blob: 4e8985acdab8b5c234a6486b3c84caf5fb89dca0 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/page_alloc.c
3 *
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
6 *
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
15 */
16
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/stddef.h>
18#include <linux/mm.h>
19#include <linux/swap.h>
20#include <linux/interrupt.h>
21#include <linux/pagemap.h>
KOSAKI Motohiro10ed2732008-03-04 14:28:32 -080022#include <linux/jiffies.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/bootmem.h>
Yinghai Luedbe7d22010-08-25 13:39:16 -070024#include <linux/memblock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/compiler.h>
Randy Dunlap9f158332005-09-13 01:25:16 -070026#include <linux/kernel.h>
Vegard Nossumb1eeab62008-11-25 16:55:53 +010027#include <linux/kmemcheck.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/module.h>
29#include <linux/suspend.h>
30#include <linux/pagevec.h>
31#include <linux/blkdev.h>
32#include <linux/slab.h>
Dave Hansena238ab52011-05-24 17:12:16 -070033#include <linux/ratelimit.h>
David Rientjes5a3135c22007-10-16 23:25:53 -070034#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <linux/notifier.h>
36#include <linux/topology.h>
37#include <linux/sysctl.h>
38#include <linux/cpu.h>
39#include <linux/cpuset.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070040#include <linux/memory_hotplug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include <linux/nodemask.h>
42#include <linux/vmalloc.h>
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -070043#include <linux/vmstat.h>
Christoph Lameter4be38e32006-01-06 00:11:17 -080044#include <linux/mempolicy.h>
Yasunori Goto68113782006-06-23 02:03:11 -070045#include <linux/stop_machine.h>
Mel Gormanc7132162006-09-27 01:49:43 -070046#include <linux/sort.h>
47#include <linux/pfn.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070048#include <linux/backing-dev.h>
Akinobu Mita933e3122006-12-08 02:39:45 -080049#include <linux/fault-inject.h>
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -070050#include <linux/page-isolation.h>
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070051#include <linux/page_cgroup.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070052#include <linux/debugobjects.h>
Catalin Marinasdbb1f812009-06-11 13:23:19 +010053#include <linux/kmemleak.h>
Robert Jennings925cc712009-12-17 14:44:38 +000054#include <linux/memory.h>
Mel Gorman56de7262010-05-24 14:32:30 -070055#include <linux/compaction.h>
Mel Gorman0d3d0622009-09-21 17:02:44 -070056#include <trace/events/kmem.h>
Wu Fengguang718a3822010-03-10 15:20:43 -080057#include <linux/ftrace_event.h>
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -070058#include <linux/memcontrol.h>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070059#include <linux/prefetch.h>
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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800322out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800323 /* Leave bad fields for debug, except PageBuddy could make trouble */
Andrea Arcangelief2b4b92011-03-18 00:16:35 +0100324 reset_page_mapcount(page); /* remove PageBuddy */
Randy Dunlap9f158332005-09-13 01:25:16 -0700325 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326}
327
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328/*
329 * Higher-order pages are called "compound pages". They are structured thusly:
330 *
331 * The first PAGE_SIZE page is called the "head page".
332 *
333 * The remaining PAGE_SIZE pages are called "tail pages".
334 *
335 * All pages have PG_compound set. All pages have their ->private pointing at
336 * the head page (even the head page has this).
337 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800338 * The first tail page's ->lru.next holds the address of the compound page's
339 * put_page() function. Its ->lru.prev holds the order of allocation.
340 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800342
343static void free_compound_page(struct page *page)
344{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700345 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800346}
347
Andi Kleen01ad1c02008-07-23 21:27:46 -0700348void prep_compound_page(struct page *page, unsigned long order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349{
350 int i;
351 int nr_pages = 1 << order;
352
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800353 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700354 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700355 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800356 for (i = 1; i < nr_pages; i++) {
357 struct page *p = page + i;
358
Christoph Lameterd85f3382007-05-06 14:49:39 -0700359 __SetPageTail(p);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700360 p->first_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 }
362}
363
Andrea Arcangeli59ff4212011-01-13 15:46:38 -0800364/* update __split_huge_page_refcount if you change this function */
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800365static int destroy_compound_page(struct page *page, unsigned long order)
Andy Whitcroft18229df2008-11-06 12:53:27 -0800366{
367 int i;
368 int nr_pages = 1 << order;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800369 int bad = 0;
Andy Whitcroft18229df2008-11-06 12:53:27 -0800370
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800371 if (unlikely(compound_order(page) != order) ||
372 unlikely(!PageHead(page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800373 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800374 bad++;
375 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376
Christoph Lameter6d777952007-05-06 14:49:40 -0700377 __ClearPageHead(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800378
Andy Whitcroft18229df2008-11-06 12:53:27 -0800379 for (i = 1; i < nr_pages; i++) {
380 struct page *p = page + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381
Alexey Zaytseve713a212009-01-10 02:47:57 +0300382 if (unlikely(!PageTail(p) || (p->first_page != page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800383 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800384 bad++;
385 }
Christoph Lameterd85f3382007-05-06 14:49:39 -0700386 __ClearPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800388
389 return bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
Nick Piggin17cf4402006-03-22 00:08:41 -0800392static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
393{
394 int i;
395
Andrew Morton6626c5d2006-03-22 00:08:42 -0800396 /*
397 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
398 * and __GFP_HIGHMEM from hard or soft interrupt context.
399 */
Nick Piggin725d7042006-09-25 23:30:55 -0700400 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800401 for (i = 0; i < (1 << order); i++)
402 clear_highpage(page + i);
403}
404
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700405static inline void set_page_order(struct page *page, int order)
406{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700407 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000408 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409}
410
411static inline void rmv_page_order(struct page *page)
412{
Nick Piggin676165a2006-04-10 11:21:48 +1000413 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700414 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415}
416
417/*
418 * Locate the struct page for both the matching buddy in our
419 * pair (buddy1) and the combined O(n+1) page they form (page).
420 *
421 * 1) Any buddy B1 will have an order O twin B2 which satisfies
422 * the following equation:
423 * B2 = B1 ^ (1 << O)
424 * For example, if the starting buddy (buddy2) is #8 its order
425 * 1 buddy is #10:
426 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
427 *
428 * 2) Any buddy B will have an order O+1 parent P which
429 * satisfies the following equation:
430 * P = B & ~(1 << O)
431 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200432 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434static inline unsigned long
KyongHo Cho43506fa2011-01-13 15:47:24 -0800435__find_buddy_index(unsigned long page_idx, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436{
KyongHo Cho43506fa2011-01-13 15:47:24 -0800437 return page_idx ^ (1 << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438}
439
440/*
441 * This function checks whether a page is free && is the buddy
442 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800443 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000444 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700445 * (c) a page and its buddy have the same order &&
446 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 *
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800448 * For recording whether a page is in the buddy system, we set ->_mapcount -2.
449 * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock.
Nick Piggin676165a2006-04-10 11:21:48 +1000450 *
451 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700453static inline int page_is_buddy(struct page *page, struct page *buddy,
454 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700456 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800457 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800458
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700459 if (page_zone_id(page) != page_zone_id(buddy))
460 return 0;
461
462 if (PageBuddy(buddy) && page_order(buddy) == order) {
Nick Piggina3af9c32009-06-16 15:32:10 -0700463 VM_BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700464 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000465 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700466 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467}
468
469/*
470 * Freeing function for a buddy system allocator.
471 *
472 * The concept of a buddy system is to maintain direct-mapped table
473 * (containing bit values) for memory blocks of various "orders".
474 * The bottom level table contains the map for the smallest allocatable
475 * units of memory (here, pages), and each level above it describes
476 * pairs of units from the levels below, hence, "buddies".
477 * At a high level, all that happens here is marking the table entry
478 * at the bottom level available, and propagating the changes upward
479 * as necessary, plus some accounting needed to play nicely with other
480 * parts of the VM system.
481 * At each level, we keep a list of pages, which are heads of continuous
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800482 * free pages of length of (1 << order) and marked with _mapcount -2. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700483 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 * So when we are allocating or freeing one, we can derive the state of the
485 * other. That is, if we allocate a small block, and both were
486 * free, the remainder of the region must be split into blocks.
487 * If a block is freed, and its buddy is also free, then this
488 * triggers coalescing into a block of larger size.
489 *
490 * -- wli
491 */
492
Nick Piggin48db57f2006-01-08 01:00:42 -0800493static inline void __free_one_page(struct page *page,
Mel Gormaned0ae212009-06-16 15:32:07 -0700494 struct zone *zone, unsigned int order,
495 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496{
497 unsigned long page_idx;
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700498 unsigned long combined_idx;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800499 unsigned long uninitialized_var(buddy_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700500 struct page *buddy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501
Nick Piggin224abf92006-01-06 00:11:11 -0800502 if (unlikely(PageCompound(page)))
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800503 if (unlikely(destroy_compound_page(page, order)))
504 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505
Mel Gormaned0ae212009-06-16 15:32:07 -0700506 VM_BUG_ON(migratetype == -1);
507
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
509
Mel Gormanf2260e62009-06-16 15:32:13 -0700510 VM_BUG_ON(page_idx & ((1 << order) - 1));
Nick Piggin725d7042006-09-25 23:30:55 -0700511 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 while (order < MAX_ORDER-1) {
KyongHo Cho43506fa2011-01-13 15:47:24 -0800514 buddy_idx = __find_buddy_index(page_idx, order);
515 buddy = page + (buddy_idx - page_idx);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700516 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700517 break;
Nick Piggin13e74442006-01-06 00:10:58 -0800518
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700519 /* Our buddy is free, merge with it and move up one order. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 list_del(&buddy->lru);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700521 zone->free_area[order].nr_free--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 rmv_page_order(buddy);
KyongHo Cho43506fa2011-01-13 15:47:24 -0800523 combined_idx = buddy_idx & page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 page = page + (combined_idx - page_idx);
525 page_idx = combined_idx;
526 order++;
527 }
528 set_page_order(page, order);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700529
530 /*
531 * If this is not the largest possible page, check if the buddy
532 * of the next-highest order is free. If it is, it's possible
533 * that pages are being freed that will coalesce soon. In case,
534 * that is happening, add the free page to the tail of the list
535 * so it's less likely to be used soon and more likely to be merged
536 * as a higher order page
537 */
Mel Gormanb7f50cf2010-10-26 14:21:11 -0700538 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700539 struct page *higher_page, *higher_buddy;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800540 combined_idx = buddy_idx & page_idx;
541 higher_page = page + (combined_idx - page_idx);
542 buddy_idx = __find_buddy_index(combined_idx, order + 1);
543 higher_buddy = page + (buddy_idx - combined_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700544 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
545 list_add_tail(&page->lru,
546 &zone->free_area[order].free_list[migratetype]);
547 goto out;
548 }
549 }
550
551 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
552out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 zone->free_area[order].nr_free++;
554}
555
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700556/*
557 * free_page_mlock() -- clean up attempts to free and mlocked() page.
558 * Page should not be on lru, so no need to fix that up.
559 * free_pages_check() will verify...
560 */
561static inline void free_page_mlock(struct page *page)
562{
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700563 __dec_zone_page_state(page, NR_MLOCK);
564 __count_vm_event(UNEVICTABLE_MLOCKFREED);
565}
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700566
Nick Piggin224abf92006-01-06 00:11:11 -0800567static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568{
Nick Piggin92be2e332006-01-06 00:10:57 -0800569 if (unlikely(page_mapcount(page) |
570 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700571 (atomic_read(&page->_count) != 0) |
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -0700572 (page->flags & PAGE_FLAGS_CHECK_AT_FREE) |
573 (mem_cgroup_bad_page_check(page)))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800574 bad_page(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800575 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800576 }
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800577 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
578 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
579 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580}
581
582/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700583 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700585 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 *
587 * If the zone was previously in an "all pages pinned" state then look to
588 * see if this freeing clears that state.
589 *
590 * And clear the zone's pages_scanned counter, to hold off the "all pages are
591 * pinned" detection logic.
592 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700593static void free_pcppages_bulk(struct zone *zone, int count,
594 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595{
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700596 int migratetype = 0;
Mel Gormana6f9edd2009-09-21 17:03:20 -0700597 int batch_free = 0;
Mel Gorman72853e22010-09-09 16:38:16 -0700598 int to_free = count;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700599
Nick Pigginc54ad302006-01-06 00:10:56 -0800600 spin_lock(&zone->lock);
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -0800601 zone->all_unreclaimable = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700603
Mel Gorman72853e22010-09-09 16:38:16 -0700604 while (to_free) {
Nick Piggin48db57f2006-01-08 01:00:42 -0800605 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700606 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -0800607
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700608 /*
Mel Gormana6f9edd2009-09-21 17:03:20 -0700609 * Remove pages from lists in a round-robin fashion. A
610 * batch_free count is maintained that is incremented when an
611 * empty list is encountered. This is so more pages are freed
612 * off fuller lists instead of spinning excessively around empty
613 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700614 */
615 do {
Mel Gormana6f9edd2009-09-21 17:03:20 -0700616 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700617 if (++migratetype == MIGRATE_PCPTYPES)
618 migratetype = 0;
619 list = &pcp->lists[migratetype];
620 } while (list_empty(list));
621
Namhyung Kim1d168712011-03-22 16:32:45 -0700622 /* This is the only non-empty list. Free them all. */
623 if (batch_free == MIGRATE_PCPTYPES)
624 batch_free = to_free;
625
Mel Gormana6f9edd2009-09-21 17:03:20 -0700626 do {
627 page = list_entry(list->prev, struct page, lru);
628 /* must delete as __free_one_page list manipulates */
629 list_del(&page->lru);
Hugh Dickinsa7016232010-01-29 17:46:34 +0000630 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
631 __free_one_page(page, zone, 0, page_private(page));
632 trace_mm_page_pcpu_drain(page, 0, page_private(page));
Mel Gorman72853e22010-09-09 16:38:16 -0700633 } while (--to_free && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 }
Mel Gorman72853e22010-09-09 16:38:16 -0700635 __mod_zone_page_state(zone, NR_FREE_PAGES, count);
Nick Pigginc54ad302006-01-06 00:10:56 -0800636 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637}
638
Mel Gormaned0ae212009-06-16 15:32:07 -0700639static void free_one_page(struct zone *zone, struct page *page, int order,
640 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800641{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700642 spin_lock(&zone->lock);
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -0800643 zone->all_unreclaimable = 0;
Christoph Lameter006d22d2006-09-25 23:31:48 -0700644 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700645
Mel Gormaned0ae212009-06-16 15:32:07 -0700646 __free_one_page(page, zone, order, migratetype);
Mel Gorman72853e22010-09-09 16:38:16 -0700647 __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700648 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800649}
650
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700651static bool free_pages_prepare(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 int i;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800654 int bad = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655
Li Hongf6503162010-03-05 13:41:52 -0800656 trace_mm_page_free_direct(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100657 kmemcheck_free_shadow(page, order);
658
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -0800659 if (PageAnon(page))
660 page->mapping = NULL;
661 for (i = 0; i < (1 << order); i++)
662 bad += free_pages_check(page + i);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800663 if (bad)
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700664 return false;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800665
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700666 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -0700667 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700668 debug_check_no_obj_freed(page_address(page),
669 PAGE_SIZE << order);
670 }
Nick Piggindafb1362006-10-11 01:21:30 -0700671 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800672 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700673
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700674 return true;
675}
676
677static void __free_pages_ok(struct page *page, unsigned int order)
678{
679 unsigned long flags;
680 int wasMlocked = __TestClearPageMlocked(page);
681
682 if (!free_pages_prepare(page, order))
683 return;
684
Nick Pigginc54ad302006-01-06 00:10:56 -0800685 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +0200686 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -0700687 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700688 __count_vm_events(PGFREE, 1 << order);
Mel Gormaned0ae212009-06-16 15:32:07 -0700689 free_one_page(page_zone(page), page, order,
690 get_pageblock_migratetype(page));
Nick Pigginc54ad302006-01-06 00:10:56 -0800691 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692}
693
David Howellsa226f6c2006-01-06 00:11:08 -0800694/*
695 * permit the bootmem allocator to evade page validation on high-order frees
696 */
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700697void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800698{
699 if (order == 0) {
700 __ClearPageReserved(page);
701 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800702 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800703 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800704 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800705 int loop;
706
Nick Piggin545b1ea2006-03-22 00:08:07 -0800707 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800708 for (loop = 0; loop < BITS_PER_LONG; loop++) {
709 struct page *p = &page[loop];
710
Nick Piggin545b1ea2006-03-22 00:08:07 -0800711 if (loop + 1 < BITS_PER_LONG)
712 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800713 __ClearPageReserved(p);
714 set_page_count(p, 0);
715 }
716
Nick Piggin7835e982006-03-22 00:08:40 -0800717 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800718 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800719 }
720}
721
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722
723/*
724 * The order of subdivision here is critical for the IO subsystem.
725 * Please do not alter this order without good reasons and regression
726 * testing. Specifically, as large blocks of memory are subdivided,
727 * the order in which smaller blocks are delivered depends on the order
728 * they're subdivided in this function. This is the primary factor
729 * influencing the order in which pages are delivered to the IO
730 * subsystem according to empirical testing, and this is also justified
731 * by considering the behavior of a buddy system containing a single
732 * large block of memory acted on by a series of small allocations.
733 * This behavior is a critical factor in sglist merging's success.
734 *
735 * -- wli
736 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800737static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700738 int low, int high, struct free_area *area,
739 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740{
741 unsigned long size = 1 << high;
742
743 while (high > low) {
744 area--;
745 high--;
746 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700747 VM_BUG_ON(bad_range(zone, &page[size]));
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700748 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 area->nr_free++;
750 set_page_order(&page[size], high);
751 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752}
753
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754/*
755 * This page is about to be returned from the page allocator
756 */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200757static inline int check_new_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758{
Nick Piggin92be2e332006-01-06 00:10:57 -0800759 if (unlikely(page_mapcount(page) |
760 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700761 (atomic_read(&page->_count) != 0) |
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -0700762 (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
763 (mem_cgroup_bad_page_check(page)))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800764 bad_page(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800765 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800766 }
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200767 return 0;
768}
769
770static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
771{
772 int i;
773
774 for (i = 0; i < (1 << order); i++) {
775 struct page *p = page + i;
776 if (unlikely(check_new_page(p)))
777 return 1;
778 }
Hugh Dickins689bceb2005-11-21 21:32:20 -0800779
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700780 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800781 set_page_refcounted(page);
Nick Piggincc1025092006-12-06 20:32:00 -0800782
783 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800785
786 if (gfp_flags & __GFP_ZERO)
787 prep_zero_page(page, order, gfp_flags);
788
789 if (order && (gfp_flags & __GFP_COMP))
790 prep_compound_page(page, order);
791
Hugh Dickins689bceb2005-11-21 21:32:20 -0800792 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793}
794
Mel Gorman56fd56b2007-10-16 01:25:58 -0700795/*
796 * Go through the free lists for the given migratetype and remove
797 * the smallest available page from the freelists
798 */
Mel Gorman728ec982009-06-16 15:32:04 -0700799static inline
800struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -0700801 int migratetype)
802{
803 unsigned int current_order;
804 struct free_area * area;
805 struct page *page;
806
807 /* Find a page of the appropriate size in the preferred list */
808 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
809 area = &(zone->free_area[current_order]);
810 if (list_empty(&area->free_list[migratetype]))
811 continue;
812
813 page = list_entry(area->free_list[migratetype].next,
814 struct page, lru);
815 list_del(&page->lru);
816 rmv_page_order(page);
817 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -0700818 expand(zone, page, order, current_order, area, migratetype);
819 return page;
820 }
821
822 return NULL;
823}
824
825
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700826/*
827 * This array describes the order lists are fallen back to when
828 * the free lists for the desirable migrate type are depleted
829 */
830static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
Mel Gorman64c5e132007-10-16 01:25:59 -0700831 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
832 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
833 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
834 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700835};
836
Mel Gormanc361be52007-10-16 01:25:51 -0700837/*
838 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -0700839 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -0700840 * boundary. If alignment is required, use move_freepages_block()
841 */
Adrian Bunkb69a7282008-07-23 21:28:12 -0700842static int move_freepages(struct zone *zone,
843 struct page *start_page, struct page *end_page,
844 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700845{
846 struct page *page;
847 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -0700848 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -0700849
850#ifndef CONFIG_HOLES_IN_ZONE
851 /*
852 * page_zone is not safe to call in this context when
853 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
854 * anyway as we check zone boundaries in move_freepages_block().
855 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -0700856 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -0700857 */
858 BUG_ON(page_zone(start_page) != page_zone(end_page));
859#endif
860
861 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -0700862 /* Make sure we are not inadvertently changing nodes */
863 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
864
Mel Gormanc361be52007-10-16 01:25:51 -0700865 if (!pfn_valid_within(page_to_pfn(page))) {
866 page++;
867 continue;
868 }
869
870 if (!PageBuddy(page)) {
871 page++;
872 continue;
873 }
874
875 order = page_order(page);
Kirill A. Shutemov84be48d2011-03-22 16:33:41 -0700876 list_move(&page->lru,
877 &zone->free_area[order].free_list[migratetype]);
Mel Gormanc361be52007-10-16 01:25:51 -0700878 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -0700879 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -0700880 }
881
Mel Gormand1003132007-10-16 01:26:00 -0700882 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -0700883}
884
Adrian Bunkb69a7282008-07-23 21:28:12 -0700885static int move_freepages_block(struct zone *zone, struct page *page,
886 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700887{
888 unsigned long start_pfn, end_pfn;
889 struct page *start_page, *end_page;
890
891 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -0700892 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -0700893 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -0700894 end_page = start_page + pageblock_nr_pages - 1;
895 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -0700896
897 /* Do not cross zone boundaries */
898 if (start_pfn < zone->zone_start_pfn)
899 start_page = page;
900 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
901 return 0;
902
903 return move_freepages(zone, start_page, end_page, migratetype);
904}
905
Mel Gorman2f66a682009-09-21 17:02:31 -0700906static void change_pageblock_range(struct page *pageblock_page,
907 int start_order, int migratetype)
908{
909 int nr_pageblocks = 1 << (start_order - pageblock_order);
910
911 while (nr_pageblocks--) {
912 set_pageblock_migratetype(pageblock_page, migratetype);
913 pageblock_page += pageblock_nr_pages;
914 }
915}
916
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700917/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -0700918static inline struct page *
919__rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700920{
921 struct free_area * area;
922 int current_order;
923 struct page *page;
924 int migratetype, i;
925
926 /* Find the largest possible block of pages in the other list */
927 for (current_order = MAX_ORDER-1; current_order >= order;
928 --current_order) {
929 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
930 migratetype = fallbacks[start_migratetype][i];
931
Mel Gorman56fd56b2007-10-16 01:25:58 -0700932 /* MIGRATE_RESERVE handled later if necessary */
933 if (migratetype == MIGRATE_RESERVE)
934 continue;
Mel Gormane0104872007-10-16 01:25:53 -0700935
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700936 area = &(zone->free_area[current_order]);
937 if (list_empty(&area->free_list[migratetype]))
938 continue;
939
940 page = list_entry(area->free_list[migratetype].next,
941 struct page, lru);
942 area->nr_free--;
943
944 /*
Mel Gormanc361be52007-10-16 01:25:51 -0700945 * If breaking a large block of pages, move all free
Mel Gorman46dafbc2007-10-16 01:25:55 -0700946 * pages to the preferred allocation list. If falling
947 * back for a reclaimable kernel allocation, be more
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300948 * aggressive about taking ownership of free pages
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700949 */
Mel Gormand9c23402007-10-16 01:26:01 -0700950 if (unlikely(current_order >= (pageblock_order >> 1)) ||
Mel Gormandd5d2412009-09-05 11:17:11 -0700951 start_migratetype == MIGRATE_RECLAIMABLE ||
952 page_group_by_mobility_disabled) {
Mel Gorman46dafbc2007-10-16 01:25:55 -0700953 unsigned long pages;
954 pages = move_freepages_block(zone, page,
955 start_migratetype);
956
957 /* Claim the whole block if over half of it is free */
Mel Gormandd5d2412009-09-05 11:17:11 -0700958 if (pages >= (1 << (pageblock_order-1)) ||
959 page_group_by_mobility_disabled)
Mel Gorman46dafbc2007-10-16 01:25:55 -0700960 set_pageblock_migratetype(page,
961 start_migratetype);
962
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700963 migratetype = start_migratetype;
Mel Gormanc361be52007-10-16 01:25:51 -0700964 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700965
966 /* Remove the page from the freelists */
967 list_del(&page->lru);
968 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700969
Mel Gorman2f66a682009-09-21 17:02:31 -0700970 /* Take ownership for orders >= pageblock_order */
971 if (current_order >= pageblock_order)
972 change_pageblock_range(page, current_order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700973 start_migratetype);
974
975 expand(zone, page, order, current_order, area, migratetype);
Mel Gormane0fff1b2009-09-21 17:02:42 -0700976
977 trace_mm_page_alloc_extfrag(page, order, current_order,
978 start_migratetype, migratetype);
979
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700980 return page;
981 }
982 }
983
Mel Gorman728ec982009-06-16 15:32:04 -0700984 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700985}
986
Mel Gorman56fd56b2007-10-16 01:25:58 -0700987/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 * Do the hard work of removing an element from the buddy allocator.
