blob: 2b8ba3aebf6e2c6b46b0d12dfea058ee3ab022fe [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/page_alloc.c
3 *
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
6 *
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
15 */
16
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/stddef.h>
18#include <linux/mm.h>
19#include <linux/swap.h>
20#include <linux/interrupt.h>
21#include <linux/pagemap.h>
KOSAKI Motohiro10ed2732008-03-04 14:28:32 -080022#include <linux/jiffies.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/bootmem.h>
Yinghai Luedbe7d22010-08-25 13:39:16 -070024#include <linux/memblock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/compiler.h>
Randy Dunlap9f158332005-09-13 01:25:16 -070026#include <linux/kernel.h>
Vegard Nossumb1eeab62008-11-25 16:55:53 +010027#include <linux/kmemcheck.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/module.h>
29#include <linux/suspend.h>
30#include <linux/pagevec.h>
31#include <linux/blkdev.h>
32#include <linux/slab.h>
Dave Hansena238ab52011-05-24 17:12:16 -070033#include <linux/ratelimit.h>
David Rientjes5a3135c22007-10-16 23:25:53 -070034#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <linux/notifier.h>
36#include <linux/topology.h>
37#include <linux/sysctl.h>
38#include <linux/cpu.h>
39#include <linux/cpuset.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070040#include <linux/memory_hotplug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include <linux/nodemask.h>
42#include <linux/vmalloc.h>
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -070043#include <linux/vmstat.h>
Christoph Lameter4be38e32006-01-06 00:11:17 -080044#include <linux/mempolicy.h>
Yasunori Goto68113782006-06-23 02:03:11 -070045#include <linux/stop_machine.h>
Mel Gormanc7132162006-09-27 01:49:43 -070046#include <linux/sort.h>
47#include <linux/pfn.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070048#include <linux/backing-dev.h>
Akinobu Mita933e3122006-12-08 02:39:45 -080049#include <linux/fault-inject.h>
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -070050#include <linux/page-isolation.h>
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070051#include <linux/page_cgroup.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070052#include <linux/debugobjects.h>
Catalin Marinasdbb1f812009-06-11 13:23:19 +010053#include <linux/kmemleak.h>
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;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -080099int percpu_pagelist_fraction;
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000100gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800102#ifdef CONFIG_PM_SLEEP
103/*
104 * The following functions are used by the suspend/hibernate code to temporarily
105 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
106 * while devices are suspended. To avoid races with the suspend/hibernate code,
107 * they should always be called with pm_mutex held (gfp_allowed_mask also should
108 * only be modified with pm_mutex held, unless the suspend/hibernate code is
109 * guaranteed not to run in parallel with that modification).
110 */
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100111
112static gfp_t saved_gfp_mask;
113
114void pm_restore_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800115{
116 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100117 if (saved_gfp_mask) {
118 gfp_allowed_mask = saved_gfp_mask;
119 saved_gfp_mask = 0;
120 }
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800121}
122
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100123void pm_restrict_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800124{
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800125 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100126 WARN_ON(saved_gfp_mask);
127 saved_gfp_mask = gfp_allowed_mask;
128 gfp_allowed_mask &= ~GFP_IOFS;
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800129}
130#endif /* CONFIG_PM_SLEEP */
131
Mel Gormand9c23402007-10-16 01:26:01 -0700132#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
133int pageblock_order __read_mostly;
134#endif
135
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800136static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -0800137
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138/*
139 * results with 256, 32 in the lowmem_reserve sysctl:
140 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
141 * 1G machine -> (16M dma, 784M normal, 224M high)
142 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
143 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
144 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +0100145 *
146 * TBD: should special case ZONE_DMA32 machines here - in those we normally
147 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700149int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800150#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700151 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -0800152#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700153#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700154 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700155#endif
Christoph Lametere53ef382006-09-25 23:31:14 -0700156#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700157 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700158#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700159 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700160};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
162EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163
Helge Deller15ad7cd2006-12-06 20:40:36 -0800164static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800165#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700166 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800167#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700168#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700169 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700170#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700171 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700172#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700173 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700174#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700175 "Movable",
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700176};
177
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178int min_free_kbytes = 1024;
179
Jan Beulich2c85f512009-09-21 17:03:07 -0700180static unsigned long __meminitdata nr_kernel_pages;
181static unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700182static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183
Mel Gormanc7132162006-09-27 01:49:43 -0700184#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
185 /*
Simon Arlott183ff222007-10-20 01:27:18 +0200186 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
Mel Gormanc7132162006-09-27 01:49:43 -0700187 * ranges of memory (RAM) that may be registered with add_active_range().
188 * Ranges passed to add_active_range() will be merged if possible
189 * so the number of times add_active_range() can be called is
190 * related to the number of nodes and the number of holes
191 */
192 #ifdef CONFIG_MAX_ACTIVE_REGIONS
193 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
194 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
195 #else
196 #if MAX_NUMNODES >= 32
197 /* If there can be many nodes, allow up to 50 holes per node */
198 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
199 #else
200 /* By default, allow up to 256 distinct regions */
201 #define MAX_ACTIVE_REGIONS 256
202 #endif
203 #endif
204
Jan Beulich98011f52007-07-15 23:38:17 -0700205 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
206 static int __meminitdata nr_nodemap_entries;
207 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
208 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
Adrian Bunkb69a7282008-07-23 21:28:12 -0700209 static unsigned long __initdata required_kernelcore;
Adrian Bunk484f51f2007-10-16 01:26:03 -0700210 static unsigned long __initdata required_movablecore;
Adrian Bunkb69a7282008-07-23 21:28:12 -0700211 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Mel Gorman2a1e2742007-07-17 04:03:12 -0700212
213 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
214 int movable_zone;
215 EXPORT_SYMBOL(movable_zone);
Mel Gormanc7132162006-09-27 01:49:43 -0700216#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
217
Miklos Szeredi418508c2007-05-23 13:57:55 -0700218#if MAX_NUMNODES > 1
219int nr_node_ids __read_mostly = MAX_NUMNODES;
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700220int nr_online_nodes __read_mostly = 1;
Miklos Szeredi418508c2007-05-23 13:57:55 -0700221EXPORT_SYMBOL(nr_node_ids);
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700222EXPORT_SYMBOL(nr_online_nodes);
Miklos Szeredi418508c2007-05-23 13:57:55 -0700223#endif
224
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700225int page_group_by_mobility_disabled __read_mostly;
226
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700227static void set_pageblock_migratetype(struct page *page, int migratetype)
228{
Mel Gorman49255c62009-06-16 15:31:58 -0700229
230 if (unlikely(page_group_by_mobility_disabled))
231 migratetype = MIGRATE_UNMOVABLE;
232
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700233 set_pageblock_flags_group(page, (unsigned long)migratetype,
234 PB_migrate, PB_migrate_end);
235}
236
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -0700237bool oom_killer_disabled __read_mostly;
238
Nick Piggin13e74442006-01-06 00:10:58 -0800239#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700240static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700242 int ret = 0;
243 unsigned seq;
244 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -0700245
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700246 do {
247 seq = zone_span_seqbegin(zone);
248 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
249 ret = 1;
250 else if (pfn < zone->zone_start_pfn)
251 ret = 1;
252 } while (zone_span_seqretry(zone, seq));
253
254 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700255}
256
257static int page_is_consistent(struct zone *zone, struct page *page)
258{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700259 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700260 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700262 return 0;
263
264 return 1;
265}
266/*
267 * Temporary debugging check for pages not lying within a given zone.
268 */
269static int bad_range(struct zone *zone, struct page *page)
270{
271 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700273 if (!page_is_consistent(zone, page))
274 return 1;
275
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 return 0;
277}
Nick Piggin13e74442006-01-06 00:10:58 -0800278#else
279static inline int bad_range(struct zone *zone, struct page *page)
280{
281 return 0;
282}
283#endif
284
Nick Piggin224abf92006-01-06 00:11:11 -0800285static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800287 static unsigned long resume;
288 static unsigned long nr_shown;
289 static unsigned long nr_unshown;
290
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200291 /* Don't complain about poisoned pages */
292 if (PageHWPoison(page)) {
Andrea Arcangelief2b4b92011-03-18 00:16:35 +0100293 reset_page_mapcount(page); /* remove PageBuddy */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200294 return;
295 }
296
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800297 /*
298 * Allow a burst of 60 reports, then keep quiet for that minute;
299 * or allow a steady drip of one report per second.
300 */
301 if (nr_shown == 60) {
302 if (time_before(jiffies, resume)) {
303 nr_unshown++;
304 goto out;
305 }
306 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800307 printk(KERN_ALERT
308 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800309 nr_unshown);
310 nr_unshown = 0;
311 }
312 nr_shown = 0;
313 }
314 if (nr_shown++ == 0)
315 resume = jiffies + 60 * HZ;
316
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800317 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800318 current->comm, page_to_pfn(page));
Wu Fengguang718a3822010-03-10 15:20:43 -0800319 dump_page(page);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700320
Dave Jones4f318882011-10-31 17:07:24 -0700321 print_modules();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800323out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800324 /* Leave bad fields for debug, except PageBuddy could make trouble */
Andrea Arcangelief2b4b92011-03-18 00:16:35 +0100325 reset_page_mapcount(page); /* remove PageBuddy */
Randy Dunlap9f158332005-09-13 01:25:16 -0700326 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327}
328
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329/*
330 * Higher-order pages are called "compound pages". They are structured thusly:
331 *
332 * The first PAGE_SIZE page is called the "head page".
333 *
334 * The remaining PAGE_SIZE pages are called "tail pages".
335 *
336 * All pages have PG_compound set. All pages have their ->private pointing at
337 * the head page (even the head page has this).
338 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800339 * The first tail page's ->lru.next holds the address of the compound page's
340 * put_page() function. Its ->lru.prev holds the order of allocation.
341 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800343
344static void free_compound_page(struct page *page)
345{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700346 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800347}
348
Andi Kleen01ad1c02008-07-23 21:27:46 -0700349void prep_compound_page(struct page *page, unsigned long order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350{
351 int i;
352 int nr_pages = 1 << order;
353
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800354 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700355 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700356 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800357 for (i = 1; i < nr_pages; i++) {
358 struct page *p = page + i;
Christoph Lameterd85f3382007-05-06 14:49:39 -0700359 __SetPageTail(p);
Youquan Song58a84aa2011-12-08 14:34:18 -0800360 set_page_count(p, 0);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700361 p->first_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 }
363}
364
Andrea Arcangeli59ff4212011-01-13 15:46:38 -0800365/* update __split_huge_page_refcount if you change this function */
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800366static int destroy_compound_page(struct page *page, unsigned long order)
Andy Whitcroft18229df2008-11-06 12:53:27 -0800367{
368 int i;
369 int nr_pages = 1 << order;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800370 int bad = 0;
Andy Whitcroft18229df2008-11-06 12:53:27 -0800371
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800372 if (unlikely(compound_order(page) != order) ||
373 unlikely(!PageHead(page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800374 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800375 bad++;
376 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377
Christoph Lameter6d777952007-05-06 14:49:40 -0700378 __ClearPageHead(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800379
Andy Whitcroft18229df2008-11-06 12:53:27 -0800380 for (i = 1; i < nr_pages; i++) {
381 struct page *p = page + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382
Alexey Zaytseve713a212009-01-10 02:47:57 +0300383 if (unlikely(!PageTail(p) || (p->first_page != page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800384 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800385 bad++;
386 }
Christoph Lameterd85f3382007-05-06 14:49:39 -0700387 __ClearPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800389
390 return bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392
Nick Piggin17cf4402006-03-22 00:08:41 -0800393static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
394{
395 int i;
396
Andrew Morton6626c5d2006-03-22 00:08:42 -0800397 /*
398 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
399 * and __GFP_HIGHMEM from hard or soft interrupt context.
400 */
Nick Piggin725d7042006-09-25 23:30:55 -0700401 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800402 for (i = 0; i < (1 << order); i++)
403 clear_highpage(page + i);
404}
405
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700406static inline void set_page_order(struct page *page, int order)
407{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700408 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000409 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410}
411
412static inline void rmv_page_order(struct page *page)
413{
Nick Piggin676165a2006-04-10 11:21:48 +1000414 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700415 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416}
417
418/*
419 * Locate the struct page for both the matching buddy in our
420 * pair (buddy1) and the combined O(n+1) page they form (page).
421 *
422 * 1) Any buddy B1 will have an order O twin B2 which satisfies
423 * the following equation:
424 * B2 = B1 ^ (1 << O)
425 * For example, if the starting buddy (buddy2) is #8 its order
426 * 1 buddy is #10:
427 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
428 *
429 * 2) Any buddy B will have an order O+1 parent P which
430 * satisfies the following equation:
431 * P = B & ~(1 << O)
432 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200433 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435static inline unsigned long
KyongHo Cho43506fa2011-01-13 15:47:24 -0800436__find_buddy_index(unsigned long page_idx, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437{
KyongHo Cho43506fa2011-01-13 15:47:24 -0800438 return page_idx ^ (1 << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439}
440
441/*
442 * This function checks whether a page is free && is the buddy
443 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800444 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000445 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700446 * (c) a page and its buddy have the same order &&
447 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 *
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800449 * For recording whether a page is in the buddy system, we set ->_mapcount -2.
450 * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock.
Nick Piggin676165a2006-04-10 11:21:48 +1000451 *
452 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700454static inline int page_is_buddy(struct page *page, struct page *buddy,
455 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700457 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800458 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800459
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700460 if (page_zone_id(page) != page_zone_id(buddy))
461 return 0;
462
463 if (PageBuddy(buddy) && page_order(buddy) == order) {
Nick Piggina3af9c32009-06-16 15:32:10 -0700464 VM_BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700465 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000466 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700467 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468}
469
470/*
471 * Freeing function for a buddy system allocator.
472 *
473 * The concept of a buddy system is to maintain direct-mapped table
474 * (containing bit values) for memory blocks of various "orders".
475 * The bottom level table contains the map for the smallest allocatable
476 * units of memory (here, pages), and each level above it describes
477 * pairs of units from the levels below, hence, "buddies".
478 * At a high level, all that happens here is marking the table entry
479 * at the bottom level available, and propagating the changes upward
480 * as necessary, plus some accounting needed to play nicely with other
481 * parts of the VM system.
482 * At each level, we keep a list of pages, which are heads of continuous
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800483 * free pages of length of (1 << order) and marked with _mapcount -2. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700484 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 * So when we are allocating or freeing one, we can derive the state of the
486 * other. That is, if we allocate a small block, and both were
487 * free, the remainder of the region must be split into blocks.
488 * If a block is freed, and its buddy is also free, then this
489 * triggers coalescing into a block of larger size.
490 *
491 * -- wli
492 */
493
Nick Piggin48db57f2006-01-08 01:00:42 -0800494static inline void __free_one_page(struct page *page,
Mel Gormaned0ae212009-06-16 15:32:07 -0700495 struct zone *zone, unsigned int order,
496 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497{
498 unsigned long page_idx;
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700499 unsigned long combined_idx;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800500 unsigned long uninitialized_var(buddy_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700501 struct page *buddy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502
Nick Piggin224abf92006-01-06 00:11:11 -0800503 if (unlikely(PageCompound(page)))
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800504 if (unlikely(destroy_compound_page(page, order)))
505 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506
Mel Gormaned0ae212009-06-16 15:32:07 -0700507 VM_BUG_ON(migratetype == -1);
508
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
510
Mel Gormanf2260e62009-06-16 15:32:13 -0700511 VM_BUG_ON(page_idx & ((1 << order) - 1));
Nick Piggin725d7042006-09-25 23:30:55 -0700512 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 while (order < MAX_ORDER-1) {
KyongHo Cho43506fa2011-01-13 15:47:24 -0800515 buddy_idx = __find_buddy_index(page_idx, order);
516 buddy = page + (buddy_idx - page_idx);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700517 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700518 break;
Nick Piggin13e74442006-01-06 00:10:58 -0800519
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700520 /* Our buddy is free, merge with it and move up one order. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 list_del(&buddy->lru);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700522 zone->free_area[order].nr_free--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 rmv_page_order(buddy);
KyongHo Cho43506fa2011-01-13 15:47:24 -0800524 combined_idx = buddy_idx & page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 page = page + (combined_idx - page_idx);
526 page_idx = combined_idx;
527 order++;
528 }
529 set_page_order(page, order);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700530
531 /*
532 * If this is not the largest possible page, check if the buddy
533 * of the next-highest order is free. If it is, it's possible
534 * that pages are being freed that will coalesce soon. In case,
535 * that is happening, add the free page to the tail of the list
536 * so it's less likely to be used soon and more likely to be merged
537 * as a higher order page
538 */
Mel Gormanb7f50cf2010-10-26 14:21:11 -0700539 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700540 struct page *higher_page, *higher_buddy;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800541 combined_idx = buddy_idx & page_idx;
542 higher_page = page + (combined_idx - page_idx);
543 buddy_idx = __find_buddy_index(combined_idx, order + 1);
544 higher_buddy = page + (buddy_idx - combined_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700545 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
546 list_add_tail(&page->lru,
547 &zone->free_area[order].free_list[migratetype]);
548 goto out;
549 }
550 }
551
552 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
553out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 zone->free_area[order].nr_free++;
555}
556
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700557/*
558 * free_page_mlock() -- clean up attempts to free and mlocked() page.
559 * Page should not be on lru, so no need to fix that up.
560 * free_pages_check() will verify...
561 */
562static inline void free_page_mlock(struct page *page)
563{
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700564 __dec_zone_page_state(page, NR_MLOCK);
565 __count_vm_event(UNEVICTABLE_MLOCKFREED);
566}
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700567
Nick Piggin224abf92006-01-06 00:11:11 -0800568static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569{
Nick Piggin92be2e332006-01-06 00:10:57 -0800570 if (unlikely(page_mapcount(page) |
571 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700572 (atomic_read(&page->_count) != 0) |
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -0700573 (page->flags & PAGE_FLAGS_CHECK_AT_FREE) |
574 (mem_cgroup_bad_page_check(page)))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800575 bad_page(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800576 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800577 }
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800578 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
579 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
580 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581}
582
583/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700584 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700586 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 *
588 * If the zone was previously in an "all pages pinned" state then look to
589 * see if this freeing clears that state.
590 *
591 * And clear the zone's pages_scanned counter, to hold off the "all pages are
592 * pinned" detection logic.
593 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700594static void free_pcppages_bulk(struct zone *zone, int count,
595 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596{
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700597 int migratetype = 0;
Mel Gormana6f9edd62009-09-21 17:03:20 -0700598 int batch_free = 0;
Mel Gorman72853e22010-09-09 16:38:16 -0700599 int to_free = count;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700600
Nick Pigginc54ad302006-01-06 00:10:56 -0800601 spin_lock(&zone->lock);
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -0800602 zone->all_unreclaimable = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700604
Mel Gorman72853e22010-09-09 16:38:16 -0700605 while (to_free) {
Nick Piggin48db57f2006-01-08 01:00:42 -0800606 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700607 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -0800608
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700609 /*
Mel Gormana6f9edd62009-09-21 17:03:20 -0700610 * Remove pages from lists in a round-robin fashion. A
611 * batch_free count is maintained that is incremented when an
612 * empty list is encountered. This is so more pages are freed
613 * off fuller lists instead of spinning excessively around empty
614 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700615 */
616 do {
Mel Gormana6f9edd62009-09-21 17:03:20 -0700617 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700618 if (++migratetype == MIGRATE_PCPTYPES)
619 migratetype = 0;
620 list = &pcp->lists[migratetype];
621 } while (list_empty(list));
622
Namhyung Kim1d168712011-03-22 16:32:45 -0700623 /* This is the only non-empty list. Free them all. */
624 if (batch_free == MIGRATE_PCPTYPES)
625 batch_free = to_free;
626
Mel Gormana6f9edd62009-09-21 17:03:20 -0700627 do {
628 page = list_entry(list->prev, struct page, lru);
629 /* must delete as __free_one_page list manipulates */
630 list_del(&page->lru);
Hugh Dickinsa7016232010-01-29 17:46:34 +0000631 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
632 __free_one_page(page, zone, 0, page_private(page));
633 trace_mm_page_pcpu_drain(page, 0, page_private(page));
Mel Gorman72853e22010-09-09 16:38:16 -0700634 } while (--to_free && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 }
Mel Gorman72853e22010-09-09 16:38:16 -0700636 __mod_zone_page_state(zone, NR_FREE_PAGES, count);
Nick Pigginc54ad302006-01-06 00:10:56 -0800637 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638}
639
Mel Gormaned0ae212009-06-16 15:32:07 -0700640static void free_one_page(struct zone *zone, struct page *page, int order,
641 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800642{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700643 spin_lock(&zone->lock);
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -0800644 zone->all_unreclaimable = 0;
Christoph Lameter006d22d2006-09-25 23:31:48 -0700645 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700646
Mel Gormaned0ae212009-06-16 15:32:07 -0700647 __free_one_page(page, zone, order, migratetype);
Mel Gorman72853e22010-09-09 16:38:16 -0700648 __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700649 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800650}
651
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700652static bool free_pages_prepare(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 int i;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800655 int bad = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
Li Hongf6503162010-03-05 13:41:52 -0800657 trace_mm_page_free_direct(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100658 kmemcheck_free_shadow(page, order);
659
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -0800660 if (PageAnon(page))
661 page->mapping = NULL;
662 for (i = 0; i < (1 << order); i++)
663 bad += free_pages_check(page + i);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800664 if (bad)
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700665 return false;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800666
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700667 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -0700668 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700669 debug_check_no_obj_freed(page_address(page),
670 PAGE_SIZE << order);
671 }
Nick Piggindafb1362006-10-11 01:21:30 -0700672 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800673 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700674
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700675 return true;
676}
677
678static void __free_pages_ok(struct page *page, unsigned int order)
679{
680 unsigned long flags;
681 int wasMlocked = __TestClearPageMlocked(page);
682
683 if (!free_pages_prepare(page, order))
684 return;
685
Nick Pigginc54ad302006-01-06 00:10:56 -0800686 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +0200687 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -0700688 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700689 __count_vm_events(PGFREE, 1 << order);
Mel Gormaned0ae212009-06-16 15:32:07 -0700690 free_one_page(page_zone(page), page, order,
691 get_pageblock_migratetype(page));
Nick Pigginc54ad302006-01-06 00:10:56 -0800692 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693}
694
David Howellsa226f6c2006-01-06 00:11:08 -0800695/*
696 * permit the bootmem allocator to evade page validation on high-order frees
697 */
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700698void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800699{
700 if (order == 0) {
701 __ClearPageReserved(page);
702 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800703 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800704 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800705 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800706 int loop;
707
Nick Piggin545b1ea2006-03-22 00:08:07 -0800708 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800709 for (loop = 0; loop < BITS_PER_LONG; loop++) {
710 struct page *p = &page[loop];
711
Nick Piggin545b1ea2006-03-22 00:08:07 -0800712 if (loop + 1 < BITS_PER_LONG)
713 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800714 __ClearPageReserved(p);
715 set_page_count(p, 0);
716 }
717
Nick Piggin7835e982006-03-22 00:08:40 -0800718 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800719 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800720 }
721}
722
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723
724/*
725 * The order of subdivision here is critical for the IO subsystem.
726 * Please do not alter this order without good reasons and regression
727 * testing. Specifically, as large blocks of memory are subdivided,
728 * the order in which smaller blocks are delivered depends on the order
729 * they're subdivided in this function. This is the primary factor
730 * influencing the order in which pages are delivered to the IO
731 * subsystem according to empirical testing, and this is also justified
732 * by considering the behavior of a buddy system containing a single
733 * large block of memory acted on by a series of small allocations.
734 * This behavior is a critical factor in sglist merging's success.
