocked = __TestClearPageMlocked(page);
702
703 if (!free_pages_prepare(page, order))
704 return;
705
Nick Pigginc54ad302006-01-06 00:10:56 -0800706 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +0200707 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -0700708 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700709 __count_vm_events(PGFREE, 1 << order);
Mel Gormaned0ae212009-06-16 15:32:07 -0700710 free_one_page(page_zone(page), page, order,
711 get_pageblock_migratetype(page));
Nick Pigginc54ad302006-01-06 00:10:56 -0800712 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713}
714
David Howellsa226f6c2006-01-06 00:11:08 -0800715/*
716 * permit the bootmem allocator to evade page validation on high-order frees
717 */
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700718void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800719{
720 if (order == 0) {
721 __ClearPageReserved(page);
722 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800723 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800724 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800725 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800726 int loop;
727
Nick Piggin545b1ea2006-03-22 00:08:07 -0800728 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800729 for (loop = 0; loop < BITS_PER_LONG; loop++) {
730 struct page *p = &page[loop];
731
Nick Piggin545b1ea2006-03-22 00:08:07 -0800732 if (loop + 1 < BITS_PER_LONG)
733 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800734 __ClearPageReserved(p);
735 set_page_count(p, 0);
736 }
737
Nick Piggin7835e982006-03-22 00:08:40 -0800738 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800739 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800740 }
741}
742
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743
744/*
745 * The order of subdivision here is critical for the IO subsystem.
746 * Please do not alter this order without good reasons and regression
747 * testing. Specifically, as large blocks of memory are subdivided,
748 * the order in which smaller blocks are delivered depends on the order
749 * they're subdivided in this function. This is the primary factor
750 * influencing the order in which pages are delivered to the IO
751 * subsystem according to empirical testing, and this is also justified
752 * by considering the behavior of a buddy system containing a single
753 * large block of memory acted on by a series of small allocations.
754 * This behavior is a critical factor in sglist merging's success.
755 *
756 * -- wli
757 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800758static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700759 int low, int high, struct free_area *area,
760 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761{
762 unsigned long size = 1 << high;
763
764 while (high > low) {
765 area--;
766 high--;
767 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700768 VM_BUG_ON(bad_range(zone, &page[size]));
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700769 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 area->nr_free++;
771 set_page_order(&page[size], high);
772 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773}
774
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775/*
776 * This page is about to be returned from the page allocator
777 */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200778static inline int check_new_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779{
Nick Piggin92be2e32006-01-06 00:10:57 -0800780 if (unlikely(page_mapcount(page) |
781 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700782 (atomic_read(&page->_count) != 0) |
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -0700783 (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
784 (mem_cgroup_bad_page_check(page)))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800785 bad_page(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800786 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800787 }
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200788 return 0;
789}
790
791static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
792{
793 int i;
794
795 for (i = 0; i < (1 << order); i++) {
796 struct page *p = page + i;
797 if (unlikely(check_new_page(p)))
798 return 1;
799 }
Hugh Dickins689bceb2005-11-21 21:32:20 -0800800
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700801 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800802 set_page_refcounted(page);
Nick Piggincc102502006-12-06 20:32:00 -0800803
804 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800806
807 if (gfp_flags & __GFP_ZERO)
808 prep_zero_page(page, order, gfp_flags);
809
810 if (order && (gfp_flags & __GFP_COMP))
811 prep_compound_page(page, order);
812
Hugh Dickins689bceb2005-11-21 21:32:20 -0800813 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814}
815
Mel Gorman56fd56b2007-10-16 01:25:58 -0700816/*
817 * Go through the free lists for the given migratetype and remove
818 * the smallest available page from the freelists
819 */
Mel Gorman728ec982009-06-16 15:32:04 -0700820static inline
821struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -0700822 int migratetype)
823{
824 unsigned int current_order;
825 struct free_area * area;
826 struct page *page;
827
828 /* Find a page of the appropriate size in the preferred list */
829 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
830 area = &(zone->free_area[current_order]);
831 if (list_empty(&area->free_list[migratetype]))
832 continue;
833
834 page = list_entry(area->free_list[migratetype].next,
835 struct page, lru);
836 list_del(&page->lru);
837 rmv_page_order(page);
838 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -0700839 expand(zone, page, order, current_order, area, migratetype);
840 return page;
841 }
842
843 return NULL;
844}
845
846
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700847/*
848 * This array describes the order lists are fallen back to when
849 * the free lists for the desirable migrate type are depleted
850 */
851static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
Mel Gorman64c5e132007-10-16 01:25:59 -0700852 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
853 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
854 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
855 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700856};
857
Mel Gormanc361be52007-10-16 01:25:51 -0700858/*
859 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -0700860 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -0700861 * boundary. If alignment is required, use move_freepages_block()
862 */
Adrian Bunkb69a7282008-07-23 21:28:12 -0700863static int move_freepages(struct zone *zone,
864 struct page *start_page, struct page *end_page,
865 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700866{
867 struct page *page;
868 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -0700869 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -0700870
871#ifndef CONFIG_HOLES_IN_ZONE
872 /*
873 * page_zone is not safe to call in this context when
874 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
875 * anyway as we check zone boundaries in move_freepages_block().
876 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -0700877 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -0700878 */
879 BUG_ON(page_zone(start_page) != page_zone(end_page));
880#endif
881
882 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -0700883 /* Make sure we are not inadvertently changing nodes */
884 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
885
Mel Gormanc361be52007-10-16 01:25:51 -0700886 if (!pfn_valid_within(page_to_pfn(page))) {
887 page++;
888 continue;
889 }
890
891 if (!PageBuddy(page)) {
892 page++;
893 continue;
894 }
895
896 order = page_order(page);
Kirill A. Shutemov84be48d2011-03-22 16:33:41 -0700897 list_move(&page->lru,
898 &zone->free_area[order].free_list[migratetype]);
Mel Gormanc361be52007-10-16 01:25:51 -0700899 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -0700900 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -0700901 }
902
Mel Gormand1003132007-10-16 01:26:00 -0700903 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -0700904}
905
Adrian Bunkb69a7282008-07-23 21:28:12 -0700906static int move_freepages_block(struct zone *zone, struct page *page,
907 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700908{
909 unsigned long start_pfn, end_pfn;
910 struct page *start_page, *end_page;
911
912 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -0700913 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -0700914 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -0700915 end_page = start_page + pageblock_nr_pages - 1;
916 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -0700917
918 /* Do not cross zone boundaries */
919 if (start_pfn < zone->zone_start_pfn)
920 start_page = page;
921 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
922 return 0;
923
924 return move_freepages(zone, start_page, end_page, migratetype);
925}
926
Mel Gorman2f66a682009-09-21 17:02:31 -0700927static void change_pageblock_range(struct page *pageblock_page,
928 int start_order, int migratetype)
929{
930 int nr_pageblocks = 1 << (start_order - pageblock_order);
931
932 while (nr_pageblocks--) {
933 set_pageblock_migratetype(pageblock_page, migratetype);
934 pageblock_page += pageblock_nr_pages;
935 }
936}
937
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700938/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -0700939static inline struct page *
940__rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700941{
942 struct free_area * area;
943 int current_order;
944 struct page *page;
945 int migratetype, i;
946
947 /* Find the largest possible block of pages in the other list */
948 for (current_order = MAX_ORDER-1; current_order >= order;
949 --current_order) {
950 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
951 migratetype = fallbacks[start_migratetype][i];
952
Mel Gorman56fd56b2007-10-16 01:25:58 -0700953 /* MIGRATE_RESERVE handled later if necessary */
954 if (migratetype == MIGRATE_RESERVE)
955 continue;
Mel Gormane0104872007-10-16 01:25:53 -0700956
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700957 area = &(zone->free_area[current_order]);
958 if (list_empty(&area->free_list[migratetype]))
959 continue;
960
961 page = list_entry(area->free_list[migratetype].next,
962 struct page, lru);
963 area->nr_free--;
964
965 /*
Mel Gormanc361be52007-10-16 01:25:51 -0700966 * If breaking a large block of pages, move all free
Mel Gorman46dafbc2007-10-16 01:25:55 -0700967 * pages to the preferred allocation list. If falling
968 * back for a reclaimable kernel allocation, be more
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300969 * aggressive about taking ownership of free pages
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700970 */
Mel Gormand9c23402007-10-16 01:26:01 -0700971 if (unlikely(current_order >= (pageblock_order >> 1)) ||
Mel Gormandd5d2412009-09-05 11:17:11 -0700972 start_migratetype == MIGRATE_RECLAIMABLE ||
973 page_group_by_mobility_disabled) {
Mel Gorman46dafbc2007-10-16 01:25:55 -0700974 unsigned long pages;
975 pages = move_freepages_block(zone, page,
976 start_migratetype);
977
978 /* Claim the whole block if over half of it is free */
Mel Gormandd5d2412009-09-05 11:17:11 -0700979 if (pages >= (1 << (pageblock_order-1)) ||
980 page_group_by_mobility_disabled)
Mel Gorman46dafbc2007-10-16 01:25:55 -0700981 set_pageblock_migratetype(page,
982 start_migratetype);
983
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700984 migratetype = start_migratetype;
Mel Gormanc361be52007-10-16 01:25:51 -0700985 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700986
987 /* Remove the page from the freelists */
988 list_del(&page->lru);
989 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700990
Mel Gorman2f66a682009-09-21 17:02:31 -0700991 /* Take ownership for orders >= pageblock_order */
992 if (current_order >= pageblock_order)
993 change_pageblock_range(page, current_order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700994 start_migratetype);
995
996 expand(zone, page, order, current_order, area, migratetype);
Mel Gormane0fff1b2009-09-21 17:02:42 -0700997
998 trace_mm_page_alloc_extfrag(page, order, current_order,
999 start_migratetype, migratetype);
1000
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001001 return page;
1002 }
1003 }
1004
Mel Gorman728ec982009-06-16 15:32:04 -07001005 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001006}
1007
Mel Gorman56fd56b2007-10-16 01:25:58 -07001008/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 * Do the hard work of removing an element from the buddy allocator.
1010 * Call me with the zone->lock already held.
1011 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001012static struct page *__rmqueue(struct zone *zone, unsigned int order,
1013 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 struct page *page;
1016
Mel Gorman728ec982009-06-16 15:32:04 -07001017retry_reserve:
Mel Gorman56fd56b2007-10-16 01:25:58 -07001018 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019
Mel Gorman728ec982009-06-16 15:32:04 -07001020 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07001021 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001022
Mel Gorman728ec982009-06-16 15:32:04 -07001023 /*
1024 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1025 * is used because __rmqueue_smallest is an inline function
1026 * and we want just one call site
1027 */
1028 if (!page) {
1029 migratetype = MIGRATE_RESERVE;
1030 goto retry_reserve;
1031 }
1032 }
1033
Mel Gorman0d3d0622009-09-21 17:02:44 -07001034 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001035 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036}
1037
1038/*
1039 * Obtain a specified number of elements from the buddy allocator, all under
1040 * a single hold of the lock, for efficiency. Add them to the supplied list.
1041 * Returns the number of new pages which were placed at *list.
1042 */
1043static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001044 unsigned long count, struct list_head *list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001045 int migratetype, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048
Nick Pigginc54ad302006-01-06 00:10:56 -08001049 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001051 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001052 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08001054
1055 /*
1056 * Split buddy pages returned by expand() are received here
1057 * in physical page order. The page is added to the callers and
1058 * list and the list head then moves forward. From the callers
1059 * perspective, the linked list is ordered by page number in
1060 * some conditions. This is useful for IO devices that can
1061 * merge IO requests if the physical pages are ordered
1062 * properly.
1063 */
Mel Gormane084b2d2009-07-29 15:02:04 -07001064 if (likely(cold == 0))
1065 list_add(&page->lru, list);
1066 else
1067 list_add_tail(&page->lru, list);
Mel Gorman535131e2007-10-16 01:25:49 -07001068 set_page_private(page, migratetype);
Mel Gorman81eabcb2007-12-17 16:20:05 -08001069 list = &page->lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 }
Mel Gormanf2260e62009-06-16 15:32:13 -07001071 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08001072 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001073 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074}
1075
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001076#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001077/*
Christoph Lameter4037d452007-05-09 02:35:14 -07001078 * Called from the vmstat counter updater to drain pagesets of this
1079 * currently executing processor on remote nodes after they have
1080 * expired.
1081 *
Christoph Lameter879336c2006-03-22 00:09:08 -08001082 * Note that this function must be called with the thread pinned to
1083 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001084 */
Christoph Lameter4037d452007-05-09 02:35:14 -07001085void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001086{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001087 unsigned long flags;
Christoph Lameter4037d452007-05-09 02:35:14 -07001088 int to_drain;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001089
Christoph Lameter4037d452007-05-09 02:35:14 -07001090 local_irq_save(flags);
1091 if (pcp->count >= pcp->batch)
1092 to_drain = pcp->batch;
1093 else
1094 to_drain = pcp->count;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001095 free_pcppages_bulk(zone, to_drain, pcp);
Christoph Lameter4037d452007-05-09 02:35:14 -07001096 pcp->count -= to_drain;
1097 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001098}
1099#endif
1100
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001101/*
1102 * Drain pages of the indicated processor.
1103 *
1104 * The processor must either be the current processor and the
1105 * thread pinned to the current processor or a processor that
1106 * is not online.
1107 */
1108static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109{
Nick Pigginc54ad302006-01-06 00:10:56 -08001110 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07001113 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 struct per_cpu_pageset *pset;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001115 struct per_cpu_pages *pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001117 local_irq_save(flags);
1118 pset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001120 pcp = &pset->pcp;
David Rientjes2ff754f2011-01-25 15:07:23 -08001121 if (pcp->count) {
1122 free_pcppages_bulk(zone, pcp->count, pcp);
1123 pcp->count = 0;
1124 }
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001125 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 }
1127}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001129/*
1130 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1131 */
1132void drain_local_pages(void *arg)
1133{
1134 drain_pages(smp_processor_id());
1135}
1136
1137/*
1138 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
1139 */
1140void drain_all_pages(void)
1141{
Jens Axboe15c8b6c2008-05-09 09:39:44 +02001142 on_each_cpu(drain_local_pages, NULL, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001143}
1144
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02001145#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
1147void mark_free_pages(struct zone *zone)
1148{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001149 unsigned long pfn, max_zone_pfn;
1150 unsigned long flags;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001151 int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 struct list_head *curr;
1153
1154 if (!zone->spanned_pages)
1155 return;
1156
1157 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001158
1159 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1160 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1161 if (pfn_valid(pfn)) {
1162 struct page *page = pfn_to_page(pfn);
1163
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001164 if (!swsusp_page_is_forbidden(page))
1165 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001166 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001168 for_each_migratetype_order(order, t) {
1169 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001170 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001172 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1173 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001174 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001175 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001176 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 spin_unlock_irqrestore(&zone->lock, flags);
1178}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001179#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180
1181/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 * Free a 0-order page
Li Hongfc916682010-03-05 13:41:54 -08001183 * cold == 1 ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 */
Li Hongfc916682010-03-05 13:41:54 -08001185void free_hot_cold_page(struct page *page, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186{
1187 struct zone *zone = page_zone(page);
1188 struct per_cpu_pages *pcp;
1189 unsigned long flags;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001190 int migratetype;
Johannes Weiner451ea252009-09-21 17:01:48 -07001191 int wasMlocked = __TestClearPageMlocked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001193 if (!free_pages_prepare(page, 0))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001194 return;
1195
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001196 migratetype = get_pageblock_migratetype(page);
1197 set_page_private(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +02001199 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -07001200 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001201 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001202
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001203 /*
1204 * We only track unmovable, reclaimable and movable on pcp lists.
1205 * Free ISOLATE pages back to the allocator because they are being
1206 * offlined but treat RESERVE as movable pages so we can get those
1207 * areas back if necessary. Otherwise, we may have to free
1208 * excessively into the page allocator
1209 */
1210 if (migratetype >= MIGRATE_PCPTYPES) {
1211 if (unlikely(migratetype == MIGRATE_ISOLATE)) {
1212 free_one_page(zone, page, 0, migratetype);
1213 goto out;
1214 }
1215 migratetype = MIGRATE_MOVABLE;
1216 }
1217
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001218 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001219 if (cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001220 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001221 else
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001222 list_add(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001224 if (pcp->count >= pcp->high) {
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001225 free_pcppages_bulk(zone, pcp->batch, pcp);
Nick Piggin48db57f2006-01-08 01:00:42 -08001226 pcp->count -= pcp->batch;
1227 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001228
1229out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231}
1232
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001233/*
1234 * split_page takes a non-compound higher-order page, and splits it into
1235 * n (1<<order) sub-pages: page[0..n]
1236 * Each sub-page must be freed individually.
1237 *
1238 * Note: this is probably too low level an operation for use in drivers.
1239 * Please consult with lkml before using this in your driver.
1240 */
1241void split_page(struct page *page, unsigned int order)
1242{
1243 int i;
1244
Nick Piggin725d7042006-09-25 23:30:55 -07001245 VM_BUG_ON(PageCompound(page));
1246 VM_BUG_ON(!page_count(page));
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001247
1248#ifdef CONFIG_KMEMCHECK
1249 /*
1250 * Split shadow pages too, because free(page[0]) would
1251 * otherwise free the whole shadow.
1252 */
1253 if (kmemcheck_page_is_tracked(page))
1254 split_page(virt_to_page(page[0].shadow), order);
1255#endif
1256
Nick Piggin7835e982006-03-22 00:08:40 -08001257 for (i = 1; i < (1 << order); i++)
1258 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001259}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001260
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261/*
Mel Gorman748446b2010-05-24 14:32:27 -07001262 * Similar to split_page except the page is already free. As this is only
1263 * being used for migration, the migratetype of the block also changes.
1264 * As this is called with interrupts disabled, the caller is responsible
1265 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1266 * are enabled.
1267 *
1268 * Note: this is probably too low level an operation for use in drivers.
1269 * Please consult with lkml before using this in your driver.
1270 */
1271int split_free_page(struct page *page)
1272{
1273 unsigned int order;
1274 unsigned long watermark;
1275 struct zone *zone;
1276
1277 BUG_ON(!PageBuddy(page));
1278
1279 zone = page_zone(page);
1280 order = page_order(page);
1281
1282 /* Obey watermarks as if the page was being allocated */
1283 watermark = low_wmark_pages(zone) + (1 << order);
1284 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1285 return 0;
1286
1287 /* Remove page from free list */
1288 list_del(&page->lru);
1289 zone->free_area[order].nr_free--;
1290 rmv_page_order(page);
1291 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1UL << order));
1292
1293 /* Split into individual pages */
1294 set_page_refcounted(page);
1295 split_page(page, order);
1296
1297 if (order >= pageblock_order - 1) {
1298 struct page *endpage = page + (1 << order) - 1;
1299 for (; page < endpage; page += pageblock_nr_pages)
1300 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1301 }
1302
1303 return 1 << order;
1304}
1305
1306/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1308 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1309 * or two.
1310 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07001311static inline
1312struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001313 struct zone *zone, int order, gfp_t gfp_flags,
1314 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315{
1316 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001317 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 int cold = !!(gfp_flags & __GFP_COLD);
1319
Hugh Dickins689bceb2005-11-21 21:32:20 -08001320again:
Nick Piggin48db57f2006-01-08 01:00:42 -08001321 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001323 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 local_irq_save(flags);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001326 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1327 list = &pcp->lists[migratetype];
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001328 if (list_empty(list)) {
Mel Gorman535131e2007-10-16 01:25:49 -07001329 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001330 pcp->batch, list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001331 migratetype, cold);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001332 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07001333 goto failed;
Mel Gorman535131e2007-10-16 01:25:49 -07001334 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001335
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001336 if (cold)
1337 page = list_entry(list->prev, struct page, lru);
1338 else
1339 page = list_entry(list->next, struct page, lru);
1340
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001341 list_del(&page->lru);
1342 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001343 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07001344 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1345 /*
1346 * __GFP_NOFAIL is not to be used in new code.
1347 *
1348 * All __GFP_NOFAIL callers should be fixed so that they
1349 * properly detect and handle allocation failures.
1350 *
1351 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07001352 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07001353 * __GFP_NOFAIL.