989 * Call me with the zone->lock already held.
990 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700991static struct page *__rmqueue(struct zone *zone, unsigned int order,
992 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 struct page *page;
995
Mel Gorman728ec982009-06-16 15:32:04 -0700996retry_reserve:
Mel Gorman56fd56b2007-10-16 01:25:58 -0700997 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998
Mel Gorman728ec982009-06-16 15:32:04 -0700999 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07001000 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001001
Mel Gorman728ec982009-06-16 15:32:04 -07001002 /*
1003 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1004 * is used because __rmqueue_smallest is an inline function
1005 * and we want just one call site
1006 */
1007 if (!page) {
1008 migratetype = MIGRATE_RESERVE;
1009 goto retry_reserve;
1010 }
1011 }
1012
Mel Gorman0d3d0622009-09-21 17:02:44 -07001013 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001014 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015}
1016
1017/*
1018 * Obtain a specified number of elements from the buddy allocator, all under
1019 * a single hold of the lock, for efficiency. Add them to the supplied list.
1020 * Returns the number of new pages which were placed at *list.
1021 */
1022static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001023 unsigned long count, struct list_head *list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001024 int migratetype, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027
Nick Pigginc54ad302006-01-06 00:10:56 -08001028 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001030 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001031 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08001033
1034 /*
1035 * Split buddy pages returned by expand() are received here
1036 * in physical page order. The page is added to the callers and
1037 * list and the list head then moves forward. From the callers
1038 * perspective, the linked list is ordered by page number in
1039 * some conditions. This is useful for IO devices that can
1040 * merge IO requests if the physical pages are ordered
1041 * properly.
1042 */
Mel Gormane084b2d2009-07-29 15:02:04 -07001043 if (likely(cold == 0))
1044 list_add(&page->lru, list);
1045 else
1046 list_add_tail(&page->lru, list);
Mel Gorman535131e62007-10-16 01:25:49 -07001047 set_page_private(page, migratetype);
Mel Gorman81eabcb2007-12-17 16:20:05 -08001048 list = &page->lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 }
Mel Gormanf2260e62009-06-16 15:32:13 -07001050 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08001051 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001052 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053}
1054
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001055#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001056/*
Christoph Lameter4037d452007-05-09 02:35:14 -07001057 * Called from the vmstat counter updater to drain pagesets of this
1058 * currently executing processor on remote nodes after they have
1059 * expired.
1060 *
Christoph Lameter879336c2006-03-22 00:09:08 -08001061 * Note that this function must be called with the thread pinned to
1062 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001063 */
Christoph Lameter4037d452007-05-09 02:35:14 -07001064void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001065{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001066 unsigned long flags;
Christoph Lameter4037d452007-05-09 02:35:14 -07001067 int to_drain;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001068
Christoph Lameter4037d452007-05-09 02:35:14 -07001069 local_irq_save(flags);
1070 if (pcp->count >= pcp->batch)
1071 to_drain = pcp->batch;
1072 else
1073 to_drain = pcp->count;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001074 free_pcppages_bulk(zone, to_drain, pcp);
Christoph Lameter4037d452007-05-09 02:35:14 -07001075 pcp->count -= to_drain;
1076 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001077}
1078#endif
1079
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001080/*
1081 * Drain pages of the indicated processor.
1082 *
1083 * The processor must either be the current processor and the
1084 * thread pinned to the current processor or a processor that
1085 * is not online.
1086 */
1087static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088{
Nick Pigginc54ad302006-01-06 00:10:56 -08001089 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07001092 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 struct per_cpu_pageset *pset;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001094 struct per_cpu_pages *pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001096 local_irq_save(flags);
1097 pset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001099 pcp = &pset->pcp;
David Rientjes2ff754f2011-01-25 15:07:23 -08001100 if (pcp->count) {
1101 free_pcppages_bulk(zone, pcp->count, pcp);
1102 pcp->count = 0;
1103 }
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001104 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 }
1106}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001108/*
1109 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1110 */
1111void drain_local_pages(void *arg)
1112{
1113 drain_pages(smp_processor_id());
1114}
1115
1116/*
1117 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
1118 */
1119void drain_all_pages(void)
1120{
Jens Axboe15c8b6c2008-05-09 09:39:44 +02001121 on_each_cpu(drain_local_pages, NULL, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001122}
1123
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02001124#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125
1126void mark_free_pages(struct zone *zone)
1127{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001128 unsigned long pfn, max_zone_pfn;
1129 unsigned long flags;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001130 int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 struct list_head *curr;
1132
1133 if (!zone->spanned_pages)
1134 return;
1135
1136 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001137
1138 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1139 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1140 if (pfn_valid(pfn)) {
1141 struct page *page = pfn_to_page(pfn);
1142
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001143 if (!swsusp_page_is_forbidden(page))
1144 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001145 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001147 for_each_migratetype_order(order, t) {
1148 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001149 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001151 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1152 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001153 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001154 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001155 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 spin_unlock_irqrestore(&zone->lock, flags);
1157}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001158#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159
1160/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 * Free a 0-order page
Li Hongfc916682010-03-05 13:41:54 -08001162 * cold == 1 ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 */
Li Hongfc916682010-03-05 13:41:54 -08001164void free_hot_cold_page(struct page *page, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165{
1166 struct zone *zone = page_zone(page);
1167 struct per_cpu_pages *pcp;
1168 unsigned long flags;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001169 int migratetype;
Johannes Weiner451ea252009-09-21 17:01:48 -07001170 int wasMlocked = __TestClearPageMlocked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001172 if (!free_pages_prepare(page, 0))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001173 return;
1174
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001175 migratetype = get_pageblock_migratetype(page);
1176 set_page_private(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +02001178 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -07001179 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001180 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001181
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001182 /*
1183 * We only track unmovable, reclaimable and movable on pcp lists.
1184 * Free ISOLATE pages back to the allocator because they are being
1185 * offlined but treat RESERVE as movable pages so we can get those
1186 * areas back if necessary. Otherwise, we may have to free
1187 * excessively into the page allocator
1188 */
1189 if (migratetype >= MIGRATE_PCPTYPES) {
1190 if (unlikely(migratetype == MIGRATE_ISOLATE)) {
1191 free_one_page(zone, page, 0, migratetype);
1192 goto out;
1193 }
1194 migratetype = MIGRATE_MOVABLE;
1195 }
1196
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001197 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001198 if (cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001199 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001200 else
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001201 list_add(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001203 if (pcp->count >= pcp->high) {
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001204 free_pcppages_bulk(zone, pcp->batch, pcp);
Nick Piggin48db57f2006-01-08 01:00:42 -08001205 pcp->count -= pcp->batch;
1206 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001207
1208out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210}
1211
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001212/*
1213 * split_page takes a non-compound higher-order page, and splits it into
1214 * n (1<<order) sub-pages: page[0..n]
1215 * Each sub-page must be freed individually.
1216 *
1217 * Note: this is probably too low level an operation for use in drivers.
1218 * Please consult with lkml before using this in your driver.
1219 */
1220void split_page(struct page *page, unsigned int order)
1221{
1222 int i;
1223
Nick Piggin725d7042006-09-25 23:30:55 -07001224 VM_BUG_ON(PageCompound(page));
1225 VM_BUG_ON(!page_count(page));
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001226
1227#ifdef CONFIG_KMEMCHECK
1228 /*
1229 * Split shadow pages too, because free(page[0]) would
1230 * otherwise free the whole shadow.
1231 */
1232 if (kmemcheck_page_is_tracked(page))
1233 split_page(virt_to_page(page[0].shadow), order);
1234#endif
1235
Nick Piggin7835e982006-03-22 00:08:40 -08001236 for (i = 1; i < (1 << order); i++)
1237 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001238}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001239
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240/*
Mel Gorman748446b2010-05-24 14:32:27 -07001241 * Similar to split_page except the page is already free. As this is only
1242 * being used for migration, the migratetype of the block also changes.
1243 * As this is called with interrupts disabled, the caller is responsible
1244 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1245 * are enabled.
1246 *
1247 * Note: this is probably too low level an operation for use in drivers.
1248 * Please consult with lkml before using this in your driver.
1249 */
1250int split_free_page(struct page *page)
1251{
1252 unsigned int order;
1253 unsigned long watermark;
1254 struct zone *zone;
1255
1256 BUG_ON(!PageBuddy(page));
1257
1258 zone = page_zone(page);
1259 order = page_order(page);
1260
1261 /* Obey watermarks as if the page was being allocated */
1262 watermark = low_wmark_pages(zone) + (1 << order);
1263 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1264 return 0;
1265
1266 /* Remove page from free list */
1267 list_del(&page->lru);
1268 zone->free_area[order].nr_free--;
1269 rmv_page_order(page);
1270 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1UL << order));
1271
1272 /* Split into individual pages */
1273 set_page_refcounted(page);
1274 split_page(page, order);
1275
1276 if (order >= pageblock_order - 1) {
1277 struct page *endpage = page + (1 << order) - 1;
1278 for (; page < endpage; page += pageblock_nr_pages)
1279 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1280 }
1281
1282 return 1 << order;
1283}
1284
1285/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1287 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1288 * or two.
1289 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07001290static inline
1291struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001292 struct zone *zone, int order, gfp_t gfp_flags,
1293 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294{
1295 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001296 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 int cold = !!(gfp_flags & __GFP_COLD);
1298
Hugh Dickins689bceb2005-11-21 21:32:20 -08001299again:
Nick Piggin48db57f2006-01-08 01:00:42 -08001300 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001302 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 local_irq_save(flags);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001305 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1306 list = &pcp->lists[migratetype];
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001307 if (list_empty(list)) {
Mel Gorman535131e62007-10-16 01:25:49 -07001308 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001309 pcp->batch, list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001310 migratetype, cold);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001311 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07001312 goto failed;
Mel Gorman535131e62007-10-16 01:25:49 -07001313 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001314
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001315 if (cold)
1316 page = list_entry(list->prev, struct page, lru);
1317 else
1318 page = list_entry(list->next, struct page, lru);
1319
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001320 list_del(&page->lru);
1321 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001322 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07001323 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1324 /*
1325 * __GFP_NOFAIL is not to be used in new code.
1326 *
1327 * All __GFP_NOFAIL callers should be fixed so that they
1328 * properly detect and handle allocation failures.
1329 *
1330 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07001331 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07001332 * __GFP_NOFAIL.
1333 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07001334 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07001335 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001337 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001338 spin_unlock(&zone->lock);
1339 if (!page)
1340 goto failed;
KOSAKI Motohiro6ccf80e2010-01-15 17:01:18 -08001341 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 }
1343
Christoph Lameterf8891e52006-06-30 01:55:45 -07001344 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Andi Kleen78afd562011-03-22 16:33:12 -07001345 zone_statistics(preferred_zone, zone, gfp_flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001346 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347
Nick Piggin725d7042006-09-25 23:30:55 -07001348 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -08001349 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001350 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001352
1353failed:
1354 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001355 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356}
1357
Mel Gorman41858962009-06-16 15:32:12 -07001358/* The ALLOC_WMARK bits are used as an index to zone->watermark */
1359#define ALLOC_WMARK_MIN WMARK_MIN
1360#define ALLOC_WMARK_LOW WMARK_LOW
1361#define ALLOC_WMARK_HIGH WMARK_HIGH
1362#define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1363
1364/* Mask to get the watermark bits */
1365#define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1366
Nick Piggin31488902005-11-28 13:44:03 -08001367#define ALLOC_HARDER 0x10 /* try to alloc harder */
1368#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1369#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001370
Akinobu Mita933e3122006-12-08 02:39:45 -08001371#ifdef CONFIG_FAIL_PAGE_ALLOC
1372
1373static struct fail_page_alloc_attr {
1374 struct fault_attr attr;
1375
1376 u32 ignore_gfp_highmem;
1377 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001378 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001379
1380#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1381
1382 struct dentry *ignore_gfp_highmem_file;
1383 struct dentry *ignore_gfp_wait_file;
Akinobu Mita54114992007-07-15 23:40:23 -07001384 struct dentry *min_order_file;
Akinobu Mita933e3122006-12-08 02:39:45 -08001385
1386#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1387
1388} fail_page_alloc = {
1389 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001390 .ignore_gfp_wait = 1,
1391 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001392 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001393};
1394
1395static int __init setup_fail_page_alloc(char *str)
1396{
1397 return setup_fault_attr(&fail_page_alloc.attr, str);
1398}
1399__setup("fail_page_alloc=", setup_fail_page_alloc);
1400
1401static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1402{
Akinobu Mita54114992007-07-15 23:40:23 -07001403 if (order < fail_page_alloc.min_order)
1404 return 0;
Akinobu Mita933e3122006-12-08 02:39:45 -08001405 if (gfp_mask & __GFP_NOFAIL)
1406 return 0;
1407 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1408 return 0;
1409 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1410 return 0;
1411
1412 return should_fail(&fail_page_alloc.attr, 1 << order);
1413}
1414
1415#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1416
1417static int __init fail_page_alloc_debugfs(void)
1418{
1419 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1420 struct dentry *dir;
1421 int err;
1422
1423 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1424 "fail_page_alloc");
1425 if (err)
1426 return err;
1427 dir = fail_page_alloc.attr.dentries.dir;
1428
1429 fail_page_alloc.ignore_gfp_wait_file =
1430 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1431 &fail_page_alloc.ignore_gfp_wait);
1432
1433 fail_page_alloc.ignore_gfp_highmem_file =
1434 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1435 &fail_page_alloc.ignore_gfp_highmem);
Akinobu Mita54114992007-07-15 23:40:23 -07001436 fail_page_alloc.min_order_file =
1437 debugfs_create_u32("min-order", mode, dir,
1438 &fail_page_alloc.min_order);
Akinobu Mita933e3122006-12-08 02:39:45 -08001439
1440 if (!fail_page_alloc.ignore_gfp_wait_file ||
Akinobu Mita54114992007-07-15 23:40:23 -07001441 !fail_page_alloc.ignore_gfp_highmem_file ||
1442 !fail_page_alloc.min_order_file) {
Akinobu Mita933e3122006-12-08 02:39:45 -08001443 err = -ENOMEM;
1444 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1445 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
Akinobu Mita54114992007-07-15 23:40:23 -07001446 debugfs_remove(fail_page_alloc.min_order_file);
Akinobu Mita933e3122006-12-08 02:39:45 -08001447 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1448 }
1449
1450 return err;
1451}
1452
1453late_initcall(fail_page_alloc_debugfs);
1454
1455#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1456
1457#else /* CONFIG_FAIL_PAGE_ALLOC */
1458
1459static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1460{
1461 return 0;
1462}
1463
1464#endif /* CONFIG_FAIL_PAGE_ALLOC */
1465
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466/*
Mel Gorman88f5acf2011-01-13 15:45:41 -08001467 * Return true if free pages are above 'mark'. This takes into account the order
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 * of the allocation.
1469 */
Mel Gorman88f5acf2011-01-13 15:45:41 -08001470static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1471 int classzone_idx, int alloc_flags, long free_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472{
1473 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001474 long min = mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 int o;
1476
Mel Gorman88f5acf2011-01-13 15:45:41 -08001477 free_pages -= (1 << order) + 1;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001478 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001480 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 min -= min / 4;
1482
1483 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
Mel Gorman88f5acf2011-01-13 15:45:41 -08001484 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 for (o = 0; o < order; o++) {
1486 /* At the next order, this order's pages become unavailable */
1487 free_pages -= z->free_area[o].nr_free << o;
1488
1489 /* Require fewer higher order pages to be free */
1490 min >>= 1;
1491
1492 if (free_pages <= min)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001493 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 }
Mel Gorman88f5acf2011-01-13 15:45:41 -08001495 return true;
1496}
1497
1498bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1499 int classzone_idx, int alloc_flags)
1500{
1501 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1502 zone_page_state(z, NR_FREE_PAGES));
1503}
1504
1505bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
1506 int classzone_idx, int alloc_flags)
1507{
1508 long free_pages = zone_page_state(z, NR_FREE_PAGES);
1509
1510 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1511 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1512
1513 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1514 free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515}
1516
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001517#ifdef CONFIG_NUMA
1518/*
1519 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1520 * skip over zones that are not allowed by the cpuset, or that have
1521 * been recently (in last second) found to be nearly full. See further
1522 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001523 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001524 *
1525 * If the zonelist cache is present in the passed in zonelist, then
1526 * returns a pointer to the allowed node mask (either the current
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001527 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001528 *
1529 * If the zonelist cache is not available for this zonelist, does
1530 * nothing and returns NULL.
1531 *
1532 * If the fullzones BITMAP in the zonelist cache is stale (more than
1533 * a second since last zap'd) then we zap it out (clear its bits.)
1534 *
1535 * We hold off even calling zlc_setup, until after we've checked the
1536 * first zone in the zonelist, on the theory that most allocations will
1537 * be satisfied from that first zone, so best to examine that zone as
1538 * quickly as we can.
1539 */
1540static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1541{
1542 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1543 nodemask_t *allowednodes; /* zonelist_cache approximation */
1544
1545 zlc = zonelist->zlcache_ptr;
1546 if (!zlc)
1547 return NULL;
1548
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001549 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001550 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1551 zlc->last_full_zap = jiffies;
1552 }
1553
1554 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1555 &cpuset_current_mems_allowed :
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001556 &node_states[N_HIGH_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001557 return allowednodes;
1558}
1559
1560/*
1561 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1562 * if it is worth looking at further for free memory:
1563 * 1) Check that the zone isn't thought to be full (doesn't have its
1564 * bit set in the zonelist_cache fullzones BITMAP).
1565 * 2) Check that the zones node (obtained from the zonelist_cache
1566 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1567 * Return true (non-zero) if zone is worth looking at further, or
1568 * else return false (zero) if it is not.
1569 *
1570 * This check -ignores- the distinction between various watermarks,
1571 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1572 * found to be full for any variation of these watermarks, it will
1573 * be considered full for up to one second by all requests, unless
1574 * we are so low on memory on all allowed nodes that we are forced
1575 * into the second scan of the zonelist.
1576 *
1577 * In the second scan we ignore this zonelist cache and exactly
1578 * apply the watermarks to all zones, even it is slower to do so.
1579 * We are low on memory in the second scan, and should leave no stone
1580 * unturned looking for a free page.
1581 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001582static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001583 nodemask_t *allowednodes)
1584{
1585 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1586 int i; /* index of *z in zonelist zones */
1587 int n; /* node that zone *z is on */
1588
1589 zlc = zonelist->zlcache_ptr;
1590 if (!zlc)
1591 return 1;
1592
Mel Gormandd1a2392008-04-28 02:12:17 -07001593 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001594 n = zlc->z_to_n[i];
1595
1596 /* This zone is worth trying if it is allowed but not full */
1597 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1598}
1599
1600/*
1601 * Given 'z' scanning a zonelist, set the corresponding bit in
1602 * zlc->fullzones, so that subsequent attempts to allocate a page
1603 * from that zone don't waste time re-examining it.
1604 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001605static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001606{
1607 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1608 int i; /* index of *z in zonelist zones */
1609
1610 zlc = zonelist->zlcache_ptr;
1611 if (!zlc)
1612 return;
1613
Mel Gormandd1a2392008-04-28 02:12:17 -07001614 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001615
1616 set_bit(i, zlc->fullzones);
1617}
1618
1619#else /* CONFIG_NUMA */
1620
1621static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1622{
1623 return NULL;
1624}
1625
Mel Gormandd1a2392008-04-28 02:12:17 -07001626static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001627 nodemask_t *allowednodes)
1628{
1629 return 1;
1630}
1631
Mel Gormandd1a2392008-04-28 02:12:17 -07001632static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001633{
1634}
1635#endif /* CONFIG_NUMA */
1636
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001637/*
Paul Jackson0798e512006-12-06 20:31:38 -08001638 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001639 * a page.
1640 */
1641static struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001642get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
Mel Gorman5117f452009-06-16 15:31:59 -07001643 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
Mel Gorman3dd28262009-06-16 15:32:00 -07001644 struct zone *preferred_zone, int migratetype)
Martin Hicks753ee722005-06-21 17:14:41 -07001645{
Mel Gormandd1a2392008-04-28 02:12:17 -07001646 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001647 struct page *page = NULL;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001648 int classzone_idx;
Mel Gorman5117f452009-06-16 15:31:59 -07001649 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001650 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1651 int zlc_active = 0; /* set if using zonelist_cache */
1652 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gorman54a6eb52008-04-28 02:12:16 -07001653
Mel Gorman19770b32008-04-28 02:12:18 -07001654 classzone_idx = zone_idx(preferred_zone);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001655zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001656 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001657 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001658 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1659 */
Mel Gorman19770b32008-04-28 02:12:18 -07001660 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1661 high_zoneidx, nodemask) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001662 if (NUMA_BUILD && zlc_active &&
1663 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1664 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001665 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001666 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001667 goto try_next_zone;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001668
Mel Gorman41858962009-06-16 15:32:12 -07001669 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001670 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001671 unsigned long mark;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001672 int ret;
1673
Mel Gorman41858962009-06-16 15:32:12 -07001674 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
Mel Gormanfa5e0842009-06-16 15:33:22 -07001675 if (zone_watermark_ok(zone, order, mark,
1676 classzone_idx, alloc_flags))
1677 goto try_this_zone;
1678
1679 if (zone_reclaim_mode == 0)
1680 goto this_zone_full;
1681
1682 ret = zone_reclaim(zone, gfp_mask, order);
1683 switch (ret) {
1684 case ZONE_RECLAIM_NOSCAN:
1685 /* did not scan */
1686 goto try_next_zone;
1687 case ZONE_RECLAIM_FULL:
1688 /* scanned but unreclaimable */
1689 goto this_zone_full;
1690 default:
1691 /* did we reclaim enough */
1692 if (!zone_watermark_ok(zone, order, mark,
1693 classzone_idx, alloc_flags))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001694 goto this_zone_full;
Paul Jackson0798e512006-12-06 20:31:38 -08001695 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001696 }
1697
Mel Gormanfa5e0842009-06-16 15:33:22 -07001698try_this_zone:
Mel Gorman3dd28262009-06-16 15:32:00 -07001699 page = buffered_rmqueue(preferred_zone, zone, order,
1700 gfp_mask, migratetype);
Paul Jackson0798e512006-12-06 20:31:38 -08001701 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001702 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001703this_zone_full:
1704 if (NUMA_BUILD)
1705 zlc_mark_zone_full(zonelist, z);
1706try_next_zone:
Christoph Lameter62bc62a2009-06-16 15:32:15 -07001707 if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) {
Mel Gormand395b732009-06-16 15:32:09 -07001708 /*
1709 * we do zlc_setup after the first zone is tried but only
1710 * if there are multiple nodes make it worthwhile
1711 */
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001712 allowednodes = zlc_setup(zonelist, alloc_flags);
1713 zlc_active = 1;
1714 did_zlc_setup = 1;
1715 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07001716 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001717
1718 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1719 /* Disable zlc cache for second zonelist scan */
1720 zlc_active = 0;
1721 goto zonelist_scan;
1722 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001723 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07001724}
1725
David Rientjes29423e772011-03-22 16:30:47 -07001726/*
1727 * Large machines with many possible nodes should not always dump per-node
1728 * meminfo in irq context.