735 *
736 * -- wli
737 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800738static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700739 int low, int high, struct free_area *area,
740 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741{
742 unsigned long size = 1 << high;
743
744 while (high > low) {
745 area--;
746 high--;
747 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700748 VM_BUG_ON(bad_range(zone, &page[size]));
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700749 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 area->nr_free++;
751 set_page_order(&page[size], high);
752 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753}
754
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755/*
756 * This page is about to be returned from the page allocator
757 */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200758static inline int check_new_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759{
Nick Piggin92be2e332006-01-06 00:10:57 -0800760 if (unlikely(page_mapcount(page) |
761 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700762 (atomic_read(&page->_count) != 0) |
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -0700763 (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
764 (mem_cgroup_bad_page_check(page)))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800765 bad_page(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800766 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800767 }
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200768 return 0;
769}
770
771static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
772{
773 int i;
774
775 for (i = 0; i < (1 << order); i++) {
776 struct page *p = page + i;
777 if (unlikely(check_new_page(p)))
778 return 1;
779 }
Hugh Dickins689bceb2005-11-21 21:32:20 -0800780
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700781 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800782 set_page_refcounted(page);
Nick Piggincc1025092006-12-06 20:32:00 -0800783
784 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800786
787 if (gfp_flags & __GFP_ZERO)
788 prep_zero_page(page, order, gfp_flags);
789
790 if (order && (gfp_flags & __GFP_COMP))
791 prep_compound_page(page, order);
792
Hugh Dickins689bceb2005-11-21 21:32:20 -0800793 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794}
795
Mel Gorman56fd56b2007-10-16 01:25:58 -0700796/*
797 * Go through the free lists for the given migratetype and remove
798 * the smallest available page from the freelists
799 */
Mel Gorman728ec982009-06-16 15:32:04 -0700800static inline
801struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -0700802 int migratetype)
803{
804 unsigned int current_order;
805 struct free_area * area;
806 struct page *page;
807
808 /* Find a page of the appropriate size in the preferred list */
809 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
810 area = &(zone->free_area[current_order]);
811 if (list_empty(&area->free_list[migratetype]))
812 continue;
813
814 page = list_entry(area->free_list[migratetype].next,
815 struct page, lru);
816 list_del(&page->lru);
817 rmv_page_order(page);
818 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -0700819 expand(zone, page, order, current_order, area, migratetype);
820 return page;
821 }
822
823 return NULL;
824}
825
826
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700827/*
828 * This array describes the order lists are fallen back to when
829 * the free lists for the desirable migrate type are depleted
830 */
831static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
Mel Gorman64c5e132007-10-16 01:25:59 -0700832 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
833 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
834 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
835 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700836};
837
Mel Gormanc361be52007-10-16 01:25:51 -0700838/*
839 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -0700840 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -0700841 * boundary. If alignment is required, use move_freepages_block()
842 */
Adrian Bunkb69a7282008-07-23 21:28:12 -0700843static int move_freepages(struct zone *zone,
844 struct page *start_page, struct page *end_page,
845 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700846{
847 struct page *page;
848 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -0700849 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -0700850
851#ifndef CONFIG_HOLES_IN_ZONE
852 /*
853 * page_zone is not safe to call in this context when
854 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
855 * anyway as we check zone boundaries in move_freepages_block().
856 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -0700857 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -0700858 */
859 BUG_ON(page_zone(start_page) != page_zone(end_page));
860#endif
861
862 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -0700863 /* Make sure we are not inadvertently changing nodes */
864 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
865
Mel Gormanc361be52007-10-16 01:25:51 -0700866 if (!pfn_valid_within(page_to_pfn(page))) {
867 page++;
868 continue;
869 }
870
871 if (!PageBuddy(page)) {
872 page++;
873 continue;
874 }
875
876 order = page_order(page);
Kirill A. Shutemov84be48d2011-03-22 16:33:41 -0700877 list_move(&page->lru,
878 &zone->free_area[order].free_list[migratetype]);
Mel Gormanc361be52007-10-16 01:25:51 -0700879 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -0700880 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -0700881 }
882
Mel Gormand1003132007-10-16 01:26:00 -0700883 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -0700884}
885
Adrian Bunkb69a7282008-07-23 21:28:12 -0700886static int move_freepages_block(struct zone *zone, struct page *page,
887 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700888{
889 unsigned long start_pfn, end_pfn;
890 struct page *start_page, *end_page;
891
892 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -0700893 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -0700894 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -0700895 end_page = start_page + pageblock_nr_pages - 1;
896 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -0700897
898 /* Do not cross zone boundaries */
899 if (start_pfn < zone->zone_start_pfn)
900 start_page = page;
901 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
902 return 0;
903
904 return move_freepages(zone, start_page, end_page, migratetype);
905}
906
Mel Gorman2f66a682009-09-21 17:02:31 -0700907static void change_pageblock_range(struct page *pageblock_page,
908 int start_order, int migratetype)
909{
910 int nr_pageblocks = 1 << (start_order - pageblock_order);
911
912 while (nr_pageblocks--) {
913 set_pageblock_migratetype(pageblock_page, migratetype);
914 pageblock_page += pageblock_nr_pages;
915 }
916}
917
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700918/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -0700919static inline struct page *
920__rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700921{
922 struct free_area * area;
923 int current_order;
924 struct page *page;
925 int migratetype, i;
926
927 /* Find the largest possible block of pages in the other list */
928 for (current_order = MAX_ORDER-1; current_order >= order;
929 --current_order) {
930 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
931 migratetype = fallbacks[start_migratetype][i];
932
Mel Gorman56fd56b2007-10-16 01:25:58 -0700933 /* MIGRATE_RESERVE handled later if necessary */
934 if (migratetype == MIGRATE_RESERVE)
935 continue;
Mel Gormane0104872007-10-16 01:25:53 -0700936
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700937 area = &(zone->free_area[current_order]);
938 if (list_empty(&area->free_list[migratetype]))
939 continue;
940
941 page = list_entry(area->free_list[migratetype].next,
942 struct page, lru);
943 area->nr_free--;
944
945 /*
Mel Gormanc361be52007-10-16 01:25:51 -0700946 * If breaking a large block of pages, move all free
Mel Gorman46dafbc2007-10-16 01:25:55 -0700947 * pages to the preferred allocation list. If falling
948 * back for a reclaimable kernel allocation, be more
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300949 * aggressive about taking ownership of free pages
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700950 */
Mel Gormand9c23402007-10-16 01:26:01 -0700951 if (unlikely(current_order >= (pageblock_order >> 1)) ||
Mel Gormandd5d2412009-09-05 11:17:11 -0700952 start_migratetype == MIGRATE_RECLAIMABLE ||
953 page_group_by_mobility_disabled) {
Mel Gorman46dafbc2007-10-16 01:25:55 -0700954 unsigned long pages;
955 pages = move_freepages_block(zone, page,
956 start_migratetype);
957
958 /* Claim the whole block if over half of it is free */
Mel Gormandd5d2412009-09-05 11:17:11 -0700959 if (pages >= (1 << (pageblock_order-1)) ||
960 page_group_by_mobility_disabled)
Mel Gorman46dafbc2007-10-16 01:25:55 -0700961 set_pageblock_migratetype(page,
962 start_migratetype);
963
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700964 migratetype = start_migratetype;
Mel Gormanc361be52007-10-16 01:25:51 -0700965 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700966
967 /* Remove the page from the freelists */
968 list_del(&page->lru);
969 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700970
Mel Gorman2f66a682009-09-21 17:02:31 -0700971 /* Take ownership for orders >= pageblock_order */
972 if (current_order >= pageblock_order)
973 change_pageblock_range(page, current_order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700974 start_migratetype);
975
976 expand(zone, page, order, current_order, area, migratetype);
Mel Gormane0fff1b2009-09-21 17:02:42 -0700977
978 trace_mm_page_alloc_extfrag(page, order, current_order,
979 start_migratetype, migratetype);
980
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700981 return page;
982 }
983 }
984
Mel Gorman728ec982009-06-16 15:32:04 -0700985 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700986}
987
Mel Gorman56fd56b2007-10-16 01:25:58 -0700988/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 * Do the hard work of removing an element from the buddy allocator.
990 * Call me with the zone->lock already held.
991 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700992static struct page *__rmqueue(struct zone *zone, unsigned int order,
993 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 struct page *page;
996
Mel Gorman728ec982009-06-16 15:32:04 -0700997retry_reserve:
Mel Gorman56fd56b2007-10-16 01:25:58 -0700998 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999
Mel Gorman728ec982009-06-16 15:32:04 -07001000 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07001001 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001002
Mel Gorman728ec982009-06-16 15:32:04 -07001003 /*
1004 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1005 * is used because __rmqueue_smallest is an inline function
1006 * and we want just one call site
1007 */
1008 if (!page) {
1009 migratetype = MIGRATE_RESERVE;
1010 goto retry_reserve;
1011 }
1012 }
1013
Mel Gorman0d3d0622009-09-21 17:02:44 -07001014 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001015 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016}
1017
1018/*
1019 * Obtain a specified number of elements from the buddy allocator, all under
1020 * a single hold of the lock, for efficiency. Add them to the supplied list.
1021 * Returns the number of new pages which were placed at *list.
1022 */
1023static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001024 unsigned long count, struct list_head *list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001025 int migratetype, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028
Nick Pigginc54ad302006-01-06 00:10:56 -08001029 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001031 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001032 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08001034
1035 /*
1036 * Split buddy pages returned by expand() are received here
1037 * in physical page order. The page is added to the callers and
1038 * list and the list head then moves forward. From the callers
1039 * perspective, the linked list is ordered by page number in
1040 * some conditions. This is useful for IO devices that can
1041 * merge IO requests if the physical pages are ordered
1042 * properly.
1043 */
Mel Gormane084b2d2009-07-29 15:02:04 -07001044 if (likely(cold == 0))
1045 list_add(&page->lru, list);
1046 else
1047 list_add_tail(&page->lru, list);
Mel Gorman535131e62007-10-16 01:25:49 -07001048 set_page_private(page, migratetype);
Mel Gorman81eabcb2007-12-17 16:20:05 -08001049 list = &page->lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 }
Mel Gormanf2260e62009-06-16 15:32:13 -07001051 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08001052 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001053 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054}
1055
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001056#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001057/*
Christoph Lameter4037d452007-05-09 02:35:14 -07001058 * Called from the vmstat counter updater to drain pagesets of this
1059 * currently executing processor on remote nodes after they have
1060 * expired.
1061 *
Christoph Lameter879336c2006-03-22 00:09:08 -08001062 * Note that this function must be called with the thread pinned to
1063 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001064 */
Christoph Lameter4037d452007-05-09 02:35:14 -07001065void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001066{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001067 unsigned long flags;
Christoph Lameter4037d452007-05-09 02:35:14 -07001068 int to_drain;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001069
Christoph Lameter4037d452007-05-09 02:35:14 -07001070 local_irq_save(flags);
1071 if (pcp->count >= pcp->batch)
1072 to_drain = pcp->batch;
1073 else
1074 to_drain = pcp->count;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001075 free_pcppages_bulk(zone, to_drain, pcp);
Christoph Lameter4037d452007-05-09 02:35:14 -07001076 pcp->count -= to_drain;
1077 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001078}
1079#endif
1080
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001081/*
1082 * Drain pages of the indicated processor.
1083 *
1084 * The processor must either be the current processor and the
1085 * thread pinned to the current processor or a processor that
1086 * is not online.
1087 */
1088static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089{
Nick Pigginc54ad302006-01-06 00:10:56 -08001090 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07001093 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 struct per_cpu_pageset *pset;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001095 struct per_cpu_pages *pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001097 local_irq_save(flags);
1098 pset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001100 pcp = &pset->pcp;
David Rientjes2ff754f2011-01-25 15:07:23 -08001101 if (pcp->count) {
1102 free_pcppages_bulk(zone, pcp->count, pcp);
1103 pcp->count = 0;
1104 }
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001105 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 }
1107}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001109/*
1110 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1111 */
1112void drain_local_pages(void *arg)
1113{
1114 drain_pages(smp_processor_id());
1115}
1116
1117/*
1118 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
1119 */
1120void drain_all_pages(void)
1121{
Jens Axboe15c8b6c2008-05-09 09:39:44 +02001122 on_each_cpu(drain_local_pages, NULL, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001123}
1124
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02001125#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126
1127void mark_free_pages(struct zone *zone)
1128{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001129 unsigned long pfn, max_zone_pfn;
1130 unsigned long flags;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001131 int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 struct list_head *curr;
1133
1134 if (!zone->spanned_pages)
1135 return;
1136
1137 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001138
1139 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1140 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1141 if (pfn_valid(pfn)) {
1142 struct page *page = pfn_to_page(pfn);
1143
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001144 if (!swsusp_page_is_forbidden(page))
1145 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001146 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001148 for_each_migratetype_order(order, t) {
1149 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001150 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001152 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1153 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001154 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001155 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001156 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 spin_unlock_irqrestore(&zone->lock, flags);
1158}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001159#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160
1161/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 * Free a 0-order page
Li Hongfc916682010-03-05 13:41:54 -08001163 * cold == 1 ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 */
Li Hongfc916682010-03-05 13:41:54 -08001165void free_hot_cold_page(struct page *page, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166{
1167 struct zone *zone = page_zone(page);
1168 struct per_cpu_pages *pcp;
1169 unsigned long flags;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001170 int migratetype;
Johannes Weiner451ea252009-09-21 17:01:48 -07001171 int wasMlocked = __TestClearPageMlocked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001173 if (!free_pages_prepare(page, 0))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001174 return;
1175
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001176 migratetype = get_pageblock_migratetype(page);
1177 set_page_private(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +02001179 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -07001180 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001181 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001182
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001183 /*
1184 * We only track unmovable, reclaimable and movable on pcp lists.
1185 * Free ISOLATE pages back to the allocator because they are being
1186 * offlined but treat RESERVE as movable pages so we can get those
1187 * areas back if necessary. Otherwise, we may have to free
1188 * excessively into the page allocator
1189 */
1190 if (migratetype >= MIGRATE_PCPTYPES) {
1191 if (unlikely(migratetype == MIGRATE_ISOLATE)) {
1192 free_one_page(zone, page, 0, migratetype);
1193 goto out;
1194 }
1195 migratetype = MIGRATE_MOVABLE;
1196 }
1197
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001198 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001199 if (cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001200 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001201 else
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001202 list_add(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001204 if (pcp->count >= pcp->high) {
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001205 free_pcppages_bulk(zone, pcp->batch, pcp);
Nick Piggin48db57f2006-01-08 01:00:42 -08001206 pcp->count -= pcp->batch;
1207 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001208
1209out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211}
1212
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001213/*
1214 * split_page takes a non-compound higher-order page, and splits it into
1215 * n (1<<order) sub-pages: page[0..n]
1216 * Each sub-page must be freed individually.
1217 *
1218 * Note: this is probably too low level an operation for use in drivers.
1219 * Please consult with lkml before using this in your driver.
1220 */
1221void split_page(struct page *page, unsigned int order)
1222{
1223 int i;
1224
Nick Piggin725d7042006-09-25 23:30:55 -07001225 VM_BUG_ON(PageCompound(page));
1226 VM_BUG_ON(!page_count(page));
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001227
1228#ifdef CONFIG_KMEMCHECK
1229 /*
1230 * Split shadow pages too, because free(page[0]) would
1231 * otherwise free the whole shadow.
1232 */
1233 if (kmemcheck_page_is_tracked(page))
1234 split_page(virt_to_page(page[0].shadow), order);
1235#endif
1236
Nick Piggin7835e982006-03-22 00:08:40 -08001237 for (i = 1; i < (1 << order); i++)
1238 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001239}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001240
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241/*
Mel Gorman748446b2010-05-24 14:32:27 -07001242 * Similar to split_page except the page is already free. As this is only
1243 * being used for migration, the migratetype of the block also changes.
1244 * As this is called with interrupts disabled, the caller is responsible
1245 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1246 * are enabled.
1247 *
1248 * Note: this is probably too low level an operation for use in drivers.
1249 * Please consult with lkml before using this in your driver.
1250 */
1251int split_free_page(struct page *page)
1252{
1253 unsigned int order;
1254 unsigned long watermark;
1255 struct zone *zone;
1256
1257 BUG_ON(!PageBuddy(page));
1258
1259 zone = page_zone(page);
1260 order = page_order(page);
1261
1262 /* Obey watermarks as if the page was being allocated */
1263 watermark = low_wmark_pages(zone) + (1 << order);
1264 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1265 return 0;
1266
1267 /* Remove page from free list */
1268 list_del(&page->lru);
1269 zone->free_area[order].nr_free--;
1270 rmv_page_order(page);
1271 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1UL << order));
1272
1273 /* Split into individual pages */
1274 set_page_refcounted(page);
1275 split_page(page, order);
1276
1277 if (order >= pageblock_order - 1) {
1278 struct page *endpage = page + (1 << order) - 1;
1279 for (; page < endpage; page += pageblock_nr_pages)
1280 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1281 }
1282
1283 return 1 << order;
1284}
1285
1286/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1288 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1289 * or two.
1290 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07001291static inline
1292struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001293 struct zone *zone, int order, gfp_t gfp_flags,
1294 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295{
1296 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001297 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 int cold = !!(gfp_flags & __GFP_COLD);
1299
Hugh Dickins689bceb2005-11-21 21:32:20 -08001300again:
Nick Piggin48db57f2006-01-08 01:00:42 -08001301 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001303 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 local_irq_save(flags);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001306 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1307 list = &pcp->lists[migratetype];
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001308 if (list_empty(list)) {
Mel Gorman535131e62007-10-16 01:25:49 -07001309 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001310 pcp->batch, list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001311 migratetype, cold);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001312 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07001313 goto failed;
Mel Gorman535131e62007-10-16 01:25:49 -07001314 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001315
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001316 if (cold)
1317 page = list_entry(list->prev, struct page, lru);
1318 else
1319 page = list_entry(list->next, struct page, lru);
1320
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001321 list_del(&page->lru);
1322 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001323 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07001324 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1325 /*
1326 * __GFP_NOFAIL is not to be used in new code.
1327 *
1328 * All __GFP_NOFAIL callers should be fixed so that they
1329 * properly detect and handle allocation failures.
1330 *
1331 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07001332 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07001333 * __GFP_NOFAIL.
1334 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07001335 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07001336 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001338 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001339 spin_unlock(&zone->lock);
1340 if (!page)
1341 goto failed;
KOSAKI Motohiro6ccf80e2010-01-15 17:01:18 -08001342 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 }
1344
Christoph Lameterf8891e52006-06-30 01:55:45 -07001345 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Andi Kleen78afd562011-03-22 16:33:12 -07001346 zone_statistics(preferred_zone, zone, gfp_flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001347 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348
Nick Piggin725d7042006-09-25 23:30:55 -07001349 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -08001350 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001351 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001353
1354failed:
1355 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001356 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357}
1358
Mel Gorman41858962009-06-16 15:32:12 -07001359/* The ALLOC_WMARK bits are used as an index to zone->watermark */
1360#define ALLOC_WMARK_MIN WMARK_MIN
1361#define ALLOC_WMARK_LOW WMARK_LOW
1362#define ALLOC_WMARK_HIGH WMARK_HIGH
1363#define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1364
1365/* Mask to get the watermark bits */
1366#define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1367
Nick Piggin31488902005-11-28 13:44:03 -08001368#define ALLOC_HARDER 0x10 /* try to alloc harder */
1369#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1370#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001371
Akinobu Mita933e3122006-12-08 02:39:45 -08001372#ifdef CONFIG_FAIL_PAGE_ALLOC
1373
Akinobu Mitab2588c42011-07-26 16:09:03 -07001374static struct {
Akinobu Mita933e3122006-12-08 02:39:45 -08001375 struct fault_attr attr;
1376
1377 u32 ignore_gfp_highmem;
1378 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001379 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001380} fail_page_alloc = {
1381 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001382 .ignore_gfp_wait = 1,
1383 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001384 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001385};
1386
1387static int __init setup_fail_page_alloc(char *str)
1388{
1389 return setup_fault_attr(&fail_page_alloc.attr, str);
1390}
1391__setup("fail_page_alloc=", setup_fail_page_alloc);
1392
1393static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1394{
Akinobu Mita54114992007-07-15 23:40:23 -07001395 if (order < fail_page_alloc.min_order)
1396 return 0;
Akinobu Mita933e3122006-12-08 02:39:45 -08001397 if (gfp_mask & __GFP_NOFAIL)
1398 return 0;
1399 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1400 return 0;
1401 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1402 return 0;
1403
1404 return should_fail(&fail_page_alloc.attr, 1 << order);
1405}
1406
1407#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1408
1409static int __init fail_page_alloc_debugfs(void)
1410{
1411 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1412 struct dentry *dir;
Akinobu Mita933e3122006-12-08 02:39:45 -08001413
Akinobu Mitadd48c082011-08-03 16:21:01 -07001414 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
1415 &fail_page_alloc.attr);
1416 if (IS_ERR(dir))
1417 return PTR_ERR(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08001418
Akinobu Mitab2588c42011-07-26 16:09:03 -07001419 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
1420 &fail_page_alloc.ignore_gfp_wait))
1421 goto fail;
1422 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1423 &fail_page_alloc.ignore_gfp_highmem))
1424 goto fail;
1425 if (!debugfs_create_u32("min-order", mode, dir,
1426 &fail_page_alloc.min_order))
1427 goto fail;
Akinobu Mita933e3122006-12-08 02:39:45 -08001428
Akinobu Mitab2588c42011-07-26 16:09:03 -07001429 return 0;
1430fail:
Akinobu Mitadd48c082011-08-03 16:21:01 -07001431 debugfs_remove_recursive(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08001432
Akinobu Mitab2588c42011-07-26 16:09:03 -07001433 return -ENOMEM;
Akinobu Mita933e3122006-12-08 02:39:45 -08001434}
1435
1436late_initcall(fail_page_alloc_debugfs);
1437
1438#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1439
1440#else /* CONFIG_FAIL_PAGE_ALLOC */
1441
1442static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1443{
1444 return 0;
1445}
1446
1447#endif /* CONFIG_FAIL_PAGE_ALLOC */
1448
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449/*
Mel Gorman88f5acf2011-01-13 15:45:41 -08001450 * Return true if free pages are above 'mark'. This takes into account the order
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 * of the allocation.
1452 */
Mel Gorman88f5acf2011-01-13 15:45:41 -08001453static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1454 int classzone_idx, int alloc_flags, long free_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455{
1456 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001457 long min = mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 int o;
1459
Mel Gorman88f5acf2011-01-13 15:45:41 -08001460 free_pages -= (1 << order) + 1;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001461 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001463 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 min -= min / 4;
1465
1466 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
Mel Gorman88f5acf2011-01-13 15:45:41 -08001467 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 for (o = 0; o < order; o++) {
1469 /* At the next order, this order's pages become unavailable */
1470 free_pages -= z->free_area[o].nr_free << o;
1471
1472 /* Require fewer higher order pages to be free */
1473 min >>= 1;
1474
1475 if (free_pages <= min)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001476 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 }
Mel Gorman88f5acf2011-01-13 15:45:41 -08001478 return true;
1479}
1480
1481bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1482 int classzone_idx, int alloc_flags)
1483{
1484 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1485 zone_page_state(z, NR_FREE_PAGES));
1486}
1487
1488bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
1489 int classzone_idx, int alloc_flags)
1490{
1491 long free_pages = zone_page_state(z, NR_FREE_PAGES);
1492
1493 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1494 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1495
1496 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1497 free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498}
1499
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001500#ifdef CONFIG_NUMA
1501/*
1502 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1503 * skip over zones that are not allowed by the cpuset, or that have
1504 * been recently (in last second) found to be nearly full. See further
1505 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001506 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001507 *
1508 * If the zonelist cache is present in the passed in zonelist, then
1509 * returns a pointer to the allowed node mask (either the current
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001510 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001511 *
1512 * If the zonelist cache is not available for this zonelist, does
1513 * nothing and returns NULL.
1514 *
1515 * If the fullzones BITMAP in the zonelist cache is stale (more than
1516 * a second since last zap'd) then we zap it out (clear its bits.)
1517 *
1518 * We hold off even calling zlc_setup, until after we've checked the
1519 * first zone in the zonelist, on the theory that most allocations will
1520 * be satisfied from that first zone, so best to examine that zone as
1521 * quickly as we can.
1522 */
1523static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1524{
1525 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1526 nodemask_t *allowednodes; /* zonelist_cache approximation */
1527
1528 zlc = zonelist->zlcache_ptr;
1529 if (!zlc)
1530 return NULL;
1531
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001532 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001533 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1534 zlc->last_full_zap = jiffies;
1535 }
1536
1537 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1538 &cpuset_current_mems_allowed :
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001539 &node_states[N_HIGH_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001540 return allowednodes;
1541}
1542
1543/*
1544 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1545 * if it is worth looking at further for free memory:
1546 * 1) Check that the zone isn't thought to be full (doesn't have its
1547 * bit set in the zonelist_cache fullzones BITMAP).
1548 * 2) Check that the zones node (obtained from the zonelist_cache
1549 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1550 * Return true (non-zero) if zone is worth looking at further, or
1551 * else return false (zero) if it is not.
1552 *
1553 * This check -ignores- the distinction between various watermarks,
1554 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1555 * found to be full for any variation of these watermarks, it will
1556 * be considered full for up to one second by all requests, unless
1557 * we are so low on memory on all allowed nodes that we are forced
1558 * into the second scan of the zonelist.
1559 *
1560 * In the second scan we ignore this zonelist cache and exactly
1561 * apply the watermarks to all zones, even it is slower to do so.
1562 * We are low on memory in the second scan, and should leave no stone
1563 * unturned looking for a free page.