1354 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07001355 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07001356 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001358 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001359 spin_unlock(&zone->lock);
1360 if (!page)
1361 goto failed;
KOSAKI Motohiro6ccf80e2010-01-15 17:01:18 -08001362 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 }
1364
Christoph Lameterf8891e52006-06-30 01:55:45 -07001365 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Andi Kleen78afd562011-03-22 16:33:12 -07001366 zone_statistics(preferred_zone, zone, gfp_flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001367 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368
Nick Piggin725d7042006-09-25 23:30:55 -07001369 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -08001370 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001371 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001373
1374failed:
1375 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001376 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377}
1378
Mel Gorman41858962009-06-16 15:32:12 -07001379/* The ALLOC_WMARK bits are used as an index to zone->watermark */
1380#define ALLOC_WMARK_MIN WMARK_MIN
1381#define ALLOC_WMARK_LOW WMARK_LOW
1382#define ALLOC_WMARK_HIGH WMARK_HIGH
1383#define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1384
1385/* Mask to get the watermark bits */
1386#define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1387
Nick Piggin31488902005-11-28 13:44:03 -08001388#define ALLOC_HARDER 0x10 /* try to alloc harder */
1389#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1390#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001391
Akinobu Mita933e3122006-12-08 02:39:45 -08001392#ifdef CONFIG_FAIL_PAGE_ALLOC
1393
1394static struct fail_page_alloc_attr {
1395 struct fault_attr attr;
1396
1397 u32 ignore_gfp_highmem;
1398 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001399 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001400
1401#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1402
1403 struct dentry *ignore_gfp_highmem_file;
1404 struct dentry *ignore_gfp_wait_file;
Akinobu Mita54114992007-07-15 23:40:23 -07001405 struct dentry *min_order_file;
Akinobu Mita933e3122006-12-08 02:39:45 -08001406
1407#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1408
1409} fail_page_alloc = {
1410 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001411 .ignore_gfp_wait = 1,
1412 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001413 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001414};
1415
1416static int __init setup_fail_page_alloc(char *str)
1417{
1418 return setup_fault_attr(&fail_page_alloc.attr, str);
1419}
1420__setup("fail_page_alloc=", setup_fail_page_alloc);
1421
1422static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1423{
Akinobu Mita54114992007-07-15 23:40:23 -07001424 if (order < fail_page_alloc.min_order)
1425 return 0;
Akinobu Mita933e3122006-12-08 02:39:45 -08001426 if (gfp_mask & __GFP_NOFAIL)
1427 return 0;
1428 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1429 return 0;
1430 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1431 return 0;
1432
1433 return should_fail(&fail_page_alloc.attr, 1 << order);
1434}
1435
1436#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1437
1438static int __init fail_page_alloc_debugfs(void)
1439{
1440 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1441 struct dentry *dir;
1442 int err;
1443
1444 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1445 "fail_page_alloc");
1446 if (err)
1447 return err;
1448 dir = fail_page_alloc.attr.dentries.dir;
1449
1450 fail_page_alloc.ignore_gfp_wait_file =
1451 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1452 &fail_page_alloc.ignore_gfp_wait);
1453
1454 fail_page_alloc.ignore_gfp_highmem_file =
1455 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1456 &fail_page_alloc.ignore_gfp_highmem);
Akinobu Mita54114992007-07-15 23:40:23 -07001457 fail_page_alloc.min_order_file =
1458 debugfs_create_u32("min-order", mode, dir,
1459 &fail_page_alloc.min_order);
Akinobu Mita933e3122006-12-08 02:39:45 -08001460
1461 if (!fail_page_alloc.ignore_gfp_wait_file ||
Akinobu Mita54114992007-07-15 23:40:23 -07001462 !fail_page_alloc.ignore_gfp_highmem_file ||
1463 !fail_page_alloc.min_order_file) {
Akinobu Mita933e3122006-12-08 02:39:45 -08001464 err = -ENOMEM;
1465 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1466 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
Akinobu Mita54114992007-07-15 23:40:23 -07001467 debugfs_remove(fail_page_alloc.min_order_file);
Akinobu Mita933e3122006-12-08 02:39:45 -08001468 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1469 }
1470
1471 return err;
1472}
1473
1474late_initcall(fail_page_alloc_debugfs);
1475
1476#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1477
1478#else /* CONFIG_FAIL_PAGE_ALLOC */
1479
1480static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1481{
1482 return 0;
1483}
1484
1485#endif /* CONFIG_FAIL_PAGE_ALLOC */
1486
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487/*
Mel Gorman88f5acf2011-01-13 15:45:41 -08001488 * Return true if free pages are above 'mark'. This takes into account the order
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 * of the allocation.
1490 */
Mel Gorman88f5acf2011-01-13 15:45:41 -08001491static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1492 int classzone_idx, int alloc_flags, long free_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493{
Jack Cheung9f41da82011-11-28 16:41:28 -08001494 /* free_pages may go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001495 long min = mark;
Jack Cheung9f41da82011-11-28 16:41:28 -08001496 long lowmem_reserve = z->lowmem_reserve[classzone_idx];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 int o;
1498
Mel Gorman88f5acf2011-01-13 15:45:41 -08001499 free_pages -= (1 << order) + 1;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001500 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001502 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 min -= min / 4;
1504
Jack Cheung9f41da82011-11-28 16:41:28 -08001505 if (free_pages <= min + lowmem_reserve)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001506 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 for (o = 0; o < order; o++) {
1508 /* At the next order, this order's pages become unavailable */
1509 free_pages -= z->free_area[o].nr_free << o;
1510
1511 /* Require fewer higher order pages to be free */
Arve Hjønnevåg83de7312009-02-17 14:51:02 -08001512 min >>= min_free_order_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513
1514 if (free_pages <= min)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001515 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 }
Mel Gorman88f5acf2011-01-13 15:45:41 -08001517 return true;
1518}
1519
1520bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1521 int classzone_idx, int alloc_flags)
1522{
1523 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1524 zone_page_state(z, NR_FREE_PAGES));
1525}
1526
1527bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
1528 int classzone_idx, int alloc_flags)
1529{
1530 long free_pages = zone_page_state(z, NR_FREE_PAGES);
1531
1532 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1533 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1534
1535 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1536 free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537}
1538
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001539#ifdef CONFIG_NUMA
1540/*
1541 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1542 * skip over zones that are not allowed by the cpuset, or that have
1543 * been recently (in last second) found to be nearly full. See further
1544 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001545 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001546 *
1547 * If the zonelist cache is present in the passed in zonelist, then
1548 * returns a pointer to the allowed node mask (either the current
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001549 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001550 *
1551 * If the zonelist cache is not available for this zonelist, does
1552 * nothing and returns NULL.
1553 *
1554 * If the fullzones BITMAP in the zonelist cache is stale (more than
1555 * a second since last zap'd) then we zap it out (clear its bits.)
1556 *
1557 * We hold off even calling zlc_setup, until after we've checked the
1558 * first zone in the zonelist, on the theory that most allocations will
1559 * be satisfied from that first zone, so best to examine that zone as
1560 * quickly as we can.
1561 */
1562static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1563{
1564 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1565 nodemask_t *allowednodes; /* zonelist_cache approximation */
1566
1567 zlc = zonelist->zlcache_ptr;
1568 if (!zlc)
1569 return NULL;
1570
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001571 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001572 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1573 zlc->last_full_zap = jiffies;
1574 }
1575
1576 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1577 &cpuset_current_mems_allowed :
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001578 &node_states[N_HIGH_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001579 return allowednodes;
1580}
1581
1582/*
1583 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1584 * if it is worth looking at further for free memory:
1585 * 1) Check that the zone isn't thought to be full (doesn't have its
1586 * bit set in the zonelist_cache fullzones BITMAP).
1587 * 2) Check that the zones node (obtained from the zonelist_cache
1588 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1589 * Return true (non-zero) if zone is worth looking at further, or
1590 * else return false (zero) if it is not.
1591 *
1592 * This check -ignores- the distinction between various watermarks,
1593 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1594 * found to be full for any variation of these watermarks, it will
1595 * be considered full for up to one second by all requests, unless
1596 * we are so low on memory on all allowed nodes that we are forced
1597 * into the second scan of the zonelist.
1598 *
1599 * In the second scan we ignore this zonelist cache and exactly
1600 * apply the watermarks to all zones, even it is slower to do so.
1601 * We are low on memory in the second scan, and should leave no stone
1602 * unturned looking for a free page.
1603 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001604static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001605 nodemask_t *allowednodes)
1606{
1607 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1608 int i; /* index of *z in zonelist zones */
1609 int n; /* node that zone *z is on */
1610
1611 zlc = zonelist->zlcache_ptr;
1612 if (!zlc)
1613 return 1;
1614
Mel Gormandd1a2392008-04-28 02:12:17 -07001615 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001616 n = zlc->z_to_n[i];
1617
1618 /* This zone is worth trying if it is allowed but not full */
1619 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1620}
1621
1622/*
1623 * Given 'z' scanning a zonelist, set the corresponding bit in
1624 * zlc->fullzones, so that subsequent attempts to allocate a page
1625 * from that zone don't waste time re-examining it.
1626 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001627static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001628{
1629 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1630 int i; /* index of *z in zonelist zones */
1631
1632 zlc = zonelist->zlcache_ptr;
1633 if (!zlc)
1634 return;
1635
Mel Gormandd1a2392008-04-28 02:12:17 -07001636 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001637
1638 set_bit(i, zlc->fullzones);
1639}
1640
Mel Gorman66d52cb2011-07-25 17:12:30 -07001641/*
1642 * clear all zones full, called after direct reclaim makes progress so that
1643 * a zone that was recently full is not skipped over for up to a second
1644 */
1645static void zlc_clear_zones_full(struct zonelist *zonelist)
1646{
1647 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1648
1649 zlc = zonelist->zlcache_ptr;
1650 if (!zlc)
1651 return;
1652
1653 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1654}
1655
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001656#else /* CONFIG_NUMA */
1657
1658static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1659{
1660 return NULL;
1661}
1662
Mel Gormandd1a2392008-04-28 02:12:17 -07001663static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001664 nodemask_t *allowednodes)
1665{
1666 return 1;
1667}
1668
Mel Gormandd1a2392008-04-28 02:12:17 -07001669static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001670{
1671}
Mel Gorman66d52cb2011-07-25 17:12:30 -07001672
1673static void zlc_clear_zones_full(struct zonelist *zonelist)
1674{
1675}
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001676#endif /* CONFIG_NUMA */
1677
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001678/*
Paul Jackson0798e512006-12-06 20:31:38 -08001679 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001680 * a page.
1681 */
1682static struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001683get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
Mel Gorman5117f452009-06-16 15:31:59 -07001684 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
Mel Gorman3dd28262009-06-16 15:32:00 -07001685 struct zone *preferred_zone, int migratetype)
Martin Hicks753ee722005-06-21 17:14:41 -07001686{
Mel Gormandd1a2392008-04-28 02:12:17 -07001687 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001688 struct page *page = NULL;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001689 int classzone_idx;
Mel Gorman5117f452009-06-16 15:31:59 -07001690 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001691 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1692 int zlc_active = 0; /* set if using zonelist_cache */
1693 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gorman54a6eb52008-04-28 02:12:16 -07001694
Mel Gorman19770b32008-04-28 02:12:18 -07001695 classzone_idx = zone_idx(preferred_zone);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001696zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001697 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001698 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001699 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1700 */
Mel Gorman19770b32008-04-28 02:12:18 -07001701 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1702 high_zoneidx, nodemask) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001703 if (NUMA_BUILD && zlc_active &&
1704 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1705 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001706 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001707 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Mel Gorman42274b52011-07-25 17:12:29 -07001708 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001709
Mel Gorman41858962009-06-16 15:32:12 -07001710 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001711 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001712 unsigned long mark;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001713 int ret;
1714
Mel Gorman41858962009-06-16 15:32:12 -07001715 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
Mel Gormanfa5e0842009-06-16 15:33:22 -07001716 if (zone_watermark_ok(zone, order, mark,
1717 classzone_idx, alloc_flags))
1718 goto try_this_zone;
1719
Mel Gorman42274b52011-07-25 17:12:29 -07001720 if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) {
1721 /*
1722 * we do zlc_setup if there are multiple nodes
1723 * and before considering the first zone allowed
1724 * by the cpuset.
1725 */
1726 allowednodes = zlc_setup(zonelist, alloc_flags);
1727 zlc_active = 1;
1728 did_zlc_setup = 1;
1729 }
1730
Mel Gormanfa5e0842009-06-16 15:33:22 -07001731 if (zone_reclaim_mode == 0)
1732 goto this_zone_full;
1733
Mel Gorman42274b52011-07-25 17:12:29 -07001734 /*
1735 * As we may have just activated ZLC, check if the first
1736 * eligible zone has failed zone_reclaim recently.
1737 */
1738 if (NUMA_BUILD && zlc_active &&
1739 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1740 continue;
1741
Mel Gormanfa5e0842009-06-16 15:33:22 -07001742 ret = zone_reclaim(zone, gfp_mask, order);
1743 switch (ret) {
1744 case ZONE_RECLAIM_NOSCAN:
1745 /* did not scan */
Mel Gorman42274b52011-07-25 17:12:29 -07001746 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001747 case ZONE_RECLAIM_FULL:
1748 /* scanned but unreclaimable */
Mel Gorman42274b52011-07-25 17:12:29 -07001749 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001750 default:
1751 /* did we reclaim enough */
1752 if (!zone_watermark_ok(zone, order, mark,
1753 classzone_idx, alloc_flags))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001754 goto this_zone_full;
Paul Jackson0798e512006-12-06 20:31:38 -08001755 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001756 }
1757
Mel Gormanfa5e0842009-06-16 15:33:22 -07001758try_this_zone:
Mel Gorman3dd28262009-06-16 15:32:00 -07001759 page = buffered_rmqueue(preferred_zone, zone, order,
1760 gfp_mask, migratetype);
Paul Jackson0798e512006-12-06 20:31:38 -08001761 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001762 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001763this_zone_full:
1764 if (NUMA_BUILD)
1765 zlc_mark_zone_full(zonelist, z);
Mel Gorman54a6eb52008-04-28 02:12:16 -07001766 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001767
1768 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1769 /* Disable zlc cache for second zonelist scan */
1770 zlc_active = 0;
1771 goto zonelist_scan;
1772 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001773 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07001774}
1775
David Rientjes29423e72011-03-22 16:30:47 -07001776/*
1777 * Large machines with many possible nodes should not always dump per-node
1778 * meminfo in irq context.
1779 */
1780static inline bool should_suppress_show_mem(void)
1781{
1782 bool ret = false;
1783
1784#if NODES_SHIFT > 8
1785 ret = in_interrupt();
1786#endif
1787 return ret;
1788}
1789
Dave Hansena238ab52011-05-24 17:12:16 -07001790static DEFINE_RATELIMIT_STATE(nopage_rs,
1791 DEFAULT_RATELIMIT_INTERVAL,
1792 DEFAULT_RATELIMIT_BURST);
1793
1794void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
1795{
1796 va_list args;
1797 unsigned int filter = SHOW_MEM_FILTER_NODES;
1798
1799 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs))
1800 return;
1801
1802 /*
1803 * This documents exceptions given to allocations in certain
1804 * contexts that are allowed to allocate outside current's set
1805 * of allowed nodes.
1806 */
1807 if (!(gfp_mask & __GFP_NOMEMALLOC))
1808 if (test_thread_flag(TIF_MEMDIE) ||
1809 (current->flags & (PF_MEMALLOC | PF_EXITING)))
1810 filter &= ~SHOW_MEM_FILTER_NODES;
1811 if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
1812 filter &= ~SHOW_MEM_FILTER_NODES;
1813
1814 if (fmt) {
1815 printk(KERN_WARNING);
1816 va_start(args, fmt);
1817 vprintk(fmt, args);
1818 va_end(args);
1819 }
1820
1821 pr_warning("%s: page allocation failure: order:%d, mode:0x%x\n",
1822 current->comm, order, gfp_mask);
1823
1824 dump_stack();
1825 if (!should_suppress_show_mem())
1826 show_mem(filter);
1827}
1828
Mel Gorman11e33f62009-06-16 15:31:57 -07001829static inline int
1830should_alloc_retry(gfp_t gfp_mask, unsigned int order,
1831 unsigned long pages_reclaimed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832{
Mel Gorman11e33f62009-06-16 15:31:57 -07001833 /* Do not loop if specifically requested */
1834 if (gfp_mask & __GFP_NORETRY)
1835 return 0;
1836
1837 /*
1838 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1839 * means __GFP_NOFAIL, but that may not be true in other
1840 * implementations.
1841 */
1842 if (order <= PAGE_ALLOC_COSTLY_ORDER)
1843 return 1;
1844
1845 /*
1846 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1847 * specified, then we retry until we no longer reclaim any pages
1848 * (above), or we've reclaimed an order of pages at least as
1849 * large as the allocation's order. In both cases, if the
1850 * allocation still fails, we stop retrying.
1851 */
1852 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
1853 return 1;
1854
1855 /*
1856 * Don't let big-order allocations loop unless the caller
1857 * explicitly requests that.
1858 */
1859 if (gfp_mask & __GFP_NOFAIL)
1860 return 1;
1861
1862 return 0;
1863}
1864
1865static inline struct page *
1866__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1867 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001868 nodemask_t *nodemask, struct zone *preferred_zone,
1869 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001870{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872
Mel Gorman11e33f62009-06-16 15:31:57 -07001873 /* Acquire the OOM killer lock for the zones in zonelist */
Minchan Kimff321fe2010-08-09 17:18:57 -07001874 if (!try_set_zonelist_oom(zonelist, gfp_mask)) {
Mel Gorman11e33f62009-06-16 15:31:57 -07001875 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 return NULL;
1877 }
Jens Axboe6b1de912005-11-17 21:35:02 +01001878
Mel Gorman11e33f62009-06-16 15:31:57 -07001879 /*
1880 * Go through the zonelist yet one more time, keep very high watermark
1881 * here, this is only to catch a parallel oom killing, we must fail if
1882 * we're still under heavy pressure.
1883 */
1884 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1885 order, zonelist, high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001886 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07001887 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001888 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07001889 goto out;
1890
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001891 if (!(gfp_mask & __GFP_NOFAIL)) {
1892 /* The OOM killer will not help higher order allocs */
1893 if (order > PAGE_ALLOC_COSTLY_ORDER)
1894 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07001895 /* The OOM killer does not needlessly kill tasks for lowmem */
1896 if (high_zoneidx < ZONE_NORMAL)
1897 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001898 /*
1899 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
1900 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
1901 * The caller should handle page allocation failure by itself if
1902 * it specifies __GFP_THISNODE.
1903 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
1904 */
1905 if (gfp_mask & __GFP_THISNODE)
1906 goto out;
1907 }
Mel Gorman11e33f62009-06-16 15:31:57 -07001908 /* Exhausted what can be done so it's blamo time */
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001909 out_of_memory(zonelist, gfp_mask, order, nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07001910
1911out:
1912 clear_zonelist_oom(zonelist, gfp_mask);
1913 return page;
1914}
1915
Mel Gorman56de7262010-05-24 14:32:30 -07001916#ifdef CONFIG_COMPACTION
1917/* Try memory compaction for high-order allocations before reclaim */
1918static struct page *
1919__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
1920 struct zonelist *zonelist, enum zone_type high_zoneidx,
1921 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman77f1fe62011-01-13 15:45:57 -08001922 int migratetype, unsigned long *did_some_progress,
1923 bool sync_migration)
Mel Gorman56de7262010-05-24 14:32:30 -07001924{
1925 struct page *page;
1926
Mel Gorman4f92e252010-05-24 14:32:32 -07001927 if (!order || compaction_deferred(preferred_zone))
Mel Gorman56de7262010-05-24 14:32:30 -07001928 return NULL;
1929
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001930 current->flags |= PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07001931 *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
Mel Gorman77f1fe62011-01-13 15:45:57 -08001932 nodemask, sync_migration);
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001933 current->flags &= ~PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07001934 if (*did_some_progress != COMPACT_SKIPPED) {
1935
1936 /* Page migration frees to the PCP lists but we want merging */
1937 drain_pages(get_cpu());
1938 put_cpu();
1939
1940 page = get_page_from_freelist(gfp_mask, nodemask,
1941 order, zonelist, high_zoneidx,
1942 alloc_flags, preferred_zone,
1943 migratetype);
1944 if (page) {
Mel Gorman4f92e252010-05-24 14:32:32 -07001945 preferred_zone->compact_considered = 0;
1946 preferred_zone->compact_defer_shift = 0;
Mel Gorman56de7262010-05-24 14:32:30 -07001947 count_vm_event(COMPACTSUCCESS);
1948 return page;
1949 }
1950
1951 /*
1952 * It's bad if compaction run occurs and fails.
1953 * The most likely reason is that pages exist,
1954 * but not enough to satisfy watermarks.
1955 */
1956 count_vm_event(COMPACTFAIL);
Mel Gorman4f92e252010-05-24 14:32:32 -07001957 defer_compaction(preferred_zone);
Mel Gorman56de7262010-05-24 14:32:30 -07001958
1959 cond_resched();
1960 }
1961
1962 return NULL;
1963}
1964#else
1965static inline struct page *
1966__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
1967 struct zonelist *zonelist, enum zone_type high_zoneidx,
1968 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman77f1fe62011-01-13 15:45:57 -08001969 int migratetype, unsigned long *did_some_progress,
1970 bool sync_migration)
Mel Gorman56de7262010-05-24 14:32:30 -07001971{
1972 return NULL;
1973}
1974#endif /* CONFIG_COMPACTION */
1975
Mel Gorman11e33f62009-06-16 15:31:57 -07001976/* The really slow allocator path where we enter direct reclaim */
1977static inline struct page *
1978__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
1979 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001980 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001981 int migratetype, unsigned long *did_some_progress)
Mel Gorman11e33f62009-06-16 15:31:57 -07001982{
1983 struct page *page = NULL;
1984 struct reclaim_state reclaim_state;
Mel Gorman9ee493c2010-09-09 16:38:18 -07001985 bool drained = false;
Mel Gorman11e33f62009-06-16 15:31:57 -07001986
1987 cond_resched();
1988
1989 /* We now go into synchronous reclaim */
1990 cpuset_memory_pressure_bump();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001991 current->flags |= PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07001992 lockdep_set_current_reclaim_state(gfp_mask);
1993 reclaim_state.reclaimed_slab = 0;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001994 current->reclaim_state = &reclaim_state;
Mel Gorman11e33f62009-06-16 15:31:57 -07001995
1996 *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
1997
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001998 current->reclaim_state = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07001999 lockdep_clear_current_reclaim_state();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002000 current->flags &= ~PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002001
2002 cond_resched();
2003
Mel Gorman9ee493c2010-09-09 16:38:18 -07002004 if (unlikely(!(*did_some_progress)))
2005 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002006
Mel Gorman66d52cb2011-07-25 17:12:30 -07002007 /* After successful reclaim, reconsider all zones for allocation */
2008 if (NUMA_BUILD)
2009 zlc_clear_zones_full(zonelist);
2010
Mel Gorman9ee493c2010-09-09 16:38:18 -07002011retry:
2012 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002013 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002014 alloc_flags, preferred_zone,
2015 migratetype);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002016
2017 /*
2018 * If an allocation failed after direct reclaim, it could be because
2019 * pages are pinned on the per-cpu lists. Drain them and try again
2020 */
2021 if (!page && !drained) {
2022 drain_all_pages();
2023 drained = true;
2024 goto retry;
2025 }
2026
Mel Gorman11e33f62009-06-16 15:31:57 -07002027 return page;
2028}
2029
Mel Gorman11e33f62009-06-16 15:31:57 -07002030/*
2031 * This is called in the allocator slow-path if the allocation request is of
2032 * sufficient urgency to ignore watermarks and take other desperate measures
2033 */
2034static inline struct page *
2035__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
2036 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002037 nodemask_t *nodemask, struct zone *preferred_zone,
2038 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002039{
2040 struct page *page;
2041
2042 do {
2043 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002044 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
Mel Gorman3dd28262009-06-16 15:32:00 -07002045 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002046
2047 if (!page && gfp_mask & __GFP_NOFAIL)
Mel Gorman0e093d992010-10-26 14:21:45 -07002048 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07002049 } while (!page && (gfp_mask & __GFP_NOFAIL));
2050
2051 return page;
2052}
2053
2054static inline
2055void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
Mel Gorman99504742011-01-13 15:46:20 -08002056 enum zone_type high_zoneidx,
2057 enum zone_type classzone_idx)
Mel Gorman11e33f62009-06-16 15:31:57 -07002058{
2059 struct zoneref *z;
2060 struct zone *zone;
2061
2062 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
Mel Gorman99504742011-01-13 15:46:20 -08002063 wakeup_kswapd(zone, order, classzone_idx);
Mel Gorman11e33f62009-06-16 15:31:57 -07002064}
2065
Peter Zijlstra341ce062009-06-16 15:32:02 -07002066static inline int
2067gfp_to_alloc_flags(gfp_t gfp_mask)
2068{
Peter Zijlstra341ce062009-06-16 15:32:02 -07002069 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
2070 const gfp_t wait = gfp_mask & __GFP_WAIT;
2071
Mel Gormana56f57f2009-06-16 15:32:02 -07002072 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07002073 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07002074
Peter Zijlstra341ce062009-06-16 15:32:02 -07002075 /*
2076 * The caller may dip into page reserves a bit more if the caller
2077 * cannot run direct reclaim, or if the caller has realtime scheduling
2078 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2079 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
2080 */
Namhyung Kime6223a32010-10-26 14:21:59 -07002081 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07002082
2083 if (!wait) {
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002084 /*
2085 * Not worth trying to allocate harder for
2086 * __GFP_NOMEMALLOC even if it can't schedule.