1729 */
1730static inline bool should_suppress_show_mem(void)
1731{
1732 bool ret = false;
1733
1734#if NODES_SHIFT > 8
1735 ret = in_interrupt();
1736#endif
1737 return ret;
1738}
1739
Dave Hansena238ab52011-05-24 17:12:16 -07001740static DEFINE_RATELIMIT_STATE(nopage_rs,
1741 DEFAULT_RATELIMIT_INTERVAL,
1742 DEFAULT_RATELIMIT_BURST);
1743
1744void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
1745{
1746 va_list args;
1747 unsigned int filter = SHOW_MEM_FILTER_NODES;
1748
1749 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs))
1750 return;
1751
1752 /*
1753 * This documents exceptions given to allocations in certain
1754 * contexts that are allowed to allocate outside current's set
1755 * of allowed nodes.
1756 */
1757 if (!(gfp_mask & __GFP_NOMEMALLOC))
1758 if (test_thread_flag(TIF_MEMDIE) ||
1759 (current->flags & (PF_MEMALLOC | PF_EXITING)))
1760 filter &= ~SHOW_MEM_FILTER_NODES;
1761 if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
1762 filter &= ~SHOW_MEM_FILTER_NODES;
1763
1764 if (fmt) {
1765 printk(KERN_WARNING);
1766 va_start(args, fmt);
1767 vprintk(fmt, args);
1768 va_end(args);
1769 }
1770
1771 pr_warning("%s: page allocation failure: order:%d, mode:0x%x\n",
1772 current->comm, order, gfp_mask);
1773
1774 dump_stack();
1775 if (!should_suppress_show_mem())
1776 show_mem(filter);
1777}
1778
Mel Gorman11e33f62009-06-16 15:31:57 -07001779static inline int
1780should_alloc_retry(gfp_t gfp_mask, unsigned int order,
1781 unsigned long pages_reclaimed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782{
Mel Gorman11e33f62009-06-16 15:31:57 -07001783 /* Do not loop if specifically requested */
1784 if (gfp_mask & __GFP_NORETRY)
1785 return 0;
1786
1787 /*
1788 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1789 * means __GFP_NOFAIL, but that may not be true in other
1790 * implementations.
1791 */
1792 if (order <= PAGE_ALLOC_COSTLY_ORDER)
1793 return 1;
1794
1795 /*
1796 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1797 * specified, then we retry until we no longer reclaim any pages
1798 * (above), or we've reclaimed an order of pages at least as
1799 * large as the allocation's order. In both cases, if the
1800 * allocation still fails, we stop retrying.
1801 */
1802 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
1803 return 1;
1804
1805 /*
1806 * Don't let big-order allocations loop unless the caller
1807 * explicitly requests that.
1808 */
1809 if (gfp_mask & __GFP_NOFAIL)
1810 return 1;
1811
1812 return 0;
1813}
1814
1815static inline struct page *
1816__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1817 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001818 nodemask_t *nodemask, struct zone *preferred_zone,
1819 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001820{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822
Mel Gorman11e33f62009-06-16 15:31:57 -07001823 /* Acquire the OOM killer lock for the zones in zonelist */
Minchan Kimff321fe2010-08-09 17:18:57 -07001824 if (!try_set_zonelist_oom(zonelist, gfp_mask)) {
Mel Gorman11e33f62009-06-16 15:31:57 -07001825 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 return NULL;
1827 }
Jens Axboe6b1de912005-11-17 21:35:02 +01001828
Mel Gorman11e33f62009-06-16 15:31:57 -07001829 /*
1830 * Go through the zonelist yet one more time, keep very high watermark
1831 * here, this is only to catch a parallel oom killing, we must fail if
1832 * we're still under heavy pressure.
1833 */
1834 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1835 order, zonelist, high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001836 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07001837 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001838 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07001839 goto out;
1840
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001841 if (!(gfp_mask & __GFP_NOFAIL)) {
1842 /* The OOM killer will not help higher order allocs */
1843 if (order > PAGE_ALLOC_COSTLY_ORDER)
1844 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07001845 /* The OOM killer does not needlessly kill tasks for lowmem */
1846 if (high_zoneidx < ZONE_NORMAL)
1847 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001848 /*
1849 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
1850 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
1851 * The caller should handle page allocation failure by itself if
1852 * it specifies __GFP_THISNODE.
1853 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
1854 */
1855 if (gfp_mask & __GFP_THISNODE)
1856 goto out;
1857 }
Mel Gorman11e33f62009-06-16 15:31:57 -07001858 /* Exhausted what can be done so it's blamo time */
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001859 out_of_memory(zonelist, gfp_mask, order, nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07001860
1861out:
1862 clear_zonelist_oom(zonelist, gfp_mask);
1863 return page;
1864}
1865
Mel Gorman56de7262010-05-24 14:32:30 -07001866#ifdef CONFIG_COMPACTION
1867/* Try memory compaction for high-order allocations before reclaim */
1868static struct page *
1869__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
1870 struct zonelist *zonelist, enum zone_type high_zoneidx,
1871 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman77f1fe62011-01-13 15:45:57 -08001872 int migratetype, unsigned long *did_some_progress,
1873 bool sync_migration)
Mel Gorman56de7262010-05-24 14:32:30 -07001874{
1875 struct page *page;
1876
Mel Gorman4f92e252010-05-24 14:32:32 -07001877 if (!order || compaction_deferred(preferred_zone))
Mel Gorman56de7262010-05-24 14:32:30 -07001878 return NULL;
1879
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001880 current->flags |= PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07001881 *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
Mel Gorman77f1fe62011-01-13 15:45:57 -08001882 nodemask, sync_migration);
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001883 current->flags &= ~PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07001884 if (*did_some_progress != COMPACT_SKIPPED) {
1885
1886 /* Page migration frees to the PCP lists but we want merging */
1887 drain_pages(get_cpu());
1888 put_cpu();
1889
1890 page = get_page_from_freelist(gfp_mask, nodemask,
1891 order, zonelist, high_zoneidx,
1892 alloc_flags, preferred_zone,
1893 migratetype);
1894 if (page) {
Mel Gorman4f92e252010-05-24 14:32:32 -07001895 preferred_zone->compact_considered = 0;
1896 preferred_zone->compact_defer_shift = 0;
Mel Gorman56de7262010-05-24 14:32:30 -07001897 count_vm_event(COMPACTSUCCESS);
1898 return page;
1899 }
1900
1901 /*
1902 * It's bad if compaction run occurs and fails.
1903 * The most likely reason is that pages exist,
1904 * but not enough to satisfy watermarks.
1905 */
1906 count_vm_event(COMPACTFAIL);
Mel Gorman4f92e252010-05-24 14:32:32 -07001907 defer_compaction(preferred_zone);
Mel Gorman56de7262010-05-24 14:32:30 -07001908
1909 cond_resched();
1910 }
1911
1912 return NULL;
1913}
1914#else
1915static inline struct page *
1916__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
1917 struct zonelist *zonelist, enum zone_type high_zoneidx,
1918 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman77f1fe62011-01-13 15:45:57 -08001919 int migratetype, unsigned long *did_some_progress,
1920 bool sync_migration)
Mel Gorman56de7262010-05-24 14:32:30 -07001921{
1922 return NULL;
1923}
1924#endif /* CONFIG_COMPACTION */
1925
Mel Gorman11e33f62009-06-16 15:31:57 -07001926/* The really slow allocator path where we enter direct reclaim */
1927static inline struct page *
1928__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
1929 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001930 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001931 int migratetype, unsigned long *did_some_progress)
Mel Gorman11e33f62009-06-16 15:31:57 -07001932{
1933 struct page *page = NULL;
1934 struct reclaim_state reclaim_state;
Mel Gorman9ee493c2010-09-09 16:38:18 -07001935 bool drained = false;
Mel Gorman11e33f62009-06-16 15:31:57 -07001936
1937 cond_resched();
1938
1939 /* We now go into synchronous reclaim */
1940 cpuset_memory_pressure_bump();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001941 current->flags |= PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07001942 lockdep_set_current_reclaim_state(gfp_mask);
1943 reclaim_state.reclaimed_slab = 0;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001944 current->reclaim_state = &reclaim_state;
Mel Gorman11e33f62009-06-16 15:31:57 -07001945
1946 *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
1947
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001948 current->reclaim_state = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07001949 lockdep_clear_current_reclaim_state();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08001950 current->flags &= ~PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07001951
1952 cond_resched();
1953
Mel Gorman9ee493c2010-09-09 16:38:18 -07001954 if (unlikely(!(*did_some_progress)))
1955 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07001956
Mel Gorman9ee493c2010-09-09 16:38:18 -07001957retry:
1958 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001959 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001960 alloc_flags, preferred_zone,
1961 migratetype);
Mel Gorman9ee493c2010-09-09 16:38:18 -07001962
1963 /*
1964 * If an allocation failed after direct reclaim, it could be because
1965 * pages are pinned on the per-cpu lists. Drain them and try again
1966 */
1967 if (!page && !drained) {
1968 drain_all_pages();
1969 drained = true;
1970 goto retry;
1971 }
1972
Mel Gorman11e33f62009-06-16 15:31:57 -07001973 return page;
1974}
1975
Mel Gorman11e33f62009-06-16 15:31:57 -07001976/*
1977 * This is called in the allocator slow-path if the allocation request is of
1978 * sufficient urgency to ignore watermarks and take other desperate measures
1979 */
1980static inline struct page *
1981__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
1982 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001983 nodemask_t *nodemask, struct zone *preferred_zone,
1984 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001985{
1986 struct page *page;
1987
1988 do {
1989 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001990 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
Mel Gorman3dd28262009-06-16 15:32:00 -07001991 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001992
1993 if (!page && gfp_mask & __GFP_NOFAIL)
Mel Gorman0e093d992010-10-26 14:21:45 -07001994 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07001995 } while (!page && (gfp_mask & __GFP_NOFAIL));
1996
1997 return page;
1998}
1999
2000static inline
2001void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
Mel Gorman99504742011-01-13 15:46:20 -08002002 enum zone_type high_zoneidx,
2003 enum zone_type classzone_idx)
Mel Gorman11e33f62009-06-16 15:31:57 -07002004{
2005 struct zoneref *z;
2006 struct zone *zone;
2007
2008 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
Mel Gorman99504742011-01-13 15:46:20 -08002009 wakeup_kswapd(zone, order, classzone_idx);
Mel Gorman11e33f62009-06-16 15:31:57 -07002010}
2011
Peter Zijlstra341ce062009-06-16 15:32:02 -07002012static inline int
2013gfp_to_alloc_flags(gfp_t gfp_mask)
2014{
Peter Zijlstra341ce062009-06-16 15:32:02 -07002015 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
2016 const gfp_t wait = gfp_mask & __GFP_WAIT;
2017
Mel Gormana56f57f2009-06-16 15:32:02 -07002018 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07002019 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07002020
Peter Zijlstra341ce062009-06-16 15:32:02 -07002021 /*
2022 * The caller may dip into page reserves a bit more if the caller
2023 * cannot run direct reclaim, or if the caller has realtime scheduling
2024 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2025 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
2026 */
Namhyung Kime6223a32010-10-26 14:21:59 -07002027 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07002028
2029 if (!wait) {
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002030 /*
2031 * Not worth trying to allocate harder for
2032 * __GFP_NOMEMALLOC even if it can't schedule.
2033 */
2034 if (!(gfp_mask & __GFP_NOMEMALLOC))
2035 alloc_flags |= ALLOC_HARDER;
Peter Zijlstra341ce062009-06-16 15:32:02 -07002036 /*
2037 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
2038 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
2039 */
2040 alloc_flags &= ~ALLOC_CPUSET;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002041 } else if (unlikely(rt_task(current)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07002042 alloc_flags |= ALLOC_HARDER;
2043
2044 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2045 if (!in_interrupt() &&
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002046 ((current->flags & PF_MEMALLOC) ||
Peter Zijlstra341ce062009-06-16 15:32:02 -07002047 unlikely(test_thread_flag(TIF_MEMDIE))))
2048 alloc_flags |= ALLOC_NO_WATERMARKS;
2049 }
2050
2051 return alloc_flags;
2052}
2053
Mel Gorman11e33f62009-06-16 15:31:57 -07002054static inline struct page *
2055__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
2056 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002057 nodemask_t *nodemask, struct zone *preferred_zone,
2058 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002059{
2060 const gfp_t wait = gfp_mask & __GFP_WAIT;
2061 struct page *page = NULL;
2062 int alloc_flags;
2063 unsigned long pages_reclaimed = 0;
2064 unsigned long did_some_progress;
Mel Gorman77f1fe62011-01-13 15:45:57 -08002065 bool sync_migration = false;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002066
Christoph Lameter952f3b52006-12-06 20:33:26 -08002067 /*
Mel Gorman72807a72009-06-16 15:32:18 -07002068 * In the slowpath, we sanity check order to avoid ever trying to
2069 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2070 * be using allocators in order of preference for an area that is
2071 * too large.
2072 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07002073 if (order >= MAX_ORDER) {
2074 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07002075 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07002076 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002077
Christoph Lameter952f3b52006-12-06 20:33:26 -08002078 /*
2079 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2080 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2081 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2082 * using a larger set of nodes after it has established that the
2083 * allowed per node queues are empty and that nodes are
2084 * over allocated.
2085 */
2086 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
2087 goto nopage;
2088
Mel Gormancc4a6852009-11-11 14:26:14 -08002089restart:
Andrea Arcangeli32dba982011-01-13 15:46:49 -08002090 if (!(gfp_mask & __GFP_NO_KSWAPD))
2091 wake_all_kswapd(order, zonelist, high_zoneidx,
Mel Gorman99504742011-01-13 15:46:20 -08002092 zone_idx(preferred_zone));
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002093
Paul Jackson9bf22292005-09-06 15:18:12 -07002094 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002095 * OK, we're below the kswapd watermark and have kicked background
2096 * reclaim. Now things get more complex, so set up alloc_flags according
2097 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07002098 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002099 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100
David Rientjesf33261d2011-01-25 15:07:20 -08002101 /*
2102 * Find the true preferred zone if the allocation is unconstrained by
2103 * cpusets.
2104 */
2105 if (!(alloc_flags & ALLOC_CPUSET) && !nodemask)
2106 first_zones_zonelist(zonelist, high_zoneidx, NULL,
2107 &preferred_zone);
2108
Andrew Barrycfa54a02011-05-24 17:12:52 -07002109rebalance:
Peter Zijlstra341ce062009-06-16 15:32:02 -07002110 /* This is the last chance, in general, before the goto nopage. */
Mel Gorman19770b32008-04-28 02:12:18 -07002111 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
Peter Zijlstra341ce062009-06-16 15:32:02 -07002112 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
2113 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002114 if (page)
2115 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116
Mel Gorman11e33f62009-06-16 15:31:57 -07002117 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002118 if (alloc_flags & ALLOC_NO_WATERMARKS) {
2119 page = __alloc_pages_high_priority(gfp_mask, order,
2120 zonelist, high_zoneidx, nodemask,
2121 preferred_zone, migratetype);
2122 if (page)
2123 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 }
2125
2126 /* Atomic allocations - we can't balance anything */
2127 if (!wait)
2128 goto nopage;
2129
Peter Zijlstra341ce062009-06-16 15:32:02 -07002130 /* Avoid recursion of direct reclaim */
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002131 if (current->flags & PF_MEMALLOC)
Peter Zijlstra341ce062009-06-16 15:32:02 -07002132 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133
David Rientjes6583bb62009-07-29 15:02:06 -07002134 /* Avoid allocations with no watermarks from looping endlessly */
2135 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2136 goto nopage;
2137
Mel Gorman77f1fe62011-01-13 15:45:57 -08002138 /*
2139 * Try direct compaction. The first pass is asynchronous. Subsequent
2140 * attempts after direct reclaim are synchronous
2141 */
Mel Gorman56de7262010-05-24 14:32:30 -07002142 page = __alloc_pages_direct_compact(gfp_mask, order,
2143 zonelist, high_zoneidx,
2144 nodemask,
2145 alloc_flags, preferred_zone,
Mel Gorman77f1fe62011-01-13 15:45:57 -08002146 migratetype, &did_some_progress,
2147 sync_migration);
Mel Gorman56de7262010-05-24 14:32:30 -07002148 if (page)
2149 goto got_pg;
Andrea Arcangelic6a140b2011-05-24 17:11:38 -07002150 sync_migration = true;
Mel Gorman56de7262010-05-24 14:32:30 -07002151
Mel Gorman11e33f62009-06-16 15:31:57 -07002152 /* Try direct reclaim and then allocating */
2153 page = __alloc_pages_direct_reclaim(gfp_mask, order,
2154 zonelist, high_zoneidx,
2155 nodemask,
Mel Gorman5117f452009-06-16 15:31:59 -07002156 alloc_flags, preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07002157 migratetype, &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07002158 if (page)
2159 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 /*
Mel Gorman11e33f62009-06-16 15:31:57 -07002162 * If we failed to make any progress reclaiming, then we are
2163 * running out of options and have to consider going OOM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 */
Mel Gorman11e33f62009-06-16 15:31:57 -07002165 if (!did_some_progress) {
2166 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -07002167 if (oom_killer_disabled)
2168 goto nopage;
Mel Gorman11e33f62009-06-16 15:31:57 -07002169 page = __alloc_pages_may_oom(gfp_mask, order,
2170 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002171 nodemask, preferred_zone,
2172 migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002173 if (page)
2174 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175
David Rientjes03668b32010-08-09 17:18:54 -07002176 if (!(gfp_mask & __GFP_NOFAIL)) {
2177 /*
2178 * The oom killer is not called for high-order
2179 * allocations that may fail, so if no progress
2180 * is being made, there are no other options and
2181 * retrying is unlikely to help.
2182 */
2183 if (order > PAGE_ALLOC_COSTLY_ORDER)
2184 goto nopage;
2185 /*
2186 * The oom killer is not called for lowmem
2187 * allocations to prevent needlessly killing
2188 * innocent tasks.
2189 */
2190 if (high_zoneidx < ZONE_NORMAL)
2191 goto nopage;
2192 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 goto restart;
2195 }
2196 }
2197
Mel Gorman11e33f62009-06-16 15:31:57 -07002198 /* Check if we should retry the allocation */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07002199 pages_reclaimed += did_some_progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07002200 if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
2201 /* Wait for some write requests to complete then retry */
Mel Gorman0e093d992010-10-26 14:21:45 -07002202 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 goto rebalance;
Mel Gorman3e7d3442011-01-13 15:45:56 -08002204 } else {
2205 /*
2206 * High-order allocations do not necessarily loop after
2207 * direct reclaim and reclaim/compaction depends on compaction
2208 * being called after reclaim so call directly if necessary
2209 */
2210 page = __alloc_pages_direct_compact(gfp_mask, order,
2211 zonelist, high_zoneidx,
2212 nodemask,
2213 alloc_flags, preferred_zone,
Mel Gorman77f1fe62011-01-13 15:45:57 -08002214 migratetype, &did_some_progress,
2215 sync_migration);
Mel Gorman3e7d3442011-01-13 15:45:56 -08002216 if (page)
2217 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 }
2219
2220nopage:
Dave Hansena238ab52011-05-24 17:12:16 -07002221 warn_alloc_failed(gfp_mask, order, NULL);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002222 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223got_pg:
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002224 if (kmemcheck_enabled)
2225 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 return page;
Mel Gorman11e33f62009-06-16 15:31:57 -07002227
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228}
Mel Gorman11e33f62009-06-16 15:31:57 -07002229
2230/*
2231 * This is the 'heart' of the zoned buddy allocator.
2232 */
2233struct page *
2234__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2235 struct zonelist *zonelist, nodemask_t *nodemask)
2236{
2237 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Mel Gorman5117f452009-06-16 15:31:59 -07002238 struct zone *preferred_zone;
Mel Gorman11e33f62009-06-16 15:31:57 -07002239 struct page *page;
Mel Gorman3dd28262009-06-16 15:32:00 -07002240 int migratetype = allocflags_to_migratetype(gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002241
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10002242 gfp_mask &= gfp_allowed_mask;
2243
Mel Gorman11e33f62009-06-16 15:31:57 -07002244 lockdep_trace_alloc(gfp_mask);
2245
2246 might_sleep_if(gfp_mask & __GFP_WAIT);
2247
2248 if (should_fail_alloc_page(gfp_mask, order))
2249 return NULL;
2250
2251 /*
2252 * Check the zones suitable for the gfp_mask contain at least one
2253 * valid zone. It's possible to have an empty zonelist as a result
2254 * of GFP_THISNODE and a memoryless node
2255 */
2256 if (unlikely(!zonelist->_zonerefs->zone))
2257 return NULL;
2258
Miao Xiec0ff7452010-05-24 14:32:08 -07002259 get_mems_allowed();
Mel Gorman5117f452009-06-16 15:31:59 -07002260 /* The preferred zone is used for statistics later */
David Rientjesf33261d2011-01-25 15:07:20 -08002261 first_zones_zonelist(zonelist, high_zoneidx,
2262 nodemask ? : &cpuset_current_mems_allowed,
2263 &preferred_zone);
Miao Xiec0ff7452010-05-24 14:32:08 -07002264 if (!preferred_zone) {
2265 put_mems_allowed();
Mel Gorman5117f452009-06-16 15:31:59 -07002266 return NULL;
Miao Xiec0ff7452010-05-24 14:32:08 -07002267 }
Mel Gorman5117f452009-06-16 15:31:59 -07002268
2269 /* First allocation attempt */
Mel Gorman11e33f62009-06-16 15:31:57 -07002270 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002271 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07002272 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002273 if (unlikely(!page))
2274 page = __alloc_pages_slowpath(gfp_mask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002275 zonelist, high_zoneidx, nodemask,
Mel Gorman3dd28262009-06-16 15:32:00 -07002276 preferred_zone, migratetype);
Miao Xiec0ff7452010-05-24 14:32:08 -07002277 put_mems_allowed();
Mel Gorman11e33f62009-06-16 15:31:57 -07002278
Mel Gorman4b4f2782009-09-21 17:02:41 -07002279 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002280 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281}
Mel Gormand2391712009-06-16 15:31:52 -07002282EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283
2284/*
2285 * Common helper functions.