1564 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001565static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001566 nodemask_t *allowednodes)
1567{
1568 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1569 int i; /* index of *z in zonelist zones */
1570 int n; /* node that zone *z is on */
1571
1572 zlc = zonelist->zlcache_ptr;
1573 if (!zlc)
1574 return 1;
1575
Mel Gormandd1a2392008-04-28 02:12:17 -07001576 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001577 n = zlc->z_to_n[i];
1578
1579 /* This zone is worth trying if it is allowed but not full */
1580 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1581}
1582
1583/*
1584 * Given 'z' scanning a zonelist, set the corresponding bit in
1585 * zlc->fullzones, so that subsequent attempts to allocate a page
1586 * from that zone don't waste time re-examining it.
1587 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001588static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001589{
1590 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1591 int i; /* index of *z in zonelist zones */
1592
1593 zlc = zonelist->zlcache_ptr;
1594 if (!zlc)
1595 return;
1596
Mel Gormandd1a2392008-04-28 02:12:17 -07001597 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001598
1599 set_bit(i, zlc->fullzones);
1600}
1601
Mel Gorman76d3fbf2011-07-25 17:12:30 -07001602/*
1603 * clear all zones full, called after direct reclaim makes progress so that
1604 * a zone that was recently full is not skipped over for up to a second
1605 */
1606static void zlc_clear_zones_full(struct zonelist *zonelist)
1607{
1608 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1609
1610 zlc = zonelist->zlcache_ptr;
1611 if (!zlc)
1612 return;
1613
1614 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1615}
1616
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001617#else /* CONFIG_NUMA */
1618
1619static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1620{
1621 return NULL;
1622}
1623
Mel Gormandd1a2392008-04-28 02:12:17 -07001624static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001625 nodemask_t *allowednodes)
1626{
1627 return 1;
1628}
1629
Mel Gormandd1a2392008-04-28 02:12:17 -07001630static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001631{
1632}
Mel Gorman76d3fbf2011-07-25 17:12:30 -07001633
1634static void zlc_clear_zones_full(struct zonelist *zonelist)
1635{
1636}
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001637#endif /* CONFIG_NUMA */
1638
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001639/*
Paul Jackson0798e512006-12-06 20:31:38 -08001640 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001641 * a page.
1642 */
1643static struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001644get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
Mel Gorman5117f452009-06-16 15:31:59 -07001645 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
Mel Gorman3dd28262009-06-16 15:32:00 -07001646 struct zone *preferred_zone, int migratetype)
Martin Hicks753ee722005-06-21 17:14:41 -07001647{
Mel Gormandd1a2392008-04-28 02:12:17 -07001648 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001649 struct page *page = NULL;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001650 int classzone_idx;
Mel Gorman5117f452009-06-16 15:31:59 -07001651 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001652 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1653 int zlc_active = 0; /* set if using zonelist_cache */
1654 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gorman54a6eb52008-04-28 02:12:16 -07001655
Mel Gorman19770b32008-04-28 02:12:18 -07001656 classzone_idx = zone_idx(preferred_zone);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001657zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001658 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001659 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001660 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1661 */
Mel Gorman19770b32008-04-28 02:12:18 -07001662 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1663 high_zoneidx, nodemask) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001664 if (NUMA_BUILD && zlc_active &&
1665 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1666 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001667 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001668 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Mel Gormancd38b112011-07-25 17:12:29 -07001669 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001670
Mel Gorman41858962009-06-16 15:32:12 -07001671 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001672 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001673 unsigned long mark;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001674 int ret;
1675
Mel Gorman41858962009-06-16 15:32:12 -07001676 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
Mel Gormanfa5e0842009-06-16 15:33:22 -07001677 if (zone_watermark_ok(zone, order, mark,
1678 classzone_idx, alloc_flags))
1679 goto try_this_zone;
1680
Mel Gormancd38b112011-07-25 17:12:29 -07001681 if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) {
1682 /*
1683 * we do zlc_setup if there are multiple nodes
1684 * and before considering the first zone allowed
1685 * by the cpuset.
1686 */
1687 allowednodes = zlc_setup(zonelist, alloc_flags);
1688 zlc_active = 1;
1689 did_zlc_setup = 1;
1690 }
1691
Mel Gormanfa5e0842009-06-16 15:33:22 -07001692 if (zone_reclaim_mode == 0)
1693 goto this_zone_full;
1694
Mel Gormancd38b112011-07-25 17:12:29 -07001695 /*
1696 * As we may have just activated ZLC, check if the first
1697 * eligible zone has failed zone_reclaim recently.
1698 */
1699 if (NUMA_BUILD && zlc_active &&
1700 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1701 continue;
1702
Mel Gormanfa5e0842009-06-16 15:33:22 -07001703 ret = zone_reclaim(zone, gfp_mask, order);
1704 switch (ret) {
1705 case ZONE_RECLAIM_NOSCAN:
1706 /* did not scan */
Mel Gormancd38b112011-07-25 17:12:29 -07001707 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001708 case ZONE_RECLAIM_FULL:
1709 /* scanned but unreclaimable */
Mel Gormancd38b112011-07-25 17:12:29 -07001710 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001711 default:
1712 /* did we reclaim enough */
1713 if (!zone_watermark_ok(zone, order, mark,
1714 classzone_idx, alloc_flags))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001715 goto this_zone_full;
Paul Jackson0798e512006-12-06 20:31:38 -08001716 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001717 }
1718
Mel Gormanfa5e0842009-06-16 15:33:22 -07001719try_this_zone:
Mel Gorman3dd28262009-06-16 15:32:00 -07001720 page = buffered_rmqueue(preferred_zone, zone, order,
1721 gfp_mask, migratetype);
Paul Jackson0798e512006-12-06 20:31:38 -08001722 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001723 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001724this_zone_full:
1725 if (NUMA_BUILD)
1726 zlc_mark_zone_full(zonelist, z);
Mel Gorman54a6eb52008-04-28 02:12:16 -07001727 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001728
1729 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1730 /* Disable zlc cache for second zonelist scan */
1731 zlc_active = 0;
1732 goto zonelist_scan;
1733 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001734 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07001735}
1736
David Rientjes29423e772011-03-22 16:30:47 -07001737/*
1738 * Large machines with many possible nodes should not always dump per-node
1739 * meminfo in irq context.
1740 */
1741static inline bool should_suppress_show_mem(void)
1742{
1743 bool ret = false;
1744
1745#if NODES_SHIFT > 8
1746 ret = in_interrupt();
1747#endif
1748 return ret;
1749}
1750
Dave Hansena238ab52011-05-24 17:12:16 -07001751static DEFINE_RATELIMIT_STATE(nopage_rs,
1752 DEFAULT_RATELIMIT_INTERVAL,
1753 DEFAULT_RATELIMIT_BURST);
1754
1755void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
1756{
Dave Hansena238ab52011-05-24 17:12:16 -07001757 unsigned int filter = SHOW_MEM_FILTER_NODES;
1758
1759 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs))
1760 return;
1761
1762 /*
1763 * This documents exceptions given to allocations in certain
1764 * contexts that are allowed to allocate outside current's set
1765 * of allowed nodes.
1766 */
1767 if (!(gfp_mask & __GFP_NOMEMALLOC))
1768 if (test_thread_flag(TIF_MEMDIE) ||
1769 (current->flags & (PF_MEMALLOC | PF_EXITING)))
1770 filter &= ~SHOW_MEM_FILTER_NODES;
1771 if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
1772 filter &= ~SHOW_MEM_FILTER_NODES;
1773
1774 if (fmt) {
Joe Perches3ee9a4f2011-10-31 17:08:35 -07001775 struct va_format vaf;
1776 va_list args;
1777
Dave Hansena238ab52011-05-24 17:12:16 -07001778 va_start(args, fmt);
Joe Perches3ee9a4f2011-10-31 17:08:35 -07001779
1780 vaf.fmt = fmt;
1781 vaf.va = &args;
1782
1783 pr_warn("%pV", &vaf);
1784
Dave Hansena238ab52011-05-24 17:12:16 -07001785 va_end(args);
1786 }
1787
Joe Perches3ee9a4f2011-10-31 17:08:35 -07001788 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
1789 current->comm, order, gfp_mask);
Dave Hansena238ab52011-05-24 17:12:16 -07001790
1791 dump_stack();
1792 if (!should_suppress_show_mem())
1793 show_mem(filter);
1794}
1795
Mel Gorman11e33f62009-06-16 15:31:57 -07001796static inline int
1797should_alloc_retry(gfp_t gfp_mask, unsigned int order,
1798 unsigned long pages_reclaimed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799{
Mel Gorman11e33f62009-06-16 15:31:57 -07001800 /* Do not loop if specifically requested */
1801 if (gfp_mask & __GFP_NORETRY)
1802 return 0;
1803
1804 /*
1805 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1806 * means __GFP_NOFAIL, but that may not be true in other
1807 * implementations.
1808 */
1809 if (order <= PAGE_ALLOC_COSTLY_ORDER)
1810 return 1;
1811
1812 /*
1813 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1814 * specified, then we retry until we no longer reclaim any pages
1815 * (above), or we've reclaimed an order of pages at least as
1816 * large as the allocation's order. In both cases, if the
1817 * allocation still fails, we stop retrying.
1818 */
1819 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
1820 return 1;
1821
1822 /*
1823 * Don't let big-order allocations loop unless the caller
1824 * explicitly requests that.
1825 */
1826 if (gfp_mask & __GFP_NOFAIL)
1827 return 1;
1828
1829 return 0;
1830}
1831
1832static inline struct page *
1833__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1834 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001835 nodemask_t *nodemask, struct zone *preferred_zone,
1836 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001837{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839
Mel Gorman11e33f62009-06-16 15:31:57 -07001840 /* Acquire the OOM killer lock for the zones in zonelist */
Minchan Kimff321fe2010-08-09 17:18:57 -07001841 if (!try_set_zonelist_oom(zonelist, gfp_mask)) {
Mel Gorman11e33f62009-06-16 15:31:57 -07001842 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843 return NULL;
1844 }
Jens Axboe6b1de912005-11-17 21:35:02 +01001845
Mel Gorman11e33f62009-06-16 15:31:57 -07001846 /*
1847 * Go through the zonelist yet one more time, keep very high watermark
1848 * here, this is only to catch a parallel oom killing, we must fail if
1849 * we're still under heavy pressure.
1850 */
1851 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1852 order, zonelist, high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001853 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07001854 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001855 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07001856 goto out;
1857
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001858 if (!(gfp_mask & __GFP_NOFAIL)) {
1859 /* The OOM killer will not help higher order allocs */
1860 if (order > PAGE_ALLOC_COSTLY_ORDER)
1861 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07001862 /* The OOM killer does not needlessly kill tasks for lowmem */
1863 if (high_zoneidx < ZONE_NORMAL)
1864 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001865 /*
1866 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
1867 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
1868 * The caller should handle page allocation failure by itself if
1869 * it specifies __GFP_THISNODE.
1870 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
1871 */
1872 if (gfp_mask & __GFP_THISNODE)
1873 goto out;
1874 }
Mel Gorman11e33f62009-06-16 15:31:57 -07001875 /* Exhausted what can be done so it's blamo time */
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001876 out_of_memory(zonelist, gfp_mask, order, nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07001877
1878out:
1879 clear_zonelist_oom(zonelist, gfp_mask);
1880 return page;
1881}
1882
Mel Gorman56de7262010-05-24 14:32:30 -07001883#ifdef CONFIG_COMPACTION
1884/* Try memory compaction for high-order allocations before reclaim */
1885static struct page *
1886__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
1887 struct zonelist *zonelist, enum zone_type high_zoneidx,
1888 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman77f1fe62011-01-13 15:45:57 -08001889 int migratetype, unsigned long *did_some_progress,
1890 bool sync_migration)
Mel Gorman56de7262010-05-24 14:32:30 -07001891{
1892 struct page *page;
1893
Mel Gorman4f92e252010-05-24 14:32:32 -07001894 if (!order || compaction_deferred(preferred_zone))
Mel Gorman56de7262010-05-24 14:32:30 -07001895 return NULL;
1896
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001897 current->flags |= PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07001898 *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
Mel Gorman77f1fe62011-01-13 15:45:57 -08001899 nodemask, sync_migration);
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001900 current->flags &= ~PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07001901 if (*did_some_progress != COMPACT_SKIPPED) {
1902
1903 /* Page migration frees to the PCP lists but we want merging */
1904 drain_pages(get_cpu());
1905 put_cpu();
1906
1907 page = get_page_from_freelist(gfp_mask, nodemask,
1908 order, zonelist, high_zoneidx,
1909 alloc_flags, preferred_zone,
1910 migratetype);
1911 if (page) {
Mel Gorman4f92e252010-05-24 14:32:32 -07001912 preferred_zone->compact_considered = 0;
1913 preferred_zone->compact_defer_shift = 0;
Mel Gorman56de7262010-05-24 14:32:30 -07001914 count_vm_event(COMPACTSUCCESS);
1915 return page;
1916 }
1917
1918 /*
1919 * It's bad if compaction run occurs and fails.
1920 * The most likely reason is that pages exist,
1921 * but not enough to satisfy watermarks.
1922 */
1923 count_vm_event(COMPACTFAIL);
Mel Gorman4f92e252010-05-24 14:32:32 -07001924 defer_compaction(preferred_zone);
Mel Gorman56de7262010-05-24 14:32:30 -07001925
1926 cond_resched();
1927 }
1928
1929 return NULL;
1930}
1931#else
1932static inline struct page *
1933__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
1934 struct zonelist *zonelist, enum zone_type high_zoneidx,
1935 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman77f1fe62011-01-13 15:45:57 -08001936 int migratetype, unsigned long *did_some_progress,
1937 bool sync_migration)
Mel Gorman56de7262010-05-24 14:32:30 -07001938{
1939 return NULL;
1940}
1941#endif /* CONFIG_COMPACTION */
1942
Mel Gorman11e33f62009-06-16 15:31:57 -07001943/* The really slow allocator path where we enter direct reclaim */
1944static inline struct page *
1945__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
1946 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001947 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001948 int migratetype, unsigned long *did_some_progress)
Mel Gorman11e33f62009-06-16 15:31:57 -07001949{
1950 struct page *page = NULL;
1951 struct reclaim_state reclaim_state;
Mel Gorman9ee493c2010-09-09 16:38:18 -07001952 bool drained = false;
Mel Gorman11e33f62009-06-16 15:31:57 -07001953
1954 cond_resched();
1955
1956 /* We now go into synchronous reclaim */
1957 cpuset_memory_pressure_bump();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001958 current->flags |= PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07001959 lockdep_set_current_reclaim_state(gfp_mask);
1960 reclaim_state.reclaimed_slab = 0;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001961 current->reclaim_state = &reclaim_state;
Mel Gorman11e33f62009-06-16 15:31:57 -07001962
1963 *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
1964
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001965 current->reclaim_state = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07001966 lockdep_clear_current_reclaim_state();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001967 current->flags &= ~PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07001968
1969 cond_resched();
1970
Mel Gorman9ee493c2010-09-09 16:38:18 -07001971 if (unlikely(!(*did_some_progress)))
1972 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07001973
Mel Gorman76d3fbf2011-07-25 17:12:30 -07001974 /* After successful reclaim, reconsider all zones for allocation */
1975 if (NUMA_BUILD)
1976 zlc_clear_zones_full(zonelist);
1977
Mel Gorman9ee493c2010-09-09 16:38:18 -07001978retry:
1979 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001980 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001981 alloc_flags, preferred_zone,
1982 migratetype);
Mel Gorman9ee493c2010-09-09 16:38:18 -07001983
1984 /*
1985 * If an allocation failed after direct reclaim, it could be because
1986 * pages are pinned on the per-cpu lists. Drain them and try again
1987 */
1988 if (!page && !drained) {
1989 drain_all_pages();
1990 drained = true;
1991 goto retry;
1992 }
1993
Mel Gorman11e33f62009-06-16 15:31:57 -07001994 return page;
1995}
1996
Mel Gorman11e33f62009-06-16 15:31:57 -07001997/*
1998 * This is called in the allocator slow-path if the allocation request is of
1999 * sufficient urgency to ignore watermarks and take other desperate measures
2000 */
2001static inline struct page *
2002__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
2003 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002004 nodemask_t *nodemask, struct zone *preferred_zone,
2005 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002006{
2007 struct page *page;
2008
2009 do {
2010 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002011 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
Mel Gorman3dd28262009-06-16 15:32:00 -07002012 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002013
2014 if (!page && gfp_mask & __GFP_NOFAIL)
Mel Gorman0e093d992010-10-26 14:21:45 -07002015 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07002016 } while (!page && (gfp_mask & __GFP_NOFAIL));
2017
2018 return page;
2019}
2020
2021static inline
2022void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
Mel Gorman99504742011-01-13 15:46:20 -08002023 enum zone_type high_zoneidx,
2024 enum zone_type classzone_idx)
Mel Gorman11e33f62009-06-16 15:31:57 -07002025{
2026 struct zoneref *z;
2027 struct zone *zone;
2028
2029 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
Mel Gorman99504742011-01-13 15:46:20 -08002030 wakeup_kswapd(zone, order, classzone_idx);
Mel Gorman11e33f62009-06-16 15:31:57 -07002031}
2032
Peter Zijlstra341ce062009-06-16 15:32:02 -07002033static inline int
2034gfp_to_alloc_flags(gfp_t gfp_mask)
2035{
Peter Zijlstra341ce062009-06-16 15:32:02 -07002036 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
2037 const gfp_t wait = gfp_mask & __GFP_WAIT;
2038
Mel Gormana56f57f2009-06-16 15:32:02 -07002039 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07002040 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07002041
Peter Zijlstra341ce062009-06-16 15:32:02 -07002042 /*
2043 * The caller may dip into page reserves a bit more if the caller
2044 * cannot run direct reclaim, or if the caller has realtime scheduling
2045 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2046 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
2047 */
Namhyung Kime6223a32010-10-26 14:21:59 -07002048 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07002049
2050 if (!wait) {
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002051 /*
2052 * Not worth trying to allocate harder for
2053 * __GFP_NOMEMALLOC even if it can't schedule.
2054 */
2055 if (!(gfp_mask & __GFP_NOMEMALLOC))
2056 alloc_flags |= ALLOC_HARDER;
Peter Zijlstra341ce062009-06-16 15:32:02 -07002057 /*
2058 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
2059 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
2060 */
2061 alloc_flags &= ~ALLOC_CPUSET;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002062 } else if (unlikely(rt_task(current)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07002063 alloc_flags |= ALLOC_HARDER;
2064
2065 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2066 if (!in_interrupt() &&
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002067 ((current->flags & PF_MEMALLOC) ||
Peter Zijlstra341ce062009-06-16 15:32:02 -07002068 unlikely(test_thread_flag(TIF_MEMDIE))))
2069 alloc_flags |= ALLOC_NO_WATERMARKS;
2070 }
2071
2072 return alloc_flags;
2073}
2074
Mel Gorman11e33f62009-06-16 15:31:57 -07002075static inline struct page *
2076__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
2077 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002078 nodemask_t *nodemask, struct zone *preferred_zone,
2079 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002080{
2081 const gfp_t wait = gfp_mask & __GFP_WAIT;
2082 struct page *page = NULL;
2083 int alloc_flags;
2084 unsigned long pages_reclaimed = 0;
2085 unsigned long did_some_progress;
Mel Gorman77f1fe62011-01-13 15:45:57 -08002086 bool sync_migration = false;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002087
Christoph Lameter952f3b52006-12-06 20:33:26 -08002088 /*
Mel Gorman72807a72009-06-16 15:32:18 -07002089 * In the slowpath, we sanity check order to avoid ever trying to
2090 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2091 * be using allocators in order of preference for an area that is
2092 * too large.
2093 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07002094 if (order >= MAX_ORDER) {
2095 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07002096 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07002097 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002098
Christoph Lameter952f3b52006-12-06 20:33:26 -08002099 /*
2100 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2101 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2102 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2103 * using a larger set of nodes after it has established that the
2104 * allowed per node queues are empty and that nodes are
2105 * over allocated.
2106 */
2107 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
2108 goto nopage;
2109
Mel Gormancc4a6852009-11-11 14:26:14 -08002110restart:
Andrea Arcangeli32dba982011-01-13 15:46:49 -08002111 if (!(gfp_mask & __GFP_NO_KSWAPD))
2112 wake_all_kswapd(order, zonelist, high_zoneidx,
Mel Gorman99504742011-01-13 15:46:20 -08002113 zone_idx(preferred_zone));
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002114
Paul Jackson9bf22292005-09-06 15:18:12 -07002115 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002116 * OK, we're below the kswapd watermark and have kicked background
2117 * reclaim. Now things get more complex, so set up alloc_flags according
2118 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07002119 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002120 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121
David Rientjesf33261d2011-01-25 15:07:20 -08002122 /*
2123 * Find the true preferred zone if the allocation is unconstrained by
2124 * cpusets.
2125 */
2126 if (!(alloc_flags & ALLOC_CPUSET) && !nodemask)
2127 first_zones_zonelist(zonelist, high_zoneidx, NULL,
2128 &preferred_zone);
2129
Andrew Barrycfa54a02011-05-24 17:12:52 -07002130rebalance:
Peter Zijlstra341ce062009-06-16 15:32:02 -07002131 /* This is the last chance, in general, before the goto nopage. */
Mel Gorman19770b32008-04-28 02:12:18 -07002132 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
Peter Zijlstra341ce062009-06-16 15:32:02 -07002133 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
2134 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002135 if (page)
2136 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137
Mel Gorman11e33f62009-06-16 15:31:57 -07002138 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002139 if (alloc_flags & ALLOC_NO_WATERMARKS) {
2140 page = __alloc_pages_high_priority(gfp_mask, order,
2141 zonelist, high_zoneidx, nodemask,
2142 preferred_zone, migratetype);
2143 if (page)
2144 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 }
2146
2147 /* Atomic allocations - we can't balance anything */
2148 if (!wait)
2149 goto nopage;
2150
Peter Zijlstra341ce062009-06-16 15:32:02 -07002151 /* Avoid recursion of direct reclaim */
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002152 if (current->flags & PF_MEMALLOC)
Peter Zijlstra341ce062009-06-16 15:32:02 -07002153 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154
David Rientjes6583bb62009-07-29 15:02:06 -07002155 /* Avoid allocations with no watermarks from looping endlessly */
2156 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2157 goto nopage;
2158
Mel Gorman77f1fe62011-01-13 15:45:57 -08002159 /*
2160 * Try direct compaction. The first pass is asynchronous. Subsequent
2161 * attempts after direct reclaim are synchronous
2162 */
Mel Gorman56de7262010-05-24 14:32:30 -07002163 page = __alloc_pages_direct_compact(gfp_mask, order,
2164 zonelist, high_zoneidx,
2165 nodemask,
2166 alloc_flags, preferred_zone,
Mel Gorman77f1fe62011-01-13 15:45:57 -08002167 migratetype, &did_some_progress,
2168 sync_migration);
Mel Gorman56de7262010-05-24 14:32:30 -07002169 if (page)
2170 goto got_pg;
Andrea Arcangelic6a140b2011-05-24 17:11:38 -07002171 sync_migration = true;
Mel Gorman56de7262010-05-24 14:32:30 -07002172
Mel Gorman11e33f62009-06-16 15:31:57 -07002173 /* Try direct reclaim and then allocating */
2174 page = __alloc_pages_direct_reclaim(gfp_mask, order,
2175 zonelist, high_zoneidx,
2176 nodemask,
Mel Gorman5117f452009-06-16 15:31:59 -07002177 alloc_flags, preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07002178 migratetype, &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07002179 if (page)
2180 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 /*
Mel Gorman11e33f62009-06-16 15:31:57 -07002183 * If we failed to make any progress reclaiming, then we are
2184 * running out of options and have to consider going OOM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 */
Mel Gorman11e33f62009-06-16 15:31:57 -07002186 if (!did_some_progress) {
2187 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -07002188 if (oom_killer_disabled)
2189 goto nopage;
Mel Gorman11e33f62009-06-16 15:31:57 -07002190 page = __alloc_pages_may_oom(gfp_mask, order,
2191 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002192 nodemask, preferred_zone,
2193 migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002194 if (page)
2195 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196
David Rientjes03668b32010-08-09 17:18:54 -07002197 if (!(gfp_mask & __GFP_NOFAIL)) {
2198 /*
2199 * The oom killer is not called for high-order
2200 * allocations that may fail, so if no progress
2201 * is being made, there are no other options and
2202 * retrying is unlikely to help.
2203 */
2204 if (order > PAGE_ALLOC_COSTLY_ORDER)
2205 goto nopage;
2206 /*
2207 * The oom killer is not called for lowmem
2208 * allocations to prevent needlessly killing
2209 * innocent tasks.