2087 */
2088 if (!(gfp_mask & __GFP_NOMEMALLOC))
2089 alloc_flags |= ALLOC_HARDER;
Peter Zijlstra341ce062009-06-16 15:32:02 -07002090 /*
2091 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
2092 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
2093 */
2094 alloc_flags &= ~ALLOC_CPUSET;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002095 } else if (unlikely(rt_task(current)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07002096 alloc_flags |= ALLOC_HARDER;
2097
2098 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2099 if (!in_interrupt() &&
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002100 ((current->flags & PF_MEMALLOC) ||
Peter Zijlstra341ce062009-06-16 15:32:02 -07002101 unlikely(test_thread_flag(TIF_MEMDIE))))
2102 alloc_flags |= ALLOC_NO_WATERMARKS;
2103 }
2104
2105 return alloc_flags;
2106}
2107
Mel Gorman11e33f62009-06-16 15:31:57 -07002108static inline struct page *
2109__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
2110 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002111 nodemask_t *nodemask, struct zone *preferred_zone,
2112 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002113{
2114 const gfp_t wait = gfp_mask & __GFP_WAIT;
2115 struct page *page = NULL;
2116 int alloc_flags;
2117 unsigned long pages_reclaimed = 0;
2118 unsigned long did_some_progress;
Mel Gorman77f1fe62011-01-13 15:45:57 -08002119 bool sync_migration = false;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002120
Christoph Lameter952f3b52006-12-06 20:33:26 -08002121 /*
Mel Gorman72807a72009-06-16 15:32:18 -07002122 * In the slowpath, we sanity check order to avoid ever trying to
2123 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2124 * be using allocators in order of preference for an area that is
2125 * too large.
2126 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07002127 if (order >= MAX_ORDER) {
2128 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07002129 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07002130 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002131
Christoph Lameter952f3b52006-12-06 20:33:26 -08002132 /*
2133 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2134 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2135 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2136 * using a larger set of nodes after it has established that the
2137 * allowed per node queues are empty and that nodes are
2138 * over allocated.
2139 */
2140 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
2141 goto nopage;
2142
Mel Gormancc4a6852009-11-11 14:26:14 -08002143restart:
Andrea Arcangeli32dba982011-01-13 15:46:49 -08002144 if (!(gfp_mask & __GFP_NO_KSWAPD))
2145 wake_all_kswapd(order, zonelist, high_zoneidx,
Mel Gorman99504742011-01-13 15:46:20 -08002146 zone_idx(preferred_zone));
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002147
Paul Jackson9bf22292005-09-06 15:18:12 -07002148 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002149 * OK, we're below the kswapd watermark and have kicked background
2150 * reclaim. Now things get more complex, so set up alloc_flags according
2151 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07002152 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002153 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154
David Rientjesf33261d2011-01-25 15:07:20 -08002155 /*
2156 * Find the true preferred zone if the allocation is unconstrained by
2157 * cpusets.
2158 */
2159 if (!(alloc_flags & ALLOC_CPUSET) && !nodemask)
2160 first_zones_zonelist(zonelist, high_zoneidx, NULL,
2161 &preferred_zone);
2162
Andrew Barrycfa54a02011-05-24 17:12:52 -07002163rebalance:
Peter Zijlstra341ce062009-06-16 15:32:02 -07002164 /* This is the last chance, in general, before the goto nopage. */
Mel Gorman19770b32008-04-28 02:12:18 -07002165 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
Peter Zijlstra341ce062009-06-16 15:32:02 -07002166 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
2167 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002168 if (page)
2169 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170
Mel Gorman11e33f62009-06-16 15:31:57 -07002171 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002172 if (alloc_flags & ALLOC_NO_WATERMARKS) {
2173 page = __alloc_pages_high_priority(gfp_mask, order,
2174 zonelist, high_zoneidx, nodemask,
2175 preferred_zone, migratetype);
2176 if (page)
2177 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178 }
2179
2180 /* Atomic allocations - we can't balance anything */
2181 if (!wait)
2182 goto nopage;
2183
Peter Zijlstra341ce062009-06-16 15:32:02 -07002184 /* Avoid recursion of direct reclaim */
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002185 if (current->flags & PF_MEMALLOC)
Peter Zijlstra341ce062009-06-16 15:32:02 -07002186 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187
David Rientjes6583bb62009-07-29 15:02:06 -07002188 /* Avoid allocations with no watermarks from looping endlessly */
2189 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2190 goto nopage;
2191
Mel Gorman77f1fe62011-01-13 15:45:57 -08002192 /*
2193 * Try direct compaction. The first pass is asynchronous. Subsequent
2194 * attempts after direct reclaim are synchronous
2195 */
Mel Gorman56de7262010-05-24 14:32:30 -07002196 page = __alloc_pages_direct_compact(gfp_mask, order,
2197 zonelist, high_zoneidx,
2198 nodemask,
2199 alloc_flags, preferred_zone,
Mel Gorman77f1fe62011-01-13 15:45:57 -08002200 migratetype, &did_some_progress,
2201 sync_migration);
Mel Gorman56de7262010-05-24 14:32:30 -07002202 if (page)
2203 goto got_pg;
Andrea Arcangelic6a140b2011-05-24 17:11:38 -07002204 sync_migration = true;
Mel Gorman56de7262010-05-24 14:32:30 -07002205
Mel Gorman11e33f62009-06-16 15:31:57 -07002206 /* Try direct reclaim and then allocating */
2207 page = __alloc_pages_direct_reclaim(gfp_mask, order,
2208 zonelist, high_zoneidx,
2209 nodemask,
Mel Gorman5117f452009-06-16 15:31:59 -07002210 alloc_flags, preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07002211 migratetype, &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07002212 if (page)
2213 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215 /*
Mel Gorman11e33f62009-06-16 15:31:57 -07002216 * If we failed to make any progress reclaiming, then we are
2217 * running out of options and have to consider going OOM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 */
Mel Gorman11e33f62009-06-16 15:31:57 -07002219 if (!did_some_progress) {
2220 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -07002221 if (oom_killer_disabled)
2222 goto nopage;
Mel Gorman11e33f62009-06-16 15:31:57 -07002223 page = __alloc_pages_may_oom(gfp_mask, order,
2224 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002225 nodemask, preferred_zone,
2226 migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002227 if (page)
2228 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229
David Rientjes03668b32010-08-09 17:18:54 -07002230 if (!(gfp_mask & __GFP_NOFAIL)) {
2231 /*
2232 * The oom killer is not called for high-order
2233 * allocations that may fail, so if no progress
2234 * is being made, there are no other options and
2235 * retrying is unlikely to help.
2236 */
2237 if (order > PAGE_ALLOC_COSTLY_ORDER)
2238 goto nopage;
2239 /*
2240 * The oom killer is not called for lowmem
2241 * allocations to prevent needlessly killing
2242 * innocent tasks.
2243 */
2244 if (high_zoneidx < ZONE_NORMAL)
2245 goto nopage;
2246 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 goto restart;
2249 }
Mel Gorman2f53cb72011-10-24 16:33:42 -07002250
2251 /*
2252 * Suspend converts GFP_KERNEL to __GFP_WAIT which can
2253 * prevent reclaim making forward progress without
2254 * invoking OOM. Bail if we are suspending
2255 */
2256 if (pm_suspending())
2257 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 }
2259
Mel Gorman11e33f62009-06-16 15:31:57 -07002260 /* Check if we should retry the allocation */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07002261 pages_reclaimed += did_some_progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07002262 if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
2263 /* Wait for some write requests to complete then retry */
Mel Gorman0e093d992010-10-26 14:21:45 -07002264 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265 goto rebalance;
Mel Gorman3e7d3442011-01-13 15:45:56 -08002266 } else {
2267 /*
2268 * High-order allocations do not necessarily loop after
2269 * direct reclaim and reclaim/compaction depends on compaction
2270 * being called after reclaim so call directly if necessary
2271 */
2272 page = __alloc_pages_direct_compact(gfp_mask, order,
2273 zonelist, high_zoneidx,
2274 nodemask,
2275 alloc_flags, preferred_zone,
Mel Gorman77f1fe62011-01-13 15:45:57 -08002276 migratetype, &did_some_progress,
2277 sync_migration);
Mel Gorman3e7d3442011-01-13 15:45:56 -08002278 if (page)
2279 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280 }
2281
2282nopage:
Dave Hansena238ab52011-05-24 17:12:16 -07002283 warn_alloc_failed(gfp_mask, order, NULL);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002284 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285got_pg:
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002286 if (kmemcheck_enabled)
2287 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288 return page;
Mel Gorman11e33f62009-06-16 15:31:57 -07002289
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290}
Mel Gorman11e33f62009-06-16 15:31:57 -07002291
2292/*
2293 * This is the 'heart' of the zoned buddy allocator.
2294 */
2295struct page *
2296__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2297 struct zonelist *zonelist, nodemask_t *nodemask)
2298{
2299 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Mel Gorman5117f452009-06-16 15:31:59 -07002300 struct zone *preferred_zone;
Mel Gorman11e33f62009-06-16 15:31:57 -07002301 struct page *page;
Mel Gorman3dd28262009-06-16 15:32:00 -07002302 int migratetype = allocflags_to_migratetype(gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002303
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10002304 gfp_mask &= gfp_allowed_mask;
2305
Mel Gorman11e33f62009-06-16 15:31:57 -07002306 lockdep_trace_alloc(gfp_mask);
2307
2308 might_sleep_if(gfp_mask & __GFP_WAIT);
2309
2310 if (should_fail_alloc_page(gfp_mask, order))
2311 return NULL;
2312
2313 /*
2314 * Check the zones suitable for the gfp_mask contain at least one
2315 * valid zone. It's possible to have an empty zonelist as a result
2316 * of GFP_THISNODE and a memoryless node
2317 */
2318 if (unlikely(!zonelist->_zonerefs->zone))
2319 return NULL;
2320
Miao Xiec0ff7452010-05-24 14:32:08 -07002321 get_mems_allowed();
Mel Gorman5117f452009-06-16 15:31:59 -07002322 /* The preferred zone is used for statistics later */
David Rientjesf33261d2011-01-25 15:07:20 -08002323 first_zones_zonelist(zonelist, high_zoneidx,
2324 nodemask ? : &cpuset_current_mems_allowed,
2325 &preferred_zone);
Miao Xiec0ff7452010-05-24 14:32:08 -07002326 if (!preferred_zone) {
2327 put_mems_allowed();
Mel Gorman5117f452009-06-16 15:31:59 -07002328 return NULL;
Miao Xiec0ff7452010-05-24 14:32:08 -07002329 }
Mel Gorman5117f452009-06-16 15:31:59 -07002330
2331 /* First allocation attempt */
Mel Gorman11e33f62009-06-16 15:31:57 -07002332 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002333 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07002334 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002335 if (unlikely(!page))
2336 page = __alloc_pages_slowpath(gfp_mask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002337 zonelist, high_zoneidx, nodemask,
Mel Gorman3dd28262009-06-16 15:32:00 -07002338 preferred_zone, migratetype);
Miao Xiec0ff7452010-05-24 14:32:08 -07002339 put_mems_allowed();
Mel Gorman11e33f62009-06-16 15:31:57 -07002340
Mel Gorman4b4f2782009-09-21 17:02:41 -07002341 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002342 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343}
Mel Gormand2391712009-06-16 15:31:52 -07002344EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345
2346/*
2347 * Common helper functions.
2348 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08002349unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350{
Akinobu Mita945a1112009-09-21 17:01:47 -07002351 struct page *page;
2352
2353 /*
2354 * __get_free_pages() returns a 32-bit address, which cannot represent
2355 * a highmem page
2356 */
2357 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2358
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 page = alloc_pages(gfp_mask, order);
2360 if (!page)
2361 return 0;
2362 return (unsigned long) page_address(page);
2363}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364EXPORT_SYMBOL(__get_free_pages);
2365
Harvey Harrison920c7a52008-02-04 22:29:26 -08002366unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367{
Akinobu Mita945a1112009-09-21 17:01:47 -07002368 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370EXPORT_SYMBOL(get_zeroed_page);
2371
2372void __pagevec_free(struct pagevec *pvec)
2373{
2374 int i = pagevec_count(pvec);
2375
Mel Gorman4b4f2782009-09-21 17:02:41 -07002376 while (--i >= 0) {
2377 trace_mm_pagevec_free(pvec->pages[i], pvec->cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378 free_hot_cold_page(pvec->pages[i], pvec->cold);
Mel Gorman4b4f2782009-09-21 17:02:41 -07002379 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380}
2381
Harvey Harrison920c7a52008-02-04 22:29:26 -08002382void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383{
Nick Pigginb5810032005-10-29 18:16:12 -07002384 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 if (order == 0)
Li Hongfc916682010-03-05 13:41:54 -08002386 free_hot_cold_page(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387 else
2388 __free_pages_ok(page, order);
2389 }
2390}
2391
2392EXPORT_SYMBOL(__free_pages);
2393
Harvey Harrison920c7a52008-02-04 22:29:26 -08002394void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395{
2396 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07002397 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398 __free_pages(virt_to_page((void *)addr), order);
2399 }
2400}
2401
2402EXPORT_SYMBOL(free_pages);
2403
Andi Kleenee85c2e2011-05-11 15:13:34 -07002404static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
2405{
2406 if (addr) {
2407 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2408 unsigned long used = addr + PAGE_ALIGN(size);
2409
2410 split_page(virt_to_page((void *)addr), order);
2411 while (used < alloc_end) {
2412 free_page(used);
2413 used += PAGE_SIZE;
2414 }
2415 }
2416 return (void *)addr;
2417}
2418
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002419/**
2420 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2421 * @size: the number of bytes to allocate
2422 * @gfp_mask: GFP flags for the allocation
2423 *
2424 * This function is similar to alloc_pages(), except that it allocates the
2425 * minimum number of pages to satisfy the request. alloc_pages() can only
2426 * allocate memory in power-of-two pages.
2427 *
2428 * This function is also limited by MAX_ORDER.
2429 *
2430 * Memory allocated by this function must be released by free_pages_exact().
2431 */
2432void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2433{
2434 unsigned int order = get_order(size);
2435 unsigned long addr;
2436
2437 addr = __get_free_pages(gfp_mask, order);
Andi Kleenee85c2e2011-05-11 15:13:34 -07002438 return make_alloc_exact(addr, order, size);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002439}
2440EXPORT_SYMBOL(alloc_pages_exact);
2441
2442/**
Andi Kleenee85c2e2011-05-11 15:13:34 -07002443 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2444 * pages on a node.
Randy Dunlapb5e6ab52011-05-16 13:16:54 -07002445 * @nid: the preferred node ID where memory should be allocated
Andi Kleenee85c2e2011-05-11 15:13:34 -07002446 * @size: the number of bytes to allocate
2447 * @gfp_mask: GFP flags for the allocation
2448 *
2449 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2450 * back.
2451 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2452 * but is not exact.
2453 */
2454void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
2455{
2456 unsigned order = get_order(size);
2457 struct page *p = alloc_pages_node(nid, gfp_mask, order);
2458 if (!p)
2459 return NULL;
2460 return make_alloc_exact((unsigned long)page_address(p), order, size);
2461}
2462EXPORT_SYMBOL(alloc_pages_exact_nid);
2463
2464/**
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002465 * free_pages_exact - release memory allocated via alloc_pages_exact()
2466 * @virt: the value returned by alloc_pages_exact.
2467 * @size: size of allocation, same value as passed to alloc_pages_exact().
2468 *
2469 * Release the memory allocated by a previous call to alloc_pages_exact.
2470 */
2471void free_pages_exact(void *virt, size_t size)
2472{
2473 unsigned long addr = (unsigned long)virt;
2474 unsigned long end = addr + PAGE_ALIGN(size);
2475
2476 while (addr < end) {
2477 free_page(addr);
2478 addr += PAGE_SIZE;
2479 }
2480}
2481EXPORT_SYMBOL(free_pages_exact);
2482
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483static unsigned int nr_free_zone_pages(int offset)
2484{
Mel Gormandd1a2392008-04-28 02:12:17 -07002485 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002486 struct zone *zone;
2487
Martin J. Blighe310fd42005-07-29 22:59:18 -07002488 /* Just pick one node, since fallback list is circular */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489 unsigned int sum = 0;
2490
Mel Gorman0e884602008-04-28 02:12:14 -07002491 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492
Mel Gorman54a6eb52008-04-28 02:12:16 -07002493 for_each_zone_zonelist(zone, z, zonelist, offset) {
Martin J. Blighe310fd42005-07-29 22:59:18 -07002494 unsigned long size = zone->present_pages;
Mel Gorman41858962009-06-16 15:32:12 -07002495 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07002496 if (size > high)
2497 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498 }
2499
2500 return sum;
2501}
2502
2503/*
2504 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2505 */
2506unsigned int nr_free_buffer_pages(void)
2507{
Al Viroaf4ca452005-10-21 02:55:38 -04002508 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509}
Meelap Shahc2f1a552007-07-17 04:04:39 -07002510EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511
2512/*
2513 * Amount of free RAM allocatable within all zones
2514 */
2515unsigned int nr_free_pagecache_pages(void)
2516{
Mel Gorman2a1e2742007-07-17 04:03:12 -07002517 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002519
2520static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521{
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002522 if (NUMA_BUILD)
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08002523 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526void si_meminfo(struct sysinfo *val)
2527{
2528 val->totalram = totalram_pages;
2529 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002530 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 val->totalhigh = totalhigh_pages;
2533 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534 val->mem_unit = PAGE_SIZE;
2535}
2536
2537EXPORT_SYMBOL(si_meminfo);
2538
2539#ifdef CONFIG_NUMA
2540void si_meminfo_node(struct sysinfo *val, int nid)
2541{
2542 pg_data_t *pgdat = NODE_DATA(nid);
2543
2544 val->totalram = pgdat->node_present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002545 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002546#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002548 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2549 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002550#else
2551 val->totalhigh = 0;
2552 val->freehigh = 0;
2553#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554 val->mem_unit = PAGE_SIZE;
2555}
2556#endif
2557
David Rientjesddd588b2011-03-22 16:30:46 -07002558/*
David Rientjes7bf02ea2011-05-24 17:11:16 -07002559 * Determine whether the node should be displayed or not, depending on whether
2560 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
David Rientjesddd588b2011-03-22 16:30:46 -07002561 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07002562bool skip_free_areas_node(unsigned int flags, int nid)
David Rientjesddd588b2011-03-22 16:30:46 -07002563{
2564 bool ret = false;
2565
2566 if (!(flags & SHOW_MEM_FILTER_NODES))
2567 goto out;
2568
2569 get_mems_allowed();
David Rientjes7bf02ea2011-05-24 17:11:16 -07002570 ret = !node_isset(nid, cpuset_current_mems_allowed);
David Rientjesddd588b2011-03-22 16:30:46 -07002571 put_mems_allowed();
2572out:
2573 return ret;
2574}
2575
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576#define K(x) ((x) << (PAGE_SHIFT-10))
2577
2578/*
2579 * Show free area list (used inside shift_scroll-lock stuff)
2580 * We also calculate the percentage fragmentation. We do this by counting the
2581 * memory on each free list with the exception of the first item on the list.