2286 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08002287unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288{
Akinobu Mita945a1112009-09-21 17:01:47 -07002289 struct page *page;
2290
2291 /*
2292 * __get_free_pages() returns a 32-bit address, which cannot represent
2293 * a highmem page
2294 */
2295 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2296
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 page = alloc_pages(gfp_mask, order);
2298 if (!page)
2299 return 0;
2300 return (unsigned long) page_address(page);
2301}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302EXPORT_SYMBOL(__get_free_pages);
2303
Harvey Harrison920c7a52008-02-04 22:29:26 -08002304unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305{
Akinobu Mita945a1112009-09-21 17:01:47 -07002306 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308EXPORT_SYMBOL(get_zeroed_page);
2309
2310void __pagevec_free(struct pagevec *pvec)
2311{
2312 int i = pagevec_count(pvec);
2313
Mel Gorman4b4f2782009-09-21 17:02:41 -07002314 while (--i >= 0) {
2315 trace_mm_pagevec_free(pvec->pages[i], pvec->cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 free_hot_cold_page(pvec->pages[i], pvec->cold);
Mel Gorman4b4f2782009-09-21 17:02:41 -07002317 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318}
2319
Harvey Harrison920c7a52008-02-04 22:29:26 -08002320void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321{
Nick Pigginb5810032005-10-29 18:16:12 -07002322 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 if (order == 0)
Li Hongfc916682010-03-05 13:41:54 -08002324 free_hot_cold_page(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 else
2326 __free_pages_ok(page, order);
2327 }
2328}
2329
2330EXPORT_SYMBOL(__free_pages);
2331
Harvey Harrison920c7a52008-02-04 22:29:26 -08002332void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333{
2334 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07002335 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 __free_pages(virt_to_page((void *)addr), order);
2337 }
2338}
2339
2340EXPORT_SYMBOL(free_pages);
2341
Andi Kleenee85c2e2011-05-11 15:13:34 -07002342static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
2343{
2344 if (addr) {
2345 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2346 unsigned long used = addr + PAGE_ALIGN(size);
2347
2348 split_page(virt_to_page((void *)addr), order);
2349 while (used < alloc_end) {
2350 free_page(used);
2351 used += PAGE_SIZE;
2352 }
2353 }
2354 return (void *)addr;
2355}
2356
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002357/**
2358 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2359 * @size: the number of bytes to allocate
2360 * @gfp_mask: GFP flags for the allocation
2361 *
2362 * This function is similar to alloc_pages(), except that it allocates the
2363 * minimum number of pages to satisfy the request. alloc_pages() can only
2364 * allocate memory in power-of-two pages.
2365 *
2366 * This function is also limited by MAX_ORDER.
2367 *
2368 * Memory allocated by this function must be released by free_pages_exact().
2369 */
2370void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2371{
2372 unsigned int order = get_order(size);
2373 unsigned long addr;
2374
2375 addr = __get_free_pages(gfp_mask, order);
Andi Kleenee85c2e2011-05-11 15:13:34 -07002376 return make_alloc_exact(addr, order, size);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002377}
2378EXPORT_SYMBOL(alloc_pages_exact);
2379
2380/**
Andi Kleenee85c2e2011-05-11 15:13:34 -07002381 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2382 * pages on a node.
Randy Dunlapb5e6ab52011-05-16 13:16:54 -07002383 * @nid: the preferred node ID where memory should be allocated
Andi Kleenee85c2e2011-05-11 15:13:34 -07002384 * @size: the number of bytes to allocate
2385 * @gfp_mask: GFP flags for the allocation
2386 *
2387 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2388 * back.
2389 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2390 * but is not exact.
2391 */
2392void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
2393{
2394 unsigned order = get_order(size);
2395 struct page *p = alloc_pages_node(nid, gfp_mask, order);
2396 if (!p)
2397 return NULL;
2398 return make_alloc_exact((unsigned long)page_address(p), order, size);
2399}
2400EXPORT_SYMBOL(alloc_pages_exact_nid);
2401
2402/**
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002403 * free_pages_exact - release memory allocated via alloc_pages_exact()
2404 * @virt: the value returned by alloc_pages_exact.
2405 * @size: size of allocation, same value as passed to alloc_pages_exact().
2406 *
2407 * Release the memory allocated by a previous call to alloc_pages_exact.
2408 */
2409void free_pages_exact(void *virt, size_t size)
2410{
2411 unsigned long addr = (unsigned long)virt;
2412 unsigned long end = addr + PAGE_ALIGN(size);
2413
2414 while (addr < end) {
2415 free_page(addr);
2416 addr += PAGE_SIZE;
2417 }
2418}
2419EXPORT_SYMBOL(free_pages_exact);
2420
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421static unsigned int nr_free_zone_pages(int offset)
2422{
Mel Gormandd1a2392008-04-28 02:12:17 -07002423 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002424 struct zone *zone;
2425
Martin J. Blighe310fd42005-07-29 22:59:18 -07002426 /* Just pick one node, since fallback list is circular */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 unsigned int sum = 0;
2428
Mel Gorman0e884602008-04-28 02:12:14 -07002429 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430
Mel Gorman54a6eb52008-04-28 02:12:16 -07002431 for_each_zone_zonelist(zone, z, zonelist, offset) {
Martin J. Blighe310fd42005-07-29 22:59:18 -07002432 unsigned long size = zone->present_pages;
Mel Gorman41858962009-06-16 15:32:12 -07002433 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07002434 if (size > high)
2435 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 }
2437
2438 return sum;
2439}
2440
2441/*
2442 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2443 */
2444unsigned int nr_free_buffer_pages(void)
2445{
Al Viroaf4ca452005-10-21 02:55:38 -04002446 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447}
Meelap Shahc2f1a552007-07-17 04:04:39 -07002448EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449
2450/*
2451 * Amount of free RAM allocatable within all zones
2452 */
2453unsigned int nr_free_pagecache_pages(void)
2454{
Mel Gorman2a1e2742007-07-17 04:03:12 -07002455 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002457
2458static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459{
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002460 if (NUMA_BUILD)
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08002461 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464void si_meminfo(struct sysinfo *val)
2465{
2466 val->totalram = totalram_pages;
2467 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002468 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470 val->totalhigh = totalhigh_pages;
2471 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 val->mem_unit = PAGE_SIZE;
2473}
2474
2475EXPORT_SYMBOL(si_meminfo);
2476
2477#ifdef CONFIG_NUMA
2478void si_meminfo_node(struct sysinfo *val, int nid)
2479{
2480 pg_data_t *pgdat = NODE_DATA(nid);
2481
2482 val->totalram = pgdat->node_present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002483 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002484#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002486 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2487 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002488#else
2489 val->totalhigh = 0;
2490 val->freehigh = 0;
2491#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492 val->mem_unit = PAGE_SIZE;
2493}
2494#endif
2495
David Rientjesddd588b2011-03-22 16:30:46 -07002496/*
David Rientjes7bf02ea2011-05-24 17:11:16 -07002497 * Determine whether the node should be displayed or not, depending on whether
2498 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
David Rientjesddd588b2011-03-22 16:30:46 -07002499 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07002500bool skip_free_areas_node(unsigned int flags, int nid)
David Rientjesddd588b2011-03-22 16:30:46 -07002501{
2502 bool ret = false;
2503
2504 if (!(flags & SHOW_MEM_FILTER_NODES))
2505 goto out;
2506
2507 get_mems_allowed();
David Rientjes7bf02ea2011-05-24 17:11:16 -07002508 ret = !node_isset(nid, cpuset_current_mems_allowed);
David Rientjesddd588b2011-03-22 16:30:46 -07002509 put_mems_allowed();
2510out:
2511 return ret;
2512}
2513
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514#define K(x) ((x) << (PAGE_SHIFT-10))
2515
2516/*
2517 * Show free area list (used inside shift_scroll-lock stuff)
2518 * We also calculate the percentage fragmentation. We do this by counting the
2519 * memory on each free list with the exception of the first item on the list.
David Rientjesddd588b2011-03-22 16:30:46 -07002520 * Suppresses nodes that are not allowed by current's cpuset if
2521 * SHOW_MEM_FILTER_NODES is passed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07002523void show_free_areas(unsigned int filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524{
Jes Sorensenc7241912006-09-27 01:50:05 -07002525 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526 struct zone *zone;
2527
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002528 for_each_populated_zone(zone) {
David Rientjes7bf02ea2011-05-24 17:11:16 -07002529 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07002530 continue;
Jes Sorensenc7241912006-09-27 01:50:05 -07002531 show_node(zone);
2532 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533
Dave Jones6b482c62005-11-10 15:45:56 -05002534 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535 struct per_cpu_pageset *pageset;
2536
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002537 pageset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002539 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2540 cpu, pageset->pcp.high,
2541 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 }
2543 }
2544
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002545 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2546 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002547 " unevictable:%lu"
Andrew Mortonb76146e2009-10-26 16:49:52 -07002548 " dirty:%lu writeback:%lu unstable:%lu\n"
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07002549 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002550 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002551 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002552 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002553 global_page_state(NR_ISOLATED_ANON),
2554 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002555 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002556 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002557 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07002558 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07002559 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07002560 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08002561 global_page_state(NR_FREE_PAGES),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07002562 global_page_state(NR_SLAB_RECLAIMABLE),
2563 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07002564 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002565 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08002566 global_page_state(NR_PAGETABLE),
2567 global_page_state(NR_BOUNCE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002569 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 int i;
2571
David Rientjes7bf02ea2011-05-24 17:11:16 -07002572 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07002573 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574 show_node(zone);
2575 printk("%s"
2576 " free:%lukB"
2577 " min:%lukB"
2578 " low:%lukB"
2579 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002580 " active_anon:%lukB"
2581 " inactive_anon:%lukB"
2582 " active_file:%lukB"
2583 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002584 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002585 " isolated(anon):%lukB"
2586 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587 " present:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002588 " mlocked:%lukB"
2589 " dirty:%lukB"
2590 " writeback:%lukB"
2591 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002592 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002593 " slab_reclaimable:%lukB"
2594 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07002595 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002596 " pagetables:%lukB"
2597 " unstable:%lukB"
2598 " bounce:%lukB"
2599 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 " pages_scanned:%lu"
2601 " all_unreclaimable? %s"
2602 "\n",
2603 zone->name,
Mel Gorman88f5acf2011-01-13 15:45:41 -08002604 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07002605 K(min_wmark_pages(zone)),
2606 K(low_wmark_pages(zone)),
2607 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002608 K(zone_page_state(zone, NR_ACTIVE_ANON)),
2609 K(zone_page_state(zone, NR_INACTIVE_ANON)),
2610 K(zone_page_state(zone, NR_ACTIVE_FILE)),
2611 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002612 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002613 K(zone_page_state(zone, NR_ISOLATED_ANON)),
2614 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 K(zone->present_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002616 K(zone_page_state(zone, NR_MLOCK)),
2617 K(zone_page_state(zone, NR_FILE_DIRTY)),
2618 K(zone_page_state(zone, NR_WRITEBACK)),
2619 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002620 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002621 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
2622 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07002623 zone_page_state(zone, NR_KERNEL_STACK) *
2624 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002625 K(zone_page_state(zone, NR_PAGETABLE)),
2626 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
2627 K(zone_page_state(zone, NR_BOUNCE)),
2628 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629 zone->pages_scanned,
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -08002630 (zone->all_unreclaimable ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631 );
2632 printk("lowmem_reserve[]:");
2633 for (i = 0; i < MAX_NR_ZONES; i++)
2634 printk(" %lu", zone->lowmem_reserve[i]);
2635 printk("\n");
2636 }
2637
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002638 for_each_populated_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002639 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640
David Rientjes7bf02ea2011-05-24 17:11:16 -07002641 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07002642 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643 show_node(zone);
2644 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645
2646 spin_lock_irqsave(&zone->lock, flags);
2647 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002648 nr[order] = zone->free_area[order].nr_free;
2649 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650 }
2651 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002652 for (order = 0; order < MAX_ORDER; order++)
2653 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 printk("= %lukB\n", K(total));
2655 }
2656
Larry Woodmane6f36022008-02-04 22:29:30 -08002657 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2658
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659 show_swap_cache_info();
2660}
2661
Mel Gorman19770b32008-04-28 02:12:18 -07002662static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2663{
2664 zoneref->zone = zone;
2665 zoneref->zone_idx = zone_idx(zone);
2666}
2667
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668/*
2669 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08002670 *
2671 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002673static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2674 int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675{
Christoph Lameter1a932052006-01-06 00:11:16 -08002676 struct zone *zone;
2677
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002678 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002679 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08002680
2681 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002682 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08002683 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08002684 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002685 zoneref_set_zone(zone,
2686 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08002687 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08002689
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002690 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08002691 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692}
2693
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002694
2695/*
2696 * zonelist_order:
2697 * 0 = automatic detection of better ordering.
2698 * 1 = order by ([node] distance, -zonetype)
2699 * 2 = order by (-zonetype, [node] distance)
2700 *
2701 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2702 * the same zonelist. So only NUMA can configure this param.
2703 */
2704#define ZONELIST_ORDER_DEFAULT 0
2705#define ZONELIST_ORDER_NODE 1
2706#define ZONELIST_ORDER_ZONE 2
2707
2708/* zonelist order in the kernel.
2709 * set_zonelist_order() will set this to NODE or ZONE.
2710 */
2711static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2712static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2713
2714
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002716/* The value user specified ....changed by config */
2717static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2718/* string for sysctl */
2719#define NUMA_ZONELIST_ORDER_LEN 16
2720char numa_zonelist_order[16] = "default";
2721
2722/*
2723 * interface for configure zonelist ordering.
2724 * command line option "numa_zonelist_order"
2725 * = "[dD]efault - default, automatic configuration.
2726 * = "[nN]ode - order by node locality, then by zone within node
2727 * = "[zZ]one - order by zone, then by locality within zone
2728 */
2729
2730static int __parse_numa_zonelist_order(char *s)
2731{
2732 if (*s == 'd' || *s == 'D') {
2733 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2734 } else if (*s == 'n' || *s == 'N') {
2735 user_zonelist_order = ZONELIST_ORDER_NODE;
2736 } else if (*s == 'z' || *s == 'Z') {
2737 user_zonelist_order = ZONELIST_ORDER_ZONE;
2738 } else {
2739 printk(KERN_WARNING
2740 "Ignoring invalid numa_zonelist_order value: "
2741 "%s\n", s);
2742 return -EINVAL;
2743 }
2744 return 0;
2745}
2746
2747static __init int setup_numa_zonelist_order(char *s)
2748{
Volodymyr G. Lukiianykecb256f2011-01-13 15:46:26 -08002749 int ret;
2750
2751 if (!s)
2752 return 0;
2753
2754 ret = __parse_numa_zonelist_order(s);
2755 if (ret == 0)
2756 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
2757
2758 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002759}
2760early_param("numa_zonelist_order", setup_numa_zonelist_order);
2761
2762/*
2763 * sysctl handler for numa_zonelist_order
2764 */
2765int numa_zonelist_order_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07002766 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002767 loff_t *ppos)
2768{
2769 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2770 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01002771 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002772
Andi Kleen443c6f12009-12-23 21:00:47 +01002773 mutex_lock(&zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002774 if (write)
Andi Kleen443c6f12009-12-23 21:00:47 +01002775 strcpy(saved_string, (char*)table->data);
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07002776 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002777 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01002778 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002779 if (write) {
2780 int oldval = user_zonelist_order;
2781 if (__parse_numa_zonelist_order((char*)table->data)) {
2782 /*
2783 * bogus value. restore saved string
2784 */
2785 strncpy((char*)table->data, saved_string,
2786 NUMA_ZONELIST_ORDER_LEN);
2787 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07002788 } else if (oldval != user_zonelist_order) {
2789 mutex_lock(&zonelists_mutex);
Haicheng Li1f522502010-05-24 14:32:51 -07002790 build_all_zonelists(NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07002791 mutex_unlock(&zonelists_mutex);
2792 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002793 }
Andi Kleen443c6f12009-12-23 21:00:47 +01002794out:
2795 mutex_unlock(&zl_order_mutex);
2796 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002797}
2798
2799
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002800#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002801static int node_load[MAX_NUMNODES];
2802
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07002804 * 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 -07002805 * @node: node whose fallback list we're appending
2806 * @used_node_mask: nodemask_t of already used nodes
2807 *
2808 * We use a number of factors to determine which is the next node that should
2809 * appear on a given node's fallback list. The node should not have appeared
2810 * already in @node's fallback list, and it should be the next closest node
2811 * according to the distance array (which contains arbitrary distance values
2812 * from each node to each node in the system), and should also prefer nodes
2813 * with no CPUs, since presumably they'll have very little allocation pressure
2814 * on them otherwise.
2815 * It returns -1 if no node is found.
2816 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002817static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002819 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820 int min_val = INT_MAX;
2821 int best_node = -1;
Rusty Russella70f7302009-03-13 14:49:46 +10302822 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002823
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002824 /* Use the local node if we haven't already */
2825 if (!node_isset(node, *used_node_mask)) {
2826 node_set(node, *used_node_mask);
2827 return node;
2828 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002830 for_each_node_state(n, N_HIGH_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831
2832 /* Don't want a node to appear more than once */
2833 if (node_isset(n, *used_node_mask))
2834 continue;
2835
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836 /* Use the distance array to find the distance */
2837 val = node_distance(node, n);
2838
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002839 /* Penalize nodes under us ("prefer the next node") */
2840 val += (n < node);
2841
Linus Torvalds1da177e2005-04-16 15:20:36 -07002842 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10302843 tmp = cpumask_of_node(n);
2844 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845 val += PENALTY_FOR_NODE_WITH_CPUS;
2846
2847 /* Slight preference for less loaded node */
2848 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2849 val += node_load[n];
2850
2851 if (val < min_val) {
2852 min_val = val;
2853 best_node = n;
2854 }
2855 }
2856
2857 if (best_node >= 0)
2858 node_set(best_node, *used_node_mask);
2859
2860 return best_node;
2861}
2862
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002863
2864/*
2865 * Build zonelists ordered by node and zones within node.
2866 * This results in maximum locality--normal zone overflows into local
2867 * DMA zone, if any--but risks exhausting DMA zone.
2868 */
2869static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002871 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002873
Mel Gorman54a6eb52008-04-28 02:12:16 -07002874 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07002875 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07002876 ;
2877 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2878 MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002879 zonelist->_zonerefs[j].zone = NULL;
2880 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002881}
2882
2883/*
Christoph Lameter523b9452007-10-16 01:25:37 -07002884 * Build gfp_thisnode zonelists
2885 */
2886static void build_thisnode_zonelists(pg_data_t *pgdat)
2887{
Christoph Lameter523b9452007-10-16 01:25:37 -07002888 int j;
2889 struct zonelist *zonelist;
2890
Mel Gorman54a6eb52008-04-28 02:12:16 -07002891 zonelist = &pgdat->node_zonelists[1];
2892 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002893 zonelist->_zonerefs[j].zone = NULL;
2894 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07002895}
2896
2897/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002898 * Build zonelists ordered by zone and nodes within zones.
2899 * This results in conserving DMA zone[s] until all Normal memory is
2900 * exhausted, but results in overflowing to remote node while memory
2901 * may still exist in local DMA zone.
2902 */
2903static int node_order[MAX_NUMNODES];
2904
2905static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2906{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002907 int pos, j, node;
2908 int zone_type; /* needs to be signed */
2909 struct zone *z;
2910 struct zonelist *zonelist;
2911
Mel Gorman54a6eb52008-04-28 02:12:16 -07002912 zonelist = &pgdat->node_zonelists[0];
2913 pos = 0;
2914 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2915 for (j = 0; j < nr_nodes; j++) {
2916 node = node_order[j];
2917 z = &NODE_DATA(node)->node_zones[zone_type];
2918 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002919 zoneref_set_zone(z,
2920 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002921 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002922 }
2923 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002924 }
Mel Gormandd1a2392008-04-28 02:12:17 -07002925 zonelist->_zonerefs[pos].zone = NULL;
2926 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002927}
2928
2929static int default_zonelist_order(void)
2930{
2931 int nid, zone_type;
2932 unsigned long low_kmem_size,total_size;
2933 struct zone *z;
2934 int average_size;
2935 /*
Thomas Weber88393162010-03-16 11:47:56 +01002936 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002937 * If they are really small and used heavily, the system can fall
2938 * into OOM very easily.
David Rientjese325c902010-05-24 14:32:13 -07002939 * This function detect ZONE_DMA/DMA32 size and configures zone order.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002940 */
2941 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2942 low_kmem_size = 0;
2943 total_size = 0;
2944 for_each_online_node(nid) {
2945 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2946 z = &NODE_DATA(nid)->node_zones[zone_type];
2947 if (populated_zone(z)) {
2948 if (zone_type < ZONE_NORMAL)
2949 low_kmem_size += z->present_pages;
2950 total_size += z->present_pages;
David Rientjese325c902010-05-24 14:32:13 -07002951 } else if (zone_type == ZONE_NORMAL) {
2952 /*
2953 * If any node has only lowmem, then node order
2954 * is preferred to allow kernel allocations
2955 * locally; otherwise, they can easily infringe
2956 * on other nodes when there is an abundance of
2957 * lowmem available to allocate from.
2958 */
2959 return ZONELIST_ORDER_NODE;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002960 }
2961 }
2962 }
2963 if (!low_kmem_size || /* there are no DMA area. */
2964 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2965 return ZONELIST_ORDER_NODE;
2966 /*
2967 * look into each node's config.
2968 * If there is a node whose DMA/DMA32 memory is very big area on
2969 * local memory, NODE_ORDER may be suitable.
2970 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002971 average_size = total_size /
2972 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002973 for_each_online_node(nid) {
2974 low_kmem_size = 0;
2975 total_size = 0;
2976 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2977 z = &NODE_DATA(nid)->node_zones[zone_type];
2978 if (populated_zone(z)) {
2979 if (zone_type < ZONE_NORMAL)
2980 low_kmem_size += z->present_pages;
2981 total_size += z->present_pages;
2982 }
2983 }
2984 if (low_kmem_size &&
2985 total_size > average_size && /* ignore small node */
2986 low_kmem_size > total_size * 70/100)
2987 return ZONELIST_ORDER_NODE;
2988 }
2989 return ZONELIST_ORDER_ZONE;
2990}
2991
2992static void set_zonelist_order(void)
2993{
2994 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2995 current_zonelist_order = default_zonelist_order();
2996 else
2997 current_zonelist_order = user_zonelist_order;
2998}
2999
3000static void build_zonelists(pg_data_t *pgdat)
3001{
3002 int j, node, load;
3003 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003005 int local_node, prev_node;
3006 struct zonelist *zonelist;
3007 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003008
3009 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07003010 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07003012 zonelist->_zonerefs[0].zone = NULL;
3013 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014 }
3015
3016 /* NUMA-aware ordering of nodes */
3017 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003018 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003019 prev_node = local_node;
3020 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003021
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003022 memset(node_order, 0, sizeof(node_order));
3023 j = 0;
3024
Linus Torvalds1da177e2005-04-16 15:20:36 -07003025 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08003026 int distance = node_distance(local_node, node);
3027
3028 /*
3029 * If another node is sufficiently far away then it is better
3030 * to reclaim pages in a zone before going off node.