2210 */
2211 if (high_zoneidx < ZONE_NORMAL)
2212 goto nopage;
2213 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215 goto restart;
2216 }
2217 }
2218
Mel Gorman11e33f62009-06-16 15:31:57 -07002219 /* Check if we should retry the allocation */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07002220 pages_reclaimed += did_some_progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07002221 if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
2222 /* Wait for some write requests to complete then retry */
Mel Gorman0e093d992010-10-26 14:21:45 -07002223 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 goto rebalance;
Mel Gorman3e7d3442011-01-13 15:45:56 -08002225 } else {
2226 /*
2227 * High-order allocations do not necessarily loop after
2228 * direct reclaim and reclaim/compaction depends on compaction
2229 * being called after reclaim so call directly if necessary
2230 */
2231 page = __alloc_pages_direct_compact(gfp_mask, order,
2232 zonelist, high_zoneidx,
2233 nodemask,
2234 alloc_flags, preferred_zone,
Mel Gorman77f1fe62011-01-13 15:45:57 -08002235 migratetype, &did_some_progress,
2236 sync_migration);
Mel Gorman3e7d3442011-01-13 15:45:56 -08002237 if (page)
2238 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 }
2240
2241nopage:
Dave Hansena238ab52011-05-24 17:12:16 -07002242 warn_alloc_failed(gfp_mask, order, NULL);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002243 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244got_pg:
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002245 if (kmemcheck_enabled)
2246 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 return page;
Mel Gorman11e33f62009-06-16 15:31:57 -07002248
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249}
Mel Gorman11e33f62009-06-16 15:31:57 -07002250
2251/*
2252 * This is the 'heart' of the zoned buddy allocator.
2253 */
2254struct page *
2255__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2256 struct zonelist *zonelist, nodemask_t *nodemask)
2257{
2258 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Mel Gorman5117f452009-06-16 15:31:59 -07002259 struct zone *preferred_zone;
Mel Gorman11e33f62009-06-16 15:31:57 -07002260 struct page *page;
Mel Gorman3dd28262009-06-16 15:32:00 -07002261 int migratetype = allocflags_to_migratetype(gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002262
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10002263 gfp_mask &= gfp_allowed_mask;
2264
Mel Gorman11e33f62009-06-16 15:31:57 -07002265 lockdep_trace_alloc(gfp_mask);
2266
2267 might_sleep_if(gfp_mask & __GFP_WAIT);
2268
2269 if (should_fail_alloc_page(gfp_mask, order))
2270 return NULL;
2271
2272 /*
2273 * Check the zones suitable for the gfp_mask contain at least one
2274 * valid zone. It's possible to have an empty zonelist as a result
2275 * of GFP_THISNODE and a memoryless node
2276 */
2277 if (unlikely(!zonelist->_zonerefs->zone))
2278 return NULL;
2279
Miao Xiec0ff7452010-05-24 14:32:08 -07002280 get_mems_allowed();
Mel Gorman5117f452009-06-16 15:31:59 -07002281 /* The preferred zone is used for statistics later */
David Rientjesf33261d2011-01-25 15:07:20 -08002282 first_zones_zonelist(zonelist, high_zoneidx,
2283 nodemask ? : &cpuset_current_mems_allowed,
2284 &preferred_zone);
Miao Xiec0ff7452010-05-24 14:32:08 -07002285 if (!preferred_zone) {
2286 put_mems_allowed();
Mel Gorman5117f452009-06-16 15:31:59 -07002287 return NULL;
Miao Xiec0ff7452010-05-24 14:32:08 -07002288 }
Mel Gorman5117f452009-06-16 15:31:59 -07002289
2290 /* First allocation attempt */
Mel Gorman11e33f62009-06-16 15:31:57 -07002291 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002292 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07002293 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002294 if (unlikely(!page))
2295 page = __alloc_pages_slowpath(gfp_mask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002296 zonelist, high_zoneidx, nodemask,
Mel Gorman3dd28262009-06-16 15:32:00 -07002297 preferred_zone, migratetype);
Miao Xiec0ff7452010-05-24 14:32:08 -07002298 put_mems_allowed();
Mel Gorman11e33f62009-06-16 15:31:57 -07002299
Mel Gorman4b4f2782009-09-21 17:02:41 -07002300 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002301 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302}
Mel Gormand2391712009-06-16 15:31:52 -07002303EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304
2305/*
2306 * Common helper functions.
2307 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08002308unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309{
Akinobu Mita945a1112009-09-21 17:01:47 -07002310 struct page *page;
2311
2312 /*
2313 * __get_free_pages() returns a 32-bit address, which cannot represent
2314 * a highmem page
2315 */
2316 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2317
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318 page = alloc_pages(gfp_mask, order);
2319 if (!page)
2320 return 0;
2321 return (unsigned long) page_address(page);
2322}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323EXPORT_SYMBOL(__get_free_pages);
2324
Harvey Harrison920c7a52008-02-04 22:29:26 -08002325unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326{
Akinobu Mita945a1112009-09-21 17:01:47 -07002327 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329EXPORT_SYMBOL(get_zeroed_page);
2330
2331void __pagevec_free(struct pagevec *pvec)
2332{
2333 int i = pagevec_count(pvec);
2334
Mel Gorman4b4f2782009-09-21 17:02:41 -07002335 while (--i >= 0) {
2336 trace_mm_pagevec_free(pvec->pages[i], pvec->cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337 free_hot_cold_page(pvec->pages[i], pvec->cold);
Mel Gorman4b4f2782009-09-21 17:02:41 -07002338 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339}
2340
Harvey Harrison920c7a52008-02-04 22:29:26 -08002341void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342{
Nick Pigginb5810032005-10-29 18:16:12 -07002343 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344 if (order == 0)
Li Hongfc916682010-03-05 13:41:54 -08002345 free_hot_cold_page(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 else
2347 __free_pages_ok(page, order);
2348 }
2349}
2350
2351EXPORT_SYMBOL(__free_pages);
2352
Harvey Harrison920c7a52008-02-04 22:29:26 -08002353void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354{
2355 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07002356 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357 __free_pages(virt_to_page((void *)addr), order);
2358 }
2359}
2360
2361EXPORT_SYMBOL(free_pages);
2362
Andi Kleenee85c2e2011-05-11 15:13:34 -07002363static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
2364{
2365 if (addr) {
2366 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2367 unsigned long used = addr + PAGE_ALIGN(size);
2368
2369 split_page(virt_to_page((void *)addr), order);
2370 while (used < alloc_end) {
2371 free_page(used);
2372 used += PAGE_SIZE;
2373 }
2374 }
2375 return (void *)addr;
2376}
2377
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002378/**
2379 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2380 * @size: the number of bytes to allocate
2381 * @gfp_mask: GFP flags for the allocation
2382 *
2383 * This function is similar to alloc_pages(), except that it allocates the
2384 * minimum number of pages to satisfy the request. alloc_pages() can only
2385 * allocate memory in power-of-two pages.
2386 *
2387 * This function is also limited by MAX_ORDER.
2388 *
2389 * Memory allocated by this function must be released by free_pages_exact().
2390 */
2391void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2392{
2393 unsigned int order = get_order(size);
2394 unsigned long addr;
2395
2396 addr = __get_free_pages(gfp_mask, order);
Andi Kleenee85c2e2011-05-11 15:13:34 -07002397 return make_alloc_exact(addr, order, size);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002398}
2399EXPORT_SYMBOL(alloc_pages_exact);
2400
2401/**
Andi Kleenee85c2e2011-05-11 15:13:34 -07002402 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2403 * pages on a node.
Randy Dunlapb5e6ab52011-05-16 13:16:54 -07002404 * @nid: the preferred node ID where memory should be allocated
Andi Kleenee85c2e2011-05-11 15:13:34 -07002405 * @size: the number of bytes to allocate
2406 * @gfp_mask: GFP flags for the allocation
2407 *
2408 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2409 * back.
2410 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2411 * but is not exact.
2412 */
2413void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
2414{
2415 unsigned order = get_order(size);
2416 struct page *p = alloc_pages_node(nid, gfp_mask, order);
2417 if (!p)
2418 return NULL;
2419 return make_alloc_exact((unsigned long)page_address(p), order, size);
2420}
2421EXPORT_SYMBOL(alloc_pages_exact_nid);
2422
2423/**
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002424 * free_pages_exact - release memory allocated via alloc_pages_exact()
2425 * @virt: the value returned by alloc_pages_exact.
2426 * @size: size of allocation, same value as passed to alloc_pages_exact().
2427 *
2428 * Release the memory allocated by a previous call to alloc_pages_exact.
2429 */
2430void free_pages_exact(void *virt, size_t size)
2431{
2432 unsigned long addr = (unsigned long)virt;
2433 unsigned long end = addr + PAGE_ALIGN(size);
2434
2435 while (addr < end) {
2436 free_page(addr);
2437 addr += PAGE_SIZE;
2438 }
2439}
2440EXPORT_SYMBOL(free_pages_exact);
2441
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442static unsigned int nr_free_zone_pages(int offset)
2443{
Mel Gormandd1a2392008-04-28 02:12:17 -07002444 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002445 struct zone *zone;
2446
Martin J. Blighe310fd42005-07-29 22:59:18 -07002447 /* Just pick one node, since fallback list is circular */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448 unsigned int sum = 0;
2449
Mel Gorman0e884602008-04-28 02:12:14 -07002450 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451
Mel Gorman54a6eb52008-04-28 02:12:16 -07002452 for_each_zone_zonelist(zone, z, zonelist, offset) {
Martin J. Blighe310fd42005-07-29 22:59:18 -07002453 unsigned long size = zone->present_pages;
Mel Gorman41858962009-06-16 15:32:12 -07002454 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07002455 if (size > high)
2456 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 }
2458
2459 return sum;
2460}
2461
2462/*
2463 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2464 */
2465unsigned int nr_free_buffer_pages(void)
2466{
Al Viroaf4ca452005-10-21 02:55:38 -04002467 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468}
Meelap Shahc2f1a552007-07-17 04:04:39 -07002469EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470
2471/*
2472 * Amount of free RAM allocatable within all zones
2473 */
2474unsigned int nr_free_pagecache_pages(void)
2475{
Mel Gorman2a1e2742007-07-17 04:03:12 -07002476 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002478
2479static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480{
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002481 if (NUMA_BUILD)
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08002482 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485void si_meminfo(struct sysinfo *val)
2486{
2487 val->totalram = totalram_pages;
2488 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002489 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491 val->totalhigh = totalhigh_pages;
2492 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493 val->mem_unit = PAGE_SIZE;
2494}
2495
2496EXPORT_SYMBOL(si_meminfo);
2497
2498#ifdef CONFIG_NUMA
2499void si_meminfo_node(struct sysinfo *val, int nid)
2500{
2501 pg_data_t *pgdat = NODE_DATA(nid);
2502
2503 val->totalram = pgdat->node_present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002504 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002505#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002507 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2508 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002509#else
2510 val->totalhigh = 0;
2511 val->freehigh = 0;
2512#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513 val->mem_unit = PAGE_SIZE;
2514}
2515#endif
2516
David Rientjesddd588b2011-03-22 16:30:46 -07002517/*
David Rientjes7bf02ea2011-05-24 17:11:16 -07002518 * Determine whether the node should be displayed or not, depending on whether
2519 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
David Rientjesddd588b2011-03-22 16:30:46 -07002520 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07002521bool skip_free_areas_node(unsigned int flags, int nid)
David Rientjesddd588b2011-03-22 16:30:46 -07002522{
2523 bool ret = false;
2524
2525 if (!(flags & SHOW_MEM_FILTER_NODES))
2526 goto out;
2527
2528 get_mems_allowed();
David Rientjes7bf02ea2011-05-24 17:11:16 -07002529 ret = !node_isset(nid, cpuset_current_mems_allowed);
David Rientjesddd588b2011-03-22 16:30:46 -07002530 put_mems_allowed();
2531out:
2532 return ret;
2533}
2534
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535#define K(x) ((x) << (PAGE_SHIFT-10))
2536
2537/*
2538 * Show free area list (used inside shift_scroll-lock stuff)
2539 * We also calculate the percentage fragmentation. We do this by counting the
2540 * memory on each free list with the exception of the first item on the list.
David Rientjesddd588b2011-03-22 16:30:46 -07002541 * Suppresses nodes that are not allowed by current's cpuset if
2542 * SHOW_MEM_FILTER_NODES is passed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07002544void show_free_areas(unsigned int filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545{
Jes Sorensenc7241912006-09-27 01:50:05 -07002546 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547 struct zone *zone;
2548
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002549 for_each_populated_zone(zone) {
David Rientjes7bf02ea2011-05-24 17:11:16 -07002550 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07002551 continue;
Jes Sorensenc7241912006-09-27 01:50:05 -07002552 show_node(zone);
2553 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554
Dave Jones6b482c62005-11-10 15:45:56 -05002555 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 struct per_cpu_pageset *pageset;
2557
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002558 pageset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002560 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2561 cpu, pageset->pcp.high,
2562 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563 }
2564 }
2565
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002566 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2567 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002568 " unevictable:%lu"
Andrew Mortonb76146e2009-10-26 16:49:52 -07002569 " dirty:%lu writeback:%lu unstable:%lu\n"
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07002570 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002571 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002572 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002573 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002574 global_page_state(NR_ISOLATED_ANON),
2575 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002576 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002577 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002578 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07002579 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07002580 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07002581 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08002582 global_page_state(NR_FREE_PAGES),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07002583 global_page_state(NR_SLAB_RECLAIMABLE),
2584 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07002585 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002586 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08002587 global_page_state(NR_PAGETABLE),
2588 global_page_state(NR_BOUNCE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002590 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591 int i;
2592
David Rientjes7bf02ea2011-05-24 17:11:16 -07002593 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07002594 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595 show_node(zone);
2596 printk("%s"
2597 " free:%lukB"
2598 " min:%lukB"
2599 " low:%lukB"
2600 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002601 " active_anon:%lukB"
2602 " inactive_anon:%lukB"
2603 " active_file:%lukB"
2604 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002605 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002606 " isolated(anon):%lukB"
2607 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608 " present:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002609 " mlocked:%lukB"
2610 " dirty:%lukB"
2611 " writeback:%lukB"
2612 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002613 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002614 " slab_reclaimable:%lukB"
2615 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07002616 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002617 " pagetables:%lukB"
2618 " unstable:%lukB"
2619 " bounce:%lukB"
2620 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621 " pages_scanned:%lu"
2622 " all_unreclaimable? %s"
2623 "\n",
2624 zone->name,
Mel Gorman88f5acf2011-01-13 15:45:41 -08002625 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07002626 K(min_wmark_pages(zone)),
2627 K(low_wmark_pages(zone)),
2628 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002629 K(zone_page_state(zone, NR_ACTIVE_ANON)),
2630 K(zone_page_state(zone, NR_INACTIVE_ANON)),
2631 K(zone_page_state(zone, NR_ACTIVE_FILE)),
2632 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002633 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002634 K(zone_page_state(zone, NR_ISOLATED_ANON)),
2635 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636 K(zone->present_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002637 K(zone_page_state(zone, NR_MLOCK)),
2638 K(zone_page_state(zone, NR_FILE_DIRTY)),
2639 K(zone_page_state(zone, NR_WRITEBACK)),
2640 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002641 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002642 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
2643 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07002644 zone_page_state(zone, NR_KERNEL_STACK) *
2645 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002646 K(zone_page_state(zone, NR_PAGETABLE)),
2647 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
2648 K(zone_page_state(zone, NR_BOUNCE)),
2649 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650 zone->pages_scanned,
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -08002651 (zone->all_unreclaimable ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652 );
2653 printk("lowmem_reserve[]:");
2654 for (i = 0; i < MAX_NR_ZONES; i++)
2655 printk(" %lu", zone->lowmem_reserve[i]);
2656 printk("\n");
2657 }
2658
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002659 for_each_populated_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002660 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661
David Rientjes7bf02ea2011-05-24 17:11:16 -07002662 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07002663 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664 show_node(zone);
2665 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666
2667 spin_lock_irqsave(&zone->lock, flags);
2668 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002669 nr[order] = zone->free_area[order].nr_free;
2670 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671 }
2672 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002673 for (order = 0; order < MAX_ORDER; order++)
2674 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675 printk("= %lukB\n", K(total));
2676 }
2677
Larry Woodmane6f36022008-02-04 22:29:30 -08002678 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2679
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680 show_swap_cache_info();
2681}
2682
Mel Gorman19770b32008-04-28 02:12:18 -07002683static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2684{
2685 zoneref->zone = zone;
2686 zoneref->zone_idx = zone_idx(zone);
2687}
2688
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689/*
2690 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08002691 *
2692 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002694static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2695 int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696{
Christoph Lameter1a932052006-01-06 00:11:16 -08002697 struct zone *zone;
2698
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002699 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002700 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08002701
2702 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002703 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08002704 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08002705 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002706 zoneref_set_zone(zone,
2707 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08002708 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08002710
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002711 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08002712 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713}
2714
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002715
2716/*
2717 * zonelist_order:
2718 * 0 = automatic detection of better ordering.
2719 * 1 = order by ([node] distance, -zonetype)
2720 * 2 = order by (-zonetype, [node] distance)
2721 *
2722 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2723 * the same zonelist. So only NUMA can configure this param.
2724 */
2725#define ZONELIST_ORDER_DEFAULT 0
2726#define ZONELIST_ORDER_NODE 1
2727#define ZONELIST_ORDER_ZONE 2
2728
2729/* zonelist order in the kernel.
2730 * set_zonelist_order() will set this to NODE or ZONE.
2731 */
2732static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2733static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2734
2735
Linus Torvalds1da177e2005-04-16 15:20:36 -07002736#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002737/* The value user specified ....changed by config */
2738static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2739/* string for sysctl */
2740#define NUMA_ZONELIST_ORDER_LEN 16
2741char numa_zonelist_order[16] = "default";
2742
2743/*
2744 * interface for configure zonelist ordering.
2745 * command line option "numa_zonelist_order"
2746 * = "[dD]efault - default, automatic configuration.
2747 * = "[nN]ode - order by node locality, then by zone within node
2748 * = "[zZ]one - order by zone, then by locality within zone
2749 */
2750
2751static int __parse_numa_zonelist_order(char *s)
2752{
2753 if (*s == 'd' || *s == 'D') {
2754 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2755 } else if (*s == 'n' || *s == 'N') {
2756 user_zonelist_order = ZONELIST_ORDER_NODE;
2757 } else if (*s == 'z' || *s == 'Z') {
2758 user_zonelist_order = ZONELIST_ORDER_ZONE;
2759 } else {
2760 printk(KERN_WARNING
2761 "Ignoring invalid numa_zonelist_order value: "
2762 "%s\n", s);
2763 return -EINVAL;
2764 }
2765 return 0;
2766}
2767
2768static __init int setup_numa_zonelist_order(char *s)
2769{
Volodymyr G. Lukiianykecb256f2011-01-13 15:46:26 -08002770 int ret;
2771
2772 if (!s)
2773 return 0;
2774
2775 ret = __parse_numa_zonelist_order(s);
2776 if (ret == 0)
2777 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
2778
2779 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002780}
2781early_param("numa_zonelist_order", setup_numa_zonelist_order);
2782
2783/*
2784 * sysctl handler for numa_zonelist_order
2785 */
2786int numa_zonelist_order_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07002787 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002788 loff_t *ppos)
2789{
2790 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2791 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01002792 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002793
Andi Kleen443c6f12009-12-23 21:00:47 +01002794 mutex_lock(&zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002795 if (write)
Andi Kleen443c6f12009-12-23 21:00:47 +01002796 strcpy(saved_string, (char*)table->data);
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07002797 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002798 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01002799 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002800 if (write) {
2801 int oldval = user_zonelist_order;
2802 if (__parse_numa_zonelist_order((char*)table->data)) {
2803 /*
2804 * bogus value. restore saved string
2805 */
2806 strncpy((char*)table->data, saved_string,
2807 NUMA_ZONELIST_ORDER_LEN);
2808 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07002809 } else if (oldval != user_zonelist_order) {
2810 mutex_lock(&zonelists_mutex);
Haicheng Li1f522502010-05-24 14:32:51 -07002811 build_all_zonelists(NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07002812 mutex_unlock(&zonelists_mutex);
2813 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002814 }
Andi Kleen443c6f12009-12-23 21:00:47 +01002815out:
2816 mutex_unlock(&zl_order_mutex);
2817 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002818}
2819
2820
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002821#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002822static int node_load[MAX_NUMNODES];
2823
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07002825 * 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 -07002826 * @node: node whose fallback list we're appending
2827 * @used_node_mask: nodemask_t of already used nodes
2828 *
2829 * We use a number of factors to determine which is the next node that should
2830 * appear on a given node's fallback list. The node should not have appeared
2831 * already in @node's fallback list, and it should be the next closest node
2832 * according to the distance array (which contains arbitrary distance values
2833 * from each node to each node in the system), and should also prefer nodes
2834 * with no CPUs, since presumably they'll have very little allocation pressure
2835 * on them otherwise.
2836 * It returns -1 if no node is found.
2837 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002838static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839{
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01002840 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841 int min_val = INT_MAX;
2842 int best_node = -1;
Rusty Russella70f7302009-03-13 14:49:46 +10302843 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01002845 /* Use the local node if we haven't already */
2846 if (!node_isset(node, *used_node_mask)) {
2847 node_set(node, *used_node_mask);
2848 return node;
2849 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002850
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002851 for_each_node_state(n, N_HIGH_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852
2853 /* Don't want a node to appear more than once */
2854 if (node_isset(n, *used_node_mask))
2855 continue;
2856
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857 /* Use the distance array to find the distance */
2858 val = node_distance(node, n);
2859
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01002860 /* Penalize nodes under us ("prefer the next node") */
2861 val += (n < node);
2862
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10302864 tmp = cpumask_of_node(n);
2865 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002866 val += PENALTY_FOR_NODE_WITH_CPUS;
2867
2868 /* Slight preference for less loaded node */
2869 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2870 val += node_load[n];
2871
2872 if (val < min_val) {
2873 min_val = val;
2874 best_node = n;
2875 }
2876 }
2877
2878 if (best_node >= 0)
2879 node_set(best_node, *used_node_mask);
2880
2881 return best_node;
2882}
2883
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002884
2885/*
2886 * Build zonelists ordered by node and zones within node.
2887 * This results in maximum locality--normal zone overflows into local
2888 * DMA zone, if any--but risks exhausting DMA zone.
2889 */
2890static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002892 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002893 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002894
Mel Gorman54a6eb52008-04-28 02:12:16 -07002895 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07002896 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07002897 ;
2898 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2899 MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002900 zonelist->_zonerefs[j].zone = NULL;
2901 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002902}
2903
2904/*
Christoph Lameter523b9452007-10-16 01:25:37 -07002905 * Build gfp_thisnode zonelists
2906 */
2907static void build_thisnode_zonelists(pg_data_t *pgdat)
2908{
Christoph Lameter523b9452007-10-16 01:25:37 -07002909 int j;
2910 struct zonelist *zonelist;
2911
Mel Gorman54a6eb52008-04-28 02:12:16 -07002912 zonelist = &pgdat->node_zonelists[1];
2913 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002914 zonelist->_zonerefs[j].zone = NULL;
2915 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07002916}
2917
2918/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002919 * Build zonelists ordered by zone and nodes within zones.
2920 * This results in conserving DMA zone[s] until all Normal memory is
2921 * exhausted, but results in overflowing to remote node while memory
2922 * may still exist in local DMA zone.
2923 */
2924static int node_order[MAX_NUMNODES];
2925
2926static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2927{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002928 int pos, j, node;
2929 int zone_type; /* needs to be signed */
2930 struct zone *z;
2931 struct zonelist *zonelist;
2932
Mel Gorman54a6eb52008-04-28 02:12:16 -07002933 zonelist = &pgdat->node_zonelists[0];
2934 pos = 0;
2935 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2936 for (j = 0; j < nr_nodes; j++) {
2937 node = node_order[j];
2938 z = &NODE_DATA(node)->node_zones[zone_type];
2939 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002940 zoneref_set_zone(z,
2941 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002942 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002943 }
2944 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002945 }
Mel Gormandd1a2392008-04-28 02:12:17 -07002946 zonelist->_zonerefs[pos].zone = NULL;
2947 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002948}
2949
2950static int default_zonelist_order(void)
2951{
2952 int nid, zone_type;
2953 unsigned long low_kmem_size,total_size;
2954 struct zone *z;
2955 int average_size;
2956 /*
Thomas Weber88393162010-03-16 11:47:56 +01002957 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002958 * If they are really small and used heavily, the system can fall
2959 * into OOM very easily.