David Rientjesddd588b2011-03-22 16:30:46 -07002582 * Suppresses nodes that are not allowed by current's cpuset if
2583 * SHOW_MEM_FILTER_NODES is passed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07002585void show_free_areas(unsigned int filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586{
Jes Sorensenc7241912006-09-27 01:50:05 -07002587 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588 struct zone *zone;
2589
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002590 for_each_populated_zone(zone) {
David Rientjes7bf02ea2011-05-24 17:11:16 -07002591 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07002592 continue;
Jes Sorensenc7241912006-09-27 01:50:05 -07002593 show_node(zone);
2594 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595
Dave Jones6b482c62005-11-10 15:45:56 -05002596 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597 struct per_cpu_pageset *pageset;
2598
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002599 pageset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002601 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2602 cpu, pageset->pcp.high,
2603 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604 }
2605 }
2606
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002607 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2608 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002609 " unevictable:%lu"
Andrew Mortonb76146e2009-10-26 16:49:52 -07002610 " dirty:%lu writeback:%lu unstable:%lu\n"
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07002611 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002612 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002613 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002614 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002615 global_page_state(NR_ISOLATED_ANON),
2616 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002617 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002618 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002619 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07002620 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07002621 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07002622 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08002623 global_page_state(NR_FREE_PAGES),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07002624 global_page_state(NR_SLAB_RECLAIMABLE),
2625 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07002626 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002627 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08002628 global_page_state(NR_PAGETABLE),
2629 global_page_state(NR_BOUNCE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002631 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632 int i;
2633
David Rientjes7bf02ea2011-05-24 17:11:16 -07002634 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07002635 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636 show_node(zone);
2637 printk("%s"
2638 " free:%lukB"
2639 " min:%lukB"
2640 " low:%lukB"
2641 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002642 " active_anon:%lukB"
2643 " inactive_anon:%lukB"
2644 " active_file:%lukB"
2645 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002646 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002647 " isolated(anon):%lukB"
2648 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649 " present:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002650 " mlocked:%lukB"
2651 " dirty:%lukB"
2652 " writeback:%lukB"
2653 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002654 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002655 " slab_reclaimable:%lukB"
2656 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07002657 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002658 " pagetables:%lukB"
2659 " unstable:%lukB"
2660 " bounce:%lukB"
2661 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 " pages_scanned:%lu"
2663 " all_unreclaimable? %s"
2664 "\n",
2665 zone->name,
Mel Gorman88f5acf2011-01-13 15:45:41 -08002666 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07002667 K(min_wmark_pages(zone)),
2668 K(low_wmark_pages(zone)),
2669 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002670 K(zone_page_state(zone, NR_ACTIVE_ANON)),
2671 K(zone_page_state(zone, NR_INACTIVE_ANON)),
2672 K(zone_page_state(zone, NR_ACTIVE_FILE)),
2673 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002674 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002675 K(zone_page_state(zone, NR_ISOLATED_ANON)),
2676 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 K(zone->present_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002678 K(zone_page_state(zone, NR_MLOCK)),
2679 K(zone_page_state(zone, NR_FILE_DIRTY)),
2680 K(zone_page_state(zone, NR_WRITEBACK)),
2681 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002682 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002683 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
2684 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07002685 zone_page_state(zone, NR_KERNEL_STACK) *
2686 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002687 K(zone_page_state(zone, NR_PAGETABLE)),
2688 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
2689 K(zone_page_state(zone, NR_BOUNCE)),
2690 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691 zone->pages_scanned,
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -08002692 (zone->all_unreclaimable ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693 );
2694 printk("lowmem_reserve[]:");
2695 for (i = 0; i < MAX_NR_ZONES; i++)
2696 printk(" %lu", zone->lowmem_reserve[i]);
2697 printk("\n");
2698 }
2699
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002700 for_each_populated_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002701 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702
David Rientjes7bf02ea2011-05-24 17:11:16 -07002703 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07002704 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705 show_node(zone);
2706 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707
2708 spin_lock_irqsave(&zone->lock, flags);
2709 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002710 nr[order] = zone->free_area[order].nr_free;
2711 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712 }
2713 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002714 for (order = 0; order < MAX_ORDER; order++)
2715 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716 printk("= %lukB\n", K(total));
2717 }
2718
Larry Woodmane6f36022008-02-04 22:29:30 -08002719 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2720
Linus Torvalds1da177e2005-04-16 15:20:36 -07002721 show_swap_cache_info();
2722}
2723
Mel Gorman19770b32008-04-28 02:12:18 -07002724static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2725{
2726 zoneref->zone = zone;
2727 zoneref->zone_idx = zone_idx(zone);
2728}
2729
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730/*
2731 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08002732 *
2733 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002734 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002735static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2736 int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737{
Christoph Lameter1a932052006-01-06 00:11:16 -08002738 struct zone *zone;
2739
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002740 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002741 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08002742
2743 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002744 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08002745 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08002746 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002747 zoneref_set_zone(zone,
2748 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08002749 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08002751
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002752 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08002753 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002754}
2755
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002756
2757/*
2758 * zonelist_order:
2759 * 0 = automatic detection of better ordering.
2760 * 1 = order by ([node] distance, -zonetype)
2761 * 2 = order by (-zonetype, [node] distance)
2762 *
2763 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2764 * the same zonelist. So only NUMA can configure this param.
2765 */
2766#define ZONELIST_ORDER_DEFAULT 0
2767#define ZONELIST_ORDER_NODE 1
2768#define ZONELIST_ORDER_ZONE 2
2769
2770/* zonelist order in the kernel.
2771 * set_zonelist_order() will set this to NODE or ZONE.
2772 */
2773static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2774static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2775
2776
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002778/* The value user specified ....changed by config */
2779static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2780/* string for sysctl */
2781#define NUMA_ZONELIST_ORDER_LEN 16
2782char numa_zonelist_order[16] = "default";
2783
2784/*
2785 * interface for configure zonelist ordering.
2786 * command line option "numa_zonelist_order"
2787 * = "[dD]efault - default, automatic configuration.
2788 * = "[nN]ode - order by node locality, then by zone within node
2789 * = "[zZ]one - order by zone, then by locality within zone
2790 */
2791
2792static int __parse_numa_zonelist_order(char *s)
2793{
2794 if (*s == 'd' || *s == 'D') {
2795 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2796 } else if (*s == 'n' || *s == 'N') {
2797 user_zonelist_order = ZONELIST_ORDER_NODE;
2798 } else if (*s == 'z' || *s == 'Z') {
2799 user_zonelist_order = ZONELIST_ORDER_ZONE;
2800 } else {
2801 printk(KERN_WARNING
2802 "Ignoring invalid numa_zonelist_order value: "
2803 "%s\n", s);
2804 return -EINVAL;
2805 }
2806 return 0;
2807}
2808
2809static __init int setup_numa_zonelist_order(char *s)
2810{
Volodymyr G. Lukiianykecb256f2011-01-13 15:46:26 -08002811 int ret;
2812
2813 if (!s)
2814 return 0;
2815
2816 ret = __parse_numa_zonelist_order(s);
2817 if (ret == 0)
2818 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
2819
2820 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002821}
2822early_param("numa_zonelist_order", setup_numa_zonelist_order);
2823
2824/*
2825 * sysctl handler for numa_zonelist_order
2826 */
2827int numa_zonelist_order_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07002828 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002829 loff_t *ppos)
2830{
2831 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2832 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01002833 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002834
Andi Kleen443c6f12009-12-23 21:00:47 +01002835 mutex_lock(&zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002836 if (write)
Andi Kleen443c6f12009-12-23 21:00:47 +01002837 strcpy(saved_string, (char*)table->data);
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07002838 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002839 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01002840 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002841 if (write) {
2842 int oldval = user_zonelist_order;
2843 if (__parse_numa_zonelist_order((char*)table->data)) {
2844 /*
2845 * bogus value. restore saved string
2846 */
2847 strncpy((char*)table->data, saved_string,
2848 NUMA_ZONELIST_ORDER_LEN);
2849 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07002850 } else if (oldval != user_zonelist_order) {
2851 mutex_lock(&zonelists_mutex);
Haicheng Li1f522502010-05-24 14:32:51 -07002852 build_all_zonelists(NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07002853 mutex_unlock(&zonelists_mutex);
2854 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002855 }
Andi Kleen443c6f12009-12-23 21:00:47 +01002856out:
2857 mutex_unlock(&zl_order_mutex);
2858 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002859}
2860
2861
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002862#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002863static int node_load[MAX_NUMNODES];
2864
Linus Torvalds1da177e2005-04-16 15:20:36 -07002865/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07002866 * 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 -07002867 * @node: node whose fallback list we're appending
2868 * @used_node_mask: nodemask_t of already used nodes
2869 *
2870 * We use a number of factors to determine which is the next node that should
2871 * appear on a given node's fallback list. The node should not have appeared
2872 * already in @node's fallback list, and it should be the next closest node
2873 * according to the distance array (which contains arbitrary distance values
2874 * from each node to each node in the system), and should also prefer nodes
2875 * with no CPUs, since presumably they'll have very little allocation pressure
2876 * on them otherwise.
2877 * It returns -1 if no node is found.
2878 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002879static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002881 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882 int min_val = INT_MAX;
2883 int best_node = -1;
Rusty Russella70f7302009-03-13 14:49:46 +10302884 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002886 /* Use the local node if we haven't already */
2887 if (!node_isset(node, *used_node_mask)) {
2888 node_set(node, *used_node_mask);
2889 return node;
2890 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002892 for_each_node_state(n, N_HIGH_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002893
2894 /* Don't want a node to appear more than once */
2895 if (node_isset(n, *used_node_mask))
2896 continue;
2897
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898 /* Use the distance array to find the distance */
2899 val = node_distance(node, n);
2900
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002901 /* Penalize nodes under us ("prefer the next node") */
2902 val += (n < node);
2903
Linus Torvalds1da177e2005-04-16 15:20:36 -07002904 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10302905 tmp = cpumask_of_node(n);
2906 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907 val += PENALTY_FOR_NODE_WITH_CPUS;
2908
2909 /* Slight preference for less loaded node */
2910 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2911 val += node_load[n];
2912
2913 if (val < min_val) {
2914 min_val = val;
2915 best_node = n;
2916 }
2917 }
2918
2919 if (best_node >= 0)
2920 node_set(best_node, *used_node_mask);
2921
2922 return best_node;
2923}
2924
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002925
2926/*
2927 * Build zonelists ordered by node and zones within node.
2928 * This results in maximum locality--normal zone overflows into local
2929 * DMA zone, if any--but risks exhausting DMA zone.
2930 */
2931static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002932{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002933 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002935
Mel Gorman54a6eb52008-04-28 02:12:16 -07002936 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07002937 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07002938 ;
2939 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2940 MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002941 zonelist->_zonerefs[j].zone = NULL;
2942 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002943}
2944
2945/*
Christoph Lameter523b9452007-10-16 01:25:37 -07002946 * Build gfp_thisnode zonelists
2947 */
2948static void build_thisnode_zonelists(pg_data_t *pgdat)
2949{
Christoph Lameter523b9452007-10-16 01:25:37 -07002950 int j;
2951 struct zonelist *zonelist;
2952
Mel Gorman54a6eb52008-04-28 02:12:16 -07002953 zonelist = &pgdat->node_zonelists[1];
2954 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002955 zonelist->_zonerefs[j].zone = NULL;
2956 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07002957}
2958
2959/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002960 * Build zonelists ordered by zone and nodes within zones.
2961 * This results in conserving DMA zone[s] until all Normal memory is
2962 * exhausted, but results in overflowing to remote node while memory
2963 * may still exist in local DMA zone.
2964 */
2965static int node_order[MAX_NUMNODES];
2966
2967static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2968{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002969 int pos, j, node;
2970 int zone_type; /* needs to be signed */
2971 struct zone *z;
2972 struct zonelist *zonelist;
2973
Mel Gorman54a6eb52008-04-28 02:12:16 -07002974 zonelist = &pgdat->node_zonelists[0];
2975 pos = 0;
2976 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2977 for (j = 0; j < nr_nodes; j++) {
2978 node = node_order[j];
2979 z = &NODE_DATA(node)->node_zones[zone_type];
2980 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002981 zoneref_set_zone(z,
2982 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002983 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002984 }
2985 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002986 }
Mel Gormandd1a2392008-04-28 02:12:17 -07002987 zonelist->_zonerefs[pos].zone = NULL;
2988 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002989}
2990
2991static int default_zonelist_order(void)
2992{
2993 int nid, zone_type;
2994 unsigned long low_kmem_size,total_size;
2995 struct zone *z;
2996 int average_size;
2997 /*
Thomas Weber88393162010-03-16 11:47:56 +01002998 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002999 * If they are really small and used heavily, the system can fall
3000 * into OOM very easily.
David Rientjese325c902010-05-24 14:32:13 -07003001 * This function detect ZONE_DMA/DMA32 size and configures zone order.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003002 */
3003 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
3004 low_kmem_size = 0;
3005 total_size = 0;
3006 for_each_online_node(nid) {
3007 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3008 z = &NODE_DATA(nid)->node_zones[zone_type];
3009 if (populated_zone(z)) {
3010 if (zone_type < ZONE_NORMAL)
3011 low_kmem_size += z->present_pages;
3012 total_size += z->present_pages;
David Rientjese325c902010-05-24 14:32:13 -07003013 } else if (zone_type == ZONE_NORMAL) {
3014 /*
3015 * If any node has only lowmem, then node order
3016 * is preferred to allow kernel allocations
3017 * locally; otherwise, they can easily infringe
3018 * on other nodes when there is an abundance of
3019 * lowmem available to allocate from.
3020 */
3021 return ZONELIST_ORDER_NODE;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003022 }
3023 }
3024 }
3025 if (!low_kmem_size || /* there are no DMA area. */
3026 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
3027 return ZONELIST_ORDER_NODE;
3028 /*
3029 * look into each node's config.
3030 * If there is a node whose DMA/DMA32 memory is very big area on
3031 * local memory, NODE_ORDER may be suitable.
3032 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003033 average_size = total_size /
3034 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003035 for_each_online_node(nid) {
3036 low_kmem_size = 0;
3037 total_size = 0;
3038 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3039 z = &NODE_DATA(nid)->node_zones[zone_type];
3040 if (populated_zone(z)) {
3041 if (zone_type < ZONE_NORMAL)
3042 low_kmem_size += z->present_pages;
3043 total_size += z->present_pages;
3044 }
3045 }
3046 if (low_kmem_size &&
3047 total_size > average_size && /* ignore small node */
3048 low_kmem_size > total_size * 70/100)
3049 return ZONELIST_ORDER_NODE;
3050 }
3051 return ZONELIST_ORDER_ZONE;
3052}
3053
3054static void set_zonelist_order(void)
3055{
3056 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
3057 current_zonelist_order = default_zonelist_order();
3058 else
3059 current_zonelist_order = user_zonelist_order;
3060}
3061
3062static void build_zonelists(pg_data_t *pgdat)
3063{
3064 int j, node, load;
3065 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003066 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003067 int local_node, prev_node;
3068 struct zonelist *zonelist;
3069 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003070
3071 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07003072 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003073 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07003074 zonelist->_zonerefs[0].zone = NULL;
3075 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003076 }
3077
3078 /* NUMA-aware ordering of nodes */
3079 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003080 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003081 prev_node = local_node;
3082 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003083
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003084 memset(node_order, 0, sizeof(node_order));
3085 j = 0;
3086
Linus Torvalds1da177e2005-04-16 15:20:36 -07003087 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08003088 int distance = node_distance(local_node, node);
3089
3090 /*
3091 * If another node is sufficiently far away then it is better
3092 * to reclaim pages in a zone before going off node.
3093 */
3094 if (distance > RECLAIM_DISTANCE)
3095 zone_reclaim_mode = 1;
3096
Linus Torvalds1da177e2005-04-16 15:20:36 -07003097 /*
3098 * We don't want to pressure a particular node.
3099 * So adding penalty to the first node in same
3100 * distance group to make it round-robin.
3101 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08003102 if (distance != node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003103 node_load[node] = load;
3104
Linus Torvalds1da177e2005-04-16 15:20:36 -07003105 prev_node = node;
3106 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003107 if (order == ZONELIST_ORDER_NODE)
3108 build_zonelists_in_node_order(pgdat, node);
3109 else
3110 node_order[j++] = node; /* remember order */
3111 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003112
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003113 if (order == ZONELIST_ORDER_ZONE) {
3114 /* calculate node order -- i.e., DMA last! */
3115 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003116 }
Christoph Lameter523b9452007-10-16 01:25:37 -07003117
3118 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003119}
3120
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003121/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003122static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003123{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003124 struct zonelist *zonelist;
3125 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07003126 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003127
Mel Gorman54a6eb52008-04-28 02:12:16 -07003128 zonelist = &pgdat->node_zonelists[0];
3129 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3130 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07003131 for (z = zonelist->_zonerefs; z->zone; z++)
3132 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003133}
3134
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003135#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3136/*
3137 * Return node id of node used for "local" allocations.
3138 * I.e., first node id of first zone in arg node's generic zonelist.
3139 * Used for initializing percpu 'numa_mem', which is used primarily
3140 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3141 */
3142int local_memory_node(int node)
3143{
3144 struct zone *zone;
3145
3146 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3147 gfp_zone(GFP_KERNEL),
3148 NULL,
3149 &zone);
3150 return zone->node;
3151}
3152#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003153
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154#else /* CONFIG_NUMA */
3155
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003156static void set_zonelist_order(void)
3157{
3158 current_zonelist_order = ZONELIST_ORDER_ZONE;
3159}
3160
3161static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003162{
Christoph Lameter19655d32006-09-25 23:31:19 -07003163 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003164 enum zone_type j;
3165 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003166
3167 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003168
Mel Gorman54a6eb52008-04-28 02:12:16 -07003169 zonelist = &pgdat->node_zonelists[0];
3170 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003171
Mel Gorman54a6eb52008-04-28 02:12:16 -07003172 /*
3173 * Now we build the zonelist so that it contains the zones
3174 * of all the other nodes.
3175 * We don't want to pressure a particular node, so when
3176 * building the zones for node N, we make sure that the
3177 * zones coming right after the local ones are those from
3178 * node N+1 (modulo N)
3179 */
3180 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3181 if (!node_online(node))
3182 continue;
3183 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3184 MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003185 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07003186 for (node = 0; node < local_node; node++) {
3187 if (!node_online(node))
3188 continue;
3189 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3190 MAX_NR_ZONES - 1);
3191 }
3192
Mel Gormandd1a2392008-04-28 02:12:17 -07003193 zonelist->_zonerefs[j].zone = NULL;
3194 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003195}
3196
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003197/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003198static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003199{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003200 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003201}
3202
Linus Torvalds1da177e2005-04-16 15:20:36 -07003203#endif /* CONFIG_NUMA */
3204
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003205/*
3206 * Boot pageset table. One per cpu which is going to be used for all
3207 * zones and all nodes. The parameters will be set in such a way
3208 * that an item put on a list will immediately be handed over to
3209 * the buddy list. This is safe since pageset manipulation is done
3210 * with interrupts disabled.
3211 *
3212 * The boot_pagesets must be kept even after bootup is complete for
3213 * unused processors and/or zones. They do play a role for bootstrapping
3214 * hotplugged processors.
3215 *
3216 * zoneinfo_show() and maybe other functions do
3217 * not check if the processor is online before following the pageset pointer.
3218 * Other parts of the kernel may not check if the zone is available.
3219 */
3220static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3221static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07003222static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003223
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003224/*
3225 * Global mutex to protect against size modification of zonelists
3226 * as well as to serialize pageset setup for the new populated zone.
3227 */
3228DEFINE_MUTEX(zonelists_mutex);
3229
Rusty Russell9b1a4d32008-07-28 12:16:30 -05003230/* return values int ....just for stop_machine() */
Haicheng Li1f522502010-05-24 14:32:51 -07003231static __init_refok int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003232{
Yasunori Goto68113782006-06-23 02:03:11 -07003233 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003234 int cpu;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003235
Bo Liu7f9cfb32009-08-18 14:11:19 -07003236#ifdef CONFIG_NUMA
3237 memset(node_load, 0, sizeof(node_load));
3238#endif
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003239 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07003240 pg_data_t *pgdat = NODE_DATA(nid);
3241
3242 build_zonelists(pgdat);
3243 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003244 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003245
3246 /*
3247 * Initialize the boot_pagesets that are going to be used
3248 * for bootstrapping processors. The real pagesets for
3249 * each zone will be allocated later when the per cpu
3250 * allocator is available.
3251 *
3252 * boot_pagesets are used also for bootstrapping offline
3253 * cpus if the system is already booted because the pagesets
3254 * are needed to initialize allocators on a specific cpu too.
3255 * F.e. the percpu allocator needs the page allocator which
3256 * needs the percpu allocator in order to allocate its pagesets
3257 * (a chicken-egg dilemma).
3258 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003259 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003260 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3261
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003262#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3263 /*
3264 * We now know the "local memory node" for each node--
3265 * i.e., the node of the first zone in the generic zonelist.
3266 * Set up numa_mem percpu variable for on-line cpus. During
3267 * boot, only the boot cpu should be on-line; we'll init the
3268 * secondary cpus' numa_mem as they come on-line. During
3269 * node/memory hotplug, we'll fixup all on-line cpus.
3270 */
3271 if (cpu_online(cpu))
3272 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3273#endif
3274 }
3275
Yasunori Goto68113782006-06-23 02:03:11 -07003276 return 0;
3277}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003278
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003279/*
3280 * Called with zonelists_mutex held always
3281 * unless system_state == SYSTEM_BOOTING.