3031 */
3032 if (distance > RECLAIM_DISTANCE)
3033 zone_reclaim_mode = 1;
3034
Linus Torvalds1da177e2005-04-16 15:20:36 -07003035 /*
3036 * We don't want to pressure a particular node.
3037 * So adding penalty to the first node in same
3038 * distance group to make it round-robin.
3039 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08003040 if (distance != node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003041 node_load[node] = load;
3042
Linus Torvalds1da177e2005-04-16 15:20:36 -07003043 prev_node = node;
3044 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003045 if (order == ZONELIST_ORDER_NODE)
3046 build_zonelists_in_node_order(pgdat, node);
3047 else
3048 node_order[j++] = node; /* remember order */
3049 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003050
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003051 if (order == ZONELIST_ORDER_ZONE) {
3052 /* calculate node order -- i.e., DMA last! */
3053 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003054 }
Christoph Lameter523b9452007-10-16 01:25:37 -07003055
3056 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057}
3058
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003059/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003060static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003061{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003062 struct zonelist *zonelist;
3063 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07003064 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003065
Mel Gorman54a6eb52008-04-28 02:12:16 -07003066 zonelist = &pgdat->node_zonelists[0];
3067 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3068 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07003069 for (z = zonelist->_zonerefs; z->zone; z++)
3070 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003071}
3072
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003073#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3074/*
3075 * Return node id of node used for "local" allocations.
3076 * I.e., first node id of first zone in arg node's generic zonelist.
3077 * Used for initializing percpu 'numa_mem', which is used primarily
3078 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3079 */
3080int local_memory_node(int node)
3081{
3082 struct zone *zone;
3083
3084 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3085 gfp_zone(GFP_KERNEL),
3086 NULL,
3087 &zone);
3088 return zone->node;
3089}
3090#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003091
Linus Torvalds1da177e2005-04-16 15:20:36 -07003092#else /* CONFIG_NUMA */
3093
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003094static void set_zonelist_order(void)
3095{
3096 current_zonelist_order = ZONELIST_ORDER_ZONE;
3097}
3098
3099static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003100{
Christoph Lameter19655d32006-09-25 23:31:19 -07003101 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003102 enum zone_type j;
3103 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003104
3105 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003106
Mel Gorman54a6eb52008-04-28 02:12:16 -07003107 zonelist = &pgdat->node_zonelists[0];
3108 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003109
Mel Gorman54a6eb52008-04-28 02:12:16 -07003110 /*
3111 * Now we build the zonelist so that it contains the zones
3112 * of all the other nodes.
3113 * We don't want to pressure a particular node, so when
3114 * building the zones for node N, we make sure that the
3115 * zones coming right after the local ones are those from
3116 * node N+1 (modulo N)
3117 */
3118 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3119 if (!node_online(node))
3120 continue;
3121 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3122 MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003123 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07003124 for (node = 0; node < local_node; node++) {
3125 if (!node_online(node))
3126 continue;
3127 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3128 MAX_NR_ZONES - 1);
3129 }
3130
Mel Gormandd1a2392008-04-28 02:12:17 -07003131 zonelist->_zonerefs[j].zone = NULL;
3132 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003133}
3134
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003135/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003136static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003137{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003138 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003139}
3140
Linus Torvalds1da177e2005-04-16 15:20:36 -07003141#endif /* CONFIG_NUMA */
3142
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003143/*
3144 * Boot pageset table. One per cpu which is going to be used for all
3145 * zones and all nodes. The parameters will be set in such a way
3146 * that an item put on a list will immediately be handed over to
3147 * the buddy list. This is safe since pageset manipulation is done
3148 * with interrupts disabled.
3149 *
3150 * The boot_pagesets must be kept even after bootup is complete for
3151 * unused processors and/or zones. They do play a role for bootstrapping
3152 * hotplugged processors.
3153 *
3154 * zoneinfo_show() and maybe other functions do
3155 * not check if the processor is online before following the pageset pointer.
3156 * Other parts of the kernel may not check if the zone is available.
3157 */
3158static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3159static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07003160static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003161
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003162/*
3163 * Global mutex to protect against size modification of zonelists
3164 * as well as to serialize pageset setup for the new populated zone.
3165 */
3166DEFINE_MUTEX(zonelists_mutex);
3167
Rusty Russell9b1a4d32008-07-28 12:16:30 -05003168/* return values int ....just for stop_machine() */
Haicheng Li1f522502010-05-24 14:32:51 -07003169static __init_refok int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003170{
Yasunori Goto68113782006-06-23 02:03:11 -07003171 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003172 int cpu;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003173
Bo Liu7f9cfb32009-08-18 14:11:19 -07003174#ifdef CONFIG_NUMA
3175 memset(node_load, 0, sizeof(node_load));
3176#endif
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003177 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07003178 pg_data_t *pgdat = NODE_DATA(nid);
3179
3180 build_zonelists(pgdat);
3181 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003182 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003183
3184 /*
3185 * Initialize the boot_pagesets that are going to be used
3186 * for bootstrapping processors. The real pagesets for
3187 * each zone will be allocated later when the per cpu
3188 * allocator is available.
3189 *
3190 * boot_pagesets are used also for bootstrapping offline
3191 * cpus if the system is already booted because the pagesets
3192 * are needed to initialize allocators on a specific cpu too.
3193 * F.e. the percpu allocator needs the page allocator which
3194 * needs the percpu allocator in order to allocate its pagesets
3195 * (a chicken-egg dilemma).
3196 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003197 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003198 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3199
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003200#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3201 /*
3202 * We now know the "local memory node" for each node--
3203 * i.e., the node of the first zone in the generic zonelist.
3204 * Set up numa_mem percpu variable for on-line cpus. During
3205 * boot, only the boot cpu should be on-line; we'll init the
3206 * secondary cpus' numa_mem as they come on-line. During
3207 * node/memory hotplug, we'll fixup all on-line cpus.
3208 */
3209 if (cpu_online(cpu))
3210 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3211#endif
3212 }
3213
Yasunori Goto68113782006-06-23 02:03:11 -07003214 return 0;
3215}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003216
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003217/*
3218 * Called with zonelists_mutex held always
3219 * unless system_state == SYSTEM_BOOTING.
3220 */
Paul Mundt9f6ae442011-04-14 15:21:57 -07003221void __ref build_all_zonelists(void *data)
Yasunori Goto68113782006-06-23 02:03:11 -07003222{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003223 set_zonelist_order();
3224
Yasunori Goto68113782006-06-23 02:03:11 -07003225 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07003226 __build_all_zonelists(NULL);
Mel Gorman68ad8df2008-07-23 21:26:52 -07003227 mminit_verify_zonelist();
Yasunori Goto68113782006-06-23 02:03:11 -07003228 cpuset_init_current_mems_allowed();
3229 } else {
Simon Arlott183ff222007-10-20 01:27:18 +02003230 /* we have to stop all cpus to guarantee there is no user
Yasunori Goto68113782006-06-23 02:03:11 -07003231 of zonelist */
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003232#ifdef CONFIG_MEMORY_HOTPLUG
3233 if (data)
3234 setup_zone_pageset((struct zone *)data);
3235#endif
3236 stop_machine(__build_all_zonelists, NULL, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07003237 /* cpuset refresh routine should be here */
3238 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07003239 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003240 /*
3241 * Disable grouping by mobility if the number of pages in the
3242 * system is too low to allow the mechanism to work. It would be
3243 * more accurate, but expensive to check per-zone. This check is
3244 * made on memory-hotadd so a system can start with mobility
3245 * disabled and enable it later
3246 */
Mel Gormand9c23402007-10-16 01:26:01 -07003247 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003248 page_group_by_mobility_disabled = 1;
3249 else
3250 page_group_by_mobility_disabled = 0;
3251
3252 printk("Built %i zonelists in %s order, mobility grouping %s. "
3253 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003254 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003255 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003256 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003257 vm_total_pages);
3258#ifdef CONFIG_NUMA
3259 printk("Policy zone: %s\n", zone_names[policy_zone]);
3260#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003261}
3262
3263/*
3264 * Helper functions to size the waitqueue hash table.
3265 * Essentially these want to choose hash table sizes sufficiently
3266 * large so that collisions trying to wait on pages are rare.
3267 * But in fact, the number of active page waitqueues on typical
3268 * systems is ridiculously low, less than 200. So this is even
3269 * conservative, even though it seems large.
3270 *
3271 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3272 * waitqueues, i.e. the size of the waitq table given the number of pages.
3273 */
3274#define PAGES_PER_WAITQUEUE 256
3275
Yasunori Gotocca448f2006-06-23 02:03:10 -07003276#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07003277static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003278{
3279 unsigned long size = 1;
3280
3281 pages /= PAGES_PER_WAITQUEUE;
3282
3283 while (size < pages)
3284 size <<= 1;
3285
3286 /*
3287 * Once we have dozens or even hundreds of threads sleeping
3288 * on IO we've got bigger problems than wait queue collision.
3289 * Limit the size of the wait table to a reasonable size.
3290 */
3291 size = min(size, 4096UL);
3292
3293 return max(size, 4UL);
3294}
Yasunori Gotocca448f2006-06-23 02:03:10 -07003295#else
3296/*
3297 * A zone's size might be changed by hot-add, so it is not possible to determine
3298 * a suitable size for its wait_table. So we use the maximum size now.
3299 *
3300 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3301 *
3302 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3303 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3304 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3305 *
3306 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3307 * or more by the traditional way. (See above). It equals:
3308 *
3309 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3310 * ia64(16K page size) : = ( 8G + 4M)byte.
3311 * powerpc (64K page size) : = (32G +16M)byte.
3312 */
3313static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3314{
3315 return 4096UL;
3316}
3317#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003318
3319/*
3320 * This is an integer logarithm so that shifts can be used later
3321 * to extract the more random high bits from the multiplicative
3322 * hash function before the remainder is taken.
3323 */
3324static inline unsigned long wait_table_bits(unsigned long size)
3325{
3326 return ffz(~size);
3327}
3328
3329#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3330
Mel Gorman56fd56b2007-10-16 01:25:58 -07003331/*
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07003332 * Check if a pageblock contains reserved pages
3333 */
3334static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
3335{
3336 unsigned long pfn;
3337
3338 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
3339 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
3340 return 1;
3341 }
3342 return 0;
3343}
3344
3345/*
Mel Gormand9c23402007-10-16 01:26:01 -07003346 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07003347 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3348 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07003349 * higher will lead to a bigger reserve which will get freed as contiguous
3350 * blocks as reclaim kicks in
3351 */
3352static void setup_zone_migrate_reserve(struct zone *zone)
3353{
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07003354 unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003355 struct page *page;
Mel Gorman78986a62009-09-21 17:03:02 -07003356 unsigned long block_migratetype;
3357 int reserve;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003358
3359 /* Get the start pfn, end pfn and the number of blocks to reserve */
3360 start_pfn = zone->zone_start_pfn;
3361 end_pfn = start_pfn + zone->spanned_pages;
Mel Gorman41858962009-06-16 15:32:12 -07003362 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07003363 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003364
Mel Gorman78986a62009-09-21 17:03:02 -07003365 /*
3366 * Reserve blocks are generally in place to help high-order atomic
3367 * allocations that are short-lived. A min_free_kbytes value that
3368 * would result in more than 2 reserve blocks for atomic allocations
3369 * is assumed to be in place to help anti-fragmentation for the
3370 * future allocation of hugepages at runtime.
3371 */
3372 reserve = min(2, reserve);
3373
Mel Gormand9c23402007-10-16 01:26:01 -07003374 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07003375 if (!pfn_valid(pfn))
3376 continue;
3377 page = pfn_to_page(pfn);
3378
Adam Litke344c7902008-09-02 14:35:38 -07003379 /* Watch out for overlapping nodes */
3380 if (page_to_nid(page) != zone_to_nid(zone))
3381 continue;
3382
Mel Gorman56fd56b2007-10-16 01:25:58 -07003383 /* Blocks with reserved pages will never free, skip them. */
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07003384 block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
3385 if (pageblock_is_reserved(pfn, block_end_pfn))
Mel Gorman56fd56b2007-10-16 01:25:58 -07003386 continue;
3387
3388 block_migratetype = get_pageblock_migratetype(page);
3389
3390 /* If this block is reserved, account for it */
3391 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
3392 reserve--;
3393 continue;
3394 }
3395
3396 /* Suitable for reserving if this block is movable */
3397 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
3398 set_pageblock_migratetype(page, MIGRATE_RESERVE);
3399 move_freepages_block(zone, page, MIGRATE_RESERVE);
3400 reserve--;
3401 continue;
3402 }
3403
3404 /*
3405 * If the reserve is met and this is a previous reserved block,
3406 * take it back
3407 */
3408 if (block_migratetype == MIGRATE_RESERVE) {
3409 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
3410 move_freepages_block(zone, page, MIGRATE_MOVABLE);
3411 }
3412 }
3413}
Mel Gormanac0e5b72007-10-16 01:25:58 -07003414
Linus Torvalds1da177e2005-04-16 15:20:36 -07003415/*
3416 * Initially all pages are reserved - free ones are freed
3417 * up by free_all_bootmem() once the early boot process is
3418 * done. Non-atomic initialization, single-pass.
3419 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003420void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08003421 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003422{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003423 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07003424 unsigned long end_pfn = start_pfn + size;
3425 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003426 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003427
Hugh Dickins22b31ee2009-01-06 14:40:09 -08003428 if (highest_memmap_pfn < end_pfn - 1)
3429 highest_memmap_pfn = end_pfn - 1;
3430
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003431 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08003432 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08003433 /*
3434 * There can be holes in boot-time mem_map[]s
3435 * handed to this function. They do not
3436 * exist on hotplugged memory.
3437 */
3438 if (context == MEMMAP_EARLY) {
3439 if (!early_pfn_valid(pfn))
3440 continue;
3441 if (!early_pfn_in_nid(pfn, nid))
3442 continue;
3443 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003444 page = pfn_to_page(pfn);
3445 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07003446 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08003447 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003448 reset_page_mapcount(page);
3449 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003450 /*
3451 * Mark the block movable so that blocks are reserved for
3452 * movable at startup. This will force kernel allocations
3453 * to reserve their blocks rather than leaking throughout
3454 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07003455 * kernel allocations are made. Later some blocks near
3456 * the start are marked MIGRATE_RESERVE by
3457 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003458 *
3459 * bitmap is created for zone's valid pfn range. but memmap
3460 * can be created for invalid pages (for alignment)
3461 * check here not to call set_pageblock_migratetype() against
3462 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003463 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003464 if ((z->zone_start_pfn <= pfn)
3465 && (pfn < z->zone_start_pfn + z->spanned_pages)
3466 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07003467 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003468
Linus Torvalds1da177e2005-04-16 15:20:36 -07003469 INIT_LIST_HEAD(&page->lru);
3470#ifdef WANT_PAGE_VIRTUAL
3471 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3472 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07003473 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003474#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003475 }
3476}
3477
Andi Kleen1e548de2008-02-04 22:29:26 -08003478static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003479{
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003480 int order, t;
3481 for_each_migratetype_order(order, t) {
3482 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003483 zone->free_area[order].nr_free = 0;
3484 }
3485}
3486
3487#ifndef __HAVE_ARCH_MEMMAP_INIT
3488#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08003489 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003490#endif
3491
Sam Ravnborg1d6f4e62008-01-17 15:21:12 -08003492static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003493{
David Howells3a6be872009-05-06 16:03:03 -07003494#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003495 int batch;
3496
3497 /*
3498 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07003499 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003500 *
3501 * OK, so we don't know how big the cache is. So guess.
3502 */
3503 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07003504 if (batch * PAGE_SIZE > 512 * 1024)
3505 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003506 batch /= 4; /* We effectively *= 4 below */
3507 if (batch < 1)
3508 batch = 1;
3509
3510 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11003511 * Clamp the batch to a 2^n - 1 value. Having a power
3512 * of 2 value was found to be more likely to have
3513 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003514 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11003515 * For example if 2 tasks are alternately allocating
3516 * batches of pages, one task can end up with a lot
3517 * of pages of one half of the possible page colors
3518 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003519 */
David Howells91552032009-05-06 16:03:02 -07003520 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07003521
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003522 return batch;
David Howells3a6be872009-05-06 16:03:03 -07003523
3524#else
3525 /* The deferral and batching of frees should be suppressed under NOMMU
3526 * conditions.
3527 *
3528 * The problem is that NOMMU needs to be able to allocate large chunks
3529 * of contiguous memory as there's no hardware page translation to
3530 * assemble apparent contiguous memory from discontiguous pages.
3531 *
3532 * Queueing large contiguous runs of pages for batching, however,
3533 * causes the pages to actually be freed in smaller chunks. As there
3534 * can be a significant delay between the individual batches being
3535 * recycled, this leads to the once large chunks of space being
3536 * fragmented and becoming unavailable for high-order allocations.
3537 */
3538 return 0;
3539#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003540}
3541
Adrian Bunkb69a7282008-07-23 21:28:12 -07003542static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
Christoph Lameter2caaad42005-06-21 17:15:00 -07003543{
3544 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003545 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003546
Magnus Damm1c6fe942005-10-26 01:58:59 -07003547 memset(p, 0, sizeof(*p));
3548
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003549 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003550 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003551 pcp->high = 6 * batch;
3552 pcp->batch = max(1UL, 1 * batch);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003553 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
3554 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07003555}
3556
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003557/*
3558 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3559 * to the value high for the pageset p.
3560 */
3561
3562static void setup_pagelist_highmark(struct per_cpu_pageset *p,
3563 unsigned long high)
3564{
3565 struct per_cpu_pages *pcp;
3566
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003567 pcp = &p->pcp;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003568 pcp->high = high;
3569 pcp->batch = max(1UL, high/4);
3570 if ((high/4) > (PAGE_SHIFT * 8))
3571 pcp->batch = PAGE_SHIFT * 8;
3572}
3573
Nikanth Karthikesan58c2ee42011-03-15 10:59:02 +05303574static void setup_zone_pageset(struct zone *zone)
Wu Fengguang319774e2010-05-24 14:32:49 -07003575{
3576 int cpu;
3577
3578 zone->pageset = alloc_percpu(struct per_cpu_pageset);
3579
3580 for_each_possible_cpu(cpu) {
3581 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
3582
3583 setup_pageset(pcp, zone_batchsize(zone));
3584
3585 if (percpu_pagelist_fraction)
3586 setup_pagelist_highmark(pcp,
3587 (zone->present_pages /
3588 percpu_pagelist_fraction));
3589 }
3590}
3591
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003592/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003593 * Allocate per cpu pagesets and initialize them.
3594 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07003595 */
Al Viro78d99552005-12-15 09:18:25 +00003596void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003597{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003598 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003599
Wu Fengguang319774e2010-05-24 14:32:49 -07003600 for_each_populated_zone(zone)
3601 setup_zone_pageset(zone);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003602}
3603
Sam Ravnborg577a32f2007-05-17 23:29:25 +02003604static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07003605int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07003606{
3607 int i;
3608 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003609 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07003610
3611 /*
3612 * The per-page waitqueue mechanism uses hashed waitqueues
3613 * per zone.
3614 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07003615 zone->wait_table_hash_nr_entries =
3616 wait_table_hash_nr_entries(zone_size_pages);
3617 zone->wait_table_bits =
3618 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003619 alloc_size = zone->wait_table_hash_nr_entries
3620 * sizeof(wait_queue_head_t);
3621
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07003622 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07003623 zone->wait_table = (wait_queue_head_t *)
Yinghai Lu8f389a992011-05-11 15:13:32 -07003624 alloc_bootmem_node_nopanic(pgdat, alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003625 } else {
3626 /*
3627 * This case means that a zone whose size was 0 gets new memory
3628 * via memory hot-add.
3629 * But it may be the case that a new node was hot-added. In
3630 * this case vmalloc() will not be able to use this new node's
3631 * memory - this wait_table must be initialized to use this new
3632 * node itself as well.
3633 * To use this new node's memory, further consideration will be
3634 * necessary.
3635 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07003636 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003637 }
3638 if (!zone->wait_table)
3639 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07003640
Yasunori Goto02b694d2006-06-23 02:03:08 -07003641 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07003642 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003643
3644 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003645}
3646
Shaohua Li112067f2009-09-21 17:01:16 -07003647static int __zone_pcp_update(void *data)
3648{
3649 struct zone *zone = data;
3650 int cpu;
3651 unsigned long batch = zone_batchsize(zone), flags;
3652
Thomas Gleixner2d30a1f2010-03-10 15:20:40 -08003653 for_each_possible_cpu(cpu) {
Shaohua Li112067f2009-09-21 17:01:16 -07003654 struct per_cpu_pageset *pset;
3655 struct per_cpu_pages *pcp;
3656
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003657 pset = per_cpu_ptr(zone->pageset, cpu);
Shaohua Li112067f2009-09-21 17:01:16 -07003658 pcp = &pset->pcp;
3659
3660 local_irq_save(flags);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003661 free_pcppages_bulk(zone, pcp->count, pcp);
Shaohua Li112067f2009-09-21 17:01:16 -07003662 setup_pageset(pset, batch);
3663 local_irq_restore(flags);
3664 }
3665 return 0;
3666}
3667
3668void zone_pcp_update(struct zone *zone)
3669{
3670 stop_machine(__zone_pcp_update, zone, NULL);
3671}
3672
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003673static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07003674{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003675 /*
3676 * per cpu subsystem is not up at this point. The following code
3677 * relies on the ability of the linker to provide the
3678 * offset of a (static) per cpu variable into the per cpu area.
3679 */
3680 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07003681
Anton Blanchardf5335c02006-03-25 03:06:49 -08003682 if (zone->present_pages)
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003683 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
3684 zone->name, zone->present_pages,
3685 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07003686}
3687
Yasunori Goto718127c2006-06-23 02:03:10 -07003688__meminit int init_currently_empty_zone(struct zone *zone,
3689 unsigned long zone_start_pfn,
Dave Hansena2f3aa022007-01-10 23:15:30 -08003690 unsigned long size,
3691 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07003692{
3693 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003694 int ret;
3695 ret = zone_wait_table_init(zone, size);
3696 if (ret)
3697 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07003698 pgdat->nr_zones = zone_idx(zone) + 1;
3699
Dave Hansened8ece22005-10-29 18:16:50 -07003700 zone->zone_start_pfn = zone_start_pfn;
3701
Mel Gorman708614e2008-07-23 21:26:51 -07003702 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3703 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3704 pgdat->node_id,
3705 (unsigned long)zone_idx(zone),
3706 zone_start_pfn, (zone_start_pfn + size));
3707
Andi Kleen1e548de2008-02-04 22:29:26 -08003708 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07003709
3710 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003711}
3712
Mel Gormanc7132162006-09-27 01:49:43 -07003713#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3714/*
3715 * Basic iterator support. Return the first range of PFNs for a node
3716 * Note: nid == MAX_NUMNODES returns first region regardless of node
3717 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003718static int __meminit first_active_region_index_in_nid(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003719{
3720 int i;
3721
3722 for (i = 0; i < nr_nodemap_entries; i++)
3723 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3724 return i;
3725
3726 return -1;
3727}
3728
3729/*
3730 * Basic iterator support. Return the next active range of PFNs for a node
Simon Arlott183ff222007-10-20 01:27:18 +02003731 * Note: nid == MAX_NUMNODES returns next region regardless of node
Mel Gormanc7132162006-09-27 01:49:43 -07003732 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003733static int __meminit next_active_region_index_in_nid(int index, int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003734{
3735 for (index = index + 1; index < nr_nodemap_entries; index++)
3736 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3737 return index;
3738
3739 return -1;
3740}
3741
3742#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3743/*
3744 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3745 * Architectures may implement their own version but if add_active_range()
3746 * was used and there are no special requirements, this is a convenient
3747 * alternative
3748 */
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003749int __meminit __early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07003750{
3751 int i;
3752
3753 for (i = 0; i < nr_nodemap_entries; i++) {
3754 unsigned long start_pfn = early_node_map[i].start_pfn;
3755 unsigned long end_pfn = early_node_map[i].end_pfn;
3756
3757 if (start_pfn <= pfn && pfn < end_pfn)
3758 return early_node_map[i].nid;
3759 }
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003760 /* This is a memory hole */
3761 return -1;
Mel Gormanc7132162006-09-27 01:49:43 -07003762}
3763#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3764
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003765int __meminit early_pfn_to_nid(unsigned long pfn)
3766{
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003767 int nid;
3768
3769 nid = __early_pfn_to_nid(pfn);
3770 if (nid >= 0)
3771 return nid;
3772 /* just returns 0 */
3773 return 0;
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003774}
3775
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003776#ifdef CONFIG_NODES_SPAN_OTHER_NODES
3777bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3778{
3779 int nid;
3780
3781 nid = __early_pfn_to_nid(pfn);
3782 if (nid >= 0 && nid != node)
3783 return false;
3784 return true;
3785}
3786#endif
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003787
Mel Gormanc7132162006-09-27 01:49:43 -07003788/* Basic iterator support to walk early_node_map[] */
3789#define for_each_active_range_index_in_nid(i, nid) \
3790 for (i = first_active_region_index_in_nid(nid); i != -1; \
3791 i = next_active_region_index_in_nid(i, nid))
3792
3793/**
3794 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003795 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3796 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07003797 *
3798 * If an architecture guarantees that all ranges registered with
3799 * add_active_ranges() contain no holes and may be freed, this
3800 * this function may be used instead of calling free_bootmem() manually.