David Rientjese325c902010-05-24 14:32:13 -07002960 * This function detect ZONE_DMA/DMA32 size and configures zone order.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002961 */
2962 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2963 low_kmem_size = 0;
2964 total_size = 0;
2965 for_each_online_node(nid) {
2966 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2967 z = &NODE_DATA(nid)->node_zones[zone_type];
2968 if (populated_zone(z)) {
2969 if (zone_type < ZONE_NORMAL)
2970 low_kmem_size += z->present_pages;
2971 total_size += z->present_pages;
David Rientjese325c902010-05-24 14:32:13 -07002972 } else if (zone_type == ZONE_NORMAL) {
2973 /*
2974 * If any node has only lowmem, then node order
2975 * is preferred to allow kernel allocations
2976 * locally; otherwise, they can easily infringe
2977 * on other nodes when there is an abundance of
2978 * lowmem available to allocate from.
2979 */
2980 return ZONELIST_ORDER_NODE;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002981 }
2982 }
2983 }
2984 if (!low_kmem_size || /* there are no DMA area. */
2985 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2986 return ZONELIST_ORDER_NODE;
2987 /*
2988 * look into each node's config.
2989 * If there is a node whose DMA/DMA32 memory is very big area on
2990 * local memory, NODE_ORDER may be suitable.
2991 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002992 average_size = total_size /
2993 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002994 for_each_online_node(nid) {
2995 low_kmem_size = 0;
2996 total_size = 0;
2997 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2998 z = &NODE_DATA(nid)->node_zones[zone_type];
2999 if (populated_zone(z)) {
3000 if (zone_type < ZONE_NORMAL)
3001 low_kmem_size += z->present_pages;
3002 total_size += z->present_pages;
3003 }
3004 }
3005 if (low_kmem_size &&
3006 total_size > average_size && /* ignore small node */
3007 low_kmem_size > total_size * 70/100)
3008 return ZONELIST_ORDER_NODE;
3009 }
3010 return ZONELIST_ORDER_ZONE;
3011}
3012
3013static void set_zonelist_order(void)
3014{
3015 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
3016 current_zonelist_order = default_zonelist_order();
3017 else
3018 current_zonelist_order = user_zonelist_order;
3019}
3020
3021static void build_zonelists(pg_data_t *pgdat)
3022{
3023 int j, node, load;
3024 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003025 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003026 int local_node, prev_node;
3027 struct zonelist *zonelist;
3028 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003029
3030 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07003031 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003032 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07003033 zonelist->_zonerefs[0].zone = NULL;
3034 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003035 }
3036
3037 /* NUMA-aware ordering of nodes */
3038 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003039 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040 prev_node = local_node;
3041 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003042
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003043 memset(node_order, 0, sizeof(node_order));
3044 j = 0;
3045
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08003047 int distance = node_distance(local_node, node);
3048
3049 /*
3050 * If another node is sufficiently far away then it is better
3051 * to reclaim pages in a zone before going off node.
3052 */
3053 if (distance > RECLAIM_DISTANCE)
3054 zone_reclaim_mode = 1;
3055
Linus Torvalds1da177e2005-04-16 15:20:36 -07003056 /*
3057 * We don't want to pressure a particular node.
3058 * So adding penalty to the first node in same
3059 * distance group to make it round-robin.
3060 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08003061 if (distance != node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003062 node_load[node] = load;
3063
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064 prev_node = node;
3065 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003066 if (order == ZONELIST_ORDER_NODE)
3067 build_zonelists_in_node_order(pgdat, node);
3068 else
3069 node_order[j++] = node; /* remember order */
3070 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003072 if (order == ZONELIST_ORDER_ZONE) {
3073 /* calculate node order -- i.e., DMA last! */
3074 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075 }
Christoph Lameter523b9452007-10-16 01:25:37 -07003076
3077 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003078}
3079
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003080/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003081static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003082{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003083 struct zonelist *zonelist;
3084 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07003085 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003086
Mel Gorman54a6eb52008-04-28 02:12:16 -07003087 zonelist = &pgdat->node_zonelists[0];
3088 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3089 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07003090 for (z = zonelist->_zonerefs; z->zone; z++)
3091 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003092}
3093
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003094#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3095/*
3096 * Return node id of node used for "local" allocations.
3097 * I.e., first node id of first zone in arg node's generic zonelist.
3098 * Used for initializing percpu 'numa_mem', which is used primarily
3099 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3100 */
3101int local_memory_node(int node)
3102{
3103 struct zone *zone;
3104
3105 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3106 gfp_zone(GFP_KERNEL),
3107 NULL,
3108 &zone);
3109 return zone->node;
3110}
3111#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003112
Linus Torvalds1da177e2005-04-16 15:20:36 -07003113#else /* CONFIG_NUMA */
3114
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003115static void set_zonelist_order(void)
3116{
3117 current_zonelist_order = ZONELIST_ORDER_ZONE;
3118}
3119
3120static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003121{
Christoph Lameter19655d32006-09-25 23:31:19 -07003122 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003123 enum zone_type j;
3124 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003125
3126 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003127
Mel Gorman54a6eb52008-04-28 02:12:16 -07003128 zonelist = &pgdat->node_zonelists[0];
3129 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003130
Mel Gorman54a6eb52008-04-28 02:12:16 -07003131 /*
3132 * Now we build the zonelist so that it contains the zones
3133 * of all the other nodes.
3134 * We don't want to pressure a particular node, so when
3135 * building the zones for node N, we make sure that the
3136 * zones coming right after the local ones are those from
3137 * node N+1 (modulo N)
3138 */
3139 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3140 if (!node_online(node))
3141 continue;
3142 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3143 MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003144 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07003145 for (node = 0; node < local_node; node++) {
3146 if (!node_online(node))
3147 continue;
3148 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3149 MAX_NR_ZONES - 1);
3150 }
3151
Mel Gormandd1a2392008-04-28 02:12:17 -07003152 zonelist->_zonerefs[j].zone = NULL;
3153 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154}
3155
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003156/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003157static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003158{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003159 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003160}
3161
Linus Torvalds1da177e2005-04-16 15:20:36 -07003162#endif /* CONFIG_NUMA */
3163
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003164/*
3165 * Boot pageset table. One per cpu which is going to be used for all
3166 * zones and all nodes. The parameters will be set in such a way
3167 * that an item put on a list will immediately be handed over to
3168 * the buddy list. This is safe since pageset manipulation is done
3169 * with interrupts disabled.
3170 *
3171 * The boot_pagesets must be kept even after bootup is complete for
3172 * unused processors and/or zones. They do play a role for bootstrapping
3173 * hotplugged processors.
3174 *
3175 * zoneinfo_show() and maybe other functions do
3176 * not check if the processor is online before following the pageset pointer.
3177 * Other parts of the kernel may not check if the zone is available.
3178 */
3179static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3180static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07003181static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003182
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003183/*
3184 * Global mutex to protect against size modification of zonelists
3185 * as well as to serialize pageset setup for the new populated zone.
3186 */
3187DEFINE_MUTEX(zonelists_mutex);
3188
Rusty Russell9b1a4d32008-07-28 12:16:30 -05003189/* return values int ....just for stop_machine() */
Haicheng Li1f522502010-05-24 14:32:51 -07003190static __init_refok int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003191{
Yasunori Goto68113782006-06-23 02:03:11 -07003192 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003193 int cpu;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003194
Bo Liu7f9cfb32009-08-18 14:11:19 -07003195#ifdef CONFIG_NUMA
3196 memset(node_load, 0, sizeof(node_load));
3197#endif
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003198 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07003199 pg_data_t *pgdat = NODE_DATA(nid);
3200
3201 build_zonelists(pgdat);
3202 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003203 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003204
3205 /*
3206 * Initialize the boot_pagesets that are going to be used
3207 * for bootstrapping processors. The real pagesets for
3208 * each zone will be allocated later when the per cpu
3209 * allocator is available.
3210 *
3211 * boot_pagesets are used also for bootstrapping offline
3212 * cpus if the system is already booted because the pagesets
3213 * are needed to initialize allocators on a specific cpu too.
3214 * F.e. the percpu allocator needs the page allocator which
3215 * needs the percpu allocator in order to allocate its pagesets
3216 * (a chicken-egg dilemma).
3217 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003218 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003219 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3220
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003221#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3222 /*
3223 * We now know the "local memory node" for each node--
3224 * i.e., the node of the first zone in the generic zonelist.
3225 * Set up numa_mem percpu variable for on-line cpus. During
3226 * boot, only the boot cpu should be on-line; we'll init the
3227 * secondary cpus' numa_mem as they come on-line. During
3228 * node/memory hotplug, we'll fixup all on-line cpus.
3229 */
3230 if (cpu_online(cpu))
3231 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3232#endif
3233 }
3234
Yasunori Goto68113782006-06-23 02:03:11 -07003235 return 0;
3236}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003237
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003238/*
3239 * Called with zonelists_mutex held always
3240 * unless system_state == SYSTEM_BOOTING.
3241 */
Paul Mundt9f6ae442011-04-14 15:21:57 -07003242void __ref build_all_zonelists(void *data)
Yasunori Goto68113782006-06-23 02:03:11 -07003243{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003244 set_zonelist_order();
3245
Yasunori Goto68113782006-06-23 02:03:11 -07003246 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07003247 __build_all_zonelists(NULL);
Mel Gorman68ad8df2008-07-23 21:26:52 -07003248 mminit_verify_zonelist();
Yasunori Goto68113782006-06-23 02:03:11 -07003249 cpuset_init_current_mems_allowed();
3250 } else {
Simon Arlott183ff222007-10-20 01:27:18 +02003251 /* we have to stop all cpus to guarantee there is no user
Yasunori Goto68113782006-06-23 02:03:11 -07003252 of zonelist */
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003253#ifdef CONFIG_MEMORY_HOTPLUG
3254 if (data)
3255 setup_zone_pageset((struct zone *)data);
3256#endif
3257 stop_machine(__build_all_zonelists, NULL, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07003258 /* cpuset refresh routine should be here */
3259 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07003260 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003261 /*
3262 * Disable grouping by mobility if the number of pages in the
3263 * system is too low to allow the mechanism to work. It would be
3264 * more accurate, but expensive to check per-zone. This check is
3265 * made on memory-hotadd so a system can start with mobility
3266 * disabled and enable it later
3267 */
Mel Gormand9c23402007-10-16 01:26:01 -07003268 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003269 page_group_by_mobility_disabled = 1;
3270 else
3271 page_group_by_mobility_disabled = 0;
3272
3273 printk("Built %i zonelists in %s order, mobility grouping %s. "
3274 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003275 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003276 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003277 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003278 vm_total_pages);
3279#ifdef CONFIG_NUMA
3280 printk("Policy zone: %s\n", zone_names[policy_zone]);
3281#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003282}
3283
3284/*
3285 * Helper functions to size the waitqueue hash table.
3286 * Essentially these want to choose hash table sizes sufficiently
3287 * large so that collisions trying to wait on pages are rare.
3288 * But in fact, the number of active page waitqueues on typical
3289 * systems is ridiculously low, less than 200. So this is even
3290 * conservative, even though it seems large.
3291 *
3292 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3293 * waitqueues, i.e. the size of the waitq table given the number of pages.
3294 */
3295#define PAGES_PER_WAITQUEUE 256
3296
Yasunori Gotocca448f2006-06-23 02:03:10 -07003297#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07003298static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003299{
3300 unsigned long size = 1;
3301
3302 pages /= PAGES_PER_WAITQUEUE;
3303
3304 while (size < pages)
3305 size <<= 1;
3306
3307 /*
3308 * Once we have dozens or even hundreds of threads sleeping
3309 * on IO we've got bigger problems than wait queue collision.
3310 * Limit the size of the wait table to a reasonable size.
3311 */
3312 size = min(size, 4096UL);
3313
3314 return max(size, 4UL);
3315}
Yasunori Gotocca448f2006-06-23 02:03:10 -07003316#else
3317/*
3318 * A zone's size might be changed by hot-add, so it is not possible to determine
3319 * a suitable size for its wait_table. So we use the maximum size now.
3320 *
3321 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3322 *
3323 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3324 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3325 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3326 *
3327 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3328 * or more by the traditional way. (See above). It equals:
3329 *
3330 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3331 * ia64(16K page size) : = ( 8G + 4M)byte.
3332 * powerpc (64K page size) : = (32G +16M)byte.
3333 */
3334static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3335{
3336 return 4096UL;
3337}
3338#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003339
3340/*
3341 * This is an integer logarithm so that shifts can be used later
3342 * to extract the more random high bits from the multiplicative
3343 * hash function before the remainder is taken.
3344 */
3345static inline unsigned long wait_table_bits(unsigned long size)
3346{
3347 return ffz(~size);
3348}
3349
3350#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3351
Mel Gorman56fd56b2007-10-16 01:25:58 -07003352/*
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07003353 * Check if a pageblock contains reserved pages
3354 */
3355static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
3356{
3357 unsigned long pfn;
3358
3359 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
3360 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
3361 return 1;
3362 }
3363 return 0;
3364}
3365
3366/*
Mel Gormand9c23402007-10-16 01:26:01 -07003367 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07003368 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3369 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07003370 * higher will lead to a bigger reserve which will get freed as contiguous
3371 * blocks as reclaim kicks in
3372 */
3373static void setup_zone_migrate_reserve(struct zone *zone)
3374{
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07003375 unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003376 struct page *page;
Mel Gorman78986a62009-09-21 17:03:02 -07003377 unsigned long block_migratetype;
3378 int reserve;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003379
Michal Hockod02156382011-12-08 14:34:27 -08003380 /*
3381 * Get the start pfn, end pfn and the number of blocks to reserve
3382 * We have to be careful to be aligned to pageblock_nr_pages to
3383 * make sure that we always check pfn_valid for the first page in
3384 * the block.
3385 */
Mel Gorman56fd56b2007-10-16 01:25:58 -07003386 start_pfn = zone->zone_start_pfn;
3387 end_pfn = start_pfn + zone->spanned_pages;
Michal Hockod02156382011-12-08 14:34:27 -08003388 start_pfn = roundup(start_pfn, pageblock_nr_pages);
Mel Gorman41858962009-06-16 15:32:12 -07003389 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07003390 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003391
Mel Gorman78986a62009-09-21 17:03:02 -07003392 /*
3393 * Reserve blocks are generally in place to help high-order atomic
3394 * allocations that are short-lived. A min_free_kbytes value that
3395 * would result in more than 2 reserve blocks for atomic allocations
3396 * is assumed to be in place to help anti-fragmentation for the
3397 * future allocation of hugepages at runtime.
3398 */
3399 reserve = min(2, reserve);
3400
Mel Gormand9c23402007-10-16 01:26:01 -07003401 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07003402 if (!pfn_valid(pfn))
3403 continue;
3404 page = pfn_to_page(pfn);
3405
Adam Litke344c7902008-09-02 14:35:38 -07003406 /* Watch out for overlapping nodes */
3407 if (page_to_nid(page) != zone_to_nid(zone))
3408 continue;
3409
Mel Gorman56fd56b2007-10-16 01:25:58 -07003410 /* Blocks with reserved pages will never free, skip them. */
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07003411 block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
3412 if (pageblock_is_reserved(pfn, block_end_pfn))
Mel Gorman56fd56b2007-10-16 01:25:58 -07003413 continue;
3414
3415 block_migratetype = get_pageblock_migratetype(page);
3416
3417 /* If this block is reserved, account for it */
3418 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
3419 reserve--;
3420 continue;
3421 }
3422
3423 /* Suitable for reserving if this block is movable */
3424 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
3425 set_pageblock_migratetype(page, MIGRATE_RESERVE);
3426 move_freepages_block(zone, page, MIGRATE_RESERVE);
3427 reserve--;
3428 continue;
3429 }
3430
3431 /*
3432 * If the reserve is met and this is a previous reserved block,
3433 * take it back
3434 */
3435 if (block_migratetype == MIGRATE_RESERVE) {
3436 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
3437 move_freepages_block(zone, page, MIGRATE_MOVABLE);
3438 }
3439 }
3440}
Mel Gormanac0e5b72007-10-16 01:25:58 -07003441
Linus Torvalds1da177e2005-04-16 15:20:36 -07003442/*
3443 * Initially all pages are reserved - free ones are freed
3444 * up by free_all_bootmem() once the early boot process is
3445 * done. Non-atomic initialization, single-pass.
3446 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003447void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08003448 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003449{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003450 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07003451 unsigned long end_pfn = start_pfn + size;
3452 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003453 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003454
Hugh Dickins22b31ee2009-01-06 14:40:09 -08003455 if (highest_memmap_pfn < end_pfn - 1)
3456 highest_memmap_pfn = end_pfn - 1;
3457
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003458 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08003459 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08003460 /*
3461 * There can be holes in boot-time mem_map[]s
3462 * handed to this function. They do not
3463 * exist on hotplugged memory.
3464 */
3465 if (context == MEMMAP_EARLY) {
3466 if (!early_pfn_valid(pfn))
3467 continue;
3468 if (!early_pfn_in_nid(pfn, nid))
3469 continue;
3470 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003471 page = pfn_to_page(pfn);
3472 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07003473 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08003474 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003475 reset_page_mapcount(page);
3476 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003477 /*
3478 * Mark the block movable so that blocks are reserved for
3479 * movable at startup. This will force kernel allocations
3480 * to reserve their blocks rather than leaking throughout
3481 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07003482 * kernel allocations are made. Later some blocks near
3483 * the start are marked MIGRATE_RESERVE by
3484 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003485 *
3486 * bitmap is created for zone's valid pfn range. but memmap
3487 * can be created for invalid pages (for alignment)
3488 * check here not to call set_pageblock_migratetype() against
3489 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003490 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003491 if ((z->zone_start_pfn <= pfn)
3492 && (pfn < z->zone_start_pfn + z->spanned_pages)
3493 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07003494 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003495
Linus Torvalds1da177e2005-04-16 15:20:36 -07003496 INIT_LIST_HEAD(&page->lru);
3497#ifdef WANT_PAGE_VIRTUAL
3498 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3499 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07003500 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003501#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003502 }
3503}
3504
Andi Kleen1e548de2008-02-04 22:29:26 -08003505static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003506{
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003507 int order, t;
3508 for_each_migratetype_order(order, t) {
3509 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003510 zone->free_area[order].nr_free = 0;
3511 }
3512}
3513
3514#ifndef __HAVE_ARCH_MEMMAP_INIT
3515#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08003516 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003517#endif
3518
Sam Ravnborg1d6f4e62008-01-17 15:21:12 -08003519static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003520{
David Howells3a6be872009-05-06 16:03:03 -07003521#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003522 int batch;
3523
3524 /*
3525 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07003526 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003527 *
3528 * OK, so we don't know how big the cache is. So guess.
3529 */
3530 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07003531 if (batch * PAGE_SIZE > 512 * 1024)
3532 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003533 batch /= 4; /* We effectively *= 4 below */
3534 if (batch < 1)
3535 batch = 1;
3536
3537 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11003538 * Clamp the batch to a 2^n - 1 value. Having a power
3539 * of 2 value was found to be more likely to have
3540 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003541 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11003542 * For example if 2 tasks are alternately allocating
3543 * batches of pages, one task can end up with a lot
3544 * of pages of one half of the possible page colors
3545 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003546 */
David Howells91552032009-05-06 16:03:02 -07003547 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07003548
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003549 return batch;
David Howells3a6be872009-05-06 16:03:03 -07003550
3551#else
3552 /* The deferral and batching of frees should be suppressed under NOMMU
3553 * conditions.
3554 *
3555 * The problem is that NOMMU needs to be able to allocate large chunks
3556 * of contiguous memory as there's no hardware page translation to
3557 * assemble apparent contiguous memory from discontiguous pages.
3558 *
3559 * Queueing large contiguous runs of pages for batching, however,
3560 * causes the pages to actually be freed in smaller chunks. As there
3561 * can be a significant delay between the individual batches being
3562 * recycled, this leads to the once large chunks of space being
3563 * fragmented and becoming unavailable for high-order allocations.
3564 */
3565 return 0;
3566#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003567}
3568
Adrian Bunkb69a7282008-07-23 21:28:12 -07003569static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
Christoph Lameter2caaad42005-06-21 17:15:00 -07003570{
3571 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003572 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003573
Magnus Damm1c6fe942005-10-26 01:58:59 -07003574 memset(p, 0, sizeof(*p));
3575
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003576 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003577 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003578 pcp->high = 6 * batch;
3579 pcp->batch = max(1UL, 1 * batch);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003580 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
3581 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07003582}
3583
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003584/*
3585 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3586 * to the value high for the pageset p.
3587 */
3588
3589static void setup_pagelist_highmark(struct per_cpu_pageset *p,
3590 unsigned long high)
3591{
3592 struct per_cpu_pages *pcp;
3593
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003594 pcp = &p->pcp;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003595 pcp->high = high;
3596 pcp->batch = max(1UL, high/4);
3597 if ((high/4) > (PAGE_SHIFT * 8))
3598 pcp->batch = PAGE_SHIFT * 8;
3599}
3600
Nikanth Karthikesan58c2ee42011-03-15 10:59:02 +05303601static void setup_zone_pageset(struct zone *zone)
Wu Fengguang319774e2010-05-24 14:32:49 -07003602{
3603 int cpu;
3604
3605 zone->pageset = alloc_percpu(struct per_cpu_pageset);
3606
3607 for_each_possible_cpu(cpu) {
3608 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
3609
3610 setup_pageset(pcp, zone_batchsize(zone));
3611
3612 if (percpu_pagelist_fraction)
3613 setup_pagelist_highmark(pcp,
3614 (zone->present_pages /
3615 percpu_pagelist_fraction));
3616 }
3617}
3618
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003619/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003620 * Allocate per cpu pagesets and initialize them.
3621 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07003622 */
Al Viro78d99552005-12-15 09:18:25 +00003623void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003624{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003625 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003626
Wu Fengguang319774e2010-05-24 14:32:49 -07003627 for_each_populated_zone(zone)
3628 setup_zone_pageset(zone);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003629}
3630
Sam Ravnborg577a32f2007-05-17 23:29:25 +02003631static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07003632int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07003633{
3634 int i;
3635 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003636 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07003637
3638 /*
3639 * The per-page waitqueue mechanism uses hashed waitqueues
3640 * per zone.
3641 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07003642 zone->wait_table_hash_nr_entries =
3643 wait_table_hash_nr_entries(zone_size_pages);
3644 zone->wait_table_bits =
3645 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003646 alloc_size = zone->wait_table_hash_nr_entries
3647 * sizeof(wait_queue_head_t);
3648
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07003649 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07003650 zone->wait_table = (wait_queue_head_t *)
Yinghai Lu8f389a992011-05-11 15:13:32 -07003651 alloc_bootmem_node_nopanic(pgdat, alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003652 } else {
3653 /*
3654 * This case means that a zone whose size was 0 gets new memory
3655 * via memory hot-add.
3656 * But it may be the case that a new node was hot-added. In
3657 * this case vmalloc() will not be able to use this new node's
3658 * memory - this wait_table must be initialized to use this new
3659 * node itself as well.
3660 * To use this new node's memory, further consideration will be
3661 * necessary.
3662 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07003663 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003664 }
3665 if (!zone->wait_table)
3666 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07003667
Yasunori Goto02b694d2006-06-23 02:03:08 -07003668 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07003669 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003670
3671 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003672}
3673
Shaohua Li112067f2009-09-21 17:01:16 -07003674static int __zone_pcp_update(void *data)
3675{
3676 struct zone *zone = data;
3677 int cpu;
3678 unsigned long batch = zone_batchsize(zone), flags;
3679
Thomas Gleixner2d30a1f2010-03-10 15:20:40 -08003680 for_each_possible_cpu(cpu) {
Shaohua Li112067f2009-09-21 17:01:16 -07003681 struct per_cpu_pageset *pset;
3682 struct per_cpu_pages *pcp;
3683
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003684 pset = per_cpu_ptr(zone->pageset, cpu);
Shaohua Li112067f2009-09-21 17:01:16 -07003685 pcp = &pset->pcp;
3686
3687 local_irq_save(flags);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003688 free_pcppages_bulk(zone, pcp->count, pcp);
Shaohua Li112067f2009-09-21 17:01:16 -07003689 setup_pageset(pset, batch);
3690 local_irq_restore(flags);
3691 }
3692 return 0;
3693}
3694
3695void zone_pcp_update(struct zone *zone)
3696{
3697 stop_machine(__zone_pcp_update, zone, NULL);
3698}
3699
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003700static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07003701{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003702 /*
3703 * per cpu subsystem is not up at this point. The following code
3704 * relies on the ability of the linker to provide the
3705 * offset of a (static) per cpu variable into the per cpu area.