3282 */
Paul Mundt9f6ae442011-04-14 15:21:57 -07003283void __ref build_all_zonelists(void *data)
Yasunori Goto68113782006-06-23 02:03:11 -07003284{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003285 set_zonelist_order();
3286
Yasunori Goto68113782006-06-23 02:03:11 -07003287 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07003288 __build_all_zonelists(NULL);
Mel Gorman68ad8df2008-07-23 21:26:52 -07003289 mminit_verify_zonelist();
Yasunori Goto68113782006-06-23 02:03:11 -07003290 cpuset_init_current_mems_allowed();
3291 } else {
Simon Arlott183ff222007-10-20 01:27:18 +02003292 /* we have to stop all cpus to guarantee there is no user
Yasunori Goto68113782006-06-23 02:03:11 -07003293 of zonelist */
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003294#ifdef CONFIG_MEMORY_HOTPLUG
3295 if (data)
3296 setup_zone_pageset((struct zone *)data);
3297#endif
3298 stop_machine(__build_all_zonelists, NULL, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07003299 /* cpuset refresh routine should be here */
3300 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07003301 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003302 /*
3303 * Disable grouping by mobility if the number of pages in the
3304 * system is too low to allow the mechanism to work. It would be
3305 * more accurate, but expensive to check per-zone. This check is
3306 * made on memory-hotadd so a system can start with mobility
3307 * disabled and enable it later
3308 */
Mel Gormand9c23402007-10-16 01:26:01 -07003309 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003310 page_group_by_mobility_disabled = 1;
3311 else
3312 page_group_by_mobility_disabled = 0;
3313
3314 printk("Built %i zonelists in %s order, mobility grouping %s. "
3315 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003316 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003317 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003318 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003319 vm_total_pages);
3320#ifdef CONFIG_NUMA
3321 printk("Policy zone: %s\n", zone_names[policy_zone]);
3322#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003323}
3324
3325/*
3326 * Helper functions to size the waitqueue hash table.
3327 * Essentially these want to choose hash table sizes sufficiently
3328 * large so that collisions trying to wait on pages are rare.
3329 * But in fact, the number of active page waitqueues on typical
3330 * systems is ridiculously low, less than 200. So this is even
3331 * conservative, even though it seems large.
3332 *
3333 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3334 * waitqueues, i.e. the size of the waitq table given the number of pages.
3335 */
3336#define PAGES_PER_WAITQUEUE 256
3337
Yasunori Gotocca448f2006-06-23 02:03:10 -07003338#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07003339static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003340{
3341 unsigned long size = 1;
3342
3343 pages /= PAGES_PER_WAITQUEUE;
3344
3345 while (size < pages)
3346 size <<= 1;
3347
3348 /*
3349 * Once we have dozens or even hundreds of threads sleeping
3350 * on IO we've got bigger problems than wait queue collision.
3351 * Limit the size of the wait table to a reasonable size.
3352 */
3353 size = min(size, 4096UL);
3354
3355 return max(size, 4UL);
3356}
Yasunori Gotocca448f2006-06-23 02:03:10 -07003357#else
3358/*
3359 * A zone's size might be changed by hot-add, so it is not possible to determine
3360 * a suitable size for its wait_table. So we use the maximum size now.
3361 *
3362 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3363 *
3364 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3365 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3366 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3367 *
3368 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3369 * or more by the traditional way. (See above). It equals:
3370 *
3371 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3372 * ia64(16K page size) : = ( 8G + 4M)byte.
3373 * powerpc (64K page size) : = (32G +16M)byte.
3374 */
3375static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3376{
3377 return 4096UL;
3378}
3379#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003380
3381/*
3382 * This is an integer logarithm so that shifts can be used later
3383 * to extract the more random high bits from the multiplicative
3384 * hash function before the remainder is taken.
3385 */
3386static inline unsigned long wait_table_bits(unsigned long size)
3387{
3388 return ffz(~size);
3389}
3390
3391#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3392
Mel Gorman56fd56b2007-10-16 01:25:58 -07003393/*
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07003394 * Check if a pageblock contains reserved pages
3395 */
3396static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
3397{
3398 unsigned long pfn;
3399
3400 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
3401 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
3402 return 1;
3403 }
3404 return 0;
3405}
3406
3407/*
Mel Gormand9c23402007-10-16 01:26:01 -07003408 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07003409 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3410 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07003411 * higher will lead to a bigger reserve which will get freed as contiguous
3412 * blocks as reclaim kicks in
3413 */
3414static void setup_zone_migrate_reserve(struct zone *zone)
3415{
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07003416 unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003417 struct page *page;
Mel Gorman78986a62009-09-21 17:03:02 -07003418 unsigned long block_migratetype;
3419 int reserve;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003420
3421 /* Get the start pfn, end pfn and the number of blocks to reserve */
3422 start_pfn = zone->zone_start_pfn;
3423 end_pfn = start_pfn + zone->spanned_pages;
Mel Gorman41858962009-06-16 15:32:12 -07003424 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07003425 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003426
Mel Gorman78986a62009-09-21 17:03:02 -07003427 /*
3428 * Reserve blocks are generally in place to help high-order atomic
3429 * allocations that are short-lived. A min_free_kbytes value that
3430 * would result in more than 2 reserve blocks for atomic allocations
3431 * is assumed to be in place to help anti-fragmentation for the
3432 * future allocation of hugepages at runtime.
3433 */
3434 reserve = min(2, reserve);
3435
Mel Gormand9c23402007-10-16 01:26:01 -07003436 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07003437 if (!pfn_valid(pfn))
3438 continue;
3439 page = pfn_to_page(pfn);
3440
Adam Litke344c7902008-09-02 14:35:38 -07003441 /* Watch out for overlapping nodes */
3442 if (page_to_nid(page) != zone_to_nid(zone))
3443 continue;
3444
Mel Gorman56fd56b2007-10-16 01:25:58 -07003445 /* Blocks with reserved pages will never free, skip them. */
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07003446 block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
3447 if (pageblock_is_reserved(pfn, block_end_pfn))
Mel Gorman56fd56b2007-10-16 01:25:58 -07003448 continue;
3449
3450 block_migratetype = get_pageblock_migratetype(page);
3451
3452 /* If this block is reserved, account for it */
3453 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
3454 reserve--;
3455 continue;
3456 }
3457
3458 /* Suitable for reserving if this block is movable */
3459 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
3460 set_pageblock_migratetype(page, MIGRATE_RESERVE);
3461 move_freepages_block(zone, page, MIGRATE_RESERVE);
3462 reserve--;
3463 continue;
3464 }
3465
3466 /*
3467 * If the reserve is met and this is a previous reserved block,
3468 * take it back
3469 */
3470 if (block_migratetype == MIGRATE_RESERVE) {
3471 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
3472 move_freepages_block(zone, page, MIGRATE_MOVABLE);
3473 }
3474 }
3475}
Mel Gormanac0e5b72007-10-16 01:25:58 -07003476
Linus Torvalds1da177e2005-04-16 15:20:36 -07003477/*
3478 * Initially all pages are reserved - free ones are freed
3479 * up by free_all_bootmem() once the early boot process is
3480 * done. Non-atomic initialization, single-pass.
3481 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003482void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa02007-01-10 23:15:30 -08003483 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003484{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003485 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07003486 unsigned long end_pfn = start_pfn + size;
3487 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003488 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003489
Hugh Dickins22b31ee2009-01-06 14:40:09 -08003490 if (highest_memmap_pfn < end_pfn - 1)
3491 highest_memmap_pfn = end_pfn - 1;
3492
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003493 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08003494 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa02007-01-10 23:15:30 -08003495 /*
3496 * There can be holes in boot-time mem_map[]s
3497 * handed to this function. They do not
3498 * exist on hotplugged memory.
3499 */
3500 if (context == MEMMAP_EARLY) {
3501 if (!early_pfn_valid(pfn))
3502 continue;
3503 if (!early_pfn_in_nid(pfn, nid))
3504 continue;
3505 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003506 page = pfn_to_page(pfn);
3507 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07003508 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08003509 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003510 reset_page_mapcount(page);
3511 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003512 /*
3513 * Mark the block movable so that blocks are reserved for
3514 * movable at startup. This will force kernel allocations
3515 * to reserve their blocks rather than leaking throughout
3516 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07003517 * kernel allocations are made. Later some blocks near
3518 * the start are marked MIGRATE_RESERVE by
3519 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003520 *
3521 * bitmap is created for zone's valid pfn range. but memmap
3522 * can be created for invalid pages (for alignment)
3523 * check here not to call set_pageblock_migratetype() against
3524 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003525 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003526 if ((z->zone_start_pfn <= pfn)
3527 && (pfn < z->zone_start_pfn + z->spanned_pages)
3528 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07003529 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003530
Linus Torvalds1da177e2005-04-16 15:20:36 -07003531 INIT_LIST_HEAD(&page->lru);
3532#ifdef WANT_PAGE_VIRTUAL
3533 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3534 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07003535 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003536#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003537 }
3538}
3539
Andi Kleen1e548de2008-02-04 22:29:26 -08003540static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003541{
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003542 int order, t;
3543 for_each_migratetype_order(order, t) {
3544 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003545 zone->free_area[order].nr_free = 0;
3546 }
3547}
3548
3549#ifndef __HAVE_ARCH_MEMMAP_INIT
3550#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa02007-01-10 23:15:30 -08003551 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003552#endif
3553
Sam Ravnborg1d6f4e62008-01-17 15:21:12 -08003554static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003555{
David Howells3a6be872009-05-06 16:03:03 -07003556#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003557 int batch;
3558
3559 /*
3560 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07003561 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003562 *
3563 * OK, so we don't know how big the cache is. So guess.
3564 */
3565 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07003566 if (batch * PAGE_SIZE > 512 * 1024)
3567 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003568 batch /= 4; /* We effectively *= 4 below */
3569 if (batch < 1)
3570 batch = 1;
3571
3572 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11003573 * Clamp the batch to a 2^n - 1 value. Having a power
3574 * of 2 value was found to be more likely to have
3575 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003576 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11003577 * For example if 2 tasks are alternately allocating
3578 * batches of pages, one task can end up with a lot
3579 * of pages of one half of the possible page colors
3580 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003581 */
David Howells91552032009-05-06 16:03:02 -07003582 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07003583
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003584 return batch;
David Howells3a6be872009-05-06 16:03:03 -07003585
3586#else
3587 /* The deferral and batching of frees should be suppressed under NOMMU
3588 * conditions.
3589 *
3590 * The problem is that NOMMU needs to be able to allocate large chunks
3591 * of contiguous memory as there's no hardware page translation to
3592 * assemble apparent contiguous memory from discontiguous pages.
3593 *
3594 * Queueing large contiguous runs of pages for batching, however,
3595 * causes the pages to actually be freed in smaller chunks. As there
3596 * can be a significant delay between the individual batches being
3597 * recycled, this leads to the once large chunks of space being
3598 * fragmented and becoming unavailable for high-order allocations.
3599 */
3600 return 0;
3601#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003602}
3603
Adrian Bunkb69a7282008-07-23 21:28:12 -07003604static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
Christoph Lameter2caaad42005-06-21 17:15:00 -07003605{
3606 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003607 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003608
Magnus Damm1c6fe942005-10-26 01:58:59 -07003609 memset(p, 0, sizeof(*p));
3610
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003611 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003612 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003613 pcp->high = 6 * batch;
3614 pcp->batch = max(1UL, 1 * batch);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003615 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
3616 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07003617}
3618
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003619/*
3620 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3621 * to the value high for the pageset p.
3622 */
3623
3624static void setup_pagelist_highmark(struct per_cpu_pageset *p,
3625 unsigned long high)
3626{
3627 struct per_cpu_pages *pcp;
3628
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003629 pcp = &p->pcp;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003630 pcp->high = high;
3631 pcp->batch = max(1UL, high/4);
3632 if ((high/4) > (PAGE_SHIFT * 8))
3633 pcp->batch = PAGE_SHIFT * 8;
3634}
3635
Nikanth Karthikesan58c2ee42011-03-15 10:59:02 +05303636static void setup_zone_pageset(struct zone *zone)
Wu Fengguang319774e2010-05-24 14:32:49 -07003637{
3638 int cpu;
3639
3640 zone->pageset = alloc_percpu(struct per_cpu_pageset);
3641
3642 for_each_possible_cpu(cpu) {
3643 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
3644
3645 setup_pageset(pcp, zone_batchsize(zone));
3646
3647 if (percpu_pagelist_fraction)
3648 setup_pagelist_highmark(pcp,
3649 (zone->present_pages /
3650 percpu_pagelist_fraction));
3651 }
3652}
3653
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003654/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003655 * Allocate per cpu pagesets and initialize them.
3656 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07003657 */
Al Viro78d99552005-12-15 09:18:25 +00003658void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003659{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003660 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003661
Wu Fengguang319774e2010-05-24 14:32:49 -07003662 for_each_populated_zone(zone)
3663 setup_zone_pageset(zone);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003664}
3665
Sam Ravnborg577a32f2007-05-17 23:29:25 +02003666static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07003667int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07003668{
3669 int i;
3670 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003671 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07003672
3673 /*
3674 * The per-page waitqueue mechanism uses hashed waitqueues
3675 * per zone.
3676 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07003677 zone->wait_table_hash_nr_entries =
3678 wait_table_hash_nr_entries(zone_size_pages);
3679 zone->wait_table_bits =
3680 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003681 alloc_size = zone->wait_table_hash_nr_entries
3682 * sizeof(wait_queue_head_t);
3683
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07003684 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07003685 zone->wait_table = (wait_queue_head_t *)
Yinghai Lu8f389a992011-05-11 15:13:32 -07003686 alloc_bootmem_node_nopanic(pgdat, alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003687 } else {
3688 /*
3689 * This case means that a zone whose size was 0 gets new memory
3690 * via memory hot-add.
3691 * But it may be the case that a new node was hot-added. In
3692 * this case vmalloc() will not be able to use this new node's
3693 * memory - this wait_table must be initialized to use this new
3694 * node itself as well.
3695 * To use this new node's memory, further consideration will be
3696 * necessary.
3697 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07003698 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003699 }
3700 if (!zone->wait_table)
3701 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07003702
Yasunori Goto02b694d2006-06-23 02:03:08 -07003703 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07003704 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003705
3706 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003707}
3708
Shaohua Li112067f2009-09-21 17:01:16 -07003709static int __zone_pcp_update(void *data)
3710{
3711 struct zone *zone = data;
3712 int cpu;
3713 unsigned long batch = zone_batchsize(zone), flags;
3714
Thomas Gleixner2d30a1f2010-03-10 15:20:40 -08003715 for_each_possible_cpu(cpu) {
Shaohua Li112067f2009-09-21 17:01:16 -07003716 struct per_cpu_pageset *pset;
3717 struct per_cpu_pages *pcp;
3718
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003719 pset = per_cpu_ptr(zone->pageset, cpu);
Shaohua Li112067f2009-09-21 17:01:16 -07003720 pcp = &pset->pcp;
3721
3722 local_irq_save(flags);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003723 free_pcppages_bulk(zone, pcp->count, pcp);
Shaohua Li112067f2009-09-21 17:01:16 -07003724 setup_pageset(pset, batch);
3725 local_irq_restore(flags);
3726 }
3727 return 0;
3728}
3729
3730void zone_pcp_update(struct zone *zone)
3731{
3732 stop_machine(__zone_pcp_update, zone, NULL);
3733}
3734
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003735static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07003736{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003737 /*
3738 * per cpu subsystem is not up at this point. The following code
3739 * relies on the ability of the linker to provide the
3740 * offset of a (static) per cpu variable into the per cpu area.
3741 */
3742 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07003743
Anton Blanchardf5335c02006-03-25 03:06:49 -08003744 if (zone->present_pages)
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003745 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
3746 zone->name, zone->present_pages,
3747 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07003748}
3749
Yasunori Goto718127c2006-06-23 02:03:10 -07003750__meminit int init_currently_empty_zone(struct zone *zone,
3751 unsigned long zone_start_pfn,
Dave Hansena2f3aa02007-01-10 23:15:30 -08003752 unsigned long size,
3753 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07003754{
3755 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003756 int ret;
3757 ret = zone_wait_table_init(zone, size);
3758 if (ret)
3759 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07003760 pgdat->nr_zones = zone_idx(zone) + 1;
3761
Dave Hansened8ece22005-10-29 18:16:50 -07003762 zone->zone_start_pfn = zone_start_pfn;
3763
Mel Gorman708614e2008-07-23 21:26:51 -07003764 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3765 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3766 pgdat->node_id,
3767 (unsigned long)zone_idx(zone),
3768 zone_start_pfn, (zone_start_pfn + size));
3769
Andi Kleen1e548de2008-02-04 22:29:26 -08003770 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07003771
3772 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003773}
3774
Mel Gormanc7132162006-09-27 01:49:43 -07003775#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3776/*
3777 * Basic iterator support. Return the first range of PFNs for a node
3778 * Note: nid == MAX_NUMNODES returns first region regardless of node
3779 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003780static int __meminit first_active_region_index_in_nid(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003781{
3782 int i;
3783
3784 for (i = 0; i < nr_nodemap_entries; i++)
3785 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3786 return i;
3787
3788 return -1;
3789}
3790
3791/*
3792 * Basic iterator support. Return the next active range of PFNs for a node
Simon Arlott183ff222007-10-20 01:27:18 +02003793 * Note: nid == MAX_NUMNODES returns next region regardless of node
Mel Gormanc7132162006-09-27 01:49:43 -07003794 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003795static int __meminit next_active_region_index_in_nid(int index, int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003796{
3797 for (index = index + 1; index < nr_nodemap_entries; index++)
3798 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3799 return index;
3800
3801 return -1;
3802}
3803
3804#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3805/*
3806 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3807 * Architectures may implement their own version but if add_active_range()
3808 * was used and there are no special requirements, this is a convenient
3809 * alternative
3810 */
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003811int __meminit __early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07003812{
3813 int i;
3814
3815 for (i = 0; i < nr_nodemap_entries; i++) {
3816 unsigned long start_pfn = early_node_map[i].start_pfn;
3817 unsigned long end_pfn = early_node_map[i].end_pfn;
3818
3819 if (start_pfn <= pfn && pfn < end_pfn)
3820 return early_node_map[i].nid;
3821 }
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003822 /* This is a memory hole */
3823 return -1;
Mel Gormanc7132162006-09-27 01:49:43 -07003824}
3825#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3826
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003827int __meminit early_pfn_to_nid(unsigned long pfn)
3828{
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003829 int nid;
3830
3831 nid = __early_pfn_to_nid(pfn);
3832 if (nid >= 0)
3833 return nid;
3834 /* just returns 0 */
3835 return 0;
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003836}
3837
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003838#ifdef CONFIG_NODES_SPAN_OTHER_NODES
3839bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3840{
3841 int nid;
3842
3843 nid = __early_pfn_to_nid(pfn);
3844 if (nid >= 0 && nid != node)
3845 return false;
3846 return true;
3847}
3848#endif
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003849
Mel Gormanc7132162006-09-27 01:49:43 -07003850/* Basic iterator support to walk early_node_map[] */
3851#define for_each_active_range_index_in_nid(i, nid) \
3852 for (i = first_active_region_index_in_nid(nid); i != -1; \
3853 i = next_active_region_index_in_nid(i, nid))
3854
3855/**
3856 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003857 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3858 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07003859 *
3860 * If an architecture guarantees that all ranges registered with
3861 * add_active_ranges() contain no holes and may be freed, this
3862 * this function may be used instead of calling free_bootmem() manually.