3801 */
3802void __init free_bootmem_with_active_regions(int nid,
3803 unsigned long max_low_pfn)
3804{
3805 int i;
3806
3807 for_each_active_range_index_in_nid(i, nid) {
3808 unsigned long size_pages = 0;
3809 unsigned long end_pfn = early_node_map[i].end_pfn;
3810
3811 if (early_node_map[i].start_pfn >= max_low_pfn)
3812 continue;
3813
3814 if (end_pfn > max_low_pfn)
3815 end_pfn = max_low_pfn;
3816
3817 size_pages = end_pfn - early_node_map[i].start_pfn;
3818 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3819 PFN_PHYS(early_node_map[i].start_pfn),
3820 size_pages << PAGE_SHIFT);
3821 }
3822}
3823
Yinghai Luedbe7d22010-08-25 13:39:16 -07003824#ifdef CONFIG_HAVE_MEMBLOCK
Yinghai Lucc289892011-02-26 13:05:43 +01003825/*
3826 * Basic iterator support. Return the last range of PFNs for a node
3827 * Note: nid == MAX_NUMNODES returns last region regardless of node
3828 */
3829static int __meminit last_active_region_index_in_nid(int nid)
3830{
3831 int i;
3832
3833 for (i = nr_nodemap_entries - 1; i >= 0; i--)
3834 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3835 return i;
3836
3837 return -1;
3838}
3839
3840/*
3841 * Basic iterator support. Return the previous active range of PFNs for a node
3842 * Note: nid == MAX_NUMNODES returns next region regardless of node
3843 */
3844static int __meminit previous_active_region_index_in_nid(int index, int nid)
3845{
3846 for (index = index - 1; index >= 0; index--)
3847 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3848 return index;
3849
3850 return -1;
3851}
3852
3853#define for_each_active_range_index_in_nid_reverse(i, nid) \
3854 for (i = last_active_region_index_in_nid(nid); i != -1; \
3855 i = previous_active_region_index_in_nid(i, nid))
3856
Yinghai Luedbe7d22010-08-25 13:39:16 -07003857u64 __init find_memory_core_early(int nid, u64 size, u64 align,
3858 u64 goal, u64 limit)
3859{
3860 int i;
3861
3862 /* Need to go over early_node_map to find out good range for node */
Yinghai Lu1a4a6782010-12-17 16:59:07 -08003863 for_each_active_range_index_in_nid_reverse(i, nid) {
Yinghai Luedbe7d22010-08-25 13:39:16 -07003864 u64 addr;
3865 u64 ei_start, ei_last;
3866 u64 final_start, final_end;
3867
3868 ei_last = early_node_map[i].end_pfn;
3869 ei_last <<= PAGE_SHIFT;
3870 ei_start = early_node_map[i].start_pfn;
3871 ei_start <<= PAGE_SHIFT;
3872
3873 final_start = max(ei_start, goal);
3874 final_end = min(ei_last, limit);
3875
3876 if (final_start >= final_end)
3877 continue;
3878
3879 addr = memblock_find_in_range(final_start, final_end, size, align);
3880
3881 if (addr == MEMBLOCK_ERROR)
3882 continue;
3883
3884 return addr;
3885 }
3886
3887 return MEMBLOCK_ERROR;
3888}
3889#endif
3890
Yinghai Lu08677212010-02-10 01:20:20 -08003891int __init add_from_early_node_map(struct range *range, int az,
3892 int nr_range, int nid)
3893{
3894 int i;
3895 u64 start, end;
3896
3897 /* need to go over early_node_map to find out good range for node */
3898 for_each_active_range_index_in_nid(i, nid) {
3899 start = early_node_map[i].start_pfn;
3900 end = early_node_map[i].end_pfn;
3901 nr_range = add_range(range, az, nr_range, start, end);
3902 }
3903 return nr_range;
3904}
3905
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003906void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3907{
3908 int i;
Yinghai Lud52d53b2008-06-16 20:10:55 -07003909 int ret;
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003910
Yinghai Lud52d53b2008-06-16 20:10:55 -07003911 for_each_active_range_index_in_nid(i, nid) {
3912 ret = work_fn(early_node_map[i].start_pfn,
3913 early_node_map[i].end_pfn, data);
3914 if (ret)
3915 break;
3916 }
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003917}
Mel Gormanc7132162006-09-27 01:49:43 -07003918/**
3919 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003920 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07003921 *
3922 * If an architecture guarantees that all ranges registered with
3923 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003924 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07003925 */
3926void __init sparse_memory_present_with_active_regions(int nid)
3927{
3928 int i;
3929
3930 for_each_active_range_index_in_nid(i, nid)
3931 memory_present(early_node_map[i].nid,
3932 early_node_map[i].start_pfn,
3933 early_node_map[i].end_pfn);
3934}
3935
3936/**
3937 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003938 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3939 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3940 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07003941 *
3942 * It returns the start and end page frame of a node based on information
3943 * provided by an arch calling add_active_range(). If called for a node
3944 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003945 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07003946 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003947void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003948 unsigned long *start_pfn, unsigned long *end_pfn)
3949{
3950 int i;
3951 *start_pfn = -1UL;
3952 *end_pfn = 0;
3953
3954 for_each_active_range_index_in_nid(i, nid) {
3955 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3956 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3957 }
3958
Christoph Lameter633c0662007-10-16 01:25:37 -07003959 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07003960 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07003961}
3962
3963/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07003964 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3965 * assumption is made that zones within a node are ordered in monotonic
3966 * increasing memory addresses so that the "highest" populated zone is used
3967 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003968static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07003969{
3970 int zone_index;
3971 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3972 if (zone_index == ZONE_MOVABLE)
3973 continue;
3974
3975 if (arch_zone_highest_possible_pfn[zone_index] >
3976 arch_zone_lowest_possible_pfn[zone_index])
3977 break;
3978 }
3979
3980 VM_BUG_ON(zone_index == -1);
3981 movable_zone = zone_index;
3982}
3983
3984/*
3985 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003986 * because it is sized independent of architecture. Unlike the other zones,
Mel Gorman2a1e2742007-07-17 04:03:12 -07003987 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3988 * in each node depending on the size of each node and how evenly kernelcore
3989 * is distributed. This helper function adjusts the zone ranges
3990 * provided by the architecture for a given node by using the end of the
3991 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3992 * zones within a node are in order of monotonic increases memory addresses
3993 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003994static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07003995 unsigned long zone_type,
3996 unsigned long node_start_pfn,
3997 unsigned long node_end_pfn,
3998 unsigned long *zone_start_pfn,
3999 unsigned long *zone_end_pfn)
4000{
4001 /* Only adjust if ZONE_MOVABLE is on this node */
4002 if (zone_movable_pfn[nid]) {
4003 /* Size ZONE_MOVABLE */
4004 if (zone_type == ZONE_MOVABLE) {
4005 *zone_start_pfn = zone_movable_pfn[nid];
4006 *zone_end_pfn = min(node_end_pfn,
4007 arch_zone_highest_possible_pfn[movable_zone]);
4008
4009 /* Adjust for ZONE_MOVABLE starting within this range */
4010 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
4011 *zone_end_pfn > zone_movable_pfn[nid]) {
4012 *zone_end_pfn = zone_movable_pfn[nid];
4013
4014 /* Check if this whole range is within ZONE_MOVABLE */
4015 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
4016 *zone_start_pfn = *zone_end_pfn;
4017 }
4018}
4019
4020/*
Mel Gormanc7132162006-09-27 01:49:43 -07004021 * Return the number of pages a zone spans in a node, including holes
4022 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4023 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004024static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004025 unsigned long zone_type,
4026 unsigned long *ignored)
4027{
4028 unsigned long node_start_pfn, node_end_pfn;
4029 unsigned long zone_start_pfn, zone_end_pfn;
4030
4031 /* Get the start and end of the node and zone */
4032 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
4033 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4034 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07004035 adjust_zone_range_for_zone_movable(nid, zone_type,
4036 node_start_pfn, node_end_pfn,
4037 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004038
4039 /* Check that this node has pages within the zone's required range */
4040 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4041 return 0;
4042
4043 /* Move the zone boundaries inside the node if necessary */
4044 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4045 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4046
4047 /* Return the spanned pages */
4048 return zone_end_pfn - zone_start_pfn;
4049}
4050
4051/*
4052 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004053 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07004054 */
Yinghai Lu32996252009-12-15 17:59:02 -08004055unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004056 unsigned long range_start_pfn,
4057 unsigned long range_end_pfn)
4058{
4059 int i = 0;
4060 unsigned long prev_end_pfn = 0, hole_pages = 0;
4061 unsigned long start_pfn;
4062
4063 /* Find the end_pfn of the first active range of pfns in the node */
4064 i = first_active_region_index_in_nid(nid);
4065 if (i == -1)
4066 return 0;
4067
Mel Gormanb5445f92007-07-26 10:41:18 -07004068 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
4069
Mel Gorman9c7cd682006-09-27 01:49:58 -07004070 /* Account for ranges before physical memory on this node */
4071 if (early_node_map[i].start_pfn > range_start_pfn)
Mel Gormanb5445f92007-07-26 10:41:18 -07004072 hole_pages = prev_end_pfn - range_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004073
4074 /* Find all holes for the zone within the node */
4075 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
4076
4077 /* No need to continue if prev_end_pfn is outside the zone */
4078 if (prev_end_pfn >= range_end_pfn)
4079 break;
4080
4081 /* Make sure the end of the zone is not within the hole */
4082 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
4083 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
4084
4085 /* Update the hole size cound and move on */
4086 if (start_pfn > range_start_pfn) {
4087 BUG_ON(prev_end_pfn > start_pfn);
4088 hole_pages += start_pfn - prev_end_pfn;
4089 }
4090 prev_end_pfn = early_node_map[i].end_pfn;
4091 }
4092
Mel Gorman9c7cd682006-09-27 01:49:58 -07004093 /* Account for ranges past physical memory on this node */
4094 if (range_end_pfn > prev_end_pfn)
Mel Gorman0c6cb972006-10-28 10:38:59 -07004095 hole_pages += range_end_pfn -
Mel Gorman9c7cd682006-09-27 01:49:58 -07004096 max(range_start_pfn, prev_end_pfn);
4097
Mel Gormanc7132162006-09-27 01:49:43 -07004098 return hole_pages;
4099}
4100
4101/**
4102 * absent_pages_in_range - Return number of page frames in holes within a range
4103 * @start_pfn: The start PFN to start searching for holes
4104 * @end_pfn: The end PFN to stop searching for holes
4105 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004106 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07004107 */
4108unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4109 unsigned long end_pfn)
4110{
4111 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4112}
4113
4114/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004115static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004116 unsigned long zone_type,
4117 unsigned long *ignored)
4118{
Mel Gorman9c7cd682006-09-27 01:49:58 -07004119 unsigned long node_start_pfn, node_end_pfn;
4120 unsigned long zone_start_pfn, zone_end_pfn;
4121
4122 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
4123 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
4124 node_start_pfn);
4125 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
4126 node_end_pfn);
4127
Mel Gorman2a1e2742007-07-17 04:03:12 -07004128 adjust_zone_range_for_zone_movable(nid, zone_type,
4129 node_start_pfn, node_end_pfn,
4130 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004131 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004132}
Mel Gorman0e0b8642006-09-27 01:49:56 -07004133
Mel Gormanc7132162006-09-27 01:49:43 -07004134#else
Paul Mundt6ea6e682007-07-15 23:38:20 -07004135static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004136 unsigned long zone_type,
4137 unsigned long *zones_size)
4138{
4139 return zones_size[zone_type];
4140}
4141
Paul Mundt6ea6e682007-07-15 23:38:20 -07004142static inline 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 *zholes_size)
4145{
4146 if (!zholes_size)
4147 return 0;
4148
4149 return zholes_size[zone_type];
4150}
Mel Gorman0e0b8642006-09-27 01:49:56 -07004151
Mel Gormanc7132162006-09-27 01:49:43 -07004152#endif
4153
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004154static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Mel Gormanc7132162006-09-27 01:49:43 -07004155 unsigned long *zones_size, unsigned long *zholes_size)
4156{
4157 unsigned long realtotalpages, totalpages = 0;
4158 enum zone_type i;
4159
4160 for (i = 0; i < MAX_NR_ZONES; i++)
4161 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
4162 zones_size);
4163 pgdat->node_spanned_pages = totalpages;
4164
4165 realtotalpages = totalpages;
4166 for (i = 0; i < MAX_NR_ZONES; i++)
4167 realtotalpages -=
4168 zone_absent_pages_in_node(pgdat->node_id, i,
4169 zholes_size);
4170 pgdat->node_present_pages = realtotalpages;
4171 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4172 realtotalpages);
4173}
4174
Mel Gorman835c1342007-10-16 01:25:47 -07004175#ifndef CONFIG_SPARSEMEM
4176/*
4177 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07004178 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4179 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07004180 * round what is now in bits to nearest long in bits, then return it in
4181 * bytes.
4182 */
4183static unsigned long __init usemap_size(unsigned long zonesize)
4184{
4185 unsigned long usemapsize;
4186
Mel Gormand9c23402007-10-16 01:26:01 -07004187 usemapsize = roundup(zonesize, pageblock_nr_pages);
4188 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07004189 usemapsize *= NR_PAGEBLOCK_BITS;
4190 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4191
4192 return usemapsize / 8;
4193}
4194
4195static void __init setup_usemap(struct pglist_data *pgdat,
4196 struct zone *zone, unsigned long zonesize)
4197{
4198 unsigned long usemapsize = usemap_size(zonesize);
4199 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08004200 if (usemapsize)
Yinghai Lu8f389a992011-05-11 15:13:32 -07004201 zone->pageblock_flags = alloc_bootmem_node_nopanic(pgdat,
4202 usemapsize);
Mel Gorman835c1342007-10-16 01:25:47 -07004203}
4204#else
Jesper Juhlfa9f90b2010-11-28 21:39:34 +01004205static inline void setup_usemap(struct pglist_data *pgdat,
Mel Gorman835c1342007-10-16 01:25:47 -07004206 struct zone *zone, unsigned long zonesize) {}
4207#endif /* CONFIG_SPARSEMEM */
4208
Mel Gormand9c23402007-10-16 01:26:01 -07004209#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08004210
4211/* Return a sensible default order for the pageblock size. */
4212static inline int pageblock_default_order(void)
4213{
4214 if (HPAGE_SHIFT > PAGE_SHIFT)
4215 return HUGETLB_PAGE_ORDER;
4216
4217 return MAX_ORDER-1;
4218}
4219
Mel Gormand9c23402007-10-16 01:26:01 -07004220/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4221static inline void __init set_pageblock_order(unsigned int order)
4222{
4223 /* Check that pageblock_nr_pages has not already been setup */
4224 if (pageblock_order)
4225 return;
4226
4227 /*
4228 * Assume the largest contiguous order of interest is a huge page.
4229 * This value may be variable depending on boot parameters on IA64
4230 */
4231 pageblock_order = order;
4232}
4233#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4234
Mel Gormanba72cb82007-11-28 16:21:13 -08004235/*
4236 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4237 * and pageblock_default_order() are unused as pageblock_order is set
4238 * at compile-time. See include/linux/pageblock-flags.h for the values of
4239 * pageblock_order based on the kernel config
4240 */
4241static inline int pageblock_default_order(unsigned int order)
4242{
4243 return MAX_ORDER-1;
4244}
Mel Gormand9c23402007-10-16 01:26:01 -07004245#define set_pageblock_order(x) do {} while (0)
4246
4247#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4248
Linus Torvalds1da177e2005-04-16 15:20:36 -07004249/*
4250 * Set up the zone data structures:
4251 * - mark all pages reserved
4252 * - mark all memory queues empty
4253 * - clear the memory bitmaps
4254 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08004255static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004256 unsigned long *zones_size, unsigned long *zholes_size)
4257{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07004258 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07004259 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004260 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07004261 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004262
Dave Hansen208d54e2005-10-29 18:16:52 -07004263 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004264 pgdat->nr_zones = 0;
4265 init_waitqueue_head(&pgdat->kswapd_wait);
4266 pgdat->kswapd_max_order = 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004267 pgdat_page_cgroup_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004268
4269 for (j = 0; j < MAX_NR_ZONES; j++) {
4270 struct zone *zone = pgdat->node_zones + j;
Mel Gorman0e0b8642006-09-27 01:49:56 -07004271 unsigned long size, realsize, memmap_pages;
Christoph Lameterb69408e2008-10-18 20:26:14 -07004272 enum lru_list l;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004273
Mel Gormanc7132162006-09-27 01:49:43 -07004274 size = zone_spanned_pages_in_node(nid, j, zones_size);
4275 realsize = size - zone_absent_pages_in_node(nid, j,
4276 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004277
Mel Gorman0e0b8642006-09-27 01:49:56 -07004278 /*
4279 * Adjust realsize so that it accounts for how much memory
4280 * is used by this zone for memmap. This affects the watermark
4281 * and per-cpu initialisations
4282 */
Johannes Weinerf7232152008-05-23 13:04:21 -07004283 memmap_pages =
4284 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
Mel Gorman0e0b8642006-09-27 01:49:56 -07004285 if (realsize >= memmap_pages) {
4286 realsize -= memmap_pages;
Yinghai Lu5594c8c2009-01-06 14:39:14 -08004287 if (memmap_pages)
4288 printk(KERN_DEBUG
4289 " %s zone: %lu pages used for memmap\n",
4290 zone_names[j], memmap_pages);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004291 } else
4292 printk(KERN_WARNING
4293 " %s zone: %lu pages exceeds realsize %lu\n",
4294 zone_names[j], memmap_pages, realsize);
4295
Christoph Lameter62672762007-02-10 01:43:07 -08004296 /* Account for reserved pages */
4297 if (j == 0 && realsize > dma_reserve) {
Mel Gorman0e0b8642006-09-27 01:49:56 -07004298 realsize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07004299 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08004300 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004301 }
4302
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004303 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004304 nr_kernel_pages += realsize;
4305 nr_all_pages += realsize;
4306
4307 zone->spanned_pages = size;
4308 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07004309#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07004310 zone->node = nid;
Christoph Lameter8417bba2006-09-25 23:31:51 -07004311 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07004312 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07004313 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07004314#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004315 zone->name = zone_names[j];
4316 spin_lock_init(&zone->lock);
4317 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07004318 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004319 zone->zone_pgdat = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004320
Dave Hansened8ece22005-10-29 18:16:50 -07004321 zone_pcp_init(zone);
KAMEZAWA Hiroyuki246e87a2011-05-26 16:25:34 -07004322 for_each_lru(l)
Christoph Lameterb69408e2008-10-18 20:26:14 -07004323 INIT_LIST_HEAD(&zone->lru[l].list);
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08004324 zone->reclaim_stat.recent_rotated[0] = 0;
4325 zone->reclaim_stat.recent_rotated[1] = 0;
4326 zone->reclaim_stat.recent_scanned[0] = 0;
4327 zone->reclaim_stat.recent_scanned[1] = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07004328 zap_zone_vm_stats(zone);
David Rientjese815af92007-10-16 23:25:54 -07004329 zone->flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004330 if (!size)
4331 continue;
4332
Mel Gormanba72cb82007-11-28 16:21:13 -08004333 set_pageblock_order(pageblock_default_order());
Mel Gorman835c1342007-10-16 01:25:47 -07004334 setup_usemap(pgdat, zone, size);
Dave Hansena2f3aa022007-01-10 23:15:30 -08004335 ret = init_currently_empty_zone(zone, zone_start_pfn,
4336 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07004337 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07004338 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004339 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004340 }
4341}
4342
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004343static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004344{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004345 /* Skip empty nodes */
4346 if (!pgdat->node_spanned_pages)
4347 return;
4348
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004349#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07004350 /* ia64 gets its own node_mem_map, before this, without bootmem */
4351 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07004352 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004353 struct page *map;
4354
Bob Piccoe984bb42006-05-20 15:00:31 -07004355 /*
4356 * The zone's endpoints aren't required to be MAX_ORDER
4357 * aligned but the node_mem_map endpoints must be in order
4358 * for the buddy allocator to function correctly.
4359 */
4360 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
4361 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
4362 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4363 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07004364 map = alloc_remap(pgdat->node_id, size);
4365 if (!map)
Yinghai Lu8f389a992011-05-11 15:13:32 -07004366 map = alloc_bootmem_node_nopanic(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07004367 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004368 }
Roman Zippel12d810c2007-05-31 00:40:54 -07004369#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07004370 /*
4371 * With no DISCONTIG, the global mem_map is just set as node 0's
4372 */
Mel Gormanc7132162006-09-27 01:49:43 -07004373 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004374 mem_map = NODE_DATA(0)->node_mem_map;
Mel Gormanc7132162006-09-27 01:49:43 -07004375#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
4376 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08004377 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Mel Gormanc7132162006-09-27 01:49:43 -07004378#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4379 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004380#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004381#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004382}
4383
Johannes Weiner9109fb72008-07-23 21:27:20 -07004384void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4385 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004386{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004387 pg_data_t *pgdat = NODE_DATA(nid);
4388
Linus Torvalds1da177e2005-04-16 15:20:36 -07004389 pgdat->node_id = nid;
4390 pgdat->node_start_pfn = node_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004391 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004392
4393 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07004394#ifdef CONFIG_FLAT_NODE_MEM_MAP
4395 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4396 nid, (unsigned long)pgdat,
4397 (unsigned long)pgdat->node_mem_map);
4398#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004399
4400 free_area_init_core(pgdat, zones_size, zholes_size);
4401}
4402
Mel Gormanc7132162006-09-27 01:49:43 -07004403#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07004404
4405#if MAX_NUMNODES > 1
4406/*
4407 * Figure out the number of possible node ids.