3706 */
3707 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07003708
Anton Blanchardf5335c02006-03-25 03:06:49 -08003709 if (zone->present_pages)
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003710 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
3711 zone->name, zone->present_pages,
3712 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07003713}
3714
Yasunori Goto718127c2006-06-23 02:03:10 -07003715__meminit int init_currently_empty_zone(struct zone *zone,
3716 unsigned long zone_start_pfn,
Dave Hansena2f3aa022007-01-10 23:15:30 -08003717 unsigned long size,
3718 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07003719{
3720 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003721 int ret;
3722 ret = zone_wait_table_init(zone, size);
3723 if (ret)
3724 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07003725 pgdat->nr_zones = zone_idx(zone) + 1;
3726
Dave Hansened8ece22005-10-29 18:16:50 -07003727 zone->zone_start_pfn = zone_start_pfn;
3728
Mel Gorman708614e2008-07-23 21:26:51 -07003729 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3730 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3731 pgdat->node_id,
3732 (unsigned long)zone_idx(zone),
3733 zone_start_pfn, (zone_start_pfn + size));
3734
Andi Kleen1e548de2008-02-04 22:29:26 -08003735 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07003736
3737 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003738}
3739
Mel Gormanc7132162006-09-27 01:49:43 -07003740#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3741/*
3742 * Basic iterator support. Return the first range of PFNs for a node
3743 * Note: nid == MAX_NUMNODES returns first region regardless of node
3744 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003745static int __meminit first_active_region_index_in_nid(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003746{
3747 int i;
3748
3749 for (i = 0; i < nr_nodemap_entries; i++)
3750 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3751 return i;
3752
3753 return -1;
3754}
3755
3756/*
3757 * Basic iterator support. Return the next active range of PFNs for a node
Simon Arlott183ff222007-10-20 01:27:18 +02003758 * Note: nid == MAX_NUMNODES returns next region regardless of node
Mel Gormanc7132162006-09-27 01:49:43 -07003759 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003760static int __meminit next_active_region_index_in_nid(int index, int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003761{
3762 for (index = index + 1; index < nr_nodemap_entries; index++)
3763 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3764 return index;
3765
3766 return -1;
3767}
3768
3769#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3770/*
3771 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3772 * Architectures may implement their own version but if add_active_range()
3773 * was used and there are no special requirements, this is a convenient
3774 * alternative
3775 */
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003776int __meminit __early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07003777{
3778 int i;
3779
3780 for (i = 0; i < nr_nodemap_entries; i++) {
3781 unsigned long start_pfn = early_node_map[i].start_pfn;
3782 unsigned long end_pfn = early_node_map[i].end_pfn;
3783
3784 if (start_pfn <= pfn && pfn < end_pfn)
3785 return early_node_map[i].nid;
3786 }
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003787 /* This is a memory hole */
3788 return -1;
Mel Gormanc7132162006-09-27 01:49:43 -07003789}
3790#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3791
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003792int __meminit early_pfn_to_nid(unsigned long pfn)
3793{
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003794 int nid;
3795
3796 nid = __early_pfn_to_nid(pfn);
3797 if (nid >= 0)
3798 return nid;
3799 /* just returns 0 */
3800 return 0;
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003801}
3802
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003803#ifdef CONFIG_NODES_SPAN_OTHER_NODES
3804bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3805{
3806 int nid;
3807
3808 nid = __early_pfn_to_nid(pfn);
3809 if (nid >= 0 && nid != node)
3810 return false;
3811 return true;
3812}
3813#endif
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003814
Mel Gormanc7132162006-09-27 01:49:43 -07003815/* Basic iterator support to walk early_node_map[] */
3816#define for_each_active_range_index_in_nid(i, nid) \
3817 for (i = first_active_region_index_in_nid(nid); i != -1; \
3818 i = next_active_region_index_in_nid(i, nid))
3819
3820/**
3821 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003822 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3823 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07003824 *
3825 * If an architecture guarantees that all ranges registered with
3826 * add_active_ranges() contain no holes and may be freed, this
3827 * this function may be used instead of calling free_bootmem() manually.
3828 */
3829void __init free_bootmem_with_active_regions(int nid,
3830 unsigned long max_low_pfn)
3831{
3832 int i;
3833
3834 for_each_active_range_index_in_nid(i, nid) {
3835 unsigned long size_pages = 0;
3836 unsigned long end_pfn = early_node_map[i].end_pfn;
3837
3838 if (early_node_map[i].start_pfn >= max_low_pfn)
3839 continue;
3840
3841 if (end_pfn > max_low_pfn)
3842 end_pfn = max_low_pfn;
3843
3844 size_pages = end_pfn - early_node_map[i].start_pfn;
3845 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3846 PFN_PHYS(early_node_map[i].start_pfn),
3847 size_pages << PAGE_SHIFT);
3848 }
3849}
3850
Yinghai Luedbe7d22010-08-25 13:39:16 -07003851#ifdef CONFIG_HAVE_MEMBLOCK
Yinghai Lucc289892011-02-26 13:05:43 +01003852/*
3853 * Basic iterator support. Return the last range of PFNs for a node
3854 * Note: nid == MAX_NUMNODES returns last region regardless of node
3855 */
3856static int __meminit last_active_region_index_in_nid(int nid)
3857{
3858 int i;
3859
3860 for (i = nr_nodemap_entries - 1; i >= 0; i--)
3861 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3862 return i;
3863
3864 return -1;
3865}
3866
3867/*
3868 * Basic iterator support. Return the previous active range of PFNs for a node
3869 * Note: nid == MAX_NUMNODES returns next region regardless of node
3870 */
3871static int __meminit previous_active_region_index_in_nid(int index, int nid)
3872{
3873 for (index = index - 1; index >= 0; index--)
3874 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3875 return index;
3876
3877 return -1;
3878}
3879
3880#define for_each_active_range_index_in_nid_reverse(i, nid) \
3881 for (i = last_active_region_index_in_nid(nid); i != -1; \
3882 i = previous_active_region_index_in_nid(i, nid))
3883
Yinghai Luedbe7d22010-08-25 13:39:16 -07003884u64 __init find_memory_core_early(int nid, u64 size, u64 align,
3885 u64 goal, u64 limit)
3886{
3887 int i;
3888
3889 /* Need to go over early_node_map to find out good range for node */
Yinghai Lu1a4a6782010-12-17 16:59:07 -08003890 for_each_active_range_index_in_nid_reverse(i, nid) {
Yinghai Luedbe7d22010-08-25 13:39:16 -07003891 u64 addr;
3892 u64 ei_start, ei_last;
3893 u64 final_start, final_end;
3894
3895 ei_last = early_node_map[i].end_pfn;
3896 ei_last <<= PAGE_SHIFT;
3897 ei_start = early_node_map[i].start_pfn;
3898 ei_start <<= PAGE_SHIFT;
3899
3900 final_start = max(ei_start, goal);
3901 final_end = min(ei_last, limit);
3902
3903 if (final_start >= final_end)
3904 continue;
3905
3906 addr = memblock_find_in_range(final_start, final_end, size, align);
3907
3908 if (addr == MEMBLOCK_ERROR)
3909 continue;
3910
3911 return addr;
3912 }
3913
3914 return MEMBLOCK_ERROR;
3915}
3916#endif
3917
Yinghai Lu08677212010-02-10 01:20:20 -08003918int __init add_from_early_node_map(struct range *range, int az,
3919 int nr_range, int nid)
3920{
3921 int i;
3922 u64 start, end;
3923
3924 /* need to go over early_node_map to find out good range for node */
3925 for_each_active_range_index_in_nid(i, nid) {
3926 start = early_node_map[i].start_pfn;
3927 end = early_node_map[i].end_pfn;
3928 nr_range = add_range(range, az, nr_range, start, end);
3929 }
3930 return nr_range;
3931}
3932
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003933void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3934{
3935 int i;
Yinghai Lud52d53b2008-06-16 20:10:55 -07003936 int ret;
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003937
Yinghai Lud52d53b2008-06-16 20:10:55 -07003938 for_each_active_range_index_in_nid(i, nid) {
3939 ret = work_fn(early_node_map[i].start_pfn,
3940 early_node_map[i].end_pfn, data);
3941 if (ret)
3942 break;
3943 }
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003944}
Mel Gormanc7132162006-09-27 01:49:43 -07003945/**
3946 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003947 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07003948 *
3949 * If an architecture guarantees that all ranges registered with
3950 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003951 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07003952 */
3953void __init sparse_memory_present_with_active_regions(int nid)
3954{
3955 int i;
3956
3957 for_each_active_range_index_in_nid(i, nid)
3958 memory_present(early_node_map[i].nid,
3959 early_node_map[i].start_pfn,
3960 early_node_map[i].end_pfn);
3961}
3962
3963/**
3964 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003965 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3966 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3967 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07003968 *
3969 * It returns the start and end page frame of a node based on information
3970 * provided by an arch calling add_active_range(). If called for a node
3971 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003972 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07003973 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003974void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003975 unsigned long *start_pfn, unsigned long *end_pfn)
3976{
3977 int i;
3978 *start_pfn = -1UL;
3979 *end_pfn = 0;
3980
3981 for_each_active_range_index_in_nid(i, nid) {
3982 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3983 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3984 }
3985
Christoph Lameter633c0662007-10-16 01:25:37 -07003986 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07003987 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07003988}
3989
3990/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07003991 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3992 * assumption is made that zones within a node are ordered in monotonic
3993 * increasing memory addresses so that the "highest" populated zone is used
3994 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003995static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07003996{
3997 int zone_index;
3998 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3999 if (zone_index == ZONE_MOVABLE)
4000 continue;
4001
4002 if (arch_zone_highest_possible_pfn[zone_index] >
4003 arch_zone_lowest_possible_pfn[zone_index])
4004 break;
4005 }
4006
4007 VM_BUG_ON(zone_index == -1);
4008 movable_zone = zone_index;
4009}
4010
4011/*
4012 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004013 * because it is sized independent of architecture. Unlike the other zones,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004014 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4015 * in each node depending on the size of each node and how evenly kernelcore
4016 * is distributed. This helper function adjusts the zone ranges
4017 * provided by the architecture for a given node by using the end of the
4018 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4019 * zones within a node are in order of monotonic increases memory addresses
4020 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004021static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004022 unsigned long zone_type,
4023 unsigned long node_start_pfn,
4024 unsigned long node_end_pfn,
4025 unsigned long *zone_start_pfn,
4026 unsigned long *zone_end_pfn)
4027{
4028 /* Only adjust if ZONE_MOVABLE is on this node */
4029 if (zone_movable_pfn[nid]) {
4030 /* Size ZONE_MOVABLE */
4031 if (zone_type == ZONE_MOVABLE) {
4032 *zone_start_pfn = zone_movable_pfn[nid];
4033 *zone_end_pfn = min(node_end_pfn,
4034 arch_zone_highest_possible_pfn[movable_zone]);
4035
4036 /* Adjust for ZONE_MOVABLE starting within this range */
4037 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
4038 *zone_end_pfn > zone_movable_pfn[nid]) {
4039 *zone_end_pfn = zone_movable_pfn[nid];
4040
4041 /* Check if this whole range is within ZONE_MOVABLE */
4042 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
4043 *zone_start_pfn = *zone_end_pfn;
4044 }
4045}
4046
4047/*
Mel Gormanc7132162006-09-27 01:49:43 -07004048 * Return the number of pages a zone spans in a node, including holes
4049 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4050 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004051static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004052 unsigned long zone_type,
4053 unsigned long *ignored)
4054{
4055 unsigned long node_start_pfn, node_end_pfn;
4056 unsigned long zone_start_pfn, zone_end_pfn;
4057
4058 /* Get the start and end of the node and zone */
4059 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
4060 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4061 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07004062 adjust_zone_range_for_zone_movable(nid, zone_type,
4063 node_start_pfn, node_end_pfn,
4064 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004065
4066 /* Check that this node has pages within the zone's required range */
4067 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4068 return 0;
4069
4070 /* Move the zone boundaries inside the node if necessary */
4071 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4072 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4073
4074 /* Return the spanned pages */
4075 return zone_end_pfn - zone_start_pfn;
4076}
4077
4078/*
4079 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004080 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07004081 */
Yinghai Lu32996252009-12-15 17:59:02 -08004082unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004083 unsigned long range_start_pfn,
4084 unsigned long range_end_pfn)
4085{
4086 int i = 0;
4087 unsigned long prev_end_pfn = 0, hole_pages = 0;
4088 unsigned long start_pfn;
4089
4090 /* Find the end_pfn of the first active range of pfns in the node */
4091 i = first_active_region_index_in_nid(nid);
4092 if (i == -1)
4093 return 0;
4094
Mel Gormanb5445f92007-07-26 10:41:18 -07004095 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
4096
Mel Gorman9c7cd682006-09-27 01:49:58 -07004097 /* Account for ranges before physical memory on this node */
4098 if (early_node_map[i].start_pfn > range_start_pfn)
Mel Gormanb5445f92007-07-26 10:41:18 -07004099 hole_pages = prev_end_pfn - range_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004100
4101 /* Find all holes for the zone within the node */
4102 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
4103
4104 /* No need to continue if prev_end_pfn is outside the zone */
4105 if (prev_end_pfn >= range_end_pfn)
4106 break;
4107
4108 /* Make sure the end of the zone is not within the hole */
4109 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
4110 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
4111
4112 /* Update the hole size cound and move on */
4113 if (start_pfn > range_start_pfn) {
4114 BUG_ON(prev_end_pfn > start_pfn);
4115 hole_pages += start_pfn - prev_end_pfn;
4116 }
4117 prev_end_pfn = early_node_map[i].end_pfn;
4118 }
4119
Mel Gorman9c7cd682006-09-27 01:49:58 -07004120 /* Account for ranges past physical memory on this node */
4121 if (range_end_pfn > prev_end_pfn)
Mel Gorman0c6cb972006-10-28 10:38:59 -07004122 hole_pages += range_end_pfn -
Mel Gorman9c7cd682006-09-27 01:49:58 -07004123 max(range_start_pfn, prev_end_pfn);
4124
Mel Gormanc7132162006-09-27 01:49:43 -07004125 return hole_pages;
4126}
4127
4128/**
4129 * absent_pages_in_range - Return number of page frames in holes within a range
4130 * @start_pfn: The start PFN to start searching for holes
4131 * @end_pfn: The end PFN to stop searching for holes
4132 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004133 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07004134 */
4135unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4136 unsigned long end_pfn)
4137{
4138 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4139}
4140
4141/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004142static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004143 unsigned long zone_type,
4144 unsigned long *ignored)
4145{
Mel Gorman9c7cd682006-09-27 01:49:58 -07004146 unsigned long node_start_pfn, node_end_pfn;
4147 unsigned long zone_start_pfn, zone_end_pfn;
4148
4149 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
4150 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
4151 node_start_pfn);
4152 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
4153 node_end_pfn);
4154
Mel Gorman2a1e2742007-07-17 04:03:12 -07004155 adjust_zone_range_for_zone_movable(nid, zone_type,
4156 node_start_pfn, node_end_pfn,
4157 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004158 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004159}
Mel Gorman0e0b8642006-09-27 01:49:56 -07004160
Mel Gormanc7132162006-09-27 01:49:43 -07004161#else
Paul Mundt6ea6e682007-07-15 23:38:20 -07004162static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004163 unsigned long zone_type,
4164 unsigned long *zones_size)
4165{
4166 return zones_size[zone_type];
4167}
4168
Paul Mundt6ea6e682007-07-15 23:38:20 -07004169static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004170 unsigned long zone_type,
4171 unsigned long *zholes_size)
4172{
4173 if (!zholes_size)
4174 return 0;
4175
4176 return zholes_size[zone_type];
4177}
Mel Gorman0e0b8642006-09-27 01:49:56 -07004178
Mel Gormanc7132162006-09-27 01:49:43 -07004179#endif
4180
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004181static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Mel Gormanc7132162006-09-27 01:49:43 -07004182 unsigned long *zones_size, unsigned long *zholes_size)
4183{
4184 unsigned long realtotalpages, totalpages = 0;
4185 enum zone_type i;
4186
4187 for (i = 0; i < MAX_NR_ZONES; i++)
4188 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
4189 zones_size);
4190 pgdat->node_spanned_pages = totalpages;
4191
4192 realtotalpages = totalpages;
4193 for (i = 0; i < MAX_NR_ZONES; i++)
4194 realtotalpages -=
4195 zone_absent_pages_in_node(pgdat->node_id, i,
4196 zholes_size);
4197 pgdat->node_present_pages = realtotalpages;
4198 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4199 realtotalpages);
4200}
4201
Mel Gorman835c1342007-10-16 01:25:47 -07004202#ifndef CONFIG_SPARSEMEM
4203/*
4204 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07004205 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4206 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07004207 * round what is now in bits to nearest long in bits, then return it in
4208 * bytes.
4209 */
4210static unsigned long __init usemap_size(unsigned long zonesize)
4211{
4212 unsigned long usemapsize;
4213
Mel Gormand9c23402007-10-16 01:26:01 -07004214 usemapsize = roundup(zonesize, pageblock_nr_pages);
4215 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07004216 usemapsize *= NR_PAGEBLOCK_BITS;
4217 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4218
4219 return usemapsize / 8;
4220}
4221
4222static void __init setup_usemap(struct pglist_data *pgdat,
4223 struct zone *zone, unsigned long zonesize)
4224{
4225 unsigned long usemapsize = usemap_size(zonesize);
4226 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08004227 if (usemapsize)
Yinghai Lu8f389a992011-05-11 15:13:32 -07004228 zone->pageblock_flags = alloc_bootmem_node_nopanic(pgdat,
4229 usemapsize);
Mel Gorman835c1342007-10-16 01:25:47 -07004230}
4231#else
Jesper Juhlfa9f90b2010-11-28 21:39:34 +01004232static inline void setup_usemap(struct pglist_data *pgdat,
Mel Gorman835c1342007-10-16 01:25:47 -07004233 struct zone *zone, unsigned long zonesize) {}
4234#endif /* CONFIG_SPARSEMEM */
4235
Mel Gormand9c23402007-10-16 01:26:01 -07004236#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08004237
4238/* Return a sensible default order for the pageblock size. */
4239static inline int pageblock_default_order(void)
4240{
4241 if (HPAGE_SHIFT > PAGE_SHIFT)
4242 return HUGETLB_PAGE_ORDER;
4243
4244 return MAX_ORDER-1;
4245}
4246
Mel Gormand9c23402007-10-16 01:26:01 -07004247/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4248static inline void __init set_pageblock_order(unsigned int order)
4249{
4250 /* Check that pageblock_nr_pages has not already been setup */
4251 if (pageblock_order)
4252 return;
4253
4254 /*
4255 * Assume the largest contiguous order of interest is a huge page.
4256 * This value may be variable depending on boot parameters on IA64
4257 */
4258 pageblock_order = order;
4259}
4260#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4261
Mel Gormanba72cb82007-11-28 16:21:13 -08004262/*
4263 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4264 * and pageblock_default_order() are unused as pageblock_order is set
4265 * at compile-time. See include/linux/pageblock-flags.h for the values of
4266 * pageblock_order based on the kernel config
4267 */
4268static inline int pageblock_default_order(unsigned int order)
4269{
4270 return MAX_ORDER-1;
4271}
Mel Gormand9c23402007-10-16 01:26:01 -07004272#define set_pageblock_order(x) do {} while (0)
4273
4274#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4275
Linus Torvalds1da177e2005-04-16 15:20:36 -07004276/*
4277 * Set up the zone data structures:
4278 * - mark all pages reserved
4279 * - mark all memory queues empty
4280 * - clear the memory bitmaps
4281 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08004282static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004283 unsigned long *zones_size, unsigned long *zholes_size)
4284{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07004285 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07004286 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004287 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07004288 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004289
Dave Hansen208d54e2005-10-29 18:16:52 -07004290 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004291 pgdat->nr_zones = 0;
4292 init_waitqueue_head(&pgdat->kswapd_wait);
4293 pgdat->kswapd_max_order = 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004294 pgdat_page_cgroup_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004295
4296 for (j = 0; j < MAX_NR_ZONES; j++) {
4297 struct zone *zone = pgdat->node_zones + j;
Mel Gorman0e0b8642006-09-27 01:49:56 -07004298 unsigned long size, realsize, memmap_pages;
Christoph Lameterb69408e2008-10-18 20:26:14 -07004299 enum lru_list l;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004300
Mel Gormanc7132162006-09-27 01:49:43 -07004301 size = zone_spanned_pages_in_node(nid, j, zones_size);
4302 realsize = size - zone_absent_pages_in_node(nid, j,
4303 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004304
Mel Gorman0e0b8642006-09-27 01:49:56 -07004305 /*
4306 * Adjust realsize so that it accounts for how much memory
4307 * is used by this zone for memmap. This affects the watermark
4308 * and per-cpu initialisations
4309 */
Johannes Weinerf7232152008-05-23 13:04:21 -07004310 memmap_pages =
4311 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
Mel Gorman0e0b8642006-09-27 01:49:56 -07004312 if (realsize >= memmap_pages) {
4313 realsize -= memmap_pages;
Yinghai Lu5594c8c2009-01-06 14:39:14 -08004314 if (memmap_pages)
4315 printk(KERN_DEBUG
4316 " %s zone: %lu pages used for memmap\n",
4317 zone_names[j], memmap_pages);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004318 } else
4319 printk(KERN_WARNING
4320 " %s zone: %lu pages exceeds realsize %lu\n",
4321 zone_names[j], memmap_pages, realsize);
4322
Christoph Lameter62672762007-02-10 01:43:07 -08004323 /* Account for reserved pages */
4324 if (j == 0 && realsize > dma_reserve) {
Mel Gorman0e0b8642006-09-27 01:49:56 -07004325 realsize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07004326 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08004327 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004328 }
4329
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004330 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004331 nr_kernel_pages += realsize;
4332 nr_all_pages += realsize;
4333
4334 zone->spanned_pages = size;
4335 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07004336#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07004337 zone->node = nid;
Christoph Lameter8417bba2006-09-25 23:31:51 -07004338 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07004339 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07004340 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07004341#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004342 zone->name = zone_names[j];
4343 spin_lock_init(&zone->lock);
4344 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07004345 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004346 zone->zone_pgdat = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004347
Dave Hansened8ece22005-10-29 18:16:50 -07004348 zone_pcp_init(zone);
KAMEZAWA Hiroyuki246e87a2011-05-26 16:25:34 -07004349 for_each_lru(l)
Christoph Lameterb69408e2008-10-18 20:26:14 -07004350 INIT_LIST_HEAD(&zone->lru[l].list);
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08004351 zone->reclaim_stat.recent_rotated[0] = 0;
4352 zone->reclaim_stat.recent_rotated[1] = 0;
4353 zone->reclaim_stat.recent_scanned[0] = 0;
4354 zone->reclaim_stat.recent_scanned[1] = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07004355 zap_zone_vm_stats(zone);
David Rientjese815af92007-10-16 23:25:54 -07004356 zone->flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004357 if (!size)
4358 continue;
4359
Mel Gormanba72cb82007-11-28 16:21:13 -08004360 set_pageblock_order(pageblock_default_order());
Mel Gorman835c1342007-10-16 01:25:47 -07004361 setup_usemap(pgdat, zone, size);
Dave Hansena2f3aa022007-01-10 23:15:30 -08004362 ret = init_currently_empty_zone(zone, zone_start_pfn,
4363 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07004364 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07004365 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004366 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004367 }
4368}
4369
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004370static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004371{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004372 /* Skip empty nodes */
4373 if (!pgdat->node_spanned_pages)
4374 return;
4375
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004376#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07004377 /* ia64 gets its own node_mem_map, before this, without bootmem */
4378 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07004379 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004380 struct page *map;
4381
Bob Piccoe984bb42006-05-20 15:00:31 -07004382 /*
4383 * The zone's endpoints aren't required to be MAX_ORDER
4384 * aligned but the node_mem_map endpoints must be in order
4385 * for the buddy allocator to function correctly.
4386 */
4387 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
4388 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
4389 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4390 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07004391 map = alloc_remap(pgdat->node_id, size);
4392 if (!map)
Yinghai Lu8f389a992011-05-11 15:13:32 -07004393 map = alloc_bootmem_node_nopanic(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07004394 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004395 }
Roman Zippel12d810c2007-05-31 00:40:54 -07004396#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07004397 /*
4398 * With no DISCONTIG, the global mem_map is just set as node 0's
4399 */
Mel Gormanc7132162006-09-27 01:49:43 -07004400 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004401 mem_map = NODE_DATA(0)->node_mem_map;
Mel Gormanc7132162006-09-27 01:49:43 -07004402#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
4403 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08004404 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Mel Gormanc7132162006-09-27 01:49:43 -07004405#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4406 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004407#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004408#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004409}
4410
Johannes Weiner9109fb72008-07-23 21:27:20 -07004411void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4412 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004413{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004414 pg_data_t *pgdat = NODE_DATA(nid);
4415
Linus Torvalds1da177e2005-04-16 15:20:36 -07004416 pgdat->node_id = nid;
4417 pgdat->node_start_pfn = node_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004418 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004419
4420 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07004421#ifdef CONFIG_FLAT_NODE_MEM_MAP
4422 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4423 nid, (unsigned long)pgdat,
4424 (unsigned long)pgdat->node_mem_map);
4425#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004426
4427 free_area_init_core(pgdat, zones_size, zholes_size);
4428}
4429
Mel Gormanc7132162006-09-27 01:49:43 -07004430#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07004431
4432#if MAX_NUMNODES > 1
4433/*
4434 * Figure out the number of possible node ids.