3863 */
3864void __init free_bootmem_with_active_regions(int nid,
3865 unsigned long max_low_pfn)
3866{
3867 int i;
3868
3869 for_each_active_range_index_in_nid(i, nid) {
3870 unsigned long size_pages = 0;
3871 unsigned long end_pfn = early_node_map[i].end_pfn;
3872
3873 if (early_node_map[i].start_pfn >= max_low_pfn)
3874 continue;
3875
3876 if (end_pfn > max_low_pfn)
3877 end_pfn = max_low_pfn;
3878
3879 size_pages = end_pfn - early_node_map[i].start_pfn;
3880 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3881 PFN_PHYS(early_node_map[i].start_pfn),
3882 size_pages << PAGE_SHIFT);
3883 }
3884}
3885
Yinghai Luedbe7d22010-08-25 13:39:16 -07003886#ifdef CONFIG_HAVE_MEMBLOCK
Yinghai Lucc289892011-02-26 13:05:43 +01003887/*
3888 * Basic iterator support. Return the last range of PFNs for a node
3889 * Note: nid == MAX_NUMNODES returns last region regardless of node
3890 */
3891static int __meminit last_active_region_index_in_nid(int nid)
3892{
3893 int i;
3894
3895 for (i = nr_nodemap_entries - 1; i >= 0; i--)
3896 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3897 return i;
3898
3899 return -1;
3900}
3901
3902/*
3903 * Basic iterator support. Return the previous active range of PFNs for a node
3904 * Note: nid == MAX_NUMNODES returns next region regardless of node
3905 */
3906static int __meminit previous_active_region_index_in_nid(int index, int nid)
3907{
3908 for (index = index - 1; index >= 0; index--)
3909 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3910 return index;
3911
3912 return -1;
3913}
3914
3915#define for_each_active_range_index_in_nid_reverse(i, nid) \
3916 for (i = last_active_region_index_in_nid(nid); i != -1; \
3917 i = previous_active_region_index_in_nid(i, nid))
3918
Yinghai Luedbe7d22010-08-25 13:39:16 -07003919u64 __init find_memory_core_early(int nid, u64 size, u64 align,
3920 u64 goal, u64 limit)
3921{
3922 int i;
3923
3924 /* Need to go over early_node_map to find out good range for node */
Yinghai Lu1a4a6782010-12-17 16:59:07 -08003925 for_each_active_range_index_in_nid_reverse(i, nid) {
Yinghai Luedbe7d22010-08-25 13:39:16 -07003926 u64 addr;
3927 u64 ei_start, ei_last;
3928 u64 final_start, final_end;
3929
3930 ei_last = early_node_map[i].end_pfn;
3931 ei_last <<= PAGE_SHIFT;
3932 ei_start = early_node_map[i].start_pfn;
3933 ei_start <<= PAGE_SHIFT;
3934
3935 final_start = max(ei_start, goal);
3936 final_end = min(ei_last, limit);
3937
3938 if (final_start >= final_end)
3939 continue;
3940
3941 addr = memblock_find_in_range(final_start, final_end, size, align);
3942
3943 if (addr == MEMBLOCK_ERROR)
3944 continue;
3945
3946 return addr;
3947 }
3948
3949 return MEMBLOCK_ERROR;
3950}
3951#endif
3952
Yinghai Lu08677212010-02-10 01:20:20 -08003953int __init add_from_early_node_map(struct range *range, int az,
3954 int nr_range, int nid)
3955{
3956 int i;
3957 u64 start, end;
3958
3959 /* need to go over early_node_map to find out good range for node */
3960 for_each_active_range_index_in_nid(i, nid) {
3961 start = early_node_map[i].start_pfn;
3962 end = early_node_map[i].end_pfn;
3963 nr_range = add_range(range, az, nr_range, start, end);
3964 }
3965 return nr_range;
3966}
3967
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003968void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3969{
3970 int i;
Yinghai Lud52d53b2008-06-16 20:10:55 -07003971 int ret;
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003972
Yinghai Lud52d53b2008-06-16 20:10:55 -07003973 for_each_active_range_index_in_nid(i, nid) {
3974 ret = work_fn(early_node_map[i].start_pfn,
3975 early_node_map[i].end_pfn, data);
3976 if (ret)
3977 break;
3978 }
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003979}
Mel Gormanc7132162006-09-27 01:49:43 -07003980/**
3981 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003982 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07003983 *
3984 * If an architecture guarantees that all ranges registered with
3985 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003986 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07003987 */
3988void __init sparse_memory_present_with_active_regions(int nid)
3989{
3990 int i;
3991
3992 for_each_active_range_index_in_nid(i, nid)
3993 memory_present(early_node_map[i].nid,
3994 early_node_map[i].start_pfn,
3995 early_node_map[i].end_pfn);
3996}
3997
3998/**
3999 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004000 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4001 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4002 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07004003 *
4004 * It returns the start and end page frame of a node based on information
4005 * provided by an arch calling add_active_range(). If called for a node
4006 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004007 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07004008 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004009void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004010 unsigned long *start_pfn, unsigned long *end_pfn)
4011{
4012 int i;
4013 *start_pfn = -1UL;
4014 *end_pfn = 0;
4015
4016 for_each_active_range_index_in_nid(i, nid) {
4017 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
4018 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
4019 }
4020
Christoph Lameter633c0662007-10-16 01:25:37 -07004021 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07004022 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07004023}
4024
4025/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07004026 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4027 * assumption is made that zones within a node are ordered in monotonic
4028 * increasing memory addresses so that the "highest" populated zone is used
4029 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004030static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004031{
4032 int zone_index;
4033 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
4034 if (zone_index == ZONE_MOVABLE)
4035 continue;
4036
4037 if (arch_zone_highest_possible_pfn[zone_index] >
4038 arch_zone_lowest_possible_pfn[zone_index])
4039 break;
4040 }
4041
4042 VM_BUG_ON(zone_index == -1);
4043 movable_zone = zone_index;
4044}
4045
4046/*
4047 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004048 * because it is sized independent of architecture. Unlike the other zones,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004049 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4050 * in each node depending on the size of each node and how evenly kernelcore
4051 * is distributed. This helper function adjusts the zone ranges
4052 * provided by the architecture for a given node by using the end of the
4053 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4054 * zones within a node are in order of monotonic increases memory addresses
4055 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004056static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004057 unsigned long zone_type,
4058 unsigned long node_start_pfn,
4059 unsigned long node_end_pfn,
4060 unsigned long *zone_start_pfn,
4061 unsigned long *zone_end_pfn)
4062{
4063 /* Only adjust if ZONE_MOVABLE is on this node */
4064 if (zone_movable_pfn[nid]) {
4065 /* Size ZONE_MOVABLE */
4066 if (zone_type == ZONE_MOVABLE) {
4067 *zone_start_pfn = zone_movable_pfn[nid];
4068 *zone_end_pfn = min(node_end_pfn,
4069 arch_zone_highest_possible_pfn[movable_zone]);
4070
4071 /* Adjust for ZONE_MOVABLE starting within this range */
4072 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
4073 *zone_end_pfn > zone_movable_pfn[nid]) {
4074 *zone_end_pfn = zone_movable_pfn[nid];
4075
4076 /* Check if this whole range is within ZONE_MOVABLE */
4077 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
4078 *zone_start_pfn = *zone_end_pfn;
4079 }
4080}
4081
4082/*
Mel Gormanc7132162006-09-27 01:49:43 -07004083 * Return the number of pages a zone spans in a node, including holes
4084 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4085 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004086static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004087 unsigned long zone_type,
4088 unsigned long *ignored)
4089{
4090 unsigned long node_start_pfn, node_end_pfn;
4091 unsigned long zone_start_pfn, zone_end_pfn;
4092
4093 /* Get the start and end of the node and zone */
4094 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
4095 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4096 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07004097 adjust_zone_range_for_zone_movable(nid, zone_type,
4098 node_start_pfn, node_end_pfn,
4099 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004100
4101 /* Check that this node has pages within the zone's required range */
4102 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4103 return 0;
4104
4105 /* Move the zone boundaries inside the node if necessary */
4106 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4107 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4108
4109 /* Return the spanned pages */
4110 return zone_end_pfn - zone_start_pfn;
4111}
4112
4113/*
4114 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004115 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07004116 */
Yinghai Lu32996252009-12-15 17:59:02 -08004117unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004118 unsigned long range_start_pfn,
4119 unsigned long range_end_pfn)
4120{
4121 int i = 0;
4122 unsigned long prev_end_pfn = 0, hole_pages = 0;
4123 unsigned long start_pfn;
4124
4125 /* Find the end_pfn of the first active range of pfns in the node */
4126 i = first_active_region_index_in_nid(nid);
4127 if (i == -1)
4128 return 0;
4129
Mel Gormanb5445f92007-07-26 10:41:18 -07004130 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
4131
Mel Gorman9c7cd682006-09-27 01:49:58 -07004132 /* Account for ranges before physical memory on this node */
4133 if (early_node_map[i].start_pfn > range_start_pfn)
Mel Gormanb5445f92007-07-26 10:41:18 -07004134 hole_pages = prev_end_pfn - range_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004135
4136 /* Find all holes for the zone within the node */
4137 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
4138
4139 /* No need to continue if prev_end_pfn is outside the zone */
4140 if (prev_end_pfn >= range_end_pfn)
4141 break;
4142
4143 /* Make sure the end of the zone is not within the hole */
4144 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
4145 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
4146
4147 /* Update the hole size cound and move on */
4148 if (start_pfn > range_start_pfn) {
4149 BUG_ON(prev_end_pfn > start_pfn);
4150 hole_pages += start_pfn - prev_end_pfn;
4151 }
4152 prev_end_pfn = early_node_map[i].end_pfn;
4153 }
4154
Mel Gorman9c7cd682006-09-27 01:49:58 -07004155 /* Account for ranges past physical memory on this node */
4156 if (range_end_pfn > prev_end_pfn)
Mel Gorman0c6cb972006-10-28 10:38:59 -07004157 hole_pages += range_end_pfn -
Mel Gorman9c7cd682006-09-27 01:49:58 -07004158 max(range_start_pfn, prev_end_pfn);
4159
Mel Gormanc7132162006-09-27 01:49:43 -07004160 return hole_pages;
4161}
4162
4163/**
4164 * absent_pages_in_range - Return number of page frames in holes within a range
4165 * @start_pfn: The start PFN to start searching for holes
4166 * @end_pfn: The end PFN to stop searching for holes
4167 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004168 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07004169 */
4170unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4171 unsigned long end_pfn)
4172{
4173 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4174}
4175
4176/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004177static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004178 unsigned long zone_type,
4179 unsigned long *ignored)
4180{
Mel Gorman9c7cd682006-09-27 01:49:58 -07004181 unsigned long node_start_pfn, node_end_pfn;
4182 unsigned long zone_start_pfn, zone_end_pfn;
4183
4184 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
4185 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
4186 node_start_pfn);
4187 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
4188 node_end_pfn);
4189
Mel Gorman2a1e2742007-07-17 04:03:12 -07004190 adjust_zone_range_for_zone_movable(nid, zone_type,
4191 node_start_pfn, node_end_pfn,
4192 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004193 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004194}
Mel Gorman0e0b8642006-09-27 01:49:56 -07004195
Mel Gormanc7132162006-09-27 01:49:43 -07004196#else
Paul Mundt6ea6e682007-07-15 23:38:20 -07004197static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004198 unsigned long zone_type,
4199 unsigned long *zones_size)
4200{
4201 return zones_size[zone_type];
4202}
4203
Paul Mundt6ea6e682007-07-15 23:38:20 -07004204static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004205 unsigned long zone_type,
4206 unsigned long *zholes_size)
4207{
4208 if (!zholes_size)
4209 return 0;
4210
4211 return zholes_size[zone_type];
4212}
Mel Gorman0e0b8642006-09-27 01:49:56 -07004213
Mel Gormanc7132162006-09-27 01:49:43 -07004214#endif
4215
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004216static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Mel Gormanc7132162006-09-27 01:49:43 -07004217 unsigned long *zones_size, unsigned long *zholes_size)
4218{
4219 unsigned long realtotalpages, totalpages = 0;
4220 enum zone_type i;
4221
4222 for (i = 0; i < MAX_NR_ZONES; i++)
4223 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
4224 zones_size);
4225 pgdat->node_spanned_pages = totalpages;
4226
4227 realtotalpages = totalpages;
4228 for (i = 0; i < MAX_NR_ZONES; i++)
4229 realtotalpages -=
4230 zone_absent_pages_in_node(pgdat->node_id, i,
4231 zholes_size);
4232 pgdat->node_present_pages = realtotalpages;
4233 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4234 realtotalpages);
4235}
4236
Mel Gorman835c1342007-10-16 01:25:47 -07004237#ifndef CONFIG_SPARSEMEM
4238/*
4239 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07004240 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4241 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07004242 * round what is now in bits to nearest long in bits, then return it in
4243 * bytes.
4244 */
4245static unsigned long __init usemap_size(unsigned long zonesize)
4246{
4247 unsigned long usemapsize;
4248
Mel Gormand9c23402007-10-16 01:26:01 -07004249 usemapsize = roundup(zonesize, pageblock_nr_pages);
4250 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07004251 usemapsize *= NR_PAGEBLOCK_BITS;
4252 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4253
4254 return usemapsize / 8;
4255}
4256
4257static void __init setup_usemap(struct pglist_data *pgdat,
4258 struct zone *zone, unsigned long zonesize)
4259{
4260 unsigned long usemapsize = usemap_size(zonesize);
4261 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08004262 if (usemapsize)
Yinghai Lu8f389a992011-05-11 15:13:32 -07004263 zone->pageblock_flags = alloc_bootmem_node_nopanic(pgdat,
4264 usemapsize);
Mel Gorman835c1342007-10-16 01:25:47 -07004265}
4266#else
Jesper Juhlfa9f90b2010-11-28 21:39:34 +01004267static inline void setup_usemap(struct pglist_data *pgdat,
Mel Gorman835c1342007-10-16 01:25:47 -07004268 struct zone *zone, unsigned long zonesize) {}
4269#endif /* CONFIG_SPARSEMEM */
4270
Mel Gormand9c23402007-10-16 01:26:01 -07004271#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08004272
4273/* Return a sensible default order for the pageblock size. */
4274static inline int pageblock_default_order(void)
4275{
4276 if (HPAGE_SHIFT > PAGE_SHIFT)
4277 return HUGETLB_PAGE_ORDER;
4278
4279 return MAX_ORDER-1;
4280}
4281
Mel Gormand9c23402007-10-16 01:26:01 -07004282/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4283static inline void __init set_pageblock_order(unsigned int order)
4284{
4285 /* Check that pageblock_nr_pages has not already been setup */
4286 if (pageblock_order)
4287 return;
4288
4289 /*
4290 * Assume the largest contiguous order of interest is a huge page.
4291 * This value may be variable depending on boot parameters on IA64
4292 */
4293 pageblock_order = order;
4294}
4295#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4296
Mel Gormanba72cb82007-11-28 16:21:13 -08004297/*
4298 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4299 * and pageblock_default_order() are unused as pageblock_order is set
4300 * at compile-time. See include/linux/pageblock-flags.h for the values of
4301 * pageblock_order based on the kernel config
4302 */
4303static inline int pageblock_default_order(unsigned int order)
4304{
4305 return MAX_ORDER-1;
4306}
Mel Gormand9c23402007-10-16 01:26:01 -07004307#define set_pageblock_order(x) do {} while (0)
4308
4309#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4310
Linus Torvalds1da177e2005-04-16 15:20:36 -07004311/*
4312 * Set up the zone data structures:
4313 * - mark all pages reserved
4314 * - mark all memory queues empty
4315 * - clear the memory bitmaps
4316 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08004317static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004318 unsigned long *zones_size, unsigned long *zholes_size)
4319{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07004320 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07004321 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004322 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07004323 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004324
Dave Hansen208d54e2005-10-29 18:16:52 -07004325 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004326 pgdat->nr_zones = 0;
4327 init_waitqueue_head(&pgdat->kswapd_wait);
4328 pgdat->kswapd_max_order = 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004329 pgdat_page_cgroup_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004330
4331 for (j = 0; j < MAX_NR_ZONES; j++) {
4332 struct zone *zone = pgdat->node_zones + j;
Mel Gorman0e0b8642006-09-27 01:49:56 -07004333 unsigned long size, realsize, memmap_pages;
Christoph Lameterb69408e2008-10-18 20:26:14 -07004334 enum lru_list l;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004335
Mel Gormanc7132162006-09-27 01:49:43 -07004336 size = zone_spanned_pages_in_node(nid, j, zones_size);
4337 realsize = size - zone_absent_pages_in_node(nid, j,
4338 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004339
Mel Gorman0e0b8642006-09-27 01:49:56 -07004340 /*
4341 * Adjust realsize so that it accounts for how much memory
4342 * is used by this zone for memmap. This affects the watermark
4343 * and per-cpu initialisations
4344 */
Johannes Weinerf7232152008-05-23 13:04:21 -07004345 memmap_pages =
4346 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
Mel Gorman0e0b8642006-09-27 01:49:56 -07004347 if (realsize >= memmap_pages) {
4348 realsize -= memmap_pages;
Yinghai Lu5594c8c2009-01-06 14:39:14 -08004349 if (memmap_pages)
4350 printk(KERN_DEBUG
4351 " %s zone: %lu pages used for memmap\n",
4352 zone_names[j], memmap_pages);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004353 } else
4354 printk(KERN_WARNING
4355 " %s zone: %lu pages exceeds realsize %lu\n",
4356 zone_names[j], memmap_pages, realsize);
4357
Christoph Lameter62672762007-02-10 01:43:07 -08004358 /* Account for reserved pages */
4359 if (j == 0 && realsize > dma_reserve) {
Mel Gorman0e0b8642006-09-27 01:49:56 -07004360 realsize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07004361 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08004362 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004363 }
4364
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004365 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004366 nr_kernel_pages += realsize;
4367 nr_all_pages += realsize;
4368
4369 zone->spanned_pages = size;
4370 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07004371#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07004372 zone->node = nid;
Christoph Lameter8417bba2006-09-25 23:31:51 -07004373 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07004374 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07004375 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07004376#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004377 zone->name = zone_names[j];
4378 spin_lock_init(&zone->lock);
4379 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07004380 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004381 zone->zone_pgdat = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004382
Dave Hansened8ece22005-10-29 18:16:50 -07004383 zone_pcp_init(zone);
KAMEZAWA Hiroyuki246e87a2011-05-26 16:25:34 -07004384 for_each_lru(l)
Christoph Lameterb69408e2008-10-18 20:26:14 -07004385 INIT_LIST_HEAD(&zone->lru[l].list);
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08004386 zone->reclaim_stat.recent_rotated[0] = 0;
4387 zone->reclaim_stat.recent_rotated[1] = 0;
4388 zone->reclaim_stat.recent_scanned[0] = 0;
4389 zone->reclaim_stat.recent_scanned[1] = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07004390 zap_zone_vm_stats(zone);
David Rientjese815af92007-10-16 23:25:54 -07004391 zone->flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004392 if (!size)
4393 continue;
4394
Mel Gormanba72cb82007-11-28 16:21:13 -08004395 set_pageblock_order(pageblock_default_order());
Mel Gorman835c1342007-10-16 01:25:47 -07004396 setup_usemap(pgdat, zone, size);
Dave Hansena2f3aa02007-01-10 23:15:30 -08004397 ret = init_currently_empty_zone(zone, zone_start_pfn,
4398 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07004399 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07004400 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004401 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004402 }
4403}
4404
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004405static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004406{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004407 /* Skip empty nodes */
4408 if (!pgdat->node_spanned_pages)
4409 return;
4410
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004411#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07004412 /* ia64 gets its own node_mem_map, before this, without bootmem */
4413 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07004414 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004415 struct page *map;
4416
Bob Piccoe984bb42006-05-20 15:00:31 -07004417 /*
4418 * The zone's endpoints aren't required to be MAX_ORDER
4419 * aligned but the node_mem_map endpoints must be in order
4420 * for the buddy allocator to function correctly.
4421 */
4422 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
4423 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
4424 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4425 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07004426 map = alloc_remap(pgdat->node_id, size);
4427 if (!map)
Yinghai Lu8f389a992011-05-11 15:13:32 -07004428 map = alloc_bootmem_node_nopanic(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07004429 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004430 }
Roman Zippel12d810c2007-05-31 00:40:54 -07004431#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07004432 /*
4433 * With no DISCONTIG, the global mem_map is just set as node 0's
4434 */
Mel Gormanc7132162006-09-27 01:49:43 -07004435 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004436 mem_map = NODE_DATA(0)->node_mem_map;
Mel Gormanc7132162006-09-27 01:49:43 -07004437#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
4438 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08004439 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Mel Gormanc7132162006-09-27 01:49:43 -07004440#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4441 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004442#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004443#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004444}
4445
Johannes Weiner9109fb72008-07-23 21:27:20 -07004446void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4447 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004448{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004449 pg_data_t *pgdat = NODE_DATA(nid);
4450
Linus Torvalds1da177e2005-04-16 15:20:36 -07004451 pgdat->node_id = nid;
4452 pgdat->node_start_pfn = node_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004453 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004454
4455 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07004456#ifdef CONFIG_FLAT_NODE_MEM_MAP
4457 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4458 nid, (unsigned long)pgdat,
4459 (unsigned long)pgdat->node_mem_map);
4460#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004461
4462 free_area_init_core(pgdat, zones_size, zholes_size);
4463}
4464
Mel Gormanc7132162006-09-27 01:49:43 -07004465#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07004466
4467#if MAX_NUMNODES > 1
4468/*
4469 * Figure out the number of possible node ids.
4470 */
4471static void __init setup_nr_node_ids(void)
4472{
4473 unsigned int node;
4474 unsigned int highest = 0;
4475
4476 for_each_node_mask(node, node_possible_map)
4477 highest = node;
4478 nr_node_ids = highest + 1;
4479}
4480#else
4481static inline void setup_nr_node_ids(void)
4482{
4483}
4484#endif
4485
Mel Gormanc7132162006-09-27 01:49:43 -07004486/**
4487 * add_active_range - Register a range of PFNs backed by physical memory
4488 * @nid: The node ID the range resides on
4489 * @start_pfn: The start PFN of the available physical memory
4490 * @end_pfn: The end PFN of the available physical memory
4491 *
4492 * These ranges are stored in an early_node_map[] and later used by
4493 * free_area_init_nodes() to calculate zone sizes and holes. If the
4494 * range spans a memory hole, it is up to the architecture to ensure
4495 * the memory is not freed by the bootmem allocator. If possible
4496 * the range being registered will be merged with existing ranges.
4497 */
4498void __init add_active_range(unsigned int nid, unsigned long start_pfn,
4499 unsigned long end_pfn)
4500{
4501 int i;
4502
Mel Gorman6b74ab92008-07-23 21:26:49 -07004503 mminit_dprintk(MMINIT_TRACE, "memory_register",
4504 "Entering add_active_range(%d, %#lx, %#lx) "
4505 "%d entries of %d used\n",
4506 nid, start_pfn, end_pfn,
4507 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
Mel Gormanc7132162006-09-27 01:49:43 -07004508
Mel Gorman2dbb51c2008-07-23 21:26:52 -07004509 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
4510
Mel Gormanc7132162006-09-27 01:49:43 -07004511 /* Merge with existing active regions if possible */
4512 for (i = 0; i < nr_nodemap_entries; i++) {
4513 if (early_node_map[i].nid != nid)
4514 continue;
4515
4516 /* Skip if an existing region covers this new one */
4517 if (start_pfn >= early_node_map[i].start_pfn &&
4518 end_pfn <= early_node_map[i].end_pfn)
4519 return;
4520
4521 /* Merge forward if suitable */
4522 if (start_pfn <= early_node_map[i].end_pfn &&
4523 end_pfn > early_node_map[i].end_pfn) {
4524 early_node_map[i].end_pfn = end_pfn;
4525 return;
4526 }
4527
4528 /* Merge backward if suitable */
Kazuhisa Ichikawad2dbe082010-01-15 17:01:20 -08004529 if (start_pfn < early_node_map[i].start_pfn &&
Mel Gormanc7132162006-09-27 01:49:43 -07004530 end_pfn >= early_node_map[i].start_pfn) {
4531 early_node_map[i].start_pfn = start_pfn;
4532 return;
4533 }
4534 }
4535
4536 /* Check that early_node_map is large enough */
4537 if (i >= MAX_ACTIVE_REGIONS) {
4538 printk(KERN_CRIT "More than %d memory regions, truncating\n",
4539 MAX_ACTIVE_REGIONS);
4540 return;
4541 }
4542
4543 early_node_map[i].nid = nid;
4544 early_node_map[i].start_pfn = start_pfn;
4545 early_node_map[i].end_pfn = end_pfn;
4546 nr_nodemap_entries = i + 1;
4547}
4548
4549/**
Yinghai Lucc1050b2008-06-13 19:08:52 -07004550 * remove_active_range - Shrink an existing registered range of PFNs
Mel Gormanc7132162006-09-27 01:49:43 -07004551 * @nid: The node id the range is on that should be shrunk
Yinghai Lucc1050b2008-06-13 19:08:52 -07004552 * @start_pfn: The new PFN of the range
4553 * @end_pfn: The new PFN of the range
Mel Gormanc7132162006-09-27 01:49:43 -07004554 *
4555 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004556 * The map is kept near the end physical page range that has already been
4557 * registered. This function allows an arch to shrink an existing registered
4558 * range.