4408 */
4409static void __init setup_nr_node_ids(void)
4410{
4411 unsigned int node;
4412 unsigned int highest = 0;
4413
4414 for_each_node_mask(node, node_possible_map)
4415 highest = node;
4416 nr_node_ids = highest + 1;
4417}
4418#else
4419static inline void setup_nr_node_ids(void)
4420{
4421}
4422#endif
4423
Mel Gormanc7132162006-09-27 01:49:43 -07004424/**
4425 * add_active_range - Register a range of PFNs backed by physical memory
4426 * @nid: The node ID the range resides on
4427 * @start_pfn: The start PFN of the available physical memory
4428 * @end_pfn: The end PFN of the available physical memory
4429 *
4430 * These ranges are stored in an early_node_map[] and later used by
4431 * free_area_init_nodes() to calculate zone sizes and holes. If the
4432 * range spans a memory hole, it is up to the architecture to ensure
4433 * the memory is not freed by the bootmem allocator. If possible
4434 * the range being registered will be merged with existing ranges.
4435 */
4436void __init add_active_range(unsigned int nid, unsigned long start_pfn,
4437 unsigned long end_pfn)
4438{
4439 int i;
4440
Mel Gorman6b74ab92008-07-23 21:26:49 -07004441 mminit_dprintk(MMINIT_TRACE, "memory_register",
4442 "Entering add_active_range(%d, %#lx, %#lx) "
4443 "%d entries of %d used\n",
4444 nid, start_pfn, end_pfn,
4445 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
Mel Gormanc7132162006-09-27 01:49:43 -07004446
Mel Gorman2dbb51c2008-07-23 21:26:52 -07004447 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
4448
Mel Gormanc7132162006-09-27 01:49:43 -07004449 /* Merge with existing active regions if possible */
4450 for (i = 0; i < nr_nodemap_entries; i++) {
4451 if (early_node_map[i].nid != nid)
4452 continue;
4453
4454 /* Skip if an existing region covers this new one */
4455 if (start_pfn >= early_node_map[i].start_pfn &&
4456 end_pfn <= early_node_map[i].end_pfn)
4457 return;
4458
4459 /* Merge forward if suitable */
4460 if (start_pfn <= early_node_map[i].end_pfn &&
4461 end_pfn > early_node_map[i].end_pfn) {
4462 early_node_map[i].end_pfn = end_pfn;
4463 return;
4464 }
4465
4466 /* Merge backward if suitable */
Kazuhisa Ichikawad2dbe082010-01-15 17:01:20 -08004467 if (start_pfn < early_node_map[i].start_pfn &&
Mel Gormanc7132162006-09-27 01:49:43 -07004468 end_pfn >= early_node_map[i].start_pfn) {
4469 early_node_map[i].start_pfn = start_pfn;
4470 return;
4471 }
4472 }
4473
4474 /* Check that early_node_map is large enough */
4475 if (i >= MAX_ACTIVE_REGIONS) {
4476 printk(KERN_CRIT "More than %d memory regions, truncating\n",
4477 MAX_ACTIVE_REGIONS);
4478 return;
4479 }
4480
4481 early_node_map[i].nid = nid;
4482 early_node_map[i].start_pfn = start_pfn;
4483 early_node_map[i].end_pfn = end_pfn;
4484 nr_nodemap_entries = i + 1;
4485}
4486
4487/**
Yinghai Lucc1050b2008-06-13 19:08:52 -07004488 * remove_active_range - Shrink an existing registered range of PFNs
Mel Gormanc7132162006-09-27 01:49:43 -07004489 * @nid: The node id the range is on that should be shrunk
Yinghai Lucc1050b2008-06-13 19:08:52 -07004490 * @start_pfn: The new PFN of the range
4491 * @end_pfn: The new PFN of the range
Mel Gormanc7132162006-09-27 01:49:43 -07004492 *
4493 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004494 * The map is kept near the end physical page range that has already been
4495 * registered. This function allows an arch to shrink an existing registered
4496 * range.
Mel Gormanc7132162006-09-27 01:49:43 -07004497 */
Yinghai Lucc1050b2008-06-13 19:08:52 -07004498void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
4499 unsigned long end_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004500{
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004501 int i, j;
4502 int removed = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07004503
Yinghai Lucc1050b2008-06-13 19:08:52 -07004504 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
4505 nid, start_pfn, end_pfn);
4506
Mel Gormanc7132162006-09-27 01:49:43 -07004507 /* Find the old active region end and shrink */
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004508 for_each_active_range_index_in_nid(i, nid) {
Yinghai Lucc1050b2008-06-13 19:08:52 -07004509 if (early_node_map[i].start_pfn >= start_pfn &&
4510 early_node_map[i].end_pfn <= end_pfn) {
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004511 /* clear it */
Yinghai Lucc1050b2008-06-13 19:08:52 -07004512 early_node_map[i].start_pfn = 0;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004513 early_node_map[i].end_pfn = 0;
4514 removed = 1;
4515 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004516 }
Yinghai Lucc1050b2008-06-13 19:08:52 -07004517 if (early_node_map[i].start_pfn < start_pfn &&
4518 early_node_map[i].end_pfn > start_pfn) {
4519 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
4520 early_node_map[i].end_pfn = start_pfn;
4521 if (temp_end_pfn > end_pfn)
4522 add_active_range(nid, end_pfn, temp_end_pfn);
4523 continue;
4524 }
4525 if (early_node_map[i].start_pfn >= start_pfn &&
4526 early_node_map[i].end_pfn > end_pfn &&
4527 early_node_map[i].start_pfn < end_pfn) {
4528 early_node_map[i].start_pfn = end_pfn;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004529 continue;
4530 }
4531 }
4532
4533 if (!removed)
4534 return;
4535
4536 /* remove the blank ones */
4537 for (i = nr_nodemap_entries - 1; i > 0; i--) {
4538 if (early_node_map[i].nid != nid)
4539 continue;
4540 if (early_node_map[i].end_pfn)
4541 continue;
4542 /* we found it, get rid of it */
4543 for (j = i; j < nr_nodemap_entries - 1; j++)
4544 memcpy(&early_node_map[j], &early_node_map[j+1],
4545 sizeof(early_node_map[j]));
4546 j = nr_nodemap_entries - 1;
4547 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
4548 nr_nodemap_entries--;
4549 }
Mel Gormanc7132162006-09-27 01:49:43 -07004550}
4551
4552/**
4553 * remove_all_active_ranges - Remove all currently registered regions
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004554 *
Mel Gormanc7132162006-09-27 01:49:43 -07004555 * During discovery, it may be found that a table like SRAT is invalid
4556 * and an alternative discovery method must be used. This function removes
4557 * all currently registered regions.
4558 */
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004559void __init remove_all_active_ranges(void)
Mel Gormanc7132162006-09-27 01:49:43 -07004560{
4561 memset(early_node_map, 0, sizeof(early_node_map));
4562 nr_nodemap_entries = 0;
4563}
4564
4565/* Compare two active node_active_regions */
4566static int __init cmp_node_active_region(const void *a, const void *b)
4567{
4568 struct node_active_region *arange = (struct node_active_region *)a;
4569 struct node_active_region *brange = (struct node_active_region *)b;
4570
4571 /* Done this way to avoid overflows */
4572 if (arange->start_pfn > brange->start_pfn)
4573 return 1;
4574 if (arange->start_pfn < brange->start_pfn)
4575 return -1;
4576
4577 return 0;
4578}
4579
4580/* sort the node_map by start_pfn */
Yinghai Lu32996252009-12-15 17:59:02 -08004581void __init sort_node_map(void)
Mel Gormanc7132162006-09-27 01:49:43 -07004582{
4583 sort(early_node_map, (size_t)nr_nodemap_entries,
4584 sizeof(struct node_active_region),
4585 cmp_node_active_region, NULL);
4586}
4587
Mel Gormana6af2bc2007-02-10 01:42:57 -08004588/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004589static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07004590{
4591 int i;
Mel Gormana6af2bc2007-02-10 01:42:57 -08004592 unsigned long min_pfn = ULONG_MAX;
Mel Gorman1abbfb42006-11-23 12:01:41 +00004593
Mel Gormanc7132162006-09-27 01:49:43 -07004594 /* Assuming a sorted map, the first range found has the starting pfn */
4595 for_each_active_range_index_in_nid(i, nid)
Mel Gormana6af2bc2007-02-10 01:42:57 -08004596 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004597
Mel Gormana6af2bc2007-02-10 01:42:57 -08004598 if (min_pfn == ULONG_MAX) {
4599 printk(KERN_WARNING
Paul Jackson2bc0d262008-06-22 07:22:17 -07004600 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08004601 return 0;
4602 }
4603
4604 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004605}
4606
4607/**
4608 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4609 *
4610 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004611 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07004612 */
4613unsigned long __init find_min_pfn_with_active_regions(void)
4614{
4615 return find_min_pfn_for_node(MAX_NUMNODES);
4616}
4617
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004618/*
4619 * early_calculate_totalpages()
4620 * Sum pages in active regions for movable zone.
4621 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4622 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07004623static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07004624{
4625 int i;
4626 unsigned long totalpages = 0;
4627
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004628 for (i = 0; i < nr_nodemap_entries; i++) {
4629 unsigned long pages = early_node_map[i].end_pfn -
Mel Gorman7e63efe2007-07-17 04:03:15 -07004630 early_node_map[i].start_pfn;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004631 totalpages += pages;
4632 if (pages)
4633 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
4634 }
4635 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07004636}
4637
Mel Gorman2a1e2742007-07-17 04:03:12 -07004638/*
4639 * Find the PFN the Movable zone begins in each node. Kernel memory
4640 * is spread evenly between nodes as long as the nodes have enough
4641 * memory. When they don't, some nodes will have more kernelcore than
4642 * others
4643 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004644static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004645{
4646 int i, nid;
4647 unsigned long usable_startpfn;
4648 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07004649 /* save the state before borrow the nodemask */
4650 nodemask_t saved_node_state = node_states[N_HIGH_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004651 unsigned long totalpages = early_calculate_totalpages();
4652 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004653
Mel Gorman7e63efe2007-07-17 04:03:15 -07004654 /*
4655 * If movablecore was specified, calculate what size of
4656 * kernelcore that corresponds so that memory usable for
4657 * any allocation type is evenly spread. If both kernelcore
4658 * and movablecore are specified, then the value of kernelcore
4659 * will be used for required_kernelcore if it's greater than
4660 * what movablecore would have allowed.
4661 */
4662 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07004663 unsigned long corepages;
4664
4665 /*
4666 * Round-up so that ZONE_MOVABLE is at least as large as what
4667 * was requested by the user
4668 */
4669 required_movablecore =
4670 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
4671 corepages = totalpages - required_movablecore;
4672
4673 required_kernelcore = max(required_kernelcore, corepages);
4674 }
4675
Mel Gorman2a1e2742007-07-17 04:03:12 -07004676 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4677 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07004678 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004679
4680 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4681 find_usable_zone_for_movable();
4682 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4683
4684restart:
4685 /* Spread kernelcore memory as evenly as possible throughout nodes */
4686 kernelcore_node = required_kernelcore / usable_nodes;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004687 for_each_node_state(nid, N_HIGH_MEMORY) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004688 /*
4689 * Recalculate kernelcore_node if the division per node
4690 * now exceeds what is necessary to satisfy the requested
4691 * amount of memory for the kernel
4692 */
4693 if (required_kernelcore < kernelcore_node)
4694 kernelcore_node = required_kernelcore / usable_nodes;
4695
4696 /*
4697 * As the map is walked, we track how much memory is usable
4698 * by the kernel using kernelcore_remaining. When it is
4699 * 0, the rest of the node is usable by ZONE_MOVABLE
4700 */
4701 kernelcore_remaining = kernelcore_node;
4702
4703 /* Go through each range of PFNs within this node */
4704 for_each_active_range_index_in_nid(i, nid) {
4705 unsigned long start_pfn, end_pfn;
4706 unsigned long size_pages;
4707
4708 start_pfn = max(early_node_map[i].start_pfn,
4709 zone_movable_pfn[nid]);
4710 end_pfn = early_node_map[i].end_pfn;
4711 if (start_pfn >= end_pfn)
4712 continue;
4713
4714 /* Account for what is only usable for kernelcore */
4715 if (start_pfn < usable_startpfn) {
4716 unsigned long kernel_pages;
4717 kernel_pages = min(end_pfn, usable_startpfn)
4718 - start_pfn;
4719
4720 kernelcore_remaining -= min(kernel_pages,
4721 kernelcore_remaining);
4722 required_kernelcore -= min(kernel_pages,
4723 required_kernelcore);
4724
4725 /* Continue if range is now fully accounted */
4726 if (end_pfn <= usable_startpfn) {
4727
4728 /*
4729 * Push zone_movable_pfn to the end so
4730 * that if we have to rebalance
4731 * kernelcore across nodes, we will
4732 * not double account here
4733 */
4734 zone_movable_pfn[nid] = end_pfn;
4735 continue;
4736 }
4737 start_pfn = usable_startpfn;
4738 }
4739
4740 /*
4741 * The usable PFN range for ZONE_MOVABLE is from
4742 * start_pfn->end_pfn. Calculate size_pages as the
4743 * number of pages used as kernelcore
4744 */
4745 size_pages = end_pfn - start_pfn;
4746 if (size_pages > kernelcore_remaining)
4747 size_pages = kernelcore_remaining;
4748 zone_movable_pfn[nid] = start_pfn + size_pages;
4749
4750 /*
4751 * Some kernelcore has been met, update counts and
4752 * break if the kernelcore for this node has been
4753 * satisified
4754 */
4755 required_kernelcore -= min(required_kernelcore,
4756 size_pages);
4757 kernelcore_remaining -= size_pages;
4758 if (!kernelcore_remaining)
4759 break;
4760 }
4761 }
4762
4763 /*
4764 * If there is still required_kernelcore, we do another pass with one
4765 * less node in the count. This will push zone_movable_pfn[nid] further
4766 * along on the nodes that still have memory until kernelcore is
4767 * satisified
4768 */
4769 usable_nodes--;
4770 if (usable_nodes && required_kernelcore > usable_nodes)
4771 goto restart;
4772
4773 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4774 for (nid = 0; nid < MAX_NUMNODES; nid++)
4775 zone_movable_pfn[nid] =
4776 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07004777
4778out:
4779 /* restore the node_state */
4780 node_states[N_HIGH_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004781}
4782
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004783/* Any regular memory on that node ? */
4784static void check_for_regular_memory(pg_data_t *pgdat)
4785{
4786#ifdef CONFIG_HIGHMEM
4787 enum zone_type zone_type;
4788
4789 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4790 struct zone *zone = &pgdat->node_zones[zone_type];
4791 if (zone->present_pages)
4792 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4793 }
4794#endif
4795}
4796
Mel Gormanc7132162006-09-27 01:49:43 -07004797/**
4798 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004799 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07004800 *
4801 * This will call free_area_init_node() for each active node in the system.
4802 * Using the page ranges provided by add_active_range(), the size of each
4803 * zone in each node and their holes is calculated. If the maximum PFN
4804 * between two adjacent zones match, it is assumed that the zone is empty.
4805 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4806 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4807 * starts where the previous one ended. For example, ZONE_DMA32 starts
4808 * at arch_max_dma_pfn.
4809 */
4810void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4811{
4812 unsigned long nid;
Andrew Mortondb991002008-10-15 22:01:07 -07004813 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004814
Mel Gormana6af2bc2007-02-10 01:42:57 -08004815 /* Sort early_node_map as initialisation assumes it is sorted */
4816 sort_node_map();
4817
Mel Gormanc7132162006-09-27 01:49:43 -07004818 /* Record where the zone boundaries are */
4819 memset(arch_zone_lowest_possible_pfn, 0,
4820 sizeof(arch_zone_lowest_possible_pfn));
4821 memset(arch_zone_highest_possible_pfn, 0,
4822 sizeof(arch_zone_highest_possible_pfn));
4823 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4824 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4825 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004826 if (i == ZONE_MOVABLE)
4827 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004828 arch_zone_lowest_possible_pfn[i] =
4829 arch_zone_highest_possible_pfn[i-1];
4830 arch_zone_highest_possible_pfn[i] =
4831 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4832 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07004833 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4834 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4835
4836 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4837 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4838 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004839
Mel Gormanc7132162006-09-27 01:49:43 -07004840 /* Print out the zone ranges */
4841 printk("Zone PFN ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07004842 for (i = 0; i < MAX_NR_ZONES; i++) {
4843 if (i == ZONE_MOVABLE)
4844 continue;
David Rientjes72f0ba02010-03-05 13:42:14 -08004845 printk(" %-8s ", zone_names[i]);
4846 if (arch_zone_lowest_possible_pfn[i] ==
4847 arch_zone_highest_possible_pfn[i])
4848 printk("empty\n");
4849 else
4850 printk("%0#10lx -> %0#10lx\n",
Mel Gormanc7132162006-09-27 01:49:43 -07004851 arch_zone_lowest_possible_pfn[i],
4852 arch_zone_highest_possible_pfn[i]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004853 }
4854
4855 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4856 printk("Movable zone start PFN for each node\n");
4857 for (i = 0; i < MAX_NUMNODES; i++) {
4858 if (zone_movable_pfn[i])
4859 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4860 }
Mel Gormanc7132162006-09-27 01:49:43 -07004861
4862 /* Print out the early_node_map[] */
4863 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4864 for (i = 0; i < nr_nodemap_entries; i++)
Paul Jackson5dab8ec2008-06-25 05:44:40 -07004865 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004866 early_node_map[i].start_pfn,
4867 early_node_map[i].end_pfn);
4868
4869 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07004870 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08004871 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07004872 for_each_online_node(nid) {
4873 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07004874 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07004875 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004876
4877 /* Any memory on that node */
4878 if (pgdat->node_present_pages)
4879 node_set_state(nid, N_HIGH_MEMORY);
4880 check_for_regular_memory(pgdat);
Mel Gormanc7132162006-09-27 01:49:43 -07004881 }
4882}
Mel Gorman2a1e2742007-07-17 04:03:12 -07004883
Mel Gorman7e63efe2007-07-17 04:03:15 -07004884static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004885{
4886 unsigned long long coremem;
4887 if (!p)
4888 return -EINVAL;
4889
4890 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004891 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004892
Mel Gorman7e63efe2007-07-17 04:03:15 -07004893 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07004894 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4895
4896 return 0;
4897}
Mel Gormaned7ed362007-07-17 04:03:14 -07004898
Mel Gorman7e63efe2007-07-17 04:03:15 -07004899/*
4900 * kernelcore=size sets the amount of memory for use for allocations that
4901 * cannot be reclaimed or migrated.
4902 */
4903static int __init cmdline_parse_kernelcore(char *p)
4904{
4905 return cmdline_parse_core(p, &required_kernelcore);
4906}
4907
4908/*
4909 * movablecore=size sets the amount of memory for use for allocations that
4910 * can be reclaimed or migrated.
4911 */
4912static int __init cmdline_parse_movablecore(char *p)
4913{
4914 return cmdline_parse_core(p, &required_movablecore);
4915}
4916
Mel Gormaned7ed362007-07-17 04:03:14 -07004917early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004918early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07004919
Mel Gormanc7132162006-09-27 01:49:43 -07004920#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4921
Mel Gorman0e0b8642006-09-27 01:49:56 -07004922/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004923 * set_dma_reserve - set the specified number of pages reserved in the first zone
4924 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07004925 *
4926 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4927 * In the DMA zone, a significant percentage may be consumed by kernel image
4928 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004929 * function may optionally be used to account for unfreeable pages in the
4930 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4931 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07004932 */
4933void __init set_dma_reserve(unsigned long new_dma_reserve)
4934{
4935 dma_reserve = new_dma_reserve;
4936}
4937
Linus Torvalds1da177e2005-04-16 15:20:36 -07004938void __init free_area_init(unsigned long *zones_size)
4939{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004940 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004941 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4942}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004943
Linus Torvalds1da177e2005-04-16 15:20:36 -07004944static int page_alloc_cpu_notify(struct notifier_block *self,
4945 unsigned long action, void *hcpu)
4946{
4947 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004948
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07004949 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004950 drain_pages(cpu);
4951
4952 /*
4953 * Spill the event counters of the dead processor
4954 * into the current processors event counters.
4955 * This artificially elevates the count of the current
4956 * processor.
4957 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07004958 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004959
4960 /*
4961 * Zero the differential counters of the dead processor
4962 * so that the vm statistics are consistent.
4963 *
4964 * This is only okay since the processor is dead and cannot
4965 * race with what we are doing.
4966 */
Christoph Lameter2244b952006-06-30 01:55:33 -07004967 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004968 }
4969 return NOTIFY_OK;
4970}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004971
4972void __init page_alloc_init(void)
4973{
4974 hotcpu_notifier(page_alloc_cpu_notify, 0);
4975}
4976
4977/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004978 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4979 * or min_free_kbytes changes.
4980 */
4981static void calculate_totalreserve_pages(void)
4982{
4983 struct pglist_data *pgdat;
4984 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004985 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004986
4987 for_each_online_pgdat(pgdat) {
4988 for (i = 0; i < MAX_NR_ZONES; i++) {
4989 struct zone *zone = pgdat->node_zones + i;
4990 unsigned long max = 0;
4991
4992 /* Find valid and maximum lowmem_reserve in the zone */
4993 for (j = i; j < MAX_NR_ZONES; j++) {
4994 if (zone->lowmem_reserve[j] > max)
4995 max = zone->lowmem_reserve[j];
4996 }
4997
Mel Gorman41858962009-06-16 15:32:12 -07004998 /* we treat the high watermark as reserved pages. */
4999 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005000
5001 if (max > zone->present_pages)
5002 max = zone->present_pages;
5003 reserve_pages += max;
5004 }
5005 }
5006 totalreserve_pages = reserve_pages;
5007}
5008
5009/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07005010 * setup_per_zone_lowmem_reserve - called whenever
5011 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
5012 * has a correct pages reserved value, so an adequate number of
5013 * pages are left in the zone after a successful __alloc_pages().
5014 */
5015static void setup_per_zone_lowmem_reserve(void)
5016{
5017 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005018 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005019
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08005020 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005021 for (j = 0; j < MAX_NR_ZONES; j++) {
5022 struct zone *zone = pgdat->node_zones + j;
5023 unsigned long present_pages = zone->present_pages;
5024
5025 zone->lowmem_reserve[j] = 0;
5026
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005027 idx = j;
5028 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005029 struct zone *lower_zone;
5030
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005031 idx--;
5032
Linus Torvalds1da177e2005-04-16 15:20:36 -07005033 if (sysctl_lowmem_reserve_ratio[idx] < 1)
5034 sysctl_lowmem_reserve_ratio[idx] = 1;
5035
5036 lower_zone = pgdat->node_zones + idx;
5037 lower_zone->lowmem_reserve[j] = present_pages /
5038 sysctl_lowmem_reserve_ratio[idx];
5039 present_pages += lower_zone->present_pages;
5040 }
5041 }
5042 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005043
5044 /* update totalreserve_pages */
5045 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005046}
5047
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005048/**
Minchan Kimbc75d332009-06-16 15:32:48 -07005049 * setup_per_zone_wmarks - called when min_free_kbytes changes
Minchan Kimbce73942009-06-16 15:32:50 -07005050 * or when memory is hot-{added|removed}
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005051 *
Minchan Kimbc75d332009-06-16 15:32:48 -07005052 * Ensures that the watermark[min,low,high] values for each zone are set
5053 * correctly with respect to min_free_kbytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005054 */
Minchan Kimbc75d332009-06-16 15:32:48 -07005055void setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005056{
5057 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
5058 unsigned long lowmem_pages = 0;
5059 struct zone *zone;
5060 unsigned long flags;
5061
5062 /* Calculate total number of !ZONE_HIGHMEM pages */
5063 for_each_zone(zone) {
5064 if (!is_highmem(zone))
5065 lowmem_pages += zone->present_pages;
5066 }
5067
5068 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07005069 u64 tmp;
5070
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005071 spin_lock_irqsave(&zone->lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07005072 tmp = (u64)pages_min * zone->present_pages;
5073 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005074 if (is_highmem(zone)) {
5075 /*
Nick Piggin669ed172005-11-13 16:06:45 -08005076 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5077 * need highmem pages, so cap pages_min to a small
5078 * value here.