4435 */
4436static void __init setup_nr_node_ids(void)
4437{
4438 unsigned int node;
4439 unsigned int highest = 0;
4440
4441 for_each_node_mask(node, node_possible_map)
4442 highest = node;
4443 nr_node_ids = highest + 1;
4444}
4445#else
4446static inline void setup_nr_node_ids(void)
4447{
4448}
4449#endif
4450
Mel Gormanc7132162006-09-27 01:49:43 -07004451/**
4452 * add_active_range - Register a range of PFNs backed by physical memory
4453 * @nid: The node ID the range resides on
4454 * @start_pfn: The start PFN of the available physical memory
4455 * @end_pfn: The end PFN of the available physical memory
4456 *
4457 * These ranges are stored in an early_node_map[] and later used by
4458 * free_area_init_nodes() to calculate zone sizes and holes. If the
4459 * range spans a memory hole, it is up to the architecture to ensure
4460 * the memory is not freed by the bootmem allocator. If possible
4461 * the range being registered will be merged with existing ranges.
4462 */
4463void __init add_active_range(unsigned int nid, unsigned long start_pfn,
4464 unsigned long end_pfn)
4465{
4466 int i;
4467
Mel Gorman6b74ab92008-07-23 21:26:49 -07004468 mminit_dprintk(MMINIT_TRACE, "memory_register",
4469 "Entering add_active_range(%d, %#lx, %#lx) "
4470 "%d entries of %d used\n",
4471 nid, start_pfn, end_pfn,
4472 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
Mel Gormanc7132162006-09-27 01:49:43 -07004473
Mel Gorman2dbb51c2008-07-23 21:26:52 -07004474 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
4475
Mel Gormanc7132162006-09-27 01:49:43 -07004476 /* Merge with existing active regions if possible */
4477 for (i = 0; i < nr_nodemap_entries; i++) {
4478 if (early_node_map[i].nid != nid)
4479 continue;
4480
4481 /* Skip if an existing region covers this new one */
4482 if (start_pfn >= early_node_map[i].start_pfn &&
4483 end_pfn <= early_node_map[i].end_pfn)
4484 return;
4485
4486 /* Merge forward if suitable */
4487 if (start_pfn <= early_node_map[i].end_pfn &&
4488 end_pfn > early_node_map[i].end_pfn) {
4489 early_node_map[i].end_pfn = end_pfn;
4490 return;
4491 }
4492
4493 /* Merge backward if suitable */
Kazuhisa Ichikawad2dbe082010-01-15 17:01:20 -08004494 if (start_pfn < early_node_map[i].start_pfn &&
Mel Gormanc7132162006-09-27 01:49:43 -07004495 end_pfn >= early_node_map[i].start_pfn) {
4496 early_node_map[i].start_pfn = start_pfn;
4497 return;
4498 }
4499 }
4500
4501 /* Check that early_node_map is large enough */
4502 if (i >= MAX_ACTIVE_REGIONS) {
4503 printk(KERN_CRIT "More than %d memory regions, truncating\n",
4504 MAX_ACTIVE_REGIONS);
4505 return;
4506 }
4507
4508 early_node_map[i].nid = nid;
4509 early_node_map[i].start_pfn = start_pfn;
4510 early_node_map[i].end_pfn = end_pfn;
4511 nr_nodemap_entries = i + 1;
4512}
4513
4514/**
Yinghai Lucc1050b2008-06-13 19:08:52 -07004515 * remove_active_range - Shrink an existing registered range of PFNs
Mel Gormanc7132162006-09-27 01:49:43 -07004516 * @nid: The node id the range is on that should be shrunk
Yinghai Lucc1050b2008-06-13 19:08:52 -07004517 * @start_pfn: The new PFN of the range
4518 * @end_pfn: The new PFN of the range
Mel Gormanc7132162006-09-27 01:49:43 -07004519 *
4520 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004521 * The map is kept near the end physical page range that has already been
4522 * registered. This function allows an arch to shrink an existing registered
4523 * range.
Mel Gormanc7132162006-09-27 01:49:43 -07004524 */
Yinghai Lucc1050b2008-06-13 19:08:52 -07004525void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
4526 unsigned long end_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004527{
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004528 int i, j;
4529 int removed = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07004530
Yinghai Lucc1050b2008-06-13 19:08:52 -07004531 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
4532 nid, start_pfn, end_pfn);
4533
Mel Gormanc7132162006-09-27 01:49:43 -07004534 /* Find the old active region end and shrink */
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004535 for_each_active_range_index_in_nid(i, nid) {
Yinghai Lucc1050b2008-06-13 19:08:52 -07004536 if (early_node_map[i].start_pfn >= start_pfn &&
4537 early_node_map[i].end_pfn <= end_pfn) {
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004538 /* clear it */
Yinghai Lucc1050b2008-06-13 19:08:52 -07004539 early_node_map[i].start_pfn = 0;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004540 early_node_map[i].end_pfn = 0;
4541 removed = 1;
4542 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004543 }
Yinghai Lucc1050b2008-06-13 19:08:52 -07004544 if (early_node_map[i].start_pfn < start_pfn &&
4545 early_node_map[i].end_pfn > start_pfn) {
4546 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
4547 early_node_map[i].end_pfn = start_pfn;
4548 if (temp_end_pfn > end_pfn)
4549 add_active_range(nid, end_pfn, temp_end_pfn);
4550 continue;
4551 }
4552 if (early_node_map[i].start_pfn >= start_pfn &&
4553 early_node_map[i].end_pfn > end_pfn &&
4554 early_node_map[i].start_pfn < end_pfn) {
4555 early_node_map[i].start_pfn = end_pfn;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004556 continue;
4557 }
4558 }
4559
4560 if (!removed)
4561 return;
4562
4563 /* remove the blank ones */
4564 for (i = nr_nodemap_entries - 1; i > 0; i--) {
4565 if (early_node_map[i].nid != nid)
4566 continue;
4567 if (early_node_map[i].end_pfn)
4568 continue;
4569 /* we found it, get rid of it */
4570 for (j = i; j < nr_nodemap_entries - 1; j++)
4571 memcpy(&early_node_map[j], &early_node_map[j+1],
4572 sizeof(early_node_map[j]));
4573 j = nr_nodemap_entries - 1;
4574 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
4575 nr_nodemap_entries--;
4576 }
Mel Gormanc7132162006-09-27 01:49:43 -07004577}
4578
4579/**
4580 * remove_all_active_ranges - Remove all currently registered regions
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004581 *
Mel Gormanc7132162006-09-27 01:49:43 -07004582 * During discovery, it may be found that a table like SRAT is invalid
4583 * and an alternative discovery method must be used. This function removes
4584 * all currently registered regions.
4585 */
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004586void __init remove_all_active_ranges(void)
Mel Gormanc7132162006-09-27 01:49:43 -07004587{
4588 memset(early_node_map, 0, sizeof(early_node_map));
4589 nr_nodemap_entries = 0;
4590}
4591
4592/* Compare two active node_active_regions */
4593static int __init cmp_node_active_region(const void *a, const void *b)
4594{
4595 struct node_active_region *arange = (struct node_active_region *)a;
4596 struct node_active_region *brange = (struct node_active_region *)b;
4597
4598 /* Done this way to avoid overflows */
4599 if (arange->start_pfn > brange->start_pfn)
4600 return 1;
4601 if (arange->start_pfn < brange->start_pfn)
4602 return -1;
4603
4604 return 0;
4605}
4606
4607/* sort the node_map by start_pfn */
Yinghai Lu32996252009-12-15 17:59:02 -08004608void __init sort_node_map(void)
Mel Gormanc7132162006-09-27 01:49:43 -07004609{
4610 sort(early_node_map, (size_t)nr_nodemap_entries,
4611 sizeof(struct node_active_region),
4612 cmp_node_active_region, NULL);
4613}
4614
Tejun Heo1e019792011-07-12 09:45:34 +02004615/**
4616 * node_map_pfn_alignment - determine the maximum internode alignment
4617 *
4618 * This function should be called after node map is populated and sorted.
4619 * It calculates the maximum power of two alignment which can distinguish
4620 * all the nodes.
4621 *
4622 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
4623 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
4624 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
4625 * shifted, 1GiB is enough and this function will indicate so.
4626 *
4627 * This is used to test whether pfn -> nid mapping of the chosen memory
4628 * model has fine enough granularity to avoid incorrect mapping for the
4629 * populated node map.
4630 *
4631 * Returns the determined alignment in pfn's. 0 if there is no alignment
4632 * requirement (single node).
4633 */
4634unsigned long __init node_map_pfn_alignment(void)
4635{
4636 unsigned long accl_mask = 0, last_end = 0;
4637 int last_nid = -1;
4638 int i;
4639
4640 for_each_active_range_index_in_nid(i, MAX_NUMNODES) {
4641 int nid = early_node_map[i].nid;
4642 unsigned long start = early_node_map[i].start_pfn;
4643 unsigned long end = early_node_map[i].end_pfn;
4644 unsigned long mask;
4645
4646 if (!start || last_nid < 0 || last_nid == nid) {
4647 last_nid = nid;
4648 last_end = end;
4649 continue;
4650 }
4651
4652 /*
4653 * Start with a mask granular enough to pin-point to the
4654 * start pfn and tick off bits one-by-one until it becomes
4655 * too coarse to separate the current node from the last.
4656 */
4657 mask = ~((1 << __ffs(start)) - 1);
4658 while (mask && last_end <= (start & (mask << 1)))
4659 mask <<= 1;
4660
4661 /* accumulate all internode masks */
4662 accl_mask |= mask;
4663 }
4664
4665 /* convert mask to number of pages */
4666 return ~accl_mask + 1;
4667}
4668
Mel Gormana6af2bc2007-02-10 01:42:57 -08004669/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004670static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07004671{
4672 int i;
Mel Gormana6af2bc2007-02-10 01:42:57 -08004673 unsigned long min_pfn = ULONG_MAX;
Mel Gorman1abbfb42006-11-23 12:01:41 +00004674
Mel Gormanc7132162006-09-27 01:49:43 -07004675 /* Assuming a sorted map, the first range found has the starting pfn */
4676 for_each_active_range_index_in_nid(i, nid)
Mel Gormana6af2bc2007-02-10 01:42:57 -08004677 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004678
Mel Gormana6af2bc2007-02-10 01:42:57 -08004679 if (min_pfn == ULONG_MAX) {
4680 printk(KERN_WARNING
Paul Jackson2bc0d2612008-06-22 07:22:17 -07004681 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08004682 return 0;
4683 }
4684
4685 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004686}
4687
4688/**
4689 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4690 *
4691 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004692 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07004693 */
4694unsigned long __init find_min_pfn_with_active_regions(void)
4695{
4696 return find_min_pfn_for_node(MAX_NUMNODES);
4697}
4698
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004699/*
4700 * early_calculate_totalpages()
4701 * Sum pages in active regions for movable zone.
4702 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4703 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07004704static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07004705{
4706 int i;
4707 unsigned long totalpages = 0;
4708
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004709 for (i = 0; i < nr_nodemap_entries; i++) {
4710 unsigned long pages = early_node_map[i].end_pfn -
Mel Gorman7e63efe2007-07-17 04:03:15 -07004711 early_node_map[i].start_pfn;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004712 totalpages += pages;
4713 if (pages)
4714 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
4715 }
4716 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07004717}
4718
Mel Gorman2a1e2742007-07-17 04:03:12 -07004719/*
4720 * Find the PFN the Movable zone begins in each node. Kernel memory
4721 * is spread evenly between nodes as long as the nodes have enough
4722 * memory. When they don't, some nodes will have more kernelcore than
4723 * others
4724 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004725static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004726{
4727 int i, nid;
4728 unsigned long usable_startpfn;
4729 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07004730 /* save the state before borrow the nodemask */
4731 nodemask_t saved_node_state = node_states[N_HIGH_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004732 unsigned long totalpages = early_calculate_totalpages();
4733 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004734
Mel Gorman7e63efe2007-07-17 04:03:15 -07004735 /*
4736 * If movablecore was specified, calculate what size of
4737 * kernelcore that corresponds so that memory usable for
4738 * any allocation type is evenly spread. If both kernelcore
4739 * and movablecore are specified, then the value of kernelcore
4740 * will be used for required_kernelcore if it's greater than
4741 * what movablecore would have allowed.
4742 */
4743 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07004744 unsigned long corepages;
4745
4746 /*
4747 * Round-up so that ZONE_MOVABLE is at least as large as what
4748 * was requested by the user
4749 */
4750 required_movablecore =
4751 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
4752 corepages = totalpages - required_movablecore;
4753
4754 required_kernelcore = max(required_kernelcore, corepages);
4755 }
4756
Mel Gorman2a1e2742007-07-17 04:03:12 -07004757 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4758 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07004759 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004760
4761 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4762 find_usable_zone_for_movable();
4763 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4764
4765restart:
4766 /* Spread kernelcore memory as evenly as possible throughout nodes */
4767 kernelcore_node = required_kernelcore / usable_nodes;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004768 for_each_node_state(nid, N_HIGH_MEMORY) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004769 /*
4770 * Recalculate kernelcore_node if the division per node
4771 * now exceeds what is necessary to satisfy the requested
4772 * amount of memory for the kernel
4773 */
4774 if (required_kernelcore < kernelcore_node)
4775 kernelcore_node = required_kernelcore / usable_nodes;
4776
4777 /*
4778 * As the map is walked, we track how much memory is usable
4779 * by the kernel using kernelcore_remaining. When it is
4780 * 0, the rest of the node is usable by ZONE_MOVABLE
4781 */
4782 kernelcore_remaining = kernelcore_node;
4783
4784 /* Go through each range of PFNs within this node */
4785 for_each_active_range_index_in_nid(i, nid) {
4786 unsigned long start_pfn, end_pfn;
4787 unsigned long size_pages;
4788
4789 start_pfn = max(early_node_map[i].start_pfn,
4790 zone_movable_pfn[nid]);
4791 end_pfn = early_node_map[i].end_pfn;
4792 if (start_pfn >= end_pfn)
4793 continue;
4794
4795 /* Account for what is only usable for kernelcore */
4796 if (start_pfn < usable_startpfn) {
4797 unsigned long kernel_pages;
4798 kernel_pages = min(end_pfn, usable_startpfn)
4799 - start_pfn;
4800
4801 kernelcore_remaining -= min(kernel_pages,
4802 kernelcore_remaining);
4803 required_kernelcore -= min(kernel_pages,
4804 required_kernelcore);
4805
4806 /* Continue if range is now fully accounted */
4807 if (end_pfn <= usable_startpfn) {
4808
4809 /*
4810 * Push zone_movable_pfn to the end so
4811 * that if we have to rebalance
4812 * kernelcore across nodes, we will
4813 * not double account here
4814 */
4815 zone_movable_pfn[nid] = end_pfn;
4816 continue;
4817 }
4818 start_pfn = usable_startpfn;
4819 }
4820
4821 /*
4822 * The usable PFN range for ZONE_MOVABLE is from
4823 * start_pfn->end_pfn. Calculate size_pages as the
4824 * number of pages used as kernelcore
4825 */
4826 size_pages = end_pfn - start_pfn;
4827 if (size_pages > kernelcore_remaining)
4828 size_pages = kernelcore_remaining;
4829 zone_movable_pfn[nid] = start_pfn + size_pages;
4830
4831 /*
4832 * Some kernelcore has been met, update counts and
4833 * break if the kernelcore for this node has been
4834 * satisified
4835 */
4836 required_kernelcore -= min(required_kernelcore,
4837 size_pages);
4838 kernelcore_remaining -= size_pages;
4839 if (!kernelcore_remaining)
4840 break;
4841 }
4842 }
4843
4844 /*
4845 * If there is still required_kernelcore, we do another pass with one
4846 * less node in the count. This will push zone_movable_pfn[nid] further
4847 * along on the nodes that still have memory until kernelcore is
4848 * satisified
4849 */
4850 usable_nodes--;
4851 if (usable_nodes && required_kernelcore > usable_nodes)
4852 goto restart;
4853
4854 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4855 for (nid = 0; nid < MAX_NUMNODES; nid++)
4856 zone_movable_pfn[nid] =
4857 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07004858
4859out:
4860 /* restore the node_state */
4861 node_states[N_HIGH_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004862}
4863
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004864/* Any regular memory on that node ? */
4865static void check_for_regular_memory(pg_data_t *pgdat)
4866{
4867#ifdef CONFIG_HIGHMEM
4868 enum zone_type zone_type;
4869
4870 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4871 struct zone *zone = &pgdat->node_zones[zone_type];
4872 if (zone->present_pages)
4873 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4874 }
4875#endif
4876}
4877
Mel Gormanc7132162006-09-27 01:49:43 -07004878/**
4879 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004880 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07004881 *
4882 * This will call free_area_init_node() for each active node in the system.
4883 * Using the page ranges provided by add_active_range(), the size of each
4884 * zone in each node and their holes is calculated. If the maximum PFN
4885 * between two adjacent zones match, it is assumed that the zone is empty.
4886 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4887 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4888 * starts where the previous one ended. For example, ZONE_DMA32 starts
4889 * at arch_max_dma_pfn.
4890 */
4891void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4892{
4893 unsigned long nid;
Andrew Mortondb991002008-10-15 22:01:07 -07004894 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004895
Mel Gormana6af2bc2007-02-10 01:42:57 -08004896 /* Sort early_node_map as initialisation assumes it is sorted */
4897 sort_node_map();
4898
Mel Gormanc7132162006-09-27 01:49:43 -07004899 /* Record where the zone boundaries are */
4900 memset(arch_zone_lowest_possible_pfn, 0,
4901 sizeof(arch_zone_lowest_possible_pfn));
4902 memset(arch_zone_highest_possible_pfn, 0,
4903 sizeof(arch_zone_highest_possible_pfn));
4904 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4905 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4906 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004907 if (i == ZONE_MOVABLE)
4908 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004909 arch_zone_lowest_possible_pfn[i] =
4910 arch_zone_highest_possible_pfn[i-1];
4911 arch_zone_highest_possible_pfn[i] =
4912 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4913 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07004914 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4915 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4916
4917 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4918 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4919 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004920
Mel Gormanc7132162006-09-27 01:49:43 -07004921 /* Print out the zone ranges */
4922 printk("Zone PFN ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07004923 for (i = 0; i < MAX_NR_ZONES; i++) {
4924 if (i == ZONE_MOVABLE)
4925 continue;
David Rientjes72f0ba02010-03-05 13:42:14 -08004926 printk(" %-8s ", zone_names[i]);
4927 if (arch_zone_lowest_possible_pfn[i] ==
4928 arch_zone_highest_possible_pfn[i])
4929 printk("empty\n");
4930 else
4931 printk("%0#10lx -> %0#10lx\n",
Mel Gormanc7132162006-09-27 01:49:43 -07004932 arch_zone_lowest_possible_pfn[i],
4933 arch_zone_highest_possible_pfn[i]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004934 }
4935
4936 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4937 printk("Movable zone start PFN for each node\n");
4938 for (i = 0; i < MAX_NUMNODES; i++) {
4939 if (zone_movable_pfn[i])
4940 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4941 }
Mel Gormanc7132162006-09-27 01:49:43 -07004942
4943 /* Print out the early_node_map[] */
4944 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4945 for (i = 0; i < nr_nodemap_entries; i++)
Paul Jackson5dab8ec2008-06-25 05:44:40 -07004946 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004947 early_node_map[i].start_pfn,
4948 early_node_map[i].end_pfn);
4949
4950 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07004951 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08004952 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07004953 for_each_online_node(nid) {
4954 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07004955 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07004956 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004957
4958 /* Any memory on that node */
4959 if (pgdat->node_present_pages)
4960 node_set_state(nid, N_HIGH_MEMORY);
4961 check_for_regular_memory(pgdat);
Mel Gormanc7132162006-09-27 01:49:43 -07004962 }
4963}
Mel Gorman2a1e2742007-07-17 04:03:12 -07004964
Mel Gorman7e63efe2007-07-17 04:03:15 -07004965static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004966{
4967 unsigned long long coremem;
4968 if (!p)
4969 return -EINVAL;
4970
4971 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004972 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004973
Mel Gorman7e63efe2007-07-17 04:03:15 -07004974 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07004975 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4976
4977 return 0;
4978}
Mel Gormaned7ed362007-07-17 04:03:14 -07004979
Mel Gorman7e63efe2007-07-17 04:03:15 -07004980/*
4981 * kernelcore=size sets the amount of memory for use for allocations that
4982 * cannot be reclaimed or migrated.
4983 */
4984static int __init cmdline_parse_kernelcore(char *p)
4985{
4986 return cmdline_parse_core(p, &required_kernelcore);
4987}
4988
4989/*
4990 * movablecore=size sets the amount of memory for use for allocations that
4991 * can be reclaimed or migrated.
4992 */
4993static int __init cmdline_parse_movablecore(char *p)
4994{
4995 return cmdline_parse_core(p, &required_movablecore);
4996}
4997
Mel Gormaned7ed362007-07-17 04:03:14 -07004998early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004999early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07005000
Mel Gormanc7132162006-09-27 01:49:43 -07005001#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
5002
Mel Gorman0e0b8642006-09-27 01:49:56 -07005003/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005004 * set_dma_reserve - set the specified number of pages reserved in the first zone
5005 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07005006 *
5007 * The per-cpu batchsize and zone watermarks are determined by present_pages.
5008 * In the DMA zone, a significant percentage may be consumed by kernel image
5009 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005010 * function may optionally be used to account for unfreeable pages in the
5011 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5012 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07005013 */
5014void __init set_dma_reserve(unsigned long new_dma_reserve)
5015{
5016 dma_reserve = new_dma_reserve;
5017}
5018
Linus Torvalds1da177e2005-04-16 15:20:36 -07005019void __init free_area_init(unsigned long *zones_size)
5020{
Johannes Weiner9109fb72008-07-23 21:27:20 -07005021 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005022 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
5023}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005024
Linus Torvalds1da177e2005-04-16 15:20:36 -07005025static int page_alloc_cpu_notify(struct notifier_block *self,
5026 unsigned long action, void *hcpu)
5027{
5028 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005029
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07005030 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005031 drain_pages(cpu);
5032
5033 /*
5034 * Spill the event counters of the dead processor
5035 * into the current processors event counters.
5036 * This artificially elevates the count of the current
5037 * processor.
5038 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07005039 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005040
5041 /*
5042 * Zero the differential counters of the dead processor
5043 * so that the vm statistics are consistent.
5044 *
5045 * This is only okay since the processor is dead and cannot
5046 * race with what we are doing.
5047 */
Christoph Lameter2244b952006-06-30 01:55:33 -07005048 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005049 }
5050 return NOTIFY_OK;
5051}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005052
5053void __init page_alloc_init(void)
5054{
5055 hotcpu_notifier(page_alloc_cpu_notify, 0);
5056}
5057
5058/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005059 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5060 * or min_free_kbytes changes.
5061 */
5062static void calculate_totalreserve_pages(void)
5063{
5064 struct pglist_data *pgdat;
5065 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005066 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005067
5068 for_each_online_pgdat(pgdat) {
5069 for (i = 0; i < MAX_NR_ZONES; i++) {
5070 struct zone *zone = pgdat->node_zones + i;
5071 unsigned long max = 0;
5072
5073 /* Find valid and maximum lowmem_reserve in the zone */
5074 for (j = i; j < MAX_NR_ZONES; j++) {
5075 if (zone->lowmem_reserve[j] > max)
5076 max = zone->lowmem_reserve[j];
5077 }
5078
Mel Gorman41858962009-06-16 15:32:12 -07005079 /* we treat the high watermark as reserved pages. */
5080 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005081
5082 if (max > zone->present_pages)
5083 max = zone->present_pages;
5084 reserve_pages += max;
5085 }
5086 }
5087 totalreserve_pages = reserve_pages;
5088}
5089
5090/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07005091 * setup_per_zone_lowmem_reserve - called whenever
5092 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
5093 * has a correct pages reserved value, so an adequate number of
5094 * pages are left in the zone after a successful __alloc_pages().