Mel Gormanc7132162006-09-27 01:49:43 -07004559 */
Yinghai Lucc1050b2008-06-13 19:08:52 -07004560void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
4561 unsigned long end_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004562{
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004563 int i, j;
4564 int removed = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07004565
Yinghai Lucc1050b2008-06-13 19:08:52 -07004566 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
4567 nid, start_pfn, end_pfn);
4568
Mel Gormanc7132162006-09-27 01:49:43 -07004569 /* Find the old active region end and shrink */
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004570 for_each_active_range_index_in_nid(i, nid) {
Yinghai Lucc1050b2008-06-13 19:08:52 -07004571 if (early_node_map[i].start_pfn >= start_pfn &&
4572 early_node_map[i].end_pfn <= end_pfn) {
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004573 /* clear it */
Yinghai Lucc1050b2008-06-13 19:08:52 -07004574 early_node_map[i].start_pfn = 0;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004575 early_node_map[i].end_pfn = 0;
4576 removed = 1;
4577 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004578 }
Yinghai Lucc1050b2008-06-13 19:08:52 -07004579 if (early_node_map[i].start_pfn < start_pfn &&
4580 early_node_map[i].end_pfn > start_pfn) {
4581 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
4582 early_node_map[i].end_pfn = start_pfn;
4583 if (temp_end_pfn > end_pfn)
4584 add_active_range(nid, end_pfn, temp_end_pfn);
4585 continue;
4586 }
4587 if (early_node_map[i].start_pfn >= start_pfn &&
4588 early_node_map[i].end_pfn > end_pfn &&
4589 early_node_map[i].start_pfn < end_pfn) {
4590 early_node_map[i].start_pfn = end_pfn;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004591 continue;
4592 }
4593 }
4594
4595 if (!removed)
4596 return;
4597
4598 /* remove the blank ones */
4599 for (i = nr_nodemap_entries - 1; i > 0; i--) {
4600 if (early_node_map[i].nid != nid)
4601 continue;
4602 if (early_node_map[i].end_pfn)
4603 continue;
4604 /* we found it, get rid of it */
4605 for (j = i; j < nr_nodemap_entries - 1; j++)
4606 memcpy(&early_node_map[j], &early_node_map[j+1],
4607 sizeof(early_node_map[j]));
4608 j = nr_nodemap_entries - 1;
4609 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
4610 nr_nodemap_entries--;
4611 }
Mel Gormanc7132162006-09-27 01:49:43 -07004612}
4613
4614/**
4615 * remove_all_active_ranges - Remove all currently registered regions
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004616 *
Mel Gormanc7132162006-09-27 01:49:43 -07004617 * During discovery, it may be found that a table like SRAT is invalid
4618 * and an alternative discovery method must be used. This function removes
4619 * all currently registered regions.
4620 */
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004621void __init remove_all_active_ranges(void)
Mel Gormanc7132162006-09-27 01:49:43 -07004622{
4623 memset(early_node_map, 0, sizeof(early_node_map));
4624 nr_nodemap_entries = 0;
4625}
4626
4627/* Compare two active node_active_regions */
4628static int __init cmp_node_active_region(const void *a, const void *b)
4629{
4630 struct node_active_region *arange = (struct node_active_region *)a;
4631 struct node_active_region *brange = (struct node_active_region *)b;
4632
4633 /* Done this way to avoid overflows */
4634 if (arange->start_pfn > brange->start_pfn)
4635 return 1;
4636 if (arange->start_pfn < brange->start_pfn)
4637 return -1;
4638
4639 return 0;
4640}
4641
4642/* sort the node_map by start_pfn */
Yinghai Lu32996252009-12-15 17:59:02 -08004643void __init sort_node_map(void)
Mel Gormanc7132162006-09-27 01:49:43 -07004644{
4645 sort(early_node_map, (size_t)nr_nodemap_entries,
4646 sizeof(struct node_active_region),
4647 cmp_node_active_region, NULL);
4648}
4649
Mel Gormana6af2bc2007-02-10 01:42:57 -08004650/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004651static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07004652{
4653 int i;
Mel Gormana6af2bc2007-02-10 01:42:57 -08004654 unsigned long min_pfn = ULONG_MAX;
Mel Gorman1abbfb42006-11-23 12:01:41 +00004655
Mel Gormanc7132162006-09-27 01:49:43 -07004656 /* Assuming a sorted map, the first range found has the starting pfn */
4657 for_each_active_range_index_in_nid(i, nid)
Mel Gormana6af2bc2007-02-10 01:42:57 -08004658 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004659
Mel Gormana6af2bc2007-02-10 01:42:57 -08004660 if (min_pfn == ULONG_MAX) {
4661 printk(KERN_WARNING
Paul Jackson2bc0d262008-06-22 07:22:17 -07004662 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08004663 return 0;
4664 }
4665
4666 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004667}
4668
4669/**
4670 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4671 *
4672 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004673 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07004674 */
4675unsigned long __init find_min_pfn_with_active_regions(void)
4676{
4677 return find_min_pfn_for_node(MAX_NUMNODES);
4678}
4679
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004680/*
4681 * early_calculate_totalpages()
4682 * Sum pages in active regions for movable zone.
4683 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4684 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07004685static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07004686{
4687 int i;
4688 unsigned long totalpages = 0;
4689
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004690 for (i = 0; i < nr_nodemap_entries; i++) {
4691 unsigned long pages = early_node_map[i].end_pfn -
Mel Gorman7e63efe2007-07-17 04:03:15 -07004692 early_node_map[i].start_pfn;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004693 totalpages += pages;
4694 if (pages)
4695 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
4696 }
4697 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07004698}
4699
Mel Gorman2a1e2742007-07-17 04:03:12 -07004700/*
4701 * Find the PFN the Movable zone begins in each node. Kernel memory
4702 * is spread evenly between nodes as long as the nodes have enough
4703 * memory. When they don't, some nodes will have more kernelcore than
4704 * others
4705 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004706static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004707{
4708 int i, nid;
4709 unsigned long usable_startpfn;
4710 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07004711 /* save the state before borrow the nodemask */
4712 nodemask_t saved_node_state = node_states[N_HIGH_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004713 unsigned long totalpages = early_calculate_totalpages();
4714 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004715
Larry Bassele6436862011-10-14 10:59:07 -07004716#ifdef CONFIG_FIX_MOVABLE_ZONE
4717 required_movablecore = movable_reserved_size >> PAGE_SHIFT;
4718#endif
Mel Gorman7e63efe2007-07-17 04:03:15 -07004719 /*
4720 * If movablecore was specified, calculate what size of
4721 * kernelcore that corresponds so that memory usable for
4722 * any allocation type is evenly spread. If both kernelcore
4723 * and movablecore are specified, then the value of kernelcore
4724 * will be used for required_kernelcore if it's greater than
4725 * what movablecore would have allowed.
4726 */
4727 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07004728 unsigned long corepages;
4729
4730 /*
4731 * Round-up so that ZONE_MOVABLE is at least as large as what
4732 * was requested by the user
4733 */
4734 required_movablecore =
4735 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
4736 corepages = totalpages - required_movablecore;
4737
4738 required_kernelcore = max(required_kernelcore, corepages);
4739 }
4740
Mel Gorman2a1e2742007-07-17 04:03:12 -07004741 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4742 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07004743 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004744
4745 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4746 find_usable_zone_for_movable();
4747 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4748
4749restart:
4750 /* Spread kernelcore memory as evenly as possible throughout nodes */
4751 kernelcore_node = required_kernelcore / usable_nodes;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004752 for_each_node_state(nid, N_HIGH_MEMORY) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004753 /*
4754 * Recalculate kernelcore_node if the division per node
4755 * now exceeds what is necessary to satisfy the requested
4756 * amount of memory for the kernel
4757 */
4758 if (required_kernelcore < kernelcore_node)
4759 kernelcore_node = required_kernelcore / usable_nodes;
4760
4761 /*
4762 * As the map is walked, we track how much memory is usable
4763 * by the kernel using kernelcore_remaining. When it is
4764 * 0, the rest of the node is usable by ZONE_MOVABLE
4765 */
4766 kernelcore_remaining = kernelcore_node;
4767
4768 /* Go through each range of PFNs within this node */
4769 for_each_active_range_index_in_nid(i, nid) {
4770 unsigned long start_pfn, end_pfn;
4771 unsigned long size_pages;
4772
4773 start_pfn = max(early_node_map[i].start_pfn,
4774 zone_movable_pfn[nid]);
4775 end_pfn = early_node_map[i].end_pfn;
4776 if (start_pfn >= end_pfn)
4777 continue;
4778
4779 /* Account for what is only usable for kernelcore */
4780 if (start_pfn < usable_startpfn) {
4781 unsigned long kernel_pages;
4782 kernel_pages = min(end_pfn, usable_startpfn)
4783 - start_pfn;
4784
4785 kernelcore_remaining -= min(kernel_pages,
4786 kernelcore_remaining);
4787 required_kernelcore -= min(kernel_pages,
4788 required_kernelcore);
4789
4790 /* Continue if range is now fully accounted */
4791 if (end_pfn <= usable_startpfn) {
4792
4793 /*
4794 * Push zone_movable_pfn to the end so
4795 * that if we have to rebalance
4796 * kernelcore across nodes, we will
4797 * not double account here
4798 */
4799 zone_movable_pfn[nid] = end_pfn;
4800 continue;
4801 }
4802 start_pfn = usable_startpfn;
4803 }
4804
4805 /*
4806 * The usable PFN range for ZONE_MOVABLE is from
4807 * start_pfn->end_pfn. Calculate size_pages as the
4808 * number of pages used as kernelcore
4809 */
4810 size_pages = end_pfn - start_pfn;
4811 if (size_pages > kernelcore_remaining)
4812 size_pages = kernelcore_remaining;
4813 zone_movable_pfn[nid] = start_pfn + size_pages;
4814
4815 /*
4816 * Some kernelcore has been met, update counts and
4817 * break if the kernelcore for this node has been
4818 * satisified
4819 */
4820 required_kernelcore -= min(required_kernelcore,
4821 size_pages);
4822 kernelcore_remaining -= size_pages;
4823 if (!kernelcore_remaining)
4824 break;
4825 }
4826 }
4827
4828 /*
4829 * If there is still required_kernelcore, we do another pass with one
4830 * less node in the count. This will push zone_movable_pfn[nid] further
4831 * along on the nodes that still have memory until kernelcore is
4832 * satisified
4833 */
4834 usable_nodes--;
4835 if (usable_nodes && required_kernelcore > usable_nodes)
4836 goto restart;
4837
4838 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4839 for (nid = 0; nid < MAX_NUMNODES; nid++)
4840 zone_movable_pfn[nid] =
4841 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07004842
4843out:
4844 /* restore the node_state */
4845 node_states[N_HIGH_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004846}
4847
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004848/* Any regular memory on that node ? */
4849static void check_for_regular_memory(pg_data_t *pgdat)
4850{
4851#ifdef CONFIG_HIGHMEM
4852 enum zone_type zone_type;
4853
4854 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4855 struct zone *zone = &pgdat->node_zones[zone_type];
4856 if (zone->present_pages)
4857 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4858 }
4859#endif
4860}
4861
Mel Gormanc7132162006-09-27 01:49:43 -07004862/**
4863 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004864 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07004865 *
4866 * This will call free_area_init_node() for each active node in the system.
4867 * Using the page ranges provided by add_active_range(), the size of each
4868 * zone in each node and their holes is calculated. If the maximum PFN
4869 * between two adjacent zones match, it is assumed that the zone is empty.
4870 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4871 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4872 * starts where the previous one ended. For example, ZONE_DMA32 starts
4873 * at arch_max_dma_pfn.
4874 */
4875void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4876{
4877 unsigned long nid;
Andrew Mortondb991002008-10-15 22:01:07 -07004878 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004879
Mel Gormana6af2bc2007-02-10 01:42:57 -08004880 /* Sort early_node_map as initialisation assumes it is sorted */
4881 sort_node_map();
4882
Mel Gormanc7132162006-09-27 01:49:43 -07004883 /* Record where the zone boundaries are */
4884 memset(arch_zone_lowest_possible_pfn, 0,
4885 sizeof(arch_zone_lowest_possible_pfn));
4886 memset(arch_zone_highest_possible_pfn, 0,
4887 sizeof(arch_zone_highest_possible_pfn));
4888 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4889 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4890 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004891 if (i == ZONE_MOVABLE)
4892 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004893 arch_zone_lowest_possible_pfn[i] =
4894 arch_zone_highest_possible_pfn[i-1];
4895 arch_zone_highest_possible_pfn[i] =
4896 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4897 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07004898 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4899 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4900
4901 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4902 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4903 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004904
Mel Gormanc7132162006-09-27 01:49:43 -07004905 /* Print out the zone ranges */
4906 printk("Zone PFN ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07004907 for (i = 0; i < MAX_NR_ZONES; i++) {
4908 if (i == ZONE_MOVABLE)
4909 continue;
David Rientjes72f0ba02010-03-05 13:42:14 -08004910 printk(" %-8s ", zone_names[i]);
4911 if (arch_zone_lowest_possible_pfn[i] ==
4912 arch_zone_highest_possible_pfn[i])
4913 printk("empty\n");
4914 else
4915 printk("%0#10lx -> %0#10lx\n",
Mel Gormanc7132162006-09-27 01:49:43 -07004916 arch_zone_lowest_possible_pfn[i],
4917 arch_zone_highest_possible_pfn[i]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004918 }
4919
4920 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4921 printk("Movable zone start PFN for each node\n");
4922 for (i = 0; i < MAX_NUMNODES; i++) {
4923 if (zone_movable_pfn[i])
4924 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4925 }
Mel Gormanc7132162006-09-27 01:49:43 -07004926
4927 /* Print out the early_node_map[] */
4928 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4929 for (i = 0; i < nr_nodemap_entries; i++)
Paul Jackson5dab8ec2008-06-25 05:44:40 -07004930 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004931 early_node_map[i].start_pfn,
4932 early_node_map[i].end_pfn);
4933
4934 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07004935 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08004936 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07004937 for_each_online_node(nid) {
4938 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07004939 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07004940 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004941
4942 /* Any memory on that node */
4943 if (pgdat->node_present_pages)
4944 node_set_state(nid, N_HIGH_MEMORY);
4945 check_for_regular_memory(pgdat);
Mel Gormanc7132162006-09-27 01:49:43 -07004946 }
4947}
Mel Gorman2a1e2742007-07-17 04:03:12 -07004948
Mel Gorman7e63efe2007-07-17 04:03:15 -07004949static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004950{
4951 unsigned long long coremem;
4952 if (!p)
4953 return -EINVAL;
4954
4955 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004956 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004957
Mel Gorman7e63efe2007-07-17 04:03:15 -07004958 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07004959 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4960
4961 return 0;
4962}
Mel Gormaned7ed362007-07-17 04:03:14 -07004963
Mel Gorman7e63efe2007-07-17 04:03:15 -07004964/*
4965 * kernelcore=size sets the amount of memory for use for allocations that
4966 * cannot be reclaimed or migrated.
4967 */
4968static int __init cmdline_parse_kernelcore(char *p)
4969{
4970 return cmdline_parse_core(p, &required_kernelcore);
4971}
4972
4973/*
4974 * movablecore=size sets the amount of memory for use for allocations that
4975 * can be reclaimed or migrated.
4976 */
4977static int __init cmdline_parse_movablecore(char *p)
4978{
4979 return cmdline_parse_core(p, &required_movablecore);
4980}
4981
Mel Gormaned7ed362007-07-17 04:03:14 -07004982early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004983early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07004984
Mel Gormanc7132162006-09-27 01:49:43 -07004985#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4986
Mel Gorman0e0b8642006-09-27 01:49:56 -07004987/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004988 * set_dma_reserve - set the specified number of pages reserved in the first zone
4989 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07004990 *
4991 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4992 * In the DMA zone, a significant percentage may be consumed by kernel image
4993 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004994 * function may optionally be used to account for unfreeable pages in the
4995 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4996 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07004997 */
4998void __init set_dma_reserve(unsigned long new_dma_reserve)
4999{
5000 dma_reserve = new_dma_reserve;
5001}
5002
Linus Torvalds1da177e2005-04-16 15:20:36 -07005003void __init free_area_init(unsigned long *zones_size)
5004{
Johannes Weiner9109fb72008-07-23 21:27:20 -07005005 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005006 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
5007}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005008
Linus Torvalds1da177e2005-04-16 15:20:36 -07005009static int page_alloc_cpu_notify(struct notifier_block *self,
5010 unsigned long action, void *hcpu)
5011{
5012 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005013
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07005014 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005015 drain_pages(cpu);
5016
5017 /*
5018 * Spill the event counters of the dead processor
5019 * into the current processors event counters.
5020 * This artificially elevates the count of the current
5021 * processor.
5022 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07005023 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005024
5025 /*
5026 * Zero the differential counters of the dead processor
5027 * so that the vm statistics are consistent.
5028 *
5029 * This is only okay since the processor is dead and cannot
5030 * race with what we are doing.
5031 */
Christoph Lameter2244b952006-06-30 01:55:33 -07005032 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005033 }
5034 return NOTIFY_OK;
5035}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005036
5037void __init page_alloc_init(void)
5038{
5039 hotcpu_notifier(page_alloc_cpu_notify, 0);
5040}
5041
5042/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005043 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5044 * or min_free_kbytes changes.
5045 */
5046static void calculate_totalreserve_pages(void)
5047{
5048 struct pglist_data *pgdat;
5049 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005050 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005051
5052 for_each_online_pgdat(pgdat) {
5053 for (i = 0; i < MAX_NR_ZONES; i++) {
5054 struct zone *zone = pgdat->node_zones + i;
5055 unsigned long max = 0;
5056
5057 /* Find valid and maximum lowmem_reserve in the zone */
5058 for (j = i; j < MAX_NR_ZONES; j++) {
5059 if (zone->lowmem_reserve[j] > max)
5060 max = zone->lowmem_reserve[j];
5061 }
5062
Mel Gorman41858962009-06-16 15:32:12 -07005063 /* we treat the high watermark as reserved pages. */
5064 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005065
5066 if (max > zone->present_pages)
5067 max = zone->present_pages;
5068 reserve_pages += max;
5069 }
5070 }
5071 totalreserve_pages = reserve_pages;
5072}
5073
5074/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07005075 * setup_per_zone_lowmem_reserve - called whenever
5076 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
5077 * has a correct pages reserved value, so an adequate number of
5078 * pages are left in the zone after a successful __alloc_pages().
5079 */
5080static void setup_per_zone_lowmem_reserve(void)
5081{
5082 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005083 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005084
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08005085 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005086 for (j = 0; j < MAX_NR_ZONES; j++) {
5087 struct zone *zone = pgdat->node_zones + j;
5088 unsigned long present_pages = zone->present_pages;
5089
5090 zone->lowmem_reserve[j] = 0;
5091
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005092 idx = j;
5093 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005094 struct zone *lower_zone;
5095
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005096 idx--;
5097
Linus Torvalds1da177e2005-04-16 15:20:36 -07005098 if (sysctl_lowmem_reserve_ratio[idx] < 1)
5099 sysctl_lowmem_reserve_ratio[idx] = 1;
5100
5101 lower_zone = pgdat->node_zones + idx;
5102 lower_zone->lowmem_reserve[j] = present_pages /
5103 sysctl_lowmem_reserve_ratio[idx];
5104 present_pages += lower_zone->present_pages;
5105 }
5106 }
5107 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005108
5109 /* update totalreserve_pages */
5110 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005111}
5112
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005113/**
Minchan Kimbc75d332009-06-16 15:32:48 -07005114 * setup_per_zone_wmarks - called when min_free_kbytes changes
Minchan Kimbce73942009-06-16 15:32:50 -07005115 * or when memory is hot-{added|removed}
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005116 *
Minchan Kimbc75d332009-06-16 15:32:48 -07005117 * Ensures that the watermark[min,low,high] values for each zone are set
5118 * correctly with respect to min_free_kbytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005119 */
Minchan Kimbc75d332009-06-16 15:32:48 -07005120void setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005121{
5122 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
5123 unsigned long lowmem_pages = 0;
5124 struct zone *zone;
5125 unsigned long flags;
5126
5127 /* Calculate total number of !ZONE_HIGHMEM pages */
5128 for_each_zone(zone) {
5129 if (!is_highmem(zone))
5130 lowmem_pages += zone->present_pages;
5131 }
5132
5133 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07005134 u64 tmp;
5135
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005136 spin_lock_irqsave(&zone->lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07005137 tmp = (u64)pages_min * zone->present_pages;
5138 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005139 if (is_highmem(zone)) {
5140 /*
Nick Piggin669ed172005-11-13 16:06:45 -08005141 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5142 * need highmem pages, so cap pages_min to a small
5143 * value here.
5144 *
Mel Gorman41858962009-06-16 15:32:12 -07005145 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Nick Piggin669ed172005-11-13 16:06:45 -08005146 * deltas controls asynch page reclaim, and so should
5147 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005148 */
5149 int min_pages;
5150
5151 min_pages = zone->present_pages / 1024;
5152 if (min_pages < SWAP_CLUSTER_MAX)
5153 min_pages = SWAP_CLUSTER_MAX;
5154 if (min_pages > 128)
5155 min_pages = 128;
Mel Gorman41858962009-06-16 15:32:12 -07005156 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005157 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08005158 /*
5159 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07005160 * proportionate to the zone's size.