5079 *
Mel Gorman41858962009-06-16 15:32:12 -07005080 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Nick Piggin669ed172005-11-13 16:06:45 -08005081 * deltas controls asynch page reclaim, and so should
5082 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005083 */
5084 int min_pages;
5085
5086 min_pages = zone->present_pages / 1024;
5087 if (min_pages < SWAP_CLUSTER_MAX)
5088 min_pages = SWAP_CLUSTER_MAX;
5089 if (min_pages > 128)
5090 min_pages = 128;
Mel Gorman41858962009-06-16 15:32:12 -07005091 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005092 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08005093 /*
5094 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07005095 * proportionate to the zone's size.
5096 */
Mel Gorman41858962009-06-16 15:32:12 -07005097 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005098 }
5099
Mel Gorman41858962009-06-16 15:32:12 -07005100 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
5101 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Mel Gorman56fd56b2007-10-16 01:25:58 -07005102 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005103 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005104 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005105
5106 /* update totalreserve_pages */
5107 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005108}
5109
Randy Dunlap55a44622009-09-21 17:01:20 -07005110/*
Rik van Riel556adec2008-10-18 20:26:34 -07005111 * The inactive anon list should be small enough that the VM never has to
5112 * do too much work, but large enough that each inactive page has a chance
5113 * to be referenced again before it is swapped out.
5114 *
5115 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5116 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5117 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5118 * the anonymous pages are kept on the inactive list.
5119 *
5120 * total target max
5121 * memory ratio inactive anon
5122 * -------------------------------------
5123 * 10MB 1 5MB
5124 * 100MB 1 50MB
5125 * 1GB 3 250MB
5126 * 10GB 10 0.9GB
5127 * 100GB 31 3GB
5128 * 1TB 101 10GB
5129 * 10TB 320 32GB
5130 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005131static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
Minchan Kim96cb4df2009-06-16 15:32:49 -07005132{
5133 unsigned int gb, ratio;
5134
5135 /* Zone size in gigabytes */
5136 gb = zone->present_pages >> (30 - PAGE_SHIFT);
5137 if (gb)
5138 ratio = int_sqrt(10 * gb);
5139 else
5140 ratio = 1;
5141
5142 zone->inactive_ratio = ratio;
5143}
5144
KOSAKI Motohiro839a4fc2011-05-24 17:11:31 -07005145static void __meminit setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07005146{
5147 struct zone *zone;
5148
Minchan Kim96cb4df2009-06-16 15:32:49 -07005149 for_each_zone(zone)
5150 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07005151}
5152
Linus Torvalds1da177e2005-04-16 15:20:36 -07005153/*
5154 * Initialise min_free_kbytes.
5155 *
5156 * For small machines we want it small (128k min). For large machines
5157 * we want it large (64MB max). But it is not linear, because network
5158 * bandwidth does not increase linearly with machine size. We use
5159 *
5160 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5161 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5162 *
5163 * which yields
5164 *
5165 * 16MB: 512k
5166 * 32MB: 724k
5167 * 64MB: 1024k
5168 * 128MB: 1448k
5169 * 256MB: 2048k
5170 * 512MB: 2896k
5171 * 1024MB: 4096k
5172 * 2048MB: 5792k
5173 * 4096MB: 8192k
5174 * 8192MB: 11584k
5175 * 16384MB: 16384k
5176 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005177int __meminit init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005178{
5179 unsigned long lowmem_kbytes;
5180
5181 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
5182
5183 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
5184 if (min_free_kbytes < 128)
5185 min_free_kbytes = 128;
5186 if (min_free_kbytes > 65536)
5187 min_free_kbytes = 65536;
Minchan Kimbc75d332009-06-16 15:32:48 -07005188 setup_per_zone_wmarks();
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -07005189 refresh_zone_stat_thresholds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005190 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07005191 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005192 return 0;
5193}
Minchan Kimbc75d332009-06-16 15:32:48 -07005194module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005195
5196/*
5197 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5198 * that we can call two helper functions whenever min_free_kbytes
5199 * changes.
5200 */
5201int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005202 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005203{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005204 proc_dointvec(table, write, buffer, length, ppos);
Mel Gorman3b1d92c2007-05-06 14:49:30 -07005205 if (write)
Minchan Kimbc75d332009-06-16 15:32:48 -07005206 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005207 return 0;
5208}
5209
Christoph Lameter96146342006-07-03 00:24:13 -07005210#ifdef CONFIG_NUMA
5211int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005212 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07005213{
5214 struct zone *zone;
5215 int rc;
5216
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005217 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07005218 if (rc)
5219 return rc;
5220
5221 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07005222 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07005223 sysctl_min_unmapped_ratio) / 100;
5224 return 0;
5225}
Christoph Lameter0ff38492006-09-25 23:31:52 -07005226
5227int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005228 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07005229{
5230 struct zone *zone;
5231 int rc;
5232
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005233 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07005234 if (rc)
5235 return rc;
5236
5237 for_each_zone(zone)
5238 zone->min_slab_pages = (zone->present_pages *
5239 sysctl_min_slab_ratio) / 100;
5240 return 0;
5241}
Christoph Lameter96146342006-07-03 00:24:13 -07005242#endif
5243
Linus Torvalds1da177e2005-04-16 15:20:36 -07005244/*
5245 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5246 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5247 * whenever sysctl_lowmem_reserve_ratio changes.
5248 *
5249 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07005250 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07005251 * if in function of the boot time zone sizes.
5252 */
5253int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005254 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005255{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005256 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005257 setup_per_zone_lowmem_reserve();
5258 return 0;
5259}
5260
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005261/*
5262 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5263 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5264 * can have before it gets flushed back to buddy allocator.
5265 */
5266
5267int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005268 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005269{
5270 struct zone *zone;
5271 unsigned int cpu;
5272 int ret;
5273
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005274 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005275 if (!write || (ret == -EINVAL))
5276 return ret;
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07005277 for_each_populated_zone(zone) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005278 for_each_possible_cpu(cpu) {
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005279 unsigned long high;
5280 high = zone->present_pages / percpu_pagelist_fraction;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005281 setup_pagelist_highmark(
5282 per_cpu_ptr(zone->pageset, cpu), high);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005283 }
5284 }
5285 return 0;
5286}
5287
David S. Millerf034b5d2006-08-24 03:08:07 -07005288int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005289
5290#ifdef CONFIG_NUMA
5291static int __init set_hashdist(char *str)
5292{
5293 if (!str)
5294 return 0;
5295 hashdist = simple_strtoul(str, &str, 0);
5296 return 1;
5297}
5298__setup("hashdist=", set_hashdist);
5299#endif
5300
5301/*
5302 * allocate a large system hash table from bootmem
5303 * - it is assumed that the hash table must contain an exact power-of-2
5304 * quantity of entries
5305 * - limit is the number of hash buckets, not the total allocation size
5306 */
5307void *__init alloc_large_system_hash(const char *tablename,
5308 unsigned long bucketsize,
5309 unsigned long numentries,
5310 int scale,
5311 int flags,
5312 unsigned int *_hash_shift,
5313 unsigned int *_hash_mask,
5314 unsigned long limit)
5315{
5316 unsigned long long max = limit;
5317 unsigned long log2qty, size;
5318 void *table = NULL;
5319
5320 /* allow the kernel cmdline to have a say */
5321 if (!numentries) {
5322 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08005323 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005324 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
5325 numentries >>= 20 - PAGE_SHIFT;
5326 numentries <<= 20 - PAGE_SHIFT;
5327
5328 /* limit to 1 bucket per 2^scale bytes of low memory */
5329 if (scale > PAGE_SHIFT)
5330 numentries >>= (scale - PAGE_SHIFT);
5331 else
5332 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08005333
5334 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07005335 if (unlikely(flags & HASH_SMALL)) {
5336 /* Makes no sense without HASH_EARLY */
5337 WARN_ON(!(flags & HASH_EARLY));
5338 if (!(numentries >> *_hash_shift)) {
5339 numentries = 1UL << *_hash_shift;
5340 BUG_ON(!numentries);
5341 }
5342 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08005343 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005344 }
John Hawkes6e692ed2006-03-25 03:08:02 -08005345 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005346
5347 /* limit allocation size to 1/16 total memory by default */
5348 if (max == 0) {
5349 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
5350 do_div(max, bucketsize);
5351 }
5352
5353 if (numentries > max)
5354 numentries = max;
5355
David Howellsf0d1b0b2006-12-08 02:37:49 -08005356 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005357
5358 do {
5359 size = bucketsize << log2qty;
5360 if (flags & HASH_EARLY)
Jan Beulich74768ed2008-08-12 15:08:39 -07005361 table = alloc_bootmem_nopanic(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005362 else if (hashdist)
5363 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
5364 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07005365 /*
5366 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07005367 * some pages at the end of hash table which
5368 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07005369 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005370 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07005371 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005372 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
5373 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005374 }
5375 } while (!table && size > PAGE_SIZE && --log2qty);
5376
5377 if (!table)
5378 panic("Failed to allocate %s hash table\n", tablename);
5379
Robin Holtf241e6602010-10-07 12:59:26 -07005380 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07005381 tablename,
Robin Holtf241e6602010-10-07 12:59:26 -07005382 (1UL << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08005383 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005384 size);
5385
5386 if (_hash_shift)
5387 *_hash_shift = log2qty;
5388 if (_hash_mask)
5389 *_hash_mask = (1 << log2qty) - 1;
5390
5391 return table;
5392}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08005393
Mel Gorman835c1342007-10-16 01:25:47 -07005394/* Return a pointer to the bitmap storing bits affecting a block of pages */
5395static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
5396 unsigned long pfn)
5397{
5398#ifdef CONFIG_SPARSEMEM
5399 return __pfn_to_section(pfn)->pageblock_flags;
5400#else
5401 return zone->pageblock_flags;
5402#endif /* CONFIG_SPARSEMEM */
5403}
Andrew Morton6220ec72006-10-19 23:29:05 -07005404
Mel Gorman835c1342007-10-16 01:25:47 -07005405static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
5406{
5407#ifdef CONFIG_SPARSEMEM
5408 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07005409 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005410#else
5411 pfn = pfn - zone->zone_start_pfn;
Mel Gormand9c23402007-10-16 01:26:01 -07005412 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005413#endif /* CONFIG_SPARSEMEM */
5414}
5415
5416/**
Mel Gormand9c23402007-10-16 01:26:01 -07005417 * 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 -07005418 * @page: The page within the block of interest
5419 * @start_bitidx: The first bit of interest to retrieve
5420 * @end_bitidx: The last bit of interest
5421 * returns pageblock_bits flags
5422 */
5423unsigned long get_pageblock_flags_group(struct page *page,
5424 int start_bitidx, int end_bitidx)
5425{
5426 struct zone *zone;
5427 unsigned long *bitmap;
5428 unsigned long pfn, bitidx;
5429 unsigned long flags = 0;
5430 unsigned long value = 1;
5431
5432 zone = page_zone(page);
5433 pfn = page_to_pfn(page);
5434 bitmap = get_pageblock_bitmap(zone, pfn);
5435 bitidx = pfn_to_bitidx(zone, pfn);
5436
5437 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5438 if (test_bit(bitidx + start_bitidx, bitmap))
5439 flags |= value;
5440
5441 return flags;
5442}
5443
5444/**
Mel Gormand9c23402007-10-16 01:26:01 -07005445 * 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 -07005446 * @page: The page within the block of interest
5447 * @start_bitidx: The first bit of interest
5448 * @end_bitidx: The last bit of interest
5449 * @flags: The flags to set
5450 */
5451void set_pageblock_flags_group(struct page *page, unsigned long flags,
5452 int start_bitidx, int end_bitidx)
5453{
5454 struct zone *zone;
5455 unsigned long *bitmap;
5456 unsigned long pfn, bitidx;
5457 unsigned long value = 1;
5458
5459 zone = page_zone(page);
5460 pfn = page_to_pfn(page);
5461 bitmap = get_pageblock_bitmap(zone, pfn);
5462 bitidx = pfn_to_bitidx(zone, pfn);
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07005463 VM_BUG_ON(pfn < zone->zone_start_pfn);
5464 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
Mel Gorman835c1342007-10-16 01:25:47 -07005465
5466 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5467 if (flags & value)
5468 __set_bit(bitidx + start_bitidx, bitmap);
5469 else
5470 __clear_bit(bitidx + start_bitidx, bitmap);
5471}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005472
5473/*
5474 * This is designed as sub function...plz see page_isolation.c also.
5475 * set/clear page block's type to be ISOLATE.
5476 * page allocater never alloc memory from ISOLATE block.
5477 */
5478
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005479static int
5480__count_immobile_pages(struct zone *zone, struct page *page, int count)
5481{
5482 unsigned long pfn, iter, found;
5483 /*
5484 * For avoiding noise data, lru_add_drain_all() should be called
5485 * If ZONE_MOVABLE, the zone never contains immobile pages
5486 */
5487 if (zone_idx(zone) == ZONE_MOVABLE)
5488 return true;
5489
5490 if (get_pageblock_migratetype(page) == MIGRATE_MOVABLE)
5491 return true;
5492
5493 pfn = page_to_pfn(page);
5494 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
5495 unsigned long check = pfn + iter;
5496
Namhyung Kim29723fc2011-02-25 14:44:25 -08005497 if (!pfn_valid_within(check))
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005498 continue;
Namhyung Kim29723fc2011-02-25 14:44:25 -08005499
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005500 page = pfn_to_page(check);
5501 if (!page_count(page)) {
5502 if (PageBuddy(page))
5503 iter += (1 << page_order(page)) - 1;
5504 continue;
5505 }
5506 if (!PageLRU(page))
5507 found++;
5508 /*
5509 * If there are RECLAIMABLE pages, we need to check it.
5510 * But now, memory offline itself doesn't call shrink_slab()
5511 * and it still to be fixed.
5512 */
5513 /*
5514 * If the page is not RAM, page_count()should be 0.
5515 * we don't need more check. This is an _used_ not-movable page.
5516 *
5517 * The problematic thing here is PG_reserved pages. PG_reserved
5518 * is set to both of a memory hole page and a _used_ kernel
5519 * page at boot.
5520 */
5521 if (found > count)
5522 return false;
5523 }
5524 return true;
5525}
5526
5527bool is_pageblock_removable_nolock(struct page *page)
5528{
5529 struct zone *zone = page_zone(page);
5530 return __count_immobile_pages(zone, page, 0);
5531}
5532
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005533int set_migratetype_isolate(struct page *page)
5534{
5535 struct zone *zone;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005536 unsigned long flags, pfn;
Robert Jennings925cc712009-12-17 14:44:38 +00005537 struct memory_isolate_notify arg;
5538 int notifier_ret;
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005539 int ret = -EBUSY;
5540
5541 zone = page_zone(page);
Robert Jennings925cc712009-12-17 14:44:38 +00005542
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005543 spin_lock_irqsave(&zone->lock, flags);
Robert Jennings925cc712009-12-17 14:44:38 +00005544
5545 pfn = page_to_pfn(page);
5546 arg.start_pfn = pfn;
5547 arg.nr_pages = pageblock_nr_pages;
5548 arg.pages_found = 0;
5549
5550 /*
5551 * It may be possible to isolate a pageblock even if the
5552 * migratetype is not MIGRATE_MOVABLE. The memory isolation
5553 * notifier chain is used by balloon drivers to return the
5554 * number of pages in a range that are held by the balloon
5555 * driver to shrink memory. If all the pages are accounted for
5556 * by balloons, are free, or on the LRU, isolation can continue.
5557 * Later, for example, when memory hotplug notifier runs, these
5558 * pages reported as "can be isolated" should be isolated(freed)
5559 * by the balloon driver through the memory notifier chain.
5560 */
5561 notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg);
5562 notifier_ret = notifier_to_errno(notifier_ret);
KAMEZAWA Hiroyuki4b204772010-10-26 14:21:29 -07005563 if (notifier_ret)
Robert Jennings925cc712009-12-17 14:44:38 +00005564 goto out;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005565 /*
5566 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
5567 * We just check MOVABLE pages.
5568 */
5569 if (__count_immobile_pages(zone, page, arg.pages_found))
Robert Jennings925cc712009-12-17 14:44:38 +00005570 ret = 0;
5571
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005572 /*
5573 * immobile means "not-on-lru" paes. If immobile is larger than
5574 * removable-by-driver pages reported by notifier, we'll fail.
5575 */
5576
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005577out:
Robert Jennings925cc712009-12-17 14:44:38 +00005578 if (!ret) {
5579 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
5580 move_freepages_block(zone, page, MIGRATE_ISOLATE);
5581 }
5582
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005583 spin_unlock_irqrestore(&zone->lock, flags);
5584 if (!ret)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005585 drain_all_pages();
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005586 return ret;
5587}
5588
5589void unset_migratetype_isolate(struct page *page)
5590{
5591 struct zone *zone;
5592 unsigned long flags;
5593 zone = page_zone(page);
5594 spin_lock_irqsave(&zone->lock, flags);
5595 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
5596 goto out;
5597 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5598 move_freepages_block(zone, page, MIGRATE_MOVABLE);
5599out:
5600 spin_unlock_irqrestore(&zone->lock, flags);
5601}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07005602
5603#ifdef CONFIG_MEMORY_HOTREMOVE
5604/*
5605 * All pages in the range must be isolated before calling this.
5606 */
5607void
5608__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
5609{
5610 struct page *page;
5611 struct zone *zone;
5612 int order, i;
5613 unsigned long pfn;
5614 unsigned long flags;
5615 /* find the first valid pfn */
5616 for (pfn = start_pfn; pfn < end_pfn; pfn++)
5617 if (pfn_valid(pfn))
5618 break;
5619 if (pfn == end_pfn)
5620 return;
5621 zone = page_zone(pfn_to_page(pfn));
5622 spin_lock_irqsave(&zone->lock, flags);
5623 pfn = start_pfn;
5624 while (pfn < end_pfn) {
5625 if (!pfn_valid(pfn)) {
5626 pfn++;
5627 continue;
5628 }
5629 page = pfn_to_page(pfn);
5630 BUG_ON(page_count(page));
5631 BUG_ON(!PageBuddy(page));
5632 order = page_order(page);
5633#ifdef CONFIG_DEBUG_VM
5634 printk(KERN_INFO "remove from free list %lx %d %lx\n",
5635 pfn, 1 << order, end_pfn);
5636#endif
5637 list_del(&page->lru);
5638 rmv_page_order(page);
5639 zone->free_area[order].nr_free--;
5640 __mod_zone_page_state(zone, NR_FREE_PAGES,
5641 - (1UL << order));
5642 for (i = 0; i < (1 << order); i++)
5643 SetPageReserved((page+i));
5644 pfn += (1 << order);
5645 }
5646 spin_unlock_irqrestore(&zone->lock, flags);
5647}
5648#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01005649
5650#ifdef CONFIG_MEMORY_FAILURE
5651bool is_free_buddy_page(struct page *page)
5652{
5653 struct zone *zone = page_zone(page);
5654 unsigned long pfn = page_to_pfn(page);
5655 unsigned long flags;
5656 int order;
5657
5658 spin_lock_irqsave(&zone->lock, flags);
5659 for (order = 0; order < MAX_ORDER; order++) {
5660 struct page *page_head = page - (pfn & ((1 << order) - 1));
5661
5662 if (PageBuddy(page_head) && page_order(page_head) >= order)
5663 break;
5664 }
5665 spin_unlock_irqrestore(&zone->lock, flags);
5666
5667 return order < MAX_ORDER;
5668}
5669#endif
Wu Fengguang718a3822010-03-10 15:20:43 -08005670
5671static struct trace_print_flags pageflag_names[] = {
5672 {1UL << PG_locked, "locked" },
5673 {1UL << PG_error, "error" },
5674 {1UL << PG_referenced, "referenced" },
5675 {1UL << PG_uptodate, "uptodate" },
5676 {1UL << PG_dirty, "dirty" },
5677 {1UL << PG_lru, "lru" },
5678 {1UL << PG_active, "active" },
5679 {1UL << PG_slab, "slab" },
5680 {1UL << PG_owner_priv_1, "owner_priv_1" },
5681 {1UL << PG_arch_1, "arch_1" },
5682 {1UL << PG_reserved, "reserved" },
5683 {1UL << PG_private, "private" },
5684 {1UL << PG_private_2, "private_2" },
5685 {1UL << PG_writeback, "writeback" },
5686#ifdef CONFIG_PAGEFLAGS_EXTENDED
5687 {1UL << PG_head, "head" },
5688 {1UL << PG_tail, "tail" },
5689#else
5690 {1UL << PG_compound, "compound" },
5691#endif
5692 {1UL << PG_swapcache, "swapcache" },
5693 {1UL << PG_mappedtodisk, "mappedtodisk" },
5694 {1UL << PG_reclaim, "reclaim" },
Wu Fengguang718a3822010-03-10 15:20:43 -08005695 {1UL << PG_swapbacked, "swapbacked" },
5696 {1UL << PG_unevictable, "unevictable" },
5697#ifdef CONFIG_MMU
5698 {1UL << PG_mlocked, "mlocked" },
5699#endif
5700#ifdef CONFIG_ARCH_USES_PG_UNCACHED
5701 {1UL << PG_uncached, "uncached" },
5702#endif
5703#ifdef CONFIG_MEMORY_FAILURE
5704 {1UL << PG_hwpoison, "hwpoison" },
5705#endif
5706 {-1UL, NULL },
5707};
5708
5709static void dump_page_flags(unsigned long flags)
5710{
5711 const char *delim = "";
5712 unsigned long mask;
5713 int i;
5714
5715 printk(KERN_ALERT "page flags: %#lx(", flags);
5716
5717 /* remove zone id */
5718 flags &= (1UL << NR_PAGEFLAGS) - 1;
5719
5720 for (i = 0; pageflag_names[i].name && flags; i++) {
5721
5722 mask = pageflag_names[i].mask;
5723 if ((flags & mask) != mask)
5724 continue;
5725
5726 flags &= ~mask;
5727 printk("%s%s", delim, pageflag_names[i].name);
5728 delim = "|";
5729 }
5730
5731 /* check for left over flags */
5732 if (flags)
5733 printk("%s%#lx", delim, flags);
5734
5735 printk(")\n");
5736}
5737
5738void dump_page(struct page *page)
5739{
5740 printk(KERN_ALERT
5741 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
Andrea Arcangeli4e9f64c2011-01-13 15:46:29 -08005742 page, atomic_read(&page->_count), page_mapcount(page),
Wu Fengguang718a3822010-03-10 15:20:43 -08005743 page->mapping, page->index);
5744 dump_page_flags(page->flags);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07005745 mem_cgroup_print_bad_page(page);
Wu Fengguang718a3822010-03-10 15:20:43 -08005746}