5095 */
5096static void setup_per_zone_lowmem_reserve(void)
5097{
5098 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005099 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005100
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08005101 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005102 for (j = 0; j < MAX_NR_ZONES; j++) {
5103 struct zone *zone = pgdat->node_zones + j;
5104 unsigned long present_pages = zone->present_pages;
5105
5106 zone->lowmem_reserve[j] = 0;
5107
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005108 idx = j;
5109 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005110 struct zone *lower_zone;
5111
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005112 idx--;
5113
Linus Torvalds1da177e2005-04-16 15:20:36 -07005114 if (sysctl_lowmem_reserve_ratio[idx] < 1)
5115 sysctl_lowmem_reserve_ratio[idx] = 1;
5116
5117 lower_zone = pgdat->node_zones + idx;
5118 lower_zone->lowmem_reserve[j] = present_pages /
5119 sysctl_lowmem_reserve_ratio[idx];
5120 present_pages += lower_zone->present_pages;
5121 }
5122 }
5123 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005124
5125 /* update totalreserve_pages */
5126 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005127}
5128
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005129/**
Minchan Kimbc75d332009-06-16 15:32:48 -07005130 * setup_per_zone_wmarks - called when min_free_kbytes changes
Minchan Kimbce73942009-06-16 15:32:50 -07005131 * or when memory is hot-{added|removed}
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005132 *
Minchan Kimbc75d332009-06-16 15:32:48 -07005133 * Ensures that the watermark[min,low,high] values for each zone are set
5134 * correctly with respect to min_free_kbytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005135 */
Minchan Kimbc75d332009-06-16 15:32:48 -07005136void setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005137{
5138 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
5139 unsigned long lowmem_pages = 0;
5140 struct zone *zone;
5141 unsigned long flags;
5142
5143 /* Calculate total number of !ZONE_HIGHMEM pages */
5144 for_each_zone(zone) {
5145 if (!is_highmem(zone))
5146 lowmem_pages += zone->present_pages;
5147 }
5148
5149 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07005150 u64 tmp;
5151
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005152 spin_lock_irqsave(&zone->lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07005153 tmp = (u64)pages_min * zone->present_pages;
5154 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005155 if (is_highmem(zone)) {
5156 /*
Nick Piggin669ed172005-11-13 16:06:45 -08005157 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5158 * need highmem pages, so cap pages_min to a small
5159 * value here.
5160 *
Mel Gorman41858962009-06-16 15:32:12 -07005161 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Nick Piggin669ed172005-11-13 16:06:45 -08005162 * deltas controls asynch page reclaim, and so should
5163 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005164 */
5165 int min_pages;
5166
5167 min_pages = zone->present_pages / 1024;
5168 if (min_pages < SWAP_CLUSTER_MAX)
5169 min_pages = SWAP_CLUSTER_MAX;
5170 if (min_pages > 128)
5171 min_pages = 128;
Mel Gorman41858962009-06-16 15:32:12 -07005172 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005173 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08005174 /*
5175 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07005176 * proportionate to the zone's size.
5177 */
Mel Gorman41858962009-06-16 15:32:12 -07005178 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005179 }
5180
Mel Gorman41858962009-06-16 15:32:12 -07005181 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
5182 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Mel Gorman56fd56b2007-10-16 01:25:58 -07005183 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005184 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005185 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005186
5187 /* update totalreserve_pages */
5188 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005189}
5190
Randy Dunlap55a44622009-09-21 17:01:20 -07005191/*
Rik van Riel556adec2008-10-18 20:26:34 -07005192 * The inactive anon list should be small enough that the VM never has to
5193 * do too much work, but large enough that each inactive page has a chance
5194 * to be referenced again before it is swapped out.
5195 *
5196 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5197 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5198 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5199 * the anonymous pages are kept on the inactive list.
5200 *
5201 * total target max
5202 * memory ratio inactive anon
5203 * -------------------------------------
5204 * 10MB 1 5MB
5205 * 100MB 1 50MB
5206 * 1GB 3 250MB
5207 * 10GB 10 0.9GB
5208 * 100GB 31 3GB
5209 * 1TB 101 10GB
5210 * 10TB 320 32GB
5211 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005212static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
Minchan Kim96cb4df2009-06-16 15:32:49 -07005213{
5214 unsigned int gb, ratio;
5215
5216 /* Zone size in gigabytes */
5217 gb = zone->present_pages >> (30 - PAGE_SHIFT);
5218 if (gb)
5219 ratio = int_sqrt(10 * gb);
5220 else
5221 ratio = 1;
5222
5223 zone->inactive_ratio = ratio;
5224}
5225
KOSAKI Motohiro839a4fc2011-05-24 17:11:31 -07005226static void __meminit setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07005227{
5228 struct zone *zone;
5229
Minchan Kim96cb4df2009-06-16 15:32:49 -07005230 for_each_zone(zone)
5231 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07005232}
5233
Linus Torvalds1da177e2005-04-16 15:20:36 -07005234/*
5235 * Initialise min_free_kbytes.
5236 *
5237 * For small machines we want it small (128k min). For large machines
5238 * we want it large (64MB max). But it is not linear, because network
5239 * bandwidth does not increase linearly with machine size. We use
5240 *
5241 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5242 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5243 *
5244 * which yields
5245 *
5246 * 16MB: 512k
5247 * 32MB: 724k
5248 * 64MB: 1024k
5249 * 128MB: 1448k
5250 * 256MB: 2048k
5251 * 512MB: 2896k
5252 * 1024MB: 4096k
5253 * 2048MB: 5792k
5254 * 4096MB: 8192k
5255 * 8192MB: 11584k
5256 * 16384MB: 16384k
5257 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005258int __meminit init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005259{
5260 unsigned long lowmem_kbytes;
5261
5262 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
5263
5264 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
5265 if (min_free_kbytes < 128)
5266 min_free_kbytes = 128;
5267 if (min_free_kbytes > 65536)
5268 min_free_kbytes = 65536;
Minchan Kimbc75d332009-06-16 15:32:48 -07005269 setup_per_zone_wmarks();
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -07005270 refresh_zone_stat_thresholds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005271 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07005272 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005273 return 0;
5274}
Minchan Kimbc75d332009-06-16 15:32:48 -07005275module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005276
5277/*
5278 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5279 * that we can call two helper functions whenever min_free_kbytes
5280 * changes.
5281 */
5282int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005283 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005284{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005285 proc_dointvec(table, write, buffer, length, ppos);
Mel Gorman3b1d92c2007-05-06 14:49:30 -07005286 if (write)
Minchan Kimbc75d332009-06-16 15:32:48 -07005287 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005288 return 0;
5289}
5290
Christoph Lameter96146342006-07-03 00:24:13 -07005291#ifdef CONFIG_NUMA
5292int sysctl_min_unmapped_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 Lameter96146342006-07-03 00:24:13 -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 Lameter96146342006-07-03 00:24:13 -07005299 if (rc)
5300 return rc;
5301
5302 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07005303 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07005304 sysctl_min_unmapped_ratio) / 100;
5305 return 0;
5306}
Christoph Lameter0ff38492006-09-25 23:31:52 -07005307
5308int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005309 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07005310{
5311 struct zone *zone;
5312 int rc;
5313
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005314 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07005315 if (rc)
5316 return rc;
5317
5318 for_each_zone(zone)
5319 zone->min_slab_pages = (zone->present_pages *
5320 sysctl_min_slab_ratio) / 100;
5321 return 0;
5322}
Christoph Lameter96146342006-07-03 00:24:13 -07005323#endif
5324
Linus Torvalds1da177e2005-04-16 15:20:36 -07005325/*
5326 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5327 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5328 * whenever sysctl_lowmem_reserve_ratio changes.
5329 *
5330 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07005331 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07005332 * if in function of the boot time zone sizes.
5333 */
5334int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005335 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005336{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005337 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005338 setup_per_zone_lowmem_reserve();
5339 return 0;
5340}
5341
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005342/*
5343 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5344 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5345 * can have before it gets flushed back to buddy allocator.
5346 */
5347
5348int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005349 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005350{
5351 struct zone *zone;
5352 unsigned int cpu;
5353 int ret;
5354
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005355 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005356 if (!write || (ret == -EINVAL))
5357 return ret;
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07005358 for_each_populated_zone(zone) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005359 for_each_possible_cpu(cpu) {
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005360 unsigned long high;
5361 high = zone->present_pages / percpu_pagelist_fraction;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005362 setup_pagelist_highmark(
5363 per_cpu_ptr(zone->pageset, cpu), high);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005364 }
5365 }
5366 return 0;
5367}
5368
David S. Millerf034b5d2006-08-24 03:08:07 -07005369int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005370
5371#ifdef CONFIG_NUMA
5372static int __init set_hashdist(char *str)
5373{
5374 if (!str)
5375 return 0;
5376 hashdist = simple_strtoul(str, &str, 0);
5377 return 1;
5378}
5379__setup("hashdist=", set_hashdist);
5380#endif
5381
5382/*
5383 * allocate a large system hash table from bootmem
5384 * - it is assumed that the hash table must contain an exact power-of-2
5385 * quantity of entries
5386 * - limit is the number of hash buckets, not the total allocation size
5387 */
5388void *__init alloc_large_system_hash(const char *tablename,
5389 unsigned long bucketsize,
5390 unsigned long numentries,
5391 int scale,
5392 int flags,
5393 unsigned int *_hash_shift,
5394 unsigned int *_hash_mask,
5395 unsigned long limit)
5396{
5397 unsigned long long max = limit;
5398 unsigned long log2qty, size;
5399 void *table = NULL;
5400
5401 /* allow the kernel cmdline to have a say */
5402 if (!numentries) {
5403 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08005404 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005405 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
5406 numentries >>= 20 - PAGE_SHIFT;
5407 numentries <<= 20 - PAGE_SHIFT;
5408
5409 /* limit to 1 bucket per 2^scale bytes of low memory */
5410 if (scale > PAGE_SHIFT)
5411 numentries >>= (scale - PAGE_SHIFT);
5412 else
5413 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08005414
5415 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07005416 if (unlikely(flags & HASH_SMALL)) {
5417 /* Makes no sense without HASH_EARLY */
5418 WARN_ON(!(flags & HASH_EARLY));
5419 if (!(numentries >> *_hash_shift)) {
5420 numentries = 1UL << *_hash_shift;
5421 BUG_ON(!numentries);
5422 }
5423 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08005424 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005425 }
John Hawkes6e692ed2006-03-25 03:08:02 -08005426 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005427
5428 /* limit allocation size to 1/16 total memory by default */
5429 if (max == 0) {
5430 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
5431 do_div(max, bucketsize);
5432 }
5433
5434 if (numentries > max)
5435 numentries = max;
5436
David Howellsf0d1b0b2006-12-08 02:37:49 -08005437 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005438
5439 do {
5440 size = bucketsize << log2qty;
5441 if (flags & HASH_EARLY)
Jan Beulich74768ed2008-08-12 15:08:39 -07005442 table = alloc_bootmem_nopanic(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005443 else if (hashdist)
5444 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
5445 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07005446 /*
5447 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07005448 * some pages at the end of hash table which
5449 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07005450 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005451 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07005452 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005453 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
5454 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005455 }
5456 } while (!table && size > PAGE_SIZE && --log2qty);
5457
5458 if (!table)
5459 panic("Failed to allocate %s hash table\n", tablename);
5460
Robin Holtf241e6602010-10-07 12:59:26 -07005461 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07005462 tablename,
Robin Holtf241e6602010-10-07 12:59:26 -07005463 (1UL << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08005464 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005465 size);
5466
5467 if (_hash_shift)
5468 *_hash_shift = log2qty;
5469 if (_hash_mask)
5470 *_hash_mask = (1 << log2qty) - 1;
5471
5472 return table;
5473}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08005474
Mel Gorman835c1342007-10-16 01:25:47 -07005475/* Return a pointer to the bitmap storing bits affecting a block of pages */
5476static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
5477 unsigned long pfn)
5478{
5479#ifdef CONFIG_SPARSEMEM
5480 return __pfn_to_section(pfn)->pageblock_flags;
5481#else
5482 return zone->pageblock_flags;
5483#endif /* CONFIG_SPARSEMEM */
5484}
Andrew Morton6220ec72006-10-19 23:29:05 -07005485
Mel Gorman835c1342007-10-16 01:25:47 -07005486static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
5487{
5488#ifdef CONFIG_SPARSEMEM
5489 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07005490 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005491#else
5492 pfn = pfn - zone->zone_start_pfn;
Mel Gormand9c23402007-10-16 01:26:01 -07005493 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005494#endif /* CONFIG_SPARSEMEM */
5495}
5496
5497/**
Mel Gormand9c23402007-10-16 01:26:01 -07005498 * 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 -07005499 * @page: The page within the block of interest
5500 * @start_bitidx: The first bit of interest to retrieve
5501 * @end_bitidx: The last bit of interest
5502 * returns pageblock_bits flags
5503 */
5504unsigned long get_pageblock_flags_group(struct page *page,
5505 int start_bitidx, int end_bitidx)
5506{
5507 struct zone *zone;
5508 unsigned long *bitmap;
5509 unsigned long pfn, bitidx;
5510 unsigned long flags = 0;
5511 unsigned long value = 1;
5512
5513 zone = page_zone(page);
5514 pfn = page_to_pfn(page);
5515 bitmap = get_pageblock_bitmap(zone, pfn);
5516 bitidx = pfn_to_bitidx(zone, pfn);
5517
5518 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5519 if (test_bit(bitidx + start_bitidx, bitmap))
5520 flags |= value;
5521
5522 return flags;
5523}
5524
5525/**
Mel Gormand9c23402007-10-16 01:26:01 -07005526 * 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 -07005527 * @page: The page within the block of interest
5528 * @start_bitidx: The first bit of interest
5529 * @end_bitidx: The last bit of interest
5530 * @flags: The flags to set
5531 */
5532void set_pageblock_flags_group(struct page *page, unsigned long flags,
5533 int start_bitidx, int end_bitidx)
5534{
5535 struct zone *zone;
5536 unsigned long *bitmap;
5537 unsigned long pfn, bitidx;
5538 unsigned long value = 1;
5539
5540 zone = page_zone(page);
5541 pfn = page_to_pfn(page);
5542 bitmap = get_pageblock_bitmap(zone, pfn);
5543 bitidx = pfn_to_bitidx(zone, pfn);
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07005544 VM_BUG_ON(pfn < zone->zone_start_pfn);
5545 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
Mel Gorman835c1342007-10-16 01:25:47 -07005546
5547 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5548 if (flags & value)
5549 __set_bit(bitidx + start_bitidx, bitmap);
5550 else
5551 __clear_bit(bitidx + start_bitidx, bitmap);
5552}
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07005553
5554/*
5555 * This is designed as sub function...plz see page_isolation.c also.
5556 * set/clear page block's type to be ISOLATE.
5557 * page allocater never alloc memory from ISOLATE block.
5558 */
5559
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005560static int
5561__count_immobile_pages(struct zone *zone, struct page *page, int count)
5562{
5563 unsigned long pfn, iter, found;
5564 /*
5565 * For avoiding noise data, lru_add_drain_all() should be called
5566 * If ZONE_MOVABLE, the zone never contains immobile pages
5567 */
5568 if (zone_idx(zone) == ZONE_MOVABLE)
5569 return true;
5570
5571 if (get_pageblock_migratetype(page) == MIGRATE_MOVABLE)
5572 return true;
5573
5574 pfn = page_to_pfn(page);
5575 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
5576 unsigned long check = pfn + iter;
5577
Namhyung Kim29723fc2011-02-25 14:44:25 -08005578 if (!pfn_valid_within(check))
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005579 continue;
Namhyung Kim29723fc2011-02-25 14:44:25 -08005580
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005581 page = pfn_to_page(check);
5582 if (!page_count(page)) {
5583 if (PageBuddy(page))
5584 iter += (1 << page_order(page)) - 1;
5585 continue;
5586 }
5587 if (!PageLRU(page))
5588 found++;
5589 /*
5590 * If there are RECLAIMABLE pages, we need to check it.
5591 * But now, memory offline itself doesn't call shrink_slab()
5592 * and it still to be fixed.
5593 */
5594 /*
5595 * If the page is not RAM, page_count()should be 0.
5596 * we don't need more check. This is an _used_ not-movable page.
5597 *
5598 * The problematic thing here is PG_reserved pages. PG_reserved
5599 * is set to both of a memory hole page and a _used_ kernel
5600 * page at boot.
5601 */
5602 if (found > count)
5603 return false;
5604 }
5605 return true;
5606}
5607
5608bool is_pageblock_removable_nolock(struct page *page)
5609{
5610 struct zone *zone = page_zone(page);
5611 return __count_immobile_pages(zone, page, 0);
5612}
5613
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07005614int set_migratetype_isolate(struct page *page)
5615{
5616 struct zone *zone;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005617 unsigned long flags, pfn;
Robert Jennings925cc712009-12-17 14:44:38 +00005618 struct memory_isolate_notify arg;
5619 int notifier_ret;
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07005620 int ret = -EBUSY;
5621
5622 zone = page_zone(page);
Robert Jennings925cc712009-12-17 14:44:38 +00005623
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07005624 spin_lock_irqsave(&zone->lock, flags);
Robert Jennings925cc712009-12-17 14:44:38 +00005625
5626 pfn = page_to_pfn(page);
5627 arg.start_pfn = pfn;
5628 arg.nr_pages = pageblock_nr_pages;
5629 arg.pages_found = 0;
5630
5631 /*
5632 * It may be possible to isolate a pageblock even if the
5633 * migratetype is not MIGRATE_MOVABLE. The memory isolation
5634 * notifier chain is used by balloon drivers to return the
5635 * number of pages in a range that are held by the balloon
5636 * driver to shrink memory. If all the pages are accounted for
5637 * by balloons, are free, or on the LRU, isolation can continue.
5638 * Later, for example, when memory hotplug notifier runs, these
5639 * pages reported as "can be isolated" should be isolated(freed)
5640 * by the balloon driver through the memory notifier chain.
5641 */
5642 notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg);
5643 notifier_ret = notifier_to_errno(notifier_ret);
KAMEZAWA Hiroyuki4b204772010-10-26 14:21:29 -07005644 if (notifier_ret)
Robert Jennings925cc712009-12-17 14:44:38 +00005645 goto out;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005646 /*
5647 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
5648 * We just check MOVABLE pages.
5649 */
5650 if (__count_immobile_pages(zone, page, arg.pages_found))
Robert Jennings925cc712009-12-17 14:44:38 +00005651 ret = 0;
5652
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005653 /*
5654 * immobile means "not-on-lru" paes. If immobile is larger than
5655 * removable-by-driver pages reported by notifier, we'll fail.
5656 */
5657
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07005658out:
Robert Jennings925cc712009-12-17 14:44:38 +00005659 if (!ret) {
5660 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
5661 move_freepages_block(zone, page, MIGRATE_ISOLATE);
5662 }
5663
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07005664 spin_unlock_irqrestore(&zone->lock, flags);
5665 if (!ret)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005666 drain_all_pages();
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07005667 return ret;
5668}
5669
5670void unset_migratetype_isolate(struct page *page)
5671{
5672 struct zone *zone;
5673 unsigned long flags;
5674 zone = page_zone(page);
5675 spin_lock_irqsave(&zone->lock, flags);
5676 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
5677 goto out;
5678 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5679 move_freepages_block(zone, page, MIGRATE_MOVABLE);
5680out:
5681 spin_unlock_irqrestore(&zone->lock, flags);
5682}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07005683
5684#ifdef CONFIG_MEMORY_HOTREMOVE
5685/*
5686 * All pages in the range must be isolated before calling this.
5687 */
5688void
5689__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
5690{
5691 struct page *page;
5692 struct zone *zone;
5693 int order, i;
5694 unsigned long pfn;
5695 unsigned long flags;
5696 /* find the first valid pfn */
5697 for (pfn = start_pfn; pfn < end_pfn; pfn++)
5698 if (pfn_valid(pfn))
5699 break;
5700 if (pfn == end_pfn)
5701 return;
5702 zone = page_zone(pfn_to_page(pfn));
5703 spin_lock_irqsave(&zone->lock, flags);
5704 pfn = start_pfn;
5705 while (pfn < end_pfn) {
5706 if (!pfn_valid(pfn)) {
5707 pfn++;
5708 continue;
5709 }
5710 page = pfn_to_page(pfn);
5711 BUG_ON(page_count(page));
5712 BUG_ON(!PageBuddy(page));
5713 order = page_order(page);
5714#ifdef CONFIG_DEBUG_VM
5715 printk(KERN_INFO "remove from free list %lx %d %lx\n",
5716 pfn, 1 << order, end_pfn);
5717#endif
5718 list_del(&page->lru);
5719 rmv_page_order(page);
5720 zone->free_area[order].nr_free--;
5721 __mod_zone_page_state(zone, NR_FREE_PAGES,
5722 - (1UL << order));
5723 for (i = 0; i < (1 << order); i++)
5724 SetPageReserved((page+i));
5725 pfn += (1 << order);
5726 }
5727 spin_unlock_irqrestore(&zone->lock, flags);
5728}
5729#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01005730
5731#ifdef CONFIG_MEMORY_FAILURE
5732bool is_free_buddy_page(struct page *page)
5733{
5734 struct zone *zone = page_zone(page);
5735 unsigned long pfn = page_to_pfn(page);
5736 unsigned long flags;
5737 int order;
5738
5739 spin_lock_irqsave(&zone->lock, flags);
5740 for (order = 0; order < MAX_ORDER; order++) {
5741 struct page *page_head = page - (pfn & ((1 << order) - 1));
5742
5743 if (PageBuddy(page_head) && page_order(page_head) >= order)
5744 break;
5745 }
5746 spin_unlock_irqrestore(&zone->lock, flags);
5747
5748 return order < MAX_ORDER;
5749}
5750#endif
Wu Fengguang718a3822010-03-10 15:20:43 -08005751
5752static struct trace_print_flags pageflag_names[] = {
5753 {1UL << PG_locked, "locked" },
5754 {1UL << PG_error, "error" },
5755 {1UL << PG_referenced, "referenced" },
5756 {1UL << PG_uptodate, "uptodate" },
5757 {1UL << PG_dirty, "dirty" },
5758 {1UL << PG_lru, "lru" },
5759 {1UL << PG_active, "active" },
5760 {1UL << PG_slab, "slab" },
5761 {1UL << PG_owner_priv_1, "owner_priv_1" },
5762 {1UL << PG_arch_1, "arch_1" },
5763 {1UL << PG_reserved, "reserved" },
5764 {1UL << PG_private, "private" },
5765 {1UL << PG_private_2, "private_2" },
5766 {1UL << PG_writeback, "writeback" },
5767#ifdef CONFIG_PAGEFLAGS_EXTENDED
5768 {1UL << PG_head, "head" },
5769 {1UL << PG_tail, "tail" },
5770#else
5771 {1UL << PG_compound, "compound" },
5772#endif
5773 {1UL << PG_swapcache, "swapcache" },
5774 {1UL << PG_mappedtodisk, "mappedtodisk" },
5775 {1UL << PG_reclaim, "reclaim" },
Wu Fengguang718a3822010-03-10 15:20:43 -08005776 {1UL << PG_swapbacked, "swapbacked" },
5777 {1UL << PG_unevictable, "unevictable" },
5778#ifdef CONFIG_MMU
5779 {1UL << PG_mlocked, "mlocked" },
5780#endif
5781#ifdef CONFIG_ARCH_USES_PG_UNCACHED
5782 {1UL << PG_uncached, "uncached" },
5783#endif
5784#ifdef CONFIG_MEMORY_FAILURE
5785 {1UL << PG_hwpoison, "hwpoison" },
5786#endif
5787 {-1UL, NULL },
5788};
5789
5790static void dump_page_flags(unsigned long flags)
5791{
5792 const char *delim = "";
5793 unsigned long mask;
5794 int i;
5795
5796 printk(KERN_ALERT "page flags: %#lx(", flags);
5797
5798 /* remove zone id */
5799 flags &= (1UL << NR_PAGEFLAGS) - 1;
5800
5801 for (i = 0; pageflag_names[i].name && flags; i++) {
5802
5803 mask = pageflag_names[i].mask;
5804 if ((flags & mask) != mask)
5805 continue;
5806
5807 flags &= ~mask;
5808 printk("%s%s", delim, pageflag_names[i].name);
5809 delim = "|";
5810 }
5811
5812 /* check for left over flags */
5813 if (flags)
5814 printk("%s%#lx", delim, flags);
5815
5816 printk(")\n");
5817}
5818
5819void dump_page(struct page *page)
5820{
5821 printk(KERN_ALERT
5822 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
Andrea Arcangeli4e9f64c2011-01-13 15:46:29 -08005823 page, atomic_read(&page->_count), page_mapcount(page),
Wu Fengguang718a3822010-03-10 15:20:43 -08005824 page->mapping, page->index);
5825 dump_page_flags(page->flags);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07005826 mem_cgroup_print_bad_page(page);
Wu Fengguang718a3822010-03-10 15:20:43 -08005827}