5161 */
Mel Gorman41858962009-06-16 15:32:12 -07005162 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005163 }
5164
Mel Gorman41858962009-06-16 15:32:12 -07005165 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
5166 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Mel Gorman56fd56b2007-10-16 01:25:58 -07005167 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005168 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005169 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005170
5171 /* update totalreserve_pages */
5172 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005173}
5174
Randy Dunlap55a44622009-09-21 17:01:20 -07005175/*
Rik van Riel556adec2008-10-18 20:26:34 -07005176 * The inactive anon list should be small enough that the VM never has to
5177 * do too much work, but large enough that each inactive page has a chance
5178 * to be referenced again before it is swapped out.
5179 *
5180 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5181 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5182 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5183 * the anonymous pages are kept on the inactive list.
5184 *
5185 * total target max
5186 * memory ratio inactive anon
5187 * -------------------------------------
5188 * 10MB 1 5MB
5189 * 100MB 1 50MB
5190 * 1GB 3 250MB
5191 * 10GB 10 0.9GB
5192 * 100GB 31 3GB
5193 * 1TB 101 10GB
5194 * 10TB 320 32GB
5195 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005196static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
Minchan Kim96cb4df2009-06-16 15:32:49 -07005197{
5198 unsigned int gb, ratio;
5199
5200 /* Zone size in gigabytes */
5201 gb = zone->present_pages >> (30 - PAGE_SHIFT);
5202 if (gb)
5203 ratio = int_sqrt(10 * gb);
5204 else
5205 ratio = 1;
5206
5207 zone->inactive_ratio = ratio;
5208}
5209
KOSAKI Motohiro839a4fc2011-05-24 17:11:31 -07005210static void __meminit setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07005211{
5212 struct zone *zone;
5213
Minchan Kim96cb4df2009-06-16 15:32:49 -07005214 for_each_zone(zone)
5215 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07005216}
5217
Linus Torvalds1da177e2005-04-16 15:20:36 -07005218/*
5219 * Initialise min_free_kbytes.
5220 *
5221 * For small machines we want it small (128k min). For large machines
5222 * we want it large (64MB max). But it is not linear, because network
5223 * bandwidth does not increase linearly with machine size. We use
5224 *
5225 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5226 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5227 *
5228 * which yields
5229 *
5230 * 16MB: 512k
5231 * 32MB: 724k
5232 * 64MB: 1024k
5233 * 128MB: 1448k
5234 * 256MB: 2048k
5235 * 512MB: 2896k
5236 * 1024MB: 4096k
5237 * 2048MB: 5792k
5238 * 4096MB: 8192k
5239 * 8192MB: 11584k
5240 * 16384MB: 16384k
5241 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005242int __meminit init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005243{
5244 unsigned long lowmem_kbytes;
5245
5246 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
5247
5248 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
5249 if (min_free_kbytes < 128)
5250 min_free_kbytes = 128;
5251 if (min_free_kbytes > 65536)
5252 min_free_kbytes = 65536;
Minchan Kimbc75d332009-06-16 15:32:48 -07005253 setup_per_zone_wmarks();
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -07005254 refresh_zone_stat_thresholds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005255 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07005256 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005257 return 0;
5258}
Minchan Kimbc75d332009-06-16 15:32:48 -07005259module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005260
5261/*
5262 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5263 * that we can call two helper functions whenever min_free_kbytes
5264 * changes.
5265 */
5266int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005267 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005268{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005269 proc_dointvec(table, write, buffer, length, ppos);
Mel Gorman3b1d92c2007-05-06 14:49:30 -07005270 if (write)
Minchan Kimbc75d332009-06-16 15:32:48 -07005271 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005272 return 0;
5273}
5274
Christoph Lameter96146342006-07-03 00:24:13 -07005275#ifdef CONFIG_NUMA
5276int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005277 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07005278{
5279 struct zone *zone;
5280 int rc;
5281
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005282 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07005283 if (rc)
5284 return rc;
5285
5286 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07005287 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07005288 sysctl_min_unmapped_ratio) / 100;
5289 return 0;
5290}
Christoph Lameter0ff38492006-09-25 23:31:52 -07005291
5292int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005293 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07005294{
5295 struct zone *zone;
5296 int rc;
5297
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005298 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07005299 if (rc)
5300 return rc;
5301
5302 for_each_zone(zone)
5303 zone->min_slab_pages = (zone->present_pages *
5304 sysctl_min_slab_ratio) / 100;
5305 return 0;
5306}
Christoph Lameter96146342006-07-03 00:24:13 -07005307#endif
5308
Linus Torvalds1da177e2005-04-16 15:20:36 -07005309/*
5310 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5311 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5312 * whenever sysctl_lowmem_reserve_ratio changes.
5313 *
5314 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07005315 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07005316 * if in function of the boot time zone sizes.
5317 */
5318int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005319 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005320{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005321 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005322 setup_per_zone_lowmem_reserve();
5323 return 0;
5324}
5325
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005326/*
5327 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5328 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5329 * can have before it gets flushed back to buddy allocator.
5330 */
5331
5332int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005333 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005334{
5335 struct zone *zone;
5336 unsigned int cpu;
5337 int ret;
5338
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005339 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005340 if (!write || (ret == -EINVAL))
5341 return ret;
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07005342 for_each_populated_zone(zone) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005343 for_each_possible_cpu(cpu) {
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005344 unsigned long high;
5345 high = zone->present_pages / percpu_pagelist_fraction;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005346 setup_pagelist_highmark(
5347 per_cpu_ptr(zone->pageset, cpu), high);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005348 }
5349 }
5350 return 0;
5351}
5352
David S. Millerf034b5d2006-08-24 03:08:07 -07005353int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005354
5355#ifdef CONFIG_NUMA
5356static int __init set_hashdist(char *str)
5357{
5358 if (!str)
5359 return 0;
5360 hashdist = simple_strtoul(str, &str, 0);
5361 return 1;
5362}
5363__setup("hashdist=", set_hashdist);
5364#endif
5365
5366/*
5367 * allocate a large system hash table from bootmem
5368 * - it is assumed that the hash table must contain an exact power-of-2
5369 * quantity of entries
5370 * - limit is the number of hash buckets, not the total allocation size
5371 */
5372void *__init alloc_large_system_hash(const char *tablename,
5373 unsigned long bucketsize,
5374 unsigned long numentries,
5375 int scale,
5376 int flags,
5377 unsigned int *_hash_shift,
5378 unsigned int *_hash_mask,
5379 unsigned long limit)
5380{
5381 unsigned long long max = limit;
5382 unsigned long log2qty, size;
5383 void *table = NULL;
5384
5385 /* allow the kernel cmdline to have a say */
5386 if (!numentries) {
5387 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08005388 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005389 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
5390 numentries >>= 20 - PAGE_SHIFT;
5391 numentries <<= 20 - PAGE_SHIFT;
5392
5393 /* limit to 1 bucket per 2^scale bytes of low memory */
5394 if (scale > PAGE_SHIFT)
5395 numentries >>= (scale - PAGE_SHIFT);
5396 else
5397 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08005398
5399 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07005400 if (unlikely(flags & HASH_SMALL)) {
5401 /* Makes no sense without HASH_EARLY */
5402 WARN_ON(!(flags & HASH_EARLY));
5403 if (!(numentries >> *_hash_shift)) {
5404 numentries = 1UL << *_hash_shift;
5405 BUG_ON(!numentries);
5406 }
5407 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08005408 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005409 }
John Hawkes6e692ed2006-03-25 03:08:02 -08005410 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005411
5412 /* limit allocation size to 1/16 total memory by default */
5413 if (max == 0) {
5414 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
5415 do_div(max, bucketsize);
5416 }
5417
5418 if (numentries > max)
5419 numentries = max;
5420
David Howellsf0d1b0b2006-12-08 02:37:49 -08005421 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005422
5423 do {
5424 size = bucketsize << log2qty;
5425 if (flags & HASH_EARLY)
Jan Beulich74768ed2008-08-12 15:08:39 -07005426 table = alloc_bootmem_nopanic(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005427 else if (hashdist)
5428 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
5429 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07005430 /*
5431 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07005432 * some pages at the end of hash table which
5433 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07005434 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005435 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07005436 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005437 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
5438 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005439 }
5440 } while (!table && size > PAGE_SIZE && --log2qty);
5441
5442 if (!table)
5443 panic("Failed to allocate %s hash table\n", tablename);
5444
Robin Holtf241e662010-10-07 12:59:26 -07005445 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07005446 tablename,
Robin Holtf241e662010-10-07 12:59:26 -07005447 (1UL << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08005448 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005449 size);
5450
5451 if (_hash_shift)
5452 *_hash_shift = log2qty;
5453 if (_hash_mask)
5454 *_hash_mask = (1 << log2qty) - 1;
5455
5456 return table;
5457}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08005458
Mel Gorman835c1342007-10-16 01:25:47 -07005459/* Return a pointer to the bitmap storing bits affecting a block of pages */
5460static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
5461 unsigned long pfn)
5462{
5463#ifdef CONFIG_SPARSEMEM
5464 return __pfn_to_section(pfn)->pageblock_flags;
5465#else
5466 return zone->pageblock_flags;
5467#endif /* CONFIG_SPARSEMEM */
5468}
Andrew Morton6220ec72006-10-19 23:29:05 -07005469
Mel Gorman835c1342007-10-16 01:25:47 -07005470static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
5471{
5472#ifdef CONFIG_SPARSEMEM
5473 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07005474 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005475#else
5476 pfn = pfn - zone->zone_start_pfn;
Mel Gormand9c23402007-10-16 01:26:01 -07005477 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005478#endif /* CONFIG_SPARSEMEM */
5479}
5480
5481/**
Mel Gormand9c23402007-10-16 01:26:01 -07005482 * 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 -07005483 * @page: The page within the block of interest
5484 * @start_bitidx: The first bit of interest to retrieve
5485 * @end_bitidx: The last bit of interest
5486 * returns pageblock_bits flags
5487 */
5488unsigned long get_pageblock_flags_group(struct page *page,
5489 int start_bitidx, int end_bitidx)
5490{
5491 struct zone *zone;
5492 unsigned long *bitmap;
5493 unsigned long pfn, bitidx;
5494 unsigned long flags = 0;
5495 unsigned long value = 1;
5496
5497 zone = page_zone(page);
5498 pfn = page_to_pfn(page);
5499 bitmap = get_pageblock_bitmap(zone, pfn);
5500 bitidx = pfn_to_bitidx(zone, pfn);
5501
5502 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5503 if (test_bit(bitidx + start_bitidx, bitmap))
5504 flags |= value;
5505
5506 return flags;
5507}
5508
5509/**
Mel Gormand9c23402007-10-16 01:26:01 -07005510 * 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 -07005511 * @page: The page within the block of interest
5512 * @start_bitidx: The first bit of interest
5513 * @end_bitidx: The last bit of interest
5514 * @flags: The flags to set
5515 */
5516void set_pageblock_flags_group(struct page *page, unsigned long flags,
5517 int start_bitidx, int end_bitidx)
5518{
5519 struct zone *zone;
5520 unsigned long *bitmap;
5521 unsigned long pfn, bitidx;
5522 unsigned long value = 1;
5523
5524 zone = page_zone(page);
5525 pfn = page_to_pfn(page);
5526 bitmap = get_pageblock_bitmap(zone, pfn);
5527 bitidx = pfn_to_bitidx(zone, pfn);
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07005528 VM_BUG_ON(pfn < zone->zone_start_pfn);
5529 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
Mel Gorman835c1342007-10-16 01:25:47 -07005530
5531 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5532 if (flags & value)
5533 __set_bit(bitidx + start_bitidx, bitmap);
5534 else
5535 __clear_bit(bitidx + start_bitidx, bitmap);
5536}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005537
5538/*
5539 * This is designed as sub function...plz see page_isolation.c also.
5540 * set/clear page block's type to be ISOLATE.
5541 * page allocater never alloc memory from ISOLATE block.
5542 */
5543
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005544static int
5545__count_immobile_pages(struct zone *zone, struct page *page, int count)
5546{
5547 unsigned long pfn, iter, found;
5548 /*
5549 * For avoiding noise data, lru_add_drain_all() should be called
5550 * If ZONE_MOVABLE, the zone never contains immobile pages
5551 */
5552 if (zone_idx(zone) == ZONE_MOVABLE)
5553 return true;
5554
5555 if (get_pageblock_migratetype(page) == MIGRATE_MOVABLE)
5556 return true;
5557
5558 pfn = page_to_pfn(page);
5559 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
5560 unsigned long check = pfn + iter;
5561
Namhyung Kim29723fc2011-02-25 14:44:25 -08005562 if (!pfn_valid_within(check))
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005563 continue;
Namhyung Kim29723fc2011-02-25 14:44:25 -08005564
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005565 page = pfn_to_page(check);
5566 if (!page_count(page)) {
5567 if (PageBuddy(page))
5568 iter += (1 << page_order(page)) - 1;
5569 continue;
5570 }
5571 if (!PageLRU(page))
5572 found++;
5573 /*
5574 * If there are RECLAIMABLE pages, we need to check it.
5575 * But now, memory offline itself doesn't call shrink_slab()
5576 * and it still to be fixed.
5577 */
5578 /*
5579 * If the page is not RAM, page_count()should be 0.
5580 * we don't need more check. This is an _used_ not-movable page.
5581 *
5582 * The problematic thing here is PG_reserved pages. PG_reserved
5583 * is set to both of a memory hole page and a _used_ kernel
5584 * page at boot.
5585 */
5586 if (found > count)
5587 return false;
5588 }
5589 return true;
5590}
5591
5592bool is_pageblock_removable_nolock(struct page *page)
5593{
5594 struct zone *zone = page_zone(page);
5595 return __count_immobile_pages(zone, page, 0);
5596}
5597
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005598int set_migratetype_isolate(struct page *page)
5599{
5600 struct zone *zone;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005601 unsigned long flags, pfn;
Robert Jennings925cc712009-12-17 14:44:38 +00005602 struct memory_isolate_notify arg;
5603 int notifier_ret;
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005604 int ret = -EBUSY;
5605
5606 zone = page_zone(page);
Robert Jennings925cc712009-12-17 14:44:38 +00005607
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005608 spin_lock_irqsave(&zone->lock, flags);
Robert Jennings925cc712009-12-17 14:44:38 +00005609
5610 pfn = page_to_pfn(page);
5611 arg.start_pfn = pfn;
5612 arg.nr_pages = pageblock_nr_pages;
5613 arg.pages_found = 0;
5614
5615 /*
5616 * It may be possible to isolate a pageblock even if the
5617 * migratetype is not MIGRATE_MOVABLE. The memory isolation
5618 * notifier chain is used by balloon drivers to return the
5619 * number of pages in a range that are held by the balloon
5620 * driver to shrink memory. If all the pages are accounted for
5621 * by balloons, are free, or on the LRU, isolation can continue.
5622 * Later, for example, when memory hotplug notifier runs, these
5623 * pages reported as "can be isolated" should be isolated(freed)
5624 * by the balloon driver through the memory notifier chain.
5625 */
5626 notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg);
5627 notifier_ret = notifier_to_errno(notifier_ret);
KAMEZAWA Hiroyuki4b204772010-10-26 14:21:29 -07005628 if (notifier_ret)
Robert Jennings925cc712009-12-17 14:44:38 +00005629 goto out;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005630 /*
5631 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
5632 * We just check MOVABLE pages.
5633 */
5634 if (__count_immobile_pages(zone, page, arg.pages_found))
Robert Jennings925cc712009-12-17 14:44:38 +00005635 ret = 0;
5636
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005637 /*
5638 * immobile means "not-on-lru" paes. If immobile is larger than
5639 * removable-by-driver pages reported by notifier, we'll fail.
5640 */
5641
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005642out:
Robert Jennings925cc712009-12-17 14:44:38 +00005643 if (!ret) {
5644 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
5645 move_freepages_block(zone, page, MIGRATE_ISOLATE);
5646 }
5647
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005648 spin_unlock_irqrestore(&zone->lock, flags);
5649 if (!ret)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005650 drain_all_pages();
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005651 return ret;
5652}
5653
5654void unset_migratetype_isolate(struct page *page)
5655{
5656 struct zone *zone;
5657 unsigned long flags;
5658 zone = page_zone(page);
5659 spin_lock_irqsave(&zone->lock, flags);
5660 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
5661 goto out;
5662 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5663 move_freepages_block(zone, page, MIGRATE_MOVABLE);
5664out:
5665 spin_unlock_irqrestore(&zone->lock, flags);
5666}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07005667
5668#ifdef CONFIG_MEMORY_HOTREMOVE
5669/*
5670 * All pages in the range must be isolated before calling this.
5671 */
5672void
5673__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
5674{
5675 struct page *page;
5676 struct zone *zone;
5677 int order, i;
5678 unsigned long pfn;
5679 unsigned long flags;
5680 /* find the first valid pfn */
5681 for (pfn = start_pfn; pfn < end_pfn; pfn++)
5682 if (pfn_valid(pfn))
5683 break;
5684 if (pfn == end_pfn)
5685 return;
5686 zone = page_zone(pfn_to_page(pfn));
5687 spin_lock_irqsave(&zone->lock, flags);
5688 pfn = start_pfn;
5689 while (pfn < end_pfn) {
5690 if (!pfn_valid(pfn)) {
5691 pfn++;
5692 continue;
5693 }
5694 page = pfn_to_page(pfn);
5695 BUG_ON(page_count(page));
5696 BUG_ON(!PageBuddy(page));
5697 order = page_order(page);
5698#ifdef CONFIG_DEBUG_VM
5699 printk(KERN_INFO "remove from free list %lx %d %lx\n",
5700 pfn, 1 << order, end_pfn);
5701#endif
5702 list_del(&page->lru);
5703 rmv_page_order(page);
5704 zone->free_area[order].nr_free--;
5705 __mod_zone_page_state(zone, NR_FREE_PAGES,
5706 - (1UL << order));
5707 for (i = 0; i < (1 << order); i++)
5708 SetPageReserved((page+i));
5709 pfn += (1 << order);
5710 }
5711 spin_unlock_irqrestore(&zone->lock, flags);
5712}
5713#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01005714
5715#ifdef CONFIG_MEMORY_FAILURE
5716bool is_free_buddy_page(struct page *page)
5717{
5718 struct zone *zone = page_zone(page);
5719 unsigned long pfn = page_to_pfn(page);
5720 unsigned long flags;
5721 int order;
5722
5723 spin_lock_irqsave(&zone->lock, flags);
5724 for (order = 0; order < MAX_ORDER; order++) {
5725 struct page *page_head = page - (pfn & ((1 << order) - 1));
5726
5727 if (PageBuddy(page_head) && page_order(page_head) >= order)
5728 break;
5729 }
5730 spin_unlock_irqrestore(&zone->lock, flags);
5731
5732 return order < MAX_ORDER;
5733}
5734#endif
Wu Fengguang718a3822010-03-10 15:20:43 -08005735
5736static struct trace_print_flags pageflag_names[] = {
5737 {1UL << PG_locked, "locked" },
5738 {1UL << PG_error, "error" },
5739 {1UL << PG_referenced, "referenced" },
5740 {1UL << PG_uptodate, "uptodate" },
5741 {1UL << PG_dirty, "dirty" },
5742 {1UL << PG_lru, "lru" },
5743 {1UL << PG_active, "active" },
5744 {1UL << PG_slab, "slab" },
5745 {1UL << PG_owner_priv_1, "owner_priv_1" },
5746 {1UL << PG_arch_1, "arch_1" },
5747 {1UL << PG_reserved, "reserved" },
5748 {1UL << PG_private, "private" },
5749 {1UL << PG_private_2, "private_2" },
5750 {1UL << PG_writeback, "writeback" },
5751#ifdef CONFIG_PAGEFLAGS_EXTENDED
5752 {1UL << PG_head, "head" },
5753 {1UL << PG_tail, "tail" },
5754#else
5755 {1UL << PG_compound, "compound" },
5756#endif
5757 {1UL << PG_swapcache, "swapcache" },
5758 {1UL << PG_mappedtodisk, "mappedtodisk" },
5759 {1UL << PG_reclaim, "reclaim" },
Wu Fengguang718a3822010-03-10 15:20:43 -08005760 {1UL << PG_swapbacked, "swapbacked" },
5761 {1UL << PG_unevictable, "unevictable" },
5762#ifdef CONFIG_MMU
5763 {1UL << PG_mlocked, "mlocked" },
5764#endif
5765#ifdef CONFIG_ARCH_USES_PG_UNCACHED
5766 {1UL << PG_uncached, "uncached" },
5767#endif
5768#ifdef CONFIG_MEMORY_FAILURE
5769 {1UL << PG_hwpoison, "hwpoison" },
5770#endif
5771 {-1UL, NULL },
5772};
5773
5774static void dump_page_flags(unsigned long flags)
5775{
5776 const char *delim = "";
5777 unsigned long mask;
5778 int i;
5779
5780 printk(KERN_ALERT "page flags: %#lx(", flags);
5781
5782 /* remove zone id */
5783 flags &= (1UL << NR_PAGEFLAGS) - 1;
5784
5785 for (i = 0; pageflag_names[i].name && flags; i++) {
5786
5787 mask = pageflag_names[i].mask;
5788 if ((flags & mask) != mask)
5789 continue;
5790
5791 flags &= ~mask;
5792 printk("%s%s", delim, pageflag_names[i].name);
5793 delim = "|";
5794 }
5795
5796 /* check for left over flags */
5797 if (flags)
5798 printk("%s%#lx", delim, flags);
5799
5800 printk(")\n");
5801}
5802
5803void dump_page(struct page *page)
5804{
5805 printk(KERN_ALERT
5806 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
Andrea Arcangeli4e9f64c2011-01-13 15:46:29 -08005807 page, atomic_read(&page->_count), page_mapcount(page),
Wu Fengguang718a3822010-03-10 15:20:43 -08005808 page->mapping, page->index);
5809 dump_page_flags(page->flags);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07005810 mem_cgroup_print_bad_page(page);
Wu Fengguang718a3822010-03-10 15:20:43 -08005811}